2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
35 #include <linux/khugepaged.h>
37 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
39 static struct vfsmount
*shm_mnt
;
43 * This virtual memory filesystem is heavily based on the ramfs. It
44 * extends ramfs by the ability to use swap and honor resource limits
45 * which makes it a completely usable filesystem.
48 #include <linux/xattr.h>
49 #include <linux/exportfs.h>
50 #include <linux/posix_acl.h>
51 #include <linux/posix_acl_xattr.h>
52 #include <linux/mman.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/backing-dev.h>
56 #include <linux/shmem_fs.h>
57 #include <linux/writeback.h>
58 #include <linux/blkdev.h>
59 #include <linux/pagevec.h>
60 #include <linux/percpu_counter.h>
61 #include <linux/falloc.h>
62 #include <linux/splice.h>
63 #include <linux/security.h>
64 #include <linux/swapops.h>
65 #include <linux/mempolicy.h>
66 #include <linux/namei.h>
67 #include <linux/ctype.h>
68 #include <linux/migrate.h>
69 #include <linux/highmem.h>
70 #include <linux/seq_file.h>
71 #include <linux/magic.h>
72 #include <linux/syscalls.h>
73 #include <linux/fcntl.h>
74 #include <uapi/linux/memfd.h>
75 #include <linux/userfaultfd_k.h>
76 #include <linux/rmap.h>
78 #include <linux/uaccess.h>
79 #include <asm/pgtable.h>
83 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
84 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
86 /* Pretend that each entry is of this size in directory's i_size */
87 #define BOGO_DIRENT_SIZE 20
89 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
90 #define SHORT_SYMLINK_LEN 128
93 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
94 * inode->i_private (with i_mutex making sure that it has only one user at
95 * a time): we would prefer not to enlarge the shmem inode just for that.
98 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
99 pgoff_t start
; /* start of range currently being fallocated */
100 pgoff_t next
; /* the next page offset to be fallocated */
101 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
102 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
106 static unsigned long shmem_default_max_blocks(void)
108 return totalram_pages
/ 2;
111 static unsigned long shmem_default_max_inodes(void)
113 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
117 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
118 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
119 struct shmem_inode_info
*info
, pgoff_t index
);
120 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
121 struct page
**pagep
, enum sgp_type sgp
,
122 gfp_t gfp
, struct vm_area_struct
*vma
,
123 struct vm_fault
*vmf
, int *fault_type
);
125 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
126 struct page
**pagep
, enum sgp_type sgp
)
128 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
129 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
132 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
134 return sb
->s_fs_info
;
138 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
139 * for shared memory and for shared anonymous (/dev/zero) mappings
140 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
141 * consistent with the pre-accounting of private mappings ...
143 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
145 return (flags
& VM_NORESERVE
) ?
146 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
149 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
151 if (!(flags
& VM_NORESERVE
))
152 vm_unacct_memory(VM_ACCT(size
));
155 static inline int shmem_reacct_size(unsigned long flags
,
156 loff_t oldsize
, loff_t newsize
)
158 if (!(flags
& VM_NORESERVE
)) {
159 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
160 return security_vm_enough_memory_mm(current
->mm
,
161 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
162 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
163 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
169 * ... whereas tmpfs objects are accounted incrementally as
170 * pages are allocated, in order to allow large sparse files.
171 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
172 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
174 static inline int shmem_acct_block(unsigned long flags
, long pages
)
176 if (!(flags
& VM_NORESERVE
))
179 return security_vm_enough_memory_mm(current
->mm
,
180 pages
* VM_ACCT(PAGE_SIZE
));
183 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
185 if (flags
& VM_NORESERVE
)
186 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
189 static const struct super_operations shmem_ops
;
190 static const struct address_space_operations shmem_aops
;
191 static const struct file_operations shmem_file_operations
;
192 static const struct inode_operations shmem_inode_operations
;
193 static const struct inode_operations shmem_dir_inode_operations
;
194 static const struct inode_operations shmem_special_inode_operations
;
195 static const struct vm_operations_struct shmem_vm_ops
;
196 static struct file_system_type shmem_fs_type
;
198 bool vma_is_shmem(struct vm_area_struct
*vma
)
200 return vma
->vm_ops
== &shmem_vm_ops
;
203 static LIST_HEAD(shmem_swaplist
);
204 static DEFINE_MUTEX(shmem_swaplist_mutex
);
206 static int shmem_reserve_inode(struct super_block
*sb
)
208 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
209 if (sbinfo
->max_inodes
) {
210 spin_lock(&sbinfo
->stat_lock
);
211 if (!sbinfo
->free_inodes
) {
212 spin_unlock(&sbinfo
->stat_lock
);
215 sbinfo
->free_inodes
--;
216 spin_unlock(&sbinfo
->stat_lock
);
221 static void shmem_free_inode(struct super_block
*sb
)
223 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
224 if (sbinfo
->max_inodes
) {
225 spin_lock(&sbinfo
->stat_lock
);
226 sbinfo
->free_inodes
++;
227 spin_unlock(&sbinfo
->stat_lock
);
232 * shmem_recalc_inode - recalculate the block usage of an inode
233 * @inode: inode to recalc
235 * We have to calculate the free blocks since the mm can drop
236 * undirtied hole pages behind our back.
238 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
239 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
241 * It has to be called with the spinlock held.
243 static void shmem_recalc_inode(struct inode
*inode
)
245 struct shmem_inode_info
*info
= SHMEM_I(inode
);
248 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
250 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
251 if (sbinfo
->max_blocks
)
252 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
253 info
->alloced
-= freed
;
254 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
255 shmem_unacct_blocks(info
->flags
, freed
);
259 bool shmem_charge(struct inode
*inode
, long pages
)
261 struct shmem_inode_info
*info
= SHMEM_I(inode
);
262 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
265 if (shmem_acct_block(info
->flags
, pages
))
267 spin_lock_irqsave(&info
->lock
, flags
);
268 info
->alloced
+= pages
;
269 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
270 shmem_recalc_inode(inode
);
271 spin_unlock_irqrestore(&info
->lock
, flags
);
272 inode
->i_mapping
->nrpages
+= pages
;
274 if (!sbinfo
->max_blocks
)
276 if (percpu_counter_compare(&sbinfo
->used_blocks
,
277 sbinfo
->max_blocks
- pages
) > 0) {
278 inode
->i_mapping
->nrpages
-= pages
;
279 spin_lock_irqsave(&info
->lock
, flags
);
280 info
->alloced
-= pages
;
281 shmem_recalc_inode(inode
);
282 spin_unlock_irqrestore(&info
->lock
, flags
);
283 shmem_unacct_blocks(info
->flags
, pages
);
286 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
290 void shmem_uncharge(struct inode
*inode
, long pages
)
292 struct shmem_inode_info
*info
= SHMEM_I(inode
);
293 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
296 spin_lock_irqsave(&info
->lock
, flags
);
297 info
->alloced
-= pages
;
298 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
299 shmem_recalc_inode(inode
);
300 spin_unlock_irqrestore(&info
->lock
, flags
);
302 if (sbinfo
->max_blocks
)
303 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
304 shmem_unacct_blocks(info
->flags
, pages
);
308 * Replace item expected in radix tree by a new item, while holding tree lock.
310 static int shmem_radix_tree_replace(struct address_space
*mapping
,
311 pgoff_t index
, void *expected
, void *replacement
)
313 struct radix_tree_node
*node
;
317 VM_BUG_ON(!expected
);
318 VM_BUG_ON(!replacement
);
319 item
= __radix_tree_lookup(&mapping
->page_tree
, index
, &node
, &pslot
);
322 if (item
!= expected
)
324 __radix_tree_replace(&mapping
->page_tree
, node
, pslot
,
325 replacement
, NULL
, NULL
);
330 * Sometimes, before we decide whether to proceed or to fail, we must check
331 * that an entry was not already brought back from swap by a racing thread.
333 * Checking page is not enough: by the time a SwapCache page is locked, it
334 * might be reused, and again be SwapCache, using the same swap as before.
336 static bool shmem_confirm_swap(struct address_space
*mapping
,
337 pgoff_t index
, swp_entry_t swap
)
342 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
344 return item
== swp_to_radix_entry(swap
);
348 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
351 * disables huge pages for the mount;
353 * enables huge pages for the mount;
354 * SHMEM_HUGE_WITHIN_SIZE:
355 * only allocate huge pages if the page will be fully within i_size,
356 * also respect fadvise()/madvise() hints;
358 * only allocate huge pages if requested with fadvise()/madvise();
361 #define SHMEM_HUGE_NEVER 0
362 #define SHMEM_HUGE_ALWAYS 1
363 #define SHMEM_HUGE_WITHIN_SIZE 2
364 #define SHMEM_HUGE_ADVISE 3
368 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
371 * disables huge on shm_mnt and all mounts, for emergency use;
373 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
376 #define SHMEM_HUGE_DENY (-1)
377 #define SHMEM_HUGE_FORCE (-2)
379 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
380 /* ifdef here to avoid bloating shmem.o when not necessary */
382 int shmem_huge __read_mostly
;
384 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
385 static int shmem_parse_huge(const char *str
)
387 if (!strcmp(str
, "never"))
388 return SHMEM_HUGE_NEVER
;
389 if (!strcmp(str
, "always"))
390 return SHMEM_HUGE_ALWAYS
;
391 if (!strcmp(str
, "within_size"))
392 return SHMEM_HUGE_WITHIN_SIZE
;
393 if (!strcmp(str
, "advise"))
394 return SHMEM_HUGE_ADVISE
;
395 if (!strcmp(str
, "deny"))
396 return SHMEM_HUGE_DENY
;
397 if (!strcmp(str
, "force"))
398 return SHMEM_HUGE_FORCE
;
402 static const char *shmem_format_huge(int huge
)
405 case SHMEM_HUGE_NEVER
:
407 case SHMEM_HUGE_ALWAYS
:
409 case SHMEM_HUGE_WITHIN_SIZE
:
410 return "within_size";
411 case SHMEM_HUGE_ADVISE
:
413 case SHMEM_HUGE_DENY
:
415 case SHMEM_HUGE_FORCE
:
424 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
425 struct shrink_control
*sc
, unsigned long nr_to_split
)
427 LIST_HEAD(list
), *pos
, *next
;
428 LIST_HEAD(to_remove
);
430 struct shmem_inode_info
*info
;
432 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
433 int removed
= 0, split
= 0;
435 if (list_empty(&sbinfo
->shrinklist
))
438 spin_lock(&sbinfo
->shrinklist_lock
);
439 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
440 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
443 inode
= igrab(&info
->vfs_inode
);
445 /* inode is about to be evicted */
447 list_del_init(&info
->shrinklist
);
452 /* Check if there's anything to gain */
453 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
454 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
455 list_move(&info
->shrinklist
, &to_remove
);
460 list_move(&info
->shrinklist
, &list
);
465 spin_unlock(&sbinfo
->shrinklist_lock
);
467 list_for_each_safe(pos
, next
, &to_remove
) {
468 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
469 inode
= &info
->vfs_inode
;
470 list_del_init(&info
->shrinklist
);
474 list_for_each_safe(pos
, next
, &list
) {
477 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
478 inode
= &info
->vfs_inode
;
480 if (nr_to_split
&& split
>= nr_to_split
) {
485 page
= find_lock_page(inode
->i_mapping
,
486 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
490 if (!PageTransHuge(page
)) {
496 ret
= split_huge_page(page
);
501 /* split failed: leave it on the list */
508 list_del_init(&info
->shrinklist
);
513 spin_lock(&sbinfo
->shrinklist_lock
);
514 list_splice_tail(&list
, &sbinfo
->shrinklist
);
515 sbinfo
->shrinklist_len
-= removed
;
516 spin_unlock(&sbinfo
->shrinklist_lock
);
521 static long shmem_unused_huge_scan(struct super_block
*sb
,
522 struct shrink_control
*sc
)
524 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
526 if (!READ_ONCE(sbinfo
->shrinklist_len
))
529 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
532 static long shmem_unused_huge_count(struct super_block
*sb
,
533 struct shrink_control
*sc
)
535 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
536 return READ_ONCE(sbinfo
->shrinklist_len
);
538 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
540 #define shmem_huge SHMEM_HUGE_DENY
542 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
543 struct shrink_control
*sc
, unsigned long nr_to_split
)
547 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
550 * Like add_to_page_cache_locked, but error if expected item has gone.
