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/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
42 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
44 static struct vfsmount
*shm_mnt
;
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/blkdev.h>
64 #include <linux/pagevec.h>
65 #include <linux/percpu_counter.h>
66 #include <linux/falloc.h>
67 #include <linux/splice.h>
68 #include <linux/security.h>
69 #include <linux/swapops.h>
70 #include <linux/mempolicy.h>
71 #include <linux/namei.h>
72 #include <linux/ctype.h>
73 #include <linux/migrate.h>
74 #include <linux/highmem.h>
75 #include <linux/seq_file.h>
76 #include <linux/magic.h>
77 #include <linux/syscalls.h>
78 #include <linux/fcntl.h>
79 #include <uapi/linux/memfd.h>
80 #include <linux/userfaultfd_k.h>
81 #include <linux/rmap.h>
82 #include <linux/uuid.h>
84 #include <linux/uaccess.h>
85 #include <asm/pgtable.h>
89 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
90 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
92 /* Pretend that each entry is of this size in directory's i_size */
93 #define BOGO_DIRENT_SIZE 20
95 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
96 #define SHORT_SYMLINK_LEN 128
99 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
100 * inode->i_private (with i_mutex making sure that it has only one user at
101 * a time): we would prefer not to enlarge the shmem inode just for that.
103 struct shmem_falloc
{
104 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
105 pgoff_t start
; /* start of range currently being fallocated */
106 pgoff_t next
; /* the next page offset to be fallocated */
107 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
108 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
111 struct shmem_options
{
112 unsigned long long blocks
;
113 unsigned long long inodes
;
114 struct mempolicy
*mpol
;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
126 static unsigned long shmem_default_max_blocks(void)
128 return totalram_pages() / 2;
131 static unsigned long shmem_default_max_inodes(void)
133 unsigned long nr_pages
= totalram_pages();
135 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
139 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
140 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
141 struct shmem_inode_info
*info
, pgoff_t index
);
142 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
143 struct page
**pagep
, enum sgp_type sgp
,
144 gfp_t gfp
, struct vm_area_struct
*vma
,
145 vm_fault_t
*fault_type
);
146 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
147 struct page
**pagep
, enum sgp_type sgp
,
148 gfp_t gfp
, struct vm_area_struct
*vma
,
149 struct vm_fault
*vmf
, vm_fault_t
*fault_type
);
151 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
152 struct page
**pagep
, enum sgp_type sgp
)
154 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
155 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
158 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
160 return sb
->s_fs_info
;
164 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
165 * for shared memory and for shared anonymous (/dev/zero) mappings
166 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
167 * consistent with the pre-accounting of private mappings ...
169 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
171 return (flags
& VM_NORESERVE
) ?
172 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
175 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
177 if (!(flags
& VM_NORESERVE
))
178 vm_unacct_memory(VM_ACCT(size
));
181 static inline int shmem_reacct_size(unsigned long flags
,
182 loff_t oldsize
, loff_t newsize
)
184 if (!(flags
& VM_NORESERVE
)) {
185 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
186 return security_vm_enough_memory_mm(current
->mm
,
187 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
188 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
189 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
195 * ... whereas tmpfs objects are accounted incrementally as
196 * pages are allocated, in order to allow large sparse files.
197 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
198 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
200 static inline int shmem_acct_block(unsigned long flags
, long pages
)
202 if (!(flags
& VM_NORESERVE
))
205 return security_vm_enough_memory_mm(current
->mm
,
206 pages
* VM_ACCT(PAGE_SIZE
));
209 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
211 if (flags
& VM_NORESERVE
)
212 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
215 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
217 struct shmem_inode_info
*info
= SHMEM_I(inode
);
218 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
220 if (shmem_acct_block(info
->flags
, pages
))
223 if (sbinfo
->max_blocks
) {
224 if (percpu_counter_compare(&sbinfo
->used_blocks
,
225 sbinfo
->max_blocks
- pages
) > 0)
227 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
233 shmem_unacct_blocks(info
->flags
, pages
);
237 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
239 struct shmem_inode_info
*info
= SHMEM_I(inode
);
240 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
242 if (sbinfo
->max_blocks
)
243 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
244 shmem_unacct_blocks(info
->flags
, pages
);
247 static const struct super_operations shmem_ops
;
248 static const struct address_space_operations shmem_aops
;
249 static const struct file_operations shmem_file_operations
;
250 static const struct inode_operations shmem_inode_operations
;
251 static const struct inode_operations shmem_dir_inode_operations
;
252 static const struct inode_operations shmem_special_inode_operations
;
253 static const struct vm_operations_struct shmem_vm_ops
;
254 static struct file_system_type shmem_fs_type
;
256 bool vma_is_shmem(struct vm_area_struct
*vma
)
258 return vma
->vm_ops
== &shmem_vm_ops
;
261 static LIST_HEAD(shmem_swaplist
);
262 static DEFINE_MUTEX(shmem_swaplist_mutex
);
264 static int shmem_reserve_inode(struct super_block
*sb
)
266 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
267 if (sbinfo
->max_inodes
) {
268 spin_lock(&sbinfo
->stat_lock
);
269 if (!sbinfo
->free_inodes
) {
270 spin_unlock(&sbinfo
->stat_lock
);
273 sbinfo
->free_inodes
--;
274 spin_unlock(&sbinfo
->stat_lock
);
279 static void shmem_free_inode(struct super_block
*sb
)
281 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
282 if (sbinfo
->max_inodes
) {
283 spin_lock(&sbinfo
->stat_lock
);
284 sbinfo
->free_inodes
++;
285 spin_unlock(&sbinfo
->stat_lock
);
290 * shmem_recalc_inode - recalculate the block usage of an inode
291 * @inode: inode to recalc
293 * We have to calculate the free blocks since the mm can drop
294 * undirtied hole pages behind our back.
296 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
297 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
299 * It has to be called with the spinlock held.
301 static void shmem_recalc_inode(struct inode
*inode
)
303 struct shmem_inode_info
*info
= SHMEM_I(inode
);
306 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
308 info
->alloced
-= freed
;
309 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
310 shmem_inode_unacct_blocks(inode
, freed
);
314 bool shmem_charge(struct inode
*inode
, long pages
)
316 struct shmem_inode_info
*info
= SHMEM_I(inode
);
319 if (!shmem_inode_acct_block(inode
, pages
))
322 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
323 inode
->i_mapping
->nrpages
+= pages
;
325 spin_lock_irqsave(&info
->lock
, flags
);
326 info
->alloced
+= pages
;
327 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
328 shmem_recalc_inode(inode
);
329 spin_unlock_irqrestore(&info
->lock
, flags
);
334 void shmem_uncharge(struct inode
*inode
, long pages
)
336 struct shmem_inode_info
*info
= SHMEM_I(inode
);
339 /* nrpages adjustment done by __delete_from_page_cache() or caller */
341 spin_lock_irqsave(&info
->lock
, flags
);
342 info
->alloced
-= pages
;
343 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
344 shmem_recalc_inode(inode
);
345 spin_unlock_irqrestore(&info
->lock
, flags
);
347 shmem_inode_unacct_blocks(inode
, pages
);
351 * Replace item expected in xarray by a new item, while holding xa_lock.
353 static int shmem_replace_entry(struct address_space
*mapping
,
354 pgoff_t index
, void *expected
, void *replacement
)
356 XA_STATE(xas
, &mapping
->i_pages
, index
);
359 VM_BUG_ON(!expected
);
360 VM_BUG_ON(!replacement
);
361 item
= xas_load(&xas
);
362 if (item
!= expected
)
364 xas_store(&xas
, replacement
);
369 * Sometimes, before we decide whether to proceed or to fail, we must check
370 * that an entry was not already brought back from swap by a racing thread.
372 * Checking page is not enough: by the time a SwapCache page is locked, it
373 * might be reused, and again be SwapCache, using the same swap as before.
375 static bool shmem_confirm_swap(struct address_space
*mapping
,
376 pgoff_t index
, swp_entry_t swap
)
378 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
382 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
385 * disables huge pages for the mount;
387 * enables huge pages for the mount;
388 * SHMEM_HUGE_WITHIN_SIZE:
389 * only allocate huge pages if the page will be fully within i_size,
390 * also respect fadvise()/madvise() hints;
392 * only allocate huge pages if requested with fadvise()/madvise();
395 #define SHMEM_HUGE_NEVER 0
396 #define SHMEM_HUGE_ALWAYS 1
397 #define SHMEM_HUGE_WITHIN_SIZE 2
398 #define SHMEM_HUGE_ADVISE 3
402 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
405 * disables huge on shm_mnt and all mounts, for emergency use;
407 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
410 #define SHMEM_HUGE_DENY (-1)
411 #define SHMEM_HUGE_FORCE (-2)
413 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
414 /* ifdef here to avoid bloating shmem.o when not necessary */
416 static int shmem_huge __read_mostly
;
418 #if defined(CONFIG_SYSFS)
419 static int shmem_parse_huge(const char *str
)
421 if (!strcmp(str
, "never"))
422 return SHMEM_HUGE_NEVER
;
423 if (!strcmp(str
, "always"))
424 return SHMEM_HUGE_ALWAYS
;
425 if (!strcmp(str
, "within_size"))
426 return SHMEM_HUGE_WITHIN_SIZE
;
427 if (!strcmp(str
, "advise"))
428 return SHMEM_HUGE_ADVISE
;
429 if (!strcmp(str
, "deny"))
430 return SHMEM_HUGE_DENY
;
431 if (!strcmp(str
, "force"))
432 return SHMEM_HUGE_FORCE
;
437 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
438 static const char *shmem_format_huge(int huge
)
441 case SHMEM_HUGE_NEVER
:
443 case SHMEM_HUGE_ALWAYS
:
445 case SHMEM_HUGE_WITHIN_SIZE
:
446 return "within_size";
447 case SHMEM_HUGE_ADVISE
:
449 case SHMEM_HUGE_DENY
:
451 case SHMEM_HUGE_FORCE
:
460 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
461 struct shrink_control
*sc
, unsigned long nr_to_split
)
463 LIST_HEAD(list
), *pos
, *next
;
464 LIST_HEAD(to_remove
);
466 struct shmem_inode_info
*info
;
468 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
469 int removed
= 0, split
= 0;
471 if (list_empty(&sbinfo
->shrinklist
))
474 spin_lock(&sbinfo
->shrinklist_lock
);
475 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
476 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
479 inode
= igrab(&info
->vfs_inode
);
481 /* inode is about to be evicted */
483 list_del_init(&info
->shrinklist
);
488 /* Check if there's anything to gain */
489 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
490 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
491 list_move(&info
->shrinklist
, &to_remove
);
496 list_move(&info
->shrinklist
, &list
);
501 spin_unlock(&sbinfo
->shrinklist_lock
);
503 list_for_each_safe(pos
, next
, &to_remove
) {
504 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
505 inode
= &info
->vfs_inode
;
506 list_del_init(&info
->shrinklist
);
510 list_for_each_safe(pos
, next
, &list
) {
513 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
514 inode
= &info
->vfs_inode
;
516 if (nr_to_split
&& split
>= nr_to_split
)
519 page
= find_get_page(inode
->i_mapping
,
520 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
524 /* No huge page at the end of the file: nothing to split */
525 if (!PageTransHuge(page
)) {
531 * Leave the inode on the list if we failed to lock
532 * the page at this time.
