2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/sched/signal.h>
33 #include <linux/export.h>
34 #include <linux/swap.h>
35 #include <linux/uio.h>
36 #include <linux/khugepaged.h>
37 #include <linux/hugetlb.h>
39 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
41 static struct vfsmount
*shm_mnt
;
45 * This virtual memory filesystem is heavily based on the ramfs. It
46 * extends ramfs by the ability to use swap and honor resource limits
47 * which makes it a completely usable filesystem.
50 #include <linux/xattr.h>
51 #include <linux/exportfs.h>
52 #include <linux/posix_acl.h>
53 #include <linux/posix_acl_xattr.h>
54 #include <linux/mman.h>
55 #include <linux/string.h>
56 #include <linux/slab.h>
57 #include <linux/backing-dev.h>
58 #include <linux/shmem_fs.h>
59 #include <linux/writeback.h>
60 #include <linux/blkdev.h>
61 #include <linux/pagevec.h>
62 #include <linux/percpu_counter.h>
63 #include <linux/falloc.h>
64 #include <linux/splice.h>
65 #include <linux/security.h>
66 #include <linux/swapops.h>
67 #include <linux/mempolicy.h>
68 #include <linux/namei.h>
69 #include <linux/ctype.h>
70 #include <linux/migrate.h>
71 #include <linux/highmem.h>
72 #include <linux/seq_file.h>
73 #include <linux/magic.h>
74 #include <linux/syscalls.h>
75 #include <linux/fcntl.h>
76 #include <uapi/linux/memfd.h>
77 #include <linux/userfaultfd_k.h>
78 #include <linux/rmap.h>
79 #include <linux/uuid.h>
81 #include <linux/uaccess.h>
82 #include <asm/pgtable.h>
86 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 /* Pretend that each entry is of this size in directory's i_size */
90 #define BOGO_DIRENT_SIZE 20
92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93 #define SHORT_SYMLINK_LEN 128
96 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_mutex making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
100 struct shmem_falloc
{
101 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
102 pgoff_t start
; /* start of range currently being fallocated */
103 pgoff_t next
; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
109 static unsigned long shmem_default_max_blocks(void)
111 return totalram_pages
/ 2;
114 static int shmem_default_max_inodes(void)
119 ul
= min3(ul
, totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
124 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
125 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
126 struct shmem_inode_info
*info
, pgoff_t index
);
127 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
128 struct page
**pagep
, enum sgp_type sgp
,
129 gfp_t gfp
, struct vm_area_struct
*vma
,
130 struct vm_fault
*vmf
, int *fault_type
);
132 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
133 struct page
**pagep
, enum sgp_type sgp
)
135 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
136 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
139 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
141 return sb
->s_fs_info
;
145 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
146 * for shared memory and for shared anonymous (/dev/zero) mappings
147 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
148 * consistent with the pre-accounting of private mappings ...
150 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
152 return (flags
& VM_NORESERVE
) ?
153 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
156 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
158 if (!(flags
& VM_NORESERVE
))
159 vm_unacct_memory(VM_ACCT(size
));
162 static inline int shmem_reacct_size(unsigned long flags
,
163 loff_t oldsize
, loff_t newsize
)
165 if (!(flags
& VM_NORESERVE
)) {
166 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
167 return security_vm_enough_memory_mm(current
->mm
,
168 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
169 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
170 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
176 * ... whereas tmpfs objects are accounted incrementally as
177 * pages are allocated, in order to allow large sparse files.
178 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
179 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
181 static inline int shmem_acct_block(unsigned long flags
, long pages
)
183 if (!(flags
& VM_NORESERVE
))
186 return security_vm_enough_memory_mm(current
->mm
,
187 pages
* VM_ACCT(PAGE_SIZE
));
190 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
192 if (flags
& VM_NORESERVE
)
193 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
196 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
198 struct shmem_inode_info
*info
= SHMEM_I(inode
);
199 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
201 if (shmem_acct_block(info
->flags
, pages
))
204 if (sbinfo
->max_blocks
) {
205 if (percpu_counter_compare(&sbinfo
->used_blocks
,
206 sbinfo
->max_blocks
- pages
) > 0)
208 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
214 shmem_unacct_blocks(info
->flags
, pages
);
218 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
220 struct shmem_inode_info
*info
= SHMEM_I(inode
);
221 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
223 if (sbinfo
->max_blocks
)
224 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
225 shmem_unacct_blocks(info
->flags
, pages
);
228 static const struct super_operations shmem_ops
;
229 static const struct address_space_operations shmem_aops
;
230 static const struct file_operations shmem_file_operations
;
231 static const struct inode_operations shmem_inode_operations
;
232 static const struct inode_operations shmem_dir_inode_operations
;
233 static const struct inode_operations shmem_special_inode_operations
;
234 static const struct vm_operations_struct shmem_vm_ops
;
235 static struct file_system_type shmem_fs_type
;
237 bool vma_is_shmem(struct vm_area_struct
*vma
)
239 return vma
->vm_ops
== &shmem_vm_ops
;
242 static LIST_HEAD(shmem_swaplist
);
243 static DEFINE_MUTEX(shmem_swaplist_mutex
);
245 static int shmem_reserve_inode(struct super_block
*sb
)
247 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
248 if (sbinfo
->max_inodes
) {
249 spin_lock(&sbinfo
->stat_lock
);
250 if (!sbinfo
->free_inodes
) {
251 spin_unlock(&sbinfo
->stat_lock
);
254 sbinfo
->free_inodes
--;
255 spin_unlock(&sbinfo
->stat_lock
);
260 static void shmem_free_inode(struct super_block
*sb
)
262 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
263 if (sbinfo
->max_inodes
) {
264 spin_lock(&sbinfo
->stat_lock
);
265 sbinfo
->free_inodes
++;
266 spin_unlock(&sbinfo
->stat_lock
);
271 * shmem_recalc_inode - recalculate the block usage of an inode
272 * @inode: inode to recalc
274 * We have to calculate the free blocks since the mm can drop
275 * undirtied hole pages behind our back.
277 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
278 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
280 * It has to be called with the spinlock held.
282 static void shmem_recalc_inode(struct inode
*inode
)
284 struct shmem_inode_info
*info
= SHMEM_I(inode
);
287 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
289 info
->alloced
-= freed
;
290 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
291 shmem_inode_unacct_blocks(inode
, freed
);
295 bool shmem_charge(struct inode
*inode
, long pages
)
297 struct shmem_inode_info
*info
= SHMEM_I(inode
);
300 if (!shmem_inode_acct_block(inode
, pages
))
303 spin_lock_irqsave(&info
->lock
, flags
);
304 info
->alloced
+= pages
;
305 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
306 shmem_recalc_inode(inode
);
307 spin_unlock_irqrestore(&info
->lock
, flags
);
308 inode
->i_mapping
->nrpages
+= pages
;
313 void shmem_uncharge(struct inode
*inode
, long pages
)
315 struct shmem_inode_info
*info
= SHMEM_I(inode
);
318 spin_lock_irqsave(&info
->lock
, flags
);
319 info
->alloced
-= pages
;
320 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
321 shmem_recalc_inode(inode
);
322 spin_unlock_irqrestore(&info
->lock
, flags
);
324 shmem_inode_unacct_blocks(inode
, pages
);
328 * Replace item expected in radix tree by a new item, while holding tree lock.
330 static int shmem_radix_tree_replace(struct address_space
*mapping
,
331 pgoff_t index
, void *expected
, void *replacement
)
333 struct radix_tree_node
*node
;
337 VM_BUG_ON(!expected
);
338 VM_BUG_ON(!replacement
);
339 item
= __radix_tree_lookup(&mapping
->page_tree
, index
, &node
, &pslot
);
342 if (item
!= expected
)
344 __radix_tree_replace(&mapping
->page_tree
, node
, pslot
,
350 * Sometimes, before we decide whether to proceed or to fail, we must check
351 * that an entry was not already brought back from swap by a racing thread.
353 * Checking page is not enough: by the time a SwapCache page is locked, it
354 * might be reused, and again be SwapCache, using the same swap as before.
356 static bool shmem_confirm_swap(struct address_space
*mapping
,
357 pgoff_t index
, swp_entry_t swap
)
362 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
364 return item
== swp_to_radix_entry(swap
);
368 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
371 * disables huge pages for the mount;
373 * enables huge pages for the mount;
374 * SHMEM_HUGE_WITHIN_SIZE:
375 * only allocate huge pages if the page will be fully within i_size,
376 * also respect fadvise()/madvise() hints;
378 * only allocate huge pages if requested with fadvise()/madvise();
381 #define SHMEM_HUGE_NEVER 0
382 #define SHMEM_HUGE_ALWAYS 1
383 #define SHMEM_HUGE_WITHIN_SIZE 2
384 #define SHMEM_HUGE_ADVISE 3
388 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
391 * disables huge on shm_mnt and all mounts, for emergency use;
393 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
396 #define SHMEM_HUGE_DENY (-1)
397 #define SHMEM_HUGE_FORCE (-2)
399 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
400 /* ifdef here to avoid bloating shmem.o when not necessary */
402 int shmem_huge __read_mostly
;
404 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
405 static int shmem_parse_huge(const char *str
)
407 if (!strcmp(str
, "never"))
408 return SHMEM_HUGE_NEVER
;
409 if (!strcmp(str
, "always"))
410 return SHMEM_HUGE_ALWAYS
;
411 if (!strcmp(str
, "within_size"))
412 return SHMEM_HUGE_WITHIN_SIZE
;
413 if (!strcmp(str
, "advise"))
414 return SHMEM_HUGE_ADVISE
;
415 if (!strcmp(str
, "deny"))
416 return SHMEM_HUGE_DENY
;
417 if (!strcmp(str
, "force"))
418 return SHMEM_HUGE_FORCE
;
422 static const char *shmem_format_huge(int huge
)
425 case SHMEM_HUGE_NEVER
:
427 case SHMEM_HUGE_ALWAYS
:
429 case SHMEM_HUGE_WITHIN_SIZE
:
430 return "within_size";
431 case SHMEM_HUGE_ADVISE
:
433 case SHMEM_HUGE_DENY
:
435 case SHMEM_HUGE_FORCE
:
444 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
445 struct shrink_control
*sc
, unsigned long nr_to_split
)
447 LIST_HEAD(list
), *pos
, *next
;
448 LIST_HEAD(to_remove
);
450 struct shmem_inode_info
*info
;
452 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
453 int removed
= 0, split
= 0;
455 if (list_empty(&sbinfo
->shrinklist
))
458 spin_lock(&sbinfo
->shrinklist_lock
);
459 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
460 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
463 inode
= igrab(&info
->vfs_inode
);
465 /* inode is about to be evicted */
467 list_del_init(&info
->shrinklist
);
472 /* Check if there's anything to gain */
473 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
474 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
475 list_move(&info
->shrinklist
, &to_remove
);
480 list_move(&info
->shrinklist
, &list
);
485 spin_unlock(&sbinfo
->shrinklist_lock
);
487 list_for_each_safe(pos
, next
, &to_remove
) {
488 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
489 inode
= &info
->vfs_inode
;
490 list_del_init(&info
->shrinklist
);
494 list_for_each_safe(pos
, next
, &list
) {
497 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
498 inode
= &info
->vfs_inode
;
500 if (nr_to_split
&& split
>= nr_to_split
) {
505 page
= find_lock_page(inode
->i_mapping
,
506 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
510 if (!PageTransHuge(page
)) {
516 ret
= split_huge_page(page
);
521 /* split failed: leave it on the list */
528 list_del_init(&info
->shrinklist
);
533 spin_lock(&sbinfo
->shrinklist_lock
);
534 list_splice_tail(&list
, &sbinfo
->shrinklist
);
535 sbinfo
->shrinklist_len
-= removed
;
536 spin_unlock(&sbinfo
->shrinklist_lock
);
541 static long shmem_unused_huge_scan(struct super_block
*sb
,
542 struct shrink_control
*sc
)
544 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
546 if (!READ_ONCE(sbinfo
->shrinklist_len
))
549 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
552 static long shmem_unused_huge_count(struct super_block
*sb
,
553 struct shrink_control
*sc
)
555 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
556 return READ_ONCE(sbinfo
->shrinklist_len
);
558 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
560 #define shmem_huge SHMEM_HUGE_DENY
562 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
563 struct shrink_control
*sc
, unsigned long nr_to_split
)
567 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
570 * Like add_to_page_cache_locked, but error if expected item has gone.
