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 unsigned long shmem_default_max_inodes(void)
116 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
120 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
121 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
122 struct shmem_inode_info
*info
, pgoff_t index
);
123 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
124 struct page
**pagep
, enum sgp_type sgp
,
125 gfp_t gfp
, struct vm_area_struct
*vma
,
126 struct vm_fault
*vmf
, int *fault_type
);
128 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
129 struct page
**pagep
, enum sgp_type sgp
)
131 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
132 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
135 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
137 return sb
->s_fs_info
;
141 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
142 * for shared memory and for shared anonymous (/dev/zero) mappings
143 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
144 * consistent with the pre-accounting of private mappings ...
146 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
148 return (flags
& VM_NORESERVE
) ?
149 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
152 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
154 if (!(flags
& VM_NORESERVE
))
155 vm_unacct_memory(VM_ACCT(size
));
158 static inline int shmem_reacct_size(unsigned long flags
,
159 loff_t oldsize
, loff_t newsize
)
161 if (!(flags
& VM_NORESERVE
)) {
162 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
163 return security_vm_enough_memory_mm(current
->mm
,
164 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
165 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
166 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
172 * ... whereas tmpfs objects are accounted incrementally as
173 * pages are allocated, in order to allow large sparse files.
174 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
175 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
177 static inline int shmem_acct_block(unsigned long flags
, long pages
)
179 if (!(flags
& VM_NORESERVE
))
182 return security_vm_enough_memory_mm(current
->mm
,
183 pages
* VM_ACCT(PAGE_SIZE
));
186 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
188 if (flags
& VM_NORESERVE
)
189 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
192 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
194 struct shmem_inode_info
*info
= SHMEM_I(inode
);
195 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
197 if (shmem_acct_block(info
->flags
, pages
))
200 if (sbinfo
->max_blocks
) {
201 if (percpu_counter_compare(&sbinfo
->used_blocks
,
202 sbinfo
->max_blocks
- pages
) > 0)
204 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
210 shmem_unacct_blocks(info
->flags
, pages
);
214 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
216 struct shmem_inode_info
*info
= SHMEM_I(inode
);
217 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
219 if (sbinfo
->max_blocks
)
220 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
221 shmem_unacct_blocks(info
->flags
, pages
);
224 static const struct super_operations shmem_ops
;
225 static const struct address_space_operations shmem_aops
;
226 static const struct file_operations shmem_file_operations
;
227 static const struct inode_operations shmem_inode_operations
;
228 static const struct inode_operations shmem_dir_inode_operations
;
229 static const struct inode_operations shmem_special_inode_operations
;
230 static const struct vm_operations_struct shmem_vm_ops
;
231 static struct file_system_type shmem_fs_type
;
233 bool vma_is_shmem(struct vm_area_struct
*vma
)
235 return vma
->vm_ops
== &shmem_vm_ops
;
238 static LIST_HEAD(shmem_swaplist
);
239 static DEFINE_MUTEX(shmem_swaplist_mutex
);
241 static int shmem_reserve_inode(struct super_block
*sb
)
243 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
244 if (sbinfo
->max_inodes
) {
245 spin_lock(&sbinfo
->stat_lock
);
246 if (!sbinfo
->free_inodes
) {
247 spin_unlock(&sbinfo
->stat_lock
);
250 sbinfo
->free_inodes
--;
251 spin_unlock(&sbinfo
->stat_lock
);
256 static void shmem_free_inode(struct super_block
*sb
)
258 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
259 if (sbinfo
->max_inodes
) {
260 spin_lock(&sbinfo
->stat_lock
);
261 sbinfo
->free_inodes
++;
262 spin_unlock(&sbinfo
->stat_lock
);
267 * shmem_recalc_inode - recalculate the block usage of an inode
268 * @inode: inode to recalc
270 * We have to calculate the free blocks since the mm can drop
271 * undirtied hole pages behind our back.
273 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
274 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
276 * It has to be called with the spinlock held.
278 static void shmem_recalc_inode(struct inode
*inode
)
280 struct shmem_inode_info
*info
= SHMEM_I(inode
);
283 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
285 info
->alloced
-= freed
;
286 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
287 shmem_inode_unacct_blocks(inode
, freed
);
291 bool shmem_charge(struct inode
*inode
, long pages
)
293 struct shmem_inode_info
*info
= SHMEM_I(inode
);
296 if (!shmem_inode_acct_block(inode
, pages
))
299 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
300 inode
->i_mapping
->nrpages
+= pages
;
302 spin_lock_irqsave(&info
->lock
, flags
);
303 info
->alloced
+= pages
;
304 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
305 shmem_recalc_inode(inode
);
306 spin_unlock_irqrestore(&info
->lock
, flags
);
311 void shmem_uncharge(struct inode
*inode
, long pages
)
313 struct shmem_inode_info
*info
= SHMEM_I(inode
);
316 /* nrpages adjustment done by __delete_from_page_cache() or caller */
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
,
345 replacement
, NULL
, NULL
);
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
];
763 pagevec_init(&pvec
, 0);
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 */
806 pagevec_init(&pvec
, 0);
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 shmem_free_inode(inode
->i_sb
);
1102 static unsigned long find_swap_entry(struct radix_tree_root
*root
, void *item
)
1104 struct radix_tree_iter iter
;
1106 unsigned long found
= -1;
1107 unsigned int checked
= 0;
1110 radix_tree_for_each_slot(slot
, root
, &iter
, 0) {
1111 if (*slot
== item
) {
1116 if ((checked
% 4096) != 0)
1118 slot
= radix_tree_iter_resume(slot
, &iter
);
1127 * If swap found in inode, free it and move page from swapcache to filecache.
1129 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
1130 swp_entry_t swap
, struct page
**pagep
)
1132 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1138 radswap
= swp_to_radix_entry(swap
);
1139 index
= find_swap_entry(&mapping
->page_tree
, radswap
);
1141 return -EAGAIN
; /* tell shmem_unuse we found nothing */
1144 * Move _head_ to start search for next from here.
1145 * But be careful: shmem_evict_inode checks list_empty without taking
1146 * mutex, and there's an instant in list_move_tail when info->swaplist
1147 * would appear empty, if it were the only one on shmem_swaplist.
1149 if (shmem_swaplist
.next
!= &info
->swaplist
)
1150 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
1152 gfp
= mapping_gfp_mask(mapping
);
1153 if (shmem_should_replace_page(*pagep
, gfp
)) {
1154 mutex_unlock(&shmem_swaplist_mutex
);
1155 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
1156 mutex_lock(&shmem_swaplist_mutex
);
1158 * We needed to drop mutex to make that restrictive page
1159 * allocation, but the inode might have been freed while we
1160 * dropped it: although a racing shmem_evict_inode() cannot
1161 * complete without emptying the radix_tree, our page lock
1162 * on this swapcache page is not enough to prevent that -
1163 * free_swap_and_cache() of our swap entry will only
1164 * trylock_page(), removing swap from radix_tree whatever.
1166 * We must not proceed to shmem_add_to_page_cache() if the
1167 * inode has been freed, but of course we cannot rely on
1168 * inode or mapping or info to check that. However, we can
1169 * safely check if our swap entry is still in use (and here
1170 * it can't have got reused for another page): if it's still
1171 * in use, then the inode cannot have been freed yet, and we
1172 * can safely proceed (if it's no longer in use, that tells
1173 * nothing about the inode, but we don't need to unuse swap).
1175 if (!page_swapcount(*pagep
))
1180 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1181 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1182 * beneath us (pagelock doesn't help until the page is in pagecache).
1185 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
1187 if (error
!= -ENOMEM
) {
1189 * Truncation and eviction use free_swap_and_cache(), which
1190 * only does trylock page: if we raced, best clean up here.
1192 delete_from_swap_cache(*pagep
);
1193 set_page_dirty(*pagep
);
1195 spin_lock_irq(&info
->lock
);
1197 spin_unlock_irq(&info
->lock
);
1205 * Search through swapped inodes to find and replace swap by page.
1207 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1209 struct list_head
*this, *next
;
1210 struct shmem_inode_info
*info
;
1211 struct mem_cgroup
*memcg
;
1215 * There's a faint possibility that swap page was replaced before
1216 * caller locked it: caller will come back later with the right page.
1218 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1222 * Charge page using GFP_KERNEL while we can wait, before taking
1223 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1224 * Charged back to the user (not to caller) when swap account is used.
1226 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1230 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1233 mutex_lock(&shmem_swaplist_mutex
);
1234 list_for_each_safe(this, next
, &shmem_swaplist
) {
1235 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1237 error
= shmem_unuse_inode(info
, swap
, &page
);
1239 list_del_init(&info
->swaplist
);
1241 if (error
!= -EAGAIN
)
1243 /* found nothing in this: move on to search the next */
1245 mutex_unlock(&shmem_swaplist_mutex
);
1248 if (error
!= -ENOMEM
)
1250 mem_cgroup_cancel_charge(page
, memcg
, false);
1252 mem_cgroup_commit_charge(page
, memcg
, true, false);
1260 * Move the page from the page cache to the swap cache.
1262 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1264 struct shmem_inode_info
*info
;
1265 struct address_space
*mapping
;
1266 struct inode
*inode
;
1270 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1271 BUG_ON(!PageLocked(page
));
1272 mapping
= page
->mapping
;
1273 index
= page
->index
;
1274 inode
= mapping
->host
;
1275 info
= SHMEM_I(inode
);
1276 if (info
->flags
& VM_LOCKED
)
1278 if (!total_swap_pages
)
1282 * Our capabilities prevent regular writeback or sync from ever calling
1283 * shmem_writepage; but a stacking filesystem might use ->writepage of
1284 * its underlying filesystem, in which case tmpfs should write out to
1285 * swap only in response to memory pressure, and not for the writeback
1288 if (!wbc
->for_reclaim
) {
1289 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1294 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1295 * value into swapfile.c, the only way we can correctly account for a
1296 * fallocated page arriving here is now to initialize it and write it.
