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>
35 #include <linux/khugepaged.h>
37 static struct vfsmount
*shm_mnt
;
41 * This virtual memory filesystem is heavily based on the ramfs. It
42 * extends ramfs by the ability to use swap and honor resource limits
43 * which makes it a completely usable filesystem.
46 #include <linux/xattr.h>
47 #include <linux/exportfs.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/mman.h>
51 #include <linux/string.h>
52 #include <linux/slab.h>
53 #include <linux/backing-dev.h>
54 #include <linux/shmem_fs.h>
55 #include <linux/writeback.h>
56 #include <linux/blkdev.h>
57 #include <linux/pagevec.h>
58 #include <linux/percpu_counter.h>
59 #include <linux/falloc.h>
60 #include <linux/splice.h>
61 #include <linux/security.h>
62 #include <linux/swapops.h>
63 #include <linux/mempolicy.h>
64 #include <linux/namei.h>
65 #include <linux/ctype.h>
66 #include <linux/migrate.h>
67 #include <linux/highmem.h>
68 #include <linux/seq_file.h>
69 #include <linux/magic.h>
70 #include <linux/syscalls.h>
71 #include <linux/fcntl.h>
72 #include <uapi/linux/memfd.h>
74 #include <asm/uaccess.h>
75 #include <asm/pgtable.h>
79 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
80 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
82 /* Pretend that each entry is of this size in directory's i_size */
83 #define BOGO_DIRENT_SIZE 20
85 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
86 #define SHORT_SYMLINK_LEN 128
89 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
90 * inode->i_private (with i_mutex making sure that it has only one user at
91 * a time): we would prefer not to enlarge the shmem inode just for that.
94 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
95 pgoff_t start
; /* start of range currently being fallocated */
96 pgoff_t next
; /* the next page offset to be fallocated */
97 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
98 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
102 static unsigned long shmem_default_max_blocks(void)
104 return totalram_pages
/ 2;
107 static unsigned long shmem_default_max_inodes(void)
109 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
113 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
114 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
115 struct shmem_inode_info
*info
, pgoff_t index
);
116 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
117 struct page
**pagep
, enum sgp_type sgp
,
118 gfp_t gfp
, struct mm_struct
*fault_mm
, int *fault_type
);
120 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
121 struct page
**pagep
, enum sgp_type sgp
)
123 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
124 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
);
127 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
129 return sb
->s_fs_info
;
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
138 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
140 return (flags
& VM_NORESERVE
) ?
141 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
144 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
146 if (!(flags
& VM_NORESERVE
))
147 vm_unacct_memory(VM_ACCT(size
));
150 static inline int shmem_reacct_size(unsigned long flags
,
151 loff_t oldsize
, loff_t newsize
)
153 if (!(flags
& VM_NORESERVE
)) {
154 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
155 return security_vm_enough_memory_mm(current
->mm
,
156 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
157 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
158 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
164 * ... whereas tmpfs objects are accounted incrementally as
165 * pages are allocated, in order to allow large sparse files.
166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169 static inline int shmem_acct_block(unsigned long flags
, long pages
)
171 if (!(flags
& VM_NORESERVE
))
174 return security_vm_enough_memory_mm(current
->mm
,
175 pages
* VM_ACCT(PAGE_SIZE
));
178 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
180 if (flags
& VM_NORESERVE
)
181 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
184 static const struct super_operations shmem_ops
;
185 static const struct address_space_operations shmem_aops
;
186 static const struct file_operations shmem_file_operations
;
187 static const struct inode_operations shmem_inode_operations
;
188 static const struct inode_operations shmem_dir_inode_operations
;
189 static const struct inode_operations shmem_special_inode_operations
;
190 static const struct vm_operations_struct shmem_vm_ops
;
191 static struct file_system_type shmem_fs_type
;
193 static LIST_HEAD(shmem_swaplist
);
194 static DEFINE_MUTEX(shmem_swaplist_mutex
);
196 static int shmem_reserve_inode(struct super_block
*sb
)
198 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
199 if (sbinfo
->max_inodes
) {
200 spin_lock(&sbinfo
->stat_lock
);
201 if (!sbinfo
->free_inodes
) {
202 spin_unlock(&sbinfo
->stat_lock
);
205 sbinfo
->free_inodes
--;
206 spin_unlock(&sbinfo
->stat_lock
);
211 static void shmem_free_inode(struct super_block
*sb
)
213 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
214 if (sbinfo
->max_inodes
) {
215 spin_lock(&sbinfo
->stat_lock
);
216 sbinfo
->free_inodes
++;
217 spin_unlock(&sbinfo
->stat_lock
);
222 * shmem_recalc_inode - recalculate the block usage of an inode
223 * @inode: inode to recalc
225 * We have to calculate the free blocks since the mm can drop
226 * undirtied hole pages behind our back.
228 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
229 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
231 * It has to be called with the spinlock held.
233 static void shmem_recalc_inode(struct inode
*inode
)
235 struct shmem_inode_info
*info
= SHMEM_I(inode
);
238 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
240 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
241 if (sbinfo
->max_blocks
)
242 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
243 info
->alloced
-= freed
;
244 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
245 shmem_unacct_blocks(info
->flags
, freed
);
249 bool shmem_charge(struct inode
*inode
, long pages
)
251 struct shmem_inode_info
*info
= SHMEM_I(inode
);
252 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
255 if (shmem_acct_block(info
->flags
, pages
))
257 spin_lock_irqsave(&info
->lock
, flags
);
258 info
->alloced
+= pages
;
259 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
260 shmem_recalc_inode(inode
);
261 spin_unlock_irqrestore(&info
->lock
, flags
);
262 inode
->i_mapping
->nrpages
+= pages
;
264 if (!sbinfo
->max_blocks
)
266 if (percpu_counter_compare(&sbinfo
->used_blocks
,
267 sbinfo
->max_blocks
- pages
) > 0) {
268 inode
->i_mapping
->nrpages
-= pages
;
269 spin_lock_irqsave(&info
->lock
, flags
);
270 info
->alloced
-= pages
;
271 shmem_recalc_inode(inode
);
272 spin_unlock_irqrestore(&info
->lock
, flags
);
276 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
280 void shmem_uncharge(struct inode
*inode
, long pages
)
282 struct shmem_inode_info
*info
= SHMEM_I(inode
);
283 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
286 spin_lock_irqsave(&info
->lock
, flags
);
287 info
->alloced
-= pages
;
288 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
289 shmem_recalc_inode(inode
);
290 spin_unlock_irqrestore(&info
->lock
, flags
);
292 if (sbinfo
->max_blocks
)
293 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
297 * Replace item expected in radix tree by a new item, while holding tree lock.
299 static int shmem_radix_tree_replace(struct address_space
*mapping
,
300 pgoff_t index
, void *expected
, void *replacement
)
305 VM_BUG_ON(!expected
);
306 VM_BUG_ON(!replacement
);
307 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
310 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
311 if (item
!= expected
)
313 radix_tree_replace_slot(pslot
, replacement
);
318 * Sometimes, before we decide whether to proceed or to fail, we must check
319 * that an entry was not already brought back from swap by a racing thread.
321 * Checking page is not enough: by the time a SwapCache page is locked, it
322 * might be reused, and again be SwapCache, using the same swap as before.
324 static bool shmem_confirm_swap(struct address_space
*mapping
,
325 pgoff_t index
, swp_entry_t swap
)
330 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
332 return item
== swp_to_radix_entry(swap
);
336 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
339 * disables huge pages for the mount;
341 * enables huge pages for the mount;
342 * SHMEM_HUGE_WITHIN_SIZE:
343 * only allocate huge pages if the page will be fully within i_size,
344 * also respect fadvise()/madvise() hints;
346 * only allocate huge pages if requested with fadvise()/madvise();
349 #define SHMEM_HUGE_NEVER 0
350 #define SHMEM_HUGE_ALWAYS 1
351 #define SHMEM_HUGE_WITHIN_SIZE 2
352 #define SHMEM_HUGE_ADVISE 3
356 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
359 * disables huge on shm_mnt and all mounts, for emergency use;
361 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
364 #define SHMEM_HUGE_DENY (-1)
365 #define SHMEM_HUGE_FORCE (-2)
367 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
368 /* ifdef here to avoid bloating shmem.o when not necessary */
370 int shmem_huge __read_mostly
;
372 static int shmem_parse_huge(const char *str
)
374 if (!strcmp(str
, "never"))
375 return SHMEM_HUGE_NEVER
;
376 if (!strcmp(str
, "always"))
377 return SHMEM_HUGE_ALWAYS
;
378 if (!strcmp(str
, "within_size"))
379 return SHMEM_HUGE_WITHIN_SIZE
;
380 if (!strcmp(str
, "advise"))
381 return SHMEM_HUGE_ADVISE
;
382 if (!strcmp(str
, "deny"))
383 return SHMEM_HUGE_DENY
;
384 if (!strcmp(str
, "force"))
385 return SHMEM_HUGE_FORCE
;
389 static const char *shmem_format_huge(int huge
)
392 case SHMEM_HUGE_NEVER
:
394 case SHMEM_HUGE_ALWAYS
:
396 case SHMEM_HUGE_WITHIN_SIZE
:
397 return "within_size";
398 case SHMEM_HUGE_ADVISE
:
400 case SHMEM_HUGE_DENY
:
402 case SHMEM_HUGE_FORCE
:
410 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
411 struct shrink_control
*sc
, unsigned long nr_to_split
)
413 LIST_HEAD(list
), *pos
, *next
;
415 struct shmem_inode_info
*info
;
417 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
418 int removed
= 0, split
= 0;
420 if (list_empty(&sbinfo
->shrinklist
))
423 spin_lock(&sbinfo
->shrinklist_lock
);
424 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
425 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
428 inode
= igrab(&info
->vfs_inode
);
430 /* inode is about to be evicted */
432 list_del_init(&info
->shrinklist
);
437 /* Check if there's anything to gain */
438 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
439 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
440 list_del_init(&info
->shrinklist
);
446 list_move(&info
->shrinklist
, &list
);
451 spin_unlock(&sbinfo
->shrinklist_lock
);
453 list_for_each_safe(pos
, next
, &list
) {
456 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
457 inode
= &info
->vfs_inode
;
459 if (nr_to_split
&& split
>= nr_to_split
) {
464 page
= find_lock_page(inode
->i_mapping
,
465 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
469 if (!PageTransHuge(page
)) {
475 ret
= split_huge_page(page
);
480 /* split failed: leave it on the list */
487 list_del_init(&info
->shrinklist
);
492 spin_lock(&sbinfo
->shrinklist_lock
);
493 list_splice_tail(&list
, &sbinfo
->shrinklist
);
494 sbinfo
->shrinklist_len
-= removed
;
495 spin_unlock(&sbinfo
->shrinklist_lock
);
500 static long shmem_unused_huge_scan(struct super_block
*sb
,
501 struct shrink_control
*sc
)
503 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
505 if (!READ_ONCE(sbinfo
->shrinklist_len
))
508 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
511 static long shmem_unused_huge_count(struct super_block
*sb
,
512 struct shrink_control
*sc
)
514 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
515 return READ_ONCE(sbinfo
->shrinklist_len
);
517 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
519 #define shmem_huge SHMEM_HUGE_DENY
521 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
522 struct shrink_control
*sc
, unsigned long nr_to_split
)
526 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
529 * Like add_to_page_cache_locked, but error if expected item has gone.