552 static int shmem_add_to_page_cache(struct page
*page
,
553 struct address_space
*mapping
,
554 pgoff_t index
, void *expected
)
556 int error
, nr
= hpage_nr_pages(page
);
558 VM_BUG_ON_PAGE(PageTail(page
), page
);
559 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
560 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
561 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
562 VM_BUG_ON(expected
&& PageTransHuge(page
));
564 page_ref_add(page
, nr
);
565 page
->mapping
= mapping
;
568 spin_lock_irq(&mapping
->tree_lock
);
569 if (PageTransHuge(page
)) {
570 void __rcu
**results
;
575 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
,
576 &results
, &idx
, index
, 1) &&
577 idx
< index
+ HPAGE_PMD_NR
) {
582 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
583 error
= radix_tree_insert(&mapping
->page_tree
,
584 index
+ i
, page
+ i
);
587 count_vm_event(THP_FILE_ALLOC
);
589 } else if (!expected
) {
590 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
592 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
597 mapping
->nrpages
+= nr
;
598 if (PageTransHuge(page
))
599 __inc_node_page_state(page
, NR_SHMEM_THPS
);
600 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
601 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
602 spin_unlock_irq(&mapping
->tree_lock
);
604 page
->mapping
= NULL
;
605 spin_unlock_irq(&mapping
->tree_lock
);
606 page_ref_sub(page
, nr
);
612 * Like delete_from_page_cache, but substitutes swap for page.
614 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
616 struct address_space
*mapping
= page
->mapping
;
619 VM_BUG_ON_PAGE(PageCompound(page
), page
);
621 spin_lock_irq(&mapping
->tree_lock
);
622 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
623 page
->mapping
= NULL
;
625 __dec_node_page_state(page
, NR_FILE_PAGES
);
626 __dec_node_page_state(page
, NR_SHMEM
);
627 spin_unlock_irq(&mapping
->tree_lock
);
633 * Remove swap entry from radix tree, free the swap and its page cache.
635 static int shmem_free_swap(struct address_space
*mapping
,
636 pgoff_t index
, void *radswap
)
640 spin_lock_irq(&mapping
->tree_lock
);
641 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
642 spin_unlock_irq(&mapping
->tree_lock
);
645 free_swap_and_cache(radix_to_swp_entry(radswap
));
650 * Determine (in bytes) how many of the shmem object's pages mapped by the
651 * given offsets are swapped out.
653 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
654 * as long as the inode doesn't go away and racy results are not a problem.
656 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
657 pgoff_t start
, pgoff_t end
)
659 struct radix_tree_iter iter
;
662 unsigned long swapped
= 0;
666 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
667 if (iter
.index
>= end
)
670 page
= radix_tree_deref_slot(slot
);
672 if (radix_tree_deref_retry(page
)) {
673 slot
= radix_tree_iter_retry(&iter
);
677 if (radix_tree_exceptional_entry(page
))
680 if (need_resched()) {
681 slot
= radix_tree_iter_resume(slot
, &iter
);
688 return swapped
<< PAGE_SHIFT
;
692 * Determine (in bytes) how many of the shmem object's pages mapped by the
693 * given vma is swapped out.
695 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
696 * as long as the inode doesn't go away and racy results are not a problem.
698 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
700 struct inode
*inode
= file_inode(vma
->vm_file
);
701 struct shmem_inode_info
*info
= SHMEM_I(inode
);
702 struct address_space
*mapping
= inode
->i_mapping
;
703 unsigned long swapped
;
705 /* Be careful as we don't hold info->lock */
706 swapped
= READ_ONCE(info
->swapped
);
709 * The easier cases are when the shmem object has nothing in swap, or
710 * the vma maps it whole. Then we can simply use the stats that we
716 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
717 return swapped
<< PAGE_SHIFT
;
719 /* Here comes the more involved part */
720 return shmem_partial_swap_usage(mapping
,
721 linear_page_index(vma
, vma
->vm_start
),
722 linear_page_index(vma
, vma
->vm_end
));
726 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
728 void shmem_unlock_mapping(struct address_space
*mapping
)
731 pgoff_t indices
[PAGEVEC_SIZE
];
734 pagevec_init(&pvec
, 0);
736 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
738 while (!mapping_unevictable(mapping
)) {
740 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
741 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
743 pvec
.nr
= find_get_entries(mapping
, index
,
744 PAGEVEC_SIZE
, pvec
.pages
, indices
);
747 index
= indices
[pvec
.nr
- 1] + 1;
748 pagevec_remove_exceptionals(&pvec
);
749 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
750 pagevec_release(&pvec
);
756 * Remove range of pages and swap entries from radix tree, and free them.
757 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
759 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
762 struct address_space
*mapping
= inode
->i_mapping
;
763 struct shmem_inode_info
*info
= SHMEM_I(inode
);
764 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
765 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
766 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
767 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
769 pgoff_t indices
[PAGEVEC_SIZE
];
770 long nr_swaps_freed
= 0;
775 end
= -1; /* unsigned, so actually very big */
777 pagevec_init(&pvec
, 0);
779 while (index
< end
) {
780 pvec
.nr
= find_get_entries(mapping
, index
,
781 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
782 pvec
.pages
, indices
);
785 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
786 struct page
*page
= pvec
.pages
[i
];
792 if (radix_tree_exceptional_entry(page
)) {
795 nr_swaps_freed
+= !shmem_free_swap(mapping
,
800 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
802 if (!trylock_page(page
))
805 if (PageTransTail(page
)) {
806 /* Middle of THP: zero out the page */
807 clear_highpage(page
);
810 } else if (PageTransHuge(page
)) {
811 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
813 * Range ends in the middle of THP:
816 clear_highpage(page
);
820 index
+= HPAGE_PMD_NR
- 1;
821 i
+= HPAGE_PMD_NR
- 1;
824 if (!unfalloc
|| !PageUptodate(page
)) {
825 VM_BUG_ON_PAGE(PageTail(page
), page
);
826 if (page_mapping(page
) == mapping
) {
827 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
828 truncate_inode_page(mapping
, page
);
833 pagevec_remove_exceptionals(&pvec
);
834 pagevec_release(&pvec
);
840 struct page
*page
= NULL
;
841 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
843 unsigned int top
= PAGE_SIZE
;
848 zero_user_segment(page
, partial_start
, top
);
849 set_page_dirty(page
);
855 struct page
*page
= NULL
;
856 shmem_getpage(inode
, end
, &page
, SGP_READ
);
858 zero_user_segment(page
, 0, partial_end
);
859 set_page_dirty(page
);
868 while (index
< end
) {
871 pvec
.nr
= find_get_entries(mapping
, index
,
872 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
873 pvec
.pages
, indices
);
875 /* If all gone or hole-punch or unfalloc, we're done */
876 if (index
== start
|| end
!= -1)
878 /* But if truncating, restart to make sure all gone */
882 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
883 struct page
*page
= pvec
.pages
[i
];
889 if (radix_tree_exceptional_entry(page
)) {
892 if (shmem_free_swap(mapping
, index
, page
)) {
893 /* Swap was replaced by page: retry */
903 if (PageTransTail(page
)) {
904 /* Middle of THP: zero out the page */
905 clear_highpage(page
);
908 * Partial thp truncate due 'start' in middle
909 * of THP: don't need to look on these pages
910 * again on !pvec.nr restart.
912 if (index
!= round_down(end
, HPAGE_PMD_NR
))
915 } else if (PageTransHuge(page
)) {
916 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
918 * Range ends in the middle of THP:
921 clear_highpage(page
);
925 index
+= HPAGE_PMD_NR
- 1;
926 i
+= HPAGE_PMD_NR
- 1;
929 if (!unfalloc
|| !PageUptodate(page
)) {
930 VM_BUG_ON_PAGE(PageTail(page
), page
);
931 if (page_mapping(page
) == mapping
) {
932 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
933 truncate_inode_page(mapping
, page
);
935 /* Page was replaced by swap: retry */
943 pagevec_remove_exceptionals(&pvec
);
944 pagevec_release(&pvec
);
948 spin_lock_irq(&info
->lock
);
949 info
->swapped
-= nr_swaps_freed
;
950 shmem_recalc_inode(inode
);
951 spin_unlock_irq(&info
->lock
);
954 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
956 shmem_undo_range(inode
, lstart
, lend
, false);
957 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
959 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
961 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
964 struct inode
*inode
= dentry
->d_inode
;
965 struct shmem_inode_info
*info
= SHMEM_I(inode
);
967 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
968 spin_lock_irq(&info
->lock
);
969 shmem_recalc_inode(inode
);
970 spin_unlock_irq(&info
->lock
);
972 generic_fillattr(inode
, stat
);
976 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
978 struct inode
*inode
= d_inode(dentry
);
979 struct shmem_inode_info
*info
= SHMEM_I(inode
);
980 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
983 error
= setattr_prepare(dentry
, attr
);
987 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
988 loff_t oldsize
= inode
->i_size
;
989 loff_t newsize
= attr
->ia_size
;
991 /* protected by i_mutex */
992 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
993 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
996 if (newsize
!= oldsize
) {
997 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1001 i_size_write(inode
, newsize
);
1002 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1004 if (newsize
<= oldsize
) {
1005 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1006 if (oldsize
> holebegin
)
1007 unmap_mapping_range(inode
->i_mapping
,
1010 shmem_truncate_range(inode
,
1011 newsize
, (loff_t
)-1);
1012 /* unmap again to remove racily COWed private pages */
1013 if (oldsize
> holebegin
)
1014 unmap_mapping_range(inode
->i_mapping
,
1018 * Part of the huge page can be beyond i_size: subject
1019 * to shrink under memory pressure.
1021 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1022 spin_lock(&sbinfo
->shrinklist_lock
);
1023 if (list_empty(&info
->shrinklist
)) {
1024 list_add_tail(&info
->shrinklist
,
1025 &sbinfo
->shrinklist
);
1026 sbinfo
->shrinklist_len
++;
1028 spin_unlock(&sbinfo
->shrinklist_lock
);
1033 setattr_copy(inode
, attr
);
1034 if (attr
->ia_valid
& ATTR_MODE
)
1035 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1039 static void shmem_evict_inode(struct inode
*inode
)
1041 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1042 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1044 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1045 shmem_unacct_size(info
->flags
, inode
->i_size
);
1047 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1048 if (!list_empty(&info
->shrinklist
)) {
1049 spin_lock(&sbinfo
->shrinklist_lock
);
1050 if (!list_empty(&info
->shrinklist
)) {
1051 list_del_init(&info
->shrinklist
);
1052 sbinfo
->shrinklist_len
--;
1054 spin_unlock(&sbinfo
->shrinklist_lock
);
1056 if (!list_empty(&info
->swaplist
)) {
1057 mutex_lock(&shmem_swaplist_mutex
);
1058 list_del_init(&info
->swaplist
);
1059 mutex_unlock(&shmem_swaplist_mutex
);
1063 simple_xattrs_free(&info
->xattrs
);
1064 WARN_ON(inode
->i_blocks
);
1065 shmem_free_inode(inode
->i_sb
);
1069 static unsigned long find_swap_entry(struct radix_tree_root
*root
, void *item
)
1071 struct radix_tree_iter iter
;
1073 unsigned long found
= -1;
1074 unsigned int checked
= 0;
1077 radix_tree_for_each_slot(slot
, root
, &iter
, 0) {
1078 if (*slot
== item
) {
1083 if ((checked
% 4096) != 0)
1085 slot
= radix_tree_iter_resume(slot
, &iter
);
1094 * If swap found in inode, free it and move page from swapcache to filecache.
