2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
42 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
44 static struct vfsmount
*shm_mnt
;
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/blkdev.h>
64 #include <linux/pagevec.h>
65 #include <linux/percpu_counter.h>
66 #include <linux/falloc.h>
67 #include <linux/splice.h>
68 #include <linux/security.h>
69 #include <linux/swapops.h>
70 #include <linux/mempolicy.h>
71 #include <linux/namei.h>
72 #include <linux/ctype.h>
73 #include <linux/migrate.h>
74 #include <linux/highmem.h>
75 #include <linux/seq_file.h>
76 #include <linux/magic.h>
77 #include <linux/syscalls.h>
78 #include <linux/fcntl.h>
79 #include <uapi/linux/memfd.h>
80 #include <linux/userfaultfd_k.h>
81 #include <linux/rmap.h>
82 #include <linux/uuid.h>
84 #include <linux/uaccess.h>
85 #include <asm/pgtable.h>
89 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
90 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
92 /* Pretend that each entry is of this size in directory's i_size */
93 #define BOGO_DIRENT_SIZE 20
95 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
96 #define SHORT_SYMLINK_LEN 128
99 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
100 * inode->i_private (with i_mutex making sure that it has only one user at
101 * a time): we would prefer not to enlarge the shmem inode just for that.
103 struct shmem_falloc
{
104 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
105 pgoff_t start
; /* start of range currently being fallocated */
106 pgoff_t next
; /* the next page offset to be fallocated */
107 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
108 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
111 struct shmem_options
{
112 unsigned long long blocks
;
113 unsigned long long inodes
;
114 struct mempolicy
*mpol
;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
126 static unsigned long shmem_default_max_blocks(void)
128 return totalram_pages() / 2;
131 static unsigned long shmem_default_max_inodes(void)
133 unsigned long nr_pages
= totalram_pages();
135 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
139 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
140 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
141 struct shmem_inode_info
*info
, pgoff_t index
);
142 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
143 struct page
**pagep
, enum sgp_type sgp
,
144 gfp_t gfp
, struct vm_area_struct
*vma
,
145 vm_fault_t
*fault_type
);
146 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
147 struct page
**pagep
, enum sgp_type sgp
,
148 gfp_t gfp
, struct vm_area_struct
*vma
,
149 struct vm_fault
*vmf
, vm_fault_t
*fault_type
);
151 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
152 struct page
**pagep
, enum sgp_type sgp
)
154 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
155 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
158 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
160 return sb
->s_fs_info
;
164 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
165 * for shared memory and for shared anonymous (/dev/zero) mappings
166 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
167 * consistent with the pre-accounting of private mappings ...
169 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
171 return (flags
& VM_NORESERVE
) ?
172 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
175 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
177 if (!(flags
& VM_NORESERVE
))
178 vm_unacct_memory(VM_ACCT(size
));
181 static inline int shmem_reacct_size(unsigned long flags
,
182 loff_t oldsize
, loff_t newsize
)
184 if (!(flags
& VM_NORESERVE
)) {
185 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
186 return security_vm_enough_memory_mm(current
->mm
,
187 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
188 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
189 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
195 * ... whereas tmpfs objects are accounted incrementally as
196 * pages are allocated, in order to allow large sparse files.
197 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
198 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
200 static inline int shmem_acct_block(unsigned long flags
, long pages
)
202 if (!(flags
& VM_NORESERVE
))
205 return security_vm_enough_memory_mm(current
->mm
,
206 pages
* VM_ACCT(PAGE_SIZE
));
209 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
211 if (flags
& VM_NORESERVE
)
212 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
215 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
217 struct shmem_inode_info
*info
= SHMEM_I(inode
);
218 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
220 if (shmem_acct_block(info
->flags
, pages
))
223 if (sbinfo
->max_blocks
) {
224 if (percpu_counter_compare(&sbinfo
->used_blocks
,
225 sbinfo
->max_blocks
- pages
) > 0)
227 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
233 shmem_unacct_blocks(info
->flags
, pages
);
237 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
239 struct shmem_inode_info
*info
= SHMEM_I(inode
);
240 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
242 if (sbinfo
->max_blocks
)
243 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
244 shmem_unacct_blocks(info
->flags
, pages
);
247 static const struct super_operations shmem_ops
;
248 static const struct address_space_operations shmem_aops
;
249 static const struct file_operations shmem_file_operations
;
250 static const struct inode_operations shmem_inode_operations
;
251 static const struct inode_operations shmem_dir_inode_operations
;
252 static const struct inode_operations shmem_special_inode_operations
;
253 static const struct vm_operations_struct shmem_vm_ops
;
254 static struct file_system_type shmem_fs_type
;
256 bool vma_is_shmem(struct vm_area_struct
*vma
)
258 return vma
->vm_ops
== &shmem_vm_ops
;
261 static LIST_HEAD(shmem_swaplist
);
262 static DEFINE_MUTEX(shmem_swaplist_mutex
);
264 static int shmem_reserve_inode(struct super_block
*sb
)
266 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
267 if (sbinfo
->max_inodes
) {
268 spin_lock(&sbinfo
->stat_lock
);
269 if (!sbinfo
->free_inodes
) {
270 spin_unlock(&sbinfo
->stat_lock
);
273 sbinfo
->free_inodes
--;
274 spin_unlock(&sbinfo
->stat_lock
);
279 static void shmem_free_inode(struct super_block
*sb
)
281 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
282 if (sbinfo
->max_inodes
) {
283 spin_lock(&sbinfo
->stat_lock
);
284 sbinfo
->free_inodes
++;
285 spin_unlock(&sbinfo
->stat_lock
);
290 * shmem_recalc_inode - recalculate the block usage of an inode
291 * @inode: inode to recalc
293 * We have to calculate the free blocks since the mm can drop
294 * undirtied hole pages behind our back.
296 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
297 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
299 * It has to be called with the spinlock held.
301 static void shmem_recalc_inode(struct inode
*inode
)
303 struct shmem_inode_info
*info
= SHMEM_I(inode
);
306 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
308 info
->alloced
-= freed
;
309 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
310 shmem_inode_unacct_blocks(inode
, freed
);
314 bool shmem_charge(struct inode
*inode
, long pages
)
316 struct shmem_inode_info
*info
= SHMEM_I(inode
);
319 if (!shmem_inode_acct_block(inode
, pages
))
322 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
323 inode
->i_mapping
->nrpages
+= pages
;
325 spin_lock_irqsave(&info
->lock
, flags
);
326 info
->alloced
+= pages
;
327 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
328 shmem_recalc_inode(inode
);
329 spin_unlock_irqrestore(&info
->lock
, flags
);
334 void shmem_uncharge(struct inode
*inode
, long pages
)
336 struct shmem_inode_info
*info
= SHMEM_I(inode
);
339 /* nrpages adjustment done by __delete_from_page_cache() or caller */
341 spin_lock_irqsave(&info
->lock
, flags
);
342 info
->alloced
-= pages
;
343 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
344 shmem_recalc_inode(inode
);
345 spin_unlock_irqrestore(&info
->lock
, flags
);
347 shmem_inode_unacct_blocks(inode
, pages
);
351 * Replace item expected in xarray by a new item, while holding xa_lock.
353 static int shmem_replace_entry(struct address_space
*mapping
,
354 pgoff_t index
, void *expected
, void *replacement
)
356 XA_STATE(xas
, &mapping
->i_pages
, index
);
359 VM_BUG_ON(!expected
);
360 VM_BUG_ON(!replacement
);
361 item
= xas_load(&xas
);
362 if (item
!= expected
)
364 xas_store(&xas
, replacement
);
369 * Sometimes, before we decide whether to proceed or to fail, we must check
370 * that an entry was not already brought back from swap by a racing thread.
372 * Checking page is not enough: by the time a SwapCache page is locked, it
373 * might be reused, and again be SwapCache, using the same swap as before.
375 static bool shmem_confirm_swap(struct address_space
*mapping
,
376 pgoff_t index
, swp_entry_t swap
)
378 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
382 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
385 * disables huge pages for the mount;
387 * enables huge pages for the mount;
388 * SHMEM_HUGE_WITHIN_SIZE:
389 * only allocate huge pages if the page will be fully within i_size,
390 * also respect fadvise()/madvise() hints;
392 * only allocate huge pages if requested with fadvise()/madvise();
395 #define SHMEM_HUGE_NEVER 0
396 #define SHMEM_HUGE_ALWAYS 1
397 #define SHMEM_HUGE_WITHIN_SIZE 2
398 #define SHMEM_HUGE_ADVISE 3
402 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
405 * disables huge on shm_mnt and all mounts, for emergency use;
407 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
410 #define SHMEM_HUGE_DENY (-1)
411 #define SHMEM_HUGE_FORCE (-2)
413 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
414 /* ifdef here to avoid bloating shmem.o when not necessary */
416 static int shmem_huge __read_mostly
;
418 #if defined(CONFIG_SYSFS)
419 static int shmem_parse_huge(const char *str
)
421 if (!strcmp(str
, "never"))
422 return SHMEM_HUGE_NEVER
;
423 if (!strcmp(str
, "always"))
424 return SHMEM_HUGE_ALWAYS
;
425 if (!strcmp(str
, "within_size"))
426 return SHMEM_HUGE_WITHIN_SIZE
;
427 if (!strcmp(str
, "advise"))
428 return SHMEM_HUGE_ADVISE
;
429 if (!strcmp(str
, "deny"))
430 return SHMEM_HUGE_DENY
;
431 if (!strcmp(str
, "force"))
432 return SHMEM_HUGE_FORCE
;
437 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
438 static const char *shmem_format_huge(int huge
)
441 case SHMEM_HUGE_NEVER
:
443 case SHMEM_HUGE_ALWAYS
:
445 case SHMEM_HUGE_WITHIN_SIZE
:
446 return "within_size";
447 case SHMEM_HUGE_ADVISE
:
449 case SHMEM_HUGE_DENY
:
451 case SHMEM_HUGE_FORCE
:
460 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
461 struct shrink_control
*sc
, unsigned long nr_to_split
)
463 LIST_HEAD(list
), *pos
, *next
;
464 LIST_HEAD(to_remove
);
466 struct shmem_inode_info
*info
;
468 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
469 int removed
= 0, split
= 0;
471 if (list_empty(&sbinfo
->shrinklist
))
474 spin_lock(&sbinfo
->shrinklist_lock
);
475 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
476 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
479 inode
= igrab(&info
->vfs_inode
);
481 /* inode is about to be evicted */
483 list_del_init(&info
->shrinklist
);
488 /* Check if there's anything to gain */
489 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
490 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
491 list_move(&info
->shrinklist
, &to_remove
);
496 list_move(&info
->shrinklist
, &list
);
501 spin_unlock(&sbinfo
->shrinklist_lock
);
503 list_for_each_safe(pos
, next
, &to_remove
) {
504 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
505 inode
= &info
->vfs_inode
;
506 list_del_init(&info
->shrinklist
);
510 list_for_each_safe(pos
, next
, &list
) {
513 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
514 inode
= &info
->vfs_inode
;
516 if (nr_to_split
&& split
>= nr_to_split
)
519 page
= find_get_page(inode
->i_mapping
,
520 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
524 /* No huge page at the end of the file: nothing to split */
525 if (!PageTransHuge(page
)) {
531 * Leave the inode on the list if we failed to lock
532 * the page at this time.
