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
;
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 int shmem_default_max_inodes(void)
133 unsigned long nr_pages
= totalram_pages();
137 ul
= min3(ul
, nr_pages
- totalhigh_pages(), nr_pages
/ 2);
142 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
143 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
144 struct shmem_inode_info
*info
, pgoff_t index
);
145 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
146 struct page
**pagep
, enum sgp_type sgp
,
147 gfp_t gfp
, struct vm_area_struct
*vma
,
148 vm_fault_t
*fault_type
);
149 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
150 struct page
**pagep
, enum sgp_type sgp
,
151 gfp_t gfp
, struct vm_area_struct
*vma
,
152 struct vm_fault
*vmf
, vm_fault_t
*fault_type
);
154 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
155 struct page
**pagep
, enum sgp_type sgp
)
157 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
158 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
161 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
163 return sb
->s_fs_info
;
167 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
168 * for shared memory and for shared anonymous (/dev/zero) mappings
169 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
170 * consistent with the pre-accounting of private mappings ...
172 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
174 return (flags
& VM_NORESERVE
) ?
175 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
178 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
180 if (!(flags
& VM_NORESERVE
))
181 vm_unacct_memory(VM_ACCT(size
));
184 static inline int shmem_reacct_size(unsigned long flags
,
185 loff_t oldsize
, loff_t newsize
)
187 if (!(flags
& VM_NORESERVE
)) {
188 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
189 return security_vm_enough_memory_mm(current
->mm
,
190 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
191 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
192 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
198 * ... whereas tmpfs objects are accounted incrementally as
199 * pages are allocated, in order to allow large sparse files.
200 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
201 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
203 static inline int shmem_acct_block(unsigned long flags
, long pages
)
205 if (!(flags
& VM_NORESERVE
))
208 return security_vm_enough_memory_mm(current
->mm
,
209 pages
* VM_ACCT(PAGE_SIZE
));
212 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
214 if (flags
& VM_NORESERVE
)
215 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
218 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
220 struct shmem_inode_info
*info
= SHMEM_I(inode
);
221 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
223 if (shmem_acct_block(info
->flags
, pages
))
226 if (sbinfo
->max_blocks
) {
227 if (percpu_counter_compare(&sbinfo
->used_blocks
,
228 sbinfo
->max_blocks
- pages
) > 0)
230 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
236 shmem_unacct_blocks(info
->flags
, pages
);
240 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
242 struct shmem_inode_info
*info
= SHMEM_I(inode
);
243 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
245 if (sbinfo
->max_blocks
)
246 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
247 shmem_unacct_blocks(info
->flags
, pages
);
250 static const struct super_operations shmem_ops
;
251 static const struct address_space_operations shmem_aops
;
252 static const struct file_operations shmem_file_operations
;
253 static const struct inode_operations shmem_inode_operations
;
254 static const struct inode_operations shmem_dir_inode_operations
;
255 static const struct inode_operations shmem_special_inode_operations
;
256 static const struct vm_operations_struct shmem_vm_ops
;
257 static struct file_system_type shmem_fs_type
;
259 bool vma_is_shmem(struct vm_area_struct
*vma
)
261 return vma
->vm_ops
== &shmem_vm_ops
;
264 static LIST_HEAD(shmem_swaplist
);
265 static DEFINE_MUTEX(shmem_swaplist_mutex
);
267 static int shmem_reserve_inode(struct super_block
*sb
)
269 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
270 if (sbinfo
->max_inodes
) {
271 spin_lock(&sbinfo
->stat_lock
);
272 if (!sbinfo
->free_inodes
) {
273 spin_unlock(&sbinfo
->stat_lock
);
276 sbinfo
->free_inodes
--;
277 spin_unlock(&sbinfo
->stat_lock
);
282 static void shmem_free_inode(struct super_block
*sb
)
284 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
285 if (sbinfo
->max_inodes
) {
286 spin_lock(&sbinfo
->stat_lock
);
287 sbinfo
->free_inodes
++;
288 spin_unlock(&sbinfo
->stat_lock
);
293 * shmem_recalc_inode - recalculate the block usage of an inode
294 * @inode: inode to recalc
296 * We have to calculate the free blocks since the mm can drop
297 * undirtied hole pages behind our back.
299 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
300 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
302 * It has to be called with the spinlock held.
304 static void shmem_recalc_inode(struct inode
*inode
)
306 struct shmem_inode_info
*info
= SHMEM_I(inode
);
309 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
311 info
->alloced
-= freed
;
312 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
313 shmem_inode_unacct_blocks(inode
, freed
);
317 bool shmem_charge(struct inode
*inode
, long pages
)
319 struct shmem_inode_info
*info
= SHMEM_I(inode
);
322 if (!shmem_inode_acct_block(inode
, pages
))
325 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
326 inode
->i_mapping
->nrpages
+= pages
;
328 spin_lock_irqsave(&info
->lock
, flags
);
329 info
->alloced
+= pages
;
330 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
331 shmem_recalc_inode(inode
);
332 spin_unlock_irqrestore(&info
->lock
, flags
);
337 void shmem_uncharge(struct inode
*inode
, long pages
)
339 struct shmem_inode_info
*info
= SHMEM_I(inode
);
342 /* nrpages adjustment done by __delete_from_page_cache() or caller */
344 spin_lock_irqsave(&info
->lock
, flags
);
345 info
->alloced
-= pages
;
346 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
347 shmem_recalc_inode(inode
);
348 spin_unlock_irqrestore(&info
->lock
, flags
);
350 shmem_inode_unacct_blocks(inode
, pages
);
354 * Replace item expected in xarray by a new item, while holding xa_lock.
356 static int shmem_replace_entry(struct address_space
*mapping
,
357 pgoff_t index
, void *expected
, void *replacement
)
359 XA_STATE(xas
, &mapping
->i_pages
, index
);
362 VM_BUG_ON(!expected
);
363 VM_BUG_ON(!replacement
);
364 item
= xas_load(&xas
);
365 if (item
!= expected
)
367 xas_store(&xas
, replacement
);
372 * Sometimes, before we decide whether to proceed or to fail, we must check
373 * that an entry was not already brought back from swap by a racing thread.
375 * Checking page is not enough: by the time a SwapCache page is locked, it
376 * might be reused, and again be SwapCache, using the same swap as before.
378 static bool shmem_confirm_swap(struct address_space
*mapping
,
379 pgoff_t index
, swp_entry_t swap
)
381 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
385 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
388 * disables huge pages for the mount;
390 * enables huge pages for the mount;
391 * SHMEM_HUGE_WITHIN_SIZE:
392 * only allocate huge pages if the page will be fully within i_size,
393 * also respect fadvise()/madvise() hints;
395 * only allocate huge pages if requested with fadvise()/madvise();
398 #define SHMEM_HUGE_NEVER 0
399 #define SHMEM_HUGE_ALWAYS 1
400 #define SHMEM_HUGE_WITHIN_SIZE 2
401 #define SHMEM_HUGE_ADVISE 3
405 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
408 * disables huge on shm_mnt and all mounts, for emergency use;
410 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
413 #define SHMEM_HUGE_DENY (-1)
414 #define SHMEM_HUGE_FORCE (-2)
416 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
417 /* ifdef here to avoid bloating shmem.o when not necessary */
419 static int shmem_huge __read_mostly
;
421 #if defined(CONFIG_SYSFS)
422 static int shmem_parse_huge(const char *str
)
424 if (!strcmp(str
, "never"))
425 return SHMEM_HUGE_NEVER
;
426 if (!strcmp(str
, "always"))
427 return SHMEM_HUGE_ALWAYS
;
428 if (!strcmp(str
, "within_size"))
429 return SHMEM_HUGE_WITHIN_SIZE
;
430 if (!strcmp(str
, "advise"))
431 return SHMEM_HUGE_ADVISE
;
432 if (!strcmp(str
, "deny"))
433 return SHMEM_HUGE_DENY
;
434 if (!strcmp(str
, "force"))
435 return SHMEM_HUGE_FORCE
;
440 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
441 static const char *shmem_format_huge(int huge
)
444 case SHMEM_HUGE_NEVER
:
446 case SHMEM_HUGE_ALWAYS
:
448 case SHMEM_HUGE_WITHIN_SIZE
:
449 return "within_size";
450 case SHMEM_HUGE_ADVISE
:
452 case SHMEM_HUGE_DENY
:
454 case SHMEM_HUGE_FORCE
:
463 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
464 struct shrink_control
*sc
, unsigned long nr_to_split
)
466 LIST_HEAD(list
), *pos
, *next
;
467 LIST_HEAD(to_remove
);
469 struct shmem_inode_info
*info
;
471 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
472 int removed
= 0, split
= 0;
474 if (list_empty(&sbinfo
->shrinklist
))
477 spin_lock(&sbinfo
->shrinklist_lock
);
478 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
479 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
482 inode
= igrab(&info
->vfs_inode
);
484 /* inode is about to be evicted */
486 list_del_init(&info
->shrinklist
);
491 /* Check if there's anything to gain */
492 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
493 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
494 list_move(&info
->shrinklist
, &to_remove
);
499 list_move(&info
->shrinklist
, &list
);
504 spin_unlock(&sbinfo
->shrinklist_lock
);
506 list_for_each_safe(pos
, next
, &to_remove
) {
507 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
508 inode
= &info
->vfs_inode
;
509 list_del_init(&info
->shrinklist
);
513 list_for_each_safe(pos
, next
, &list
) {
516 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
517 inode
= &info
->vfs_inode
;
519 if (nr_to_split
&& split
>= nr_to_split
)
522 page
= find_get_page(inode
->i_mapping
,
523 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
527 /* No huge page at the end of the file: nothing to split */
528 if (!PageTransHuge(page
)) {
534 * Leave the inode on the list if we failed to lock
535 * the page at this time.
537 * Waiting for the lock may lead to deadlock in the
540 if (!trylock_page(page
)) {
545 ret
= split_huge_page(page
);
549 /* If split failed leave the inode on the list */
555 list_del_init(&info
->shrinklist
);
561 spin_lock(&sbinfo
->shrinklist_lock
);
562 list_splice_tail(&list
, &sbinfo
->shrinklist
);
563 sbinfo
->shrinklist_len
-= removed
;
564 spin_unlock(&sbinfo
->shrinklist_lock
);
569 static long shmem_unused_huge_scan(struct super_block
*sb
,
570 struct shrink_control
*sc
)
572 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
574 if (!READ_ONCE(sbinfo
->shrinklist_len
))
577 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
580 static long shmem_unused_huge_count(struct super_block
*sb
,
581 struct shrink_control
*sc
)
583 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
584 return READ_ONCE(sbinfo
->shrinklist_len
);
586 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
588 #define shmem_huge SHMEM_HUGE_DENY
590 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
591 struct shrink_control
*sc
, unsigned long nr_to_split
)
595 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
597 static inline bool is_huge_enabled(struct shmem_sb_info
*sbinfo
)
599 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
600 (shmem_huge
== SHMEM_HUGE_FORCE
|| sbinfo
->huge
) &&
601 shmem_huge
!= SHMEM_HUGE_DENY
)
607 * Like add_to_page_cache_locked, but error if expected item has gone.