1096 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
1097 swp_entry_t swap
, struct page
**pagep
)
1099 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1105 radswap
= swp_to_radix_entry(swap
);
1106 index
= find_swap_entry(&mapping
->page_tree
, radswap
);
1108 return -EAGAIN
; /* tell shmem_unuse we found nothing */
1111 * Move _head_ to start search for next from here.
1112 * But be careful: shmem_evict_inode checks list_empty without taking
1113 * mutex, and there's an instant in list_move_tail when info->swaplist
1114 * would appear empty, if it were the only one on shmem_swaplist.
1116 if (shmem_swaplist
.next
!= &info
->swaplist
)
1117 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
1119 gfp
= mapping_gfp_mask(mapping
);
1120 if (shmem_should_replace_page(*pagep
, gfp
)) {
1121 mutex_unlock(&shmem_swaplist_mutex
);
1122 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
1123 mutex_lock(&shmem_swaplist_mutex
);
1125 * We needed to drop mutex to make that restrictive page
1126 * allocation, but the inode might have been freed while we
1127 * dropped it: although a racing shmem_evict_inode() cannot
1128 * complete without emptying the radix_tree, our page lock
1129 * on this swapcache page is not enough to prevent that -
1130 * free_swap_and_cache() of our swap entry will only
1131 * trylock_page(), removing swap from radix_tree whatever.
1133 * We must not proceed to shmem_add_to_page_cache() if the
1134 * inode has been freed, but of course we cannot rely on
1135 * inode or mapping or info to check that. However, we can
1136 * safely check if our swap entry is still in use (and here
1137 * it can't have got reused for another page): if it's still
1138 * in use, then the inode cannot have been freed yet, and we
1139 * can safely proceed (if it's no longer in use, that tells
1140 * nothing about the inode, but we don't need to unuse swap).
1142 if (!page_swapcount(*pagep
))
1147 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1148 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1149 * beneath us (pagelock doesn't help until the page is in pagecache).
1152 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
1154 if (error
!= -ENOMEM
) {
1156 * Truncation and eviction use free_swap_and_cache(), which
1157 * only does trylock page: if we raced, best clean up here.
1159 delete_from_swap_cache(*pagep
);
1160 set_page_dirty(*pagep
);
1162 spin_lock_irq(&info
->lock
);
1164 spin_unlock_irq(&info
->lock
);
1172 * Search through swapped inodes to find and replace swap by page.
1174 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1176 struct list_head
*this, *next
;
1177 struct shmem_inode_info
*info
;
1178 struct mem_cgroup
*memcg
;
1182 * There's a faint possibility that swap page was replaced before
1183 * caller locked it: caller will come back later with the right page.
1185 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1189 * Charge page using GFP_KERNEL while we can wait, before taking
1190 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1191 * Charged back to the user (not to caller) when swap account is used.
1193 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1197 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1200 mutex_lock(&shmem_swaplist_mutex
);
1201 list_for_each_safe(this, next
, &shmem_swaplist
) {
1202 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1204 error
= shmem_unuse_inode(info
, swap
, &page
);
1206 list_del_init(&info
->swaplist
);
1208 if (error
!= -EAGAIN
)
1210 /* found nothing in this: move on to search the next */
1212 mutex_unlock(&shmem_swaplist_mutex
);
1215 if (error
!= -ENOMEM
)
1217 mem_cgroup_cancel_charge(page
, memcg
, false);
1219 mem_cgroup_commit_charge(page
, memcg
, true, false);
1227 * Move the page from the page cache to the swap cache.
1229 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1231 struct shmem_inode_info
*info
;
1232 struct address_space
*mapping
;
1233 struct inode
*inode
;
1237 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1238 BUG_ON(!PageLocked(page
));
1239 mapping
= page
->mapping
;
1240 index
= page
->index
;
1241 inode
= mapping
->host
;
1242 info
= SHMEM_I(inode
);
1243 if (info
->flags
& VM_LOCKED
)
1245 if (!total_swap_pages
)
1249 * Our capabilities prevent regular writeback or sync from ever calling
1250 * shmem_writepage; but a stacking filesystem might use ->writepage of
1251 * its underlying filesystem, in which case tmpfs should write out to
1252 * swap only in response to memory pressure, and not for the writeback
1255 if (!wbc
->for_reclaim
) {
1256 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1261 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1262 * value into swapfile.c, the only way we can correctly account for a
1263 * fallocated page arriving here is now to initialize it and write it.
1265 * That's okay for a page already fallocated earlier, but if we have
1266 * not yet completed the fallocation, then (a) we want to keep track
1267 * of this page in case we have to undo it, and (b) it may not be a
1268 * good idea to continue anyway, once we're pushing into swap. So
1269 * reactivate the page, and let shmem_fallocate() quit when too many.
1271 if (!PageUptodate(page
)) {
1272 if (inode
->i_private
) {
1273 struct shmem_falloc
*shmem_falloc
;
1274 spin_lock(&inode
->i_lock
);
1275 shmem_falloc
= inode
->i_private
;
1277 !shmem_falloc
->waitq
&&
1278 index
>= shmem_falloc
->start
&&
1279 index
< shmem_falloc
->next
)
1280 shmem_falloc
->nr_unswapped
++;
1282 shmem_falloc
= NULL
;
1283 spin_unlock(&inode
->i_lock
);
1287 clear_highpage(page
);
1288 flush_dcache_page(page
);
1289 SetPageUptodate(page
);
1292 swap
= get_swap_page();
1296 if (mem_cgroup_try_charge_swap(page
, swap
))
1300 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1301 * if it's not already there. Do it now before the page is
1302 * moved to swap cache, when its pagelock no longer protects
1303 * the inode from eviction. But don't unlock the mutex until
1304 * we've incremented swapped, because shmem_unuse_inode() will
1305 * prune a !swapped inode from the swaplist under this mutex.
1307 mutex_lock(&shmem_swaplist_mutex
);
1308 if (list_empty(&info
->swaplist
))
1309 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1311 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1312 spin_lock_irq(&info
->lock
);
1313 shmem_recalc_inode(inode
);
1315 spin_unlock_irq(&info
->lock
);
1317 swap_shmem_alloc(swap
);
1318 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1320 mutex_unlock(&shmem_swaplist_mutex
);
1321 BUG_ON(page_mapped(page
));
1322 swap_writepage(page
, wbc
);
1326 mutex_unlock(&shmem_swaplist_mutex
);
1328 swapcache_free(swap
);
1330 set_page_dirty(page
);
1331 if (wbc
->for_reclaim
)
1332 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1337 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1338 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1342 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1343 return; /* show nothing */
1345 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1347 seq_printf(seq
, ",mpol=%s", buffer
);
1350 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1352 struct mempolicy
*mpol
= NULL
;
1354 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1355 mpol
= sbinfo
->mpol
;
1357 spin_unlock(&sbinfo
->stat_lock
);
1361 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1362 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1365 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1369 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1371 #define vm_policy vm_private_data
1374 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1375 struct shmem_inode_info
*info
, pgoff_t index
)
1377 /* Create a pseudo vma that just contains the policy */
1379 /* Bias interleave by inode number to distribute better across nodes */
1380 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1382 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1385 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1387 /* Drop reference taken by mpol_shared_policy_lookup() */
1388 mpol_cond_put(vma
->vm_policy
);
1391 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1392 struct shmem_inode_info
*info
, pgoff_t index
)
1394 struct vm_area_struct pvma
;
1397 shmem_pseudo_vma_init(&pvma
, info
, index
);
1398 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1399 shmem_pseudo_vma_destroy(&pvma
);
1404 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1405 struct shmem_inode_info
*info
, pgoff_t index
)
1407 struct vm_area_struct pvma
;
1408 struct inode
*inode
= &info
->vfs_inode
;
1409 struct address_space
*mapping
= inode
->i_mapping
;
1410 pgoff_t idx
, hindex
;
1411 void __rcu
**results
;
1414 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1417 hindex
= round_down(index
, HPAGE_PMD_NR
);
1419 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1420 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1426 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1427 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1428 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1429 shmem_pseudo_vma_destroy(&pvma
);
1431 prep_transhuge_page(page
);
1435 static struct page
*shmem_alloc_page(gfp_t gfp
,
1436 struct shmem_inode_info
*info
, pgoff_t index
)
1438 struct vm_area_struct pvma
;
1441 shmem_pseudo_vma_init(&pvma
, info
, index
);
1442 page
= alloc_page_vma(gfp
, &pvma
, 0);
1443 shmem_pseudo_vma_destroy(&pvma
);
1448 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1449 struct shmem_inode_info
*info
, struct shmem_sb_info
*sbinfo
,
1450 pgoff_t index
, bool huge
)
1456 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1458 nr
= huge
? HPAGE_PMD_NR
: 1;
1460 if (shmem_acct_block(info
->flags
, nr
))
1462 if (sbinfo
->max_blocks
) {
1463 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1464 sbinfo
->max_blocks
- nr
) > 0)
1466 percpu_counter_add(&sbinfo
->used_blocks
, nr
);
1470 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1472 page
= shmem_alloc_page(gfp
, info
, index
);
1474 __SetPageLocked(page
);
1475 __SetPageSwapBacked(page
);
1480 if (sbinfo
->max_blocks
)
1481 percpu_counter_add(&sbinfo
->used_blocks
, -nr
);
1483 shmem_unacct_blocks(info
->flags
, nr
);
1485 return ERR_PTR(err
);
1489 * When a page is moved from swapcache to shmem filecache (either by the
1490 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1491 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1492 * ignorance of the mapping it belongs to. If that mapping has special
1493 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1494 * we may need to copy to a suitable page before moving to filecache.
1496 * In a future release, this may well be extended to respect cpuset and
1497 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1498 * but for now it is a simple matter of zone.
1500 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1502 return page_zonenum(page
) > gfp_zone(gfp
);
1505 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1506 struct shmem_inode_info
*info
, pgoff_t index
)
1508 struct page
*oldpage
, *newpage
;
1509 struct address_space
*swap_mapping
;
1514 swap_index
= page_private(oldpage
);
1515 swap_mapping
= page_mapping(oldpage
);
1518 * We have arrived here because our zones are constrained, so don't
1519 * limit chance of success by further cpuset and node constraints.
1521 gfp
&= ~GFP_CONSTRAINT_MASK
;
1522 newpage
= shmem_alloc_page(gfp
, info
, index
);
1527 copy_highpage(newpage
, oldpage
);
1528 flush_dcache_page(newpage
);
1530 __SetPageLocked(newpage
);
1531 __SetPageSwapBacked(newpage
);
1532 SetPageUptodate(newpage
);
1533 set_page_private(newpage
, swap_index
);
1534 SetPageSwapCache(newpage
);
1537 * Our caller will very soon move newpage out of swapcache, but it's
1538 * a nice clean interface for us to replace oldpage by newpage there.
1540 spin_lock_irq(&swap_mapping
->tree_lock
);
1541 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1544 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1545 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1547 spin_unlock_irq(&swap_mapping
->tree_lock
);
1549 if (unlikely(error
)) {
1551 * Is this possible? I think not, now that our callers check
1552 * both PageSwapCache and page_private after getting page lock;
1553 * but be defensive. Reverse old to newpage for clear and free.
1557 mem_cgroup_migrate(oldpage
, newpage
);
1558 lru_cache_add_anon(newpage
);
1562 ClearPageSwapCache(oldpage
);
1563 set_page_private(oldpage
, 0);
1565 unlock_page(oldpage
);
1572 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1574 * If we allocate a new one we do not mark it dirty. That's up to the
1575 * vm. If we swap it in we mark it dirty since we also free the swap
1576 * entry since a page cannot live in both the swap and page cache.
