2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/sched/signal.h>
33 #include <linux/export.h>
34 #include <linux/swap.h>
35 #include <linux/uio.h>
36 #include <linux/khugepaged.h>
38 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
40 static struct vfsmount
*shm_mnt
;
44 * This virtual memory filesystem is heavily based on the ramfs. It
45 * extends ramfs by the ability to use swap and honor resource limits
46 * which makes it a completely usable filesystem.
49 #include <linux/xattr.h>
50 #include <linux/exportfs.h>
51 #include <linux/posix_acl.h>
52 #include <linux/posix_acl_xattr.h>
53 #include <linux/mman.h>
54 #include <linux/string.h>
55 #include <linux/slab.h>
56 #include <linux/backing-dev.h>
57 #include <linux/shmem_fs.h>
58 #include <linux/writeback.h>
59 #include <linux/blkdev.h>
60 #include <linux/pagevec.h>
61 #include <linux/percpu_counter.h>
62 #include <linux/falloc.h>
63 #include <linux/splice.h>
64 #include <linux/security.h>
65 #include <linux/swapops.h>
66 #include <linux/mempolicy.h>
67 #include <linux/namei.h>
68 #include <linux/ctype.h>
69 #include <linux/migrate.h>
70 #include <linux/highmem.h>
71 #include <linux/seq_file.h>
72 #include <linux/magic.h>
73 #include <linux/syscalls.h>
74 #include <linux/fcntl.h>
75 #include <uapi/linux/memfd.h>
76 #include <linux/userfaultfd_k.h>
77 #include <linux/rmap.h>
78 #include <linux/uuid.h>
80 #include <linux/uaccess.h>
81 #include <asm/pgtable.h>
85 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
86 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
88 /* Pretend that each entry is of this size in directory's i_size */
89 #define BOGO_DIRENT_SIZE 20
91 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
92 #define SHORT_SYMLINK_LEN 128
95 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
96 * inode->i_private (with i_mutex making sure that it has only one user at
97 * a time): we would prefer not to enlarge the shmem inode just for that.
100 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
101 pgoff_t start
; /* start of range currently being fallocated */
102 pgoff_t next
; /* the next page offset to be fallocated */
103 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
104 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
108 static unsigned long shmem_default_max_blocks(void)
110 return totalram_pages
/ 2;
113 static int shmem_default_max_inodes(void)
118 ul
= min3(ul
, totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
123 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
124 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
125 struct shmem_inode_info
*info
, pgoff_t index
);
126 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
127 struct page
**pagep
, enum sgp_type sgp
,
128 gfp_t gfp
, struct vm_area_struct
*vma
,
129 struct vm_fault
*vmf
, int *fault_type
);
131 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
132 struct page
**pagep
, enum sgp_type sgp
)
134 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
135 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
138 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
140 return sb
->s_fs_info
;
144 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
145 * for shared memory and for shared anonymous (/dev/zero) mappings
146 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
147 * consistent with the pre-accounting of private mappings ...
149 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
151 return (flags
& VM_NORESERVE
) ?
152 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
155 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
157 if (!(flags
& VM_NORESERVE
))
158 vm_unacct_memory(VM_ACCT(size
));
161 static inline int shmem_reacct_size(unsigned long flags
,
162 loff_t oldsize
, loff_t newsize
)
164 if (!(flags
& VM_NORESERVE
)) {
165 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
166 return security_vm_enough_memory_mm(current
->mm
,
167 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
168 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
169 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
175 * ... whereas tmpfs objects are accounted incrementally as
176 * pages are allocated, in order to allow large sparse files.
177 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
178 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
180 static inline int shmem_acct_block(unsigned long flags
, long pages
)
182 if (!(flags
& VM_NORESERVE
))
185 return security_vm_enough_memory_mm(current
->mm
,
186 pages
* VM_ACCT(PAGE_SIZE
));
189 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
191 if (flags
& VM_NORESERVE
)
192 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
195 static const struct super_operations shmem_ops
;
196 static const struct address_space_operations shmem_aops
;
197 static const struct file_operations shmem_file_operations
;
198 static const struct inode_operations shmem_inode_operations
;
199 static const struct inode_operations shmem_dir_inode_operations
;
200 static const struct inode_operations shmem_special_inode_operations
;
201 static const struct vm_operations_struct shmem_vm_ops
;
202 static struct file_system_type shmem_fs_type
;
204 bool vma_is_shmem(struct vm_area_struct
*vma
)
206 return vma
->vm_ops
== &shmem_vm_ops
;
209 static LIST_HEAD(shmem_swaplist
);
210 static DEFINE_MUTEX(shmem_swaplist_mutex
);
212 static int shmem_reserve_inode(struct super_block
*sb
)
214 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
215 if (sbinfo
->max_inodes
) {
216 spin_lock(&sbinfo
->stat_lock
);
217 if (!sbinfo
->free_inodes
) {
218 spin_unlock(&sbinfo
->stat_lock
);
221 sbinfo
->free_inodes
--;
222 spin_unlock(&sbinfo
->stat_lock
);
227 static void shmem_free_inode(struct super_block
*sb
)
229 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
230 if (sbinfo
->max_inodes
) {
231 spin_lock(&sbinfo
->stat_lock
);
232 sbinfo
->free_inodes
++;
233 spin_unlock(&sbinfo
->stat_lock
);
238 * shmem_recalc_inode - recalculate the block usage of an inode
239 * @inode: inode to recalc
241 * We have to calculate the free blocks since the mm can drop
242 * undirtied hole pages behind our back.
244 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
245 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
247 * It has to be called with the spinlock held.
249 static void shmem_recalc_inode(struct inode
*inode
)
251 struct shmem_inode_info
*info
= SHMEM_I(inode
);
254 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
256 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
257 if (sbinfo
->max_blocks
)
258 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
259 info
->alloced
-= freed
;
260 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
261 shmem_unacct_blocks(info
->flags
, freed
);
265 bool shmem_charge(struct inode
*inode
, long pages
)
267 struct shmem_inode_info
*info
= SHMEM_I(inode
);
268 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
271 if (shmem_acct_block(info
->flags
, pages
))
273 spin_lock_irqsave(&info
->lock
, flags
);
274 info
->alloced
+= pages
;
275 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
276 shmem_recalc_inode(inode
);
277 spin_unlock_irqrestore(&info
->lock
, flags
);
278 inode
->i_mapping
->nrpages
+= pages
;
280 if (!sbinfo
->max_blocks
)
282 if (percpu_counter_compare(&sbinfo
->used_blocks
,
283 sbinfo
->max_blocks
- pages
) > 0) {
284 inode
->i_mapping
->nrpages
-= pages
;
285 spin_lock_irqsave(&info
->lock
, flags
);
286 info
->alloced
-= pages
;
287 shmem_recalc_inode(inode
);
288 spin_unlock_irqrestore(&info
->lock
, flags
);
289 shmem_unacct_blocks(info
->flags
, pages
);
292 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
296 void shmem_uncharge(struct inode
*inode
, long pages
)
298 struct shmem_inode_info
*info
= SHMEM_I(inode
);
299 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
302 spin_lock_irqsave(&info
->lock
, flags
);
303 info
->alloced
-= pages
;
304 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
305 shmem_recalc_inode(inode
);
306 spin_unlock_irqrestore(&info
->lock
, flags
);
308 if (sbinfo
->max_blocks
)
309 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
310 shmem_unacct_blocks(info
->flags
, pages
);
314 * Replace item expected in radix tree by a new item, while holding tree lock.
316 static int shmem_radix_tree_replace(struct address_space
*mapping
,
317 pgoff_t index
, void *expected
, void *replacement
)
319 struct radix_tree_node
*node
;
323 VM_BUG_ON(!expected
);
324 VM_BUG_ON(!replacement
);
325 item
= __radix_tree_lookup(&mapping
->page_tree
, index
, &node
, &pslot
);
328 if (item
!= expected
)
330 __radix_tree_replace(&mapping
->page_tree
, node
, pslot
,
331 replacement
, NULL
, NULL
);
336 * Sometimes, before we decide whether to proceed or to fail, we must check
337 * that an entry was not already brought back from swap by a racing thread.
339 * Checking page is not enough: by the time a SwapCache page is locked, it
340 * might be reused, and again be SwapCache, using the same swap as before.
342 static bool shmem_confirm_swap(struct address_space
*mapping
,
343 pgoff_t index
, swp_entry_t swap
)
348 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
350 return item
== swp_to_radix_entry(swap
);
354 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
357 * disables huge pages for the mount;
359 * enables huge pages for the mount;
360 * SHMEM_HUGE_WITHIN_SIZE:
361 * only allocate huge pages if the page will be fully within i_size,
362 * also respect fadvise()/madvise() hints;
364 * only allocate huge pages if requested with fadvise()/madvise();
367 #define SHMEM_HUGE_NEVER 0
368 #define SHMEM_HUGE_ALWAYS 1
369 #define SHMEM_HUGE_WITHIN_SIZE 2
370 #define SHMEM_HUGE_ADVISE 3
374 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
377 * disables huge on shm_mnt and all mounts, for emergency use;
379 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
382 #define SHMEM_HUGE_DENY (-1)
383 #define SHMEM_HUGE_FORCE (-2)
385 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
386 /* ifdef here to avoid bloating shmem.o when not necessary */
388 int shmem_huge __read_mostly
;
390 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
391 static int shmem_parse_huge(const char *str
)
393 if (!strcmp(str
, "never"))
394 return SHMEM_HUGE_NEVER
;
395 if (!strcmp(str
, "always"))
396 return SHMEM_HUGE_ALWAYS
;
397 if (!strcmp(str
, "within_size"))
398 return SHMEM_HUGE_WITHIN_SIZE
;
399 if (!strcmp(str
, "advise"))
400 return SHMEM_HUGE_ADVISE
;
401 if (!strcmp(str
, "deny"))
402 return SHMEM_HUGE_DENY
;
403 if (!strcmp(str
, "force"))
404 return SHMEM_HUGE_FORCE
;
408 static const char *shmem_format_huge(int huge
)
411 case SHMEM_HUGE_NEVER
:
413 case SHMEM_HUGE_ALWAYS
:
415 case SHMEM_HUGE_WITHIN_SIZE
:
416 return "within_size";
417 case SHMEM_HUGE_ADVISE
:
419 case SHMEM_HUGE_DENY
:
421 case SHMEM_HUGE_FORCE
:
430 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
431 struct shrink_control
*sc
, unsigned long nr_to_split
)
433 LIST_HEAD(list
), *pos
, *next
;
434 LIST_HEAD(to_remove
);
436 struct shmem_inode_info
*info
;
438 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
439 int removed
= 0, split
= 0;
441 if (list_empty(&sbinfo
->shrinklist
))
444 spin_lock(&sbinfo
->shrinklist_lock
);
445 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
446 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
449 inode
= igrab(&info
->vfs_inode
);
451 /* inode is about to be evicted */
453 list_del_init(&info
->shrinklist
);
458 /* Check if there's anything to gain */
459 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
460 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
461 list_move(&info
->shrinklist
, &to_remove
);
466 list_move(&info
->shrinklist
, &list
);
471 spin_unlock(&sbinfo
->shrinklist_lock
);
473 list_for_each_safe(pos
, next
, &to_remove
) {
474 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
475 inode
= &info
->vfs_inode
;
476 list_del_init(&info
->shrinklist
);
480 list_for_each_safe(pos
, next
, &list
) {
483 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
484 inode
= &info
->vfs_inode
;
486 if (nr_to_split
&& split
>= nr_to_split
) {
491 page
= find_lock_page(inode
->i_mapping
,
492 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
496 if (!PageTransHuge(page
)) {
502 ret
= split_huge_page(page
);
507 /* split failed: leave it on the list */
514 list_del_init(&info
->shrinklist
);
519 spin_lock(&sbinfo
->shrinklist_lock
);
520 list_splice_tail(&list
, &sbinfo
->shrinklist
);
521 sbinfo
->shrinklist_len
-= removed
;
522 spin_unlock(&sbinfo
->shrinklist_lock
);
527 static long shmem_unused_huge_scan(struct super_block
*sb
,
528 struct shrink_control
*sc
)
530 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
532 if (!READ_ONCE(sbinfo
->shrinklist_len
))
535 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
538 static long shmem_unused_huge_count(struct super_block
*sb
,
539 struct shrink_control
*sc
)
541 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
542 return READ_ONCE(sbinfo
->shrinklist_len
);
544 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
546 #define shmem_huge SHMEM_HUGE_DENY
548 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
549 struct shrink_control
*sc
, unsigned long nr_to_split
)
553 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
556 * Like add_to_page_cache_locked, but error if expected item has gone.