572 static int shmem_add_to_page_cache(struct page
*page
,
573 struct address_space
*mapping
,
574 pgoff_t index
, void *expected
)
576 int error
, nr
= hpage_nr_pages(page
);
578 VM_BUG_ON_PAGE(PageTail(page
), page
);
579 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
580 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
581 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
582 VM_BUG_ON(expected
&& PageTransHuge(page
));
584 page_ref_add(page
, nr
);
585 page
->mapping
= mapping
;
588 spin_lock_irq(&mapping
->tree_lock
);
589 if (PageTransHuge(page
)) {
590 void __rcu
**results
;
595 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
,
596 &results
, &idx
, index
, 1) &&
597 idx
< index
+ HPAGE_PMD_NR
) {
602 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
603 error
= radix_tree_insert(&mapping
->page_tree
,
604 index
+ i
, page
+ i
);
607 count_vm_event(THP_FILE_ALLOC
);
609 } else if (!expected
) {
610 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
612 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
617 mapping
->nrpages
+= nr
;
618 if (PageTransHuge(page
))
619 __inc_node_page_state(page
, NR_SHMEM_THPS
);
620 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
621 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
622 spin_unlock_irq(&mapping
->tree_lock
);
624 page
->mapping
= NULL
;
625 spin_unlock_irq(&mapping
->tree_lock
);
626 page_ref_sub(page
, nr
);
632 * Like delete_from_page_cache, but substitutes swap for page.
634 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
636 struct address_space
*mapping
= page
->mapping
;
639 VM_BUG_ON_PAGE(PageCompound(page
), page
);
641 spin_lock_irq(&mapping
->tree_lock
);
642 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
643 page
->mapping
= NULL
;
645 __dec_node_page_state(page
, NR_FILE_PAGES
);
646 __dec_node_page_state(page
, NR_SHMEM
);
647 spin_unlock_irq(&mapping
->tree_lock
);
653 * Remove swap entry from radix tree, free the swap and its page cache.
655 static int shmem_free_swap(struct address_space
*mapping
,
656 pgoff_t index
, void *radswap
)
660 spin_lock_irq(&mapping
->tree_lock
);
661 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
662 spin_unlock_irq(&mapping
->tree_lock
);
665 free_swap_and_cache(radix_to_swp_entry(radswap
));
670 * Determine (in bytes) how many of the shmem object's pages mapped by the
671 * given offsets are swapped out.
673 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
674 * as long as the inode doesn't go away and racy results are not a problem.
676 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
677 pgoff_t start
, pgoff_t end
)
679 struct radix_tree_iter iter
;
682 unsigned long swapped
= 0;
686 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
687 if (iter
.index
>= end
)
690 page
= radix_tree_deref_slot(slot
);
692 if (radix_tree_deref_retry(page
)) {
693 slot
= radix_tree_iter_retry(&iter
);
697 if (radix_tree_exceptional_entry(page
))
700 if (need_resched()) {
701 slot
= radix_tree_iter_resume(slot
, &iter
);
708 return swapped
<< PAGE_SHIFT
;
712 * Determine (in bytes) how many of the shmem object's pages mapped by the
713 * given vma is swapped out.
715 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
716 * as long as the inode doesn't go away and racy results are not a problem.
718 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
720 struct inode
*inode
= file_inode(vma
->vm_file
);
721 struct shmem_inode_info
*info
= SHMEM_I(inode
);
722 struct address_space
*mapping
= inode
->i_mapping
;
723 unsigned long swapped
;
725 /* Be careful as we don't hold info->lock */
726 swapped
= READ_ONCE(info
->swapped
);
729 * The easier cases are when the shmem object has nothing in swap, or
730 * the vma maps it whole. Then we can simply use the stats that we
736 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
737 return swapped
<< PAGE_SHIFT
;
739 /* Here comes the more involved part */
740 return shmem_partial_swap_usage(mapping
,
741 linear_page_index(vma
, vma
->vm_start
),
742 linear_page_index(vma
, vma
->vm_end
));
746 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
748 void shmem_unlock_mapping(struct address_space
*mapping
)
751 pgoff_t indices
[PAGEVEC_SIZE
];
756 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
758 while (!mapping_unevictable(mapping
)) {
760 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
761 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
763 pvec
.nr
= find_get_entries(mapping
, index
,
764 PAGEVEC_SIZE
, pvec
.pages
, indices
);
767 index
= indices
[pvec
.nr
- 1] + 1;
768 pagevec_remove_exceptionals(&pvec
);
769 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
770 pagevec_release(&pvec
);
776 * Remove range of pages and swap entries from radix tree, and free them.
777 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
779 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
782 struct address_space
*mapping
= inode
->i_mapping
;
783 struct shmem_inode_info
*info
= SHMEM_I(inode
);
784 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
785 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
786 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
787 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
789 pgoff_t indices
[PAGEVEC_SIZE
];
790 long nr_swaps_freed
= 0;
795 end
= -1; /* unsigned, so actually very big */
799 while (index
< end
) {
800 pvec
.nr
= find_get_entries(mapping
, index
,
801 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
802 pvec
.pages
, indices
);
805 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
806 struct page
*page
= pvec
.pages
[i
];
812 if (radix_tree_exceptional_entry(page
)) {
815 nr_swaps_freed
+= !shmem_free_swap(mapping
,
820 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
822 if (!trylock_page(page
))
825 if (PageTransTail(page
)) {
826 /* Middle of THP: zero out the page */
827 clear_highpage(page
);
830 } else if (PageTransHuge(page
)) {
831 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
833 * Range ends in the middle of THP:
836 clear_highpage(page
);
840 index
+= HPAGE_PMD_NR
- 1;
841 i
+= HPAGE_PMD_NR
- 1;
844 if (!unfalloc
|| !PageUptodate(page
)) {
845 VM_BUG_ON_PAGE(PageTail(page
), page
);
846 if (page_mapping(page
) == mapping
) {
847 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
848 truncate_inode_page(mapping
, page
);
853 pagevec_remove_exceptionals(&pvec
);
854 pagevec_release(&pvec
);
860 struct page
*page
= NULL
;
861 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
863 unsigned int top
= PAGE_SIZE
;
868 zero_user_segment(page
, partial_start
, top
);
869 set_page_dirty(page
);
875 struct page
*page
= NULL
;
876 shmem_getpage(inode
, end
, &page
, SGP_READ
);
878 zero_user_segment(page
, 0, partial_end
);
879 set_page_dirty(page
);
888 while (index
< end
) {
891 pvec
.nr
= find_get_entries(mapping
, index
,
892 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
893 pvec
.pages
, indices
);
895 /* If all gone or hole-punch or unfalloc, we're done */
896 if (index
== start
|| end
!= -1)
898 /* But if truncating, restart to make sure all gone */
902 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
903 struct page
*page
= pvec
.pages
[i
];
909 if (radix_tree_exceptional_entry(page
)) {
912 if (shmem_free_swap(mapping
, index
, page
)) {
913 /* Swap was replaced by page: retry */
923 if (PageTransTail(page
)) {
924 /* Middle of THP: zero out the page */
925 clear_highpage(page
);
928 * Partial thp truncate due 'start' in middle
929 * of THP: don't need to look on these pages
930 * again on !pvec.nr restart.
932 if (index
!= round_down(end
, HPAGE_PMD_NR
))
935 } else if (PageTransHuge(page
)) {
936 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
938 * Range ends in the middle of THP:
941 clear_highpage(page
);
945 index
+= HPAGE_PMD_NR
- 1;
946 i
+= HPAGE_PMD_NR
- 1;
949 if (!unfalloc
|| !PageUptodate(page
)) {
950 VM_BUG_ON_PAGE(PageTail(page
), page
);
951 if (page_mapping(page
) == mapping
) {
952 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
953 truncate_inode_page(mapping
, page
);
955 /* Page was replaced by swap: retry */
963 pagevec_remove_exceptionals(&pvec
);
964 pagevec_release(&pvec
);
968 spin_lock_irq(&info
->lock
);
969 info
->swapped
-= nr_swaps_freed
;
970 shmem_recalc_inode(inode
);
971 spin_unlock_irq(&info
->lock
);
974 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
976 shmem_undo_range(inode
, lstart
, lend
, false);
977 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
979 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
981 static int shmem_getattr(const struct path
*path
, struct kstat
*stat
,
982 u32 request_mask
, unsigned int query_flags
)
984 struct inode
*inode
= path
->dentry
->d_inode
;
985 struct shmem_inode_info
*info
= SHMEM_I(inode
);
987 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
988 spin_lock_irq(&info
->lock
);
989 shmem_recalc_inode(inode
);
990 spin_unlock_irq(&info
->lock
);
992 generic_fillattr(inode
, stat
);
996 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
998 struct inode
*inode
= d_inode(dentry
);
999 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1000 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1003 error
= setattr_prepare(dentry
, attr
);
1007 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1008 loff_t oldsize
= inode
->i_size
;
1009 loff_t newsize
= attr
->ia_size
;
1011 /* protected by i_mutex */
1012 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1013 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1016 if (newsize
!= oldsize
) {
1017 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1021 i_size_write(inode
, newsize
);
1022 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1024 if (newsize
<= oldsize
) {
1025 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1026 if (oldsize
> holebegin
)
1027 unmap_mapping_range(inode
->i_mapping
,
1030 shmem_truncate_range(inode
,
1031 newsize
, (loff_t
)-1);
1032 /* unmap again to remove racily COWed private pages */
1033 if (oldsize
> holebegin
)
1034 unmap_mapping_range(inode
->i_mapping
,
1038 * Part of the huge page can be beyond i_size: subject
1039 * to shrink under memory pressure.
1041 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1042 spin_lock(&sbinfo
->shrinklist_lock
);
1044 * _careful to defend against unlocked access to
1045 * ->shrink_list in shmem_unused_huge_shrink()
1047 if (list_empty_careful(&info
->shrinklist
)) {
1048 list_add_tail(&info
->shrinklist
,
1049 &sbinfo
->shrinklist
);
1050 sbinfo
->shrinklist_len
++;
1052 spin_unlock(&sbinfo
->shrinklist_lock
);
1057 setattr_copy(inode
, attr
);
1058 if (attr
->ia_valid
& ATTR_MODE
)
1059 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1063 static void shmem_evict_inode(struct inode
*inode
)
1065 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1066 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1068 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1069 shmem_unacct_size(info
->flags
, inode
->i_size
);
1071 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1072 if (!list_empty(&info
->shrinklist
)) {
1073 spin_lock(&sbinfo
->shrinklist_lock
);
1074 if (!list_empty(&info
->shrinklist
)) {
1075 list_del_init(&info
->shrinklist
);
1076 sbinfo
->shrinklist_len
--;
1078 spin_unlock(&sbinfo
->shrinklist_lock
);
1080 if (!list_empty(&info
->swaplist
)) {
1081 mutex_lock(&shmem_swaplist_mutex
);
1082 list_del_init(&info
->swaplist
);
1083 mutex_unlock(&shmem_swaplist_mutex
);
1087 simple_xattrs_free(&info
->xattrs
);
1088 WARN_ON(inode
->i_blocks
);
1089 if (!sbinfo
->idr_nouse
&& inode
->i_ino
) {
1090 mutex_lock(&sbinfo
->idr_lock
);
1091 idr_remove(&sbinfo
->idr
, inode
->i_ino
);
1092 mutex_unlock(&sbinfo
->idr_lock
);
1094 shmem_free_inode(inode
->i_sb
);
1098 static unsigned long find_swap_entry(struct radix_tree_root
*root
, void *item
)
1100 struct radix_tree_iter iter
;
1102 unsigned long found
= -1;
1103 unsigned int checked
= 0;
1106 radix_tree_for_each_slot(slot
, root
, &iter
, 0) {
1107 if (*slot
== item
) {
1112 if ((checked
% 4096) != 0)
1114 slot
= radix_tree_iter_resume(slot
, &iter
);
1123 * If swap found in inode, free it and move page from swapcache to filecache.