534 * Waiting for the lock may lead to deadlock in the
537 if (!trylock_page(page
)) {
542 ret
= split_huge_page(page
);
546 /* If split failed leave the inode on the list */
552 list_del_init(&info
->shrinklist
);
558 spin_lock(&sbinfo
->shrinklist_lock
);
559 list_splice_tail(&list
, &sbinfo
->shrinklist
);
560 sbinfo
->shrinklist_len
-= removed
;
561 spin_unlock(&sbinfo
->shrinklist_lock
);
566 static long shmem_unused_huge_scan(struct super_block
*sb
,
567 struct shrink_control
*sc
)
569 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
571 if (!READ_ONCE(sbinfo
->shrinklist_len
))
574 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
577 static long shmem_unused_huge_count(struct super_block
*sb
,
578 struct shrink_control
*sc
)
580 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
581 return READ_ONCE(sbinfo
->shrinklist_len
);
583 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
585 #define shmem_huge SHMEM_HUGE_DENY
587 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
588 struct shrink_control
*sc
, unsigned long nr_to_split
)
592 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
594 static inline bool is_huge_enabled(struct shmem_sb_info
*sbinfo
)
596 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
597 (shmem_huge
== SHMEM_HUGE_FORCE
|| sbinfo
->huge
) &&
598 shmem_huge
!= SHMEM_HUGE_DENY
)
604 * Like add_to_page_cache_locked, but error if expected item has gone.
606 static int shmem_add_to_page_cache(struct page
*page
,
607 struct address_space
*mapping
,
608 pgoff_t index
, void *expected
, gfp_t gfp
)
610 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, compound_order(page
));
612 unsigned long nr
= compound_nr(page
);
614 VM_BUG_ON_PAGE(PageTail(page
), page
);
615 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
616 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
617 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
618 VM_BUG_ON(expected
&& PageTransHuge(page
));
620 page_ref_add(page
, nr
);
621 page
->mapping
= mapping
;
627 entry
= xas_find_conflict(&xas
);
628 if (entry
!= expected
)
629 xas_set_err(&xas
, -EEXIST
);
630 xas_create_range(&xas
);
634 xas_store(&xas
, page
);
639 if (PageTransHuge(page
)) {
640 count_vm_event(THP_FILE_ALLOC
);
641 __inc_node_page_state(page
, NR_SHMEM_THPS
);
643 mapping
->nrpages
+= nr
;
644 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
645 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
647 xas_unlock_irq(&xas
);
648 } while (xas_nomem(&xas
, gfp
));
650 if (xas_error(&xas
)) {
651 page
->mapping
= NULL
;
652 page_ref_sub(page
, nr
);
653 return xas_error(&xas
);
660 * Like delete_from_page_cache, but substitutes swap for page.
662 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
664 struct address_space
*mapping
= page
->mapping
;
667 VM_BUG_ON_PAGE(PageCompound(page
), page
);
669 xa_lock_irq(&mapping
->i_pages
);
670 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
671 page
->mapping
= NULL
;
673 __dec_node_page_state(page
, NR_FILE_PAGES
);
674 __dec_node_page_state(page
, NR_SHMEM
);
675 xa_unlock_irq(&mapping
->i_pages
);
681 * Remove swap entry from page cache, free the swap and its page cache.
683 static int shmem_free_swap(struct address_space
*mapping
,
684 pgoff_t index
, void *radswap
)
688 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
691 free_swap_and_cache(radix_to_swp_entry(radswap
));
696 * Determine (in bytes) how many of the shmem object's pages mapped by the
697 * given offsets are swapped out.
699 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
700 * as long as the inode doesn't go away and racy results are not a problem.
702 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
703 pgoff_t start
, pgoff_t end
)
705 XA_STATE(xas
, &mapping
->i_pages
, start
);
707 unsigned long swapped
= 0;
710 xas_for_each(&xas
, page
, end
- 1) {
711 if (xas_retry(&xas
, page
))
713 if (xa_is_value(page
))
716 if (need_resched()) {
724 return swapped
<< PAGE_SHIFT
;
728 * Determine (in bytes) how many of the shmem object's pages mapped by the
729 * given vma is swapped out.
731 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
732 * as long as the inode doesn't go away and racy results are not a problem.
734 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
736 struct inode
*inode
= file_inode(vma
->vm_file
);
737 struct shmem_inode_info
*info
= SHMEM_I(inode
);
738 struct address_space
*mapping
= inode
->i_mapping
;
739 unsigned long swapped
;
741 /* Be careful as we don't hold info->lock */
742 swapped
= READ_ONCE(info
->swapped
);
745 * The easier cases are when the shmem object has nothing in swap, or
746 * the vma maps it whole. Then we can simply use the stats that we
752 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
753 return swapped
<< PAGE_SHIFT
;
755 /* Here comes the more involved part */
756 return shmem_partial_swap_usage(mapping
,
757 linear_page_index(vma
, vma
->vm_start
),
758 linear_page_index(vma
, vma
->vm_end
));
762 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
764 void shmem_unlock_mapping(struct address_space
*mapping
)
767 pgoff_t indices
[PAGEVEC_SIZE
];
772 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
774 while (!mapping_unevictable(mapping
)) {
776 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
777 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
779 pvec
.nr
= find_get_entries(mapping
, index
,
780 PAGEVEC_SIZE
, pvec
.pages
, indices
);
783 index
= indices
[pvec
.nr
- 1] + 1;
784 pagevec_remove_exceptionals(&pvec
);
785 check_move_unevictable_pages(&pvec
);
786 pagevec_release(&pvec
);
792 * Remove range of pages and swap entries from page cache, and free them.
793 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
795 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
798 struct address_space
*mapping
= inode
->i_mapping
;
799 struct shmem_inode_info
*info
= SHMEM_I(inode
);
800 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
801 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
802 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
803 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
805 pgoff_t indices
[PAGEVEC_SIZE
];
806 long nr_swaps_freed
= 0;
811 end
= -1; /* unsigned, so actually very big */
815 while (index
< end
) {
816 pvec
.nr
= find_get_entries(mapping
, index
,
817 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
818 pvec
.pages
, indices
);
821 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
822 struct page
*page
= pvec
.pages
[i
];
828 if (xa_is_value(page
)) {
831 nr_swaps_freed
+= !shmem_free_swap(mapping
,
836 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
838 if (!trylock_page(page
))
841 if (PageTransTail(page
)) {
842 /* Middle of THP: zero out the page */
843 clear_highpage(page
);
846 } else if (PageTransHuge(page
)) {
847 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
849 * Range ends in the middle of THP:
852 clear_highpage(page
);
856 index
+= HPAGE_PMD_NR
- 1;
857 i
+= HPAGE_PMD_NR
- 1;
860 if (!unfalloc
|| !PageUptodate(page
)) {
861 VM_BUG_ON_PAGE(PageTail(page
), page
);
862 if (page_mapping(page
) == mapping
) {
863 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
864 truncate_inode_page(mapping
, page
);
869 pagevec_remove_exceptionals(&pvec
);
870 pagevec_release(&pvec
);
876 struct page
*page
= NULL
;
877 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
879 unsigned int top
= PAGE_SIZE
;
884 zero_user_segment(page
, partial_start
, top
);
885 set_page_dirty(page
);
891 struct page
*page
= NULL
;
892 shmem_getpage(inode
, end
, &page
, SGP_READ
);
894 zero_user_segment(page
, 0, partial_end
);
895 set_page_dirty(page
);
904 while (index
< end
) {
907 pvec
.nr
= find_get_entries(mapping
, index
,
908 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
909 pvec
.pages
, indices
);
911 /* If all gone or hole-punch or unfalloc, we're done */
912 if (index
== start
|| end
!= -1)
914 /* But if truncating, restart to make sure all gone */
918 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
919 struct page
*page
= pvec
.pages
[i
];
925 if (xa_is_value(page
)) {
928 if (shmem_free_swap(mapping
, index
, page
)) {
929 /* Swap was replaced by page: retry */
939 if (PageTransTail(page
)) {
940 /* Middle of THP: zero out the page */
941 clear_highpage(page
);
944 * Partial thp truncate due 'start' in middle
945 * of THP: don't need to look on these pages
946 * again on !pvec.nr restart.
948 if (index
!= round_down(end
, HPAGE_PMD_NR
))
951 } else if (PageTransHuge(page
)) {
952 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
954 * Range ends in the middle of THP:
957 clear_highpage(page
);
961 index
+= HPAGE_PMD_NR
- 1;
962 i
+= HPAGE_PMD_NR
- 1;
965 if (!unfalloc
|| !PageUptodate(page
)) {
966 VM_BUG_ON_PAGE(PageTail(page
), page
);
967 if (page_mapping(page
) == mapping
) {
968 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
969 truncate_inode_page(mapping
, page
);
971 /* Page was replaced by swap: retry */
979 pagevec_remove_exceptionals(&pvec
);
980 pagevec_release(&pvec
);
984 spin_lock_irq(&info
->lock
);
985 info
->swapped
-= nr_swaps_freed
;
986 shmem_recalc_inode(inode
);
987 spin_unlock_irq(&info
->lock
);
990 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
992 shmem_undo_range(inode
, lstart
, lend
, false);
993 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
995 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
997 static int shmem_getattr(const struct path
*path
, struct kstat
*stat
,
998 u32 request_mask
, unsigned int query_flags
)
1000 struct inode
*inode
= path
->dentry
->d_inode
;
1001 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1002 struct shmem_sb_info
*sb_info
= SHMEM_SB(inode
->i_sb
);
1004 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1005 spin_lock_irq(&info
->lock
);
1006 shmem_recalc_inode(inode
);
1007 spin_unlock_irq(&info
->lock
);
1009 generic_fillattr(inode
, stat
);
1011 if (is_huge_enabled(sb_info
))
1012 stat
->blksize
= HPAGE_PMD_SIZE
;
1017 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1019 struct inode
*inode
= d_inode(dentry
);
1020 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1021 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1024 error
= setattr_prepare(dentry
, attr
);
1028 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1029 loff_t oldsize
= inode
->i_size
;
1030 loff_t newsize
= attr
->ia_size
;
1032 /* protected by i_mutex */
1033 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1034 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1037 if (newsize
!= oldsize
) {
1038 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1042 i_size_write(inode
, newsize
);
1043 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1045 if (newsize
<= oldsize
) {
1046 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1047 if (oldsize
> holebegin
)
1048 unmap_mapping_range(inode
->i_mapping
,
1051 shmem_truncate_range(inode
,
1052 newsize
, (loff_t
)-1);
1053 /* unmap again to remove racily COWed private pages */
1054 if (oldsize
> holebegin
)
1055 unmap_mapping_range(inode
->i_mapping
,
1059 * Part of the huge page can be beyond i_size: subject
1060 * to shrink under memory pressure.