534 * Waiting for the lock may lead to deadlock in the
537 if (!trylock_page(page
)) {
542 ret
= split_huge_page(page
);
546 /* If split failed leave the inode on the list */
552 list_del_init(&info
->shrinklist
);
558 spin_lock(&sbinfo
->shrinklist_lock
);
559 list_splice_tail(&list
, &sbinfo
->shrinklist
);
560 sbinfo
->shrinklist_len
-= removed
;
561 spin_unlock(&sbinfo
->shrinklist_lock
);
566 static long shmem_unused_huge_scan(struct super_block
*sb
,
567 struct shrink_control
*sc
)
569 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
571 if (!READ_ONCE(sbinfo
->shrinklist_len
))
574 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
577 static long shmem_unused_huge_count(struct super_block
*sb
,
578 struct shrink_control
*sc
)
580 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
581 return READ_ONCE(sbinfo
->shrinklist_len
);
583 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
585 #define shmem_huge SHMEM_HUGE_DENY
587 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
588 struct shrink_control
*sc
, unsigned long nr_to_split
)
592 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
594 static inline bool is_huge_enabled(struct shmem_sb_info
*sbinfo
)
596 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
597 (shmem_huge
== SHMEM_HUGE_FORCE
|| sbinfo
->huge
) &&
598 shmem_huge
!= SHMEM_HUGE_DENY
)
604 * Like add_to_page_cache_locked, but error if expected item has gone.
606 static int shmem_add_to_page_cache(struct page
*page
,
607 struct address_space
*mapping
,
608 pgoff_t index
, void *expected
, gfp_t gfp
,
609 struct mm_struct
*charge_mm
)
611 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, compound_order(page
));
613 unsigned long nr
= compound_nr(page
);
616 VM_BUG_ON_PAGE(PageTail(page
), page
);
617 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
618 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
619 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
620 VM_BUG_ON(expected
&& PageTransHuge(page
));
622 page_ref_add(page
, nr
);
623 page
->mapping
= mapping
;
626 if (!PageSwapCache(page
)) {
627 error
= mem_cgroup_charge(page
, charge_mm
, gfp
);
629 if (PageTransHuge(page
)) {
630 count_vm_event(THP_FILE_FALLBACK
);
631 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
636 cgroup_throttle_swaprate(page
, gfp
);
641 entry
= xas_find_conflict(&xas
);
642 if (entry
!= expected
)
643 xas_set_err(&xas
, -EEXIST
);
644 xas_create_range(&xas
);
648 xas_store(&xas
, page
);
653 if (PageTransHuge(page
)) {
654 count_vm_event(THP_FILE_ALLOC
);
655 __inc_node_page_state(page
, NR_SHMEM_THPS
);
657 mapping
->nrpages
+= nr
;
658 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, nr
);
659 __mod_lruvec_page_state(page
, NR_SHMEM
, nr
);
661 xas_unlock_irq(&xas
);
662 } while (xas_nomem(&xas
, gfp
));
664 if (xas_error(&xas
)) {
665 error
= xas_error(&xas
);
671 page
->mapping
= NULL
;
672 page_ref_sub(page
, nr
);
677 * Like delete_from_page_cache, but substitutes swap for page.
679 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
681 struct address_space
*mapping
= page
->mapping
;
684 VM_BUG_ON_PAGE(PageCompound(page
), page
);
686 xa_lock_irq(&mapping
->i_pages
);
687 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
688 page
->mapping
= NULL
;
690 __dec_lruvec_page_state(page
, NR_FILE_PAGES
);
691 __dec_lruvec_page_state(page
, NR_SHMEM
);
692 xa_unlock_irq(&mapping
->i_pages
);
698 * Remove swap entry from page cache, free the swap and its page cache.
700 static int shmem_free_swap(struct address_space
*mapping
,
701 pgoff_t index
, void *radswap
)
705 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
708 free_swap_and_cache(radix_to_swp_entry(radswap
));
713 * Determine (in bytes) how many of the shmem object's pages mapped by the
714 * given offsets are swapped out.
716 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
717 * as long as the inode doesn't go away and racy results are not a problem.
719 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
720 pgoff_t start
, pgoff_t end
)
722 XA_STATE(xas
, &mapping
->i_pages
, start
);
724 unsigned long swapped
= 0;
727 xas_for_each(&xas
, page
, end
- 1) {
728 if (xas_retry(&xas
, page
))
730 if (xa_is_value(page
))
733 if (need_resched()) {
741 return swapped
<< PAGE_SHIFT
;
745 * Determine (in bytes) how many of the shmem object's pages mapped by the
746 * given vma is swapped out.
748 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
749 * as long as the inode doesn't go away and racy results are not a problem.
751 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
753 struct inode
*inode
= file_inode(vma
->vm_file
);
754 struct shmem_inode_info
*info
= SHMEM_I(inode
);
755 struct address_space
*mapping
= inode
->i_mapping
;
756 unsigned long swapped
;
758 /* Be careful as we don't hold info->lock */
759 swapped
= READ_ONCE(info
->swapped
);
762 * The easier cases are when the shmem object has nothing in swap, or
763 * the vma maps it whole. Then we can simply use the stats that we
769 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
770 return swapped
<< PAGE_SHIFT
;
772 /* Here comes the more involved part */
773 return shmem_partial_swap_usage(mapping
,
774 linear_page_index(vma
, vma
->vm_start
),
775 linear_page_index(vma
, vma
->vm_end
));
779 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
781 void shmem_unlock_mapping(struct address_space
*mapping
)
784 pgoff_t indices
[PAGEVEC_SIZE
];
789 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
791 while (!mapping_unevictable(mapping
)) {
793 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
794 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
796 pvec
.nr
= find_get_entries(mapping
, index
,
797 PAGEVEC_SIZE
, pvec
.pages
, indices
);
800 index
= indices
[pvec
.nr
- 1] + 1;
801 pagevec_remove_exceptionals(&pvec
);
802 check_move_unevictable_pages(&pvec
);
803 pagevec_release(&pvec
);
809 * Check whether a hole-punch or truncation needs to split a huge page,
810 * returning true if no split was required, or the split has been successful.
812 * Eviction (or truncation to 0 size) should never need to split a huge page;
813 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
814 * head, and then succeeded to trylock on tail.
816 * A split can only succeed when there are no additional references on the
817 * huge page: so the split below relies upon find_get_entries() having stopped
818 * when it found a subpage of the huge page, without getting further references.
820 static bool shmem_punch_compound(struct page
*page
, pgoff_t start
, pgoff_t end
)
822 if (!PageTransCompound(page
))
825 /* Just proceed to delete a huge page wholly within the range punched */
826 if (PageHead(page
) &&
827 page
->index
>= start
&& page
->index
+ HPAGE_PMD_NR
<= end
)
830 /* Try to split huge page, so we can truly punch the hole or truncate */
831 return split_huge_page(page
) >= 0;
835 * Remove range of pages and swap entries from page cache, and free them.
836 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
838 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
841 struct address_space
*mapping
= inode
->i_mapping
;
842 struct shmem_inode_info
*info
= SHMEM_I(inode
);
843 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
844 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
845 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
846 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
848 pgoff_t indices
[PAGEVEC_SIZE
];
849 long nr_swaps_freed
= 0;
854 end
= -1; /* unsigned, so actually very big */
858 while (index
< end
) {
859 pvec
.nr
= find_get_entries(mapping
, index
,
860 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
861 pvec
.pages
, indices
);
864 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
865 struct page
*page
= pvec
.pages
[i
];
871 if (xa_is_value(page
)) {
874 nr_swaps_freed
+= !shmem_free_swap(mapping
,
879 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
881 if (!trylock_page(page
))
884 if ((!unfalloc
|| !PageUptodate(page
)) &&
885 page_mapping(page
) == mapping
) {
886 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
887 if (shmem_punch_compound(page
, start
, end
))
888 truncate_inode_page(mapping
, page
);
892 pagevec_remove_exceptionals(&pvec
);
893 pagevec_release(&pvec
);
899 struct page
*page
= NULL
;
900 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
902 unsigned int top
= PAGE_SIZE
;
907 zero_user_segment(page
, partial_start
, top
);
908 set_page_dirty(page
);
914 struct page
*page
= NULL
;
915 shmem_getpage(inode
, end
, &page
, SGP_READ
);
917 zero_user_segment(page
, 0, partial_end
);
918 set_page_dirty(page
);
927 while (index
< end
) {
930 pvec
.nr
= find_get_entries(mapping
, index
,
931 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
932 pvec
.pages
, indices
);
934 /* If all gone or hole-punch or unfalloc, we're done */
935 if (index
== start
|| end
!= -1)
937 /* But if truncating, restart to make sure all gone */
941 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
942 struct page
*page
= pvec
.pages
[i
];
948 if (xa_is_value(page
)) {
951 if (shmem_free_swap(mapping
, index
, page
)) {
952 /* Swap was replaced by page: retry */
962 if (!unfalloc
|| !PageUptodate(page
)) {
963 if (page_mapping(page
) != mapping
) {
964 /* Page was replaced by swap: retry */
969 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
970 if (shmem_punch_compound(page
, start
, end
))
971 truncate_inode_page(mapping
, page
);
972 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
973 /* Wipe the page and don't get stuck */
974 clear_highpage(page
);
975 flush_dcache_page(page
);
976 set_page_dirty(page
);
978 round_up(start
, HPAGE_PMD_NR
))
984 pagevec_remove_exceptionals(&pvec
);
985 pagevec_release(&pvec
);
989 spin_lock_irq(&info
->lock
);
990 info
->swapped
-= nr_swaps_freed
;
991 shmem_recalc_inode(inode
);
992 spin_unlock_irq(&info
->lock
);
995 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
997 shmem_undo_range(inode
, lstart
, lend
, false);
998 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1000 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1002 static int shmem_getattr(const struct path
*path
, struct kstat
*stat
,
1003 u32 request_mask
, unsigned int query_flags
)
1005 struct inode
*inode
= path
->dentry
->d_inode
;
1006 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1007 struct shmem_sb_info
*sb_info
= SHMEM_SB(inode
->i_sb
);
1009 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1010 spin_lock_irq(&info
->lock
);
1011 shmem_recalc_inode(inode
);
1012 spin_unlock_irq(&info
->lock
);
1014 generic_fillattr(inode
, stat
);
1016 if (is_huge_enabled(sb_info
))
1017 stat
->blksize
= HPAGE_PMD_SIZE
;
1022 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1024 struct inode
*inode
= d_inode(dentry
);
1025 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1026 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1029 error
= setattr_prepare(dentry
, attr
);
1033 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1034 loff_t oldsize
= inode
->i_size
;
1035 loff_t newsize
= attr
->ia_size
;
1037 /* protected by i_mutex */
1038 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1039 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1042 if (newsize
!= oldsize
) {
1043 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1047 i_size_write(inode
, newsize
);
1048 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1050 if (newsize
<= oldsize
) {
1051 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1052 if (oldsize
> holebegin
)
1053 unmap_mapping_range(inode
->i_mapping
,
1056 shmem_truncate_range(inode
,
1057 newsize
, (loff_t
)-1);
1058 /* unmap again to remove racily COWed private pages */
1059 if (oldsize
> holebegin
)
1060 unmap_mapping_range(inode
->i_mapping
,
1064 * Part of the huge page can be beyond i_size: subject
1065 * to shrink under memory pressure.
