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/sched/signal.h>
33 #include <linux/export.h>
34 #include <linux/swap.h>
35 #include <linux/uio.h>
36 #include <linux/khugepaged.h>
37 #include <linux/hugetlb.h>
39 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
41 static struct vfsmount
*shm_mnt
;
45 * This virtual memory filesystem is heavily based on the ramfs. It
46 * extends ramfs by the ability to use swap and honor resource limits
47 * which makes it a completely usable filesystem.
50 #include <linux/xattr.h>
51 #include <linux/exportfs.h>
52 #include <linux/posix_acl.h>
53 #include <linux/posix_acl_xattr.h>
54 #include <linux/mman.h>
55 #include <linux/string.h>
56 #include <linux/slab.h>
57 #include <linux/backing-dev.h>
58 #include <linux/shmem_fs.h>
59 #include <linux/writeback.h>
60 #include <linux/blkdev.h>
61 #include <linux/pagevec.h>
62 #include <linux/percpu_counter.h>
63 #include <linux/falloc.h>
64 #include <linux/splice.h>
65 #include <linux/security.h>
66 #include <linux/swapops.h>
67 #include <linux/mempolicy.h>
68 #include <linux/namei.h>
69 #include <linux/ctype.h>
70 #include <linux/migrate.h>
71 #include <linux/highmem.h>
72 #include <linux/seq_file.h>
73 #include <linux/magic.h>
74 #include <linux/syscalls.h>
75 #include <linux/fcntl.h>
76 #include <uapi/linux/memfd.h>
77 #include <linux/userfaultfd_k.h>
78 #include <linux/rmap.h>
79 #include <linux/uuid.h>
81 #include <linux/uaccess.h>
82 #include <asm/pgtable.h>
86 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 /* Pretend that each entry is of this size in directory's i_size */
90 #define BOGO_DIRENT_SIZE 20
92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93 #define SHORT_SYMLINK_LEN 128
96 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_mutex making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
100 struct shmem_falloc
{
101 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
102 pgoff_t start
; /* start of range currently being fallocated */
103 pgoff_t next
; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
109 static unsigned long shmem_default_max_blocks(void)
111 return totalram_pages
/ 2;
114 static int shmem_default_max_inodes(void)
119 ul
= min3(ul
, totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
124 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
125 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
126 struct shmem_inode_info
*info
, pgoff_t index
);
127 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
128 struct page
**pagep
, enum sgp_type sgp
,
129 gfp_t gfp
, struct vm_area_struct
*vma
,
130 struct vm_fault
*vmf
, int *fault_type
);
132 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
133 struct page
**pagep
, enum sgp_type sgp
)
135 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
136 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
139 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
141 return sb
->s_fs_info
;
145 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
146 * for shared memory and for shared anonymous (/dev/zero) mappings
147 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
148 * consistent with the pre-accounting of private mappings ...
150 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
152 return (flags
& VM_NORESERVE
) ?
153 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
156 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
158 if (!(flags
& VM_NORESERVE
))
159 vm_unacct_memory(VM_ACCT(size
));
162 static inline int shmem_reacct_size(unsigned long flags
,
163 loff_t oldsize
, loff_t newsize
)
165 if (!(flags
& VM_NORESERVE
)) {
166 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
167 return security_vm_enough_memory_mm(current
->mm
,
168 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
169 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
170 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
176 * ... whereas tmpfs objects are accounted incrementally as
177 * pages are allocated, in order to allow large sparse files.
178 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
179 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
181 static inline int shmem_acct_block(unsigned long flags
, long pages
)
183 if (!(flags
& VM_NORESERVE
))
186 return security_vm_enough_memory_mm(current
->mm
,
187 pages
* VM_ACCT(PAGE_SIZE
));
190 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
192 if (flags
& VM_NORESERVE
)
193 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
196 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
198 struct shmem_inode_info
*info
= SHMEM_I(inode
);
199 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
201 if (shmem_acct_block(info
->flags
, pages
))
204 if (sbinfo
->max_blocks
) {
205 if (percpu_counter_compare(&sbinfo
->used_blocks
,
206 sbinfo
->max_blocks
- pages
) > 0)
208 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
214 shmem_unacct_blocks(info
->flags
, pages
);
218 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
220 struct shmem_inode_info
*info
= SHMEM_I(inode
);
221 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
223 if (sbinfo
->max_blocks
)
224 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
225 shmem_unacct_blocks(info
->flags
, pages
);
228 static const struct super_operations shmem_ops
;
229 static const struct address_space_operations shmem_aops
;
230 static const struct file_operations shmem_file_operations
;
231 static const struct inode_operations shmem_inode_operations
;
232 static const struct inode_operations shmem_dir_inode_operations
;
233 static const struct inode_operations shmem_special_inode_operations
;
234 static const struct vm_operations_struct shmem_vm_ops
;
235 static struct file_system_type shmem_fs_type
;
237 bool vma_is_shmem(struct vm_area_struct
*vma
)
239 return vma
->vm_ops
== &shmem_vm_ops
;
242 static LIST_HEAD(shmem_swaplist
);
243 static DEFINE_MUTEX(shmem_swaplist_mutex
);
245 static int shmem_reserve_inode(struct super_block
*sb
)
247 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
248 if (sbinfo
->max_inodes
) {
249 spin_lock(&sbinfo
->stat_lock
);
250 if (!sbinfo
->free_inodes
) {
251 spin_unlock(&sbinfo
->stat_lock
);
254 sbinfo
->free_inodes
--;
255 spin_unlock(&sbinfo
->stat_lock
);
260 static void shmem_free_inode(struct super_block
*sb
)
262 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
263 if (sbinfo
->max_inodes
) {
264 spin_lock(&sbinfo
->stat_lock
);
265 sbinfo
->free_inodes
++;
266 spin_unlock(&sbinfo
->stat_lock
);
271 * shmem_recalc_inode - recalculate the block usage of an inode
272 * @inode: inode to recalc
274 * We have to calculate the free blocks since the mm can drop
275 * undirtied hole pages behind our back.
277 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
278 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
280 * It has to be called with the spinlock held.
282 static void shmem_recalc_inode(struct inode
*inode
)
284 struct shmem_inode_info
*info
= SHMEM_I(inode
);
287 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
289 info
->alloced
-= freed
;
290 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
291 shmem_inode_unacct_blocks(inode
, freed
);
295 bool shmem_charge(struct inode
*inode
, long pages
)
297 struct shmem_inode_info
*info
= SHMEM_I(inode
);
300 if (!shmem_inode_acct_block(inode
, pages
))
303 spin_lock_irqsave(&info
->lock
, flags
);
304 info
->alloced
+= pages
;
305 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
306 shmem_recalc_inode(inode
);
307 spin_unlock_irqrestore(&info
->lock
, flags
);
308 inode
->i_mapping
->nrpages
+= pages
;
313 void shmem_uncharge(struct inode
*inode
, long pages
)
315 struct shmem_inode_info
*info
= SHMEM_I(inode
);
318 spin_lock_irqsave(&info
->lock
, flags
);
319 info
->alloced
-= pages
;
320 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
321 shmem_recalc_inode(inode
);
322 spin_unlock_irqrestore(&info
->lock
, flags
);
324 shmem_inode_unacct_blocks(inode
, pages
);
328 * Replace item expected in radix tree by a new item, while holding tree lock.
330 static int shmem_radix_tree_replace(struct address_space
*mapping
,
331 pgoff_t index
, void *expected
, void *replacement
)
333 struct radix_tree_node
*node
;
337 VM_BUG_ON(!expected
);
338 VM_BUG_ON(!replacement
);
339 item
= __radix_tree_lookup(&mapping
->page_tree
, index
, &node
, &pslot
);
342 if (item
!= expected
)
344 __radix_tree_replace(&mapping
->page_tree
, node
, pslot
,
350 * Sometimes, before we decide whether to proceed or to fail, we must check
351 * that an entry was not already brought back from swap by a racing thread.
353 * Checking page is not enough: by the time a SwapCache page is locked, it
354 * might be reused, and again be SwapCache, using the same swap as before.
356 static bool shmem_confirm_swap(struct address_space
*mapping
,
357 pgoff_t index
, swp_entry_t swap
)
362 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
364 return item
== swp_to_radix_entry(swap
);
368 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
371 * disables huge pages for the mount;
373 * enables huge pages for the mount;
374 * SHMEM_HUGE_WITHIN_SIZE:
375 * only allocate huge pages if the page will be fully within i_size,
376 * also respect fadvise()/madvise() hints;
378 * only allocate huge pages if requested with fadvise()/madvise();
381 #define SHMEM_HUGE_NEVER 0
382 #define SHMEM_HUGE_ALWAYS 1
383 #define SHMEM_HUGE_WITHIN_SIZE 2
384 #define SHMEM_HUGE_ADVISE 3
388 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
391 * disables huge on shm_mnt and all mounts, for emergency use;
393 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
396 #define SHMEM_HUGE_DENY (-1)
397 #define SHMEM_HUGE_FORCE (-2)
399 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
400 /* ifdef here to avoid bloating shmem.o when not necessary */
402 int shmem_huge __read_mostly
;
404 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
405 static int shmem_parse_huge(const char *str
)
407 if (!strcmp(str
, "never"))
408 return SHMEM_HUGE_NEVER
;
409 if (!strcmp(str
, "always"))
410 return SHMEM_HUGE_ALWAYS
;
411 if (!strcmp(str
, "within_size"))
412 return SHMEM_HUGE_WITHIN_SIZE
;
413 if (!strcmp(str
, "advise"))
414 return SHMEM_HUGE_ADVISE
;
415 if (!strcmp(str
, "deny"))
416 return SHMEM_HUGE_DENY
;
417 if (!strcmp(str
, "force"))
418 return SHMEM_HUGE_FORCE
;
422 static const char *shmem_format_huge(int huge
)
425 case SHMEM_HUGE_NEVER
:
427 case SHMEM_HUGE_ALWAYS
:
429 case SHMEM_HUGE_WITHIN_SIZE
:
430 return "within_size";
431 case SHMEM_HUGE_ADVISE
:
433 case SHMEM_HUGE_DENY
:
435 case SHMEM_HUGE_FORCE
:
444 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
445 struct shrink_control
*sc
, unsigned long nr_to_split
)
447 LIST_HEAD(list
), *pos
, *next
;
448 LIST_HEAD(to_remove
);
450 struct shmem_inode_info
*info
;
452 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
453 int removed
= 0, split
= 0;
455 if (list_empty(&sbinfo
->shrinklist
))
458 spin_lock(&sbinfo
->shrinklist_lock
);
459 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
460 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
463 inode
= igrab(&info
->vfs_inode
);
465 /* inode is about to be evicted */
467 list_del_init(&info
->shrinklist
);
472 /* Check if there's anything to gain */
473 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
474 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
475 list_move(&info
->shrinklist
, &to_remove
);
480 list_move(&info
->shrinklist
, &list
);
485 spin_unlock(&sbinfo
->shrinklist_lock
);
487 list_for_each_safe(pos
, next
, &to_remove
) {
488 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
489 inode
= &info
->vfs_inode
;
490 list_del_init(&info
->shrinklist
);
494 list_for_each_safe(pos
, next
, &list
) {
497 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
498 inode
= &info
->vfs_inode
;
500 if (nr_to_split
&& split
>= nr_to_split
)
503 page
= find_get_page(inode
->i_mapping
,
504 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
508 /* No huge page at the end of the file: nothing to split */
509 if (!PageTransHuge(page
)) {
515 * Leave the inode on the list if we failed to lock
516 * the page at this time.
518 * Waiting for the lock may lead to deadlock in the
521 if (!trylock_page(page
)) {
526 ret
= split_huge_page(page
);
530 /* If split failed leave the inode on the list */
536 list_del_init(&info
->shrinklist
);
542 spin_lock(&sbinfo
->shrinklist_lock
);
543 list_splice_tail(&list
, &sbinfo
->shrinklist
);
544 sbinfo
->shrinklist_len
-= removed
;
545 spin_unlock(&sbinfo
->shrinklist_lock
);
550 static long shmem_unused_huge_scan(struct super_block
*sb
,
551 struct shrink_control
*sc
)
553 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
555 if (!READ_ONCE(sbinfo
->shrinklist_len
))
558 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
561 static long shmem_unused_huge_count(struct super_block
*sb
,
562 struct shrink_control
*sc
)
564 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
565 return READ_ONCE(sbinfo
->shrinklist_len
);
567 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
569 #define shmem_huge SHMEM_HUGE_DENY
571 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
572 struct shrink_control
*sc
, unsigned long nr_to_split
)
576 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
579 * Like add_to_page_cache_locked, but error if expected item has gone.