1062 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1063 spin_lock(&sbinfo
->shrinklist_lock
);
1065 * _careful to defend against unlocked access to
1066 * ->shrink_list in shmem_unused_huge_shrink()
1068 if (list_empty_careful(&info
->shrinklist
)) {
1069 list_add_tail(&info
->shrinklist
,
1070 &sbinfo
->shrinklist
);
1071 sbinfo
->shrinklist_len
++;
1073 spin_unlock(&sbinfo
->shrinklist_lock
);
1078 setattr_copy(inode
, attr
);
1079 if (attr
->ia_valid
& ATTR_MODE
)
1080 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1084 static void shmem_evict_inode(struct inode
*inode
)
1086 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1087 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1089 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1090 shmem_unacct_size(info
->flags
, inode
->i_size
);
1092 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1093 if (!list_empty(&info
->shrinklist
)) {
1094 spin_lock(&sbinfo
->shrinklist_lock
);
1095 if (!list_empty(&info
->shrinklist
)) {
1096 list_del_init(&info
->shrinklist
);
1097 sbinfo
->shrinklist_len
--;
1099 spin_unlock(&sbinfo
->shrinklist_lock
);
1101 while (!list_empty(&info
->swaplist
)) {
1102 /* Wait while shmem_unuse() is scanning this inode... */
1103 wait_var_event(&info
->stop_eviction
,
1104 !atomic_read(&info
->stop_eviction
));
1105 mutex_lock(&shmem_swaplist_mutex
);
1106 /* ...but beware of the race if we peeked too early */
1107 if (!atomic_read(&info
->stop_eviction
))
1108 list_del_init(&info
->swaplist
);
1109 mutex_unlock(&shmem_swaplist_mutex
);
1113 simple_xattrs_free(&info
->xattrs
);
1114 WARN_ON(inode
->i_blocks
);
1115 shmem_free_inode(inode
->i_sb
);
1119 extern struct swap_info_struct
*swap_info
[];
1121 static int shmem_find_swap_entries(struct address_space
*mapping
,
1122 pgoff_t start
, unsigned int nr_entries
,
1123 struct page
**entries
, pgoff_t
*indices
,
1124 unsigned int type
, bool frontswap
)
1126 XA_STATE(xas
, &mapping
->i_pages
, start
);
1129 unsigned int ret
= 0;
1135 xas_for_each(&xas
, page
, ULONG_MAX
) {
1136 if (xas_retry(&xas
, page
))
1139 if (!xa_is_value(page
))
1142 entry
= radix_to_swp_entry(page
);
1143 if (swp_type(entry
) != type
)
1146 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1149 indices
[ret
] = xas
.xa_index
;
1150 entries
[ret
] = page
;
1152 if (need_resched()) {
1156 if (++ret
== nr_entries
)
1165 * Move the swapped pages for an inode to page cache. Returns the count
1166 * of pages swapped in, or the error in case of failure.
1168 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1174 struct address_space
*mapping
= inode
->i_mapping
;
1176 for (i
= 0; i
< pvec
.nr
; i
++) {
1177 struct page
*page
= pvec
.pages
[i
];
1179 if (!xa_is_value(page
))
1181 error
= shmem_swapin_page(inode
, indices
[i
],
1183 mapping_gfp_mask(mapping
),
1190 if (error
== -ENOMEM
)
1194 return error
? error
: ret
;
1198 * If swap found in inode, free it and move page from swapcache to filecache.
1200 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1201 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1203 struct address_space
*mapping
= inode
->i_mapping
;
1205 struct pagevec pvec
;
1206 pgoff_t indices
[PAGEVEC_SIZE
];
1207 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1210 pagevec_init(&pvec
);
1212 unsigned int nr_entries
= PAGEVEC_SIZE
;
1214 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1215 nr_entries
= *fs_pages_to_unuse
;
1217 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1218 pvec
.pages
, indices
,
1225 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1229 if (frontswap_partial
) {
1230 *fs_pages_to_unuse
-= ret
;
1231 if (*fs_pages_to_unuse
== 0) {
1232 ret
= FRONTSWAP_PAGES_UNUSED
;
1237 start
= indices
[pvec
.nr
- 1];
1244 * Read all the shared memory data that resides in the swap
1245 * device 'type' back into memory, so the swap device can be
1248 int shmem_unuse(unsigned int type
, bool frontswap
,
1249 unsigned long *fs_pages_to_unuse
)
1251 struct shmem_inode_info
*info
, *next
;
1254 if (list_empty(&shmem_swaplist
))
1257 mutex_lock(&shmem_swaplist_mutex
);
1258 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1259 if (!info
->swapped
) {
1260 list_del_init(&info
->swaplist
);
1264 * Drop the swaplist mutex while searching the inode for swap;
1265 * but before doing so, make sure shmem_evict_inode() will not
1266 * remove placeholder inode from swaplist, nor let it be freed
1267 * (igrab() would protect from unlink, but not from unmount).
1269 atomic_inc(&info
->stop_eviction
);
1270 mutex_unlock(&shmem_swaplist_mutex
);
1272 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1276 mutex_lock(&shmem_swaplist_mutex
);
1277 next
= list_next_entry(info
, swaplist
);
1279 list_del_init(&info
->swaplist
);
1280 if (atomic_dec_and_test(&info
->stop_eviction
))
1281 wake_up_var(&info
->stop_eviction
);
1285 mutex_unlock(&shmem_swaplist_mutex
);
1291 * Move the page from the page cache to the swap cache.
1293 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1295 struct shmem_inode_info
*info
;
1296 struct address_space
*mapping
;
1297 struct inode
*inode
;
1301 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1302 BUG_ON(!PageLocked(page
));
1303 mapping
= page
->mapping
;
1304 index
= page
->index
;
1305 inode
= mapping
->host
;
1306 info
= SHMEM_I(inode
);
1307 if (info
->flags
& VM_LOCKED
)
1309 if (!total_swap_pages
)
1313 * Our capabilities prevent regular writeback or sync from ever calling
1314 * shmem_writepage; but a stacking filesystem might use ->writepage of
1315 * its underlying filesystem, in which case tmpfs should write out to
1316 * swap only in response to memory pressure, and not for the writeback
1319 if (!wbc
->for_reclaim
) {
1320 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1325 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1326 * value into swapfile.c, the only way we can correctly account for a
1327 * fallocated page arriving here is now to initialize it and write it.
1329 * That's okay for a page already fallocated earlier, but if we have
1330 * not yet completed the fallocation, then (a) we want to keep track
1331 * of this page in case we have to undo it, and (b) it may not be a
1332 * good idea to continue anyway, once we're pushing into swap. So
1333 * reactivate the page, and let shmem_fallocate() quit when too many.
1335 if (!PageUptodate(page
)) {
1336 if (inode
->i_private
) {
1337 struct shmem_falloc
*shmem_falloc
;
1338 spin_lock(&inode
->i_lock
);
1339 shmem_falloc
= inode
->i_private
;
1341 !shmem_falloc
->waitq
&&
1342 index
>= shmem_falloc
->start
&&
1343 index
< shmem_falloc
->next
)
1344 shmem_falloc
->nr_unswapped
++;
1346 shmem_falloc
= NULL
;
1347 spin_unlock(&inode
->i_lock
);
1351 clear_highpage(page
);
1352 flush_dcache_page(page
);
1353 SetPageUptodate(page
);
1356 swap
= get_swap_page(page
);
1361 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1362 * if it's not already there. Do it now before the page is
1363 * moved to swap cache, when its pagelock no longer protects
1364 * the inode from eviction. But don't unlock the mutex until
1365 * we've incremented swapped, because shmem_unuse_inode() will
1366 * prune a !swapped inode from the swaplist under this mutex.
1368 mutex_lock(&shmem_swaplist_mutex
);
1369 if (list_empty(&info
->swaplist
))
1370 list_add(&info
->swaplist
, &shmem_swaplist
);
1372 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1373 spin_lock_irq(&info
->lock
);
1374 shmem_recalc_inode(inode
);
1376 spin_unlock_irq(&info
->lock
);
1378 swap_shmem_alloc(swap
);
1379 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1381 mutex_unlock(&shmem_swaplist_mutex
);
1382 BUG_ON(page_mapped(page
));
1383 swap_writepage(page
, wbc
);
1387 mutex_unlock(&shmem_swaplist_mutex
);
1388 put_swap_page(page
, swap
);
1390 set_page_dirty(page
);
1391 if (wbc
->for_reclaim
)
1392 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1397 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1398 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1402 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1403 return; /* show nothing */
1405 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1407 seq_printf(seq
, ",mpol=%s", buffer
);
1410 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1412 struct mempolicy
*mpol
= NULL
;
1414 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1415 mpol
= sbinfo
->mpol
;
1417 spin_unlock(&sbinfo
->stat_lock
);
1421 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1422 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1425 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1429 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1431 #define vm_policy vm_private_data
1434 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1435 struct shmem_inode_info
*info
, pgoff_t index
)
1437 /* Create a pseudo vma that just contains the policy */
1438 vma_init(vma
, NULL
);
1439 /* Bias interleave by inode number to distribute better across nodes */
1440 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1441 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1444 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1446 /* Drop reference taken by mpol_shared_policy_lookup() */
1447 mpol_cond_put(vma
->vm_policy
);
1450 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1451 struct shmem_inode_info
*info
, pgoff_t index
)
1453 struct vm_area_struct pvma
;
1455 struct vm_fault vmf
;
1457 shmem_pseudo_vma_init(&pvma
, info
, index
);
1460 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1461 shmem_pseudo_vma_destroy(&pvma
);
1466 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1467 struct shmem_inode_info
*info
, pgoff_t index
)
1469 struct vm_area_struct pvma
;
1470 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1474 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1477 hindex
= round_down(index
, HPAGE_PMD_NR
);
1478 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1482 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1483 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1484 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1485 shmem_pseudo_vma_destroy(&pvma
);
1487 prep_transhuge_page(page
);
1491 static struct page
*shmem_alloc_page(gfp_t gfp
,
1492 struct shmem_inode_info
*info
, pgoff_t index
)
1494 struct vm_area_struct pvma
;
1497 shmem_pseudo_vma_init(&pvma
, info
, index
);
1498 page
= alloc_page_vma(gfp
, &pvma
, 0);
1499 shmem_pseudo_vma_destroy(&pvma
);
1504 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1505 struct inode
*inode
,
1506 pgoff_t index
, bool huge
)
1508 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1513 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1515 nr
= huge
? HPAGE_PMD_NR
: 1;
1517 if (!shmem_inode_acct_block(inode
, nr
))
1521 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1523 page
= shmem_alloc_page(gfp
, info
, index
);
1525 __SetPageLocked(page
);
1526 __SetPageSwapBacked(page
);
1531 shmem_inode_unacct_blocks(inode
, nr
);
1533 return ERR_PTR(err
);
1537 * When a page is moved from swapcache to shmem filecache (either by the
1538 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1539 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1540 * ignorance of the mapping it belongs to. If that mapping has special
1541 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1542 * we may need to copy to a suitable page before moving to filecache.
1544 * In a future release, this may well be extended to respect cpuset and
1545 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1546 * but for now it is a simple matter of zone.
1548 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1550 return page_zonenum(page
) > gfp_zone(gfp
);
1553 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1554 struct shmem_inode_info
*info
, pgoff_t index
)
1556 struct page
*oldpage
, *newpage
;
1557 struct address_space
*swap_mapping
;
1563 entry
.val
= page_private(oldpage
);
1564 swap_index
= swp_offset(entry
);
1565 swap_mapping
= page_mapping(oldpage
);
1568 * We have arrived here because our zones are constrained, so don't
1569 * limit chance of success by further cpuset and node constraints.
1571 gfp
&= ~GFP_CONSTRAINT_MASK
;
1572 newpage
= shmem_alloc_page(gfp
, info
, index
);
1577 copy_highpage(newpage
, oldpage
);
1578 flush_dcache_page(newpage
);
1580 __SetPageLocked(newpage
);
1581 __SetPageSwapBacked(newpage
);
1582 SetPageUptodate(newpage
);
1583 set_page_private(newpage
, entry
.val
);
1584 SetPageSwapCache(newpage
);
1587 * Our caller will very soon move newpage out of swapcache, but it's
1588 * a nice clean interface for us to replace oldpage by newpage there.
1590 xa_lock_irq(&swap_mapping
->i_pages
);
1591 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1593 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1594 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1596 xa_unlock_irq(&swap_mapping
->i_pages
);
1598 if (unlikely(error
)) {
1600 * Is this possible? I think not, now that our callers check
1601 * both PageSwapCache and page_private after getting page lock;
1602 * but be defensive. Reverse old to newpage for clear and free.
1606 mem_cgroup_migrate(oldpage
, newpage
);
1607 lru_cache_add_anon(newpage
);
1611 ClearPageSwapCache(oldpage
);
1612 set_page_private(oldpage
, 0);
1614 unlock_page(oldpage
);
1621 * Swap in the page pointed to by *pagep.
