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>
88 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
91 /* Pretend that each entry is of this size in directory's i_size */
92 #define BOGO_DIRENT_SIZE 20
94 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
95 #define SHORT_SYMLINK_LEN 128
98 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
99 * inode->i_private (with i_mutex making sure that it has only one user at
100 * a time): we would prefer not to enlarge the shmem inode just for that.
102 struct shmem_falloc
{
103 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
104 pgoff_t start
; /* start of range currently being fallocated */
105 pgoff_t next
; /* the next page offset to be fallocated */
106 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
107 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
110 struct shmem_options
{
111 unsigned long long blocks
;
112 unsigned long long inodes
;
113 struct mempolicy
*mpol
;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
123 #define SHMEM_SEEN_INUMS 8
127 static unsigned long shmem_default_max_blocks(void)
129 return totalram_pages() / 2;
132 static unsigned long shmem_default_max_inodes(void)
134 unsigned long nr_pages
= totalram_pages();
136 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
140 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
141 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
142 struct shmem_inode_info
*info
, pgoff_t index
);
143 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
144 struct page
**pagep
, enum sgp_type sgp
,
145 gfp_t gfp
, struct vm_area_struct
*vma
,
146 vm_fault_t
*fault_type
);
147 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
148 struct page
**pagep
, enum sgp_type sgp
,
149 gfp_t gfp
, struct vm_area_struct
*vma
,
150 struct vm_fault
*vmf
, vm_fault_t
*fault_type
);
152 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
153 struct page
**pagep
, enum sgp_type sgp
)
155 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
156 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
159 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
161 return sb
->s_fs_info
;
165 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
166 * for shared memory and for shared anonymous (/dev/zero) mappings
167 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
168 * consistent with the pre-accounting of private mappings ...
170 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
172 return (flags
& VM_NORESERVE
) ?
173 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
176 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
178 if (!(flags
& VM_NORESERVE
))
179 vm_unacct_memory(VM_ACCT(size
));
182 static inline int shmem_reacct_size(unsigned long flags
,
183 loff_t oldsize
, loff_t newsize
)
185 if (!(flags
& VM_NORESERVE
)) {
186 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
187 return security_vm_enough_memory_mm(current
->mm
,
188 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
189 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
190 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
196 * ... whereas tmpfs objects are accounted incrementally as
197 * pages are allocated, in order to allow large sparse files.
198 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
199 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201 static inline int shmem_acct_block(unsigned long flags
, long pages
)
203 if (!(flags
& VM_NORESERVE
))
206 return security_vm_enough_memory_mm(current
->mm
,
207 pages
* VM_ACCT(PAGE_SIZE
));
210 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
212 if (flags
& VM_NORESERVE
)
213 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
216 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
218 struct shmem_inode_info
*info
= SHMEM_I(inode
);
219 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
221 if (shmem_acct_block(info
->flags
, pages
))
224 if (sbinfo
->max_blocks
) {
225 if (percpu_counter_compare(&sbinfo
->used_blocks
,
226 sbinfo
->max_blocks
- pages
) > 0)
228 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
234 shmem_unacct_blocks(info
->flags
, pages
);
238 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
240 struct shmem_inode_info
*info
= SHMEM_I(inode
);
241 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
243 if (sbinfo
->max_blocks
)
244 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
245 shmem_unacct_blocks(info
->flags
, pages
);
248 static const struct super_operations shmem_ops
;
249 const struct address_space_operations shmem_aops
;
250 static const struct file_operations shmem_file_operations
;
251 static const struct inode_operations shmem_inode_operations
;
252 static const struct inode_operations shmem_dir_inode_operations
;
253 static const struct inode_operations shmem_special_inode_operations
;
254 static const struct vm_operations_struct shmem_vm_ops
;
255 static struct file_system_type shmem_fs_type
;
257 bool vma_is_shmem(struct vm_area_struct
*vma
)
259 return vma
->vm_ops
== &shmem_vm_ops
;
262 static LIST_HEAD(shmem_swaplist
);
263 static DEFINE_MUTEX(shmem_swaplist_mutex
);
266 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
267 * produces a novel ino for the newly allocated inode.
269 * It may also be called when making a hard link to permit the space needed by
270 * each dentry. However, in that case, no new inode number is needed since that
271 * internally draws from another pool of inode numbers (currently global
272 * get_next_ino()). This case is indicated by passing NULL as inop.
274 #define SHMEM_INO_BATCH 1024
275 static int shmem_reserve_inode(struct super_block
*sb
, ino_t
*inop
)
277 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
280 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
281 spin_lock(&sbinfo
->stat_lock
);
282 if (sbinfo
->max_inodes
) {
283 if (!sbinfo
->free_inodes
) {
284 spin_unlock(&sbinfo
->stat_lock
);
287 sbinfo
->free_inodes
--;
290 ino
= sbinfo
->next_ino
++;
291 if (unlikely(is_zero_ino(ino
)))
292 ino
= sbinfo
->next_ino
++;
293 if (unlikely(!sbinfo
->full_inums
&&
296 * Emulate get_next_ino uint wraparound for
299 if (IS_ENABLED(CONFIG_64BIT
))
300 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
301 __func__
, MINOR(sb
->s_dev
));
302 sbinfo
->next_ino
= 1;
303 ino
= sbinfo
->next_ino
++;
307 spin_unlock(&sbinfo
->stat_lock
);
310 * __shmem_file_setup, one of our callers, is lock-free: it
311 * doesn't hold stat_lock in shmem_reserve_inode since
312 * max_inodes is always 0, and is called from potentially
313 * unknown contexts. As such, use a per-cpu batched allocator
314 * which doesn't require the per-sb stat_lock unless we are at
315 * the batch boundary.
317 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
318 * shmem mounts are not exposed to userspace, so we don't need
319 * to worry about things like glibc compatibility.
322 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
324 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
325 spin_lock(&sbinfo
->stat_lock
);
326 ino
= sbinfo
->next_ino
;
327 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
328 spin_unlock(&sbinfo
->stat_lock
);
329 if (unlikely(is_zero_ino(ino
)))
340 static void shmem_free_inode(struct super_block
*sb
)
342 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
343 if (sbinfo
->max_inodes
) {
344 spin_lock(&sbinfo
->stat_lock
);
345 sbinfo
->free_inodes
++;
346 spin_unlock(&sbinfo
->stat_lock
);
351 * shmem_recalc_inode - recalculate the block usage of an inode
352 * @inode: inode to recalc
354 * We have to calculate the free blocks since the mm can drop
355 * undirtied hole pages behind our back.
357 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
358 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
360 * It has to be called with the spinlock held.
362 static void shmem_recalc_inode(struct inode
*inode
)
364 struct shmem_inode_info
*info
= SHMEM_I(inode
);
367 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
369 info
->alloced
-= freed
;
370 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
371 shmem_inode_unacct_blocks(inode
, freed
);
375 bool shmem_charge(struct inode
*inode
, long pages
)
377 struct shmem_inode_info
*info
= SHMEM_I(inode
);
380 if (!shmem_inode_acct_block(inode
, pages
))
383 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
384 inode
->i_mapping
->nrpages
+= pages
;
386 spin_lock_irqsave(&info
->lock
, flags
);
387 info
->alloced
+= pages
;
388 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
389 shmem_recalc_inode(inode
);
390 spin_unlock_irqrestore(&info
->lock
, flags
);
395 void shmem_uncharge(struct inode
*inode
, long pages
)
397 struct shmem_inode_info
*info
= SHMEM_I(inode
);
400 /* nrpages adjustment done by __delete_from_page_cache() or caller */
402 spin_lock_irqsave(&info
->lock
, flags
);
403 info
->alloced
-= pages
;
404 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
405 shmem_recalc_inode(inode
);
406 spin_unlock_irqrestore(&info
->lock
, flags
);
408 shmem_inode_unacct_blocks(inode
, pages
);
412 * Replace item expected in xarray by a new item, while holding xa_lock.
414 static int shmem_replace_entry(struct address_space
*mapping
,
415 pgoff_t index
, void *expected
, void *replacement
)
417 XA_STATE(xas
, &mapping
->i_pages
, index
);
420 VM_BUG_ON(!expected
);
421 VM_BUG_ON(!replacement
);
422 item
= xas_load(&xas
);
423 if (item
!= expected
)
425 xas_store(&xas
, replacement
);
430 * Sometimes, before we decide whether to proceed or to fail, we must check
431 * that an entry was not already brought back from swap by a racing thread.
433 * Checking page is not enough: by the time a SwapCache page is locked, it
434 * might be reused, and again be SwapCache, using the same swap as before.
436 static bool shmem_confirm_swap(struct address_space
*mapping
,
437 pgoff_t index
, swp_entry_t swap
)
439 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
443 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
446 * disables huge pages for the mount;
448 * enables huge pages for the mount;
449 * SHMEM_HUGE_WITHIN_SIZE:
450 * only allocate huge pages if the page will be fully within i_size,
451 * also respect fadvise()/madvise() hints;
453 * only allocate huge pages if requested with fadvise()/madvise();
456 #define SHMEM_HUGE_NEVER 0
457 #define SHMEM_HUGE_ALWAYS 1
458 #define SHMEM_HUGE_WITHIN_SIZE 2
459 #define SHMEM_HUGE_ADVISE 3
463 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
466 * disables huge on shm_mnt and all mounts, for emergency use;
468 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
471 #define SHMEM_HUGE_DENY (-1)
472 #define SHMEM_HUGE_FORCE (-2)
474 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
475 /* ifdef here to avoid bloating shmem.o when not necessary */
477 static int shmem_huge __read_mostly
;
479 #if defined(CONFIG_SYSFS)
480 static int shmem_parse_huge(const char *str
)
482 if (!strcmp(str
, "never"))
483 return SHMEM_HUGE_NEVER
;
484 if (!strcmp(str
, "always"))
485 return SHMEM_HUGE_ALWAYS
;
486 if (!strcmp(str
, "within_size"))
487 return SHMEM_HUGE_WITHIN_SIZE
;
488 if (!strcmp(str
, "advise"))
489 return SHMEM_HUGE_ADVISE
;
490 if (!strcmp(str
, "deny"))
491 return SHMEM_HUGE_DENY
;
492 if (!strcmp(str
, "force"))
493 return SHMEM_HUGE_FORCE
;
498 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
499 static const char *shmem_format_huge(int huge
)
502 case SHMEM_HUGE_NEVER
:
504 case SHMEM_HUGE_ALWAYS
:
506 case SHMEM_HUGE_WITHIN_SIZE
:
507 return "within_size";
508 case SHMEM_HUGE_ADVISE
:
510 case SHMEM_HUGE_DENY
:
512 case SHMEM_HUGE_FORCE
:
521 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
522 struct shrink_control
*sc
, unsigned long nr_to_split
)
524 LIST_HEAD(list
), *pos
, *next
;
525 LIST_HEAD(to_remove
);
527 struct shmem_inode_info
*info
;
529 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
530 int removed
= 0, split
= 0;
532 if (list_empty(&sbinfo
->shrinklist
))
535 spin_lock(&sbinfo
->shrinklist_lock
);
536 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
537 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
540 inode
= igrab(&info
->vfs_inode
);
542 /* inode is about to be evicted */
544 list_del_init(&info
->shrinklist
);
549 /* Check if there's anything to gain */
550 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
551 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
552 list_move(&info
->shrinklist
, &to_remove
);
557 list_move(&info
->shrinklist
, &list
);
562 spin_unlock(&sbinfo
->shrinklist_lock
);
564 list_for_each_safe(pos
, next
, &to_remove
) {
565 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
566 inode
= &info
->vfs_inode
;
567 list_del_init(&info
->shrinklist
);
571 list_for_each_safe(pos
, next
, &list
) {
574 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
575 inode
= &info
->vfs_inode
;
577 if (nr_to_split
&& split
>= nr_to_split
)
580 page
= find_get_page(inode
->i_mapping
,
581 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
585 /* No huge page at the end of the file: nothing to split */
586 if (!PageTransHuge(page
)) {
592 * Leave the inode on the list if we failed to lock
593 * the page at this time.