1578 * fault_mm and fault_type are only supplied by shmem_fault:
1579 * otherwise they are NULL.
1581 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1582 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1583 struct vm_area_struct
*vma
, struct vm_fault
*vmf
, int *fault_type
)
1585 struct address_space
*mapping
= inode
->i_mapping
;
1586 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1587 struct shmem_sb_info
*sbinfo
;
1588 struct mm_struct
*charge_mm
;
1589 struct mem_cgroup
*memcg
;
1592 enum sgp_type sgp_huge
= sgp
;
1593 pgoff_t hindex
= index
;
1598 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1600 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1604 page
= find_lock_entry(mapping
, index
);
1605 if (radix_tree_exceptional_entry(page
)) {
1606 swap
= radix_to_swp_entry(page
);
1610 if (sgp
<= SGP_CACHE
&&
1611 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1616 if (page
&& sgp
== SGP_WRITE
)
1617 mark_page_accessed(page
);
1619 /* fallocated page? */
1620 if (page
&& !PageUptodate(page
)) {
1621 if (sgp
!= SGP_READ
)
1627 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1633 * Fast cache lookup did not find it:
1634 * bring it back from swap or allocate.
1636 sbinfo
= SHMEM_SB(inode
->i_sb
);
1637 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1640 /* Look it up and read it in.. */
1641 page
= lookup_swap_cache(swap
);
1643 /* Or update major stats only when swapin succeeds?? */
1645 *fault_type
|= VM_FAULT_MAJOR
;
1646 count_vm_event(PGMAJFAULT
);
1647 mem_cgroup_count_vm_event(charge_mm
,
1650 /* Here we actually start the io */
1651 page
= shmem_swapin(swap
, gfp
, info
, index
);
1658 /* We have to do this with page locked to prevent races */
1660 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1661 !shmem_confirm_swap(mapping
, index
, swap
)) {
1662 error
= -EEXIST
; /* try again */
1665 if (!PageUptodate(page
)) {
1669 wait_on_page_writeback(page
);
1671 if (shmem_should_replace_page(page
, gfp
)) {
1672 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1677 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1680 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1681 swp_to_radix_entry(swap
));
1683 * We already confirmed swap under page lock, and make
1684 * no memory allocation here, so usually no possibility
1685 * of error; but free_swap_and_cache() only trylocks a
1686 * page, so it is just possible that the entry has been
1687 * truncated or holepunched since swap was confirmed.
1688 * shmem_undo_range() will have done some of the
1689 * unaccounting, now delete_from_swap_cache() will do
1691 * Reset swap.val? No, leave it so "failed" goes back to
1692 * "repeat": reading a hole and writing should succeed.
1695 mem_cgroup_cancel_charge(page
, memcg
, false);
1696 delete_from_swap_cache(page
);
1702 mem_cgroup_commit_charge(page
, memcg
, true, false);
1704 spin_lock_irq(&info
->lock
);
1706 shmem_recalc_inode(inode
);
1707 spin_unlock_irq(&info
->lock
);
1709 if (sgp
== SGP_WRITE
)
1710 mark_page_accessed(page
);
1712 delete_from_swap_cache(page
);
1713 set_page_dirty(page
);
1717 if (vma
&& userfaultfd_missing(vma
)) {
1718 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1722 /* shmem_symlink() */
1723 if (mapping
->a_ops
!= &shmem_aops
)
1725 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1727 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1729 switch (sbinfo
->huge
) {
1732 case SHMEM_HUGE_NEVER
:
1734 case SHMEM_HUGE_WITHIN_SIZE
:
1735 off
= round_up(index
, HPAGE_PMD_NR
);
1736 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1737 if (i_size
>= HPAGE_PMD_SIZE
&&
1738 i_size
>> PAGE_SHIFT
>= off
)
1741 case SHMEM_HUGE_ADVISE
:
1742 if (sgp_huge
== SGP_HUGE
)
1744 /* TODO: implement fadvise() hints */
1749 page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1752 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1757 error
= PTR_ERR(page
);
1759 if (error
!= -ENOSPC
)
1762 * Try to reclaim some spece by splitting a huge page
1763 * beyond i_size on the filesystem.
1767 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1768 if (ret
== SHRINK_STOP
)
1776 if (PageTransHuge(page
))
1777 hindex
= round_down(index
, HPAGE_PMD_NR
);
1781 if (sgp
== SGP_WRITE
)
1782 __SetPageReferenced(page
);
1784 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1785 PageTransHuge(page
));
1788 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1789 compound_order(page
));
1791 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1793 radix_tree_preload_end();
1796 mem_cgroup_cancel_charge(page
, memcg
,
1797 PageTransHuge(page
));
1800 mem_cgroup_commit_charge(page
, memcg
, false,
1801 PageTransHuge(page
));
1802 lru_cache_add_anon(page
);
1804 spin_lock_irq(&info
->lock
);
1805 info
->alloced
+= 1 << compound_order(page
);
1806 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1807 shmem_recalc_inode(inode
);
1808 spin_unlock_irq(&info
->lock
);
1811 if (PageTransHuge(page
) &&
1812 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1813 hindex
+ HPAGE_PMD_NR
- 1) {
1815 * Part of the huge page is beyond i_size: subject
1816 * to shrink under memory pressure.
1818 spin_lock(&sbinfo
->shrinklist_lock
);
1819 if (list_empty(&info
->shrinklist
)) {
1820 list_add_tail(&info
->shrinklist
,
1821 &sbinfo
->shrinklist
);
1822 sbinfo
->shrinklist_len
++;
1824 spin_unlock(&sbinfo
->shrinklist_lock
);
1828 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1830 if (sgp
== SGP_FALLOC
)
1834 * Let SGP_WRITE caller clear ends if write does not fill page;
1835 * but SGP_FALLOC on a page fallocated earlier must initialize
1836 * it now, lest undo on failure cancel our earlier guarantee.
1838 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1839 struct page
*head
= compound_head(page
);
1842 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1843 clear_highpage(head
+ i
);
1844 flush_dcache_page(head
+ i
);
1846 SetPageUptodate(head
);
1850 /* Perhaps the file has been truncated since we checked */
1851 if (sgp
<= SGP_CACHE
&&
1852 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1854 ClearPageDirty(page
);
1855 delete_from_page_cache(page
);
1856 spin_lock_irq(&info
->lock
);
1857 shmem_recalc_inode(inode
);
1858 spin_unlock_irq(&info
->lock
);
1863 *pagep
= page
+ index
- hindex
;
1870 if (sbinfo
->max_blocks
)
1871 percpu_counter_sub(&sbinfo
->used_blocks
,
1872 1 << compound_order(page
));
1873 shmem_unacct_blocks(info
->flags
, 1 << compound_order(page
));
1875 if (PageTransHuge(page
)) {
1881 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1888 if (error
== -ENOSPC
&& !once
++) {
1889 spin_lock_irq(&info
->lock
);
1890 shmem_recalc_inode(inode
);
1891 spin_unlock_irq(&info
->lock
);
1894 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1900 * This is like autoremove_wake_function, but it removes the wait queue
1901 * entry unconditionally - even if something else had already woken the
1904 static int synchronous_wake_function(wait_queue_t
*wait
, unsigned mode
, int sync
, void *key
)
1906 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1907 list_del_init(&wait
->task_list
);
1911 static int shmem_fault(struct vm_fault
*vmf
)
1913 struct vm_area_struct
*vma
= vmf
->vma
;
1914 struct inode
*inode
= file_inode(vma
->vm_file
);
1915 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1918 int ret
= VM_FAULT_LOCKED
;
1921 * Trinity finds that probing a hole which tmpfs is punching can
1922 * prevent the hole-punch from ever completing: which in turn
1923 * locks writers out with its hold on i_mutex. So refrain from
1924 * faulting pages into the hole while it's being punched. Although
1925 * shmem_undo_range() does remove the additions, it may be unable to
1926 * keep up, as each new page needs its own unmap_mapping_range() call,
1927 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1929 * It does not matter if we sometimes reach this check just before the
1930 * hole-punch begins, so that one fault then races with the punch:
1931 * we just need to make racing faults a rare case.
1933 * The implementation below would be much simpler if we just used a
1934 * standard mutex or completion: but we cannot take i_mutex in fault,
1935 * and bloating every shmem inode for this unlikely case would be sad.
1937 if (unlikely(inode
->i_private
)) {
1938 struct shmem_falloc
*shmem_falloc
;
1940 spin_lock(&inode
->i_lock
);
1941 shmem_falloc
= inode
->i_private
;
1943 shmem_falloc
->waitq
&&
1944 vmf
->pgoff
>= shmem_falloc
->start
&&
1945 vmf
->pgoff
< shmem_falloc
->next
) {
1946 wait_queue_head_t
*shmem_falloc_waitq
;
1947 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
1949 ret
= VM_FAULT_NOPAGE
;
1950 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1951 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1952 /* It's polite to up mmap_sem if we can */
1953 up_read(&vma
->vm_mm
->mmap_sem
);
1954 ret
= VM_FAULT_RETRY
;
1957 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1958 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1959 TASK_UNINTERRUPTIBLE
);
1960 spin_unlock(&inode
->i_lock
);
1964 * shmem_falloc_waitq points into the shmem_fallocate()
1965 * stack of the hole-punching task: shmem_falloc_waitq
1966 * is usually invalid by the time we reach here, but
1967 * finish_wait() does not dereference it in that case;
1968 * though i_lock needed lest racing with wake_up_all().
1970 spin_lock(&inode
->i_lock
);
1971 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1972 spin_unlock(&inode
->i_lock
);
1975 spin_unlock(&inode
->i_lock
);
1979 if (vma
->vm_flags
& VM_HUGEPAGE
)
1981 else if (vma
->vm_flags
& VM_NOHUGEPAGE
)
1984 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
1985 gfp
, vma
, vmf
, &ret
);
1987 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1991 unsigned long shmem_get_unmapped_area(struct file
*file
,
1992 unsigned long uaddr
, unsigned long len
,
1993 unsigned long pgoff
, unsigned long flags
)
1995 unsigned long (*get_area
)(struct file
*,
1996 unsigned long, unsigned long, unsigned long, unsigned long);
1998 unsigned long offset
;
1999 unsigned long inflated_len
;
2000 unsigned long inflated_addr
;
2001 unsigned long inflated_offset
;
2003 if (len
> TASK_SIZE
)
2006 get_area
= current
->mm
->get_unmapped_area
;
2007 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2009 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
2011 if (IS_ERR_VALUE(addr
))
2013 if (addr
& ~PAGE_MASK
)
2015 if (addr
> TASK_SIZE
- len
)
2018 if (shmem_huge
== SHMEM_HUGE_DENY
)
2020 if (len
< HPAGE_PMD_SIZE
)
2022 if (flags
& MAP_FIXED
)
2025 * Our priority is to support MAP_SHARED mapped hugely;
2026 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2027 * But if caller specified an address hint, respect that as before.
2032 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2033 struct super_block
*sb
;
2036 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2037 sb
= file_inode(file
)->i_sb
;
2040 * Called directly from mm/mmap.c, or drivers/char/mem.c
2041 * for "/dev/zero", to create a shared anonymous object.