1298 * That's okay for a page already fallocated earlier, but if we have
1299 * not yet completed the fallocation, then (a) we want to keep track
1300 * of this page in case we have to undo it, and (b) it may not be a
1301 * good idea to continue anyway, once we're pushing into swap. So
1302 * reactivate the page, and let shmem_fallocate() quit when too many.
1304 if (!PageUptodate(page
)) {
1305 if (inode
->i_private
) {
1306 struct shmem_falloc
*shmem_falloc
;
1307 spin_lock(&inode
->i_lock
);
1308 shmem_falloc
= inode
->i_private
;
1310 !shmem_falloc
->waitq
&&
1311 index
>= shmem_falloc
->start
&&
1312 index
< shmem_falloc
->next
)
1313 shmem_falloc
->nr_unswapped
++;
1315 shmem_falloc
= NULL
;
1316 spin_unlock(&inode
->i_lock
);
1320 clear_highpage(page
);
1321 flush_dcache_page(page
);
1322 SetPageUptodate(page
);
1325 swap
= get_swap_page(page
);
1329 if (mem_cgroup_try_charge_swap(page
, swap
))
1333 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1334 * if it's not already there. Do it now before the page is
1335 * moved to swap cache, when its pagelock no longer protects
1336 * the inode from eviction. But don't unlock the mutex until
1337 * we've incremented swapped, because shmem_unuse_inode() will
1338 * prune a !swapped inode from the swaplist under this mutex.
1340 mutex_lock(&shmem_swaplist_mutex
);
1341 if (list_empty(&info
->swaplist
))
1342 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1344 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1345 spin_lock_irq(&info
->lock
);
1346 shmem_recalc_inode(inode
);
1348 spin_unlock_irq(&info
->lock
);
1350 swap_shmem_alloc(swap
);
1351 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1353 mutex_unlock(&shmem_swaplist_mutex
);
1354 BUG_ON(page_mapped(page
));
1355 swap_writepage(page
, wbc
);
1359 mutex_unlock(&shmem_swaplist_mutex
);
1361 put_swap_page(page
, swap
);
1363 set_page_dirty(page
);
1364 if (wbc
->for_reclaim
)
1365 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1370 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1371 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1375 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1376 return; /* show nothing */
1378 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1380 seq_printf(seq
, ",mpol=%s", buffer
);
1383 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1385 struct mempolicy
*mpol
= NULL
;
1387 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1388 mpol
= sbinfo
->mpol
;
1390 spin_unlock(&sbinfo
->stat_lock
);
1394 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1395 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1398 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1402 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1404 #define vm_policy vm_private_data
1407 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1408 struct shmem_inode_info
*info
, pgoff_t index
)
1410 /* Create a pseudo vma that just contains the policy */
1412 /* Bias interleave by inode number to distribute better across nodes */
1413 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1415 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1418 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1420 /* Drop reference taken by mpol_shared_policy_lookup() */
1421 mpol_cond_put(vma
->vm_policy
);
1424 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1425 struct shmem_inode_info
*info
, pgoff_t index
)
1427 struct vm_area_struct pvma
;
1430 shmem_pseudo_vma_init(&pvma
, info
, index
);
1431 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1432 shmem_pseudo_vma_destroy(&pvma
);
1437 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1438 struct shmem_inode_info
*info
, pgoff_t index
)
1440 struct vm_area_struct pvma
;
1441 struct inode
*inode
= &info
->vfs_inode
;
1442 struct address_space
*mapping
= inode
->i_mapping
;
1443 pgoff_t idx
, hindex
;
1444 void __rcu
**results
;
1447 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1450 hindex
= round_down(index
, HPAGE_PMD_NR
);
1452 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1453 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1459 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1460 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1461 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1462 shmem_pseudo_vma_destroy(&pvma
);
1464 prep_transhuge_page(page
);
1468 static struct page
*shmem_alloc_page(gfp_t gfp
,
1469 struct shmem_inode_info
*info
, pgoff_t index
)
1471 struct vm_area_struct pvma
;
1474 shmem_pseudo_vma_init(&pvma
, info
, index
);
1475 page
= alloc_page_vma(gfp
, &pvma
, 0);
1476 shmem_pseudo_vma_destroy(&pvma
);
1481 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1482 struct inode
*inode
,
1483 pgoff_t index
, bool huge
)
1485 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1490 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1492 nr
= huge
? HPAGE_PMD_NR
: 1;
1494 if (!shmem_inode_acct_block(inode
, nr
))
1498 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1500 page
= shmem_alloc_page(gfp
, info
, index
);
1502 __SetPageLocked(page
);
1503 __SetPageSwapBacked(page
);
1508 shmem_inode_unacct_blocks(inode
, nr
);
1510 return ERR_PTR(err
);
1514 * When a page is moved from swapcache to shmem filecache (either by the
1515 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1516 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1517 * ignorance of the mapping it belongs to. If that mapping has special
1518 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1519 * we may need to copy to a suitable page before moving to filecache.
1521 * In a future release, this may well be extended to respect cpuset and
1522 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1523 * but for now it is a simple matter of zone.
1525 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1527 return page_zonenum(page
) > gfp_zone(gfp
);
1530 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1531 struct shmem_inode_info
*info
, pgoff_t index
)
1533 struct page
*oldpage
, *newpage
;
1534 struct address_space
*swap_mapping
;
1539 swap_index
= page_private(oldpage
);
1540 swap_mapping
= page_mapping(oldpage
);
1543 * We have arrived here because our zones are constrained, so don't
1544 * limit chance of success by further cpuset and node constraints.
1546 gfp
&= ~GFP_CONSTRAINT_MASK
;
1547 newpage
= shmem_alloc_page(gfp
, info
, index
);
1552 copy_highpage(newpage
, oldpage
);
1553 flush_dcache_page(newpage
);
1555 __SetPageLocked(newpage
);
1556 __SetPageSwapBacked(newpage
);
1557 SetPageUptodate(newpage
);
1558 set_page_private(newpage
, swap_index
);
1559 SetPageSwapCache(newpage
);
1562 * Our caller will very soon move newpage out of swapcache, but it's
1563 * a nice clean interface for us to replace oldpage by newpage there.
1565 spin_lock_irq(&swap_mapping
->tree_lock
);
1566 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1569 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1570 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1572 spin_unlock_irq(&swap_mapping
->tree_lock
);
1574 if (unlikely(error
)) {
1576 * Is this possible? I think not, now that our callers check
1577 * both PageSwapCache and page_private after getting page lock;
1578 * but be defensive. Reverse old to newpage for clear and free.
1582 mem_cgroup_migrate(oldpage
, newpage
);
1583 lru_cache_add_anon(newpage
);
1587 ClearPageSwapCache(oldpage
);
1588 set_page_private(oldpage
, 0);
1590 unlock_page(oldpage
);
1597 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1599 * If we allocate a new one we do not mark it dirty. That's up to the
1600 * vm. If we swap it in we mark it dirty since we also free the swap
1601 * entry since a page cannot live in both the swap and page cache.
1603 * fault_mm and fault_type are only supplied by shmem_fault:
1604 * otherwise they are NULL.
1606 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1607 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1608 struct vm_area_struct
*vma
, struct vm_fault
*vmf
, int *fault_type
)
1610 struct address_space
*mapping
= inode
->i_mapping
;
1611 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1612 struct shmem_sb_info
*sbinfo
;
1613 struct mm_struct
*charge_mm
;
1614 struct mem_cgroup
*memcg
;
1617 enum sgp_type sgp_huge
= sgp
;
1618 pgoff_t hindex
= index
;
1623 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1625 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1629 page
= find_lock_entry(mapping
, index
);
1630 if (radix_tree_exceptional_entry(page
)) {
1631 swap
= radix_to_swp_entry(page
);
1635 if (sgp
<= SGP_CACHE
&&
1636 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1641 if (page
&& sgp
== SGP_WRITE
)
1642 mark_page_accessed(page
);
1644 /* fallocated page? */
1645 if (page
&& !PageUptodate(page
)) {
1646 if (sgp
!= SGP_READ
)
1652 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1658 * Fast cache lookup did not find it:
1659 * bring it back from swap or allocate.
1661 sbinfo
= SHMEM_SB(inode
->i_sb
);
1662 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1665 /* Look it up and read it in.. */
1666 page
= lookup_swap_cache(swap
, NULL
, 0);
1668 /* Or update major stats only when swapin succeeds?? */
1670 *fault_type
|= VM_FAULT_MAJOR
;
1671 count_vm_event(PGMAJFAULT
);
1672 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1674 /* Here we actually start the io */
1675 page
= shmem_swapin(swap
, gfp
, info
, index
);
1682 /* We have to do this with page locked to prevent races */
1684 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1685 !shmem_confirm_swap(mapping
, index
, swap
)) {
1686 error
= -EEXIST
; /* try again */
1689 if (!PageUptodate(page
)) {
1693 wait_on_page_writeback(page
);
1695 if (shmem_should_replace_page(page
, gfp
)) {
1696 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1701 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1704 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1705 swp_to_radix_entry(swap
));
1707 * We already confirmed swap under page lock, and make
1708 * no memory allocation here, so usually no possibility
1709 * of error; but free_swap_and_cache() only trylocks a
1710 * page, so it is just possible that the entry has been
1711 * truncated or holepunched since swap was confirmed.