595 * Waiting for the lock may lead to deadlock in the
598 if (!trylock_page(page
)) {
603 ret
= split_huge_page(page
);
607 /* If split failed leave the inode on the list */
613 list_del_init(&info
->shrinklist
);
619 spin_lock(&sbinfo
->shrinklist_lock
);
620 list_splice_tail(&list
, &sbinfo
->shrinklist
);
621 sbinfo
->shrinklist_len
-= removed
;
622 spin_unlock(&sbinfo
->shrinklist_lock
);
627 static long shmem_unused_huge_scan(struct super_block
*sb
,
628 struct shrink_control
*sc
)
630 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
632 if (!READ_ONCE(sbinfo
->shrinklist_len
))
635 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
638 static long shmem_unused_huge_count(struct super_block
*sb
,
639 struct shrink_control
*sc
)
641 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
642 return READ_ONCE(sbinfo
->shrinklist_len
);
644 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
646 #define shmem_huge SHMEM_HUGE_DENY
648 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
649 struct shrink_control
*sc
, unsigned long nr_to_split
)
653 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
655 static inline bool is_huge_enabled(struct shmem_sb_info
*sbinfo
)
657 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
658 (shmem_huge
== SHMEM_HUGE_FORCE
|| sbinfo
->huge
) &&
659 shmem_huge
!= SHMEM_HUGE_DENY
)
665 * Like add_to_page_cache_locked, but error if expected item has gone.
667 static int shmem_add_to_page_cache(struct page
*page
,
668 struct address_space
*mapping
,
669 pgoff_t index
, void *expected
, gfp_t gfp
,
670 struct mm_struct
*charge_mm
)
672 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, compound_order(page
));
674 unsigned long nr
= compound_nr(page
);
677 VM_BUG_ON_PAGE(PageTail(page
), page
);
678 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
679 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
680 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
681 VM_BUG_ON(expected
&& PageTransHuge(page
));
683 page_ref_add(page
, nr
);
684 page
->mapping
= mapping
;
687 if (!PageSwapCache(page
)) {
688 error
= mem_cgroup_charge(page
, charge_mm
, gfp
);
690 if (PageTransHuge(page
)) {
691 count_vm_event(THP_FILE_FALLBACK
);
692 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
697 cgroup_throttle_swaprate(page
, gfp
);
702 entry
= xas_find_conflict(&xas
);
703 if (entry
!= expected
)
704 xas_set_err(&xas
, -EEXIST
);
705 xas_create_range(&xas
);
709 xas_store(&xas
, page
);
714 if (PageTransHuge(page
)) {
715 count_vm_event(THP_FILE_ALLOC
);
716 __mod_lruvec_page_state(page
, NR_SHMEM_THPS
, nr
);
718 mapping
->nrpages
+= nr
;
719 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, nr
);
720 __mod_lruvec_page_state(page
, NR_SHMEM
, nr
);
722 xas_unlock_irq(&xas
);
723 } while (xas_nomem(&xas
, gfp
));
725 if (xas_error(&xas
)) {
726 error
= xas_error(&xas
);
732 page
->mapping
= NULL
;
733 page_ref_sub(page
, nr
);
738 * Like delete_from_page_cache, but substitutes swap for page.
740 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
742 struct address_space
*mapping
= page
->mapping
;
745 VM_BUG_ON_PAGE(PageCompound(page
), page
);
747 xa_lock_irq(&mapping
->i_pages
);
748 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
749 page
->mapping
= NULL
;
751 __dec_lruvec_page_state(page
, NR_FILE_PAGES
);
752 __dec_lruvec_page_state(page
, NR_SHMEM
);
753 xa_unlock_irq(&mapping
->i_pages
);
759 * Remove swap entry from page cache, free the swap and its page cache.
761 static int shmem_free_swap(struct address_space
*mapping
,
762 pgoff_t index
, void *radswap
)
766 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
769 free_swap_and_cache(radix_to_swp_entry(radswap
));
774 * Determine (in bytes) how many of the shmem object's pages mapped by the
775 * given offsets are swapped out.
777 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
778 * as long as the inode doesn't go away and racy results are not a problem.
780 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
781 pgoff_t start
, pgoff_t end
)
783 XA_STATE(xas
, &mapping
->i_pages
, start
);
785 unsigned long swapped
= 0;
788 xas_for_each(&xas
, page
, end
- 1) {
789 if (xas_retry(&xas
, page
))
791 if (xa_is_value(page
))
794 if (need_resched()) {
802 return swapped
<< PAGE_SHIFT
;
806 * Determine (in bytes) how many of the shmem object's pages mapped by the
807 * given vma is swapped out.
809 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
810 * as long as the inode doesn't go away and racy results are not a problem.
812 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
814 struct inode
*inode
= file_inode(vma
->vm_file
);
815 struct shmem_inode_info
*info
= SHMEM_I(inode
);
816 struct address_space
*mapping
= inode
->i_mapping
;
817 unsigned long swapped
;
819 /* Be careful as we don't hold info->lock */
820 swapped
= READ_ONCE(info
->swapped
);
823 * The easier cases are when the shmem object has nothing in swap, or
824 * the vma maps it whole. Then we can simply use the stats that we
830 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
831 return swapped
<< PAGE_SHIFT
;
833 /* Here comes the more involved part */
834 return shmem_partial_swap_usage(mapping
,
835 linear_page_index(vma
, vma
->vm_start
),
836 linear_page_index(vma
, vma
->vm_end
));
840 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
842 void shmem_unlock_mapping(struct address_space
*mapping
)
849 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
851 while (!mapping_unevictable(mapping
)) {
852 if (!pagevec_lookup(&pvec
, mapping
, &index
))
854 check_move_unevictable_pages(&pvec
);
855 pagevec_release(&pvec
);
861 * Check whether a hole-punch or truncation needs to split a huge page,
862 * returning true if no split was required, or the split has been successful.
864 * Eviction (or truncation to 0 size) should never need to split a huge page;
865 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
866 * head, and then succeeded to trylock on tail.
868 * A split can only succeed when there are no additional references on the
869 * huge page: so the split below relies upon find_get_entries() having stopped
870 * when it found a subpage of the huge page, without getting further references.
872 static bool shmem_punch_compound(struct page
*page
, pgoff_t start
, pgoff_t end
)
874 if (!PageTransCompound(page
))
877 /* Just proceed to delete a huge page wholly within the range punched */
878 if (PageHead(page
) &&
879 page
->index
>= start
&& page
->index
+ HPAGE_PMD_NR
<= end
)
882 /* Try to split huge page, so we can truly punch the hole or truncate */
883 return split_huge_page(page
) >= 0;
887 * Remove range of pages and swap entries from page cache, and free them.
888 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
890 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
893 struct address_space
*mapping
= inode
->i_mapping
;
894 struct shmem_inode_info
*info
= SHMEM_I(inode
);
895 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
896 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
897 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
898 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
900 pgoff_t indices
[PAGEVEC_SIZE
];
901 long nr_swaps_freed
= 0;
906 end
= -1; /* unsigned, so actually very big */
910 while (index
< end
&& find_lock_entries(mapping
, index
, end
- 1,
912 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
913 struct page
*page
= pvec
.pages
[i
];
917 if (xa_is_value(page
)) {
920 nr_swaps_freed
+= !shmem_free_swap(mapping
,
924 index
+= thp_nr_pages(page
) - 1;
926 if (!unfalloc
|| !PageUptodate(page
))
927 truncate_inode_page(mapping
, page
);
930 pagevec_remove_exceptionals(&pvec
);
931 pagevec_release(&pvec
);
937 struct page
*page
= NULL
;
938 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
940 unsigned int top
= PAGE_SIZE
;
945 zero_user_segment(page
, partial_start
, top
);
946 set_page_dirty(page
);
952 struct page
*page
= NULL
;
953 shmem_getpage(inode
, end
, &page
, SGP_READ
);
955 zero_user_segment(page
, 0, partial_end
);
956 set_page_dirty(page
);
965 while (index
< end
) {
968 if (!find_get_entries(mapping
, index
, end
- 1, &pvec
,
970 /* If all gone or hole-punch or unfalloc, we're done */
971 if (index
== start
|| end
!= -1)
973 /* But if truncating, restart to make sure all gone */
977 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
978 struct page
*page
= pvec
.pages
[i
];
981 if (xa_is_value(page
)) {
984 if (shmem_free_swap(mapping
, index
, page
)) {
985 /* Swap was replaced by page: retry */
995 if (!unfalloc
|| !PageUptodate(page
)) {
996 if (page_mapping(page
) != mapping
) {
997 /* Page was replaced by swap: retry */
1002 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
1003 if (shmem_punch_compound(page
, start
, end
))
1004 truncate_inode_page(mapping
, page
);
1005 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1006 /* Wipe the page and don't get stuck */
1007 clear_highpage(page
);
1008 flush_dcache_page(page
);
1009 set_page_dirty(page
);
1011 round_up(start
, HPAGE_PMD_NR
))
1017 pagevec_remove_exceptionals(&pvec
);
1018 pagevec_release(&pvec
);
1022 spin_lock_irq(&info
->lock
);
1023 info
->swapped
-= nr_swaps_freed
;
1024 shmem_recalc_inode(inode
);
1025 spin_unlock_irq(&info
->lock
);
1028 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1030 shmem_undo_range(inode
, lstart
, lend
, false);
1031 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1033 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1035 static int shmem_getattr(struct user_namespace
*mnt_userns
,
1036 const struct path
*path
, struct kstat
*stat
,
1037 u32 request_mask
, unsigned int query_flags
)
1039 struct inode
*inode
= path
->dentry
->d_inode
;
1040 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1041 struct shmem_sb_info
*sb_info
= SHMEM_SB(inode
->i_sb
);
1043 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1044 spin_lock_irq(&info
->lock
);
1045 shmem_recalc_inode(inode
);
1046 spin_unlock_irq(&info
->lock
);
1048 generic_fillattr(&init_user_ns
, inode
, stat
);
1050 if (is_huge_enabled(sb_info
))
1051 stat
->blksize
= HPAGE_PMD_SIZE
;
1056 static int shmem_setattr(struct user_namespace
*mnt_userns
,
1057 struct dentry
*dentry
, struct iattr
*attr
)
1059 struct inode
*inode
= d_inode(dentry
);
1060 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1061 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1064 error
= setattr_prepare(&init_user_ns
, dentry
, attr
);
1068 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1069 loff_t oldsize
= inode
->i_size
;
1070 loff_t newsize
= attr
->ia_size
;
1072 /* protected by i_mutex */
1073 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1074 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1077 if (newsize
!= oldsize
) {
1078 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1082 i_size_write(inode
, newsize
);
1083 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1085 if (newsize
<= oldsize
) {
1086 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1087 if (oldsize
> holebegin
)
1088 unmap_mapping_range(inode
->i_mapping
,
1091 shmem_truncate_range(inode
,
1092 newsize
, (loff_t
)-1);
1093 /* unmap again to remove racily COWed private pages */
1094 if (oldsize
> holebegin
)
1095 unmap_mapping_range(inode
->i_mapping
,
1099 * Part of the huge page can be beyond i_size: subject
1100 * to shrink under memory pressure.
