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
);
273 shmem_unacct_blocks(info
->flags
, pages
);
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
);
294 shmem_unacct_blocks(info
->flags
, pages
);
298 * Replace item expected in radix tree by a new item, while holding tree lock.
300 static int shmem_radix_tree_replace(struct address_space
*mapping
,
301 pgoff_t index
, void *expected
, void *replacement
)
306 VM_BUG_ON(!expected
);
307 VM_BUG_ON(!replacement
);
308 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
311 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
312 if (item
!= expected
)
314 radix_tree_replace_slot(pslot
, replacement
);
319 * Sometimes, before we decide whether to proceed or to fail, we must check
320 * that an entry was not already brought back from swap by a racing thread.
322 * Checking page is not enough: by the time a SwapCache page is locked, it
323 * might be reused, and again be SwapCache, using the same swap as before.
325 static bool shmem_confirm_swap(struct address_space
*mapping
,
326 pgoff_t index
, swp_entry_t swap
)
331 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
333 return item
== swp_to_radix_entry(swap
);
337 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
340 * disables huge pages for the mount;
342 * enables huge pages for the mount;
343 * SHMEM_HUGE_WITHIN_SIZE:
344 * only allocate huge pages if the page will be fully within i_size,
345 * also respect fadvise()/madvise() hints;
347 * only allocate huge pages if requested with fadvise()/madvise();
350 #define SHMEM_HUGE_NEVER 0
351 #define SHMEM_HUGE_ALWAYS 1
352 #define SHMEM_HUGE_WITHIN_SIZE 2
353 #define SHMEM_HUGE_ADVISE 3
357 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
360 * disables huge on shm_mnt and all mounts, for emergency use;
362 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
365 #define SHMEM_HUGE_DENY (-1)
366 #define SHMEM_HUGE_FORCE (-2)
368 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
369 /* ifdef here to avoid bloating shmem.o when not necessary */
371 int shmem_huge __read_mostly
;
373 static int shmem_parse_huge(const char *str
)
375 if (!strcmp(str
, "never"))
376 return SHMEM_HUGE_NEVER
;
377 if (!strcmp(str
, "always"))
378 return SHMEM_HUGE_ALWAYS
;
379 if (!strcmp(str
, "within_size"))
380 return SHMEM_HUGE_WITHIN_SIZE
;
381 if (!strcmp(str
, "advise"))
382 return SHMEM_HUGE_ADVISE
;
383 if (!strcmp(str
, "deny"))
384 return SHMEM_HUGE_DENY
;
385 if (!strcmp(str
, "force"))
386 return SHMEM_HUGE_FORCE
;
390 static const char *shmem_format_huge(int huge
)
393 case SHMEM_HUGE_NEVER
:
395 case SHMEM_HUGE_ALWAYS
:
397 case SHMEM_HUGE_WITHIN_SIZE
:
398 return "within_size";
399 case SHMEM_HUGE_ADVISE
:
401 case SHMEM_HUGE_DENY
:
403 case SHMEM_HUGE_FORCE
:
411 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
412 struct shrink_control
*sc
, unsigned long nr_to_split
)
414 LIST_HEAD(list
), *pos
, *next
;
416 struct shmem_inode_info
*info
;
418 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
419 int removed
= 0, split
= 0;
421 if (list_empty(&sbinfo
->shrinklist
))
424 spin_lock(&sbinfo
->shrinklist_lock
);
425 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
426 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
429 inode
= igrab(&info
->vfs_inode
);
431 /* inode is about to be evicted */
433 list_del_init(&info
->shrinklist
);
438 /* Check if there's anything to gain */
439 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
440 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
441 list_del_init(&info
->shrinklist
);
447 list_move(&info
->shrinklist
, &list
);
452 spin_unlock(&sbinfo
->shrinklist_lock
);
454 list_for_each_safe(pos
, next
, &list
) {
457 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
458 inode
= &info
->vfs_inode
;
460 if (nr_to_split
&& split
>= nr_to_split
) {
465 page
= find_lock_page(inode
->i_mapping
,
466 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
470 if (!PageTransHuge(page
)) {
476 ret
= split_huge_page(page
);
481 /* split failed: leave it on the list */
488 list_del_init(&info
->shrinklist
);
493 spin_lock(&sbinfo
->shrinklist_lock
);
494 list_splice_tail(&list
, &sbinfo
->shrinklist
);
495 sbinfo
->shrinklist_len
-= removed
;
496 spin_unlock(&sbinfo
->shrinklist_lock
);
501 static long shmem_unused_huge_scan(struct super_block
*sb
,
502 struct shrink_control
*sc
)
504 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
506 if (!READ_ONCE(sbinfo
->shrinklist_len
))
509 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
512 static long shmem_unused_huge_count(struct super_block
*sb
,
513 struct shrink_control
*sc
)
515 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
516 return READ_ONCE(sbinfo
->shrinklist_len
);
518 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
520 #define shmem_huge SHMEM_HUGE_DENY
522 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
523 struct shrink_control
*sc
, unsigned long nr_to_split
)
527 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
530 * Like add_to_page_cache_locked, but error if expected item has gone.
532 static int shmem_add_to_page_cache(struct page
*page
,
533 struct address_space
*mapping
,
534 pgoff_t index
, void *expected
)
536 int error
, nr
= hpage_nr_pages(page
);
538 VM_BUG_ON_PAGE(PageTail(page
), page
);
539 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
540 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
541 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
542 VM_BUG_ON(expected
&& PageTransHuge(page
));
544 page_ref_add(page
, nr
);
545 page
->mapping
= mapping
;
548 spin_lock_irq(&mapping
->tree_lock
);
549 if (PageTransHuge(page
)) {
550 void __rcu
**results
;
555 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
,
556 &results
, &idx
, index
, 1) &&
557 idx
< index
+ HPAGE_PMD_NR
) {
562 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
563 error
= radix_tree_insert(&mapping
->page_tree
,
564 index
+ i
, page
+ i
);
567 count_vm_event(THP_FILE_ALLOC
);
569 } else if (!expected
) {
570 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
572 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
577 mapping
->nrpages
+= nr
;
578 if (PageTransHuge(page
))
579 __inc_node_page_state(page
, NR_SHMEM_THPS
);
580 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
581 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
582 spin_unlock_irq(&mapping
->tree_lock
);
584 page
->mapping
= NULL
;
585 spin_unlock_irq(&mapping
->tree_lock
);
586 page_ref_sub(page
, nr
);
592 * Like delete_from_page_cache, but substitutes swap for page.
594 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
596 struct address_space
*mapping
= page
->mapping
;
599 VM_BUG_ON_PAGE(PageCompound(page
), page
);
601 spin_lock_irq(&mapping
->tree_lock
);
602 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
603 page
->mapping
= NULL
;
605 __dec_node_page_state(page
, NR_FILE_PAGES
);
606 __dec_node_page_state(page
, NR_SHMEM
);
607 spin_unlock_irq(&mapping
->tree_lock
);
613 * Remove swap entry from radix tree, free the swap and its page cache.
615 static int shmem_free_swap(struct address_space
*mapping
,
616 pgoff_t index
, void *radswap
)
620 spin_lock_irq(&mapping
->tree_lock
);
621 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
622 spin_unlock_irq(&mapping
->tree_lock
);
625 free_swap_and_cache(radix_to_swp_entry(radswap
));
630 * Determine (in bytes) how many of the shmem object's pages mapped by the
631 * given offsets are swapped out.
633 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
634 * as long as the inode doesn't go away and racy results are not a problem.
636 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
637 pgoff_t start
, pgoff_t end
)
639 struct radix_tree_iter iter
;
642 unsigned long swapped
= 0;
646 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
647 if (iter
.index
>= end
)
650 page
= radix_tree_deref_slot(slot
);
652 if (radix_tree_deref_retry(page
)) {
653 slot
= radix_tree_iter_retry(&iter
);
657 if (radix_tree_exceptional_entry(page
))
660 if (need_resched()) {
662 slot
= radix_tree_iter_next(&iter
);
668 return swapped
<< PAGE_SHIFT
;
672 * Determine (in bytes) how many of the shmem object's pages mapped by the
673 * given vma is swapped out.
675 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
676 * as long as the inode doesn't go away and racy results are not a problem.
678 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
680 struct inode
*inode
= file_inode(vma
->vm_file
);
681 struct shmem_inode_info
*info
= SHMEM_I(inode
);
682 struct address_space
*mapping
= inode
->i_mapping
;
683 unsigned long swapped
;
685 /* Be careful as we don't hold info->lock */
686 swapped
= READ_ONCE(info
->swapped
);
689 * The easier cases are when the shmem object has nothing in swap, or
690 * the vma maps it whole. Then we can simply use the stats that we
696 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
697 return swapped
<< PAGE_SHIFT
;
699 /* Here comes the more involved part */
700 return shmem_partial_swap_usage(mapping
,
701 linear_page_index(vma
, vma
->vm_start
),
702 linear_page_index(vma
, vma
->vm_end
));
706 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
708 void shmem_unlock_mapping(struct address_space
*mapping
)
711 pgoff_t indices
[PAGEVEC_SIZE
];
714 pagevec_init(&pvec
, 0);
716 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
718 while (!mapping_unevictable(mapping
)) {
720 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
721 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
723 pvec
.nr
= find_get_entries(mapping
, index
,
724 PAGEVEC_SIZE
, pvec
.pages
, indices
);
727 index
= indices
[pvec
.nr
- 1] + 1;
728 pagevec_remove_exceptionals(&pvec
);
729 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
730 pagevec_release(&pvec
);
736 * Remove range of pages and swap entries from radix tree, and free them.