1125 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
1126 swp_entry_t swap
, struct page
**pagep
)
1128 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1134 radswap
= swp_to_radix_entry(swap
);
1135 index
= find_swap_entry(&mapping
->page_tree
, radswap
);
1137 return -EAGAIN
; /* tell shmem_unuse we found nothing */
1140 * Move _head_ to start search for next from here.
1141 * But be careful: shmem_evict_inode checks list_empty without taking
1142 * mutex, and there's an instant in list_move_tail when info->swaplist
1143 * would appear empty, if it were the only one on shmem_swaplist.
1145 if (shmem_swaplist
.next
!= &info
->swaplist
)
1146 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
1148 gfp
= mapping_gfp_mask(mapping
);
1149 if (shmem_should_replace_page(*pagep
, gfp
)) {
1150 mutex_unlock(&shmem_swaplist_mutex
);
1151 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
1152 mutex_lock(&shmem_swaplist_mutex
);
1154 * We needed to drop mutex to make that restrictive page
1155 * allocation, but the inode might have been freed while we
1156 * dropped it: although a racing shmem_evict_inode() cannot
1157 * complete without emptying the radix_tree, our page lock
1158 * on this swapcache page is not enough to prevent that -
1159 * free_swap_and_cache() of our swap entry will only
1160 * trylock_page(), removing swap from radix_tree whatever.
1162 * We must not proceed to shmem_add_to_page_cache() if the
1163 * inode has been freed, but of course we cannot rely on
1164 * inode or mapping or info to check that. However, we can
1165 * safely check if our swap entry is still in use (and here
1166 * it can't have got reused for another page): if it's still
1167 * in use, then the inode cannot have been freed yet, and we
1168 * can safely proceed (if it's no longer in use, that tells
1169 * nothing about the inode, but we don't need to unuse swap).
1171 if (!page_swapcount(*pagep
))
1176 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1177 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1178 * beneath us (pagelock doesn't help until the page is in pagecache).
1181 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
1183 if (error
!= -ENOMEM
) {
1185 * Truncation and eviction use free_swap_and_cache(), which
1186 * only does trylock page: if we raced, best clean up here.
1188 delete_from_swap_cache(*pagep
);
1189 set_page_dirty(*pagep
);
1191 spin_lock_irq(&info
->lock
);
1193 spin_unlock_irq(&info
->lock
);
1201 * Search through swapped inodes to find and replace swap by page.
1203 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1205 struct list_head
*this, *next
;
1206 struct shmem_inode_info
*info
;
1207 struct mem_cgroup
*memcg
;
1211 * There's a faint possibility that swap page was replaced before
1212 * caller locked it: caller will come back later with the right page.
1214 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1218 * Charge page using GFP_KERNEL while we can wait, before taking
1219 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1220 * Charged back to the user (not to caller) when swap account is used.
1222 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1226 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1229 mutex_lock(&shmem_swaplist_mutex
);
1230 list_for_each_safe(this, next
, &shmem_swaplist
) {
1231 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1233 error
= shmem_unuse_inode(info
, swap
, &page
);
1235 list_del_init(&info
->swaplist
);
1237 if (error
!= -EAGAIN
)
1239 /* found nothing in this: move on to search the next */
1241 mutex_unlock(&shmem_swaplist_mutex
);
1244 if (error
!= -ENOMEM
)
1246 mem_cgroup_cancel_charge(page
, memcg
, false);
1248 mem_cgroup_commit_charge(page
, memcg
, true, false);
1256 * Move the page from the page cache to the swap cache.
1258 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1260 struct shmem_inode_info
*info
;
1261 struct address_space
*mapping
;
1262 struct inode
*inode
;
1266 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1267 BUG_ON(!PageLocked(page
));
1268 mapping
= page
->mapping
;
1269 index
= page
->index
;
1270 inode
= mapping
->host
;
1271 info
= SHMEM_I(inode
);
1272 if (info
->flags
& VM_LOCKED
)
1274 if (!total_swap_pages
)
1278 * Our capabilities prevent regular writeback or sync from ever calling
1279 * shmem_writepage; but a stacking filesystem might use ->writepage of
1280 * its underlying filesystem, in which case tmpfs should write out to
1281 * swap only in response to memory pressure, and not for the writeback
1284 if (!wbc
->for_reclaim
) {
1285 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1290 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1291 * value into swapfile.c, the only way we can correctly account for a
1292 * fallocated page arriving here is now to initialize it and write it.
1294 * That's okay for a page already fallocated earlier, but if we have
1295 * not yet completed the fallocation, then (a) we want to keep track
1296 * of this page in case we have to undo it, and (b) it may not be a
1297 * good idea to continue anyway, once we're pushing into swap. So
1298 * reactivate the page, and let shmem_fallocate() quit when too many.
1300 if (!PageUptodate(page
)) {
1301 if (inode
->i_private
) {
1302 struct shmem_falloc
*shmem_falloc
;
1303 spin_lock(&inode
->i_lock
);
1304 shmem_falloc
= inode
->i_private
;
1306 !shmem_falloc
->waitq
&&
1307 index
>= shmem_falloc
->start
&&
1308 index
< shmem_falloc
->next
)
1309 shmem_falloc
->nr_unswapped
++;
1311 shmem_falloc
= NULL
;
1312 spin_unlock(&inode
->i_lock
);
1316 clear_highpage(page
);
1317 flush_dcache_page(page
);
1318 SetPageUptodate(page
);
1321 swap
= get_swap_page(page
);
1325 if (mem_cgroup_try_charge_swap(page
, swap
))
1329 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1330 * if it's not already there. Do it now before the page is
1331 * moved to swap cache, when its pagelock no longer protects
1332 * the inode from eviction. But don't unlock the mutex until
1333 * we've incremented swapped, because shmem_unuse_inode() will
1334 * prune a !swapped inode from the swaplist under this mutex.
1336 mutex_lock(&shmem_swaplist_mutex
);
1337 if (list_empty(&info
->swaplist
))
1338 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1340 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1341 spin_lock_irq(&info
->lock
);
1342 shmem_recalc_inode(inode
);
1344 spin_unlock_irq(&info
->lock
);
1346 swap_shmem_alloc(swap
);
1347 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1349 mutex_unlock(&shmem_swaplist_mutex
);
1350 BUG_ON(page_mapped(page
));
1351 swap_writepage(page
, wbc
);
1355 mutex_unlock(&shmem_swaplist_mutex
);
1357 put_swap_page(page
, swap
);
1359 set_page_dirty(page
);
1360 if (wbc
->for_reclaim
)
1361 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1366 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1367 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1371 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1372 return; /* show nothing */
1374 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1376 seq_printf(seq
, ",mpol=%s", buffer
);
1379 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1381 struct mempolicy
*mpol
= NULL
;
1383 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1384 mpol
= sbinfo
->mpol
;
1386 spin_unlock(&sbinfo
->stat_lock
);
1390 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1391 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1394 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1398 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1400 #define vm_policy vm_private_data
1403 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1404 struct shmem_inode_info
*info
, pgoff_t index
)
1406 /* Create a pseudo vma that just contains the policy */
1408 /* Bias interleave by inode number to distribute better across nodes */
1409 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1411 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1414 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1416 /* Drop reference taken by mpol_shared_policy_lookup() */
1417 mpol_cond_put(vma
->vm_policy
);
1420 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1421 struct shmem_inode_info
*info
, pgoff_t index
)
1423 struct vm_area_struct pvma
;
1426 shmem_pseudo_vma_init(&pvma
, info
, index
);
1427 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1428 shmem_pseudo_vma_destroy(&pvma
);
1433 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1434 struct shmem_inode_info
*info
, pgoff_t index
)
1436 struct vm_area_struct pvma
;
1437 struct inode
*inode
= &info
->vfs_inode
;
1438 struct address_space
*mapping
= inode
->i_mapping
;
1439 pgoff_t idx
, hindex
;
1440 void __rcu
**results
;
1443 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1446 hindex
= round_down(index
, HPAGE_PMD_NR
);
1448 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1449 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1455 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1456 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1457 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1458 shmem_pseudo_vma_destroy(&pvma
);
1460 prep_transhuge_page(page
);
1464 static struct page
*shmem_alloc_page(gfp_t gfp
,
1465 struct shmem_inode_info
*info
, pgoff_t index
)
1467 struct vm_area_struct pvma
;
1470 shmem_pseudo_vma_init(&pvma
, info
, index
);
1471 page
= alloc_page_vma(gfp
, &pvma
, 0);
1472 shmem_pseudo_vma_destroy(&pvma
);
1477 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1478 struct inode
*inode
,
1479 pgoff_t index
, bool huge
)
1481 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1486 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1488 nr
= huge
? HPAGE_PMD_NR
: 1;
1490 if (!shmem_inode_acct_block(inode
, nr
))
1494 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1496 page
= shmem_alloc_page(gfp
, info
, index
);
1498 __SetPageLocked(page
);
1499 __SetPageSwapBacked(page
);
1504 shmem_inode_unacct_blocks(inode
, nr
);
1506 return ERR_PTR(err
);
1510 * When a page is moved from swapcache to shmem filecache (either by the
1511 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1512 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1513 * ignorance of the mapping it belongs to. If that mapping has special
1514 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1515 * we may need to copy to a suitable page before moving to filecache.
1517 * In a future release, this may well be extended to respect cpuset and
1518 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1519 * but for now it is a simple matter of zone.
1521 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1523 return page_zonenum(page
) > gfp_zone(gfp
);
1526 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1527 struct shmem_inode_info
*info
, pgoff_t index
)
1529 struct page
*oldpage
, *newpage
;
1530 struct address_space
*swap_mapping
;
1535 swap_index
= page_private(oldpage
);
1536 swap_mapping
= page_mapping(oldpage
);
1539 * We have arrived here because our zones are constrained, so don't
1540 * limit chance of success by further cpuset and node constraints.
1542 gfp
&= ~GFP_CONSTRAINT_MASK
;
1543 newpage
= shmem_alloc_page(gfp
, info
, index
);
1548 copy_highpage(newpage
, oldpage
);
1549 flush_dcache_page(newpage
);
1551 __SetPageLocked(newpage
);
1552 __SetPageSwapBacked(newpage
);
1553 SetPageUptodate(newpage
);
1554 set_page_private(newpage
, swap_index
);
1555 SetPageSwapCache(newpage
);
1558 * Our caller will very soon move newpage out of swapcache, but it's
1559 * a nice clean interface for us to replace oldpage by newpage there.
1561 spin_lock_irq(&swap_mapping
->tree_lock
);
1562 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1565 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1566 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1568 spin_unlock_irq(&swap_mapping
->tree_lock
);
1570 if (unlikely(error
)) {
1572 * Is this possible? I think not, now that our callers check
1573 * both PageSwapCache and page_private after getting page lock;
1574 * but be defensive. Reverse old to newpage for clear and free.
1578 mem_cgroup_migrate(oldpage
, newpage
);
1579 lru_cache_add_anon(newpage
);
1583 ClearPageSwapCache(oldpage
);
1584 set_page_private(oldpage
, 0);
1586 unlock_page(oldpage
);
1593 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1595 * If we allocate a new one we do not mark it dirty. That's up to the
1596 * vm. If we swap it in we mark it dirty since we also free the swap
1597 * entry since a page cannot live in both the swap and page cache.
1599 * fault_mm and fault_type are only supplied by shmem_fault:
1600 * otherwise they are NULL.