1102 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1103 spin_lock(&sbinfo
->shrinklist_lock
);
1105 * _careful to defend against unlocked access to
1106 * ->shrink_list in shmem_unused_huge_shrink()
1108 if (list_empty_careful(&info
->shrinklist
)) {
1109 list_add_tail(&info
->shrinklist
,
1110 &sbinfo
->shrinklist
);
1111 sbinfo
->shrinklist_len
++;
1113 spin_unlock(&sbinfo
->shrinklist_lock
);
1118 setattr_copy(&init_user_ns
, inode
, attr
);
1119 if (attr
->ia_valid
& ATTR_MODE
)
1120 error
= posix_acl_chmod(&init_user_ns
, inode
, inode
->i_mode
);
1124 static void shmem_evict_inode(struct inode
*inode
)
1126 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1127 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1129 if (shmem_mapping(inode
->i_mapping
)) {
1130 shmem_unacct_size(info
->flags
, inode
->i_size
);
1132 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1133 if (!list_empty(&info
->shrinklist
)) {
1134 spin_lock(&sbinfo
->shrinklist_lock
);
1135 if (!list_empty(&info
->shrinklist
)) {
1136 list_del_init(&info
->shrinklist
);
1137 sbinfo
->shrinklist_len
--;
1139 spin_unlock(&sbinfo
->shrinklist_lock
);
1141 while (!list_empty(&info
->swaplist
)) {
1142 /* Wait while shmem_unuse() is scanning this inode... */
1143 wait_var_event(&info
->stop_eviction
,
1144 !atomic_read(&info
->stop_eviction
));
1145 mutex_lock(&shmem_swaplist_mutex
);
1146 /* ...but beware of the race if we peeked too early */
1147 if (!atomic_read(&info
->stop_eviction
))
1148 list_del_init(&info
->swaplist
);
1149 mutex_unlock(&shmem_swaplist_mutex
);
1153 simple_xattrs_free(&info
->xattrs
);
1154 WARN_ON(inode
->i_blocks
);
1155 shmem_free_inode(inode
->i_sb
);
1159 extern struct swap_info_struct
*swap_info
[];
1161 static int shmem_find_swap_entries(struct address_space
*mapping
,
1162 pgoff_t start
, unsigned int nr_entries
,
1163 struct page
**entries
, pgoff_t
*indices
,
1164 unsigned int type
, bool frontswap
)
1166 XA_STATE(xas
, &mapping
->i_pages
, start
);
1169 unsigned int ret
= 0;
1175 xas_for_each(&xas
, page
, ULONG_MAX
) {
1176 if (xas_retry(&xas
, page
))
1179 if (!xa_is_value(page
))
1182 entry
= radix_to_swp_entry(page
);
1183 if (swp_type(entry
) != type
)
1186 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1189 indices
[ret
] = xas
.xa_index
;
1190 entries
[ret
] = page
;
1192 if (need_resched()) {
1196 if (++ret
== nr_entries
)
1205 * Move the swapped pages for an inode to page cache. Returns the count
1206 * of pages swapped in, or the error in case of failure.
1208 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1214 struct address_space
*mapping
= inode
->i_mapping
;
1216 for (i
= 0; i
< pvec
.nr
; i
++) {
1217 struct page
*page
= pvec
.pages
[i
];
1219 if (!xa_is_value(page
))
1221 error
= shmem_swapin_page(inode
, indices
[i
],
1223 mapping_gfp_mask(mapping
),
1230 if (error
== -ENOMEM
)
1234 return error
? error
: ret
;
1238 * If swap found in inode, free it and move page from swapcache to filecache.
1240 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1241 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1243 struct address_space
*mapping
= inode
->i_mapping
;
1245 struct pagevec pvec
;
1246 pgoff_t indices
[PAGEVEC_SIZE
];
1247 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1250 pagevec_init(&pvec
);
1252 unsigned int nr_entries
= PAGEVEC_SIZE
;
1254 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1255 nr_entries
= *fs_pages_to_unuse
;
1257 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1258 pvec
.pages
, indices
,
1265 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1269 if (frontswap_partial
) {
1270 *fs_pages_to_unuse
-= ret
;
1271 if (*fs_pages_to_unuse
== 0) {
1272 ret
= FRONTSWAP_PAGES_UNUSED
;
1277 start
= indices
[pvec
.nr
- 1];
1284 * Read all the shared memory data that resides in the swap
1285 * device 'type' back into memory, so the swap device can be
1288 int shmem_unuse(unsigned int type
, bool frontswap
,
1289 unsigned long *fs_pages_to_unuse
)
1291 struct shmem_inode_info
*info
, *next
;
1294 if (list_empty(&shmem_swaplist
))
1297 mutex_lock(&shmem_swaplist_mutex
);
1298 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1299 if (!info
->swapped
) {
1300 list_del_init(&info
->swaplist
);
1304 * Drop the swaplist mutex while searching the inode for swap;
1305 * but before doing so, make sure shmem_evict_inode() will not
1306 * remove placeholder inode from swaplist, nor let it be freed
1307 * (igrab() would protect from unlink, but not from unmount).
1309 atomic_inc(&info
->stop_eviction
);
1310 mutex_unlock(&shmem_swaplist_mutex
);
1312 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1316 mutex_lock(&shmem_swaplist_mutex
);
1317 next
= list_next_entry(info
, swaplist
);
1319 list_del_init(&info
->swaplist
);
1320 if (atomic_dec_and_test(&info
->stop_eviction
))
1321 wake_up_var(&info
->stop_eviction
);
1325 mutex_unlock(&shmem_swaplist_mutex
);
1331 * Move the page from the page cache to the swap cache.
1333 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1335 struct shmem_inode_info
*info
;
1336 struct address_space
*mapping
;
1337 struct inode
*inode
;
1341 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1342 BUG_ON(!PageLocked(page
));
1343 mapping
= page
->mapping
;
1344 index
= page
->index
;
1345 inode
= mapping
->host
;
1346 info
= SHMEM_I(inode
);
1347 if (info
->flags
& VM_LOCKED
)
1349 if (!total_swap_pages
)
1353 * Our capabilities prevent regular writeback or sync from ever calling
1354 * shmem_writepage; but a stacking filesystem might use ->writepage of
1355 * its underlying filesystem, in which case tmpfs should write out to
1356 * swap only in response to memory pressure, and not for the writeback
1359 if (!wbc
->for_reclaim
) {
1360 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1365 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1366 * value into swapfile.c, the only way we can correctly account for a
1367 * fallocated page arriving here is now to initialize it and write it.
1369 * That's okay for a page already fallocated earlier, but if we have
1370 * not yet completed the fallocation, then (a) we want to keep track
1371 * of this page in case we have to undo it, and (b) it may not be a
1372 * good idea to continue anyway, once we're pushing into swap. So
1373 * reactivate the page, and let shmem_fallocate() quit when too many.
1375 if (!PageUptodate(page
)) {
1376 if (inode
->i_private
) {
1377 struct shmem_falloc
*shmem_falloc
;
1378 spin_lock(&inode
->i_lock
);
1379 shmem_falloc
= inode
->i_private
;
1381 !shmem_falloc
->waitq
&&
1382 index
>= shmem_falloc
->start
&&
1383 index
< shmem_falloc
->next
)
1384 shmem_falloc
->nr_unswapped
++;
1386 shmem_falloc
= NULL
;
1387 spin_unlock(&inode
->i_lock
);
1391 clear_highpage(page
);
1392 flush_dcache_page(page
);
1393 SetPageUptodate(page
);
1396 swap
= get_swap_page(page
);
1401 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1402 * if it's not already there. Do it now before the page is
1403 * moved to swap cache, when its pagelock no longer protects
1404 * the inode from eviction. But don't unlock the mutex until
1405 * we've incremented swapped, because shmem_unuse_inode() will
1406 * prune a !swapped inode from the swaplist under this mutex.
1408 mutex_lock(&shmem_swaplist_mutex
);
1409 if (list_empty(&info
->swaplist
))
1410 list_add(&info
->swaplist
, &shmem_swaplist
);
1412 if (add_to_swap_cache(page
, swap
,
1413 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1415 spin_lock_irq(&info
->lock
);
1416 shmem_recalc_inode(inode
);
1418 spin_unlock_irq(&info
->lock
);
1420 swap_shmem_alloc(swap
);
1421 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1423 mutex_unlock(&shmem_swaplist_mutex
);
1424 BUG_ON(page_mapped(page
));
1425 swap_writepage(page
, wbc
);
1429 mutex_unlock(&shmem_swaplist_mutex
);
1430 put_swap_page(page
, swap
);
1432 set_page_dirty(page
);
1433 if (wbc
->for_reclaim
)
1434 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1439 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1440 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1444 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1445 return; /* show nothing */
1447 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1449 seq_printf(seq
, ",mpol=%s", buffer
);
1452 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1454 struct mempolicy
*mpol
= NULL
;
1456 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1457 mpol
= sbinfo
->mpol
;
1459 spin_unlock(&sbinfo
->stat_lock
);
1463 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1464 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1467 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1471 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1473 #define vm_policy vm_private_data
1476 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1477 struct shmem_inode_info
*info
, pgoff_t index
)
1479 /* Create a pseudo vma that just contains the policy */
1480 vma_init(vma
, NULL
);
1481 /* Bias interleave by inode number to distribute better across nodes */
1482 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1483 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1486 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1488 /* Drop reference taken by mpol_shared_policy_lookup() */
1489 mpol_cond_put(vma
->vm_policy
);
1492 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1493 struct shmem_inode_info
*info
, pgoff_t index
)
1495 struct vm_area_struct pvma
;
1497 struct vm_fault vmf
= {
1501 shmem_pseudo_vma_init(&pvma
, info
, index
);
1502 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1503 shmem_pseudo_vma_destroy(&pvma
);
1509 * Make sure huge_gfp is always more limited than limit_gfp.
1510 * Some of the flags set permissions, while others set limitations.
1512 static gfp_t
limit_gfp_mask(gfp_t huge_gfp
, gfp_t limit_gfp
)
1514 gfp_t allowflags
= __GFP_IO
| __GFP_FS
| __GFP_RECLAIM
;
1515 gfp_t denyflags
= __GFP_NOWARN
| __GFP_NORETRY
;
1516 gfp_t zoneflags
= limit_gfp
& GFP_ZONEMASK
;
1517 gfp_t result
= huge_gfp
& ~(allowflags
| GFP_ZONEMASK
);
1519 /* Allow allocations only from the originally specified zones. */
1520 result
|= zoneflags
;
1523 * Minimize the result gfp by taking the union with the deny flags,
1524 * and the intersection of the allow flags.
1526 result
|= (limit_gfp
& denyflags
);
1527 result
|= (huge_gfp
& limit_gfp
) & allowflags
;
1532 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1533 struct shmem_inode_info
*info
, pgoff_t index
)
1535 struct vm_area_struct pvma
;
1536 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1540 hindex
= round_down(index
, HPAGE_PMD_NR
);
1541 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1545 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1546 page
= alloc_pages_vma(gfp
, HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(),
1548 shmem_pseudo_vma_destroy(&pvma
);
1550 prep_transhuge_page(page
);
1552 count_vm_event(THP_FILE_FALLBACK
);
1556 static struct page
*shmem_alloc_page(gfp_t gfp
,
1557 struct shmem_inode_info
*info
, pgoff_t index
)
1559 struct vm_area_struct pvma
;
1562 shmem_pseudo_vma_init(&pvma
, info
, index
);
1563 page
= alloc_page_vma(gfp
, &pvma
, 0);
1564 shmem_pseudo_vma_destroy(&pvma
);
1569 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1570 struct inode
*inode
,
1571 pgoff_t index
, bool huge
)
1573 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1578 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1580 nr
= huge
? HPAGE_PMD_NR
: 1;
1582 if (!shmem_inode_acct_block(inode
, nr
))
1586 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1588 page
= shmem_alloc_page(gfp
, info
, index
);
1590 __SetPageLocked(page
);
1591 __SetPageSwapBacked(page
);
1596 shmem_inode_unacct_blocks(inode
, nr
);
1598 return ERR_PTR(err
);
1602 * When a page is moved from swapcache to shmem filecache (either by the
1603 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1604 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1605 * ignorance of the mapping it belongs to. If that mapping has special
1606 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1607 * we may need to copy to a suitable page before moving to filecache.