581 static int shmem_add_to_page_cache(struct page
*page
,
582 struct address_space
*mapping
,
583 pgoff_t index
, void *expected
)
585 int error
, nr
= hpage_nr_pages(page
);
587 VM_BUG_ON_PAGE(PageTail(page
), page
);
588 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
589 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
590 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
591 VM_BUG_ON(expected
&& PageTransHuge(page
));
593 page_ref_add(page
, nr
);
594 page
->mapping
= mapping
;
597 spin_lock_irq(&mapping
->tree_lock
);
598 if (PageTransHuge(page
)) {
599 void __rcu
**results
;
604 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
,
605 &results
, &idx
, index
, 1) &&
606 idx
< index
+ HPAGE_PMD_NR
) {
611 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
612 error
= radix_tree_insert(&mapping
->page_tree
,
613 index
+ i
, page
+ i
);
616 count_vm_event(THP_FILE_ALLOC
);
618 } else if (!expected
) {
619 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
621 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
626 mapping
->nrpages
+= nr
;
627 if (PageTransHuge(page
))
628 __inc_node_page_state(page
, NR_SHMEM_THPS
);
629 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
630 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
631 spin_unlock_irq(&mapping
->tree_lock
);
633 page
->mapping
= NULL
;
634 spin_unlock_irq(&mapping
->tree_lock
);
635 page_ref_sub(page
, nr
);
641 * Like delete_from_page_cache, but substitutes swap for page.
643 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
645 struct address_space
*mapping
= page
->mapping
;
648 VM_BUG_ON_PAGE(PageCompound(page
), page
);
650 spin_lock_irq(&mapping
->tree_lock
);
651 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
652 page
->mapping
= NULL
;
654 __dec_node_page_state(page
, NR_FILE_PAGES
);
655 __dec_node_page_state(page
, NR_SHMEM
);
656 spin_unlock_irq(&mapping
->tree_lock
);
662 * Remove swap entry from radix tree, free the swap and its page cache.
664 static int shmem_free_swap(struct address_space
*mapping
,
665 pgoff_t index
, void *radswap
)
669 spin_lock_irq(&mapping
->tree_lock
);
670 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
671 spin_unlock_irq(&mapping
->tree_lock
);
674 free_swap_and_cache(radix_to_swp_entry(radswap
));
679 * Determine (in bytes) how many of the shmem object's pages mapped by the
680 * given offsets are swapped out.
682 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
683 * as long as the inode doesn't go away and racy results are not a problem.
685 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
686 pgoff_t start
, pgoff_t end
)
688 struct radix_tree_iter iter
;
691 unsigned long swapped
= 0;
695 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
696 if (iter
.index
>= end
)
699 page
= radix_tree_deref_slot(slot
);
701 if (radix_tree_deref_retry(page
)) {
702 slot
= radix_tree_iter_retry(&iter
);
706 if (radix_tree_exceptional_entry(page
))
709 if (need_resched()) {
710 slot
= radix_tree_iter_resume(slot
, &iter
);
717 return swapped
<< PAGE_SHIFT
;
721 * Determine (in bytes) how many of the shmem object's pages mapped by the
722 * given vma is swapped out.
724 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
725 * as long as the inode doesn't go away and racy results are not a problem.
727 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
729 struct inode
*inode
= file_inode(vma
->vm_file
);
730 struct shmem_inode_info
*info
= SHMEM_I(inode
);
731 struct address_space
*mapping
= inode
->i_mapping
;
732 unsigned long swapped
;
734 /* Be careful as we don't hold info->lock */
735 swapped
= READ_ONCE(info
->swapped
);
738 * The easier cases are when the shmem object has nothing in swap, or
739 * the vma maps it whole. Then we can simply use the stats that we
745 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
746 return swapped
<< PAGE_SHIFT
;
748 /* Here comes the more involved part */
749 return shmem_partial_swap_usage(mapping
,
750 linear_page_index(vma
, vma
->vm_start
),
751 linear_page_index(vma
, vma
->vm_end
));
755 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
757 void shmem_unlock_mapping(struct address_space
*mapping
)
760 pgoff_t indices
[PAGEVEC_SIZE
];
765 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
767 while (!mapping_unevictable(mapping
)) {
769 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
770 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
772 pvec
.nr
= find_get_entries(mapping
, index
,
773 PAGEVEC_SIZE
, pvec
.pages
, indices
);
776 index
= indices
[pvec
.nr
- 1] + 1;
777 pagevec_remove_exceptionals(&pvec
);
778 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
779 pagevec_release(&pvec
);
785 * Remove range of pages and swap entries from radix tree, and free them.
786 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
788 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
791 struct address_space
*mapping
= inode
->i_mapping
;
792 struct shmem_inode_info
*info
= SHMEM_I(inode
);
793 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
794 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
795 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
796 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
798 pgoff_t indices
[PAGEVEC_SIZE
];
799 long nr_swaps_freed
= 0;
804 end
= -1; /* unsigned, so actually very big */
808 while (index
< end
) {
809 pvec
.nr
= find_get_entries(mapping
, index
,
810 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
811 pvec
.pages
, indices
);
814 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
815 struct page
*page
= pvec
.pages
[i
];
821 if (radix_tree_exceptional_entry(page
)) {
824 nr_swaps_freed
+= !shmem_free_swap(mapping
,
829 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
831 if (!trylock_page(page
))
834 if (PageTransTail(page
)) {
835 /* Middle of THP: zero out the page */
836 clear_highpage(page
);
839 } else if (PageTransHuge(page
)) {
840 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
842 * Range ends in the middle of THP:
845 clear_highpage(page
);
849 index
+= HPAGE_PMD_NR
- 1;
850 i
+= HPAGE_PMD_NR
- 1;
853 if (!unfalloc
|| !PageUptodate(page
)) {
854 VM_BUG_ON_PAGE(PageTail(page
), page
);
855 if (page_mapping(page
) == mapping
) {
856 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
857 truncate_inode_page(mapping
, page
);
862 pagevec_remove_exceptionals(&pvec
);
863 pagevec_release(&pvec
);
869 struct page
*page
= NULL
;
870 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
872 unsigned int top
= PAGE_SIZE
;
877 zero_user_segment(page
, partial_start
, top
);
878 set_page_dirty(page
);
884 struct page
*page
= NULL
;
885 shmem_getpage(inode
, end
, &page
, SGP_READ
);
887 zero_user_segment(page
, 0, partial_end
);
888 set_page_dirty(page
);
897 while (index
< end
) {
900 pvec
.nr
= find_get_entries(mapping
, index
,
901 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
902 pvec
.pages
, indices
);
904 /* If all gone or hole-punch or unfalloc, we're done */
905 if (index
== start
|| end
!= -1)
907 /* But if truncating, restart to make sure all gone */
911 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
912 struct page
*page
= pvec
.pages
[i
];
918 if (radix_tree_exceptional_entry(page
)) {
921 if (shmem_free_swap(mapping
, index
, page
)) {
922 /* Swap was replaced by page: retry */
932 if (PageTransTail(page
)) {
933 /* Middle of THP: zero out the page */
934 clear_highpage(page
);
937 * Partial thp truncate due 'start' in middle
938 * of THP: don't need to look on these pages
939 * again on !pvec.nr restart.
941 if (index
!= round_down(end
, HPAGE_PMD_NR
))
944 } else if (PageTransHuge(page
)) {
945 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
947 * Range ends in the middle of THP:
950 clear_highpage(page
);
954 index
+= HPAGE_PMD_NR
- 1;
955 i
+= HPAGE_PMD_NR
- 1;
958 if (!unfalloc
|| !PageUptodate(page
)) {
959 VM_BUG_ON_PAGE(PageTail(page
), page
);
960 if (page_mapping(page
) == mapping
) {
961 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
962 truncate_inode_page(mapping
, page
);
964 /* Page was replaced by swap: retry */
972 pagevec_remove_exceptionals(&pvec
);
973 pagevec_release(&pvec
);
977 spin_lock_irq(&info
->lock
);
978 info
->swapped
-= nr_swaps_freed
;
979 shmem_recalc_inode(inode
);
980 spin_unlock_irq(&info
->lock
);
983 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
985 shmem_undo_range(inode
, lstart
, lend
, false);
986 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
988 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
990 static int shmem_getattr(const struct path
*path
, struct kstat
*stat
,
991 u32 request_mask
, unsigned int query_flags
)
993 struct inode
*inode
= path
->dentry
->d_inode
;
994 struct shmem_inode_info
*info
= SHMEM_I(inode
);
996 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
997 spin_lock_irq(&info
->lock
);
998 shmem_recalc_inode(inode
);
999 spin_unlock_irq(&info
->lock
);
1001 generic_fillattr(inode
, stat
);
1005 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1007 struct inode
*inode
= d_inode(dentry
);
1008 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1009 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1012 error
= setattr_prepare(dentry
, attr
);
1016 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1017 loff_t oldsize
= inode
->i_size
;
1018 loff_t newsize
= attr
->ia_size
;
1020 /* protected by i_mutex */
1021 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1022 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1025 if (newsize
!= oldsize
) {
1026 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1030 i_size_write(inode
, newsize
);
1031 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1033 if (newsize
<= oldsize
) {
1034 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1035 if (oldsize
> holebegin
)
1036 unmap_mapping_range(inode
->i_mapping
,
1039 shmem_truncate_range(inode
,
1040 newsize
, (loff_t
)-1);
1041 /* unmap again to remove racily COWed private pages */
1042 if (oldsize
> holebegin
)
1043 unmap_mapping_range(inode
->i_mapping
,
1047 * Part of the huge page can be beyond i_size: subject
1048 * to shrink under memory pressure.
1050 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1051 spin_lock(&sbinfo
->shrinklist_lock
);
1053 * _careful to defend against unlocked access to
1054 * ->shrink_list in shmem_unused_huge_shrink()
1056 if (list_empty_careful(&info
->shrinklist
)) {
1057 list_add_tail(&info
->shrinklist
,
1058 &sbinfo
->shrinklist
);
1059 sbinfo
->shrinklist_len
++;
1061 spin_unlock(&sbinfo
->shrinklist_lock
);
1066 setattr_copy(inode
, attr
);
1067 if (attr
->ia_valid
& ATTR_MODE
)
1068 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1072 static void shmem_evict_inode(struct inode
*inode
)
1074 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1075 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1077 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1078 shmem_unacct_size(info
->flags
, inode
->i_size
);
1080 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1081 if (!list_empty(&info
->shrinklist
)) {
1082 spin_lock(&sbinfo
->shrinklist_lock
);
1083 if (!list_empty(&info
->shrinklist
)) {
1084 list_del_init(&info
->shrinklist
);
1085 sbinfo
->shrinklist_len
--;
1087 spin_unlock(&sbinfo
->shrinklist_lock
);
1089 if (!list_empty(&info
->swaplist
)) {
1090 mutex_lock(&shmem_swaplist_mutex
);
1091 list_del_init(&info
->swaplist
);
1092 mutex_unlock(&shmem_swaplist_mutex
);
1096 simple_xattrs_free(&info
->xattrs
);
1097 WARN_ON(inode
->i_blocks
);
1098 if (!sbinfo
->idr_nouse
&& inode
->i_ino
) {
1099 mutex_lock(&sbinfo
->idr_lock
);
1100 idr_remove(&sbinfo
->idr
, inode
->i_ino
);
1101 mutex_unlock(&sbinfo
->idr_lock
);
1103 shmem_free_inode(inode
->i_sb
);
1107 static unsigned long find_swap_entry(struct radix_tree_root
*root
, void *item
)
1109 struct radix_tree_iter iter
;
1111 unsigned long found
= -1;
1112 unsigned int checked
= 0;
1115 radix_tree_for_each_slot(slot
, root
, &iter
, 0) {
1116 if (*slot
== item
) {
1121 if ((checked
% 4096) != 0)
1123 slot
= radix_tree_iter_resume(slot
, &iter
);
1132 * If swap found in inode, free it and move page from swapcache to filecache.
