2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
35 #include <linux/khugepaged.h>
37 static struct vfsmount
*shm_mnt
;
41 * This virtual memory filesystem is heavily based on the ramfs. It
42 * extends ramfs by the ability to use swap and honor resource limits
43 * which makes it a completely usable filesystem.
46 #include <linux/xattr.h>
47 #include <linux/exportfs.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/mman.h>
51 #include <linux/string.h>
52 #include <linux/slab.h>
53 #include <linux/backing-dev.h>
54 #include <linux/shmem_fs.h>
55 #include <linux/writeback.h>
56 #include <linux/blkdev.h>
57 #include <linux/pagevec.h>
58 #include <linux/percpu_counter.h>
59 #include <linux/falloc.h>
60 #include <linux/splice.h>
61 #include <linux/security.h>
62 #include <linux/swapops.h>
63 #include <linux/mempolicy.h>
64 #include <linux/namei.h>
65 #include <linux/ctype.h>
66 #include <linux/migrate.h>
67 #include <linux/highmem.h>
68 #include <linux/seq_file.h>
69 #include <linux/magic.h>
70 #include <linux/syscalls.h>
71 #include <linux/fcntl.h>
72 #include <uapi/linux/memfd.h>
74 #include <asm/uaccess.h>
75 #include <asm/pgtable.h>
79 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
80 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
82 /* Pretend that each entry is of this size in directory's i_size */
83 #define BOGO_DIRENT_SIZE 20
85 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
86 #define SHORT_SYMLINK_LEN 128
89 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
90 * inode->i_private (with i_mutex making sure that it has only one user at
91 * a time): we would prefer not to enlarge the shmem inode just for that.
94 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
95 pgoff_t start
; /* start of range currently being fallocated */
96 pgoff_t next
; /* the next page offset to be fallocated */
97 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
98 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
102 static unsigned long shmem_default_max_blocks(void)
104 return totalram_pages
/ 2;
107 static unsigned long shmem_default_max_inodes(void)
109 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
113 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
114 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
115 struct shmem_inode_info
*info
, pgoff_t index
);
116 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
117 struct page
**pagep
, enum sgp_type sgp
,
118 gfp_t gfp
, struct mm_struct
*fault_mm
, int *fault_type
);
120 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
121 struct page
**pagep
, enum sgp_type sgp
)
123 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
124 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
);
127 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
129 return sb
->s_fs_info
;
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
138 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
140 return (flags
& VM_NORESERVE
) ?
141 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
144 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
146 if (!(flags
& VM_NORESERVE
))
147 vm_unacct_memory(VM_ACCT(size
));
150 static inline int shmem_reacct_size(unsigned long flags
,
151 loff_t oldsize
, loff_t newsize
)
153 if (!(flags
& VM_NORESERVE
)) {
154 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
155 return security_vm_enough_memory_mm(current
->mm
,
156 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
157 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
158 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
164 * ... whereas tmpfs objects are accounted incrementally as
165 * pages are allocated, in order to allow large sparse files.
166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169 static inline int shmem_acct_block(unsigned long flags
, long pages
)
171 if (!(flags
& VM_NORESERVE
))
174 return security_vm_enough_memory_mm(current
->mm
,
175 pages
* VM_ACCT(PAGE_SIZE
));
178 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
180 if (flags
& VM_NORESERVE
)
181 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
184 static const struct super_operations shmem_ops
;
185 static const struct address_space_operations shmem_aops
;
186 static const struct file_operations shmem_file_operations
;
187 static const struct inode_operations shmem_inode_operations
;
188 static const struct inode_operations shmem_dir_inode_operations
;
189 static const struct inode_operations shmem_special_inode_operations
;
190 static const struct vm_operations_struct shmem_vm_ops
;
191 static struct file_system_type shmem_fs_type
;
193 static LIST_HEAD(shmem_swaplist
);
194 static DEFINE_MUTEX(shmem_swaplist_mutex
);
196 static int shmem_reserve_inode(struct super_block
*sb
)
198 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
199 if (sbinfo
->max_inodes
) {
200 spin_lock(&sbinfo
->stat_lock
);
201 if (!sbinfo
->free_inodes
) {
202 spin_unlock(&sbinfo
->stat_lock
);
205 sbinfo
->free_inodes
--;
206 spin_unlock(&sbinfo
->stat_lock
);
211 static void shmem_free_inode(struct super_block
*sb
)
213 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
214 if (sbinfo
->max_inodes
) {
215 spin_lock(&sbinfo
->stat_lock
);
216 sbinfo
->free_inodes
++;
217 spin_unlock(&sbinfo
->stat_lock
);
222 * shmem_recalc_inode - recalculate the block usage of an inode
223 * @inode: inode to recalc
225 * We have to calculate the free blocks since the mm can drop
226 * undirtied hole pages behind our back.
228 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
229 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
231 * It has to be called with the spinlock held.
233 static void shmem_recalc_inode(struct inode
*inode
)
235 struct shmem_inode_info
*info
= SHMEM_I(inode
);
238 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
240 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
241 if (sbinfo
->max_blocks
)
242 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
243 info
->alloced
-= freed
;
244 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
245 shmem_unacct_blocks(info
->flags
, freed
);
249 bool shmem_charge(struct inode
*inode
, long pages
)
251 struct shmem_inode_info
*info
= SHMEM_I(inode
);
252 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
255 if (shmem_acct_block(info
->flags
, pages
))
257 spin_lock_irqsave(&info
->lock
, flags
);
258 info
->alloced
+= pages
;
259 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
260 shmem_recalc_inode(inode
);
261 spin_unlock_irqrestore(&info
->lock
, flags
);
262 inode
->i_mapping
->nrpages
+= pages
;
264 if (!sbinfo
->max_blocks
)
266 if (percpu_counter_compare(&sbinfo
->used_blocks
,
267 sbinfo
->max_blocks
- pages
) > 0) {
268 inode
->i_mapping
->nrpages
-= pages
;
269 spin_lock_irqsave(&info
->lock
, flags
);
270 info
->alloced
-= pages
;
271 shmem_recalc_inode(inode
);
272 spin_unlock_irqrestore(&info
->lock
, flags
);
273 shmem_unacct_blocks(info
->flags
, pages
);
276 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
280 void shmem_uncharge(struct inode
*inode
, long pages
)
282 struct shmem_inode_info
*info
= SHMEM_I(inode
);
283 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
286 spin_lock_irqsave(&info
->lock
, flags
);
287 info
->alloced
-= pages
;
288 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
289 shmem_recalc_inode(inode
);
290 spin_unlock_irqrestore(&info
->lock
, flags
);
292 if (sbinfo
->max_blocks
)
293 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
294 shmem_unacct_blocks(info
->flags
, pages
);
298 * Replace item expected in radix tree by a new item, while holding tree lock.
300 static int shmem_radix_tree_replace(struct address_space
*mapping
,
301 pgoff_t index
, void *expected
, void *replacement
)
303 struct radix_tree_node
*node
;
307 VM_BUG_ON(!expected
);
308 VM_BUG_ON(!replacement
);
309 item
= __radix_tree_lookup(&mapping
->page_tree
, index
, &node
, &pslot
);
312 if (item
!= expected
)
314 __radix_tree_replace(&mapping
->page_tree
, node
, pslot
, replacement
);
319 * Sometimes, before we decide whether to proceed or to fail, we must check
320 * that an entry was not already brought back from swap by a racing thread.
322 * Checking page is not enough: by the time a SwapCache page is locked, it
323 * might be reused, and again be SwapCache, using the same swap as before.
325 static bool shmem_confirm_swap(struct address_space
*mapping
,
326 pgoff_t index
, swp_entry_t swap
)
331 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
333 return item
== swp_to_radix_entry(swap
);
337 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
340 * disables huge pages for the mount;
342 * enables huge pages for the mount;
343 * SHMEM_HUGE_WITHIN_SIZE:
344 * only allocate huge pages if the page will be fully within i_size,
345 * also respect fadvise()/madvise() hints;
347 * only allocate huge pages if requested with fadvise()/madvise();
350 #define SHMEM_HUGE_NEVER 0
351 #define SHMEM_HUGE_ALWAYS 1
352 #define SHMEM_HUGE_WITHIN_SIZE 2
353 #define SHMEM_HUGE_ADVISE 3
357 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
360 * disables huge on shm_mnt and all mounts, for emergency use;
362 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
365 #define SHMEM_HUGE_DENY (-1)
366 #define SHMEM_HUGE_FORCE (-2)
368 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
369 /* ifdef here to avoid bloating shmem.o when not necessary */
371 int shmem_huge __read_mostly
;
373 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
374 static int shmem_parse_huge(const char *str
)
376 if (!strcmp(str
, "never"))
377 return SHMEM_HUGE_NEVER
;
378 if (!strcmp(str
, "always"))
379 return SHMEM_HUGE_ALWAYS
;
380 if (!strcmp(str
, "within_size"))
381 return SHMEM_HUGE_WITHIN_SIZE
;
382 if (!strcmp(str
, "advise"))
383 return SHMEM_HUGE_ADVISE
;
384 if (!strcmp(str
, "deny"))
385 return SHMEM_HUGE_DENY
;
386 if (!strcmp(str
, "force"))
387 return SHMEM_HUGE_FORCE
;
391 static const char *shmem_format_huge(int huge
)
394 case SHMEM_HUGE_NEVER
:
396 case SHMEM_HUGE_ALWAYS
:
398 case SHMEM_HUGE_WITHIN_SIZE
:
399 return "within_size";
400 case SHMEM_HUGE_ADVISE
:
402 case SHMEM_HUGE_DENY
:
404 case SHMEM_HUGE_FORCE
:
413 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
414 struct shrink_control
*sc
, unsigned long nr_to_split
)
416 LIST_HEAD(list
), *pos
, *next
;
418 struct shmem_inode_info
*info
;
420 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
421 int removed
= 0, split
= 0;
423 if (list_empty(&sbinfo
->shrinklist
))
426 spin_lock(&sbinfo
->shrinklist_lock
);
427 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
428 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
431 inode
= igrab(&info
->vfs_inode
);
433 /* inode is about to be evicted */
435 list_del_init(&info
->shrinklist
);
440 /* Check if there's anything to gain */
441 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
442 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
443 list_del_init(&info
->shrinklist
);
449 list_move(&info
->shrinklist
, &list
);
454 spin_unlock(&sbinfo
->shrinklist_lock
);
456 list_for_each_safe(pos
, next
, &list
) {
459 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
460 inode
= &info
->vfs_inode
;
462 if (nr_to_split
&& split
>= nr_to_split
) {
467 page
= find_lock_page(inode
->i_mapping
,
468 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
472 if (!PageTransHuge(page
)) {
478 ret
= split_huge_page(page
);
483 /* split failed: leave it on the list */
490 list_del_init(&info
->shrinklist
);
495 spin_lock(&sbinfo
->shrinklist_lock
);
496 list_splice_tail(&list
, &sbinfo
->shrinklist
);
497 sbinfo
->shrinklist_len
-= removed
;
498 spin_unlock(&sbinfo
->shrinklist_lock
);
503 static long shmem_unused_huge_scan(struct super_block
*sb
,
504 struct shrink_control
*sc
)
506 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
508 if (!READ_ONCE(sbinfo
->shrinklist_len
))
511 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
514 static long shmem_unused_huge_count(struct super_block
*sb
,
515 struct shrink_control
*sc
)
517 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
518 return READ_ONCE(sbinfo
->shrinklist_len
);
520 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
522 #define shmem_huge SHMEM_HUGE_DENY
524 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
525 struct shrink_control
*sc
, unsigned long nr_to_split
)
529 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
532 * Like add_to_page_cache_locked, but error if expected item has gone.