558 static int shmem_add_to_page_cache(struct page
*page
,
559 struct address_space
*mapping
,
560 pgoff_t index
, void *expected
)
562 int error
, nr
= hpage_nr_pages(page
);
564 VM_BUG_ON_PAGE(PageTail(page
), page
);
565 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
566 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
567 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
568 VM_BUG_ON(expected
&& PageTransHuge(page
));
570 page_ref_add(page
, nr
);
571 page
->mapping
= mapping
;
574 spin_lock_irq(&mapping
->tree_lock
);
575 if (PageTransHuge(page
)) {
576 void __rcu
**results
;
581 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
,
582 &results
, &idx
, index
, 1) &&
583 idx
< index
+ HPAGE_PMD_NR
) {
588 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
589 error
= radix_tree_insert(&mapping
->page_tree
,
590 index
+ i
, page
+ i
);
593 count_vm_event(THP_FILE_ALLOC
);
595 } else if (!expected
) {
596 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
598 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
603 mapping
->nrpages
+= nr
;
604 if (PageTransHuge(page
))
605 __inc_node_page_state(page
, NR_SHMEM_THPS
);
606 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
607 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
608 spin_unlock_irq(&mapping
->tree_lock
);
610 page
->mapping
= NULL
;
611 spin_unlock_irq(&mapping
->tree_lock
);
612 page_ref_sub(page
, nr
);
618 * Like delete_from_page_cache, but substitutes swap for page.
620 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
622 struct address_space
*mapping
= page
->mapping
;
625 VM_BUG_ON_PAGE(PageCompound(page
), page
);
627 spin_lock_irq(&mapping
->tree_lock
);
628 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
629 page
->mapping
= NULL
;
631 __dec_node_page_state(page
, NR_FILE_PAGES
);
632 __dec_node_page_state(page
, NR_SHMEM
);
633 spin_unlock_irq(&mapping
->tree_lock
);
639 * Remove swap entry from radix tree, free the swap and its page cache.
641 static int shmem_free_swap(struct address_space
*mapping
,
642 pgoff_t index
, void *radswap
)
646 spin_lock_irq(&mapping
->tree_lock
);
647 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
648 spin_unlock_irq(&mapping
->tree_lock
);
651 free_swap_and_cache(radix_to_swp_entry(radswap
));
656 * Determine (in bytes) how many of the shmem object's pages mapped by the
657 * given offsets are swapped out.
659 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
660 * as long as the inode doesn't go away and racy results are not a problem.
662 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
663 pgoff_t start
, pgoff_t end
)
665 struct radix_tree_iter iter
;
668 unsigned long swapped
= 0;
672 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
673 if (iter
.index
>= end
)
676 page
= radix_tree_deref_slot(slot
);
678 if (radix_tree_deref_retry(page
)) {
679 slot
= radix_tree_iter_retry(&iter
);
683 if (radix_tree_exceptional_entry(page
))
686 if (need_resched()) {
687 slot
= radix_tree_iter_resume(slot
, &iter
);
694 return swapped
<< PAGE_SHIFT
;
698 * Determine (in bytes) how many of the shmem object's pages mapped by the
699 * given vma is swapped out.
701 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
702 * as long as the inode doesn't go away and racy results are not a problem.
704 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
706 struct inode
*inode
= file_inode(vma
->vm_file
);
707 struct shmem_inode_info
*info
= SHMEM_I(inode
);
708 struct address_space
*mapping
= inode
->i_mapping
;
709 unsigned long swapped
;
711 /* Be careful as we don't hold info->lock */
712 swapped
= READ_ONCE(info
->swapped
);
715 * The easier cases are when the shmem object has nothing in swap, or
716 * the vma maps it whole. Then we can simply use the stats that we
722 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
723 return swapped
<< PAGE_SHIFT
;
725 /* Here comes the more involved part */
726 return shmem_partial_swap_usage(mapping
,
727 linear_page_index(vma
, vma
->vm_start
),
728 linear_page_index(vma
, vma
->vm_end
));
732 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
734 void shmem_unlock_mapping(struct address_space
*mapping
)
737 pgoff_t indices
[PAGEVEC_SIZE
];
740 pagevec_init(&pvec
, 0);
742 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
744 while (!mapping_unevictable(mapping
)) {
746 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
747 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
749 pvec
.nr
= find_get_entries(mapping
, index
,
750 PAGEVEC_SIZE
, pvec
.pages
, indices
);
753 index
= indices
[pvec
.nr
- 1] + 1;
754 pagevec_remove_exceptionals(&pvec
);
755 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
756 pagevec_release(&pvec
);
762 * Remove range of pages and swap entries from radix tree, and free them.
763 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
765 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
768 struct address_space
*mapping
= inode
->i_mapping
;
769 struct shmem_inode_info
*info
= SHMEM_I(inode
);
770 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
771 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
772 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
773 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
775 pgoff_t indices
[PAGEVEC_SIZE
];
776 long nr_swaps_freed
= 0;
781 end
= -1; /* unsigned, so actually very big */
783 pagevec_init(&pvec
, 0);
785 while (index
< end
) {
786 pvec
.nr
= find_get_entries(mapping
, index
,
787 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
788 pvec
.pages
, indices
);
791 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
792 struct page
*page
= pvec
.pages
[i
];
798 if (radix_tree_exceptional_entry(page
)) {
801 nr_swaps_freed
+= !shmem_free_swap(mapping
,
806 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
808 if (!trylock_page(page
))
811 if (PageTransTail(page
)) {
812 /* Middle of THP: zero out the page */
813 clear_highpage(page
);
816 } else if (PageTransHuge(page
)) {
817 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
819 * Range ends in the middle of THP:
822 clear_highpage(page
);
826 index
+= HPAGE_PMD_NR
- 1;
827 i
+= HPAGE_PMD_NR
- 1;
830 if (!unfalloc
|| !PageUptodate(page
)) {
831 VM_BUG_ON_PAGE(PageTail(page
), page
);
832 if (page_mapping(page
) == mapping
) {
833 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
834 truncate_inode_page(mapping
, page
);
839 pagevec_remove_exceptionals(&pvec
);
840 pagevec_release(&pvec
);
846 struct page
*page
= NULL
;
847 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
849 unsigned int top
= PAGE_SIZE
;
854 zero_user_segment(page
, partial_start
, top
);
855 set_page_dirty(page
);
861 struct page
*page
= NULL
;
862 shmem_getpage(inode
, end
, &page
, SGP_READ
);
864 zero_user_segment(page
, 0, partial_end
);
865 set_page_dirty(page
);
874 while (index
< end
) {
877 pvec
.nr
= find_get_entries(mapping
, index
,
878 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
879 pvec
.pages
, indices
);
881 /* If all gone or hole-punch or unfalloc, we're done */
882 if (index
== start
|| end
!= -1)
884 /* But if truncating, restart to make sure all gone */
888 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
889 struct page
*page
= pvec
.pages
[i
];
895 if (radix_tree_exceptional_entry(page
)) {
898 if (shmem_free_swap(mapping
, index
, page
)) {
899 /* Swap was replaced by page: retry */
909 if (PageTransTail(page
)) {
910 /* Middle of THP: zero out the page */
911 clear_highpage(page
);
914 * Partial thp truncate due 'start' in middle
915 * of THP: don't need to look on these pages
916 * again on !pvec.nr restart.
918 if (index
!= round_down(end
, HPAGE_PMD_NR
))
921 } else if (PageTransHuge(page
)) {
922 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
924 * Range ends in the middle of THP:
927 clear_highpage(page
);
931 index
+= HPAGE_PMD_NR
- 1;
932 i
+= HPAGE_PMD_NR
- 1;
935 if (!unfalloc
|| !PageUptodate(page
)) {
936 VM_BUG_ON_PAGE(PageTail(page
), page
);
937 if (page_mapping(page
) == mapping
) {
938 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
939 truncate_inode_page(mapping
, page
);
941 /* Page was replaced by swap: retry */
949 pagevec_remove_exceptionals(&pvec
);
950 pagevec_release(&pvec
);
954 spin_lock_irq(&info
->lock
);
955 info
->swapped
-= nr_swaps_freed
;
956 shmem_recalc_inode(inode
);
957 spin_unlock_irq(&info
->lock
);
960 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
962 shmem_undo_range(inode
, lstart
, lend
, false);
963 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
965 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
967 static int shmem_getattr(const struct path
*path
, struct kstat
*stat
,
968 u32 request_mask
, unsigned int query_flags
)
970 struct inode
*inode
= path
->dentry
->d_inode
;
971 struct shmem_inode_info
*info
= SHMEM_I(inode
);
973 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
974 spin_lock_irq(&info
->lock
);
975 shmem_recalc_inode(inode
);
976 spin_unlock_irq(&info
->lock
);
978 generic_fillattr(inode
, stat
);
982 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
984 struct inode
*inode
= d_inode(dentry
);
985 struct shmem_inode_info
*info
= SHMEM_I(inode
);
986 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
989 error
= setattr_prepare(dentry
, attr
);
993 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
994 loff_t oldsize
= inode
->i_size
;
995 loff_t newsize
= attr
->ia_size
;
997 /* protected by i_mutex */
998 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
999 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1002 if (newsize
!= oldsize
) {
1003 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1007 i_size_write(inode
, newsize
);
1008 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1010 if (newsize
<= oldsize
) {
1011 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1012 if (oldsize
> holebegin
)
1013 unmap_mapping_range(inode
->i_mapping
,
1016 shmem_truncate_range(inode
,
1017 newsize
, (loff_t
)-1);
1018 /* unmap again to remove racily COWed private pages */
1019 if (oldsize
> holebegin
)
1020 unmap_mapping_range(inode
->i_mapping
,
1024 * Part of the huge page can be beyond i_size: subject
1025 * to shrink under memory pressure.
1027 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1028 spin_lock(&sbinfo
->shrinklist_lock
);
1030 * _careful to defend against unlocked access to
1031 * ->shrink_list in shmem_unused_huge_shrink()
1033 if (list_empty_careful(&info
->shrinklist
)) {
1034 list_add_tail(&info
->shrinklist
,
1035 &sbinfo
->shrinklist
);
1036 sbinfo
->shrinklist_len
++;
1038 spin_unlock(&sbinfo
->shrinklist_lock
);
1043 setattr_copy(inode
, attr
);
1044 if (attr
->ia_valid
& ATTR_MODE
)
1045 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1049 static void shmem_evict_inode(struct inode
*inode
)
1051 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1052 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1054 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1055 shmem_unacct_size(info
->flags
, inode
->i_size
);
1057 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1058 if (!list_empty(&info
->shrinklist
)) {
1059 spin_lock(&sbinfo
->shrinklist_lock
);
1060 if (!list_empty(&info
->shrinklist
)) {
1061 list_del_init(&info
->shrinklist
);
1062 sbinfo
->shrinklist_len
--;
1064 spin_unlock(&sbinfo
->shrinklist_lock
);
1066 if (!list_empty(&info
->swaplist
)) {
1067 mutex_lock(&shmem_swaplist_mutex
);
1068 list_del_init(&info
->swaplist
);
1069 mutex_unlock(&shmem_swaplist_mutex
);
1073 simple_xattrs_free(&info
->xattrs
);
1074 WARN_ON(inode
->i_blocks
);
1075 if (!sbinfo
->idr_nouse
&& inode
->i_ino
) {
1076 mutex_lock(&sbinfo
->idr_lock
);
1077 idr_remove(&sbinfo
->idr
, inode
->i_ino
);
1078 mutex_unlock(&sbinfo
->idr_lock
);
1080 shmem_free_inode(inode
->i_sb
);
1084 static unsigned long find_swap_entry(struct radix_tree_root
*root
, void *item
)
1086 struct radix_tree_iter iter
;
1088 unsigned long found
= -1;
1089 unsigned int checked
= 0;
1092 radix_tree_for_each_slot(slot
, root
, &iter
, 0) {
1093 if (*slot
== item
) {
1098 if ((checked
% 4096) != 0)
1100 slot
= radix_tree_iter_resume(slot
, &iter
);
1109 * If swap found in inode, free it and move page from swapcache to filecache.