1602 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1603 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1604 struct vm_area_struct
*vma
, struct vm_fault
*vmf
, int *fault_type
)
1606 struct address_space
*mapping
= inode
->i_mapping
;
1607 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1608 struct shmem_sb_info
*sbinfo
;
1609 struct mm_struct
*charge_mm
;
1610 struct mem_cgroup
*memcg
;
1613 enum sgp_type sgp_huge
= sgp
;
1614 pgoff_t hindex
= index
;
1619 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1621 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1625 page
= find_lock_entry(mapping
, index
);
1626 if (radix_tree_exceptional_entry(page
)) {
1627 swap
= radix_to_swp_entry(page
);
1631 if (sgp
<= SGP_CACHE
&&
1632 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1637 if (page
&& sgp
== SGP_WRITE
)
1638 mark_page_accessed(page
);
1640 /* fallocated page? */
1641 if (page
&& !PageUptodate(page
)) {
1642 if (sgp
!= SGP_READ
)
1648 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1654 * Fast cache lookup did not find it:
1655 * bring it back from swap or allocate.
1657 sbinfo
= SHMEM_SB(inode
->i_sb
);
1658 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1661 /* Look it up and read it in.. */
1662 page
= lookup_swap_cache(swap
, NULL
, 0);
1664 /* Or update major stats only when swapin succeeds?? */
1666 *fault_type
|= VM_FAULT_MAJOR
;
1667 count_vm_event(PGMAJFAULT
);
1668 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1670 /* Here we actually start the io */
1671 page
= shmem_swapin(swap
, gfp
, info
, index
);
1678 /* We have to do this with page locked to prevent races */
1680 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1681 !shmem_confirm_swap(mapping
, index
, swap
)) {
1682 error
= -EEXIST
; /* try again */
1685 if (!PageUptodate(page
)) {
1689 wait_on_page_writeback(page
);
1691 if (shmem_should_replace_page(page
, gfp
)) {
1692 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1697 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1700 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1701 swp_to_radix_entry(swap
));
1703 * We already confirmed swap under page lock, and make
1704 * no memory allocation here, so usually no possibility
1705 * of error; but free_swap_and_cache() only trylocks a
1706 * page, so it is just possible that the entry has been
1707 * truncated or holepunched since swap was confirmed.
1708 * shmem_undo_range() will have done some of the
1709 * unaccounting, now delete_from_swap_cache() will do
1711 * Reset swap.val? No, leave it so "failed" goes back to
1712 * "repeat": reading a hole and writing should succeed.
1715 mem_cgroup_cancel_charge(page
, memcg
, false);
1716 delete_from_swap_cache(page
);
1722 mem_cgroup_commit_charge(page
, memcg
, true, false);
1724 spin_lock_irq(&info
->lock
);
1726 shmem_recalc_inode(inode
);
1727 spin_unlock_irq(&info
->lock
);
1729 if (sgp
== SGP_WRITE
)
1730 mark_page_accessed(page
);
1732 delete_from_swap_cache(page
);
1733 set_page_dirty(page
);
1737 if (vma
&& userfaultfd_missing(vma
)) {
1738 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1742 /* shmem_symlink() */
1743 if (mapping
->a_ops
!= &shmem_aops
)
1745 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1747 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1749 switch (sbinfo
->huge
) {
1752 case SHMEM_HUGE_NEVER
:
1754 case SHMEM_HUGE_WITHIN_SIZE
:
1755 off
= round_up(index
, HPAGE_PMD_NR
);
1756 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1757 if (i_size
>= HPAGE_PMD_SIZE
&&
1758 i_size
>> PAGE_SHIFT
>= off
)
1761 case SHMEM_HUGE_ADVISE
:
1762 if (sgp_huge
== SGP_HUGE
)
1764 /* TODO: implement fadvise() hints */
1769 page
= shmem_alloc_and_acct_page(gfp
, inode
, index
, true);
1771 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, inode
,
1776 error
= PTR_ERR(page
);
1778 if (error
!= -ENOSPC
)
1781 * Try to reclaim some spece by splitting a huge page
1782 * beyond i_size on the filesystem.
1786 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1787 if (ret
== SHRINK_STOP
)
1795 if (PageTransHuge(page
))
1796 hindex
= round_down(index
, HPAGE_PMD_NR
);
1800 if (sgp
== SGP_WRITE
)
1801 __SetPageReferenced(page
);
1803 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1804 PageTransHuge(page
));
1807 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1808 compound_order(page
));
1810 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1812 radix_tree_preload_end();
1815 mem_cgroup_cancel_charge(page
, memcg
,
1816 PageTransHuge(page
));
1819 mem_cgroup_commit_charge(page
, memcg
, false,
1820 PageTransHuge(page
));
1821 lru_cache_add_anon(page
);
1823 spin_lock_irq(&info
->lock
);
1824 info
->alloced
+= 1 << compound_order(page
);
1825 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1826 shmem_recalc_inode(inode
);
1827 spin_unlock_irq(&info
->lock
);
1830 if (PageTransHuge(page
) &&
1831 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1832 hindex
+ HPAGE_PMD_NR
- 1) {
1834 * Part of the huge page is beyond i_size: subject
1835 * to shrink under memory pressure.
1837 spin_lock(&sbinfo
->shrinklist_lock
);
1839 * _careful to defend against unlocked access to
1840 * ->shrink_list in shmem_unused_huge_shrink()
1842 if (list_empty_careful(&info
->shrinklist
)) {
1843 list_add_tail(&info
->shrinklist
,
1844 &sbinfo
->shrinklist
);
1845 sbinfo
->shrinklist_len
++;
1847 spin_unlock(&sbinfo
->shrinklist_lock
);
1851 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1853 if (sgp
== SGP_FALLOC
)
1857 * Let SGP_WRITE caller clear ends if write does not fill page;
1858 * but SGP_FALLOC on a page fallocated earlier must initialize
1859 * it now, lest undo on failure cancel our earlier guarantee.
1861 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1862 struct page
*head
= compound_head(page
);
1865 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1866 clear_highpage(head
+ i
);
1867 flush_dcache_page(head
+ i
);
1869 SetPageUptodate(head
);
1873 /* Perhaps the file has been truncated since we checked */
1874 if (sgp
<= SGP_CACHE
&&
1875 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1877 ClearPageDirty(page
);
1878 delete_from_page_cache(page
);
1879 spin_lock_irq(&info
->lock
);
1880 shmem_recalc_inode(inode
);
1881 spin_unlock_irq(&info
->lock
);
1886 *pagep
= page
+ index
- hindex
;
1893 shmem_inode_unacct_blocks(inode
, 1 << compound_order(page
));
1895 if (PageTransHuge(page
)) {
1901 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1908 if (error
== -ENOSPC
&& !once
++) {
1909 spin_lock_irq(&info
->lock
);
1910 shmem_recalc_inode(inode
);
1911 spin_unlock_irq(&info
->lock
);
1914 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1920 * This is like autoremove_wake_function, but it removes the wait queue
1921 * entry unconditionally - even if something else had already woken the
1924 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
1926 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1927 list_del_init(&wait
->entry
);
1931 static int shmem_fault(struct vm_fault
*vmf
)
1933 struct vm_area_struct
*vma
= vmf
->vma
;
1934 struct inode
*inode
= file_inode(vma
->vm_file
);
1935 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1938 int ret
= VM_FAULT_LOCKED
;
1941 * Trinity finds that probing a hole which tmpfs is punching can
1942 * prevent the hole-punch from ever completing: which in turn
1943 * locks writers out with its hold on i_mutex. So refrain from
1944 * faulting pages into the hole while it's being punched. Although
1945 * shmem_undo_range() does remove the additions, it may be unable to
1946 * keep up, as each new page needs its own unmap_mapping_range() call,
1947 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1949 * It does not matter if we sometimes reach this check just before the
1950 * hole-punch begins, so that one fault then races with the punch:
1951 * we just need to make racing faults a rare case.
1953 * The implementation below would be much simpler if we just used a
1954 * standard mutex or completion: but we cannot take i_mutex in fault,
1955 * and bloating every shmem inode for this unlikely case would be sad.
1957 if (unlikely(inode
->i_private
)) {
1958 struct shmem_falloc
*shmem_falloc
;
1960 spin_lock(&inode
->i_lock
);
1961 shmem_falloc
= inode
->i_private
;
1963 shmem_falloc
->waitq
&&
1964 vmf
->pgoff
>= shmem_falloc
->start
&&
1965 vmf
->pgoff
< shmem_falloc
->next
) {
1966 wait_queue_head_t
*shmem_falloc_waitq
;
1967 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
1969 ret
= VM_FAULT_NOPAGE
;
1970 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1971 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1972 /* It's polite to up mmap_sem if we can */
1973 up_read(&vma
->vm_mm
->mmap_sem
);
1974 ret
= VM_FAULT_RETRY
;
1977 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1978 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1979 TASK_UNINTERRUPTIBLE
);
1980 spin_unlock(&inode
->i_lock
);
1984 * shmem_falloc_waitq points into the shmem_fallocate()
1985 * stack of the hole-punching task: shmem_falloc_waitq
1986 * is usually invalid by the time we reach here, but
1987 * finish_wait() does not dereference it in that case;
1988 * though i_lock needed lest racing with wake_up_all().
1990 spin_lock(&inode
->i_lock
);
1991 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1992 spin_unlock(&inode
->i_lock
);
1995 spin_unlock(&inode
->i_lock
);
2000 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2001 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2003 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2006 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2007 gfp
, vma
, vmf
, &ret
);
2009 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
2013 unsigned long shmem_get_unmapped_area(struct file
*file
,
2014 unsigned long uaddr
, unsigned long len
,
2015 unsigned long pgoff
, unsigned long flags
)
2017 unsigned long (*get_area
)(struct file
*,
2018 unsigned long, unsigned long, unsigned long, unsigned long);
2020 unsigned long offset
;
2021 unsigned long inflated_len
;
2022 unsigned long inflated_addr
;
2023 unsigned long inflated_offset
;
2025 if (len
> TASK_SIZE
)
2028 get_area
= current
->mm
->get_unmapped_area
;
2029 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2031 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
2033 if (IS_ERR_VALUE(addr
))
2035 if (addr
& ~PAGE_MASK
)
2037 if (addr
> TASK_SIZE
- len
)
2040 if (shmem_huge
== SHMEM_HUGE_DENY
)
2042 if (len
< HPAGE_PMD_SIZE
)
2044 if (flags
& MAP_FIXED
)
2047 * Our priority is to support MAP_SHARED mapped hugely;
2048 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2049 * But if caller specified an address hint, respect that as before.
2054 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2055 struct super_block
*sb
;
2058 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2059 sb
= file_inode(file
)->i_sb
;
2062 * Called directly from mm/mmap.c, or drivers/char/mem.c
2063 * for "/dev/zero", to create a shared anonymous object.