2043 if (IS_ERR(shm_mnt
))
2045 sb
= shm_mnt
->mnt_sb
;
2047 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2051 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2052 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2054 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2057 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2058 if (inflated_len
> TASK_SIZE
)
2060 if (inflated_len
< len
)
2063 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
2064 if (IS_ERR_VALUE(inflated_addr
))
2066 if (inflated_addr
& ~PAGE_MASK
)
2069 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2070 inflated_addr
+= offset
- inflated_offset
;
2071 if (inflated_offset
> offset
)
2072 inflated_addr
+= HPAGE_PMD_SIZE
;
2074 if (inflated_addr
> TASK_SIZE
- len
)
2076 return inflated_addr
;
2080 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2082 struct inode
*inode
= file_inode(vma
->vm_file
);
2083 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2086 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2089 struct inode
*inode
= file_inode(vma
->vm_file
);
2092 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2093 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2097 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2099 struct inode
*inode
= file_inode(file
);
2100 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2101 int retval
= -ENOMEM
;
2103 spin_lock_irq(&info
->lock
);
2104 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2105 if (!user_shm_lock(inode
->i_size
, user
))
2107 info
->flags
|= VM_LOCKED
;
2108 mapping_set_unevictable(file
->f_mapping
);
2110 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2111 user_shm_unlock(inode
->i_size
, user
);
2112 info
->flags
&= ~VM_LOCKED
;
2113 mapping_clear_unevictable(file
->f_mapping
);
2118 spin_unlock_irq(&info
->lock
);
2122 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2124 file_accessed(file
);
2125 vma
->vm_ops
= &shmem_vm_ops
;
2126 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2127 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2128 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2129 khugepaged_enter(vma
, vma
->vm_flags
);
2134 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2135 umode_t mode
, dev_t dev
, unsigned long flags
)
2137 struct inode
*inode
;
2138 struct shmem_inode_info
*info
;
2139 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2141 if (shmem_reserve_inode(sb
))
2144 inode
= new_inode(sb
);
2146 inode
->i_ino
= get_next_ino();
2147 inode_init_owner(inode
, dir
, mode
);
2148 inode
->i_blocks
= 0;
2149 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2150 inode
->i_generation
= get_seconds();
2151 info
= SHMEM_I(inode
);
2152 memset(info
, 0, (char *)inode
- (char *)info
);
2153 spin_lock_init(&info
->lock
);
2154 info
->seals
= F_SEAL_SEAL
;
2155 info
->flags
= flags
& VM_NORESERVE
;
2156 INIT_LIST_HEAD(&info
->shrinklist
);
2157 INIT_LIST_HEAD(&info
->swaplist
);
2158 simple_xattrs_init(&info
->xattrs
);
2159 cache_no_acl(inode
);
2161 switch (mode
& S_IFMT
) {
2163 inode
->i_op
= &shmem_special_inode_operations
;
2164 init_special_inode(inode
, mode
, dev
);
2167 inode
->i_mapping
->a_ops
= &shmem_aops
;
2168 inode
->i_op
= &shmem_inode_operations
;
2169 inode
->i_fop
= &shmem_file_operations
;
2170 mpol_shared_policy_init(&info
->policy
,
2171 shmem_get_sbmpol(sbinfo
));
2175 /* Some things misbehave if size == 0 on a directory */
2176 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2177 inode
->i_op
= &shmem_dir_inode_operations
;
2178 inode
->i_fop
= &simple_dir_operations
;
2182 * Must not load anything in the rbtree,
2183 * mpol_free_shared_policy will not be called.
2185 mpol_shared_policy_init(&info
->policy
, NULL
);
2189 shmem_free_inode(sb
);
2193 bool shmem_mapping(struct address_space
*mapping
)
2195 return mapping
->a_ops
== &shmem_aops
;
2198 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2200 struct vm_area_struct
*dst_vma
,
2201 unsigned long dst_addr
,
2202 unsigned long src_addr
,
2203 struct page
**pagep
)
2205 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2206 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2207 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2208 struct address_space
*mapping
= inode
->i_mapping
;
2209 gfp_t gfp
= mapping_gfp_mask(mapping
);
2210 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2211 struct mem_cgroup
*memcg
;
2215 pte_t _dst_pte
, *dst_pte
;
2219 if (shmem_acct_block(info
->flags
, 1))
2221 if (sbinfo
->max_blocks
) {
2222 if (percpu_counter_compare(&sbinfo
->used_blocks
,
2223 sbinfo
->max_blocks
) >= 0)
2224 goto out_unacct_blocks
;
2225 percpu_counter_inc(&sbinfo
->used_blocks
);
2229 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2231 goto out_dec_used_blocks
;
2233 page_kaddr
= kmap_atomic(page
);
2234 ret
= copy_from_user(page_kaddr
, (const void __user
*)src_addr
,
2236 kunmap_atomic(page_kaddr
);
2238 /* fallback to copy_from_user outside mmap_sem */
2239 if (unlikely(ret
)) {
2241 if (sbinfo
->max_blocks
)
2242 percpu_counter_add(&sbinfo
->used_blocks
, -1);
2243 shmem_unacct_blocks(info
->flags
, 1);
2244 /* don't free the page */
2252 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2253 __SetPageLocked(page
);
2254 __SetPageSwapBacked(page
);
2255 __SetPageUptodate(page
);
2257 ret
= mem_cgroup_try_charge(page
, dst_mm
, gfp
, &memcg
, false);
2261 ret
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
2263 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
);
2264 radix_tree_preload_end();
2267 goto out_release_uncharge
;
2269 mem_cgroup_commit_charge(page
, memcg
, false, false);
2271 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2272 if (dst_vma
->vm_flags
& VM_WRITE
)
2273 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2276 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2277 if (!pte_none(*dst_pte
))
2278 goto out_release_uncharge_unlock
;
2280 lru_cache_add_anon(page
);
2282 spin_lock(&info
->lock
);
2284 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2285 shmem_recalc_inode(inode
);
2286 spin_unlock(&info
->lock
);
2288 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2289 page_add_file_rmap(page
, false);
2290 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2292 /* No need to invalidate - it was non-present before */
2293 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2295 pte_unmap_unlock(dst_pte
, ptl
);
2299 out_release_uncharge_unlock
:
2300 pte_unmap_unlock(dst_pte
, ptl
);
2301 out_release_uncharge
:
2302 mem_cgroup_cancel_charge(page
, memcg
, false);
2306 out_dec_used_blocks
:
2307 if (sbinfo
->max_blocks
)
2308 percpu_counter_add(&sbinfo
->used_blocks
, -1);
2310 shmem_unacct_blocks(info
->flags
, 1);
2315 static const struct inode_operations shmem_symlink_inode_operations
;
2316 static const struct inode_operations shmem_short_symlink_operations
;
2318 #ifdef CONFIG_TMPFS_XATTR
2319 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2321 #define shmem_initxattrs NULL
2325 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2326 loff_t pos
, unsigned len
, unsigned flags
,
2327 struct page
**pagep
, void **fsdata
)
2329 struct inode
*inode
= mapping
->host
;
2330 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2331 pgoff_t index
= pos
>> PAGE_SHIFT
;
2333 /* i_mutex is held by caller */
2334 if (unlikely(info
->seals
)) {
2335 if (info
->seals
& F_SEAL_WRITE
)
2337 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2341 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2345 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2346 loff_t pos
, unsigned len
, unsigned copied
,
2347 struct page
*page
, void *fsdata
)
2349 struct inode
*inode
= mapping
->host
;
2351 if (pos
+ copied
> inode
->i_size
)
2352 i_size_write(inode
, pos
+ copied
);
2354 if (!PageUptodate(page
)) {
2355 struct page
*head
= compound_head(page
);
2356 if (PageTransCompound(page
)) {
2359 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2360 if (head
+ i
== page
)
2362 clear_highpage(head
+ i
);
2363 flush_dcache_page(head
+ i
);
2366 if (copied
< PAGE_SIZE
) {
2367 unsigned from
= pos
& (PAGE_SIZE
- 1);
2368 zero_user_segments(page
, 0, from
,
2369 from
+ copied
, PAGE_SIZE
);
2371 SetPageUptodate(head
);
2373 set_page_dirty(page
);
2380 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2382 struct file
*file
= iocb
->ki_filp
;
2383 struct inode
*inode
= file_inode(file
);
2384 struct address_space
*mapping
= inode
->i_mapping
;
2386 unsigned long offset
;
2387 enum sgp_type sgp
= SGP_READ
;
2390 loff_t
*ppos
= &iocb
->ki_pos
;
2393 * Might this read be for a stacking filesystem? Then when reading
2394 * holes of a sparse file, we actually need to allocate those pages,
2395 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2397 if (!iter_is_iovec(to
))
2400 index
= *ppos
>> PAGE_SHIFT
;
2401 offset
= *ppos
& ~PAGE_MASK
;
2404 struct page
*page
= NULL
;
2406 unsigned long nr
, ret
;
2407 loff_t i_size
= i_size_read(inode
);
2409 end_index
= i_size
>> PAGE_SHIFT
;
2410 if (index
> end_index
)
2412 if (index
== end_index
) {
2413 nr
= i_size
& ~PAGE_MASK
;
2418 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2420 if (error
== -EINVAL
)
2425 if (sgp
== SGP_CACHE
)
2426 set_page_dirty(page
);
2431 * We must evaluate after, since reads (unlike writes)
2432 * are called without i_mutex protection against truncate
2435 i_size
= i_size_read(inode
);
2436 end_index
= i_size
>> PAGE_SHIFT
;
2437 if (index
== end_index
) {
2438 nr
= i_size
& ~PAGE_MASK
;
2449 * If users can be writing to this page using arbitrary
2450 * virtual addresses, take care about potential aliasing
2451 * before reading the page on the kernel side.
2453 if (mapping_writably_mapped(mapping
))
2454 flush_dcache_page(page
);
2456 * Mark the page accessed if we read the beginning.
2459 mark_page_accessed(page
);
2461 page
= ZERO_PAGE(0);
2466 * Ok, we have the page, and it's up-to-date, so
2467 * now we can copy it to user space...
2469 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2472 index
+= offset
>> PAGE_SHIFT
;
2473 offset
&= ~PAGE_MASK
;
2476 if (!iov_iter_count(to
))
2485 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2486 file_accessed(file
);
2487 return retval
? retval
: error
;
2491 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2493 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2494 pgoff_t index
, pgoff_t end
, int whence
)
2497 struct pagevec pvec
;
2498 pgoff_t indices
[PAGEVEC_SIZE
];
2502 pagevec_init(&pvec
, 0);
2503 pvec
.nr
= 1; /* start small: we may be there already */
2505 pvec
.nr
= find_get_entries(mapping
, index
,
2506 pvec
.nr
, pvec
.pages
, indices
);
2508 if (whence
== SEEK_DATA
)
2512 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2513 if (index
< indices
[i
]) {
2514 if (whence
== SEEK_HOLE
) {
2520 page
= pvec
.pages
[i
];
2521 if (page
&& !radix_tree_exceptional_entry(page
)) {
2522 if (!PageUptodate(page
))
2526 (page
&& whence
== SEEK_DATA
) ||
2527 (!page
&& whence
== SEEK_HOLE
)) {
2532 pagevec_remove_exceptionals(&pvec
);
2533 pagevec_release(&pvec
);
2534 pvec
.nr
= PAGEVEC_SIZE
;
2540 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2542 struct address_space
*mapping
= file
->f_mapping
;
2543 struct inode
*inode
= mapping
->host
;
2547 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2548 return generic_file_llseek_size(file
, offset
, whence
,
2549 MAX_LFS_FILESIZE
, i_size_read(inode
));
2551 /* We're holding i_mutex so we can access i_size directly */
2555 else if (offset
>= inode
->i_size
)
2558 start
= offset
>> PAGE_SHIFT
;
2559 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2560 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2561 new_offset
<<= PAGE_SHIFT
;
2562 if (new_offset
> offset
) {
2563 if (new_offset
< inode
->i_size
)
2564 offset
= new_offset
;
2565 else if (whence
== SEEK_DATA
)
2568 offset
= inode
->i_size
;
2573 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2574 inode_unlock(inode
);
2579 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2580 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2582 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2583 #define LAST_SCAN 4 /* about 150ms max */
2585 static void shmem_tag_pins(struct address_space
*mapping
)
2587 struct radix_tree_iter iter
;
2596 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2597 page
= radix_tree_deref_slot(slot
);
2598 if (!page
|| radix_tree_exception(page
)) {
2599 if (radix_tree_deref_retry(page
)) {
2600 slot
= radix_tree_iter_retry(&iter
);
2603 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2604 spin_lock_irq(&mapping
->tree_lock
);
2605 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2607 spin_unlock_irq(&mapping
->tree_lock
);
2610 if (need_resched()) {
2611 slot
= radix_tree_iter_resume(slot
, &iter
);
2619 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2620 * via get_user_pages(), drivers might have some pending I/O without any active
2621 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2622 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2623 * them to be dropped.