1067 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1068 spin_lock(&sbinfo
->shrinklist_lock
);
1070 * _careful to defend against unlocked access to
1071 * ->shrink_list in shmem_unused_huge_shrink()
1073 if (list_empty_careful(&info
->shrinklist
)) {
1074 list_add_tail(&info
->shrinklist
,
1075 &sbinfo
->shrinklist
);
1076 sbinfo
->shrinklist_len
++;
1078 spin_unlock(&sbinfo
->shrinklist_lock
);
1083 setattr_copy(inode
, attr
);
1084 if (attr
->ia_valid
& ATTR_MODE
)
1085 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1089 static void shmem_evict_inode(struct inode
*inode
)
1091 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1092 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1094 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1095 shmem_unacct_size(info
->flags
, inode
->i_size
);
1097 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1098 if (!list_empty(&info
->shrinklist
)) {
1099 spin_lock(&sbinfo
->shrinklist_lock
);
1100 if (!list_empty(&info
->shrinklist
)) {
1101 list_del_init(&info
->shrinklist
);
1102 sbinfo
->shrinklist_len
--;
1104 spin_unlock(&sbinfo
->shrinklist_lock
);
1106 while (!list_empty(&info
->swaplist
)) {
1107 /* Wait while shmem_unuse() is scanning this inode... */
1108 wait_var_event(&info
->stop_eviction
,
1109 !atomic_read(&info
->stop_eviction
));
1110 mutex_lock(&shmem_swaplist_mutex
);
1111 /* ...but beware of the race if we peeked too early */
1112 if (!atomic_read(&info
->stop_eviction
))
1113 list_del_init(&info
->swaplist
);
1114 mutex_unlock(&shmem_swaplist_mutex
);
1118 simple_xattrs_free(&info
->xattrs
);
1119 WARN_ON(inode
->i_blocks
);
1120 shmem_free_inode(inode
->i_sb
);
1124 extern struct swap_info_struct
*swap_info
[];
1126 static int shmem_find_swap_entries(struct address_space
*mapping
,
1127 pgoff_t start
, unsigned int nr_entries
,
1128 struct page
**entries
, pgoff_t
*indices
,
1129 unsigned int type
, bool frontswap
)
1131 XA_STATE(xas
, &mapping
->i_pages
, start
);
1134 unsigned int ret
= 0;
1140 xas_for_each(&xas
, page
, ULONG_MAX
) {
1141 if (xas_retry(&xas
, page
))
1144 if (!xa_is_value(page
))
1147 entry
= radix_to_swp_entry(page
);
1148 if (swp_type(entry
) != type
)
1151 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1154 indices
[ret
] = xas
.xa_index
;
1155 entries
[ret
] = page
;
1157 if (need_resched()) {
1161 if (++ret
== nr_entries
)
1170 * Move the swapped pages for an inode to page cache. Returns the count
1171 * of pages swapped in, or the error in case of failure.
1173 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1179 struct address_space
*mapping
= inode
->i_mapping
;
1181 for (i
= 0; i
< pvec
.nr
; i
++) {
1182 struct page
*page
= pvec
.pages
[i
];
1184 if (!xa_is_value(page
))
1186 error
= shmem_swapin_page(inode
, indices
[i
],
1188 mapping_gfp_mask(mapping
),
1195 if (error
== -ENOMEM
)
1199 return error
? error
: ret
;
1203 * If swap found in inode, free it and move page from swapcache to filecache.
1205 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1206 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1208 struct address_space
*mapping
= inode
->i_mapping
;
1210 struct pagevec pvec
;
1211 pgoff_t indices
[PAGEVEC_SIZE
];
1212 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1215 pagevec_init(&pvec
);
1217 unsigned int nr_entries
= PAGEVEC_SIZE
;
1219 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1220 nr_entries
= *fs_pages_to_unuse
;
1222 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1223 pvec
.pages
, indices
,
1230 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1234 if (frontswap_partial
) {
1235 *fs_pages_to_unuse
-= ret
;
1236 if (*fs_pages_to_unuse
== 0) {
1237 ret
= FRONTSWAP_PAGES_UNUSED
;
1242 start
= indices
[pvec
.nr
- 1];
1249 * Read all the shared memory data that resides in the swap
1250 * device 'type' back into memory, so the swap device can be
1253 int shmem_unuse(unsigned int type
, bool frontswap
,
1254 unsigned long *fs_pages_to_unuse
)
1256 struct shmem_inode_info
*info
, *next
;
1259 if (list_empty(&shmem_swaplist
))
1262 mutex_lock(&shmem_swaplist_mutex
);
1263 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1264 if (!info
->swapped
) {
1265 list_del_init(&info
->swaplist
);
1269 * Drop the swaplist mutex while searching the inode for swap;
1270 * but before doing so, make sure shmem_evict_inode() will not
1271 * remove placeholder inode from swaplist, nor let it be freed
1272 * (igrab() would protect from unlink, but not from unmount).
1274 atomic_inc(&info
->stop_eviction
);
1275 mutex_unlock(&shmem_swaplist_mutex
);
1277 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1281 mutex_lock(&shmem_swaplist_mutex
);
1282 next
= list_next_entry(info
, swaplist
);
1284 list_del_init(&info
->swaplist
);
1285 if (atomic_dec_and_test(&info
->stop_eviction
))
1286 wake_up_var(&info
->stop_eviction
);
1290 mutex_unlock(&shmem_swaplist_mutex
);
1296 * Move the page from the page cache to the swap cache.
1298 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1300 struct shmem_inode_info
*info
;
1301 struct address_space
*mapping
;
1302 struct inode
*inode
;
1306 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1307 BUG_ON(!PageLocked(page
));
1308 mapping
= page
->mapping
;
1309 index
= page
->index
;
1310 inode
= mapping
->host
;
1311 info
= SHMEM_I(inode
);
1312 if (info
->flags
& VM_LOCKED
)
1314 if (!total_swap_pages
)
1318 * Our capabilities prevent regular writeback or sync from ever calling
1319 * shmem_writepage; but a stacking filesystem might use ->writepage of
1320 * its underlying filesystem, in which case tmpfs should write out to
1321 * swap only in response to memory pressure, and not for the writeback
1324 if (!wbc
->for_reclaim
) {
1325 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1330 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1331 * value into swapfile.c, the only way we can correctly account for a
1332 * fallocated page arriving here is now to initialize it and write it.
1334 * That's okay for a page already fallocated earlier, but if we have
1335 * not yet completed the fallocation, then (a) we want to keep track
1336 * of this page in case we have to undo it, and (b) it may not be a
1337 * good idea to continue anyway, once we're pushing into swap. So
1338 * reactivate the page, and let shmem_fallocate() quit when too many.
1340 if (!PageUptodate(page
)) {
1341 if (inode
->i_private
) {
1342 struct shmem_falloc
*shmem_falloc
;
1343 spin_lock(&inode
->i_lock
);
1344 shmem_falloc
= inode
->i_private
;
1346 !shmem_falloc
->waitq
&&
1347 index
>= shmem_falloc
->start
&&
1348 index
< shmem_falloc
->next
)
1349 shmem_falloc
->nr_unswapped
++;
1351 shmem_falloc
= NULL
;
1352 spin_unlock(&inode
->i_lock
);
1356 clear_highpage(page
);
1357 flush_dcache_page(page
);
1358 SetPageUptodate(page
);
1361 swap
= get_swap_page(page
);
1366 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1367 * if it's not already there. Do it now before the page is
1368 * moved to swap cache, when its pagelock no longer protects
1369 * the inode from eviction. But don't unlock the mutex until
1370 * we've incremented swapped, because shmem_unuse_inode() will
1371 * prune a !swapped inode from the swaplist under this mutex.
1373 mutex_lock(&shmem_swaplist_mutex
);
1374 if (list_empty(&info
->swaplist
))
1375 list_add(&info
->swaplist
, &shmem_swaplist
);
1377 if (add_to_swap_cache(page
, swap
,
1378 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
) == 0) {
1379 spin_lock_irq(&info
->lock
);
1380 shmem_recalc_inode(inode
);
1382 spin_unlock_irq(&info
->lock
);
1384 swap_shmem_alloc(swap
);
1385 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1387 mutex_unlock(&shmem_swaplist_mutex
);
1388 BUG_ON(page_mapped(page
));
1389 swap_writepage(page
, wbc
);
1393 mutex_unlock(&shmem_swaplist_mutex
);
1394 put_swap_page(page
, swap
);
1396 set_page_dirty(page
);
1397 if (wbc
->for_reclaim
)
1398 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1403 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1404 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1408 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1409 return; /* show nothing */
1411 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1413 seq_printf(seq
, ",mpol=%s", buffer
);
1416 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1418 struct mempolicy
*mpol
= NULL
;
1420 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1421 mpol
= sbinfo
->mpol
;
1423 spin_unlock(&sbinfo
->stat_lock
);
1427 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1428 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1431 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1435 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1437 #define vm_policy vm_private_data
1440 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1441 struct shmem_inode_info
*info
, pgoff_t index
)
1443 /* Create a pseudo vma that just contains the policy */
1444 vma_init(vma
, NULL
);
1445 /* Bias interleave by inode number to distribute better across nodes */
1446 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1447 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1450 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1452 /* Drop reference taken by mpol_shared_policy_lookup() */
1453 mpol_cond_put(vma
->vm_policy
);
1456 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1457 struct shmem_inode_info
*info
, pgoff_t index
)
1459 struct vm_area_struct pvma
;
1461 struct vm_fault vmf
;
1463 shmem_pseudo_vma_init(&pvma
, info
, index
);
1466 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1467 shmem_pseudo_vma_destroy(&pvma
);
1472 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1473 struct shmem_inode_info
*info
, pgoff_t index
)
1475 struct vm_area_struct pvma
;
1476 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1480 hindex
= round_down(index
, HPAGE_PMD_NR
);
1481 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1485 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1486 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1487 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1488 shmem_pseudo_vma_destroy(&pvma
);
1490 prep_transhuge_page(page
);
1492 count_vm_event(THP_FILE_FALLBACK
);
1496 static struct page
*shmem_alloc_page(gfp_t gfp
,
1497 struct shmem_inode_info
*info
, pgoff_t index
)
1499 struct vm_area_struct pvma
;
1502 shmem_pseudo_vma_init(&pvma
, info
, index
);
1503 page
= alloc_page_vma(gfp
, &pvma
, 0);
1504 shmem_pseudo_vma_destroy(&pvma
);
1509 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1510 struct inode
*inode
,
1511 pgoff_t index
, bool huge
)
1513 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1518 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1520 nr
= huge
? HPAGE_PMD_NR
: 1;
1522 if (!shmem_inode_acct_block(inode
, nr
))
1526 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1528 page
= shmem_alloc_page(gfp
, info
, index
);
1530 __SetPageLocked(page
);
1531 __SetPageSwapBacked(page
);
1536 shmem_inode_unacct_blocks(inode
, nr
);
1538 return ERR_PTR(err
);
1542 * When a page is moved from swapcache to shmem filecache (either by the
1543 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1544 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1545 * ignorance of the mapping it belongs to. If that mapping has special
1546 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1547 * we may need to copy to a suitable page before moving to filecache.
1549 * In a future release, this may well be extended to respect cpuset and
1550 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1551 * but for now it is a simple matter of zone.
1553 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1555 return page_zonenum(page
) > gfp_zone(gfp
);
1558 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1559 struct shmem_inode_info
*info
, pgoff_t index
)
1561 struct page
*oldpage
, *newpage
;
1562 struct address_space
*swap_mapping
;
1568 entry
.val
= page_private(oldpage
);
1569 swap_index
= swp_offset(entry
);
1570 swap_mapping
= page_mapping(oldpage
);
1573 * We have arrived here because our zones are constrained, so don't
1574 * limit chance of success by further cpuset and node constraints.