1111 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
1112 swp_entry_t swap
, struct page
**pagep
)
1114 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1120 radswap
= swp_to_radix_entry(swap
);
1121 index
= find_swap_entry(&mapping
->page_tree
, radswap
);
1123 return -EAGAIN
; /* tell shmem_unuse we found nothing */
1126 * Move _head_ to start search for next from here.
1127 * But be careful: shmem_evict_inode checks list_empty without taking
1128 * mutex, and there's an instant in list_move_tail when info->swaplist
1129 * would appear empty, if it were the only one on shmem_swaplist.
1131 if (shmem_swaplist
.next
!= &info
->swaplist
)
1132 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
1134 gfp
= mapping_gfp_mask(mapping
);
1135 if (shmem_should_replace_page(*pagep
, gfp
)) {
1136 mutex_unlock(&shmem_swaplist_mutex
);
1137 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
1138 mutex_lock(&shmem_swaplist_mutex
);
1140 * We needed to drop mutex to make that restrictive page
1141 * allocation, but the inode might have been freed while we
1142 * dropped it: although a racing shmem_evict_inode() cannot
1143 * complete without emptying the radix_tree, our page lock
1144 * on this swapcache page is not enough to prevent that -
1145 * free_swap_and_cache() of our swap entry will only
1146 * trylock_page(), removing swap from radix_tree whatever.
1148 * We must not proceed to shmem_add_to_page_cache() if the
1149 * inode has been freed, but of course we cannot rely on
1150 * inode or mapping or info to check that. However, we can
1151 * safely check if our swap entry is still in use (and here
1152 * it can't have got reused for another page): if it's still
1153 * in use, then the inode cannot have been freed yet, and we
1154 * can safely proceed (if it's no longer in use, that tells
1155 * nothing about the inode, but we don't need to unuse swap).
1157 if (!page_swapcount(*pagep
))
1162 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1163 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1164 * beneath us (pagelock doesn't help until the page is in pagecache).
1167 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
1169 if (error
!= -ENOMEM
) {
1171 * Truncation and eviction use free_swap_and_cache(), which
1172 * only does trylock page: if we raced, best clean up here.
1174 delete_from_swap_cache(*pagep
);
1175 set_page_dirty(*pagep
);
1177 spin_lock_irq(&info
->lock
);
1179 spin_unlock_irq(&info
->lock
);
1187 * Search through swapped inodes to find and replace swap by page.
1189 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1191 struct list_head
*this, *next
;
1192 struct shmem_inode_info
*info
;
1193 struct mem_cgroup
*memcg
;
1197 * There's a faint possibility that swap page was replaced before
1198 * caller locked it: caller will come back later with the right page.
1200 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1204 * Charge page using GFP_KERNEL while we can wait, before taking
1205 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1206 * Charged back to the user (not to caller) when swap account is used.
1208 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1212 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1215 mutex_lock(&shmem_swaplist_mutex
);
1216 list_for_each_safe(this, next
, &shmem_swaplist
) {
1217 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1219 error
= shmem_unuse_inode(info
, swap
, &page
);
1221 list_del_init(&info
->swaplist
);
1223 if (error
!= -EAGAIN
)
1225 /* found nothing in this: move on to search the next */
1227 mutex_unlock(&shmem_swaplist_mutex
);
1230 if (error
!= -ENOMEM
)
1232 mem_cgroup_cancel_charge(page
, memcg
, false);
1234 mem_cgroup_commit_charge(page
, memcg
, true, false);
1242 * Move the page from the page cache to the swap cache.
1244 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1246 struct shmem_inode_info
*info
;
1247 struct address_space
*mapping
;
1248 struct inode
*inode
;
1252 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1253 BUG_ON(!PageLocked(page
));
1254 mapping
= page
->mapping
;
1255 index
= page
->index
;
1256 inode
= mapping
->host
;
1257 info
= SHMEM_I(inode
);
1258 if (info
->flags
& VM_LOCKED
)
1260 if (!total_swap_pages
)
1264 * Our capabilities prevent regular writeback or sync from ever calling
1265 * shmem_writepage; but a stacking filesystem might use ->writepage of
1266 * its underlying filesystem, in which case tmpfs should write out to
1267 * swap only in response to memory pressure, and not for the writeback
1270 if (!wbc
->for_reclaim
) {
1271 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1276 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1277 * value into swapfile.c, the only way we can correctly account for a
1278 * fallocated page arriving here is now to initialize it and write it.
1280 * That's okay for a page already fallocated earlier, but if we have
1281 * not yet completed the fallocation, then (a) we want to keep track
1282 * of this page in case we have to undo it, and (b) it may not be a
1283 * good idea to continue anyway, once we're pushing into swap. So
1284 * reactivate the page, and let shmem_fallocate() quit when too many.
1286 if (!PageUptodate(page
)) {
1287 if (inode
->i_private
) {
1288 struct shmem_falloc
*shmem_falloc
;
1289 spin_lock(&inode
->i_lock
);
1290 shmem_falloc
= inode
->i_private
;
1292 !shmem_falloc
->waitq
&&
1293 index
>= shmem_falloc
->start
&&
1294 index
< shmem_falloc
->next
)
1295 shmem_falloc
->nr_unswapped
++;
1297 shmem_falloc
= NULL
;
1298 spin_unlock(&inode
->i_lock
);
1302 clear_highpage(page
);
1303 flush_dcache_page(page
);
1304 SetPageUptodate(page
);
1307 swap
= get_swap_page(page
);
1311 if (mem_cgroup_try_charge_swap(page
, swap
))
1315 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1316 * if it's not already there. Do it now before the page is
1317 * moved to swap cache, when its pagelock no longer protects
1318 * the inode from eviction. But don't unlock the mutex until
1319 * we've incremented swapped, because shmem_unuse_inode() will
1320 * prune a !swapped inode from the swaplist under this mutex.
1322 mutex_lock(&shmem_swaplist_mutex
);
1323 if (list_empty(&info
->swaplist
))
1324 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1326 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1327 spin_lock_irq(&info
->lock
);
1328 shmem_recalc_inode(inode
);
1330 spin_unlock_irq(&info
->lock
);
1332 swap_shmem_alloc(swap
);
1333 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1335 mutex_unlock(&shmem_swaplist_mutex
);
1336 BUG_ON(page_mapped(page
));
1337 swap_writepage(page
, wbc
);
1341 mutex_unlock(&shmem_swaplist_mutex
);
1343 put_swap_page(page
, swap
);
1345 set_page_dirty(page
);
1346 if (wbc
->for_reclaim
)
1347 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1352 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1353 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1357 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1358 return; /* show nothing */
1360 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1362 seq_printf(seq
, ",mpol=%s", buffer
);
1365 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1367 struct mempolicy
*mpol
= NULL
;
1369 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1370 mpol
= sbinfo
->mpol
;
1372 spin_unlock(&sbinfo
->stat_lock
);
1376 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1377 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1380 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1384 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1386 #define vm_policy vm_private_data
1389 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1390 struct shmem_inode_info
*info
, pgoff_t index
)
1392 /* Create a pseudo vma that just contains the policy */
1394 /* Bias interleave by inode number to distribute better across nodes */
1395 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1397 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1400 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1402 /* Drop reference taken by mpol_shared_policy_lookup() */
1403 mpol_cond_put(vma
->vm_policy
);
1406 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1407 struct shmem_inode_info
*info
, pgoff_t index
)
1409 struct vm_area_struct pvma
;
1412 shmem_pseudo_vma_init(&pvma
, info
, index
);
1413 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1414 shmem_pseudo_vma_destroy(&pvma
);
1419 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1420 struct shmem_inode_info
*info
, pgoff_t index
)
1422 struct vm_area_struct pvma
;
1423 struct inode
*inode
= &info
->vfs_inode
;
1424 struct address_space
*mapping
= inode
->i_mapping
;
1425 pgoff_t idx
, hindex
;
1426 void __rcu
**results
;
1429 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1432 hindex
= round_down(index
, HPAGE_PMD_NR
);
1434 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1435 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1441 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1442 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1443 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1444 shmem_pseudo_vma_destroy(&pvma
);
1446 prep_transhuge_page(page
);
1450 static struct page
*shmem_alloc_page(gfp_t gfp
,
1451 struct shmem_inode_info
*info
, pgoff_t index
)
1453 struct vm_area_struct pvma
;
1456 shmem_pseudo_vma_init(&pvma
, info
, index
);
1457 page
= alloc_page_vma(gfp
, &pvma
, 0);
1458 shmem_pseudo_vma_destroy(&pvma
);
1463 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1464 struct shmem_inode_info
*info
, struct shmem_sb_info
*sbinfo
,
1465 pgoff_t index
, bool huge
)
1471 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1473 nr
= huge
? HPAGE_PMD_NR
: 1;
1475 if (shmem_acct_block(info
->flags
, nr
))
1477 if (sbinfo
->max_blocks
) {
1478 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1479 sbinfo
->max_blocks
- nr
) > 0)
1481 percpu_counter_add(&sbinfo
->used_blocks
, nr
);
1485 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1487 page
= shmem_alloc_page(gfp
, info
, index
);
1489 __SetPageLocked(page
);
1490 __SetPageSwapBacked(page
);
1495 if (sbinfo
->max_blocks
)
1496 percpu_counter_add(&sbinfo
->used_blocks
, -nr
);
1498 shmem_unacct_blocks(info
->flags
, nr
);
1500 return ERR_PTR(err
);
1504 * When a page is moved from swapcache to shmem filecache (either by the
1505 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1506 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1507 * ignorance of the mapping it belongs to. If that mapping has special
1508 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1509 * we may need to copy to a suitable page before moving to filecache.
1511 * In a future release, this may well be extended to respect cpuset and
1512 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1513 * but for now it is a simple matter of zone.
1515 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1517 return page_zonenum(page
) > gfp_zone(gfp
);
1520 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1521 struct shmem_inode_info
*info
, pgoff_t index
)
1523 struct page
*oldpage
, *newpage
;
1524 struct address_space
*swap_mapping
;
1529 swap_index
= page_private(oldpage
);
1530 swap_mapping
= page_mapping(oldpage
);
1533 * We have arrived here because our zones are constrained, so don't
1534 * limit chance of success by further cpuset and node constraints.
1536 gfp
&= ~GFP_CONSTRAINT_MASK
;
1537 newpage
= shmem_alloc_page(gfp
, info
, index
);
1542 copy_highpage(newpage
, oldpage
);
1543 flush_dcache_page(newpage
);
1545 __SetPageLocked(newpage
);
1546 __SetPageSwapBacked(newpage
);
1547 SetPageUptodate(newpage
);
1548 set_page_private(newpage
, swap_index
);
1549 SetPageSwapCache(newpage
);
1552 * Our caller will very soon move newpage out of swapcache, but it's
1553 * a nice clean interface for us to replace oldpage by newpage there.
1555 spin_lock_irq(&swap_mapping
->tree_lock
);
1556 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1559 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1560 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1562 spin_unlock_irq(&swap_mapping
->tree_lock
);
1564 if (unlikely(error
)) {
1566 * Is this possible? I think not, now that our callers check
1567 * both PageSwapCache and page_private after getting page lock;
1568 * but be defensive. Reverse old to newpage for clear and free.