2065 if (IS_ERR(shm_mnt
))
2067 sb
= shm_mnt
->mnt_sb
;
2069 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2073 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2074 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2076 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2079 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2080 if (inflated_len
> TASK_SIZE
)
2082 if (inflated_len
< len
)
2085 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
2086 if (IS_ERR_VALUE(inflated_addr
))
2088 if (inflated_addr
& ~PAGE_MASK
)
2091 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2092 inflated_addr
+= offset
- inflated_offset
;
2093 if (inflated_offset
> offset
)
2094 inflated_addr
+= HPAGE_PMD_SIZE
;
2096 if (inflated_addr
> TASK_SIZE
- len
)
2098 return inflated_addr
;
2102 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2104 struct inode
*inode
= file_inode(vma
->vm_file
);
2105 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2108 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2111 struct inode
*inode
= file_inode(vma
->vm_file
);
2114 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2115 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2119 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2121 struct inode
*inode
= file_inode(file
);
2122 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2123 int retval
= -ENOMEM
;
2125 spin_lock_irq(&info
->lock
);
2126 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2127 if (!user_shm_lock(inode
->i_size
, user
))
2129 info
->flags
|= VM_LOCKED
;
2130 mapping_set_unevictable(file
->f_mapping
);
2132 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2133 user_shm_unlock(inode
->i_size
, user
);
2134 info
->flags
&= ~VM_LOCKED
;
2135 mapping_clear_unevictable(file
->f_mapping
);
2140 spin_unlock_irq(&info
->lock
);
2144 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2146 file_accessed(file
);
2147 vma
->vm_ops
= &shmem_vm_ops
;
2148 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2149 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2150 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2151 khugepaged_enter(vma
, vma
->vm_flags
);
2156 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2157 umode_t mode
, dev_t dev
, unsigned long flags
)
2159 struct inode
*inode
;
2160 struct shmem_inode_info
*info
;
2161 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2164 if (shmem_reserve_inode(sb
))
2167 inode
= new_inode(sb
);
2169 inode_init_owner(inode
, dir
, mode
);
2170 inode
->i_blocks
= 0;
2171 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2172 inode
->i_generation
= get_seconds();
2173 info
= SHMEM_I(inode
);
2174 memset(info
, 0, (char *)inode
- (char *)info
);
2175 spin_lock_init(&info
->lock
);
2176 info
->seals
= F_SEAL_SEAL
;
2177 info
->flags
= flags
& VM_NORESERVE
;
2178 INIT_LIST_HEAD(&info
->shrinklist
);
2179 INIT_LIST_HEAD(&info
->swaplist
);
2180 simple_xattrs_init(&info
->xattrs
);
2181 cache_no_acl(inode
);
2183 switch (mode
& S_IFMT
) {
2185 inode
->i_op
= &shmem_special_inode_operations
;
2186 init_special_inode(inode
, mode
, dev
);
2189 inode
->i_mapping
->a_ops
= &shmem_aops
;
2190 inode
->i_op
= &shmem_inode_operations
;
2191 inode
->i_fop
= &shmem_file_operations
;
2192 mpol_shared_policy_init(&info
->policy
,
2193 shmem_get_sbmpol(sbinfo
));
2197 /* Some things misbehave if size == 0 on a directory */
2198 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2199 inode
->i_op
= &shmem_dir_inode_operations
;
2200 inode
->i_fop
= &simple_dir_operations
;
2204 * Must not load anything in the rbtree,
2205 * mpol_free_shared_policy will not be called.
2207 mpol_shared_policy_init(&info
->policy
, NULL
);
2211 if (!sbinfo
->idr_nouse
) {
2212 /* inum 0 and 1 are unused */
2213 mutex_lock(&sbinfo
->idr_lock
);
2214 ino
= idr_alloc(&sbinfo
->idr
, inode
, 2, INT_MAX
,
2218 mutex_unlock(&sbinfo
->idr_lock
);
2219 __insert_inode_hash(inode
, inode
->i_ino
);
2222 mutex_unlock(&sbinfo
->idr_lock
);
2224 /* shmem_free_inode() will be called */
2228 inode
->i_ino
= get_next_ino();
2230 shmem_free_inode(sb
);
2234 bool shmem_mapping(struct address_space
*mapping
)
2236 return mapping
->a_ops
== &shmem_aops
;
2239 static int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2241 struct vm_area_struct
*dst_vma
,
2242 unsigned long dst_addr
,
2243 unsigned long src_addr
,
2245 struct page
**pagep
)
2247 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2248 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2249 struct address_space
*mapping
= inode
->i_mapping
;
2250 gfp_t gfp
= mapping_gfp_mask(mapping
);
2251 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2252 struct mem_cgroup
*memcg
;
2256 pte_t _dst_pte
, *dst_pte
;
2260 if (!shmem_inode_acct_block(inode
, 1))
2264 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2266 goto out_unacct_blocks
;
2268 if (!zeropage
) { /* mcopy_atomic */
2269 page_kaddr
= kmap_atomic(page
);
2270 ret
= copy_from_user(page_kaddr
,
2271 (const void __user
*)src_addr
,
2273 kunmap_atomic(page_kaddr
);
2275 /* fallback to copy_from_user outside mmap_sem */
2276 if (unlikely(ret
)) {
2278 shmem_inode_unacct_blocks(inode
, 1);
2279 /* don't free the page */
2282 } else { /* mfill_zeropage_atomic */
2283 clear_highpage(page
);
2290 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2291 __SetPageLocked(page
);
2292 __SetPageSwapBacked(page
);
2293 __SetPageUptodate(page
);
2295 ret
= mem_cgroup_try_charge(page
, dst_mm
, gfp
, &memcg
, false);
2299 ret
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
2301 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
);
2302 radix_tree_preload_end();
2305 goto out_release_uncharge
;
2307 mem_cgroup_commit_charge(page
, memcg
, false, false);
2309 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2310 if (dst_vma
->vm_flags
& VM_WRITE
)
2311 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2314 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2315 if (!pte_none(*dst_pte
))
2316 goto out_release_uncharge_unlock
;
2318 lru_cache_add_anon(page
);
2320 spin_lock(&info
->lock
);
2322 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2323 shmem_recalc_inode(inode
);
2324 spin_unlock(&info
->lock
);
2326 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2327 page_add_file_rmap(page
, false);
2328 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2330 /* No need to invalidate - it was non-present before */
2331 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2333 pte_unmap_unlock(dst_pte
, ptl
);
2337 out_release_uncharge_unlock
:
2338 pte_unmap_unlock(dst_pte
, ptl
);
2339 out_release_uncharge
:
2340 mem_cgroup_cancel_charge(page
, memcg
, false);
2345 shmem_inode_unacct_blocks(inode
, 1);
2349 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2351 struct vm_area_struct
*dst_vma
,
2352 unsigned long dst_addr
,
2353 unsigned long src_addr
,
2354 struct page
**pagep
)
2356 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2357 dst_addr
, src_addr
, false, pagep
);
2360 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2362 struct vm_area_struct
*dst_vma
,
2363 unsigned long dst_addr
)
2365 struct page
*page
= NULL
;
2367 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2368 dst_addr
, 0, true, &page
);
2372 static const struct inode_operations shmem_symlink_inode_operations
;
2373 static const struct inode_operations shmem_short_symlink_operations
;
2375 #ifdef CONFIG_TMPFS_XATTR
2376 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2378 #define shmem_initxattrs NULL
2382 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2383 loff_t pos
, unsigned len
, unsigned flags
,
2384 struct page
**pagep
, void **fsdata
)
2386 struct inode
*inode
= mapping
->host
;
2387 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2388 pgoff_t index
= pos
>> PAGE_SHIFT
;
2390 /* i_mutex is held by caller */
2391 if (unlikely(info
->seals
& (F_SEAL_WRITE
| F_SEAL_GROW
))) {
2392 if (info
->seals
& F_SEAL_WRITE
)
2394 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2398 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2402 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2403 loff_t pos
, unsigned len
, unsigned copied
,
2404 struct page
*page
, void *fsdata
)
2406 struct inode
*inode
= mapping
->host
;
2408 if (pos
+ copied
> inode
->i_size
)
2409 i_size_write(inode
, pos
+ copied
);
2411 if (!PageUptodate(page
)) {
2412 struct page
*head
= compound_head(page
);
2413 if (PageTransCompound(page
)) {
2416 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2417 if (head
+ i
== page
)
2419 clear_highpage(head
+ i
);
2420 flush_dcache_page(head
+ i
);
2423 if (copied
< PAGE_SIZE
) {
2424 unsigned from
= pos
& (PAGE_SIZE
- 1);
2425 zero_user_segments(page
, 0, from
,
2426 from
+ copied
, PAGE_SIZE
);
2428 SetPageUptodate(head
);
2430 set_page_dirty(page
);
2437 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2439 struct file
*file
= iocb
->ki_filp
;
2440 struct inode
*inode
= file_inode(file
);
2441 struct address_space
*mapping
= inode
->i_mapping
;
2443 unsigned long offset
;
2444 enum sgp_type sgp
= SGP_READ
;
2447 loff_t
*ppos
= &iocb
->ki_pos
;
2450 * Might this read be for a stacking filesystem? Then when reading
2451 * holes of a sparse file, we actually need to allocate those pages,
2452 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2454 if (!iter_is_iovec(to
))
2457 index
= *ppos
>> PAGE_SHIFT
;
2458 offset
= *ppos
& ~PAGE_MASK
;
2461 struct page
*page
= NULL
;
2463 unsigned long nr
, ret
;
2464 loff_t i_size
= i_size_read(inode
);
2466 end_index
= i_size
>> PAGE_SHIFT
;
2467 if (index
> end_index
)
2469 if (index
== end_index
) {
2470 nr
= i_size
& ~PAGE_MASK
;
2475 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2477 if (error
== -EINVAL
)
2482 if (sgp
== SGP_CACHE
)
2483 set_page_dirty(page
);
2488 * We must evaluate after, since reads (unlike writes)
2489 * are called without i_mutex protection against truncate
2492 i_size
= i_size_read(inode
);
2493 end_index
= i_size
>> PAGE_SHIFT
;
2494 if (index
== end_index
) {
2495 nr
= i_size
& ~PAGE_MASK
;
2506 * If users can be writing to this page using arbitrary
2507 * virtual addresses, take care about potential aliasing
2508 * before reading the page on the kernel side.
2510 if (mapping_writably_mapped(mapping
))
2511 flush_dcache_page(page
);
2513 * Mark the page accessed if we read the beginning.
2516 mark_page_accessed(page
);
2518 page
= ZERO_PAGE(0);
2523 * Ok, we have the page, and it's up-to-date, so
2524 * now we can copy it to user space...
2526 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2529 index
+= offset
>> PAGE_SHIFT
;
2530 offset
&= ~PAGE_MASK
;
2533 if (!iov_iter_count(to
))
2542 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2543 file_accessed(file
);
2544 return retval
? retval
: error
;
2548 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2550 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2551 pgoff_t index
, pgoff_t end
, int whence
)
2554 struct pagevec pvec
;
2555 pgoff_t indices
[PAGEVEC_SIZE
];
2559 pagevec_init(&pvec
);
2560 pvec
.nr
= 1; /* start small: we may be there already */
2562 pvec
.nr
= find_get_entries(mapping
, index
,
2563 pvec
.nr
, pvec
.pages
, indices
);
2565 if (whence
== SEEK_DATA
)
2569 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2570 if (index
< indices
[i
]) {
2571 if (whence
== SEEK_HOLE
) {
2577 page
= pvec
.pages
[i
];
2578 if (page
&& !radix_tree_exceptional_entry(page
)) {
2579 if (!PageUptodate(page
))
2583 (page
&& whence
== SEEK_DATA
) ||
2584 (!page
&& whence
== SEEK_HOLE
)) {
2589 pagevec_remove_exceptionals(&pvec
);
2590 pagevec_release(&pvec
);
2591 pvec
.nr
= PAGEVEC_SIZE
;
2597 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2599 struct address_space
*mapping
= file
->f_mapping
;
2600 struct inode
*inode
= mapping
->host
;
2604 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2605 return generic_file_llseek_size(file
, offset
, whence
,
2606 MAX_LFS_FILESIZE
, i_size_read(inode
));
2608 /* We're holding i_mutex so we can access i_size directly */
2612 else if (offset
>= inode
->i_size
)
2615 start
= offset
>> PAGE_SHIFT
;
2616 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2617 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2618 new_offset
<<= PAGE_SHIFT
;
2619 if (new_offset
> offset
) {
2620 if (new_offset
< inode
->i_size
)
2621 offset
= new_offset
;
2622 else if (whence
== SEEK_DATA
)
2625 offset
= inode
->i_size
;
2630 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2631 inode_unlock(inode
);
2636 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2637 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2639 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2640 #define LAST_SCAN 4 /* about 150ms max */
2642 static void shmem_tag_pins(struct address_space
*mapping
)
2644 struct radix_tree_iter iter
;
2653 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2654 page
= radix_tree_deref_slot(slot
);
2655 if (!page
|| radix_tree_exception(page
)) {
2656 if (radix_tree_deref_retry(page
)) {
2657 slot
= radix_tree_iter_retry(&iter
);
2660 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2661 spin_lock_irq(&mapping
->tree_lock
);
2662 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2664 spin_unlock_irq(&mapping
->tree_lock
);
2667 if (need_resched()) {
2668 slot
= radix_tree_iter_resume(slot
, &iter
);
2676 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2677 * via get_user_pages(), drivers might have some pending I/O without any active
2678 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2679 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2680 * them to be dropped.