531 static int shmem_add_to_page_cache(struct page
*page
,
532 struct address_space
*mapping
,
533 pgoff_t index
, void *expected
)
535 int error
, nr
= hpage_nr_pages(page
);
537 VM_BUG_ON_PAGE(PageTail(page
), page
);
538 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
539 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
540 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
541 VM_BUG_ON(expected
&& PageTransHuge(page
));
543 page_ref_add(page
, nr
);
544 page
->mapping
= mapping
;
547 spin_lock_irq(&mapping
->tree_lock
);
548 if (PageTransHuge(page
)) {
549 void __rcu
**results
;
554 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
,
555 &results
, &idx
, index
, 1) &&
556 idx
< index
+ HPAGE_PMD_NR
) {
561 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
562 error
= radix_tree_insert(&mapping
->page_tree
,
563 index
+ i
, page
+ i
);
566 count_vm_event(THP_FILE_ALLOC
);
568 } else if (!expected
) {
569 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
571 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
576 mapping
->nrpages
+= nr
;
577 if (PageTransHuge(page
))
578 __inc_node_page_state(page
, NR_SHMEM_THPS
);
579 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
580 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
581 spin_unlock_irq(&mapping
->tree_lock
);
583 page
->mapping
= NULL
;
584 spin_unlock_irq(&mapping
->tree_lock
);
585 page_ref_sub(page
, nr
);
591 * Like delete_from_page_cache, but substitutes swap for page.
593 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
595 struct address_space
*mapping
= page
->mapping
;
598 VM_BUG_ON_PAGE(PageCompound(page
), page
);
600 spin_lock_irq(&mapping
->tree_lock
);
601 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
602 page
->mapping
= NULL
;
604 __dec_node_page_state(page
, NR_FILE_PAGES
);
605 __dec_node_page_state(page
, NR_SHMEM
);
606 spin_unlock_irq(&mapping
->tree_lock
);
612 * Remove swap entry from radix tree, free the swap and its page cache.
614 static int shmem_free_swap(struct address_space
*mapping
,
615 pgoff_t index
, void *radswap
)
619 spin_lock_irq(&mapping
->tree_lock
);
620 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
621 spin_unlock_irq(&mapping
->tree_lock
);
624 free_swap_and_cache(radix_to_swp_entry(radswap
));
629 * Determine (in bytes) how many of the shmem object's pages mapped by the
630 * given offsets are swapped out.
632 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
633 * as long as the inode doesn't go away and racy results are not a problem.
635 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
636 pgoff_t start
, pgoff_t end
)
638 struct radix_tree_iter iter
;
641 unsigned long swapped
= 0;
645 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
646 if (iter
.index
>= end
)
649 page
= radix_tree_deref_slot(slot
);
651 if (radix_tree_deref_retry(page
)) {
652 slot
= radix_tree_iter_retry(&iter
);
656 if (radix_tree_exceptional_entry(page
))
659 if (need_resched()) {
661 slot
= radix_tree_iter_next(&iter
);
667 return swapped
<< PAGE_SHIFT
;
671 * Determine (in bytes) how many of the shmem object's pages mapped by the
672 * given vma is swapped out.
674 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
675 * as long as the inode doesn't go away and racy results are not a problem.
677 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
679 struct inode
*inode
= file_inode(vma
->vm_file
);
680 struct shmem_inode_info
*info
= SHMEM_I(inode
);
681 struct address_space
*mapping
= inode
->i_mapping
;
682 unsigned long swapped
;
684 /* Be careful as we don't hold info->lock */
685 swapped
= READ_ONCE(info
->swapped
);
688 * The easier cases are when the shmem object has nothing in swap, or
689 * the vma maps it whole. Then we can simply use the stats that we
695 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
696 return swapped
<< PAGE_SHIFT
;
698 /* Here comes the more involved part */
699 return shmem_partial_swap_usage(mapping
,
700 linear_page_index(vma
, vma
->vm_start
),
701 linear_page_index(vma
, vma
->vm_end
));
705 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
707 void shmem_unlock_mapping(struct address_space
*mapping
)
710 pgoff_t indices
[PAGEVEC_SIZE
];
713 pagevec_init(&pvec
, 0);
715 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
717 while (!mapping_unevictable(mapping
)) {
719 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
720 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
722 pvec
.nr
= find_get_entries(mapping
, index
,
723 PAGEVEC_SIZE
, pvec
.pages
, indices
);
726 index
= indices
[pvec
.nr
- 1] + 1;
727 pagevec_remove_exceptionals(&pvec
);
728 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
729 pagevec_release(&pvec
);
735 * Remove range of pages and swap entries from radix tree, and free them.
736 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
738 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
741 struct address_space
*mapping
= inode
->i_mapping
;
742 struct shmem_inode_info
*info
= SHMEM_I(inode
);
743 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
744 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
745 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
746 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
748 pgoff_t indices
[PAGEVEC_SIZE
];
749 long nr_swaps_freed
= 0;
754 end
= -1; /* unsigned, so actually very big */
756 pagevec_init(&pvec
, 0);
758 while (index
< end
) {
759 pvec
.nr
= find_get_entries(mapping
, index
,
760 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
761 pvec
.pages
, indices
);
764 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
765 struct page
*page
= pvec
.pages
[i
];
771 if (radix_tree_exceptional_entry(page
)) {
774 nr_swaps_freed
+= !shmem_free_swap(mapping
,
779 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
781 if (!trylock_page(page
))
784 if (PageTransTail(page
)) {
785 /* Middle of THP: zero out the page */
786 clear_highpage(page
);
789 } else if (PageTransHuge(page
)) {
790 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
792 * Range ends in the middle of THP:
795 clear_highpage(page
);
799 index
+= HPAGE_PMD_NR
- 1;
800 i
+= HPAGE_PMD_NR
- 1;
803 if (!unfalloc
|| !PageUptodate(page
)) {
804 VM_BUG_ON_PAGE(PageTail(page
), page
);
805 if (page_mapping(page
) == mapping
) {
806 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
807 truncate_inode_page(mapping
, page
);
812 pagevec_remove_exceptionals(&pvec
);
813 pagevec_release(&pvec
);
819 struct page
*page
= NULL
;
820 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
822 unsigned int top
= PAGE_SIZE
;
827 zero_user_segment(page
, partial_start
, top
);
828 set_page_dirty(page
);
834 struct page
*page
= NULL
;
835 shmem_getpage(inode
, end
, &page
, SGP_READ
);
837 zero_user_segment(page
, 0, partial_end
);
838 set_page_dirty(page
);
847 while (index
< end
) {
850 pvec
.nr
= find_get_entries(mapping
, index
,
851 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
852 pvec
.pages
, indices
);
854 /* If all gone or hole-punch or unfalloc, we're done */
855 if (index
== start
|| end
!= -1)
857 /* But if truncating, restart to make sure all gone */
861 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
862 struct page
*page
= pvec
.pages
[i
];
868 if (radix_tree_exceptional_entry(page
)) {
871 if (shmem_free_swap(mapping
, index
, page
)) {
872 /* Swap was replaced by page: retry */
882 if (PageTransTail(page
)) {
883 /* Middle of THP: zero out the page */
884 clear_highpage(page
);
887 * Partial thp truncate due 'start' in middle
888 * of THP: don't need to look on these pages
889 * again on !pvec.nr restart.
891 if (index
!= round_down(end
, HPAGE_PMD_NR
))
894 } else if (PageTransHuge(page
)) {
895 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
897 * Range ends in the middle of THP:
900 clear_highpage(page
);
904 index
+= HPAGE_PMD_NR
- 1;
905 i
+= HPAGE_PMD_NR
- 1;
908 if (!unfalloc
|| !PageUptodate(page
)) {
909 VM_BUG_ON_PAGE(PageTail(page
), page
);
910 if (page_mapping(page
) == mapping
) {
911 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
912 truncate_inode_page(mapping
, page
);
914 /* Page was replaced by swap: retry */
922 pagevec_remove_exceptionals(&pvec
);
923 pagevec_release(&pvec
);
927 spin_lock_irq(&info
->lock
);
928 info
->swapped
-= nr_swaps_freed
;
929 shmem_recalc_inode(inode
);
930 spin_unlock_irq(&info
->lock
);
933 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
935 shmem_undo_range(inode
, lstart
, lend
, false);
936 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
938 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
940 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
943 struct inode
*inode
= dentry
->d_inode
;
944 struct shmem_inode_info
*info
= SHMEM_I(inode
);
946 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
947 spin_lock_irq(&info
->lock
);
948 shmem_recalc_inode(inode
);
949 spin_unlock_irq(&info
->lock
);
951 generic_fillattr(inode
, stat
);
955 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
957 struct inode
*inode
= d_inode(dentry
);
958 struct shmem_inode_info
*info
= SHMEM_I(inode
);
959 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
962 error
= inode_change_ok(inode
, attr
);
966 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
967 loff_t oldsize
= inode
->i_size
;
968 loff_t newsize
= attr
->ia_size
;
970 /* protected by i_mutex */
971 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
972 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
975 if (newsize
!= oldsize
) {
976 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
980 i_size_write(inode
, newsize
);
981 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
983 if (newsize
<= oldsize
) {
984 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
985 if (oldsize
> holebegin
)
986 unmap_mapping_range(inode
->i_mapping
,
989 shmem_truncate_range(inode
,
990 newsize
, (loff_t
)-1);
991 /* unmap again to remove racily COWed private pages */
992 if (oldsize
> holebegin
)
993 unmap_mapping_range(inode
->i_mapping
,
997 * Part of the huge page can be beyond i_size: subject
998 * to shrink under memory pressure.
1000 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1001 spin_lock(&sbinfo
->shrinklist_lock
);
1002 if (list_empty(&info
->shrinklist
)) {
1003 list_add_tail(&info
->shrinklist
,
1004 &sbinfo
->shrinklist
);
1005 sbinfo
->shrinklist_len
++;
1007 spin_unlock(&sbinfo
->shrinklist_lock
);
1012 setattr_copy(inode
, attr
);
1013 if (attr
->ia_valid
& ATTR_MODE
)
1014 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1018 static void shmem_evict_inode(struct inode
*inode
)
1020 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1021 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1023 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1024 shmem_unacct_size(info
->flags
, inode
->i_size
);
1026 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1027 if (!list_empty(&info
->shrinklist
)) {
1028 spin_lock(&sbinfo
->shrinklist_lock
);
1029 if (!list_empty(&info
->shrinklist
)) {
1030 list_del_init(&info
->shrinklist
);
1031 sbinfo
->shrinklist_len
--;
1033 spin_unlock(&sbinfo
->shrinklist_lock
);
1035 if (!list_empty(&info
->swaplist
)) {
1036 mutex_lock(&shmem_swaplist_mutex
);
1037 list_del_init(&info
->swaplist
);
1038 mutex_unlock(&shmem_swaplist_mutex
);
1042 simple_xattrs_free(&info
->xattrs
);
1043 WARN_ON(inode
->i_blocks
);
1044 shmem_free_inode(inode
->i_sb
);
1049 * If swap found in inode, free it and move page from swapcache to filecache.
1051 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
1052 swp_entry_t swap
, struct page
**pagep
)
1054 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1060 radswap
= swp_to_radix_entry(swap
);
1061 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
1063 return -EAGAIN
; /* tell shmem_unuse we found nothing */
1066 * Move _head_ to start search for next from here.
1067 * But be careful: shmem_evict_inode checks list_empty without taking
1068 * mutex, and there's an instant in list_move_tail when info->swaplist
1069 * would appear empty, if it were the only one on shmem_swaplist.