1572 mem_cgroup_migrate(oldpage
, newpage
);
1573 lru_cache_add_anon(newpage
);
1577 ClearPageSwapCache(oldpage
);
1578 set_page_private(oldpage
, 0);
1580 unlock_page(oldpage
);
1587 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1589 * If we allocate a new one we do not mark it dirty. That's up to the
1590 * vm. If we swap it in we mark it dirty since we also free the swap
1591 * entry since a page cannot live in both the swap and page cache.
1593 * fault_mm and fault_type are only supplied by shmem_fault:
1594 * otherwise they are NULL.
1596 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1597 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1598 struct vm_area_struct
*vma
, struct vm_fault
*vmf
, int *fault_type
)
1600 struct address_space
*mapping
= inode
->i_mapping
;
1601 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1602 struct shmem_sb_info
*sbinfo
;
1603 struct mm_struct
*charge_mm
;
1604 struct mem_cgroup
*memcg
;
1607 enum sgp_type sgp_huge
= sgp
;
1608 pgoff_t hindex
= index
;
1613 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1615 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1619 page
= find_lock_entry(mapping
, index
);
1620 if (radix_tree_exceptional_entry(page
)) {
1621 swap
= radix_to_swp_entry(page
);
1625 if (sgp
<= SGP_CACHE
&&
1626 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1631 if (page
&& sgp
== SGP_WRITE
)
1632 mark_page_accessed(page
);
1634 /* fallocated page? */
1635 if (page
&& !PageUptodate(page
)) {
1636 if (sgp
!= SGP_READ
)
1642 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1648 * Fast cache lookup did not find it:
1649 * bring it back from swap or allocate.
1651 sbinfo
= SHMEM_SB(inode
->i_sb
);
1652 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1655 /* Look it up and read it in.. */
1656 page
= lookup_swap_cache(swap
);
1658 /* Or update major stats only when swapin succeeds?? */
1660 *fault_type
|= VM_FAULT_MAJOR
;
1661 count_vm_event(PGMAJFAULT
);
1662 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1664 /* Here we actually start the io */
1665 page
= shmem_swapin(swap
, gfp
, info
, index
);
1672 /* We have to do this with page locked to prevent races */
1674 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1675 !shmem_confirm_swap(mapping
, index
, swap
)) {
1676 error
= -EEXIST
; /* try again */
1679 if (!PageUptodate(page
)) {
1683 wait_on_page_writeback(page
);
1685 if (shmem_should_replace_page(page
, gfp
)) {
1686 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1691 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1694 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1695 swp_to_radix_entry(swap
));
1697 * We already confirmed swap under page lock, and make
1698 * no memory allocation here, so usually no possibility
1699 * of error; but free_swap_and_cache() only trylocks a
1700 * page, so it is just possible that the entry has been
1701 * truncated or holepunched since swap was confirmed.
1702 * shmem_undo_range() will have done some of the
1703 * unaccounting, now delete_from_swap_cache() will do
1705 * Reset swap.val? No, leave it so "failed" goes back to
1706 * "repeat": reading a hole and writing should succeed.
1709 mem_cgroup_cancel_charge(page
, memcg
, false);
1710 delete_from_swap_cache(page
);
1716 mem_cgroup_commit_charge(page
, memcg
, true, false);
1718 spin_lock_irq(&info
->lock
);
1720 shmem_recalc_inode(inode
);
1721 spin_unlock_irq(&info
->lock
);
1723 if (sgp
== SGP_WRITE
)
1724 mark_page_accessed(page
);
1726 delete_from_swap_cache(page
);
1727 set_page_dirty(page
);
1731 if (vma
&& userfaultfd_missing(vma
)) {
1732 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1736 /* shmem_symlink() */
1737 if (mapping
->a_ops
!= &shmem_aops
)
1739 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1741 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1743 switch (sbinfo
->huge
) {
1746 case SHMEM_HUGE_NEVER
:
1748 case SHMEM_HUGE_WITHIN_SIZE
:
1749 off
= round_up(index
, HPAGE_PMD_NR
);
1750 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1751 if (i_size
>= HPAGE_PMD_SIZE
&&
1752 i_size
>> PAGE_SHIFT
>= off
)
1755 case SHMEM_HUGE_ADVISE
:
1756 if (sgp_huge
== SGP_HUGE
)
1758 /* TODO: implement fadvise() hints */
1763 page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1766 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1771 error
= PTR_ERR(page
);
1773 if (error
!= -ENOSPC
)
1776 * Try to reclaim some spece by splitting a huge page
1777 * beyond i_size on the filesystem.
1781 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1782 if (ret
== SHRINK_STOP
)
1790 if (PageTransHuge(page
))
1791 hindex
= round_down(index
, HPAGE_PMD_NR
);
1795 if (sgp
== SGP_WRITE
)
1796 __SetPageReferenced(page
);
1798 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1799 PageTransHuge(page
));
1802 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1803 compound_order(page
));
1805 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1807 radix_tree_preload_end();
1810 mem_cgroup_cancel_charge(page
, memcg
,
1811 PageTransHuge(page
));
1814 mem_cgroup_commit_charge(page
, memcg
, false,
1815 PageTransHuge(page
));
1816 lru_cache_add_anon(page
);
1818 spin_lock_irq(&info
->lock
);
1819 info
->alloced
+= 1 << compound_order(page
);
1820 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1821 shmem_recalc_inode(inode
);
1822 spin_unlock_irq(&info
->lock
);
1825 if (PageTransHuge(page
) &&
1826 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1827 hindex
+ HPAGE_PMD_NR
- 1) {
1829 * Part of the huge page is beyond i_size: subject
1830 * to shrink under memory pressure.
1832 spin_lock(&sbinfo
->shrinklist_lock
);
1834 * _careful to defend against unlocked access to
1835 * ->shrink_list in shmem_unused_huge_shrink()
1837 if (list_empty_careful(&info
->shrinklist
)) {
1838 list_add_tail(&info
->shrinklist
,
1839 &sbinfo
->shrinklist
);
1840 sbinfo
->shrinklist_len
++;
1842 spin_unlock(&sbinfo
->shrinklist_lock
);
1846 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1848 if (sgp
== SGP_FALLOC
)
1852 * Let SGP_WRITE caller clear ends if write does not fill page;
1853 * but SGP_FALLOC on a page fallocated earlier must initialize
1854 * it now, lest undo on failure cancel our earlier guarantee.
1856 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1857 struct page
*head
= compound_head(page
);
1860 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1861 clear_highpage(head
+ i
);
1862 flush_dcache_page(head
+ i
);
1864 SetPageUptodate(head
);
1868 /* Perhaps the file has been truncated since we checked */
1869 if (sgp
<= SGP_CACHE
&&
1870 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1872 ClearPageDirty(page
);
1873 delete_from_page_cache(page
);
1874 spin_lock_irq(&info
->lock
);
1875 shmem_recalc_inode(inode
);
1876 spin_unlock_irq(&info
->lock
);
1881 *pagep
= page
+ index
- hindex
;
1888 if (sbinfo
->max_blocks
)
1889 percpu_counter_sub(&sbinfo
->used_blocks
,
1890 1 << compound_order(page
));
1891 shmem_unacct_blocks(info
->flags
, 1 << compound_order(page
));
1893 if (PageTransHuge(page
)) {
1899 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1906 if (error
== -ENOSPC
&& !once
++) {
1907 spin_lock_irq(&info
->lock
);
1908 shmem_recalc_inode(inode
);
1909 spin_unlock_irq(&info
->lock
);
1912 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1918 * This is like autoremove_wake_function, but it removes the wait queue
1919 * entry unconditionally - even if something else had already woken the
1922 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
1924 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1925 list_del_init(&wait
->entry
);
1929 static int shmem_fault(struct vm_fault
*vmf
)
1931 struct vm_area_struct
*vma
= vmf
->vma
;
1932 struct inode
*inode
= file_inode(vma
->vm_file
);
1933 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1936 int ret
= VM_FAULT_LOCKED
;
1939 * Trinity finds that probing a hole which tmpfs is punching can
1940 * prevent the hole-punch from ever completing: which in turn
1941 * locks writers out with its hold on i_mutex. So refrain from
1942 * faulting pages into the hole while it's being punched. Although
1943 * shmem_undo_range() does remove the additions, it may be unable to
1944 * keep up, as each new page needs its own unmap_mapping_range() call,
1945 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1947 * It does not matter if we sometimes reach this check just before the
1948 * hole-punch begins, so that one fault then races with the punch:
1949 * we just need to make racing faults a rare case.
1951 * The implementation below would be much simpler if we just used a
1952 * standard mutex or completion: but we cannot take i_mutex in fault,
1953 * and bloating every shmem inode for this unlikely case would be sad.
1955 if (unlikely(inode
->i_private
)) {
1956 struct shmem_falloc
*shmem_falloc
;
1958 spin_lock(&inode
->i_lock
);
1959 shmem_falloc
= inode
->i_private
;
1961 shmem_falloc
->waitq
&&
1962 vmf
->pgoff
>= shmem_falloc
->start
&&
1963 vmf
->pgoff
< shmem_falloc
->next
) {
1964 wait_queue_head_t
*shmem_falloc_waitq
;
1965 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
1967 ret
= VM_FAULT_NOPAGE
;
1968 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1969 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1970 /* It's polite to up mmap_sem if we can */
1971 up_read(&vma
->vm_mm
->mmap_sem
);
1972 ret
= VM_FAULT_RETRY
;
1975 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1976 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1977 TASK_UNINTERRUPTIBLE
);
1978 spin_unlock(&inode
->i_lock
);
1982 * shmem_falloc_waitq points into the shmem_fallocate()
1983 * stack of the hole-punching task: shmem_falloc_waitq
1984 * is usually invalid by the time we reach here, but
1985 * finish_wait() does not dereference it in that case;
1986 * though i_lock needed lest racing with wake_up_all().
1988 spin_lock(&inode
->i_lock
);
1989 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1990 spin_unlock(&inode
->i_lock
);
1993 spin_unlock(&inode
->i_lock
);
1998 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
1999 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2001 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2004 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2005 gfp
, vma
, vmf
, &ret
);
2007 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
2011 unsigned long shmem_get_unmapped_area(struct file
*file
,
2012 unsigned long uaddr
, unsigned long len
,
2013 unsigned long pgoff
, unsigned long flags
)
2015 unsigned long (*get_area
)(struct file
*,
2016 unsigned long, unsigned long, unsigned long, unsigned long);
2018 unsigned long offset
;
2019 unsigned long inflated_len
;
2020 unsigned long inflated_addr
;
2021 unsigned long inflated_offset
;
2023 if (len
> TASK_SIZE
)
2026 get_area
= current
->mm
->get_unmapped_area
;
2027 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2029 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
2031 if (IS_ERR_VALUE(addr
))
2033 if (addr
& ~PAGE_MASK
)
2035 if (addr
> TASK_SIZE
- len
)
2038 if (shmem_huge
== SHMEM_HUGE_DENY
)
2040 if (len
< HPAGE_PMD_SIZE
)
2042 if (flags
& MAP_FIXED
)
2045 * Our priority is to support MAP_SHARED mapped hugely;
2046 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2047 * But if caller specified an address hint, respect that as before.
2052 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2053 struct super_block
*sb
;
2056 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2057 sb
= file_inode(file
)->i_sb
;
2060 * Called directly from mm/mmap.c, or drivers/char/mem.c
2061 * for "/dev/zero", to create a shared anonymous object.