1622 * Caller has to make sure that *pagep contains a valid swapped page.
1623 * Returns 0 and the page in pagep if success. On failure, returns the
1624 * the error code and NULL in *pagep.
1626 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1627 struct page
**pagep
, enum sgp_type sgp
,
1628 gfp_t gfp
, struct vm_area_struct
*vma
,
1629 vm_fault_t
*fault_type
)
1631 struct address_space
*mapping
= inode
->i_mapping
;
1632 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1633 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1634 struct mem_cgroup
*memcg
;
1639 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1640 swap
= radix_to_swp_entry(*pagep
);
1643 /* Look it up and read it in.. */
1644 page
= lookup_swap_cache(swap
, NULL
, 0);
1646 /* Or update major stats only when swapin succeeds?? */
1648 *fault_type
|= VM_FAULT_MAJOR
;
1649 count_vm_event(PGMAJFAULT
);
1650 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1652 /* Here we actually start the io */
1653 page
= shmem_swapin(swap
, gfp
, info
, index
);
1660 /* We have to do this with page locked to prevent races */
1662 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1663 !shmem_confirm_swap(mapping
, index
, swap
)) {
1667 if (!PageUptodate(page
)) {
1671 wait_on_page_writeback(page
);
1673 if (shmem_should_replace_page(page
, gfp
)) {
1674 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1679 error
= mem_cgroup_try_charge_delay(page
, charge_mm
, gfp
, &memcg
,
1682 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1683 swp_to_radix_entry(swap
), gfp
);
1685 * We already confirmed swap under page lock, and make
1686 * no memory allocation here, so usually no possibility
1687 * of error; but free_swap_and_cache() only trylocks a
1688 * page, so it is just possible that the entry has been
1689 * truncated or holepunched since swap was confirmed.
1690 * shmem_undo_range() will have done some of the
1691 * unaccounting, now delete_from_swap_cache() will do
1695 mem_cgroup_cancel_charge(page
, memcg
, false);
1696 delete_from_swap_cache(page
);
1702 mem_cgroup_commit_charge(page
, memcg
, true, false);
1704 spin_lock_irq(&info
->lock
);
1706 shmem_recalc_inode(inode
);
1707 spin_unlock_irq(&info
->lock
);
1709 if (sgp
== SGP_WRITE
)
1710 mark_page_accessed(page
);
1712 delete_from_swap_cache(page
);
1713 set_page_dirty(page
);
1719 if (!shmem_confirm_swap(mapping
, index
, swap
))
1731 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1733 * If we allocate a new one we do not mark it dirty. That's up to the
1734 * vm. If we swap it in we mark it dirty since we also free the swap
1735 * entry since a page cannot live in both the swap and page cache.
1737 * vmf and fault_type are only supplied by shmem_fault:
1738 * otherwise they are NULL.
1740 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1741 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1742 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1743 vm_fault_t
*fault_type
)
1745 struct address_space
*mapping
= inode
->i_mapping
;
1746 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1747 struct shmem_sb_info
*sbinfo
;
1748 struct mm_struct
*charge_mm
;
1749 struct mem_cgroup
*memcg
;
1751 enum sgp_type sgp_huge
= sgp
;
1752 pgoff_t hindex
= index
;
1757 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1759 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1762 if (sgp
<= SGP_CACHE
&&
1763 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1767 sbinfo
= SHMEM_SB(inode
->i_sb
);
1768 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1770 page
= find_lock_entry(mapping
, index
);
1771 if (xa_is_value(page
)) {
1772 error
= shmem_swapin_page(inode
, index
, &page
,
1773 sgp
, gfp
, vma
, fault_type
);
1774 if (error
== -EEXIST
)
1781 if (page
&& sgp
== SGP_WRITE
)
1782 mark_page_accessed(page
);
1784 /* fallocated page? */
1785 if (page
&& !PageUptodate(page
)) {
1786 if (sgp
!= SGP_READ
)
1792 if (page
|| sgp
== SGP_READ
) {
1798 * Fast cache lookup did not find it:
1799 * bring it back from swap or allocate.
1802 if (vma
&& userfaultfd_missing(vma
)) {
1803 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1807 /* shmem_symlink() */
1808 if (mapping
->a_ops
!= &shmem_aops
)
1810 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1812 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1814 switch (sbinfo
->huge
) {
1817 case SHMEM_HUGE_NEVER
:
1819 case SHMEM_HUGE_WITHIN_SIZE
:
1820 off
= round_up(index
, HPAGE_PMD_NR
);
1821 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1822 if (i_size
>= HPAGE_PMD_SIZE
&&
1823 i_size
>> PAGE_SHIFT
>= off
)
1826 case SHMEM_HUGE_ADVISE
:
1827 if (sgp_huge
== SGP_HUGE
)
1829 /* TODO: implement fadvise() hints */
1834 page
= shmem_alloc_and_acct_page(gfp
, inode
, index
, true);
1837 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1843 error
= PTR_ERR(page
);
1845 if (error
!= -ENOSPC
)
1848 * Try to reclaim some space by splitting a huge page
1849 * beyond i_size on the filesystem.
1854 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1855 if (ret
== SHRINK_STOP
)
1863 if (PageTransHuge(page
))
1864 hindex
= round_down(index
, HPAGE_PMD_NR
);
1868 if (sgp
== SGP_WRITE
)
1869 __SetPageReferenced(page
);
1871 error
= mem_cgroup_try_charge_delay(page
, charge_mm
, gfp
, &memcg
,
1872 PageTransHuge(page
));
1875 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1876 NULL
, gfp
& GFP_RECLAIM_MASK
);
1878 mem_cgroup_cancel_charge(page
, memcg
,
1879 PageTransHuge(page
));
1882 mem_cgroup_commit_charge(page
, memcg
, false,
1883 PageTransHuge(page
));
1884 lru_cache_add_anon(page
);
1886 spin_lock_irq(&info
->lock
);
1887 info
->alloced
+= compound_nr(page
);
1888 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1889 shmem_recalc_inode(inode
);
1890 spin_unlock_irq(&info
->lock
);
1893 if (PageTransHuge(page
) &&
1894 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1895 hindex
+ HPAGE_PMD_NR
- 1) {
1897 * Part of the huge page is beyond i_size: subject
1898 * to shrink under memory pressure.
1900 spin_lock(&sbinfo
->shrinklist_lock
);
1902 * _careful to defend against unlocked access to
1903 * ->shrink_list in shmem_unused_huge_shrink()
1905 if (list_empty_careful(&info
->shrinklist
)) {
1906 list_add_tail(&info
->shrinklist
,
1907 &sbinfo
->shrinklist
);
1908 sbinfo
->shrinklist_len
++;
1910 spin_unlock(&sbinfo
->shrinklist_lock
);
1914 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1916 if (sgp
== SGP_FALLOC
)
1920 * Let SGP_WRITE caller clear ends if write does not fill page;
1921 * but SGP_FALLOC on a page fallocated earlier must initialize
1922 * it now, lest undo on failure cancel our earlier guarantee.
1924 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1925 struct page
*head
= compound_head(page
);
1928 for (i
= 0; i
< compound_nr(head
); i
++) {
1929 clear_highpage(head
+ i
);
1930 flush_dcache_page(head
+ i
);
1932 SetPageUptodate(head
);
1935 /* Perhaps the file has been truncated since we checked */
1936 if (sgp
<= SGP_CACHE
&&
1937 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1939 ClearPageDirty(page
);
1940 delete_from_page_cache(page
);
1941 spin_lock_irq(&info
->lock
);
1942 shmem_recalc_inode(inode
);
1943 spin_unlock_irq(&info
->lock
);
1948 *pagep
= page
+ index
- hindex
;
1955 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
1957 if (PageTransHuge(page
)) {
1967 if (error
== -ENOSPC
&& !once
++) {
1968 spin_lock_irq(&info
->lock
);
1969 shmem_recalc_inode(inode
);
1970 spin_unlock_irq(&info
->lock
);
1973 if (error
== -EEXIST
)
1979 * This is like autoremove_wake_function, but it removes the wait queue
1980 * entry unconditionally - even if something else had already woken the
1983 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
1985 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1986 list_del_init(&wait
->entry
);
1990 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
1992 struct vm_area_struct
*vma
= vmf
->vma
;
1993 struct inode
*inode
= file_inode(vma
->vm_file
);
1994 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1997 vm_fault_t ret
= VM_FAULT_LOCKED
;
2000 * Trinity finds that probing a hole which tmpfs is punching can
2001 * prevent the hole-punch from ever completing: which in turn
2002 * locks writers out with its hold on i_mutex. So refrain from
2003 * faulting pages into the hole while it's being punched. Although
2004 * shmem_undo_range() does remove the additions, it may be unable to
2005 * keep up, as each new page needs its own unmap_mapping_range() call,
2006 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2008 * It does not matter if we sometimes reach this check just before the
2009 * hole-punch begins, so that one fault then races with the punch:
2010 * we just need to make racing faults a rare case.
2012 * The implementation below would be much simpler if we just used a
2013 * standard mutex or completion: but we cannot take i_mutex in fault,
2014 * and bloating every shmem inode for this unlikely case would be sad.
2016 if (unlikely(inode
->i_private
)) {
2017 struct shmem_falloc
*shmem_falloc
;
2019 spin_lock(&inode
->i_lock
);
2020 shmem_falloc
= inode
->i_private
;
2022 shmem_falloc
->waitq
&&
2023 vmf
->pgoff
>= shmem_falloc
->start
&&
2024 vmf
->pgoff
< shmem_falloc
->next
) {
2025 wait_queue_head_t
*shmem_falloc_waitq
;
2026 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2028 ret
= VM_FAULT_NOPAGE
;
2029 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
2030 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
2031 /* It's polite to up mmap_sem if we can */
2032 up_read(&vma
->vm_mm
->mmap_sem
);
2033 ret
= VM_FAULT_RETRY
;
2036 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2037 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2038 TASK_UNINTERRUPTIBLE
);
2039 spin_unlock(&inode
->i_lock
);
2043 * shmem_falloc_waitq points into the shmem_fallocate()
2044 * stack of the hole-punching task: shmem_falloc_waitq
2045 * is usually invalid by the time we reach here, but
2046 * finish_wait() does not dereference it in that case;
2047 * though i_lock needed lest racing with wake_up_all().
2049 spin_lock(&inode
->i_lock
);
2050 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2051 spin_unlock(&inode
->i_lock
);
2054 spin_unlock(&inode
->i_lock
);
2059 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2060 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2062 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2065 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2066 gfp
, vma
, vmf
, &ret
);
2068 return vmf_error(err
);
2072 unsigned long shmem_get_unmapped_area(struct file
*file
,
2073 unsigned long uaddr
, unsigned long len
,
2074 unsigned long pgoff
, unsigned long flags
)
2076 unsigned long (*get_area
)(struct file
*,
2077 unsigned long, unsigned long, unsigned long, unsigned long);
2079 unsigned long offset
;
2080 unsigned long inflated_len
;
2081 unsigned long inflated_addr
;
2082 unsigned long inflated_offset
;
2084 if (len
> TASK_SIZE
)
2087 get_area
= current
->mm
->get_unmapped_area
;
2088 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2090 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
2092 if (IS_ERR_VALUE(addr
))
2094 if (addr
& ~PAGE_MASK
)
2096 if (addr
> TASK_SIZE
- len
)
2099 if (shmem_huge
== SHMEM_HUGE_DENY
)
2101 if (len
< HPAGE_PMD_SIZE
)
2103 if (flags
& MAP_FIXED
)
2106 * Our priority is to support MAP_SHARED mapped hugely;
2107 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2108 * But if caller specified an address hint, respect that as before.
