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>
31 #include <linux/fileattr.h>
33 #include <linux/random.h>
34 #include <linux/sched/signal.h>
35 #include <linux/export.h>
36 #include <linux/shmem_fs.h>
37 #include <linux/swap.h>
38 #include <linux/uio.h>
39 #include <linux/hugetlb.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
42 #include <linux/iversion.h>
45 static struct vfsmount
*shm_mnt
;
49 * This virtual memory filesystem is heavily based on the ramfs. It
50 * extends ramfs by the ability to use swap and honor resource limits
51 * which makes it a completely usable filesystem.
54 #include <linux/xattr.h>
55 #include <linux/exportfs.h>
56 #include <linux/posix_acl.h>
57 #include <linux/posix_acl_xattr.h>
58 #include <linux/mman.h>
59 #include <linux/string.h>
60 #include <linux/slab.h>
61 #include <linux/backing-dev.h>
62 #include <linux/writeback.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/userfaultfd_k.h>
80 #include <linux/rmap.h>
81 #include <linux/uuid.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_rwsem making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc
{
102 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
103 pgoff_t start
; /* start of range currently being fallocated */
104 pgoff_t next
; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
109 struct shmem_options
{
110 unsigned long long blocks
;
111 unsigned long long inodes
;
112 struct mempolicy
*mpol
;
119 #define SHMEM_SEEN_BLOCKS 1
120 #define SHMEM_SEEN_INODES 2
121 #define SHMEM_SEEN_HUGE 4
122 #define SHMEM_SEEN_INUMS 8
126 static unsigned long shmem_default_max_blocks(void)
128 return totalram_pages() / 2;
131 static unsigned long shmem_default_max_inodes(void)
133 unsigned long nr_pages
= totalram_pages();
135 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
139 static int shmem_swapin_folio(struct inode
*inode
, pgoff_t index
,
140 struct folio
**foliop
, enum sgp_type sgp
,
141 gfp_t gfp
, struct vm_area_struct
*vma
,
142 vm_fault_t
*fault_type
);
144 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
146 return sb
->s_fs_info
;
150 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
151 * for shared memory and for shared anonymous (/dev/zero) mappings
152 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
153 * consistent with the pre-accounting of private mappings ...
155 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
157 return (flags
& VM_NORESERVE
) ?
158 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
161 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
163 if (!(flags
& VM_NORESERVE
))
164 vm_unacct_memory(VM_ACCT(size
));
167 static inline int shmem_reacct_size(unsigned long flags
,
168 loff_t oldsize
, loff_t newsize
)
170 if (!(flags
& VM_NORESERVE
)) {
171 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
172 return security_vm_enough_memory_mm(current
->mm
,
173 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
174 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
175 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
181 * ... whereas tmpfs objects are accounted incrementally as
182 * pages are allocated, in order to allow large sparse files.
183 * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
184 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
186 static inline int shmem_acct_block(unsigned long flags
, long pages
)
188 if (!(flags
& VM_NORESERVE
))
191 return security_vm_enough_memory_mm(current
->mm
,
192 pages
* VM_ACCT(PAGE_SIZE
));
195 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
197 if (flags
& VM_NORESERVE
)
198 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
201 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
203 struct shmem_inode_info
*info
= SHMEM_I(inode
);
204 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
206 if (shmem_acct_block(info
->flags
, pages
))
209 if (sbinfo
->max_blocks
) {
210 if (percpu_counter_compare(&sbinfo
->used_blocks
,
211 sbinfo
->max_blocks
- pages
) > 0)
213 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
219 shmem_unacct_blocks(info
->flags
, pages
);
223 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
225 struct shmem_inode_info
*info
= SHMEM_I(inode
);
226 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
228 if (sbinfo
->max_blocks
)
229 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
230 shmem_unacct_blocks(info
->flags
, pages
);
233 static const struct super_operations shmem_ops
;
234 const struct address_space_operations shmem_aops
;
235 static const struct file_operations shmem_file_operations
;
236 static const struct inode_operations shmem_inode_operations
;
237 static const struct inode_operations shmem_dir_inode_operations
;
238 static const struct inode_operations shmem_special_inode_operations
;
239 static const struct vm_operations_struct shmem_vm_ops
;
240 static const struct vm_operations_struct shmem_anon_vm_ops
;
241 static struct file_system_type shmem_fs_type
;
243 bool vma_is_anon_shmem(struct vm_area_struct
*vma
)
245 return vma
->vm_ops
== &shmem_anon_vm_ops
;
248 bool vma_is_shmem(struct vm_area_struct
*vma
)
250 return vma_is_anon_shmem(vma
) || vma
->vm_ops
== &shmem_vm_ops
;
253 static LIST_HEAD(shmem_swaplist
);
254 static DEFINE_MUTEX(shmem_swaplist_mutex
);
257 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
258 * produces a novel ino for the newly allocated inode.
260 * It may also be called when making a hard link to permit the space needed by
261 * each dentry. However, in that case, no new inode number is needed since that
262 * internally draws from another pool of inode numbers (currently global
263 * get_next_ino()). This case is indicated by passing NULL as inop.