737 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
739 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
742 struct address_space
*mapping
= inode
->i_mapping
;
743 struct shmem_inode_info
*info
= SHMEM_I(inode
);
744 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
745 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
746 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
747 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
749 pgoff_t indices
[PAGEVEC_SIZE
];
750 long nr_swaps_freed
= 0;
755 end
= -1; /* unsigned, so actually very big */
757 pagevec_init(&pvec
, 0);
759 while (index
< end
) {
760 pvec
.nr
= find_get_entries(mapping
, index
,
761 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
762 pvec
.pages
, indices
);
765 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
766 struct page
*page
= pvec
.pages
[i
];
772 if (radix_tree_exceptional_entry(page
)) {
775 nr_swaps_freed
+= !shmem_free_swap(mapping
,
780 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
782 if (!trylock_page(page
))
785 if (PageTransTail(page
)) {
786 /* Middle of THP: zero out the page */
787 clear_highpage(page
);
790 } else if (PageTransHuge(page
)) {
791 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
793 * Range ends in the middle of THP:
796 clear_highpage(page
);
800 index
+= HPAGE_PMD_NR
- 1;
801 i
+= HPAGE_PMD_NR
- 1;
804 if (!unfalloc
|| !PageUptodate(page
)) {
805 VM_BUG_ON_PAGE(PageTail(page
), page
);
806 if (page_mapping(page
) == mapping
) {
807 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
808 truncate_inode_page(mapping
, page
);
813 pagevec_remove_exceptionals(&pvec
);
814 pagevec_release(&pvec
);
820 struct page
*page
= NULL
;
821 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
823 unsigned int top
= PAGE_SIZE
;
828 zero_user_segment(page
, partial_start
, top
);
829 set_page_dirty(page
);
835 struct page
*page
= NULL
;
836 shmem_getpage(inode
, end
, &page
, SGP_READ
);
838 zero_user_segment(page
, 0, partial_end
);
839 set_page_dirty(page
);
848 while (index
< end
) {
851 pvec
.nr
= find_get_entries(mapping
, index
,
852 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
853 pvec
.pages
, indices
);
855 /* If all gone or hole-punch or unfalloc, we're done */
856 if (index
== start
|| end
!= -1)
858 /* But if truncating, restart to make sure all gone */
862 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
863 struct page
*page
= pvec
.pages
[i
];
869 if (radix_tree_exceptional_entry(page
)) {
872 if (shmem_free_swap(mapping
, index
, page
)) {
873 /* Swap was replaced by page: retry */
883 if (PageTransTail(page
)) {
884 /* Middle of THP: zero out the page */
885 clear_highpage(page
);
888 * Partial thp truncate due 'start' in middle
889 * of THP: don't need to look on these pages
890 * again on !pvec.nr restart.
892 if (index
!= round_down(end
, HPAGE_PMD_NR
))
895 } else if (PageTransHuge(page
)) {
896 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
898 * Range ends in the middle of THP:
901 clear_highpage(page
);
905 index
+= HPAGE_PMD_NR
- 1;
906 i
+= HPAGE_PMD_NR
- 1;
909 if (!unfalloc
|| !PageUptodate(page
)) {
910 VM_BUG_ON_PAGE(PageTail(page
), page
);
911 if (page_mapping(page
) == mapping
) {
912 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
913 truncate_inode_page(mapping
, page
);
915 /* Page was replaced by swap: retry */
923 pagevec_remove_exceptionals(&pvec
);
924 pagevec_release(&pvec
);
928 spin_lock_irq(&info
->lock
);
929 info
->swapped
-= nr_swaps_freed
;
930 shmem_recalc_inode(inode
);
931 spin_unlock_irq(&info
->lock
);
934 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
936 shmem_undo_range(inode
, lstart
, lend
, false);
937 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
939 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
941 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
944 struct inode
*inode
= dentry
->d_inode
;
945 struct shmem_inode_info
*info
= SHMEM_I(inode
);
947 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
948 spin_lock_irq(&info
->lock
);
949 shmem_recalc_inode(inode
);
950 spin_unlock_irq(&info
->lock
);
952 generic_fillattr(inode
, stat
);
956 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
958 struct inode
*inode
= d_inode(dentry
);
959 struct shmem_inode_info
*info
= SHMEM_I(inode
);
960 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
963 error
= setattr_prepare(dentry
, attr
);
967 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
968 loff_t oldsize
= inode
->i_size
;
969 loff_t newsize
= attr
->ia_size
;
971 /* protected by i_mutex */
972 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
973 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
976 if (newsize
!= oldsize
) {
977 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
981 i_size_write(inode
, newsize
);
982 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
984 if (newsize
<= oldsize
) {
985 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
986 if (oldsize
> holebegin
)
987 unmap_mapping_range(inode
->i_mapping
,
990 shmem_truncate_range(inode
,
991 newsize
, (loff_t
)-1);
992 /* unmap again to remove racily COWed private pages */
993 if (oldsize
> holebegin
)
994 unmap_mapping_range(inode
->i_mapping
,
998 * Part of the huge page can be beyond i_size: subject
999 * to shrink under memory pressure.
1001 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1002 spin_lock(&sbinfo
->shrinklist_lock
);
1003 if (list_empty(&info
->shrinklist
)) {
1004 list_add_tail(&info
->shrinklist
,
1005 &sbinfo
->shrinklist
);
1006 sbinfo
->shrinklist_len
++;
1008 spin_unlock(&sbinfo
->shrinklist_lock
);
1013 setattr_copy(inode
, attr
);
1014 if (attr
->ia_valid
& ATTR_MODE
)
1015 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1019 static void shmem_evict_inode(struct inode
*inode
)
1021 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1022 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1024 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1025 shmem_unacct_size(info
->flags
, inode
->i_size
);
1027 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1028 if (!list_empty(&info
->shrinklist
)) {
1029 spin_lock(&sbinfo
->shrinklist_lock
);
1030 if (!list_empty(&info
->shrinklist
)) {
1031 list_del_init(&info
->shrinklist
);
1032 sbinfo
->shrinklist_len
--;
1034 spin_unlock(&sbinfo
->shrinklist_lock
);
1036 if (!list_empty(&info
->swaplist
)) {
1037 mutex_lock(&shmem_swaplist_mutex
);
1038 list_del_init(&info
->swaplist
);
1039 mutex_unlock(&shmem_swaplist_mutex
);
1043 simple_xattrs_free(&info
->xattrs
);
1044 WARN_ON(inode
->i_blocks
);
1045 shmem_free_inode(inode
->i_sb
);
1050 * If swap found in inode, free it and move page from swapcache to filecache.
1052 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
1053 swp_entry_t swap
, struct page
**pagep
)
1055 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1061 radswap
= swp_to_radix_entry(swap
);
1062 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
1064 return -EAGAIN
; /* tell shmem_unuse we found nothing */
1067 * Move _head_ to start search for next from here.
1068 * But be careful: shmem_evict_inode checks list_empty without taking
1069 * mutex, and there's an instant in list_move_tail when info->swaplist
1070 * would appear empty, if it were the only one on shmem_swaplist.
1072 if (shmem_swaplist
.next
!= &info
->swaplist
)
1073 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
1075 gfp
= mapping_gfp_mask(mapping
);
1076 if (shmem_should_replace_page(*pagep
, gfp
)) {
1077 mutex_unlock(&shmem_swaplist_mutex
);
1078 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
1079 mutex_lock(&shmem_swaplist_mutex
);
1081 * We needed to drop mutex to make that restrictive page
1082 * allocation, but the inode might have been freed while we
1083 * dropped it: although a racing shmem_evict_inode() cannot
1084 * complete without emptying the radix_tree, our page lock
1085 * on this swapcache page is not enough to prevent that -
1086 * free_swap_and_cache() of our swap entry will only
1087 * trylock_page(), removing swap from radix_tree whatever.
1089 * We must not proceed to shmem_add_to_page_cache() if the
1090 * inode has been freed, but of course we cannot rely on
1091 * inode or mapping or info to check that. However, we can
1092 * safely check if our swap entry is still in use (and here
1093 * it can't have got reused for another page): if it's still
1094 * in use, then the inode cannot have been freed yet, and we
1095 * can safely proceed (if it's no longer in use, that tells
1096 * nothing about the inode, but we don't need to unuse swap).
1098 if (!page_swapcount(*pagep
))
1103 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1104 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1105 * beneath us (pagelock doesn't help until the page is in pagecache).
1108 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
1110 if (error
!= -ENOMEM
) {
1112 * Truncation and eviction use free_swap_and_cache(), which
1113 * only does trylock page: if we raced, best clean up here.
1115 delete_from_swap_cache(*pagep
);
1116 set_page_dirty(*pagep
);
1118 spin_lock_irq(&info
->lock
);
1120 spin_unlock_irq(&info
->lock
);
1128 * Search through swapped inodes to find and replace swap by page.
1130 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1132 struct list_head
*this, *next
;
1133 struct shmem_inode_info
*info
;
1134 struct mem_cgroup
*memcg
;
1138 * There's a faint possibility that swap page was replaced before
1139 * caller locked it: caller will come back later with the right page.
1141 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1145 * Charge page using GFP_KERNEL while we can wait, before taking
1146 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1147 * Charged back to the user (not to caller) when swap account is used.
1149 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1153 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1156 mutex_lock(&shmem_swaplist_mutex
);
1157 list_for_each_safe(this, next
, &shmem_swaplist
) {
1158 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1160 error
= shmem_unuse_inode(info
, swap
, &page
);
1162 list_del_init(&info
->swaplist
);
1164 if (error
!= -EAGAIN
)
1166 /* found nothing in this: move on to search the next */
1168 mutex_unlock(&shmem_swaplist_mutex
);
1171 if (error
!= -ENOMEM
)
1173 mem_cgroup_cancel_charge(page
, memcg
, false);
1175 mem_cgroup_commit_charge(page
, memcg
, true, false);
1183 * Move the page from the page cache to the swap cache.
1185 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1187 struct shmem_inode_info
*info
;
1188 struct address_space
*mapping
;
1189 struct inode
*inode
;
1193 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1194 BUG_ON(!PageLocked(page
));
1195 mapping
= page
->mapping
;
1196 index
= page
->index
;
1197 inode
= mapping
->host
;
1198 info
= SHMEM_I(inode
);
1199 if (info
->flags
& VM_LOCKED
)
1201 if (!total_swap_pages
)
1205 * Our capabilities prevent regular writeback or sync from ever calling
1206 * shmem_writepage; but a stacking filesystem might use ->writepage of
1207 * its underlying filesystem, in which case tmpfs should write out to
1208 * swap only in response to memory pressure, and not for the writeback
1211 if (!wbc
->for_reclaim
) {
1212 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1217 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1218 * value into swapfile.c, the only way we can correctly account for a
1219 * fallocated page arriving here is now to initialize it and write it.
1221 * That's okay for a page already fallocated earlier, but if we have
1222 * not yet completed the fallocation, then (a) we want to keep track
1223 * of this page in case we have to undo it, and (b) it may not be a
1224 * good idea to continue anyway, once we're pushing into swap. So
1225 * reactivate the page, and let shmem_fallocate() quit when too many.