2113 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2114 struct super_block
*sb
;
2117 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2118 sb
= file_inode(file
)->i_sb
;
2121 * Called directly from mm/mmap.c, or drivers/char/mem.c
2122 * for "/dev/zero", to create a shared anonymous object.
2124 if (IS_ERR(shm_mnt
))
2126 sb
= shm_mnt
->mnt_sb
;
2128 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2132 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2133 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2135 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2138 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2139 if (inflated_len
> TASK_SIZE
)
2141 if (inflated_len
< len
)
2144 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
2145 if (IS_ERR_VALUE(inflated_addr
))
2147 if (inflated_addr
& ~PAGE_MASK
)
2150 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2151 inflated_addr
+= offset
- inflated_offset
;
2152 if (inflated_offset
> offset
)
2153 inflated_addr
+= HPAGE_PMD_SIZE
;
2155 if (inflated_addr
> TASK_SIZE
- len
)
2157 return inflated_addr
;
2161 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2163 struct inode
*inode
= file_inode(vma
->vm_file
);
2164 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2167 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2170 struct inode
*inode
= file_inode(vma
->vm_file
);
2173 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2174 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2178 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2180 struct inode
*inode
= file_inode(file
);
2181 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2182 int retval
= -ENOMEM
;
2184 spin_lock_irq(&info
->lock
);
2185 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2186 if (!user_shm_lock(inode
->i_size
, user
))
2188 info
->flags
|= VM_LOCKED
;
2189 mapping_set_unevictable(file
->f_mapping
);
2191 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2192 user_shm_unlock(inode
->i_size
, user
);
2193 info
->flags
&= ~VM_LOCKED
;
2194 mapping_clear_unevictable(file
->f_mapping
);
2199 spin_unlock_irq(&info
->lock
);
2203 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2205 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2207 if (info
->seals
& F_SEAL_FUTURE_WRITE
) {
2209 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2210 * "future write" seal active.
2212 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_WRITE
))
2216 * Since the F_SEAL_FUTURE_WRITE seals allow for a MAP_SHARED
2217 * read-only mapping, take care to not allow mprotect to revert
2220 vma
->vm_flags
&= ~(VM_MAYWRITE
);
2223 file_accessed(file
);
2224 vma
->vm_ops
= &shmem_vm_ops
;
2225 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2226 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2227 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2228 khugepaged_enter(vma
, vma
->vm_flags
);
2233 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2234 umode_t mode
, dev_t dev
, unsigned long flags
)
2236 struct inode
*inode
;
2237 struct shmem_inode_info
*info
;
2238 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2240 if (shmem_reserve_inode(sb
))
2243 inode
= new_inode(sb
);
2245 inode
->i_ino
= get_next_ino();
2246 inode_init_owner(inode
, dir
, mode
);
2247 inode
->i_blocks
= 0;
2248 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2249 inode
->i_generation
= prandom_u32();
2250 info
= SHMEM_I(inode
);
2251 memset(info
, 0, (char *)inode
- (char *)info
);
2252 spin_lock_init(&info
->lock
);
2253 atomic_set(&info
->stop_eviction
, 0);
2254 info
->seals
= F_SEAL_SEAL
;
2255 info
->flags
= flags
& VM_NORESERVE
;
2256 INIT_LIST_HEAD(&info
->shrinklist
);
2257 INIT_LIST_HEAD(&info
->swaplist
);
2258 simple_xattrs_init(&info
->xattrs
);
2259 cache_no_acl(inode
);
2261 switch (mode
& S_IFMT
) {
2263 inode
->i_op
= &shmem_special_inode_operations
;
2264 init_special_inode(inode
, mode
, dev
);
2267 inode
->i_mapping
->a_ops
= &shmem_aops
;
2268 inode
->i_op
= &shmem_inode_operations
;
2269 inode
->i_fop
= &shmem_file_operations
;
2270 mpol_shared_policy_init(&info
->policy
,
2271 shmem_get_sbmpol(sbinfo
));
2275 /* Some things misbehave if size == 0 on a directory */
2276 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2277 inode
->i_op
= &shmem_dir_inode_operations
;
2278 inode
->i_fop
= &simple_dir_operations
;
2282 * Must not load anything in the rbtree,
2283 * mpol_free_shared_policy will not be called.
2285 mpol_shared_policy_init(&info
->policy
, NULL
);
2289 lockdep_annotate_inode_mutex_key(inode
);
2291 shmem_free_inode(sb
);
2295 bool shmem_mapping(struct address_space
*mapping
)
2297 return mapping
->a_ops
== &shmem_aops
;
2300 static int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2302 struct vm_area_struct
*dst_vma
,
2303 unsigned long dst_addr
,
2304 unsigned long src_addr
,
2306 struct page
**pagep
)
2308 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2309 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2310 struct address_space
*mapping
= inode
->i_mapping
;
2311 gfp_t gfp
= mapping_gfp_mask(mapping
);
2312 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2313 struct mem_cgroup
*memcg
;
2317 pte_t _dst_pte
, *dst_pte
;
2319 pgoff_t offset
, max_off
;
2322 if (!shmem_inode_acct_block(inode
, 1))
2326 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2328 goto out_unacct_blocks
;
2330 if (!zeropage
) { /* mcopy_atomic */
2331 page_kaddr
= kmap_atomic(page
);
2332 ret
= copy_from_user(page_kaddr
,
2333 (const void __user
*)src_addr
,
2335 kunmap_atomic(page_kaddr
);
2337 /* fallback to copy_from_user outside mmap_sem */
2338 if (unlikely(ret
)) {
2340 shmem_inode_unacct_blocks(inode
, 1);
2341 /* don't free the page */
2344 } else { /* mfill_zeropage_atomic */
2345 clear_highpage(page
);
2352 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2353 __SetPageLocked(page
);
2354 __SetPageSwapBacked(page
);
2355 __SetPageUptodate(page
);
2358 offset
= linear_page_index(dst_vma
, dst_addr
);
2359 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2360 if (unlikely(offset
>= max_off
))
2363 ret
= mem_cgroup_try_charge_delay(page
, dst_mm
, gfp
, &memcg
, false);
2367 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2368 gfp
& GFP_RECLAIM_MASK
);
2370 goto out_release_uncharge
;
2372 mem_cgroup_commit_charge(page
, memcg
, false, false);
2374 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2375 if (dst_vma
->vm_flags
& VM_WRITE
)
2376 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2379 * We don't set the pte dirty if the vma has no
2380 * VM_WRITE permission, so mark the page dirty or it
2381 * could be freed from under us. We could do it
2382 * unconditionally before unlock_page(), but doing it
2383 * only if VM_WRITE is not set is faster.
2385 set_page_dirty(page
);
2388 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2391 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2392 if (unlikely(offset
>= max_off
))
2393 goto out_release_uncharge_unlock
;
2396 if (!pte_none(*dst_pte
))
2397 goto out_release_uncharge_unlock
;
2399 lru_cache_add_anon(page
);
2401 spin_lock(&info
->lock
);
2403 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2404 shmem_recalc_inode(inode
);
2405 spin_unlock(&info
->lock
);
2407 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2408 page_add_file_rmap(page
, false);
2409 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2411 /* No need to invalidate - it was non-present before */
2412 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2413 pte_unmap_unlock(dst_pte
, ptl
);
2418 out_release_uncharge_unlock
:
2419 pte_unmap_unlock(dst_pte
, ptl
);
2420 ClearPageDirty(page
);
2421 delete_from_page_cache(page
);
2422 out_release_uncharge
:
2423 mem_cgroup_cancel_charge(page
, memcg
, false);
2428 shmem_inode_unacct_blocks(inode
, 1);
2432 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2434 struct vm_area_struct
*dst_vma
,
2435 unsigned long dst_addr
,
2436 unsigned long src_addr
,
2437 struct page
**pagep
)
2439 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2440 dst_addr
, src_addr
, false, pagep
);
2443 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2445 struct vm_area_struct
*dst_vma
,
2446 unsigned long dst_addr
)
2448 struct page
*page
= NULL
;
2450 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2451 dst_addr
, 0, true, &page
);
2455 static const struct inode_operations shmem_symlink_inode_operations
;
2456 static const struct inode_operations shmem_short_symlink_operations
;
2458 #ifdef CONFIG_TMPFS_XATTR
2459 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2461 #define shmem_initxattrs NULL
2465 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2466 loff_t pos
, unsigned len
, unsigned flags
,
2467 struct page
**pagep
, void **fsdata
)
2469 struct inode
*inode
= mapping
->host
;
2470 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2471 pgoff_t index
= pos
>> PAGE_SHIFT
;
2473 /* i_mutex is held by caller */
2474 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2475 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2476 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2478 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2482 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2486 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2487 loff_t pos
, unsigned len
, unsigned copied
,
2488 struct page
*page
, void *fsdata
)
2490 struct inode
*inode
= mapping
->host
;
2492 if (pos
+ copied
> inode
->i_size
)
2493 i_size_write(inode
, pos
+ copied
);
2495 if (!PageUptodate(page
)) {
2496 struct page
*head
= compound_head(page
);
2497 if (PageTransCompound(page
)) {
2500 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2501 if (head
+ i
== page
)
2503 clear_highpage(head
+ i
);
2504 flush_dcache_page(head
+ i
);
2507 if (copied
< PAGE_SIZE
) {
2508 unsigned from
= pos
& (PAGE_SIZE
- 1);
2509 zero_user_segments(page
, 0, from
,
2510 from
+ copied
, PAGE_SIZE
);
2512 SetPageUptodate(head
);
2514 set_page_dirty(page
);
2521 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2523 struct file
*file
= iocb
->ki_filp
;
2524 struct inode
*inode
= file_inode(file
);
2525 struct address_space
*mapping
= inode
->i_mapping
;
2527 unsigned long offset
;
2528 enum sgp_type sgp
= SGP_READ
;
2531 loff_t
*ppos
= &iocb
->ki_pos
;
2534 * Might this read be for a stacking filesystem? Then when reading
2535 * holes of a sparse file, we actually need to allocate those pages,
2536 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2538 if (!iter_is_iovec(to
))
2541 index
= *ppos
>> PAGE_SHIFT
;
2542 offset
= *ppos
& ~PAGE_MASK
;
2545 struct page
*page
= NULL
;
2547 unsigned long nr
, ret
;
2548 loff_t i_size
= i_size_read(inode
);
2550 end_index
= i_size
>> PAGE_SHIFT
;
2551 if (index
> end_index
)
2553 if (index
== end_index
) {
2554 nr
= i_size
& ~PAGE_MASK
;
2559 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2561 if (error
== -EINVAL
)
2566 if (sgp
== SGP_CACHE
)
2567 set_page_dirty(page
);
2572 * We must evaluate after, since reads (unlike writes)
2573 * are called without i_mutex protection against truncate
2576 i_size
= i_size_read(inode
);
2577 end_index
= i_size
>> PAGE_SHIFT
;
2578 if (index
== end_index
) {
2579 nr
= i_size
& ~PAGE_MASK
;
2590 * If users can be writing to this page using arbitrary
2591 * virtual addresses, take care about potential aliasing
2592 * before reading the page on the kernel side.
2594 if (mapping_writably_mapped(mapping
))
2595 flush_dcache_page(page
);
2597 * Mark the page accessed if we read the beginning.
2600 mark_page_accessed(page
);
2602 page
= ZERO_PAGE(0);
2607 * Ok, we have the page, and it's up-to-date, so
2608 * now we can copy it to user space...