609 static int shmem_add_to_page_cache(struct page
*page
,
610 struct address_space
*mapping
,
611 pgoff_t index
, void *expected
, gfp_t gfp
)
613 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, compound_order(page
));
615 unsigned long nr
= compound_nr(page
);
617 VM_BUG_ON_PAGE(PageTail(page
), page
);
618 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
619 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
620 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
621 VM_BUG_ON(expected
&& PageTransHuge(page
));
623 page_ref_add(page
, nr
);
624 page
->mapping
= mapping
;
630 entry
= xas_find_conflict(&xas
);
631 if (entry
!= expected
)
632 xas_set_err(&xas
, -EEXIST
);
633 xas_create_range(&xas
);
637 xas_store(&xas
, page
);
642 if (PageTransHuge(page
)) {
643 count_vm_event(THP_FILE_ALLOC
);
644 __inc_node_page_state(page
, NR_SHMEM_THPS
);
646 mapping
->nrpages
+= nr
;
647 __mod_node_page_state(page_pgdat(page
), NR_FILE_PAGES
, nr
);
648 __mod_node_page_state(page_pgdat(page
), NR_SHMEM
, nr
);
650 xas_unlock_irq(&xas
);
651 } while (xas_nomem(&xas
, gfp
));
653 if (xas_error(&xas
)) {
654 page
->mapping
= NULL
;
655 page_ref_sub(page
, nr
);
656 return xas_error(&xas
);
663 * Like delete_from_page_cache, but substitutes swap for page.
665 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
667 struct address_space
*mapping
= page
->mapping
;
670 VM_BUG_ON_PAGE(PageCompound(page
), page
);
672 xa_lock_irq(&mapping
->i_pages
);
673 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
674 page
->mapping
= NULL
;
676 __dec_node_page_state(page
, NR_FILE_PAGES
);
677 __dec_node_page_state(page
, NR_SHMEM
);
678 xa_unlock_irq(&mapping
->i_pages
);
684 * Remove swap entry from page cache, free the swap and its page cache.
686 static int shmem_free_swap(struct address_space
*mapping
,
687 pgoff_t index
, void *radswap
)
691 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
694 free_swap_and_cache(radix_to_swp_entry(radswap
));
699 * Determine (in bytes) how many of the shmem object's pages mapped by the
700 * given offsets are swapped out.
702 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
703 * as long as the inode doesn't go away and racy results are not a problem.
705 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
706 pgoff_t start
, pgoff_t end
)
708 XA_STATE(xas
, &mapping
->i_pages
, start
);
710 unsigned long swapped
= 0;
713 xas_for_each(&xas
, page
, end
- 1) {
714 if (xas_retry(&xas
, page
))
716 if (xa_is_value(page
))
719 if (need_resched()) {
727 return swapped
<< PAGE_SHIFT
;
731 * Determine (in bytes) how many of the shmem object's pages mapped by the
732 * given vma is swapped out.
734 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
735 * as long as the inode doesn't go away and racy results are not a problem.
737 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
739 struct inode
*inode
= file_inode(vma
->vm_file
);
740 struct shmem_inode_info
*info
= SHMEM_I(inode
);
741 struct address_space
*mapping
= inode
->i_mapping
;
742 unsigned long swapped
;
744 /* Be careful as we don't hold info->lock */
745 swapped
= READ_ONCE(info
->swapped
);
748 * The easier cases are when the shmem object has nothing in swap, or
749 * the vma maps it whole. Then we can simply use the stats that we
755 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
756 return swapped
<< PAGE_SHIFT
;
758 /* Here comes the more involved part */
759 return shmem_partial_swap_usage(mapping
,
760 linear_page_index(vma
, vma
->vm_start
),
761 linear_page_index(vma
, vma
->vm_end
));
765 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
767 void shmem_unlock_mapping(struct address_space
*mapping
)
770 pgoff_t indices
[PAGEVEC_SIZE
];
775 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
777 while (!mapping_unevictable(mapping
)) {
779 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
780 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
782 pvec
.nr
= find_get_entries(mapping
, index
,
783 PAGEVEC_SIZE
, pvec
.pages
, indices
);
786 index
= indices
[pvec
.nr
- 1] + 1;
787 pagevec_remove_exceptionals(&pvec
);
788 check_move_unevictable_pages(&pvec
);
789 pagevec_release(&pvec
);
795 * Remove range of pages and swap entries from page cache, and free them.
796 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
798 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
801 struct address_space
*mapping
= inode
->i_mapping
;
802 struct shmem_inode_info
*info
= SHMEM_I(inode
);
803 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
804 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
805 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
806 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
808 pgoff_t indices
[PAGEVEC_SIZE
];
809 long nr_swaps_freed
= 0;
814 end
= -1; /* unsigned, so actually very big */
818 while (index
< end
) {
819 pvec
.nr
= find_get_entries(mapping
, index
,
820 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
821 pvec
.pages
, indices
);
824 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
825 struct page
*page
= pvec
.pages
[i
];
831 if (xa_is_value(page
)) {
834 nr_swaps_freed
+= !shmem_free_swap(mapping
,
839 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
841 if (!trylock_page(page
))
844 if (PageTransTail(page
)) {
845 /* Middle of THP: zero out the page */
846 clear_highpage(page
);
849 } else if (PageTransHuge(page
)) {
850 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
852 * Range ends in the middle of THP:
855 clear_highpage(page
);
859 index
+= HPAGE_PMD_NR
- 1;
860 i
+= HPAGE_PMD_NR
- 1;
863 if (!unfalloc
|| !PageUptodate(page
)) {
864 VM_BUG_ON_PAGE(PageTail(page
), page
);
865 if (page_mapping(page
) == mapping
) {
866 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
867 truncate_inode_page(mapping
, page
);
872 pagevec_remove_exceptionals(&pvec
);
873 pagevec_release(&pvec
);
879 struct page
*page
= NULL
;
880 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
882 unsigned int top
= PAGE_SIZE
;
887 zero_user_segment(page
, partial_start
, top
);
888 set_page_dirty(page
);
894 struct page
*page
= NULL
;
895 shmem_getpage(inode
, end
, &page
, SGP_READ
);
897 zero_user_segment(page
, 0, partial_end
);
898 set_page_dirty(page
);
907 while (index
< end
) {
910 pvec
.nr
= find_get_entries(mapping
, index
,
911 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
912 pvec
.pages
, indices
);
914 /* If all gone or hole-punch or unfalloc, we're done */
915 if (index
== start
|| end
!= -1)
917 /* But if truncating, restart to make sure all gone */
921 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
922 struct page
*page
= pvec
.pages
[i
];
928 if (xa_is_value(page
)) {
931 if (shmem_free_swap(mapping
, index
, page
)) {
932 /* Swap was replaced by page: retry */
942 if (PageTransTail(page
)) {
943 /* Middle of THP: zero out the page */
944 clear_highpage(page
);
947 * Partial thp truncate due 'start' in middle
948 * of THP: don't need to look on these pages
949 * again on !pvec.nr restart.
951 if (index
!= round_down(end
, HPAGE_PMD_NR
))
954 } else if (PageTransHuge(page
)) {
955 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
957 * Range ends in the middle of THP:
960 clear_highpage(page
);
964 index
+= HPAGE_PMD_NR
- 1;
965 i
+= HPAGE_PMD_NR
- 1;
968 if (!unfalloc
|| !PageUptodate(page
)) {
969 VM_BUG_ON_PAGE(PageTail(page
), page
);
970 if (page_mapping(page
) == mapping
) {
971 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
972 truncate_inode_page(mapping
, page
);
974 /* Page was replaced by swap: retry */
982 pagevec_remove_exceptionals(&pvec
);
983 pagevec_release(&pvec
);
987 spin_lock_irq(&info
->lock
);
988 info
->swapped
-= nr_swaps_freed
;
989 shmem_recalc_inode(inode
);
990 spin_unlock_irq(&info
->lock
);
993 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
995 shmem_undo_range(inode
, lstart
, lend
, false);
996 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
998 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1000 static int shmem_getattr(const struct path
*path
, struct kstat
*stat
,
1001 u32 request_mask
, unsigned int query_flags
)
1003 struct inode
*inode
= path
->dentry
->d_inode
;
1004 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1005 struct shmem_sb_info
*sb_info
= SHMEM_SB(inode
->i_sb
);
1007 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1008 spin_lock_irq(&info
->lock
);
1009 shmem_recalc_inode(inode
);
1010 spin_unlock_irq(&info
->lock
);
1012 generic_fillattr(inode
, stat
);
1014 if (is_huge_enabled(sb_info
))
1015 stat
->blksize
= HPAGE_PMD_SIZE
;
1020 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1022 struct inode
*inode
= d_inode(dentry
);
1023 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1024 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1027 error
= setattr_prepare(dentry
, attr
);
1031 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1032 loff_t oldsize
= inode
->i_size
;
1033 loff_t newsize
= attr
->ia_size
;
1035 /* protected by i_mutex */
1036 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1037 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1040 if (newsize
!= oldsize
) {
1041 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1045 i_size_write(inode
, newsize
);
1046 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1048 if (newsize
<= oldsize
) {
1049 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1050 if (oldsize
> holebegin
)
1051 unmap_mapping_range(inode
->i_mapping
,
1054 shmem_truncate_range(inode
,
1055 newsize
, (loff_t
)-1);
1056 /* unmap again to remove racily COWed private pages */
1057 if (oldsize
> holebegin
)
1058 unmap_mapping_range(inode
->i_mapping
,
1062 * Part of the huge page can be beyond i_size: subject
1063 * to shrink under memory pressure.
1065 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
)) {
1066 spin_lock(&sbinfo
->shrinklist_lock
);
1068 * _careful to defend against unlocked access to
1069 * ->shrink_list in shmem_unused_huge_shrink()
1071 if (list_empty_careful(&info
->shrinklist
)) {
1072 list_add_tail(&info
->shrinklist
,
1073 &sbinfo
->shrinklist
);
1074 sbinfo
->shrinklist_len
++;
1076 spin_unlock(&sbinfo
->shrinklist_lock
);
1081 setattr_copy(inode
, attr
);
1082 if (attr
->ia_valid
& ATTR_MODE
)
1083 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1087 static void shmem_evict_inode(struct inode
*inode
)
1089 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1090 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1092 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1093 shmem_unacct_size(info
->flags
, inode
->i_size
);
1095 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1096 if (!list_empty(&info
->shrinklist
)) {
1097 spin_lock(&sbinfo
->shrinklist_lock
);
1098 if (!list_empty(&info
->shrinklist
)) {
1099 list_del_init(&info
->shrinklist
);
1100 sbinfo
->shrinklist_len
--;
1102 spin_unlock(&sbinfo
->shrinklist_lock
);
1104 while (!list_empty(&info
->swaplist
)) {
1105 /* Wait while shmem_unuse() is scanning this inode... */
1106 wait_var_event(&info
->stop_eviction
,
1107 !atomic_read(&info
->stop_eviction
));
1108 mutex_lock(&shmem_swaplist_mutex
);
1109 /* ...but beware of the race if we peeked too early */
1110 if (!atomic_read(&info
->stop_eviction
))
1111 list_del_init(&info
->swaplist
);
1112 mutex_unlock(&shmem_swaplist_mutex
);
1116 simple_xattrs_free(&info
->xattrs
);
1117 WARN_ON(inode
->i_blocks
);
1118 if (!sbinfo
->idr_nouse
&& inode
->i_ino
) {
1119 mutex_lock(&sbinfo
->idr_lock
);
1120 idr_remove(&sbinfo
->idr
, inode
->i_ino
);
1121 mutex_unlock(&sbinfo
->idr_lock
);
1123 shmem_free_inode(inode
->i_sb
);
1127 extern struct swap_info_struct
*swap_info
[];
1129 static int shmem_find_swap_entries(struct address_space
*mapping
,
1130 pgoff_t start
, unsigned int nr_entries
,
1131 struct page
**entries
, pgoff_t
*indices
,
1132 unsigned int type
, bool frontswap
)
1134 XA_STATE(xas
, &mapping
->i_pages
, start
);
1137 unsigned int ret
= 0;
1143 xas_for_each(&xas
, page
, ULONG_MAX
) {
1144 if (xas_retry(&xas
, page
))
1147 if (!xa_is_value(page
))
1150 entry
= radix_to_swp_entry(page
);
1151 if (swp_type(entry
) != type
)
1154 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1157 indices
[ret
] = xas
.xa_index
;
1158 entries
[ret
] = page
;
1160 if (need_resched()) {
1164 if (++ret
== nr_entries
)
1173 * Move the swapped pages for an inode to page cache. Returns the count
1174 * of pages swapped in, or the error in case of failure.