2063 if (IS_ERR(shm_mnt
))
2065 sb
= shm_mnt
->mnt_sb
;
2067 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2071 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2072 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2074 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2077 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2078 if (inflated_len
> TASK_SIZE
)
2080 if (inflated_len
< len
)
2083 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
2084 if (IS_ERR_VALUE(inflated_addr
))
2086 if (inflated_addr
& ~PAGE_MASK
)
2089 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2090 inflated_addr
+= offset
- inflated_offset
;
2091 if (inflated_offset
> offset
)
2092 inflated_addr
+= HPAGE_PMD_SIZE
;
2094 if (inflated_addr
> TASK_SIZE
- len
)
2096 return inflated_addr
;
2100 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2102 struct inode
*inode
= file_inode(vma
->vm_file
);
2103 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2106 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2109 struct inode
*inode
= file_inode(vma
->vm_file
);
2112 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2113 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2117 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2119 struct inode
*inode
= file_inode(file
);
2120 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2121 int retval
= -ENOMEM
;
2123 spin_lock_irq(&info
->lock
);
2124 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2125 if (!user_shm_lock(inode
->i_size
, user
))
2127 info
->flags
|= VM_LOCKED
;
2128 mapping_set_unevictable(file
->f_mapping
);
2130 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2131 user_shm_unlock(inode
->i_size
, user
);
2132 info
->flags
&= ~VM_LOCKED
;
2133 mapping_clear_unevictable(file
->f_mapping
);
2138 spin_unlock_irq(&info
->lock
);
2142 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2144 file_accessed(file
);
2145 vma
->vm_ops
= &shmem_vm_ops
;
2146 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2147 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2148 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2149 khugepaged_enter(vma
, vma
->vm_flags
);
2154 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2155 umode_t mode
, dev_t dev
, unsigned long flags
)
2157 struct inode
*inode
;
2158 struct shmem_inode_info
*info
;
2159 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2162 if (shmem_reserve_inode(sb
))
2165 inode
= new_inode(sb
);
2167 inode_init_owner(inode
, dir
, mode
);
2168 inode
->i_blocks
= 0;
2169 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2170 inode
->i_generation
= get_seconds();
2171 info
= SHMEM_I(inode
);
2172 memset(info
, 0, (char *)inode
- (char *)info
);
2173 spin_lock_init(&info
->lock
);
2174 info
->seals
= F_SEAL_SEAL
;
2175 info
->flags
= flags
& VM_NORESERVE
;
2176 INIT_LIST_HEAD(&info
->shrinklist
);
2177 INIT_LIST_HEAD(&info
->swaplist
);
2178 simple_xattrs_init(&info
->xattrs
);
2179 cache_no_acl(inode
);
2181 switch (mode
& S_IFMT
) {
2183 inode
->i_op
= &shmem_special_inode_operations
;
2184 init_special_inode(inode
, mode
, dev
);
2187 inode
->i_mapping
->a_ops
= &shmem_aops
;
2188 inode
->i_op
= &shmem_inode_operations
;
2189 inode
->i_fop
= &shmem_file_operations
;
2190 mpol_shared_policy_init(&info
->policy
,
2191 shmem_get_sbmpol(sbinfo
));
2195 /* Some things misbehave if size == 0 on a directory */
2196 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2197 inode
->i_op
= &shmem_dir_inode_operations
;
2198 inode
->i_fop
= &simple_dir_operations
;
2202 * Must not load anything in the rbtree,
2203 * mpol_free_shared_policy will not be called.
2205 mpol_shared_policy_init(&info
->policy
, NULL
);
2209 if (!sbinfo
->idr_nouse
) {
2210 /* inum 0 and 1 are unused */
2211 mutex_lock(&sbinfo
->idr_lock
);
2212 ino
= idr_alloc(&sbinfo
->idr
, inode
, 2, INT_MAX
,
2216 mutex_unlock(&sbinfo
->idr_lock
);
2217 __insert_inode_hash(inode
, inode
->i_ino
);
2220 mutex_unlock(&sbinfo
->idr_lock
);
2222 /* shmem_free_inode() will be called */
2226 inode
->i_ino
= get_next_ino();
2228 shmem_free_inode(sb
);
2232 bool shmem_mapping(struct address_space
*mapping
)
2234 return mapping
->a_ops
== &shmem_aops
;
2237 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2239 struct vm_area_struct
*dst_vma
,
2240 unsigned long dst_addr
,
2241 unsigned long src_addr
,
2242 struct page
**pagep
)
2244 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2245 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2246 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2247 struct address_space
*mapping
= inode
->i_mapping
;
2248 gfp_t gfp
= mapping_gfp_mask(mapping
);
2249 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2250 struct mem_cgroup
*memcg
;
2254 pte_t _dst_pte
, *dst_pte
;
2258 if (shmem_acct_block(info
->flags
, 1))
2260 if (sbinfo
->max_blocks
) {
2261 if (percpu_counter_compare(&sbinfo
->used_blocks
,
2262 sbinfo
->max_blocks
) >= 0)
2263 goto out_unacct_blocks
;
2264 percpu_counter_inc(&sbinfo
->used_blocks
);
2268 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2270 goto out_dec_used_blocks
;
2272 page_kaddr
= kmap_atomic(page
);
2273 ret
= copy_from_user(page_kaddr
, (const void __user
*)src_addr
,
2275 kunmap_atomic(page_kaddr
);
2277 /* fallback to copy_from_user outside mmap_sem */
2278 if (unlikely(ret
)) {
2280 if (sbinfo
->max_blocks
)
2281 percpu_counter_add(&sbinfo
->used_blocks
, -1);
2282 shmem_unacct_blocks(info
->flags
, 1);
2283 /* don't free the page */
2291 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2292 __SetPageLocked(page
);
2293 __SetPageSwapBacked(page
);
2294 __SetPageUptodate(page
);
2296 ret
= mem_cgroup_try_charge(page
, dst_mm
, gfp
, &memcg
, false);
2300 ret
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
2302 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
);
2303 radix_tree_preload_end();
2306 goto out_release_uncharge
;
2308 mem_cgroup_commit_charge(page
, memcg
, false, false);
2310 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2311 if (dst_vma
->vm_flags
& VM_WRITE
)
2312 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2315 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2316 if (!pte_none(*dst_pte
))
2317 goto out_release_uncharge_unlock
;
2319 lru_cache_add_anon(page
);
2321 spin_lock(&info
->lock
);
2323 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2324 shmem_recalc_inode(inode
);
2325 spin_unlock(&info
->lock
);
2327 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2328 page_add_file_rmap(page
, false);
2329 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2331 /* No need to invalidate - it was non-present before */
2332 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2334 pte_unmap_unlock(dst_pte
, ptl
);
2338 out_release_uncharge_unlock
:
2339 pte_unmap_unlock(dst_pte
, ptl
);
2340 out_release_uncharge
:
2341 mem_cgroup_cancel_charge(page
, memcg
, false);
2345 out_dec_used_blocks
:
2346 if (sbinfo
->max_blocks
)
2347 percpu_counter_add(&sbinfo
->used_blocks
, -1);
2349 shmem_unacct_blocks(info
->flags
, 1);
2354 static const struct inode_operations shmem_symlink_inode_operations
;
2355 static const struct inode_operations shmem_short_symlink_operations
;
2357 #ifdef CONFIG_TMPFS_XATTR
2358 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2360 #define shmem_initxattrs NULL
2364 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2365 loff_t pos
, unsigned len
, unsigned flags
,
2366 struct page
**pagep
, void **fsdata
)
2368 struct inode
*inode
= mapping
->host
;
2369 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2370 pgoff_t index
= pos
>> PAGE_SHIFT
;
2372 /* i_mutex is held by caller */
2373 if (unlikely(info
->seals
& (F_SEAL_WRITE
| F_SEAL_GROW
))) {
2374 if (info
->seals
& F_SEAL_WRITE
)
2376 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2380 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2384 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2385 loff_t pos
, unsigned len
, unsigned copied
,
2386 struct page
*page
, void *fsdata
)
2388 struct inode
*inode
= mapping
->host
;
2390 if (pos
+ copied
> inode
->i_size
)
2391 i_size_write(inode
, pos
+ copied
);
2393 if (!PageUptodate(page
)) {
2394 struct page
*head
= compound_head(page
);
2395 if (PageTransCompound(page
)) {
2398 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2399 if (head
+ i
== page
)
2401 clear_highpage(head
+ i
);
2402 flush_dcache_page(head
+ i
);
2405 if (copied
< PAGE_SIZE
) {
2406 unsigned from
= pos
& (PAGE_SIZE
- 1);
2407 zero_user_segments(page
, 0, from
,
2408 from
+ copied
, PAGE_SIZE
);
2410 SetPageUptodate(head
);
2412 set_page_dirty(page
);
2419 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2421 struct file
*file
= iocb
->ki_filp
;
2422 struct inode
*inode
= file_inode(file
);
2423 struct address_space
*mapping
= inode
->i_mapping
;
2425 unsigned long offset
;
2426 enum sgp_type sgp
= SGP_READ
;
2429 loff_t
*ppos
= &iocb
->ki_pos
;
2432 * Might this read be for a stacking filesystem? Then when reading
2433 * holes of a sparse file, we actually need to allocate those pages,
2434 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2436 if (!iter_is_iovec(to
))
2439 index
= *ppos
>> PAGE_SHIFT
;
2440 offset
= *ppos
& ~PAGE_MASK
;
2443 struct page
*page
= NULL
;
2445 unsigned long nr
, ret
;
2446 loff_t i_size
= i_size_read(inode
);
2448 end_index
= i_size
>> PAGE_SHIFT
;
2449 if (index
> end_index
)
2451 if (index
== end_index
) {
2452 nr
= i_size
& ~PAGE_MASK
;
2457 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2459 if (error
== -EINVAL
)
2464 if (sgp
== SGP_CACHE
)
2465 set_page_dirty(page
);
2470 * We must evaluate after, since reads (unlike writes)
2471 * are called without i_mutex protection against truncate
2474 i_size
= i_size_read(inode
);
2475 end_index
= i_size
>> PAGE_SHIFT
;
2476 if (index
== end_index
) {
2477 nr
= i_size
& ~PAGE_MASK
;
2488 * If users can be writing to this page using arbitrary
2489 * virtual addresses, take care about potential aliasing
2490 * before reading the page on the kernel side.
2492 if (mapping_writably_mapped(mapping
))
2493 flush_dcache_page(page
);
2495 * Mark the page accessed if we read the beginning.
2498 mark_page_accessed(page
);
2500 page
= ZERO_PAGE(0);
2505 * Ok, we have the page, and it's up-to-date, so
2506 * now we can copy it to user space...
2508 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2511 index
+= offset
>> PAGE_SHIFT
;
2512 offset
&= ~PAGE_MASK
;
2515 if (!iov_iter_count(to
))
2524 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2525 file_accessed(file
);
2526 return retval
? retval
: error
;
2530 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2532 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2533 pgoff_t index
, pgoff_t end
, int whence
)
2536 struct pagevec pvec
;
2537 pgoff_t indices
[PAGEVEC_SIZE
];
2541 pagevec_init(&pvec
, 0);
2542 pvec
.nr
= 1; /* start small: we may be there already */
2544 pvec
.nr
= find_get_entries(mapping
, index
,
2545 pvec
.nr
, pvec
.pages
, indices
);
2547 if (whence
== SEEK_DATA
)
2551 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2552 if (index
< indices
[i
]) {
2553 if (whence
== SEEK_HOLE
) {
2559 page
= pvec
.pages
[i
];
2560 if (page
&& !radix_tree_exceptional_entry(page
)) {
2561 if (!PageUptodate(page
))
2565 (page
&& whence
== SEEK_DATA
) ||
2566 (!page
&& whence
== SEEK_HOLE
)) {
2571 pagevec_remove_exceptionals(&pvec
);
2572 pagevec_release(&pvec
);
2573 pvec
.nr
= PAGEVEC_SIZE
;
2579 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2581 struct address_space
*mapping
= file
->f_mapping
;
2582 struct inode
*inode
= mapping
->host
;
2586 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2587 return generic_file_llseek_size(file
, offset
, whence
,
2588 MAX_LFS_FILESIZE
, i_size_read(inode
));
2590 /* We're holding i_mutex so we can access i_size directly */
2594 else if (offset
>= inode
->i_size
)
2597 start
= offset
>> PAGE_SHIFT
;
2598 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2599 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2600 new_offset
<<= PAGE_SHIFT
;
2601 if (new_offset
> offset
) {
2602 if (new_offset
< inode
->i_size
)
2603 offset
= new_offset
;
2604 else if (whence
== SEEK_DATA
)
2607 offset
= inode
->i_size
;
2612 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2613 inode_unlock(inode
);
2618 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2619 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2621 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2622 #define LAST_SCAN 4 /* about 150ms max */
2624 static void shmem_tag_pins(struct address_space
*mapping
)
2626 struct radix_tree_iter iter
;
2635 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2636 page
= radix_tree_deref_slot(slot
);
2637 if (!page
|| radix_tree_exception(page
)) {
2638 if (radix_tree_deref_retry(page
)) {
2639 slot
= radix_tree_iter_retry(&iter
);
2642 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2643 spin_lock_irq(&mapping
->tree_lock
);
2644 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2646 spin_unlock_irq(&mapping
->tree_lock
);
2649 if (need_resched()) {
2650 slot
= radix_tree_iter_resume(slot
, &iter
);
2658 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2659 * via get_user_pages(), drivers might have some pending I/O without any active
2660 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2661 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2662 * them to be dropped.