2610 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2613 index
+= offset
>> PAGE_SHIFT
;
2614 offset
&= ~PAGE_MASK
;
2617 if (!iov_iter_count(to
))
2626 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2627 file_accessed(file
);
2628 return retval
? retval
: error
;
2632 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2634 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2635 pgoff_t index
, pgoff_t end
, int whence
)
2638 struct pagevec pvec
;
2639 pgoff_t indices
[PAGEVEC_SIZE
];
2643 pagevec_init(&pvec
);
2644 pvec
.nr
= 1; /* start small: we may be there already */
2646 pvec
.nr
= find_get_entries(mapping
, index
,
2647 pvec
.nr
, pvec
.pages
, indices
);
2649 if (whence
== SEEK_DATA
)
2653 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2654 if (index
< indices
[i
]) {
2655 if (whence
== SEEK_HOLE
) {
2661 page
= pvec
.pages
[i
];
2662 if (page
&& !xa_is_value(page
)) {
2663 if (!PageUptodate(page
))
2667 (page
&& whence
== SEEK_DATA
) ||
2668 (!page
&& whence
== SEEK_HOLE
)) {
2673 pagevec_remove_exceptionals(&pvec
);
2674 pagevec_release(&pvec
);
2675 pvec
.nr
= PAGEVEC_SIZE
;
2681 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2683 struct address_space
*mapping
= file
->f_mapping
;
2684 struct inode
*inode
= mapping
->host
;
2688 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2689 return generic_file_llseek_size(file
, offset
, whence
,
2690 MAX_LFS_FILESIZE
, i_size_read(inode
));
2692 /* We're holding i_mutex so we can access i_size directly */
2694 if (offset
< 0 || offset
>= inode
->i_size
)
2697 start
= offset
>> PAGE_SHIFT
;
2698 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2699 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2700 new_offset
<<= PAGE_SHIFT
;
2701 if (new_offset
> offset
) {
2702 if (new_offset
< inode
->i_size
)
2703 offset
= new_offset
;
2704 else if (whence
== SEEK_DATA
)
2707 offset
= inode
->i_size
;
2712 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2713 inode_unlock(inode
);
2717 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2720 struct inode
*inode
= file_inode(file
);
2721 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2722 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2723 struct shmem_falloc shmem_falloc
;
2724 pgoff_t start
, index
, end
;
2727 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2732 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2733 struct address_space
*mapping
= file
->f_mapping
;
2734 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2735 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2736 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2738 /* protected by i_mutex */
2739 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2744 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2745 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2746 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2747 spin_lock(&inode
->i_lock
);
2748 inode
->i_private
= &shmem_falloc
;
2749 spin_unlock(&inode
->i_lock
);
2751 if ((u64
)unmap_end
> (u64
)unmap_start
)
2752 unmap_mapping_range(mapping
, unmap_start
,
2753 1 + unmap_end
- unmap_start
, 0);
2754 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2755 /* No need to unmap again: hole-punching leaves COWed pages */
2757 spin_lock(&inode
->i_lock
);
2758 inode
->i_private
= NULL
;
2759 wake_up_all(&shmem_falloc_waitq
);
2760 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2761 spin_unlock(&inode
->i_lock
);
2766 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2767 error
= inode_newsize_ok(inode
, offset
+ len
);
2771 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2776 start
= offset
>> PAGE_SHIFT
;
2777 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2778 /* Try to avoid a swapstorm if len is impossible to satisfy */
2779 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2784 shmem_falloc
.waitq
= NULL
;
2785 shmem_falloc
.start
= start
;
2786 shmem_falloc
.next
= start
;
2787 shmem_falloc
.nr_falloced
= 0;
2788 shmem_falloc
.nr_unswapped
= 0;
2789 spin_lock(&inode
->i_lock
);
2790 inode
->i_private
= &shmem_falloc
;
2791 spin_unlock(&inode
->i_lock
);
2793 for (index
= start
; index
< end
; index
++) {
2797 * Good, the fallocate(2) manpage permits EINTR: we may have
2798 * been interrupted because we are using up too much memory.
2800 if (signal_pending(current
))
2802 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2805 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2807 /* Remove the !PageUptodate pages we added */
2808 if (index
> start
) {
2809 shmem_undo_range(inode
,
2810 (loff_t
)start
<< PAGE_SHIFT
,
2811 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2817 * Inform shmem_writepage() how far we have reached.
2818 * No need for lock or barrier: we have the page lock.
2820 shmem_falloc
.next
++;
2821 if (!PageUptodate(page
))
2822 shmem_falloc
.nr_falloced
++;
2825 * If !PageUptodate, leave it that way so that freeable pages
2826 * can be recognized if we need to rollback on error later.
2827 * But set_page_dirty so that memory pressure will swap rather
2828 * than free the pages we are allocating (and SGP_CACHE pages
2829 * might still be clean: we now need to mark those dirty too).
2831 set_page_dirty(page
);
2837 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2838 i_size_write(inode
, offset
+ len
);
2839 inode
->i_ctime
= current_time(inode
);
2841 spin_lock(&inode
->i_lock
);
2842 inode
->i_private
= NULL
;
2843 spin_unlock(&inode
->i_lock
);
2845 inode_unlock(inode
);
2849 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2851 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2853 buf
->f_type
= TMPFS_MAGIC
;
2854 buf
->f_bsize
= PAGE_SIZE
;
2855 buf
->f_namelen
= NAME_MAX
;
2856 if (sbinfo
->max_blocks
) {
2857 buf
->f_blocks
= sbinfo
->max_blocks
;
2859 buf
->f_bfree
= sbinfo
->max_blocks
-
2860 percpu_counter_sum(&sbinfo
->used_blocks
);
2862 if (sbinfo
->max_inodes
) {
2863 buf
->f_files
= sbinfo
->max_inodes
;
2864 buf
->f_ffree
= sbinfo
->free_inodes
;
2866 /* else leave those fields 0 like simple_statfs */
2871 * File creation. Allocate an inode, and we're done..
2874 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2876 struct inode
*inode
;
2877 int error
= -ENOSPC
;
2879 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2881 error
= simple_acl_create(dir
, inode
);
2884 error
= security_inode_init_security(inode
, dir
,
2886 shmem_initxattrs
, NULL
);
2887 if (error
&& error
!= -EOPNOTSUPP
)
2891 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2892 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2893 d_instantiate(dentry
, inode
);
2894 dget(dentry
); /* Extra count - pin the dentry in core */
2903 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2905 struct inode
*inode
;
2906 int error
= -ENOSPC
;
2908 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2910 error
= security_inode_init_security(inode
, dir
,
2912 shmem_initxattrs
, NULL
);
2913 if (error
&& error
!= -EOPNOTSUPP
)
2915 error
= simple_acl_create(dir
, inode
);
2918 d_tmpfile(dentry
, inode
);
2926 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2930 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2936 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2939 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2945 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2947 struct inode
*inode
= d_inode(old_dentry
);
2951 * No ordinary (disk based) filesystem counts links as inodes;
2952 * but each new link needs a new dentry, pinning lowmem, and
2953 * tmpfs dentries cannot be pruned until they are unlinked.
2954 * But if an O_TMPFILE file is linked into the tmpfs, the
2955 * first link must skip that, to get the accounting right.
2957 if (inode
->i_nlink
) {
2958 ret
= shmem_reserve_inode(inode
->i_sb
);
2963 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2964 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2966 ihold(inode
); /* New dentry reference */
2967 dget(dentry
); /* Extra pinning count for the created dentry */
2968 d_instantiate(dentry
, inode
);
2973 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2975 struct inode
*inode
= d_inode(dentry
);
2977 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2978 shmem_free_inode(inode
->i_sb
);
2980 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2981 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2983 dput(dentry
); /* Undo the count from "create" - this does all the work */
2987 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2989 if (!simple_empty(dentry
))
2992 drop_nlink(d_inode(dentry
));
2994 return shmem_unlink(dir
, dentry
);
2997 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2999 bool old_is_dir
= d_is_dir(old_dentry
);
3000 bool new_is_dir
= d_is_dir(new_dentry
);
3002 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3004 drop_nlink(old_dir
);
3007 drop_nlink(new_dir
);
3011 old_dir
->i_ctime
= old_dir
->i_mtime
=
3012 new_dir
->i_ctime
= new_dir
->i_mtime
=
3013 d_inode(old_dentry
)->i_ctime
=
3014 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3019 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3021 struct dentry
*whiteout
;
3024 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3028 error
= shmem_mknod(old_dir
, whiteout
,
3029 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3035 * Cheat and hash the whiteout while the old dentry is still in
3036 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3038 * d_lookup() will consistently find one of them at this point,
3039 * not sure which one, but that isn't even important.
3046 * The VFS layer already does all the dentry stuff for rename,
3047 * we just have to decrement the usage count for the target if
3048 * it exists so that the VFS layer correctly free's it when it
3051 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3053 struct inode
*inode
= d_inode(old_dentry
);
3054 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3056 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3059 if (flags
& RENAME_EXCHANGE
)
3060 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3062 if (!simple_empty(new_dentry
))
3065 if (flags
& RENAME_WHITEOUT
) {
3068 error
= shmem_whiteout(old_dir
, old_dentry
);
3073 if (d_really_is_positive(new_dentry
)) {
3074 (void) shmem_unlink(new_dir
, new_dentry
);
3075 if (they_are_dirs
) {
3076 drop_nlink(d_inode(new_dentry
));
3077 drop_nlink(old_dir
);
3079 } else if (they_are_dirs
) {
3080 drop_nlink(old_dir
);
3084 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3085 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3086 old_dir
->i_ctime
= old_dir
->i_mtime
=
3087 new_dir
->i_ctime
= new_dir
->i_mtime
=
3088 inode
->i_ctime
= current_time(old_dir
);
3092 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3096 struct inode
*inode
;
3099 len
= strlen(symname
) + 1;
3100 if (len
> PAGE_SIZE
)
3101 return -ENAMETOOLONG
;
3103 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3108 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3109 shmem_initxattrs
, NULL
);
3111 if (error
!= -EOPNOTSUPP
) {
3118 inode
->i_size
= len
-1;
3119 if (len
<= SHORT_SYMLINK_LEN
) {
3120 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3121 if (!inode
->i_link
) {
3125 inode
->i_op
= &shmem_short_symlink_operations
;
3127 inode_nohighmem(inode
);
3128 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3133 inode
->i_mapping
->a_ops
= &shmem_aops
;
3134 inode
->i_op
= &shmem_symlink_inode_operations
;
3135 memcpy(page_address(page
), symname
, len
);
3136 SetPageUptodate(page
);
3137 set_page_dirty(page
);
3141 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3142 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3143 d_instantiate(dentry
, inode
);
3148 static void shmem_put_link(void *arg
)
3150 mark_page_accessed(arg
);
3154 static const char *shmem_get_link(struct dentry
*dentry
,
3155 struct inode
*inode
,
3156 struct delayed_call
*done
)
3158 struct page
*page
= NULL
;
3161 page
= find_get_page(inode
->i_mapping
, 0);
3163 return ERR_PTR(-ECHILD
);
3164 if (!PageUptodate(page
)) {
3166 return ERR_PTR(-ECHILD
);
3169 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3171 return ERR_PTR(error
);
3174 set_delayed_call(done
, shmem_put_link
, page
);
3175 return page_address(page
);
3178 #ifdef CONFIG_TMPFS_XATTR
3180 * Superblocks without xattr inode operations may get some security.* xattr
3181 * support from the LSM "for free". As soon as we have any other xattrs
3182 * like ACLs, we also need to implement the security.* handlers at
3183 * filesystem level, though.
3187 * Callback for security_inode_init_security() for acquiring xattrs.