534 static int shmem_add_to_page_cache(struct page
*page
,
535 struct address_space
*mapping
,
536 pgoff_t index
, void *expected
)
538 int error
, nr
= hpage_nr_pages(page
);
540 VM_BUG_ON_PAGE(PageTail(page
), page
);
541 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
542 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
543 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
544 VM_BUG_ON(expected
&& PageTransHuge(page
));
546 page_ref_add(page
, nr
);
547 page
->mapping
= mapping
;
550 spin_lock_irq(&mapping
->tree_lock
);
551 if (PageTransHuge(page
)) {
552 void __rcu
**results
;
557 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
,
558 &results
, &idx
, index
, 1) &&
559 idx
< index
+ HPAGE_PMD_NR
) {
564 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
565 error
= radix_tree_insert(&mapping
->page_tree
,
566 index
+ i
, page
+ i
);
569 count_vm_event(THP_FILE_ALLOC
);
571 } else if (!expected
) {
572 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
574 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
579 mapping
->nrpages
+= nr
;
580 if (PageTransHuge(page
))
581 __inc_node_page_state(page
, NR_SHMEM_THPS
);
582 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
583 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
584 spin_unlock_irq(&mapping
->tree_lock
);
586 page
->mapping
= NULL
;
587 spin_unlock_irq(&mapping
->tree_lock
);
588 page_ref_sub(page
, nr
);
594 * Like delete_from_page_cache, but substitutes swap for page.
596 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
598 struct address_space
*mapping
= page
->mapping
;
601 VM_BUG_ON_PAGE(PageCompound(page
), page
);
603 spin_lock_irq(&mapping
->tree_lock
);
604 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
605 page
->mapping
= NULL
;
607 __dec_node_page_state(page
, NR_FILE_PAGES
);
608 __dec_node_page_state(page
, NR_SHMEM
);
609 spin_unlock_irq(&mapping
->tree_lock
);
615 * Remove swap entry from radix tree, free the swap and its page cache.
617 static int shmem_free_swap(struct address_space
*mapping
,
618 pgoff_t index
, void *radswap
)
622 spin_lock_irq(&mapping
->tree_lock
);
623 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
624 spin_unlock_irq(&mapping
->tree_lock
);
627 free_swap_and_cache(radix_to_swp_entry(radswap
));
632 * Determine (in bytes) how many of the shmem object's pages mapped by the
633 * given offsets are swapped out.
635 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
636 * as long as the inode doesn't go away and racy results are not a problem.
638 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
639 pgoff_t start
, pgoff_t end
)
641 struct radix_tree_iter iter
;
644 unsigned long swapped
= 0;
648 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
649 if (iter
.index
>= end
)
652 page
= radix_tree_deref_slot(slot
);
654 if (radix_tree_deref_retry(page
)) {
655 slot
= radix_tree_iter_retry(&iter
);
659 if (radix_tree_exceptional_entry(page
))
662 if (need_resched()) {
664 slot
= radix_tree_iter_next(&iter
);
670 return swapped
<< PAGE_SHIFT
;
674 * Determine (in bytes) how many of the shmem object's pages mapped by the
675 * given vma is swapped out.
677 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
678 * as long as the inode doesn't go away and racy results are not a problem.
680 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
682 struct inode
*inode
= file_inode(vma
->vm_file
);
683 struct shmem_inode_info
*info
= SHMEM_I(inode
);
684 struct address_space
*mapping
= inode
->i_mapping
;
685 unsigned long swapped
;
687 /* Be careful as we don't hold info->lock */
688 swapped
= READ_ONCE(info
->swapped
);
691 * The easier cases are when the shmem object has nothing in swap, or
692 * the vma maps it whole. Then we can simply use the stats that we
698 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
699 return swapped
<< PAGE_SHIFT
;
701 /* Here comes the more involved part */
702 return shmem_partial_swap_usage(mapping
,
703 linear_page_index(vma
, vma
->vm_start
),
704 linear_page_index(vma
, vma
->vm_end
));
708 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
710 void shmem_unlock_mapping(struct address_space
*mapping
)
713 pgoff_t indices
[PAGEVEC_SIZE
];
716 pagevec_init(&pvec
, 0);
718 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
720 while (!mapping_unevictable(mapping
)) {
722 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
723 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
725 pvec
.nr
= find_get_entries(mapping
, index
,
726 PAGEVEC_SIZE
, pvec
.pages
, indices
);
729 index
= indices
[pvec
.nr
- 1] + 1;
730 pagevec_remove_exceptionals(&pvec
);
731 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
732 pagevec_release(&pvec
);
738 * Remove range of pages and swap entries from radix tree, and free them.
739 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
741 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
744 struct address_space
*mapping
= inode
->i_mapping
;
745 struct shmem_inode_info
*info
= SHMEM_I(inode
);
746 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
747 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
748 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
749 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
751 pgoff_t indices
[PAGEVEC_SIZE
];
752 long nr_swaps_freed
= 0;
757 end
= -1; /* unsigned, so actually very big */
759 pagevec_init(&pvec
, 0);
761 while (index
< end
) {
762 pvec
.nr
= find_get_entries(mapping
, index
,
763 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
764 pvec
.pages
, indices
);
767 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
768 struct page
*page
= pvec
.pages
[i
];
774 if (radix_tree_exceptional_entry(page
)) {
777 nr_swaps_freed
+= !shmem_free_swap(mapping
,
782 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
784 if (!trylock_page(page
))
787 if (PageTransTail(page
)) {
788 /* Middle of THP: zero out the page */
789 clear_highpage(page
);
792 } else if (PageTransHuge(page
)) {
793 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
795 * Range ends in the middle of THP:
798 clear_highpage(page
);
802 index
+= HPAGE_PMD_NR
- 1;
803 i
+= HPAGE_PMD_NR
- 1;
806 if (!unfalloc
|| !PageUptodate(page
)) {
807 VM_BUG_ON_PAGE(PageTail(page
), page
);
808 if (page_mapping(page
) == mapping
) {
809 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
810 truncate_inode_page(mapping
, page
);
815 pagevec_remove_exceptionals(&pvec
);
816 pagevec_release(&pvec
);
822 struct page
*page
= NULL
;
823 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
825 unsigned int top
= PAGE_SIZE
;
830 zero_user_segment(page
, partial_start
, top
);
831 set_page_dirty(page
);
837 struct page
*page
= NULL
;
838 shmem_getpage(inode
, end
, &page
, SGP_READ
);
840 zero_user_segment(page
, 0, partial_end
);
841 set_page_dirty(page
);
850 while (index
< end
) {
853 pvec
.nr
= find_get_entries(mapping
, index
,
854 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
855 pvec
.pages
, indices
);
857 /* If all gone or hole-punch or unfalloc, we're done */
858 if (index
== start
|| end
!= -1)
860 /* But if truncating, restart to make sure all gone */
864 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
865 struct page
*page
= pvec
.pages
[i
];
871 if (radix_tree_exceptional_entry(page
)) {
874 if (shmem_free_swap(mapping
, index
, page
)) {
875 /* Swap was replaced by page: retry */
885 if (PageTransTail(page
)) {
886 /* Middle of THP: zero out the page */
887 clear_highpage(page
);
890 * Partial thp truncate due 'start' in middle
891 * of THP: don't need to look on these pages
892 * again on !pvec.nr restart.
894 if (index
!= round_down(end
, HPAGE_PMD_NR
))
897 } else if (PageTransHuge(page
)) {
898 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
900 * Range ends in the middle of THP:
903 clear_highpage(page
);
907 index
+= HPAGE_PMD_NR
- 1;
908 i
+= HPAGE_PMD_NR
- 1;
911 if (!unfalloc
|| !PageUptodate(page
)) {
912 VM_BUG_ON_PAGE(PageTail(page
), page
);
913 if (page_mapping(page
) == mapping
) {
914 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
915 truncate_inode_page(mapping
, page
);
917 /* Page was replaced by swap: retry */
925 pagevec_remove_exceptionals(&pvec
);
926 pagevec_release(&pvec
);
930 spin_lock_irq(&info
->lock
);
931 info
->swapped
-= nr_swaps_freed
;
932 shmem_recalc_inode(inode
);
933 spin_unlock_irq(&info
->lock
);
936 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
938 shmem_undo_range(inode
, lstart
, lend
, false);
939 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
941 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
943 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
946 struct inode
*inode
= dentry
->d_inode
;
947 struct shmem_inode_info
*info
= SHMEM_I(inode
);
949 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
950 spin_lock_irq(&info
->lock
);
951 shmem_recalc_inode(inode
);
952 spin_unlock_irq(&info
->lock
);
954 generic_fillattr(inode
, stat
);
958 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
960 struct inode
*inode
= d_inode(dentry
);
961 struct shmem_inode_info
*info
= SHMEM_I(inode
);
962 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
965 error
= setattr_prepare(dentry
, attr
);
969 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
970 loff_t oldsize
= inode
->i_size
;
971 loff_t newsize
= attr
->ia_size
;
973 /* protected by i_mutex */
974 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
975 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
978 if (newsize
!= oldsize
) {
979 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
983 i_size_write(inode
, newsize
);
984 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
986 if (newsize
<= oldsize
) {
987 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
988 if (oldsize
> holebegin
)
989 unmap_mapping_range(inode
->i_mapping
,
992 shmem_truncate_range(inode
,
993 newsize
, (loff_t
)-1);
994 /* unmap again to remove racily COWed private pages */
995 if (oldsize
> holebegin
)
996 unmap_mapping_range(inode
->i_mapping
,
1000 * Part of the huge page can be beyond i_size: subject
1001 * to shrink under memory pressure.
1003 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1004 spin_lock(&sbinfo
->shrinklist_lock
);
1005 if (list_empty(&info
->shrinklist
)) {
1006 list_add_tail(&info
->shrinklist
,
1007 &sbinfo
->shrinklist
);
1008 sbinfo
->shrinklist_len
++;
1010 spin_unlock(&sbinfo
->shrinklist_lock
);
1015 setattr_copy(inode
, attr
);
1016 if (attr
->ia_valid
& ATTR_MODE
)
1017 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1021 static void shmem_evict_inode(struct inode
*inode
)
1023 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1024 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1026 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1027 shmem_unacct_size(info
->flags
, inode
->i_size
);
1029 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1030 if (!list_empty(&info
->shrinklist
)) {
1031 spin_lock(&sbinfo
->shrinklist_lock
);
1032 if (!list_empty(&info
->shrinklist
)) {
1033 list_del_init(&info
->shrinklist
);
1034 sbinfo
->shrinklist_len
--;
1036 spin_unlock(&sbinfo
->shrinklist_lock
);
1038 if (!list_empty(&info
->swaplist
)) {
1039 mutex_lock(&shmem_swaplist_mutex
);
1040 list_del_init(&info
->swaplist
);
1041 mutex_unlock(&shmem_swaplist_mutex
);
1045 simple_xattrs_free(&info
->xattrs
);
1046 WARN_ON(inode
->i_blocks
);
1047 shmem_free_inode(inode
->i_sb
);
1052 * If swap found in inode, free it and move page from swapcache to filecache.