1227 if (!PageUptodate(page
)) {
1228 if (inode
->i_private
) {
1229 struct shmem_falloc
*shmem_falloc
;
1230 spin_lock(&inode
->i_lock
);
1231 shmem_falloc
= inode
->i_private
;
1233 !shmem_falloc
->waitq
&&
1234 index
>= shmem_falloc
->start
&&
1235 index
< shmem_falloc
->next
)
1236 shmem_falloc
->nr_unswapped
++;
1238 shmem_falloc
= NULL
;
1239 spin_unlock(&inode
->i_lock
);
1243 clear_highpage(page
);
1244 flush_dcache_page(page
);
1245 SetPageUptodate(page
);
1248 swap
= get_swap_page();
1252 if (mem_cgroup_try_charge_swap(page
, swap
))
1256 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1257 * if it's not already there. Do it now before the page is
1258 * moved to swap cache, when its pagelock no longer protects
1259 * the inode from eviction. But don't unlock the mutex until
1260 * we've incremented swapped, because shmem_unuse_inode() will
1261 * prune a !swapped inode from the swaplist under this mutex.
1263 mutex_lock(&shmem_swaplist_mutex
);
1264 if (list_empty(&info
->swaplist
))
1265 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1267 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1268 spin_lock_irq(&info
->lock
);
1269 shmem_recalc_inode(inode
);
1271 spin_unlock_irq(&info
->lock
);
1273 swap_shmem_alloc(swap
);
1274 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1276 mutex_unlock(&shmem_swaplist_mutex
);
1277 BUG_ON(page_mapped(page
));
1278 swap_writepage(page
, wbc
);
1282 mutex_unlock(&shmem_swaplist_mutex
);
1284 swapcache_free(swap
);
1286 set_page_dirty(page
);
1287 if (wbc
->for_reclaim
)
1288 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1293 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1294 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1298 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1299 return; /* show nothing */
1301 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1303 seq_printf(seq
, ",mpol=%s", buffer
);
1306 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1308 struct mempolicy
*mpol
= NULL
;
1310 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1311 mpol
= sbinfo
->mpol
;
1313 spin_unlock(&sbinfo
->stat_lock
);
1317 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1318 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1321 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1325 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1327 #define vm_policy vm_private_data
1330 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1331 struct shmem_inode_info
*info
, pgoff_t index
)
1333 /* Create a pseudo vma that just contains the policy */
1335 /* Bias interleave by inode number to distribute better across nodes */
1336 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1338 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1341 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1343 /* Drop reference taken by mpol_shared_policy_lookup() */
1344 mpol_cond_put(vma
->vm_policy
);
1347 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1348 struct shmem_inode_info
*info
, pgoff_t index
)
1350 struct vm_area_struct pvma
;
1353 shmem_pseudo_vma_init(&pvma
, info
, index
);
1354 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1355 shmem_pseudo_vma_destroy(&pvma
);
1360 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1361 struct shmem_inode_info
*info
, pgoff_t index
)
1363 struct vm_area_struct pvma
;
1364 struct inode
*inode
= &info
->vfs_inode
;
1365 struct address_space
*mapping
= inode
->i_mapping
;
1366 pgoff_t idx
, hindex
;
1367 void __rcu
**results
;
1370 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1373 hindex
= round_down(index
, HPAGE_PMD_NR
);
1375 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1376 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1382 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1383 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1384 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1385 shmem_pseudo_vma_destroy(&pvma
);
1387 prep_transhuge_page(page
);
1391 static struct page
*shmem_alloc_page(gfp_t gfp
,
1392 struct shmem_inode_info
*info
, pgoff_t index
)
1394 struct vm_area_struct pvma
;
1397 shmem_pseudo_vma_init(&pvma
, info
, index
);
1398 page
= alloc_page_vma(gfp
, &pvma
, 0);
1399 shmem_pseudo_vma_destroy(&pvma
);
1404 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1405 struct shmem_inode_info
*info
, struct shmem_sb_info
*sbinfo
,
1406 pgoff_t index
, bool huge
)
1412 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1414 nr
= huge
? HPAGE_PMD_NR
: 1;
1416 if (shmem_acct_block(info
->flags
, nr
))
1418 if (sbinfo
->max_blocks
) {
1419 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1420 sbinfo
->max_blocks
- nr
) > 0)
1422 percpu_counter_add(&sbinfo
->used_blocks
, nr
);
1426 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1428 page
= shmem_alloc_page(gfp
, info
, index
);
1430 __SetPageLocked(page
);
1431 __SetPageSwapBacked(page
);
1436 if (sbinfo
->max_blocks
)
1437 percpu_counter_add(&sbinfo
->used_blocks
, -nr
);
1439 shmem_unacct_blocks(info
->flags
, nr
);
1441 return ERR_PTR(err
);
1445 * When a page is moved from swapcache to shmem filecache (either by the
1446 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1447 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1448 * ignorance of the mapping it belongs to. If that mapping has special
1449 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1450 * we may need to copy to a suitable page before moving to filecache.
1452 * In a future release, this may well be extended to respect cpuset and
1453 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1454 * but for now it is a simple matter of zone.
1456 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1458 return page_zonenum(page
) > gfp_zone(gfp
);
1461 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1462 struct shmem_inode_info
*info
, pgoff_t index
)
1464 struct page
*oldpage
, *newpage
;
1465 struct address_space
*swap_mapping
;
1470 swap_index
= page_private(oldpage
);
1471 swap_mapping
= page_mapping(oldpage
);
1474 * We have arrived here because our zones are constrained, so don't
1475 * limit chance of success by further cpuset and node constraints.
1477 gfp
&= ~GFP_CONSTRAINT_MASK
;
1478 newpage
= shmem_alloc_page(gfp
, info
, index
);
1483 copy_highpage(newpage
, oldpage
);
1484 flush_dcache_page(newpage
);
1486 __SetPageLocked(newpage
);
1487 __SetPageSwapBacked(newpage
);
1488 SetPageUptodate(newpage
);
1489 set_page_private(newpage
, swap_index
);
1490 SetPageSwapCache(newpage
);
1493 * Our caller will very soon move newpage out of swapcache, but it's
1494 * a nice clean interface for us to replace oldpage by newpage there.
1496 spin_lock_irq(&swap_mapping
->tree_lock
);
1497 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1500 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1501 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1503 spin_unlock_irq(&swap_mapping
->tree_lock
);
1505 if (unlikely(error
)) {
1507 * Is this possible? I think not, now that our callers check
1508 * both PageSwapCache and page_private after getting page lock;
1509 * but be defensive. Reverse old to newpage for clear and free.
1513 mem_cgroup_migrate(oldpage
, newpage
);
1514 lru_cache_add_anon(newpage
);
1518 ClearPageSwapCache(oldpage
);
1519 set_page_private(oldpage
, 0);
1521 unlock_page(oldpage
);
1528 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1530 * If we allocate a new one we do not mark it dirty. That's up to the
1531 * vm. If we swap it in we mark it dirty since we also free the swap
1532 * entry since a page cannot live in both the swap and page cache.
1534 * fault_mm and fault_type are only supplied by shmem_fault:
1535 * otherwise they are NULL.
1537 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1538 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1539 struct mm_struct
*fault_mm
, int *fault_type
)
1541 struct address_space
*mapping
= inode
->i_mapping
;
1542 struct shmem_inode_info
*info
;
1543 struct shmem_sb_info
*sbinfo
;
1544 struct mm_struct
*charge_mm
;
1545 struct mem_cgroup
*memcg
;
1548 enum sgp_type sgp_huge
= sgp
;
1549 pgoff_t hindex
= index
;
1554 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1556 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1560 page
= find_lock_entry(mapping
, index
);
1561 if (radix_tree_exceptional_entry(page
)) {
1562 swap
= radix_to_swp_entry(page
);
1566 if (sgp
<= SGP_CACHE
&&
1567 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1572 if (page
&& sgp
== SGP_WRITE
)
1573 mark_page_accessed(page
);
1575 /* fallocated page? */
1576 if (page
&& !PageUptodate(page
)) {
1577 if (sgp
!= SGP_READ
)
1583 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1589 * Fast cache lookup did not find it:
1590 * bring it back from swap or allocate.
1592 info
= SHMEM_I(inode
);
1593 sbinfo
= SHMEM_SB(inode
->i_sb
);
1594 charge_mm
= fault_mm
? : current
->mm
;
1597 /* Look it up and read it in.. */
1598 page
= lookup_swap_cache(swap
);
1600 /* Or update major stats only when swapin succeeds?? */
1602 *fault_type
|= VM_FAULT_MAJOR
;
1603 count_vm_event(PGMAJFAULT
);
1604 mem_cgroup_count_vm_event(fault_mm
, PGMAJFAULT
);
1606 /* Here we actually start the io */
1607 page
= shmem_swapin(swap
, gfp
, info
, index
);
1614 /* We have to do this with page locked to prevent races */
1616 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1617 !shmem_confirm_swap(mapping
, index
, swap
)) {
1618 error
= -EEXIST
; /* try again */
1621 if (!PageUptodate(page
)) {
1625 wait_on_page_writeback(page
);
1627 if (shmem_should_replace_page(page
, gfp
)) {
1628 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1633 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1636 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1637 swp_to_radix_entry(swap
));
1639 * We already confirmed swap under page lock, and make
1640 * no memory allocation here, so usually no possibility
1641 * of error; but free_swap_and_cache() only trylocks a
1642 * page, so it is just possible that the entry has been
1643 * truncated or holepunched since swap was confirmed.
1644 * shmem_undo_range() will have done some of the
1645 * unaccounting, now delete_from_swap_cache() will do
1647 * Reset swap.val? No, leave it so "failed" goes back to
1648 * "repeat": reading a hole and writing should succeed.
1651 mem_cgroup_cancel_charge(page
, memcg
, false);
1652 delete_from_swap_cache(page
);
1658 mem_cgroup_commit_charge(page
, memcg
, true, false);
1660 spin_lock_irq(&info
->lock
);
1662 shmem_recalc_inode(inode
);
1663 spin_unlock_irq(&info
->lock
);
1665 if (sgp
== SGP_WRITE
)
1666 mark_page_accessed(page
);
1668 delete_from_swap_cache(page
);
1669 set_page_dirty(page
);
1673 /* shmem_symlink() */
1674 if (mapping
->a_ops
!= &shmem_aops
)
1676 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1678 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1680 switch (sbinfo
->huge
) {
1683 case SHMEM_HUGE_NEVER
:
1685 case SHMEM_HUGE_WITHIN_SIZE
:
1686 off
= round_up(index
, HPAGE_PMD_NR
);
1687 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1688 if (i_size
>= HPAGE_PMD_SIZE
&&
1689 i_size
>> PAGE_SHIFT
>= off
)
1692 case SHMEM_HUGE_ADVISE
:
1693 if (sgp_huge
== SGP_HUGE
)
1695 /* TODO: implement fadvise() hints */
1700 page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1703 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1708 error
= PTR_ERR(page
);
1710 if (error
!= -ENOSPC
)
1713 * Try to reclaim some spece by splitting a huge page
1714 * beyond i_size on the filesystem.