2624 * The caller must guarantee that no new user will acquire writable references
2625 * to those pages to avoid races.
2627 static int shmem_wait_for_pins(struct address_space
*mapping
)
2629 struct radix_tree_iter iter
;
2635 shmem_tag_pins(mapping
);
2638 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2639 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2643 lru_add_drain_all();
2644 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2649 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2650 start
, SHMEM_TAG_PINNED
) {
2652 page
= radix_tree_deref_slot(slot
);
2653 if (radix_tree_exception(page
)) {
2654 if (radix_tree_deref_retry(page
)) {
2655 slot
= radix_tree_iter_retry(&iter
);
2663 page_count(page
) - page_mapcount(page
) != 1) {
2664 if (scan
< LAST_SCAN
)
2665 goto continue_resched
;
2668 * On the last scan, we clean up all those tags
2669 * we inserted; but make a note that we still
2670 * found pages pinned.
2675 spin_lock_irq(&mapping
->tree_lock
);
2676 radix_tree_tag_clear(&mapping
->page_tree
,
2677 iter
.index
, SHMEM_TAG_PINNED
);
2678 spin_unlock_irq(&mapping
->tree_lock
);
2680 if (need_resched()) {
2681 slot
= radix_tree_iter_resume(slot
, &iter
);
2691 #define F_ALL_SEALS (F_SEAL_SEAL | \
2696 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2698 struct inode
*inode
= file_inode(file
);
2699 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2704 * Sealing allows multiple parties to share a shmem-file but restrict
2705 * access to a specific subset of file operations. Seals can only be
2706 * added, but never removed. This way, mutually untrusted parties can
2707 * share common memory regions with a well-defined policy. A malicious
2708 * peer can thus never perform unwanted operations on a shared object.
2710 * Seals are only supported on special shmem-files and always affect
2711 * the whole underlying inode. Once a seal is set, it may prevent some
2712 * kinds of access to the file. Currently, the following seals are
2714 * SEAL_SEAL: Prevent further seals from being set on this file
2715 * SEAL_SHRINK: Prevent the file from shrinking
2716 * SEAL_GROW: Prevent the file from growing
2717 * SEAL_WRITE: Prevent write access to the file
2719 * As we don't require any trust relationship between two parties, we
2720 * must prevent seals from being removed. Therefore, sealing a file
2721 * only adds a given set of seals to the file, it never touches
2722 * existing seals. Furthermore, the "setting seals"-operation can be
2723 * sealed itself, which basically prevents any further seal from being
2726 * Semantics of sealing are only defined on volatile files. Only
2727 * anonymous shmem files support sealing. More importantly, seals are
2728 * never written to disk. Therefore, there's no plan to support it on
2732 if (file
->f_op
!= &shmem_file_operations
)
2734 if (!(file
->f_mode
& FMODE_WRITE
))
2736 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2741 if (info
->seals
& F_SEAL_SEAL
) {
2746 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2747 error
= mapping_deny_writable(file
->f_mapping
);
2751 error
= shmem_wait_for_pins(file
->f_mapping
);
2753 mapping_allow_writable(file
->f_mapping
);
2758 info
->seals
|= seals
;
2762 inode_unlock(inode
);
2765 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2767 int shmem_get_seals(struct file
*file
)
2769 if (file
->f_op
!= &shmem_file_operations
)
2772 return SHMEM_I(file_inode(file
))->seals
;
2774 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2776 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2782 /* disallow upper 32bit */
2786 error
= shmem_add_seals(file
, arg
);
2789 error
= shmem_get_seals(file
);
2799 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2802 struct inode
*inode
= file_inode(file
);
2803 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2804 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2805 struct shmem_falloc shmem_falloc
;
2806 pgoff_t start
, index
, end
;
2809 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2814 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2815 struct address_space
*mapping
= file
->f_mapping
;
2816 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2817 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2818 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2820 /* protected by i_mutex */
2821 if (info
->seals
& F_SEAL_WRITE
) {
2826 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2827 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2828 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2829 spin_lock(&inode
->i_lock
);
2830 inode
->i_private
= &shmem_falloc
;
2831 spin_unlock(&inode
->i_lock
);
2833 if ((u64
)unmap_end
> (u64
)unmap_start
)
2834 unmap_mapping_range(mapping
, unmap_start
,
2835 1 + unmap_end
- unmap_start
, 0);
2836 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2837 /* No need to unmap again: hole-punching leaves COWed pages */
2839 spin_lock(&inode
->i_lock
);
2840 inode
->i_private
= NULL
;
2841 wake_up_all(&shmem_falloc_waitq
);
2842 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.task_list
));
2843 spin_unlock(&inode
->i_lock
);
2848 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2849 error
= inode_newsize_ok(inode
, offset
+ len
);
2853 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2858 start
= offset
>> PAGE_SHIFT
;
2859 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2860 /* Try to avoid a swapstorm if len is impossible to satisfy */
2861 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2866 shmem_falloc
.waitq
= NULL
;
2867 shmem_falloc
.start
= start
;
2868 shmem_falloc
.next
= start
;
2869 shmem_falloc
.nr_falloced
= 0;
2870 shmem_falloc
.nr_unswapped
= 0;
2871 spin_lock(&inode
->i_lock
);
2872 inode
->i_private
= &shmem_falloc
;
2873 spin_unlock(&inode
->i_lock
);
2875 for (index
= start
; index
< end
; index
++) {
2879 * Good, the fallocate(2) manpage permits EINTR: we may have
2880 * been interrupted because we are using up too much memory.
2882 if (signal_pending(current
))
2884 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2887 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2889 /* Remove the !PageUptodate pages we added */
2890 if (index
> start
) {
2891 shmem_undo_range(inode
,
2892 (loff_t
)start
<< PAGE_SHIFT
,
2893 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2899 * Inform shmem_writepage() how far we have reached.
2900 * No need for lock or barrier: we have the page lock.
2902 shmem_falloc
.next
++;
2903 if (!PageUptodate(page
))
2904 shmem_falloc
.nr_falloced
++;
2907 * If !PageUptodate, leave it that way so that freeable pages
2908 * can be recognized if we need to rollback on error later.
2909 * But set_page_dirty so that memory pressure will swap rather
2910 * than free the pages we are allocating (and SGP_CACHE pages
2911 * might still be clean: we now need to mark those dirty too).
2913 set_page_dirty(page
);
2919 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2920 i_size_write(inode
, offset
+ len
);
2921 inode
->i_ctime
= current_time(inode
);
2923 spin_lock(&inode
->i_lock
);
2924 inode
->i_private
= NULL
;
2925 spin_unlock(&inode
->i_lock
);
2927 inode_unlock(inode
);
2931 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2933 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2935 buf
->f_type
= TMPFS_MAGIC
;
2936 buf
->f_bsize
= PAGE_SIZE
;
2937 buf
->f_namelen
= NAME_MAX
;
2938 if (sbinfo
->max_blocks
) {
2939 buf
->f_blocks
= sbinfo
->max_blocks
;
2941 buf
->f_bfree
= sbinfo
->max_blocks
-
2942 percpu_counter_sum(&sbinfo
->used_blocks
);
2944 if (sbinfo
->max_inodes
) {
2945 buf
->f_files
= sbinfo
->max_inodes
;
2946 buf
->f_ffree
= sbinfo
->free_inodes
;
2948 /* else leave those fields 0 like simple_statfs */
2953 * File creation. Allocate an inode, and we're done..
2956 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2958 struct inode
*inode
;
2959 int error
= -ENOSPC
;
2961 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2963 error
= simple_acl_create(dir
, inode
);
2966 error
= security_inode_init_security(inode
, dir
,
2968 shmem_initxattrs
, NULL
);
2969 if (error
&& error
!= -EOPNOTSUPP
)
2973 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2974 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2975 d_instantiate(dentry
, inode
);
2976 dget(dentry
); /* Extra count - pin the dentry in core */
2985 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2987 struct inode
*inode
;
2988 int error
= -ENOSPC
;
2990 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2992 error
= security_inode_init_security(inode
, dir
,
2994 shmem_initxattrs
, NULL
);
2995 if (error
&& error
!= -EOPNOTSUPP
)
2997 error
= simple_acl_create(dir
, inode
);
3000 d_tmpfile(dentry
, inode
);
3008 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3012 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
3018 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
3021 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
3027 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
3029 struct inode
*inode
= d_inode(old_dentry
);
3033 * No ordinary (disk based) filesystem counts links as inodes;
3034 * but each new link needs a new dentry, pinning lowmem, and
3035 * tmpfs dentries cannot be pruned until they are unlinked.
3037 ret
= shmem_reserve_inode(inode
->i_sb
);
3041 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3042 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3044 ihold(inode
); /* New dentry reference */
3045 dget(dentry
); /* Extra pinning count for the created dentry */
3046 d_instantiate(dentry
, inode
);
3051 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3053 struct inode
*inode
= d_inode(dentry
);
3055 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3056 shmem_free_inode(inode
->i_sb
);
3058 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3059 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3061 dput(dentry
); /* Undo the count from "create" - this does all the work */
3065 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3067 if (!simple_empty(dentry
))
3070 drop_nlink(d_inode(dentry
));
3072 return shmem_unlink(dir
, dentry
);
3075 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3077 bool old_is_dir
= d_is_dir(old_dentry
);
3078 bool new_is_dir
= d_is_dir(new_dentry
);
3080 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3082 drop_nlink(old_dir
);
3085 drop_nlink(new_dir
);
3089 old_dir
->i_ctime
= old_dir
->i_mtime
=
3090 new_dir
->i_ctime
= new_dir
->i_mtime
=
3091 d_inode(old_dentry
)->i_ctime
=
3092 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3097 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3099 struct dentry
*whiteout
;
3102 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3106 error
= shmem_mknod(old_dir
, whiteout
,
3107 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3113 * Cheat and hash the whiteout while the old dentry is still in
3114 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3116 * d_lookup() will consistently find one of them at this point,
3117 * not sure which one, but that isn't even important.