1176 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1182 struct address_space
*mapping
= inode
->i_mapping
;
1184 for (i
= 0; i
< pvec
.nr
; i
++) {
1185 struct page
*page
= pvec
.pages
[i
];
1187 if (!xa_is_value(page
))
1189 error
= shmem_swapin_page(inode
, indices
[i
],
1191 mapping_gfp_mask(mapping
),
1198 if (error
== -ENOMEM
)
1202 return error
? error
: ret
;
1206 * If swap found in inode, free it and move page from swapcache to filecache.
1208 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1209 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1211 struct address_space
*mapping
= inode
->i_mapping
;
1213 struct pagevec pvec
;
1214 pgoff_t indices
[PAGEVEC_SIZE
];
1215 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1218 pagevec_init(&pvec
);
1220 unsigned int nr_entries
= PAGEVEC_SIZE
;
1222 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1223 nr_entries
= *fs_pages_to_unuse
;
1225 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1226 pvec
.pages
, indices
,
1233 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1237 if (frontswap_partial
) {
1238 *fs_pages_to_unuse
-= ret
;
1239 if (*fs_pages_to_unuse
== 0) {
1240 ret
= FRONTSWAP_PAGES_UNUSED
;
1245 start
= indices
[pvec
.nr
- 1];
1252 * Read all the shared memory data that resides in the swap
1253 * device 'type' back into memory, so the swap device can be
1256 int shmem_unuse(unsigned int type
, bool frontswap
,
1257 unsigned long *fs_pages_to_unuse
)
1259 struct shmem_inode_info
*info
, *next
;
1262 if (list_empty(&shmem_swaplist
))
1265 mutex_lock(&shmem_swaplist_mutex
);
1266 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1267 if (!info
->swapped
) {
1268 list_del_init(&info
->swaplist
);
1272 * Drop the swaplist mutex while searching the inode for swap;
1273 * but before doing so, make sure shmem_evict_inode() will not
1274 * remove placeholder inode from swaplist, nor let it be freed
1275 * (igrab() would protect from unlink, but not from unmount).
1277 atomic_inc(&info
->stop_eviction
);
1278 mutex_unlock(&shmem_swaplist_mutex
);
1280 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1284 mutex_lock(&shmem_swaplist_mutex
);
1285 next
= list_next_entry(info
, swaplist
);
1287 list_del_init(&info
->swaplist
);
1288 if (atomic_dec_and_test(&info
->stop_eviction
))
1289 wake_up_var(&info
->stop_eviction
);
1293 mutex_unlock(&shmem_swaplist_mutex
);
1299 * Move the page from the page cache to the swap cache.
1301 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1303 struct shmem_inode_info
*info
;
1304 struct address_space
*mapping
;
1305 struct inode
*inode
;
1309 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1310 BUG_ON(!PageLocked(page
));
1311 mapping
= page
->mapping
;
1312 index
= page
->index
;
1313 inode
= mapping
->host
;
1314 info
= SHMEM_I(inode
);
1315 if (info
->flags
& VM_LOCKED
)
1317 if (!total_swap_pages
)
1321 * Our capabilities prevent regular writeback or sync from ever calling
1322 * shmem_writepage; but a stacking filesystem might use ->writepage of
1323 * its underlying filesystem, in which case tmpfs should write out to
1324 * swap only in response to memory pressure, and not for the writeback
1327 if (!wbc
->for_reclaim
) {
1328 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1333 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1334 * value into swapfile.c, the only way we can correctly account for a
1335 * fallocated page arriving here is now to initialize it and write it.
1337 * That's okay for a page already fallocated earlier, but if we have
1338 * not yet completed the fallocation, then (a) we want to keep track
1339 * of this page in case we have to undo it, and (b) it may not be a
1340 * good idea to continue anyway, once we're pushing into swap. So
1341 * reactivate the page, and let shmem_fallocate() quit when too many.
1343 if (!PageUptodate(page
)) {
1344 if (inode
->i_private
) {
1345 struct shmem_falloc
*shmem_falloc
;
1346 spin_lock(&inode
->i_lock
);
1347 shmem_falloc
= inode
->i_private
;
1349 !shmem_falloc
->waitq
&&
1350 index
>= shmem_falloc
->start
&&
1351 index
< shmem_falloc
->next
)
1352 shmem_falloc
->nr_unswapped
++;
1354 shmem_falloc
= NULL
;
1355 spin_unlock(&inode
->i_lock
);
1359 clear_highpage(page
);
1360 flush_dcache_page(page
);
1361 SetPageUptodate(page
);
1364 swap
= get_swap_page(page
);
1369 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1370 * if it's not already there. Do it now before the page is
1371 * moved to swap cache, when its pagelock no longer protects
1372 * the inode from eviction. But don't unlock the mutex until
1373 * we've incremented swapped, because shmem_unuse_inode() will
1374 * prune a !swapped inode from the swaplist under this mutex.
1376 mutex_lock(&shmem_swaplist_mutex
);
1377 if (list_empty(&info
->swaplist
))
1378 list_add(&info
->swaplist
, &shmem_swaplist
);
1380 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1381 spin_lock_irq(&info
->lock
);
1382 shmem_recalc_inode(inode
);
1384 spin_unlock_irq(&info
->lock
);
1386 swap_shmem_alloc(swap
);
1387 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1389 mutex_unlock(&shmem_swaplist_mutex
);
1390 BUG_ON(page_mapped(page
));
1391 swap_writepage(page
, wbc
);
1395 mutex_unlock(&shmem_swaplist_mutex
);
1396 put_swap_page(page
, swap
);
1398 set_page_dirty(page
);
1399 if (wbc
->for_reclaim
)
1400 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1405 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1406 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1410 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1411 return; /* show nothing */
1413 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1415 seq_printf(seq
, ",mpol=%s", buffer
);
1418 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1420 struct mempolicy
*mpol
= NULL
;
1422 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1423 mpol
= sbinfo
->mpol
;
1425 spin_unlock(&sbinfo
->stat_lock
);
1429 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1430 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1433 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1437 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1439 #define vm_policy vm_private_data
1442 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1443 struct shmem_inode_info
*info
, pgoff_t index
)
1445 /* Create a pseudo vma that just contains the policy */
1446 vma_init(vma
, NULL
);
1447 /* Bias interleave by inode number to distribute better across nodes */
1448 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1449 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1452 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1454 /* Drop reference taken by mpol_shared_policy_lookup() */
1455 mpol_cond_put(vma
->vm_policy
);
1458 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1459 struct shmem_inode_info
*info
, pgoff_t index
)
1461 struct vm_area_struct pvma
;
1463 struct vm_fault vmf
;
1465 shmem_pseudo_vma_init(&pvma
, info
, index
);
1468 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1469 shmem_pseudo_vma_destroy(&pvma
);
1474 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1475 struct shmem_inode_info
*info
, pgoff_t index
)
1477 struct vm_area_struct pvma
;
1478 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1482 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1485 hindex
= round_down(index
, HPAGE_PMD_NR
);
1486 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1490 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1491 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1492 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1493 shmem_pseudo_vma_destroy(&pvma
);
1495 prep_transhuge_page(page
);
1499 static struct page
*shmem_alloc_page(gfp_t gfp
,
1500 struct shmem_inode_info
*info
, pgoff_t index
)
1502 struct vm_area_struct pvma
;
1505 shmem_pseudo_vma_init(&pvma
, info
, index
);
1506 page
= alloc_page_vma(gfp
, &pvma
, 0);
1507 shmem_pseudo_vma_destroy(&pvma
);
1512 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1513 struct inode
*inode
,
1514 pgoff_t index
, bool huge
)
1516 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1521 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
1523 nr
= huge
? HPAGE_PMD_NR
: 1;
1525 if (!shmem_inode_acct_block(inode
, nr
))
1529 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1531 page
= shmem_alloc_page(gfp
, info
, index
);
1533 __SetPageLocked(page
);
1534 __SetPageSwapBacked(page
);
1539 shmem_inode_unacct_blocks(inode
, nr
);
1541 return ERR_PTR(err
);
1545 * When a page is moved from swapcache to shmem filecache (either by the
1546 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1547 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1548 * ignorance of the mapping it belongs to. If that mapping has special
1549 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1550 * we may need to copy to a suitable page before moving to filecache.
1552 * In a future release, this may well be extended to respect cpuset and
1553 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1554 * but for now it is a simple matter of zone.
1556 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1558 return page_zonenum(page
) > gfp_zone(gfp
);
1561 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1562 struct shmem_inode_info
*info
, pgoff_t index
)
1564 struct page
*oldpage
, *newpage
;
1565 struct address_space
*swap_mapping
;
1571 entry
.val
= page_private(oldpage
);
1572 swap_index
= swp_offset(entry
);
1573 swap_mapping
= page_mapping(oldpage
);
1576 * We have arrived here because our zones are constrained, so don't
1577 * limit chance of success by further cpuset and node constraints.
1579 gfp
&= ~GFP_CONSTRAINT_MASK
;
1580 newpage
= shmem_alloc_page(gfp
, info
, index
);
1585 copy_highpage(newpage
, oldpage
);
1586 flush_dcache_page(newpage
);
1588 __SetPageLocked(newpage
);
1589 __SetPageSwapBacked(newpage
);
1590 SetPageUptodate(newpage
);
1591 set_page_private(newpage
, entry
.val
);
1592 SetPageSwapCache(newpage
);
1595 * Our caller will very soon move newpage out of swapcache, but it's
1596 * a nice clean interface for us to replace oldpage by newpage there.
1598 xa_lock_irq(&swap_mapping
->i_pages
);
1599 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1601 __inc_node_page_state(newpage
, NR_FILE_PAGES
);
1602 __dec_node_page_state(oldpage
, NR_FILE_PAGES
);
1604 xa_unlock_irq(&swap_mapping
->i_pages
);
1606 if (unlikely(error
)) {
1608 * Is this possible? I think not, now that our callers check
1609 * both PageSwapCache and page_private after getting page lock;
1610 * but be defensive. Reverse old to newpage for clear and free.
1614 mem_cgroup_migrate(oldpage
, newpage
);
1615 lru_cache_add_anon(newpage
);
1619 ClearPageSwapCache(oldpage
);
1620 set_page_private(oldpage
, 0);
1622 unlock_page(oldpage
);
1629 * Swap in the page pointed to by *pagep.
1630 * Caller has to make sure that *pagep contains a valid swapped page.
1631 * Returns 0 and the page in pagep if success. On failure, returns the
1632 * the error code and NULL in *pagep.