1712 * shmem_undo_range() will have done some of the
1713 * unaccounting, now delete_from_swap_cache() will do
1715 * Reset swap.val? No, leave it so "failed" goes back to
1716 * "repeat": reading a hole and writing should succeed.
1719 mem_cgroup_cancel_charge(page
, memcg
, false);
1720 delete_from_swap_cache(page
);
1726 mem_cgroup_commit_charge(page
, memcg
, true, false);
1728 spin_lock_irq(&info
->lock
);
1730 shmem_recalc_inode(inode
);
1731 spin_unlock_irq(&info
->lock
);
1733 if (sgp
== SGP_WRITE
)
1734 mark_page_accessed(page
);
1736 delete_from_swap_cache(page
);
1737 set_page_dirty(page
);
1741 if (vma
&& userfaultfd_missing(vma
)) {
1742 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1746 /* shmem_symlink() */
1747 if (mapping
->a_ops
!= &shmem_aops
)
1749 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1751 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1753 switch (sbinfo
->huge
) {
1756 case SHMEM_HUGE_NEVER
:
1758 case SHMEM_HUGE_WITHIN_SIZE
:
1759 off
= round_up(index
, HPAGE_PMD_NR
);
1760 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1761 if (i_size
>= HPAGE_PMD_SIZE
&&
1762 i_size
>> PAGE_SHIFT
>= off
)
1765 case SHMEM_HUGE_ADVISE
:
1766 if (sgp_huge
== SGP_HUGE
)
1768 /* TODO: implement fadvise() hints */
1773 page
= shmem_alloc_and_acct_page(gfp
, inode
, index
, true);
1775 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, inode
,
1780 error
= PTR_ERR(page
);
1782 if (error
!= -ENOSPC
)
1785 * Try to reclaim some spece by splitting a huge page
1786 * beyond i_size on the filesystem.
1790 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1791 if (ret
== SHRINK_STOP
)
1799 if (PageTransHuge(page
))
1800 hindex
= round_down(index
, HPAGE_PMD_NR
);
1804 if (sgp
== SGP_WRITE
)
1805 __SetPageReferenced(page
);
1807 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1808 PageTransHuge(page
));
1811 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1812 compound_order(page
));
1814 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1816 radix_tree_preload_end();
1819 mem_cgroup_cancel_charge(page
, memcg
,
1820 PageTransHuge(page
));
1823 mem_cgroup_commit_charge(page
, memcg
, false,
1824 PageTransHuge(page
));
1825 lru_cache_add_anon(page
);
1827 spin_lock_irq(&info
->lock
);
1828 info
->alloced
+= 1 << compound_order(page
);
1829 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1830 shmem_recalc_inode(inode
);
1831 spin_unlock_irq(&info
->lock
);
1834 if (PageTransHuge(page
) &&
1835 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1836 hindex
+ HPAGE_PMD_NR
- 1) {
1838 * Part of the huge page is beyond i_size: subject
1839 * to shrink under memory pressure.
1841 spin_lock(&sbinfo
->shrinklist_lock
);
1843 * _careful to defend against unlocked access to
1844 * ->shrink_list in shmem_unused_huge_shrink()
1846 if (list_empty_careful(&info
->shrinklist
)) {
1847 list_add_tail(&info
->shrinklist
,
1848 &sbinfo
->shrinklist
);
1849 sbinfo
->shrinklist_len
++;
1851 spin_unlock(&sbinfo
->shrinklist_lock
);
1855 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1857 if (sgp
== SGP_FALLOC
)
1861 * Let SGP_WRITE caller clear ends if write does not fill page;
1862 * but SGP_FALLOC on a page fallocated earlier must initialize
1863 * it now, lest undo on failure cancel our earlier guarantee.
1865 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1866 struct page
*head
= compound_head(page
);
1869 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1870 clear_highpage(head
+ i
);
1871 flush_dcache_page(head
+ i
);
1873 SetPageUptodate(head
);
1877 /* Perhaps the file has been truncated since we checked */
1878 if (sgp
<= SGP_CACHE
&&
1879 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1881 ClearPageDirty(page
);
1882 delete_from_page_cache(page
);
1883 spin_lock_irq(&info
->lock
);
1884 shmem_recalc_inode(inode
);
1885 spin_unlock_irq(&info
->lock
);
1890 *pagep
= page
+ index
- hindex
;
1897 shmem_inode_unacct_blocks(inode
, 1 << compound_order(page
));
1899 if (PageTransHuge(page
)) {
1905 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1912 if (error
== -ENOSPC
&& !once
++) {
1913 spin_lock_irq(&info
->lock
);
1914 shmem_recalc_inode(inode
);
1915 spin_unlock_irq(&info
->lock
);
1918 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1924 * This is like autoremove_wake_function, but it removes the wait queue
1925 * entry unconditionally - even if something else had already woken the
1928 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
1930 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1931 list_del_init(&wait
->entry
);
1935 static int shmem_fault(struct vm_fault
*vmf
)
1937 struct vm_area_struct
*vma
= vmf
->vma
;
1938 struct inode
*inode
= file_inode(vma
->vm_file
);
1939 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1942 int ret
= VM_FAULT_LOCKED
;
1945 * Trinity finds that probing a hole which tmpfs is punching can
1946 * prevent the hole-punch from ever completing: which in turn
1947 * locks writers out with its hold on i_mutex. So refrain from
1948 * faulting pages into the hole while it's being punched. Although
1949 * shmem_undo_range() does remove the additions, it may be unable to
1950 * keep up, as each new page needs its own unmap_mapping_range() call,
1951 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1953 * It does not matter if we sometimes reach this check just before the
1954 * hole-punch begins, so that one fault then races with the punch:
1955 * we just need to make racing faults a rare case.
1957 * The implementation below would be much simpler if we just used a
1958 * standard mutex or completion: but we cannot take i_mutex in fault,
1959 * and bloating every shmem inode for this unlikely case would be sad.
1961 if (unlikely(inode
->i_private
)) {
1962 struct shmem_falloc
*shmem_falloc
;
1964 spin_lock(&inode
->i_lock
);
1965 shmem_falloc
= inode
->i_private
;
1967 shmem_falloc
->waitq
&&
1968 vmf
->pgoff
>= shmem_falloc
->start
&&
1969 vmf
->pgoff
< shmem_falloc
->next
) {
1970 wait_queue_head_t
*shmem_falloc_waitq
;
1971 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
1973 ret
= VM_FAULT_NOPAGE
;
1974 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1975 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1976 /* It's polite to up mmap_sem if we can */
1977 up_read(&vma
->vm_mm
->mmap_sem
);
1978 ret
= VM_FAULT_RETRY
;
1981 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1982 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1983 TASK_UNINTERRUPTIBLE
);
1984 spin_unlock(&inode
->i_lock
);
1988 * shmem_falloc_waitq points into the shmem_fallocate()
1989 * stack of the hole-punching task: shmem_falloc_waitq
1990 * is usually invalid by the time we reach here, but
1991 * finish_wait() does not dereference it in that case;
1992 * though i_lock needed lest racing with wake_up_all().
1994 spin_lock(&inode
->i_lock
);
1995 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1996 spin_unlock(&inode
->i_lock
);
1999 spin_unlock(&inode
->i_lock
);
2004 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2005 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2007 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2010 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2011 gfp
, vma
, vmf
, &ret
);
2013 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
2017 unsigned long shmem_get_unmapped_area(struct file
*file
,
2018 unsigned long uaddr
, unsigned long len
,
2019 unsigned long pgoff
, unsigned long flags
)
2021 unsigned long (*get_area
)(struct file
*,
2022 unsigned long, unsigned long, unsigned long, unsigned long);
2024 unsigned long offset
;
2025 unsigned long inflated_len
;
2026 unsigned long inflated_addr
;
2027 unsigned long inflated_offset
;
2029 if (len
> TASK_SIZE
)
2032 get_area
= current
->mm
->get_unmapped_area
;
2033 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2035 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
2037 if (IS_ERR_VALUE(addr
))
2039 if (addr
& ~PAGE_MASK
)
2041 if (addr
> TASK_SIZE
- len
)
2044 if (shmem_huge
== SHMEM_HUGE_DENY
)
2046 if (len
< HPAGE_PMD_SIZE
)
2048 if (flags
& MAP_FIXED
)
2051 * Our priority is to support MAP_SHARED mapped hugely;
2052 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2053 * But if caller specified an address hint, respect that as before.
2058 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2059 struct super_block
*sb
;
2062 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2063 sb
= file_inode(file
)->i_sb
;
2066 * Called directly from mm/mmap.c, or drivers/char/mem.c
2067 * for "/dev/zero", to create a shared anonymous object.