2681 * The caller must guarantee that no new user will acquire writable references
2682 * to those pages to avoid races.
2684 static int shmem_wait_for_pins(struct address_space
*mapping
)
2686 struct radix_tree_iter iter
;
2692 shmem_tag_pins(mapping
);
2695 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2696 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2700 lru_add_drain_all();
2701 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2706 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2707 start
, SHMEM_TAG_PINNED
) {
2709 page
= radix_tree_deref_slot(slot
);
2710 if (radix_tree_exception(page
)) {
2711 if (radix_tree_deref_retry(page
)) {
2712 slot
= radix_tree_iter_retry(&iter
);
2720 page_count(page
) - page_mapcount(page
) != 1) {
2721 if (scan
< LAST_SCAN
)
2722 goto continue_resched
;
2725 * On the last scan, we clean up all those tags
2726 * we inserted; but make a note that we still
2727 * found pages pinned.
2732 spin_lock_irq(&mapping
->tree_lock
);
2733 radix_tree_tag_clear(&mapping
->page_tree
,
2734 iter
.index
, SHMEM_TAG_PINNED
);
2735 spin_unlock_irq(&mapping
->tree_lock
);
2737 if (need_resched()) {
2738 slot
= radix_tree_iter_resume(slot
, &iter
);
2748 #define F_ALL_SEALS (F_SEAL_SEAL | \
2753 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2755 struct inode
*inode
= file_inode(file
);
2756 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2761 * Sealing allows multiple parties to share a shmem-file but restrict
2762 * access to a specific subset of file operations. Seals can only be
2763 * added, but never removed. This way, mutually untrusted parties can
2764 * share common memory regions with a well-defined policy. A malicious
2765 * peer can thus never perform unwanted operations on a shared object.
2767 * Seals are only supported on special shmem-files and always affect
2768 * the whole underlying inode. Once a seal is set, it may prevent some
2769 * kinds of access to the file. Currently, the following seals are
2771 * SEAL_SEAL: Prevent further seals from being set on this file
2772 * SEAL_SHRINK: Prevent the file from shrinking
2773 * SEAL_GROW: Prevent the file from growing
2774 * SEAL_WRITE: Prevent write access to the file
2776 * As we don't require any trust relationship between two parties, we
2777 * must prevent seals from being removed. Therefore, sealing a file
2778 * only adds a given set of seals to the file, it never touches
2779 * existing seals. Furthermore, the "setting seals"-operation can be
2780 * sealed itself, which basically prevents any further seal from being
2783 * Semantics of sealing are only defined on volatile files. Only
2784 * anonymous shmem files support sealing. More importantly, seals are
2785 * never written to disk. Therefore, there's no plan to support it on
2789 if (file
->f_op
!= &shmem_file_operations
)
2791 if (!(file
->f_mode
& FMODE_WRITE
))
2793 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2798 if (info
->seals
& F_SEAL_SEAL
) {
2803 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2804 error
= mapping_deny_writable(file
->f_mapping
);
2808 error
= shmem_wait_for_pins(file
->f_mapping
);
2810 mapping_allow_writable(file
->f_mapping
);
2815 info
->seals
|= seals
;
2819 inode_unlock(inode
);
2822 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2824 int shmem_get_seals(struct file
*file
)
2826 if (file
->f_op
!= &shmem_file_operations
)
2829 return SHMEM_I(file_inode(file
))->seals
;
2831 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2833 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2839 /* disallow upper 32bit */
2843 error
= shmem_add_seals(file
, arg
);
2846 error
= shmem_get_seals(file
);
2856 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2859 struct inode
*inode
= file_inode(file
);
2860 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2861 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2862 struct shmem_falloc shmem_falloc
;
2863 pgoff_t start
, index
, end
;
2866 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2871 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2872 struct address_space
*mapping
= file
->f_mapping
;
2873 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2874 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2875 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2877 /* protected by i_mutex */
2878 if (info
->seals
& F_SEAL_WRITE
) {
2883 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2884 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2885 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2886 spin_lock(&inode
->i_lock
);
2887 inode
->i_private
= &shmem_falloc
;
2888 spin_unlock(&inode
->i_lock
);
2890 if ((u64
)unmap_end
> (u64
)unmap_start
)
2891 unmap_mapping_range(mapping
, unmap_start
,
2892 1 + unmap_end
- unmap_start
, 0);
2893 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2894 /* No need to unmap again: hole-punching leaves COWed pages */
2896 spin_lock(&inode
->i_lock
);
2897 inode
->i_private
= NULL
;
2898 wake_up_all(&shmem_falloc_waitq
);
2899 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2900 spin_unlock(&inode
->i_lock
);
2905 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2906 error
= inode_newsize_ok(inode
, offset
+ len
);
2910 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2915 start
= offset
>> PAGE_SHIFT
;
2916 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2917 /* Try to avoid a swapstorm if len is impossible to satisfy */
2918 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2923 shmem_falloc
.waitq
= NULL
;
2924 shmem_falloc
.start
= start
;
2925 shmem_falloc
.next
= start
;
2926 shmem_falloc
.nr_falloced
= 0;
2927 shmem_falloc
.nr_unswapped
= 0;
2928 spin_lock(&inode
->i_lock
);
2929 inode
->i_private
= &shmem_falloc
;
2930 spin_unlock(&inode
->i_lock
);
2932 for (index
= start
; index
< end
; index
++) {
2936 * Good, the fallocate(2) manpage permits EINTR: we may have
2937 * been interrupted because we are using up too much memory.
2939 if (signal_pending(current
))
2941 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2944 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2946 /* Remove the !PageUptodate pages we added */
2947 if (index
> start
) {
2948 shmem_undo_range(inode
,
2949 (loff_t
)start
<< PAGE_SHIFT
,
2950 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2956 * Inform shmem_writepage() how far we have reached.
2957 * No need for lock or barrier: we have the page lock.
2959 shmem_falloc
.next
++;
2960 if (!PageUptodate(page
))
2961 shmem_falloc
.nr_falloced
++;
2964 * If !PageUptodate, leave it that way so that freeable pages
2965 * can be recognized if we need to rollback on error later.
2966 * But set_page_dirty so that memory pressure will swap rather
2967 * than free the pages we are allocating (and SGP_CACHE pages
2968 * might still be clean: we now need to mark those dirty too).
2970 set_page_dirty(page
);
2976 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2977 i_size_write(inode
, offset
+ len
);
2978 inode
->i_ctime
= current_time(inode
);
2980 spin_lock(&inode
->i_lock
);
2981 inode
->i_private
= NULL
;
2982 spin_unlock(&inode
->i_lock
);
2984 inode_unlock(inode
);
2988 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2990 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2992 buf
->f_type
= TMPFS_MAGIC
;
2993 buf
->f_bsize
= PAGE_SIZE
;
2994 buf
->f_namelen
= NAME_MAX
;
2995 if (sbinfo
->max_blocks
) {
2996 buf
->f_blocks
= sbinfo
->max_blocks
;
2998 buf
->f_bfree
= sbinfo
->max_blocks
-
2999 percpu_counter_sum(&sbinfo
->used_blocks
);
3001 if (sbinfo
->max_inodes
) {
3002 buf
->f_files
= sbinfo
->max_inodes
;
3003 buf
->f_ffree
= sbinfo
->free_inodes
;
3005 /* else leave those fields 0 like simple_statfs */
3010 * File creation. Allocate an inode, and we're done..
3013 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3015 struct inode
*inode
;
3016 int error
= -ENOSPC
;
3018 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
3020 error
= simple_acl_create(dir
, inode
);
3023 error
= security_inode_init_security(inode
, dir
,
3025 shmem_initxattrs
, NULL
);
3026 if (error
&& error
!= -EOPNOTSUPP
)
3030 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3031 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3032 d_instantiate(dentry
, inode
);
3033 dget(dentry
); /* Extra count - pin the dentry in core */
3042 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3044 struct inode
*inode
;
3045 int error
= -ENOSPC
;
3047 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
3049 error
= security_inode_init_security(inode
, dir
,
3051 shmem_initxattrs
, NULL
);
3052 if (error
&& error
!= -EOPNOTSUPP
)
3054 error
= simple_acl_create(dir
, inode
);
3057 d_tmpfile(dentry
, inode
);
3065 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3069 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
3075 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
3078 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
3084 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
3086 struct inode
*inode
= d_inode(old_dentry
);
3090 * No ordinary (disk based) filesystem counts links as inodes;
3091 * but each new link needs a new dentry, pinning lowmem, and
3092 * tmpfs dentries cannot be pruned until they are unlinked.
3094 ret
= shmem_reserve_inode(inode
->i_sb
);
3098 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3099 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3101 ihold(inode
); /* New dentry reference */
3102 dget(dentry
); /* Extra pinning count for the created dentry */
3103 d_instantiate(dentry
, inode
);
3108 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3110 struct inode
*inode
= d_inode(dentry
);
3112 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3113 shmem_free_inode(inode
->i_sb
);
3115 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3116 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3118 dput(dentry
); /* Undo the count from "create" - this does all the work */
3122 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3124 if (!simple_empty(dentry
))
3127 drop_nlink(d_inode(dentry
));
3129 return shmem_unlink(dir
, dentry
);
3132 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3134 bool old_is_dir
= d_is_dir(old_dentry
);
3135 bool new_is_dir
= d_is_dir(new_dentry
);
3137 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3139 drop_nlink(old_dir
);
3142 drop_nlink(new_dir
);
3146 old_dir
->i_ctime
= old_dir
->i_mtime
=
3147 new_dir
->i_ctime
= new_dir
->i_mtime
=
3148 d_inode(old_dentry
)->i_ctime
=
3149 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3154 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3156 struct dentry
*whiteout
;
3159 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3163 error
= shmem_mknod(old_dir
, whiteout
,
3164 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3170 * Cheat and hash the whiteout while the old dentry is still in
3171 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3173 * d_lookup() will consistently find one of them at this point,
3174 * not sure which one, but that isn't even important.
3181 * The VFS layer already does all the dentry stuff for rename,
3182 * we just have to decrement the usage count for the target if
3183 * it exists so that the VFS layer correctly free's it when it
3186 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3188 struct inode
*inode
= d_inode(old_dentry
);
3189 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3191 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3194 if (flags
& RENAME_EXCHANGE
)
3195 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3197 if (!simple_empty(new_dentry
))
3200 if (flags
& RENAME_WHITEOUT
) {
3203 error
= shmem_whiteout(old_dir
, old_dentry
);
3208 if (d_really_is_positive(new_dentry
)) {
3209 (void) shmem_unlink(new_dir
, new_dentry
);
3210 if (they_are_dirs
) {
3211 drop_nlink(d_inode(new_dentry
));
3212 drop_nlink(old_dir
);
3214 } else if (they_are_dirs
) {
3215 drop_nlink(old_dir
);
3219 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3220 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3221 old_dir
->i_ctime
= old_dir
->i_mtime
=
3222 new_dir
->i_ctime
= new_dir
->i_mtime
=
3223 inode
->i_ctime
= current_time(old_dir
);
3227 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3231 struct inode
*inode
;
3234 len
= strlen(symname
) + 1;
3235 if (len
> PAGE_SIZE
)
3236 return -ENAMETOOLONG
;
3238 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
3242 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3243 shmem_initxattrs
, NULL
);
3245 if (error
!= -EOPNOTSUPP
) {
3252 inode
->i_size
= len
-1;
3253 if (len
<= SHORT_SYMLINK_LEN
) {
3254 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3255 if (!inode
->i_link
) {
3259 inode
->i_op
= &shmem_short_symlink_operations
;
3261 inode_nohighmem(inode
);
3262 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3267 inode
->i_mapping
->a_ops
= &shmem_aops
;
3268 inode
->i_op
= &shmem_symlink_inode_operations
;
3269 memcpy(page_address(page
), symname
, len
);
3270 SetPageUptodate(page
);
3271 set_page_dirty(page
);
3275 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3276 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3277 d_instantiate(dentry
, inode
);
3282 static void shmem_put_link(void *arg
)
3284 mark_page_accessed(arg
);
3288 static const char *shmem_get_link(struct dentry
*dentry
,
3289 struct inode
*inode
,
3290 struct delayed_call
*done
)
3292 struct page
*page
= NULL
;
3295 page
= find_get_page(inode
->i_mapping
, 0);
3297 return ERR_PTR(-ECHILD
);
3298 if (!PageUptodate(page
)) {
3300 return ERR_PTR(-ECHILD
);
3303 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3305 return ERR_PTR(error
);
3308 set_delayed_call(done
, shmem_put_link
, page
);
3309 return page_address(page
);
3312 #ifdef CONFIG_TMPFS_XATTR
3314 * Superblocks without xattr inode operations may get some security.* xattr
3315 * support from the LSM "for free". As soon as we have any other xattrs
3316 * like ACLs, we also need to implement the security.* handlers at
3317 * filesystem level, though.