1634 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1635 struct page
**pagep
, enum sgp_type sgp
,
1636 gfp_t gfp
, struct vm_area_struct
*vma
,
1637 vm_fault_t
*fault_type
)
1639 struct address_space
*mapping
= inode
->i_mapping
;
1640 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1641 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1642 struct mem_cgroup
*memcg
;
1647 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1648 swap
= radix_to_swp_entry(*pagep
);
1651 /* Look it up and read it in.. */
1652 page
= lookup_swap_cache(swap
, NULL
, 0);
1654 /* Or update major stats only when swapin succeeds?? */
1656 *fault_type
|= VM_FAULT_MAJOR
;
1657 count_vm_event(PGMAJFAULT
);
1658 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1660 /* Here we actually start the io */
1661 page
= shmem_swapin(swap
, gfp
, info
, index
);
1668 /* We have to do this with page locked to prevent races */
1670 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1671 !shmem_confirm_swap(mapping
, index
, swap
)) {
1675 if (!PageUptodate(page
)) {
1679 wait_on_page_writeback(page
);
1681 if (shmem_should_replace_page(page
, gfp
)) {
1682 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1687 error
= mem_cgroup_try_charge_delay(page
, charge_mm
, gfp
, &memcg
,
1690 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1691 swp_to_radix_entry(swap
), gfp
);
1693 * We already confirmed swap under page lock, and make
1694 * no memory allocation here, so usually no possibility
1695 * of error; but free_swap_and_cache() only trylocks a
1696 * page, so it is just possible that the entry has been
1697 * truncated or holepunched since swap was confirmed.
1698 * shmem_undo_range() will have done some of the
1699 * unaccounting, now delete_from_swap_cache() will do
1703 mem_cgroup_cancel_charge(page
, memcg
, false);
1704 delete_from_swap_cache(page
);
1710 mem_cgroup_commit_charge(page
, memcg
, true, false);
1712 spin_lock_irq(&info
->lock
);
1714 shmem_recalc_inode(inode
);
1715 spin_unlock_irq(&info
->lock
);
1717 if (sgp
== SGP_WRITE
)
1718 mark_page_accessed(page
);
1720 delete_from_swap_cache(page
);
1721 set_page_dirty(page
);
1727 if (!shmem_confirm_swap(mapping
, index
, swap
))
1739 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1741 * If we allocate a new one we do not mark it dirty. That's up to the
1742 * vm. If we swap it in we mark it dirty since we also free the swap
1743 * entry since a page cannot live in both the swap and page cache.
1745 * vmf and fault_type are only supplied by shmem_fault:
1746 * otherwise they are NULL.
1748 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1749 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1750 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1751 vm_fault_t
*fault_type
)
1753 struct address_space
*mapping
= inode
->i_mapping
;
1754 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1755 struct shmem_sb_info
*sbinfo
;
1756 struct mm_struct
*charge_mm
;
1757 struct mem_cgroup
*memcg
;
1759 enum sgp_type sgp_huge
= sgp
;
1760 pgoff_t hindex
= index
;
1765 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1767 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1770 if (sgp
<= SGP_CACHE
&&
1771 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1775 sbinfo
= SHMEM_SB(inode
->i_sb
);
1776 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1778 page
= find_lock_entry(mapping
, index
);
1779 if (xa_is_value(page
)) {
1780 error
= shmem_swapin_page(inode
, index
, &page
,
1781 sgp
, gfp
, vma
, fault_type
);
1782 if (error
== -EEXIST
)
1789 if (page
&& sgp
== SGP_WRITE
)
1790 mark_page_accessed(page
);
1792 /* fallocated page? */
1793 if (page
&& !PageUptodate(page
)) {
1794 if (sgp
!= SGP_READ
)
1800 if (page
|| sgp
== SGP_READ
) {
1806 * Fast cache lookup did not find it:
1807 * bring it back from swap or allocate.
1810 if (vma
&& userfaultfd_missing(vma
)) {
1811 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1815 /* shmem_symlink() */
1816 if (mapping
->a_ops
!= &shmem_aops
)
1818 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1820 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1822 switch (sbinfo
->huge
) {
1825 case SHMEM_HUGE_NEVER
:
1827 case SHMEM_HUGE_WITHIN_SIZE
:
1828 off
= round_up(index
, HPAGE_PMD_NR
);
1829 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1830 if (i_size
>= HPAGE_PMD_SIZE
&&
1831 i_size
>> PAGE_SHIFT
>= off
)
1834 case SHMEM_HUGE_ADVISE
:
1835 if (sgp_huge
== SGP_HUGE
)
1837 /* TODO: implement fadvise() hints */
1842 page
= shmem_alloc_and_acct_page(gfp
, inode
, index
, true);
1845 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1851 error
= PTR_ERR(page
);
1853 if (error
!= -ENOSPC
)
1856 * Try to reclaim some space by splitting a huge page
1857 * beyond i_size on the filesystem.
1862 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1863 if (ret
== SHRINK_STOP
)
1871 if (PageTransHuge(page
))
1872 hindex
= round_down(index
, HPAGE_PMD_NR
);
1876 if (sgp
== SGP_WRITE
)
1877 __SetPageReferenced(page
);
1879 error
= mem_cgroup_try_charge_delay(page
, charge_mm
, gfp
, &memcg
,
1880 PageTransHuge(page
));
1883 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1884 NULL
, gfp
& GFP_RECLAIM_MASK
);
1886 mem_cgroup_cancel_charge(page
, memcg
,
1887 PageTransHuge(page
));
1890 mem_cgroup_commit_charge(page
, memcg
, false,
1891 PageTransHuge(page
));
1892 lru_cache_add_anon(page
);
1894 spin_lock_irq(&info
->lock
);
1895 info
->alloced
+= compound_nr(page
);
1896 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1897 shmem_recalc_inode(inode
);
1898 spin_unlock_irq(&info
->lock
);
1901 if (PageTransHuge(page
) &&
1902 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1903 hindex
+ HPAGE_PMD_NR
- 1) {
1905 * Part of the huge page is beyond i_size: subject
1906 * to shrink under memory pressure.
1908 spin_lock(&sbinfo
->shrinklist_lock
);
1910 * _careful to defend against unlocked access to
1911 * ->shrink_list in shmem_unused_huge_shrink()
1913 if (list_empty_careful(&info
->shrinklist
)) {
1914 list_add_tail(&info
->shrinklist
,
1915 &sbinfo
->shrinklist
);
1916 sbinfo
->shrinklist_len
++;
1918 spin_unlock(&sbinfo
->shrinklist_lock
);
1922 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1924 if (sgp
== SGP_FALLOC
)
1928 * Let SGP_WRITE caller clear ends if write does not fill page;
1929 * but SGP_FALLOC on a page fallocated earlier must initialize
1930 * it now, lest undo on failure cancel our earlier guarantee.
1932 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1933 struct page
*head
= compound_head(page
);
1936 for (i
= 0; i
< compound_nr(head
); i
++) {
1937 clear_highpage(head
+ i
);
1938 flush_dcache_page(head
+ i
);
1940 SetPageUptodate(head
);
1943 /* Perhaps the file has been truncated since we checked */
1944 if (sgp
<= SGP_CACHE
&&
1945 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1947 ClearPageDirty(page
);
1948 delete_from_page_cache(page
);
1949 spin_lock_irq(&info
->lock
);
1950 shmem_recalc_inode(inode
);
1951 spin_unlock_irq(&info
->lock
);
1956 *pagep
= page
+ index
- hindex
;
1963 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
1965 if (PageTransHuge(page
)) {
1975 if (error
== -ENOSPC
&& !once
++) {
1976 spin_lock_irq(&info
->lock
);
1977 shmem_recalc_inode(inode
);
1978 spin_unlock_irq(&info
->lock
);
1981 if (error
== -EEXIST
)
1987 * This is like autoremove_wake_function, but it removes the wait queue
1988 * entry unconditionally - even if something else had already woken the
1991 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
1993 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1994 list_del_init(&wait
->entry
);
1998 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2000 struct vm_area_struct
*vma
= vmf
->vma
;
2001 struct inode
*inode
= file_inode(vma
->vm_file
);
2002 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2005 vm_fault_t ret
= VM_FAULT_LOCKED
;
2008 * Trinity finds that probing a hole which tmpfs is punching can
2009 * prevent the hole-punch from ever completing: which in turn
2010 * locks writers out with its hold on i_mutex. So refrain from
2011 * faulting pages into the hole while it's being punched. Although
2012 * shmem_undo_range() does remove the additions, it may be unable to
2013 * keep up, as each new page needs its own unmap_mapping_range() call,
2014 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2016 * It does not matter if we sometimes reach this check just before the
2017 * hole-punch begins, so that one fault then races with the punch:
2018 * we just need to make racing faults a rare case.
2020 * The implementation below would be much simpler if we just used a
2021 * standard mutex or completion: but we cannot take i_mutex in fault,
2022 * and bloating every shmem inode for this unlikely case would be sad.
2024 if (unlikely(inode
->i_private
)) {
2025 struct shmem_falloc
*shmem_falloc
;
2027 spin_lock(&inode
->i_lock
);
2028 shmem_falloc
= inode
->i_private
;
2030 shmem_falloc
->waitq
&&
2031 vmf
->pgoff
>= shmem_falloc
->start
&&
2032 vmf
->pgoff
< shmem_falloc
->next
) {
2034 wait_queue_head_t
*shmem_falloc_waitq
;
2035 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2037 ret
= VM_FAULT_NOPAGE
;
2038 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2040 ret
= VM_FAULT_RETRY
;
2042 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2043 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2044 TASK_UNINTERRUPTIBLE
);
2045 spin_unlock(&inode
->i_lock
);
2049 * shmem_falloc_waitq points into the shmem_fallocate()
2050 * stack of the hole-punching task: shmem_falloc_waitq
2051 * is usually invalid by the time we reach here, but
2052 * finish_wait() does not dereference it in that case;
2053 * though i_lock needed lest racing with wake_up_all().
2055 spin_lock(&inode
->i_lock
);
2056 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2057 spin_unlock(&inode
->i_lock
);
2063 spin_unlock(&inode
->i_lock
);
2068 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2069 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2071 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2074 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2075 gfp
, vma
, vmf
, &ret
);
2077 return vmf_error(err
);
2081 unsigned long shmem_get_unmapped_area(struct file
*file
,
2082 unsigned long uaddr
, unsigned long len
,
2083 unsigned long pgoff
, unsigned long flags
)
2085 unsigned long (*get_area
)(struct file
*,
2086 unsigned long, unsigned long, unsigned long, unsigned long);
2088 unsigned long offset
;
2089 unsigned long inflated_len
;
2090 unsigned long inflated_addr
;
2091 unsigned long inflated_offset
;
2093 if (len
> TASK_SIZE
)
2096 get_area
= current
->mm
->get_unmapped_area
;
2097 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2099 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
))
2101 if (IS_ERR_VALUE(addr
))
2103 if (addr
& ~PAGE_MASK
)
2105 if (addr
> TASK_SIZE
- len
)
2108 if (shmem_huge
== SHMEM_HUGE_DENY
)
2110 if (len
< HPAGE_PMD_SIZE
)
2112 if (flags
& MAP_FIXED
)
2115 * Our priority is to support MAP_SHARED mapped hugely;
2116 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2117 * But if caller specified an address hint and we allocated area there
2118 * successfully, respect that as before.
2123 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2124 struct super_block
*sb
;
2127 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2128 sb
= file_inode(file
)->i_sb
;
2131 * Called directly from mm/mmap.c, or drivers/char/mem.c
2132 * for "/dev/zero", to create a shared anonymous object.
2134 if (IS_ERR(shm_mnt
))
2136 sb
= shm_mnt
->mnt_sb
;
2138 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2142 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2143 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2145 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2148 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2149 if (inflated_len
> TASK_SIZE
)
2151 if (inflated_len
< len
)
2154 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2155 if (IS_ERR_VALUE(inflated_addr
))
2157 if (inflated_addr
& ~PAGE_MASK
)
2160 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2161 inflated_addr
+= offset
- inflated_offset
;
2162 if (inflated_offset
> offset
)
2163 inflated_addr
+= HPAGE_PMD_SIZE
;
2165 if (inflated_addr
> TASK_SIZE
- len
)
2167 return inflated_addr
;
2171 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2173 struct inode
*inode
= file_inode(vma
->vm_file
);
2174 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2177 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2180 struct inode
*inode
= file_inode(vma
->vm_file
);
2183 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2184 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2188 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2190 struct inode
*inode
= file_inode(file
);
2191 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2192 int retval
= -ENOMEM
;
2195 * What serializes the accesses to info->flags?
