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>
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 LIST_HEAD(shmem_swaplist
);
195 static DEFINE_MUTEX(shmem_swaplist_mutex
);
197 static int shmem_reserve_inode(struct super_block
*sb
)
199 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
200 if (sbinfo
->max_inodes
) {
201 spin_lock(&sbinfo
->stat_lock
);
202 if (!sbinfo
->free_inodes
) {
203 spin_unlock(&sbinfo
->stat_lock
);
206 sbinfo
->free_inodes
--;
207 spin_unlock(&sbinfo
->stat_lock
);
212 static void shmem_free_inode(struct super_block
*sb
)
214 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
215 if (sbinfo
->max_inodes
) {
216 spin_lock(&sbinfo
->stat_lock
);
217 sbinfo
->free_inodes
++;
218 spin_unlock(&sbinfo
->stat_lock
);
223 * shmem_recalc_inode - recalculate the block usage of an inode
224 * @inode: inode to recalc
226 * We have to calculate the free blocks since the mm can drop
227 * undirtied hole pages behind our back.
229 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
230 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
232 * It has to be called with the spinlock held.
234 static void shmem_recalc_inode(struct inode
*inode
)
236 struct shmem_inode_info
*info
= SHMEM_I(inode
);
239 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
241 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
242 if (sbinfo
->max_blocks
)
243 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
244 info
->alloced
-= freed
;
245 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
246 shmem_unacct_blocks(info
->flags
, freed
);
251 * Replace item expected in radix tree by a new item, while holding tree lock.
253 static int shmem_radix_tree_replace(struct address_space
*mapping
,
254 pgoff_t index
, void *expected
, void *replacement
)
259 VM_BUG_ON(!expected
);
260 VM_BUG_ON(!replacement
);
261 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
264 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
265 if (item
!= expected
)
267 radix_tree_replace_slot(pslot
, replacement
);
272 * Sometimes, before we decide whether to proceed or to fail, we must check
273 * that an entry was not already brought back from swap by a racing thread.
275 * Checking page is not enough: by the time a SwapCache page is locked, it
276 * might be reused, and again be SwapCache, using the same swap as before.
278 static bool shmem_confirm_swap(struct address_space
*mapping
,
279 pgoff_t index
, swp_entry_t swap
)
284 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
286 return item
== swp_to_radix_entry(swap
);
290 * Like add_to_page_cache_locked, but error if expected item has gone.
292 static int shmem_add_to_page_cache(struct page
*page
,
293 struct address_space
*mapping
,
294 pgoff_t index
, void *expected
)
298 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
299 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
301 page_cache_get(page
);
302 page
->mapping
= mapping
;
305 spin_lock_irq(&mapping
->tree_lock
);
307 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
309 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
313 __inc_zone_page_state(page
, NR_FILE_PAGES
);
314 __inc_zone_page_state(page
, NR_SHMEM
);
315 spin_unlock_irq(&mapping
->tree_lock
);
317 page
->mapping
= NULL
;
318 spin_unlock_irq(&mapping
->tree_lock
);
319 page_cache_release(page
);
325 * Like delete_from_page_cache, but substitutes swap for page.
327 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
329 struct address_space
*mapping
= page
->mapping
;
332 spin_lock_irq(&mapping
->tree_lock
);
333 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
334 page
->mapping
= NULL
;
336 __dec_zone_page_state(page
, NR_FILE_PAGES
);
337 __dec_zone_page_state(page
, NR_SHMEM
);
338 spin_unlock_irq(&mapping
->tree_lock
);
339 page_cache_release(page
);
344 * Remove swap entry from radix tree, free the swap and its page cache.
346 static int shmem_free_swap(struct address_space
*mapping
,
347 pgoff_t index
, void *radswap
)
351 spin_lock_irq(&mapping
->tree_lock
);
352 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
353 spin_unlock_irq(&mapping
->tree_lock
);
356 free_swap_and_cache(radix_to_swp_entry(radswap
));
361 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
363 void shmem_unlock_mapping(struct address_space
*mapping
)
366 pgoff_t indices
[PAGEVEC_SIZE
];
369 pagevec_init(&pvec
, 0);
371 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
373 while (!mapping_unevictable(mapping
)) {
375 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
376 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
378 pvec
.nr
= find_get_entries(mapping
, index
,
379 PAGEVEC_SIZE
, pvec
.pages
, indices
);
382 index
= indices
[pvec
.nr
- 1] + 1;
383 pagevec_remove_exceptionals(&pvec
);
384 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
385 pagevec_release(&pvec
);
391 * Remove range of pages and swap entries from radix tree, and free them.
392 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
394 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
397 struct address_space
*mapping
= inode
->i_mapping
;
398 struct shmem_inode_info
*info
= SHMEM_I(inode
);
399 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
400 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
401 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
402 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
404 pgoff_t indices
[PAGEVEC_SIZE
];
405 long nr_swaps_freed
= 0;
410 end
= -1; /* unsigned, so actually very big */
412 pagevec_init(&pvec
, 0);
414 while (index
< end
) {
415 pvec
.nr
= find_get_entries(mapping
, index
,
416 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
417 pvec
.pages
, indices
);
420 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
421 struct page
*page
= pvec
.pages
[i
];
427 if (radix_tree_exceptional_entry(page
)) {
430 nr_swaps_freed
+= !shmem_free_swap(mapping
,
435 if (!trylock_page(page
))
437 if (!unfalloc
|| !PageUptodate(page
)) {
438 if (page
->mapping
== mapping
) {
439 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
440 truncate_inode_page(mapping
, page
);
445 pagevec_remove_exceptionals(&pvec
);
446 pagevec_release(&pvec
);
452 struct page
*page
= NULL
;
453 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
455 unsigned int top
= PAGE_CACHE_SIZE
;
460 zero_user_segment(page
, partial_start
, top
);
461 set_page_dirty(page
);
463 page_cache_release(page
);
467 struct page
*page
= NULL
;
468 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
470 zero_user_segment(page
, 0, partial_end
);
471 set_page_dirty(page
);
473 page_cache_release(page
);
480 while (index
< end
) {
483 pvec
.nr
= find_get_entries(mapping
, index
,
484 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
485 pvec
.pages
, indices
);
487 /* If all gone or hole-punch or unfalloc, we're done */
488 if (index
== start
|| end
!= -1)
490 /* But if truncating, restart to make sure all gone */
494 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
495 struct page
*page
= pvec
.pages
[i
];
501 if (radix_tree_exceptional_entry(page
)) {
504 if (shmem_free_swap(mapping
, index
, page
)) {
505 /* Swap was replaced by page: retry */
514 if (!unfalloc
|| !PageUptodate(page
)) {
515 if (page
->mapping
== mapping
) {
516 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
517 truncate_inode_page(mapping
, page
);
519 /* Page was replaced by swap: retry */
527 pagevec_remove_exceptionals(&pvec
);
528 pagevec_release(&pvec
);
532 spin_lock(&info
->lock
);
533 info
->swapped
-= nr_swaps_freed
;
534 shmem_recalc_inode(inode
);
535 spin_unlock(&info
->lock
);
538 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
540 shmem_undo_range(inode
, lstart
, lend
, false);
541 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
543 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
545 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
548 struct inode
*inode
= dentry
->d_inode
;
549 struct shmem_inode_info
*info
= SHMEM_I(inode
);
551 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
552 spin_lock(&info
->lock
);
553 shmem_recalc_inode(inode
);
554 spin_unlock(&info
->lock
);
556 generic_fillattr(inode
, stat
);
560 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
562 struct inode
*inode
= d_inode(dentry
);
563 struct shmem_inode_info
*info
= SHMEM_I(inode
);
566 error
= inode_change_ok(inode
, attr
);
570 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
571 loff_t oldsize
= inode
->i_size
;
572 loff_t newsize
= attr
->ia_size
;
574 /* protected by i_mutex */
575 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
576 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
579 if (newsize
!= oldsize
) {
580 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
584 i_size_write(inode
, newsize
);
585 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
587 if (newsize
<= oldsize
) {
588 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
589 if (oldsize
> holebegin
)
590 unmap_mapping_range(inode
->i_mapping
,
593 shmem_truncate_range(inode
,
594 newsize
, (loff_t
)-1);
595 /* unmap again to remove racily COWed private pages */
596 if (oldsize
> holebegin
)
597 unmap_mapping_range(inode
->i_mapping
,
602 setattr_copy(inode
, attr
);
603 if (attr
->ia_valid
& ATTR_MODE
)
604 error
= posix_acl_chmod(inode
, inode
->i_mode
);
608 static void shmem_evict_inode(struct inode
*inode
)
610 struct shmem_inode_info
*info
= SHMEM_I(inode
);
612 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
613 shmem_unacct_size(info
->flags
, inode
->i_size
);
615 shmem_truncate_range(inode
, 0, (loff_t
)-1);
616 if (!list_empty(&info
->swaplist
)) {
617 mutex_lock(&shmem_swaplist_mutex
);
618 list_del_init(&info
->swaplist
);
619 mutex_unlock(&shmem_swaplist_mutex
);
622 kfree(info
->symlink
);
624 simple_xattrs_free(&info
->xattrs
);
625 WARN_ON(inode
->i_blocks
);
626 shmem_free_inode(inode
->i_sb
);
631 * If swap found in inode, free it and move page from swapcache to filecache.