265 #define SHMEM_INO_BATCH 1024
266 static int shmem_reserve_inode(struct super_block
*sb
, ino_t
*inop
)
268 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
271 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
272 raw_spin_lock(&sbinfo
->stat_lock
);
273 if (sbinfo
->max_inodes
) {
274 if (!sbinfo
->free_inodes
) {
275 raw_spin_unlock(&sbinfo
->stat_lock
);
278 sbinfo
->free_inodes
--;
281 ino
= sbinfo
->next_ino
++;
282 if (unlikely(is_zero_ino(ino
)))
283 ino
= sbinfo
->next_ino
++;
284 if (unlikely(!sbinfo
->full_inums
&&
287 * Emulate get_next_ino uint wraparound for
290 if (IS_ENABLED(CONFIG_64BIT
))
291 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
292 __func__
, MINOR(sb
->s_dev
));
293 sbinfo
->next_ino
= 1;
294 ino
= sbinfo
->next_ino
++;
298 raw_spin_unlock(&sbinfo
->stat_lock
);
301 * __shmem_file_setup, one of our callers, is lock-free: it
302 * doesn't hold stat_lock in shmem_reserve_inode since
303 * max_inodes is always 0, and is called from potentially
304 * unknown contexts. As such, use a per-cpu batched allocator
305 * which doesn't require the per-sb stat_lock unless we are at
306 * the batch boundary.
308 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
309 * shmem mounts are not exposed to userspace, so we don't need
310 * to worry about things like glibc compatibility.
314 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
316 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
317 raw_spin_lock(&sbinfo
->stat_lock
);
318 ino
= sbinfo
->next_ino
;
319 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
320 raw_spin_unlock(&sbinfo
->stat_lock
);
321 if (unlikely(is_zero_ino(ino
)))
332 static void shmem_free_inode(struct super_block
*sb
)
334 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
335 if (sbinfo
->max_inodes
) {
336 raw_spin_lock(&sbinfo
->stat_lock
);
337 sbinfo
->free_inodes
++;
338 raw_spin_unlock(&sbinfo
->stat_lock
);
343 * shmem_recalc_inode - recalculate the block usage of an inode
344 * @inode: inode to recalc
346 * We have to calculate the free blocks since the mm can drop
347 * undirtied hole pages behind our back.
349 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
350 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
352 * It has to be called with the spinlock held.
354 static void shmem_recalc_inode(struct inode
*inode
)
356 struct shmem_inode_info
*info
= SHMEM_I(inode
);
359 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
361 info
->alloced
-= freed
;
362 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
363 shmem_inode_unacct_blocks(inode
, freed
);
367 bool shmem_charge(struct inode
*inode
, long pages
)
369 struct shmem_inode_info
*info
= SHMEM_I(inode
);
372 if (!shmem_inode_acct_block(inode
, pages
))
375 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
376 inode
->i_mapping
->nrpages
+= pages
;
378 spin_lock_irqsave(&info
->lock
, flags
);
379 info
->alloced
+= pages
;
380 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
381 shmem_recalc_inode(inode
);
382 spin_unlock_irqrestore(&info
->lock
, flags
);
387 void shmem_uncharge(struct inode
*inode
, long pages
)
389 struct shmem_inode_info
*info
= SHMEM_I(inode
);
392 /* nrpages adjustment done by __filemap_remove_folio() or caller */
394 spin_lock_irqsave(&info
->lock
, flags
);
395 info
->alloced
-= pages
;
396 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
397 shmem_recalc_inode(inode
);
398 spin_unlock_irqrestore(&info
->lock
, flags
);
400 shmem_inode_unacct_blocks(inode
, pages
);
404 * Replace item expected in xarray by a new item, while holding xa_lock.
406 static int shmem_replace_entry(struct address_space
*mapping
,
407 pgoff_t index
, void *expected
, void *replacement
)
409 XA_STATE(xas
, &mapping
->i_pages
, index
);
412 VM_BUG_ON(!expected
);
413 VM_BUG_ON(!replacement
);
414 item
= xas_load(&xas
);
415 if (item
!= expected
)
417 xas_store(&xas
, replacement
);
422 * Sometimes, before we decide whether to proceed or to fail, we must check
423 * that an entry was not already brought back from swap by a racing thread.
425 * Checking page is not enough: by the time a SwapCache page is locked, it
426 * might be reused, and again be SwapCache, using the same swap as before.