3321 * Callback for security_inode_init_security() for acquiring xattrs.
3323 static int shmem_initxattrs(struct inode
*inode
,
3324 const struct xattr
*xattr_array
,
3327 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3328 const struct xattr
*xattr
;
3329 struct simple_xattr
*new_xattr
;
3332 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3333 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3337 len
= strlen(xattr
->name
) + 1;
3338 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3340 if (!new_xattr
->name
) {
3345 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3346 XATTR_SECURITY_PREFIX_LEN
);
3347 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3350 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3356 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3357 struct dentry
*unused
, struct inode
*inode
,
3358 const char *name
, void *buffer
, size_t size
)
3360 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3362 name
= xattr_full_name(handler
, name
);
3363 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3366 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3367 struct dentry
*unused
, struct inode
*inode
,
3368 const char *name
, const void *value
,
3369 size_t size
, int flags
)
3371 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3373 name
= xattr_full_name(handler
, name
);
3374 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3377 static const struct xattr_handler shmem_security_xattr_handler
= {
3378 .prefix
= XATTR_SECURITY_PREFIX
,
3379 .get
= shmem_xattr_handler_get
,
3380 .set
= shmem_xattr_handler_set
,
3383 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3384 .prefix
= XATTR_TRUSTED_PREFIX
,
3385 .get
= shmem_xattr_handler_get
,
3386 .set
= shmem_xattr_handler_set
,
3389 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3390 #ifdef CONFIG_TMPFS_POSIX_ACL
3391 &posix_acl_access_xattr_handler
,
3392 &posix_acl_default_xattr_handler
,
3394 &shmem_security_xattr_handler
,
3395 &shmem_trusted_xattr_handler
,
3399 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3401 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3402 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3404 #endif /* CONFIG_TMPFS_XATTR */
3406 static const struct inode_operations shmem_short_symlink_operations
= {
3407 .get_link
= simple_get_link
,
3408 #ifdef CONFIG_TMPFS_XATTR
3409 .listxattr
= shmem_listxattr
,
3413 static const struct inode_operations shmem_symlink_inode_operations
= {
3414 .get_link
= shmem_get_link
,
3415 #ifdef CONFIG_TMPFS_XATTR
3416 .listxattr
= shmem_listxattr
,
3420 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3422 return ERR_PTR(-ESTALE
);
3425 static int shmem_match(struct inode
*ino
, void *vfh
)
3429 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3432 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3433 struct fid
*fid
, int fh_len
, int fh_type
)
3435 struct inode
*inode
;
3436 struct dentry
*dentry
= NULL
;
3443 inode
= ilookup5(sb
, inum
, shmem_match
, fid
->raw
);
3445 dentry
= d_find_alias(inode
);
3452 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3453 struct inode
*parent
)
3457 return FILEID_INVALID
;
3460 fh
[0] = inode
->i_generation
;
3461 fh
[1] = inode
->i_ino
;
3467 static const struct export_operations shmem_export_ops
= {
3468 .get_parent
= shmem_get_parent
,
3469 .encode_fh
= shmem_encode_fh
,
3470 .fh_to_dentry
= shmem_fh_to_dentry
,
3473 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3476 char *this_char
, *value
, *rest
;
3477 struct mempolicy
*mpol
= NULL
;
3481 while (options
!= NULL
) {
3482 this_char
= options
;
3485 * NUL-terminate this option: unfortunately,
3486 * mount options form a comma-separated list,
3487 * but mpol's nodelist may also contain commas.
3489 options
= strchr(options
, ',');
3490 if (options
== NULL
)
3493 if (!isdigit(*options
)) {
3500 if ((value
= strchr(this_char
,'=')) != NULL
) {
3503 pr_err("tmpfs: No value for mount option '%s'\n",
3508 if (!strcmp(this_char
,"size")) {
3509 unsigned long long size
;
3510 size
= memparse(value
,&rest
);
3512 size
<<= PAGE_SHIFT
;
3513 size
*= totalram_pages
;
3519 sbinfo
->max_blocks
=
3520 DIV_ROUND_UP(size
, PAGE_SIZE
);
3521 } else if (!strcmp(this_char
,"nr_blocks")) {
3522 sbinfo
->max_blocks
= memparse(value
, &rest
);
3525 } else if (!strcmp(this_char
,"nr_inodes")) {
3526 sbinfo
->max_inodes
= memparse(value
, &rest
);
3527 if (*rest
|| sbinfo
->max_inodes
< 2)
3529 } else if (!strcmp(this_char
,"mode")) {
3532 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3535 } else if (!strcmp(this_char
,"uid")) {
3538 uid
= simple_strtoul(value
, &rest
, 0);
3541 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3542 if (!uid_valid(sbinfo
->uid
))
3544 } else if (!strcmp(this_char
,"gid")) {
3547 gid
= simple_strtoul(value
, &rest
, 0);
3550 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3551 if (!gid_valid(sbinfo
->gid
))
3553 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3554 } else if (!strcmp(this_char
, "huge")) {
3556 huge
= shmem_parse_huge(value
);
3559 if (!has_transparent_hugepage() &&
3560 huge
!= SHMEM_HUGE_NEVER
)
3562 sbinfo
->huge
= huge
;
3565 } else if (!strcmp(this_char
,"mpol")) {
3568 if (mpol_parse_str(value
, &mpol
))
3572 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3576 sbinfo
->mpol
= mpol
;
3580 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3588 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3590 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3591 struct shmem_sb_info config
= *sbinfo
;
3593 int error
= -EINVAL
;
3596 if (shmem_parse_options(data
, &config
, true))
3599 spin_lock(&sbinfo
->stat_lock
);
3600 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3601 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3603 if (config
.max_inodes
< inodes
)
3606 * Those tests disallow limited->unlimited while any are in use;
3607 * but we must separately disallow unlimited->limited, because
3608 * in that case we have no record of how much is already in use.
3610 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3612 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3616 sbinfo
->huge
= config
.huge
;
3617 sbinfo
->max_blocks
= config
.max_blocks
;
3618 sbinfo
->max_inodes
= config
.max_inodes
;
3619 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3622 * Preserve previous mempolicy unless mpol remount option was specified.
3625 mpol_put(sbinfo
->mpol
);
3626 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3629 spin_unlock(&sbinfo
->stat_lock
);
3633 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3635 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3637 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3638 seq_printf(seq
, ",size=%luk",
3639 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3640 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3641 seq_printf(seq
, ",nr_inodes=%d", sbinfo
->max_inodes
);
3642 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3643 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3644 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3645 seq_printf(seq
, ",uid=%u",
3646 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3647 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3648 seq_printf(seq
, ",gid=%u",
3649 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3650 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3651 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3653 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3655 shmem_show_mpol(seq
, sbinfo
->mpol
);
3659 #define MFD_NAME_PREFIX "memfd:"
3660 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3661 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3663 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
3665 SYSCALL_DEFINE2(memfd_create
,
3666 const char __user
*, uname
,
3667 unsigned int, flags
)
3669 struct shmem_inode_info
*info
;
3675 if (!(flags
& MFD_HUGETLB
)) {
3676 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3679 /* Sealing not supported in hugetlbfs (MFD_HUGETLB) */
3680 if (flags
& MFD_ALLOW_SEALING
)
3682 /* Allow huge page size encoding in flags. */
3683 if (flags
& ~(unsigned int)(MFD_ALL_FLAGS
|
3684 (MFD_HUGE_MASK
<< MFD_HUGE_SHIFT
)))
3688 /* length includes terminating zero */
3689 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3692 if (len
> MFD_NAME_MAX_LEN
+ 1)
3695 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_KERNEL
);
3699 strcpy(name
, MFD_NAME_PREFIX
);
3700 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3705 /* terminating-zero may have changed after strnlen_user() returned */
3706 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3711 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3717 if (flags
& MFD_HUGETLB
) {
3718 struct user_struct
*user
= NULL
;
3720 file
= hugetlb_file_setup(name
, 0, VM_NORESERVE
, &user
,
3721 HUGETLB_ANONHUGE_INODE
,
3722 (flags
>> MFD_HUGE_SHIFT
) &
3725 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3727 error
= PTR_ERR(file
);
3730 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3731 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3733 if (flags
& MFD_ALLOW_SEALING
) {
3735 * flags check at beginning of function ensures
3736 * this is not a hugetlbfs (MFD_HUGETLB) file.
3738 info
= SHMEM_I(file_inode(file
));
3739 info
->seals
&= ~F_SEAL_SEAL
;
3742 fd_install(fd
, file
);
3753 #endif /* CONFIG_TMPFS */
3755 static void shmem_put_super(struct super_block
*sb
)
3757 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3759 if (!sbinfo
->idr_nouse
)
3760 idr_destroy(&sbinfo
->idr
);
3761 percpu_counter_destroy(&sbinfo
->used_blocks
);
3762 mpol_put(sbinfo
->mpol
);
3764 sb
->s_fs_info
= NULL
;
3767 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3769 struct inode
*inode
;
3770 struct shmem_sb_info
*sbinfo
;
3773 /* Round up to L1_CACHE_BYTES to resist false sharing */
3774 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3775 L1_CACHE_BYTES
), GFP_KERNEL
);
3779 mutex_init(&sbinfo
->idr_lock
);
3780 idr_init(&sbinfo
->idr
);
3781 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3782 sbinfo
->uid
= current_fsuid();
3783 sbinfo
->gid
= current_fsgid();
3784 sb
->s_fs_info
= sbinfo
;
3788 * Per default we only allow half of the physical ram per
3789 * tmpfs instance, limiting inodes to one per page of lowmem;
3790 * but the internal instance is left unlimited.