633 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
634 swp_entry_t swap
, struct page
**pagep
)
636 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
642 radswap
= swp_to_radix_entry(swap
);
643 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
645 return -EAGAIN
; /* tell shmem_unuse we found nothing */
648 * Move _head_ to start search for next from here.
649 * But be careful: shmem_evict_inode checks list_empty without taking
650 * mutex, and there's an instant in list_move_tail when info->swaplist
651 * would appear empty, if it were the only one on shmem_swaplist.
653 if (shmem_swaplist
.next
!= &info
->swaplist
)
654 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
656 gfp
= mapping_gfp_mask(mapping
);
657 if (shmem_should_replace_page(*pagep
, gfp
)) {
658 mutex_unlock(&shmem_swaplist_mutex
);
659 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
660 mutex_lock(&shmem_swaplist_mutex
);
662 * We needed to drop mutex to make that restrictive page
663 * allocation, but the inode might have been freed while we
664 * dropped it: although a racing shmem_evict_inode() cannot
665 * complete without emptying the radix_tree, our page lock
666 * on this swapcache page is not enough to prevent that -
667 * free_swap_and_cache() of our swap entry will only
668 * trylock_page(), removing swap from radix_tree whatever.
670 * We must not proceed to shmem_add_to_page_cache() if the
671 * inode has been freed, but of course we cannot rely on
672 * inode or mapping or info to check that. However, we can
673 * safely check if our swap entry is still in use (and here
674 * it can't have got reused for another page): if it's still
675 * in use, then the inode cannot have been freed yet, and we
676 * can safely proceed (if it's no longer in use, that tells
677 * nothing about the inode, but we don't need to unuse swap).
679 if (!page_swapcount(*pagep
))
684 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
685 * but also to hold up shmem_evict_inode(): so inode cannot be freed
686 * beneath us (pagelock doesn't help until the page is in pagecache).
689 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
691 if (error
!= -ENOMEM
) {
693 * Truncation and eviction use free_swap_and_cache(), which
694 * only does trylock page: if we raced, best clean up here.
696 delete_from_swap_cache(*pagep
);
697 set_page_dirty(*pagep
);
699 spin_lock(&info
->lock
);
701 spin_unlock(&info
->lock
);
709 * Search through swapped inodes to find and replace swap by page.
711 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
713 struct list_head
*this, *next
;
714 struct shmem_inode_info
*info
;
715 struct mem_cgroup
*memcg
;
719 * There's a faint possibility that swap page was replaced before
720 * caller locked it: caller will come back later with the right page.
722 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
726 * Charge page using GFP_KERNEL while we can wait, before taking
727 * the shmem_swaplist_mutex which might hold up shmem_writepage().
728 * Charged back to the user (not to caller) when swap account is used.
730 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
);
733 /* No radix_tree_preload: swap entry keeps a place for page in tree */
736 mutex_lock(&shmem_swaplist_mutex
);
737 list_for_each_safe(this, next
, &shmem_swaplist
) {
738 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
740 error
= shmem_unuse_inode(info
, swap
, &page
);
742 list_del_init(&info
->swaplist
);
744 if (error
!= -EAGAIN
)
746 /* found nothing in this: move on to search the next */
748 mutex_unlock(&shmem_swaplist_mutex
);
751 if (error
!= -ENOMEM
)
753 mem_cgroup_cancel_charge(page
, memcg
);
755 mem_cgroup_commit_charge(page
, memcg
, true);
758 page_cache_release(page
);
763 * Move the page from the page cache to the swap cache.
765 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
767 struct shmem_inode_info
*info
;
768 struct address_space
*mapping
;
773 BUG_ON(!PageLocked(page
));
774 mapping
= page
->mapping
;
776 inode
= mapping
->host
;
777 info
= SHMEM_I(inode
);
778 if (info
->flags
& VM_LOCKED
)
780 if (!total_swap_pages
)
784 * Our capabilities prevent regular writeback or sync from ever calling
785 * shmem_writepage; but a stacking filesystem might use ->writepage of
786 * its underlying filesystem, in which case tmpfs should write out to
787 * swap only in response to memory pressure, and not for the writeback
790 if (!wbc
->for_reclaim
) {
791 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
796 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
797 * value into swapfile.c, the only way we can correctly account for a
798 * fallocated page arriving here is now to initialize it and write it.
800 * That's okay for a page already fallocated earlier, but if we have
801 * not yet completed the fallocation, then (a) we want to keep track
802 * of this page in case we have to undo it, and (b) it may not be a
803 * good idea to continue anyway, once we're pushing into swap. So
804 * reactivate the page, and let shmem_fallocate() quit when too many.
806 if (!PageUptodate(page
)) {
807 if (inode
->i_private
) {
808 struct shmem_falloc
*shmem_falloc
;
809 spin_lock(&inode
->i_lock
);
810 shmem_falloc
= inode
->i_private
;
812 !shmem_falloc
->waitq
&&
813 index
>= shmem_falloc
->start
&&
814 index
< shmem_falloc
->next
)
815 shmem_falloc
->nr_unswapped
++;
818 spin_unlock(&inode
->i_lock
);
822 clear_highpage(page
);
823 flush_dcache_page(page
);
824 SetPageUptodate(page
);
827 swap
= get_swap_page();
832 * Add inode to shmem_unuse()'s list of swapped-out inodes,
833 * if it's not already there. Do it now before the page is
834 * moved to swap cache, when its pagelock no longer protects
835 * the inode from eviction. But don't unlock the mutex until
836 * we've incremented swapped, because shmem_unuse_inode() will
837 * prune a !swapped inode from the swaplist under this mutex.
839 mutex_lock(&shmem_swaplist_mutex
);
840 if (list_empty(&info
->swaplist
))
841 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
843 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
844 swap_shmem_alloc(swap
);
845 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
847 spin_lock(&info
->lock
);
849 shmem_recalc_inode(inode
);
850 spin_unlock(&info
->lock
);
852 mutex_unlock(&shmem_swaplist_mutex
);
853 BUG_ON(page_mapped(page
));
854 swap_writepage(page
, wbc
);
858 mutex_unlock(&shmem_swaplist_mutex
);
859 swapcache_free(swap
);
861 set_page_dirty(page
);
862 if (wbc
->for_reclaim
)
863 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
870 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
874 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
875 return; /* show nothing */
877 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
879 seq_printf(seq
, ",mpol=%s", buffer
);
882 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
884 struct mempolicy
*mpol
= NULL
;
886 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
889 spin_unlock(&sbinfo
->stat_lock
);
893 #endif /* CONFIG_TMPFS */
895 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
896 struct shmem_inode_info
*info
, pgoff_t index
)
898 struct vm_area_struct pvma
;
901 /* Create a pseudo vma that just contains the policy */
903 /* Bias interleave by inode number to distribute better across nodes */
904 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
906 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
908 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
910 /* Drop reference taken by mpol_shared_policy_lookup() */
911 mpol_cond_put(pvma
.vm_policy
);
916 static struct page
*shmem_alloc_page(gfp_t gfp
,
917 struct shmem_inode_info
*info
, pgoff_t index
)
919 struct vm_area_struct pvma
;
922 /* Create a pseudo vma that just contains the policy */
924 /* Bias interleave by inode number to distribute better across nodes */
925 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
927 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
929 page
= alloc_page_vma(gfp
, &pvma
, 0);
931 /* Drop reference taken by mpol_shared_policy_lookup() */
932 mpol_cond_put(pvma
.vm_policy
);
936 #else /* !CONFIG_NUMA */
938 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
941 #endif /* CONFIG_TMPFS */
943 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
944 struct shmem_inode_info
*info
, pgoff_t index
)
946 return swapin_readahead(swap
, gfp
, NULL
, 0);
949 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
950 struct shmem_inode_info
*info
, pgoff_t index
)
952 return alloc_page(gfp
);
954 #endif /* CONFIG_NUMA */
956 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
957 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
964 * When a page is moved from swapcache to shmem filecache (either by the
965 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
966 * shmem_unuse_inode()), it may have been read in earlier from swap, in
967 * ignorance of the mapping it belongs to. If that mapping has special
968 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
969 * we may need to copy to a suitable page before moving to filecache.
971 * In a future release, this may well be extended to respect cpuset and
972 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
973 * but for now it is a simple matter of zone.
975 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
977 return page_zonenum(page
) > gfp_zone(gfp
);
980 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
981 struct shmem_inode_info
*info
, pgoff_t index
)
983 struct page
*oldpage
, *newpage
;
984 struct address_space
*swap_mapping
;
989 swap_index
= page_private(oldpage
);
990 swap_mapping
= page_mapping(oldpage
);
993 * We have arrived here because our zones are constrained, so don't
994 * limit chance of success by further cpuset and node constraints.