428 static bool shmem_confirm_swap(struct address_space
*mapping
,
429 pgoff_t index
, swp_entry_t swap
)
431 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
435 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
438 * disables huge pages for the mount;
440 * enables huge pages for the mount;
441 * SHMEM_HUGE_WITHIN_SIZE:
442 * only allocate huge pages if the page will be fully within i_size,
443 * also respect fadvise()/madvise() hints;
445 * only allocate huge pages if requested with fadvise()/madvise();
448 #define SHMEM_HUGE_NEVER 0
449 #define SHMEM_HUGE_ALWAYS 1
450 #define SHMEM_HUGE_WITHIN_SIZE 2
451 #define SHMEM_HUGE_ADVISE 3
455 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
458 * disables huge on shm_mnt and all mounts, for emergency use;
460 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
463 #define SHMEM_HUGE_DENY (-1)
464 #define SHMEM_HUGE_FORCE (-2)
466 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
467 /* ifdef here to avoid bloating shmem.o when not necessary */
469 static int shmem_huge __read_mostly
= SHMEM_HUGE_NEVER
;
471 bool shmem_is_huge(struct inode
*inode
, pgoff_t index
, bool shmem_huge_force
,
472 struct mm_struct
*mm
, unsigned long vm_flags
)
476 if (!S_ISREG(inode
->i_mode
))
478 if (mm
&& ((vm_flags
& VM_NOHUGEPAGE
) || test_bit(MMF_DISABLE_THP
, &mm
->flags
)))
480 if (shmem_huge
== SHMEM_HUGE_DENY
)
482 if (shmem_huge_force
|| shmem_huge
== SHMEM_HUGE_FORCE
)
485 switch (SHMEM_SB(inode
->i_sb
)->huge
) {
486 case SHMEM_HUGE_ALWAYS
:
488 case SHMEM_HUGE_WITHIN_SIZE
:
489 index
= round_up(index
+ 1, HPAGE_PMD_NR
);
490 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
491 if (i_size
>> PAGE_SHIFT
>= index
)
494 case SHMEM_HUGE_ADVISE
:
495 if (mm
&& (vm_flags
& VM_HUGEPAGE
))
503 #if defined(CONFIG_SYSFS)
504 static int shmem_parse_huge(const char *str
)
506 if (!strcmp(str
, "never"))
507 return SHMEM_HUGE_NEVER
;
508 if (!strcmp(str
, "always"))
509 return SHMEM_HUGE_ALWAYS
;
510 if (!strcmp(str
, "within_size"))
511 return SHMEM_HUGE_WITHIN_SIZE
;
512 if (!strcmp(str
, "advise"))
513 return SHMEM_HUGE_ADVISE
;
514 if (!strcmp(str
, "deny"))
515 return SHMEM_HUGE_DENY
;
516 if (!strcmp(str
, "force"))
517 return SHMEM_HUGE_FORCE
;
522 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
523 static const char *shmem_format_huge(int huge
)
526 case SHMEM_HUGE_NEVER
:
528 case SHMEM_HUGE_ALWAYS
:
530 case SHMEM_HUGE_WITHIN_SIZE
:
531 return "within_size";
532 case SHMEM_HUGE_ADVISE
:
534 case SHMEM_HUGE_DENY
:
536 case SHMEM_HUGE_FORCE
:
545 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
546 struct shrink_control
*sc
, unsigned long nr_to_split
)
548 LIST_HEAD(list
), *pos
, *next
;
549 LIST_HEAD(to_remove
);
551 struct shmem_inode_info
*info
;
553 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
556 if (list_empty(&sbinfo
->shrinklist
))
559 spin_lock(&sbinfo
->shrinklist_lock
);
560 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
561 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
564 inode
= igrab(&info
->vfs_inode
);
566 /* inode is about to be evicted */
568 list_del_init(&info
->shrinklist
);
572 /* Check if there's anything to gain */
573 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
574 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
575 list_move(&info
->shrinklist
, &to_remove
);
579 list_move(&info
->shrinklist
, &list
);
581 sbinfo
->shrinklist_len
--;
585 spin_unlock(&sbinfo
->shrinklist_lock
);
587 list_for_each_safe(pos
, next
, &to_remove
) {
588 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
589 inode
= &info
->vfs_inode
;
590 list_del_init(&info
->shrinklist
);
594 list_for_each_safe(pos
, next
, &list
) {
598 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
599 inode
= &info
->vfs_inode
;
601 if (nr_to_split
&& split
>= nr_to_split
)
604 index
= (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
;
605 folio
= filemap_get_folio(inode
->i_mapping
, index
);
609 /* No huge page at the end of the file: nothing to split */
610 if (!folio_test_large(folio
)) {
616 * Move the inode on the list back to shrinklist if we failed
617 * to lock the page at this time.
619 * Waiting for the lock may lead to deadlock in the
622 if (!folio_trylock(folio
)) {
627 ret
= split_folio(folio
);
631 /* If split failed move the inode on the list back to shrinklist */
637 list_del_init(&info
->shrinklist
);
641 * Make sure the inode is either on the global list or deleted
642 * from any local list before iput() since it could be deleted
643 * in another thread once we put the inode (then the local list
646 spin_lock(&sbinfo
->shrinklist_lock
);
647 list_move(&info
->shrinklist
, &sbinfo
->shrinklist
);
648 sbinfo
->shrinklist_len
++;
649 spin_unlock(&sbinfo
->shrinklist_lock
);
657 static long shmem_unused_huge_scan(struct super_block
*sb
,
658 struct shrink_control
*sc
)
660 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
662 if (!READ_ONCE(sbinfo
->shrinklist_len
))
665 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
668 static long shmem_unused_huge_count(struct super_block
*sb
,
669 struct shrink_control
*sc
)
671 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
672 return READ_ONCE(sbinfo
->shrinklist_len
);
674 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
676 #define shmem_huge SHMEM_HUGE_DENY
678 bool shmem_is_huge(struct inode
*inode
, pgoff_t index
, bool shmem_huge_force
,
679 struct mm_struct
*mm
, unsigned long vm_flags
)
684 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
685 struct shrink_control
*sc
, unsigned long nr_to_split
)
689 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
692 * Like filemap_add_folio, but error if expected item has gone.