1134 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
1135 swp_entry_t swap
, struct page
**pagep
)
1137 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1143 radswap
= swp_to_radix_entry(swap
);
1144 index
= find_swap_entry(&mapping
->page_tree
, radswap
);
1146 return -EAGAIN
; /* tell shmem_unuse we found nothing */
1149 * Move _head_ to start search for next from here.
1150 * But be careful: shmem_evict_inode checks list_empty without taking
1151 * mutex, and there's an instant in list_move_tail when info->swaplist
1152 * would appear empty, if it were the only one on shmem_swaplist.
1154 if (shmem_swaplist
.next
!= &info
->swaplist
)
1155 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
1157 gfp
= mapping_gfp_mask(mapping
);
1158 if (shmem_should_replace_page(*pagep
, gfp
)) {
1159 mutex_unlock(&shmem_swaplist_mutex
);
1160 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
1161 mutex_lock(&shmem_swaplist_mutex
);
1163 * We needed to drop mutex to make that restrictive page
1164 * allocation, but the inode might have been freed while we
1165 * dropped it: although a racing shmem_evict_inode() cannot
1166 * complete without emptying the radix_tree, our page lock
1167 * on this swapcache page is not enough to prevent that -
1168 * free_swap_and_cache() of our swap entry will only
1169 * trylock_page(), removing swap from radix_tree whatever.
1171 * We must not proceed to shmem_add_to_page_cache() if the
1172 * inode has been freed, but of course we cannot rely on
1173 * inode or mapping or info to check that. However, we can
1174 * safely check if our swap entry is still in use (and here
1175 * it can't have got reused for another page): if it's still
1176 * in use, then the inode cannot have been freed yet, and we
1177 * can safely proceed (if it's no longer in use, that tells
1178 * nothing about the inode, but we don't need to unuse swap).
1180 if (!page_swapcount(*pagep
))
1185 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1186 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1187 * beneath us (pagelock doesn't help until the page is in pagecache).
1190 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
1192 if (error
!= -ENOMEM
) {
1194 * Truncation and eviction use free_swap_and_cache(), which
1195 * only does trylock page: if we raced, best clean up here.
1197 delete_from_swap_cache(*pagep
);
1198 set_page_dirty(*pagep
);
1200 spin_lock_irq(&info
->lock
);
1202 spin_unlock_irq(&info
->lock
);
1210 * Search through swapped inodes to find and replace swap by page.
1212 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1214 struct list_head
*this, *next
;
1215 struct shmem_inode_info
*info
;
1216 struct mem_cgroup
*memcg
;
1220 * There's a faint possibility that swap page was replaced before
1221 * caller locked it: caller will come back later with the right page.
1223 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1227 * Charge page using GFP_KERNEL while we can wait, before taking
1228 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1229 * Charged back to the user (not to caller) when swap account is used.
1231 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1235 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1238 mutex_lock(&shmem_swaplist_mutex
);
1239 list_for_each_safe(this, next
, &shmem_swaplist
) {
1240 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1242 error
= shmem_unuse_inode(info
, swap
, &page
);
1244 list_del_init(&info
->swaplist
);
1246 if (error
!= -EAGAIN
)
1248 /* found nothing in this: move on to search the next */
1250 mutex_unlock(&shmem_swaplist_mutex
);
1253 if (error
!= -ENOMEM
)
1255 mem_cgroup_cancel_charge(page
, memcg
, false);
1257 mem_cgroup_commit_charge(page
, memcg
, true, false);
1265 * Move the page from the page cache to the swap cache.
1267 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1269 struct shmem_inode_info
*info
;
1270 struct address_space
*mapping
;
1271 struct inode
*inode
;
1275 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1276 BUG_ON(!PageLocked(page
));
1277 mapping
= page
->mapping
;
1278 index
= page
->index
;
1279 inode
= mapping
->host
;
1280 info
= SHMEM_I(inode
);
1281 if (info
->flags
& VM_LOCKED
)
1283 if (!total_swap_pages
)
1287 * Our capabilities prevent regular writeback or sync from ever calling
1288 * shmem_writepage; but a stacking filesystem might use ->writepage of
1289 * its underlying filesystem, in which case tmpfs should write out to
1290 * swap only in response to memory pressure, and not for the writeback
1293 if (!wbc
->for_reclaim
) {
1294 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1299 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1300 * value into swapfile.c, the only way we can correctly account for a
1301 * fallocated page arriving here is now to initialize it and write it.
1303 * That's okay for a page already fallocated earlier, but if we have
1304 * not yet completed the fallocation, then (a) we want to keep track
1305 * of this page in case we have to undo it, and (b) it may not be a
1306 * good idea to continue anyway, once we're pushing into swap. So
1307 * reactivate the page, and let shmem_fallocate() quit when too many.
1309 if (!PageUptodate(page
)) {
1310 if (inode
->i_private
) {
1311 struct shmem_falloc
*shmem_falloc
;
1312 spin_lock(&inode
->i_lock
);
1313 shmem_falloc
= inode
->i_private
;
1315 !shmem_falloc
->waitq
&&
1316 index
>= shmem_falloc
->start
&&
1317 index
< shmem_falloc
->next
)
1318 shmem_falloc
->nr_unswapped
++;
1320 shmem_falloc
= NULL
;
1321 spin_unlock(&inode
->i_lock
);
1325 clear_highpage(page
);
1326 flush_dcache_page(page
);
1327 SetPageUptodate(page
);
1330 swap
= get_swap_page(page
);
1334 if (mem_cgroup_try_charge_swap(page
, swap
))
1338 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1339 * if it's not already there. Do it now before the page is
1340 * moved to swap cache, when its pagelock no longer protects
1341 * the inode from eviction. But don't unlock the mutex until
1342 * we've incremented swapped, because shmem_unuse_inode() will
1343 * prune a !swapped inode from the swaplist under this mutex.
1345 mutex_lock(&shmem_swaplist_mutex
);
1346 if (list_empty(&info
->swaplist
))
1347 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1349 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1350 spin_lock_irq(&info
->lock
);
1351 shmem_recalc_inode(inode
);
1353 spin_unlock_irq(&info
->lock
);
1355 swap_shmem_alloc(swap
);
1356 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1358 mutex_unlock(&shmem_swaplist_mutex
);
1359 BUG_ON(page_mapped(page
));
1360 swap_writepage(page
, wbc
);
1364 mutex_unlock(&shmem_swaplist_mutex
);
1366 put_swap_page(page
, swap
);
1368 set_page_dirty(page
);
1369 if (wbc
->for_reclaim
)
1370 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1375 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1376 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1380 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1381 return; /* show nothing */
1383 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1385 seq_printf(seq
, ",mpol=%s", buffer
);
1388 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1390 struct mempolicy
*mpol
= NULL
;
1392 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1393 mpol
= sbinfo
->mpol
;
1395 spin_unlock(&sbinfo
->stat_lock
);
1399 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1400 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1403 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1407 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1409 #define vm_policy vm_private_data
1412 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1413 struct shmem_inode_info
*info
, pgoff_t index
)
1415 /* Create a pseudo vma that just contains the policy */
1417 /* Bias interleave by inode number to distribute better across nodes */
1418 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1420 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1423 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1425 /* Drop reference taken by mpol_shared_policy_lookup() */
1426 mpol_cond_put(vma
->vm_policy
);
1429 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1430 struct shmem_inode_info
*info
, pgoff_t index
)
1432 struct vm_area_struct pvma
;
1435 shmem_pseudo_vma_init(&pvma
, info
, index
);
1436 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1437 shmem_pseudo_vma_destroy(&pvma
);
1442 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1443 struct shmem_inode_info
*info
, pgoff_t index
)
1445 struct vm_area_struct pvma
;
1446 struct inode
*inode
= &info
->vfs_inode
;
1447 struct address_space
*mapping
= inode
->i_mapping
;
1448 pgoff_t idx
, hindex
;
1449 void __rcu
**results
;
1452 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1455 hindex
= round_down(index
, HPAGE_PMD_NR
);
1457 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1458 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1464 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1465 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1466 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1467 shmem_pseudo_vma_destroy(&pvma
);
1469 prep_transhuge_page(page
);
1473 static struct page
*shmem_alloc_page(gfp_t gfp
,
1474 struct shmem_inode_info
*info
, pgoff_t index
)
1476 struct vm_area_struct pvma
;
1479 shmem_pseudo_vma_init(&pvma
, info
, index
);
1480 page
= alloc_page_vma(gfp
, &pvma
, 0);
1481 shmem_pseudo_vma_destroy(&pvma
);
1486 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1487 struct inode
*inode
,
1488 pgoff_t index
, bool huge
)
1490 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1495 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1497 nr
= huge
? HPAGE_PMD_NR
: 1;
1499 if (!shmem_inode_acct_block(inode
, nr
))
1503 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1505 page
= shmem_alloc_page(gfp
, info
, index
);
1507 __SetPageLocked(page
);
1508 __SetPageSwapBacked(page
);
1513 shmem_inode_unacct_blocks(inode
, nr
);
1515 return ERR_PTR(err
);
1519 * When a page is moved from swapcache to shmem filecache (either by the
1520 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1521 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1522 * ignorance of the mapping it belongs to. If that mapping has special
1523 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1524 * we may need to copy to a suitable page before moving to filecache.
1526 * In a future release, this may well be extended to respect cpuset and
1527 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1528 * but for now it is a simple matter of zone.
1530 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1532 return page_zonenum(page
) > gfp_zone(gfp
);
1535 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1536 struct shmem_inode_info
*info
, pgoff_t index
)
1538 struct page
*oldpage
, *newpage
;
1539 struct address_space
*swap_mapping
;
1544 swap_index
= page_private(oldpage
);
1545 swap_mapping
= page_mapping(oldpage
);
1548 * We have arrived here because our zones are constrained, so don't
1549 * limit chance of success by further cpuset and node constraints.
1551 gfp
&= ~GFP_CONSTRAINT_MASK
;
1552 newpage
= shmem_alloc_page(gfp
, info
, index
);
1557 copy_highpage(newpage
, oldpage
);
1558 flush_dcache_page(newpage
);
1560 __SetPageLocked(newpage
);
1561 __SetPageSwapBacked(newpage
);
1562 SetPageUptodate(newpage
);
1563 set_page_private(newpage
, swap_index
);
1564 SetPageSwapCache(newpage
);
1567 * Our caller will very soon move newpage out of swapcache, but it's
1568 * a nice clean interface for us to replace oldpage by newpage there.
1570 spin_lock_irq(&swap_mapping
->tree_lock
);
1571 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1574 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1575 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1577 spin_unlock_irq(&swap_mapping
->tree_lock
);
1579 if (unlikely(error
)) {
1581 * Is this possible? I think not, now that our callers check
1582 * both PageSwapCache and page_private after getting page lock;
1583 * but be defensive. Reverse old to newpage for clear and free.
1587 mem_cgroup_migrate(oldpage
, newpage
);
1588 lru_cache_add_anon(newpage
);
1592 ClearPageSwapCache(oldpage
);
1593 set_page_private(oldpage
, 0);
1595 unlock_page(oldpage
);
1602 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1604 * If we allocate a new one we do not mark it dirty. That's up to the
1605 * vm. If we swap it in we mark it dirty since we also free the swap
1606 * entry since a page cannot live in both the swap and page cache.
1608 * fault_mm and fault_type are only supplied by shmem_fault:
1609 * otherwise they are NULL.