996 gfp
&= ~GFP_CONSTRAINT_MASK
;
997 newpage
= shmem_alloc_page(gfp
, info
, index
);
1001 page_cache_get(newpage
);
1002 copy_highpage(newpage
, oldpage
);
1003 flush_dcache_page(newpage
);
1005 __set_page_locked(newpage
);
1006 SetPageUptodate(newpage
);
1007 SetPageSwapBacked(newpage
);
1008 set_page_private(newpage
, swap_index
);
1009 SetPageSwapCache(newpage
);
1012 * Our caller will very soon move newpage out of swapcache, but it's
1013 * a nice clean interface for us to replace oldpage by newpage there.
1015 spin_lock_irq(&swap_mapping
->tree_lock
);
1016 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1019 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1020 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1022 spin_unlock_irq(&swap_mapping
->tree_lock
);
1024 if (unlikely(error
)) {
1026 * Is this possible? I think not, now that our callers check
1027 * both PageSwapCache and page_private after getting page lock;
1028 * but be defensive. Reverse old to newpage for clear and free.
1032 mem_cgroup_replace_page(oldpage
, newpage
);
1033 lru_cache_add_anon(newpage
);
1037 ClearPageSwapCache(oldpage
);
1038 set_page_private(oldpage
, 0);
1040 unlock_page(oldpage
);
1041 page_cache_release(oldpage
);
1042 page_cache_release(oldpage
);
1047 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1049 * If we allocate a new one we do not mark it dirty. That's up to the
1050 * vm. If we swap it in we mark it dirty since we also free the swap
1051 * entry since a page cannot live in both the swap and page cache
1053 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1054 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1056 struct address_space
*mapping
= inode
->i_mapping
;
1057 struct shmem_inode_info
*info
;
1058 struct shmem_sb_info
*sbinfo
;
1059 struct mem_cgroup
*memcg
;
1066 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1070 page
= find_lock_entry(mapping
, index
);
1071 if (radix_tree_exceptional_entry(page
)) {
1072 swap
= radix_to_swp_entry(page
);
1076 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1077 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1082 if (page
&& sgp
== SGP_WRITE
)
1083 mark_page_accessed(page
);
1085 /* fallocated page? */
1086 if (page
&& !PageUptodate(page
)) {
1087 if (sgp
!= SGP_READ
)
1090 page_cache_release(page
);
1093 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1099 * Fast cache lookup did not find it:
1100 * bring it back from swap or allocate.
1102 info
= SHMEM_I(inode
);
1103 sbinfo
= SHMEM_SB(inode
->i_sb
);
1106 /* Look it up and read it in.. */
1107 page
= lookup_swap_cache(swap
);
1109 /* here we actually do the io */
1111 *fault_type
|= VM_FAULT_MAJOR
;
1112 page
= shmem_swapin(swap
, gfp
, info
, index
);
1119 /* We have to do this with page locked to prevent races */
1121 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1122 !shmem_confirm_swap(mapping
, index
, swap
)) {
1123 error
= -EEXIST
; /* try again */
1126 if (!PageUptodate(page
)) {
1130 wait_on_page_writeback(page
);
1132 if (shmem_should_replace_page(page
, gfp
)) {
1133 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1138 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1140 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1141 swp_to_radix_entry(swap
));
1143 * We already confirmed swap under page lock, and make
1144 * no memory allocation here, so usually no possibility
1145 * of error; but free_swap_and_cache() only trylocks a
1146 * page, so it is just possible that the entry has been
1147 * truncated or holepunched since swap was confirmed.
1148 * shmem_undo_range() will have done some of the
1149 * unaccounting, now delete_from_swap_cache() will do
1151 * Reset swap.val? No, leave it so "failed" goes back to
1152 * "repeat": reading a hole and writing should succeed.
1155 mem_cgroup_cancel_charge(page
, memcg
);
1156 delete_from_swap_cache(page
);
1162 mem_cgroup_commit_charge(page
, memcg
, true);
1164 spin_lock(&info
->lock
);
1166 shmem_recalc_inode(inode
);
1167 spin_unlock(&info
->lock
);
1169 if (sgp
== SGP_WRITE
)
1170 mark_page_accessed(page
);
1172 delete_from_swap_cache(page
);
1173 set_page_dirty(page
);
1177 if (shmem_acct_block(info
->flags
)) {
1181 if (sbinfo
->max_blocks
) {
1182 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1183 sbinfo
->max_blocks
) >= 0) {
1187 percpu_counter_inc(&sbinfo
->used_blocks
);
1190 page
= shmem_alloc_page(gfp
, info
, index
);
1196 __SetPageSwapBacked(page
);
1197 __set_page_locked(page
);
1198 if (sgp
== SGP_WRITE
)
1199 __SetPageReferenced(page
);
1201 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1204 error
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
1206 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1208 radix_tree_preload_end();
1211 mem_cgroup_cancel_charge(page
, memcg
);
1214 mem_cgroup_commit_charge(page
, memcg
, false);
1215 lru_cache_add_anon(page
);
1217 spin_lock(&info
->lock
);
1219 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1220 shmem_recalc_inode(inode
);
1221 spin_unlock(&info
->lock
);
1225 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1227 if (sgp
== SGP_FALLOC
)
1231 * Let SGP_WRITE caller clear ends if write does not fill page;
1232 * but SGP_FALLOC on a page fallocated earlier must initialize
1233 * it now, lest undo on failure cancel our earlier guarantee.
1235 if (sgp
!= SGP_WRITE
) {
1236 clear_highpage(page
);
1237 flush_dcache_page(page
);
1238 SetPageUptodate(page
);
1240 if (sgp
== SGP_DIRTY
)
1241 set_page_dirty(page
);
1244 /* Perhaps the file has been truncated since we checked */
1245 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1246 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1260 info
= SHMEM_I(inode
);
1261 ClearPageDirty(page
);
1262 delete_from_page_cache(page
);
1263 spin_lock(&info
->lock
);
1265 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1266 spin_unlock(&info
->lock
);
1268 sbinfo
= SHMEM_SB(inode
->i_sb
);
1269 if (sbinfo
->max_blocks
)
1270 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1272 shmem_unacct_blocks(info
->flags
, 1);
1274 if (swap
.val
&& error
!= -EINVAL
&&
1275 !shmem_confirm_swap(mapping
, index
, swap
))
1280 page_cache_release(page
);
1282 if (error
== -ENOSPC
&& !once
++) {
1283 info
= SHMEM_I(inode
);
1284 spin_lock(&info
->lock
);
1285 shmem_recalc_inode(inode
);
1286 spin_unlock(&info
->lock
);
1289 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1294 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1296 struct inode
*inode
= file_inode(vma
->vm_file
);
1298 int ret
= VM_FAULT_LOCKED
;
1301 * Trinity finds that probing a hole which tmpfs is punching can
1302 * prevent the hole-punch from ever completing: which in turn
1303 * locks writers out with its hold on i_mutex. So refrain from
1304 * faulting pages into the hole while it's being punched. Although
1305 * shmem_undo_range() does remove the additions, it may be unable to
1306 * keep up, as each new page needs its own unmap_mapping_range() call,
1307 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1309 * It does not matter if we sometimes reach this check just before the
1310 * hole-punch begins, so that one fault then races with the punch:
1311 * we just need to make racing faults a rare case.
1313 * The implementation below would be much simpler if we just used a
1314 * standard mutex or completion: but we cannot take i_mutex in fault,
1315 * and bloating every shmem inode for this unlikely case would be sad.
1317 if (unlikely(inode
->i_private
)) {
1318 struct shmem_falloc
*shmem_falloc
;
1320 spin_lock(&inode
->i_lock
);
1321 shmem_falloc
= inode
->i_private
;
1323 shmem_falloc
->waitq
&&
1324 vmf
->pgoff
>= shmem_falloc
->start
&&
1325 vmf
->pgoff
< shmem_falloc
->next
) {
1326 wait_queue_head_t
*shmem_falloc_waitq
;
1327 DEFINE_WAIT(shmem_fault_wait
);
1329 ret
= VM_FAULT_NOPAGE
;
1330 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1331 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1332 /* It's polite to up mmap_sem if we can */
1333 up_read(&vma
->vm_mm
->mmap_sem
);
1334 ret
= VM_FAULT_RETRY
;
1337 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1338 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1339 TASK_UNINTERRUPTIBLE
);
1340 spin_unlock(&inode
->i_lock
);
1344 * shmem_falloc_waitq points into the shmem_fallocate()
1345 * stack of the hole-punching task: shmem_falloc_waitq
1346 * is usually invalid by the time we reach here, but
1347 * finish_wait() does not dereference it in that case;
1348 * though i_lock needed lest racing with wake_up_all().