694 static int shmem_add_to_page_cache(struct folio
*folio
,
695 struct address_space
*mapping
,
696 pgoff_t index
, void *expected
, gfp_t gfp
,
697 struct mm_struct
*charge_mm
)
699 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, folio_order(folio
));
700 long nr
= folio_nr_pages(folio
);
703 VM_BUG_ON_FOLIO(index
!= round_down(index
, nr
), folio
);
704 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
705 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio
), folio
);
706 VM_BUG_ON(expected
&& folio_test_large(folio
));
708 folio_ref_add(folio
, nr
);
709 folio
->mapping
= mapping
;
710 folio
->index
= index
;
712 if (!folio_test_swapcache(folio
)) {
713 error
= mem_cgroup_charge(folio
, charge_mm
, gfp
);
715 if (folio_test_pmd_mappable(folio
)) {
716 count_vm_event(THP_FILE_FALLBACK
);
717 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
722 folio_throttle_swaprate(folio
, gfp
);
726 if (expected
!= xas_find_conflict(&xas
)) {
727 xas_set_err(&xas
, -EEXIST
);
730 if (expected
&& xas_find_conflict(&xas
)) {
731 xas_set_err(&xas
, -EEXIST
);
734 xas_store(&xas
, folio
);
737 if (folio_test_pmd_mappable(folio
)) {
738 count_vm_event(THP_FILE_ALLOC
);
739 __lruvec_stat_mod_folio(folio
, NR_SHMEM_THPS
, nr
);
741 mapping
->nrpages
+= nr
;
742 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, nr
);
743 __lruvec_stat_mod_folio(folio
, NR_SHMEM
, nr
);
745 xas_unlock_irq(&xas
);
746 } while (xas_nomem(&xas
, gfp
));
748 if (xas_error(&xas
)) {
749 error
= xas_error(&xas
);
755 folio
->mapping
= NULL
;
756 folio_ref_sub(folio
, nr
);
761 * Like delete_from_page_cache, but substitutes swap for @folio.
763 static void shmem_delete_from_page_cache(struct folio
*folio
, void *radswap
)
765 struct address_space
*mapping
= folio
->mapping
;
766 long nr
= folio_nr_pages(folio
);
769 xa_lock_irq(&mapping
->i_pages
);
770 error
= shmem_replace_entry(mapping
, folio
->index
, folio
, radswap
);
771 folio
->mapping
= NULL
;
772 mapping
->nrpages
-= nr
;
773 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, -nr
);
774 __lruvec_stat_mod_folio(folio
, NR_SHMEM
, -nr
);
775 xa_unlock_irq(&mapping
->i_pages
);
781 * Remove swap entry from page cache, free the swap and its page cache.
783 static int shmem_free_swap(struct address_space
*mapping
,
784 pgoff_t index
, void *radswap
)
788 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
791 free_swap_and_cache(radix_to_swp_entry(radswap
));
796 * Determine (in bytes) how many of the shmem object's pages mapped by the
797 * given offsets are swapped out.
799 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
800 * as long as the inode doesn't go away and racy results are not a problem.
802 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
803 pgoff_t start
, pgoff_t end
)
805 XA_STATE(xas
, &mapping
->i_pages
, start
);
807 unsigned long swapped
= 0;
810 xas_for_each(&xas
, page
, end
- 1) {
811 if (xas_retry(&xas
, page
))
813 if (xa_is_value(page
))
816 if (need_resched()) {
824 return swapped
<< PAGE_SHIFT
;
828 * Determine (in bytes) how many of the shmem object's pages mapped by the
829 * given vma is swapped out.
831 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
832 * as long as the inode doesn't go away and racy results are not a problem.
834 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
836 struct inode
*inode
= file_inode(vma
->vm_file
);
837 struct shmem_inode_info
*info
= SHMEM_I(inode
);
838 struct address_space
*mapping
= inode
->i_mapping
;
839 unsigned long swapped
;
841 /* Be careful as we don't hold info->lock */
842 swapped
= READ_ONCE(info
->swapped
);
845 * The easier cases are when the shmem object has nothing in swap, or
846 * the vma maps it whole. Then we can simply use the stats that we
852 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
853 return swapped
<< PAGE_SHIFT
;
855 /* Here comes the more involved part */
856 return shmem_partial_swap_usage(mapping
, vma
->vm_pgoff
,
857 vma
->vm_pgoff
+ vma_pages(vma
));
861 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
863 void shmem_unlock_mapping(struct address_space
*mapping
)
865 struct folio_batch fbatch
;
868 folio_batch_init(&fbatch
);
870 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
872 while (!mapping_unevictable(mapping
) &&
873 filemap_get_folios(mapping
, &index
, ~0UL, &fbatch
)) {
874 check_move_unevictable_folios(&fbatch
);
875 folio_batch_release(&fbatch
);
880 static struct folio
*shmem_get_partial_folio(struct inode
*inode
, pgoff_t index
)
885 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
886 * beyond i_size, and reports fallocated pages as holes.
888 folio
= __filemap_get_folio(inode
->i_mapping
, index
,
889 FGP_ENTRY
| FGP_LOCK
, 0);
890 if (!xa_is_value(folio
))
893 * But read a page back from swap if any of it is within i_size
894 * (although in some cases this is just a waste of time).
897 shmem_get_folio(inode
, index
, &folio
, SGP_READ
);
902 * Remove range of pages and swap entries from page cache, and free them.