1054 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
1055 swp_entry_t swap
, struct page
**pagep
)
1057 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1063 radswap
= swp_to_radix_entry(swap
);
1064 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
1066 return -EAGAIN
; /* tell shmem_unuse we found nothing */
1069 * Move _head_ to start search for next from here.
1070 * But be careful: shmem_evict_inode checks list_empty without taking
1071 * mutex, and there's an instant in list_move_tail when info->swaplist
1072 * would appear empty, if it were the only one on shmem_swaplist.
1074 if (shmem_swaplist
.next
!= &info
->swaplist
)
1075 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
1077 gfp
= mapping_gfp_mask(mapping
);
1078 if (shmem_should_replace_page(*pagep
, gfp
)) {
1079 mutex_unlock(&shmem_swaplist_mutex
);
1080 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
1081 mutex_lock(&shmem_swaplist_mutex
);
1083 * We needed to drop mutex to make that restrictive page
1084 * allocation, but the inode might have been freed while we
1085 * dropped it: although a racing shmem_evict_inode() cannot
1086 * complete without emptying the radix_tree, our page lock
1087 * on this swapcache page is not enough to prevent that -
1088 * free_swap_and_cache() of our swap entry will only
1089 * trylock_page(), removing swap from radix_tree whatever.
1091 * We must not proceed to shmem_add_to_page_cache() if the
1092 * inode has been freed, but of course we cannot rely on
1093 * inode or mapping or info to check that. However, we can
1094 * safely check if our swap entry is still in use (and here
1095 * it can't have got reused for another page): if it's still
1096 * in use, then the inode cannot have been freed yet, and we
1097 * can safely proceed (if it's no longer in use, that tells
1098 * nothing about the inode, but we don't need to unuse swap).
1100 if (!page_swapcount(*pagep
))
1105 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1106 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1107 * beneath us (pagelock doesn't help until the page is in pagecache).
1110 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
1112 if (error
!= -ENOMEM
) {
1114 * Truncation and eviction use free_swap_and_cache(), which
1115 * only does trylock page: if we raced, best clean up here.
1117 delete_from_swap_cache(*pagep
);
1118 set_page_dirty(*pagep
);
1120 spin_lock_irq(&info
->lock
);
1122 spin_unlock_irq(&info
->lock
);
1130 * Search through swapped inodes to find and replace swap by page.
1132 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1134 struct list_head
*this, *next
;
1135 struct shmem_inode_info
*info
;
1136 struct mem_cgroup
*memcg
;
1140 * There's a faint possibility that swap page was replaced before
1141 * caller locked it: caller will come back later with the right page.
1143 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1147 * Charge page using GFP_KERNEL while we can wait, before taking
1148 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1149 * Charged back to the user (not to caller) when swap account is used.
1151 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1155 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1158 mutex_lock(&shmem_swaplist_mutex
);
1159 list_for_each_safe(this, next
, &shmem_swaplist
) {
1160 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1162 error
= shmem_unuse_inode(info
, swap
, &page
);
1164 list_del_init(&info
->swaplist
);
1166 if (error
!= -EAGAIN
)
1168 /* found nothing in this: move on to search the next */
1170 mutex_unlock(&shmem_swaplist_mutex
);
1173 if (error
!= -ENOMEM
)
1175 mem_cgroup_cancel_charge(page
, memcg
, false);
1177 mem_cgroup_commit_charge(page
, memcg
, true, false);
1185 * Move the page from the page cache to the swap cache.
1187 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1189 struct shmem_inode_info
*info
;
1190 struct address_space
*mapping
;
1191 struct inode
*inode
;
1195 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1196 BUG_ON(!PageLocked(page
));
1197 mapping
= page
->mapping
;
1198 index
= page
->index
;
1199 inode
= mapping
->host
;
1200 info
= SHMEM_I(inode
);
1201 if (info
->flags
& VM_LOCKED
)
1203 if (!total_swap_pages
)
1207 * Our capabilities prevent regular writeback or sync from ever calling
1208 * shmem_writepage; but a stacking filesystem might use ->writepage of
1209 * its underlying filesystem, in which case tmpfs should write out to
1210 * swap only in response to memory pressure, and not for the writeback
1213 if (!wbc
->for_reclaim
) {
1214 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1219 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1220 * value into swapfile.c, the only way we can correctly account for a
1221 * fallocated page arriving here is now to initialize it and write it.
1223 * That's okay for a page already fallocated earlier, but if we have
1224 * not yet completed the fallocation, then (a) we want to keep track
1225 * of this page in case we have to undo it, and (b) it may not be a
1226 * good idea to continue anyway, once we're pushing into swap. So
1227 * reactivate the page, and let shmem_fallocate() quit when too many.
1229 if (!PageUptodate(page
)) {
1230 if (inode
->i_private
) {
1231 struct shmem_falloc
*shmem_falloc
;
1232 spin_lock(&inode
->i_lock
);
1233 shmem_falloc
= inode
->i_private
;
1235 !shmem_falloc
->waitq
&&
1236 index
>= shmem_falloc
->start
&&
1237 index
< shmem_falloc
->next
)
1238 shmem_falloc
->nr_unswapped
++;
1240 shmem_falloc
= NULL
;
1241 spin_unlock(&inode
->i_lock
);
1245 clear_highpage(page
);
1246 flush_dcache_page(page
);
1247 SetPageUptodate(page
);
1250 swap
= get_swap_page();
1254 if (mem_cgroup_try_charge_swap(page
, swap
))
1258 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1259 * if it's not already there. Do it now before the page is
1260 * moved to swap cache, when its pagelock no longer protects
1261 * the inode from eviction. But don't unlock the mutex until
1262 * we've incremented swapped, because shmem_unuse_inode() will
1263 * prune a !swapped inode from the swaplist under this mutex.
1265 mutex_lock(&shmem_swaplist_mutex
);
1266 if (list_empty(&info
->swaplist
))
1267 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1269 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1270 spin_lock_irq(&info
->lock
);
1271 shmem_recalc_inode(inode
);
1273 spin_unlock_irq(&info
->lock
);
1275 swap_shmem_alloc(swap
);
1276 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1278 mutex_unlock(&shmem_swaplist_mutex
);
1279 BUG_ON(page_mapped(page
));
1280 swap_writepage(page
, wbc
);
1284 mutex_unlock(&shmem_swaplist_mutex
);
1286 swapcache_free(swap
);
1288 set_page_dirty(page
);
1289 if (wbc
->for_reclaim
)
1290 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1295 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1296 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1300 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1301 return; /* show nothing */
1303 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1305 seq_printf(seq
, ",mpol=%s", buffer
);
1308 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1310 struct mempolicy
*mpol
= NULL
;
1312 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1313 mpol
= sbinfo
->mpol
;
1315 spin_unlock(&sbinfo
->stat_lock
);
1319 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1320 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1323 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1327 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1329 #define vm_policy vm_private_data
1332 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1333 struct shmem_inode_info
*info
, pgoff_t index
)
1335 /* Create a pseudo vma that just contains the policy */
1337 /* Bias interleave by inode number to distribute better across nodes */
1338 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1340 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1343 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1345 /* Drop reference taken by mpol_shared_policy_lookup() */
1346 mpol_cond_put(vma
->vm_policy
);
1349 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1350 struct shmem_inode_info
*info
, pgoff_t index
)
1352 struct vm_area_struct pvma
;
1355 shmem_pseudo_vma_init(&pvma
, info
, index
);
1356 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1357 shmem_pseudo_vma_destroy(&pvma
);
1362 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1363 struct shmem_inode_info
*info
, pgoff_t index
)
1365 struct vm_area_struct pvma
;
1366 struct inode
*inode
= &info
->vfs_inode
;
1367 struct address_space
*mapping
= inode
->i_mapping
;
1368 pgoff_t idx
, hindex
;
1369 void __rcu
**results
;
1372 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1375 hindex
= round_down(index
, HPAGE_PMD_NR
);
1377 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1378 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1384 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1385 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1386 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1387 shmem_pseudo_vma_destroy(&pvma
);
1389 prep_transhuge_page(page
);
1393 static struct page
*shmem_alloc_page(gfp_t gfp
,
1394 struct shmem_inode_info
*info
, pgoff_t index
)
1396 struct vm_area_struct pvma
;
1399 shmem_pseudo_vma_init(&pvma
, info
, index
);
1400 page
= alloc_page_vma(gfp
, &pvma
, 0);
1401 shmem_pseudo_vma_destroy(&pvma
);
1406 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1407 struct shmem_inode_info
*info
, struct shmem_sb_info
*sbinfo
,
1408 pgoff_t index
, bool huge
)
1414 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1416 nr
= huge
? HPAGE_PMD_NR
: 1;
1418 if (shmem_acct_block(info
->flags
, nr
))
1420 if (sbinfo
->max_blocks
) {
1421 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1422 sbinfo
->max_blocks
- nr
) > 0)
1424 percpu_counter_add(&sbinfo
->used_blocks
, nr
);
1428 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1430 page
= shmem_alloc_page(gfp
, info
, index
);
1432 __SetPageLocked(page
);
1433 __SetPageSwapBacked(page
);
1438 if (sbinfo
->max_blocks
)
1439 percpu_counter_add(&sbinfo
->used_blocks
, -nr
);
1441 shmem_unacct_blocks(info
->flags
, nr
);
1443 return ERR_PTR(err
);
1447 * When a page is moved from swapcache to shmem filecache (either by the
1448 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1449 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1450 * ignorance of the mapping it belongs to. If that mapping has special
1451 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1452 * we may need to copy to a suitable page before moving to filecache.
1454 * In a future release, this may well be extended to respect cpuset and
1455 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1456 * but for now it is a simple matter of zone.
1458 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1460 return page_zonenum(page
) > gfp_zone(gfp
);
1463 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1464 struct shmem_inode_info
*info
, pgoff_t index
)
1466 struct page
*oldpage
, *newpage
;
1467 struct address_space
*swap_mapping
;
1472 swap_index
= page_private(oldpage
);
1473 swap_mapping
= page_mapping(oldpage
);
1476 * We have arrived here because our zones are constrained, so don't
1477 * limit chance of success by further cpuset and node constraints.
1479 gfp
&= ~GFP_CONSTRAINT_MASK
;
1480 newpage
= shmem_alloc_page(gfp
, info
, index
);
1485 copy_highpage(newpage
, oldpage
);
1486 flush_dcache_page(newpage
);
1488 __SetPageLocked(newpage
);
1489 __SetPageSwapBacked(newpage
);
1490 SetPageUptodate(newpage
);
1491 set_page_private(newpage
, swap_index
);
1492 SetPageSwapCache(newpage
);
1495 * Our caller will very soon move newpage out of swapcache, but it's
1496 * a nice clean interface for us to replace oldpage by newpage there.
1498 spin_lock_irq(&swap_mapping
->tree_lock
);
1499 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1502 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1503 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1505 spin_unlock_irq(&swap_mapping
->tree_lock
);
1507 if (unlikely(error
)) {
1509 * Is this possible? I think not, now that our callers check
1510 * both PageSwapCache and page_private after getting page lock;
1511 * but be defensive. Reverse old to newpage for clear and free.