1350 spin_lock(&inode
->i_lock
);
1351 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1352 spin_unlock(&inode
->i_lock
);
1355 spin_unlock(&inode
->i_lock
);
1358 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1360 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1362 if (ret
& VM_FAULT_MAJOR
) {
1363 count_vm_event(PGMAJFAULT
);
1364 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1370 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1372 struct inode
*inode
= file_inode(vma
->vm_file
);
1373 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1376 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1379 struct inode
*inode
= file_inode(vma
->vm_file
);
1382 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1383 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1387 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1389 struct inode
*inode
= file_inode(file
);
1390 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1391 int retval
= -ENOMEM
;
1393 spin_lock(&info
->lock
);
1394 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1395 if (!user_shm_lock(inode
->i_size
, user
))
1397 info
->flags
|= VM_LOCKED
;
1398 mapping_set_unevictable(file
->f_mapping
);
1400 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1401 user_shm_unlock(inode
->i_size
, user
);
1402 info
->flags
&= ~VM_LOCKED
;
1403 mapping_clear_unevictable(file
->f_mapping
);
1408 spin_unlock(&info
->lock
);
1412 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1414 file_accessed(file
);
1415 vma
->vm_ops
= &shmem_vm_ops
;
1419 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1420 umode_t mode
, dev_t dev
, unsigned long flags
)
1422 struct inode
*inode
;
1423 struct shmem_inode_info
*info
;
1424 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1426 if (shmem_reserve_inode(sb
))
1429 inode
= new_inode(sb
);
1431 inode
->i_ino
= get_next_ino();
1432 inode_init_owner(inode
, dir
, mode
);
1433 inode
->i_blocks
= 0;
1434 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1435 inode
->i_generation
= get_seconds();
1436 info
= SHMEM_I(inode
);
1437 memset(info
, 0, (char *)inode
- (char *)info
);
1438 spin_lock_init(&info
->lock
);
1439 info
->seals
= F_SEAL_SEAL
;
1440 info
->flags
= flags
& VM_NORESERVE
;
1441 INIT_LIST_HEAD(&info
->swaplist
);
1442 simple_xattrs_init(&info
->xattrs
);
1443 cache_no_acl(inode
);
1445 switch (mode
& S_IFMT
) {
1447 inode
->i_op
= &shmem_special_inode_operations
;
1448 init_special_inode(inode
, mode
, dev
);
1451 inode
->i_mapping
->a_ops
= &shmem_aops
;
1452 inode
->i_op
= &shmem_inode_operations
;
1453 inode
->i_fop
= &shmem_file_operations
;
1454 mpol_shared_policy_init(&info
->policy
,
1455 shmem_get_sbmpol(sbinfo
));
1459 /* Some things misbehave if size == 0 on a directory */
1460 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1461 inode
->i_op
= &shmem_dir_inode_operations
;
1462 inode
->i_fop
= &simple_dir_operations
;
1466 * Must not load anything in the rbtree,
1467 * mpol_free_shared_policy will not be called.
1469 mpol_shared_policy_init(&info
->policy
, NULL
);
1473 shmem_free_inode(sb
);
1477 bool shmem_mapping(struct address_space
*mapping
)
1482 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
1486 static const struct inode_operations shmem_symlink_inode_operations
;
1487 static const struct inode_operations shmem_short_symlink_operations
;
1489 #ifdef CONFIG_TMPFS_XATTR
1490 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1492 #define shmem_initxattrs NULL
1496 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1497 loff_t pos
, unsigned len
, unsigned flags
,
1498 struct page
**pagep
, void **fsdata
)
1500 struct inode
*inode
= mapping
->host
;
1501 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1502 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1504 /* i_mutex is held by caller */
1505 if (unlikely(info
->seals
)) {
1506 if (info
->seals
& F_SEAL_WRITE
)
1508 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
1512 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1516 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1517 loff_t pos
, unsigned len
, unsigned copied
,
1518 struct page
*page
, void *fsdata
)
1520 struct inode
*inode
= mapping
->host
;
1522 if (pos
+ copied
> inode
->i_size
)
1523 i_size_write(inode
, pos
+ copied
);
1525 if (!PageUptodate(page
)) {
1526 if (copied
< PAGE_CACHE_SIZE
) {
1527 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1528 zero_user_segments(page
, 0, from
,
1529 from
+ copied
, PAGE_CACHE_SIZE
);
1531 SetPageUptodate(page
);
1533 set_page_dirty(page
);
1535 page_cache_release(page
);
1540 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1542 struct file
*file
= iocb
->ki_filp
;
1543 struct inode
*inode
= file_inode(file
);
1544 struct address_space
*mapping
= inode
->i_mapping
;
1546 unsigned long offset
;
1547 enum sgp_type sgp
= SGP_READ
;
1550 loff_t
*ppos
= &iocb
->ki_pos
;
1553 * Might this read be for a stacking filesystem? Then when reading
1554 * holes of a sparse file, we actually need to allocate those pages,
1555 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1557 if (!iter_is_iovec(to
))
1560 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1561 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1564 struct page
*page
= NULL
;
1566 unsigned long nr
, ret
;
1567 loff_t i_size
= i_size_read(inode
);
1569 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1570 if (index
> end_index
)
1572 if (index
== end_index
) {
1573 nr
= i_size
& ~PAGE_CACHE_MASK
;
1578 error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1580 if (error
== -EINVAL
)
1588 * We must evaluate after, since reads (unlike writes)
1589 * are called without i_mutex protection against truncate
1591 nr
= PAGE_CACHE_SIZE
;
1592 i_size
= i_size_read(inode
);
1593 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1594 if (index
== end_index
) {
1595 nr
= i_size
& ~PAGE_CACHE_MASK
;
1598 page_cache_release(page
);
1606 * If users can be writing to this page using arbitrary
1607 * virtual addresses, take care about potential aliasing
1608 * before reading the page on the kernel side.
1610 if (mapping_writably_mapped(mapping
))
1611 flush_dcache_page(page
);
1613 * Mark the page accessed if we read the beginning.
1616 mark_page_accessed(page
);
1618 page
= ZERO_PAGE(0);
1619 page_cache_get(page
);
1623 * Ok, we have the page, and it's up-to-date, so
1624 * now we can copy it to user space...
1626 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
1629 index
+= offset
>> PAGE_CACHE_SHIFT
;
1630 offset
&= ~PAGE_CACHE_MASK
;
1632 page_cache_release(page
);
1633 if (!iov_iter_count(to
))
1642 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1643 file_accessed(file
);
1644 return retval
? retval
: error
;
1647 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1648 struct pipe_inode_info
*pipe
, size_t len
,
1651 struct address_space
*mapping
= in
->f_mapping
;
1652 struct inode
*inode
= mapping
->host
;
1653 unsigned int loff
, nr_pages
, req_pages
;
1654 struct page
*pages
[PIPE_DEF_BUFFERS
];
1655 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1657 pgoff_t index
, end_index
;
1660 struct splice_pipe_desc spd
= {
1663 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1665 .ops
= &page_cache_pipe_buf_ops
,
1666 .spd_release
= spd_release_page
,
1669 isize
= i_size_read(inode
);
1670 if (unlikely(*ppos
>= isize
))
1673 left
= isize
- *ppos
;
1674 if (unlikely(left
< len
))
1677 if (splice_grow_spd(pipe
, &spd
))
1680 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1681 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1682 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1683 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
1685 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1686 nr_pages
, spd
.pages
);
1687 index
+= spd
.nr_pages
;
1690 while (spd
.nr_pages
< nr_pages
) {
1691 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1695 spd
.pages
[spd
.nr_pages
++] = page
;
1699 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1700 nr_pages
= spd
.nr_pages
;
1703 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1704 unsigned int this_len
;
1709 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1710 page
= spd
.pages
[page_nr
];
1712 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1713 error
= shmem_getpage(inode
, index
, &page
,
1718 page_cache_release(spd
.pages
[page_nr
]);
1719 spd
.pages
[page_nr
] = page
;
1722 isize
= i_size_read(inode
);
1723 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1724 if (unlikely(!isize
|| index
> end_index
))
1727 if (end_index
== index
) {
1730 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1734 this_len
= min(this_len
, plen
- loff
);
1738 spd
.partial
[page_nr
].offset
= loff
;
1739 spd
.partial
[page_nr
].len
= this_len
;
1746 while (page_nr
< nr_pages
)
1747 page_cache_release(spd
.pages
[page_nr
++]);
1750 error
= splice_to_pipe(pipe
, &spd
);
1752 splice_shrink_spd(&spd
);
1762 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1764 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1765 pgoff_t index
, pgoff_t end
, int whence
)
1768 struct pagevec pvec
;
1769 pgoff_t indices
[PAGEVEC_SIZE
];
1773 pagevec_init(&pvec
, 0);
1774 pvec
.nr
= 1; /* start small: we may be there already */
1776 pvec
.nr
= find_get_entries(mapping
, index
,
1777 pvec
.nr
, pvec
.pages
, indices
);
1779 if (whence
== SEEK_DATA
)
1783 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1784 if (index
< indices
[i
]) {
1785 if (whence
== SEEK_HOLE
) {
1791 page
= pvec
.pages
[i
];
1792 if (page
&& !radix_tree_exceptional_entry(page
)) {
1793 if (!PageUptodate(page
))
1797 (page
&& whence
== SEEK_DATA
) ||
1798 (!page
&& whence
== SEEK_HOLE
)) {
1803 pagevec_remove_exceptionals(&pvec
);
1804 pagevec_release(&pvec
);
1805 pvec
.nr
= PAGEVEC_SIZE
;
1811 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1813 struct address_space
*mapping
= file
->f_mapping
;
1814 struct inode
*inode
= mapping
->host
;
1818 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1819 return generic_file_llseek_size(file
, offset
, whence
,
1820 MAX_LFS_FILESIZE
, i_size_read(inode
));
1821 mutex_lock(&inode
->i_mutex
);
1822 /* We're holding i_mutex so we can access i_size directly */
1826 else if (offset
>= inode
->i_size
)
1829 start
= offset
>> PAGE_CACHE_SHIFT
;
1830 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1831 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1832 new_offset
<<= PAGE_CACHE_SHIFT
;
1833 if (new_offset
> offset
) {
1834 if (new_offset
< inode
->i_size
)
1835 offset
= new_offset
;
1836 else if (whence
== SEEK_DATA
)
1839 offset
= inode
->i_size
;
1844 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
1845 mutex_unlock(&inode
->i_mutex
);
1850 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1851 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1853 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1854 #define LAST_SCAN 4 /* about 150ms max */
1856 static void shmem_tag_pins(struct address_space
*mapping
)
1858 struct radix_tree_iter iter
;
1868 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1869 page
= radix_tree_deref_slot(slot
);
1870 if (!page
|| radix_tree_exception(page
)) {
1871 if (radix_tree_deref_retry(page
))
1873 } else if (page_count(page
) - page_mapcount(page
) > 1) {
1874 spin_lock_irq(&mapping
->tree_lock
);
1875 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
1877 spin_unlock_irq(&mapping
->tree_lock
);
1880 if (need_resched()) {
1882 start
= iter
.index
+ 1;
1890 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1891 * via get_user_pages(), drivers might have some pending I/O without any active
1892 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1893 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1894 * them to be dropped.