1609 * In a future release, this may well be extended to respect cpuset and
1610 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1611 * but for now it is a simple matter of zone.
1613 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1615 return page_zonenum(page
) > gfp_zone(gfp
);
1618 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1619 struct shmem_inode_info
*info
, pgoff_t index
)
1621 struct page
*oldpage
, *newpage
;
1622 struct address_space
*swap_mapping
;
1628 entry
.val
= page_private(oldpage
);
1629 swap_index
= swp_offset(entry
);
1630 swap_mapping
= page_mapping(oldpage
);
1633 * We have arrived here because our zones are constrained, so don't
1634 * limit chance of success by further cpuset and node constraints.
1636 gfp
&= ~GFP_CONSTRAINT_MASK
;
1637 newpage
= shmem_alloc_page(gfp
, info
, index
);
1642 copy_highpage(newpage
, oldpage
);
1643 flush_dcache_page(newpage
);
1645 __SetPageLocked(newpage
);
1646 __SetPageSwapBacked(newpage
);
1647 SetPageUptodate(newpage
);
1648 set_page_private(newpage
, entry
.val
);
1649 SetPageSwapCache(newpage
);
1652 * Our caller will very soon move newpage out of swapcache, but it's
1653 * a nice clean interface for us to replace oldpage by newpage there.
1655 xa_lock_irq(&swap_mapping
->i_pages
);
1656 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1658 mem_cgroup_migrate(oldpage
, newpage
);
1659 __inc_lruvec_page_state(newpage
, NR_FILE_PAGES
);
1660 __dec_lruvec_page_state(oldpage
, NR_FILE_PAGES
);
1662 xa_unlock_irq(&swap_mapping
->i_pages
);
1664 if (unlikely(error
)) {
1666 * Is this possible? I think not, now that our callers check
1667 * both PageSwapCache and page_private after getting page lock;
1668 * but be defensive. Reverse old to newpage for clear and free.
1672 lru_cache_add(newpage
);
1676 ClearPageSwapCache(oldpage
);
1677 set_page_private(oldpage
, 0);
1679 unlock_page(oldpage
);
1686 * Swap in the page pointed to by *pagep.
1687 * Caller has to make sure that *pagep contains a valid swapped page.
1688 * Returns 0 and the page in pagep if success. On failure, returns the
1689 * error code and NULL in *pagep.
1691 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1692 struct page
**pagep
, enum sgp_type sgp
,
1693 gfp_t gfp
, struct vm_area_struct
*vma
,
1694 vm_fault_t
*fault_type
)
1696 struct address_space
*mapping
= inode
->i_mapping
;
1697 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1698 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: NULL
;
1703 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1704 swap
= radix_to_swp_entry(*pagep
);
1707 /* Look it up and read it in.. */
1708 page
= lookup_swap_cache(swap
, NULL
, 0);
1710 /* Or update major stats only when swapin succeeds?? */
1712 *fault_type
|= VM_FAULT_MAJOR
;
1713 count_vm_event(PGMAJFAULT
);
1714 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1716 /* Here we actually start the io */
1717 page
= shmem_swapin(swap
, gfp
, info
, index
);
1724 /* We have to do this with page locked to prevent races */
1726 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1727 !shmem_confirm_swap(mapping
, index
, swap
)) {
1731 if (!PageUptodate(page
)) {
1735 wait_on_page_writeback(page
);
1738 * Some architectures may have to restore extra metadata to the
1739 * physical page after reading from swap.
1741 arch_swap_restore(swap
, page
);
1743 if (shmem_should_replace_page(page
, gfp
)) {
1744 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1749 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1750 swp_to_radix_entry(swap
), gfp
,
1755 spin_lock_irq(&info
->lock
);
1757 shmem_recalc_inode(inode
);
1758 spin_unlock_irq(&info
->lock
);
1760 if (sgp
== SGP_WRITE
)
1761 mark_page_accessed(page
);
1763 delete_from_swap_cache(page
);
1764 set_page_dirty(page
);
1770 if (!shmem_confirm_swap(mapping
, index
, swap
))
1782 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1784 * If we allocate a new one we do not mark it dirty. That's up to the
1785 * vm. If we swap it in we mark it dirty since we also free the swap
1786 * entry since a page cannot live in both the swap and page cache.
1788 * vma, vmf, and fault_type are only supplied by shmem_fault:
1789 * otherwise they are NULL.
1791 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1792 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1793 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1794 vm_fault_t
*fault_type
)
1796 struct address_space
*mapping
= inode
->i_mapping
;
1797 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1798 struct shmem_sb_info
*sbinfo
;
1799 struct mm_struct
*charge_mm
;
1801 enum sgp_type sgp_huge
= sgp
;
1802 pgoff_t hindex
= index
;
1808 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1810 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1813 if (sgp
<= SGP_CACHE
&&
1814 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1818 sbinfo
= SHMEM_SB(inode
->i_sb
);
1819 charge_mm
= vma
? vma
->vm_mm
: NULL
;
1821 page
= pagecache_get_page(mapping
, index
,
1822 FGP_ENTRY
| FGP_HEAD
| FGP_LOCK
, 0);
1824 if (page
&& vma
&& userfaultfd_minor(vma
)) {
1825 if (!xa_is_value(page
)) {
1829 *fault_type
= handle_userfault(vmf
, VM_UFFD_MINOR
);
1833 if (xa_is_value(page
)) {
1834 error
= shmem_swapin_page(inode
, index
, &page
,
1835 sgp
, gfp
, vma
, fault_type
);
1836 if (error
== -EEXIST
)
1844 hindex
= page
->index
;
1845 if (page
&& sgp
== SGP_WRITE
)
1846 mark_page_accessed(page
);
1848 /* fallocated page? */
1849 if (page
&& !PageUptodate(page
)) {
1850 if (sgp
!= SGP_READ
)
1857 if (page
|| sgp
== SGP_READ
)
1861 * Fast cache lookup did not find it:
1862 * bring it back from swap or allocate.
1865 if (vma
&& userfaultfd_missing(vma
)) {
1866 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1870 /* shmem_symlink() */
1871 if (!shmem_mapping(mapping
))
1873 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1875 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1877 switch (sbinfo
->huge
) {
1878 case SHMEM_HUGE_NEVER
:
1880 case SHMEM_HUGE_WITHIN_SIZE
: {
1884 off
= round_up(index
, HPAGE_PMD_NR
);
1885 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1886 if (i_size
>= HPAGE_PMD_SIZE
&&
1887 i_size
>> PAGE_SHIFT
>= off
)
1892 case SHMEM_HUGE_ADVISE
:
1893 if (sgp_huge
== SGP_HUGE
)
1895 /* TODO: implement fadvise() hints */
1900 huge_gfp
= vma_thp_gfp_mask(vma
);
1901 huge_gfp
= limit_gfp_mask(huge_gfp
, gfp
);
1902 page
= shmem_alloc_and_acct_page(huge_gfp
, inode
, index
, true);
1905 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1911 error
= PTR_ERR(page
);
1913 if (error
!= -ENOSPC
)
1916 * Try to reclaim some space by splitting a huge page
1917 * beyond i_size on the filesystem.
1922 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1923 if (ret
== SHRINK_STOP
)
1931 if (PageTransHuge(page
))
1932 hindex
= round_down(index
, HPAGE_PMD_NR
);
1936 if (sgp
== SGP_WRITE
)
1937 __SetPageReferenced(page
);
1939 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1940 NULL
, gfp
& GFP_RECLAIM_MASK
,
1944 lru_cache_add(page
);
1946 spin_lock_irq(&info
->lock
);
1947 info
->alloced
+= compound_nr(page
);
1948 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1949 shmem_recalc_inode(inode
);
1950 spin_unlock_irq(&info
->lock
);
1953 if (PageTransHuge(page
) &&
1954 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1955 hindex
+ HPAGE_PMD_NR
- 1) {
1957 * Part of the huge page is beyond i_size: subject
1958 * to shrink under memory pressure.
1960 spin_lock(&sbinfo
->shrinklist_lock
);
1962 * _careful to defend against unlocked access to
1963 * ->shrink_list in shmem_unused_huge_shrink()
1965 if (list_empty_careful(&info
->shrinklist
)) {
1966 list_add_tail(&info
->shrinklist
,
1967 &sbinfo
->shrinklist
);
1968 sbinfo
->shrinklist_len
++;
1970 spin_unlock(&sbinfo
->shrinklist_lock
);
1974 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1976 if (sgp
== SGP_FALLOC
)
1980 * Let SGP_WRITE caller clear ends if write does not fill page;
1981 * but SGP_FALLOC on a page fallocated earlier must initialize
1982 * it now, lest undo on failure cancel our earlier guarantee.
1984 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1987 for (i
= 0; i
< compound_nr(page
); i
++) {
1988 clear_highpage(page
+ i
);
1989 flush_dcache_page(page
+ i
);
1991 SetPageUptodate(page
);
1994 /* Perhaps the file has been truncated since we checked */
1995 if (sgp
<= SGP_CACHE
&&
1996 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1998 ClearPageDirty(page
);
1999 delete_from_page_cache(page
);
2000 spin_lock_irq(&info
->lock
);
2001 shmem_recalc_inode(inode
);
2002 spin_unlock_irq(&info
->lock
);
2008 *pagep
= page
+ index
- hindex
;
2015 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
2017 if (PageTransHuge(page
)) {
2027 if (error
== -ENOSPC
&& !once
++) {
2028 spin_lock_irq(&info
->lock
);
2029 shmem_recalc_inode(inode
);
2030 spin_unlock_irq(&info
->lock
);
2033 if (error
== -EEXIST
)
2039 * This is like autoremove_wake_function, but it removes the wait queue
2040 * entry unconditionally - even if something else had already woken the
2043 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2045 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2046 list_del_init(&wait
->entry
);
2050 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2052 struct vm_area_struct
*vma
= vmf
->vma
;
2053 struct inode
*inode
= file_inode(vma
->vm_file
);
2054 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2057 vm_fault_t ret
= VM_FAULT_LOCKED
;
2060 * Trinity finds that probing a hole which tmpfs is punching can
2061 * prevent the hole-punch from ever completing: which in turn
2062 * locks writers out with its hold on i_mutex. So refrain from
2063 * faulting pages into the hole while it's being punched. Although
2064 * shmem_undo_range() does remove the additions, it may be unable to
2065 * keep up, as each new page needs its own unmap_mapping_range() call,
2066 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2068 * It does not matter if we sometimes reach this check just before the
2069 * hole-punch begins, so that one fault then races with the punch:
2070 * we just need to make racing faults a rare case.
2072 * The implementation below would be much simpler if we just used a
2073 * standard mutex or completion: but we cannot take i_mutex in fault,
2074 * and bloating every shmem inode for this unlikely case would be sad.
2076 if (unlikely(inode
->i_private
)) {
2077 struct shmem_falloc
*shmem_falloc
;
2079 spin_lock(&inode
->i_lock
);
2080 shmem_falloc
= inode
->i_private
;
2082 shmem_falloc
->waitq
&&
2083 vmf
->pgoff
>= shmem_falloc
->start
&&
2084 vmf
->pgoff
< shmem_falloc
->next
) {
2086 wait_queue_head_t
*shmem_falloc_waitq
;
2087 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2089 ret
= VM_FAULT_NOPAGE
;
2090 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2092 ret
= VM_FAULT_RETRY
;
2094 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2095 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2096 TASK_UNINTERRUPTIBLE
);
2097 spin_unlock(&inode
->i_lock
);
2101 * shmem_falloc_waitq points into the shmem_fallocate()
2102 * stack of the hole-punching task: shmem_falloc_waitq
2103 * is usually invalid by the time we reach here, but
2104 * finish_wait() does not dereference it in that case;
2105 * though i_lock needed lest racing with wake_up_all().