3124 * The VFS layer already does all the dentry stuff for rename,
3125 * we just have to decrement the usage count for the target if
3126 * it exists so that the VFS layer correctly free's it when it
3129 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3131 struct inode
*inode
= d_inode(old_dentry
);
3132 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3134 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3137 if (flags
& RENAME_EXCHANGE
)
3138 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3140 if (!simple_empty(new_dentry
))
3143 if (flags
& RENAME_WHITEOUT
) {
3146 error
= shmem_whiteout(old_dir
, old_dentry
);
3151 if (d_really_is_positive(new_dentry
)) {
3152 (void) shmem_unlink(new_dir
, new_dentry
);
3153 if (they_are_dirs
) {
3154 drop_nlink(d_inode(new_dentry
));
3155 drop_nlink(old_dir
);
3157 } else if (they_are_dirs
) {
3158 drop_nlink(old_dir
);
3162 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3163 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3164 old_dir
->i_ctime
= old_dir
->i_mtime
=
3165 new_dir
->i_ctime
= new_dir
->i_mtime
=
3166 inode
->i_ctime
= current_time(old_dir
);
3170 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3174 struct inode
*inode
;
3176 struct shmem_inode_info
*info
;
3178 len
= strlen(symname
) + 1;
3179 if (len
> PAGE_SIZE
)
3180 return -ENAMETOOLONG
;
3182 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
3186 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3187 shmem_initxattrs
, NULL
);
3189 if (error
!= -EOPNOTSUPP
) {
3196 info
= SHMEM_I(inode
);
3197 inode
->i_size
= len
-1;
3198 if (len
<= SHORT_SYMLINK_LEN
) {
3199 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3200 if (!inode
->i_link
) {
3204 inode
->i_op
= &shmem_short_symlink_operations
;
3206 inode_nohighmem(inode
);
3207 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3212 inode
->i_mapping
->a_ops
= &shmem_aops
;
3213 inode
->i_op
= &shmem_symlink_inode_operations
;
3214 memcpy(page_address(page
), symname
, len
);
3215 SetPageUptodate(page
);
3216 set_page_dirty(page
);
3220 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3221 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3222 d_instantiate(dentry
, inode
);
3227 static void shmem_put_link(void *arg
)
3229 mark_page_accessed(arg
);
3233 static const char *shmem_get_link(struct dentry
*dentry
,
3234 struct inode
*inode
,
3235 struct delayed_call
*done
)
3237 struct page
*page
= NULL
;
3240 page
= find_get_page(inode
->i_mapping
, 0);
3242 return ERR_PTR(-ECHILD
);
3243 if (!PageUptodate(page
)) {
3245 return ERR_PTR(-ECHILD
);
3248 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3250 return ERR_PTR(error
);
3253 set_delayed_call(done
, shmem_put_link
, page
);
3254 return page_address(page
);
3257 #ifdef CONFIG_TMPFS_XATTR
3259 * Superblocks without xattr inode operations may get some security.* xattr
3260 * support from the LSM "for free". As soon as we have any other xattrs
3261 * like ACLs, we also need to implement the security.* handlers at
3262 * filesystem level, though.
3266 * Callback for security_inode_init_security() for acquiring xattrs.
3268 static int shmem_initxattrs(struct inode
*inode
,
3269 const struct xattr
*xattr_array
,
3272 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3273 const struct xattr
*xattr
;
3274 struct simple_xattr
*new_xattr
;
3277 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3278 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3282 len
= strlen(xattr
->name
) + 1;
3283 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3285 if (!new_xattr
->name
) {
3290 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3291 XATTR_SECURITY_PREFIX_LEN
);
3292 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3295 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3301 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3302 struct dentry
*unused
, struct inode
*inode
,
3303 const char *name
, void *buffer
, size_t size
)
3305 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3307 name
= xattr_full_name(handler
, name
);
3308 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3311 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3312 struct dentry
*unused
, struct inode
*inode
,
3313 const char *name
, const void *value
,
3314 size_t size
, int flags
)
3316 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3318 name
= xattr_full_name(handler
, name
);
3319 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3322 static const struct xattr_handler shmem_security_xattr_handler
= {
3323 .prefix
= XATTR_SECURITY_PREFIX
,
3324 .get
= shmem_xattr_handler_get
,
3325 .set
= shmem_xattr_handler_set
,
3328 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3329 .prefix
= XATTR_TRUSTED_PREFIX
,
3330 .get
= shmem_xattr_handler_get
,
3331 .set
= shmem_xattr_handler_set
,
3334 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3335 #ifdef CONFIG_TMPFS_POSIX_ACL
3336 &posix_acl_access_xattr_handler
,
3337 &posix_acl_default_xattr_handler
,
3339 &shmem_security_xattr_handler
,
3340 &shmem_trusted_xattr_handler
,
3344 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3346 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3347 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3349 #endif /* CONFIG_TMPFS_XATTR */
3351 static const struct inode_operations shmem_short_symlink_operations
= {
3352 .get_link
= simple_get_link
,
3353 #ifdef CONFIG_TMPFS_XATTR
3354 .listxattr
= shmem_listxattr
,
3358 static const struct inode_operations shmem_symlink_inode_operations
= {
3359 .get_link
= shmem_get_link
,
3360 #ifdef CONFIG_TMPFS_XATTR
3361 .listxattr
= shmem_listxattr
,
3365 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3367 return ERR_PTR(-ESTALE
);
3370 static int shmem_match(struct inode
*ino
, void *vfh
)
3374 inum
= (inum
<< 32) | fh
[1];
3375 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3378 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3379 struct fid
*fid
, int fh_len
, int fh_type
)
3381 struct inode
*inode
;
3382 struct dentry
*dentry
= NULL
;
3389 inum
= (inum
<< 32) | fid
->raw
[1];
3391 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3392 shmem_match
, fid
->raw
);
3394 dentry
= d_find_alias(inode
);
3401 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3402 struct inode
*parent
)
3406 return FILEID_INVALID
;
3409 if (inode_unhashed(inode
)) {
3410 /* Unfortunately insert_inode_hash is not idempotent,
3411 * so as we hash inodes here rather than at creation
3412 * time, we need a lock to ensure we only try
3415 static DEFINE_SPINLOCK(lock
);
3417 if (inode_unhashed(inode
))
3418 __insert_inode_hash(inode
,
3419 inode
->i_ino
+ inode
->i_generation
);
3423 fh
[0] = inode
->i_generation
;
3424 fh
[1] = inode
->i_ino
;
3425 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3431 static const struct export_operations shmem_export_ops
= {
3432 .get_parent
= shmem_get_parent
,
3433 .encode_fh
= shmem_encode_fh
,
3434 .fh_to_dentry
= shmem_fh_to_dentry
,
3437 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3440 char *this_char
, *value
, *rest
;
3441 struct mempolicy
*mpol
= NULL
;
3445 while (options
!= NULL
) {
3446 this_char
= options
;
3449 * NUL-terminate this option: unfortunately,
3450 * mount options form a comma-separated list,
3451 * but mpol's nodelist may also contain commas.
3453 options
= strchr(options
, ',');
3454 if (options
== NULL
)
3457 if (!isdigit(*options
)) {
3464 if ((value
= strchr(this_char
,'=')) != NULL
) {
3467 pr_err("tmpfs: No value for mount option '%s'\n",
3472 if (!strcmp(this_char
,"size")) {
3473 unsigned long long size
;
3474 size
= memparse(value
,&rest
);
3476 size
<<= PAGE_SHIFT
;
3477 size
*= totalram_pages
;
3483 sbinfo
->max_blocks
=
3484 DIV_ROUND_UP(size
, PAGE_SIZE
);
3485 } else if (!strcmp(this_char
,"nr_blocks")) {
3486 sbinfo
->max_blocks
= memparse(value
, &rest
);
3489 } else if (!strcmp(this_char
,"nr_inodes")) {
3490 sbinfo
->max_inodes
= memparse(value
, &rest
);
3493 } else if (!strcmp(this_char
,"mode")) {
3496 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3499 } else if (!strcmp(this_char
,"uid")) {
3502 uid
= simple_strtoul(value
, &rest
, 0);
3505 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3506 if (!uid_valid(sbinfo
->uid
))
3508 } else if (!strcmp(this_char
,"gid")) {
3511 gid
= simple_strtoul(value
, &rest
, 0);
3514 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3515 if (!gid_valid(sbinfo
->gid
))
3517 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3518 } else if (!strcmp(this_char
, "huge")) {
3520 huge
= shmem_parse_huge(value
);
3523 if (!has_transparent_hugepage() &&
3524 huge
!= SHMEM_HUGE_NEVER
)
3526 sbinfo
->huge
= huge
;
3529 } else if (!strcmp(this_char
,"mpol")) {
3532 if (mpol_parse_str(value
, &mpol
))
3536 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3540 sbinfo
->mpol
= mpol
;
3544 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3552 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3554 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3555 struct shmem_sb_info config
= *sbinfo
;
3556 unsigned long inodes
;
3557 int error
= -EINVAL
;
3560 if (shmem_parse_options(data
, &config
, true))
3563 spin_lock(&sbinfo
->stat_lock
);
3564 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3565 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3567 if (config
.max_inodes
< inodes
)
3570 * Those tests disallow limited->unlimited while any are in use;
3571 * but we must separately disallow unlimited->limited, because
3572 * in that case we have no record of how much is already in use.
3574 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3576 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3580 sbinfo
->huge
= config
.huge
;
3581 sbinfo
->max_blocks
= config
.max_blocks
;
3582 sbinfo
->max_inodes
= config
.max_inodes
;
3583 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3586 * Preserve previous mempolicy unless mpol remount option was specified.
3589 mpol_put(sbinfo
->mpol
);
3590 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3593 spin_unlock(&sbinfo
->stat_lock
);
3597 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3599 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3601 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3602 seq_printf(seq
, ",size=%luk",
3603 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3604 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3605 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3606 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3607 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3608 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3609 seq_printf(seq
, ",uid=%u",
3610 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3611 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3612 seq_printf(seq
, ",gid=%u",
3613 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3614 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3615 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3617 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3619 shmem_show_mpol(seq
, sbinfo
->mpol
);
3623 #define MFD_NAME_PREFIX "memfd:"
3624 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3625 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3627 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3629 SYSCALL_DEFINE2(memfd_create
,
3630 const char __user
*, uname
,
3631 unsigned int, flags
)
3633 struct shmem_inode_info
*info
;
3639 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3642 /* length includes terminating zero */
3643 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3646 if (len
> MFD_NAME_MAX_LEN
+ 1)
3649 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
3653 strcpy(name
, MFD_NAME_PREFIX
);
3654 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3659 /* terminating-zero may have changed after strnlen_user() returned */
3660 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3665 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3671 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3673 error
= PTR_ERR(file
);
3676 info
= SHMEM_I(file_inode(file
));
3677 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3678 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3679 if (flags
& MFD_ALLOW_SEALING
)
3680 info
->seals
&= ~F_SEAL_SEAL
;
3682 fd_install(fd
, file
);
3693 #endif /* CONFIG_TMPFS */
3695 static void shmem_put_super(struct super_block
*sb
)
3697 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3699 percpu_counter_destroy(&sbinfo
->used_blocks
);
3700 mpol_put(sbinfo
->mpol
);
3702 sb
->s_fs_info
= NULL
;
3705 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3707 struct inode
*inode
;
3708 struct shmem_sb_info
*sbinfo
;
3711 /* Round up to L1_CACHE_BYTES to resist false sharing */
3712 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3713 L1_CACHE_BYTES
), GFP_KERNEL
);
3717 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3718 sbinfo
->uid
= current_fsuid();
3719 sbinfo
->gid
= current_fsgid();
3720 sb
->s_fs_info
= sbinfo
;
3724 * Per default we only allow half of the physical ram per
3725 * tmpfs instance, limiting inodes to one per page of lowmem;
3726 * but the internal instance is left unlimited.