1515 mem_cgroup_migrate(oldpage
, newpage
);
1516 lru_cache_add_anon(newpage
);
1520 ClearPageSwapCache(oldpage
);
1521 set_page_private(oldpage
, 0);
1523 unlock_page(oldpage
);
1530 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1532 * If we allocate a new one we do not mark it dirty. That's up to the
1533 * vm. If we swap it in we mark it dirty since we also free the swap
1534 * entry since a page cannot live in both the swap and page cache.
1536 * fault_mm and fault_type are only supplied by shmem_fault:
1537 * otherwise they are NULL.
1539 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1540 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1541 struct mm_struct
*fault_mm
, int *fault_type
)
1543 struct address_space
*mapping
= inode
->i_mapping
;
1544 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1545 struct shmem_sb_info
*sbinfo
;
1546 struct mm_struct
*charge_mm
;
1547 struct mem_cgroup
*memcg
;
1550 enum sgp_type sgp_huge
= sgp
;
1551 pgoff_t hindex
= index
;
1556 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1558 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1562 page
= find_lock_entry(mapping
, index
);
1563 if (radix_tree_exceptional_entry(page
)) {
1564 swap
= radix_to_swp_entry(page
);
1568 if (sgp
<= SGP_CACHE
&&
1569 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1574 if (page
&& sgp
== SGP_WRITE
)
1575 mark_page_accessed(page
);
1577 /* fallocated page? */
1578 if (page
&& !PageUptodate(page
)) {
1579 if (sgp
!= SGP_READ
)
1585 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1591 * Fast cache lookup did not find it:
1592 * bring it back from swap or allocate.
1594 sbinfo
= SHMEM_SB(inode
->i_sb
);
1595 charge_mm
= fault_mm
? : current
->mm
;
1598 /* Look it up and read it in.. */
1599 page
= lookup_swap_cache(swap
);
1601 /* Or update major stats only when swapin succeeds?? */
1603 *fault_type
|= VM_FAULT_MAJOR
;
1604 count_vm_event(PGMAJFAULT
);
1605 mem_cgroup_count_vm_event(fault_mm
, PGMAJFAULT
);
1607 /* Here we actually start the io */
1608 page
= shmem_swapin(swap
, gfp
, info
, index
);
1615 /* We have to do this with page locked to prevent races */
1617 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1618 !shmem_confirm_swap(mapping
, index
, swap
)) {
1619 error
= -EEXIST
; /* try again */
1622 if (!PageUptodate(page
)) {
1626 wait_on_page_writeback(page
);
1628 if (shmem_should_replace_page(page
, gfp
)) {
1629 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1634 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1637 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1638 swp_to_radix_entry(swap
));
1640 * We already confirmed swap under page lock, and make
1641 * no memory allocation here, so usually no possibility
1642 * of error; but free_swap_and_cache() only trylocks a
1643 * page, so it is just possible that the entry has been
1644 * truncated or holepunched since swap was confirmed.
1645 * shmem_undo_range() will have done some of the
1646 * unaccounting, now delete_from_swap_cache() will do
1648 * Reset swap.val? No, leave it so "failed" goes back to
1649 * "repeat": reading a hole and writing should succeed.
1652 mem_cgroup_cancel_charge(page
, memcg
, false);
1653 delete_from_swap_cache(page
);
1659 mem_cgroup_commit_charge(page
, memcg
, true, false);
1661 spin_lock_irq(&info
->lock
);
1663 shmem_recalc_inode(inode
);
1664 spin_unlock_irq(&info
->lock
);
1666 if (sgp
== SGP_WRITE
)
1667 mark_page_accessed(page
);
1669 delete_from_swap_cache(page
);
1670 set_page_dirty(page
);
1674 /* shmem_symlink() */
1675 if (mapping
->a_ops
!= &shmem_aops
)
1677 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1679 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1681 switch (sbinfo
->huge
) {
1684 case SHMEM_HUGE_NEVER
:
1686 case SHMEM_HUGE_WITHIN_SIZE
:
1687 off
= round_up(index
, HPAGE_PMD_NR
);
1688 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1689 if (i_size
>= HPAGE_PMD_SIZE
&&
1690 i_size
>> PAGE_SHIFT
>= off
)
1693 case SHMEM_HUGE_ADVISE
:
1694 if (sgp_huge
== SGP_HUGE
)
1696 /* TODO: implement fadvise() hints */
1701 page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1704 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1709 error
= PTR_ERR(page
);
1711 if (error
!= -ENOSPC
)
1714 * Try to reclaim some spece by splitting a huge page
1715 * beyond i_size on the filesystem.
1719 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1720 if (ret
== SHRINK_STOP
)
1728 if (PageTransHuge(page
))
1729 hindex
= round_down(index
, HPAGE_PMD_NR
);
1733 if (sgp
== SGP_WRITE
)
1734 __SetPageReferenced(page
);
1736 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1737 PageTransHuge(page
));
1740 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1741 compound_order(page
));
1743 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1745 radix_tree_preload_end();
1748 mem_cgroup_cancel_charge(page
, memcg
,
1749 PageTransHuge(page
));
1752 mem_cgroup_commit_charge(page
, memcg
, false,
1753 PageTransHuge(page
));
1754 lru_cache_add_anon(page
);
1756 spin_lock_irq(&info
->lock
);
1757 info
->alloced
+= 1 << compound_order(page
);
1758 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1759 shmem_recalc_inode(inode
);
1760 spin_unlock_irq(&info
->lock
);
1763 if (PageTransHuge(page
) &&
1764 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1765 hindex
+ HPAGE_PMD_NR
- 1) {
1767 * Part of the huge page is beyond i_size: subject
1768 * to shrink under memory pressure.
1770 spin_lock(&sbinfo
->shrinklist_lock
);
1771 if (list_empty(&info
->shrinklist
)) {
1772 list_add_tail(&info
->shrinklist
,
1773 &sbinfo
->shrinklist
);
1774 sbinfo
->shrinklist_len
++;
1776 spin_unlock(&sbinfo
->shrinklist_lock
);
1780 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1782 if (sgp
== SGP_FALLOC
)
1786 * Let SGP_WRITE caller clear ends if write does not fill page;
1787 * but SGP_FALLOC on a page fallocated earlier must initialize
1788 * it now, lest undo on failure cancel our earlier guarantee.
1790 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1791 struct page
*head
= compound_head(page
);
1794 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1795 clear_highpage(head
+ i
);
1796 flush_dcache_page(head
+ i
);
1798 SetPageUptodate(head
);
1802 /* Perhaps the file has been truncated since we checked */
1803 if (sgp
<= SGP_CACHE
&&
1804 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1806 ClearPageDirty(page
);
1807 delete_from_page_cache(page
);
1808 spin_lock_irq(&info
->lock
);
1809 shmem_recalc_inode(inode
);
1810 spin_unlock_irq(&info
->lock
);
1815 *pagep
= page
+ index
- hindex
;
1822 if (sbinfo
->max_blocks
)
1823 percpu_counter_sub(&sbinfo
->used_blocks
,
1824 1 << compound_order(page
));
1825 shmem_unacct_blocks(info
->flags
, 1 << compound_order(page
));
1827 if (PageTransHuge(page
)) {
1833 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1840 if (error
== -ENOSPC
&& !once
++) {
1841 spin_lock_irq(&info
->lock
);
1842 shmem_recalc_inode(inode
);
1843 spin_unlock_irq(&info
->lock
);
1846 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1852 * This is like autoremove_wake_function, but it removes the wait queue
1853 * entry unconditionally - even if something else had already woken the
1856 static int synchronous_wake_function(wait_queue_t
*wait
, unsigned mode
, int sync
, void *key
)
1858 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1859 list_del_init(&wait
->task_list
);
1863 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1865 struct inode
*inode
= file_inode(vma
->vm_file
);
1866 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1869 int ret
= VM_FAULT_LOCKED
;
1872 * Trinity finds that probing a hole which tmpfs is punching can
1873 * prevent the hole-punch from ever completing: which in turn
1874 * locks writers out with its hold on i_mutex. So refrain from
1875 * faulting pages into the hole while it's being punched. Although
1876 * shmem_undo_range() does remove the additions, it may be unable to
1877 * keep up, as each new page needs its own unmap_mapping_range() call,
1878 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1880 * It does not matter if we sometimes reach this check just before the
1881 * hole-punch begins, so that one fault then races with the punch:
1882 * we just need to make racing faults a rare case.
1884 * The implementation below would be much simpler if we just used a
1885 * standard mutex or completion: but we cannot take i_mutex in fault,
1886 * and bloating every shmem inode for this unlikely case would be sad.
1888 if (unlikely(inode
->i_private
)) {
1889 struct shmem_falloc
*shmem_falloc
;
1891 spin_lock(&inode
->i_lock
);
1892 shmem_falloc
= inode
->i_private
;
1894 shmem_falloc
->waitq
&&
1895 vmf
->pgoff
>= shmem_falloc
->start
&&
1896 vmf
->pgoff
< shmem_falloc
->next
) {
1897 wait_queue_head_t
*shmem_falloc_waitq
;
1898 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
1900 ret
= VM_FAULT_NOPAGE
;
1901 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1902 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1903 /* It's polite to up mmap_sem if we can */
1904 up_read(&vma
->vm_mm
->mmap_sem
);
1905 ret
= VM_FAULT_RETRY
;
1908 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1909 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1910 TASK_UNINTERRUPTIBLE
);
1911 spin_unlock(&inode
->i_lock
);
1915 * shmem_falloc_waitq points into the shmem_fallocate()
1916 * stack of the hole-punching task: shmem_falloc_waitq
1917 * is usually invalid by the time we reach here, but
1918 * finish_wait() does not dereference it in that case;
1919 * though i_lock needed lest racing with wake_up_all().
1921 spin_lock(&inode
->i_lock
);
1922 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1923 spin_unlock(&inode
->i_lock
);
1926 spin_unlock(&inode
->i_lock
);
1930 if (vma
->vm_flags
& VM_HUGEPAGE
)
1932 else if (vma
->vm_flags
& VM_NOHUGEPAGE
)
1935 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
1936 gfp
, vma
->vm_mm
, &ret
);
1938 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1942 unsigned long shmem_get_unmapped_area(struct file
*file
,
1943 unsigned long uaddr
, unsigned long len
,
1944 unsigned long pgoff
, unsigned long flags
)
1946 unsigned long (*get_area
)(struct file
*,
1947 unsigned long, unsigned long, unsigned long, unsigned long);
1949 unsigned long offset
;
1950 unsigned long inflated_len
;
1951 unsigned long inflated_addr
;
1952 unsigned long inflated_offset
;
1954 if (len
> TASK_SIZE
)
1957 get_area
= current
->mm
->get_unmapped_area
;
1958 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
1960 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1962 if (IS_ERR_VALUE(addr
))
1964 if (addr
& ~PAGE_MASK
)
1966 if (addr
> TASK_SIZE
- len
)
1969 if (shmem_huge
== SHMEM_HUGE_DENY
)
1971 if (len
< HPAGE_PMD_SIZE
)
1973 if (flags
& MAP_FIXED
)
1976 * Our priority is to support MAP_SHARED mapped hugely;
1977 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1978 * But if caller specified an address hint, respect that as before.