2069 if (IS_ERR(shm_mnt
))
2071 sb
= shm_mnt
->mnt_sb
;
2073 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2077 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2078 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2080 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2083 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2084 if (inflated_len
> TASK_SIZE
)
2086 if (inflated_len
< len
)
2089 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
2090 if (IS_ERR_VALUE(inflated_addr
))
2092 if (inflated_addr
& ~PAGE_MASK
)
2095 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2096 inflated_addr
+= offset
- inflated_offset
;
2097 if (inflated_offset
> offset
)
2098 inflated_addr
+= HPAGE_PMD_SIZE
;
2100 if (inflated_addr
> TASK_SIZE
- len
)
2102 return inflated_addr
;
2106 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2108 struct inode
*inode
= file_inode(vma
->vm_file
);
2109 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2112 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2115 struct inode
*inode
= file_inode(vma
->vm_file
);
2118 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2119 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2123 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2125 struct inode
*inode
= file_inode(file
);
2126 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2127 int retval
= -ENOMEM
;
2129 spin_lock_irq(&info
->lock
);
2130 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2131 if (!user_shm_lock(inode
->i_size
, user
))
2133 info
->flags
|= VM_LOCKED
;
2134 mapping_set_unevictable(file
->f_mapping
);
2136 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2137 user_shm_unlock(inode
->i_size
, user
);
2138 info
->flags
&= ~VM_LOCKED
;
2139 mapping_clear_unevictable(file
->f_mapping
);
2144 spin_unlock_irq(&info
->lock
);
2148 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2150 file_accessed(file
);
2151 vma
->vm_ops
= &shmem_vm_ops
;
2152 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2153 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2154 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2155 khugepaged_enter(vma
, vma
->vm_flags
);
2160 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2161 umode_t mode
, dev_t dev
, unsigned long flags
)
2163 struct inode
*inode
;
2164 struct shmem_inode_info
*info
;
2165 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2167 if (shmem_reserve_inode(sb
))
2170 inode
= new_inode(sb
);
2172 inode
->i_ino
= get_next_ino();
2173 inode_init_owner(inode
, dir
, mode
);
2174 inode
->i_blocks
= 0;
2175 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2176 inode
->i_generation
= get_seconds();
2177 info
= SHMEM_I(inode
);
2178 memset(info
, 0, (char *)inode
- (char *)info
);
2179 spin_lock_init(&info
->lock
);
2180 info
->seals
= F_SEAL_SEAL
;
2181 info
->flags
= flags
& VM_NORESERVE
;
2182 INIT_LIST_HEAD(&info
->shrinklist
);
2183 INIT_LIST_HEAD(&info
->swaplist
);
2184 simple_xattrs_init(&info
->xattrs
);
2185 cache_no_acl(inode
);
2187 switch (mode
& S_IFMT
) {
2189 inode
->i_op
= &shmem_special_inode_operations
;
2190 init_special_inode(inode
, mode
, dev
);
2193 inode
->i_mapping
->a_ops
= &shmem_aops
;
2194 inode
->i_op
= &shmem_inode_operations
;
2195 inode
->i_fop
= &shmem_file_operations
;
2196 mpol_shared_policy_init(&info
->policy
,
2197 shmem_get_sbmpol(sbinfo
));
2201 /* Some things misbehave if size == 0 on a directory */
2202 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2203 inode
->i_op
= &shmem_dir_inode_operations
;
2204 inode
->i_fop
= &simple_dir_operations
;
2208 * Must not load anything in the rbtree,
2209 * mpol_free_shared_policy will not be called.
2211 mpol_shared_policy_init(&info
->policy
, NULL
);
2215 lockdep_annotate_inode_mutex_key(inode
);
2217 shmem_free_inode(sb
);
2221 bool shmem_mapping(struct address_space
*mapping
)
2223 return mapping
->a_ops
== &shmem_aops
;
2226 static int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2228 struct vm_area_struct
*dst_vma
,
2229 unsigned long dst_addr
,
2230 unsigned long src_addr
,
2232 struct page
**pagep
)
2234 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2235 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2236 struct address_space
*mapping
= inode
->i_mapping
;
2237 gfp_t gfp
= mapping_gfp_mask(mapping
);
2238 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2239 struct mem_cgroup
*memcg
;
2243 pte_t _dst_pte
, *dst_pte
;
2247 if (!shmem_inode_acct_block(inode
, 1))
2251 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2253 goto out_unacct_blocks
;
2255 if (!zeropage
) { /* mcopy_atomic */
2256 page_kaddr
= kmap_atomic(page
);
2257 ret
= copy_from_user(page_kaddr
,
2258 (const void __user
*)src_addr
,
2260 kunmap_atomic(page_kaddr
);
2262 /* fallback to copy_from_user outside mmap_sem */
2263 if (unlikely(ret
)) {
2265 shmem_inode_unacct_blocks(inode
, 1);
2266 /* don't free the page */
2269 } else { /* mfill_zeropage_atomic */
2270 clear_highpage(page
);
2277 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2278 __SetPageLocked(page
);
2279 __SetPageSwapBacked(page
);
2280 __SetPageUptodate(page
);
2282 ret
= mem_cgroup_try_charge(page
, dst_mm
, gfp
, &memcg
, false);
2286 ret
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
2288 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
);
2289 radix_tree_preload_end();
2292 goto out_release_uncharge
;
2294 mem_cgroup_commit_charge(page
, memcg
, false, false);
2296 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2297 if (dst_vma
->vm_flags
& VM_WRITE
)
2298 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2301 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2302 if (!pte_none(*dst_pte
))
2303 goto out_release_uncharge_unlock
;
2305 lru_cache_add_anon(page
);
2307 spin_lock(&info
->lock
);
2309 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2310 shmem_recalc_inode(inode
);
2311 spin_unlock(&info
->lock
);
2313 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2314 page_add_file_rmap(page
, false);
2315 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2317 /* No need to invalidate - it was non-present before */
2318 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2320 pte_unmap_unlock(dst_pte
, ptl
);
2324 out_release_uncharge_unlock
:
2325 pte_unmap_unlock(dst_pte
, ptl
);
2326 out_release_uncharge
:
2327 mem_cgroup_cancel_charge(page
, memcg
, false);
2332 shmem_inode_unacct_blocks(inode
, 1);
2336 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2338 struct vm_area_struct
*dst_vma
,
2339 unsigned long dst_addr
,
2340 unsigned long src_addr
,
2341 struct page
**pagep
)
2343 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2344 dst_addr
, src_addr
, false, pagep
);
2347 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2349 struct vm_area_struct
*dst_vma
,
2350 unsigned long dst_addr
)
2352 struct page
*page
= NULL
;
2354 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2355 dst_addr
, 0, true, &page
);
2359 static const struct inode_operations shmem_symlink_inode_operations
;
2360 static const struct inode_operations shmem_short_symlink_operations
;
2362 #ifdef CONFIG_TMPFS_XATTR
2363 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2365 #define shmem_initxattrs NULL
2369 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2370 loff_t pos
, unsigned len
, unsigned flags
,
2371 struct page
**pagep
, void **fsdata
)
2373 struct inode
*inode
= mapping
->host
;
2374 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2375 pgoff_t index
= pos
>> PAGE_SHIFT
;
2377 /* i_mutex is held by caller */
2378 if (unlikely(info
->seals
& (F_SEAL_WRITE
| F_SEAL_GROW
))) {
2379 if (info
->seals
& F_SEAL_WRITE
)
2381 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2385 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2389 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2390 loff_t pos
, unsigned len
, unsigned copied
,
2391 struct page
*page
, void *fsdata
)
2393 struct inode
*inode
= mapping
->host
;
2395 if (pos
+ copied
> inode
->i_size
)
2396 i_size_write(inode
, pos
+ copied
);
2398 if (!PageUptodate(page
)) {
2399 struct page
*head
= compound_head(page
);
2400 if (PageTransCompound(page
)) {
2403 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2404 if (head
+ i
== page
)
2406 clear_highpage(head
+ i
);
2407 flush_dcache_page(head
+ i
);
2410 if (copied
< PAGE_SIZE
) {
2411 unsigned from
= pos
& (PAGE_SIZE
- 1);
2412 zero_user_segments(page
, 0, from
,
2413 from
+ copied
, PAGE_SIZE
);
2415 SetPageUptodate(head
);
2417 set_page_dirty(page
);
2424 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2426 struct file
*file
= iocb
->ki_filp
;
2427 struct inode
*inode
= file_inode(file
);
2428 struct address_space
*mapping
= inode
->i_mapping
;
2430 unsigned long offset
;
2431 enum sgp_type sgp
= SGP_READ
;
2434 loff_t
*ppos
= &iocb
->ki_pos
;
2437 * Might this read be for a stacking filesystem? Then when reading
2438 * holes of a sparse file, we actually need to allocate those pages,
2439 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2441 if (!iter_is_iovec(to
))
2444 index
= *ppos
>> PAGE_SHIFT
;
2445 offset
= *ppos
& ~PAGE_MASK
;
2448 struct page
*page
= NULL
;
2450 unsigned long nr
, ret
;
2451 loff_t i_size
= i_size_read(inode
);
2453 end_index
= i_size
>> PAGE_SHIFT
;
2454 if (index
> end_index
)
2456 if (index
== end_index
) {
2457 nr
= i_size
& ~PAGE_MASK
;
2462 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2464 if (error
== -EINVAL
)
2469 if (sgp
== SGP_CACHE
)
2470 set_page_dirty(page
);
2475 * We must evaluate after, since reads (unlike writes)
2476 * are called without i_mutex protection against truncate
2479 i_size
= i_size_read(inode
);
2480 end_index
= i_size
>> PAGE_SHIFT
;
2481 if (index
== end_index
) {
2482 nr
= i_size
& ~PAGE_MASK
;
2493 * If users can be writing to this page using arbitrary
2494 * virtual addresses, take care about potential aliasing
2495 * before reading the page on the kernel side.
2497 if (mapping_writably_mapped(mapping
))
2498 flush_dcache_page(page
);
2500 * Mark the page accessed if we read the beginning.
2503 mark_page_accessed(page
);
2505 page
= ZERO_PAGE(0);
2510 * Ok, we have the page, and it's up-to-date, so
2511 * now we can copy it to user space...