1718 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1719 if (ret
== SHRINK_STOP
)
1727 if (PageTransHuge(page
))
1728 hindex
= round_down(index
, HPAGE_PMD_NR
);
1732 if (sgp
== SGP_WRITE
)
1733 __SetPageReferenced(page
);
1735 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1736 PageTransHuge(page
));
1739 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1740 compound_order(page
));
1742 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1744 radix_tree_preload_end();
1747 mem_cgroup_cancel_charge(page
, memcg
,
1748 PageTransHuge(page
));
1751 mem_cgroup_commit_charge(page
, memcg
, false,
1752 PageTransHuge(page
));
1753 lru_cache_add_anon(page
);
1755 spin_lock_irq(&info
->lock
);
1756 info
->alloced
+= 1 << compound_order(page
);
1757 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1758 shmem_recalc_inode(inode
);
1759 spin_unlock_irq(&info
->lock
);
1762 if (PageTransHuge(page
) &&
1763 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1764 hindex
+ HPAGE_PMD_NR
- 1) {
1766 * Part of the huge page is beyond i_size: subject
1767 * to shrink under memory pressure.
1769 spin_lock(&sbinfo
->shrinklist_lock
);
1770 if (list_empty(&info
->shrinklist
)) {
1771 list_add_tail(&info
->shrinklist
,
1772 &sbinfo
->shrinklist
);
1773 sbinfo
->shrinklist_len
++;
1775 spin_unlock(&sbinfo
->shrinklist_lock
);
1779 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1781 if (sgp
== SGP_FALLOC
)
1785 * Let SGP_WRITE caller clear ends if write does not fill page;
1786 * but SGP_FALLOC on a page fallocated earlier must initialize
1787 * it now, lest undo on failure cancel our earlier guarantee.
1789 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1790 struct page
*head
= compound_head(page
);
1793 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1794 clear_highpage(head
+ i
);
1795 flush_dcache_page(head
+ i
);
1797 SetPageUptodate(head
);
1801 /* Perhaps the file has been truncated since we checked */
1802 if (sgp
<= SGP_CACHE
&&
1803 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1805 ClearPageDirty(page
);
1806 delete_from_page_cache(page
);
1807 spin_lock_irq(&info
->lock
);
1808 shmem_recalc_inode(inode
);
1809 spin_unlock_irq(&info
->lock
);
1814 *pagep
= page
+ index
- hindex
;
1821 if (sbinfo
->max_blocks
)
1822 percpu_counter_sub(&sbinfo
->used_blocks
,
1823 1 << compound_order(page
));
1824 shmem_unacct_blocks(info
->flags
, 1 << compound_order(page
));
1826 if (PageTransHuge(page
)) {
1832 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1839 if (error
== -ENOSPC
&& !once
++) {
1840 info
= SHMEM_I(inode
);
1841 spin_lock_irq(&info
->lock
);
1842 shmem_recalc_inode(inode
);
1843 spin_unlock_irq(&info
->lock
);
1846 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1851 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1853 struct inode
*inode
= file_inode(vma
->vm_file
);
1854 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1857 int ret
= VM_FAULT_LOCKED
;
1860 * Trinity finds that probing a hole which tmpfs is punching can
1861 * prevent the hole-punch from ever completing: which in turn
1862 * locks writers out with its hold on i_mutex. So refrain from
1863 * faulting pages into the hole while it's being punched. Although
1864 * shmem_undo_range() does remove the additions, it may be unable to
1865 * keep up, as each new page needs its own unmap_mapping_range() call,
1866 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1868 * It does not matter if we sometimes reach this check just before the
1869 * hole-punch begins, so that one fault then races with the punch:
1870 * we just need to make racing faults a rare case.
1872 * The implementation below would be much simpler if we just used a
1873 * standard mutex or completion: but we cannot take i_mutex in fault,
1874 * and bloating every shmem inode for this unlikely case would be sad.
1876 if (unlikely(inode
->i_private
)) {
1877 struct shmem_falloc
*shmem_falloc
;
1879 spin_lock(&inode
->i_lock
);
1880 shmem_falloc
= inode
->i_private
;
1882 shmem_falloc
->waitq
&&
1883 vmf
->pgoff
>= shmem_falloc
->start
&&
1884 vmf
->pgoff
< shmem_falloc
->next
) {
1885 wait_queue_head_t
*shmem_falloc_waitq
;
1886 DEFINE_WAIT(shmem_fault_wait
);
1888 ret
= VM_FAULT_NOPAGE
;
1889 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1890 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1891 /* It's polite to up mmap_sem if we can */
1892 up_read(&vma
->vm_mm
->mmap_sem
);
1893 ret
= VM_FAULT_RETRY
;
1896 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1897 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1898 TASK_UNINTERRUPTIBLE
);
1899 spin_unlock(&inode
->i_lock
);
1903 * shmem_falloc_waitq points into the shmem_fallocate()
1904 * stack of the hole-punching task: shmem_falloc_waitq
1905 * is usually invalid by the time we reach here, but
1906 * finish_wait() does not dereference it in that case;
1907 * though i_lock needed lest racing with wake_up_all().
1909 spin_lock(&inode
->i_lock
);
1910 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1911 spin_unlock(&inode
->i_lock
);
1914 spin_unlock(&inode
->i_lock
);
1918 if (vma
->vm_flags
& VM_HUGEPAGE
)
1920 else if (vma
->vm_flags
& VM_NOHUGEPAGE
)
1923 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
1924 gfp
, vma
->vm_mm
, &ret
);
1926 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1930 unsigned long shmem_get_unmapped_area(struct file
*file
,
1931 unsigned long uaddr
, unsigned long len
,
1932 unsigned long pgoff
, unsigned long flags
)
1934 unsigned long (*get_area
)(struct file
*,
1935 unsigned long, unsigned long, unsigned long, unsigned long);
1937 unsigned long offset
;
1938 unsigned long inflated_len
;
1939 unsigned long inflated_addr
;
1940 unsigned long inflated_offset
;
1942 if (len
> TASK_SIZE
)
1945 get_area
= current
->mm
->get_unmapped_area
;
1946 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
1948 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1950 if (IS_ERR_VALUE(addr
))
1952 if (addr
& ~PAGE_MASK
)
1954 if (addr
> TASK_SIZE
- len
)
1957 if (shmem_huge
== SHMEM_HUGE_DENY
)
1959 if (len
< HPAGE_PMD_SIZE
)
1961 if (flags
& MAP_FIXED
)
1964 * Our priority is to support MAP_SHARED mapped hugely;
1965 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1966 * But if caller specified an address hint, respect that as before.
1971 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
1972 struct super_block
*sb
;
1975 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
1976 sb
= file_inode(file
)->i_sb
;
1979 * Called directly from mm/mmap.c, or drivers/char/mem.c
1980 * for "/dev/zero", to create a shared anonymous object.
1982 if (IS_ERR(shm_mnt
))
1984 sb
= shm_mnt
->mnt_sb
;
1986 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
1990 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
1991 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
1993 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
1996 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
1997 if (inflated_len
> TASK_SIZE
)
1999 if (inflated_len
< len
)
2002 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
2003 if (IS_ERR_VALUE(inflated_addr
))
2005 if (inflated_addr
& ~PAGE_MASK
)
2008 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2009 inflated_addr
+= offset
- inflated_offset
;
2010 if (inflated_offset
> offset
)
2011 inflated_addr
+= HPAGE_PMD_SIZE
;
2013 if (inflated_addr
> TASK_SIZE
- len
)
2015 return inflated_addr
;
2019 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2021 struct inode
*inode
= file_inode(vma
->vm_file
);
2022 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2025 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2028 struct inode
*inode
= file_inode(vma
->vm_file
);
2031 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2032 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2036 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2038 struct inode
*inode
= file_inode(file
);
2039 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2040 int retval
= -ENOMEM
;
2042 spin_lock_irq(&info
->lock
);
2043 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2044 if (!user_shm_lock(inode
->i_size
, user
))
2046 info
->flags
|= VM_LOCKED
;
2047 mapping_set_unevictable(file
->f_mapping
);
2049 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2050 user_shm_unlock(inode
->i_size
, user
);
2051 info
->flags
&= ~VM_LOCKED
;
2052 mapping_clear_unevictable(file
->f_mapping
);
2057 spin_unlock_irq(&info
->lock
);
2061 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2063 file_accessed(file
);
2064 vma
->vm_ops
= &shmem_vm_ops
;
2065 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2066 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2067 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2068 khugepaged_enter(vma
, vma
->vm_flags
);
2073 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2074 umode_t mode
, dev_t dev
, unsigned long flags
)
2076 struct inode
*inode
;
2077 struct shmem_inode_info
*info
;
2078 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2080 if (shmem_reserve_inode(sb
))
2083 inode
= new_inode(sb
);
2085 inode
->i_ino
= get_next_ino();
2086 inode_init_owner(inode
, dir
, mode
);
2087 inode
->i_blocks
= 0;
2088 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2089 inode
->i_generation
= get_seconds();
2090 info
= SHMEM_I(inode
);
2091 memset(info
, 0, (char *)inode
- (char *)info
);
2092 spin_lock_init(&info
->lock
);
2093 info
->seals
= F_SEAL_SEAL
;
2094 info
->flags
= flags
& VM_NORESERVE
;
2095 INIT_LIST_HEAD(&info
->shrinklist
);
2096 INIT_LIST_HEAD(&info
->swaplist
);
2097 simple_xattrs_init(&info
->xattrs
);
2098 cache_no_acl(inode
);
2100 switch (mode
& S_IFMT
) {
2102 inode
->i_op
= &shmem_special_inode_operations
;
2103 init_special_inode(inode
, mode
, dev
);
2106 inode
->i_mapping
->a_ops
= &shmem_aops
;
2107 inode
->i_op
= &shmem_inode_operations
;
2108 inode
->i_fop
= &shmem_file_operations
;
2109 mpol_shared_policy_init(&info
->policy
,
2110 shmem_get_sbmpol(sbinfo
));
2114 /* Some things misbehave if size == 0 on a directory */
2115 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2116 inode
->i_op
= &shmem_dir_inode_operations
;
2117 inode
->i_fop
= &simple_dir_operations
;
2121 * Must not load anything in the rbtree,
2122 * mpol_free_shared_policy will not be called.