2196 * ipc_lock_object() when called from shmctl_do_lock(),
2197 * no serialization needed when called from shm_destroy().
2199 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2200 if (!user_shm_lock(inode
->i_size
, user
))
2202 info
->flags
|= VM_LOCKED
;
2203 mapping_set_unevictable(file
->f_mapping
);
2205 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2206 user_shm_unlock(inode
->i_size
, user
);
2207 info
->flags
&= ~VM_LOCKED
;
2208 mapping_clear_unevictable(file
->f_mapping
);
2216 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2218 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2221 ret
= seal_check_future_write(info
->seals
, vma
);
2225 file_accessed(file
);
2226 vma
->vm_ops
= &shmem_vm_ops
;
2227 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
2228 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2229 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2230 khugepaged_enter(vma
, vma
->vm_flags
);
2235 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2236 umode_t mode
, dev_t dev
, unsigned long flags
)
2238 struct inode
*inode
;
2239 struct shmem_inode_info
*info
;
2240 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2243 if (shmem_reserve_inode(sb
))
2246 inode
= new_inode(sb
);
2248 inode_init_owner(inode
, dir
, mode
);
2249 inode
->i_blocks
= 0;
2250 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2251 inode
->i_generation
= prandom_u32();
2252 info
= SHMEM_I(inode
);
2253 memset(info
, 0, (char *)inode
- (char *)info
);
2254 spin_lock_init(&info
->lock
);
2255 atomic_set(&info
->stop_eviction
, 0);
2256 info
->seals
= F_SEAL_SEAL
;
2257 info
->flags
= flags
& VM_NORESERVE
;
2258 INIT_LIST_HEAD(&info
->shrinklist
);
2259 INIT_LIST_HEAD(&info
->swaplist
);
2260 simple_xattrs_init(&info
->xattrs
);
2261 cache_no_acl(inode
);
2263 switch (mode
& S_IFMT
) {
2265 inode
->i_op
= &shmem_special_inode_operations
;
2266 init_special_inode(inode
, mode
, dev
);
2269 inode
->i_mapping
->a_ops
= &shmem_aops
;
2270 inode
->i_op
= &shmem_inode_operations
;
2271 inode
->i_fop
= &shmem_file_operations
;
2272 mpol_shared_policy_init(&info
->policy
,
2273 shmem_get_sbmpol(sbinfo
));
2277 /* Some things misbehave if size == 0 on a directory */
2278 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2279 inode
->i_op
= &shmem_dir_inode_operations
;
2280 inode
->i_fop
= &simple_dir_operations
;
2284 * Must not load anything in the rbtree,
2285 * mpol_free_shared_policy will not be called.
2287 mpol_shared_policy_init(&info
->policy
, NULL
);
2291 if (!sbinfo
->idr_nouse
) {
2292 /* inum 0 and 1 are unused */
2293 mutex_lock(&sbinfo
->idr_lock
);
2294 ino
= idr_alloc(&sbinfo
->idr
, inode
, 2, INT_MAX
,
2298 mutex_unlock(&sbinfo
->idr_lock
);
2299 __insert_inode_hash(inode
, inode
->i_ino
);
2302 mutex_unlock(&sbinfo
->idr_lock
);
2304 /* shmem_free_inode() will be called */
2308 inode
->i_ino
= get_next_ino();
2310 lockdep_annotate_inode_mutex_key(inode
);
2312 shmem_free_inode(sb
);
2316 bool shmem_mapping(struct address_space
*mapping
)
2318 return mapping
->a_ops
== &shmem_aops
;
2321 static int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2323 struct vm_area_struct
*dst_vma
,
2324 unsigned long dst_addr
,
2325 unsigned long src_addr
,
2327 struct page
**pagep
)
2329 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2330 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2331 struct address_space
*mapping
= inode
->i_mapping
;
2332 gfp_t gfp
= mapping_gfp_mask(mapping
);
2333 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2334 struct mem_cgroup
*memcg
;
2338 pte_t _dst_pte
, *dst_pte
;
2340 pgoff_t offset
, max_off
;
2343 if (!shmem_inode_acct_block(inode
, 1)) {
2345 * We may have got a page, returned -ENOENT triggering a retry,
2346 * and now we find ourselves with -ENOMEM. Release the page, to
2347 * avoid a BUG_ON in our caller.
2349 if (unlikely(*pagep
)) {
2357 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2359 goto out_unacct_blocks
;
2361 if (!zeropage
) { /* mcopy_atomic */
2362 page_kaddr
= kmap_atomic(page
);
2363 ret
= copy_from_user(page_kaddr
,
2364 (const void __user
*)src_addr
,
2366 kunmap_atomic(page_kaddr
);
2368 /* fallback to copy_from_user outside mmap_sem */
2369 if (unlikely(ret
)) {
2371 shmem_inode_unacct_blocks(inode
, 1);
2372 /* don't free the page */
2375 } else { /* mfill_zeropage_atomic */
2376 clear_highpage(page
);
2383 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2384 __SetPageLocked(page
);
2385 __SetPageSwapBacked(page
);
2386 __SetPageUptodate(page
);
2389 offset
= linear_page_index(dst_vma
, dst_addr
);
2390 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2391 if (unlikely(offset
>= max_off
))
2394 ret
= mem_cgroup_try_charge_delay(page
, dst_mm
, gfp
, &memcg
, false);
2398 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2399 gfp
& GFP_RECLAIM_MASK
);
2401 goto out_release_uncharge
;
2403 mem_cgroup_commit_charge(page
, memcg
, false, false);
2405 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2406 if (dst_vma
->vm_flags
& VM_WRITE
)
2407 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2410 * We don't set the pte dirty if the vma has no
2411 * VM_WRITE permission, so mark the page dirty or it
2412 * could be freed from under us. We could do it
2413 * unconditionally before unlock_page(), but doing it
2414 * only if VM_WRITE is not set is faster.
2416 set_page_dirty(page
);
2419 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2422 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2423 if (unlikely(offset
>= max_off
))
2424 goto out_release_uncharge_unlock
;
2427 if (!pte_none(*dst_pte
))
2428 goto out_release_uncharge_unlock
;
2430 lru_cache_add_anon(page
);
2432 spin_lock_irq(&info
->lock
);
2434 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2435 shmem_recalc_inode(inode
);
2436 spin_unlock_irq(&info
->lock
);
2438 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2439 page_add_file_rmap(page
, false);
2440 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2442 /* No need to invalidate - it was non-present before */
2443 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2444 pte_unmap_unlock(dst_pte
, ptl
);
2449 out_release_uncharge_unlock
:
2450 pte_unmap_unlock(dst_pte
, ptl
);
2451 ClearPageDirty(page
);
2452 delete_from_page_cache(page
);
2453 out_release_uncharge
:
2454 mem_cgroup_cancel_charge(page
, memcg
, false);
2459 shmem_inode_unacct_blocks(inode
, 1);
2463 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2465 struct vm_area_struct
*dst_vma
,
2466 unsigned long dst_addr
,
2467 unsigned long src_addr
,
2468 struct page
**pagep
)
2470 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2471 dst_addr
, src_addr
, false, pagep
);
2474 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2476 struct vm_area_struct
*dst_vma
,
2477 unsigned long dst_addr
)
2479 struct page
*page
= NULL
;
2481 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2482 dst_addr
, 0, true, &page
);
2486 static const struct inode_operations shmem_symlink_inode_operations
;
2487 static const struct inode_operations shmem_short_symlink_operations
;
2489 #ifdef CONFIG_TMPFS_XATTR
2490 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2492 #define shmem_initxattrs NULL
2496 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2497 loff_t pos
, unsigned len
, unsigned flags
,
2498 struct page
**pagep
, void **fsdata
)
2500 struct inode
*inode
= mapping
->host
;
2501 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2502 pgoff_t index
= pos
>> PAGE_SHIFT
;
2504 /* i_mutex is held by caller */
2505 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2506 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2507 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2509 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2513 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2517 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2518 loff_t pos
, unsigned len
, unsigned copied
,
2519 struct page
*page
, void *fsdata
)
2521 struct inode
*inode
= mapping
->host
;
2523 if (pos
+ copied
> inode
->i_size
)
2524 i_size_write(inode
, pos
+ copied
);
2526 if (!PageUptodate(page
)) {
2527 struct page
*head
= compound_head(page
);
2528 if (PageTransCompound(page
)) {
2531 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2532 if (head
+ i
== page
)
2534 clear_highpage(head
+ i
);
2535 flush_dcache_page(head
+ i
);
2538 if (copied
< PAGE_SIZE
) {
2539 unsigned from
= pos
& (PAGE_SIZE
- 1);
2540 zero_user_segments(page
, 0, from
,
2541 from
+ copied
, PAGE_SIZE
);
2543 SetPageUptodate(head
);
2545 set_page_dirty(page
);
2552 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2554 struct file
*file
= iocb
->ki_filp
;
2555 struct inode
*inode
= file_inode(file
);
2556 struct address_space
*mapping
= inode
->i_mapping
;
2558 unsigned long offset
;
2559 enum sgp_type sgp
= SGP_READ
;
2562 loff_t
*ppos
= &iocb
->ki_pos
;
2565 * Might this read be for a stacking filesystem? Then when reading
2566 * holes of a sparse file, we actually need to allocate those pages,
2567 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2569 if (!iter_is_iovec(to
))
2572 index
= *ppos
>> PAGE_SHIFT
;
2573 offset
= *ppos
& ~PAGE_MASK
;
2576 struct page
*page
= NULL
;
2578 unsigned long nr
, ret
;
2579 loff_t i_size
= i_size_read(inode
);
2581 end_index
= i_size
>> PAGE_SHIFT
;
2582 if (index
> end_index
)
2584 if (index
== end_index
) {
2585 nr
= i_size
& ~PAGE_MASK
;
2590 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2592 if (error
== -EINVAL
)
2597 if (sgp
== SGP_CACHE
)
2598 set_page_dirty(page
);
2603 * We must evaluate after, since reads (unlike writes)
2604 * are called without i_mutex protection against truncate
2607 i_size
= i_size_read(inode
);
2608 end_index
= i_size
>> PAGE_SHIFT
;
2609 if (index
== end_index
) {
2610 nr
= i_size
& ~PAGE_MASK
;
2621 * If users can be writing to this page using arbitrary
2622 * virtual addresses, take care about potential aliasing
2623 * before reading the page on the kernel side.
2625 if (mapping_writably_mapped(mapping
))
2626 flush_dcache_page(page
);
2628 * Mark the page accessed if we read the beginning.
2631 mark_page_accessed(page
);
2633 page
= ZERO_PAGE(0);
2638 * Ok, we have the page, and it's up-to-date, so
2639 * now we can copy it to user space...