3792 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3793 sbinfo
->max_blocks
= shmem_default_max_blocks();
3794 sbinfo
->max_inodes
= shmem_default_max_inodes();
3795 if (shmem_parse_options(data
, sbinfo
, false)) {
3800 sb
->s_flags
|= SB_NOUSER
;
3802 sb
->s_export_op
= &shmem_export_ops
;
3803 sb
->s_flags
|= SB_NOSEC
;
3805 sb
->s_flags
|= SB_NOUSER
;
3808 spin_lock_init(&sbinfo
->stat_lock
);
3809 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3811 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3812 spin_lock_init(&sbinfo
->shrinklist_lock
);
3813 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3815 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3816 sb
->s_blocksize
= PAGE_SIZE
;
3817 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3818 sb
->s_magic
= TMPFS_MAGIC
;
3819 sb
->s_op
= &shmem_ops
;
3820 sb
->s_time_gran
= 1;
3821 #ifdef CONFIG_TMPFS_XATTR
3822 sb
->s_xattr
= shmem_xattr_handlers
;
3824 #ifdef CONFIG_TMPFS_POSIX_ACL
3825 sb
->s_flags
|= SB_POSIXACL
;
3827 uuid_gen(&sb
->s_uuid
);
3829 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3832 inode
->i_uid
= sbinfo
->uid
;
3833 inode
->i_gid
= sbinfo
->gid
;
3834 sb
->s_root
= d_make_root(inode
);
3840 shmem_put_super(sb
);
3844 static struct kmem_cache
*shmem_inode_cachep
;
3846 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3848 struct shmem_inode_info
*info
;
3849 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3852 return &info
->vfs_inode
;
3855 static void shmem_destroy_callback(struct rcu_head
*head
)
3857 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3858 if (S_ISLNK(inode
->i_mode
))
3859 kfree(inode
->i_link
);
3860 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3863 static void shmem_destroy_inode(struct inode
*inode
)
3865 if (S_ISREG(inode
->i_mode
))
3866 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3867 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3870 static void shmem_init_inode(void *foo
)
3872 struct shmem_inode_info
*info
= foo
;
3873 inode_init_once(&info
->vfs_inode
);
3876 static void shmem_init_inodecache(void)
3878 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3879 sizeof(struct shmem_inode_info
),
3880 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3883 static void shmem_destroy_inodecache(void)
3885 kmem_cache_destroy(shmem_inode_cachep
);
3888 static __init
void shmem_no_idr(struct super_block
*sb
)
3890 struct shmem_sb_info
*sbinfo
;
3892 sbinfo
= SHMEM_SB(sb
);
3893 sbinfo
->idr_nouse
= true;
3894 idr_destroy(&sbinfo
->idr
);
3897 static const struct address_space_operations shmem_aops
= {
3898 .writepage
= shmem_writepage
,
3899 .set_page_dirty
= __set_page_dirty_no_writeback
,
3901 .write_begin
= shmem_write_begin
,
3902 .write_end
= shmem_write_end
,
3904 #ifdef CONFIG_MIGRATION
3905 .migratepage
= migrate_page
,
3907 .error_remove_page
= generic_error_remove_page
,
3910 static const struct file_operations shmem_file_operations
= {
3912 .get_unmapped_area
= shmem_get_unmapped_area
,
3914 .llseek
= shmem_file_llseek
,
3915 .read_iter
= shmem_file_read_iter
,
3916 .write_iter
= generic_file_write_iter
,
3917 .fsync
= noop_fsync
,
3918 .splice_read
= generic_file_splice_read
,
3919 .splice_write
= iter_file_splice_write
,
3920 .fallocate
= shmem_fallocate
,
3924 static const struct inode_operations shmem_inode_operations
= {
3925 .getattr
= shmem_getattr
,
3926 .setattr
= shmem_setattr
,
3927 #ifdef CONFIG_TMPFS_XATTR
3928 .listxattr
= shmem_listxattr
,
3929 .set_acl
= simple_set_acl
,
3933 static const struct inode_operations shmem_dir_inode_operations
= {
3935 .create
= shmem_create
,
3936 .lookup
= simple_lookup
,
3938 .unlink
= shmem_unlink
,
3939 .symlink
= shmem_symlink
,
3940 .mkdir
= shmem_mkdir
,
3941 .rmdir
= shmem_rmdir
,
3942 .mknod
= shmem_mknod
,
3943 .rename
= shmem_rename2
,
3944 .tmpfile
= shmem_tmpfile
,
3946 #ifdef CONFIG_TMPFS_XATTR
3947 .listxattr
= shmem_listxattr
,
3949 #ifdef CONFIG_TMPFS_POSIX_ACL
3950 .setattr
= shmem_setattr
,
3951 .set_acl
= simple_set_acl
,
3955 static const struct inode_operations shmem_special_inode_operations
= {
3956 #ifdef CONFIG_TMPFS_XATTR
3957 .listxattr
= shmem_listxattr
,
3959 #ifdef CONFIG_TMPFS_POSIX_ACL
3960 .setattr
= shmem_setattr
,
3961 .set_acl
= simple_set_acl
,
3965 static const struct super_operations shmem_ops
= {
3966 .alloc_inode
= shmem_alloc_inode
,
3967 .destroy_inode
= shmem_destroy_inode
,
3969 .statfs
= shmem_statfs
,
3970 .remount_fs
= shmem_remount_fs
,
3971 .show_options
= shmem_show_options
,
3973 .evict_inode
= shmem_evict_inode
,
3974 .drop_inode
= generic_delete_inode
,
3975 .put_super
= shmem_put_super
,
3976 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3977 .nr_cached_objects
= shmem_unused_huge_count
,
3978 .free_cached_objects
= shmem_unused_huge_scan
,
3982 static const struct vm_operations_struct shmem_vm_ops
= {
3983 .fault
= shmem_fault
,
3984 .map_pages
= filemap_map_pages
,
3986 .set_policy
= shmem_set_policy
,
3987 .get_policy
= shmem_get_policy
,
3991 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3992 int flags
, const char *dev_name
, void *data
)
3994 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3997 static struct file_system_type shmem_fs_type
= {
3998 .owner
= THIS_MODULE
,
4000 .mount
= shmem_mount
,
4001 .kill_sb
= kill_litter_super
,
4002 .fs_flags
= FS_USERNS_MOUNT
,
4005 int __init
shmem_init(void)
4009 /* If rootfs called this, don't re-init */
4010 if (shmem_inode_cachep
)
4013 shmem_init_inodecache();
4015 error
= register_filesystem(&shmem_fs_type
);
4017 pr_err("Could not register tmpfs\n");
4021 shm_mnt
= kern_mount(&shmem_fs_type
);
4022 if (IS_ERR(shm_mnt
)) {
4023 error
= PTR_ERR(shm_mnt
);
4024 pr_err("Could not kern_mount tmpfs\n");
4027 shmem_no_idr(shm_mnt
->mnt_sb
);
4029 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4030 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
4031 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4033 shmem_huge
= 0; /* just in case it was patched */
4038 unregister_filesystem(&shmem_fs_type
);
4040 shmem_destroy_inodecache();
4041 shm_mnt
= ERR_PTR(error
);
4045 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
4046 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4047 struct kobj_attribute
*attr
, char *buf
)
4051 SHMEM_HUGE_WITHIN_SIZE
,
4059 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
4060 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
4062 count
+= sprintf(buf
+ count
, fmt
,
4063 shmem_format_huge(values
[i
]));
4065 buf
[count
- 1] = '\n';
4069 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4070 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4075 if (count
+ 1 > sizeof(tmp
))
4077 memcpy(tmp
, buf
, count
);
4079 if (count
&& tmp
[count
- 1] == '\n')
4080 tmp
[count
- 1] = '\0';
4082 huge
= shmem_parse_huge(tmp
);
4083 if (huge
== -EINVAL
)
4085 if (!has_transparent_hugepage() &&
4086 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4090 if (shmem_huge
> SHMEM_HUGE_DENY
)
4091 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4095 struct kobj_attribute shmem_enabled_attr
=
4096 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4097 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4099 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4100 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4102 struct inode
*inode
= file_inode(vma
->vm_file
);
4103 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4107 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4109 if (shmem_huge
== SHMEM_HUGE_DENY
)
4111 switch (sbinfo
->huge
) {
4112 case SHMEM_HUGE_NEVER
:
4114 case SHMEM_HUGE_ALWAYS
:
4116 case SHMEM_HUGE_WITHIN_SIZE
:
4117 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4118 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4119 if (i_size
>= HPAGE_PMD_SIZE
&&
4120 i_size
>> PAGE_SHIFT
>= off
)
4123 case SHMEM_HUGE_ADVISE
:
4124 /* TODO: implement fadvise() hints */
4125 return (vma
->vm_flags
& VM_HUGEPAGE
);
4131 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4133 #else /* !CONFIG_SHMEM */
4136 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4138 * This is intended for small system where the benefits of the full
4139 * shmem code (swap-backed and resource-limited) are outweighed by
4140 * their complexity. On systems without swap this code should be
4141 * effectively equivalent, but much lighter weight.
4144 static struct file_system_type shmem_fs_type
= {
4146 .mount
= ramfs_mount
,
4147 .kill_sb
= kill_litter_super
,
4148 .fs_flags
= FS_USERNS_MOUNT
,
4151 int __init
shmem_init(void)
4153 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4155 shm_mnt
= kern_mount(&shmem_fs_type
);
4156 BUG_ON(IS_ERR(shm_mnt
));
4161 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
4166 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4171 void shmem_unlock_mapping(struct address_space
*mapping
)
4176 unsigned long shmem_get_unmapped_area(struct file
*file
,
4177 unsigned long addr
, unsigned long len
,
4178 unsigned long pgoff
, unsigned long flags
)
4180 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4184 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4186 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4188 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4190 #define shmem_vm_ops generic_file_vm_ops
4191 #define shmem_file_operations ramfs_file_operations
4192 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4193 #define shmem_acct_size(flags, size) 0
4194 #define shmem_unacct_size(flags, size) do {} while (0)
4196 #endif /* CONFIG_SHMEM */
4200 static const struct dentry_operations anon_ops
= {
4201 .d_dname
= simple_dname
4204 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4205 unsigned long flags
, unsigned int i_flags
)
4208 struct inode
*inode
;
4210 struct super_block
*sb
;
4214 return ERR_CAST(mnt
);
4216 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4217 return ERR_PTR(-EINVAL
);
4219 if (shmem_acct_size(flags
, size
))
4220 return ERR_PTR(-ENOMEM
);
4222 res
= ERR_PTR(-ENOMEM
);
4224 this.len
= strlen(name
);
4225 this.hash
= 0; /* will go */
4227 path
.mnt
= mntget(mnt
);
4228 path
.dentry
= d_alloc_pseudo(sb
, &this);
4231 d_set_d_op(path
.dentry
, &anon_ops
);
4233 res
= ERR_PTR(-ENOSPC
);
4234 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
4238 inode
->i_flags
|= i_flags
;
4239 d_instantiate(path
.dentry
, inode
);
4240 inode
->i_size
= size
;
4241 clear_nlink(inode
); /* It is unlinked */
4242 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4246 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
4247 &shmem_file_operations
);
4254 shmem_unacct_size(flags
, size
);
4261 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4262 * kernel internal. There will be NO LSM permission checks against the
4263 * underlying inode. So users of this interface must do LSM checks at a
4264 * higher layer. The users are the big_key and shm implementations. LSM
4265 * checks are provided at the key or shm level rather than the inode.
4266 * @name: name for dentry (to be seen in /proc/<pid>/maps
4267 * @size: size to be set for the file
4268 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4270 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4272 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4276 * shmem_file_setup - get an unlinked file living in tmpfs
4277 * @name: name for dentry (to be seen in /proc/<pid>/maps
4278 * @size: size to be set for the file
4279 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4281 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4283 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4285 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4288 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4289 * @mnt: the tmpfs mount where the file will be created
4290 * @name: name for dentry (to be seen in /proc/<pid>/maps
4291 * @size: size to be set for the file
4292 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4294 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4295 loff_t size
, unsigned long flags
)
4297 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4299 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4302 * shmem_zero_setup - setup a shared anonymous mapping
4303 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4305 int shmem_zero_setup(struct vm_area_struct
*vma
)
4308 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4311 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4312 * between XFS directory reading and selinux: since this file is only
4313 * accessible to the user through its mapping, use S_PRIVATE flag to
4314 * bypass file security, in the same way as shmem_kernel_file_setup().
4316 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4318 return PTR_ERR(file
);
4322 vma
->vm_file
= file
;
4323 vma
->vm_ops
= &shmem_vm_ops
;
4325 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4326 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4327 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4328 khugepaged_enter(vma
, vma
->vm_flags
);
4335 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4336 * @mapping: the page's address_space
4337 * @index: the page index
4338 * @gfp: the page allocator flags to use if allocating
4340 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4341 * with any new page allocations done using the specified allocation flags.
4342 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4343 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4344 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4346 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4347 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4349 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4350 pgoff_t index
, gfp_t gfp
)
4353 struct inode
*inode
= mapping
->host
;
4357 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4358 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4359 gfp
, NULL
, NULL
, NULL
);
4361 page
= ERR_PTR(error
);
4367 * The tiny !SHMEM case uses ramfs without swap
4369 return read_cache_page_gfp(mapping
, index
, gfp
);
4372 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);