2513 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2516 index
+= offset
>> PAGE_SHIFT
;
2517 offset
&= ~PAGE_MASK
;
2520 if (!iov_iter_count(to
))
2529 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2530 file_accessed(file
);
2531 return retval
? retval
: error
;
2535 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2537 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2538 pgoff_t index
, pgoff_t end
, int whence
)
2541 struct pagevec pvec
;
2542 pgoff_t indices
[PAGEVEC_SIZE
];
2546 pagevec_init(&pvec
, 0);
2547 pvec
.nr
= 1; /* start small: we may be there already */
2549 pvec
.nr
= find_get_entries(mapping
, index
,
2550 pvec
.nr
, pvec
.pages
, indices
);
2552 if (whence
== SEEK_DATA
)
2556 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2557 if (index
< indices
[i
]) {
2558 if (whence
== SEEK_HOLE
) {
2564 page
= pvec
.pages
[i
];
2565 if (page
&& !radix_tree_exceptional_entry(page
)) {
2566 if (!PageUptodate(page
))
2570 (page
&& whence
== SEEK_DATA
) ||
2571 (!page
&& whence
== SEEK_HOLE
)) {
2576 pagevec_remove_exceptionals(&pvec
);
2577 pagevec_release(&pvec
);
2578 pvec
.nr
= PAGEVEC_SIZE
;
2584 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2586 struct address_space
*mapping
= file
->f_mapping
;
2587 struct inode
*inode
= mapping
->host
;
2591 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2592 return generic_file_llseek_size(file
, offset
, whence
,
2593 MAX_LFS_FILESIZE
, i_size_read(inode
));
2595 /* We're holding i_mutex so we can access i_size directly */
2597 if (offset
< 0 || offset
>= inode
->i_size
)
2600 start
= offset
>> PAGE_SHIFT
;
2601 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2602 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2603 new_offset
<<= PAGE_SHIFT
;
2604 if (new_offset
> offset
) {
2605 if (new_offset
< inode
->i_size
)
2606 offset
= new_offset
;
2607 else if (whence
== SEEK_DATA
)
2610 offset
= inode
->i_size
;
2615 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2616 inode_unlock(inode
);
2621 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2622 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2624 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2625 #define LAST_SCAN 4 /* about 150ms max */
2627 static void shmem_tag_pins(struct address_space
*mapping
)
2629 struct radix_tree_iter iter
;
2638 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2639 page
= radix_tree_deref_slot(slot
);
2640 if (!page
|| radix_tree_exception(page
)) {
2641 if (radix_tree_deref_retry(page
)) {
2642 slot
= radix_tree_iter_retry(&iter
);
2645 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2646 spin_lock_irq(&mapping
->tree_lock
);
2647 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2649 spin_unlock_irq(&mapping
->tree_lock
);
2652 if (need_resched()) {
2653 slot
= radix_tree_iter_resume(slot
, &iter
);
2661 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2662 * via get_user_pages(), drivers might have some pending I/O without any active
2663 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2664 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2665 * them to be dropped.
2666 * The caller must guarantee that no new user will acquire writable references
2667 * to those pages to avoid races.
2669 static int shmem_wait_for_pins(struct address_space
*mapping
)
2671 struct radix_tree_iter iter
;
2677 shmem_tag_pins(mapping
);
2680 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2681 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2685 lru_add_drain_all();
2686 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2691 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2692 start
, SHMEM_TAG_PINNED
) {
2694 page
= radix_tree_deref_slot(slot
);
2695 if (radix_tree_exception(page
)) {
2696 if (radix_tree_deref_retry(page
)) {
2697 slot
= radix_tree_iter_retry(&iter
);
2705 page_count(page
) - page_mapcount(page
) != 1) {
2706 if (scan
< LAST_SCAN
)
2707 goto continue_resched
;
2710 * On the last scan, we clean up all those tags
2711 * we inserted; but make a note that we still
2712 * found pages pinned.
2717 spin_lock_irq(&mapping
->tree_lock
);
2718 radix_tree_tag_clear(&mapping
->page_tree
,
2719 iter
.index
, SHMEM_TAG_PINNED
);
2720 spin_unlock_irq(&mapping
->tree_lock
);
2722 if (need_resched()) {
2723 slot
= radix_tree_iter_resume(slot
, &iter
);
2733 #define F_ALL_SEALS (F_SEAL_SEAL | \
2738 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2740 struct inode
*inode
= file_inode(file
);
2741 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2746 * Sealing allows multiple parties to share a shmem-file but restrict
2747 * access to a specific subset of file operations. Seals can only be
2748 * added, but never removed. This way, mutually untrusted parties can
2749 * share common memory regions with a well-defined policy. A malicious
2750 * peer can thus never perform unwanted operations on a shared object.
2752 * Seals are only supported on special shmem-files and always affect
2753 * the whole underlying inode. Once a seal is set, it may prevent some
2754 * kinds of access to the file. Currently, the following seals are
2756 * SEAL_SEAL: Prevent further seals from being set on this file
2757 * SEAL_SHRINK: Prevent the file from shrinking
2758 * SEAL_GROW: Prevent the file from growing
2759 * SEAL_WRITE: Prevent write access to the file
2761 * As we don't require any trust relationship between two parties, we
2762 * must prevent seals from being removed. Therefore, sealing a file
2763 * only adds a given set of seals to the file, it never touches
2764 * existing seals. Furthermore, the "setting seals"-operation can be
2765 * sealed itself, which basically prevents any further seal from being
2768 * Semantics of sealing are only defined on volatile files. Only
2769 * anonymous shmem files support sealing. More importantly, seals are
2770 * never written to disk. Therefore, there's no plan to support it on
2774 if (file
->f_op
!= &shmem_file_operations
)
2776 if (!(file
->f_mode
& FMODE_WRITE
))
2778 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2783 if (info
->seals
& F_SEAL_SEAL
) {
2788 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2789 error
= mapping_deny_writable(file
->f_mapping
);
2793 error
= shmem_wait_for_pins(file
->f_mapping
);
2795 mapping_allow_writable(file
->f_mapping
);
2800 info
->seals
|= seals
;
2804 inode_unlock(inode
);
2807 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2809 int shmem_get_seals(struct file
*file
)
2811 if (file
->f_op
!= &shmem_file_operations
)
2814 return SHMEM_I(file_inode(file
))->seals
;
2816 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2818 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2824 /* disallow upper 32bit */
2828 error
= shmem_add_seals(file
, arg
);
2831 error
= shmem_get_seals(file
);
2841 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2844 struct inode
*inode
= file_inode(file
);
2845 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2846 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2847 struct shmem_falloc shmem_falloc
;
2848 pgoff_t start
, index
, end
;
2851 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2856 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2857 struct address_space
*mapping
= file
->f_mapping
;
2858 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2859 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2860 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2862 /* protected by i_mutex */
2863 if (info
->seals
& F_SEAL_WRITE
) {
2868 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2869 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2870 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2871 spin_lock(&inode
->i_lock
);
2872 inode
->i_private
= &shmem_falloc
;
2873 spin_unlock(&inode
->i_lock
);
2875 if ((u64
)unmap_end
> (u64
)unmap_start
)
2876 unmap_mapping_range(mapping
, unmap_start
,
2877 1 + unmap_end
- unmap_start
, 0);
2878 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2879 /* No need to unmap again: hole-punching leaves COWed pages */
2881 spin_lock(&inode
->i_lock
);
2882 inode
->i_private
= NULL
;
2883 wake_up_all(&shmem_falloc_waitq
);
2884 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2885 spin_unlock(&inode
->i_lock
);
2890 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2891 error
= inode_newsize_ok(inode
, offset
+ len
);
2895 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2900 start
= offset
>> PAGE_SHIFT
;
2901 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2902 /* Try to avoid a swapstorm if len is impossible to satisfy */
2903 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2908 shmem_falloc
.waitq
= NULL
;
2909 shmem_falloc
.start
= start
;
2910 shmem_falloc
.next
= start
;
2911 shmem_falloc
.nr_falloced
= 0;
2912 shmem_falloc
.nr_unswapped
= 0;
2913 spin_lock(&inode
->i_lock
);
2914 inode
->i_private
= &shmem_falloc
;
2915 spin_unlock(&inode
->i_lock
);
2917 for (index
= start
; index
< end
; index
++) {
2921 * Good, the fallocate(2) manpage permits EINTR: we may have
2922 * been interrupted because we are using up too much memory.
2924 if (signal_pending(current
))
2926 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2929 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2931 /* Remove the !PageUptodate pages we added */
2932 if (index
> start
) {
2933 shmem_undo_range(inode
,
2934 (loff_t
)start
<< PAGE_SHIFT
,
2935 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2941 * Inform shmem_writepage() how far we have reached.
2942 * No need for lock or barrier: we have the page lock.
2944 shmem_falloc
.next
++;
2945 if (!PageUptodate(page
))
2946 shmem_falloc
.nr_falloced
++;
2949 * If !PageUptodate, leave it that way so that freeable pages
2950 * can be recognized if we need to rollback on error later.
2951 * But set_page_dirty so that memory pressure will swap rather
2952 * than free the pages we are allocating (and SGP_CACHE pages
2953 * might still be clean: we now need to mark those dirty too).
2955 set_page_dirty(page
);
2961 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2962 i_size_write(inode
, offset
+ len
);
2963 inode
->i_ctime
= current_time(inode
);
2965 spin_lock(&inode
->i_lock
);
2966 inode
->i_private
= NULL
;
2967 spin_unlock(&inode
->i_lock
);
2969 inode_unlock(inode
);
2973 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2975 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2977 buf
->f_type
= TMPFS_MAGIC
;
2978 buf
->f_bsize
= PAGE_SIZE
;
2979 buf
->f_namelen
= NAME_MAX
;
2980 if (sbinfo
->max_blocks
) {
2981 buf
->f_blocks
= sbinfo
->max_blocks
;
2983 buf
->f_bfree
= sbinfo
->max_blocks
-
2984 percpu_counter_sum(&sbinfo
->used_blocks
);
2986 if (sbinfo
->max_inodes
) {
2987 buf
->f_files
= sbinfo
->max_inodes
;
2988 buf
->f_ffree
= sbinfo
->free_inodes
;
2990 /* else leave those fields 0 like simple_statfs */
2995 * File creation. Allocate an inode, and we're done..