2663 * The caller must guarantee that no new user will acquire writable references
2664 * to those pages to avoid races.
2666 static int shmem_wait_for_pins(struct address_space
*mapping
)
2668 struct radix_tree_iter iter
;
2674 shmem_tag_pins(mapping
);
2677 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2678 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2682 lru_add_drain_all();
2683 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2688 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2689 start
, SHMEM_TAG_PINNED
) {
2691 page
= radix_tree_deref_slot(slot
);
2692 if (radix_tree_exception(page
)) {
2693 if (radix_tree_deref_retry(page
)) {
2694 slot
= radix_tree_iter_retry(&iter
);
2702 page_count(page
) - page_mapcount(page
) != 1) {
2703 if (scan
< LAST_SCAN
)
2704 goto continue_resched
;
2707 * On the last scan, we clean up all those tags
2708 * we inserted; but make a note that we still
2709 * found pages pinned.
2714 spin_lock_irq(&mapping
->tree_lock
);
2715 radix_tree_tag_clear(&mapping
->page_tree
,
2716 iter
.index
, SHMEM_TAG_PINNED
);
2717 spin_unlock_irq(&mapping
->tree_lock
);
2719 if (need_resched()) {
2720 slot
= radix_tree_iter_resume(slot
, &iter
);
2730 #define F_ALL_SEALS (F_SEAL_SEAL | \
2735 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2737 struct inode
*inode
= file_inode(file
);
2738 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2743 * Sealing allows multiple parties to share a shmem-file but restrict
2744 * access to a specific subset of file operations. Seals can only be
2745 * added, but never removed. This way, mutually untrusted parties can
2746 * share common memory regions with a well-defined policy. A malicious
2747 * peer can thus never perform unwanted operations on a shared object.
2749 * Seals are only supported on special shmem-files and always affect
2750 * the whole underlying inode. Once a seal is set, it may prevent some
2751 * kinds of access to the file. Currently, the following seals are
2753 * SEAL_SEAL: Prevent further seals from being set on this file
2754 * SEAL_SHRINK: Prevent the file from shrinking
2755 * SEAL_GROW: Prevent the file from growing
2756 * SEAL_WRITE: Prevent write access to the file
2758 * As we don't require any trust relationship between two parties, we
2759 * must prevent seals from being removed. Therefore, sealing a file
2760 * only adds a given set of seals to the file, it never touches
2761 * existing seals. Furthermore, the "setting seals"-operation can be
2762 * sealed itself, which basically prevents any further seal from being
2765 * Semantics of sealing are only defined on volatile files. Only
2766 * anonymous shmem files support sealing. More importantly, seals are
2767 * never written to disk. Therefore, there's no plan to support it on
2771 if (file
->f_op
!= &shmem_file_operations
)
2773 if (!(file
->f_mode
& FMODE_WRITE
))
2775 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2780 if (info
->seals
& F_SEAL_SEAL
) {
2785 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2786 error
= mapping_deny_writable(file
->f_mapping
);
2790 error
= shmem_wait_for_pins(file
->f_mapping
);
2792 mapping_allow_writable(file
->f_mapping
);
2797 info
->seals
|= seals
;
2801 inode_unlock(inode
);
2804 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2806 int shmem_get_seals(struct file
*file
)
2808 if (file
->f_op
!= &shmem_file_operations
)
2811 return SHMEM_I(file_inode(file
))->seals
;
2813 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2815 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2821 /* disallow upper 32bit */
2825 error
= shmem_add_seals(file
, arg
);
2828 error
= shmem_get_seals(file
);
2838 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2841 struct inode
*inode
= file_inode(file
);
2842 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2843 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2844 struct shmem_falloc shmem_falloc
;
2845 pgoff_t start
, index
, end
;
2848 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2853 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2854 struct address_space
*mapping
= file
->f_mapping
;
2855 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2856 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2857 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2859 /* protected by i_mutex */
2860 if (info
->seals
& F_SEAL_WRITE
) {
2865 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2866 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2867 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2868 spin_lock(&inode
->i_lock
);
2869 inode
->i_private
= &shmem_falloc
;
2870 spin_unlock(&inode
->i_lock
);
2872 if ((u64
)unmap_end
> (u64
)unmap_start
)
2873 unmap_mapping_range(mapping
, unmap_start
,
2874 1 + unmap_end
- unmap_start
, 0);
2875 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2876 /* No need to unmap again: hole-punching leaves COWed pages */
2878 spin_lock(&inode
->i_lock
);
2879 inode
->i_private
= NULL
;
2880 wake_up_all(&shmem_falloc_waitq
);
2881 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2882 spin_unlock(&inode
->i_lock
);
2887 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2888 error
= inode_newsize_ok(inode
, offset
+ len
);
2892 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2897 start
= offset
>> PAGE_SHIFT
;
2898 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2899 /* Try to avoid a swapstorm if len is impossible to satisfy */
2900 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2905 shmem_falloc
.waitq
= NULL
;
2906 shmem_falloc
.start
= start
;
2907 shmem_falloc
.next
= start
;
2908 shmem_falloc
.nr_falloced
= 0;
2909 shmem_falloc
.nr_unswapped
= 0;
2910 spin_lock(&inode
->i_lock
);
2911 inode
->i_private
= &shmem_falloc
;
2912 spin_unlock(&inode
->i_lock
);
2914 for (index
= start
; index
< end
; index
++) {
2918 * Good, the fallocate(2) manpage permits EINTR: we may have
2919 * been interrupted because we are using up too much memory.
2921 if (signal_pending(current
))
2923 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2926 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2928 /* Remove the !PageUptodate pages we added */
2929 if (index
> start
) {
2930 shmem_undo_range(inode
,
2931 (loff_t
)start
<< PAGE_SHIFT
,
2932 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2938 * Inform shmem_writepage() how far we have reached.
2939 * No need for lock or barrier: we have the page lock.
2941 shmem_falloc
.next
++;
2942 if (!PageUptodate(page
))
2943 shmem_falloc
.nr_falloced
++;
2946 * If !PageUptodate, leave it that way so that freeable pages
2947 * can be recognized if we need to rollback on error later.
2948 * But set_page_dirty so that memory pressure will swap rather
2949 * than free the pages we are allocating (and SGP_CACHE pages
2950 * might still be clean: we now need to mark those dirty too).
2952 set_page_dirty(page
);
2958 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2959 i_size_write(inode
, offset
+ len
);
2960 inode
->i_ctime
= current_time(inode
);
2962 spin_lock(&inode
->i_lock
);
2963 inode
->i_private
= NULL
;
2964 spin_unlock(&inode
->i_lock
);
2966 inode_unlock(inode
);
2970 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2972 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2974 buf
->f_type
= TMPFS_MAGIC
;
2975 buf
->f_bsize
= PAGE_SIZE
;
2976 buf
->f_namelen
= NAME_MAX
;
2977 if (sbinfo
->max_blocks
) {
2978 buf
->f_blocks
= sbinfo
->max_blocks
;
2980 buf
->f_bfree
= sbinfo
->max_blocks
-
2981 percpu_counter_sum(&sbinfo
->used_blocks
);
2983 if (sbinfo
->max_inodes
) {
2984 buf
->f_files
= sbinfo
->max_inodes
;
2985 buf
->f_ffree
= sbinfo
->free_inodes
;
2987 /* else leave those fields 0 like simple_statfs */
2992 * File creation. Allocate an inode, and we're done..
2995 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2997 struct inode
*inode
;
2998 int error
= -ENOSPC
;
3000 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
3002 error
= simple_acl_create(dir
, inode
);
3005 error
= security_inode_init_security(inode
, dir
,
3007 shmem_initxattrs
, NULL
);
3008 if (error
&& error
!= -EOPNOTSUPP
)
3012 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3013 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3014 d_instantiate(dentry
, inode
);
3015 dget(dentry
); /* Extra count - pin the dentry in core */
3024 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3026 struct inode
*inode
;
3027 int error
= -ENOSPC
;
3029 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
3031 error
= security_inode_init_security(inode
, dir
,
3033 shmem_initxattrs
, NULL
);
3034 if (error
&& error
!= -EOPNOTSUPP
)
3036 error
= simple_acl_create(dir
, inode
);
3039 d_tmpfile(dentry
, inode
);
3047 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3051 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
3057 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
3060 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
3066 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
3068 struct inode
*inode
= d_inode(old_dentry
);
3072 * No ordinary (disk based) filesystem counts links as inodes;
3073 * but each new link needs a new dentry, pinning lowmem, and
3074 * tmpfs dentries cannot be pruned until they are unlinked.
3076 ret
= shmem_reserve_inode(inode
->i_sb
);
3080 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3081 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3083 ihold(inode
); /* New dentry reference */
3084 dget(dentry
); /* Extra pinning count for the created dentry */
3085 d_instantiate(dentry
, inode
);
3090 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3092 struct inode
*inode
= d_inode(dentry
);
3094 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3095 shmem_free_inode(inode
->i_sb
);
3097 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3098 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3100 dput(dentry
); /* Undo the count from "create" - this does all the work */
3104 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3106 if (!simple_empty(dentry
))
3109 drop_nlink(d_inode(dentry
));
3111 return shmem_unlink(dir
, dentry
);
3114 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3116 bool old_is_dir
= d_is_dir(old_dentry
);
3117 bool new_is_dir
= d_is_dir(new_dentry
);
3119 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3121 drop_nlink(old_dir
);
3124 drop_nlink(new_dir
);
3128 old_dir
->i_ctime
= old_dir
->i_mtime
=
3129 new_dir
->i_ctime
= new_dir
->i_mtime
=
3130 d_inode(old_dentry
)->i_ctime
=
3131 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3136 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3138 struct dentry
*whiteout
;
3141 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3145 error
= shmem_mknod(old_dir
, whiteout
,
3146 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3152 * Cheat and hash the whiteout while the old dentry is still in
3153 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3155 * d_lookup() will consistently find one of them at this point,
3156 * not sure which one, but that isn't even important.