1611 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1612 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1613 struct vm_area_struct
*vma
, struct vm_fault
*vmf
, int *fault_type
)
1615 struct address_space
*mapping
= inode
->i_mapping
;
1616 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1617 struct shmem_sb_info
*sbinfo
;
1618 struct mm_struct
*charge_mm
;
1619 struct mem_cgroup
*memcg
;
1622 enum sgp_type sgp_huge
= sgp
;
1623 pgoff_t hindex
= index
;
1628 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1630 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1634 page
= find_lock_entry(mapping
, index
);
1635 if (radix_tree_exceptional_entry(page
)) {
1636 swap
= radix_to_swp_entry(page
);
1640 if (sgp
<= SGP_CACHE
&&
1641 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1646 if (page
&& sgp
== SGP_WRITE
)
1647 mark_page_accessed(page
);
1649 /* fallocated page? */
1650 if (page
&& !PageUptodate(page
)) {
1651 if (sgp
!= SGP_READ
)
1657 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1663 * Fast cache lookup did not find it:
1664 * bring it back from swap or allocate.
1666 sbinfo
= SHMEM_SB(inode
->i_sb
);
1667 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1670 /* Look it up and read it in.. */
1671 page
= lookup_swap_cache(swap
, NULL
, 0);
1673 /* Or update major stats only when swapin succeeds?? */
1675 *fault_type
|= VM_FAULT_MAJOR
;
1676 count_vm_event(PGMAJFAULT
);
1677 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1679 /* Here we actually start the io */
1680 page
= shmem_swapin(swap
, gfp
, info
, index
);
1687 /* We have to do this with page locked to prevent races */
1689 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1690 !shmem_confirm_swap(mapping
, index
, swap
)) {
1691 error
= -EEXIST
; /* try again */
1694 if (!PageUptodate(page
)) {
1698 wait_on_page_writeback(page
);
1700 if (shmem_should_replace_page(page
, gfp
)) {
1701 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1706 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1709 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1710 swp_to_radix_entry(swap
));
1712 * We already confirmed swap under page lock, and make
1713 * no memory allocation here, so usually no possibility
1714 * of error; but free_swap_and_cache() only trylocks a
1715 * page, so it is just possible that the entry has been
1716 * truncated or holepunched since swap was confirmed.
1717 * shmem_undo_range() will have done some of the
1718 * unaccounting, now delete_from_swap_cache() will do
1720 * Reset swap.val? No, leave it so "failed" goes back to
1721 * "repeat": reading a hole and writing should succeed.
1724 mem_cgroup_cancel_charge(page
, memcg
, false);
1725 delete_from_swap_cache(page
);
1731 mem_cgroup_commit_charge(page
, memcg
, true, false);
1733 spin_lock_irq(&info
->lock
);
1735 shmem_recalc_inode(inode
);
1736 spin_unlock_irq(&info
->lock
);
1738 if (sgp
== SGP_WRITE
)
1739 mark_page_accessed(page
);
1741 delete_from_swap_cache(page
);
1742 set_page_dirty(page
);
1746 if (vma
&& userfaultfd_missing(vma
)) {
1747 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1751 /* shmem_symlink() */
1752 if (mapping
->a_ops
!= &shmem_aops
)
1754 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1756 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1758 switch (sbinfo
->huge
) {
1761 case SHMEM_HUGE_NEVER
:
1763 case SHMEM_HUGE_WITHIN_SIZE
:
1764 off
= round_up(index
, HPAGE_PMD_NR
);
1765 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1766 if (i_size
>= HPAGE_PMD_SIZE
&&
1767 i_size
>> PAGE_SHIFT
>= off
)
1770 case SHMEM_HUGE_ADVISE
:
1771 if (sgp_huge
== SGP_HUGE
)
1773 /* TODO: implement fadvise() hints */
1778 page
= shmem_alloc_and_acct_page(gfp
, inode
, index
, true);
1780 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, inode
,
1785 error
= PTR_ERR(page
);
1787 if (error
!= -ENOSPC
)
1790 * Try to reclaim some spece by splitting a huge page
1791 * beyond i_size on the filesystem.
1795 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1796 if (ret
== SHRINK_STOP
)
1804 if (PageTransHuge(page
))
1805 hindex
= round_down(index
, HPAGE_PMD_NR
);
1809 if (sgp
== SGP_WRITE
)
1810 __SetPageReferenced(page
);
1812 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1813 PageTransHuge(page
));
1816 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1817 compound_order(page
));
1819 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1821 radix_tree_preload_end();
1824 mem_cgroup_cancel_charge(page
, memcg
,
1825 PageTransHuge(page
));
1828 mem_cgroup_commit_charge(page
, memcg
, false,
1829 PageTransHuge(page
));
1830 lru_cache_add_anon(page
);
1832 spin_lock_irq(&info
->lock
);
1833 info
->alloced
+= 1 << compound_order(page
);
1834 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1835 shmem_recalc_inode(inode
);
1836 spin_unlock_irq(&info
->lock
);
1839 if (PageTransHuge(page
) &&
1840 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1841 hindex
+ HPAGE_PMD_NR
- 1) {
1843 * Part of the huge page is beyond i_size: subject
1844 * to shrink under memory pressure.
1846 spin_lock(&sbinfo
->shrinklist_lock
);
1848 * _careful to defend against unlocked access to
1849 * ->shrink_list in shmem_unused_huge_shrink()
1851 if (list_empty_careful(&info
->shrinklist
)) {
1852 list_add_tail(&info
->shrinklist
,
1853 &sbinfo
->shrinklist
);
1854 sbinfo
->shrinklist_len
++;
1856 spin_unlock(&sbinfo
->shrinklist_lock
);
1860 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1862 if (sgp
== SGP_FALLOC
)
1866 * Let SGP_WRITE caller clear ends if write does not fill page;
1867 * but SGP_FALLOC on a page fallocated earlier must initialize
1868 * it now, lest undo on failure cancel our earlier guarantee.
1870 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1871 struct page
*head
= compound_head(page
);
1874 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1875 clear_highpage(head
+ i
);
1876 flush_dcache_page(head
+ i
);
1878 SetPageUptodate(head
);
1882 /* Perhaps the file has been truncated since we checked */
1883 if (sgp
<= SGP_CACHE
&&
1884 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1886 ClearPageDirty(page
);
1887 delete_from_page_cache(page
);
1888 spin_lock_irq(&info
->lock
);
1889 shmem_recalc_inode(inode
);
1890 spin_unlock_irq(&info
->lock
);
1895 *pagep
= page
+ index
- hindex
;
1902 shmem_inode_unacct_blocks(inode
, 1 << compound_order(page
));
1904 if (PageTransHuge(page
)) {
1910 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1917 if (error
== -ENOSPC
&& !once
++) {
1918 spin_lock_irq(&info
->lock
);
1919 shmem_recalc_inode(inode
);
1920 spin_unlock_irq(&info
->lock
);
1923 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1929 * This is like autoremove_wake_function, but it removes the wait queue
1930 * entry unconditionally - even if something else had already woken the
1933 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
1935 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1936 list_del_init(&wait
->entry
);
1940 static int shmem_fault(struct vm_fault
*vmf
)
1942 struct vm_area_struct
*vma
= vmf
->vma
;
1943 struct inode
*inode
= file_inode(vma
->vm_file
);
1944 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1947 int ret
= VM_FAULT_LOCKED
;
1950 * Trinity finds that probing a hole which tmpfs is punching can
1951 * prevent the hole-punch from ever completing: which in turn
1952 * locks writers out with its hold on i_mutex. So refrain from
1953 * faulting pages into the hole while it's being punched. Although
1954 * shmem_undo_range() does remove the additions, it may be unable to
1955 * keep up, as each new page needs its own unmap_mapping_range() call,
1956 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1958 * It does not matter if we sometimes reach this check just before the
1959 * hole-punch begins, so that one fault then races with the punch:
1960 * we just need to make racing faults a rare case.
1962 * The implementation below would be much simpler if we just used a
1963 * standard mutex or completion: but we cannot take i_mutex in fault,
1964 * and bloating every shmem inode for this unlikely case would be sad.
1966 if (unlikely(inode
->i_private
)) {
1967 struct shmem_falloc
*shmem_falloc
;
1969 spin_lock(&inode
->i_lock
);
1970 shmem_falloc
= inode
->i_private
;
1972 shmem_falloc
->waitq
&&
1973 vmf
->pgoff
>= shmem_falloc
->start
&&
1974 vmf
->pgoff
< shmem_falloc
->next
) {
1975 wait_queue_head_t
*shmem_falloc_waitq
;
1976 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
1978 ret
= VM_FAULT_NOPAGE
;
1979 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1980 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1981 /* It's polite to up mmap_sem if we can */
1982 up_read(&vma
->vm_mm
->mmap_sem
);
1983 ret
= VM_FAULT_RETRY
;
1986 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1987 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1988 TASK_UNINTERRUPTIBLE
);
1989 spin_unlock(&inode
->i_lock
);
1993 * shmem_falloc_waitq points into the shmem_fallocate()
1994 * stack of the hole-punching task: shmem_falloc_waitq
1995 * is usually invalid by the time we reach here, but
1996 * finish_wait() does not dereference it in that case;
1997 * though i_lock needed lest racing with wake_up_all().
1999 spin_lock(&inode
->i_lock
);
2000 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2001 spin_unlock(&inode
->i_lock
);
2004 spin_unlock(&inode
->i_lock
);
2009 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2010 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2012 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2015 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2016 gfp
, vma
, vmf
, &ret
);
2018 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
2022 unsigned long shmem_get_unmapped_area(struct file
*file
,
2023 unsigned long uaddr
, unsigned long len
,
2024 unsigned long pgoff
, unsigned long flags
)
2026 unsigned long (*get_area
)(struct file
*,
2027 unsigned long, unsigned long, unsigned long, unsigned long);
2029 unsigned long offset
;
2030 unsigned long inflated_len
;
2031 unsigned long inflated_addr
;
2032 unsigned long inflated_offset
;
2034 if (len
> TASK_SIZE
)
2037 get_area
= current
->mm
->get_unmapped_area
;
2038 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2040 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
2042 if (IS_ERR_VALUE(addr
))
2044 if (addr
& ~PAGE_MASK
)
2046 if (addr
> TASK_SIZE
- len
)
2049 if (shmem_huge
== SHMEM_HUGE_DENY
)
2051 if (len
< HPAGE_PMD_SIZE
)
2053 if (flags
& MAP_FIXED
)
2056 * Our priority is to support MAP_SHARED mapped hugely;
2057 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2058 * But if caller specified an address hint, respect that as before.
2063 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2064 struct super_block
*sb
;
2067 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2068 sb
= file_inode(file
)->i_sb
;
2071 * Called directly from mm/mmap.c, or drivers/char/mem.c
2072 * for "/dev/zero", to create a shared anonymous object.