1983 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
1984 struct super_block
*sb
;
1987 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
1988 sb
= file_inode(file
)->i_sb
;
1991 * Called directly from mm/mmap.c, or drivers/char/mem.c
1992 * for "/dev/zero", to create a shared anonymous object.
1994 if (IS_ERR(shm_mnt
))
1996 sb
= shm_mnt
->mnt_sb
;
1998 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2002 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2003 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2005 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2008 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2009 if (inflated_len
> TASK_SIZE
)
2011 if (inflated_len
< len
)
2014 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
2015 if (IS_ERR_VALUE(inflated_addr
))
2017 if (inflated_addr
& ~PAGE_MASK
)
2020 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2021 inflated_addr
+= offset
- inflated_offset
;
2022 if (inflated_offset
> offset
)
2023 inflated_addr
+= HPAGE_PMD_SIZE
;
2025 if (inflated_addr
> TASK_SIZE
- len
)
2027 return inflated_addr
;
2031 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2033 struct inode
*inode
= file_inode(vma
->vm_file
);
2034 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2037 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2040 struct inode
*inode
= file_inode(vma
->vm_file
);
2043 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2044 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2048 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2050 struct inode
*inode
= file_inode(file
);
2051 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2052 int retval
= -ENOMEM
;
2054 spin_lock_irq(&info
->lock
);
2055 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2056 if (!user_shm_lock(inode
->i_size
, user
))
2058 info
->flags
|= VM_LOCKED
;
2059 mapping_set_unevictable(file
->f_mapping
);
2061 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2062 user_shm_unlock(inode
->i_size
, user
);
2063 info
->flags
&= ~VM_LOCKED
;
2064 mapping_clear_unevictable(file
->f_mapping
);
2069 spin_unlock_irq(&info
->lock
);
2073 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2075 file_accessed(file
);
2076 vma
->vm_ops
= &shmem_vm_ops
;
2077 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2078 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2079 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2080 khugepaged_enter(vma
, vma
->vm_flags
);
2085 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2086 umode_t mode
, dev_t dev
, unsigned long flags
)
2088 struct inode
*inode
;
2089 struct shmem_inode_info
*info
;
2090 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2092 if (shmem_reserve_inode(sb
))
2095 inode
= new_inode(sb
);
2097 inode
->i_ino
= get_next_ino();
2098 inode_init_owner(inode
, dir
, mode
);
2099 inode
->i_blocks
= 0;
2100 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2101 inode
->i_generation
= get_seconds();
2102 info
= SHMEM_I(inode
);
2103 memset(info
, 0, (char *)inode
- (char *)info
);
2104 spin_lock_init(&info
->lock
);
2105 info
->seals
= F_SEAL_SEAL
;
2106 info
->flags
= flags
& VM_NORESERVE
;
2107 INIT_LIST_HEAD(&info
->shrinklist
);
2108 INIT_LIST_HEAD(&info
->swaplist
);
2109 simple_xattrs_init(&info
->xattrs
);
2110 cache_no_acl(inode
);
2112 switch (mode
& S_IFMT
) {
2114 inode
->i_op
= &shmem_special_inode_operations
;
2115 init_special_inode(inode
, mode
, dev
);
2118 inode
->i_mapping
->a_ops
= &shmem_aops
;
2119 inode
->i_op
= &shmem_inode_operations
;
2120 inode
->i_fop
= &shmem_file_operations
;
2121 mpol_shared_policy_init(&info
->policy
,
2122 shmem_get_sbmpol(sbinfo
));
2126 /* Some things misbehave if size == 0 on a directory */
2127 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2128 inode
->i_op
= &shmem_dir_inode_operations
;
2129 inode
->i_fop
= &simple_dir_operations
;
2133 * Must not load anything in the rbtree,
2134 * mpol_free_shared_policy will not be called.
2136 mpol_shared_policy_init(&info
->policy
, NULL
);
2140 shmem_free_inode(sb
);
2144 bool shmem_mapping(struct address_space
*mapping
)
2149 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
2153 static const struct inode_operations shmem_symlink_inode_operations
;
2154 static const struct inode_operations shmem_short_symlink_operations
;
2156 #ifdef CONFIG_TMPFS_XATTR
2157 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2159 #define shmem_initxattrs NULL
2163 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2164 loff_t pos
, unsigned len
, unsigned flags
,
2165 struct page
**pagep
, void **fsdata
)
2167 struct inode
*inode
= mapping
->host
;
2168 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2169 pgoff_t index
= pos
>> PAGE_SHIFT
;
2171 /* i_mutex is held by caller */
2172 if (unlikely(info
->seals
)) {
2173 if (info
->seals
& F_SEAL_WRITE
)
2175 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2179 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2183 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2184 loff_t pos
, unsigned len
, unsigned copied
,
2185 struct page
*page
, void *fsdata
)
2187 struct inode
*inode
= mapping
->host
;
2189 if (pos
+ copied
> inode
->i_size
)
2190 i_size_write(inode
, pos
+ copied
);
2192 if (!PageUptodate(page
)) {
2193 struct page
*head
= compound_head(page
);
2194 if (PageTransCompound(page
)) {
2197 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2198 if (head
+ i
== page
)
2200 clear_highpage(head
+ i
);
2201 flush_dcache_page(head
+ i
);
2204 if (copied
< PAGE_SIZE
) {
2205 unsigned from
= pos
& (PAGE_SIZE
- 1);
2206 zero_user_segments(page
, 0, from
,
2207 from
+ copied
, PAGE_SIZE
);
2209 SetPageUptodate(head
);
2211 set_page_dirty(page
);
2218 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2220 struct file
*file
= iocb
->ki_filp
;
2221 struct inode
*inode
= file_inode(file
);
2222 struct address_space
*mapping
= inode
->i_mapping
;
2224 unsigned long offset
;
2225 enum sgp_type sgp
= SGP_READ
;
2228 loff_t
*ppos
= &iocb
->ki_pos
;
2231 * Might this read be for a stacking filesystem? Then when reading
2232 * holes of a sparse file, we actually need to allocate those pages,
2233 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2235 if (!iter_is_iovec(to
))
2238 index
= *ppos
>> PAGE_SHIFT
;
2239 offset
= *ppos
& ~PAGE_MASK
;
2242 struct page
*page
= NULL
;
2244 unsigned long nr
, ret
;
2245 loff_t i_size
= i_size_read(inode
);
2247 end_index
= i_size
>> PAGE_SHIFT
;
2248 if (index
> end_index
)
2250 if (index
== end_index
) {
2251 nr
= i_size
& ~PAGE_MASK
;
2256 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2258 if (error
== -EINVAL
)
2263 if (sgp
== SGP_CACHE
)
2264 set_page_dirty(page
);
2269 * We must evaluate after, since reads (unlike writes)
2270 * are called without i_mutex protection against truncate
2273 i_size
= i_size_read(inode
);
2274 end_index
= i_size
>> PAGE_SHIFT
;
2275 if (index
== end_index
) {
2276 nr
= i_size
& ~PAGE_MASK
;
2287 * If users can be writing to this page using arbitrary
2288 * virtual addresses, take care about potential aliasing
2289 * before reading the page on the kernel side.
2291 if (mapping_writably_mapped(mapping
))
2292 flush_dcache_page(page
);
2294 * Mark the page accessed if we read the beginning.
2297 mark_page_accessed(page
);
2299 page
= ZERO_PAGE(0);
2304 * Ok, we have the page, and it's up-to-date, so
2305 * now we can copy it to user space...
2307 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2310 index
+= offset
>> PAGE_SHIFT
;
2311 offset
&= ~PAGE_MASK
;
2314 if (!iov_iter_count(to
))
2323 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2324 file_accessed(file
);
2325 return retval
? retval
: error
;
2329 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2331 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2332 pgoff_t index
, pgoff_t end
, int whence
)
2335 struct pagevec pvec
;
2336 pgoff_t indices
[PAGEVEC_SIZE
];
2340 pagevec_init(&pvec
, 0);
2341 pvec
.nr
= 1; /* start small: we may be there already */
2343 pvec
.nr
= find_get_entries(mapping
, index
,
2344 pvec
.nr
, pvec
.pages
, indices
);
2346 if (whence
== SEEK_DATA
)
2350 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2351 if (index
< indices
[i
]) {
2352 if (whence
== SEEK_HOLE
) {
2358 page
= pvec
.pages
[i
];
2359 if (page
&& !radix_tree_exceptional_entry(page
)) {
2360 if (!PageUptodate(page
))
2364 (page
&& whence
== SEEK_DATA
) ||
2365 (!page
&& whence
== SEEK_HOLE
)) {
2370 pagevec_remove_exceptionals(&pvec
);
2371 pagevec_release(&pvec
);
2372 pvec
.nr
= PAGEVEC_SIZE
;
2378 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2380 struct address_space
*mapping
= file
->f_mapping
;
2381 struct inode
*inode
= mapping
->host
;
2385 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2386 return generic_file_llseek_size(file
, offset
, whence
,
2387 MAX_LFS_FILESIZE
, i_size_read(inode
));
2389 /* We're holding i_mutex so we can access i_size directly */
2393 else if (offset
>= inode
->i_size
)
2396 start
= offset
>> PAGE_SHIFT
;
2397 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2398 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2399 new_offset
<<= PAGE_SHIFT
;
2400 if (new_offset
> offset
) {
2401 if (new_offset
< inode
->i_size
)
2402 offset
= new_offset
;
2403 else if (whence
== SEEK_DATA
)
2406 offset
= inode
->i_size
;
2411 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2412 inode_unlock(inode
);
2417 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2418 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2420 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2421 #define LAST_SCAN 4 /* about 150ms max */
2423 static void shmem_tag_pins(struct address_space
*mapping
)
2425 struct radix_tree_iter iter
;
2434 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2435 page
= radix_tree_deref_slot(slot
);
2436 if (!page
|| radix_tree_exception(page
)) {
2437 if (radix_tree_deref_retry(page
)) {
2438 slot
= radix_tree_iter_retry(&iter
);
2441 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2442 spin_lock_irq(&mapping
->tree_lock
);
2443 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2445 spin_unlock_irq(&mapping
->tree_lock
);
2448 if (need_resched()) {
2450 slot
= radix_tree_iter_next(&iter
);
2457 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2458 * via get_user_pages(), drivers might have some pending I/O without any active
2459 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2460 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2461 * them to be dropped.
2462 * The caller must guarantee that no new user will acquire writable references
2463 * to those pages to avoid races.
2465 static int shmem_wait_for_pins(struct address_space
*mapping
)
2467 struct radix_tree_iter iter
;
2473 shmem_tag_pins(mapping
);
2476 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2477 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2481 lru_add_drain_all();
2482 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2487 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2488 start
, SHMEM_TAG_PINNED
) {
2490 page
= radix_tree_deref_slot(slot
);
2491 if (radix_tree_exception(page
)) {
2492 if (radix_tree_deref_retry(page
)) {
2493 slot
= radix_tree_iter_retry(&iter
);
2501 page_count(page
) - page_mapcount(page
) != 1) {
2502 if (scan
< LAST_SCAN
)
2503 goto continue_resched
;
2506 * On the last scan, we clean up all those tags
2507 * we inserted; but make a note that we still
2508 * found pages pinned.