2107 spin_lock(&inode
->i_lock
);
2108 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2109 spin_unlock(&inode
->i_lock
);
2115 spin_unlock(&inode
->i_lock
);
2120 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2121 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2123 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2126 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2127 gfp
, vma
, vmf
, &ret
);
2129 return vmf_error(err
);
2133 unsigned long shmem_get_unmapped_area(struct file
*file
,
2134 unsigned long uaddr
, unsigned long len
,
2135 unsigned long pgoff
, unsigned long flags
)
2137 unsigned long (*get_area
)(struct file
*,
2138 unsigned long, unsigned long, unsigned long, unsigned long);
2140 unsigned long offset
;
2141 unsigned long inflated_len
;
2142 unsigned long inflated_addr
;
2143 unsigned long inflated_offset
;
2145 if (len
> TASK_SIZE
)
2148 get_area
= current
->mm
->get_unmapped_area
;
2149 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2151 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2153 if (IS_ERR_VALUE(addr
))
2155 if (addr
& ~PAGE_MASK
)
2157 if (addr
> TASK_SIZE
- len
)
2160 if (shmem_huge
== SHMEM_HUGE_DENY
)
2162 if (len
< HPAGE_PMD_SIZE
)
2164 if (flags
& MAP_FIXED
)
2167 * Our priority is to support MAP_SHARED mapped hugely;
2168 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2169 * But if caller specified an address hint and we allocated area there
2170 * successfully, respect that as before.
2175 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2176 struct super_block
*sb
;
2179 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2180 sb
= file_inode(file
)->i_sb
;
2183 * Called directly from mm/mmap.c, or drivers/char/mem.c
2184 * for "/dev/zero", to create a shared anonymous object.
2186 if (IS_ERR(shm_mnt
))
2188 sb
= shm_mnt
->mnt_sb
;
2190 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2194 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2195 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2197 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2200 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2201 if (inflated_len
> TASK_SIZE
)
2203 if (inflated_len
< len
)
2206 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2207 if (IS_ERR_VALUE(inflated_addr
))
2209 if (inflated_addr
& ~PAGE_MASK
)
2212 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2213 inflated_addr
+= offset
- inflated_offset
;
2214 if (inflated_offset
> offset
)
2215 inflated_addr
+= HPAGE_PMD_SIZE
;
2217 if (inflated_addr
> TASK_SIZE
- len
)
2219 return inflated_addr
;
2223 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2225 struct inode
*inode
= file_inode(vma
->vm_file
);
2226 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2229 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2232 struct inode
*inode
= file_inode(vma
->vm_file
);
2235 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2236 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2240 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
2242 struct inode
*inode
= file_inode(file
);
2243 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2244 int retval
= -ENOMEM
;
2247 * What serializes the accesses to info->flags?
2248 * ipc_lock_object() when called from shmctl_do_lock(),
2249 * no serialization needed when called from shm_destroy().
2251 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2252 if (!user_shm_lock(inode
->i_size
, ucounts
))
2254 info
->flags
|= VM_LOCKED
;
2255 mapping_set_unevictable(file
->f_mapping
);
2257 if (!lock
&& (info
->flags
& VM_LOCKED
) && ucounts
) {
2258 user_shm_unlock(inode
->i_size
, ucounts
);
2259 info
->flags
&= ~VM_LOCKED
;
2260 mapping_clear_unevictable(file
->f_mapping
);
2268 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2270 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2273 ret
= seal_check_future_write(info
->seals
, vma
);
2277 /* arm64 - allow memory tagging on RAM-based files */
2278 vma
->vm_flags
|= VM_MTE_ALLOWED
;
2280 file_accessed(file
);
2281 vma
->vm_ops
= &shmem_vm_ops
;
2282 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
2283 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2284 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2285 khugepaged_enter(vma
, vma
->vm_flags
);
2290 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2291 umode_t mode
, dev_t dev
, unsigned long flags
)
2293 struct inode
*inode
;
2294 struct shmem_inode_info
*info
;
2295 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2298 if (shmem_reserve_inode(sb
, &ino
))
2301 inode
= new_inode(sb
);
2304 inode_init_owner(&init_user_ns
, inode
, dir
, mode
);
2305 inode
->i_blocks
= 0;
2306 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2307 inode
->i_generation
= prandom_u32();
2308 info
= SHMEM_I(inode
);
2309 memset(info
, 0, (char *)inode
- (char *)info
);
2310 spin_lock_init(&info
->lock
);
2311 atomic_set(&info
->stop_eviction
, 0);
2312 info
->seals
= F_SEAL_SEAL
;
2313 info
->flags
= flags
& VM_NORESERVE
;
2314 INIT_LIST_HEAD(&info
->shrinklist
);
2315 INIT_LIST_HEAD(&info
->swaplist
);
2316 simple_xattrs_init(&info
->xattrs
);
2317 cache_no_acl(inode
);
2319 switch (mode
& S_IFMT
) {
2321 inode
->i_op
= &shmem_special_inode_operations
;
2322 init_special_inode(inode
, mode
, dev
);
2325 inode
->i_mapping
->a_ops
= &shmem_aops
;
2326 inode
->i_op
= &shmem_inode_operations
;
2327 inode
->i_fop
= &shmem_file_operations
;
2328 mpol_shared_policy_init(&info
->policy
,
2329 shmem_get_sbmpol(sbinfo
));
2333 /* Some things misbehave if size == 0 on a directory */
2334 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2335 inode
->i_op
= &shmem_dir_inode_operations
;
2336 inode
->i_fop
= &simple_dir_operations
;
2340 * Must not load anything in the rbtree,
2341 * mpol_free_shared_policy will not be called.
2343 mpol_shared_policy_init(&info
->policy
, NULL
);
2347 lockdep_annotate_inode_mutex_key(inode
);
2349 shmem_free_inode(sb
);
2353 #ifdef CONFIG_USERFAULTFD
2354 int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2356 struct vm_area_struct
*dst_vma
,
2357 unsigned long dst_addr
,
2358 unsigned long src_addr
,
2360 struct page
**pagep
)
2362 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2363 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2364 struct address_space
*mapping
= inode
->i_mapping
;
2365 gfp_t gfp
= mapping_gfp_mask(mapping
);
2366 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2372 if (!shmem_inode_acct_block(inode
, 1)) {
2374 * We may have got a page, returned -ENOENT triggering a retry,
2375 * and now we find ourselves with -ENOMEM. Release the page, to
2376 * avoid a BUG_ON in our caller.
2378 if (unlikely(*pagep
)) {
2387 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2389 goto out_unacct_blocks
;
2391 if (!zeropage
) { /* COPY */
2392 page_kaddr
= kmap_atomic(page
);
2393 ret
= copy_from_user(page_kaddr
,
2394 (const void __user
*)src_addr
,
2396 kunmap_atomic(page_kaddr
);
2398 /* fallback to copy_from_user outside mmap_lock */
2399 if (unlikely(ret
)) {
2402 /* don't free the page */
2403 goto out_unacct_blocks
;
2405 } else { /* ZEROPAGE */
2406 clear_highpage(page
);
2413 VM_BUG_ON(PageLocked(page
));
2414 VM_BUG_ON(PageSwapBacked(page
));
2415 __SetPageLocked(page
);
2416 __SetPageSwapBacked(page
);
2417 __SetPageUptodate(page
);
2420 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2421 if (unlikely(pgoff
>= max_off
))
2424 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2425 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2429 ret
= mfill_atomic_install_pte(dst_mm
, dst_pmd
, dst_vma
, dst_addr
,
2432 goto out_delete_from_cache
;
2434 spin_lock_irq(&info
->lock
);
2436 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2437 shmem_recalc_inode(inode
);
2438 spin_unlock_irq(&info
->lock
);
2443 out_delete_from_cache
:
2444 delete_from_page_cache(page
);
2449 shmem_inode_unacct_blocks(inode
, 1);
2452 #endif /* CONFIG_USERFAULTFD */
2455 static const struct inode_operations shmem_symlink_inode_operations
;
2456 static const struct inode_operations shmem_short_symlink_operations
;
2458 #ifdef CONFIG_TMPFS_XATTR
2459 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2461 #define shmem_initxattrs NULL
2465 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2466 loff_t pos
, unsigned len
, unsigned flags
,
2467 struct page
**pagep
, void **fsdata
)
2469 struct inode
*inode
= mapping
->host
;
2470 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2471 pgoff_t index
= pos
>> PAGE_SHIFT
;
2473 /* i_mutex is held by caller */
2474 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2475 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2476 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2478 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2482 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2486 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2487 loff_t pos
, unsigned len
, unsigned copied
,
2488 struct page
*page
, void *fsdata
)
2490 struct inode
*inode
= mapping
->host
;
2492 if (pos
+ copied
> inode
->i_size
)
2493 i_size_write(inode
, pos
+ copied
);
2495 if (!PageUptodate(page
)) {
2496 struct page
*head
= compound_head(page
);
2497 if (PageTransCompound(page
)) {
2500 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2501 if (head
+ i
== page
)
2503 clear_highpage(head
+ i
);
2504 flush_dcache_page(head
+ i
);
2507 if (copied
< PAGE_SIZE
) {
2508 unsigned from
= pos
& (PAGE_SIZE
- 1);
2509 zero_user_segments(page
, 0, from
,
2510 from
+ copied
, PAGE_SIZE
);
2512 SetPageUptodate(head
);
2514 set_page_dirty(page
);
2521 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2523 struct file
*file
= iocb
->ki_filp
;
2524 struct inode
*inode
= file_inode(file
);
2525 struct address_space
*mapping
= inode
->i_mapping
;
2527 unsigned long offset
;
2528 enum sgp_type sgp
= SGP_READ
;
2531 loff_t
*ppos
= &iocb
->ki_pos
;
2534 * Might this read be for a stacking filesystem? Then when reading
2535 * holes of a sparse file, we actually need to allocate those pages,
2536 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2538 if (!iter_is_iovec(to
))
2541 index
= *ppos
>> PAGE_SHIFT
;
2542 offset
= *ppos
& ~PAGE_MASK
;
2545 struct page
*page
= NULL
;
2547 unsigned long nr
, ret
;
2548 loff_t i_size
= i_size_read(inode
);
2550 end_index
= i_size
>> PAGE_SHIFT
;
2551 if (index
> end_index
)
2553 if (index
== end_index
) {
2554 nr
= i_size
& ~PAGE_MASK
;
2559 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2561 if (error
== -EINVAL
)
2566 if (sgp
== SGP_CACHE
)
2567 set_page_dirty(page
);
2572 * We must evaluate after, since reads (unlike writes)
2573 * are called without i_mutex protection against truncate
2576 i_size
= i_size_read(inode
);
2577 end_index
= i_size
>> PAGE_SHIFT
;
2578 if (index
== end_index
) {
2579 nr
= i_size
& ~PAGE_MASK
;
2590 * If users can be writing to this page using arbitrary
2591 * virtual addresses, take care about potential aliasing
2592 * before reading the page on the kernel side.
2594 if (mapping_writably_mapped(mapping
))
2595 flush_dcache_page(page
);
2597 * Mark the page accessed if we read the beginning.
2600 mark_page_accessed(page
);
2602 page
= ZERO_PAGE(0);
2607 * Ok, we have the page, and it's up-to-date, so
2608 * now we can copy it to user space...