2124 mpol_shared_policy_init(&info
->policy
, NULL
);
2128 shmem_free_inode(sb
);
2132 bool shmem_mapping(struct address_space
*mapping
)
2137 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
2141 static const struct inode_operations shmem_symlink_inode_operations
;
2142 static const struct inode_operations shmem_short_symlink_operations
;
2144 #ifdef CONFIG_TMPFS_XATTR
2145 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2147 #define shmem_initxattrs NULL
2151 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2152 loff_t pos
, unsigned len
, unsigned flags
,
2153 struct page
**pagep
, void **fsdata
)
2155 struct inode
*inode
= mapping
->host
;
2156 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2157 pgoff_t index
= pos
>> PAGE_SHIFT
;
2159 /* i_mutex is held by caller */
2160 if (unlikely(info
->seals
)) {
2161 if (info
->seals
& F_SEAL_WRITE
)
2163 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2167 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2171 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2172 loff_t pos
, unsigned len
, unsigned copied
,
2173 struct page
*page
, void *fsdata
)
2175 struct inode
*inode
= mapping
->host
;
2177 if (pos
+ copied
> inode
->i_size
)
2178 i_size_write(inode
, pos
+ copied
);
2180 if (!PageUptodate(page
)) {
2181 struct page
*head
= compound_head(page
);
2182 if (PageTransCompound(page
)) {
2185 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2186 if (head
+ i
== page
)
2188 clear_highpage(head
+ i
);
2189 flush_dcache_page(head
+ i
);
2192 if (copied
< PAGE_SIZE
) {
2193 unsigned from
= pos
& (PAGE_SIZE
- 1);
2194 zero_user_segments(page
, 0, from
,
2195 from
+ copied
, PAGE_SIZE
);
2197 SetPageUptodate(head
);
2199 set_page_dirty(page
);
2206 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2208 struct file
*file
= iocb
->ki_filp
;
2209 struct inode
*inode
= file_inode(file
);
2210 struct address_space
*mapping
= inode
->i_mapping
;
2212 unsigned long offset
;
2213 enum sgp_type sgp
= SGP_READ
;
2216 loff_t
*ppos
= &iocb
->ki_pos
;
2219 * Might this read be for a stacking filesystem? Then when reading
2220 * holes of a sparse file, we actually need to allocate those pages,
2221 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2223 if (!iter_is_iovec(to
))
2226 index
= *ppos
>> PAGE_SHIFT
;
2227 offset
= *ppos
& ~PAGE_MASK
;
2230 struct page
*page
= NULL
;
2232 unsigned long nr
, ret
;
2233 loff_t i_size
= i_size_read(inode
);
2235 end_index
= i_size
>> PAGE_SHIFT
;
2236 if (index
> end_index
)
2238 if (index
== end_index
) {
2239 nr
= i_size
& ~PAGE_MASK
;
2244 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2246 if (error
== -EINVAL
)
2251 if (sgp
== SGP_CACHE
)
2252 set_page_dirty(page
);
2257 * We must evaluate after, since reads (unlike writes)
2258 * are called without i_mutex protection against truncate
2261 i_size
= i_size_read(inode
);
2262 end_index
= i_size
>> PAGE_SHIFT
;
2263 if (index
== end_index
) {
2264 nr
= i_size
& ~PAGE_MASK
;
2275 * If users can be writing to this page using arbitrary
2276 * virtual addresses, take care about potential aliasing
2277 * before reading the page on the kernel side.
2279 if (mapping_writably_mapped(mapping
))
2280 flush_dcache_page(page
);
2282 * Mark the page accessed if we read the beginning.
2285 mark_page_accessed(page
);
2287 page
= ZERO_PAGE(0);
2292 * Ok, we have the page, and it's up-to-date, so
2293 * now we can copy it to user space...
2295 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2298 index
+= offset
>> PAGE_SHIFT
;
2299 offset
&= ~PAGE_MASK
;
2302 if (!iov_iter_count(to
))
2311 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2312 file_accessed(file
);
2313 return retval
? retval
: error
;
2317 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2319 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2320 pgoff_t index
, pgoff_t end
, int whence
)
2323 struct pagevec pvec
;
2324 pgoff_t indices
[PAGEVEC_SIZE
];
2328 pagevec_init(&pvec
, 0);
2329 pvec
.nr
= 1; /* start small: we may be there already */
2331 pvec
.nr
= find_get_entries(mapping
, index
,
2332 pvec
.nr
, pvec
.pages
, indices
);
2334 if (whence
== SEEK_DATA
)
2338 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2339 if (index
< indices
[i
]) {
2340 if (whence
== SEEK_HOLE
) {
2346 page
= pvec
.pages
[i
];
2347 if (page
&& !radix_tree_exceptional_entry(page
)) {
2348 if (!PageUptodate(page
))
2352 (page
&& whence
== SEEK_DATA
) ||
2353 (!page
&& whence
== SEEK_HOLE
)) {
2358 pagevec_remove_exceptionals(&pvec
);
2359 pagevec_release(&pvec
);
2360 pvec
.nr
= PAGEVEC_SIZE
;
2366 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2368 struct address_space
*mapping
= file
->f_mapping
;
2369 struct inode
*inode
= mapping
->host
;
2373 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2374 return generic_file_llseek_size(file
, offset
, whence
,
2375 MAX_LFS_FILESIZE
, i_size_read(inode
));
2377 /* We're holding i_mutex so we can access i_size directly */
2381 else if (offset
>= inode
->i_size
)
2384 start
= offset
>> PAGE_SHIFT
;
2385 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2386 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2387 new_offset
<<= PAGE_SHIFT
;
2388 if (new_offset
> offset
) {
2389 if (new_offset
< inode
->i_size
)
2390 offset
= new_offset
;
2391 else if (whence
== SEEK_DATA
)
2394 offset
= inode
->i_size
;
2399 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2400 inode_unlock(inode
);
2405 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2406 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2408 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2409 #define LAST_SCAN 4 /* about 150ms max */
2411 static void shmem_tag_pins(struct address_space
*mapping
)
2413 struct radix_tree_iter iter
;
2422 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2423 page
= radix_tree_deref_slot(slot
);
2424 if (!page
|| radix_tree_exception(page
)) {
2425 if (radix_tree_deref_retry(page
)) {
2426 slot
= radix_tree_iter_retry(&iter
);
2429 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2430 spin_lock_irq(&mapping
->tree_lock
);
2431 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2433 spin_unlock_irq(&mapping
->tree_lock
);
2436 if (need_resched()) {
2438 slot
= radix_tree_iter_next(&iter
);
2445 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2446 * via get_user_pages(), drivers might have some pending I/O without any active
2447 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2448 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2449 * them to be dropped.
2450 * The caller must guarantee that no new user will acquire writable references
2451 * to those pages to avoid races.
2453 static int shmem_wait_for_pins(struct address_space
*mapping
)
2455 struct radix_tree_iter iter
;
2461 shmem_tag_pins(mapping
);
2464 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2465 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2469 lru_add_drain_all();
2470 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2475 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2476 start
, SHMEM_TAG_PINNED
) {
2478 page
= radix_tree_deref_slot(slot
);
2479 if (radix_tree_exception(page
)) {
2480 if (radix_tree_deref_retry(page
)) {
2481 slot
= radix_tree_iter_retry(&iter
);
2489 page_count(page
) - page_mapcount(page
) != 1) {
2490 if (scan
< LAST_SCAN
)
2491 goto continue_resched
;
2494 * On the last scan, we clean up all those tags
2495 * we inserted; but make a note that we still
2496 * found pages pinned.
2501 spin_lock_irq(&mapping
->tree_lock
);
2502 radix_tree_tag_clear(&mapping
->page_tree
,
2503 iter
.index
, SHMEM_TAG_PINNED
);
2504 spin_unlock_irq(&mapping
->tree_lock
);
2506 if (need_resched()) {
2508 slot
= radix_tree_iter_next(&iter
);
2517 #define F_ALL_SEALS (F_SEAL_SEAL | \
2522 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2524 struct inode
*inode
= file_inode(file
);
2525 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2530 * Sealing allows multiple parties to share a shmem-file but restrict
2531 * access to a specific subset of file operations. Seals can only be
2532 * added, but never removed. This way, mutually untrusted parties can
2533 * share common memory regions with a well-defined policy. A malicious
2534 * peer can thus never perform unwanted operations on a shared object.