1071 if (shmem_swaplist
.next
!= &info
->swaplist
)
1072 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
1074 gfp
= mapping_gfp_mask(mapping
);
1075 if (shmem_should_replace_page(*pagep
, gfp
)) {
1076 mutex_unlock(&shmem_swaplist_mutex
);
1077 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
1078 mutex_lock(&shmem_swaplist_mutex
);
1080 * We needed to drop mutex to make that restrictive page
1081 * allocation, but the inode might have been freed while we
1082 * dropped it: although a racing shmem_evict_inode() cannot
1083 * complete without emptying the radix_tree, our page lock
1084 * on this swapcache page is not enough to prevent that -
1085 * free_swap_and_cache() of our swap entry will only
1086 * trylock_page(), removing swap from radix_tree whatever.
1088 * We must not proceed to shmem_add_to_page_cache() if the
1089 * inode has been freed, but of course we cannot rely on
1090 * inode or mapping or info to check that. However, we can
1091 * safely check if our swap entry is still in use (and here
1092 * it can't have got reused for another page): if it's still
1093 * in use, then the inode cannot have been freed yet, and we
1094 * can safely proceed (if it's no longer in use, that tells
1095 * nothing about the inode, but we don't need to unuse swap).
1097 if (!page_swapcount(*pagep
))
1102 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1103 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1104 * beneath us (pagelock doesn't help until the page is in pagecache).
1107 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
1109 if (error
!= -ENOMEM
) {
1111 * Truncation and eviction use free_swap_and_cache(), which
1112 * only does trylock page: if we raced, best clean up here.
1114 delete_from_swap_cache(*pagep
);
1115 set_page_dirty(*pagep
);
1117 spin_lock_irq(&info
->lock
);
1119 spin_unlock_irq(&info
->lock
);
1127 * Search through swapped inodes to find and replace swap by page.
1129 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1131 struct list_head
*this, *next
;
1132 struct shmem_inode_info
*info
;
1133 struct mem_cgroup
*memcg
;
1137 * There's a faint possibility that swap page was replaced before
1138 * caller locked it: caller will come back later with the right page.
1140 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1144 * Charge page using GFP_KERNEL while we can wait, before taking
1145 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1146 * Charged back to the user (not to caller) when swap account is used.
1148 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1152 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1155 mutex_lock(&shmem_swaplist_mutex
);
1156 list_for_each_safe(this, next
, &shmem_swaplist
) {
1157 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1159 error
= shmem_unuse_inode(info
, swap
, &page
);
1161 list_del_init(&info
->swaplist
);
1163 if (error
!= -EAGAIN
)
1165 /* found nothing in this: move on to search the next */
1167 mutex_unlock(&shmem_swaplist_mutex
);
1170 if (error
!= -ENOMEM
)
1172 mem_cgroup_cancel_charge(page
, memcg
, false);
1174 mem_cgroup_commit_charge(page
, memcg
, true, false);
1182 * Move the page from the page cache to the swap cache.
1184 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1186 struct shmem_inode_info
*info
;
1187 struct address_space
*mapping
;
1188 struct inode
*inode
;
1192 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1193 BUG_ON(!PageLocked(page
));
1194 mapping
= page
->mapping
;
1195 index
= page
->index
;
1196 inode
= mapping
->host
;
1197 info
= SHMEM_I(inode
);
1198 if (info
->flags
& VM_LOCKED
)
1200 if (!total_swap_pages
)
1204 * Our capabilities prevent regular writeback or sync from ever calling
1205 * shmem_writepage; but a stacking filesystem might use ->writepage of
1206 * its underlying filesystem, in which case tmpfs should write out to
1207 * swap only in response to memory pressure, and not for the writeback
1210 if (!wbc
->for_reclaim
) {
1211 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1216 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1217 * value into swapfile.c, the only way we can correctly account for a
1218 * fallocated page arriving here is now to initialize it and write it.
1220 * That's okay for a page already fallocated earlier, but if we have
1221 * not yet completed the fallocation, then (a) we want to keep track
1222 * of this page in case we have to undo it, and (b) it may not be a
1223 * good idea to continue anyway, once we're pushing into swap. So
1224 * reactivate the page, and let shmem_fallocate() quit when too many.
1226 if (!PageUptodate(page
)) {
1227 if (inode
->i_private
) {
1228 struct shmem_falloc
*shmem_falloc
;
1229 spin_lock(&inode
->i_lock
);
1230 shmem_falloc
= inode
->i_private
;
1232 !shmem_falloc
->waitq
&&
1233 index
>= shmem_falloc
->start
&&
1234 index
< shmem_falloc
->next
)
1235 shmem_falloc
->nr_unswapped
++;
1237 shmem_falloc
= NULL
;
1238 spin_unlock(&inode
->i_lock
);
1242 clear_highpage(page
);
1243 flush_dcache_page(page
);
1244 SetPageUptodate(page
);
1247 swap
= get_swap_page();
1251 if (mem_cgroup_try_charge_swap(page
, swap
))
1255 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1256 * if it's not already there. Do it now before the page is
1257 * moved to swap cache, when its pagelock no longer protects
1258 * the inode from eviction. But don't unlock the mutex until
1259 * we've incremented swapped, because shmem_unuse_inode() will
1260 * prune a !swapped inode from the swaplist under this mutex.
1262 mutex_lock(&shmem_swaplist_mutex
);
1263 if (list_empty(&info
->swaplist
))
1264 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1266 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1267 spin_lock_irq(&info
->lock
);
1268 shmem_recalc_inode(inode
);
1270 spin_unlock_irq(&info
->lock
);
1272 swap_shmem_alloc(swap
);
1273 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1275 mutex_unlock(&shmem_swaplist_mutex
);
1276 BUG_ON(page_mapped(page
));
1277 swap_writepage(page
, wbc
);
1281 mutex_unlock(&shmem_swaplist_mutex
);
1283 swapcache_free(swap
);
1285 set_page_dirty(page
);
1286 if (wbc
->for_reclaim
)
1287 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1292 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1293 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1297 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1298 return; /* show nothing */
1300 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1302 seq_printf(seq
, ",mpol=%s", buffer
);
1305 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1307 struct mempolicy
*mpol
= NULL
;
1309 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1310 mpol
= sbinfo
->mpol
;
1312 spin_unlock(&sbinfo
->stat_lock
);
1316 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1317 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1320 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1324 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1326 #define vm_policy vm_private_data
1329 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1330 struct shmem_inode_info
*info
, pgoff_t index
)
1332 /* Create a pseudo vma that just contains the policy */
1334 /* Bias interleave by inode number to distribute better across nodes */
1335 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1337 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1340 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1342 /* Drop reference taken by mpol_shared_policy_lookup() */
1343 mpol_cond_put(vma
->vm_policy
);
1346 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1347 struct shmem_inode_info
*info
, pgoff_t index
)
1349 struct vm_area_struct pvma
;
1352 shmem_pseudo_vma_init(&pvma
, info
, index
);
1353 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1354 shmem_pseudo_vma_destroy(&pvma
);
1359 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1360 struct shmem_inode_info
*info
, pgoff_t index
)
1362 struct vm_area_struct pvma
;
1363 struct inode
*inode
= &info
->vfs_inode
;
1364 struct address_space
*mapping
= inode
->i_mapping
;
1365 pgoff_t idx
, hindex
= round_down(index
, HPAGE_PMD_NR
);
1366 void __rcu
**results
;
1369 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1373 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1374 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1380 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1381 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1382 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1383 shmem_pseudo_vma_destroy(&pvma
);
1385 prep_transhuge_page(page
);
1389 static struct page
*shmem_alloc_page(gfp_t gfp
,
1390 struct shmem_inode_info
*info
, pgoff_t index
)
1392 struct vm_area_struct pvma
;
1395 shmem_pseudo_vma_init(&pvma
, info
, index
);
1396 page
= alloc_page_vma(gfp
, &pvma
, 0);
1397 shmem_pseudo_vma_destroy(&pvma
);
1402 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1403 struct shmem_inode_info
*info
, struct shmem_sb_info
*sbinfo
,
1404 pgoff_t index
, bool huge
)
1410 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1412 nr
= huge
? HPAGE_PMD_NR
: 1;
1414 if (shmem_acct_block(info
->flags
, nr
))
1416 if (sbinfo
->max_blocks
) {
1417 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1418 sbinfo
->max_blocks
- nr
) > 0)
1420 percpu_counter_add(&sbinfo
->used_blocks
, nr
);
1424 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1426 page
= shmem_alloc_page(gfp
, info
, index
);
1428 __SetPageLocked(page
);
1429 __SetPageSwapBacked(page
);
1434 if (sbinfo
->max_blocks
)
1435 percpu_counter_add(&sbinfo
->used_blocks
, -nr
);
1437 shmem_unacct_blocks(info
->flags
, nr
);
1439 return ERR_PTR(err
);
1443 * When a page is moved from swapcache to shmem filecache (either by the
1444 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1445 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1446 * ignorance of the mapping it belongs to. If that mapping has special
1447 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1448 * we may need to copy to a suitable page before moving to filecache.
1450 * In a future release, this may well be extended to respect cpuset and
1451 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1452 * but for now it is a simple matter of zone.
1454 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1456 return page_zonenum(page
) > gfp_zone(gfp
);
1459 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1460 struct shmem_inode_info
*info
, pgoff_t index
)
1462 struct page
*oldpage
, *newpage
;
1463 struct address_space
*swap_mapping
;
1468 swap_index
= page_private(oldpage
);
1469 swap_mapping
= page_mapping(oldpage
);
1472 * We have arrived here because our zones are constrained, so don't
1473 * limit chance of success by further cpuset and node constraints.
1475 gfp
&= ~GFP_CONSTRAINT_MASK
;
1476 newpage
= shmem_alloc_page(gfp
, info
, index
);
1481 copy_highpage(newpage
, oldpage
);
1482 flush_dcache_page(newpage
);
1484 SetPageUptodate(newpage
);
1485 set_page_private(newpage
, swap_index
);
1486 SetPageSwapCache(newpage
);
1489 * Our caller will very soon move newpage out of swapcache, but it's
1490 * a nice clean interface for us to replace oldpage by newpage there.
1492 spin_lock_irq(&swap_mapping
->tree_lock
);
1493 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1496 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1497 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1499 spin_unlock_irq(&swap_mapping
->tree_lock
);
1501 if (unlikely(error
)) {
1503 * Is this possible? I think not, now that our callers check
1504 * both PageSwapCache and page_private after getting page lock;
1505 * but be defensive. Reverse old to newpage for clear and free.
1509 mem_cgroup_migrate(oldpage
, newpage
);
1510 lru_cache_add_anon(newpage
);
1514 ClearPageSwapCache(oldpage
);
1515 set_page_private(oldpage
, 0);
1517 unlock_page(oldpage
);
1524 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1526 * If we allocate a new one we do not mark it dirty. That's up to the
1527 * vm. If we swap it in we mark it dirty since we also free the swap
1528 * entry since a page cannot live in both the swap and page cache.
1530 * fault_mm and fault_type are only supplied by shmem_fault:
1531 * otherwise they are NULL.
1533 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1534 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1535 struct mm_struct
*fault_mm
, int *fault_type
)
1537 struct address_space
*mapping
= inode
->i_mapping
;
1538 struct shmem_inode_info
*info
;
1539 struct shmem_sb_info
*sbinfo
;
1540 struct mm_struct
*charge_mm
;
1541 struct mem_cgroup
*memcg
;
1544 enum sgp_type sgp_huge
= sgp
;
1545 pgoff_t hindex
= index
;
1550 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1552 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1556 page
= find_lock_entry(mapping
, index
);
1557 if (radix_tree_exceptional_entry(page
)) {
1558 swap
= radix_to_swp_entry(page
);
1562 if (sgp
<= SGP_CACHE
&&
1563 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1568 if (page
&& sgp
== SGP_WRITE
)
1569 mark_page_accessed(page
);
1571 /* fallocated page? */
1572 if (page
&& !PageUptodate(page
)) {
1573 if (sgp
!= SGP_READ
)
1579 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1585 * Fast cache lookup did not find it:
1586 * bring it back from swap or allocate.