2074 if (IS_ERR(shm_mnt
))
2076 sb
= shm_mnt
->mnt_sb
;
2078 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2082 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2083 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2085 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2088 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2089 if (inflated_len
> TASK_SIZE
)
2091 if (inflated_len
< len
)
2094 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
2095 if (IS_ERR_VALUE(inflated_addr
))
2097 if (inflated_addr
& ~PAGE_MASK
)
2100 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2101 inflated_addr
+= offset
- inflated_offset
;
2102 if (inflated_offset
> offset
)
2103 inflated_addr
+= HPAGE_PMD_SIZE
;
2105 if (inflated_addr
> TASK_SIZE
- len
)
2107 return inflated_addr
;
2111 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2113 struct inode
*inode
= file_inode(vma
->vm_file
);
2114 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2117 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2120 struct inode
*inode
= file_inode(vma
->vm_file
);
2123 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2124 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2128 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2130 struct inode
*inode
= file_inode(file
);
2131 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2132 int retval
= -ENOMEM
;
2134 spin_lock_irq(&info
->lock
);
2135 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2136 if (!user_shm_lock(inode
->i_size
, user
))
2138 info
->flags
|= VM_LOCKED
;
2139 mapping_set_unevictable(file
->f_mapping
);
2141 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2142 user_shm_unlock(inode
->i_size
, user
);
2143 info
->flags
&= ~VM_LOCKED
;
2144 mapping_clear_unevictable(file
->f_mapping
);
2149 spin_unlock_irq(&info
->lock
);
2153 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2155 file_accessed(file
);
2156 vma
->vm_ops
= &shmem_vm_ops
;
2157 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2158 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2159 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2160 khugepaged_enter(vma
, vma
->vm_flags
);
2165 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2166 umode_t mode
, dev_t dev
, unsigned long flags
)
2168 struct inode
*inode
;
2169 struct shmem_inode_info
*info
;
2170 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2173 if (shmem_reserve_inode(sb
))
2176 inode
= new_inode(sb
);
2178 inode_init_owner(inode
, dir
, mode
);
2179 inode
->i_blocks
= 0;
2180 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2181 inode
->i_generation
= get_seconds();
2182 info
= SHMEM_I(inode
);
2183 memset(info
, 0, (char *)inode
- (char *)info
);
2184 spin_lock_init(&info
->lock
);
2185 info
->seals
= F_SEAL_SEAL
;
2186 info
->flags
= flags
& VM_NORESERVE
;
2187 INIT_LIST_HEAD(&info
->shrinklist
);
2188 INIT_LIST_HEAD(&info
->swaplist
);
2189 simple_xattrs_init(&info
->xattrs
);
2190 cache_no_acl(inode
);
2192 switch (mode
& S_IFMT
) {
2194 inode
->i_op
= &shmem_special_inode_operations
;
2195 init_special_inode(inode
, mode
, dev
);
2198 inode
->i_mapping
->a_ops
= &shmem_aops
;
2199 inode
->i_op
= &shmem_inode_operations
;
2200 inode
->i_fop
= &shmem_file_operations
;
2201 mpol_shared_policy_init(&info
->policy
,
2202 shmem_get_sbmpol(sbinfo
));
2206 /* Some things misbehave if size == 0 on a directory */
2207 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2208 inode
->i_op
= &shmem_dir_inode_operations
;
2209 inode
->i_fop
= &simple_dir_operations
;
2213 * Must not load anything in the rbtree,
2214 * mpol_free_shared_policy will not be called.
2216 mpol_shared_policy_init(&info
->policy
, NULL
);
2220 lockdep_annotate_inode_mutex_key(inode
);
2222 if (!sbinfo
->idr_nouse
) {
2223 /* inum 0 and 1 are unused */
2224 mutex_lock(&sbinfo
->idr_lock
);
2225 ino
= idr_alloc(&sbinfo
->idr
, inode
, 2, INT_MAX
,
2229 mutex_unlock(&sbinfo
->idr_lock
);
2230 __insert_inode_hash(inode
, inode
->i_ino
);
2233 mutex_unlock(&sbinfo
->idr_lock
);
2235 /* shmem_free_inode() will be called */
2239 inode
->i_ino
= get_next_ino();
2241 shmem_free_inode(sb
);
2245 bool shmem_mapping(struct address_space
*mapping
)
2247 return mapping
->a_ops
== &shmem_aops
;
2250 static int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2252 struct vm_area_struct
*dst_vma
,
2253 unsigned long dst_addr
,
2254 unsigned long src_addr
,
2256 struct page
**pagep
)
2258 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2259 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2260 struct address_space
*mapping
= inode
->i_mapping
;
2261 gfp_t gfp
= mapping_gfp_mask(mapping
);
2262 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2263 struct mem_cgroup
*memcg
;
2267 pte_t _dst_pte
, *dst_pte
;
2269 pgoff_t offset
, max_off
;
2272 if (!shmem_inode_acct_block(inode
, 1))
2276 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2278 goto out_unacct_blocks
;
2280 if (!zeropage
) { /* mcopy_atomic */
2281 page_kaddr
= kmap_atomic(page
);
2282 ret
= copy_from_user(page_kaddr
,
2283 (const void __user
*)src_addr
,
2285 kunmap_atomic(page_kaddr
);
2287 /* fallback to copy_from_user outside mmap_sem */
2288 if (unlikely(ret
)) {
2290 shmem_inode_unacct_blocks(inode
, 1);
2291 /* don't free the page */
2294 } else { /* mfill_zeropage_atomic */
2295 clear_highpage(page
);
2302 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2303 __SetPageLocked(page
);
2304 __SetPageSwapBacked(page
);
2305 __SetPageUptodate(page
);
2308 offset
= linear_page_index(dst_vma
, dst_addr
);
2309 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2310 if (unlikely(offset
>= max_off
))
2313 ret
= mem_cgroup_try_charge(page
, dst_mm
, gfp
, &memcg
, false);
2317 ret
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
2319 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
);
2320 radix_tree_preload_end();
2323 goto out_release_uncharge
;
2325 mem_cgroup_commit_charge(page
, memcg
, false, false);
2327 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2328 if (dst_vma
->vm_flags
& VM_WRITE
)
2329 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2332 * We don't set the pte dirty if the vma has no
2333 * VM_WRITE permission, so mark the page dirty or it
2334 * could be freed from under us. We could do it
2335 * unconditionally before unlock_page(), but doing it
2336 * only if VM_WRITE is not set is faster.
2338 set_page_dirty(page
);
2341 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2344 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2345 if (unlikely(offset
>= max_off
))
2346 goto out_release_uncharge_unlock
;
2349 if (!pte_none(*dst_pte
))
2350 goto out_release_uncharge_unlock
;
2352 lru_cache_add_anon(page
);
2354 spin_lock(&info
->lock
);
2356 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2357 shmem_recalc_inode(inode
);
2358 spin_unlock(&info
->lock
);
2360 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2361 page_add_file_rmap(page
, false);
2362 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2364 /* No need to invalidate - it was non-present before */
2365 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2366 pte_unmap_unlock(dst_pte
, ptl
);
2371 out_release_uncharge_unlock
:
2372 pte_unmap_unlock(dst_pte
, ptl
);
2373 ClearPageDirty(page
);
2374 delete_from_page_cache(page
);
2375 out_release_uncharge
:
2376 mem_cgroup_cancel_charge(page
, memcg
, false);
2381 shmem_inode_unacct_blocks(inode
, 1);
2385 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2387 struct vm_area_struct
*dst_vma
,
2388 unsigned long dst_addr
,
2389 unsigned long src_addr
,
2390 struct page
**pagep
)
2392 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2393 dst_addr
, src_addr
, false, pagep
);
2396 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2398 struct vm_area_struct
*dst_vma
,
2399 unsigned long dst_addr
)
2401 struct page
*page
= NULL
;
2403 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2404 dst_addr
, 0, true, &page
);
2408 static const struct inode_operations shmem_symlink_inode_operations
;
2409 static const struct inode_operations shmem_short_symlink_operations
;
2411 #ifdef CONFIG_TMPFS_XATTR
2412 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2414 #define shmem_initxattrs NULL
2418 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2419 loff_t pos
, unsigned len
, unsigned flags
,
2420 struct page
**pagep
, void **fsdata
)
2422 struct inode
*inode
= mapping
->host
;
2423 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2424 pgoff_t index
= pos
>> PAGE_SHIFT
;
2426 /* i_mutex is held by caller */
2427 if (unlikely(info
->seals
& (F_SEAL_WRITE
| F_SEAL_GROW
))) {
2428 if (info
->seals
& F_SEAL_WRITE
)
2430 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2434 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2438 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2439 loff_t pos
, unsigned len
, unsigned copied
,
2440 struct page
*page
, void *fsdata
)
2442 struct inode
*inode
= mapping
->host
;
2444 if (pos
+ copied
> inode
->i_size
)
2445 i_size_write(inode
, pos
+ copied
);
2447 if (!PageUptodate(page
)) {
2448 struct page
*head
= compound_head(page
);
2449 if (PageTransCompound(page
)) {
2452 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2453 if (head
+ i
== page
)
2455 clear_highpage(head
+ i
);
2456 flush_dcache_page(head
+ i
);
2459 if (copied
< PAGE_SIZE
) {
2460 unsigned from
= pos
& (PAGE_SIZE
- 1);
2461 zero_user_segments(page
, 0, from
,
2462 from
+ copied
, PAGE_SIZE
);
2464 SetPageUptodate(head
);
2466 set_page_dirty(page
);
2473 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2475 struct file
*file
= iocb
->ki_filp
;
2476 struct inode
*inode
= file_inode(file
);
2477 struct address_space
*mapping
= inode
->i_mapping
;
2479 unsigned long offset
;
2480 enum sgp_type sgp
= SGP_READ
;
2483 loff_t
*ppos
= &iocb
->ki_pos
;
2486 * Might this read be for a stacking filesystem? Then when reading
2487 * holes of a sparse file, we actually need to allocate those pages,
2488 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2490 if (!iter_is_iovec(to
))
2493 index
= *ppos
>> PAGE_SHIFT
;
2494 offset
= *ppos
& ~PAGE_MASK
;
2497 struct page
*page
= NULL
;
2499 unsigned long nr
, ret
;
2500 loff_t i_size
= i_size_read(inode
);
2502 end_index
= i_size
>> PAGE_SHIFT
;
2503 if (index
> end_index
)
2505 if (index
== end_index
) {
2506 nr
= i_size
& ~PAGE_MASK
;
2511 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2513 if (error
== -EINVAL
)
2518 if (sgp
== SGP_CACHE
)
2519 set_page_dirty(page
);
2524 * We must evaluate after, since reads (unlike writes)
2525 * are called without i_mutex protection against truncate
2528 i_size
= i_size_read(inode
);
2529 end_index
= i_size
>> PAGE_SHIFT
;
2530 if (index
== end_index
) {
2531 nr
= i_size
& ~PAGE_MASK
;
2542 * If users can be writing to this page using arbitrary
2543 * virtual addresses, take care about potential aliasing
2544 * before reading the page on the kernel side.
2546 if (mapping_writably_mapped(mapping
))
2547 flush_dcache_page(page
);
2549 * Mark the page accessed if we read the beginning.
2552 mark_page_accessed(page
);
2554 page
= ZERO_PAGE(0);
2559 * Ok, we have the page, and it's up-to-date, so
2560 * now we can copy it to user space...
2562 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2565 index
+= offset
>> PAGE_SHIFT
;
2566 offset
&= ~PAGE_MASK
;
2569 if (!iov_iter_count(to
))
2578 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2579 file_accessed(file
);
2580 return retval
? retval
: error
;
2584 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2586 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2587 pgoff_t index
, pgoff_t end
, int whence
)
2590 struct pagevec pvec
;
2591 pgoff_t indices
[PAGEVEC_SIZE
];
2595 pagevec_init(&pvec
);
2596 pvec
.nr
= 1; /* start small: we may be there already */
2598 pvec
.nr
= find_get_entries(mapping
, index
,
2599 pvec
.nr
, pvec
.pages
, indices
);
2601 if (whence
== SEEK_DATA
)
2605 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2606 if (index
< indices
[i
]) {
2607 if (whence
== SEEK_HOLE
) {
2613 page
= pvec
.pages
[i
];
2614 if (page
&& !radix_tree_exceptional_entry(page
)) {
2615 if (!PageUptodate(page
))
2619 (page
&& whence
== SEEK_DATA
) ||
2620 (!page
&& whence
== SEEK_HOLE
)) {
2625 pagevec_remove_exceptionals(&pvec
);
2626 pagevec_release(&pvec
);
2627 pvec
.nr
= PAGEVEC_SIZE
;
2633 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2635 struct address_space
*mapping
= file
->f_mapping
;
2636 struct inode
*inode
= mapping
->host
;
2640 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2641 return generic_file_llseek_size(file
, offset
, whence
,
2642 MAX_LFS_FILESIZE
, i_size_read(inode
));
2644 /* We're holding i_mutex so we can access i_size directly */
2648 else if (offset
>= inode
->i_size
)
2651 start
= offset
>> PAGE_SHIFT
;
2652 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2653 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2654 new_offset
<<= PAGE_SHIFT
;
2655 if (new_offset
> offset
) {
2656 if (new_offset
< inode
->i_size
)
2657 offset
= new_offset
;
2658 else if (whence
== SEEK_DATA
)
2661 offset
= inode
->i_size
;
2666 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2667 inode_unlock(inode
);
2672 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2673 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2675 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2676 #define LAST_SCAN 4 /* about 150ms max */
2678 static void shmem_tag_pins(struct address_space
*mapping
)
2680 struct radix_tree_iter iter
;
2689 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2690 page
= radix_tree_deref_slot(slot
);
2691 if (!page
|| radix_tree_exception(page
)) {
2692 if (radix_tree_deref_retry(page
)) {
2693 slot
= radix_tree_iter_retry(&iter
);
2696 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2697 spin_lock_irq(&mapping
->tree_lock
);
2698 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2700 spin_unlock_irq(&mapping
->tree_lock
);
2703 if (need_resched()) {
2704 slot
= radix_tree_iter_resume(slot
, &iter
);
2712 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2713 * via get_user_pages(), drivers might have some pending I/O without any active
2714 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2715 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2716 * them to be dropped.