903 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
905 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
908 struct address_space
*mapping
= inode
->i_mapping
;
909 struct shmem_inode_info
*info
= SHMEM_I(inode
);
910 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
911 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
912 struct folio_batch fbatch
;
913 pgoff_t indices
[PAGEVEC_SIZE
];
916 long nr_swaps_freed
= 0;
921 end
= -1; /* unsigned, so actually very big */
923 if (info
->fallocend
> start
&& info
->fallocend
<= end
&& !unfalloc
)
924 info
->fallocend
= start
;
926 folio_batch_init(&fbatch
);
928 while (index
< end
&& find_lock_entries(mapping
, &index
, end
- 1,
930 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
931 folio
= fbatch
.folios
[i
];
933 if (xa_is_value(folio
)) {
936 nr_swaps_freed
+= !shmem_free_swap(mapping
,
941 if (!unfalloc
|| !folio_test_uptodate(folio
))
942 truncate_inode_folio(mapping
, folio
);
945 folio_batch_remove_exceptionals(&fbatch
);
946 folio_batch_release(&fbatch
);
951 * When undoing a failed fallocate, we want none of the partial folio
952 * zeroing and splitting below, but shall want to truncate the whole
953 * folio when !uptodate indicates that it was added by this fallocate,
954 * even when [lstart, lend] covers only a part of the folio.
959 same_folio
= (lstart
>> PAGE_SHIFT
) == (lend
>> PAGE_SHIFT
);
960 folio
= shmem_get_partial_folio(inode
, lstart
>> PAGE_SHIFT
);
962 same_folio
= lend
< folio_pos(folio
) + folio_size(folio
);
963 folio_mark_dirty(folio
);
964 if (!truncate_inode_partial_folio(folio
, lstart
, lend
)) {
965 start
= folio
->index
+ folio_nr_pages(folio
);
975 folio
= shmem_get_partial_folio(inode
, lend
>> PAGE_SHIFT
);
977 folio_mark_dirty(folio
);
978 if (!truncate_inode_partial_folio(folio
, lstart
, lend
))
987 while (index
< end
) {
990 if (!find_get_entries(mapping
, &index
, end
- 1, &fbatch
,
992 /* If all gone or hole-punch or unfalloc, we're done */
993 if (index
== start
|| end
!= -1)
995 /* But if truncating, restart to make sure all gone */
999 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
1000 folio
= fbatch
.folios
[i
];
1002 if (xa_is_value(folio
)) {
1005 if (shmem_free_swap(mapping
, indices
[i
], folio
)) {
1006 /* Swap was replaced by page: retry */
1016 if (!unfalloc
|| !folio_test_uptodate(folio
)) {
1017 if (folio_mapping(folio
) != mapping
) {
1018 /* Page was replaced by swap: retry */
1019 folio_unlock(folio
);
1023 VM_BUG_ON_FOLIO(folio_test_writeback(folio
),
1025 truncate_inode_folio(mapping
, folio
);
1027 folio_unlock(folio
);
1029 folio_batch_remove_exceptionals(&fbatch
);
1030 folio_batch_release(&fbatch
);
1033 spin_lock_irq(&info
->lock
);
1034 info
->swapped
-= nr_swaps_freed
;
1035 shmem_recalc_inode(inode
);
1036 spin_unlock_irq(&info
->lock
);
1039 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1041 shmem_undo_range(inode
, lstart
, lend
, false);
1042 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1043 inode_inc_iversion(inode
);
1045 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1047 static int shmem_getattr(struct mnt_idmap
*idmap
,
1048 const struct path
*path
, struct kstat
*stat
,
1049 u32 request_mask
, unsigned int query_flags
)
1051 struct inode
*inode
= path
->dentry
->d_inode
;
1052 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1054 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1055 spin_lock_irq(&info
->lock
);
1056 shmem_recalc_inode(inode
);
1057 spin_unlock_irq(&info
->lock
);
1059 if (info
->fsflags
& FS_APPEND_FL
)
1060 stat
->attributes
|= STATX_ATTR_APPEND
;
1061 if (info
->fsflags
& FS_IMMUTABLE_FL
)
1062 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
1063 if (info
->fsflags
& FS_NODUMP_FL
)
1064 stat
->attributes
|= STATX_ATTR_NODUMP
;
1065 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
1066 STATX_ATTR_IMMUTABLE
|
1068 generic_fillattr(idmap
, inode
, stat
);
1070 if (shmem_is_huge(inode
, 0, false, NULL
, 0))
1071 stat
->blksize
= HPAGE_PMD_SIZE
;
1073 if (request_mask
& STATX_BTIME
) {
1074 stat
->result_mask
|= STATX_BTIME
;
1075 stat
->btime
.tv_sec
= info
->i_crtime
.tv_sec
;
1076 stat
->btime
.tv_nsec
= info
->i_crtime
.tv_nsec
;
1082 static int shmem_setattr(struct mnt_idmap
*idmap
,
1083 struct dentry
*dentry
, struct iattr
*attr
)
1085 struct inode
*inode
= d_inode(dentry
);
1086 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1088 bool update_mtime
= false;
1089 bool update_ctime
= true;
1091 error
= setattr_prepare(idmap
, dentry
, attr
);
1095 if ((info
->seals
& F_SEAL_EXEC
) && (attr
->ia_valid
& ATTR_MODE
)) {
1096 if ((inode
->i_mode
^ attr
->ia_mode
) & 0111) {
1101 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1102 loff_t oldsize
= inode
->i_size
;
1103 loff_t newsize
= attr
->ia_size
;
1105 /* protected by i_rwsem */
1106 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1107 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1110 if (newsize