1576 gfp
&= ~GFP_CONSTRAINT_MASK
;
1577 newpage
= shmem_alloc_page(gfp
, info
, index
);
1582 copy_highpage(newpage
, oldpage
);
1583 flush_dcache_page(newpage
);
1585 __SetPageLocked(newpage
);
1586 __SetPageSwapBacked(newpage
);
1587 SetPageUptodate(newpage
);
1588 set_page_private(newpage
, entry
.val
);
1589 SetPageSwapCache(newpage
);
1592 * Our caller will very soon move newpage out of swapcache, but it's
1593 * a nice clean interface for us to replace oldpage by newpage there.
1595 xa_lock_irq(&swap_mapping
->i_pages
);
1596 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1598 mem_cgroup_migrate(oldpage
, newpage
);
1599 __inc_lruvec_page_state(newpage
, NR_FILE_PAGES
);
1600 __dec_lruvec_page_state(oldpage
, NR_FILE_PAGES
);
1602 xa_unlock_irq(&swap_mapping
->i_pages
);
1604 if (unlikely(error
)) {
1606 * Is this possible? I think not, now that our callers check
1607 * both PageSwapCache and page_private after getting page lock;
1608 * but be defensive. Reverse old to newpage for clear and free.
1612 lru_cache_add(newpage
);
1616 ClearPageSwapCache(oldpage
);
1617 set_page_private(oldpage
, 0);
1619 unlock_page(oldpage
);
1626 * Swap in the page pointed to by *pagep.
1627 * Caller has to make sure that *pagep contains a valid swapped page.
1628 * Returns 0 and the page in pagep if success. On failure, returns the
1629 * the error code and NULL in *pagep.
1631 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1632 struct page
**pagep
, enum sgp_type sgp
,
1633 gfp_t gfp
, struct vm_area_struct
*vma
,
1634 vm_fault_t
*fault_type
)
1636 struct address_space
*mapping
= inode
->i_mapping
;
1637 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1638 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1643 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1644 swap
= radix_to_swp_entry(*pagep
);
1647 /* Look it up and read it in.. */
1648 page
= lookup_swap_cache(swap
, NULL
, 0);
1650 /* Or update major stats only when swapin succeeds?? */
1652 *fault_type
|= VM_FAULT_MAJOR
;
1653 count_vm_event(PGMAJFAULT
);
1654 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1656 /* Here we actually start the io */
1657 page
= shmem_swapin(swap
, gfp
, info
, index
);
1664 /* We have to do this with page locked to prevent races */
1666 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1667 !shmem_confirm_swap(mapping
, index
, swap
)) {
1671 if (!PageUptodate(page
)) {
1675 wait_on_page_writeback(page
);
1677 if (shmem_should_replace_page(page
, gfp
)) {
1678 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1683 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1684 swp_to_radix_entry(swap
), gfp
,
1689 spin_lock_irq(&info
->lock
);
1691 shmem_recalc_inode(inode
);
1692 spin_unlock_irq(&info
->lock
);
1694 if (sgp
== SGP_WRITE
)
1695 mark_page_accessed(page
);
1697 delete_from_swap_cache(page
);
1698 set_page_dirty(page
);
1704 if (!shmem_confirm_swap(mapping
, index
, swap
))
1716 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1718 * If we allocate a new one we do not mark it dirty. That's up to the
1719 * vm. If we swap it in we mark it dirty since we also free the swap
1720 * entry since a page cannot live in both the swap and page cache.
1722 * vmf and fault_type are only supplied by shmem_fault:
1723 * otherwise they are NULL.
1725 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1726 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1727 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1728 vm_fault_t
*fault_type
)
1730 struct address_space
*mapping
= inode
->i_mapping
;
1731 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1732 struct shmem_sb_info
*sbinfo
;
1733 struct mm_struct
*charge_mm
;
1735 enum sgp_type sgp_huge
= sgp
;
1736 pgoff_t hindex
= index
;
1741 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1743 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1746 if (sgp
<= SGP_CACHE
&&
1747 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1751 sbinfo
= SHMEM_SB(inode
->i_sb
);
1752 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1754 page
= find_lock_entry(mapping
, index
);
1755 if (xa_is_value(page
)) {
1756 error
= shmem_swapin_page(inode
, index
, &page
,
1757 sgp
, gfp
, vma
, fault_type
);
1758 if (error
== -EEXIST
)
1765 if (page
&& sgp
== SGP_WRITE
)
1766 mark_page_accessed(page
);
1768 /* fallocated page? */
1769 if (page
&& !PageUptodate(page
)) {
1770 if (sgp
!= SGP_READ
)
1776 if (page
|| sgp
== SGP_READ
) {
1782 * Fast cache lookup did not find it:
1783 * bring it back from swap or allocate.
1786 if (vma
&& userfaultfd_missing(vma
)) {
1787 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1791 /* shmem_symlink() */
1792 if (mapping
->a_ops
!= &shmem_aops
)
1794 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1796 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1798 switch (sbinfo
->huge
) {
1799 case SHMEM_HUGE_NEVER
:
1801 case SHMEM_HUGE_WITHIN_SIZE
: {
1805 off
= round_up(index
, HPAGE_PMD_NR
);
1806 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1807 if (i_size
>= HPAGE_PMD_SIZE
&&
1808 i_size
>> PAGE_SHIFT
>= off
)
1813 case SHMEM_HUGE_ADVISE
:
1814 if (sgp_huge
== SGP_HUGE
)
1816 /* TODO: implement fadvise() hints */
1821 page
= shmem_alloc_and_acct_page(gfp
, inode
, index
, true);
1824 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1830 error
= PTR_ERR(page
);
1832 if (error
!= -ENOSPC
)
1835 * Try to reclaim some space by splitting a huge page
1836 * beyond i_size on the filesystem.
1841 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1842 if (ret
== SHRINK_STOP
)
1850 if (PageTransHuge(page
))
1851 hindex
= round_down(index
, HPAGE_PMD_NR
);
1855 if (sgp
== SGP_WRITE
)
1856 __SetPageReferenced(page
);
1858 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1859 NULL
, gfp
& GFP_RECLAIM_MASK
,
1863 lru_cache_add(page
);
1865 spin_lock_irq(&info
->lock
);
1866 info
->alloced
+= compound_nr(page
);
1867 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1868 shmem_recalc_inode(inode
);
1869 spin_unlock_irq(&info
->lock
);
1872 if (PageTransHuge(page
) &&
1873 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1874 hindex
+ HPAGE_PMD_NR
- 1) {
1876 * Part of the huge page is beyond i_size: subject
1877 * to shrink under memory pressure.
1879 spin_lock(&sbinfo
->shrinklist_lock
);
1881 * _careful to defend against unlocked access to
1882 * ->shrink_list in shmem_unused_huge_shrink()
1884 if (list_empty_careful(&info
->shrinklist
)) {
1885 list_add_tail(&info
->shrinklist
,
1886 &sbinfo
->shrinklist
);
1887 sbinfo
->shrinklist_len
++;
1889 spin_unlock(&sbinfo
->shrinklist_lock
);
1893 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1895 if (sgp
== SGP_FALLOC
)
1899 * Let SGP_WRITE caller clear ends if write does not fill page;
1900 * but SGP_FALLOC on a page fallocated earlier must initialize
1901 * it now, lest undo on failure cancel our earlier guarantee.
1903 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1904 struct page
*head
= compound_head(page
);
1907 for (i
= 0; i
< compound_nr(head
); i
++) {
1908 clear_highpage(head
+ i
);
1909 flush_dcache_page(head
+ i
);
1911 SetPageUptodate(head
);
1914 /* Perhaps the file has been truncated since we checked */
1915 if (sgp
<= SGP_CACHE
&&
1916 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1918 ClearPageDirty(page
);
1919 delete_from_page_cache(page
);
1920 spin_lock_irq(&info
->lock
);
1921 shmem_recalc_inode(inode
);
1922 spin_unlock_irq(&info
->lock
);
1927 *pagep
= page
+ index
- hindex
;
1934 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
1936 if (PageTransHuge(page
)) {
1946 if (error
== -ENOSPC
&& !once
++) {
1947 spin_lock_irq(&info
->lock
);
1948 shmem_recalc_inode(inode
);
1949 spin_unlock_irq(&info
->lock
);
1952 if (error
== -EEXIST
)
1958 * This is like autoremove_wake_function, but it removes the wait queue
1959 * entry unconditionally - even if something else had already woken the
1962 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
1964 int ret
= default_wake_function(wait
, mode
, sync
, key
);
1965 list_del_init(&wait
->entry
);
1969 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
1971 struct vm_area_struct
*vma
= vmf
->vma
;
1972 struct inode
*inode
= file_inode(vma
->vm_file
);
1973 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1976 vm_fault_t ret
= VM_FAULT_LOCKED
;
1979 * Trinity finds that probing a hole which tmpfs is punching can
1980 * prevent the hole-punch from ever completing: which in turn
1981 * locks writers out with its hold on i_mutex. So refrain from
1982 * faulting pages into the hole while it's being punched. Although
1983 * shmem_undo_range() does remove the additions, it may be unable to
1984 * keep up, as each new page needs its own unmap_mapping_range() call,
1985 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1987 * It does not matter if we sometimes reach this check just before the
1988 * hole-punch begins, so that one fault then races with the punch:
1989 * we just need to make racing faults a rare case.
1991 * The implementation below would be much simpler if we just used a
1992 * standard mutex or completion: but we cannot take i_mutex in fault,
1993 * and bloating every shmem inode for this unlikely case would be sad.
1995 if (unlikely(inode
->i_private
)) {
1996 struct shmem_falloc
*shmem_falloc
;
1998 spin_lock(&inode
->i_lock
);
1999 shmem_falloc
= inode
->i_private
;
2001 shmem_falloc
->waitq
&&
2002 vmf
->pgoff
>= shmem_falloc
->start
&&
2003 vmf
->pgoff
< shmem_falloc
->next
) {
2005 wait_queue_head_t
*shmem_falloc_waitq
;
2006 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2008 ret
= VM_FAULT_NOPAGE
;
2009 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2011 ret
= VM_FAULT_RETRY
;
2013 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2014 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2015 TASK_UNINTERRUPTIBLE
);
2016 spin_unlock(&inode
->i_lock
);
2020 * shmem_falloc_waitq points into the shmem_fallocate()
2021 * stack of the hole-punching task: shmem_falloc_waitq
2022 * is usually invalid by the time we reach here, but
2023 * finish_wait() does not dereference it in that case;
2024 * though i_lock needed lest racing with wake_up_all().
2026 spin_lock(&inode
->i_lock
);
2027 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2028 spin_unlock(&inode
->i_lock
);
2034 spin_unlock(&inode
->i_lock
);
2039 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2040 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2042 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2045 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2046 gfp
, vma
, vmf
, &ret
);
2048 return vmf_error(err
);
2052 unsigned long shmem_get_unmapped_area(struct file
*file
,
2053 unsigned long uaddr
, unsigned long len
,
2054 unsigned long pgoff
, unsigned long flags
)
2056 unsigned long (*get_area
)(struct file
*,
2057 unsigned long, unsigned long, unsigned long, unsigned long);
2059 unsigned long offset
;
2060 unsigned long inflated_len
;
2061 unsigned long inflated_addr
;
2062 unsigned long inflated_offset
;
2064 if (len
> TASK_SIZE
)
2067 get_area
= current
->mm
->get_unmapped_area
;
2068 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2070 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2072 if (IS_ERR_VALUE(addr
))
2074 if (addr
& ~PAGE_MASK
)
2076 if (addr
> TASK_SIZE
- len
)
2079 if (shmem_huge
== SHMEM_HUGE_DENY
)
2081 if (len
< HPAGE_PMD_SIZE
)
2083 if (flags
& MAP_FIXED
)
2086 * Our priority is to support MAP_SHARED mapped hugely;
2087 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2088 * But if caller specified an address hint and we allocated area there
2089 * successfully, respect that as before.