1895 * The caller must guarantee that no new user will acquire writable references
1896 * to those pages to avoid races.
1898 static int shmem_wait_for_pins(struct address_space
*mapping
)
1900 struct radix_tree_iter iter
;
1906 shmem_tag_pins(mapping
);
1909 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
1910 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
1914 lru_add_drain_all();
1915 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
1921 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
1922 start
, SHMEM_TAG_PINNED
) {
1924 page
= radix_tree_deref_slot(slot
);
1925 if (radix_tree_exception(page
)) {
1926 if (radix_tree_deref_retry(page
))
1933 page_count(page
) - page_mapcount(page
) != 1) {
1934 if (scan
< LAST_SCAN
)
1935 goto continue_resched
;
1938 * On the last scan, we clean up all those tags
1939 * we inserted; but make a note that we still
1940 * found pages pinned.
1945 spin_lock_irq(&mapping
->tree_lock
);
1946 radix_tree_tag_clear(&mapping
->page_tree
,
1947 iter
.index
, SHMEM_TAG_PINNED
);
1948 spin_unlock_irq(&mapping
->tree_lock
);
1950 if (need_resched()) {
1952 start
= iter
.index
+ 1;
1962 #define F_ALL_SEALS (F_SEAL_SEAL | \
1967 int shmem_add_seals(struct file
*file
, unsigned int seals
)
1969 struct inode
*inode
= file_inode(file
);
1970 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1975 * Sealing allows multiple parties to share a shmem-file but restrict
1976 * access to a specific subset of file operations. Seals can only be
1977 * added, but never removed. This way, mutually untrusted parties can
1978 * share common memory regions with a well-defined policy. A malicious
1979 * peer can thus never perform unwanted operations on a shared object.
1981 * Seals are only supported on special shmem-files and always affect
1982 * the whole underlying inode. Once a seal is set, it may prevent some
1983 * kinds of access to the file. Currently, the following seals are
1985 * SEAL_SEAL: Prevent further seals from being set on this file
1986 * SEAL_SHRINK: Prevent the file from shrinking
1987 * SEAL_GROW: Prevent the file from growing
1988 * SEAL_WRITE: Prevent write access to the file
1990 * As we don't require any trust relationship between two parties, we
1991 * must prevent seals from being removed. Therefore, sealing a file
1992 * only adds a given set of seals to the file, it never touches
1993 * existing seals. Furthermore, the "setting seals"-operation can be
1994 * sealed itself, which basically prevents any further seal from being
1997 * Semantics of sealing are only defined on volatile files. Only
1998 * anonymous shmem files support sealing. More importantly, seals are
1999 * never written to disk. Therefore, there's no plan to support it on
2003 if (file
->f_op
!= &shmem_file_operations
)
2005 if (!(file
->f_mode
& FMODE_WRITE
))
2007 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2010 mutex_lock(&inode
->i_mutex
);
2012 if (info
->seals
& F_SEAL_SEAL
) {
2017 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2018 error
= mapping_deny_writable(file
->f_mapping
);
2022 error
= shmem_wait_for_pins(file
->f_mapping
);
2024 mapping_allow_writable(file
->f_mapping
);
2029 info
->seals
|= seals
;
2033 mutex_unlock(&inode
->i_mutex
);
2036 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2038 int shmem_get_seals(struct file
*file
)
2040 if (file
->f_op
!= &shmem_file_operations
)
2043 return SHMEM_I(file_inode(file
))->seals
;
2045 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2047 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2053 /* disallow upper 32bit */
2057 error
= shmem_add_seals(file
, arg
);
2060 error
= shmem_get_seals(file
);
2070 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2073 struct inode
*inode
= file_inode(file
);
2074 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2075 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2076 struct shmem_falloc shmem_falloc
;
2077 pgoff_t start
, index
, end
;
2080 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2083 mutex_lock(&inode
->i_mutex
);
2085 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2086 struct address_space
*mapping
= file
->f_mapping
;
2087 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2088 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2089 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2091 /* protected by i_mutex */
2092 if (info
->seals
& F_SEAL_WRITE
) {
2097 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2098 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2099 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2100 spin_lock(&inode
->i_lock
);
2101 inode
->i_private
= &shmem_falloc
;
2102 spin_unlock(&inode
->i_lock
);
2104 if ((u64
)unmap_end
> (u64
)unmap_start
)
2105 unmap_mapping_range(mapping
, unmap_start
,
2106 1 + unmap_end
- unmap_start
, 0);
2107 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2108 /* No need to unmap again: hole-punching leaves COWed pages */
2110 spin_lock(&inode
->i_lock
);
2111 inode
->i_private
= NULL
;
2112 wake_up_all(&shmem_falloc_waitq
);
2113 spin_unlock(&inode
->i_lock
);
2118 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2119 error
= inode_newsize_ok(inode
, offset
+ len
);
2123 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2128 start
= offset
>> PAGE_CACHE_SHIFT
;
2129 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2130 /* Try to avoid a swapstorm if len is impossible to satisfy */
2131 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2136 shmem_falloc
.waitq
= NULL
;
2137 shmem_falloc
.start
= start
;
2138 shmem_falloc
.next
= start
;
2139 shmem_falloc
.nr_falloced
= 0;
2140 shmem_falloc
.nr_unswapped
= 0;
2141 spin_lock(&inode
->i_lock
);
2142 inode
->i_private
= &shmem_falloc
;
2143 spin_unlock(&inode
->i_lock
);
2145 for (index
= start
; index
< end
; index
++) {
2149 * Good, the fallocate(2) manpage permits EINTR: we may have
2150 * been interrupted because we are using up too much memory.
2152 if (signal_pending(current
))
2154 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2157 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
2160 /* Remove the !PageUptodate pages we added */
2161 shmem_undo_range(inode
,
2162 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
2163 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
2168 * Inform shmem_writepage() how far we have reached.
2169 * No need for lock or barrier: we have the page lock.
2171 shmem_falloc
.next
++;
2172 if (!PageUptodate(page
))
2173 shmem_falloc
.nr_falloced
++;
2176 * If !PageUptodate, leave it that way so that freeable pages
2177 * can be recognized if we need to rollback on error later.
2178 * But set_page_dirty so that memory pressure will swap rather
2179 * than free the pages we are allocating (and SGP_CACHE pages
2180 * might still be clean: we now need to mark those dirty too).