3189 static int shmem_initxattrs(struct inode
*inode
,
3190 const struct xattr
*xattr_array
,
3193 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3194 const struct xattr
*xattr
;
3195 struct simple_xattr
*new_xattr
;
3198 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3199 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3203 len
= strlen(xattr
->name
) + 1;
3204 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3206 if (!new_xattr
->name
) {
3211 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3212 XATTR_SECURITY_PREFIX_LEN
);
3213 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3216 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3222 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3223 struct dentry
*unused
, struct inode
*inode
,
3224 const char *name
, void *buffer
, size_t size
)
3226 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3228 name
= xattr_full_name(handler
, name
);
3229 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3232 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3233 struct dentry
*unused
, struct inode
*inode
,
3234 const char *name
, const void *value
,
3235 size_t size
, int flags
)
3237 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3239 name
= xattr_full_name(handler
, name
);
3240 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3243 static const struct xattr_handler shmem_security_xattr_handler
= {
3244 .prefix
= XATTR_SECURITY_PREFIX
,
3245 .get
= shmem_xattr_handler_get
,
3246 .set
= shmem_xattr_handler_set
,
3249 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3250 .prefix
= XATTR_TRUSTED_PREFIX
,
3251 .get
= shmem_xattr_handler_get
,
3252 .set
= shmem_xattr_handler_set
,
3255 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3256 #ifdef CONFIG_TMPFS_POSIX_ACL
3257 &posix_acl_access_xattr_handler
,
3258 &posix_acl_default_xattr_handler
,
3260 &shmem_security_xattr_handler
,
3261 &shmem_trusted_xattr_handler
,
3265 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3267 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3268 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3270 #endif /* CONFIG_TMPFS_XATTR */
3272 static const struct inode_operations shmem_short_symlink_operations
= {
3273 .get_link
= simple_get_link
,
3274 #ifdef CONFIG_TMPFS_XATTR
3275 .listxattr
= shmem_listxattr
,
3279 static const struct inode_operations shmem_symlink_inode_operations
= {
3280 .get_link
= shmem_get_link
,
3281 #ifdef CONFIG_TMPFS_XATTR
3282 .listxattr
= shmem_listxattr
,
3286 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3288 return ERR_PTR(-ESTALE
);
3291 static int shmem_match(struct inode
*ino
, void *vfh
)
3295 inum
= (inum
<< 32) | fh
[1];
3296 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3299 /* Find any alias of inode, but prefer a hashed alias */
3300 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3302 struct dentry
*alias
= d_find_alias(inode
);
3304 return alias
?: d_find_any_alias(inode
);
3308 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3309 struct fid
*fid
, int fh_len
, int fh_type
)
3311 struct inode
*inode
;
3312 struct dentry
*dentry
= NULL
;
3319 inum
= (inum
<< 32) | fid
->raw
[1];
3321 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3322 shmem_match
, fid
->raw
);
3324 dentry
= shmem_find_alias(inode
);
3331 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3332 struct inode
*parent
)
3336 return FILEID_INVALID
;
3339 if (inode_unhashed(inode
)) {
3340 /* Unfortunately insert_inode_hash is not idempotent,
3341 * so as we hash inodes here rather than at creation
3342 * time, we need a lock to ensure we only try
3345 static DEFINE_SPINLOCK(lock
);
3347 if (inode_unhashed(inode
))
3348 __insert_inode_hash(inode
,
3349 inode
->i_ino
+ inode
->i_generation
);
3353 fh
[0] = inode
->i_generation
;
3354 fh
[1] = inode
->i_ino
;
3355 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3361 static const struct export_operations shmem_export_ops
= {
3362 .get_parent
= shmem_get_parent
,
3363 .encode_fh
= shmem_encode_fh
,
3364 .fh_to_dentry
= shmem_fh_to_dentry
,
3378 static const struct fs_parameter_spec shmem_param_specs
[] = {
3379 fsparam_u32 ("gid", Opt_gid
),
3380 fsparam_enum ("huge", Opt_huge
),
3381 fsparam_u32oct("mode", Opt_mode
),
3382 fsparam_string("mpol", Opt_mpol
),
3383 fsparam_string("nr_blocks", Opt_nr_blocks
),
3384 fsparam_string("nr_inodes", Opt_nr_inodes
),
3385 fsparam_string("size", Opt_size
),
3386 fsparam_u32 ("uid", Opt_uid
),
3390 static const struct fs_parameter_enum shmem_param_enums
[] = {
3391 { Opt_huge
, "never", SHMEM_HUGE_NEVER
},
3392 { Opt_huge
, "always", SHMEM_HUGE_ALWAYS
},
3393 { Opt_huge
, "within_size", SHMEM_HUGE_WITHIN_SIZE
},
3394 { Opt_huge
, "advise", SHMEM_HUGE_ADVISE
},
3398 const struct fs_parameter_description shmem_fs_parameters
= {
3400 .specs
= shmem_param_specs
,
3401 .enums
= shmem_param_enums
,
3404 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3406 struct shmem_options
*ctx
= fc
->fs_private
;
3407 struct fs_parse_result result
;
3408 unsigned long long size
;
3412 opt
= fs_parse(fc
, &shmem_fs_parameters
, param
, &result
);
3418 size
= memparse(param
->string
, &rest
);
3420 size
<<= PAGE_SHIFT
;
3421 size
*= totalram_pages();
3427 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3428 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3431 ctx
->blocks
= memparse(param
->string
, &rest
);
3434 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3437 ctx
->inodes
= memparse(param
->string
, &rest
);
3440 ctx
->seen
|= SHMEM_SEEN_INODES
;
3443 ctx
->mode
= result
.uint_32
& 07777;
3446 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3447 if (!uid_valid(ctx
->uid
))
3451 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3452 if (!gid_valid(ctx
->gid
))
3456 ctx
->huge
= result
.uint_32
;
3457 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3458 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
3459 has_transparent_hugepage()))
3460 goto unsupported_parameter
;
3461 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3464 if (IS_ENABLED(CONFIG_NUMA
)) {
3465 mpol_put(ctx
->mpol
);
3467 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3471 goto unsupported_parameter
;
3475 unsupported_parameter
:
3476 return invalf(fc
, "tmpfs: Unsupported parameter '%s'", param
->key
);
3478 return invalf(fc
, "tmpfs: Bad value for '%s'", param
->key
);
3481 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3483 char *options
= data
;
3485 while (options
!= NULL
) {
3486 char *this_char
= options
;
3489 * NUL-terminate this option: unfortunately,
3490 * mount options form a comma-separated list,
3491 * but mpol's nodelist may also contain commas.
3493 options
= strchr(options
, ',');
3494 if (options
== NULL
)
3497 if (!isdigit(*options
)) {
3503 char *value
= strchr(this_char
,'=');
3509 len
= strlen(value
);
3511 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3520 * Reconfigure a shmem filesystem.
3522 * Note that we disallow change from limited->unlimited blocks/inodes while any
3523 * are in use; but we must separately disallow unlimited->limited, because in
3524 * that case we have no record of how much is already in use.
3526 static int shmem_reconfigure(struct fs_context
*fc
)
3528 struct shmem_options
*ctx
= fc
->fs_private
;
3529 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3530 unsigned long inodes
;
3533 spin_lock(&sbinfo
->stat_lock
);
3534 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3535 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3536 if (!sbinfo
->max_blocks
) {
3537 err
= "Cannot retroactively limit size";
3540 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3542 err
= "Too small a size for current use";
3546 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3547 if (!sbinfo
->max_inodes
) {
3548 err
= "Cannot retroactively limit inodes";
3551 if (ctx
->inodes
< inodes
) {
3552 err
= "Too few inodes for current use";
3557 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3558 sbinfo
->huge
= ctx
->huge
;
3559 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3560 sbinfo
->max_blocks
= ctx
->blocks
;
3561 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3562 sbinfo
->max_inodes
= ctx
->inodes
;
3563 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3567 * Preserve previous mempolicy unless mpol remount option was specified.
3570 mpol_put(sbinfo
->mpol
);
3571 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3574 spin_unlock(&sbinfo
->stat_lock
);
3577 spin_unlock(&sbinfo
->stat_lock
);
3578 return invalf(fc
, "tmpfs: %s", err
);
3581 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3583 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3585 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3586 seq_printf(seq
, ",size=%luk",
3587 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3588 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3589 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3590 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3591 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3592 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3593 seq_printf(seq
, ",uid=%u",
3594 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3595 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3596 seq_printf(seq
, ",gid=%u",
3597 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3598 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3599 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3601 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3603 shmem_show_mpol(seq
, sbinfo
->mpol
);
3607 #endif /* CONFIG_TMPFS */
3609 static void shmem_put_super(struct super_block
*sb
)
3611 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3613 percpu_counter_destroy(&sbinfo
->used_blocks
);
3614 mpol_put(sbinfo
->mpol
);
3616 sb
->s_fs_info
= NULL
;
3619 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3621 struct shmem_options
*ctx
= fc
->fs_private
;
3622 struct inode
*inode
;
3623 struct shmem_sb_info
*sbinfo
;
3626 /* Round up to L1_CACHE_BYTES to resist false sharing */
3627 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3628 L1_CACHE_BYTES
), GFP_KERNEL
);
3632 sb
->s_fs_info
= sbinfo
;
3636 * Per default we only allow half of the physical ram per
3637 * tmpfs instance, limiting inodes to one per page of lowmem;
3638 * but the internal instance is left unlimited.