!= oldsize
) {
1111 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1115 i_size_write(inode
, newsize
);
1116 update_mtime
= true;
1118 update_ctime
= false;
1120 if (newsize
<= oldsize
) {
1121 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1122 if (oldsize
> holebegin
)
1123 unmap_mapping_range(inode
->i_mapping
,
1126 shmem_truncate_range(inode
,
1127 newsize
, (loff_t
)-1);
1128 /* unmap again to remove racily COWed private pages */
1129 if (oldsize
> holebegin
)
1130 unmap_mapping_range(inode
->i_mapping
,
1135 setattr_copy(idmap
, inode
, attr
);
1136 if (attr
->ia_valid
& ATTR_MODE
)
1137 error
= posix_acl_chmod(idmap
, dentry
, inode
->i_mode
);
1138 if (!error
&& update_ctime
) {
1139 inode
->i_ctime
= current_time(inode
);
1141 inode
->i_mtime
= inode
->i_ctime
;
1142 inode_inc_iversion(inode
);
1147 static void shmem_evict_inode(struct inode
*inode
)
1149 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1150 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1152 if (shmem_mapping(inode
->i_mapping
)) {
1153 shmem_unacct_size(info
->flags
, inode
->i_size
);
1155 mapping_set_exiting(inode
->i_mapping
);
1156 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1157 if (!list_empty(&info
->shrinklist
)) {
1158 spin_lock(&sbinfo
->shrinklist_lock
);
1159 if (!list_empty(&info
->shrinklist
)) {
1160 list_del_init(&info
->shrinklist
);
1161 sbinfo
->shrinklist_len
--;
1163 spin_unlock(&sbinfo
->shrinklist_lock
);
1165 while (!list_empty(&info
->swaplist
)) {
1166 /* Wait while shmem_unuse() is scanning this inode... */
1167 wait_var_event(&info
->stop_eviction
,
1168 !atomic_read(&info
->stop_eviction
));
1169 mutex_lock(&shmem_swaplist_mutex
);
1170 /* ...but beware of the race if we peeked too early */
1171 if (!atomic_read(&info
->stop_eviction
))
1172 list_del_init(&info
->swaplist
);
1173 mutex_unlock(&shmem_swaplist_mutex
);
1177 simple_xattrs_free(&info
->xattrs
);
1178 WARN_ON(inode
->i_blocks
);
1179 shmem_free_inode(inode
->i_sb
);
1183 static int shmem_find_swap_entries(struct address_space
*mapping
,
1184 pgoff_t start
, struct folio_batch
*fbatch
,
1185 pgoff_t
*indices
, unsigned int type
)
1187 XA_STATE(xas
, &mapping
->i_pages
, start
);
1188 struct folio
*folio
;
1192 xas_for_each(&xas
, folio
, ULONG_MAX
) {
1193 if (xas_retry(&xas
, folio
))
1196 if (!xa_is_value(folio
))
1199 entry
= radix_to_swp_entry(folio
);
1201 * swapin error entries can be found in the mapping. But they're
1202 * deliberately ignored here as we've done everything we can do.
1204 if (swp_type(entry
) != type
)
1207 indices
[folio_batch_count(fbatch
)] = xas
.xa_index
;
1208 if (!folio_batch_add(fbatch
, folio
))
1211 if (need_resched()) {
1218 return xas
.xa_index
;
1222 * Move the swapped pages for an inode to page cache. Returns the count
1223 * of pages swapped in, or the error in case of failure.
1225 static int shmem_unuse_swap_entries(struct inode
*inode
,
1226 struct folio_batch
*fbatch
, pgoff_t
*indices
)
1231 struct address_space
*mapping
= inode
->i_mapping
;
1233 for (i
= 0; i
< folio_batch_count(fbatch
); i
++) {
1234 struct folio
*folio
= fbatch
->folios
[i
];
1236 if (!xa_is_value(folio
))
1238 error
= shmem_swapin_folio(inode
, indices
[i
],
1240 mapping_gfp_mask(mapping
),
1243 folio_unlock(folio
);
1247 if (error
== -ENOMEM
)
1251 return error
? error
: ret
;
1255 * If swap found in inode, free it and move page from swapcache to filecache.
1257 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
)
1259 struct address_space
*mapping
= inode
->i_mapping
;
1261 struct folio_batch fbatch
;
1262 pgoff_t indices
[PAGEVEC_SIZE
];
1266 folio_batch_init(&fbatch
);
1267 shmem_find_swap_entries(mapping
, start
, &fbatch
, indices
, type
);
1268 if (folio_batch_count(&fbatch
) == 0) {
1273 ret
= shmem_unuse_swap_entries(inode
, &fbatch
, indices
);
1277 start
= indices
[folio_batch_count(&fbatch
) - 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
)
1290 struct shmem_inode_info
*info
, *next
;
1293 if (list_empty(&shmem_swaplist
))
1296 mutex_lock(&shmem_swaplist_mutex
);
1297 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1298 if (!info
->swapped
) {
1299 list_del_init(&info
->swaplist
);
1303 * Drop the swaplist mutex while searching the inode for swap;
1304 * but before doing so, make sure shmem_evict_inode() will not
1305 * remove placeholder inode from swaplist, nor let it be freed
1306 * (igrab() would protect from unlink, but not from unmount).
1308 atomic_inc(&info
->stop_eviction
);
1309 mutex_unlock(&shmem_swaplist_mutex
);
1311 error
= shmem_unuse_inode(&info
->vfs_inode
, type
);
1314 mutex_lock(&shmem_swaplist_mutex
);
1315 next
= list_next_entry(info
, swaplist
);
1317 list_del_init(&info
->swaplist
);
1318 if (atomic_dec_and_test(&info
->stop_eviction
))
1319 wake_up_var(&info
->stop_eviction
);
1323 mutex_unlock(&shmem_swaplist_mutex
);
1329 * Move the page from the page cache to the swap cache.