2182 set_page_dirty(page
);
2184 page_cache_release(page
);
2188 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2189 i_size_write(inode
, offset
+ len
);
2190 inode
->i_ctime
= CURRENT_TIME
;
2192 spin_lock(&inode
->i_lock
);
2193 inode
->i_private
= NULL
;
2194 spin_unlock(&inode
->i_lock
);
2196 mutex_unlock(&inode
->i_mutex
);
2200 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2202 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2204 buf
->f_type
= TMPFS_MAGIC
;
2205 buf
->f_bsize
= PAGE_CACHE_SIZE
;
2206 buf
->f_namelen
= NAME_MAX
;
2207 if (sbinfo
->max_blocks
) {
2208 buf
->f_blocks
= sbinfo
->max_blocks
;
2210 buf
->f_bfree
= sbinfo
->max_blocks
-
2211 percpu_counter_sum(&sbinfo
->used_blocks
);
2213 if (sbinfo
->max_inodes
) {
2214 buf
->f_files
= sbinfo
->max_inodes
;
2215 buf
->f_ffree
= sbinfo
->free_inodes
;
2217 /* else leave those fields 0 like simple_statfs */
2222 * File creation. Allocate an inode, and we're done..
2225 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2227 struct inode
*inode
;
2228 int error
= -ENOSPC
;
2230 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2232 error
= simple_acl_create(dir
, inode
);
2235 error
= security_inode_init_security(inode
, dir
,
2237 shmem_initxattrs
, NULL
);
2238 if (error
&& error
!= -EOPNOTSUPP
)
2242 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2243 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2244 d_instantiate(dentry
, inode
);
2245 dget(dentry
); /* Extra count - pin the dentry in core */
2254 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2256 struct inode
*inode
;
2257 int error
= -ENOSPC
;
2259 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2261 error
= security_inode_init_security(inode
, dir
,
2263 shmem_initxattrs
, NULL
);
2264 if (error
&& error
!= -EOPNOTSUPP
)
2266 error
= simple_acl_create(dir
, inode
);
2269 d_tmpfile(dentry
, inode
);
2277 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2281 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2287 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2290 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2296 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2298 struct inode
*inode
= d_inode(old_dentry
);
2302 * No ordinary (disk based) filesystem counts links as inodes;
2303 * but each new link needs a new dentry, pinning lowmem, and
2304 * tmpfs dentries cannot be pruned until they are unlinked.
2306 ret
= shmem_reserve_inode(inode
->i_sb
);
2310 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2311 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2313 ihold(inode
); /* New dentry reference */
2314 dget(dentry
); /* Extra pinning count for the created dentry */
2315 d_instantiate(dentry
, inode
);
2320 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2322 struct inode
*inode
= d_inode(dentry
);
2324 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2325 shmem_free_inode(inode
->i_sb
);
2327 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2328 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2330 dput(dentry
); /* Undo the count from "create" - this does all the work */
2334 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2336 if (!simple_empty(dentry
))
2339 drop_nlink(d_inode(dentry
));
2341 return shmem_unlink(dir
, dentry
);
2344 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2346 bool old_is_dir
= d_is_dir(old_dentry
);
2347 bool new_is_dir
= d_is_dir(new_dentry
);
2349 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2351 drop_nlink(old_dir
);
2354 drop_nlink(new_dir
);
2358 old_dir
->i_ctime
= old_dir
->i_mtime
=
2359 new_dir
->i_ctime
= new_dir
->i_mtime
=
2360 d_inode(old_dentry
)->i_ctime
=
2361 d_inode(new_dentry
)->i_ctime
= CURRENT_TIME
;
2366 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2368 struct dentry
*whiteout
;
2371 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2375 error
= shmem_mknod(old_dir
, whiteout
,
2376 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2382 * Cheat and hash the whiteout while the old dentry is still in
2383 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2385 * d_lookup() will consistently find one of them at this point,
2386 * not sure which one, but that isn't even important.
2393 * The VFS layer already does all the dentry stuff for rename,
2394 * we just have to decrement the usage count for the target if
2395 * it exists so that the VFS layer correctly free's it when it
2398 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2400 struct inode
*inode
= d_inode(old_dentry
);
2401 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2403 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2406 if (flags
& RENAME_EXCHANGE
)
2407 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2409 if (!simple_empty(new_dentry
))
2412 if (flags
& RENAME_WHITEOUT
) {
2415 error
= shmem_whiteout(old_dir
, old_dentry
);
2420 if (d_really_is_positive(new_dentry
)) {
2421 (void) shmem_unlink(new_dir
, new_dentry
);
2422 if (they_are_dirs
) {
2423 drop_nlink(d_inode(new_dentry
));
2424 drop_nlink(old_dir
);
2426 } else if (they_are_dirs
) {
2427 drop_nlink(old_dir
);
2431 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2432 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2433 old_dir
->i_ctime
= old_dir
->i_mtime
=
2434 new_dir
->i_ctime
= new_dir
->i_mtime
=
2435 inode
->i_ctime
= CURRENT_TIME
;
2439 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2443 struct inode
*inode
;
2446 struct shmem_inode_info
*info
;
2448 len
= strlen(symname
) + 1;
2449 if (len
> PAGE_CACHE_SIZE
)
2450 return -ENAMETOOLONG
;
2452 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2456 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2457 shmem_initxattrs
, NULL
);
2459 if (error
!= -EOPNOTSUPP
) {
2466 info
= SHMEM_I(inode
);
2467 inode
->i_size
= len
-1;
2468 if (len
<= SHORT_SYMLINK_LEN
) {
2469 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2470 if (!info
->symlink
) {
2474 inode
->i_op
= &shmem_short_symlink_operations
;
2475 inode
->i_link
= info
->symlink
;
2477 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2482 inode
->i_mapping
->a_ops
= &shmem_aops
;
2483 inode
->i_op
= &shmem_symlink_inode_operations
;
2484 kaddr
= kmap_atomic(page
);
2485 memcpy(kaddr
, symname
, len
);
2486 kunmap_atomic(kaddr
);
2487 SetPageUptodate(page
);
2488 set_page_dirty(page
);
2490 page_cache_release(page
);
2492 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2493 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2494 d_instantiate(dentry
, inode
);
2499 static const char *shmem_follow_link(struct dentry
*dentry
, void **cookie
)
2501 struct page
*page
= NULL
;
2502 int error
= shmem_getpage(d_inode(dentry
), 0, &page
, SGP_READ
, NULL
);
2504 return ERR_PTR(error
);
2510 static void shmem_put_link(struct inode
*unused
, void *cookie
)
2512 struct page
*page
= cookie
;
2514 mark_page_accessed(page
);
2515 page_cache_release(page
);
2518 #ifdef CONFIG_TMPFS_XATTR
2520 * Superblocks without xattr inode operations may get some security.* xattr
2521 * support from the LSM "for free". As soon as we have any other xattrs
2522 * like ACLs, we also need to implement the security.* handlers at
2523 * filesystem level, though.
2527 * Callback for security_inode_init_security() for acquiring xattrs.
2529 static int shmem_initxattrs(struct inode
*inode
,
2530 const struct xattr
*xattr_array
,
2533 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2534 const struct xattr
*xattr
;
2535 struct simple_xattr
*new_xattr
;
2538 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2539 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2543 len
= strlen(xattr
->name
) + 1;
2544 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2546 if (!new_xattr
->name
) {
2551 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2552 XATTR_SECURITY_PREFIX_LEN
);
2553 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2556 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2562 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2563 #ifdef CONFIG_TMPFS_POSIX_ACL
2564 &posix_acl_access_xattr_handler
,
2565 &posix_acl_default_xattr_handler
,
2570 static int shmem_xattr_validate(const char *name
)
2572 struct { const char *prefix
; size_t len
; } arr
[] = {
2573 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2574 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2578 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2579 size_t preflen
= arr
[i
].len
;
2580 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2589 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2590 void *buffer
, size_t size
)
2592 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2596 * If this is a request for a synthetic attribute in the system.*
2597 * namespace use the generic infrastructure to resolve a handler
2598 * for it via sb->s_xattr.
2600 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2601 return generic_getxattr(dentry
, name
, buffer
, size
);
2603 err
= shmem_xattr_validate(name
);
2607 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2610 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2611 const void *value
, size_t size
, int flags
)
2613 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2617 * If this is a request for a synthetic attribute in the system.*
2618 * namespace use the generic infrastructure to resolve a handler
2619 * for it via sb->s_xattr.
2621 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2622 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2624 err
= shmem_xattr_validate(name
);
2628 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2631 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2633 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2637 * If this is a request for a synthetic attribute in the system.*
2638 * namespace use the generic infrastructure to resolve a handler
2639 * for it via sb->s_xattr.