2610 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2613 index
+= offset
>> PAGE_SHIFT
;
2614 offset
&= ~PAGE_MASK
;
2617 if (!iov_iter_count(to
))
2626 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2627 file_accessed(file
);
2628 return retval
? retval
: error
;
2631 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2633 struct address_space
*mapping
= file
->f_mapping
;
2634 struct inode
*inode
= mapping
->host
;
2636 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2637 return generic_file_llseek_size(file
, offset
, whence
,
2638 MAX_LFS_FILESIZE
, i_size_read(inode
));
2643 /* We're holding i_mutex so we can access i_size directly */
2644 offset
= mapping_seek_hole_data(mapping
, offset
, inode
->i_size
, whence
);
2646 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2647 inode_unlock(inode
);
2651 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2654 struct inode
*inode
= file_inode(file
);
2655 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2656 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2657 struct shmem_falloc shmem_falloc
;
2658 pgoff_t start
, index
, end
;
2661 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2666 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2667 struct address_space
*mapping
= file
->f_mapping
;
2668 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2669 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2670 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2672 /* protected by i_mutex */
2673 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2678 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2679 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2680 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2681 spin_lock(&inode
->i_lock
);
2682 inode
->i_private
= &shmem_falloc
;
2683 spin_unlock(&inode
->i_lock
);
2685 if ((u64
)unmap_end
> (u64
)unmap_start
)
2686 unmap_mapping_range(mapping
, unmap_start
,
2687 1 + unmap_end
- unmap_start
, 0);
2688 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2689 /* No need to unmap again: hole-punching leaves COWed pages */
2691 spin_lock(&inode
->i_lock
);
2692 inode
->i_private
= NULL
;
2693 wake_up_all(&shmem_falloc_waitq
);
2694 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2695 spin_unlock(&inode
->i_lock
);
2700 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2701 error
= inode_newsize_ok(inode
, offset
+ len
);
2705 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2710 start
= offset
>> PAGE_SHIFT
;
2711 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2712 /* Try to avoid a swapstorm if len is impossible to satisfy */
2713 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2718 shmem_falloc
.waitq
= NULL
;
2719 shmem_falloc
.start
= start
;
2720 shmem_falloc
.next
= start
;
2721 shmem_falloc
.nr_falloced
= 0;
2722 shmem_falloc
.nr_unswapped
= 0;
2723 spin_lock(&inode
->i_lock
);
2724 inode
->i_private
= &shmem_falloc
;
2725 spin_unlock(&inode
->i_lock
);
2727 for (index
= start
; index
< end
; index
++) {
2731 * Good, the fallocate(2) manpage permits EINTR: we may have
2732 * been interrupted because we are using up too much memory.
2734 if (signal_pending(current
))
2736 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2739 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2741 /* Remove the !PageUptodate pages we added */
2742 if (index
> start
) {
2743 shmem_undo_range(inode
,
2744 (loff_t
)start
<< PAGE_SHIFT
,
2745 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2751 * Inform shmem_writepage() how far we have reached.
2752 * No need for lock or barrier: we have the page lock.
2754 shmem_falloc
.next
++;
2755 if (!PageUptodate(page
))
2756 shmem_falloc
.nr_falloced
++;
2759 * If !PageUptodate, leave it that way so that freeable pages
2760 * can be recognized if we need to rollback on error later.
2761 * But set_page_dirty so that memory pressure will swap rather
2762 * than free the pages we are allocating (and SGP_CACHE pages
2763 * might still be clean: we now need to mark those dirty too).
2765 set_page_dirty(page
);
2771 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2772 i_size_write(inode
, offset
+ len
);
2773 inode
->i_ctime
= current_time(inode
);
2775 spin_lock(&inode
->i_lock
);
2776 inode
->i_private
= NULL
;
2777 spin_unlock(&inode
->i_lock
);
2779 inode_unlock(inode
);
2783 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2785 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2787 buf
->f_type
= TMPFS_MAGIC
;
2788 buf
->f_bsize
= PAGE_SIZE
;
2789 buf
->f_namelen
= NAME_MAX
;
2790 if (sbinfo
->max_blocks
) {
2791 buf
->f_blocks
= sbinfo
->max_blocks
;
2793 buf
->f_bfree
= sbinfo
->max_blocks
-
2794 percpu_counter_sum(&sbinfo
->used_blocks
);
2796 if (sbinfo
->max_inodes
) {
2797 buf
->f_files
= sbinfo
->max_inodes
;
2798 buf
->f_ffree
= sbinfo
->free_inodes
;
2800 /* else leave those fields 0 like simple_statfs */
2802 buf
->f_fsid
= uuid_to_fsid(dentry
->d_sb
->s_uuid
.b
);
2808 * File creation. Allocate an inode, and we're done..
2811 shmem_mknod(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2812 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2814 struct inode
*inode
;
2815 int error
= -ENOSPC
;
2817 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2819 error
= simple_acl_create(dir
, inode
);
2822 error
= security_inode_init_security(inode
, dir
,
2824 shmem_initxattrs
, NULL
);
2825 if (error
&& error
!= -EOPNOTSUPP
)
2829 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2830 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2831 d_instantiate(dentry
, inode
);
2832 dget(dentry
); /* Extra count - pin the dentry in core */
2841 shmem_tmpfile(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2842 struct dentry
*dentry
, umode_t mode
)
2844 struct inode
*inode
;
2845 int error
= -ENOSPC
;
2847 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2849 error
= security_inode_init_security(inode
, dir
,
2851 shmem_initxattrs
, NULL
);
2852 if (error
&& error
!= -EOPNOTSUPP
)
2854 error
= simple_acl_create(dir
, inode
);
2857 d_tmpfile(dentry
, inode
);
2865 static int shmem_mkdir(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2866 struct dentry
*dentry
, umode_t mode
)
2870 if ((error
= shmem_mknod(&init_user_ns
, dir
, dentry
,
2871 mode
| S_IFDIR
, 0)))
2877 static int shmem_create(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2878 struct dentry
*dentry
, umode_t mode
, bool excl
)
2880 return shmem_mknod(&init_user_ns
, dir
, dentry
, mode
| S_IFREG
, 0);
2886 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2888 struct inode
*inode
= d_inode(old_dentry
);
2892 * No ordinary (disk based) filesystem counts links as inodes;
2893 * but each new link needs a new dentry, pinning lowmem, and
2894 * tmpfs dentries cannot be pruned until they are unlinked.
2895 * But if an O_TMPFILE file is linked into the tmpfs, the
2896 * first link must skip that, to get the accounting right.
2898 if (inode
->i_nlink
) {
2899 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
2904 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2905 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2907 ihold(inode
); /* New dentry reference */
2908 dget(dentry
); /* Extra pinning count for the created dentry */
2909 d_instantiate(dentry
, inode
);
2914 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2916 struct inode
*inode
= d_inode(dentry
);
2918 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2919 shmem_free_inode(inode
->i_sb
);
2921 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2922 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2924 dput(dentry
); /* Undo the count from "create" - this does all the work */
2928 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2930 if (!simple_empty(dentry
))
2933 drop_nlink(d_inode(dentry
));
2935 return shmem_unlink(dir
, dentry
);
2938 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2940 bool old_is_dir
= d_is_dir(old_dentry
);
2941 bool new_is_dir
= d_is_dir(new_dentry
);
2943 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2945 drop_nlink(old_dir
);
2948 drop_nlink(new_dir
);
2952 old_dir
->i_ctime
= old_dir
->i_mtime
=
2953 new_dir
->i_ctime
= new_dir
->i_mtime
=
2954 d_inode(old_dentry
)->i_ctime
=
2955 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
2960 static int shmem_whiteout(struct user_namespace
*mnt_userns
,
2961 struct inode
*old_dir
, struct dentry
*old_dentry
)
2963 struct dentry
*whiteout
;
2966 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2970 error
= shmem_mknod(&init_user_ns
, old_dir
, whiteout
,
2971 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2977 * Cheat and hash the whiteout while the old dentry is still in
2978 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2980 * d_lookup() will consistently find one of them at this point,
2981 * not sure which one, but that isn't even important.
2988 * The VFS layer already does all the dentry stuff for rename,
2989 * we just have to decrement the usage count for the target if
2990 * it exists so that the VFS layer correctly free's it when it
2993 static int shmem_rename2(struct user_namespace
*mnt_userns
,
2994 struct inode
*old_dir
, struct dentry
*old_dentry
,
2995 struct inode
*new_dir
, struct dentry
*new_dentry
,
2998 struct inode
*inode
= d_inode(old_dentry
);
2999 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3001 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3004 if (flags
& RENAME_EXCHANGE
)
3005 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3007 if (!simple_empty(new_dentry
))
3010 if (flags
& RENAME_WHITEOUT
) {
3013 error
= shmem_whiteout(&init_user_ns
, old_dir
, old_dentry
);
3018 if (d_really_is_positive(new_dentry
)) {
3019 (void) shmem_unlink(new_dir
, new_dentry
);
3020 if (they_are_dirs
) {
3021 drop_nlink(d_inode(new_dentry
));
3022 drop_nlink(old_dir
);
3024 } else if (they_are_dirs
) {
3025 drop_nlink(old_dir
);
3029 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3030 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3031 old_dir
->i_ctime
= old_dir
->i_mtime
=
3032 new_dir
->i_ctime
= new_dir
->i_mtime
=
3033 inode
->i_ctime
= current_time(old_dir
);
3037 static int shmem_symlink(struct user_namespace
*mnt_userns
, struct inode
*dir
,
3038 struct dentry
*dentry
, const char *symname
)
3042 struct inode
*inode
;
3045 len
= strlen(symname
) + 1;
3046 if (len
> PAGE_SIZE
)
3047 return -ENAMETOOLONG
;
3049 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3054 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3055 shmem_initxattrs
, NULL
);
3056 if (error
&& error
!= -EOPNOTSUPP
) {
3061 inode
->i_size
= len
-1;
3062 if (len
<= SHORT_SYMLINK_LEN
) {
3063 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3064 if (!inode
->i_link
) {
3068 inode
->i_op
= &shmem_short_symlink_operations
;
3070 inode_nohighmem(inode
);
3071 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3076 inode
->i_mapping
->a_ops
= &shmem_aops
;
3077 inode
->i_op
= &shmem_symlink_inode_operations
;
3078 memcpy(page_address(page
), symname
, len
);
3079 SetPageUptodate(page
);
3080 set_page_dirty(page
);
3084 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3085 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3086 d_instantiate(dentry
, inode
);
3091 static void shmem_put_link(void *arg
)
3093 mark_page_accessed(arg
);
3097 static const char *shmem_get_link(struct dentry
*dentry
,
3098 struct inode
*inode
,
3099 struct delayed_call
*done
)
3101 struct page
*page
= NULL
;
3104 page
= find_get_page(inode
->i_mapping
, 0);
3106 return ERR_PTR(-ECHILD
);
3107 if (!PageUptodate(page
)) {
3109 return ERR_PTR(-ECHILD
);
3112 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3114 return ERR_PTR(error
);
3117 set_delayed_call(done
, shmem_put_link
, page
);
3118 return page_address(page
);
3121 #ifdef CONFIG_TMPFS_XATTR
3123 * Superblocks without xattr inode operations may get some security.* xattr
3124 * support from the LSM "for free". As soon as we have any other xattrs
3125 * like ACLs, we also need to implement the security.* handlers at
3126 * filesystem level, though.
3130 * Callback for security_inode_init_security() for acquiring xattrs.