2513 spin_lock_irq(&mapping
->tree_lock
);
2514 radix_tree_tag_clear(&mapping
->page_tree
,
2515 iter
.index
, SHMEM_TAG_PINNED
);
2516 spin_unlock_irq(&mapping
->tree_lock
);
2518 if (need_resched()) {
2520 slot
= radix_tree_iter_next(&iter
);
2529 #define F_ALL_SEALS (F_SEAL_SEAL | \
2534 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2536 struct inode
*inode
= file_inode(file
);
2537 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2542 * Sealing allows multiple parties to share a shmem-file but restrict
2543 * access to a specific subset of file operations. Seals can only be
2544 * added, but never removed. This way, mutually untrusted parties can
2545 * share common memory regions with a well-defined policy. A malicious
2546 * peer can thus never perform unwanted operations on a shared object.
2548 * Seals are only supported on special shmem-files and always affect
2549 * the whole underlying inode. Once a seal is set, it may prevent some
2550 * kinds of access to the file. Currently, the following seals are
2552 * SEAL_SEAL: Prevent further seals from being set on this file
2553 * SEAL_SHRINK: Prevent the file from shrinking
2554 * SEAL_GROW: Prevent the file from growing
2555 * SEAL_WRITE: Prevent write access to the file
2557 * As we don't require any trust relationship between two parties, we
2558 * must prevent seals from being removed. Therefore, sealing a file
2559 * only adds a given set of seals to the file, it never touches
2560 * existing seals. Furthermore, the "setting seals"-operation can be
2561 * sealed itself, which basically prevents any further seal from being
2564 * Semantics of sealing are only defined on volatile files. Only
2565 * anonymous shmem files support sealing. More importantly, seals are
2566 * never written to disk. Therefore, there's no plan to support it on
2570 if (file
->f_op
!= &shmem_file_operations
)
2572 if (!(file
->f_mode
& FMODE_WRITE
))
2574 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2579 if (info
->seals
& F_SEAL_SEAL
) {
2584 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2585 error
= mapping_deny_writable(file
->f_mapping
);
2589 error
= shmem_wait_for_pins(file
->f_mapping
);
2591 mapping_allow_writable(file
->f_mapping
);
2596 info
->seals
|= seals
;
2600 inode_unlock(inode
);
2603 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2605 int shmem_get_seals(struct file
*file
)
2607 if (file
->f_op
!= &shmem_file_operations
)
2610 return SHMEM_I(file_inode(file
))->seals
;
2612 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2614 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2620 /* disallow upper 32bit */
2624 error
= shmem_add_seals(file
, arg
);
2627 error
= shmem_get_seals(file
);
2637 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2640 struct inode
*inode
= file_inode(file
);
2641 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2642 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2643 struct shmem_falloc shmem_falloc
;
2644 pgoff_t start
, index
, end
;
2647 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2652 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2653 struct address_space
*mapping
= file
->f_mapping
;
2654 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2655 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2656 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2658 /* protected by i_mutex */
2659 if (info
->seals
& F_SEAL_WRITE
) {
2664 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2665 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2666 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2667 spin_lock(&inode
->i_lock
);
2668 inode
->i_private
= &shmem_falloc
;
2669 spin_unlock(&inode
->i_lock
);
2671 if ((u64
)unmap_end
> (u64
)unmap_start
)
2672 unmap_mapping_range(mapping
, unmap_start
,
2673 1 + unmap_end
- unmap_start
, 0);
2674 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2675 /* No need to unmap again: hole-punching leaves COWed pages */
2677 spin_lock(&inode
->i_lock
);
2678 inode
->i_private
= NULL
;
2679 wake_up_all(&shmem_falloc_waitq
);
2680 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.task_list
));
2681 spin_unlock(&inode
->i_lock
);
2686 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2687 error
= inode_newsize_ok(inode
, offset
+ len
);
2691 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2696 start
= offset
>> PAGE_SHIFT
;
2697 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2698 /* Try to avoid a swapstorm if len is impossible to satisfy */
2699 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2704 shmem_falloc
.waitq
= NULL
;
2705 shmem_falloc
.start
= start
;
2706 shmem_falloc
.next
= start
;
2707 shmem_falloc
.nr_falloced
= 0;
2708 shmem_falloc
.nr_unswapped
= 0;
2709 spin_lock(&inode
->i_lock
);
2710 inode
->i_private
= &shmem_falloc
;
2711 spin_unlock(&inode
->i_lock
);
2713 for (index
= start
; index
< end
; index
++) {
2717 * Good, the fallocate(2) manpage permits EINTR: we may have
2718 * been interrupted because we are using up too much memory.
2720 if (signal_pending(current
))
2722 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2725 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2727 /* Remove the !PageUptodate pages we added */
2728 if (index
> start
) {
2729 shmem_undo_range(inode
,
2730 (loff_t
)start
<< PAGE_SHIFT
,
2731 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2737 * Inform shmem_writepage() how far we have reached.
2738 * No need for lock or barrier: we have the page lock.
2740 shmem_falloc
.next
++;
2741 if (!PageUptodate(page
))
2742 shmem_falloc
.nr_falloced
++;
2745 * If !PageUptodate, leave it that way so that freeable pages
2746 * can be recognized if we need to rollback on error later.
2747 * But set_page_dirty so that memory pressure will swap rather
2748 * than free the pages we are allocating (and SGP_CACHE pages
2749 * might still be clean: we now need to mark those dirty too).
2751 set_page_dirty(page
);
2757 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2758 i_size_write(inode
, offset
+ len
);
2759 inode
->i_ctime
= current_time(inode
);
2761 spin_lock(&inode
->i_lock
);
2762 inode
->i_private
= NULL
;
2763 spin_unlock(&inode
->i_lock
);
2765 inode_unlock(inode
);
2769 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2771 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2773 buf
->f_type
= TMPFS_MAGIC
;
2774 buf
->f_bsize
= PAGE_SIZE
;
2775 buf
->f_namelen
= NAME_MAX
;
2776 if (sbinfo
->max_blocks
) {
2777 buf
->f_blocks
= sbinfo
->max_blocks
;
2779 buf
->f_bfree
= sbinfo
->max_blocks
-
2780 percpu_counter_sum(&sbinfo
->used_blocks
);
2782 if (sbinfo
->max_inodes
) {
2783 buf
->f_files
= sbinfo
->max_inodes
;
2784 buf
->f_ffree
= sbinfo
->free_inodes
;
2786 /* else leave those fields 0 like simple_statfs */
2791 * File creation. Allocate an inode, and we're done..
2794 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2796 struct inode
*inode
;
2797 int error
= -ENOSPC
;
2799 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2801 error
= simple_acl_create(dir
, inode
);
2804 error
= security_inode_init_security(inode
, dir
,
2806 shmem_initxattrs
, NULL
);
2807 if (error
&& error
!= -EOPNOTSUPP
)
2811 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2812 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2813 d_instantiate(dentry
, inode
);
2814 dget(dentry
); /* Extra count - pin the dentry in core */
2823 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2825 struct inode
*inode
;
2826 int error
= -ENOSPC
;
2828 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2830 error
= security_inode_init_security(inode
, dir
,
2832 shmem_initxattrs
, NULL
);
2833 if (error
&& error
!= -EOPNOTSUPP
)
2835 error
= simple_acl_create(dir
, inode
);
2838 d_tmpfile(dentry
, inode
);
2846 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2850 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2856 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2859 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2865 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2867 struct inode
*inode
= d_inode(old_dentry
);
2871 * No ordinary (disk based) filesystem counts links as inodes;
2872 * but each new link needs a new dentry, pinning lowmem, and
2873 * tmpfs dentries cannot be pruned until they are unlinked.
2875 ret
= shmem_reserve_inode(inode
->i_sb
);
2879 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2880 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2882 ihold(inode
); /* New dentry reference */
2883 dget(dentry
); /* Extra pinning count for the created dentry */
2884 d_instantiate(dentry
, inode
);
2889 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2891 struct inode
*inode
= d_inode(dentry
);
2893 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2894 shmem_free_inode(inode
->i_sb
);
2896 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2897 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2899 dput(dentry
); /* Undo the count from "create" - this does all the work */
2903 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2905 if (!simple_empty(dentry
))
2908 drop_nlink(d_inode(dentry
));
2910 return shmem_unlink(dir
, dentry
);
2913 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2915 bool old_is_dir
= d_is_dir(old_dentry
);
2916 bool new_is_dir
= d_is_dir(new_dentry
);
2918 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2920 drop_nlink(old_dir
);
2923 drop_nlink(new_dir
);
2927 old_dir
->i_ctime
= old_dir
->i_mtime
=
2928 new_dir
->i_ctime
= new_dir
->i_mtime
=
2929 d_inode(old_dentry
)->i_ctime
=
2930 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
2935 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2937 struct dentry
*whiteout
;
2940 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2944 error
= shmem_mknod(old_dir
, whiteout
,
2945 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2951 * Cheat and hash the whiteout while the old dentry is still in
2952 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2954 * d_lookup() will consistently find one of them at this point,
2955 * not sure which one, but that isn't even important.
2962 * The VFS layer already does all the dentry stuff for rename,
2963 * we just have to decrement the usage count for the target if
2964 * it exists so that the VFS layer correctly free's it when it
2967 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2969 struct inode
*inode
= d_inode(old_dentry
);
2970 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2972 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2975 if (flags
& RENAME_EXCHANGE
)
2976 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2978 if (!simple_empty(new_dentry
))
2981 if (flags
& RENAME_WHITEOUT
) {
2984 error
= shmem_whiteout(old_dir
, old_dentry
);
2989 if (d_really_is_positive(new_dentry
)) {
2990 (void) shmem_unlink(new_dir
, new_dentry
);
2991 if (they_are_dirs
) {
2992 drop_nlink(d_inode(new_dentry
));
2993 drop_nlink(old_dir
);
2995 } else if (they_are_dirs
) {
2996 drop_nlink(old_dir
);
3000 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3001 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3002 old_dir
->i_ctime
= old_dir
->i_mtime
=
3003 new_dir
->i_ctime
= new_dir
->i_mtime
=
3004 inode
->i_ctime
= current_time(old_dir
);
3008 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3012 struct inode
*inode
;
3014 struct shmem_inode_info
*info
;
3016 len
= strlen(symname
) + 1;
3017 if (len
> PAGE_SIZE
)
3018 return -ENAMETOOLONG
;
3020 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
3024 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3025 shmem_initxattrs
, NULL
);
3027 if (error
!= -EOPNOTSUPP
) {
3034 info
= SHMEM_I(inode
);
3035 inode
->i_size
= len
-1;
3036 if (len
<= SHORT_SYMLINK_LEN
) {
3037 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3038 if (!inode
->i_link
) {
3042 inode
->i_op
= &shmem_short_symlink_operations
;
3044 inode_nohighmem(inode
);
3045 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3050 inode
->i_mapping
->a_ops
= &shmem_aops
;
3051 inode
->i_op
= &shmem_symlink_inode_operations
;
3052 memcpy(page_address(page
), symname
, len
);
3053 SetPageUptodate(page
);
3054 set_page_dirty(page
);
3058 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3059 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3060 d_instantiate(dentry
, inode
);
3065 static void shmem_put_link(void *arg
)
3067 mark_page_accessed(arg
);
3071 static const char *shmem_get_link(struct dentry
*dentry
,
3072 struct inode
*inode
,
3073 struct delayed_call
*done
)
3075 struct page
*page
= NULL
;
3078 page
= find_get_page(inode
->i_mapping
, 0);
3080 return ERR_PTR(-ECHILD
);
3081 if (!PageUptodate(page
)) {
3083 return ERR_PTR(-ECHILD
);
3086 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3088 return ERR_PTR(error
);
3091 set_delayed_call(done
, shmem_put_link
, page
);
3092 return page_address(page
);
3095 #ifdef CONFIG_TMPFS_XATTR
3097 * Superblocks without xattr inode operations may get some security.* xattr
3098 * support from the LSM "for free". As soon as we have any other xattrs
3099 * like ACLs, we also need to implement the security.* handlers at
3100 * filesystem level, though.