2717 * The caller must guarantee that no new user will acquire writable references
2718 * to those pages to avoid races.
2720 static int shmem_wait_for_pins(struct address_space
*mapping
)
2722 struct radix_tree_iter iter
;
2728 shmem_tag_pins(mapping
);
2731 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2732 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2736 lru_add_drain_all();
2737 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2742 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2743 start
, SHMEM_TAG_PINNED
) {
2745 page
= radix_tree_deref_slot(slot
);
2746 if (radix_tree_exception(page
)) {
2747 if (radix_tree_deref_retry(page
)) {
2748 slot
= radix_tree_iter_retry(&iter
);
2756 page_count(page
) - page_mapcount(page
) != 1) {
2757 if (scan
< LAST_SCAN
)
2758 goto continue_resched
;
2761 * On the last scan, we clean up all those tags
2762 * we inserted; but make a note that we still
2763 * found pages pinned.
2768 spin_lock_irq(&mapping
->tree_lock
);
2769 radix_tree_tag_clear(&mapping
->page_tree
,
2770 iter
.index
, SHMEM_TAG_PINNED
);
2771 spin_unlock_irq(&mapping
->tree_lock
);
2773 if (need_resched()) {
2774 slot
= radix_tree_iter_resume(slot
, &iter
);
2784 #define F_ALL_SEALS (F_SEAL_SEAL | \
2789 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2791 struct inode
*inode
= file_inode(file
);
2792 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2797 * Sealing allows multiple parties to share a shmem-file but restrict
2798 * access to a specific subset of file operations. Seals can only be
2799 * added, but never removed. This way, mutually untrusted parties can
2800 * share common memory regions with a well-defined policy. A malicious
2801 * peer can thus never perform unwanted operations on a shared object.
2803 * Seals are only supported on special shmem-files and always affect
2804 * the whole underlying inode. Once a seal is set, it may prevent some
2805 * kinds of access to the file. Currently, the following seals are
2807 * SEAL_SEAL: Prevent further seals from being set on this file
2808 * SEAL_SHRINK: Prevent the file from shrinking
2809 * SEAL_GROW: Prevent the file from growing
2810 * SEAL_WRITE: Prevent write access to the file
2812 * As we don't require any trust relationship between two parties, we
2813 * must prevent seals from being removed. Therefore, sealing a file
2814 * only adds a given set of seals to the file, it never touches
2815 * existing seals. Furthermore, the "setting seals"-operation can be
2816 * sealed itself, which basically prevents any further seal from being
2819 * Semantics of sealing are only defined on volatile files. Only
2820 * anonymous shmem files support sealing. More importantly, seals are
2821 * never written to disk. Therefore, there's no plan to support it on
2825 if (file
->f_op
!= &shmem_file_operations
)
2827 if (!(file
->f_mode
& FMODE_WRITE
))
2829 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2834 if (info
->seals
& F_SEAL_SEAL
) {
2839 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2840 error
= mapping_deny_writable(file
->f_mapping
);
2844 error
= shmem_wait_for_pins(file
->f_mapping
);
2846 mapping_allow_writable(file
->f_mapping
);
2851 info
->seals
|= seals
;
2855 inode_unlock(inode
);
2858 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2860 int shmem_get_seals(struct file
*file
)
2862 if (file
->f_op
!= &shmem_file_operations
)
2865 return SHMEM_I(file_inode(file
))->seals
;
2867 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2869 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2875 /* disallow upper 32bit */
2879 error
= shmem_add_seals(file
, arg
);
2882 error
= shmem_get_seals(file
);
2892 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2895 struct inode
*inode
= file_inode(file
);
2896 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2897 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2898 struct shmem_falloc shmem_falloc
;
2899 pgoff_t start
, index
, end
;
2902 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2907 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2908 struct address_space
*mapping
= file
->f_mapping
;
2909 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2910 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2911 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2913 /* protected by i_mutex */
2914 if (info
->seals
& F_SEAL_WRITE
) {
2919 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2920 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2921 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2922 spin_lock(&inode
->i_lock
);
2923 inode
->i_private
= &shmem_falloc
;
2924 spin_unlock(&inode
->i_lock
);
2926 if ((u64
)unmap_end
> (u64
)unmap_start
)
2927 unmap_mapping_range(mapping
, unmap_start
,
2928 1 + unmap_end
- unmap_start
, 0);
2929 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2930 /* No need to unmap again: hole-punching leaves COWed pages */
2932 spin_lock(&inode
->i_lock
);
2933 inode
->i_private
= NULL
;
2934 wake_up_all(&shmem_falloc_waitq
);
2935 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2936 spin_unlock(&inode
->i_lock
);
2941 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2942 error
= inode_newsize_ok(inode
, offset
+ len
);
2946 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2951 start
= offset
>> PAGE_SHIFT
;
2952 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2953 /* Try to avoid a swapstorm if len is impossible to satisfy */
2954 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2959 shmem_falloc
.waitq
= NULL
;
2960 shmem_falloc
.start
= start
;
2961 shmem_falloc
.next
= start
;
2962 shmem_falloc
.nr_falloced
= 0;
2963 shmem_falloc
.nr_unswapped
= 0;
2964 spin_lock(&inode
->i_lock
);
2965 inode
->i_private
= &shmem_falloc
;
2966 spin_unlock(&inode
->i_lock
);
2968 for (index
= start
; index
< end
; index
++) {
2972 * Good, the fallocate(2) manpage permits EINTR: we may have
2973 * been interrupted because we are using up too much memory.
2975 if (signal_pending(current
))
2977 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2980 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2982 /* Remove the !PageUptodate pages we added */
2983 if (index
> start
) {
2984 shmem_undo_range(inode
,
2985 (loff_t
)start
<< PAGE_SHIFT
,
2986 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2992 * Inform shmem_writepage() how far we have reached.
2993 * No need for lock or barrier: we have the page lock.
2995 shmem_falloc
.next
++;
2996 if (!PageUptodate(page
))
2997 shmem_falloc
.nr_falloced
++;
3000 * If !PageUptodate, leave it that way so that freeable pages
3001 * can be recognized if we need to rollback on error later.
3002 * But set_page_dirty so that memory pressure will swap rather
3003 * than free the pages we are allocating (and SGP_CACHE pages
3004 * might still be clean: we now need to mark those dirty too).
3006 set_page_dirty(page
);
3012 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
3013 i_size_write(inode
, offset
+ len
);
3014 inode
->i_ctime
= current_time(inode
);
3016 spin_lock(&inode
->i_lock
);
3017 inode
->i_private
= NULL
;
3018 spin_unlock(&inode
->i_lock
);
3020 inode_unlock(inode
);
3024 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
3026 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
3028 buf
->f_type
= TMPFS_MAGIC
;
3029 buf
->f_bsize
= PAGE_SIZE
;
3030 buf
->f_namelen
= NAME_MAX
;
3031 if (sbinfo
->max_blocks
) {
3032 buf
->f_blocks
= sbinfo
->max_blocks
;
3034 buf
->f_bfree
= sbinfo
->max_blocks
-
3035 percpu_counter_sum(&sbinfo
->used_blocks
);
3037 if (sbinfo
->max_inodes
) {
3038 buf
->f_files
= sbinfo
->max_inodes
;
3039 buf
->f_ffree
= sbinfo
->free_inodes
;
3041 /* else leave those fields 0 like simple_statfs */
3046 * File creation. Allocate an inode, and we're done..
3049 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3051 struct inode
*inode
;
3052 int error
= -ENOSPC
;
3054 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
3056 error
= simple_acl_create(dir
, inode
);
3059 error
= security_inode_init_security(inode
, dir
,
3061 shmem_initxattrs
, NULL
);
3062 if (error
&& error
!= -EOPNOTSUPP
)
3066 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3067 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3068 d_instantiate(dentry
, inode
);
3069 dget(dentry
); /* Extra count - pin the dentry in core */
3078 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3080 struct inode
*inode
;
3081 int error
= -ENOSPC
;
3083 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
3085 error
= security_inode_init_security(inode
, dir
,
3087 shmem_initxattrs
, NULL
);
3088 if (error
&& error
!= -EOPNOTSUPP
)
3090 error
= simple_acl_create(dir
, inode
);
3093 d_tmpfile(dentry
, inode
);
3101 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3105 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
3111 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
3114 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
3120 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
3122 struct inode
*inode
= d_inode(old_dentry
);
3126 * No ordinary (disk based) filesystem counts links as inodes;
3127 * but each new link needs a new dentry, pinning lowmem, and
3128 * tmpfs dentries cannot be pruned until they are unlinked.
3130 ret
= shmem_reserve_inode(inode
->i_sb
);
3134 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3135 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3137 ihold(inode
); /* New dentry reference */
3138 dget(dentry
); /* Extra pinning count for the created dentry */
3139 d_instantiate(dentry
, inode
);
3144 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3146 struct inode
*inode
= d_inode(dentry
);
3148 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3149 shmem_free_inode(inode
->i_sb
);
3151 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3152 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3154 dput(dentry
); /* Undo the count from "create" - this does all the work */
3158 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3160 if (!simple_empty(dentry
))
3163 drop_nlink(d_inode(dentry
));
3165 return shmem_unlink(dir
, dentry
);
3168 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3170 bool old_is_dir
= d_is_dir(old_dentry
);
3171 bool new_is_dir
= d_is_dir(new_dentry
);
3173 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3175 drop_nlink(old_dir
);
3178 drop_nlink(new_dir
);
3182 old_dir
->i_ctime
= old_dir
->i_mtime
=
3183 new_dir
->i_ctime
= new_dir
->i_mtime
=
3184 d_inode(old_dentry
)->i_ctime
=
3185 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3190 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3192 struct dentry
*whiteout
;
3195 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3199 error
= shmem_mknod(old_dir
, whiteout
,
3200 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3206 * Cheat and hash the whiteout while the old dentry is still in
3207 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3209 * d_lookup() will consistently find one of them at this point,
3210 * not sure which one, but that isn't even important.
3217 * The VFS layer already does all the dentry stuff for rename,
3218 * we just have to decrement the usage count for the target if
3219 * it exists so that the VFS layer correctly free's it when it
3222 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3224 struct inode
*inode
= d_inode(old_dentry
);
3225 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3227 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3230 if (flags
& RENAME_EXCHANGE
)
3231 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3233 if (!simple_empty(new_dentry
))
3236 if (flags
& RENAME_WHITEOUT
) {
3239 error
= shmem_whiteout(old_dir
, old_dentry
);
3244 if (d_really_is_positive(new_dentry
)) {
3245 (void) shmem_unlink(new_dir
, new_dentry
);
3246 if (they_are_dirs
) {
3247 drop_nlink(d_inode(new_dentry
));
3248 drop_nlink(old_dir
);
3250 } else if (they_are_dirs
) {
3251 drop_nlink(old_dir
);
3255 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3256 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3257 old_dir
->i_ctime
= old_dir
->i_mtime
=
3258 new_dir
->i_ctime
= new_dir
->i_mtime
=
3259 inode
->i_ctime
= current_time(old_dir
);
3263 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3267 struct inode
*inode
;
3270 len
= strlen(symname
) + 1;
3271 if (len
> PAGE_SIZE
)
3272 return -ENAMETOOLONG
;
3274 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
3278 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3279 shmem_initxattrs
, NULL
);
3281 if (error
!= -EOPNOTSUPP
) {
3288 inode
->i_size
= len
-1;
3289 if (len
<= SHORT_SYMLINK_LEN
) {
3290 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3291 if (!inode
->i_link
) {
3295 inode
->i_op
= &shmem_short_symlink_operations
;
3297 inode_nohighmem(inode
);
3298 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3303 inode
->i_mapping
->a_ops
= &shmem_aops
;
3304 inode
->i_op
= &shmem_symlink_inode_operations
;
3305 memcpy(page_address(page
), symname
, len
);
3306 SetPageUptodate(page
);
3307 set_page_dirty(page
);
3311 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3312 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3313 d_instantiate(dentry
, inode
);
3318 static void shmem_put_link(void *arg
)
3320 mark_page_accessed(arg
);
3324 static const char *shmem_get_link(struct dentry
*dentry
,
3325 struct inode
*inode
,
3326 struct delayed_call
*done
)
3328 struct page
*page
= NULL
;
3331 page
= find_get_page(inode
->i_mapping
, 0);
3333 return ERR_PTR(-ECHILD
);
3334 if (!PageUptodate(page
)) {
3336 return ERR_PTR(-ECHILD
);
3339 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3341 return ERR_PTR(error
);
3344 set_delayed_call(done
, shmem_put_link
, page
);
3345 return page_address(page
);
3348 #ifdef CONFIG_TMPFS_XATTR
3350 * Superblocks without xattr inode operations may get some security.* xattr
3351 * support from the LSM "for free". As soon as we have any other xattrs
3352 * like ACLs, we also need to implement the security.* handlers at
3353 * filesystem level, though.