2094 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2095 struct super_block
*sb
;
2098 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2099 sb
= file_inode(file
)->i_sb
;
2102 * Called directly from mm/mmap.c, or drivers/char/mem.c
2103 * for "/dev/zero", to create a shared anonymous object.
2105 if (IS_ERR(shm_mnt
))
2107 sb
= shm_mnt
->mnt_sb
;
2109 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2113 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2114 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2116 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2119 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2120 if (inflated_len
> TASK_SIZE
)
2122 if (inflated_len
< len
)
2125 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2126 if (IS_ERR_VALUE(inflated_addr
))
2128 if (inflated_addr
& ~PAGE_MASK
)
2131 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2132 inflated_addr
+= offset
- inflated_offset
;
2133 if (inflated_offset
> offset
)
2134 inflated_addr
+= HPAGE_PMD_SIZE
;
2136 if (inflated_addr
> TASK_SIZE
- len
)
2138 return inflated_addr
;
2142 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2144 struct inode
*inode
= file_inode(vma
->vm_file
);
2145 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2148 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2151 struct inode
*inode
= file_inode(vma
->vm_file
);
2154 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2155 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2159 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2161 struct inode
*inode
= file_inode(file
);
2162 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2163 int retval
= -ENOMEM
;
2166 * What serializes the accesses to info->flags?
2167 * ipc_lock_object() when called from shmctl_do_lock(),
2168 * no serialization needed when called from shm_destroy().
2170 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2171 if (!user_shm_lock(inode
->i_size
, user
))
2173 info
->flags
|= VM_LOCKED
;
2174 mapping_set_unevictable(file
->f_mapping
);
2176 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2177 user_shm_unlock(inode
->i_size
, user
);
2178 info
->flags
&= ~VM_LOCKED
;
2179 mapping_clear_unevictable(file
->f_mapping
);
2187 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2189 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2191 if (info
->seals
& F_SEAL_FUTURE_WRITE
) {
2193 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2194 * "future write" seal active.
2196 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_WRITE
))
2200 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2201 * MAP_SHARED and read-only, take care to not allow mprotect to
2202 * revert protections on such mappings. Do this only for shared
2203 * mappings. For private mappings, don't need to mask
2204 * VM_MAYWRITE as we still want them to be COW-writable.
2206 if (vma
->vm_flags
& VM_SHARED
)
2207 vma
->vm_flags
&= ~(VM_MAYWRITE
);
2210 file_accessed(file
);
2211 vma
->vm_ops
= &shmem_vm_ops
;
2212 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
2213 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2214 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2215 khugepaged_enter(vma
, vma
->vm_flags
);
2220 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2221 umode_t mode
, dev_t dev
, unsigned long flags
)
2223 struct inode
*inode
;
2224 struct shmem_inode_info
*info
;
2225 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2227 if (shmem_reserve_inode(sb
))
2230 inode
= new_inode(sb
);
2232 inode
->i_ino
= get_next_ino();
2233 inode_init_owner(inode
, dir
, mode
);
2234 inode
->i_blocks
= 0;
2235 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2236 inode
->i_generation
= prandom_u32();
2237 info
= SHMEM_I(inode
);
2238 memset(info
, 0, (char *)inode
- (char *)info
);
2239 spin_lock_init(&info
->lock
);
2240 atomic_set(&info
->stop_eviction
, 0);
2241 info
->seals
= F_SEAL_SEAL
;
2242 info
->flags
= flags
& VM_NORESERVE
;
2243 INIT_LIST_HEAD(&info
->shrinklist
);
2244 INIT_LIST_HEAD(&info
->swaplist
);
2245 simple_xattrs_init(&info
->xattrs
);
2246 cache_no_acl(inode
);
2248 switch (mode
& S_IFMT
) {
2250 inode
->i_op
= &shmem_special_inode_operations
;
2251 init_special_inode(inode
, mode
, dev
);
2254 inode
->i_mapping
->a_ops
= &shmem_aops
;
2255 inode
->i_op
= &shmem_inode_operations
;
2256 inode
->i_fop
= &shmem_file_operations
;
2257 mpol_shared_policy_init(&info
->policy
,
2258 shmem_get_sbmpol(sbinfo
));
2262 /* Some things misbehave if size == 0 on a directory */
2263 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2264 inode
->i_op
= &shmem_dir_inode_operations
;
2265 inode
->i_fop
= &simple_dir_operations
;
2269 * Must not load anything in the rbtree,
2270 * mpol_free_shared_policy will not be called.
2272 mpol_shared_policy_init(&info
->policy
, NULL
);
2276 lockdep_annotate_inode_mutex_key(inode
);
2278 shmem_free_inode(sb
);
2282 bool shmem_mapping(struct address_space
*mapping
)
2284 return mapping
->a_ops
== &shmem_aops
;
2287 static int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2289 struct vm_area_struct
*dst_vma
,
2290 unsigned long dst_addr
,
2291 unsigned long src_addr
,
2293 struct page
**pagep
)
2295 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2296 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2297 struct address_space
*mapping
= inode
->i_mapping
;
2298 gfp_t gfp
= mapping_gfp_mask(mapping
);
2299 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2303 pte_t _dst_pte
, *dst_pte
;
2305 pgoff_t offset
, max_off
;
2308 if (!shmem_inode_acct_block(inode
, 1))
2312 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2314 goto out_unacct_blocks
;
2316 if (!zeropage
) { /* mcopy_atomic */
2317 page_kaddr
= kmap_atomic(page
);
2318 ret
= copy_from_user(page_kaddr
,
2319 (const void __user
*)src_addr
,
2321 kunmap_atomic(page_kaddr
);
2323 /* fallback to copy_from_user outside mmap_sem */
2324 if (unlikely(ret
)) {
2326 shmem_inode_unacct_blocks(inode
, 1);
2327 /* don't free the page */
2330 } else { /* mfill_zeropage_atomic */
2331 clear_highpage(page
);
2338 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2339 __SetPageLocked(page
);
2340 __SetPageSwapBacked(page
);
2341 __SetPageUptodate(page
);
2344 offset
= linear_page_index(dst_vma
, dst_addr
);
2345 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2346 if (unlikely(offset
>= max_off
))
2349 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2350 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2354 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2355 if (dst_vma
->vm_flags
& VM_WRITE
)
2356 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2359 * We don't set the pte dirty if the vma has no
2360 * VM_WRITE permission, so mark the page dirty or it
2361 * could be freed from under us. We could do it
2362 * unconditionally before unlock_page(), but doing it
2363 * only if VM_WRITE is not set is faster.
2365 set_page_dirty(page
);
2368 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2371 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2372 if (unlikely(offset
>= max_off
))
2373 goto out_release_unlock
;
2376 if (!pte_none(*dst_pte
))
2377 goto out_release_unlock
;
2379 lru_cache_add(page
);
2381 spin_lock_irq(&info
->lock
);
2383 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2384 shmem_recalc_inode(inode
);
2385 spin_unlock_irq(&info
->lock
);
2387 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2388 page_add_file_rmap(page
, false);
2389 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2391 /* No need to invalidate - it was non-present before */
2392 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2393 pte_unmap_unlock(dst_pte
, ptl
);
2399 pte_unmap_unlock(dst_pte
, ptl
);
2400 ClearPageDirty(page
);
2401 delete_from_page_cache(page
);
2406 shmem_inode_unacct_blocks(inode
, 1);
2410 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2412 struct vm_area_struct
*dst_vma
,
2413 unsigned long dst_addr
,
2414 unsigned long src_addr
,
2415 struct page
**pagep
)
2417 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2418 dst_addr
, src_addr
, false, pagep
);
2421 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2423 struct vm_area_struct
*dst_vma
,
2424 unsigned long dst_addr
)
2426 struct page
*page
= NULL
;
2428 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2429 dst_addr
, 0, true, &page
);
2433 static const struct inode_operations shmem_symlink_inode_operations
;
2434 static const struct inode_operations shmem_short_symlink_operations
;
2436 #ifdef CONFIG_TMPFS_XATTR
2437 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2439 #define shmem_initxattrs NULL
2443 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2444 loff_t pos
, unsigned len
, unsigned flags
,
2445 struct page
**pagep
, void **fsdata
)
2447 struct inode
*inode
= mapping
->host
;
2448 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2449 pgoff_t index
= pos
>> PAGE_SHIFT
;
2451 /* i_mutex is held by caller */
2452 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2453 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2454 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2456 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2460 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2464 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2465 loff_t pos
, unsigned len
, unsigned copied
,
2466 struct page
*page
, void *fsdata
)
2468 struct inode
*inode
= mapping
->host
;
2470 if (pos
+ copied
> inode
->i_size
)
2471 i_size_write(inode
, pos
+ copied
);
2473 if (!PageUptodate(page
)) {
2474 struct page
*head
= compound_head(page
);
2475 if (PageTransCompound(page
)) {
2478 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2479 if (head
+ i
== page
)
2481 clear_highpage(head
+ i
);
2482 flush_dcache_page(head
+ i
);
2485 if (copied
< PAGE_SIZE
) {
2486 unsigned from
= pos
& (PAGE_SIZE
- 1);
2487 zero_user_segments(page
, 0, from
,
2488 from
+ copied
, PAGE_SIZE
);
2490 SetPageUptodate(head
);
2492 set_page_dirty(page
);
2499 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2501 struct file
*file
= iocb
->ki_filp
;
2502 struct inode
*inode
= file_inode(file
);
2503 struct address_space
*mapping
= inode
->i_mapping
;
2505 unsigned long offset
;
2506 enum sgp_type sgp
= SGP_READ
;
2509 loff_t
*ppos
= &iocb
->ki_pos
;
2512 * Might this read be for a stacking filesystem? Then when reading
2513 * holes of a sparse file, we actually need to allocate those pages,
2514 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2516 if (!iter_is_iovec(to
))
2519 index
= *ppos
>> PAGE_SHIFT
;
2520 offset
= *ppos
& ~PAGE_MASK
;
2523 struct page
*page
= NULL
;
2525 unsigned long nr
, ret
;
2526 loff_t i_size
= i_size_read(inode
);
2528 end_index
= i_size
>> PAGE_SHIFT
;
2529 if (index
> end_index
)
2531 if (index
== end_index
) {
2532 nr
= i_size
& ~PAGE_MASK
;
2537 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2539 if (error
== -EINVAL
)
2544 if (sgp
== SGP_CACHE
)
2545 set_page_dirty(page
);
2550 * We must evaluate after, since reads (unlike writes)
2551 * are called without i_mutex protection against truncate
2554 i_size
= i_size_read(inode
);
2555 end_index
= i_size
>> PAGE_SHIFT
;
2556 if (index
== end_index
) {
2557 nr
= i_size
& ~PAGE_MASK
;
2568 * If users can be writing to this page using arbitrary
2569 * virtual addresses, take care about potential aliasing
2570 * before reading the page on the kernel side.
2572 if (mapping_writably_mapped(mapping
))
2573 flush_dcache_page(page
);
2575 * Mark the page accessed if we read the beginning.
2578 mark_page_accessed(page
);
2580 page
= ZERO_PAGE(0);
2585 * Ok, we have the page, and it's up-to-date, so
2586 * now we can copy it to user space...