3728 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3729 sbinfo
->max_blocks
= shmem_default_max_blocks();
3730 sbinfo
->max_inodes
= shmem_default_max_inodes();
3731 if (shmem_parse_options(data
, sbinfo
, false)) {
3736 sb
->s_flags
|= MS_NOUSER
;
3738 sb
->s_export_op
= &shmem_export_ops
;
3739 sb
->s_flags
|= MS_NOSEC
;
3741 sb
->s_flags
|= MS_NOUSER
;
3744 spin_lock_init(&sbinfo
->stat_lock
);
3745 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3747 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3748 spin_lock_init(&sbinfo
->shrinklist_lock
);
3749 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3751 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3752 sb
->s_blocksize
= PAGE_SIZE
;
3753 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3754 sb
->s_magic
= TMPFS_MAGIC
;
3755 sb
->s_op
= &shmem_ops
;
3756 sb
->s_time_gran
= 1;
3757 #ifdef CONFIG_TMPFS_XATTR
3758 sb
->s_xattr
= shmem_xattr_handlers
;
3760 #ifdef CONFIG_TMPFS_POSIX_ACL
3761 sb
->s_flags
|= MS_POSIXACL
;
3764 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3767 inode
->i_uid
= sbinfo
->uid
;
3768 inode
->i_gid
= sbinfo
->gid
;
3769 sb
->s_root
= d_make_root(inode
);
3775 shmem_put_super(sb
);
3779 static struct kmem_cache
*shmem_inode_cachep
;
3781 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3783 struct shmem_inode_info
*info
;
3784 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3787 return &info
->vfs_inode
;
3790 static void shmem_destroy_callback(struct rcu_head
*head
)
3792 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3793 if (S_ISLNK(inode
->i_mode
))
3794 kfree(inode
->i_link
);
3795 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3798 static void shmem_destroy_inode(struct inode
*inode
)
3800 if (S_ISREG(inode
->i_mode
))
3801 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3802 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3805 static void shmem_init_inode(void *foo
)
3807 struct shmem_inode_info
*info
= foo
;
3808 inode_init_once(&info
->vfs_inode
);
3811 static int shmem_init_inodecache(void)
3813 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3814 sizeof(struct shmem_inode_info
),
3815 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3819 static void shmem_destroy_inodecache(void)
3821 kmem_cache_destroy(shmem_inode_cachep
);
3824 static const struct address_space_operations shmem_aops
= {
3825 .writepage
= shmem_writepage
,
3826 .set_page_dirty
= __set_page_dirty_no_writeback
,
3828 .write_begin
= shmem_write_begin
,
3829 .write_end
= shmem_write_end
,
3831 #ifdef CONFIG_MIGRATION
3832 .migratepage
= migrate_page
,
3834 .error_remove_page
= generic_error_remove_page
,
3837 static const struct file_operations shmem_file_operations
= {
3839 .get_unmapped_area
= shmem_get_unmapped_area
,
3841 .llseek
= shmem_file_llseek
,
3842 .read_iter
= shmem_file_read_iter
,
3843 .write_iter
= generic_file_write_iter
,
3844 .fsync
= noop_fsync
,
3845 .splice_read
= generic_file_splice_read
,
3846 .splice_write
= iter_file_splice_write
,
3847 .fallocate
= shmem_fallocate
,
3851 static const struct inode_operations shmem_inode_operations
= {
3852 .getattr
= shmem_getattr
,
3853 .setattr
= shmem_setattr
,
3854 #ifdef CONFIG_TMPFS_XATTR
3855 .listxattr
= shmem_listxattr
,
3856 .set_acl
= simple_set_acl
,
3860 static const struct inode_operations shmem_dir_inode_operations
= {
3862 .create
= shmem_create
,
3863 .lookup
= simple_lookup
,
3865 .unlink
= shmem_unlink
,
3866 .symlink
= shmem_symlink
,
3867 .mkdir
= shmem_mkdir
,
3868 .rmdir
= shmem_rmdir
,
3869 .mknod
= shmem_mknod
,
3870 .rename
= shmem_rename2
,
3871 .tmpfile
= shmem_tmpfile
,
3873 #ifdef CONFIG_TMPFS_XATTR
3874 .listxattr
= shmem_listxattr
,
3876 #ifdef CONFIG_TMPFS_POSIX_ACL
3877 .setattr
= shmem_setattr
,
3878 .set_acl
= simple_set_acl
,
3882 static const struct inode_operations shmem_special_inode_operations
= {
3883 #ifdef CONFIG_TMPFS_XATTR
3884 .listxattr
= shmem_listxattr
,
3886 #ifdef CONFIG_TMPFS_POSIX_ACL
3887 .setattr
= shmem_setattr
,
3888 .set_acl
= simple_set_acl
,
3892 static const struct super_operations shmem_ops
= {
3893 .alloc_inode
= shmem_alloc_inode
,
3894 .destroy_inode
= shmem_destroy_inode
,
3896 .statfs
= shmem_statfs
,
3897 .remount_fs
= shmem_remount_fs
,
3898 .show_options
= shmem_show_options
,
3900 .evict_inode
= shmem_evict_inode
,
3901 .drop_inode
= generic_delete_inode
,
3902 .put_super
= shmem_put_super
,
3903 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3904 .nr_cached_objects
= shmem_unused_huge_count
,
3905 .free_cached_objects
= shmem_unused_huge_scan
,
3909 static const struct vm_operations_struct shmem_vm_ops
= {
3910 .fault
= shmem_fault
,
3911 .map_pages
= filemap_map_pages
,
3913 .set_policy
= shmem_set_policy
,
3914 .get_policy
= shmem_get_policy
,
3918 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3919 int flags
, const char *dev_name
, void *data
)
3921 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3924 static struct file_system_type shmem_fs_type
= {
3925 .owner
= THIS_MODULE
,
3927 .mount
= shmem_mount
,
3928 .kill_sb
= kill_litter_super
,
3929 .fs_flags
= FS_USERNS_MOUNT
,
3932 int __init
shmem_init(void)
3936 /* If rootfs called this, don't re-init */
3937 if (shmem_inode_cachep
)
3940 error
= shmem_init_inodecache();
3944 error
= register_filesystem(&shmem_fs_type
);
3946 pr_err("Could not register tmpfs\n");
3950 shm_mnt
= kern_mount(&shmem_fs_type
);
3951 if (IS_ERR(shm_mnt
)) {
3952 error
= PTR_ERR(shm_mnt
);
3953 pr_err("Could not kern_mount tmpfs\n");
3957 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3958 if (has_transparent_hugepage() && shmem_huge
< SHMEM_HUGE_DENY
)
3959 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3961 shmem_huge
= 0; /* just in case it was patched */
3966 unregister_filesystem(&shmem_fs_type
);
3968 shmem_destroy_inodecache();
3970 shm_mnt
= ERR_PTR(error
);
3974 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3975 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3976 struct kobj_attribute
*attr
, char *buf
)
3980 SHMEM_HUGE_WITHIN_SIZE
,
3988 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3989 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3991 count
+= sprintf(buf
+ count
, fmt
,
3992 shmem_format_huge(values
[i
]));
3994 buf
[count
- 1] = '\n';
3998 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3999 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4004 if (count
+ 1 > sizeof(tmp
))
4006 memcpy(tmp
, buf
, count
);
4008 if (count
&& tmp
[count
- 1] == '\n')
4009 tmp
[count
- 1] = '\0';
4011 huge
= shmem_parse_huge(tmp
);
4012 if (huge
== -EINVAL
)
4014 if (!has_transparent_hugepage() &&
4015 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4019 if (shmem_huge
< SHMEM_HUGE_DENY
)
4020 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4024 struct kobj_attribute shmem_enabled_attr
=
4025 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4026 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4028 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4029 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4031 struct inode
*inode
= file_inode(vma
->vm_file
);
4032 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4036 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4038 if (shmem_huge
== SHMEM_HUGE_DENY
)
4040 switch (sbinfo
->huge
) {
4041 case SHMEM_HUGE_NEVER
:
4043 case SHMEM_HUGE_ALWAYS
:
4045 case SHMEM_HUGE_WITHIN_SIZE
:
4046 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4047 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4048 if (i_size
>= HPAGE_PMD_SIZE
&&
4049 i_size
>> PAGE_SHIFT
>= off
)
4051 case SHMEM_HUGE_ADVISE
:
4052 /* TODO: implement fadvise() hints */
4053 return (vma
->vm_flags
& VM_HUGEPAGE
);
4059 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4061 #else /* !CONFIG_SHMEM */
4064 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4066 * This is intended for small system where the benefits of the full
4067 * shmem code (swap-backed and resource-limited) are outweighed by
4068 * their complexity. On systems without swap this code should be
4069 * effectively equivalent, but much lighter weight.
4072 static struct file_system_type shmem_fs_type
= {
4074 .mount
= ramfs_mount
,
4075 .kill_sb
= kill_litter_super
,
4076 .fs_flags
= FS_USERNS_MOUNT
,
4079 int __init
shmem_init(void)
4081 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4083 shm_mnt
= kern_mount(&shmem_fs_type
);
4084 BUG_ON(IS_ERR(shm_mnt
));
4089 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
4094 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4099 void shmem_unlock_mapping(struct address_space
*mapping
)
4104 unsigned long shmem_get_unmapped_area(struct file
*file
,
4105 unsigned long addr
, unsigned long len
,
4106 unsigned long pgoff
, unsigned long flags
)
4108 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4112 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4114 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4116 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4118 #define shmem_vm_ops generic_file_vm_ops
4119 #define shmem_file_operations ramfs_file_operations
4120 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4121 #define shmem_acct_size(flags, size) 0
4122 #define shmem_unacct_size(flags, size) do {} while (0)
4124 #endif /* CONFIG_SHMEM */
4128 static const struct dentry_operations anon_ops
= {
4129 .d_dname
= simple_dname
4132 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
4133 unsigned long flags
, unsigned int i_flags
)
4136 struct inode
*inode
;
4138 struct super_block
*sb
;
4141 if (IS_ERR(shm_mnt
))
4142 return ERR_CAST(shm_mnt
);
4144 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4145 return ERR_PTR(-EINVAL
);
4147 if (shmem_acct_size(flags
, size
))
4148 return ERR_PTR(-ENOMEM
);
4150 res
= ERR_PTR(-ENOMEM
);
4152 this.len
= strlen(name
);
4153 this.hash
= 0; /* will go */
4154 sb
= shm_mnt
->mnt_sb
;
4155 path
.mnt
= mntget(shm_mnt
);
4156 path
.dentry
= d_alloc_pseudo(sb
, &this);
4159 d_set_d_op(path
.dentry
, &anon_ops
);
4161 res
= ERR_PTR(-ENOSPC
);
4162 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
4166 inode
->i_flags
|= i_flags
;
4167 d_instantiate(path
.dentry
, inode
);
4168 inode
->i_size
= size
;
4169 clear_nlink(inode
); /* It is unlinked */
4170 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4174 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
4175 &shmem_file_operations
);
4182 shmem_unacct_size(flags
, size
);
4189 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4190 * kernel internal. There will be NO LSM permission checks against the
4191 * underlying inode. So users of this interface must do LSM checks at a
4192 * higher layer. The users are the big_key and shm implementations. LSM
4193 * checks are provided at the key or shm level rather than the inode.
4194 * @name: name for dentry (to be seen in /proc/<pid>/maps
4195 * @size: size to be set for the file
4196 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4198 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4200 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
4204 * shmem_file_setup - get an unlinked file living in tmpfs
4205 * @name: name for dentry (to be seen in /proc/<pid>/maps
4206 * @size: size to be set for the file
4207 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4209 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4211 return __shmem_file_setup(name
, size
, flags
, 0);
4213 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4216 * shmem_zero_setup - setup a shared anonymous mapping
4217 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4219 int shmem_zero_setup(struct vm_area_struct
*vma
)
4222 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4225 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4226 * between XFS directory reading and selinux: since this file is only
4227 * accessible to the user through its mapping, use S_PRIVATE flag to
4228 * bypass file security, in the same way as shmem_kernel_file_setup().
4230 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
4232 return PTR_ERR(file
);
4236 vma
->vm_file
= file
;
4237 vma
->vm_ops
= &shmem_vm_ops
;
4239 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4240 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4241 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4242 khugepaged_enter(vma
, vma
->vm_flags
);
4249 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4250 * @mapping: the page's address_space
4251 * @index: the page index
4252 * @gfp: the page allocator flags to use if allocating
4254 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4255 * with any new page allocations done using the specified allocation flags.
4256 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4257 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4258 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4260 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4261 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4263 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4264 pgoff_t index
, gfp_t gfp
)
4267 struct inode
*inode
= mapping
->host
;
4271 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4272 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4273 gfp
, NULL
, NULL
, NULL
);
4275 page
= ERR_PTR(error
);
4281 * The tiny !SHMEM case uses ramfs without swap
4283 return read_cache_page_gfp(mapping
, index
, gfp
);
4286 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);