1588 info
= SHMEM_I(inode
);
1589 sbinfo
= SHMEM_SB(inode
->i_sb
);
1590 charge_mm
= fault_mm
? : current
->mm
;
1593 /* Look it up and read it in.. */
1594 page
= lookup_swap_cache(swap
);
1596 /* Or update major stats only when swapin succeeds?? */
1598 *fault_type
|= VM_FAULT_MAJOR
;
1599 count_vm_event(PGMAJFAULT
);
1600 mem_cgroup_count_vm_event(fault_mm
, PGMAJFAULT
);
1602 /* Here we actually start the io */
1603 page
= shmem_swapin(swap
, gfp
, info
, index
);
1610 /* We have to do this with page locked to prevent races */
1612 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1613 !shmem_confirm_swap(mapping
, index
, swap
)) {
1614 error
= -EEXIST
; /* try again */
1617 if (!PageUptodate(page
)) {
1621 wait_on_page_writeback(page
);
1623 if (shmem_should_replace_page(page
, gfp
)) {
1624 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1629 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1632 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1633 swp_to_radix_entry(swap
));
1635 * We already confirmed swap under page lock, and make
1636 * no memory allocation here, so usually no possibility
1637 * of error; but free_swap_and_cache() only trylocks a
1638 * page, so it is just possible that the entry has been
1639 * truncated or holepunched since swap was confirmed.
1640 * shmem_undo_range() will have done some of the
1641 * unaccounting, now delete_from_swap_cache() will do
1643 * Reset swap.val? No, leave it so "failed" goes back to
1644 * "repeat": reading a hole and writing should succeed.
1647 mem_cgroup_cancel_charge(page
, memcg
, false);
1648 delete_from_swap_cache(page
);
1654 mem_cgroup_commit_charge(page
, memcg
, true, false);
1656 spin_lock_irq(&info
->lock
);
1658 shmem_recalc_inode(inode
);
1659 spin_unlock_irq(&info
->lock
);
1661 if (sgp
== SGP_WRITE
)
1662 mark_page_accessed(page
);
1664 delete_from_swap_cache(page
);
1665 set_page_dirty(page
);
1669 /* shmem_symlink() */
1670 if (mapping
->a_ops
!= &shmem_aops
)
1672 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1674 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1676 switch (sbinfo
->huge
) {
1679 case SHMEM_HUGE_NEVER
:
1681 case SHMEM_HUGE_WITHIN_SIZE
:
1682 off
= round_up(index
, HPAGE_PMD_NR
);
1683 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1684 if (i_size
>= HPAGE_PMD_SIZE
&&
1685 i_size
>> PAGE_SHIFT
>= off
)
1688 case SHMEM_HUGE_ADVISE
:
1689 if (sgp_huge
== SGP_HUGE
)
1691 /* TODO: implement fadvise() hints */
1696 page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1699 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1704 error
= PTR_ERR(page
);
1706 if (error
!= -ENOSPC
)
1709 * Try to reclaim some spece by splitting a huge page
1710 * beyond i_size on the filesystem.
1714 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1715 if (ret
== SHRINK_STOP
)
1723 if (PageTransHuge(page
))
1724 hindex
= round_down(index
, HPAGE_PMD_NR
);
1728 if (sgp
== SGP_WRITE
)
1729 __SetPageReferenced(page
);
1731 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1732 PageTransHuge(page
));
1735 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1736 compound_order(page
));
1738 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1740 radix_tree_preload_end();
1743 mem_cgroup_cancel_charge(page
, memcg
,
1744 PageTransHuge(page
));
1747 mem_cgroup_commit_charge(page
, memcg
, false,
1748 PageTransHuge(page
));
1749 lru_cache_add_anon(page
);
1751 spin_lock_irq(&info
->lock
);
1752 info
->alloced
+= 1 << compound_order(page
);
1753 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1754 shmem_recalc_inode(inode
);
1755 spin_unlock_irq(&info
->lock
);
1758 if (PageTransHuge(page
) &&
1759 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1760 hindex
+ HPAGE_PMD_NR
- 1) {
1762 * Part of the huge page is beyond i_size: subject
1763 * to shrink under memory pressure.
1765 spin_lock(&sbinfo
->shrinklist_lock
);
1766 if (list_empty(&info
->shrinklist
)) {
1767 list_add_tail(&info
->shrinklist
,
1768 &sbinfo
->shrinklist
);
1769 sbinfo
->shrinklist_len
++;
1771 spin_unlock(&sbinfo
->shrinklist_lock
);
1775 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1777 if (sgp
== SGP_FALLOC
)
1781 * Let SGP_WRITE caller clear ends if write does not fill page;
1782 * but SGP_FALLOC on a page fallocated earlier must initialize
1783 * it now, lest undo on failure cancel our earlier guarantee.
1785 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1786 struct page
*head
= compound_head(page
);
1789 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1790 clear_highpage(head
+ i
);
1791 flush_dcache_page(head
+ i
);
1793 SetPageUptodate(head
);
1797 /* Perhaps the file has been truncated since we checked */
1798 if (sgp
<= SGP_CACHE
&&
1799 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1801 ClearPageDirty(page
);
1802 delete_from_page_cache(page
);
1803 spin_lock_irq(&info
->lock
);
1804 shmem_recalc_inode(inode
);
1805 spin_unlock_irq(&info
->lock
);
1810 *pagep
= page
+ index
- hindex
;
1817 if (sbinfo
->max_blocks
)
1818 percpu_counter_sub(&sbinfo
->used_blocks
,
1819 1 << compound_order(page
));
1820 shmem_unacct_blocks(info
->flags
, 1 << compound_order(page
));
1822 if (PageTransHuge(page
)) {
1828 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1835 if (error
== -ENOSPC
&& !once
++) {
1836 info
= SHMEM_I(inode
);
1837 spin_lock_irq(&info
->lock
);
1838 shmem_recalc_inode(inode
);
1839 spin_unlock_irq(&info
->lock
);
1842 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1847 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1849 struct inode
*inode
= file_inode(vma
->vm_file
);
1850 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1853 int ret
= VM_FAULT_LOCKED
;
1856 * Trinity finds that probing a hole which tmpfs is punching can
1857 * prevent the hole-punch from ever completing: which in turn
1858 * locks writers out with its hold on i_mutex. So refrain from
1859 * faulting pages into the hole while it's being punched. Although
1860 * shmem_undo_range() does remove the additions, it may be unable to
1861 * keep up, as each new page needs its own unmap_mapping_range() call,
1862 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1864 * It does not matter if we sometimes reach this check just before the
1865 * hole-punch begins, so that one fault then races with the punch:
1866 * we just need to make racing faults a rare case.
1868 * The implementation below would be much simpler if we just used a
1869 * standard mutex or completion: but we cannot take i_mutex in fault,
1870 * and bloating every shmem inode for this unlikely case would be sad.
1872 if (unlikely(inode
->i_private
)) {
1873 struct shmem_falloc
*shmem_falloc
;
1875 spin_lock(&inode
->i_lock
);
1876 shmem_falloc
= inode
->i_private
;
1878 shmem_falloc
->waitq
&&
1879 vmf
->pgoff
>= shmem_falloc
->start
&&
1880 vmf
->pgoff
< shmem_falloc
->next
) {
1881 wait_queue_head_t
*shmem_falloc_waitq
;
1882 DEFINE_WAIT(shmem_fault_wait
);
1884 ret
= VM_FAULT_NOPAGE
;
1885 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1886 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1887 /* It's polite to up mmap_sem if we can */
1888 up_read(&vma
->vm_mm
->mmap_sem
);
1889 ret
= VM_FAULT_RETRY
;
1892 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1893 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1894 TASK_UNINTERRUPTIBLE
);
1895 spin_unlock(&inode
->i_lock
);
1899 * shmem_falloc_waitq points into the shmem_fallocate()
1900 * stack of the hole-punching task: shmem_falloc_waitq
1901 * is usually invalid by the time we reach here, but
1902 * finish_wait() does not dereference it in that case;
1903 * though i_lock needed lest racing with wake_up_all().
1905 spin_lock(&inode
->i_lock
);
1906 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1907 spin_unlock(&inode
->i_lock
);
1910 spin_unlock(&inode
->i_lock
);
1914 if (vma
->vm_flags
& VM_HUGEPAGE
)
1916 else if (vma
->vm_flags
& VM_NOHUGEPAGE
)
1919 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
1920 gfp
, vma
->vm_mm
, &ret
);
1922 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1926 unsigned long shmem_get_unmapped_area(struct file
*file
,
1927 unsigned long uaddr
, unsigned long len
,
1928 unsigned long pgoff
, unsigned long flags
)
1930 unsigned long (*get_area
)(struct file
*,
1931 unsigned long, unsigned long, unsigned long, unsigned long);
1933 unsigned long offset
;
1934 unsigned long inflated_len
;
1935 unsigned long inflated_addr
;
1936 unsigned long inflated_offset
;
1938 if (len
> TASK_SIZE
)
1941 get_area
= current
->mm
->get_unmapped_area
;
1942 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
1944 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1946 if (IS_ERR_VALUE(addr
))
1948 if (addr
& ~PAGE_MASK
)
1950 if (addr
> TASK_SIZE
- len
)
1953 if (shmem_huge
== SHMEM_HUGE_DENY
)
1955 if (len
< HPAGE_PMD_SIZE
)
1957 if (flags
& MAP_FIXED
)
1960 * Our priority is to support MAP_SHARED mapped hugely;
1961 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1962 * But if caller specified an address hint, respect that as before.
1967 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
1968 struct super_block
*sb
;
1971 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
1972 sb
= file_inode(file
)->i_sb
;
1975 * Called directly from mm/mmap.c, or drivers/char/mem.c
1976 * for "/dev/zero", to create a shared anonymous object.