2998 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3000 struct inode
*inode
;
3001 int error
= -ENOSPC
;
3003 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
3005 error
= simple_acl_create(dir
, inode
);
3008 error
= security_inode_init_security(inode
, dir
,
3010 shmem_initxattrs
, NULL
);
3011 if (error
&& error
!= -EOPNOTSUPP
)
3015 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3016 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3017 d_instantiate(dentry
, inode
);
3018 dget(dentry
); /* Extra count - pin the dentry in core */
3027 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3029 struct inode
*inode
;
3030 int error
= -ENOSPC
;
3032 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
3034 error
= security_inode_init_security(inode
, dir
,
3036 shmem_initxattrs
, NULL
);
3037 if (error
&& error
!= -EOPNOTSUPP
)
3039 error
= simple_acl_create(dir
, inode
);
3042 d_tmpfile(dentry
, inode
);
3050 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3054 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
3060 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
3063 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
3069 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
3071 struct inode
*inode
= d_inode(old_dentry
);
3075 * No ordinary (disk based) filesystem counts links as inodes;
3076 * but each new link needs a new dentry, pinning lowmem, and
3077 * tmpfs dentries cannot be pruned until they are unlinked.
3079 ret
= shmem_reserve_inode(inode
->i_sb
);
3083 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3084 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3086 ihold(inode
); /* New dentry reference */
3087 dget(dentry
); /* Extra pinning count for the created dentry */
3088 d_instantiate(dentry
, inode
);
3093 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3095 struct inode
*inode
= d_inode(dentry
);
3097 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3098 shmem_free_inode(inode
->i_sb
);
3100 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3101 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3103 dput(dentry
); /* Undo the count from "create" - this does all the work */
3107 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3109 if (!simple_empty(dentry
))
3112 drop_nlink(d_inode(dentry
));
3114 return shmem_unlink(dir
, dentry
);
3117 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3119 bool old_is_dir
= d_is_dir(old_dentry
);
3120 bool new_is_dir
= d_is_dir(new_dentry
);
3122 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3124 drop_nlink(old_dir
);
3127 drop_nlink(new_dir
);
3131 old_dir
->i_ctime
= old_dir
->i_mtime
=
3132 new_dir
->i_ctime
= new_dir
->i_mtime
=
3133 d_inode(old_dentry
)->i_ctime
=
3134 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3139 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3141 struct dentry
*whiteout
;
3144 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3148 error
= shmem_mknod(old_dir
, whiteout
,
3149 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3155 * Cheat and hash the whiteout while the old dentry is still in
3156 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3158 * d_lookup() will consistently find one of them at this point,
3159 * not sure which one, but that isn't even important.
3166 * The VFS layer already does all the dentry stuff for rename,
3167 * we just have to decrement the usage count for the target if
3168 * it exists so that the VFS layer correctly free's it when it
3171 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3173 struct inode
*inode
= d_inode(old_dentry
);
3174 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3176 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3179 if (flags
& RENAME_EXCHANGE
)
3180 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3182 if (!simple_empty(new_dentry
))
3185 if (flags
& RENAME_WHITEOUT
) {
3188 error
= shmem_whiteout(old_dir
, old_dentry
);
3193 if (d_really_is_positive(new_dentry
)) {
3194 (void) shmem_unlink(new_dir
, new_dentry
);
3195 if (they_are_dirs
) {
3196 drop_nlink(d_inode(new_dentry
));
3197 drop_nlink(old_dir
);
3199 } else if (they_are_dirs
) {
3200 drop_nlink(old_dir
);
3204 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3205 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3206 old_dir
->i_ctime
= old_dir
->i_mtime
=
3207 new_dir
->i_ctime
= new_dir
->i_mtime
=
3208 inode
->i_ctime
= current_time(old_dir
);
3212 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3216 struct inode
*inode
;
3218 struct shmem_inode_info
*info
;
3220 len
= strlen(symname
) + 1;
3221 if (len
> PAGE_SIZE
)
3222 return -ENAMETOOLONG
;
3224 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
3228 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3229 shmem_initxattrs
, NULL
);
3231 if (error
!= -EOPNOTSUPP
) {
3238 info
= SHMEM_I(inode
);
3239 inode
->i_size
= len
-1;
3240 if (len
<= SHORT_SYMLINK_LEN
) {
3241 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3242 if (!inode
->i_link
) {
3246 inode
->i_op
= &shmem_short_symlink_operations
;
3248 inode_nohighmem(inode
);
3249 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3254 inode
->i_mapping
->a_ops
= &shmem_aops
;
3255 inode
->i_op
= &shmem_symlink_inode_operations
;
3256 memcpy(page_address(page
), symname
, len
);
3257 SetPageUptodate(page
);
3258 set_page_dirty(page
);
3262 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3263 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3264 d_instantiate(dentry
, inode
);
3269 static void shmem_put_link(void *arg
)
3271 mark_page_accessed(arg
);
3275 static const char *shmem_get_link(struct dentry
*dentry
,
3276 struct inode
*inode
,
3277 struct delayed_call
*done
)
3279 struct page
*page
= NULL
;
3282 page
= find_get_page(inode
->i_mapping
, 0);
3284 return ERR_PTR(-ECHILD
);
3285 if (!PageUptodate(page
)) {
3287 return ERR_PTR(-ECHILD
);
3290 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3292 return ERR_PTR(error
);
3295 set_delayed_call(done
, shmem_put_link
, page
);
3296 return page_address(page
);
3299 #ifdef CONFIG_TMPFS_XATTR
3301 * Superblocks without xattr inode operations may get some security.* xattr
3302 * support from the LSM "for free". As soon as we have any other xattrs
3303 * like ACLs, we also need to implement the security.* handlers at
3304 * filesystem level, though.
3308 * Callback for security_inode_init_security() for acquiring xattrs.
3310 static int shmem_initxattrs(struct inode
*inode
,
3311 const struct xattr
*xattr_array
,
3314 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3315 const struct xattr
*xattr
;
3316 struct simple_xattr
*new_xattr
;
3319 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3320 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3324 len
= strlen(xattr
->name
) + 1;
3325 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3327 if (!new_xattr
->name
) {
3332 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3333 XATTR_SECURITY_PREFIX_LEN
);
3334 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3337 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3343 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3344 struct dentry
*unused
, struct inode
*inode
,
3345 const char *name
, void *buffer
, size_t size
)
3347 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3349 name
= xattr_full_name(handler
, name
);
3350 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3353 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3354 struct dentry
*unused
, struct inode
*inode
,
3355 const char *name
, const void *value
,
3356 size_t size
, int flags
)
3358 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3360 name
= xattr_full_name(handler
, name
);
3361 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3364 static const struct xattr_handler shmem_security_xattr_handler
= {
3365 .prefix
= XATTR_SECURITY_PREFIX
,
3366 .get
= shmem_xattr_handler_get
,
3367 .set
= shmem_xattr_handler_set
,
3370 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3371 .prefix
= XATTR_TRUSTED_PREFIX
,
3372 .get
= shmem_xattr_handler_get
,
3373 .set
= shmem_xattr_handler_set
,
3376 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3377 #ifdef CONFIG_TMPFS_POSIX_ACL
3378 &posix_acl_access_xattr_handler
,
3379 &posix_acl_default_xattr_handler
,
3381 &shmem_security_xattr_handler
,
3382 &shmem_trusted_xattr_handler
,
3386 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3388 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3389 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3391 #endif /* CONFIG_TMPFS_XATTR */
3393 static const struct inode_operations shmem_short_symlink_operations
= {
3394 .get_link
= simple_get_link
,
3395 #ifdef CONFIG_TMPFS_XATTR
3396 .listxattr
= shmem_listxattr
,
3400 static const struct inode_operations shmem_symlink_inode_operations
= {
3401 .get_link
= shmem_get_link
,
3402 #ifdef CONFIG_TMPFS_XATTR
3403 .listxattr
= shmem_listxattr
,
3407 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3409 return ERR_PTR(-ESTALE
);
3412 static int shmem_match(struct inode
*ino
, void *vfh
)
3416 inum
= (inum
<< 32) | fh
[1];
3417 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3420 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3421 struct fid
*fid
, int fh_len
, int fh_type
)
3423 struct inode
*inode
;
3424 struct dentry
*dentry
= NULL
;
3431 inum
= (inum
<< 32) | fid
->raw
[1];
3433 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3434 shmem_match
, fid
->raw
);
3436 dentry
= d_find_alias(inode
);
3443 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3444 struct inode
*parent
)
3448 return FILEID_INVALID
;
3451 if (inode_unhashed(inode
)) {
3452 /* Unfortunately insert_inode_hash is not idempotent,
3453 * so as we hash inodes here rather than at creation
3454 * time, we need a lock to ensure we only try
3457 static DEFINE_SPINLOCK(lock
);
3459 if (inode_unhashed(inode
))
3460 __insert_inode_hash(inode
,
3461 inode
->i_ino
+ inode
->i_generation
);
3465 fh
[0] = inode
->i_generation
;
3466 fh
[1] = inode
->i_ino
;
3467 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3473 static const struct export_operations shmem_export_ops
= {
3474 .get_parent
= shmem_get_parent
,
3475 .encode_fh
= shmem_encode_fh
,
3476 .fh_to_dentry
= shmem_fh_to_dentry
,
3479 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3482 char *this_char
, *value
, *rest
;
3483 struct mempolicy
*mpol
= NULL
;
3487 while (options
!= NULL
) {
3488 this_char
= options
;
3491 * NUL-terminate this option: unfortunately,
3492 * mount options form a comma-separated list,
3493 * but mpol's nodelist may also contain commas.