2641 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2642 return generic_removexattr(dentry
, name
);
2644 err
= shmem_xattr_validate(name
);
2648 return simple_xattr_remove(&info
->xattrs
, name
);
2651 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2653 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2654 return simple_xattr_list(&info
->xattrs
, buffer
, size
);
2656 #endif /* CONFIG_TMPFS_XATTR */
2658 static const struct inode_operations shmem_short_symlink_operations
= {
2659 .readlink
= generic_readlink
,
2660 .follow_link
= simple_follow_link
,
2661 #ifdef CONFIG_TMPFS_XATTR
2662 .setxattr
= shmem_setxattr
,
2663 .getxattr
= shmem_getxattr
,
2664 .listxattr
= shmem_listxattr
,
2665 .removexattr
= shmem_removexattr
,
2669 static const struct inode_operations shmem_symlink_inode_operations
= {
2670 .readlink
= generic_readlink
,
2671 .follow_link
= shmem_follow_link
,
2672 .put_link
= shmem_put_link
,
2673 #ifdef CONFIG_TMPFS_XATTR
2674 .setxattr
= shmem_setxattr
,
2675 .getxattr
= shmem_getxattr
,
2676 .listxattr
= shmem_listxattr
,
2677 .removexattr
= shmem_removexattr
,
2681 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2683 return ERR_PTR(-ESTALE
);
2686 static int shmem_match(struct inode
*ino
, void *vfh
)
2690 inum
= (inum
<< 32) | fh
[1];
2691 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2694 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2695 struct fid
*fid
, int fh_len
, int fh_type
)
2697 struct inode
*inode
;
2698 struct dentry
*dentry
= NULL
;
2705 inum
= (inum
<< 32) | fid
->raw
[1];
2707 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2708 shmem_match
, fid
->raw
);
2710 dentry
= d_find_alias(inode
);
2717 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2718 struct inode
*parent
)
2722 return FILEID_INVALID
;
2725 if (inode_unhashed(inode
)) {
2726 /* Unfortunately insert_inode_hash is not idempotent,
2727 * so as we hash inodes here rather than at creation
2728 * time, we need a lock to ensure we only try
2731 static DEFINE_SPINLOCK(lock
);
2733 if (inode_unhashed(inode
))
2734 __insert_inode_hash(inode
,
2735 inode
->i_ino
+ inode
->i_generation
);
2739 fh
[0] = inode
->i_generation
;
2740 fh
[1] = inode
->i_ino
;
2741 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2747 static const struct export_operations shmem_export_ops
= {
2748 .get_parent
= shmem_get_parent
,
2749 .encode_fh
= shmem_encode_fh
,
2750 .fh_to_dentry
= shmem_fh_to_dentry
,
2753 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2756 char *this_char
, *value
, *rest
;
2757 struct mempolicy
*mpol
= NULL
;
2761 while (options
!= NULL
) {
2762 this_char
= options
;
2765 * NUL-terminate this option: unfortunately,
2766 * mount options form a comma-separated list,
2767 * but mpol's nodelist may also contain commas.
2769 options
= strchr(options
, ',');
2770 if (options
== NULL
)
2773 if (!isdigit(*options
)) {
2780 if ((value
= strchr(this_char
,'=')) != NULL
) {
2784 "tmpfs: No value for mount option '%s'\n",
2789 if (!strcmp(this_char
,"size")) {
2790 unsigned long long size
;
2791 size
= memparse(value
,&rest
);
2793 size
<<= PAGE_SHIFT
;
2794 size
*= totalram_pages
;
2800 sbinfo
->max_blocks
=
2801 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2802 } else if (!strcmp(this_char
,"nr_blocks")) {
2803 sbinfo
->max_blocks
= memparse(value
, &rest
);
2806 } else if (!strcmp(this_char
,"nr_inodes")) {
2807 sbinfo
->max_inodes
= memparse(value
, &rest
);
2810 } else if (!strcmp(this_char
,"mode")) {
2813 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2816 } else if (!strcmp(this_char
,"uid")) {
2819 uid
= simple_strtoul(value
, &rest
, 0);
2822 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2823 if (!uid_valid(sbinfo
->uid
))
2825 } else if (!strcmp(this_char
,"gid")) {
2828 gid
= simple_strtoul(value
, &rest
, 0);
2831 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2832 if (!gid_valid(sbinfo
->gid
))
2834 } else if (!strcmp(this_char
,"mpol")) {
2837 if (mpol_parse_str(value
, &mpol
))
2840 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2845 sbinfo
->mpol
= mpol
;
2849 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2857 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2859 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2860 struct shmem_sb_info config
= *sbinfo
;
2861 unsigned long inodes
;
2862 int error
= -EINVAL
;
2865 if (shmem_parse_options(data
, &config
, true))
2868 spin_lock(&sbinfo
->stat_lock
);
2869 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2870 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2872 if (config
.max_inodes
< inodes
)
2875 * Those tests disallow limited->unlimited while any are in use;
2876 * but we must separately disallow unlimited->limited, because
2877 * in that case we have no record of how much is already in use.
2879 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2881 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2885 sbinfo
->max_blocks
= config
.max_blocks
;
2886 sbinfo
->max_inodes
= config
.max_inodes
;
2887 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2890 * Preserve previous mempolicy unless mpol remount option was specified.
2893 mpol_put(sbinfo
->mpol
);
2894 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2897 spin_unlock(&sbinfo
->stat_lock
);
2901 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2903 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2905 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2906 seq_printf(seq
, ",size=%luk",
2907 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2908 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2909 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2910 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2911 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2912 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2913 seq_printf(seq
, ",uid=%u",
2914 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2915 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2916 seq_printf(seq
, ",gid=%u",
2917 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2918 shmem_show_mpol(seq
, sbinfo
->mpol
);
2922 #define MFD_NAME_PREFIX "memfd:"
2923 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2924 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2926 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2928 SYSCALL_DEFINE2(memfd_create
,
2929 const char __user
*, uname
,
2930 unsigned int, flags
)
2932 struct shmem_inode_info
*info
;
2938 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
2941 /* length includes terminating zero */
2942 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
2945 if (len
> MFD_NAME_MAX_LEN
+ 1)
2948 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
2952 strcpy(name
, MFD_NAME_PREFIX
);
2953 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
2958 /* terminating-zero may have changed after strnlen_user() returned */
2959 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
2964 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
2970 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
2972 error
= PTR_ERR(file
);
2975 info
= SHMEM_I(file_inode(file
));
2976 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
2977 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
2978 if (flags
& MFD_ALLOW_SEALING
)
2979 info
->seals
&= ~F_SEAL_SEAL
;
2981 fd_install(fd
, file
);
2992 #endif /* CONFIG_TMPFS */
2994 static void shmem_put_super(struct super_block
*sb
)
2996 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2998 percpu_counter_destroy(&sbinfo
->used_blocks
);
2999 mpol_put(sbinfo
->mpol
);
3001 sb
->s_fs_info
= NULL
;
3004 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3006 struct inode
*inode
;
3007 struct shmem_sb_info
*sbinfo
;
3010 /* Round up to L1_CACHE_BYTES to resist false sharing */
3011 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3012 L1_CACHE_BYTES
), GFP_KERNEL
);
3016 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3017 sbinfo
->uid
= current_fsuid();
3018 sbinfo
->gid
= current_fsgid();
3019 sb
->s_fs_info
= sbinfo
;
3023 * Per default we only allow half of the physical ram per
3024 * tmpfs instance, limiting inodes to one per page of lowmem;
3025 * but the internal instance is left unlimited.