1978 if (IS_ERR(shm_mnt
))
1980 sb
= shm_mnt
->mnt_sb
;
1982 if (SHMEM_SB(sb
)->huge
!= SHMEM_HUGE_NEVER
)
1986 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
1987 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
1989 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
1992 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
1993 if (inflated_len
> TASK_SIZE
)
1995 if (inflated_len
< len
)
1998 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
1999 if (IS_ERR_VALUE(inflated_addr
))
2001 if (inflated_addr
& ~PAGE_MASK
)
2004 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2005 inflated_addr
+= offset
- inflated_offset
;
2006 if (inflated_offset
> offset
)
2007 inflated_addr
+= HPAGE_PMD_SIZE
;
2009 if (inflated_addr
> TASK_SIZE
- len
)
2011 return inflated_addr
;
2015 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2017 struct inode
*inode
= file_inode(vma
->vm_file
);
2018 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2021 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2024 struct inode
*inode
= file_inode(vma
->vm_file
);
2027 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2028 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2032 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2034 struct inode
*inode
= file_inode(file
);
2035 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2036 int retval
= -ENOMEM
;
2038 spin_lock_irq(&info
->lock
);
2039 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2040 if (!user_shm_lock(inode
->i_size
, user
))
2042 info
->flags
|= VM_LOCKED
;
2043 mapping_set_unevictable(file
->f_mapping
);
2045 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2046 user_shm_unlock(inode
->i_size
, user
);
2047 info
->flags
&= ~VM_LOCKED
;
2048 mapping_clear_unevictable(file
->f_mapping
);
2053 spin_unlock_irq(&info
->lock
);
2057 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2059 file_accessed(file
);
2060 vma
->vm_ops
= &shmem_vm_ops
;
2061 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2062 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2063 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2064 khugepaged_enter(vma
, vma
->vm_flags
);
2069 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2070 umode_t mode
, dev_t dev
, unsigned long flags
)
2072 struct inode
*inode
;
2073 struct shmem_inode_info
*info
;
2074 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2076 if (shmem_reserve_inode(sb
))
2079 inode
= new_inode(sb
);
2081 inode
->i_ino
= get_next_ino();
2082 inode_init_owner(inode
, dir
, mode
);
2083 inode
->i_blocks
= 0;
2084 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2085 inode
->i_generation
= get_seconds();
2086 info
= SHMEM_I(inode
);
2087 memset(info
, 0, (char *)inode
- (char *)info
);
2088 spin_lock_init(&info
->lock
);
2089 info
->seals
= F_SEAL_SEAL
;
2090 info
->flags
= flags
& VM_NORESERVE
;
2091 INIT_LIST_HEAD(&info
->shrinklist
);
2092 INIT_LIST_HEAD(&info
->swaplist
);
2093 simple_xattrs_init(&info
->xattrs
);
2094 cache_no_acl(inode
);
2096 switch (mode
& S_IFMT
) {
2098 inode
->i_op
= &shmem_special_inode_operations
;
2099 init_special_inode(inode
, mode
, dev
);
2102 inode
->i_mapping
->a_ops
= &shmem_aops
;
2103 inode
->i_op
= &shmem_inode_operations
;
2104 inode
->i_fop
= &shmem_file_operations
;
2105 mpol_shared_policy_init(&info
->policy
,
2106 shmem_get_sbmpol(sbinfo
));
2110 /* Some things misbehave if size == 0 on a directory */
2111 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2112 inode
->i_op
= &shmem_dir_inode_operations
;
2113 inode
->i_fop
= &simple_dir_operations
;
2117 * Must not load anything in the rbtree,
2118 * mpol_free_shared_policy will not be called.
2120 mpol_shared_policy_init(&info
->policy
, NULL
);
2124 shmem_free_inode(sb
);
2128 bool shmem_mapping(struct address_space
*mapping
)
2133 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
2137 static const struct inode_operations shmem_symlink_inode_operations
;
2138 static const struct inode_operations shmem_short_symlink_operations
;
2140 #ifdef CONFIG_TMPFS_XATTR
2141 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2143 #define shmem_initxattrs NULL
2147 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2148 loff_t pos
, unsigned len
, unsigned flags
,
2149 struct page
**pagep
, void **fsdata
)
2151 struct inode
*inode
= mapping
->host
;
2152 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2153 pgoff_t index
= pos
>> PAGE_SHIFT
;
2155 /* i_mutex is held by caller */
2156 if (unlikely(info
->seals
)) {
2157 if (info
->seals
& F_SEAL_WRITE
)
2159 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2163 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2167 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2168 loff_t pos
, unsigned len
, unsigned copied
,
2169 struct page
*page
, void *fsdata
)
2171 struct inode
*inode
= mapping
->host
;
2173 if (pos
+ copied
> inode
->i_size
)
2174 i_size_write(inode
, pos
+ copied
);
2176 if (!PageUptodate(page
)) {
2177 struct page
*head
= compound_head(page
);
2178 if (PageTransCompound(page
)) {
2181 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2182 if (head
+ i
== page
)
2184 clear_highpage(head
+ i
);
2185 flush_dcache_page(head
+ i
);
2188 if (copied
< PAGE_SIZE
) {
2189 unsigned from
= pos
& (PAGE_SIZE
- 1);
2190 zero_user_segments(page
, 0, from
,
2191 from
+ copied
, PAGE_SIZE
);
2193 SetPageUptodate(head
);
2195 set_page_dirty(page
);
2202 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2204 struct file
*file
= iocb
->ki_filp
;
2205 struct inode
*inode
= file_inode(file
);
2206 struct address_space
*mapping
= inode
->i_mapping
;
2208 unsigned long offset
;
2209 enum sgp_type sgp
= SGP_READ
;
2212 loff_t
*ppos
= &iocb
->ki_pos
;
2215 * Might this read be for a stacking filesystem? Then when reading
2216 * holes of a sparse file, we actually need to allocate those pages,
2217 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2219 if (!iter_is_iovec(to
))
2222 index
= *ppos
>> PAGE_SHIFT
;
2223 offset
= *ppos
& ~PAGE_MASK
;
2226 struct page
*page
= NULL
;
2228 unsigned long nr
, ret
;
2229 loff_t i_size
= i_size_read(inode
);
2231 end_index
= i_size
>> PAGE_SHIFT
;
2232 if (index
> end_index
)
2234 if (index
== end_index
) {
2235 nr
= i_size
& ~PAGE_MASK
;
2240 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2242 if (error
== -EINVAL
)
2247 if (sgp
== SGP_CACHE
)
2248 set_page_dirty(page
);
2253 * We must evaluate after, since reads (unlike writes)
2254 * are called without i_mutex protection against truncate
2257 i_size
= i_size_read(inode
);
2258 end_index
= i_size
>> PAGE_SHIFT
;
2259 if (index
== end_index
) {
2260 nr
= i_size
& ~PAGE_MASK
;
2271 * If users can be writing to this page using arbitrary
2272 * virtual addresses, take care about potential aliasing
2273 * before reading the page on the kernel side.
2275 if (mapping_writably_mapped(mapping
))
2276 flush_dcache_page(page
);
2278 * Mark the page accessed if we read the beginning.
2281 mark_page_accessed(page
);
2283 page
= ZERO_PAGE(0);
2288 * Ok, we have the page, and it's up-to-date, so
2289 * now we can copy it to user space...
2291 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2294 index
+= offset
>> PAGE_SHIFT
;
2295 offset
&= ~PAGE_MASK
;
2298 if (!iov_iter_count(to
))
2307 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2308 file_accessed(file
);
2309 return retval
? retval
: error
;
2312 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
2313 struct pipe_inode_info
*pipe
, size_t len
,
2316 struct address_space
*mapping
= in
->f_mapping
;
2317 struct inode
*inode
= mapping
->host
;
2318 unsigned int loff
, nr_pages
, req_pages
;
2319 struct page
*pages
[PIPE_DEF_BUFFERS
];
2320 struct partial_page partial
[PIPE_DEF_BUFFERS
];
2322 pgoff_t index
, end_index
;
2325 struct splice_pipe_desc spd
= {
2328 .nr_pages_max
= PIPE_DEF_BUFFERS
,
2330 .ops
= &page_cache_pipe_buf_ops
,
2331 .spd_release
= spd_release_page
,
2334 isize
= i_size_read(inode
);
2335 if (unlikely(*ppos
>= isize
))
2338 left
= isize
- *ppos
;
2339 if (unlikely(left
< len
))
2342 if (splice_grow_spd(pipe
, &spd
))
2345 index
= *ppos
>> PAGE_SHIFT
;
2346 loff
= *ppos
& ~PAGE_MASK
;
2347 req_pages
= (len
+ loff
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2348 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
2350 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
2351 nr_pages
, spd
.pages
);
2352 index
+= spd
.nr_pages
;
2355 while (spd
.nr_pages
< nr_pages
) {
2356 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
);
2360 spd
.pages
[spd
.nr_pages
++] = page
;
2364 index
= *ppos
>> PAGE_SHIFT
;
2365 nr_pages
= spd
.nr_pages
;
2368 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
2369 unsigned int this_len
;
2374 this_len
= min_t(unsigned long, len
, PAGE_SIZE
- loff
);
2375 page
= spd
.pages
[page_nr
];
2377 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
2378 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
);
2382 put_page(spd
.pages
[page_nr
]);
2383 spd
.pages
[page_nr
] = page
;
2386 isize
= i_size_read(inode
);
2387 end_index
= (isize
- 1) >> PAGE_SHIFT
;
2388 if (unlikely(!isize
|| index
> end_index
))
2391 if (end_index
== index
) {
2394 plen
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2398 this_len
= min(this_len
, plen
- loff
);
2402 spd
.partial
[page_nr
].offset
= loff
;
2403 spd
.partial
[page_nr
].len
= this_len
;
2410 while (page_nr
< nr_pages
)
2411 put_page(spd
.pages
[page_nr
++]);
2414 error
= splice_to_pipe(pipe
, &spd
);
2416 splice_shrink_spd(&spd
);
2426 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2428 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2429 pgoff_t index
, pgoff_t end
, int whence
)
2432 struct pagevec pvec
;
2433 pgoff_t indices
[PAGEVEC_SIZE
];
2437 pagevec_init(&pvec
, 0);
2438 pvec
.nr
= 1; /* start small: we may be there already */
2440 pvec
.nr
= find_get_entries(mapping
, index
,
2441 pvec
.nr
, pvec
.pages
, indices
);
2443 if (whence
== SEEK_DATA
)
2447 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2448 if (index
< indices
[i
]) {
2449 if (whence
== SEEK_HOLE
) {
2455 page
= pvec
.pages
[i
];
2456 if (page
&& !radix_tree_exceptional_entry(page
)) {
2457 if (!PageUptodate(page
))
2461 (page
&& whence
== SEEK_DATA
) ||
2462 (!page
&& whence
== SEEK_HOLE
)) {
2467 pagevec_remove_exceptionals(&pvec
);
2468 pagevec_release(&pvec
);
2469 pvec
.nr
= PAGEVEC_SIZE
;
2475 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2477 struct address_space
*mapping
= file
->f_mapping
;
2478 struct inode
*inode
= mapping
->host
;
2482 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2483 return generic_file_llseek_size(file
, offset
, whence
,
2484 MAX_LFS_FILESIZE
, i_size_read(inode
));
2486 /* We're holding i_mutex so we can access i_size directly */
2490 else if (offset
>= inode
->i_size
)
2493 start
= offset
>> PAGE_SHIFT
;
2494 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2495 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2496 new_offset
<<= PAGE_SHIFT
;
2497 if (new_offset
> offset
) {
2498 if (new_offset
< inode
->i_size
)
2499 offset
= new_offset
;
2500 else if (whence
== SEEK_DATA
)
2503 offset
= inode
->i_size
;
2508 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2509 inode_unlock(inode
);
2514 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2515 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2517 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2518 #define LAST_SCAN 4 /* about 150ms max */
2520 static void shmem_tag_pins(struct address_space
*mapping
)
2522 struct radix_tree_iter iter
;
2531 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2532 page
= radix_tree_deref_slot(slot
);
2533 if (!page
|| radix_tree_exception(page
)) {
2534 if (radix_tree_deref_retry(page
)) {
2535 slot
= radix_tree_iter_retry(&iter
);
2538 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2539 spin_lock_irq(&mapping
->tree_lock
);
2540 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2542 spin_unlock_irq(&mapping
->tree_lock
);
2545 if (need_resched()) {
2547 slot
= radix_tree_iter_next(&iter
);
2554 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2555 * via get_user_pages(), drivers might have some pending I/O without any active
2556 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2557 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2558 * them to be dropped.
2559 * The caller must guarantee that no new user will acquire writable references
2560 * to those pages to avoid races.
2562 static int shmem_wait_for_pins(struct address_space
*mapping
)
2564 struct radix_tree_iter iter
;
2570 shmem_tag_pins(mapping
);
2573 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2574 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2578 lru_add_drain_all();
2579 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2584 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2585 start
, SHMEM_TAG_PINNED
) {
2587 page
= radix_tree_deref_slot(slot
);
2588 if (radix_tree_exception(page
)) {
2589 if (radix_tree_deref_retry(page
)) {
2590 slot
= radix_tree_iter_retry(&iter
);
2598 page_count(page
) - page_mapcount(page
) != 1) {
2599 if (scan
< LAST_SCAN
)
2600 goto continue_resched
;
2603 * On the last scan, we clean up all those tags
2604 * we inserted; but make a note that we still
2605 * found pages pinned.