3132 static int shmem_initxattrs(struct inode
*inode
,
3133 const struct xattr
*xattr_array
,
3136 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3137 const struct xattr
*xattr
;
3138 struct simple_xattr
*new_xattr
;
3141 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3142 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3146 len
= strlen(xattr
->name
) + 1;
3147 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3149 if (!new_xattr
->name
) {
3154 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3155 XATTR_SECURITY_PREFIX_LEN
);
3156 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3159 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3165 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3166 struct dentry
*unused
, struct inode
*inode
,
3167 const char *name
, void *buffer
, size_t size
)
3169 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3171 name
= xattr_full_name(handler
, name
);
3172 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3175 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3176 struct user_namespace
*mnt_userns
,
3177 struct dentry
*unused
, struct inode
*inode
,
3178 const char *name
, const void *value
,
3179 size_t size
, int flags
)
3181 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3183 name
= xattr_full_name(handler
, name
);
3184 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3187 static const struct xattr_handler shmem_security_xattr_handler
= {
3188 .prefix
= XATTR_SECURITY_PREFIX
,
3189 .get
= shmem_xattr_handler_get
,
3190 .set
= shmem_xattr_handler_set
,
3193 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3194 .prefix
= XATTR_TRUSTED_PREFIX
,
3195 .get
= shmem_xattr_handler_get
,
3196 .set
= shmem_xattr_handler_set
,
3199 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3200 #ifdef CONFIG_TMPFS_POSIX_ACL
3201 &posix_acl_access_xattr_handler
,
3202 &posix_acl_default_xattr_handler
,
3204 &shmem_security_xattr_handler
,
3205 &shmem_trusted_xattr_handler
,
3209 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3211 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3212 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3214 #endif /* CONFIG_TMPFS_XATTR */
3216 static const struct inode_operations shmem_short_symlink_operations
= {
3217 .get_link
= simple_get_link
,
3218 #ifdef CONFIG_TMPFS_XATTR
3219 .listxattr
= shmem_listxattr
,
3223 static const struct inode_operations shmem_symlink_inode_operations
= {
3224 .get_link
= shmem_get_link
,
3225 #ifdef CONFIG_TMPFS_XATTR
3226 .listxattr
= shmem_listxattr
,
3230 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3232 return ERR_PTR(-ESTALE
);
3235 static int shmem_match(struct inode
*ino
, void *vfh
)
3239 inum
= (inum
<< 32) | fh
[1];
3240 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3243 /* Find any alias of inode, but prefer a hashed alias */
3244 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3246 struct dentry
*alias
= d_find_alias(inode
);
3248 return alias
?: d_find_any_alias(inode
);
3252 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3253 struct fid
*fid
, int fh_len
, int fh_type
)
3255 struct inode
*inode
;
3256 struct dentry
*dentry
= NULL
;
3263 inum
= (inum
<< 32) | fid
->raw
[1];
3265 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3266 shmem_match
, fid
->raw
);
3268 dentry
= shmem_find_alias(inode
);
3275 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3276 struct inode
*parent
)
3280 return FILEID_INVALID
;
3283 if (inode_unhashed(inode
)) {
3284 /* Unfortunately insert_inode_hash is not idempotent,
3285 * so as we hash inodes here rather than at creation
3286 * time, we need a lock to ensure we only try
3289 static DEFINE_SPINLOCK(lock
);
3291 if (inode_unhashed(inode
))
3292 __insert_inode_hash(inode
,
3293 inode
->i_ino
+ inode
->i_generation
);
3297 fh
[0] = inode
->i_generation
;
3298 fh
[1] = inode
->i_ino
;
3299 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3305 static const struct export_operations shmem_export_ops
= {
3306 .get_parent
= shmem_get_parent
,
3307 .encode_fh
= shmem_encode_fh
,
3308 .fh_to_dentry
= shmem_fh_to_dentry
,
3324 static const struct constant_table shmem_param_enums_huge
[] = {
3325 {"never", SHMEM_HUGE_NEVER
},
3326 {"always", SHMEM_HUGE_ALWAYS
},
3327 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3328 {"advise", SHMEM_HUGE_ADVISE
},
3332 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3333 fsparam_u32 ("gid", Opt_gid
),
3334 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3335 fsparam_u32oct("mode", Opt_mode
),
3336 fsparam_string("mpol", Opt_mpol
),
3337 fsparam_string("nr_blocks", Opt_nr_blocks
),
3338 fsparam_string("nr_inodes", Opt_nr_inodes
),
3339 fsparam_string("size", Opt_size
),
3340 fsparam_u32 ("uid", Opt_uid
),
3341 fsparam_flag ("inode32", Opt_inode32
),
3342 fsparam_flag ("inode64", Opt_inode64
),
3346 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3348 struct shmem_options
*ctx
= fc
->fs_private
;
3349 struct fs_parse_result result
;
3350 unsigned long long size
;
3354 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3360 size
= memparse(param
->string
, &rest
);
3362 size
<<= PAGE_SHIFT
;
3363 size
*= totalram_pages();
3369 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3370 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3373 ctx
->blocks
= memparse(param
->string
, &rest
);
3376 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3379 ctx
->inodes
= memparse(param
->string
, &rest
);
3382 ctx
->seen
|= SHMEM_SEEN_INODES
;
3385 ctx
->mode
= result
.uint_32
& 07777;
3388 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3389 if (!uid_valid(ctx
->uid
))
3393 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3394 if (!gid_valid(ctx
->gid
))
3398 ctx
->huge
= result
.uint_32
;
3399 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3400 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3401 has_transparent_hugepage()))
3402 goto unsupported_parameter
;
3403 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3406 if (IS_ENABLED(CONFIG_NUMA
)) {
3407 mpol_put(ctx
->mpol
);
3409 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3413 goto unsupported_parameter
;
3415 ctx
->full_inums
= false;
3416 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3419 if (sizeof(ino_t
) < 8) {
3421 "Cannot use inode64 with <64bit inums in kernel\n");
3423 ctx
->full_inums
= true;
3424 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3429 unsupported_parameter
:
3430 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3432 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3435 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3437 char *options
= data
;
3440 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3445 while (options
!= NULL
) {
3446 char *this_char
= options
;
3449 * NUL-terminate this option: unfortunately,
3450 * mount options form a comma-separated list,
3451 * but mpol's nodelist may also contain commas.
3453 options
= strchr(options
, ',');
3454 if (options
== NULL
)
3457 if (!isdigit(*options
)) {
3463 char *value
= strchr(this_char
, '=');
3469 len
= strlen(value
);
3471 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3480 * Reconfigure a shmem filesystem.
3482 * Note that we disallow change from limited->unlimited blocks/inodes while any
3483 * are in use; but we must separately disallow unlimited->limited, because in
3484 * that case we have no record of how much is already in use.
3486 static int shmem_reconfigure(struct fs_context
*fc
)
3488 struct shmem_options
*ctx
= fc
->fs_private
;
3489 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3490 unsigned long inodes
;
3493 spin_lock(&sbinfo
->stat_lock
);
3494 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3495 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3496 if (!sbinfo
->max_blocks
) {
3497 err
= "Cannot retroactively limit size";
3500 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3502 err
= "Too small a size for current use";
3506 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3507 if (!sbinfo
->max_inodes
) {
3508 err
= "Cannot retroactively limit inodes";
3511 if (ctx
->inodes
< inodes
) {
3512 err
= "Too few inodes for current use";
3517 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
3518 sbinfo
->next_ino
> UINT_MAX
) {
3519 err
= "Current inum too high to switch to 32-bit inums";
3523 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3524 sbinfo
->huge
= ctx
->huge
;
3525 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
3526 sbinfo
->full_inums
= ctx
->full_inums
;
3527 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3528 sbinfo
->max_blocks
= ctx
->blocks
;
3529 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3530 sbinfo
->max_inodes
= ctx
->inodes
;
3531 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3535 * Preserve previous mempolicy unless mpol remount option was specified.
3538 mpol_put(sbinfo
->mpol
);
3539 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3542 spin_unlock(&sbinfo
->stat_lock
);
3545 spin_unlock(&sbinfo
->stat_lock
);
3546 return invalfc(fc
, "%s", err
);
3549 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3551 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3553 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3554 seq_printf(seq
, ",size=%luk",
3555 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3556 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3557 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3558 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3559 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3560 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3561 seq_printf(seq
, ",uid=%u",
3562 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3563 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3564 seq_printf(seq
, ",gid=%u",
3565 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3568 * Showing inode{64,32} might be useful even if it's the system default,
3569 * since then people don't have to resort to checking both here and
3570 * /proc/config.gz to confirm 64-bit inums were successfully applied
3571 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3573 * We hide it when inode64 isn't the default and we are using 32-bit
3574 * inodes, since that probably just means the feature isn't even under
3579 * +-----------------+-----------------+
3580 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3581 * +------------------+-----------------+-----------------+
3582 * | full_inums=true | show | show |
3583 * | full_inums=false | show | hide |
3584 * +------------------+-----------------+-----------------+
3587 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
3588 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
3589 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3590 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3592 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3594 shmem_show_mpol(seq
, sbinfo
->mpol
);
3598 #endif /* CONFIG_TMPFS */
3600 static void shmem_put_super(struct super_block
*sb
)
3602 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3604 free_percpu(sbinfo
->ino_batch
);
3605 percpu_counter_destroy(&sbinfo
->used_blocks
);
3606 mpol_put(sbinfo
->mpol
);
3608 sb
->s_fs_info
= NULL
;
3611 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3613 struct shmem_options
*ctx
= fc
->fs_private
;
3614 struct inode
*inode
;
3615 struct shmem_sb_info
*sbinfo
;
3618 /* Round up to L1_CACHE_BYTES to resist false sharing */
3619 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3620 L1_CACHE_BYTES
), GFP_KERNEL
);
3624 sb
->s_fs_info
= sbinfo
;
3628 * Per default we only allow half of the physical ram per
3629 * tmpfs instance, limiting inodes to one per page of lowmem;
3630 * but the internal instance is left unlimited.