3357 * Callback for security_inode_init_security() for acquiring xattrs.
3359 static int shmem_initxattrs(struct inode
*inode
,
3360 const struct xattr
*xattr_array
,
3363 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3364 const struct xattr
*xattr
;
3365 struct simple_xattr
*new_xattr
;
3368 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3369 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3373 len
= strlen(xattr
->name
) + 1;
3374 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3376 if (!new_xattr
->name
) {
3381 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3382 XATTR_SECURITY_PREFIX_LEN
);
3383 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3386 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3392 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3393 struct dentry
*unused
, struct inode
*inode
,
3394 const char *name
, void *buffer
, size_t size
)
3396 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3398 name
= xattr_full_name(handler
, name
);
3399 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3402 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3403 struct dentry
*unused
, struct inode
*inode
,
3404 const char *name
, const void *value
,
3405 size_t size
, int flags
)
3407 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3409 name
= xattr_full_name(handler
, name
);
3410 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3413 static const struct xattr_handler shmem_security_xattr_handler
= {
3414 .prefix
= XATTR_SECURITY_PREFIX
,
3415 .get
= shmem_xattr_handler_get
,
3416 .set
= shmem_xattr_handler_set
,
3419 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3420 .prefix
= XATTR_TRUSTED_PREFIX
,
3421 .get
= shmem_xattr_handler_get
,
3422 .set
= shmem_xattr_handler_set
,
3425 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3426 #ifdef CONFIG_TMPFS_POSIX_ACL
3427 &posix_acl_access_xattr_handler
,
3428 &posix_acl_default_xattr_handler
,
3430 &shmem_security_xattr_handler
,
3431 &shmem_trusted_xattr_handler
,
3435 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3437 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3438 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3440 #endif /* CONFIG_TMPFS_XATTR */
3442 static const struct inode_operations shmem_short_symlink_operations
= {
3443 .get_link
= simple_get_link
,
3444 #ifdef CONFIG_TMPFS_XATTR
3445 .listxattr
= shmem_listxattr
,
3449 static const struct inode_operations shmem_symlink_inode_operations
= {
3450 .get_link
= shmem_get_link
,
3451 #ifdef CONFIG_TMPFS_XATTR
3452 .listxattr
= shmem_listxattr
,
3456 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3458 return ERR_PTR(-ESTALE
);
3461 static int shmem_match(struct inode
*ino
, void *vfh
)
3465 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3468 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3469 struct fid
*fid
, int fh_len
, int fh_type
)
3471 struct inode
*inode
;
3472 struct dentry
*dentry
= NULL
;
3479 inode
= ilookup5(sb
, inum
, shmem_match
, fid
->raw
);
3481 dentry
= d_find_alias(inode
);
3488 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3489 struct inode
*parent
)
3493 return FILEID_INVALID
;
3496 fh
[0] = inode
->i_generation
;
3497 fh
[1] = inode
->i_ino
;
3503 static const struct export_operations shmem_export_ops
= {
3504 .get_parent
= shmem_get_parent
,
3505 .encode_fh
= shmem_encode_fh
,
3506 .fh_to_dentry
= shmem_fh_to_dentry
,
3509 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3512 char *this_char
, *value
, *rest
;
3513 struct mempolicy
*mpol
= NULL
;
3517 while (options
!= NULL
) {
3518 this_char
= options
;
3521 * NUL-terminate this option: unfortunately,
3522 * mount options form a comma-separated list,
3523 * but mpol's nodelist may also contain commas.
3525 options
= strchr(options
, ',');
3526 if (options
== NULL
)
3529 if (!isdigit(*options
)) {
3536 if ((value
= strchr(this_char
,'=')) != NULL
) {
3539 pr_err("tmpfs: No value for mount option '%s'\n",
3544 if (!strcmp(this_char
,"size")) {
3545 unsigned long long size
;
3546 size
= memparse(value
,&rest
);
3548 size
<<= PAGE_SHIFT
;
3549 size
*= totalram_pages
;
3555 sbinfo
->max_blocks
=
3556 DIV_ROUND_UP(size
, PAGE_SIZE
);
3557 } else if (!strcmp(this_char
,"nr_blocks")) {
3558 sbinfo
->max_blocks
= memparse(value
, &rest
);
3561 } else if (!strcmp(this_char
,"nr_inodes")) {
3562 sbinfo
->max_inodes
= memparse(value
, &rest
);
3563 if (*rest
|| sbinfo
->max_inodes
< 2)
3565 } else if (!strcmp(this_char
,"mode")) {
3568 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3571 } else if (!strcmp(this_char
,"uid")) {
3574 uid
= simple_strtoul(value
, &rest
, 0);
3577 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3578 if (!uid_valid(sbinfo
->uid
))
3580 } else if (!strcmp(this_char
,"gid")) {
3583 gid
= simple_strtoul(value
, &rest
, 0);
3586 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3587 if (!gid_valid(sbinfo
->gid
))
3589 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3590 } else if (!strcmp(this_char
, "huge")) {
3592 huge
= shmem_parse_huge(value
);
3595 if (!has_transparent_hugepage() &&
3596 huge
!= SHMEM_HUGE_NEVER
)
3598 sbinfo
->huge
= huge
;
3601 } else if (!strcmp(this_char
,"mpol")) {
3604 if (mpol_parse_str(value
, &mpol
))
3608 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3612 sbinfo
->mpol
= mpol
;
3616 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3624 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3626 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3627 struct shmem_sb_info config
= *sbinfo
;
3629 int error
= -EINVAL
;
3632 if (shmem_parse_options(data
, &config
, true))
3635 spin_lock(&sbinfo
->stat_lock
);
3636 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3637 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3639 if (config
.max_inodes
< inodes
)
3642 * Those tests disallow limited->unlimited while any are in use;
3643 * but we must separately disallow unlimited->limited, because
3644 * in that case we have no record of how much is already in use.
3646 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3648 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3652 sbinfo
->huge
= config
.huge
;
3653 sbinfo
->max_blocks
= config
.max_blocks
;
3654 sbinfo
->max_inodes
= config
.max_inodes
;
3655 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3658 * Preserve previous mempolicy unless mpol remount option was specified.
3661 mpol_put(sbinfo
->mpol
);
3662 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3665 spin_unlock(&sbinfo
->stat_lock
);
3669 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3671 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3673 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3674 seq_printf(seq
, ",size=%luk",
3675 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3676 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3677 seq_printf(seq
, ",nr_inodes=%d", sbinfo
->max_inodes
);
3678 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3679 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3680 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3681 seq_printf(seq
, ",uid=%u",
3682 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3683 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3684 seq_printf(seq
, ",gid=%u",
3685 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3686 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3687 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3689 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3691 shmem_show_mpol(seq
, sbinfo
->mpol
);
3695 #define MFD_NAME_PREFIX "memfd:"
3696 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3697 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3699 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
3701 SYSCALL_DEFINE2(memfd_create
,
3702 const char __user
*, uname
,
3703 unsigned int, flags
)
3705 struct shmem_inode_info
*info
;
3711 if (!(flags
& MFD_HUGETLB
)) {
3712 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3715 /* Sealing not supported in hugetlbfs (MFD_HUGETLB) */
3716 if (flags
& MFD_ALLOW_SEALING
)
3718 /* Allow huge page size encoding in flags. */
3719 if (flags
& ~(unsigned int)(MFD_ALL_FLAGS
|
3720 (MFD_HUGE_MASK
<< MFD_HUGE_SHIFT
)))
3724 /* length includes terminating zero */
3725 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3728 if (len
> MFD_NAME_MAX_LEN
+ 1)
3731 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_KERNEL
);
3735 strcpy(name
, MFD_NAME_PREFIX
);
3736 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3741 /* terminating-zero may have changed after strnlen_user() returned */
3742 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3747 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3753 if (flags
& MFD_HUGETLB
) {
3754 struct user_struct
*user
= NULL
;
3756 file
= hugetlb_file_setup(name
, 0, VM_NORESERVE
, &user
,
3757 HUGETLB_ANONHUGE_INODE
,
3758 (flags
>> MFD_HUGE_SHIFT
) &
3761 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3763 error
= PTR_ERR(file
);
3766 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3767 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3769 if (flags
& MFD_ALLOW_SEALING
) {
3771 * flags check at beginning of function ensures
3772 * this is not a hugetlbfs (MFD_HUGETLB) file.
3774 info
= SHMEM_I(file_inode(file
));
3775 info
->seals
&= ~F_SEAL_SEAL
;
3778 fd_install(fd
, file
);
3789 #endif /* CONFIG_TMPFS */
3791 static void shmem_put_super(struct super_block
*sb
)
3793 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3795 if (!sbinfo
->idr_nouse
)
3796 idr_destroy(&sbinfo
->idr
);
3797 percpu_counter_destroy(&sbinfo
->used_blocks
);
3798 mpol_put(sbinfo
->mpol
);
3800 sb
->s_fs_info
= NULL
;
3803 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3805 struct inode
*inode
;
3806 struct shmem_sb_info
*sbinfo
;
3809 /* Round up to L1_CACHE_BYTES to resist false sharing */
3810 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3811 L1_CACHE_BYTES
), GFP_KERNEL
);
3815 mutex_init(&sbinfo
->idr_lock
);
3816 idr_init(&sbinfo
->idr
);
3817 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3818 sbinfo
->uid
= current_fsuid();
3819 sbinfo
->gid
= current_fsgid();
3820 sb
->s_fs_info
= sbinfo
;
3824 * Per default we only allow half of the physical ram per
3825 * tmpfs instance, limiting inodes to one per page of lowmem;
3826 * but the internal instance is left unlimited.