3495 options
= strchr(options
, ',');
3496 if (options
== NULL
)
3499 if (!isdigit(*options
)) {
3506 if ((value
= strchr(this_char
,'=')) != NULL
) {
3509 pr_err("tmpfs: No value for mount option '%s'\n",
3514 if (!strcmp(this_char
,"size")) {
3515 unsigned long long size
;
3516 size
= memparse(value
,&rest
);
3518 size
<<= PAGE_SHIFT
;
3519 size
*= totalram_pages
;
3525 sbinfo
->max_blocks
=
3526 DIV_ROUND_UP(size
, PAGE_SIZE
);
3527 } else if (!strcmp(this_char
,"nr_blocks")) {
3528 sbinfo
->max_blocks
= memparse(value
, &rest
);
3531 } else if (!strcmp(this_char
,"nr_inodes")) {
3532 sbinfo
->max_inodes
= memparse(value
, &rest
);
3535 } else if (!strcmp(this_char
,"mode")) {
3538 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3541 } else if (!strcmp(this_char
,"uid")) {
3544 uid
= simple_strtoul(value
, &rest
, 0);
3547 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3548 if (!uid_valid(sbinfo
->uid
))
3550 } else if (!strcmp(this_char
,"gid")) {
3553 gid
= simple_strtoul(value
, &rest
, 0);
3556 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3557 if (!gid_valid(sbinfo
->gid
))
3559 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3560 } else if (!strcmp(this_char
, "huge")) {
3562 huge
= shmem_parse_huge(value
);
3565 if (!has_transparent_hugepage() &&
3566 huge
!= SHMEM_HUGE_NEVER
)
3568 sbinfo
->huge
= huge
;
3571 } else if (!strcmp(this_char
,"mpol")) {
3574 if (mpol_parse_str(value
, &mpol
))
3578 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3582 sbinfo
->mpol
= mpol
;
3586 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3594 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3596 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3597 struct shmem_sb_info config
= *sbinfo
;
3598 unsigned long inodes
;
3599 int error
= -EINVAL
;
3602 if (shmem_parse_options(data
, &config
, true))
3605 spin_lock(&sbinfo
->stat_lock
);
3606 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3607 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3609 if (config
.max_inodes
< inodes
)
3612 * Those tests disallow limited->unlimited while any are in use;
3613 * but we must separately disallow unlimited->limited, because
3614 * in that case we have no record of how much is already in use.
3616 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3618 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3622 sbinfo
->huge
= config
.huge
;
3623 sbinfo
->max_blocks
= config
.max_blocks
;
3624 sbinfo
->max_inodes
= config
.max_inodes
;
3625 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3628 * Preserve previous mempolicy unless mpol remount option was specified.
3631 mpol_put(sbinfo
->mpol
);
3632 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3635 spin_unlock(&sbinfo
->stat_lock
);
3639 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3641 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3643 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3644 seq_printf(seq
, ",size=%luk",
3645 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3646 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3647 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3648 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3649 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3650 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3651 seq_printf(seq
, ",uid=%u",
3652 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3653 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3654 seq_printf(seq
, ",gid=%u",
3655 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3656 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3657 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3659 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3661 shmem_show_mpol(seq
, sbinfo
->mpol
);
3665 #define MFD_NAME_PREFIX "memfd:"
3666 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3667 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3669 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
3671 SYSCALL_DEFINE2(memfd_create
,
3672 const char __user
*, uname
,
3673 unsigned int, flags
)
3675 struct shmem_inode_info
*info
;
3681 if (!(flags
& MFD_HUGETLB
)) {
3682 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3685 /* Sealing not supported in hugetlbfs (MFD_HUGETLB) */
3686 if (flags
& MFD_ALLOW_SEALING
)
3688 /* Allow huge page size encoding in flags. */
3689 if (flags
& ~(unsigned int)(MFD_ALL_FLAGS
|
3690 (MFD_HUGE_MASK
<< MFD_HUGE_SHIFT
)))
3694 /* length includes terminating zero */
3695 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3698 if (len
> MFD_NAME_MAX_LEN
+ 1)
3701 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_KERNEL
);
3705 strcpy(name
, MFD_NAME_PREFIX
);
3706 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3711 /* terminating-zero may have changed after strnlen_user() returned */
3712 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3717 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3723 if (flags
& MFD_HUGETLB
) {
3724 struct user_struct
*user
= NULL
;
3726 file
= hugetlb_file_setup(name
, 0, VM_NORESERVE
, &user
,
3727 HUGETLB_ANONHUGE_INODE
,
3728 (flags
>> MFD_HUGE_SHIFT
) &
3731 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3733 error
= PTR_ERR(file
);
3736 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3737 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3739 if (flags
& MFD_ALLOW_SEALING
) {
3741 * flags check at beginning of function ensures
3742 * this is not a hugetlbfs (MFD_HUGETLB) file.
3744 info
= SHMEM_I(file_inode(file
));
3745 info
->seals
&= ~F_SEAL_SEAL
;
3748 fd_install(fd
, file
);
3759 #endif /* CONFIG_TMPFS */
3761 static void shmem_put_super(struct super_block
*sb
)
3763 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3765 percpu_counter_destroy(&sbinfo
->used_blocks
);
3766 mpol_put(sbinfo
->mpol
);
3768 sb
->s_fs_info
= NULL
;
3771 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3773 struct inode
*inode
;
3774 struct shmem_sb_info
*sbinfo
;
3777 /* Round up to L1_CACHE_BYTES to resist false sharing */
3778 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3779 L1_CACHE_BYTES
), GFP_KERNEL
);
3783 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3784 sbinfo
->uid
= current_fsuid();
3785 sbinfo
->gid
= current_fsgid();
3786 sb
->s_fs_info
= sbinfo
;
3790 * Per default we only allow half of the physical ram per
3791 * tmpfs instance, limiting inodes to one per page of lowmem;
3792 * but the internal instance is left unlimited.
3794 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3795 sbinfo
->max_blocks
= shmem_default_max_blocks();
3796 sbinfo
->max_inodes
= shmem_default_max_inodes();
3797 if (shmem_parse_options(data
, sbinfo
, false)) {
3802 sb
->s_flags
|= MS_NOUSER
;
3804 sb
->s_export_op
= &shmem_export_ops
;
3805 sb
->s_flags
|= MS_NOSEC
;
3807 sb
->s_flags
|= MS_NOUSER
;
3810 spin_lock_init(&sbinfo
->stat_lock
);
3811 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3813 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3814 spin_lock_init(&sbinfo
->shrinklist_lock
);
3815 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3817 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3818 sb
->s_blocksize
= PAGE_SIZE
;
3819 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3820 sb
->s_magic
= TMPFS_MAGIC
;
3821 sb
->s_op
= &shmem_ops
;
3822 sb
->s_time_gran
= 1;
3823 #ifdef CONFIG_TMPFS_XATTR
3824 sb
->s_xattr
= shmem_xattr_handlers
;
3826 #ifdef CONFIG_TMPFS_POSIX_ACL
3827 sb
->s_flags
|= MS_POSIXACL
;
3829 uuid_gen(&sb
->s_uuid
);
3831 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3834 inode
->i_uid
= sbinfo
->uid
;
3835 inode
->i_gid
= sbinfo
->gid
;
3836 sb
->s_root
= d_make_root(inode
);
3842 shmem_put_super(sb
);
3846 static struct kmem_cache
*shmem_inode_cachep
;
3848 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3850 struct shmem_inode_info
*info
;
3851 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3854 return &info
->vfs_inode
;
3857 static void shmem_destroy_callback(struct rcu_head
*head
)
3859 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3860 if (S_ISLNK(inode
->i_mode
))
3861 kfree(inode
->i_link
);
3862 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3865 static void shmem_destroy_inode(struct inode
*inode
)
3867 if (S_ISREG(inode
->i_mode
))
3868 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3869 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3872 static void shmem_init_inode(void *foo
)
3874 struct shmem_inode_info
*info
= foo
;
3875 inode_init_once(&info
->vfs_inode
);
3878 static int shmem_init_inodecache(void)
3880 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3881 sizeof(struct shmem_inode_info
),
3882 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3886 static void shmem_destroy_inodecache(void)
3888 kmem_cache_destroy(shmem_inode_cachep
);
3891 static const struct address_space_operations shmem_aops
= {
3892 .writepage
= shmem_writepage
,
3893 .set_page_dirty
= __set_page_dirty_no_writeback
,
3895 .write_begin
= shmem_write_begin
,
3896 .write_end
= shmem_write_end
,
3898 #ifdef CONFIG_MIGRATION
3899 .migratepage
= migrate_page
,
3901 .error_remove_page
= generic_error_remove_page
,
3904 static const struct file_operations shmem_file_operations
= {
3906 .get_unmapped_area
= shmem_get_unmapped_area
,
3908 .llseek
= shmem_file_llseek
,
3909 .read_iter
= shmem_file_read_iter
,
3910 .write_iter
= generic_file_write_iter
,
3911 .fsync
= noop_fsync
,
3912 .splice_read
= generic_file_splice_read
,
3913 .splice_write
= iter_file_splice_write
,
3914 .fallocate
= shmem_fallocate
,
3918 static const struct inode_operations shmem_inode_operations
= {
3919 .getattr
= shmem_getattr
,
3920 .setattr
= shmem_setattr
,
3921 #ifdef CONFIG_TMPFS_XATTR
3922 .listxattr
= shmem_listxattr
,
3923 .set_acl
= simple_set_acl
,
3927 static const struct inode_operations shmem_dir_inode_operations
= {
3929 .create
= shmem_create
,
3930 .lookup
= simple_lookup
,
3932 .unlink
= shmem_unlink
,
3933 .symlink