2641 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2644 index
+= offset
>> PAGE_SHIFT
;
2645 offset
&= ~PAGE_MASK
;
2648 if (!iov_iter_count(to
))
2657 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2658 file_accessed(file
);
2659 return retval
? retval
: error
;
2663 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2665 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2666 pgoff_t index
, pgoff_t end
, int whence
)
2669 struct pagevec pvec
;
2670 pgoff_t indices
[PAGEVEC_SIZE
];
2674 pagevec_init(&pvec
);
2675 pvec
.nr
= 1; /* start small: we may be there already */
2677 pvec
.nr
= find_get_entries(mapping
, index
,
2678 pvec
.nr
, pvec
.pages
, indices
);
2680 if (whence
== SEEK_DATA
)
2684 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2685 if (index
< indices
[i
]) {
2686 if (whence
== SEEK_HOLE
) {
2692 page
= pvec
.pages
[i
];
2693 if (page
&& !xa_is_value(page
)) {
2694 if (!PageUptodate(page
))
2698 (page
&& whence
== SEEK_DATA
) ||
2699 (!page
&& whence
== SEEK_HOLE
)) {
2704 pagevec_remove_exceptionals(&pvec
);
2705 pagevec_release(&pvec
);
2706 pvec
.nr
= PAGEVEC_SIZE
;
2712 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2714 struct address_space
*mapping
= file
->f_mapping
;
2715 struct inode
*inode
= mapping
->host
;
2719 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2720 return generic_file_llseek_size(file
, offset
, whence
,
2721 MAX_LFS_FILESIZE
, i_size_read(inode
));
2723 /* We're holding i_mutex so we can access i_size directly */
2725 if (offset
< 0 || offset
>= inode
->i_size
)
2728 start
= offset
>> PAGE_SHIFT
;
2729 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2730 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2731 new_offset
<<= PAGE_SHIFT
;
2732 if (new_offset
> offset
) {
2733 if (new_offset
< inode
->i_size
)
2734 offset
= new_offset
;
2735 else if (whence
== SEEK_DATA
)
2738 offset
= inode
->i_size
;
2743 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2744 inode_unlock(inode
);
2748 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2751 struct inode
*inode
= file_inode(file
);
2752 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2753 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2754 struct shmem_falloc shmem_falloc
;
2755 pgoff_t start
, index
, end
;
2758 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2763 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2764 struct address_space
*mapping
= file
->f_mapping
;
2765 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2766 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2767 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2769 /* protected by i_mutex */
2770 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2775 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2776 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2777 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2778 spin_lock(&inode
->i_lock
);
2779 inode
->i_private
= &shmem_falloc
;
2780 spin_unlock(&inode
->i_lock
);
2782 if ((u64
)unmap_end
> (u64
)unmap_start
)
2783 unmap_mapping_range(mapping
, unmap_start
,
2784 1 + unmap_end
- unmap_start
, 0);
2785 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2786 /* No need to unmap again: hole-punching leaves COWed pages */
2788 spin_lock(&inode
->i_lock
);
2789 inode
->i_private
= NULL
;
2790 wake_up_all(&shmem_falloc_waitq
);
2791 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2792 spin_unlock(&inode
->i_lock
);
2797 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2798 error
= inode_newsize_ok(inode
, offset
+ len
);
2802 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2807 start
= offset
>> PAGE_SHIFT
;
2808 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2809 /* Try to avoid a swapstorm if len is impossible to satisfy */
2810 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2815 shmem_falloc
.waitq
= NULL
;
2816 shmem_falloc
.start
= start
;
2817 shmem_falloc
.next
= start
;
2818 shmem_falloc
.nr_falloced
= 0;
2819 shmem_falloc
.nr_unswapped
= 0;
2820 spin_lock(&inode
->i_lock
);
2821 inode
->i_private
= &shmem_falloc
;
2822 spin_unlock(&inode
->i_lock
);
2824 for (index
= start
; index
< end
; index
++) {
2828 * Good, the fallocate(2) manpage permits EINTR: we may have
2829 * been interrupted because we are using up too much memory.
2831 if (signal_pending(current
))
2833 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2836 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2838 /* Remove the !PageUptodate pages we added */
2839 if (index
> start
) {
2840 shmem_undo_range(inode
,
2841 (loff_t
)start
<< PAGE_SHIFT
,
2842 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2848 * Inform shmem_writepage() how far we have reached.
2849 * No need for lock or barrier: we have the page lock.
2851 shmem_falloc
.next
++;
2852 if (!PageUptodate(page
))
2853 shmem_falloc
.nr_falloced
++;
2856 * If !PageUptodate, leave it that way so that freeable pages
2857 * can be recognized if we need to rollback on error later.
2858 * But set_page_dirty so that memory pressure will swap rather
2859 * than free the pages we are allocating (and SGP_CACHE pages
2860 * might still be clean: we now need to mark those dirty too).
2862 set_page_dirty(page
);
2868 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2869 i_size_write(inode
, offset
+ len
);
2870 inode
->i_ctime
= current_time(inode
);
2872 spin_lock(&inode
->i_lock
);
2873 inode
->i_private
= NULL
;
2874 spin_unlock(&inode
->i_lock
);
2876 inode_unlock(inode
);
2880 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2882 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2884 buf
->f_type
= TMPFS_MAGIC
;
2885 buf
->f_bsize
= PAGE_SIZE
;
2886 buf
->f_namelen
= NAME_MAX
;
2887 if (sbinfo
->max_blocks
) {
2888 buf
->f_blocks
= sbinfo
->max_blocks
;
2890 buf
->f_bfree
= sbinfo
->max_blocks
-
2891 percpu_counter_sum(&sbinfo
->used_blocks
);
2893 if (sbinfo
->max_inodes
) {
2894 buf
->f_files
= sbinfo
->max_inodes
;
2895 buf
->f_ffree
= sbinfo
->free_inodes
;
2897 /* else leave those fields 0 like simple_statfs */
2902 * File creation. Allocate an inode, and we're done..
2905 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2907 struct inode
*inode
;
2908 int error
= -ENOSPC
;
2910 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2912 error
= simple_acl_create(dir
, inode
);
2915 error
= security_inode_init_security(inode
, dir
,
2917 shmem_initxattrs
, NULL
);
2918 if (error
&& error
!= -EOPNOTSUPP
)
2922 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2923 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2924 d_instantiate(dentry
, inode
);
2925 dget(dentry
); /* Extra count - pin the dentry in core */
2934 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2936 struct inode
*inode
;
2937 int error
= -ENOSPC
;
2939 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2941 error
= security_inode_init_security(inode
, dir
,
2943 shmem_initxattrs
, NULL
);
2944 if (error
&& error
!= -EOPNOTSUPP
)
2946 error
= simple_acl_create(dir
, inode
);
2949 d_tmpfile(dentry
, inode
);
2957 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2961 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2967 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2970 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2976 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2978 struct inode
*inode
= d_inode(old_dentry
);
2982 * No ordinary (disk based) filesystem counts links as inodes;
2983 * but each new link needs a new dentry, pinning lowmem, and
2984 * tmpfs dentries cannot be pruned until they are unlinked.
2985 * But if an O_TMPFILE file is linked into the tmpfs, the
2986 * first link must skip that, to get the accounting right.
2988 if (inode
->i_nlink
) {
2989 ret
= shmem_reserve_inode(inode
->i_sb
);
2994 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2995 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2997 ihold(inode
); /* New dentry reference */
2998 dget(dentry
); /* Extra pinning count for the created dentry */
2999 d_instantiate(dentry
, inode
);
3004 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3006 struct inode
*inode
= d_inode(dentry
);
3008 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3009 shmem_free_inode(inode
->i_sb
);
3011 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3012 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3014 dput(dentry
); /* Undo the count from "create" - this does all the work */
3018 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3020 if (!simple_empty(dentry
))
3023 drop_nlink(d_inode(dentry
));
3025 return shmem_unlink(dir
, dentry
);
3028 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3030 bool old_is_dir
= d_is_dir(old_dentry
);
3031 bool new_is_dir
= d_is_dir(new_dentry
);
3033 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3035 drop_nlink(old_dir
);
3038 drop_nlink(new_dir
);
3042 old_dir
->i_ctime
= old_dir
->i_mtime
=
3043 new_dir
->i_ctime
= new_dir
->i_mtime
=
3044 d_inode(old_dentry
)->i_ctime
=
3045 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3050 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3052 struct dentry
*whiteout
;
3055 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3059 error
= shmem_mknod(old_dir
, whiteout
,
3060 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3066 * Cheat and hash the whiteout while the old dentry is still in
3067 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3069 * d_lookup() will consistently find one of them at this point,
3070 * not sure which one, but that isn't even important.
3077 * The VFS layer already does all the dentry stuff for rename,
3078 * we just have to decrement the usage count for the target if
3079 * it exists so that the VFS layer correctly free's it when it
3082 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3084 struct inode
*inode
= d_inode(old_dentry
);
3085 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3087 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3090 if (flags
& RENAME_EXCHANGE
)
3091 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3093 if (!simple_empty(new_dentry
))
3096 if (flags
& RENAME_WHITEOUT
) {
3099 error
= shmem_whiteout(old_dir
, old_dentry
);
3104 if (d_really_is_positive(new_dentry
)) {
3105 (void) shmem_unlink(new_dir
, new_dentry
);
3106 if (they_are_dirs
) {
3107 drop_nlink(d_inode(new_dentry
));
3108 drop_nlink(old_dir
);
3110 } else if (they_are_dirs
) {
3111 drop_nlink(old_dir
);
3115 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3116 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3117 old_dir
->i_ctime
= old_dir
->i_mtime
=
3118 new_dir
->i_ctime
= new_dir
->i_mtime
=
3119 inode
->i_ctime
= current_time(old_dir
);
3123 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3127 struct inode
*inode
;
3130 len
= strlen(symname
) + 1;
3131 if (len
> PAGE_SIZE
)
3132 return -ENAMETOOLONG
;
3134 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3139 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3140 shmem_initxattrs
, NULL
);
3142 if (error
!= -EOPNOTSUPP
) {
3149 inode
->i_size
= len
-1;
3150 if (len
<= SHORT_SYMLINK_LEN
) {
3151 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3152 if (!inode
->i_link
) {
3156 inode
->i_op
= &shmem_short_symlink_operations
;
3158 inode_nohighmem(inode
);
3159 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3164 inode
->i_mapping
->a_ops
= &shmem_aops
;
3165 inode
->i_op
= &shmem_symlink_inode_operations
;
3166 memcpy(page_address(page
), symname
, len
);
3167 SetPageUptodate(page
);
3168 set_page_dirty(page
);
3172 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3173 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3174 d_instantiate(dentry
, inode
);
3179 static void shmem_put_link(void *arg
)
3181 mark_page_accessed(arg
);
3185 static const char *shmem_get_link(struct dentry
*dentry
,
3186 struct inode
*inode
,
3187 struct delayed_call
*done
)
3189 struct page
*page
= NULL
;
3192 page
= find_get_page(inode
->i_mapping
, 0);
3194 return ERR_PTR(-ECHILD
);
3195 if (!PageUptodate(page
)) {
3197 return ERR_PTR(-ECHILD
);
3200 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3202 return ERR_PTR(error
);
3205 set_delayed_call(done
, shmem_put_link
, page
);
3206 return page_address(page
);
3209 #ifdef CONFIG_TMPFS_XATTR
3211 * Superblocks without xattr inode operations may get some security.* xattr
3212 * support from the LSM "for free". As soon as we have any other xattrs
3213 * like ACLs, we also need to implement the security.* handlers at
3214 * filesystem level, though.
3218 * Callback for security_inode_init_security() for acquiring xattrs.