3163 * The VFS layer already does all the dentry stuff for rename,
3164 * we just have to decrement the usage count for the target if
3165 * it exists so that the VFS layer correctly free's it when it
3168 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3170 struct inode
*inode
= d_inode(old_dentry
);
3171 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3173 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3176 if (flags
& RENAME_EXCHANGE
)
3177 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3179 if (!simple_empty(new_dentry
))
3182 if (flags
& RENAME_WHITEOUT
) {
3185 error
= shmem_whiteout(old_dir
, old_dentry
);
3190 if (d_really_is_positive(new_dentry
)) {
3191 (void) shmem_unlink(new_dir
, new_dentry
);
3192 if (they_are_dirs
) {
3193 drop_nlink(d_inode(new_dentry
));
3194 drop_nlink(old_dir
);
3196 } else if (they_are_dirs
) {
3197 drop_nlink(old_dir
);
3201 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3202 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3203 old_dir
->i_ctime
= old_dir
->i_mtime
=
3204 new_dir
->i_ctime
= new_dir
->i_mtime
=
3205 inode
->i_ctime
= current_time(old_dir
);
3209 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3213 struct inode
*inode
;
3215 struct shmem_inode_info
*info
;
3217 len
= strlen(symname
) + 1;
3218 if (len
> PAGE_SIZE
)
3219 return -ENAMETOOLONG
;
3221 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
3225 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3226 shmem_initxattrs
, NULL
);
3228 if (error
!= -EOPNOTSUPP
) {
3235 info
= SHMEM_I(inode
);
3236 inode
->i_size
= len
-1;
3237 if (len
<= SHORT_SYMLINK_LEN
) {
3238 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3239 if (!inode
->i_link
) {
3243 inode
->i_op
= &shmem_short_symlink_operations
;
3245 inode_nohighmem(inode
);
3246 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3251 inode
->i_mapping
->a_ops
= &shmem_aops
;
3252 inode
->i_op
= &shmem_symlink_inode_operations
;
3253 memcpy(page_address(page
), symname
, len
);
3254 SetPageUptodate(page
);
3255 set_page_dirty(page
);
3259 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3260 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3261 d_instantiate(dentry
, inode
);
3266 static void shmem_put_link(void *arg
)
3268 mark_page_accessed(arg
);
3272 static const char *shmem_get_link(struct dentry
*dentry
,
3273 struct inode
*inode
,
3274 struct delayed_call
*done
)
3276 struct page
*page
= NULL
;
3279 page
= find_get_page(inode
->i_mapping
, 0);
3281 return ERR_PTR(-ECHILD
);
3282 if (!PageUptodate(page
)) {
3284 return ERR_PTR(-ECHILD
);
3287 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3289 return ERR_PTR(error
);
3292 set_delayed_call(done
, shmem_put_link
, page
);
3293 return page_address(page
);
3296 #ifdef CONFIG_TMPFS_XATTR
3298 * Superblocks without xattr inode operations may get some security.* xattr
3299 * support from the LSM "for free". As soon as we have any other xattrs
3300 * like ACLs, we also need to implement the security.* handlers at
3301 * filesystem level, though.
3305 * Callback for security_inode_init_security() for acquiring xattrs.
3307 static int shmem_initxattrs(struct inode
*inode
,
3308 const struct xattr
*xattr_array
,
3311 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3312 const struct xattr
*xattr
;
3313 struct simple_xattr
*new_xattr
;
3316 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3317 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3321 len
= strlen(xattr
->name
) + 1;
3322 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3324 if (!new_xattr
->name
) {
3329 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3330 XATTR_SECURITY_PREFIX_LEN
);
3331 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3334 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3340 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3341 struct dentry
*unused
, struct inode
*inode
,
3342 const char *name
, void *buffer
, size_t size
)
3344 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3346 name
= xattr_full_name(handler
, name
);
3347 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3350 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3351 struct dentry
*unused
, struct inode
*inode
,
3352 const char *name
, const void *value
,
3353 size_t size
, int flags
)
3355 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3357 name
= xattr_full_name(handler
, name
);
3358 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3361 static const struct xattr_handler shmem_security_xattr_handler
= {
3362 .prefix
= XATTR_SECURITY_PREFIX
,
3363 .get
= shmem_xattr_handler_get
,
3364 .set
= shmem_xattr_handler_set
,
3367 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3368 .prefix
= XATTR_TRUSTED_PREFIX
,
3369 .get
= shmem_xattr_handler_get
,
3370 .set
= shmem_xattr_handler_set
,
3373 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3374 #ifdef CONFIG_TMPFS_POSIX_ACL
3375 &posix_acl_access_xattr_handler
,
3376 &posix_acl_default_xattr_handler
,
3378 &shmem_security_xattr_handler
,
3379 &shmem_trusted_xattr_handler
,
3383 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3385 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3386 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3388 #endif /* CONFIG_TMPFS_XATTR */
3390 static const struct inode_operations shmem_short_symlink_operations
= {
3391 .get_link
= simple_get_link
,
3392 #ifdef CONFIG_TMPFS_XATTR
3393 .listxattr
= shmem_listxattr
,
3397 static const struct inode_operations shmem_symlink_inode_operations
= {
3398 .get_link
= shmem_get_link
,
3399 #ifdef CONFIG_TMPFS_XATTR
3400 .listxattr
= shmem_listxattr
,
3404 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3406 return ERR_PTR(-ESTALE
);
3409 static int shmem_match(struct inode
*ino
, void *vfh
)
3413 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3416 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3417 struct fid
*fid
, int fh_len
, int fh_type
)
3419 struct inode
*inode
;
3420 struct dentry
*dentry
= NULL
;
3427 inode
= ilookup5(sb
, inum
, shmem_match
, fid
->raw
);
3429 dentry
= d_find_alias(inode
);
3436 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3437 struct inode
*parent
)
3441 return FILEID_INVALID
;
3444 fh
[0] = inode
->i_generation
;
3445 fh
[1] = inode
->i_ino
;
3451 static const struct export_operations shmem_export_ops
= {
3452 .get_parent
= shmem_get_parent
,
3453 .encode_fh
= shmem_encode_fh
,
3454 .fh_to_dentry
= shmem_fh_to_dentry
,
3457 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3460 char *this_char
, *value
, *rest
;
3461 struct mempolicy
*mpol
= NULL
;
3465 while (options
!= NULL
) {
3466 this_char
= options
;
3469 * NUL-terminate this option: unfortunately,
3470 * mount options form a comma-separated list,
3471 * but mpol's nodelist may also contain commas.
3473 options
= strchr(options
, ',');
3474 if (options
== NULL
)
3477 if (!isdigit(*options
)) {
3484 if ((value
= strchr(this_char
,'=')) != NULL
) {
3487 pr_err("tmpfs: No value for mount option '%s'\n",
3492 if (!strcmp(this_char
,"size")) {
3493 unsigned long long size
;
3494 size
= memparse(value
,&rest
);
3496 size
<<= PAGE_SHIFT
;
3497 size
*= totalram_pages
;
3503 sbinfo
->max_blocks
=
3504 DIV_ROUND_UP(size
, PAGE_SIZE
);
3505 } else if (!strcmp(this_char
,"nr_blocks")) {
3506 sbinfo
->max_blocks
= memparse(value
, &rest
);
3509 } else if (!strcmp(this_char
,"nr_inodes")) {
3510 sbinfo
->max_inodes
= memparse(value
, &rest
);
3511 if (*rest
|| sbinfo
->max_inodes
< 2)
3513 } else if (!strcmp(this_char
,"mode")) {
3516 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3519 } else if (!strcmp(this_char
,"uid")) {
3522 uid
= simple_strtoul(value
, &rest
, 0);
3525 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3526 if (!uid_valid(sbinfo
->uid
))
3528 } else if (!strcmp(this_char
,"gid")) {
3531 gid
= simple_strtoul(value
, &rest
, 0);
3534 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3535 if (!gid_valid(sbinfo
->gid
))
3537 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3538 } else if (!strcmp(this_char
, "huge")) {
3540 huge
= shmem_parse_huge(value
);
3543 if (!has_transparent_hugepage() &&
3544 huge
!= SHMEM_HUGE_NEVER
)
3546 sbinfo
->huge
= huge
;
3549 } else if (!strcmp(this_char
,"mpol")) {
3552 if (mpol_parse_str(value
, &mpol
))
3556 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3560 sbinfo
->mpol
= mpol
;
3564 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3572 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3574 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3575 struct shmem_sb_info config
= *sbinfo
;
3577 int error
= -EINVAL
;
3580 if (shmem_parse_options(data
, &config
, true))
3583 spin_lock(&sbinfo
->stat_lock
);
3584 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3585 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3587 if (config
.max_inodes
< inodes
)
3590 * Those tests disallow limited->unlimited while any are in use;
3591 * but we must separately disallow unlimited->limited, because
3592 * in that case we have no record of how much is already in use.
3594 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3596 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3600 sbinfo
->huge
= config
.huge
;
3601 sbinfo
->max_blocks
= config
.max_blocks
;
3602 sbinfo
->max_inodes
= config
.max_inodes
;
3603 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3606 * Preserve previous mempolicy unless mpol remount option was specified.
3609 mpol_put(sbinfo
->mpol
);
3610 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3613 spin_unlock(&sbinfo
->stat_lock
);
3617 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3619 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3621 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3622 seq_printf(seq
, ",size=%luk",
3623 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3624 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3625 seq_printf(seq
, ",nr_inodes=%d", sbinfo
->max_inodes
);
3626 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3627 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3628 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3629 seq_printf(seq
, ",uid=%u",
3630 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3631 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3632 seq_printf(seq
, ",gid=%u",
3633 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3634 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3635 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3637 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3639 shmem_show_mpol(seq
, sbinfo
->mpol
);
3643 #define MFD_NAME_PREFIX "memfd:"
3644 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3645 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3647 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3649 SYSCALL_DEFINE2(memfd_create
,
3650 const char __user
*, uname
,
3651 unsigned int, flags
)
3653 struct shmem_inode_info
*info
;
3659 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3662 /* length includes terminating zero */
3663 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3666 if (len
> MFD_NAME_MAX_LEN
+ 1)
3669 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
3673 strcpy(name
, MFD_NAME_PREFIX
);
3674 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3679 /* terminating-zero may have changed after strnlen_user() returned */
3680 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3685 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3691 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3693 error
= PTR_ERR(file
);
3696 info
= SHMEM_I(file_inode(file
));
3697 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3698 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3699 if (flags
& MFD_ALLOW_SEALING
)
3700 info
->seals
&= ~F_SEAL_SEAL
;
3702 fd_install(fd
, file
);
3713 #endif /* CONFIG_TMPFS */
3715 static void shmem_put_super(struct super_block
*sb
)
3717 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3719 if (!sbinfo
->idr_nouse
)
3720 idr_destroy(&sbinfo
->idr
);
3721 percpu_counter_destroy(&sbinfo
->used_blocks
);
3722 mpol_put(sbinfo
->mpol
);
3724 sb
->s_fs_info
= NULL
;
3727 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3729 struct inode
*inode
;
3730 struct shmem_sb_info
*sbinfo
;
3733 /* Round up to L1_CACHE_BYTES to resist false sharing */
3734 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3735 L1_CACHE_BYTES
), GFP_KERNEL
);
3739 mutex_init(&sbinfo
->idr_lock
);
3740 idr_init(&sbinfo
->idr
);
3741 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3742 sbinfo
->uid
= current_fsuid();
3743 sbinfo
->gid
= current_fsgid();
3744 sb
->s_fs_info
= sbinfo
;
3748 * Per default we only allow half of the physical ram per
3749 * tmpfs instance, limiting inodes to one per page of lowmem;
3750 * but the internal instance is left unlimited.