2610 spin_lock_irq(&mapping
->tree_lock
);
2611 radix_tree_tag_clear(&mapping
->page_tree
,
2612 iter
.index
, SHMEM_TAG_PINNED
);
2613 spin_unlock_irq(&mapping
->tree_lock
);
2615 if (need_resched()) {
2617 slot
= radix_tree_iter_next(&iter
);
2626 #define F_ALL_SEALS (F_SEAL_SEAL | \
2631 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2633 struct inode
*inode
= file_inode(file
);
2634 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2639 * Sealing allows multiple parties to share a shmem-file but restrict
2640 * access to a specific subset of file operations. Seals can only be
2641 * added, but never removed. This way, mutually untrusted parties can
2642 * share common memory regions with a well-defined policy. A malicious
2643 * peer can thus never perform unwanted operations on a shared object.
2645 * Seals are only supported on special shmem-files and always affect
2646 * the whole underlying inode. Once a seal is set, it may prevent some
2647 * kinds of access to the file. Currently, the following seals are
2649 * SEAL_SEAL: Prevent further seals from being set on this file
2650 * SEAL_SHRINK: Prevent the file from shrinking
2651 * SEAL_GROW: Prevent the file from growing
2652 * SEAL_WRITE: Prevent write access to the file
2654 * As we don't require any trust relationship between two parties, we
2655 * must prevent seals from being removed. Therefore, sealing a file
2656 * only adds a given set of seals to the file, it never touches
2657 * existing seals. Furthermore, the "setting seals"-operation can be
2658 * sealed itself, which basically prevents any further seal from being
2661 * Semantics of sealing are only defined on volatile files. Only
2662 * anonymous shmem files support sealing. More importantly, seals are
2663 * never written to disk. Therefore, there's no plan to support it on
2667 if (file
->f_op
!= &shmem_file_operations
)
2669 if (!(file
->f_mode
& FMODE_WRITE
))
2671 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2676 if (info
->seals
& F_SEAL_SEAL
) {
2681 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2682 error
= mapping_deny_writable(file
->f_mapping
);
2686 error
= shmem_wait_for_pins(file
->f_mapping
);
2688 mapping_allow_writable(file
->f_mapping
);
2693 info
->seals
|= seals
;
2697 inode_unlock(inode
);
2700 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2702 int shmem_get_seals(struct file
*file
)
2704 if (file
->f_op
!= &shmem_file_operations
)
2707 return SHMEM_I(file_inode(file
))->seals
;
2709 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2711 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2717 /* disallow upper 32bit */
2721 error
= shmem_add_seals(file
, arg
);
2724 error
= shmem_get_seals(file
);
2734 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2737 struct inode
*inode
= file_inode(file
);
2738 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2739 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2740 struct shmem_falloc shmem_falloc
;
2741 pgoff_t start
, index
, end
;
2744 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2749 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2750 struct address_space
*mapping
= file
->f_mapping
;
2751 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2752 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2753 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2755 /* protected by i_mutex */
2756 if (info
->seals
& F_SEAL_WRITE
) {
2761 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2762 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2763 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2764 spin_lock(&inode
->i_lock
);
2765 inode
->i_private
= &shmem_falloc
;
2766 spin_unlock(&inode
->i_lock
);
2768 if ((u64
)unmap_end
> (u64
)unmap_start
)
2769 unmap_mapping_range(mapping
, unmap_start
,
2770 1 + unmap_end
- unmap_start
, 0);
2771 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2772 /* No need to unmap again: hole-punching leaves COWed pages */
2774 spin_lock(&inode
->i_lock
);
2775 inode
->i_private
= NULL
;
2776 wake_up_all(&shmem_falloc_waitq
);
2777 spin_unlock(&inode
->i_lock
);
2782 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2783 error
= inode_newsize_ok(inode
, offset
+ len
);
2787 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2792 start
= offset
>> PAGE_SHIFT
;
2793 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2794 /* Try to avoid a swapstorm if len is impossible to satisfy */
2795 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2800 shmem_falloc
.waitq
= NULL
;
2801 shmem_falloc
.start
= start
;
2802 shmem_falloc
.next
= start
;
2803 shmem_falloc
.nr_falloced
= 0;
2804 shmem_falloc
.nr_unswapped
= 0;
2805 spin_lock(&inode
->i_lock
);
2806 inode
->i_private
= &shmem_falloc
;
2807 spin_unlock(&inode
->i_lock
);
2809 for (index
= start
; index
< end
; index
++) {
2813 * Good, the fallocate(2) manpage permits EINTR: we may have
2814 * been interrupted because we are using up too much memory.
2816 if (signal_pending(current
))
2818 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2821 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2823 /* Remove the !PageUptodate pages we added */
2824 if (index
> start
) {
2825 shmem_undo_range(inode
,
2826 (loff_t
)start
<< PAGE_SHIFT
,
2827 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2833 * Inform shmem_writepage() how far we have reached.
2834 * No need for lock or barrier: we have the page lock.
2836 shmem_falloc
.next
++;
2837 if (!PageUptodate(page
))
2838 shmem_falloc
.nr_falloced
++;
2841 * If !PageUptodate, leave it that way so that freeable pages
2842 * can be recognized if we need to rollback on error later.
2843 * But set_page_dirty so that memory pressure will swap rather
2844 * than free the pages we are allocating (and SGP_CACHE pages
2845 * might still be clean: we now need to mark those dirty too).
2847 set_page_dirty(page
);
2853 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2854 i_size_write(inode
, offset
+ len
);
2855 inode
->i_ctime
= CURRENT_TIME
;
2857 spin_lock(&inode
->i_lock
);
2858 inode
->i_private
= NULL
;
2859 spin_unlock(&inode
->i_lock
);
2861 inode_unlock(inode
);
2865 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2867 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2869 buf
->f_type
= TMPFS_MAGIC
;
2870 buf
->f_bsize
= PAGE_SIZE
;
2871 buf
->f_namelen
= NAME_MAX
;
2872 if (sbinfo
->max_blocks
) {
2873 buf
->f_blocks
= sbinfo
->max_blocks
;
2875 buf
->f_bfree
= sbinfo
->max_blocks
-
2876 percpu_counter_sum(&sbinfo
->used_blocks
);
2878 if (sbinfo
->max_inodes
) {
2879 buf
->f_files
= sbinfo
->max_inodes
;
2880 buf
->f_ffree
= sbinfo
->free_inodes
;
2882 /* else leave those fields 0 like simple_statfs */
2887 * File creation. Allocate an inode, and we're done..
2890 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2892 struct inode
*inode
;
2893 int error
= -ENOSPC
;
2895 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2897 error
= simple_acl_create(dir
, inode
);
2900 error
= security_inode_init_security(inode
, dir
,
2902 shmem_initxattrs
, NULL
);
2903 if (error
&& error
!= -EOPNOTSUPP
)
2907 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2908 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2909 d_instantiate(dentry
, inode
);
2910 dget(dentry
); /* Extra count - pin the dentry in core */
2919 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2921 struct inode
*inode
;
2922 int error
= -ENOSPC
;
2924 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2926 error
= security_inode_init_security(inode
, dir
,
2928 shmem_initxattrs
, NULL
);
2929 if (error
&& error
!= -EOPNOTSUPP
)
2931 error
= simple_acl_create(dir
, inode
);
2934 d_tmpfile(dentry
, inode
);
2942 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2946 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2952 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2955 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2961 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2963 struct inode
*inode
= d_inode(old_dentry
);
2967 * No ordinary (disk based) filesystem counts links as inodes;
2968 * but each new link needs a new dentry, pinning lowmem, and
2969 * tmpfs dentries cannot be pruned until they are unlinked.
2971 ret
= shmem_reserve_inode(inode
->i_sb
);
2975 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2976 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2978 ihold(inode
); /* New dentry reference */
2979 dget(dentry
); /* Extra pinning count for the created dentry */
2980 d_instantiate(dentry
, inode
);
2985 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2987 struct inode
*inode
= d_inode(dentry
);
2989 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2990 shmem_free_inode(inode
->i_sb
);
2992 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2993 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2995 dput(dentry
); /* Undo the count from "create" - this does all the work */
2999 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3001 if (!simple_empty(dentry
))
3004 drop_nlink(d_inode(dentry
));
3006 return shmem_unlink(dir
, dentry
);
3009 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3011 bool old_is_dir
= d_is_dir(old_dentry
);
3012 bool new_is_dir
= d_is_dir(new_dentry
);
3014 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3016 drop_nlink(old_dir
);
3019 drop_nlink(new_dir
);
3023 old_dir
->i_ctime
= old_dir
->i_mtime
=
3024 new_dir
->i_ctime
= new_dir
->i_mtime
=
3025 d_inode(old_dentry
)->i_ctime
=
3026 d_inode(new_dentry
)->i_ctime
= CURRENT_TIME
;
3031 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3033 struct dentry
*whiteout
;
3036 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3040 error
= shmem_mknod(old_dir
, whiteout
,
3041 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3047 * Cheat and hash the whiteout while the old dentry is still in
3048 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3050 * d_lookup() will consistently find one of them at this point,
3051 * not sure which one, but that isn't even important.
3058 * The VFS layer already does all the dentry stuff for rename,
3059 * we just have to decrement the usage count for the target if
3060 * it exists so that the VFS layer correctly free's it when it
3063 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3065 struct inode
*inode
= d_inode(old_dentry
);
3066 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3068 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3071 if (flags
& RENAME_EXCHANGE
)
3072 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3074 if (!simple_empty(new_dentry
))
3077 if (flags
& RENAME_WHITEOUT
) {
3080 error
= shmem_whiteout(old_dir
, old_dentry
);
3085 if (d_really_is_positive(new_dentry
)) {
3086 (void) shmem_unlink(new_dir
, new_dentry
);
3087 if (they_are_dirs
) {
3088 drop_nlink(d_inode(new_dentry
));
3089 drop_nlink(old_dir
);
3091 } else if (they_are_dirs
) {
3092 drop_nlink(old_dir
);
3096 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3097 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3098 old_dir
->i_ctime
= old_dir
->i_mtime
=
3099 new_dir
->i_ctime
= new_dir
->i_mtime
=
3100 inode
->i_ctime
= CURRENT_TIME
;
3104 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3108 struct inode
*inode
;
3110 struct shmem_inode_info
*info
;
3112 len
= strlen(symname
) + 1;
3113 if (len
> PAGE_SIZE
)
3114 return -ENAMETOOLONG
;
3116 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
3120 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3121 shmem_initxattrs
, NULL
);
3123 if (error
!= -EOPNOTSUPP
) {
3130 info
= SHMEM_I(inode
);
3131 inode
->i_size
= len
-1;
3132 if (len
<= SHORT_SYMLINK_LEN
) {
3133 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3134 if (!inode
->i_link
) {
3138 inode
->i_op
= &shmem_short_symlink_operations
;
3140 inode_nohighmem(inode
);
3141 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3146 inode
->i_mapping
->a_ops
= &shmem_aops
;
3147 inode
->i_op
= &shmem_symlink_inode_operations
;
3148 memcpy(page_address(page
), symname
, len
);
3149 SetPageUptodate(page
);
3150 set_page_dirty(page
);
3154 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3155 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
3156 d_instantiate(dentry
, inode
);
3161 static void shmem_put_link(void *arg
)
3163 mark_page_accessed(arg
);
3167 static const char *shmem_get_link(struct dentry
*dentry
,
3168 struct inode
*inode
,
3169 struct delayed_call
*done
)
3171 struct page
*page
= NULL
;
3174 page
= find_get_page(inode
->i_mapping
, 0);
3176 return ERR_PTR(-ECHILD
);
3177 if (!PageUptodate(page
)) {
3179 return ERR_PTR(-ECHILD
);
3182 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3184 return ERR_PTR(error
);
3187 set_delayed_call(done
, shmem_put_link
, page
);
3188 return page_address(page
);
3191 #ifdef CONFIG_TMPFS_XATTR
3193 * Superblocks without xattr inode operations may get some security.* xattr
3194 * support from the LSM "for free". As soon as we have any other xattrs
3195 * like ACLs, we also need to implement the security.* handlers at
3196 * filesystem level, though.