2588 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2591 index
+= offset
>> PAGE_SHIFT
;
2592 offset
&= ~PAGE_MASK
;
2595 if (!iov_iter_count(to
))
2604 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2605 file_accessed(file
);
2606 return retval
? retval
: error
;
2610 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2612 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2613 pgoff_t index
, pgoff_t end
, int whence
)
2616 struct pagevec pvec
;
2617 pgoff_t indices
[PAGEVEC_SIZE
];
2621 pagevec_init(&pvec
);
2622 pvec
.nr
= 1; /* start small: we may be there already */
2624 pvec
.nr
= find_get_entries(mapping
, index
,
2625 pvec
.nr
, pvec
.pages
, indices
);
2627 if (whence
== SEEK_DATA
)
2631 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2632 if (index
< indices
[i
]) {
2633 if (whence
== SEEK_HOLE
) {
2639 page
= pvec
.pages
[i
];
2640 if (page
&& !xa_is_value(page
)) {
2641 if (!PageUptodate(page
))
2645 (page
&& whence
== SEEK_DATA
) ||
2646 (!page
&& whence
== SEEK_HOLE
)) {
2651 pagevec_remove_exceptionals(&pvec
);
2652 pagevec_release(&pvec
);
2653 pvec
.nr
= PAGEVEC_SIZE
;
2659 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2661 struct address_space
*mapping
= file
->f_mapping
;
2662 struct inode
*inode
= mapping
->host
;
2666 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2667 return generic_file_llseek_size(file
, offset
, whence
,
2668 MAX_LFS_FILESIZE
, i_size_read(inode
));
2670 /* We're holding i_mutex so we can access i_size directly */
2672 if (offset
< 0 || offset
>= inode
->i_size
)
2675 start
= offset
>> PAGE_SHIFT
;
2676 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2677 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2678 new_offset
<<= PAGE_SHIFT
;
2679 if (new_offset
> offset
) {
2680 if (new_offset
< inode
->i_size
)
2681 offset
= new_offset
;
2682 else if (whence
== SEEK_DATA
)
2685 offset
= inode
->i_size
;
2690 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2691 inode_unlock(inode
);
2695 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2698 struct inode
*inode
= file_inode(file
);
2699 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2700 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2701 struct shmem_falloc shmem_falloc
;
2702 pgoff_t start
, index
, end
;
2705 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2710 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2711 struct address_space
*mapping
= file
->f_mapping
;
2712 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2713 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2714 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2716 /* protected by i_mutex */
2717 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2722 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2723 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2724 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2725 spin_lock(&inode
->i_lock
);
2726 inode
->i_private
= &shmem_falloc
;
2727 spin_unlock(&inode
->i_lock
);
2729 if ((u64
)unmap_end
> (u64
)unmap_start
)
2730 unmap_mapping_range(mapping
, unmap_start
,
2731 1 + unmap_end
- unmap_start
, 0);
2732 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2733 /* No need to unmap again: hole-punching leaves COWed pages */
2735 spin_lock(&inode
->i_lock
);
2736 inode
->i_private
= NULL
;
2737 wake_up_all(&shmem_falloc_waitq
);
2738 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2739 spin_unlock(&inode
->i_lock
);
2744 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2745 error
= inode_newsize_ok(inode
, offset
+ len
);
2749 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2754 start
= offset
>> PAGE_SHIFT
;
2755 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2756 /* Try to avoid a swapstorm if len is impossible to satisfy */
2757 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2762 shmem_falloc
.waitq
= NULL
;
2763 shmem_falloc
.start
= start
;
2764 shmem_falloc
.next
= start
;
2765 shmem_falloc
.nr_falloced
= 0;
2766 shmem_falloc
.nr_unswapped
= 0;
2767 spin_lock(&inode
->i_lock
);
2768 inode
->i_private
= &shmem_falloc
;
2769 spin_unlock(&inode
->i_lock
);
2771 for (index
= start
; index
< end
; index
++) {
2775 * Good, the fallocate(2) manpage permits EINTR: we may have
2776 * been interrupted because we are using up too much memory.
2778 if (signal_pending(current
))
2780 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2783 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2785 /* Remove the !PageUptodate pages we added */
2786 if (index
> start
) {
2787 shmem_undo_range(inode
,
2788 (loff_t
)start
<< PAGE_SHIFT
,
2789 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2795 * Inform shmem_writepage() how far we have reached.
2796 * No need for lock or barrier: we have the page lock.
2798 shmem_falloc
.next
++;
2799 if (!PageUptodate(page
))
2800 shmem_falloc
.nr_falloced
++;
2803 * If !PageUptodate, leave it that way so that freeable pages
2804 * can be recognized if we need to rollback on error later.
2805 * But set_page_dirty so that memory pressure will swap rather
2806 * than free the pages we are allocating (and SGP_CACHE pages
2807 * might still be clean: we now need to mark those dirty too).
2809 set_page_dirty(page
);
2815 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2816 i_size_write(inode
, offset
+ len
);
2817 inode
->i_ctime
= current_time(inode
);
2819 spin_lock(&inode
->i_lock
);
2820 inode
->i_private
= NULL
;
2821 spin_unlock(&inode
->i_lock
);
2823 inode_unlock(inode
);
2827 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2829 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2831 buf
->f_type
= TMPFS_MAGIC
;
2832 buf
->f_bsize
= PAGE_SIZE
;
2833 buf
->f_namelen
= NAME_MAX
;
2834 if (sbinfo
->max_blocks
) {
2835 buf
->f_blocks
= sbinfo
->max_blocks
;
2837 buf
->f_bfree
= sbinfo
->max_blocks
-
2838 percpu_counter_sum(&sbinfo
->used_blocks
);
2840 if (sbinfo
->max_inodes
) {
2841 buf
->f_files
= sbinfo
->max_inodes
;
2842 buf
->f_ffree
= sbinfo
->free_inodes
;
2844 /* else leave those fields 0 like simple_statfs */
2849 * File creation. Allocate an inode, and we're done..
2852 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2854 struct inode
*inode
;
2855 int error
= -ENOSPC
;
2857 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2859 error
= simple_acl_create(dir
, inode
);
2862 error
= security_inode_init_security(inode
, dir
,
2864 shmem_initxattrs
, NULL
);
2865 if (error
&& error
!= -EOPNOTSUPP
)
2869 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2870 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2871 d_instantiate(dentry
, inode
);
2872 dget(dentry
); /* Extra count - pin the dentry in core */
2881 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2883 struct inode
*inode
;
2884 int error
= -ENOSPC
;
2886 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2888 error
= security_inode_init_security(inode
, dir
,
2890 shmem_initxattrs
, NULL
);
2891 if (error
&& error
!= -EOPNOTSUPP
)
2893 error
= simple_acl_create(dir
, inode
);
2896 d_tmpfile(dentry
, inode
);
2904 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2908 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2914 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2917 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2923 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2925 struct inode
*inode
= d_inode(old_dentry
);
2929 * No ordinary (disk based) filesystem counts links as inodes;
2930 * but each new link needs a new dentry, pinning lowmem, and
2931 * tmpfs dentries cannot be pruned until they are unlinked.
2932 * But if an O_TMPFILE file is linked into the tmpfs, the
2933 * first link must skip that, to get the accounting right.
2935 if (inode
->i_nlink
) {
2936 ret
= shmem_reserve_inode(inode
->i_sb
);
2941 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2942 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2944 ihold(inode
); /* New dentry reference */
2945 dget(dentry
); /* Extra pinning count for the created dentry */
2946 d_instantiate(dentry
, inode
);
2951 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2953 struct inode
*inode
= d_inode(dentry
);
2955 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2956 shmem_free_inode(inode
->i_sb
);
2958 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2959 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2961 dput(dentry
); /* Undo the count from "create" - this does all the work */
2965 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2967 if (!simple_empty(dentry
))
2970 drop_nlink(d_inode(dentry
));
2972 return shmem_unlink(dir
, dentry
);
2975 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2977 bool old_is_dir
= d_is_dir(old_dentry
);
2978 bool new_is_dir
= d_is_dir(new_dentry
);
2980 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2982 drop_nlink(old_dir
);
2985 drop_nlink(new_dir
);
2989 old_dir
->i_ctime
= old_dir
->i_mtime
=
2990 new_dir
->i_ctime
= new_dir
->i_mtime
=
2991 d_inode(old_dentry
)->i_ctime
=
2992 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
2997 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2999 struct dentry
*whiteout
;
3002 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3006 error
= shmem_mknod(old_dir
, whiteout
,
3007 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3013 * Cheat and hash the whiteout while the old dentry is still in
3014 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3016 * d_lookup() will consistently find one of them at this point,
3017 * not sure which one, but that isn't even important.
3024 * The VFS layer already does all the dentry stuff for rename,
3025 * we just have to decrement the usage count for the target if
3026 * it exists so that the VFS layer correctly free's it when it
3029 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3031 struct inode
*inode
= d_inode(old_dentry
);
3032 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3034 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3037 if (flags
& RENAME_EXCHANGE
)
3038 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3040 if (!simple_empty(new_dentry
))
3043 if (flags
& RENAME_WHITEOUT
) {
3046 error
= shmem_whiteout(old_dir
, old_dentry
);
3051 if (d_really_is_positive(new_dentry
)) {
3052 (void) shmem_unlink(new_dir
, new_dentry
);
3053 if (they_are_dirs
) {
3054 drop_nlink(d_inode(new_dentry
));
3055 drop_nlink(old_dir
);
3057 } else if (they_are_dirs
) {
3058 drop_nlink(old_dir
);
3062 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3063 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3064 old_dir
->i_ctime
= old_dir
->i_mtime
=
3065 new_dir
->i_ctime
= new_dir
->i_mtime
=
3066 inode
->i_ctime
= current_time(old_dir
);
3070 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3074 struct inode
*inode
;
3077 len
= strlen(symname
) + 1;
3078 if (len
> PAGE_SIZE
)
3079 return -ENAMETOOLONG
;
3081 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3086 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3087 shmem_initxattrs
, NULL
);
3088 if (error
&& error
!= -EOPNOTSUPP
) {
3093 inode
->i_size
= len
-1;
3094 if (len
<= SHORT_SYMLINK_LEN
) {
3095 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3096 if (!inode
->i_link
) {
3100 inode
->i_op
= &shmem_short_symlink_operations
;
3102 inode_nohighmem(inode
);
3103 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3108 inode
->i_mapping
->a_ops
= &shmem_aops
;
3109 inode
->i_op
= &shmem_symlink_inode_operations
;
3110 memcpy(page_address(page
), symname
, len
);
3111 SetPageUptodate(page
);
3112 set_page_dirty(page
);
3116 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3117 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3118 d_instantiate(dentry
, inode
);
3123 static void shmem_put_link(void *arg
)
3125 mark_page_accessed(arg
);
3129 static const char *shmem_get_link(struct dentry
*dentry
,
3130 struct inode
*inode
,
3131 struct delayed_call
*done
)
3133 struct page
*page
= NULL
;
3136 page
= find_get_page(inode
->i_mapping
, 0);
3138 return ERR_PTR(-ECHILD
);
3139 if (!PageUptodate(page
)) {
3141 return ERR_PTR(-ECHILD
);
3144 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3146 return ERR_PTR(error
);
3149 set_delayed_call(done
, shmem_put_link
, page
);
3150 return page_address(page
);
3153 #ifdef CONFIG_TMPFS_XATTR
3155 * Superblocks without xattr inode operations may get some security.* xattr
3156 * support from the LSM "for free". As soon as we have any other xattrs
3157 * like ACLs, we also need to implement the security.* handlers at
3158 * filesystem level, though.