3027 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3028 sbinfo
->max_blocks
= shmem_default_max_blocks();
3029 sbinfo
->max_inodes
= shmem_default_max_inodes();
3030 if (shmem_parse_options(data
, sbinfo
, false)) {
3035 sb
->s_flags
|= MS_NOUSER
;
3037 sb
->s_export_op
= &shmem_export_ops
;
3038 sb
->s_flags
|= MS_NOSEC
;
3040 sb
->s_flags
|= MS_NOUSER
;
3043 spin_lock_init(&sbinfo
->stat_lock
);
3044 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3046 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3048 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3049 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
3050 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
3051 sb
->s_magic
= TMPFS_MAGIC
;
3052 sb
->s_op
= &shmem_ops
;
3053 sb
->s_time_gran
= 1;
3054 #ifdef CONFIG_TMPFS_XATTR
3055 sb
->s_xattr
= shmem_xattr_handlers
;
3057 #ifdef CONFIG_TMPFS_POSIX_ACL
3058 sb
->s_flags
|= MS_POSIXACL
;
3061 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3064 inode
->i_uid
= sbinfo
->uid
;
3065 inode
->i_gid
= sbinfo
->gid
;
3066 sb
->s_root
= d_make_root(inode
);
3072 shmem_put_super(sb
);
3076 static struct kmem_cache
*shmem_inode_cachep
;
3078 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3080 struct shmem_inode_info
*info
;
3081 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3084 return &info
->vfs_inode
;
3087 static void shmem_destroy_callback(struct rcu_head
*head
)
3089 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3090 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3093 static void shmem_destroy_inode(struct inode
*inode
)
3095 if (S_ISREG(inode
->i_mode
))
3096 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3097 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3100 static void shmem_init_inode(void *foo
)
3102 struct shmem_inode_info
*info
= foo
;
3103 inode_init_once(&info
->vfs_inode
);
3106 static int shmem_init_inodecache(void)
3108 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3109 sizeof(struct shmem_inode_info
),
3110 0, SLAB_PANIC
, shmem_init_inode
);
3114 static void shmem_destroy_inodecache(void)
3116 kmem_cache_destroy(shmem_inode_cachep
);
3119 static const struct address_space_operations shmem_aops
= {
3120 .writepage
= shmem_writepage
,
3121 .set_page_dirty
= __set_page_dirty_no_writeback
,
3123 .write_begin
= shmem_write_begin
,
3124 .write_end
= shmem_write_end
,
3126 #ifdef CONFIG_MIGRATION
3127 .migratepage
= migrate_page
,
3129 .error_remove_page
= generic_error_remove_page
,
3132 static const struct file_operations shmem_file_operations
= {
3135 .llseek
= shmem_file_llseek
,
3136 .read_iter
= shmem_file_read_iter
,
3137 .write_iter
= generic_file_write_iter
,
3138 .fsync
= noop_fsync
,
3139 .splice_read
= shmem_file_splice_read
,
3140 .splice_write
= iter_file_splice_write
,
3141 .fallocate
= shmem_fallocate
,
3145 static const struct inode_operations shmem_inode_operations
= {
3146 .getattr
= shmem_getattr
,
3147 .setattr
= shmem_setattr
,
3148 #ifdef CONFIG_TMPFS_XATTR
3149 .setxattr
= shmem_setxattr
,
3150 .getxattr
= shmem_getxattr
,
3151 .listxattr
= shmem_listxattr
,
3152 .removexattr
= shmem_removexattr
,
3153 .set_acl
= simple_set_acl
,
3157 static const struct inode_operations shmem_dir_inode_operations
= {
3159 .create
= shmem_create
,
3160 .lookup
= simple_lookup
,
3162 .unlink
= shmem_unlink
,
3163 .symlink
= shmem_symlink
,
3164 .mkdir
= shmem_mkdir
,
3165 .rmdir
= shmem_rmdir
,
3166 .mknod
= shmem_mknod
,
3167 .rename2
= shmem_rename2
,
3168 .tmpfile
= shmem_tmpfile
,
3170 #ifdef CONFIG_TMPFS_XATTR
3171 .setxattr
= shmem_setxattr
,
3172 .getxattr
= shmem_getxattr
,
3173 .listxattr
= shmem_listxattr
,
3174 .removexattr
= shmem_removexattr
,
3176 #ifdef CONFIG_TMPFS_POSIX_ACL
3177 .setattr
= shmem_setattr
,
3178 .set_acl
= simple_set_acl
,
3182 static const struct inode_operations shmem_special_inode_operations
= {
3183 #ifdef CONFIG_TMPFS_XATTR
3184 .setxattr
= shmem_setxattr
,
3185 .getxattr
= shmem_getxattr
,
3186 .listxattr
= shmem_listxattr
,
3187 .removexattr
= shmem_removexattr
,
3189 #ifdef CONFIG_TMPFS_POSIX_ACL
3190 .setattr
= shmem_setattr
,
3191 .set_acl
= simple_set_acl
,
3195 static const struct super_operations shmem_ops
= {
3196 .alloc_inode
= shmem_alloc_inode
,
3197 .destroy_inode
= shmem_destroy_inode
,
3199 .statfs
= shmem_statfs
,
3200 .remount_fs
= shmem_remount_fs
,
3201 .show_options
= shmem_show_options
,
3203 .evict_inode
= shmem_evict_inode
,
3204 .drop_inode
= generic_delete_inode
,
3205 .put_super
= shmem_put_super
,
3208 static const struct vm_operations_struct shmem_vm_ops
= {
3209 .fault
= shmem_fault
,
3210 .map_pages
= filemap_map_pages
,
3212 .set_policy
= shmem_set_policy
,
3213 .get_policy
= shmem_get_policy
,
3217 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3218 int flags
, const char *dev_name
, void *data
)
3220 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3223 static struct file_system_type shmem_fs_type
= {
3224 .owner
= THIS_MODULE
,
3226 .mount
= shmem_mount
,
3227 .kill_sb
= kill_litter_super
,
3228 .fs_flags
= FS_USERNS_MOUNT
,
3231 int __init
shmem_init(void)
3235 /* If rootfs called this, don't re-init */
3236 if (shmem_inode_cachep
)
3239 error
= shmem_init_inodecache();
3243 error
= register_filesystem(&shmem_fs_type
);
3245 printk(KERN_ERR
"Could not register tmpfs\n");
3249 shm_mnt
= kern_mount(&shmem_fs_type
);
3250 if (IS_ERR(shm_mnt
)) {
3251 error
= PTR_ERR(shm_mnt
);
3252 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
3258 unregister_filesystem(&shmem_fs_type
);
3260 shmem_destroy_inodecache();
3262 shm_mnt
= ERR_PTR(error
);
3266 #else /* !CONFIG_SHMEM */
3269 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3271 * This is intended for small system where the benefits of the full
3272 * shmem code (swap-backed and resource-limited) are outweighed by
3273 * their complexity. On systems without swap this code should be
3274 * effectively equivalent, but much lighter weight.
3277 static struct file_system_type shmem_fs_type
= {
3279 .mount
= ramfs_mount
,
3280 .kill_sb
= kill_litter_super
,
3281 .fs_flags
= FS_USERNS_MOUNT
,
3284 int __init
shmem_init(void)
3286 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3288 shm_mnt
= kern_mount(&shmem_fs_type
);
3289 BUG_ON(IS_ERR(shm_mnt
));
3294 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3299 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3304 void shmem_unlock_mapping(struct address_space
*mapping
)
3308 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3310 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3312 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3314 #define shmem_vm_ops generic_file_vm_ops
3315 #define shmem_file_operations ramfs_file_operations
3316 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3317 #define shmem_acct_size(flags, size) 0
3318 #define shmem_unacct_size(flags, size) do {} while (0)
3320 #endif /* CONFIG_SHMEM */
3324 static struct dentry_operations anon_ops
= {
3325 .d_dname
= simple_dname
3328 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3329 unsigned long flags
, unsigned int i_flags
)
3332 struct inode
*inode
;
3334 struct super_block
*sb
;
3337 if (IS_ERR(shm_mnt
))
3338 return ERR_CAST(shm_mnt
);
3340 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3341 return ERR_PTR(-EINVAL
);
3343 if (shmem_acct_size(flags
, size
))
3344 return ERR_PTR(-ENOMEM
);
3346 res
= ERR_PTR(-ENOMEM
);
3348 this.len
= strlen(name
);
3349 this.hash
= 0; /* will go */
3350 sb
= shm_mnt
->mnt_sb
;
3351 path
.mnt
= mntget(shm_mnt
);
3352 path
.dentry
= d_alloc_pseudo(sb
, &this);
3355 d_set_d_op(path
.dentry
, &anon_ops
);
3357 res
= ERR_PTR(-ENOSPC
);
3358 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3362 inode
->i_flags
|= i_flags
;
3363 d_instantiate(path
.dentry
, inode
);
3364 inode
->i_size
= size
;
3365 clear_nlink(inode
); /* It is unlinked */
3366 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3370 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3371 &shmem_file_operations
);
3378 shmem_unacct_size(flags
, size
);
3385 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3386 * kernel internal. There will be NO LSM permission checks against the
3387 * underlying inode. So users of this interface must do LSM checks at a
3388 * higher layer. The users are the big_key and shm implementations. LSM
3389 * checks are provided at the key or shm level rather than the inode.
3390 * @name: name for dentry (to be seen in /proc/<pid>/maps
3391 * @size: size to be set for the file
3392 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3394 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3396 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3400 * shmem_file_setup - get an unlinked file living in tmpfs
3401 * @name: name for dentry (to be seen in /proc/<pid>/maps
3402 * @size: size to be set for the file
3403 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3405 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3407 return __shmem_file_setup(name
, size
, flags
, 0);
3409 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3412 * shmem_zero_setup - setup a shared anonymous mapping
3413 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3415 int shmem_zero_setup(struct vm_area_struct
*vma
)
3418 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3421 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3422 * between XFS directory reading and selinux: since this file is only
3423 * accessible to the user through its mapping, use S_PRIVATE flag to
3424 * bypass file security, in the same way as shmem_kernel_file_setup().
3426 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
3428 return PTR_ERR(file
);
3432 vma
->vm_file
= file
;
3433 vma
->vm_ops
= &shmem_vm_ops
;
3438 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3439 * @mapping: the page's address_space
3440 * @index: the page index
3441 * @gfp: the page allocator flags to use if allocating
3443 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3444 * with any new page allocations done using the specified allocation flags.
3445 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3446 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3447 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3449 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3450 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3452 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3453 pgoff_t index
, gfp_t gfp
)
3456 struct inode
*inode
= mapping
->host
;
3460 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3461 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3463 page
= ERR_PTR(error
);
3469 * The tiny !SHMEM case uses ramfs without swap
3471 return read_cache_page_gfp(mapping
, index
, gfp
);
3474 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);