3752 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3753 sbinfo
->max_blocks
= shmem_default_max_blocks();
3754 sbinfo
->max_inodes
= shmem_default_max_inodes();
3755 if (shmem_parse_options(data
, sbinfo
, false)) {
3760 sb
->s_flags
|= MS_NOUSER
;
3762 sb
->s_export_op
= &shmem_export_ops
;
3763 sb
->s_flags
|= MS_NOSEC
;
3765 sb
->s_flags
|= MS_NOUSER
;
3768 spin_lock_init(&sbinfo
->stat_lock
);
3769 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3771 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3772 spin_lock_init(&sbinfo
->shrinklist_lock
);
3773 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3775 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3776 sb
->s_blocksize
= PAGE_SIZE
;
3777 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3778 sb
->s_magic
= TMPFS_MAGIC
;
3779 sb
->s_op
= &shmem_ops
;
3780 sb
->s_time_gran
= 1;
3781 #ifdef CONFIG_TMPFS_XATTR
3782 sb
->s_xattr
= shmem_xattr_handlers
;
3784 #ifdef CONFIG_TMPFS_POSIX_ACL
3785 sb
->s_flags
|= MS_POSIXACL
;
3787 uuid_gen(&sb
->s_uuid
);
3789 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3792 inode
->i_uid
= sbinfo
->uid
;
3793 inode
->i_gid
= sbinfo
->gid
;
3794 sb
->s_root
= d_make_root(inode
);
3800 shmem_put_super(sb
);
3804 static struct kmem_cache
*shmem_inode_cachep
;
3806 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3808 struct shmem_inode_info
*info
;
3809 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3812 return &info
->vfs_inode
;
3815 static void shmem_destroy_callback(struct rcu_head
*head
)
3817 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3818 if (S_ISLNK(inode
->i_mode
))
3819 kfree(inode
->i_link
);
3820 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3823 static void shmem_destroy_inode(struct inode
*inode
)
3825 if (S_ISREG(inode
->i_mode
))
3826 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3827 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3830 static void shmem_init_inode(void *foo
)
3832 struct shmem_inode_info
*info
= foo
;
3833 inode_init_once(&info
->vfs_inode
);
3836 static int shmem_init_inodecache(void)
3838 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3839 sizeof(struct shmem_inode_info
),
3840 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3844 static void shmem_destroy_inodecache(void)
3846 kmem_cache_destroy(shmem_inode_cachep
);
3849 static __init
void shmem_no_idr(struct super_block
*sb
)
3851 struct shmem_sb_info
*sbinfo
;
3853 sbinfo
= SHMEM_SB(sb
);
3854 sbinfo
->idr_nouse
= true;
3855 idr_destroy(&sbinfo
->idr
);
3858 static const struct address_space_operations shmem_aops
= {
3859 .writepage
= shmem_writepage
,
3860 .set_page_dirty
= __set_page_dirty_no_writeback
,
3862 .write_begin
= shmem_write_begin
,
3863 .write_end
= shmem_write_end
,
3865 #ifdef CONFIG_MIGRATION
3866 .migratepage
= migrate_page
,
3868 .error_remove_page
= generic_error_remove_page
,
3871 static const struct file_operations shmem_file_operations
= {
3873 .get_unmapped_area
= shmem_get_unmapped_area
,
3875 .llseek
= shmem_file_llseek
,
3876 .read_iter
= shmem_file_read_iter
,
3877 .write_iter
= generic_file_write_iter
,
3878 .fsync
= noop_fsync
,
3879 .splice_read
= generic_file_splice_read
,
3880 .splice_write
= iter_file_splice_write
,
3881 .fallocate
= shmem_fallocate
,
3885 static const struct inode_operations shmem_inode_operations
= {
3886 .getattr
= shmem_getattr
,
3887 .setattr
= shmem_setattr
,
3888 #ifdef CONFIG_TMPFS_XATTR
3889 .listxattr
= shmem_listxattr
,
3890 .set_acl
= simple_set_acl
,
3894 static const struct inode_operations shmem_dir_inode_operations
= {
3896 .create
= shmem_create
,
3897 .lookup
= simple_lookup
,
3899 .unlink
= shmem_unlink
,
3900 .symlink
= shmem_symlink
,
3901 .mkdir
= shmem_mkdir
,
3902 .rmdir
= shmem_rmdir
,
3903 .mknod
= shmem_mknod
,
3904 .rename
= shmem_rename2
,
3905 .tmpfile
= shmem_tmpfile
,
3907 #ifdef CONFIG_TMPFS_XATTR
3908 .listxattr
= shmem_listxattr
,
3910 #ifdef CONFIG_TMPFS_POSIX_ACL
3911 .setattr
= shmem_setattr
,
3912 .set_acl
= simple_set_acl
,
3916 static const struct inode_operations shmem_special_inode_operations
= {
3917 #ifdef CONFIG_TMPFS_XATTR
3918 .listxattr
= shmem_listxattr
,
3920 #ifdef CONFIG_TMPFS_POSIX_ACL
3921 .setattr
= shmem_setattr
,
3922 .set_acl
= simple_set_acl
,
3926 static const struct super_operations shmem_ops
= {
3927 .alloc_inode
= shmem_alloc_inode
,
3928 .destroy_inode
= shmem_destroy_inode
,
3930 .statfs
= shmem_statfs
,
3931 .remount_fs
= shmem_remount_fs
,
3932 .show_options
= shmem_show_options
,
3934 .evict_inode
= shmem_evict_inode
,
3935 .drop_inode
= generic_delete_inode
,
3936 .put_super
= shmem_put_super
,
3937 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3938 .nr_cached_objects
= shmem_unused_huge_count
,
3939 .free_cached_objects
= shmem_unused_huge_scan
,
3943 static const struct vm_operations_struct shmem_vm_ops
= {
3944 .fault
= shmem_fault
,
3945 .map_pages
= filemap_map_pages
,
3947 .set_policy
= shmem_set_policy
,
3948 .get_policy
= shmem_get_policy
,
3952 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3953 int flags
, const char *dev_name
, void *data
)
3955 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3958 static struct file_system_type shmem_fs_type
= {
3959 .owner
= THIS_MODULE
,
3961 .mount
= shmem_mount
,
3962 .kill_sb
= kill_litter_super
,
3963 .fs_flags
= FS_USERNS_MOUNT
,
3966 int __init
shmem_init(void)
3970 /* If rootfs called this, don't re-init */
3971 if (shmem_inode_cachep
)
3974 error
= shmem_init_inodecache();
3978 error
= register_filesystem(&shmem_fs_type
);
3980 pr_err("Could not register tmpfs\n");
3984 shm_mnt
= kern_mount(&shmem_fs_type
);
3985 if (IS_ERR(shm_mnt
)) {
3986 error
= PTR_ERR(shm_mnt
);
3987 pr_err("Could not kern_mount tmpfs\n");
3990 shmem_no_idr(shm_mnt
->mnt_sb
);
3992 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3993 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3994 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3996 shmem_huge
= 0; /* just in case it was patched */
4001 unregister_filesystem(&shmem_fs_type
);
4003 shmem_destroy_inodecache();
4005 shm_mnt
= ERR_PTR(error
);
4009 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
4010 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4011 struct kobj_attribute
*attr
, char *buf
)
4015 SHMEM_HUGE_WITHIN_SIZE
,
4023 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
4024 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
4026 count
+= sprintf(buf
+ count
, fmt
,
4027 shmem_format_huge(values
[i
]));
4029 buf
[count
- 1] = '\n';
4033 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4034 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4039 if (count
+ 1 > sizeof(tmp
))
4041 memcpy(tmp
, buf
, count
);
4043 if (count
&& tmp
[count
- 1] == '\n')
4044 tmp
[count
- 1] = '\0';
4046 huge
= shmem_parse_huge(tmp
);
4047 if (huge
== -EINVAL
)
4049 if (!has_transparent_hugepage() &&
4050 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4054 if (shmem_huge
> SHMEM_HUGE_DENY
)
4055 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4059 struct kobj_attribute shmem_enabled_attr
=
4060 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4061 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4063 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4064 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4066 struct inode
*inode
= file_inode(vma
->vm_file
);
4067 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4071 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4073 if (shmem_huge
== SHMEM_HUGE_DENY
)
4075 switch (sbinfo
->huge
) {
4076 case SHMEM_HUGE_NEVER
:
4078 case SHMEM_HUGE_ALWAYS
:
4080 case SHMEM_HUGE_WITHIN_SIZE
:
4081 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4082 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4083 if (i_size
>= HPAGE_PMD_SIZE
&&
4084 i_size
>> PAGE_SHIFT
>= off
)
4086 case SHMEM_HUGE_ADVISE
:
4087 /* TODO: implement fadvise() hints */
4088 return (vma
->vm_flags
& VM_HUGEPAGE
);
4094 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4096 #else /* !CONFIG_SHMEM */
4099 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4101 * This is intended for small system where the benefits of the full
4102 * shmem code (swap-backed and resource-limited) are outweighed by
4103 * their complexity. On systems without swap this code should be
4104 * effectively equivalent, but much lighter weight.
4107 static struct file_system_type shmem_fs_type
= {
4109 .mount
= ramfs_mount
,
4110 .kill_sb
= kill_litter_super
,
4111 .fs_flags
= FS_USERNS_MOUNT
,
4114 int __init
shmem_init(void)
4116 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4118 shm_mnt
= kern_mount(&shmem_fs_type
);
4119 BUG_ON(IS_ERR(shm_mnt
));
4124 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
4129 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4134 void shmem_unlock_mapping(struct address_space
*mapping
)
4139 unsigned long shmem_get_unmapped_area(struct file
*file
,
4140 unsigned long addr
, unsigned long len
,
4141 unsigned long pgoff
, unsigned long flags
)
4143 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4147 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4149 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4151 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4153 #define shmem_vm_ops generic_file_vm_ops
4154 #define shmem_file_operations ramfs_file_operations
4155 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4156 #define shmem_acct_size(flags, size) 0
4157 #define shmem_unacct_size(flags, size) do {} while (0)
4159 #endif /* CONFIG_SHMEM */
4163 static const struct dentry_operations anon_ops
= {
4164 .d_dname
= simple_dname
4167 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
4168 unsigned long flags
, unsigned int i_flags
)
4171 struct inode
*inode
;
4173 struct super_block
*sb
;
4176 if (IS_ERR(shm_mnt
))
4177 return ERR_CAST(shm_mnt
);
4179 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4180 return ERR_PTR(-EINVAL
);
4182 if (shmem_acct_size(flags
, size
))
4183 return ERR_PTR(-ENOMEM
);
4185 res
= ERR_PTR(-ENOMEM
);
4187 this.len
= strlen(name
);
4188 this.hash
= 0; /* will go */
4189 sb
= shm_mnt
->mnt_sb
;
4190 path
.mnt
= mntget(shm_mnt
);
4191 path
.dentry
= d_alloc_pseudo(sb
, &this);
4194 d_set_d_op(path
.dentry
, &anon_ops
);
4196 res
= ERR_PTR(-ENOSPC
);
4197 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
4201 inode
->i_flags
|= i_flags
;
4202 d_instantiate(path
.dentry
, inode
);
4203 inode
->i_size
= size
;
4204 clear_nlink(inode
); /* It is unlinked */
4205 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4209 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
4210 &shmem_file_operations
);
4217 shmem_unacct_size(flags
, size
);
4224 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4225 * kernel internal. There will be NO LSM permission checks against the
4226 * underlying inode. So users of this interface must do LSM checks at a
4227 * higher layer. The users are the big_key and shm implementations. LSM
4228 * checks are provided at the key or shm level rather than the inode.
4229 * @name: name for dentry (to be seen in /proc/<pid>/maps
4230 * @size: size to be set for the file
4231 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4233 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4235 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
4239 * shmem_file_setup - get an unlinked file living in tmpfs
4240 * @name: name for dentry (to be seen in /proc/<pid>/maps
4241 * @size: size to be set for the file
4242 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4244 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4246 return __shmem_file_setup(name
, size
, flags
, 0);
4248 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4251 * shmem_zero_setup - setup a shared anonymous mapping
4252 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4254 int shmem_zero_setup(struct vm_area_struct
*vma
)
4257 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4260 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4261 * between XFS directory reading and selinux: since this file is only
4262 * accessible to the user through its mapping, use S_PRIVATE flag to
4263 * bypass file security, in the same way as shmem_kernel_file_setup().
4265 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
4267 return PTR_ERR(file
);
4271 vma
->vm_file
= file
;
4272 vma
->vm_ops
= &shmem_vm_ops
;
4274 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4275 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4276 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4277 khugepaged_enter(vma
, vma
->vm_flags
);
4284 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4285 * @mapping: the page's address_space
4286 * @index: the page index
4287 * @gfp: the page allocator flags to use if allocating
4289 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4290 * with any new page allocations done using the specified allocation flags.
4291 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4292 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4293 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4295 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4296 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4298 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4299 pgoff_t index
, gfp_t gfp
)
4302 struct inode
*inode
= mapping
->host
;
4306 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4307 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4308 gfp
, NULL
, NULL
, NULL
);
4310 page
= ERR_PTR(error
);
4316 * The tiny !SHMEM case uses ramfs without swap
4318 return read_cache_page_gfp(mapping
, index
, gfp
);
4321 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);