3162 * Callback for security_inode_init_security() for acquiring xattrs.
3164 static int shmem_initxattrs(struct inode
*inode
,
3165 const struct xattr
*xattr_array
,
3168 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3169 const struct xattr
*xattr
;
3170 struct simple_xattr
*new_xattr
;
3173 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3174 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3178 len
= strlen(xattr
->name
) + 1;
3179 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3181 if (!new_xattr
->name
) {
3186 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3187 XATTR_SECURITY_PREFIX_LEN
);
3188 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3191 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3197 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3198 struct dentry
*unused
, struct inode
*inode
,
3199 const char *name
, void *buffer
, size_t size
)
3201 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3203 name
= xattr_full_name(handler
, name
);
3204 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3207 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3208 struct dentry
*unused
, struct inode
*inode
,
3209 const char *name
, const void *value
,
3210 size_t size
, int flags
)
3212 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3214 name
= xattr_full_name(handler
, name
);
3215 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3218 static const struct xattr_handler shmem_security_xattr_handler
= {
3219 .prefix
= XATTR_SECURITY_PREFIX
,
3220 .get
= shmem_xattr_handler_get
,
3221 .set
= shmem_xattr_handler_set
,
3224 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3225 .prefix
= XATTR_TRUSTED_PREFIX
,
3226 .get
= shmem_xattr_handler_get
,
3227 .set
= shmem_xattr_handler_set
,
3230 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3231 #ifdef CONFIG_TMPFS_POSIX_ACL
3232 &posix_acl_access_xattr_handler
,
3233 &posix_acl_default_xattr_handler
,
3235 &shmem_security_xattr_handler
,
3236 &shmem_trusted_xattr_handler
,
3240 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3242 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3243 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3245 #endif /* CONFIG_TMPFS_XATTR */
3247 static const struct inode_operations shmem_short_symlink_operations
= {
3248 .get_link
= simple_get_link
,
3249 #ifdef CONFIG_TMPFS_XATTR
3250 .listxattr
= shmem_listxattr
,
3254 static const struct inode_operations shmem_symlink_inode_operations
= {
3255 .get_link
= shmem_get_link
,
3256 #ifdef CONFIG_TMPFS_XATTR
3257 .listxattr
= shmem_listxattr
,
3261 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3263 return ERR_PTR(-ESTALE
);
3266 static int shmem_match(struct inode
*ino
, void *vfh
)
3270 inum
= (inum
<< 32) | fh
[1];
3271 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3274 /* Find any alias of inode, but prefer a hashed alias */
3275 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3277 struct dentry
*alias
= d_find_alias(inode
);
3279 return alias
?: d_find_any_alias(inode
);
3283 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3284 struct fid
*fid
, int fh_len
, int fh_type
)
3286 struct inode
*inode
;
3287 struct dentry
*dentry
= NULL
;
3294 inum
= (inum
<< 32) | fid
->raw
[1];
3296 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3297 shmem_match
, fid
->raw
);
3299 dentry
= shmem_find_alias(inode
);
3306 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3307 struct inode
*parent
)
3311 return FILEID_INVALID
;
3314 if (inode_unhashed(inode
)) {
3315 /* Unfortunately insert_inode_hash is not idempotent,
3316 * so as we hash inodes here rather than at creation
3317 * time, we need a lock to ensure we only try
3320 static DEFINE_SPINLOCK(lock
);
3322 if (inode_unhashed(inode
))
3323 __insert_inode_hash(inode
,
3324 inode
->i_ino
+ inode
->i_generation
);
3328 fh
[0] = inode
->i_generation
;
3329 fh
[1] = inode
->i_ino
;
3330 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3336 static const struct export_operations shmem_export_ops
= {
3337 .get_parent
= shmem_get_parent
,
3338 .encode_fh
= shmem_encode_fh
,
3339 .fh_to_dentry
= shmem_fh_to_dentry
,
3353 static const struct constant_table shmem_param_enums_huge
[] = {
3354 {"never", SHMEM_HUGE_NEVER
},
3355 {"always", SHMEM_HUGE_ALWAYS
},
3356 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3357 {"advise", SHMEM_HUGE_ADVISE
},
3361 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3362 fsparam_u32 ("gid", Opt_gid
),
3363 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3364 fsparam_u32oct("mode", Opt_mode
),
3365 fsparam_string("mpol", Opt_mpol
),
3366 fsparam_string("nr_blocks", Opt_nr_blocks
),
3367 fsparam_string("nr_inodes", Opt_nr_inodes
),
3368 fsparam_string("size", Opt_size
),
3369 fsparam_u32 ("uid", Opt_uid
),
3373 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3375 struct shmem_options
*ctx
= fc
->fs_private
;
3376 struct fs_parse_result result
;
3377 unsigned long long size
;
3381 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3387 size
= memparse(param
->string
, &rest
);
3389 size
<<= PAGE_SHIFT
;
3390 size
*= totalram_pages();
3396 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3397 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3400 ctx
->blocks
= memparse(param
->string
, &rest
);
3403 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3406 ctx
->inodes
= memparse(param
->string
, &rest
);
3409 ctx
->seen
|= SHMEM_SEEN_INODES
;
3412 ctx
->mode
= result
.uint_32
& 07777;
3415 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3416 if (!uid_valid(ctx
->uid
))
3420 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3421 if (!gid_valid(ctx
->gid
))
3425 ctx
->huge
= result
.uint_32
;
3426 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3427 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3428 has_transparent_hugepage()))
3429 goto unsupported_parameter
;
3430 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3433 if (IS_ENABLED(CONFIG_NUMA
)) {
3434 mpol_put(ctx
->mpol
);
3436 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3440 goto unsupported_parameter
;
3444 unsupported_parameter
:
3445 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3447 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3450 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3452 char *options
= data
;
3455 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3460 while (options
!= NULL
) {
3461 char *this_char
= options
;
3464 * NUL-terminate this option: unfortunately,
3465 * mount options form a comma-separated list,
3466 * but mpol's nodelist may also contain commas.
3468 options
= strchr(options
, ',');
3469 if (options
== NULL
)
3472 if (!isdigit(*options
)) {
3478 char *value
= strchr(this_char
,'=');
3484 len
= strlen(value
);
3486 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3495 * Reconfigure a shmem filesystem.
3497 * Note that we disallow change from limited->unlimited blocks/inodes while any
3498 * are in use; but we must separately disallow unlimited->limited, because in
3499 * that case we have no record of how much is already in use.
3501 static int shmem_reconfigure(struct fs_context
*fc
)
3503 struct shmem_options
*ctx
= fc
->fs_private
;
3504 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3505 unsigned long inodes
;
3508 spin_lock(&sbinfo
->stat_lock
);
3509 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3510 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3511 if (!sbinfo
->max_blocks
) {
3512 err
= "Cannot retroactively limit size";
3515 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3517 err
= "Too small a size for current use";
3521 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3522 if (!sbinfo
->max_inodes
) {
3523 err
= "Cannot retroactively limit inodes";
3526 if (ctx
->inodes
< inodes
) {
3527 err
= "Too few inodes for current use";
3532 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3533 sbinfo
->huge
= ctx
->huge
;
3534 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3535 sbinfo
->max_blocks
= ctx
->blocks
;
3536 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3537 sbinfo
->max_inodes
= ctx
->inodes
;
3538 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3542 * Preserve previous mempolicy unless mpol remount option was specified.
3545 mpol_put(sbinfo
->mpol
);
3546 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3549 spin_unlock(&sbinfo
->stat_lock
);
3552 spin_unlock(&sbinfo
->stat_lock
);
3553 return invalfc(fc
, "%s", err
);
3556 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3558 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3560 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3561 seq_printf(seq
, ",size=%luk",
3562 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3563 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3564 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3565 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3566 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3567 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3568 seq_printf(seq
, ",uid=%u",
3569 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3570 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3571 seq_printf(seq
, ",gid=%u",
3572 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3573 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3574 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3576 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3578 shmem_show_mpol(seq
, sbinfo
->mpol
);
3582 #endif /* CONFIG_TMPFS */
3584 static void shmem_put_super(struct super_block
*sb
)
3586 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3588 percpu_counter_destroy(&sbinfo
->used_blocks
);
3589 mpol_put(sbinfo
->mpol
);
3591 sb
->s_fs_info
= NULL
;
3594 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3596 struct shmem_options
*ctx
= fc
->fs_private
;
3597 struct inode
*inode
;
3598 struct shmem_sb_info
*sbinfo
;
3601 /* Round up to L1_CACHE_BYTES to resist false sharing */
3602 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3603 L1_CACHE_BYTES
), GFP_KERNEL
);
3607 sb
->s_fs_info
= sbinfo
;
3611 * Per default we only allow half of the physical ram per
3612 * tmpfs instance, limiting inodes to one per page of lowmem;
3613 * but the internal instance is left unlimited.