1331 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1333 struct folio
*folio
= page_folio(page
);
1334 struct shmem_inode_info
*info
;
1335 struct address_space
*mapping
;
1336 struct inode
*inode
;
1341 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1342 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1343 * and its shmem_writeback() needs them to be split when swapping.
1345 if (folio_test_large(folio
)) {
1346 /* Ensure the subpages are still dirty */
1347 folio_test_set_dirty(folio
);
1348 if (split_huge_page(page
) < 0)
1350 folio
= page_folio(page
);
1351 folio_clear_dirty(folio
);
1354 BUG_ON(!folio_test_locked(folio
));
1355 mapping
= folio
->mapping
;
1356 index
= folio
->index
;
1357 inode
= mapping
->host
;
1358 info
= SHMEM_I(inode
);
1359 if (info
->flags
& VM_LOCKED
)
1361 if (!total_swap_pages
)
1365 * Our capabilities prevent regular writeback or sync from ever calling
1366 * shmem_writepage; but a stacking filesystem might use ->writepage of
1367 * its underlying filesystem, in which case tmpfs should write out to
1368 * swap only in response to memory pressure, and not for the writeback
1371 if (!wbc
->for_reclaim
) {
1372 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1377 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1378 * value into swapfile.c, the only way we can correctly account for a
1379 * fallocated folio arriving here is now to initialize it and write it.
1381 * That's okay for a folio already fallocated earlier, but if we have
1382 * not yet completed the fallocation, then (a) we want to keep track
1383 * of this folio in case we have to undo it, and (b) it may not be a
1384 * good idea to continue anyway, once we're pushing into swap. So
1385 * reactivate the folio, and let shmem_fallocate() quit when too many.
1387 if (!folio_test_uptodate(folio
)) {
1388 if (inode
->i_private
) {
1389 struct shmem_falloc
*shmem_falloc
;
1390 spin_lock(&inode
->i_lock
);
1391 shmem_falloc
= inode
->i_private
;
1393 !shmem_falloc
->waitq
&&
1394 index
>= shmem_falloc
->start
&&
1395 index
< shmem_falloc
->next
)
1396 shmem_falloc
->nr_unswapped
++;
1398 shmem_falloc
= NULL
;
1399 spin_unlock(&inode
->i_lock
);
1403 folio_zero_range(folio
, 0, folio_size(folio
));
1404 flush_dcache_folio(folio
);
1405 folio_mark_uptodate(folio
);
1408 swap
= folio_alloc_swap(folio
);
1413 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1414 * if it's not already there. Do it now before the folio is
1415 * moved to swap cache, when its pagelock no longer protects
1416 * the inode from eviction. But don't unlock the mutex until
1417 * we've incremented swapped, because shmem_unuse_inode() will
1418 * prune a !swapped inode from the swaplist under this mutex.
1420 mutex_lock(&shmem_swaplist_mutex
);
1421 if (list_empty(&info
->swaplist
))
1422 list_add(&info
->swaplist
, &shmem_swaplist
);
1424 if (add_to_swap_cache(folio
, swap
,
1425 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1427 spin_lock_irq(&info
->lock
);
1428 shmem_recalc_inode(inode
);
1430 spin_unlock_irq(&info
->lock
);
1432 swap_shmem_alloc(swap
);
1433 shmem_delete_from_page_cache(folio
, swp_to_radix_entry(swap
));
1435 mutex_unlock(&shmem_swaplist_mutex
);
1436 BUG_ON(folio_mapped(folio
));
1437 swap_writepage(&folio
->page
, wbc
);
1441 mutex_unlock(&shmem_swaplist_mutex
);
1442 put_swap_folio(folio
, swap
);
1444 folio_mark_dirty(folio
);
1445 if (wbc
->for_reclaim
)
1446 return AOP_WRITEPAGE_ACTIVATE
; /* Return with folio locked */
1447 folio_unlock(folio
);
1451 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1452 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1456 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1457 return; /* show nothing */
1459 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1461 seq_printf(seq
, ",mpol=%s", buffer
);
1464 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1466 struct mempolicy
*mpol
= NULL
;
1468 raw_spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1469 mpol
= sbinfo
->mpol
;
1471 raw_spin_unlock(&sbinfo
->stat_lock
);
1475 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1476 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1479 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1483 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1485 #define vm_policy vm_private_data
1488 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1489 struct shmem_inode_info
*info
, pgoff_t index
)
1491 /* Create a pseudo vma that just contains the policy */
1492 vma_init(vma
, NULL
);
1493 /* Bias interleave by inode number to distribute better across nodes */
1494 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1495 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1498 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1500 /* Drop reference taken by mpol_shared_policy_lookup() */
1501 mpol_cond_put(vma
->vm_policy
);
1504 static struct folio
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1505 struct shmem_inode_info
*info
, pgoff_t index
)
1507 struct vm_area_struct pvma
;
1509 struct vm_fault vmf
= {
1513 shmem_pseudo_vma_init(&pvma
, info
, index
);
1514 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1515 shmem_pseudo_vma_destroy(&pvma
);
1519 return page_folio(page
);
1523 * Make sure huge_gfp is always more limited than limit_gfp.
1524 * Some of the flags set permissions, while others set limitations.