2536 * Seals are only supported on special shmem-files and always affect
2537 * the whole underlying inode. Once a seal is set, it may prevent some
2538 * kinds of access to the file. Currently, the following seals are
2540 * SEAL_SEAL: Prevent further seals from being set on this file
2541 * SEAL_SHRINK: Prevent the file from shrinking
2542 * SEAL_GROW: Prevent the file from growing
2543 * SEAL_WRITE: Prevent write access to the file
2545 * As we don't require any trust relationship between two parties, we
2546 * must prevent seals from being removed. Therefore, sealing a file
2547 * only adds a given set of seals to the file, it never touches
2548 * existing seals. Furthermore, the "setting seals"-operation can be
2549 * sealed itself, which basically prevents any further seal from being
2552 * Semantics of sealing are only defined on volatile files. Only
2553 * anonymous shmem files support sealing. More importantly, seals are
2554 * never written to disk. Therefore, there's no plan to support it on
2558 if (file
->f_op
!= &shmem_file_operations
)
2560 if (!(file
->f_mode
& FMODE_WRITE
))
2562 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2567 if (info
->seals
& F_SEAL_SEAL
) {
2572 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2573 error
= mapping_deny_writable(file
->f_mapping
);
2577 error
= shmem_wait_for_pins(file
->f_mapping
);
2579 mapping_allow_writable(file
->f_mapping
);
2584 info
->seals
|= seals
;
2588 inode_unlock(inode
);
2591 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2593 int shmem_get_seals(struct file
*file
)
2595 if (file
->f_op
!= &shmem_file_operations
)
2598 return SHMEM_I(file_inode(file
))->seals
;
2600 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2602 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2608 /* disallow upper 32bit */
2612 error
= shmem_add_seals(file
, arg
);
2615 error
= shmem_get_seals(file
);
2625 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2628 struct inode
*inode
= file_inode(file
);
2629 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2630 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2631 struct shmem_falloc shmem_falloc
;
2632 pgoff_t start
, index
, end
;
2635 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2640 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2641 struct address_space
*mapping
= file
->f_mapping
;
2642 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2643 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2644 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2646 /* protected by i_mutex */
2647 if (info
->seals
& F_SEAL_WRITE
) {
2652 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2653 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2654 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2655 spin_lock(&inode
->i_lock
);
2656 inode
->i_private
= &shmem_falloc
;
2657 spin_unlock(&inode
->i_lock
);
2659 if ((u64
)unmap_end
> (u64
)unmap_start
)
2660 unmap_mapping_range(mapping
, unmap_start
,
2661 1 + unmap_end
- unmap_start
, 0);
2662 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2663 /* No need to unmap again: hole-punching leaves COWed pages */
2665 spin_lock(&inode
->i_lock
);
2666 inode
->i_private
= NULL
;
2667 wake_up_all(&shmem_falloc_waitq
);
2668 spin_unlock(&inode
->i_lock
);
2673 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2674 error
= inode_newsize_ok(inode
, offset
+ len
);
2678 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2683 start
= offset
>> PAGE_SHIFT
;
2684 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2685 /* Try to avoid a swapstorm if len is impossible to satisfy */
2686 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2691 shmem_falloc
.waitq
= NULL
;
2692 shmem_falloc
.start
= start
;
2693 shmem_falloc
.next
= start
;
2694 shmem_falloc
.nr_falloced
= 0;
2695 shmem_falloc
.nr_unswapped
= 0;
2696 spin_lock(&inode
->i_lock
);
2697 inode
->i_private
= &shmem_falloc
;
2698 spin_unlock(&inode
->i_lock
);
2700 for (index
= start
; index
< end
; index
++) {
2704 * Good, the fallocate(2) manpage permits EINTR: we may have
2705 * been interrupted because we are using up too much memory.
2707 if (signal_pending(current
))
2709 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2712 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2714 /* Remove the !PageUptodate pages we added */
2715 if (index
> start
) {
2716 shmem_undo_range(inode
,
2717 (loff_t
)start
<< PAGE_SHIFT
,
2718 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2724 * Inform shmem_writepage() how far we have reached.
2725 * No need for lock or barrier: we have the page lock.
2727 shmem_falloc
.next
++;
2728 if (!PageUptodate(page
))
2729 shmem_falloc
.nr_falloced
++;
2732 * If !PageUptodate, leave it that way so that freeable pages
2733 * can be recognized if we need to rollback on error later.
2734 * But set_page_dirty so that memory pressure will swap rather
2735 * than free the pages we are allocating (and SGP_CACHE pages
2736 * might still be clean: we now need to mark those dirty too).
2738 set_page_dirty(page
);
2744 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2745 i_size_write(inode
, offset
+ len
);
2746 inode
->i_ctime
= current_time(inode
);
2748 spin_lock(&inode
->i_lock
);
2749 inode
->i_private
= NULL
;
2750 spin_unlock(&inode
->i_lock
);
2752 inode_unlock(inode
);
2756 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2758 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2760 buf
->f_type
= TMPFS_MAGIC
;
2761 buf
->f_bsize
= PAGE_SIZE
;
2762 buf
->f_namelen
= NAME_MAX
;
2763 if (sbinfo
->max_blocks
) {
2764 buf
->f_blocks
= sbinfo
->max_blocks
;
2766 buf
->f_bfree
= sbinfo
->max_blocks
-
2767 percpu_counter_sum(&sbinfo
->used_blocks
);
2769 if (sbinfo
->max_inodes
) {
2770 buf
->f_files
= sbinfo
->max_inodes
;
2771 buf
->f_ffree
= sbinfo
->free_inodes
;
2773 /* else leave those fields 0 like simple_statfs */
2778 * File creation. Allocate an inode, and we're done..
2781 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2783 struct inode
*inode
;
2784 int error
= -ENOSPC
;
2786 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2788 error
= simple_acl_create(dir
, inode
);
2791 error
= security_inode_init_security(inode
, dir
,
2793 shmem_initxattrs
, NULL
);
2794 if (error
&& error
!= -EOPNOTSUPP
)
2798 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2799 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2800 d_instantiate(dentry
, inode
);
2801 dget(dentry
); /* Extra count - pin the dentry in core */
2810 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2812 struct inode
*inode
;
2813 int error
= -ENOSPC
;
2815 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2817 error
= security_inode_init_security(inode
, dir
,
2819 shmem_initxattrs
, NULL
);
2820 if (error
&& error
!= -EOPNOTSUPP
)
2822 error
= simple_acl_create(dir
, inode
);
2825 d_tmpfile(dentry
, inode
);
2833 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2837 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2843 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2846 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2852 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2854 struct inode
*inode
= d_inode(old_dentry
);
2858 * No ordinary (disk based) filesystem counts links as inodes;
2859 * but each new link needs a new dentry, pinning lowmem, and
2860 * tmpfs dentries cannot be pruned until they are unlinked.
2862 ret
= shmem_reserve_inode(inode
->i_sb
);
2866 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2867 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2869 ihold(inode
); /* New dentry reference */
2870 dget(dentry
); /* Extra pinning count for the created dentry */
2871 d_instantiate(dentry
, inode
);
2876 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2878 struct inode
*inode
= d_inode(dentry
);
2880 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2881 shmem_free_inode(inode
->i_sb
);
2883 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2884 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2886 dput(dentry
); /* Undo the count from "create" - this does all the work */
2890 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2892 if (!simple_empty(dentry
))
2895 drop_nlink(d_inode(dentry
));
2897 return shmem_unlink(dir
, dentry
);
2900 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2902 bool old_is_dir
= d_is_dir(old_dentry
);
2903 bool new_is_dir
= d_is_dir(new_dentry
);
2905 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2907 drop_nlink(old_dir
);
2910 drop_nlink(new_dir
);
2914 old_dir
->i_ctime
= old_dir
->i_mtime
=
2915 new_dir
->i_ctime
= new_dir
->i_mtime
=
2916 d_inode(old_dentry
)->i_ctime
=
2917 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
2922 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2924 struct dentry
*whiteout
;
2927 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2931 error
= shmem_mknod(old_dir
, whiteout
,
2932 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2938 * Cheat and hash the whiteout while the old dentry is still in
2939 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2941 * d_lookup() will consistently find one of them at this point,
2942 * not sure which one, but that isn't even important.
2949 * The VFS layer already does all the dentry stuff for rename,
2950 * we just have to decrement the usage count for the target if
2951 * it exists so that the VFS layer correctly free's it when it
2954 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2956 struct inode
*inode
= d_inode(old_dentry
);
2957 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2959 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2962 if (flags
& RENAME_EXCHANGE
)
2963 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2965 if (!simple_empty(new_dentry
))
2968 if (flags
& RENAME_WHITEOUT
) {
2971 error
= shmem_whiteout(old_dir
, old_dentry
);
2976 if (d_really_is_positive(new_dentry
)) {
2977 (void) shmem_unlink(new_dir
, new_dentry
);
2978 if (they_are_dirs
) {
2979 drop_nlink(d_inode(new_dentry
));
2980 drop_nlink(old_dir
);
2982 } else if (they_are_dirs
) {
2983 drop_nlink(old_dir
);
2987 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2988 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2989 old_dir
->i_ctime
= old_dir
->i_mtime
=
2990 new_dir
->i_ctime
= new_dir
->i_mtime
=
2991 inode
->i_ctime
= current_time(old_dir
);
2995 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2999 struct inode
*inode
;
3001 struct shmem_inode_info
*info
;
3003 len
= strlen(symname
) + 1;
3004 if (len
> PAGE_SIZE
)
3005 return -ENAMETOOLONG
;
3007 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
3011 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3012 shmem_initxattrs
, NULL
);
3014 if (error
!= -EOPNOTSUPP
) {
3021 info
= SHMEM_I(inode
);
3022 inode
->i_size
= len
-1;
3023 if (len
<= SHORT_SYMLINK_LEN
) {
3024 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3025 if (!inode
->i_link
) {
3029 inode
->i_op
= &shmem_short_symlink_operations
;
3031 inode_nohighmem(inode
);
3032 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3037 inode
->i_mapping
->a_ops
= &shmem_aops
;
3038 inode
->i_op
= &shmem_symlink_inode_operations
;
3039 memcpy(page_address(page
), symname
, len
);
3040 SetPageUptodate(page
);
3041 set_page_dirty(page
);
3045 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3046 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3047 d_instantiate(dentry
, inode
);
3052 static void shmem_put_link(void *arg
)
3054 mark_page_accessed(arg
);
3058 static const char *shmem_get_link(struct dentry
*dentry
,
3059 struct inode
*inode
,
3060 struct delayed_call
*done
)
3062 struct page
*page
= NULL
;
3065 page
= find_get_page(inode
->i_mapping
, 0);
3067 return ERR_PTR(-ECHILD
);
3068 if (!PageUptodate(page
)) {
3070 return ERR_PTR(-ECHILD
);
3073 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3075 return ERR_PTR(error
);
3078 set_delayed_call(done
, shmem_put_link
, page
);
3079 return page_address(page
);
3082 #ifdef CONFIG_TMPFS_XATTR
3084 * Superblocks without xattr inode operations may get some security.* xattr
3085 * support from the LSM "for free". As soon as we have any other xattrs
3086 * like ACLs, we also need to implement the security.* handlers at
3087 * filesystem level, though.
3091 * Callback for security_inode_init_security() for acquiring xattrs.