3200 * Callback for security_inode_init_security() for acquiring xattrs.
3202 static int shmem_initxattrs(struct inode
*inode
,
3203 const struct xattr
*xattr_array
,
3206 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3207 const struct xattr
*xattr
;
3208 struct simple_xattr
*new_xattr
;
3211 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3212 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3216 len
= strlen(xattr
->name
) + 1;
3217 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3219 if (!new_xattr
->name
) {
3224 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3225 XATTR_SECURITY_PREFIX_LEN
);
3226 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3229 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3235 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3236 struct dentry
*unused
, struct inode
*inode
,
3237 const char *name
, void *buffer
, size_t size
)
3239 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3241 name
= xattr_full_name(handler
, name
);
3242 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3245 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3246 struct dentry
*unused
, struct inode
*inode
,
3247 const char *name
, const void *value
,
3248 size_t size
, int flags
)
3250 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3252 name
= xattr_full_name(handler
, name
);
3253 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3256 static const struct xattr_handler shmem_security_xattr_handler
= {
3257 .prefix
= XATTR_SECURITY_PREFIX
,
3258 .get
= shmem_xattr_handler_get
,
3259 .set
= shmem_xattr_handler_set
,
3262 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3263 .prefix
= XATTR_TRUSTED_PREFIX
,
3264 .get
= shmem_xattr_handler_get
,
3265 .set
= shmem_xattr_handler_set
,
3268 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3269 #ifdef CONFIG_TMPFS_POSIX_ACL
3270 &posix_acl_access_xattr_handler
,
3271 &posix_acl_default_xattr_handler
,
3273 &shmem_security_xattr_handler
,
3274 &shmem_trusted_xattr_handler
,
3278 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3280 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3281 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3283 #endif /* CONFIG_TMPFS_XATTR */
3285 static const struct inode_operations shmem_short_symlink_operations
= {
3286 .readlink
= generic_readlink
,
3287 .get_link
= simple_get_link
,
3288 #ifdef CONFIG_TMPFS_XATTR
3289 .setxattr
= generic_setxattr
,
3290 .getxattr
= generic_getxattr
,
3291 .listxattr
= shmem_listxattr
,
3292 .removexattr
= generic_removexattr
,
3296 static const struct inode_operations shmem_symlink_inode_operations
= {
3297 .readlink
= generic_readlink
,
3298 .get_link
= shmem_get_link
,
3299 #ifdef CONFIG_TMPFS_XATTR
3300 .setxattr
= generic_setxattr
,
3301 .getxattr
= generic_getxattr
,
3302 .listxattr
= shmem_listxattr
,
3303 .removexattr
= generic_removexattr
,
3307 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3309 return ERR_PTR(-ESTALE
);
3312 static int shmem_match(struct inode
*ino
, void *vfh
)
3316 inum
= (inum
<< 32) | fh
[1];
3317 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3320 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3321 struct fid
*fid
, int fh_len
, int fh_type
)
3323 struct inode
*inode
;
3324 struct dentry
*dentry
= NULL
;
3331 inum
= (inum
<< 32) | fid
->raw
[1];
3333 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3334 shmem_match
, fid
->raw
);
3336 dentry
= d_find_alias(inode
);
3343 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3344 struct inode
*parent
)
3348 return FILEID_INVALID
;
3351 if (inode_unhashed(inode
)) {
3352 /* Unfortunately insert_inode_hash is not idempotent,
3353 * so as we hash inodes here rather than at creation
3354 * time, we need a lock to ensure we only try
3357 static DEFINE_SPINLOCK(lock
);
3359 if (inode_unhashed(inode
))
3360 __insert_inode_hash(inode
,
3361 inode
->i_ino
+ inode
->i_generation
);
3365 fh
[0] = inode
->i_generation
;
3366 fh
[1] = inode
->i_ino
;
3367 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3373 static const struct export_operations shmem_export_ops
= {
3374 .get_parent
= shmem_get_parent
,
3375 .encode_fh
= shmem_encode_fh
,
3376 .fh_to_dentry
= shmem_fh_to_dentry
,
3379 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3382 char *this_char
, *value
, *rest
;
3383 struct mempolicy
*mpol
= NULL
;
3387 while (options
!= NULL
) {
3388 this_char
= options
;
3391 * NUL-terminate this option: unfortunately,
3392 * mount options form a comma-separated list,
3393 * but mpol's nodelist may also contain commas.
3395 options
= strchr(options
, ',');
3396 if (options
== NULL
)
3399 if (!isdigit(*options
)) {
3406 if ((value
= strchr(this_char
,'=')) != NULL
) {
3409 pr_err("tmpfs: No value for mount option '%s'\n",
3414 if (!strcmp(this_char
,"size")) {
3415 unsigned long long size
;
3416 size
= memparse(value
,&rest
);
3418 size
<<= PAGE_SHIFT
;
3419 size
*= totalram_pages
;
3425 sbinfo
->max_blocks
=
3426 DIV_ROUND_UP(size
, PAGE_SIZE
);
3427 } else if (!strcmp(this_char
,"nr_blocks")) {
3428 sbinfo
->max_blocks
= memparse(value
, &rest
);
3431 } else if (!strcmp(this_char
,"nr_inodes")) {
3432 sbinfo
->max_inodes
= memparse(value
, &rest
);
3435 } else if (!strcmp(this_char
,"mode")) {
3438 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3441 } else if (!strcmp(this_char
,"uid")) {
3444 uid
= simple_strtoul(value
, &rest
, 0);
3447 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3448 if (!uid_valid(sbinfo
->uid
))
3450 } else if (!strcmp(this_char
,"gid")) {
3453 gid
= simple_strtoul(value
, &rest
, 0);
3456 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3457 if (!gid_valid(sbinfo
->gid
))
3459 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3460 } else if (!strcmp(this_char
, "huge")) {
3462 huge
= shmem_parse_huge(value
);
3465 if (!has_transparent_hugepage() &&
3466 huge
!= SHMEM_HUGE_NEVER
)
3468 sbinfo
->huge
= huge
;
3471 } else if (!strcmp(this_char
,"mpol")) {
3474 if (mpol_parse_str(value
, &mpol
))
3478 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3482 sbinfo
->mpol
= mpol
;
3486 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3494 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3496 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3497 struct shmem_sb_info config
= *sbinfo
;
3498 unsigned long inodes
;
3499 int error
= -EINVAL
;
3502 if (shmem_parse_options(data
, &config
, true))
3505 spin_lock(&sbinfo
->stat_lock
);
3506 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3507 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3509 if (config
.max_inodes
< inodes
)
3512 * Those tests disallow limited->unlimited while any are in use;
3513 * but we must separately disallow unlimited->limited, because
3514 * in that case we have no record of how much is already in use.
3516 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3518 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3522 sbinfo
->huge
= config
.huge
;
3523 sbinfo
->max_blocks
= config
.max_blocks
;
3524 sbinfo
->max_inodes
= config
.max_inodes
;
3525 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3528 * Preserve previous mempolicy unless mpol remount option was specified.
3531 mpol_put(sbinfo
->mpol
);
3532 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3535 spin_unlock(&sbinfo
->stat_lock
);
3539 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3541 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3543 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3544 seq_printf(seq
, ",size=%luk",
3545 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3546 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3547 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3548 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3549 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3550 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3551 seq_printf(seq
, ",uid=%u",
3552 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3553 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3554 seq_printf(seq
, ",gid=%u",
3555 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3556 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3557 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3559 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3561 shmem_show_mpol(seq
, sbinfo
->mpol
);
3565 #define MFD_NAME_PREFIX "memfd:"
3566 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3567 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3569 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3571 SYSCALL_DEFINE2(memfd_create
,
3572 const char __user
*, uname
,
3573 unsigned int, flags
)
3575 struct shmem_inode_info
*info
;
3581 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3584 /* length includes terminating zero */
3585 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3588 if (len
> MFD_NAME_MAX_LEN
+ 1)
3591 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
3595 strcpy(name
, MFD_NAME_PREFIX
);
3596 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3601 /* terminating-zero may have changed after strnlen_user() returned */
3602 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3607 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3613 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3615 error
= PTR_ERR(file
);
3618 info
= SHMEM_I(file_inode(file
));
3619 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3620 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3621 if (flags
& MFD_ALLOW_SEALING
)
3622 info
->seals
&= ~F_SEAL_SEAL
;
3624 fd_install(fd
, file
);
3635 #endif /* CONFIG_TMPFS */
3637 static void shmem_put_super(struct super_block
*sb
)
3639 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3641 percpu_counter_destroy(&sbinfo
->used_blocks
);
3642 mpol_put(sbinfo
->mpol
);
3644 sb
->s_fs_info
= NULL
;
3647 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3649 struct inode
*inode
;
3650 struct shmem_sb_info
*sbinfo
;
3653 /* Round up to L1_CACHE_BYTES to resist false sharing */
3654 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3655 L1_CACHE_BYTES
), GFP_KERNEL
);
3659 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3660 sbinfo
->uid
= current_fsuid();
3661 sbinfo
->gid
= current_fsgid();
3662 sb
->s_fs_info
= sbinfo
;
3666 * Per default we only allow half of the physical ram per
3667 * tmpfs instance, limiting inodes to one per page of lowmem;
3668 * but the internal instance is left unlimited.