3220 static int shmem_initxattrs(struct inode
*inode
,
3221 const struct xattr
*xattr_array
,
3224 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3225 const struct xattr
*xattr
;
3226 struct simple_xattr
*new_xattr
;
3229 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3230 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3234 len
= strlen(xattr
->name
) + 1;
3235 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3237 if (!new_xattr
->name
) {
3242 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3243 XATTR_SECURITY_PREFIX_LEN
);
3244 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3247 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3253 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3254 struct dentry
*unused
, struct inode
*inode
,
3255 const char *name
, void *buffer
, size_t size
)
3257 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3259 name
= xattr_full_name(handler
, name
);
3260 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3263 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3264 struct dentry
*unused
, struct inode
*inode
,
3265 const char *name
, const void *value
,
3266 size_t size
, int flags
)
3268 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3270 name
= xattr_full_name(handler
, name
);
3271 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3274 static const struct xattr_handler shmem_security_xattr_handler
= {
3275 .prefix
= XATTR_SECURITY_PREFIX
,
3276 .get
= shmem_xattr_handler_get
,
3277 .set
= shmem_xattr_handler_set
,
3280 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3281 .prefix
= XATTR_TRUSTED_PREFIX
,
3282 .get
= shmem_xattr_handler_get
,
3283 .set
= shmem_xattr_handler_set
,
3286 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3287 #ifdef CONFIG_TMPFS_POSIX_ACL
3288 &posix_acl_access_xattr_handler
,
3289 &posix_acl_default_xattr_handler
,
3291 &shmem_security_xattr_handler
,
3292 &shmem_trusted_xattr_handler
,
3296 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3298 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3299 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3301 #endif /* CONFIG_TMPFS_XATTR */
3303 static const struct inode_operations shmem_short_symlink_operations
= {
3304 .get_link
= simple_get_link
,
3305 #ifdef CONFIG_TMPFS_XATTR
3306 .listxattr
= shmem_listxattr
,
3310 static const struct inode_operations shmem_symlink_inode_operations
= {
3311 .get_link
= shmem_get_link
,
3312 #ifdef CONFIG_TMPFS_XATTR
3313 .listxattr
= shmem_listxattr
,
3317 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3319 return ERR_PTR(-ESTALE
);
3322 static int shmem_match(struct inode
*ino
, void *vfh
)
3326 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3329 /* Find any alias of inode, but prefer a hashed alias */
3330 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3332 struct dentry
*alias
= d_find_alias(inode
);
3334 return alias
?: d_find_any_alias(inode
);
3338 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3339 struct fid
*fid
, int fh_len
, int fh_type
)
3341 struct inode
*inode
;
3342 struct dentry
*dentry
= NULL
;
3349 inode
= ilookup5(sb
, inum
, shmem_match
, fid
->raw
);
3351 dentry
= shmem_find_alias(inode
);
3358 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3359 struct inode
*parent
)
3363 return FILEID_INVALID
;
3366 fh
[0] = inode
->i_generation
;
3367 fh
[1] = inode
->i_ino
;
3373 static const struct export_operations shmem_export_ops
= {
3374 .get_parent
= shmem_get_parent
,
3375 .encode_fh
= shmem_encode_fh
,
3376 .fh_to_dentry
= shmem_fh_to_dentry
,
3390 static const struct fs_parameter_spec shmem_param_specs
[] = {
3391 fsparam_u32 ("gid", Opt_gid
),
3392 fsparam_enum ("huge", Opt_huge
),
3393 fsparam_u32oct("mode", Opt_mode
),
3394 fsparam_string("mpol", Opt_mpol
),
3395 fsparam_string("nr_blocks", Opt_nr_blocks
),
3396 fsparam_string("nr_inodes", Opt_nr_inodes
),
3397 fsparam_string("size", Opt_size
),
3398 fsparam_u32 ("uid", Opt_uid
),
3402 static const struct fs_parameter_enum shmem_param_enums
[] = {
3403 { Opt_huge
, "never", SHMEM_HUGE_NEVER
},
3404 { Opt_huge
, "always", SHMEM_HUGE_ALWAYS
},
3405 { Opt_huge
, "within_size", SHMEM_HUGE_WITHIN_SIZE
},
3406 { Opt_huge
, "advise", SHMEM_HUGE_ADVISE
},
3410 const struct fs_parameter_description shmem_fs_parameters
= {
3412 .specs
= shmem_param_specs
,
3413 .enums
= shmem_param_enums
,
3416 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3418 struct shmem_options
*ctx
= fc
->fs_private
;
3419 struct fs_parse_result result
;
3420 unsigned long long size
;
3424 opt
= fs_parse(fc
, &shmem_fs_parameters
, param
, &result
);
3430 size
= memparse(param
->string
, &rest
);
3432 size
<<= PAGE_SHIFT
;
3433 size
*= totalram_pages();
3439 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3440 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3443 ctx
->blocks
= memparse(param
->string
, &rest
);
3446 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3449 ctx
->inodes
= memparse(param
->string
, &rest
);
3450 if (*rest
|| ctx
->inodes
< 2)
3452 ctx
->seen
|= SHMEM_SEEN_INODES
;
3455 ctx
->mode
= result
.uint_32
& 07777;
3458 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3459 if (!uid_valid(ctx
->uid
))
3463 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3464 if (!gid_valid(ctx
->gid
))
3468 ctx
->huge
= result
.uint_32
;
3469 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3470 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
3471 has_transparent_hugepage()))
3472 goto unsupported_parameter
;
3473 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3476 if (IS_ENABLED(CONFIG_NUMA
)) {
3477 mpol_put(ctx
->mpol
);
3479 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3483 goto unsupported_parameter
;
3487 unsupported_parameter
:
3488 return invalf(fc
, "tmpfs: Unsupported parameter '%s'", param
->key
);
3490 return invalf(fc
, "tmpfs: Bad value for '%s'", param
->key
);
3493 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3495 char *options
= data
;
3498 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3503 while (options
!= NULL
) {
3504 char *this_char
= options
;
3507 * NUL-terminate this option: unfortunately,
3508 * mount options form a comma-separated list,
3509 * but mpol's nodelist may also contain commas.
3511 options
= strchr(options
, ',');
3512 if (options
== NULL
)
3515 if (!isdigit(*options
)) {
3521 char *value
= strchr(this_char
,'=');
3527 len
= strlen(value
);
3529 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3538 * Reconfigure a shmem filesystem.
3540 * Note that we disallow change from limited->unlimited blocks/inodes while any
3541 * are in use; but we must separately disallow unlimited->limited, because in
3542 * that case we have no record of how much is already in use.
3544 static int shmem_reconfigure(struct fs_context
*fc
)
3546 struct shmem_options
*ctx
= fc
->fs_private
;
3547 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3551 spin_lock(&sbinfo
->stat_lock
);
3552 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3553 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3554 if (!sbinfo
->max_blocks
) {
3555 err
= "Cannot retroactively limit size";
3558 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3560 err
= "Too small a size for current use";
3564 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3565 if (!sbinfo
->max_inodes
) {
3566 err
= "Cannot retroactively limit inodes";
3569 if (ctx
->inodes
< inodes
) {
3570 err
= "Too few inodes for current use";
3575 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3576 sbinfo
->huge
= ctx
->huge
;
3577 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3578 sbinfo
->max_blocks
= ctx
->blocks
;
3579 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3580 sbinfo
->max_inodes
= ctx
->inodes
;
3581 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3585 * Preserve previous mempolicy unless mpol remount option was specified.
3588 mpol_put(sbinfo
->mpol
);
3589 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3592 spin_unlock(&sbinfo
->stat_lock
);
3595 spin_unlock(&sbinfo
->stat_lock
);
3596 return invalf(fc
, "tmpfs: %s", err
);
3599 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3601 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3603 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3604 seq_printf(seq
, ",size=%luk",
3605 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3606 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3607 seq_printf(seq
, ",nr_inodes=%d", sbinfo
->max_inodes
);
3608 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3609 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3610 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3611 seq_printf(seq
, ",uid=%u",
3612 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3613 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3614 seq_printf(seq
, ",gid=%u",
3615 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3616 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3617 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3619 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3621 shmem_show_mpol(seq
, sbinfo
->mpol
);
3625 #endif /* CONFIG_TMPFS */
3627 static void shmem_put_super(struct super_block
*sb
)
3629 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3631 if (!sbinfo
->idr_nouse
)
3632 idr_destroy(&sbinfo
->idr
);
3633 percpu_counter_destroy(&sbinfo
->used_blocks
);
3634 mpol_put(sbinfo
->mpol
);
3636 sb
->s_fs_info
= NULL
;
3639 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3641 struct shmem_options
*ctx
= fc
->fs_private
;
3642 struct inode
*inode
;
3643 struct shmem_sb_info
*sbinfo
;
3646 /* Round up to L1_CACHE_BYTES to resist false sharing */
3647 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3648 L1_CACHE_BYTES
), GFP_KERNEL
);
3652 sb
->s_fs_info
= sbinfo
;
3656 * Per default we only allow half of the physical ram per
3657 * tmpfs instance, limiting inodes to one per page of lowmem;
3658 * but the internal instance is left unlimited.