3828 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3829 sbinfo
->max_blocks
= shmem_default_max_blocks();
3830 sbinfo
->max_inodes
= shmem_default_max_inodes();
3831 if (shmem_parse_options(data
, sbinfo
, false)) {
3836 sb
->s_flags
|= SB_NOUSER
;
3838 sb
->s_export_op
= &shmem_export_ops
;
3839 sb
->s_flags
|= SB_NOSEC
;
3841 sb
->s_flags
|= SB_NOUSER
;
3844 spin_lock_init(&sbinfo
->stat_lock
);
3845 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3847 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3848 spin_lock_init(&sbinfo
->shrinklist_lock
);
3849 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3851 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3852 sb
->s_blocksize
= PAGE_SIZE
;
3853 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3854 sb
->s_magic
= TMPFS_MAGIC
;
3855 sb
->s_op
= &shmem_ops
;
3856 sb
->s_time_gran
= 1;
3857 #ifdef CONFIG_TMPFS_XATTR
3858 sb
->s_xattr
= shmem_xattr_handlers
;
3860 #ifdef CONFIG_TMPFS_POSIX_ACL
3861 sb
->s_flags
|= SB_POSIXACL
;
3863 uuid_gen(&sb
->s_uuid
);
3865 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3868 inode
->i_uid
= sbinfo
->uid
;
3869 inode
->i_gid
= sbinfo
->gid
;
3870 sb
->s_root
= d_make_root(inode
);
3876 shmem_put_super(sb
);
3880 static struct kmem_cache
*shmem_inode_cachep
;
3882 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3884 struct shmem_inode_info
*info
;
3885 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3888 return &info
->vfs_inode
;
3891 static void shmem_destroy_callback(struct rcu_head
*head
)
3893 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3894 if (S_ISLNK(inode
->i_mode
))
3895 kfree(inode
->i_link
);
3896 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3899 static void shmem_destroy_inode(struct inode
*inode
)
3901 if (S_ISREG(inode
->i_mode
))
3902 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3903 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3906 static void shmem_init_inode(void *foo
)
3908 struct shmem_inode_info
*info
= foo
;
3909 inode_init_once(&info
->vfs_inode
);
3912 static void shmem_init_inodecache(void)
3914 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3915 sizeof(struct shmem_inode_info
),
3916 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3919 static void shmem_destroy_inodecache(void)
3921 kmem_cache_destroy(shmem_inode_cachep
);
3924 static __init
void shmem_no_idr(struct super_block
*sb
)
3926 struct shmem_sb_info
*sbinfo
;
3928 sbinfo
= SHMEM_SB(sb
);
3929 sbinfo
->idr_nouse
= true;
3930 idr_destroy(&sbinfo
->idr
);
3933 static const struct address_space_operations shmem_aops
= {
3934 .writepage
= shmem_writepage
,
3935 .set_page_dirty
= __set_page_dirty_no_writeback
,
3937 .write_begin
= shmem_write_begin
,
3938 .write_end
= shmem_write_end
,
3940 #ifdef CONFIG_MIGRATION
3941 .migratepage
= migrate_page
,
3943 .error_remove_page
= generic_error_remove_page
,
3946 static const struct file_operations shmem_file_operations
= {
3948 .get_unmapped_area
= shmem_get_unmapped_area
,
3950 .llseek
= shmem_file_llseek
,
3951 .read_iter
= shmem_file_read_iter
,
3952 .write_iter
= generic_file_write_iter
,
3953 .fsync
= noop_fsync
,
3954 .splice_read
= generic_file_splice_read
,
3955 .splice_write
= iter_file_splice_write
,
3956 .fallocate
= shmem_fallocate
,
3960 static const struct inode_operations shmem_inode_operations
= {
3961 .getattr
= shmem_getattr
,
3962 .setattr
= shmem_setattr
,
3963 #ifdef CONFIG_TMPFS_XATTR
3964 .listxattr
= shmem_listxattr
,
3965 .set_acl
= simple_set_acl
,
3969 static const struct inode_operations shmem_dir_inode_operations
= {
3971 .create
= shmem_create
,
3972 .lookup
= simple_lookup
,
3974 .unlink
= shmem_unlink
,
3975 .symlink
= shmem_symlink
,
3976 .mkdir
= shmem_mkdir
,
3977 .rmdir
= shmem_rmdir
,
3978 .mknod
= shmem_mknod
,
3979 .rename
= shmem_rename2
,
3980 .tmpfile
= shmem_tmpfile
,
3982 #ifdef CONFIG_TMPFS_XATTR
3983 .listxattr
= shmem_listxattr
,
3985 #ifdef CONFIG_TMPFS_POSIX_ACL
3986 .setattr
= shmem_setattr
,
3987 .set_acl
= simple_set_acl
,
3991 static const struct inode_operations shmem_special_inode_operations
= {
3992 #ifdef CONFIG_TMPFS_XATTR
3993 .listxattr
= shmem_listxattr
,
3995 #ifdef CONFIG_TMPFS_POSIX_ACL
3996 .setattr
= shmem_setattr
,
3997 .set_acl
= simple_set_acl
,
4001 static const struct super_operations shmem_ops
= {
4002 .alloc_inode
= shmem_alloc_inode
,
4003 .destroy_inode
= shmem_destroy_inode
,
4005 .statfs
= shmem_statfs
,
4006 .remount_fs
= shmem_remount_fs
,
4007 .show_options
= shmem_show_options
,
4009 .evict_inode
= shmem_evict_inode
,
4010 .drop_inode
= generic_delete_inode
,
4011 .put_super
= shmem_put_super
,
4012 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4013 .nr_cached_objects
= shmem_unused_huge_count
,
4014 .free_cached_objects
= shmem_unused_huge_scan
,
4018 static const struct vm_operations_struct shmem_vm_ops
= {
4019 .fault
= shmem_fault
,
4020 .map_pages
= filemap_map_pages
,
4022 .set_policy
= shmem_set_policy
,
4023 .get_policy
= shmem_get_policy
,
4027 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
4028 int flags
, const char *dev_name
, void *data
)
4030 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
4033 static struct file_system_type shmem_fs_type
= {
4034 .owner
= THIS_MODULE
,
4036 .mount
= shmem_mount
,
4037 .kill_sb
= kill_litter_super
,
4038 .fs_flags
= FS_USERNS_MOUNT
,
4041 int __init
shmem_init(void)
4045 /* If rootfs called this, don't re-init */
4046 if (shmem_inode_cachep
)
4049 shmem_init_inodecache();
4051 error
= register_filesystem(&shmem_fs_type
);
4053 pr_err("Could not register tmpfs\n");
4057 shm_mnt
= kern_mount(&shmem_fs_type
);
4058 if (IS_ERR(shm_mnt
)) {
4059 error
= PTR_ERR(shm_mnt
);
4060 pr_err("Could not kern_mount tmpfs\n");
4063 shmem_no_idr(shm_mnt
->mnt_sb
);
4065 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4066 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
4067 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4069 shmem_huge
= 0; /* just in case it was patched */
4074 unregister_filesystem(&shmem_fs_type
);
4076 shmem_destroy_inodecache();
4077 shm_mnt
= ERR_PTR(error
);
4081 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
4082 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4083 struct kobj_attribute
*attr
, char *buf
)
4087 SHMEM_HUGE_WITHIN_SIZE
,
4095 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
4096 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
4098 count
+= sprintf(buf
+ count
, fmt
,
4099 shmem_format_huge(values
[i
]));
4101 buf
[count
- 1] = '\n';
4105 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4106 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4111 if (count
+ 1 > sizeof(tmp
))
4113 memcpy(tmp
, buf
, count
);
4115 if (count
&& tmp
[count
- 1] == '\n')
4116 tmp
[count
- 1] = '\0';
4118 huge
= shmem_parse_huge(tmp
);
4119 if (huge
== -EINVAL
)
4121 if (!has_transparent_hugepage() &&
4122 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4126 if (shmem_huge
> SHMEM_HUGE_DENY
)
4127 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4131 struct kobj_attribute shmem_enabled_attr
=
4132 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4133 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4135 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4136 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4138 struct inode
*inode
= file_inode(vma
->vm_file
);
4139 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4143 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4145 if (shmem_huge
== SHMEM_HUGE_DENY
)
4147 switch (sbinfo
->huge
) {
4148 case SHMEM_HUGE_NEVER
:
4150 case SHMEM_HUGE_ALWAYS
:
4152 case SHMEM_HUGE_WITHIN_SIZE
:
4153 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4154 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4155 if (i_size
>= HPAGE_PMD_SIZE
&&
4156 i_size
>> PAGE_SHIFT
>= off
)
4159 case SHMEM_HUGE_ADVISE
:
4160 /* TODO: implement fadvise() hints */
4161 return (vma
->vm_flags
& VM_HUGEPAGE
);
4167 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4169 #else /* !CONFIG_SHMEM */
4172 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4174 * This is intended for small system where the benefits of the full
4175 * shmem code (swap-backed and resource-limited) are outweighed by
4176 * their complexity. On systems without swap this code should be
4177 * effectively equivalent, but much lighter weight.
4180 static struct file_system_type shmem_fs_type
= {
4182 .mount
= ramfs_mount
,
4183 .kill_sb
= kill_litter_super
,
4184 .fs_flags
= FS_USERNS_MOUNT
,
4187 int __init
shmem_init(void)
4189 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4191 shm_mnt
= kern_mount(&shmem_fs_type
);
4192 BUG_ON(IS_ERR(shm_mnt
));
4197 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
4202 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4207 void shmem_unlock_mapping(struct address_space
*mapping
)
4212 unsigned long shmem_get_unmapped_area(struct file
*file
,
4213 unsigned long addr
, unsigned long len
,
4214 unsigned long pgoff
, unsigned long flags
)
4216 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4220 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4222 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4224 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4226 #define shmem_vm_ops generic_file_vm_ops
4227 #define shmem_file_operations ramfs_file_operations
4228 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4229 #define shmem_acct_size(flags, size) 0
4230 #define shmem_unacct_size(flags, size) do {} while (0)
4232 #endif /* CONFIG_SHMEM */
4236 static const struct dentry_operations anon_ops
= {
4237 .d_dname
= simple_dname
4240 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4241 unsigned long flags
, unsigned int i_flags
)
4244 struct inode
*inode
;
4246 struct super_block
*sb
;
4250 return ERR_CAST(mnt
);
4252 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4253 return ERR_PTR(-EINVAL
);
4255 if (shmem_acct_size(flags
, size
))
4256 return ERR_PTR(-ENOMEM
);
4258 res
= ERR_PTR(-ENOMEM
);
4260 this.len
= strlen(name
);
4261 this.hash
= 0; /* will go */
4263 path
.mnt
= mntget(mnt
);
4264 path
.dentry
= d_alloc_pseudo(sb
, &this);
4267 d_set_d_op(path
.dentry
, &anon_ops
);
4269 res
= ERR_PTR(-ENOSPC
);
4270 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
4274 inode
->i_flags
|= i_flags
;
4275 d_instantiate(path
.dentry
, inode
);
4276 inode
->i_size
= size
;
4277 clear_nlink(inode
); /* It is unlinked */
4278 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4282 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
4283 &shmem_file_operations
);
4290 shmem_unacct_size(flags
, size
);
4297 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4298 * kernel internal. There will be NO LSM permission checks against the
4299 * underlying inode. So users of this interface must do LSM checks at a
4300 * higher layer. The users are the big_key and shm implementations. LSM
4301 * checks are provided at the key or shm level rather than the inode.
4302 * @name: name for dentry (to be seen in /proc/<pid>/maps
4303 * @size: size to be set for the file
4304 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4306 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4308 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4312 * shmem_file_setup - get an unlinked file living in tmpfs
4313 * @name: name for dentry (to be seen in /proc/<pid>/maps
4314 * @size: size to be set for the file
4315 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4317 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4319 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4321 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4324 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4325 * @mnt: the tmpfs mount where the file will be created
4326 * @name: name for dentry (to be seen in /proc/<pid>/maps
4327 * @size: size to be set for the file
4328 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4330 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4331 loff_t size
, unsigned long flags
)
4333 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4335 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4338 * shmem_zero_setup - setup a shared anonymous mapping
4339 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4341 int shmem_zero_setup(struct vm_area_struct
*vma
)
4344 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4347 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4348 * between XFS directory reading and selinux: since this file is only
4349 * accessible to the user through its mapping, use S_PRIVATE flag to
4350 * bypass file security, in the same way as shmem_kernel_file_setup().
4352 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4354 return PTR_ERR(file
);
4358 vma
->vm_file
= file
;
4359 vma
->vm_ops
= &shmem_vm_ops
;
4361 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4362 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4363 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4364 khugepaged_enter(vma
, vma
->vm_flags
);
4371 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4372 * @mapping: the page's address_space
4373 * @index: the page index
4374 * @gfp: the page allocator flags to use if allocating
4376 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4377 * with any new page allocations done using the specified allocation flags.
4378 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4379 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4380 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4382 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4383 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4385 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4386 pgoff_t index
, gfp_t gfp
)
4389 struct inode
*inode
= mapping
->host
;
4393 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4394 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4395 gfp
, NULL
, NULL
, NULL
);
4397 page
= ERR_PTR(error
);
4403 * The tiny !SHMEM case uses ramfs without swap
4405 return read_cache_page_gfp(mapping
, index
, gfp
);
4408 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);