= shmem_symlink
,
3934 .mkdir
= shmem_mkdir
,
3935 .rmdir
= shmem_rmdir
,
3936 .mknod
= shmem_mknod
,
3937 .rename
= shmem_rename2
,
3938 .tmpfile
= shmem_tmpfile
,
3940 #ifdef CONFIG_TMPFS_XATTR
3941 .listxattr
= shmem_listxattr
,
3943 #ifdef CONFIG_TMPFS_POSIX_ACL
3944 .setattr
= shmem_setattr
,
3945 .set_acl
= simple_set_acl
,
3949 static const struct inode_operations shmem_special_inode_operations
= {
3950 #ifdef CONFIG_TMPFS_XATTR
3951 .listxattr
= shmem_listxattr
,
3953 #ifdef CONFIG_TMPFS_POSIX_ACL
3954 .setattr
= shmem_setattr
,
3955 .set_acl
= simple_set_acl
,
3959 static const struct super_operations shmem_ops
= {
3960 .alloc_inode
= shmem_alloc_inode
,
3961 .destroy_inode
= shmem_destroy_inode
,
3963 .statfs
= shmem_statfs
,
3964 .remount_fs
= shmem_remount_fs
,
3965 .show_options
= shmem_show_options
,
3967 .evict_inode
= shmem_evict_inode
,
3968 .drop_inode
= generic_delete_inode
,
3969 .put_super
= shmem_put_super
,
3970 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3971 .nr_cached_objects
= shmem_unused_huge_count
,
3972 .free_cached_objects
= shmem_unused_huge_scan
,
3976 static const struct vm_operations_struct shmem_vm_ops
= {
3977 .fault
= shmem_fault
,
3978 .map_pages
= filemap_map_pages
,
3980 .set_policy
= shmem_set_policy
,
3981 .get_policy
= shmem_get_policy
,
3985 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3986 int flags
, const char *dev_name
, void *data
)
3988 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3991 static struct file_system_type shmem_fs_type
= {
3992 .owner
= THIS_MODULE
,
3994 .mount
= shmem_mount
,
3995 .kill_sb
= kill_litter_super
,
3996 .fs_flags
= FS_USERNS_MOUNT
,
3999 int __init
shmem_init(void)
4003 /* If rootfs called this, don't re-init */
4004 if (shmem_inode_cachep
)
4007 error
= shmem_init_inodecache();
4011 error
= register_filesystem(&shmem_fs_type
);
4013 pr_err("Could not register tmpfs\n");
4017 shm_mnt
= kern_mount(&shmem_fs_type
);
4018 if (IS_ERR(shm_mnt
)) {
4019 error
= PTR_ERR(shm_mnt
);
4020 pr_err("Could not kern_mount tmpfs\n");
4024 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4025 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
4026 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4028 shmem_huge
= 0; /* just in case it was patched */
4033 unregister_filesystem(&shmem_fs_type
);
4035 shmem_destroy_inodecache();
4037 shm_mnt
= ERR_PTR(error
);
4041 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
4042 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4043 struct kobj_attribute
*attr
, char *buf
)
4047 SHMEM_HUGE_WITHIN_SIZE
,
4055 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
4056 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
4058 count
+= sprintf(buf
+ count
, fmt
,
4059 shmem_format_huge(values
[i
]));
4061 buf
[count
- 1] = '\n';
4065 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4066 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4071 if (count
+ 1 > sizeof(tmp
))
4073 memcpy(tmp
, buf
, count
);
4075 if (count
&& tmp
[count
- 1] == '\n')
4076 tmp
[count
- 1] = '\0';
4078 huge
= shmem_parse_huge(tmp
);
4079 if (huge
== -EINVAL
)
4081 if (!has_transparent_hugepage() &&
4082 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4086 if (shmem_huge
> SHMEM_HUGE_DENY
)
4087 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4091 struct kobj_attribute shmem_enabled_attr
=
4092 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4093 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4095 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4096 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4098 struct inode
*inode
= file_inode(vma
->vm_file
);
4099 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4103 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4105 if (shmem_huge
== SHMEM_HUGE_DENY
)
4107 switch (sbinfo
->huge
) {
4108 case SHMEM_HUGE_NEVER
:
4110 case SHMEM_HUGE_ALWAYS
:
4112 case SHMEM_HUGE_WITHIN_SIZE
:
4113 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4114 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4115 if (i_size
>= HPAGE_PMD_SIZE
&&
4116 i_size
>> PAGE_SHIFT
>= off
)
4118 case SHMEM_HUGE_ADVISE
:
4119 /* TODO: implement fadvise() hints */
4120 return (vma
->vm_flags
& VM_HUGEPAGE
);
4126 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4128 #else /* !CONFIG_SHMEM */
4131 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4133 * This is intended for small system where the benefits of the full
4134 * shmem code (swap-backed and resource-limited) are outweighed by
4135 * their complexity. On systems without swap this code should be
4136 * effectively equivalent, but much lighter weight.
4139 static struct file_system_type shmem_fs_type
= {
4141 .mount
= ramfs_mount
,
4142 .kill_sb
= kill_litter_super
,
4143 .fs_flags
= FS_USERNS_MOUNT
,
4146 int __init
shmem_init(void)
4148 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4150 shm_mnt
= kern_mount(&shmem_fs_type
);
4151 BUG_ON(IS_ERR(shm_mnt
));
4156 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
4161 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4166 void shmem_unlock_mapping(struct address_space
*mapping
)
4171 unsigned long shmem_get_unmapped_area(struct file
*file
,
4172 unsigned long addr
, unsigned long len
,
4173 unsigned long pgoff
, unsigned long flags
)
4175 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4179 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4181 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4183 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4185 #define shmem_vm_ops generic_file_vm_ops
4186 #define shmem_file_operations ramfs_file_operations
4187 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4188 #define shmem_acct_size(flags, size) 0
4189 #define shmem_unacct_size(flags, size) do {} while (0)
4191 #endif /* CONFIG_SHMEM */
4195 static const struct dentry_operations anon_ops
= {
4196 .d_dname
= simple_dname
4199 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
4200 unsigned long flags
, unsigned int i_flags
)
4203 struct inode
*inode
;
4205 struct super_block
*sb
;
4208 if (IS_ERR(shm_mnt
))
4209 return ERR_CAST(shm_mnt
);
4211 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4212 return ERR_PTR(-EINVAL
);
4214 if (shmem_acct_size(flags
, size
))
4215 return ERR_PTR(-ENOMEM
);
4217 res
= ERR_PTR(-ENOMEM
);
4219 this.len
= strlen(name
);
4220 this.hash
= 0; /* will go */
4221 sb
= shm_mnt
->mnt_sb
;
4222 path
.mnt
= mntget(shm_mnt
);
4223 path
.dentry
= d_alloc_pseudo(sb
, &this);
4226 d_set_d_op(path
.dentry
, &anon_ops
);
4228 res
= ERR_PTR(-ENOSPC
);
4229 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
4233 inode
->i_flags
|= i_flags
;
4234 d_instantiate(path
.dentry
, inode
);
4235 inode
->i_size
= size
;
4236 clear_nlink(inode
); /* It is unlinked */
4237 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4241 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
4242 &shmem_file_operations
);
4249 shmem_unacct_size(flags
, size
);
4256 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4257 * kernel internal. There will be NO LSM permission checks against the
4258 * underlying inode. So users of this interface must do LSM checks at a
4259 * higher layer. The users are the big_key and shm implementations. LSM
4260 * checks are provided at the key or shm level rather than the inode.
4261 * @name: name for dentry (to be seen in /proc/<pid>/maps
4262 * @size: size to be set for the file
4263 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4265 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4267 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
4271 * shmem_file_setup - get an unlinked file living in tmpfs
4272 * @name: name for dentry (to be seen in /proc/<pid>/maps
4273 * @size: size to be set for the file
4274 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4276 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4278 return __shmem_file_setup(name
, size
, flags
, 0);
4280 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4283 * shmem_zero_setup - setup a shared anonymous mapping
4284 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4286 int shmem_zero_setup(struct vm_area_struct
*vma
)
4289 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4292 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4293 * between XFS directory reading and selinux: since this file is only
4294 * accessible to the user through its mapping, use S_PRIVATE flag to
4295 * bypass file security, in the same way as shmem_kernel_file_setup().
4297 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
4299 return PTR_ERR(file
);
4303 vma
->vm_file
= file
;
4304 vma
->vm_ops
= &shmem_vm_ops
;
4306 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4307 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4308 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4309 khugepaged_enter(vma
, vma
->vm_flags
);
4316 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4317 * @mapping: the page's address_space
4318 * @index: the page index
4319 * @gfp: the page allocator flags to use if allocating
4321 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4322 * with any new page allocations done using the specified allocation flags.
4323 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4324 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4325 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4327 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4328 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4330 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4331 pgoff_t index
, gfp_t gfp
)
4334 struct inode
*inode
= mapping
->host
;
4338 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4339 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4340 gfp
, NULL
, NULL
, NULL
);
4342 page
= ERR_PTR(error
);
4348 * The tiny !SHMEM case uses ramfs without swap
4350 return read_cache_page_gfp(mapping
, index
, gfp
);
4353 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);