3632 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3633 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3634 ctx
->blocks
= shmem_default_max_blocks();
3635 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3636 ctx
->inodes
= shmem_default_max_inodes();
3637 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
3638 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
3640 sb
->s_flags
|= SB_NOUSER
;
3642 sb
->s_export_op
= &shmem_export_ops
;
3643 sb
->s_flags
|= SB_NOSEC
;
3645 sb
->s_flags
|= SB_NOUSER
;
3647 sbinfo
->max_blocks
= ctx
->blocks
;
3648 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3649 if (sb
->s_flags
& SB_KERNMOUNT
) {
3650 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
3651 if (!sbinfo
->ino_batch
)
3654 sbinfo
->uid
= ctx
->uid
;
3655 sbinfo
->gid
= ctx
->gid
;
3656 sbinfo
->full_inums
= ctx
->full_inums
;
3657 sbinfo
->mode
= ctx
->mode
;
3658 sbinfo
->huge
= ctx
->huge
;
3659 sbinfo
->mpol
= ctx
->mpol
;
3662 spin_lock_init(&sbinfo
->stat_lock
);
3663 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3665 spin_lock_init(&sbinfo
->shrinklist_lock
);
3666 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3668 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3669 sb
->s_blocksize
= PAGE_SIZE
;
3670 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3671 sb
->s_magic
= TMPFS_MAGIC
;
3672 sb
->s_op
= &shmem_ops
;
3673 sb
->s_time_gran
= 1;
3674 #ifdef CONFIG_TMPFS_XATTR
3675 sb
->s_xattr
= shmem_xattr_handlers
;
3677 #ifdef CONFIG_TMPFS_POSIX_ACL
3678 sb
->s_flags
|= SB_POSIXACL
;
3680 uuid_gen(&sb
->s_uuid
);
3682 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3685 inode
->i_uid
= sbinfo
->uid
;
3686 inode
->i_gid
= sbinfo
->gid
;
3687 sb
->s_root
= d_make_root(inode
);
3693 shmem_put_super(sb
);
3697 static int shmem_get_tree(struct fs_context
*fc
)
3699 return get_tree_nodev(fc
, shmem_fill_super
);
3702 static void shmem_free_fc(struct fs_context
*fc
)
3704 struct shmem_options
*ctx
= fc
->fs_private
;
3707 mpol_put(ctx
->mpol
);
3712 static const struct fs_context_operations shmem_fs_context_ops
= {
3713 .free
= shmem_free_fc
,
3714 .get_tree
= shmem_get_tree
,
3716 .parse_monolithic
= shmem_parse_options
,
3717 .parse_param
= shmem_parse_one
,
3718 .reconfigure
= shmem_reconfigure
,
3722 static struct kmem_cache
*shmem_inode_cachep
;
3724 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3726 struct shmem_inode_info
*info
;
3727 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3730 return &info
->vfs_inode
;
3733 static void shmem_free_in_core_inode(struct inode
*inode
)
3735 if (S_ISLNK(inode
->i_mode
))
3736 kfree(inode
->i_link
);
3737 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3740 static void shmem_destroy_inode(struct inode
*inode
)
3742 if (S_ISREG(inode
->i_mode
))
3743 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3746 static void shmem_init_inode(void *foo
)
3748 struct shmem_inode_info
*info
= foo
;
3749 inode_init_once(&info
->vfs_inode
);
3752 static void shmem_init_inodecache(void)
3754 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3755 sizeof(struct shmem_inode_info
),
3756 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3759 static void shmem_destroy_inodecache(void)
3761 kmem_cache_destroy(shmem_inode_cachep
);
3764 const struct address_space_operations shmem_aops
= {
3765 .writepage
= shmem_writepage
,
3766 .set_page_dirty
= __set_page_dirty_no_writeback
,
3768 .write_begin
= shmem_write_begin
,
3769 .write_end
= shmem_write_end
,
3771 #ifdef CONFIG_MIGRATION
3772 .migratepage
= migrate_page
,
3774 .error_remove_page
= generic_error_remove_page
,
3776 EXPORT_SYMBOL(shmem_aops
);
3778 static const struct file_operations shmem_file_operations
= {
3780 .get_unmapped_area
= shmem_get_unmapped_area
,
3782 .llseek
= shmem_file_llseek
,
3783 .read_iter
= shmem_file_read_iter
,
3784 .write_iter
= generic_file_write_iter
,
3785 .fsync
= noop_fsync
,
3786 .splice_read
= generic_file_splice_read
,
3787 .splice_write
= iter_file_splice_write
,
3788 .fallocate
= shmem_fallocate
,
3792 static const struct inode_operations shmem_inode_operations
= {
3793 .getattr
= shmem_getattr
,
3794 .setattr
= shmem_setattr
,
3795 #ifdef CONFIG_TMPFS_XATTR
3796 .listxattr
= shmem_listxattr
,
3797 .set_acl
= simple_set_acl
,
3801 static const struct inode_operations shmem_dir_inode_operations
= {
3803 .create
= shmem_create
,
3804 .lookup
= simple_lookup
,
3806 .unlink
= shmem_unlink
,
3807 .symlink
= shmem_symlink
,
3808 .mkdir
= shmem_mkdir
,
3809 .rmdir
= shmem_rmdir
,
3810 .mknod
= shmem_mknod
,
3811 .rename
= shmem_rename2
,
3812 .tmpfile
= shmem_tmpfile
,
3814 #ifdef CONFIG_TMPFS_XATTR
3815 .listxattr
= shmem_listxattr
,
3817 #ifdef CONFIG_TMPFS_POSIX_ACL
3818 .setattr
= shmem_setattr
,
3819 .set_acl
= simple_set_acl
,
3823 static const struct inode_operations shmem_special_inode_operations
= {
3824 #ifdef CONFIG_TMPFS_XATTR
3825 .listxattr
= shmem_listxattr
,
3827 #ifdef CONFIG_TMPFS_POSIX_ACL
3828 .setattr
= shmem_setattr
,
3829 .set_acl
= simple_set_acl
,
3833 static const struct super_operations shmem_ops
= {
3834 .alloc_inode
= shmem_alloc_inode
,
3835 .free_inode
= shmem_free_in_core_inode
,
3836 .destroy_inode
= shmem_destroy_inode
,
3838 .statfs
= shmem_statfs
,
3839 .show_options
= shmem_show_options
,
3841 .evict_inode
= shmem_evict_inode
,
3842 .drop_inode
= generic_delete_inode
,
3843 .put_super
= shmem_put_super
,
3844 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3845 .nr_cached_objects
= shmem_unused_huge_count
,
3846 .free_cached_objects
= shmem_unused_huge_scan
,
3850 static const struct vm_operations_struct shmem_vm_ops
= {
3851 .fault
= shmem_fault
,
3852 .map_pages
= filemap_map_pages
,
3854 .set_policy
= shmem_set_policy
,
3855 .get_policy
= shmem_get_policy
,
3859 int shmem_init_fs_context(struct fs_context
*fc
)
3861 struct shmem_options
*ctx
;
3863 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3867 ctx
->mode
= 0777 | S_ISVTX
;
3868 ctx
->uid
= current_fsuid();
3869 ctx
->gid
= current_fsgid();
3871 fc
->fs_private
= ctx
;
3872 fc
->ops
= &shmem_fs_context_ops
;
3876 static struct file_system_type shmem_fs_type
= {
3877 .owner
= THIS_MODULE
,
3879 .init_fs_context
= shmem_init_fs_context
,
3881 .parameters
= shmem_fs_parameters
,
3883 .kill_sb
= kill_litter_super
,
3884 .fs_flags
= FS_USERNS_MOUNT
| FS_THP_SUPPORT
,
3887 int __init
shmem_init(void)
3891 shmem_init_inodecache();
3893 error
= register_filesystem(&shmem_fs_type
);
3895 pr_err("Could not register tmpfs\n");
3899 shm_mnt
= kern_mount(&shmem_fs_type
);
3900 if (IS_ERR(shm_mnt
)) {
3901 error
= PTR_ERR(shm_mnt
);
3902 pr_err("Could not kern_mount tmpfs\n");
3906 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3907 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3908 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3910 shmem_huge
= 0; /* just in case it was patched */
3915 unregister_filesystem(&shmem_fs_type
);
3917 shmem_destroy_inodecache();
3918 shm_mnt
= ERR_PTR(error
);
3922 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3923 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3924 struct kobj_attribute
*attr
, char *buf
)
3926 static const int values
[] = {
3928 SHMEM_HUGE_WITHIN_SIZE
,
3937 for (i
= 0; i
< ARRAY_SIZE(values
); i
++) {
3938 len
+= sysfs_emit_at(buf
, len
,
3939 shmem_huge
== values
[i
] ? "%s[%s]" : "%s%s",
3941 shmem_format_huge(values
[i
]));
3944 len
+= sysfs_emit_at(buf
, len
, "\n");
3949 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3950 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3955 if (count
+ 1 > sizeof(tmp
))
3957 memcpy(tmp
, buf
, count
);
3959 if (count
&& tmp
[count
- 1] == '\n')
3960 tmp
[count
- 1] = '\0';
3962 huge
= shmem_parse_huge(tmp
);
3963 if (huge
== -EINVAL
)
3965 if (!has_transparent_hugepage() &&
3966 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3970 if (shmem_huge
> SHMEM_HUGE_DENY
)
3971 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3975 struct kobj_attribute shmem_enabled_attr
=
3976 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3977 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3979 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3980 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
3982 struct inode
*inode
= file_inode(vma
->vm_file
);
3983 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3987 if (!transhuge_vma_enabled(vma
, vma
->vm_flags
))
3989 if (shmem_huge
== SHMEM_HUGE_FORCE
)
3991 if (shmem_huge
== SHMEM_HUGE_DENY
)
3993 switch (sbinfo
->huge
) {
3994 case SHMEM_HUGE_NEVER
:
3996 case SHMEM_HUGE_ALWAYS
:
3998 case SHMEM_HUGE_WITHIN_SIZE
:
3999 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4000 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4001 if (i_size
>= HPAGE_PMD_SIZE
&&
4002 i_size
>> PAGE_SHIFT
>= off
)
4005 case SHMEM_HUGE_ADVISE
:
4006 /* TODO: implement fadvise() hints */
4007 return (vma
->vm_flags
& VM_HUGEPAGE
);
4013 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4015 #else /* !CONFIG_SHMEM */
4018 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4020 * This is intended for small system where the benefits of the full
4021 * shmem code (swap-backed and resource-limited) are outweighed by
4022 * their complexity. On systems without swap this code should be
4023 * effectively equivalent, but much lighter weight.
4026 static struct file_system_type shmem_fs_type
= {
4028 .init_fs_context
= ramfs_init_fs_context
,
4029 .parameters
= ramfs_fs_parameters
,
4030 .kill_sb
= kill_litter_super
,
4031 .fs_flags
= FS_USERNS_MOUNT
,
4034 int __init
shmem_init(void)
4036 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4038 shm_mnt
= kern_mount(&shmem_fs_type
);
4039 BUG_ON(IS_ERR(shm_mnt
));
4044 int shmem_unuse(unsigned int type
, bool frontswap
,
4045 unsigned long *fs_pages_to_unuse
)
4050 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
4055 void shmem_unlock_mapping(struct address_space
*mapping
)
4060 unsigned long shmem_get_unmapped_area(struct file
*file
,
4061 unsigned long addr
, unsigned long len
,
4062 unsigned long pgoff
, unsigned long flags
)
4064 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4068 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4070 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4072 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4074 #define shmem_vm_ops generic_file_vm_ops
4075 #define shmem_file_operations ramfs_file_operations
4076 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4077 #define shmem_acct_size(flags, size) 0
4078 #define shmem_unacct_size(flags, size) do {} while (0)
4080 #endif /* CONFIG_SHMEM */
4084 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4085 unsigned long flags
, unsigned int i_flags
)
4087 struct inode
*inode
;
4091 return ERR_CAST(mnt
);
4093 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4094 return ERR_PTR(-EINVAL
);
4096 if (shmem_acct_size(flags
, size
))
4097 return ERR_PTR(-ENOMEM
);
4099 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4101 if (unlikely(!inode
)) {
4102 shmem_unacct_size(flags
, size
);
4103 return ERR_PTR(-ENOSPC
);
4105 inode
->i_flags
|= i_flags
;
4106 inode
->i_size
= size
;
4107 clear_nlink(inode
); /* It is unlinked */
4108 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4110 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4111 &shmem_file_operations
);
4118 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4119 * kernel internal. There will be NO LSM permission checks against the
4120 * underlying inode. So users of this interface must do LSM checks at a
4121 * higher layer. The users are the big_key and shm implementations. LSM
4122 * checks are provided at the key or shm level rather than the inode.
4123 * @name: name for dentry (to be seen in /proc/<pid>/maps
4124 * @size: size to be set for the file
4125 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4127 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4129 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4133 * shmem_file_setup - get an unlinked file living in tmpfs
4134 * @name: name for dentry (to be seen in /proc/<pid>/maps
4135 * @size: size to be set for the file
4136 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4138 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4140 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4142 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4145 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4146 * @mnt: the tmpfs mount where the file will be created
4147 * @name: name for dentry (to be seen in /proc/<pid>/maps
4148 * @size: size to be set for the file
4149 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4151 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4152 loff_t size
, unsigned long flags
)
4154 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4156 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4159 * shmem_zero_setup - setup a shared anonymous mapping
4160 * @vma: the vma to be mmapped is prepared by do_mmap
4162 int shmem_zero_setup(struct vm_area_struct
*vma
)
4165 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4168 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4169 * between XFS directory reading and selinux: since this file is only
4170 * accessible to the user through its mapping, use S_PRIVATE flag to
4171 * bypass file security, in the same way as shmem_kernel_file_setup().
4173 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4175 return PTR_ERR(file
);
4179 vma
->vm_file
= file
;
4180 vma
->vm_ops
= &shmem_vm_ops
;
4182 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
4183 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4184 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4185 khugepaged_enter(vma
, vma
->vm_flags
);
4192 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4193 * @mapping: the page's address_space
4194 * @index: the page index
4195 * @gfp: the page allocator flags to use if allocating
4197 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4198 * with any new page allocations done using the specified allocation flags.
4199 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4200 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4201 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4203 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4204 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4206 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4207 pgoff_t index
, gfp_t gfp
)
4210 struct inode
*inode
= mapping
->host
;
4214 BUG_ON(!shmem_mapping(mapping
));
4215 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4216 gfp
, NULL
, NULL
, NULL
);
4218 page
= ERR_PTR(error
);
4224 * The tiny !SHMEM case uses ramfs without swap
4226 return read_cache_page_gfp(mapping
, index
, gfp
);
4229 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);