3093 static int shmem_initxattrs(struct inode
*inode
,
3094 const struct xattr
*xattr_array
,
3097 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3098 const struct xattr
*xattr
;
3099 struct simple_xattr
*new_xattr
;
3102 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3103 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3107 len
= strlen(xattr
->name
) + 1;
3108 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3110 if (!new_xattr
->name
) {
3115 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3116 XATTR_SECURITY_PREFIX_LEN
);
3117 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3120 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3126 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3127 struct dentry
*unused
, struct inode
*inode
,
3128 const char *name
, void *buffer
, size_t size
)
3130 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3132 name
= xattr_full_name(handler
, name
);
3133 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3136 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3137 struct dentry
*unused
, struct inode
*inode
,
3138 const char *name
, const void *value
,
3139 size_t size
, int flags
)
3141 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3143 name
= xattr_full_name(handler
, name
);
3144 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3147 static const struct xattr_handler shmem_security_xattr_handler
= {
3148 .prefix
= XATTR_SECURITY_PREFIX
,
3149 .get
= shmem_xattr_handler_get
,
3150 .set
= shmem_xattr_handler_set
,
3153 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3154 .prefix
= XATTR_TRUSTED_PREFIX
,
3155 .get
= shmem_xattr_handler_get
,
3156 .set
= shmem_xattr_handler_set
,
3159 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3160 #ifdef CONFIG_TMPFS_POSIX_ACL
3161 &posix_acl_access_xattr_handler
,
3162 &posix_acl_default_xattr_handler
,
3164 &shmem_security_xattr_handler
,
3165 &shmem_trusted_xattr_handler
,
3169 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3171 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3172 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3174 #endif /* CONFIG_TMPFS_XATTR */
3176 static const struct inode_operations shmem_short_symlink_operations
= {
3177 .readlink
= generic_readlink
,
3178 .get_link
= simple_get_link
,
3179 #ifdef CONFIG_TMPFS_XATTR
3180 .listxattr
= shmem_listxattr
,
3184 static const struct inode_operations shmem_symlink_inode_operations
= {
3185 .readlink
= generic_readlink
,
3186 .get_link
= shmem_get_link
,
3187 #ifdef CONFIG_TMPFS_XATTR
3188 .listxattr
= shmem_listxattr
,
3192 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3194 return ERR_PTR(-ESTALE
);
3197 static int shmem_match(struct inode
*ino
, void *vfh
)
3201 inum
= (inum
<< 32) | fh
[1];
3202 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3205 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3206 struct fid
*fid
, int fh_len
, int fh_type
)
3208 struct inode
*inode
;
3209 struct dentry
*dentry
= NULL
;
3216 inum
= (inum
<< 32) | fid
->raw
[1];
3218 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3219 shmem_match
, fid
->raw
);
3221 dentry
= d_find_alias(inode
);
3228 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3229 struct inode
*parent
)
3233 return FILEID_INVALID
;
3236 if (inode_unhashed(inode
)) {
3237 /* Unfortunately insert_inode_hash is not idempotent,
3238 * so as we hash inodes here rather than at creation
3239 * time, we need a lock to ensure we only try
3242 static DEFINE_SPINLOCK(lock
);
3244 if (inode_unhashed(inode
))
3245 __insert_inode_hash(inode
,
3246 inode
->i_ino
+ inode
->i_generation
);
3250 fh
[0] = inode
->i_generation
;
3251 fh
[1] = inode
->i_ino
;
3252 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3258 static const struct export_operations shmem_export_ops
= {
3259 .get_parent
= shmem_get_parent
,
3260 .encode_fh
= shmem_encode_fh
,
3261 .fh_to_dentry
= shmem_fh_to_dentry
,
3264 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3267 char *this_char
, *value
, *rest
;
3268 struct mempolicy
*mpol
= NULL
;
3272 while (options
!= NULL
) {
3273 this_char
= options
;
3276 * NUL-terminate this option: unfortunately,
3277 * mount options form a comma-separated list,
3278 * but mpol's nodelist may also contain commas.
3280 options
= strchr(options
, ',');
3281 if (options
== NULL
)
3284 if (!isdigit(*options
)) {
3291 if ((value
= strchr(this_char
,'=')) != NULL
) {
3294 pr_err("tmpfs: No value for mount option '%s'\n",
3299 if (!strcmp(this_char
,"size")) {
3300 unsigned long long size
;
3301 size
= memparse(value
,&rest
);
3303 size
<<= PAGE_SHIFT
;
3304 size
*= totalram_pages
;
3310 sbinfo
->max_blocks
=
3311 DIV_ROUND_UP(size
, PAGE_SIZE
);
3312 } else if (!strcmp(this_char
,"nr_blocks")) {
3313 sbinfo
->max_blocks
= memparse(value
, &rest
);
3316 } else if (!strcmp(this_char
,"nr_inodes")) {
3317 sbinfo
->max_inodes
= memparse(value
, &rest
);
3320 } else if (!strcmp(this_char
,"mode")) {
3323 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3326 } else if (!strcmp(this_char
,"uid")) {
3329 uid
= simple_strtoul(value
, &rest
, 0);
3332 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3333 if (!uid_valid(sbinfo
->uid
))
3335 } else if (!strcmp(this_char
,"gid")) {
3338 gid
= simple_strtoul(value
, &rest
, 0);
3341 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3342 if (!gid_valid(sbinfo
->gid
))
3344 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3345 } else if (!strcmp(this_char
, "huge")) {
3347 huge
= shmem_parse_huge(value
);
3350 if (!has_transparent_hugepage() &&
3351 huge
!= SHMEM_HUGE_NEVER
)
3353 sbinfo
->huge
= huge
;
3356 } else if (!strcmp(this_char
,"mpol")) {
3359 if (mpol_parse_str(value
, &mpol
))
3363 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3367 sbinfo
->mpol
= mpol
;
3371 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3379 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3381 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3382 struct shmem_sb_info config
= *sbinfo
;
3383 unsigned long inodes
;
3384 int error
= -EINVAL
;
3387 if (shmem_parse_options(data
, &config
, true))
3390 spin_lock(&sbinfo
->stat_lock
);
3391 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3392 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3394 if (config
.max_inodes
< inodes
)
3397 * Those tests disallow limited->unlimited while any are in use;
3398 * but we must separately disallow unlimited->limited, because
3399 * in that case we have no record of how much is already in use.
3401 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3403 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3407 sbinfo
->huge
= config
.huge
;
3408 sbinfo
->max_blocks
= config
.max_blocks
;
3409 sbinfo
->max_inodes
= config
.max_inodes
;
3410 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3413 * Preserve previous mempolicy unless mpol remount option was specified.
3416 mpol_put(sbinfo
->mpol
);
3417 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3420 spin_unlock(&sbinfo
->stat_lock
);
3424 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3426 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3428 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3429 seq_printf(seq
, ",size=%luk",
3430 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3431 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3432 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3433 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3434 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3435 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3436 seq_printf(seq
, ",uid=%u",
3437 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3438 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3439 seq_printf(seq
, ",gid=%u",
3440 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3441 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3442 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3444 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3446 shmem_show_mpol(seq
, sbinfo
->mpol
);
3450 #define MFD_NAME_PREFIX "memfd:"
3451 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3452 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3454 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3456 SYSCALL_DEFINE2(memfd_create
,
3457 const char __user
*, uname
,
3458 unsigned int, flags
)
3460 struct shmem_inode_info
*info
;
3466 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3469 /* length includes terminating zero */
3470 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3473 if (len
> MFD_NAME_MAX_LEN
+ 1)
3476 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
3480 strcpy(name
, MFD_NAME_PREFIX
);
3481 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3486 /* terminating-zero may have changed after strnlen_user() returned */
3487 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3492 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3498 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3500 error
= PTR_ERR(file
);
3503 info
= SHMEM_I(file_inode(file
));
3504 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3505 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3506 if (flags
& MFD_ALLOW_SEALING
)
3507 info
->seals
&= ~F_SEAL_SEAL
;
3509 fd_install(fd
, file
);
3520 #endif /* CONFIG_TMPFS */
3522 static void shmem_put_super(struct super_block
*sb
)
3524 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3526 percpu_counter_destroy(&sbinfo
->used_blocks
);
3527 mpol_put(sbinfo
->mpol
);
3529 sb
->s_fs_info
= NULL
;
3532 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3534 struct inode
*inode
;
3535 struct shmem_sb_info
*sbinfo
;
3538 /* Round up to L1_CACHE_BYTES to resist false sharing */
3539 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3540 L1_CACHE_BYTES
), GFP_KERNEL
);
3544 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3545 sbinfo
->uid
= current_fsuid();
3546 sbinfo
->gid
= current_fsgid();
3547 sb
->s_fs_info
= sbinfo
;
3551 * Per default we only allow half of the physical ram per
3552 * tmpfs instance, limiting inodes to one per page of lowmem;
3553 * but the internal instance is left unlimited.