3104 * Callback for security_inode_init_security() for acquiring xattrs.
3106 static int shmem_initxattrs(struct inode
*inode
,
3107 const struct xattr
*xattr_array
,
3110 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3111 const struct xattr
*xattr
;
3112 struct simple_xattr
*new_xattr
;
3115 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3116 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3120 len
= strlen(xattr
->name
) + 1;
3121 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3123 if (!new_xattr
->name
) {
3128 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3129 XATTR_SECURITY_PREFIX_LEN
);
3130 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3133 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3139 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3140 struct dentry
*unused
, struct inode
*inode
,
3141 const char *name
, void *buffer
, size_t size
)
3143 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3145 name
= xattr_full_name(handler
, name
);
3146 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3149 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3150 struct dentry
*unused
, struct inode
*inode
,
3151 const char *name
, const void *value
,
3152 size_t size
, int flags
)
3154 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3156 name
= xattr_full_name(handler
, name
);
3157 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3160 static const struct xattr_handler shmem_security_xattr_handler
= {
3161 .prefix
= XATTR_SECURITY_PREFIX
,
3162 .get
= shmem_xattr_handler_get
,
3163 .set
= shmem_xattr_handler_set
,
3166 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3167 .prefix
= XATTR_TRUSTED_PREFIX
,
3168 .get
= shmem_xattr_handler_get
,
3169 .set
= shmem_xattr_handler_set
,
3172 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3173 #ifdef CONFIG_TMPFS_POSIX_ACL
3174 &posix_acl_access_xattr_handler
,
3175 &posix_acl_default_xattr_handler
,
3177 &shmem_security_xattr_handler
,
3178 &shmem_trusted_xattr_handler
,
3182 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3184 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3185 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3187 #endif /* CONFIG_TMPFS_XATTR */
3189 static const struct inode_operations shmem_short_symlink_operations
= {
3190 .readlink
= generic_readlink
,
3191 .get_link
= simple_get_link
,
3192 #ifdef CONFIG_TMPFS_XATTR
3193 .listxattr
= shmem_listxattr
,
3197 static const struct inode_operations shmem_symlink_inode_operations
= {
3198 .readlink
= generic_readlink
,
3199 .get_link
= shmem_get_link
,
3200 #ifdef CONFIG_TMPFS_XATTR
3201 .listxattr
= shmem_listxattr
,
3205 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3207 return ERR_PTR(-ESTALE
);
3210 static int shmem_match(struct inode
*ino
, void *vfh
)
3214 inum
= (inum
<< 32) | fh
[1];
3215 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3218 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3219 struct fid
*fid
, int fh_len
, int fh_type
)
3221 struct inode
*inode
;
3222 struct dentry
*dentry
= NULL
;
3229 inum
= (inum
<< 32) | fid
->raw
[1];
3231 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3232 shmem_match
, fid
->raw
);
3234 dentry
= d_find_alias(inode
);
3241 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3242 struct inode
*parent
)
3246 return FILEID_INVALID
;
3249 if (inode_unhashed(inode
)) {
3250 /* Unfortunately insert_inode_hash is not idempotent,
3251 * so as we hash inodes here rather than at creation
3252 * time, we need a lock to ensure we only try
3255 static DEFINE_SPINLOCK(lock
);
3257 if (inode_unhashed(inode
))
3258 __insert_inode_hash(inode
,
3259 inode
->i_ino
+ inode
->i_generation
);
3263 fh
[0] = inode
->i_generation
;
3264 fh
[1] = inode
->i_ino
;
3265 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3271 static const struct export_operations shmem_export_ops
= {
3272 .get_parent
= shmem_get_parent
,
3273 .encode_fh
= shmem_encode_fh
,
3274 .fh_to_dentry
= shmem_fh_to_dentry
,
3277 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3280 char *this_char
, *value
, *rest
;
3281 struct mempolicy
*mpol
= NULL
;
3285 while (options
!= NULL
) {
3286 this_char
= options
;
3289 * NUL-terminate this option: unfortunately,
3290 * mount options form a comma-separated list,
3291 * but mpol's nodelist may also contain commas.
3293 options
= strchr(options
, ',');
3294 if (options
== NULL
)
3297 if (!isdigit(*options
)) {
3304 if ((value
= strchr(this_char
,'=')) != NULL
) {
3307 pr_err("tmpfs: No value for mount option '%s'\n",
3312 if (!strcmp(this_char
,"size")) {
3313 unsigned long long size
;
3314 size
= memparse(value
,&rest
);
3316 size
<<= PAGE_SHIFT
;
3317 size
*= totalram_pages
;
3323 sbinfo
->max_blocks
=
3324 DIV_ROUND_UP(size
, PAGE_SIZE
);
3325 } else if (!strcmp(this_char
,"nr_blocks")) {
3326 sbinfo
->max_blocks
= memparse(value
, &rest
);
3329 } else if (!strcmp(this_char
,"nr_inodes")) {
3330 sbinfo
->max_inodes
= memparse(value
, &rest
);
3333 } else if (!strcmp(this_char
,"mode")) {
3336 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3339 } else if (!strcmp(this_char
,"uid")) {
3342 uid
= simple_strtoul(value
, &rest
, 0);
3345 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3346 if (!uid_valid(sbinfo
->uid
))
3348 } else if (!strcmp(this_char
,"gid")) {
3351 gid
= simple_strtoul(value
, &rest
, 0);
3354 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3355 if (!gid_valid(sbinfo
->gid
))
3357 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3358 } else if (!strcmp(this_char
, "huge")) {
3360 huge
= shmem_parse_huge(value
);
3363 if (!has_transparent_hugepage() &&
3364 huge
!= SHMEM_HUGE_NEVER
)
3366 sbinfo
->huge
= huge
;
3369 } else if (!strcmp(this_char
,"mpol")) {
3372 if (mpol_parse_str(value
, &mpol
))
3376 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3380 sbinfo
->mpol
= mpol
;
3384 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3392 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3394 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3395 struct shmem_sb_info config
= *sbinfo
;
3396 unsigned long inodes
;
3397 int error
= -EINVAL
;
3400 if (shmem_parse_options(data
, &config
, true))
3403 spin_lock(&sbinfo
->stat_lock
);
3404 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3405 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3407 if (config
.max_inodes
< inodes
)
3410 * Those tests disallow limited->unlimited while any are in use;
3411 * but we must separately disallow unlimited->limited, because
3412 * in that case we have no record of how much is already in use.
3414 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3416 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3420 sbinfo
->huge
= config
.huge
;
3421 sbinfo
->max_blocks
= config
.max_blocks
;
3422 sbinfo
->max_inodes
= config
.max_inodes
;
3423 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3426 * Preserve previous mempolicy unless mpol remount option was specified.
3429 mpol_put(sbinfo
->mpol
);
3430 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3433 spin_unlock(&sbinfo
->stat_lock
);
3437 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3439 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3441 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3442 seq_printf(seq
, ",size=%luk",
3443 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3444 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3445 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3446 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3447 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3448 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3449 seq_printf(seq
, ",uid=%u",
3450 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3451 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3452 seq_printf(seq
, ",gid=%u",
3453 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3454 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3455 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3457 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3459 shmem_show_mpol(seq
, sbinfo
->mpol
);
3463 #define MFD_NAME_PREFIX "memfd:"
3464 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3465 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3467 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3469 SYSCALL_DEFINE2(memfd_create
,
3470 const char __user
*, uname
,
3471 unsigned int, flags
)
3473 struct shmem_inode_info
*info
;
3479 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3482 /* length includes terminating zero */
3483 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3486 if (len
> MFD_NAME_MAX_LEN
+ 1)
3489 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
3493 strcpy(name
, MFD_NAME_PREFIX
);
3494 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3499 /* terminating-zero may have changed after strnlen_user() returned */
3500 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3505 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3511 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3513 error
= PTR_ERR(file
);
3516 info
= SHMEM_I(file_inode(file
));
3517 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3518 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3519 if (flags
& MFD_ALLOW_SEALING
)
3520 info
->seals
&= ~F_SEAL_SEAL
;
3522 fd_install(fd
, file
);
3533 #endif /* CONFIG_TMPFS */
3535 static void shmem_put_super(struct super_block
*sb
)
3537 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3539 percpu_counter_destroy(&sbinfo
->used_blocks
);
3540 mpol_put(sbinfo
->mpol
);
3542 sb
->s_fs_info
= NULL
;
3545 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3547 struct inode
*inode
;
3548 struct shmem_sb_info
*sbinfo
;
3551 /* Round up to L1_CACHE_BYTES to resist false sharing */
3552 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3553 L1_CACHE_BYTES
), GFP_KERNEL
);
3557 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3558 sbinfo
->uid
= current_fsuid();
3559 sbinfo
->gid
= current_fsgid();
3560 sb
->s_fs_info
= sbinfo
;
3564 * Per default we only allow half of the physical ram per
3565 * tmpfs instance, limiting inodes to one per page of lowmem;
3566 * but the internal instance is left unlimited.