3660 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3661 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3662 ctx
->blocks
= shmem_default_max_blocks();
3663 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3664 ctx
->inodes
= shmem_default_max_inodes();
3666 sb
->s_flags
|= SB_NOUSER
;
3668 sb
->s_export_op
= &shmem_export_ops
;
3669 sb
->s_flags
|= SB_NOSEC
;
3671 sb
->s_flags
|= SB_NOUSER
;
3673 mutex_init(&sbinfo
->idr_lock
);
3674 idr_init(&sbinfo
->idr
);
3675 sbinfo
->max_blocks
= ctx
->blocks
;
3676 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3677 sbinfo
->uid
= ctx
->uid
;
3678 sbinfo
->gid
= ctx
->gid
;
3679 sbinfo
->mode
= ctx
->mode
;
3680 sbinfo
->huge
= ctx
->huge
;
3681 sbinfo
->mpol
= ctx
->mpol
;
3684 spin_lock_init(&sbinfo
->stat_lock
);
3685 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3687 spin_lock_init(&sbinfo
->shrinklist_lock
);
3688 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3690 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3691 sb
->s_blocksize
= PAGE_SIZE
;
3692 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3693 sb
->s_magic
= TMPFS_MAGIC
;
3694 sb
->s_op
= &shmem_ops
;
3695 sb
->s_time_gran
= 1;
3696 #ifdef CONFIG_TMPFS_XATTR
3697 sb
->s_xattr
= shmem_xattr_handlers
;
3699 #ifdef CONFIG_TMPFS_POSIX_ACL
3700 sb
->s_flags
|= SB_POSIXACL
;
3702 uuid_gen(&sb
->s_uuid
);
3704 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3707 inode
->i_uid
= sbinfo
->uid
;
3708 inode
->i_gid
= sbinfo
->gid
;
3709 sb
->s_root
= d_make_root(inode
);
3715 shmem_put_super(sb
);
3719 static int shmem_get_tree(struct fs_context
*fc
)
3721 return get_tree_nodev(fc
, shmem_fill_super
);
3724 static void shmem_free_fc(struct fs_context
*fc
)
3726 struct shmem_options
*ctx
= fc
->fs_private
;
3729 mpol_put(ctx
->mpol
);
3734 static const struct fs_context_operations shmem_fs_context_ops
= {
3735 .free
= shmem_free_fc
,
3736 .get_tree
= shmem_get_tree
,
3738 .parse_monolithic
= shmem_parse_options
,
3739 .parse_param
= shmem_parse_one
,
3740 .reconfigure
= shmem_reconfigure
,
3744 static struct kmem_cache
*shmem_inode_cachep
;
3746 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3748 struct shmem_inode_info
*info
;
3749 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3752 return &info
->vfs_inode
;
3755 static void shmem_free_in_core_inode(struct inode
*inode
)
3757 if (S_ISLNK(inode
->i_mode
))
3758 kfree(inode
->i_link
);
3759 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3762 static void shmem_destroy_inode(struct inode
*inode
)
3764 if (S_ISREG(inode
->i_mode
))
3765 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3768 static void shmem_init_inode(void *foo
)
3770 struct shmem_inode_info
*info
= foo
;
3771 inode_init_once(&info
->vfs_inode
);
3774 static void shmem_init_inodecache(void)
3776 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3777 sizeof(struct shmem_inode_info
),
3778 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3781 static void shmem_destroy_inodecache(void)
3783 kmem_cache_destroy(shmem_inode_cachep
);
3786 static __init
void shmem_no_idr(struct super_block
*sb
)
3788 struct shmem_sb_info
*sbinfo
;
3790 sbinfo
= SHMEM_SB(sb
);
3791 sbinfo
->idr_nouse
= true;
3792 idr_destroy(&sbinfo
->idr
);
3795 static const struct address_space_operations shmem_aops
= {
3796 .writepage
= shmem_writepage
,
3797 .set_page_dirty
= __set_page_dirty_no_writeback
,
3799 .write_begin
= shmem_write_begin
,
3800 .write_end
= shmem_write_end
,
3802 #ifdef CONFIG_MIGRATION
3803 .migratepage
= migrate_page
,
3805 .error_remove_page
= generic_error_remove_page
,
3808 static const struct file_operations shmem_file_operations
= {
3810 .get_unmapped_area
= shmem_get_unmapped_area
,
3812 .llseek
= shmem_file_llseek
,
3813 .read_iter
= shmem_file_read_iter
,
3814 .write_iter
= generic_file_write_iter
,
3815 .fsync
= noop_fsync
,
3816 .splice_read
= generic_file_splice_read
,
3817 .splice_write
= iter_file_splice_write
,
3818 .fallocate
= shmem_fallocate
,
3822 static const struct inode_operations shmem_inode_operations
= {
3823 .getattr
= shmem_getattr
,
3824 .setattr
= shmem_setattr
,
3825 #ifdef CONFIG_TMPFS_XATTR
3826 .listxattr
= shmem_listxattr
,
3827 .set_acl
= simple_set_acl
,
3831 static const struct inode_operations shmem_dir_inode_operations
= {
3833 .create
= shmem_create
,
3834 .lookup
= simple_lookup
,
3836 .unlink
= shmem_unlink
,
3837 .symlink
= shmem_symlink
,
3838 .mkdir
= shmem_mkdir
,
3839 .rmdir
= shmem_rmdir
,
3840 .mknod
= shmem_mknod
,
3841 .rename
= shmem_rename2
,
3842 .tmpfile
= shmem_tmpfile
,
3844 #ifdef CONFIG_TMPFS_XATTR
3845 .listxattr
= shmem_listxattr
,
3847 #ifdef CONFIG_TMPFS_POSIX_ACL
3848 .setattr
= shmem_setattr
,
3849 .set_acl
= simple_set_acl
,
3853 static const struct inode_operations shmem_special_inode_operations
= {
3854 #ifdef CONFIG_TMPFS_XATTR
3855 .listxattr
= shmem_listxattr
,
3857 #ifdef CONFIG_TMPFS_POSIX_ACL
3858 .setattr
= shmem_setattr
,
3859 .set_acl
= simple_set_acl
,
3863 static const struct super_operations shmem_ops
= {
3864 .alloc_inode
= shmem_alloc_inode
,
3865 .free_inode
= shmem_free_in_core_inode
,
3866 .destroy_inode
= shmem_destroy_inode
,
3868 .statfs
= shmem_statfs
,
3869 .show_options
= shmem_show_options
,
3871 .evict_inode
= shmem_evict_inode
,
3872 .drop_inode
= generic_delete_inode
,
3873 .put_super
= shmem_put_super
,
3874 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3875 .nr_cached_objects
= shmem_unused_huge_count
,
3876 .free_cached_objects
= shmem_unused_huge_scan
,
3880 static const struct vm_operations_struct shmem_vm_ops
= {
3881 .fault
= shmem_fault
,
3882 .map_pages
= filemap_map_pages
,
3884 .set_policy
= shmem_set_policy
,
3885 .get_policy
= shmem_get_policy
,
3889 int shmem_init_fs_context(struct fs_context
*fc
)
3891 struct shmem_options
*ctx
;
3893 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3897 ctx
->mode
= 0777 | S_ISVTX
;
3898 ctx
->uid
= current_fsuid();
3899 ctx
->gid
= current_fsgid();
3901 fc
->fs_private
= ctx
;
3902 fc
->ops
= &shmem_fs_context_ops
;
3906 static struct file_system_type shmem_fs_type
= {
3907 .owner
= THIS_MODULE
,
3909 .init_fs_context
= shmem_init_fs_context
,
3911 .parameters
= &shmem_fs_parameters
,
3913 .kill_sb
= kill_litter_super
,
3914 .fs_flags
= FS_USERNS_MOUNT
,
3917 int __init
shmem_init(void)
3921 shmem_init_inodecache();
3923 error
= register_filesystem(&shmem_fs_type
);
3925 pr_err("Could not register tmpfs\n");
3929 shm_mnt
= kern_mount(&shmem_fs_type
);
3930 if (IS_ERR(shm_mnt
)) {
3931 error
= PTR_ERR(shm_mnt
);
3932 pr_err("Could not kern_mount tmpfs\n");
3935 shmem_no_idr(shm_mnt
->mnt_sb
);
3937 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3938 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3939 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3941 shmem_huge
= 0; /* just in case it was patched */
3946 unregister_filesystem(&shmem_fs_type
);
3948 shmem_destroy_inodecache();
3949 shm_mnt
= ERR_PTR(error
);
3953 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3954 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3955 struct kobj_attribute
*attr
, char *buf
)
3959 SHMEM_HUGE_WITHIN_SIZE
,
3967 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3968 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3970 count
+= sprintf(buf
+ count
, fmt
,
3971 shmem_format_huge(values
[i
]));
3973 buf
[count
- 1] = '\n';
3977 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3978 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3983 if (count
+ 1 > sizeof(tmp
))
3985 memcpy(tmp
, buf
, count
);
3987 if (count
&& tmp
[count
- 1] == '\n')
3988 tmp
[count
- 1] = '\0';
3990 huge
= shmem_parse_huge(tmp
);
3991 if (huge
== -EINVAL
)
3993 if (!has_transparent_hugepage() &&
3994 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3998 if (shmem_huge
> SHMEM_HUGE_DENY
)
3999 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4003 struct kobj_attribute shmem_enabled_attr
=
4004 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4005 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4007 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4008 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4010 struct inode
*inode
= file_inode(vma
->vm_file
);
4011 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4015 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
4016 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
4018 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4020 if (shmem_huge
== SHMEM_HUGE_DENY
)
4022 switch (sbinfo
->huge
) {
4023 case SHMEM_HUGE_NEVER
:
4025 case SHMEM_HUGE_ALWAYS
:
4027 case SHMEM_HUGE_WITHIN_SIZE
:
4028 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4029 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4030 if (i_size
>= HPAGE_PMD_SIZE
&&
4031 i_size
>> PAGE_SHIFT
>= off
)
4034 case SHMEM_HUGE_ADVISE
:
4035 /* TODO: implement fadvise() hints */
4036 return (vma
->vm_flags
& VM_HUGEPAGE
);
4042 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4044 #else /* !CONFIG_SHMEM */
4047 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4049 * This is intended for small system where the benefits of the full
4050 * shmem code (swap-backed and resource-limited) are outweighed by
4051 * their complexity. On systems without swap this code should be
4052 * effectively equivalent, but much lighter weight.
4055 static struct file_system_type shmem_fs_type
= {
4057 .init_fs_context
= ramfs_init_fs_context
,
4058 .parameters
= &ramfs_fs_parameters
,
4059 .kill_sb
= kill_litter_super
,
4060 .fs_flags
= FS_USERNS_MOUNT
,
4063 int __init
shmem_init(void)
4065 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4067 shm_mnt
= kern_mount(&shmem_fs_type
);
4068 BUG_ON(IS_ERR(shm_mnt
));
4073 int shmem_unuse(unsigned int type
, bool frontswap
,
4074 unsigned long *fs_pages_to_unuse
)
4079 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4084 void shmem_unlock_mapping(struct address_space
*mapping
)
4089 unsigned long shmem_get_unmapped_area(struct file
*file
,
4090 unsigned long addr
, unsigned long len
,
4091 unsigned long pgoff
, unsigned long flags
)
4093 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4097 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4099 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4101 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4103 #define shmem_vm_ops generic_file_vm_ops
4104 #define shmem_file_operations ramfs_file_operations
4105 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4106 #define shmem_acct_size(flags, size) 0
4107 #define shmem_unacct_size(flags, size) do {} while (0)
4109 #endif /* CONFIG_SHMEM */
4113 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4114 unsigned long flags
, unsigned int i_flags
)
4116 struct inode
*inode
;
4120 return ERR_CAST(mnt
);
4122 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4123 return ERR_PTR(-EINVAL
);
4125 if (shmem_acct_size(flags
, size
))
4126 return ERR_PTR(-ENOMEM
);
4128 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4130 if (unlikely(!inode
)) {
4131 shmem_unacct_size(flags
, size
);
4132 return ERR_PTR(-ENOSPC
);
4134 inode
->i_flags
|= i_flags
;
4135 inode
->i_size
= size
;
4136 clear_nlink(inode
); /* It is unlinked */
4137 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4139 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4140 &shmem_file_operations
);
4147 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4148 * kernel internal. There will be NO LSM permission checks against the
4149 * underlying inode. So users of this interface must do LSM checks at a
4150 * higher layer. The users are the big_key and shm implementations. LSM
4151 * checks are provided at the key or shm level rather than the inode.
4152 * @name: name for dentry (to be seen in /proc/<pid>/maps
4153 * @size: size to be set for the file
4154 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4156 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4158 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4162 * shmem_file_setup - get an unlinked file living in tmpfs
4163 * @name: name for dentry (to be seen in /proc/<pid>/maps
4164 * @size: size to be set for the file
4165 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4167 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4169 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4171 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4174 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4175 * @mnt: the tmpfs mount where the file will be created
4176 * @name: name for dentry (to be seen in /proc/<pid>/maps
4177 * @size: size to be set for the file
4178 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4180 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4181 loff_t size
, unsigned long flags
)
4183 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4185 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4188 * shmem_zero_setup - setup a shared anonymous mapping
4189 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4191 int shmem_zero_setup(struct vm_area_struct
*vma
)
4194 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4197 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4198 * between XFS directory reading and selinux: since this file is only
4199 * accessible to the user through its mapping, use S_PRIVATE flag to
4200 * bypass file security, in the same way as shmem_kernel_file_setup().
4202 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4204 return PTR_ERR(file
);
4208 vma
->vm_file
= file
;
4209 vma
->vm_ops
= &shmem_vm_ops
;
4211 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE
) &&
4212 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4213 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4214 khugepaged_enter(vma
, vma
->vm_flags
);
4219 EXPORT_SYMBOL_GPL(shmem_zero_setup
);
4222 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4223 * @mapping: the page's address_space
4224 * @index: the page index
4225 * @gfp: the page allocator flags to use if allocating
4227 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4228 * with any new page allocations done using the specified allocation flags.
4229 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4230 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4231 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4233 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4234 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4236 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4237 pgoff_t index
, gfp_t gfp
)
4240 struct inode
*inode
= mapping
->host
;
4244 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4245 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4246 gfp
, NULL
, NULL
, NULL
);
4248 page
= ERR_PTR(error
);
4254 * The tiny !SHMEM case uses ramfs without swap
4256 return read_cache_page_gfp(mapping
, index
, gfp
);
4259 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);