3555 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3556 sbinfo
->max_blocks
= shmem_default_max_blocks();
3557 sbinfo
->max_inodes
= shmem_default_max_inodes();
3558 if (shmem_parse_options(data
, sbinfo
, false)) {
3563 sb
->s_flags
|= MS_NOUSER
;
3565 sb
->s_export_op
= &shmem_export_ops
;
3566 sb
->s_flags
|= MS_NOSEC
;
3568 sb
->s_flags
|= MS_NOUSER
;
3571 spin_lock_init(&sbinfo
->stat_lock
);
3572 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3574 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3575 spin_lock_init(&sbinfo
->shrinklist_lock
);
3576 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3578 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3579 sb
->s_blocksize
= PAGE_SIZE
;
3580 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3581 sb
->s_magic
= TMPFS_MAGIC
;
3582 sb
->s_op
= &shmem_ops
;
3583 sb
->s_time_gran
= 1;
3584 #ifdef CONFIG_TMPFS_XATTR
3585 sb
->s_xattr
= shmem_xattr_handlers
;
3587 #ifdef CONFIG_TMPFS_POSIX_ACL
3588 sb
->s_flags
|= MS_POSIXACL
;
3591 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3594 inode
->i_uid
= sbinfo
->uid
;
3595 inode
->i_gid
= sbinfo
->gid
;
3596 sb
->s_root
= d_make_root(inode
);
3602 shmem_put_super(sb
);
3606 static struct kmem_cache
*shmem_inode_cachep
;
3608 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3610 struct shmem_inode_info
*info
;
3611 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3614 return &info
->vfs_inode
;
3617 static void shmem_destroy_callback(struct rcu_head
*head
)
3619 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3620 if (S_ISLNK(inode
->i_mode
))
3621 kfree(inode
->i_link
);
3622 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3625 static void shmem_destroy_inode(struct inode
*inode
)
3627 if (S_ISREG(inode
->i_mode
))
3628 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3629 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3632 static void shmem_init_inode(void *foo
)
3634 struct shmem_inode_info
*info
= foo
;
3635 inode_init_once(&info
->vfs_inode
);
3638 static int shmem_init_inodecache(void)
3640 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3641 sizeof(struct shmem_inode_info
),
3642 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3646 static void shmem_destroy_inodecache(void)
3648 kmem_cache_destroy(shmem_inode_cachep
);
3651 static const struct address_space_operations shmem_aops
= {
3652 .writepage
= shmem_writepage
,
3653 .set_page_dirty
= __set_page_dirty_no_writeback
,
3655 .write_begin
= shmem_write_begin
,
3656 .write_end
= shmem_write_end
,
3658 #ifdef CONFIG_MIGRATION
3659 .migratepage
= migrate_page
,
3661 .error_remove_page
= generic_error_remove_page
,
3664 static const struct file_operations shmem_file_operations
= {
3666 .get_unmapped_area
= shmem_get_unmapped_area
,
3668 .llseek
= shmem_file_llseek
,
3669 .read_iter
= shmem_file_read_iter
,
3670 .write_iter
= generic_file_write_iter
,
3671 .fsync
= noop_fsync
,
3672 .splice_read
= generic_file_splice_read
,
3673 .splice_write
= iter_file_splice_write
,
3674 .fallocate
= shmem_fallocate
,
3678 static const struct inode_operations shmem_inode_operations
= {
3679 .getattr
= shmem_getattr
,
3680 .setattr
= shmem_setattr
,
3681 #ifdef CONFIG_TMPFS_XATTR
3682 .listxattr
= shmem_listxattr
,
3683 .set_acl
= simple_set_acl
,
3687 static const struct inode_operations shmem_dir_inode_operations
= {
3689 .create
= shmem_create
,
3690 .lookup
= simple_lookup
,
3692 .unlink
= shmem_unlink
,
3693 .symlink
= shmem_symlink
,
3694 .mkdir
= shmem_mkdir
,
3695 .rmdir
= shmem_rmdir
,
3696 .mknod
= shmem_mknod
,
3697 .rename
= shmem_rename2
,
3698 .tmpfile
= shmem_tmpfile
,
3700 #ifdef CONFIG_TMPFS_XATTR
3701 .listxattr
= shmem_listxattr
,
3703 #ifdef CONFIG_TMPFS_POSIX_ACL
3704 .setattr
= shmem_setattr
,
3705 .set_acl
= simple_set_acl
,
3709 static const struct inode_operations shmem_special_inode_operations
= {
3710 #ifdef CONFIG_TMPFS_XATTR
3711 .listxattr
= shmem_listxattr
,
3713 #ifdef CONFIG_TMPFS_POSIX_ACL
3714 .setattr
= shmem_setattr
,
3715 .set_acl
= simple_set_acl
,
3719 static const struct super_operations shmem_ops
= {
3720 .alloc_inode
= shmem_alloc_inode
,
3721 .destroy_inode
= shmem_destroy_inode
,
3723 .statfs
= shmem_statfs
,
3724 .remount_fs
= shmem_remount_fs
,
3725 .show_options
= shmem_show_options
,
3727 .evict_inode
= shmem_evict_inode
,
3728 .drop_inode
= generic_delete_inode
,
3729 .put_super
= shmem_put_super
,
3730 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3731 .nr_cached_objects
= shmem_unused_huge_count
,
3732 .free_cached_objects
= shmem_unused_huge_scan
,
3736 static const struct vm_operations_struct shmem_vm_ops
= {
3737 .fault
= shmem_fault
,
3738 .map_pages
= filemap_map_pages
,
3740 .set_policy
= shmem_set_policy
,
3741 .get_policy
= shmem_get_policy
,
3745 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3746 int flags
, const char *dev_name
, void *data
)
3748 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3751 static struct file_system_type shmem_fs_type
= {
3752 .owner
= THIS_MODULE
,
3754 .mount
= shmem_mount
,
3755 .kill_sb
= kill_litter_super
,
3756 .fs_flags
= FS_USERNS_MOUNT
,
3759 int __init
shmem_init(void)
3763 /* If rootfs called this, don't re-init */
3764 if (shmem_inode_cachep
)
3767 error
= shmem_init_inodecache();
3771 error
= register_filesystem(&shmem_fs_type
);
3773 pr_err("Could not register tmpfs\n");
3777 shm_mnt
= kern_mount(&shmem_fs_type
);
3778 if (IS_ERR(shm_mnt
)) {
3779 error
= PTR_ERR(shm_mnt
);
3780 pr_err("Could not kern_mount tmpfs\n");
3784 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3785 if (has_transparent_hugepage() && shmem_huge
< SHMEM_HUGE_DENY
)
3786 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3788 shmem_huge
= 0; /* just in case it was patched */
3793 unregister_filesystem(&shmem_fs_type
);
3795 shmem_destroy_inodecache();
3797 shm_mnt
= ERR_PTR(error
);
3801 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3802 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3803 struct kobj_attribute
*attr
, char *buf
)
3807 SHMEM_HUGE_WITHIN_SIZE
,
3815 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3816 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3818 count
+= sprintf(buf
+ count
, fmt
,
3819 shmem_format_huge(values
[i
]));
3821 buf
[count
- 1] = '\n';
3825 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3826 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3831 if (count
+ 1 > sizeof(tmp
))
3833 memcpy(tmp
, buf
, count
);
3835 if (count
&& tmp
[count
- 1] == '\n')
3836 tmp
[count
- 1] = '\0';
3838 huge
= shmem_parse_huge(tmp
);
3839 if (huge
== -EINVAL
)
3841 if (!has_transparent_hugepage() &&
3842 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3846 if (shmem_huge
< SHMEM_HUGE_DENY
)
3847 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3851 struct kobj_attribute shmem_enabled_attr
=
3852 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3853 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3855 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3856 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
3858 struct inode
*inode
= file_inode(vma
->vm_file
);
3859 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3863 if (shmem_huge
== SHMEM_HUGE_FORCE
)
3865 if (shmem_huge
== SHMEM_HUGE_DENY
)
3867 switch (sbinfo
->huge
) {
3868 case SHMEM_HUGE_NEVER
:
3870 case SHMEM_HUGE_ALWAYS
:
3872 case SHMEM_HUGE_WITHIN_SIZE
:
3873 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
3874 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
3875 if (i_size
>= HPAGE_PMD_SIZE
&&
3876 i_size
>> PAGE_SHIFT
>= off
)
3878 case SHMEM_HUGE_ADVISE
:
3879 /* TODO: implement fadvise() hints */
3880 return (vma
->vm_flags
& VM_HUGEPAGE
);
3886 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3888 #else /* !CONFIG_SHMEM */
3891 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3893 * This is intended for small system where the benefits of the full
3894 * shmem code (swap-backed and resource-limited) are outweighed by
3895 * their complexity. On systems without swap this code should be
3896 * effectively equivalent, but much lighter weight.
3899 static struct file_system_type shmem_fs_type
= {
3901 .mount
= ramfs_mount
,
3902 .kill_sb
= kill_litter_super
,
3903 .fs_flags
= FS_USERNS_MOUNT
,
3906 int __init
shmem_init(void)
3908 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3910 shm_mnt
= kern_mount(&shmem_fs_type
);
3911 BUG_ON(IS_ERR(shm_mnt
));
3916 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3921 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3926 void shmem_unlock_mapping(struct address_space
*mapping
)
3931 unsigned long shmem_get_unmapped_area(struct file
*file
,
3932 unsigned long addr
, unsigned long len
,
3933 unsigned long pgoff
, unsigned long flags
)
3935 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
3939 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3941 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3943 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3945 #define shmem_vm_ops generic_file_vm_ops
3946 #define shmem_file_operations ramfs_file_operations
3947 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3948 #define shmem_acct_size(flags, size) 0
3949 #define shmem_unacct_size(flags, size) do {} while (0)
3951 #endif /* CONFIG_SHMEM */
3955 static const struct dentry_operations anon_ops
= {
3956 .d_dname
= simple_dname
3959 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3960 unsigned long flags
, unsigned int i_flags
)
3963 struct inode
*inode
;
3965 struct super_block
*sb
;
3968 if (IS_ERR(shm_mnt
))
3969 return ERR_CAST(shm_mnt
);
3971 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3972 return ERR_PTR(-EINVAL
);
3974 if (shmem_acct_size(flags
, size
))
3975 return ERR_PTR(-ENOMEM
);
3977 res
= ERR_PTR(-ENOMEM
);
3979 this.len
= strlen(name
);
3980 this.hash
= 0; /* will go */
3981 sb
= shm_mnt
->mnt_sb
;
3982 path
.mnt
= mntget(shm_mnt
);
3983 path
.dentry
= d_alloc_pseudo(sb
, &this);
3986 d_set_d_op(path
.dentry
, &anon_ops
);
3988 res
= ERR_PTR(-ENOSPC
);
3989 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3993 inode
->i_flags
|= i_flags
;
3994 d_instantiate(path
.dentry
, inode
);
3995 inode
->i_size
= size
;
3996 clear_nlink(inode
); /* It is unlinked */
3997 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4001 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
4002 &shmem_file_operations
);
4009 shmem_unacct_size(flags
, size
);
4016 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4017 * kernel internal. There will be NO LSM permission checks against the
4018 * underlying inode. So users of this interface must do LSM checks at a
4019 * higher layer. The users are the big_key and shm implementations. LSM
4020 * checks are provided at the key or shm level rather than the inode.
4021 * @name: name for dentry (to be seen in /proc/<pid>/maps
4022 * @size: size to be set for the file
4023 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4025 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4027 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
4031 * shmem_file_setup - get an unlinked file living in tmpfs
4032 * @name: name for dentry (to be seen in /proc/<pid>/maps
4033 * @size: size to be set for the file
4034 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4036 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4038 return __shmem_file_setup(name
, size
, flags
, 0);
4040 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4043 * shmem_zero_setup - setup a shared anonymous mapping
4044 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4046 int shmem_zero_setup(struct vm_area_struct
*vma
)
4049 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4052 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4053 * between XFS directory reading and selinux: since this file is only
4054 * accessible to the user through its mapping, use S_PRIVATE flag to
4055 * bypass file security, in the same way as shmem_kernel_file_setup().
4057 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
4059 return PTR_ERR(file
);
4063 vma
->vm_file
= file
;
4064 vma
->vm_ops
= &shmem_vm_ops
;
4066 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4067 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4068 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4069 khugepaged_enter(vma
, vma
->vm_flags
);
4076 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4077 * @mapping: the page's address_space
4078 * @index: the page index
4079 * @gfp: the page allocator flags to use if allocating
4081 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4082 * with any new page allocations done using the specified allocation flags.
4083 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4084 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4085 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4087 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4088 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4090 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4091 pgoff_t index
, gfp_t gfp
)
4094 struct inode
*inode
= mapping
->host
;
4098 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4099 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4102 page
= ERR_PTR(error
);
4108 * The tiny !SHMEM case uses ramfs without swap
4110 return read_cache_page_gfp(mapping
, index
, gfp
);
4113 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);