3568 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3569 sbinfo
->max_blocks
= shmem_default_max_blocks();
3570 sbinfo
->max_inodes
= shmem_default_max_inodes();
3571 if (shmem_parse_options(data
, sbinfo
, false)) {
3576 sb
->s_flags
|= MS_NOUSER
;
3578 sb
->s_export_op
= &shmem_export_ops
;
3579 sb
->s_flags
|= MS_NOSEC
;
3581 sb
->s_flags
|= MS_NOUSER
;
3584 spin_lock_init(&sbinfo
->stat_lock
);
3585 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3587 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3588 spin_lock_init(&sbinfo
->shrinklist_lock
);
3589 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3591 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3592 sb
->s_blocksize
= PAGE_SIZE
;
3593 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3594 sb
->s_magic
= TMPFS_MAGIC
;
3595 sb
->s_op
= &shmem_ops
;
3596 sb
->s_time_gran
= 1;
3597 #ifdef CONFIG_TMPFS_XATTR
3598 sb
->s_xattr
= shmem_xattr_handlers
;
3600 #ifdef CONFIG_TMPFS_POSIX_ACL
3601 sb
->s_flags
|= MS_POSIXACL
;
3604 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3607 inode
->i_uid
= sbinfo
->uid
;
3608 inode
->i_gid
= sbinfo
->gid
;
3609 sb
->s_root
= d_make_root(inode
);
3615 shmem_put_super(sb
);
3619 static struct kmem_cache
*shmem_inode_cachep
;
3621 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3623 struct shmem_inode_info
*info
;
3624 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3627 return &info
->vfs_inode
;
3630 static void shmem_destroy_callback(struct rcu_head
*head
)
3632 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3633 if (S_ISLNK(inode
->i_mode
))
3634 kfree(inode
->i_link
);
3635 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3638 static void shmem_destroy_inode(struct inode
*inode
)
3640 if (S_ISREG(inode
->i_mode
))
3641 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3642 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3645 static void shmem_init_inode(void *foo
)
3647 struct shmem_inode_info
*info
= foo
;
3648 inode_init_once(&info
->vfs_inode
);
3651 static int shmem_init_inodecache(void)
3653 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3654 sizeof(struct shmem_inode_info
),
3655 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3659 static void shmem_destroy_inodecache(void)
3661 kmem_cache_destroy(shmem_inode_cachep
);
3664 static const struct address_space_operations shmem_aops
= {
3665 .writepage
= shmem_writepage
,
3666 .set_page_dirty
= __set_page_dirty_no_writeback
,
3668 .write_begin
= shmem_write_begin
,
3669 .write_end
= shmem_write_end
,
3671 #ifdef CONFIG_MIGRATION
3672 .migratepage
= migrate_page
,
3674 .error_remove_page
= generic_error_remove_page
,
3677 static const struct file_operations shmem_file_operations
= {
3679 .get_unmapped_area
= shmem_get_unmapped_area
,
3681 .llseek
= shmem_file_llseek
,
3682 .read_iter
= shmem_file_read_iter
,
3683 .write_iter
= generic_file_write_iter
,
3684 .fsync
= noop_fsync
,
3685 .splice_read
= generic_file_splice_read
,
3686 .splice_write
= iter_file_splice_write
,
3687 .fallocate
= shmem_fallocate
,
3691 static const struct inode_operations shmem_inode_operations
= {
3692 .getattr
= shmem_getattr
,
3693 .setattr
= shmem_setattr
,
3694 #ifdef CONFIG_TMPFS_XATTR
3695 .listxattr
= shmem_listxattr
,
3696 .set_acl
= simple_set_acl
,
3700 static const struct inode_operations shmem_dir_inode_operations
= {
3702 .create
= shmem_create
,
3703 .lookup
= simple_lookup
,
3705 .unlink
= shmem_unlink
,
3706 .symlink
= shmem_symlink
,
3707 .mkdir
= shmem_mkdir
,
3708 .rmdir
= shmem_rmdir
,
3709 .mknod
= shmem_mknod
,
3710 .rename
= shmem_rename2
,
3711 .tmpfile
= shmem_tmpfile
,
3713 #ifdef CONFIG_TMPFS_XATTR
3714 .listxattr
= shmem_listxattr
,
3716 #ifdef CONFIG_TMPFS_POSIX_ACL
3717 .setattr
= shmem_setattr
,
3718 .set_acl
= simple_set_acl
,
3722 static const struct inode_operations shmem_special_inode_operations
= {
3723 #ifdef CONFIG_TMPFS_XATTR
3724 .listxattr
= shmem_listxattr
,
3726 #ifdef CONFIG_TMPFS_POSIX_ACL
3727 .setattr
= shmem_setattr
,
3728 .set_acl
= simple_set_acl
,
3732 static const struct super_operations shmem_ops
= {
3733 .alloc_inode
= shmem_alloc_inode
,
3734 .destroy_inode
= shmem_destroy_inode
,
3736 .statfs
= shmem_statfs
,
3737 .remount_fs
= shmem_remount_fs
,
3738 .show_options
= shmem_show_options
,
3740 .evict_inode
= shmem_evict_inode
,
3741 .drop_inode
= generic_delete_inode
,
3742 .put_super
= shmem_put_super
,
3743 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3744 .nr_cached_objects
= shmem_unused_huge_count
,
3745 .free_cached_objects
= shmem_unused_huge_scan
,
3749 static const struct vm_operations_struct shmem_vm_ops
= {
3750 .fault
= shmem_fault
,
3751 .map_pages
= filemap_map_pages
,
3753 .set_policy
= shmem_set_policy
,
3754 .get_policy
= shmem_get_policy
,
3758 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3759 int flags
, const char *dev_name
, void *data
)
3761 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3764 static struct file_system_type shmem_fs_type
= {
3765 .owner
= THIS_MODULE
,
3767 .mount
= shmem_mount
,
3768 .kill_sb
= kill_litter_super
,
3769 .fs_flags
= FS_USERNS_MOUNT
,
3772 int __init
shmem_init(void)
3776 /* If rootfs called this, don't re-init */
3777 if (shmem_inode_cachep
)
3780 error
= shmem_init_inodecache();
3784 error
= register_filesystem(&shmem_fs_type
);
3786 pr_err("Could not register tmpfs\n");
3790 shm_mnt
= kern_mount(&shmem_fs_type
);
3791 if (IS_ERR(shm_mnt
)) {
3792 error
= PTR_ERR(shm_mnt
);
3793 pr_err("Could not kern_mount tmpfs\n");
3797 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3798 if (has_transparent_hugepage() && shmem_huge
< SHMEM_HUGE_DENY
)
3799 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3801 shmem_huge
= 0; /* just in case it was patched */
3806 unregister_filesystem(&shmem_fs_type
);
3808 shmem_destroy_inodecache();
3810 shm_mnt
= ERR_PTR(error
);
3814 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3815 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3816 struct kobj_attribute
*attr
, char *buf
)
3820 SHMEM_HUGE_WITHIN_SIZE
,
3828 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3829 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3831 count
+= sprintf(buf
+ count
, fmt
,
3832 shmem_format_huge(values
[i
]));
3834 buf
[count
- 1] = '\n';
3838 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3839 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3844 if (count
+ 1 > sizeof(tmp
))
3846 memcpy(tmp
, buf
, count
);
3848 if (count
&& tmp
[count
- 1] == '\n')
3849 tmp
[count
- 1] = '\0';
3851 huge
= shmem_parse_huge(tmp
);
3852 if (huge
== -EINVAL
)
3854 if (!has_transparent_hugepage() &&
3855 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3859 if (shmem_huge
< SHMEM_HUGE_DENY
)
3860 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3864 struct kobj_attribute shmem_enabled_attr
=
3865 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3866 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3868 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3869 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
3871 struct inode
*inode
= file_inode(vma
->vm_file
);
3872 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3876 if (shmem_huge
== SHMEM_HUGE_FORCE
)
3878 if (shmem_huge
== SHMEM_HUGE_DENY
)
3880 switch (sbinfo
->huge
) {
3881 case SHMEM_HUGE_NEVER
:
3883 case SHMEM_HUGE_ALWAYS
:
3885 case SHMEM_HUGE_WITHIN_SIZE
:
3886 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
3887 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
3888 if (i_size
>= HPAGE_PMD_SIZE
&&
3889 i_size
>> PAGE_SHIFT
>= off
)
3891 case SHMEM_HUGE_ADVISE
:
3892 /* TODO: implement fadvise() hints */
3893 return (vma
->vm_flags
& VM_HUGEPAGE
);
3899 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3901 #else /* !CONFIG_SHMEM */
3904 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3906 * This is intended for small system where the benefits of the full
3907 * shmem code (swap-backed and resource-limited) are outweighed by
3908 * their complexity. On systems without swap this code should be
3909 * effectively equivalent, but much lighter weight.
3912 static struct file_system_type shmem_fs_type
= {
3914 .mount
= ramfs_mount
,
3915 .kill_sb
= kill_litter_super
,
3916 .fs_flags
= FS_USERNS_MOUNT
,
3919 int __init
shmem_init(void)
3921 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3923 shm_mnt
= kern_mount(&shmem_fs_type
);
3924 BUG_ON(IS_ERR(shm_mnt
));
3929 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3934 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3939 void shmem_unlock_mapping(struct address_space
*mapping
)
3944 unsigned long shmem_get_unmapped_area(struct file
*file
,
3945 unsigned long addr
, unsigned long len
,
3946 unsigned long pgoff
, unsigned long flags
)
3948 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
3952 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3954 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3956 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3958 #define shmem_vm_ops generic_file_vm_ops
3959 #define shmem_file_operations ramfs_file_operations
3960 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3961 #define shmem_acct_size(flags, size) 0
3962 #define shmem_unacct_size(flags, size) do {} while (0)
3964 #endif /* CONFIG_SHMEM */
3968 static const struct dentry_operations anon_ops
= {
3969 .d_dname
= simple_dname
3972 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3973 unsigned long flags
, unsigned int i_flags
)
3976 struct inode
*inode
;
3978 struct super_block
*sb
;
3981 if (IS_ERR(shm_mnt
))
3982 return ERR_CAST(shm_mnt
);
3984 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3985 return ERR_PTR(-EINVAL
);
3987 if (shmem_acct_size(flags
, size
))
3988 return ERR_PTR(-ENOMEM
);
3990 res
= ERR_PTR(-ENOMEM
);
3992 this.len
= strlen(name
);
3993 this.hash
= 0; /* will go */
3994 sb
= shm_mnt
->mnt_sb
;
3995 path
.mnt
= mntget(shm_mnt
);
3996 path
.dentry
= d_alloc_pseudo(sb
, &this);
3999 d_set_d_op(path
.dentry
, &anon_ops
);
4001 res
= ERR_PTR(-ENOSPC
);
4002 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
4006 inode
->i_flags
|= i_flags
;
4007 d_instantiate(path
.dentry
, inode
);
4008 inode
->i_size
= size
;
4009 clear_nlink(inode
); /* It is unlinked */
4010 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4014 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
4015 &shmem_file_operations
);
4022 shmem_unacct_size(flags
, size
);
4029 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4030 * kernel internal. There will be NO LSM permission checks against the
4031 * underlying inode. So users of this interface must do LSM checks at a
4032 * higher layer. The users are the big_key and shm implementations. LSM
4033 * checks are provided at the key or shm level rather than the inode.
4034 * @name: name for dentry (to be seen in /proc/<pid>/maps
4035 * @size: size to be set for the file
4036 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4038 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4040 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
4044 * shmem_file_setup - get an unlinked file living in tmpfs
4045 * @name: name for dentry (to be seen in /proc/<pid>/maps
4046 * @size: size to be set for the file
4047 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4049 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4051 return __shmem_file_setup(name
, size
, flags
, 0);
4053 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4056 * shmem_zero_setup - setup a shared anonymous mapping
4057 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4059 int shmem_zero_setup(struct vm_area_struct
*vma
)
4062 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4065 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4066 * between XFS directory reading and selinux: since this file is only
4067 * accessible to the user through its mapping, use S_PRIVATE flag to
4068 * bypass file security, in the same way as shmem_kernel_file_setup().
4070 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
4072 return PTR_ERR(file
);
4076 vma
->vm_file
= file
;
4077 vma
->vm_ops
= &shmem_vm_ops
;
4079 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4080 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4081 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4082 khugepaged_enter(vma
, vma
->vm_flags
);
4089 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4090 * @mapping: the page's address_space
4091 * @index: the page index
4092 * @gfp: the page allocator flags to use if allocating
4094 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4095 * with any new page allocations done using the specified allocation flags.
4096 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4097 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4098 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4100 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4101 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4103 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4104 pgoff_t index
, gfp_t gfp
)
4107 struct inode
*inode
= mapping
->host
;
4111 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4112 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4115 page
= ERR_PTR(error
);
4121 * The tiny !SHMEM case uses ramfs without swap
4123 return read_cache_page_gfp(mapping
, index
, gfp
);
4126 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);