3615 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3616 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3617 ctx
->blocks
= shmem_default_max_blocks();
3618 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3619 ctx
->inodes
= shmem_default_max_inodes();
3621 sb
->s_flags
|= SB_NOUSER
;
3623 sb
->s_export_op
= &shmem_export_ops
;
3624 sb
->s_flags
|= SB_NOSEC
;
3626 sb
->s_flags
|= SB_NOUSER
;
3628 sbinfo
->max_blocks
= ctx
->blocks
;
3629 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3630 sbinfo
->uid
= ctx
->uid
;
3631 sbinfo
->gid
= ctx
->gid
;
3632 sbinfo
->mode
= ctx
->mode
;
3633 sbinfo
->huge
= ctx
->huge
;
3634 sbinfo
->mpol
= ctx
->mpol
;
3637 spin_lock_init(&sbinfo
->stat_lock
);
3638 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3640 spin_lock_init(&sbinfo
->shrinklist_lock
);
3641 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3643 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3644 sb
->s_blocksize
= PAGE_SIZE
;
3645 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3646 sb
->s_magic
= TMPFS_MAGIC
;
3647 sb
->s_op
= &shmem_ops
;
3648 sb
->s_time_gran
= 1;
3649 #ifdef CONFIG_TMPFS_XATTR
3650 sb
->s_xattr
= shmem_xattr_handlers
;
3652 #ifdef CONFIG_TMPFS_POSIX_ACL
3653 sb
->s_flags
|= SB_POSIXACL
;
3655 uuid_gen(&sb
->s_uuid
);
3657 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3660 inode
->i_uid
= sbinfo
->uid
;
3661 inode
->i_gid
= sbinfo
->gid
;
3662 sb
->s_root
= d_make_root(inode
);
3668 shmem_put_super(sb
);
3672 static int shmem_get_tree(struct fs_context
*fc
)
3674 return get_tree_nodev(fc
, shmem_fill_super
);
3677 static void shmem_free_fc(struct fs_context
*fc
)
3679 struct shmem_options
*ctx
= fc
->fs_private
;
3682 mpol_put(ctx
->mpol
);
3687 static const struct fs_context_operations shmem_fs_context_ops
= {
3688 .free
= shmem_free_fc
,
3689 .get_tree
= shmem_get_tree
,
3691 .parse_monolithic
= shmem_parse_options
,
3692 .parse_param
= shmem_parse_one
,
3693 .reconfigure
= shmem_reconfigure
,
3697 static struct kmem_cache
*shmem_inode_cachep
;
3699 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3701 struct shmem_inode_info
*info
;
3702 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3705 return &info
->vfs_inode
;
3708 static void shmem_free_in_core_inode(struct inode
*inode
)
3710 if (S_ISLNK(inode
->i_mode
))
3711 kfree(inode
->i_link
);
3712 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3715 static void shmem_destroy_inode(struct inode
*inode
)
3717 if (S_ISREG(inode
->i_mode
))
3718 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3721 static void shmem_init_inode(void *foo
)
3723 struct shmem_inode_info
*info
= foo
;
3724 inode_init_once(&info
->vfs_inode
);
3727 static void shmem_init_inodecache(void)
3729 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3730 sizeof(struct shmem_inode_info
),
3731 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3734 static void shmem_destroy_inodecache(void)
3736 kmem_cache_destroy(shmem_inode_cachep
);
3739 static const struct address_space_operations shmem_aops
= {
3740 .writepage
= shmem_writepage
,
3741 .set_page_dirty
= __set_page_dirty_no_writeback
,
3743 .write_begin
= shmem_write_begin
,
3744 .write_end
= shmem_write_end
,
3746 #ifdef CONFIG_MIGRATION
3747 .migratepage
= migrate_page
,
3749 .error_remove_page
= generic_error_remove_page
,
3752 static const struct file_operations shmem_file_operations
= {
3754 .get_unmapped_area
= shmem_get_unmapped_area
,
3756 .llseek
= shmem_file_llseek
,
3757 .read_iter
= shmem_file_read_iter
,
3758 .write_iter
= generic_file_write_iter
,
3759 .fsync
= noop_fsync
,
3760 .splice_read
= generic_file_splice_read
,
3761 .splice_write
= iter_file_splice_write
,
3762 .fallocate
= shmem_fallocate
,
3766 static const struct inode_operations shmem_inode_operations
= {
3767 .getattr
= shmem_getattr
,
3768 .setattr
= shmem_setattr
,
3769 #ifdef CONFIG_TMPFS_XATTR
3770 .listxattr
= shmem_listxattr
,
3771 .set_acl
= simple_set_acl
,
3775 static const struct inode_operations shmem_dir_inode_operations
= {
3777 .create
= shmem_create
,
3778 .lookup
= simple_lookup
,
3780 .unlink
= shmem_unlink
,
3781 .symlink
= shmem_symlink
,
3782 .mkdir
= shmem_mkdir
,
3783 .rmdir
= shmem_rmdir
,
3784 .mknod
= shmem_mknod
,
3785 .rename
= shmem_rename2
,
3786 .tmpfile
= shmem_tmpfile
,
3788 #ifdef CONFIG_TMPFS_XATTR
3789 .listxattr
= shmem_listxattr
,
3791 #ifdef CONFIG_TMPFS_POSIX_ACL
3792 .setattr
= shmem_setattr
,
3793 .set_acl
= simple_set_acl
,
3797 static const struct inode_operations shmem_special_inode_operations
= {
3798 #ifdef CONFIG_TMPFS_XATTR
3799 .listxattr
= shmem_listxattr
,
3801 #ifdef CONFIG_TMPFS_POSIX_ACL
3802 .setattr
= shmem_setattr
,
3803 .set_acl
= simple_set_acl
,
3807 static const struct super_operations shmem_ops
= {
3808 .alloc_inode
= shmem_alloc_inode
,
3809 .free_inode
= shmem_free_in_core_inode
,
3810 .destroy_inode
= shmem_destroy_inode
,
3812 .statfs
= shmem_statfs
,
3813 .show_options
= shmem_show_options
,
3815 .evict_inode
= shmem_evict_inode
,
3816 .drop_inode
= generic_delete_inode
,
3817 .put_super
= shmem_put_super
,
3818 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3819 .nr_cached_objects
= shmem_unused_huge_count
,
3820 .free_cached_objects
= shmem_unused_huge_scan
,
3824 static const struct vm_operations_struct shmem_vm_ops
= {
3825 .fault
= shmem_fault
,
3826 .map_pages
= filemap_map_pages
,
3828 .set_policy
= shmem_set_policy
,
3829 .get_policy
= shmem_get_policy
,
3833 int shmem_init_fs_context(struct fs_context
*fc
)
3835 struct shmem_options
*ctx
;
3837 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3841 ctx
->mode
= 0777 | S_ISVTX
;
3842 ctx
->uid
= current_fsuid();
3843 ctx
->gid
= current_fsgid();
3845 fc
->fs_private
= ctx
;
3846 fc
->ops
= &shmem_fs_context_ops
;
3850 static struct file_system_type shmem_fs_type
= {
3851 .owner
= THIS_MODULE
,
3853 .init_fs_context
= shmem_init_fs_context
,
3855 .parameters
= shmem_fs_parameters
,
3857 .kill_sb
= kill_litter_super
,
3858 .fs_flags
= FS_USERNS_MOUNT
,
3861 int __init
shmem_init(void)
3865 shmem_init_inodecache();
3867 error
= register_filesystem(&shmem_fs_type
);
3869 pr_err("Could not register tmpfs\n");
3873 shm_mnt
= kern_mount(&shmem_fs_type
);
3874 if (IS_ERR(shm_mnt
)) {
3875 error
= PTR_ERR(shm_mnt
);
3876 pr_err("Could not kern_mount tmpfs\n");
3880 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3881 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3882 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3884 shmem_huge
= 0; /* just in case it was patched */
3889 unregister_filesystem(&shmem_fs_type
);
3891 shmem_destroy_inodecache();
3892 shm_mnt
= ERR_PTR(error
);
3896 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3897 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3898 struct kobj_attribute
*attr
, char *buf
)
3900 static const int values
[] = {
3902 SHMEM_HUGE_WITHIN_SIZE
,
3910 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3911 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3913 count
+= sprintf(buf
+ count
, fmt
,
3914 shmem_format_huge(values
[i
]));
3916 buf
[count
- 1] = '\n';
3920 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3921 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3926 if (count
+ 1 > sizeof(tmp
))
3928 memcpy(tmp
, buf
, count
);
3930 if (count
&& tmp
[count
- 1] == '\n')
3931 tmp
[count
- 1] = '\0';
3933 huge
= shmem_parse_huge(tmp
);
3934 if (huge
== -EINVAL
)
3936 if (!has_transparent_hugepage() &&
3937 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3941 if (shmem_huge
> SHMEM_HUGE_DENY
)
3942 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3946 struct kobj_attribute shmem_enabled_attr
=
3947 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3948 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3950 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3951 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
3953 struct inode
*inode
= file_inode(vma
->vm_file
);
3954 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3958 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
3959 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
3961 if (shmem_huge
== SHMEM_HUGE_FORCE
)
3963 if (shmem_huge
== SHMEM_HUGE_DENY
)
3965 switch (sbinfo
->huge
) {
3966 case SHMEM_HUGE_NEVER
:
3968 case SHMEM_HUGE_ALWAYS
:
3970 case SHMEM_HUGE_WITHIN_SIZE
:
3971 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
3972 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
3973 if (i_size
>= HPAGE_PMD_SIZE
&&
3974 i_size
>> PAGE_SHIFT
>= off
)
3977 case SHMEM_HUGE_ADVISE
:
3978 /* TODO: implement fadvise() hints */
3979 return (vma
->vm_flags
& VM_HUGEPAGE
);
3985 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
3987 #else /* !CONFIG_SHMEM */
3990 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3992 * This is intended for small system where the benefits of the full
3993 * shmem code (swap-backed and resource-limited) are outweighed by
3994 * their complexity. On systems without swap this code should be
3995 * effectively equivalent, but much lighter weight.
3998 static struct file_system_type shmem_fs_type
= {
4000 .init_fs_context
= ramfs_init_fs_context
,
4001 .parameters
= ramfs_fs_parameters
,
4002 .kill_sb
= kill_litter_super
,
4003 .fs_flags
= FS_USERNS_MOUNT
,
4006 int __init
shmem_init(void)
4008 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4010 shm_mnt
= kern_mount(&shmem_fs_type
);
4011 BUG_ON(IS_ERR(shm_mnt
));
4016 int shmem_unuse(unsigned int type
, bool frontswap
,
4017 unsigned long *fs_pages_to_unuse
)
4022 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4027 void shmem_unlock_mapping(struct address_space
*mapping
)
4032 unsigned long shmem_get_unmapped_area(struct file
*file
,
4033 unsigned long addr
, unsigned long len
,
4034 unsigned long pgoff
, unsigned long flags
)
4036 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4040 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4042 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4044 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4046 #define shmem_vm_ops generic_file_vm_ops
4047 #define shmem_file_operations ramfs_file_operations
4048 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4049 #define shmem_acct_size(flags, size) 0
4050 #define shmem_unacct_size(flags, size) do {} while (0)
4052 #endif /* CONFIG_SHMEM */
4056 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4057 unsigned long flags
, unsigned int i_flags
)
4059 struct inode
*inode
;
4063 return ERR_CAST(mnt
);
4065 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4066 return ERR_PTR(-EINVAL
);
4068 if (shmem_acct_size(flags
, size
))
4069 return ERR_PTR(-ENOMEM
);
4071 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4073 if (unlikely(!inode
)) {
4074 shmem_unacct_size(flags
, size
);
4075 return ERR_PTR(-ENOSPC
);
4077 inode
->i_flags
|= i_flags
;
4078 inode
->i_size
= size
;
4079 clear_nlink(inode
); /* It is unlinked */
4080 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4082 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4083 &shmem_file_operations
);
4090 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4091 * kernel internal. There will be NO LSM permission checks against the
4092 * underlying inode. So users of this interface must do LSM checks at a
4093 * higher layer. The users are the big_key and shm implementations. LSM
4094 * checks are provided at the key or shm level rather than the inode.
4095 * @name: name for dentry (to be seen in /proc/<pid>/maps
4096 * @size: size to be set for the file
4097 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4099 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4101 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4105 * shmem_file_setup - get an unlinked file living in tmpfs
4106 * @name: name for dentry (to be seen in /proc/<pid>/maps
4107 * @size: size to be set for the file
4108 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4110 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4112 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4114 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4117 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4118 * @mnt: the tmpfs mount where the file will be created
4119 * @name: name for dentry (to be seen in /proc/<pid>/maps
4120 * @size: size to be set for the file
4121 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4123 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4124 loff_t size
, unsigned long flags
)
4126 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4128 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4131 * shmem_zero_setup - setup a shared anonymous mapping
4132 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4134 int shmem_zero_setup(struct vm_area_struct
*vma
)
4137 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4140 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4141 * between XFS directory reading and selinux: since this file is only
4142 * accessible to the user through its mapping, use S_PRIVATE flag to
4143 * bypass file security, in the same way as shmem_kernel_file_setup().
4145 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4147 return PTR_ERR(file
);
4151 vma
->vm_file
= file
;
4152 vma
->vm_ops
= &shmem_vm_ops
;
4154 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
4155 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4156 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4157 khugepaged_enter(vma
, vma
->vm_flags
);
4164 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4165 * @mapping: the page's address_space
4166 * @index: the page index
4167 * @gfp: the page allocator flags to use if allocating
4169 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4170 * with any new page allocations done using the specified allocation flags.
4171 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4172 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4173 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4175 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4176 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4178 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4179 pgoff_t index
, gfp_t gfp
)
4182 struct inode
*inode
= mapping
->host
;
4186 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4187 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4188 gfp
, NULL
, NULL
, NULL
);
4190 page
= ERR_PTR(error
);
4196 * The tiny !SHMEM case uses ramfs without swap
4198 return read_cache_page_gfp(mapping
, index
, gfp
);
4201 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);