3640 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3641 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3642 ctx
->blocks
= shmem_default_max_blocks();
3643 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3644 ctx
->inodes
= shmem_default_max_inodes();
3646 sb
->s_flags
|= SB_NOUSER
;
3648 sb
->s_export_op
= &shmem_export_ops
;
3649 sb
->s_flags
|= SB_NOSEC
;
3651 sb
->s_flags
|= SB_NOUSER
;
3653 sbinfo
->max_blocks
= ctx
->blocks
;
3654 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3655 sbinfo
->uid
= ctx
->uid
;
3656 sbinfo
->gid
= ctx
->gid
;
3657 sbinfo
->mode
= ctx
->mode
;
3658 sbinfo
->huge
= ctx
->huge
;
3659 sbinfo
->mpol
= ctx
->mpol
;
3662 spin_lock_init(&sbinfo
->stat_lock
);
3663 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3665 spin_lock_init(&sbinfo
->shrinklist_lock
);
3666 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3668 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3669 sb
->s_blocksize
= PAGE_SIZE
;
3670 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3671 sb
->s_magic
= TMPFS_MAGIC
;
3672 sb
->s_op
= &shmem_ops
;
3673 sb
->s_time_gran
= 1;
3674 #ifdef CONFIG_TMPFS_XATTR
3675 sb
->s_xattr
= shmem_xattr_handlers
;
3677 #ifdef CONFIG_TMPFS_POSIX_ACL
3678 sb
->s_flags
|= SB_POSIXACL
;
3680 uuid_gen(&sb
->s_uuid
);
3682 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3685 inode
->i_uid
= sbinfo
->uid
;
3686 inode
->i_gid
= sbinfo
->gid
;
3687 sb
->s_root
= d_make_root(inode
);
3693 shmem_put_super(sb
);
3697 static int shmem_get_tree(struct fs_context
*fc
)
3699 return get_tree_nodev(fc
, shmem_fill_super
);
3702 static void shmem_free_fc(struct fs_context
*fc
)
3704 struct shmem_options
*ctx
= fc
->fs_private
;
3707 mpol_put(ctx
->mpol
);
3712 static const struct fs_context_operations shmem_fs_context_ops
= {
3713 .free
= shmem_free_fc
,
3714 .get_tree
= shmem_get_tree
,
3716 .parse_monolithic
= shmem_parse_options
,
3717 .parse_param
= shmem_parse_one
,
3718 .reconfigure
= shmem_reconfigure
,
3722 static struct kmem_cache
*shmem_inode_cachep
;
3724 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3726 struct shmem_inode_info
*info
;
3727 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3730 return &info
->vfs_inode
;
3733 static void shmem_free_in_core_inode(struct inode
*inode
)
3735 if (S_ISLNK(inode
->i_mode
))
3736 kfree(inode
->i_link
);
3737 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3740 static void shmem_destroy_inode(struct inode
*inode
)
3742 if (S_ISREG(inode
->i_mode
))
3743 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3746 static void shmem_init_inode(void *foo
)
3748 struct shmem_inode_info
*info
= foo
;
3749 inode_init_once(&info
->vfs_inode
);
3752 static void shmem_init_inodecache(void)
3754 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3755 sizeof(struct shmem_inode_info
),
3756 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3759 static void shmem_destroy_inodecache(void)
3761 kmem_cache_destroy(shmem_inode_cachep
);
3764 static const struct address_space_operations shmem_aops
= {
3765 .writepage
= shmem_writepage
,
3766 .set_page_dirty
= __set_page_dirty_no_writeback
,
3768 .write_begin
= shmem_write_begin
,
3769 .write_end
= shmem_write_end
,
3771 #ifdef CONFIG_MIGRATION
3772 .migratepage
= migrate_page
,
3774 .error_remove_page
= generic_error_remove_page
,
3777 static const struct file_operations shmem_file_operations
= {
3779 .get_unmapped_area
= shmem_get_unmapped_area
,
3781 .llseek
= shmem_file_llseek
,
3782 .read_iter
= shmem_file_read_iter
,
3783 .write_iter
= generic_file_write_iter
,
3784 .fsync
= noop_fsync
,
3785 .splice_read
= generic_file_splice_read
,
3786 .splice_write
= iter_file_splice_write
,
3787 .fallocate
= shmem_fallocate
,
3791 static const struct inode_operations shmem_inode_operations
= {
3792 .getattr
= shmem_getattr
,
3793 .setattr
= shmem_setattr
,
3794 #ifdef CONFIG_TMPFS_XATTR
3795 .listxattr
= shmem_listxattr
,
3796 .set_acl
= simple_set_acl
,
3800 static const struct inode_operations shmem_dir_inode_operations
= {
3802 .create
= shmem_create
,
3803 .lookup
= simple_lookup
,
3805 .unlink
= shmem_unlink
,
3806 .symlink
= shmem_symlink
,
3807 .mkdir
= shmem_mkdir
,
3808 .rmdir
= shmem_rmdir
,
3809 .mknod
= shmem_mknod
,
3810 .rename
= shmem_rename2
,
3811 .tmpfile
= shmem_tmpfile
,
3813 #ifdef CONFIG_TMPFS_XATTR
3814 .listxattr
= shmem_listxattr
,
3816 #ifdef CONFIG_TMPFS_POSIX_ACL
3817 .setattr
= shmem_setattr
,
3818 .set_acl
= simple_set_acl
,
3822 static const struct inode_operations shmem_special_inode_operations
= {
3823 #ifdef CONFIG_TMPFS_XATTR
3824 .listxattr
= shmem_listxattr
,
3826 #ifdef CONFIG_TMPFS_POSIX_ACL
3827 .setattr
= shmem_setattr
,
3828 .set_acl
= simple_set_acl
,
3832 static const struct super_operations shmem_ops
= {
3833 .alloc_inode
= shmem_alloc_inode
,
3834 .free_inode
= shmem_free_in_core_inode
,
3835 .destroy_inode
= shmem_destroy_inode
,
3837 .statfs
= shmem_statfs
,
3838 .show_options
= shmem_show_options
,
3840 .evict_inode
= shmem_evict_inode
,
3841 .drop_inode
= generic_delete_inode
,
3842 .put_super
= shmem_put_super
,
3843 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3844 .nr_cached_objects
= shmem_unused_huge_count
,
3845 .free_cached_objects
= shmem_unused_huge_scan
,
3849 static const struct vm_operations_struct shmem_vm_ops
= {
3850 .fault
= shmem_fault
,
3851 .map_pages
= filemap_map_pages
,
3853 .set_policy
= shmem_set_policy
,
3854 .get_policy
= shmem_get_policy
,
3858 int shmem_init_fs_context(struct fs_context
*fc
)
3860 struct shmem_options
*ctx
;
3862 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3866 ctx
->mode
= 0777 | S_ISVTX
;
3867 ctx
->uid
= current_fsuid();
3868 ctx
->gid
= current_fsgid();
3870 fc
->fs_private
= ctx
;
3871 fc
->ops
= &shmem_fs_context_ops
;
3875 static struct file_system_type shmem_fs_type
= {
3876 .owner
= THIS_MODULE
,
3878 .init_fs_context
= shmem_init_fs_context
,
3880 .parameters
= &shmem_fs_parameters
,
3882 .kill_sb
= kill_litter_super
,
3883 .fs_flags
= FS_USERNS_MOUNT
,
3886 int __init
shmem_init(void)
3890 shmem_init_inodecache();
3892 error
= register_filesystem(&shmem_fs_type
);
3894 pr_err("Could not register tmpfs\n");
3898 shm_mnt
= kern_mount(&shmem_fs_type
);
3899 if (IS_ERR(shm_mnt
)) {
3900 error
= PTR_ERR(shm_mnt
);
3901 pr_err("Could not kern_mount tmpfs\n");
3905 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3906 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3907 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3909 shmem_huge
= 0; /* just in case it was patched */
3914 unregister_filesystem(&shmem_fs_type
);
3916 shmem_destroy_inodecache();
3917 shm_mnt
= ERR_PTR(error
);
3921 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3922 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3923 struct kobj_attribute
*attr
, char *buf
)
3927 SHMEM_HUGE_WITHIN_SIZE
,
3935 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3936 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3938 count
+= sprintf(buf
+ count
, fmt
,
3939 shmem_format_huge(values
[i
]));
3941 buf
[count
- 1] = '\n';
3945 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3946 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3951 if (count
+ 1 > sizeof(tmp
))
3953 memcpy(tmp
, buf
, count
);
3955 if (count
&& tmp
[count
- 1] == '\n')
3956 tmp
[count
- 1] = '\0';
3958 huge
= shmem_parse_huge(tmp
);
3959 if (huge
== -EINVAL
)
3961 if (!has_transparent_hugepage() &&
3962 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3966 if (shmem_huge
> SHMEM_HUGE_DENY
)
3967 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3971 struct kobj_attribute shmem_enabled_attr
=
3972 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3973 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3975 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3976 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
3978 struct inode
*inode
= file_inode(vma
->vm_file
);
3979 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3983 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
3984 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
3986 if (shmem_huge
== SHMEM_HUGE_FORCE
)
3988 if (shmem_huge
== SHMEM_HUGE_DENY
)
3990 switch (sbinfo
->huge
) {
3991 case SHMEM_HUGE_NEVER
:
3993 case SHMEM_HUGE_ALWAYS
:
3995 case SHMEM_HUGE_WITHIN_SIZE
:
3996 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
3997 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
3998 if (i_size
>= HPAGE_PMD_SIZE
&&
3999 i_size
>> PAGE_SHIFT
>= off
)
4002 case SHMEM_HUGE_ADVISE
:
4003 /* TODO: implement fadvise() hints */
4004 return (vma
->vm_flags
& VM_HUGEPAGE
);
4010 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4012 #else /* !CONFIG_SHMEM */
4015 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4017 * This is intended for small system where the benefits of the full
4018 * shmem code (swap-backed and resource-limited) are outweighed by
4019 * their complexity. On systems without swap this code should be
4020 * effectively equivalent, but much lighter weight.
4023 static struct file_system_type shmem_fs_type
= {
4025 .init_fs_context
= ramfs_init_fs_context
,
4026 .parameters
= &ramfs_fs_parameters
,
4027 .kill_sb
= kill_litter_super
,
4028 .fs_flags
= FS_USERNS_MOUNT
,
4031 int __init
shmem_init(void)
4033 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4035 shm_mnt
= kern_mount(&shmem_fs_type
);
4036 BUG_ON(IS_ERR(shm_mnt
));
4041 int shmem_unuse(unsigned int type
, bool frontswap
,
4042 unsigned long *fs_pages_to_unuse
)
4047 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4052 void shmem_unlock_mapping(struct address_space
*mapping
)
4057 unsigned long shmem_get_unmapped_area(struct file
*file
,
4058 unsigned long addr
, unsigned long len
,
4059 unsigned long pgoff
, unsigned long flags
)
4061 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4065 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4067 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4069 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4071 #define shmem_vm_ops generic_file_vm_ops
4072 #define shmem_file_operations ramfs_file_operations
4073 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4074 #define shmem_acct_size(flags, size) 0
4075 #define shmem_unacct_size(flags, size) do {} while (0)
4077 #endif /* CONFIG_SHMEM */
4081 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4082 unsigned long flags
, unsigned int i_flags
)
4084 struct inode
*inode
;
4088 return ERR_CAST(mnt
);
4090 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4091 return ERR_PTR(-EINVAL
);
4093 if (shmem_acct_size(flags
, size
))
4094 return ERR_PTR(-ENOMEM
);
4096 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4098 if (unlikely(!inode
)) {
4099 shmem_unacct_size(flags
, size
);
4100 return ERR_PTR(-ENOSPC
);
4102 inode
->i_flags
|= i_flags
;
4103 inode
->i_size
= size
;
4104 clear_nlink(inode
); /* It is unlinked */
4105 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4107 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4108 &shmem_file_operations
);
4115 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4116 * kernel internal. There will be NO LSM permission checks against the
4117 * underlying inode. So users of this interface must do LSM checks at a
4118 * higher layer. The users are the big_key and shm implementations. LSM
4119 * checks are provided at the key or shm level rather than the inode.
4120 * @name: name for dentry (to be seen in /proc/<pid>/maps
4121 * @size: size to be set for the file
4122 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4124 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4126 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4130 * shmem_file_setup - get an unlinked file living in tmpfs
4131 * @name: name for dentry (to be seen in /proc/<pid>/maps
4132 * @size: size to be set for the file
4133 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4135 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4137 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4139 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4142 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4143 * @mnt: the tmpfs mount where the file will be created
4144 * @name: name for dentry (to be seen in /proc/<pid>/maps
4145 * @size: size to be set for the file
4146 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4148 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4149 loff_t size
, unsigned long flags
)
4151 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4153 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4156 * shmem_zero_setup - setup a shared anonymous mapping
4157 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4159 int shmem_zero_setup(struct vm_area_struct
*vma
)
4162 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4165 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4166 * between XFS directory reading and selinux: since this file is only
4167 * accessible to the user through its mapping, use S_PRIVATE flag to
4168 * bypass file security, in the same way as shmem_kernel_file_setup().
4170 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4172 return PTR_ERR(file
);
4176 vma
->vm_file
= file
;
4177 vma
->vm_ops
= &shmem_vm_ops
;
4179 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4180 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4181 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4182 khugepaged_enter(vma
, vma
->vm_flags
);
4189 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4190 * @mapping: the page's address_space
4191 * @index: the page index
4192 * @gfp: the page allocator flags to use if allocating
4194 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4195 * with any new page allocations done using the specified allocation flags.
4196 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4197 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4198 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4200 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4201 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4203 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4204 pgoff_t index
, gfp_t gfp
)
4207 struct inode
*inode
= mapping
->host
;
4211 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4212 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4213 gfp
, NULL
, NULL
, NULL
);
4215 page
= ERR_PTR(error
);
4221 * The tiny !SHMEM case uses ramfs without swap
4223 return read_cache_page_gfp(mapping
, index
, gfp
);
4226 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);