3670 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3671 sbinfo
->max_blocks
= shmem_default_max_blocks();
3672 sbinfo
->max_inodes
= shmem_default_max_inodes();
3673 if (shmem_parse_options(data
, sbinfo
, false)) {
3678 sb
->s_flags
|= MS_NOUSER
;
3680 sb
->s_export_op
= &shmem_export_ops
;
3681 sb
->s_flags
|= MS_NOSEC
;
3683 sb
->s_flags
|= MS_NOUSER
;
3686 spin_lock_init(&sbinfo
->stat_lock
);
3687 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3689 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3690 spin_lock_init(&sbinfo
->shrinklist_lock
);
3691 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3693 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3694 sb
->s_blocksize
= PAGE_SIZE
;
3695 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3696 sb
->s_magic
= TMPFS_MAGIC
;
3697 sb
->s_op
= &shmem_ops
;
3698 sb
->s_time_gran
= 1;
3699 #ifdef CONFIG_TMPFS_XATTR
3700 sb
->s_xattr
= shmem_xattr_handlers
;
3702 #ifdef CONFIG_TMPFS_POSIX_ACL
3703 sb
->s_flags
|= MS_POSIXACL
;
3706 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3709 inode
->i_uid
= sbinfo
->uid
;
3710 inode
->i_gid
= sbinfo
->gid
;
3711 sb
->s_root
= d_make_root(inode
);
3717 shmem_put_super(sb
);
3721 static struct kmem_cache
*shmem_inode_cachep
;
3723 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3725 struct shmem_inode_info
*info
;
3726 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3729 return &info
->vfs_inode
;
3732 static void shmem_destroy_callback(struct rcu_head
*head
)
3734 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3735 if (S_ISLNK(inode
->i_mode
))
3736 kfree(inode
->i_link
);
3737 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3740 static void shmem_destroy_inode(struct inode
*inode
)
3742 if (S_ISREG(inode
->i_mode
))
3743 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3744 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3747 static void shmem_init_inode(void *foo
)
3749 struct shmem_inode_info
*info
= foo
;
3750 inode_init_once(&info
->vfs_inode
);
3753 static int shmem_init_inodecache(void)
3755 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3756 sizeof(struct shmem_inode_info
),
3757 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3761 static void shmem_destroy_inodecache(void)
3763 kmem_cache_destroy(shmem_inode_cachep
);
3766 static const struct address_space_operations shmem_aops
= {
3767 .writepage
= shmem_writepage
,
3768 .set_page_dirty
= __set_page_dirty_no_writeback
,
3770 .write_begin
= shmem_write_begin
,
3771 .write_end
= shmem_write_end
,
3773 #ifdef CONFIG_MIGRATION
3774 .migratepage
= migrate_page
,
3776 .error_remove_page
= generic_error_remove_page
,
3779 static const struct file_operations shmem_file_operations
= {
3781 .get_unmapped_area
= shmem_get_unmapped_area
,
3783 .llseek
= shmem_file_llseek
,
3784 .read_iter
= shmem_file_read_iter
,
3785 .write_iter
= generic_file_write_iter
,
3786 .fsync
= noop_fsync
,
3787 .splice_read
= shmem_file_splice_read
,
3788 .splice_write
= iter_file_splice_write
,
3789 .fallocate
= shmem_fallocate
,
3793 static const struct inode_operations shmem_inode_operations
= {
3794 .getattr
= shmem_getattr
,
3795 .setattr
= shmem_setattr
,
3796 #ifdef CONFIG_TMPFS_XATTR
3797 .setxattr
= generic_setxattr
,
3798 .getxattr
= generic_getxattr
,
3799 .listxattr
= shmem_listxattr
,
3800 .removexattr
= generic_removexattr
,
3801 .set_acl
= simple_set_acl
,
3805 static const struct inode_operations shmem_dir_inode_operations
= {
3807 .create
= shmem_create
,
3808 .lookup
= simple_lookup
,
3810 .unlink
= shmem_unlink
,
3811 .symlink
= shmem_symlink
,
3812 .mkdir
= shmem_mkdir
,
3813 .rmdir
= shmem_rmdir
,
3814 .mknod
= shmem_mknod
,
3815 .rename2
= shmem_rename2
,
3816 .tmpfile
= shmem_tmpfile
,
3818 #ifdef CONFIG_TMPFS_XATTR
3819 .setxattr
= generic_setxattr
,
3820 .getxattr
= generic_getxattr
,
3821 .listxattr
= shmem_listxattr
,
3822 .removexattr
= generic_removexattr
,
3824 #ifdef CONFIG_TMPFS_POSIX_ACL
3825 .setattr
= shmem_setattr
,
3826 .set_acl
= simple_set_acl
,
3830 static const struct inode_operations shmem_special_inode_operations
= {
3831 #ifdef CONFIG_TMPFS_XATTR
3832 .setxattr
= generic_setxattr
,
3833 .getxattr
= generic_getxattr
,
3834 .listxattr
= shmem_listxattr
,
3835 .removexattr
= generic_removexattr
,
3837 #ifdef CONFIG_TMPFS_POSIX_ACL
3838 .setattr
= shmem_setattr
,
3839 .set_acl
= simple_set_acl
,
3843 static const struct super_operations shmem_ops
= {
3844 .alloc_inode
= shmem_alloc_inode
,
3845 .destroy_inode
= shmem_destroy_inode
,
3847 .statfs
= shmem_statfs
,
3848 .remount_fs
= shmem_remount_fs
,
3849 .show_options
= shmem_show_options
,
3851 .evict_inode
= shmem_evict_inode
,
3852 .drop_inode
= generic_delete_inode
,
3853 .put_super
= shmem_put_super
,
3854 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3855 .nr_cached_objects
= shmem_unused_huge_count
,
3856 .free_cached_objects
= shmem_unused_huge_scan
,
3860 static const struct vm_operations_struct shmem_vm_ops
= {
3861 .fault
= shmem_fault
,
3862 .map_pages
= filemap_map_pages
,
3864 .set_policy
= shmem_set_policy
,
3865 .get_policy
= shmem_get_policy
,
3869 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3870 int flags
, const char *dev_name
, void *data
)
3872 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3875 static struct file_system_type shmem_fs_type
= {
3876 .owner
= THIS_MODULE
,
3878 .mount
= shmem_mount
,
3879 .kill_sb
= kill_litter_super
,
3880 .fs_flags
= FS_USERNS_MOUNT
,
3883 int __init
shmem_init(void)
3887 /* If rootfs called this, don't re-init */
3888 if (shmem_inode_cachep
)
3891 error
= shmem_init_inodecache();
3895 error
= register_filesystem(&shmem_fs_type
);
3897 pr_err("Could not register tmpfs\n");
3901 shm_mnt
= kern_mount(&shmem_fs_type
);
3902 if (IS_ERR(shm_mnt
)) {
3903 error
= PTR_ERR(shm_mnt
);
3904 pr_err("Could not kern_mount tmpfs\n");
3908 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3909 if (has_transparent_hugepage() && shmem_huge
< SHMEM_HUGE_DENY
)
3910 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3912 shmem_huge
= 0; /* just in case it was patched */
3917 unregister_filesystem(&shmem_fs_type
);
3919 shmem_destroy_inodecache();
3921 shm_mnt
= ERR_PTR(error
);
3925 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3926 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3927 struct kobj_attribute
*attr
, char *buf
)
3931 SHMEM_HUGE_WITHIN_SIZE
,
3939 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3940 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3942 count
+= sprintf(buf
+ count
, fmt
,
3943 shmem_format_huge(values
[i
]));
3945 buf
[count
- 1] = '\n';
3949 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3950 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3955 if (count
+ 1 > sizeof(tmp
))
3957 memcpy(tmp
, buf
, count
);
3959 if (count
&& tmp
[count
- 1] == '\n')
3960 tmp
[count
- 1] = '\0';
3962 huge
= shmem_parse_huge(tmp
);
3963 if (huge
== -EINVAL
)
3965 if (!has_transparent_hugepage() &&
3966 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3970 if (shmem_huge
< SHMEM_HUGE_DENY
)
3971 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3975 struct kobj_attribute shmem_enabled_attr
=
3976 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3978 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
3980 struct inode
*inode
= file_inode(vma
->vm_file
);
3981 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3985 if (shmem_huge
== SHMEM_HUGE_FORCE
)
3987 if (shmem_huge
== SHMEM_HUGE_DENY
)
3989 switch (sbinfo
->huge
) {
3990 case SHMEM_HUGE_NEVER
:
3992 case SHMEM_HUGE_ALWAYS
:
3994 case SHMEM_HUGE_WITHIN_SIZE
:
3995 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
3996 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
3997 if (i_size
>= HPAGE_PMD_SIZE
&&
3998 i_size
>> PAGE_SHIFT
>= off
)
4000 case SHMEM_HUGE_ADVISE
:
4001 /* TODO: implement fadvise() hints */
4002 return (vma
->vm_flags
& VM_HUGEPAGE
);
4008 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4010 #else /* !CONFIG_SHMEM */
4013 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4015 * This is intended for small system where the benefits of the full
4016 * shmem code (swap-backed and resource-limited) are outweighed by
4017 * their complexity. On systems without swap this code should be
4018 * effectively equivalent, but much lighter weight.
4021 static struct file_system_type shmem_fs_type
= {
4023 .mount
= ramfs_mount
,
4024 .kill_sb
= kill_litter_super
,
4025 .fs_flags
= FS_USERNS_MOUNT
,
4028 int __init
shmem_init(void)
4030 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4032 shm_mnt
= kern_mount(&shmem_fs_type
);
4033 BUG_ON(IS_ERR(shm_mnt
));
4038 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
4043 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4048 void shmem_unlock_mapping(struct address_space
*mapping
)
4053 unsigned long shmem_get_unmapped_area(struct file
*file
,
4054 unsigned long addr
, unsigned long len
,
4055 unsigned long pgoff
, unsigned long flags
)
4057 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4061 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4063 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4065 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4067 #define shmem_vm_ops generic_file_vm_ops
4068 #define shmem_file_operations ramfs_file_operations
4069 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4070 #define shmem_acct_size(flags, size) 0
4071 #define shmem_unacct_size(flags, size) do {} while (0)
4073 #endif /* CONFIG_SHMEM */
4077 static struct dentry_operations anon_ops
= {
4078 .d_dname
= simple_dname
4081 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
4082 unsigned long flags
, unsigned int i_flags
)
4085 struct inode
*inode
;
4087 struct super_block
*sb
;
4090 if (IS_ERR(shm_mnt
))
4091 return ERR_CAST(shm_mnt
);
4093 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4094 return ERR_PTR(-EINVAL
);
4096 if (shmem_acct_size(flags
, size
))
4097 return ERR_PTR(-ENOMEM
);
4099 res
= ERR_PTR(-ENOMEM
);
4101 this.len
= strlen(name
);
4102 this.hash
= 0; /* will go */
4103 sb
= shm_mnt
->mnt_sb
;
4104 path
.mnt
= mntget(shm_mnt
);
4105 path
.dentry
= d_alloc_pseudo(sb
, &this);
4108 d_set_d_op(path
.dentry
, &anon_ops
);
4110 res
= ERR_PTR(-ENOSPC
);
4111 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
4115 inode
->i_flags
|= i_flags
;
4116 d_instantiate(path
.dentry
, inode
);
4117 inode
->i_size
= size
;
4118 clear_nlink(inode
); /* It is unlinked */
4119 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4123 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
4124 &shmem_file_operations
);
4131 shmem_unacct_size(flags
, size
);
4138 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4139 * kernel internal. There will be NO LSM permission checks against the
4140 * underlying inode. So users of this interface must do LSM checks at a
4141 * higher layer. The users are the big_key and shm implementations. LSM
4142 * checks are provided at the key or shm level rather than the inode.
4143 * @name: name for dentry (to be seen in /proc/<pid>/maps
4144 * @size: size to be set for the file
4145 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4147 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4149 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
4153 * shmem_file_setup - get an unlinked file living in tmpfs
4154 * @name: name for dentry (to be seen in /proc/<pid>/maps
4155 * @size: size to be set for the file
4156 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4158 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4160 return __shmem_file_setup(name
, size
, flags
, 0);
4162 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4165 * shmem_zero_setup - setup a shared anonymous mapping
4166 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4168 int shmem_zero_setup(struct vm_area_struct
*vma
)
4171 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4174 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4175 * between XFS directory reading and selinux: since this file is only
4176 * accessible to the user through its mapping, use S_PRIVATE flag to
4177 * bypass file security, in the same way as shmem_kernel_file_setup().
4179 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
4181 return PTR_ERR(file
);
4185 vma
->vm_file
= file
;
4186 vma
->vm_ops
= &shmem_vm_ops
;
4188 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4189 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4190 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4191 khugepaged_enter(vma
, vma
->vm_flags
);
4198 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4199 * @mapping: the page's address_space
4200 * @index: the page index
4201 * @gfp: the page allocator flags to use if allocating
4203 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4204 * with any new page allocations done using the specified allocation flags.
4205 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4206 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4207 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4209 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4210 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4212 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4213 pgoff_t index
, gfp_t gfp
)
4216 struct inode
*inode
= mapping
->host
;
4220 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4221 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4224 page
= ERR_PTR(error
);
4230 * The tiny !SHMEM case uses ramfs without swap
4232 return read_cache_page_gfp(mapping
, index
, gfp
);
4235 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);