1526 static gfp_t
limit_gfp_mask(gfp_t huge_gfp
, gfp_t limit_gfp
)
1528 gfp_t allowflags
= __GFP_IO
| __GFP_FS
| __GFP_RECLAIM
;
1529 gfp_t denyflags
= __GFP_NOWARN
| __GFP_NORETRY
;
1530 gfp_t zoneflags
= limit_gfp
& GFP_ZONEMASK
;
1531 gfp_t result
= huge_gfp
& ~(allowflags
| GFP_ZONEMASK
);
1533 /* Allow allocations only from the originally specified zones. */
1534 result
|= zoneflags
;
1537 * Minimize the result gfp by taking the union with the deny flags,
1538 * and the intersection of the allow flags.
1540 result
|= (limit_gfp
& denyflags
);
1541 result
|= (huge_gfp
& limit_gfp
) & allowflags
;
1546 static struct folio
*shmem_alloc_hugefolio(gfp_t gfp
,
1547 struct shmem_inode_info
*info
, pgoff_t index
)
1549 struct vm_area_struct pvma
;
1550 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1552 struct folio
*folio
;
1554 hindex
= round_down(index
, HPAGE_PMD_NR
);
1555 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1559 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1560 folio
= vma_alloc_folio(gfp
, HPAGE_PMD_ORDER
, &pvma
, 0, true);
1561 shmem_pseudo_vma_destroy(&pvma
);
1563 count_vm_event(THP_FILE_FALLBACK
);
1567 static struct folio
*shmem_alloc_folio(gfp_t gfp
,
1568 struct shmem_inode_info
*info
, pgoff_t index
)
1570 struct vm_area_struct pvma
;
1571 struct folio
*folio
;
1573 shmem_pseudo_vma_init(&pvma
, info
, index
);
1574 folio
= vma_alloc_folio(gfp
, 0, &pvma
, 0, false);
1575 shmem_pseudo_vma_destroy(&pvma
);
1580 static struct folio
*shmem_alloc_and_acct_folio(gfp_t gfp
, struct inode
*inode
,
1581 pgoff_t index
, bool huge
)
1583 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1584 struct folio
*folio
;
1588 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1590 nr
= huge
? HPAGE_PMD_NR
: 1;
1592 if (!shmem_inode_acct_block(inode
, nr
))
1596 folio
= shmem_alloc_hugefolio(gfp
, info
, index
);
1598 folio
= shmem_alloc_folio(gfp
, info
, index
);
1600 __folio_set_locked(folio
);
1601 __folio_set_swapbacked(folio
);
1606 shmem_inode_unacct_blocks(inode
, nr
);
1608 return ERR_PTR(err
);
1612 * When a page is moved from swapcache to shmem filecache (either by the
1613 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1614 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1615 * ignorance of the mapping it belongs to. If that mapping has special
1616 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1617 * we may need to copy to a suitable page before moving to filecache.
1619 * In a future release, this may well be extended to respect cpuset and
1620 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1621 * but for now it is a simple matter of zone.
1623 static bool shmem_should_replace_folio(struct folio
*folio
, gfp_t gfp
)
1625 return folio_zonenum(folio
) > gfp_zone(gfp
);
1628 static int shmem_replace_folio(struct folio
**foliop
, gfp_t gfp
,
1629 struct shmem_inode_info
*info
, pgoff_t index
)
1631 struct folio
*old
, *new;
1632 struct address_space
*swap_mapping
;
1638 entry
= folio_swap_entry(old
);
1639 swap_index
= swp_offset(entry
);
1640 swap_mapping
= swap_address_space(entry
);
1643 * We have arrived here because our zones are constrained, so don't
1644 * limit chance of success by further cpuset and node constraints.
1646 gfp
&= ~GFP_CONSTRAINT_MASK
;
1647 VM_BUG_ON_FOLIO(folio_test_large(old
), old
);
1648 new = shmem_alloc_folio(gfp
, info
, index
);
1653 folio_copy(new, old
);
1654 flush_dcache_folio(new);
1656 __folio_set_locked(new);
1657 __folio_set_swapbacked(new);
1658 folio_mark_uptodate(new);
1659 folio_set_swap_entry(new, entry
);
1660 folio_set_swapcache(new);
1663 * Our caller will very soon move newpage out of swapcache, but it's
1664 * a nice clean interface for us to replace oldpage by newpage there.
1666 xa_lock_irq(&swap_mapping
->i_pages
);
1667 error
= shmem_replace_entry(swap_mapping
, swap_index
, old
, new);
1669 mem_cgroup_migrate(old
, new);
1670 __lruvec_stat_mod_folio(new, NR_FILE_PAGES
, 1);
1671 __lruvec_stat_mod_folio(new, NR_SHMEM
, 1);
1672 __lruvec_stat_mod_folio(old
, NR_FILE_PAGES
, -1);
1673 __lruvec_stat_mod_folio(old
, NR_SHMEM
, -1);
1675 xa_unlock_irq(&swap_mapping
->i_pages
);
1677 if (unlikely(error
)) {
1679 * Is this possible? I think not, now that our callers check
1680 * both PageSwapCache and page_private after getting page lock;
1681 * but be defensive. Reverse old to newpage for clear and free.
1689 folio_clear_swapcache(old
);
1690 old
->private = NULL
;
1693 folio_put_refs(old
, 2);
1697 static void shmem_set_folio_swapin_error(struct inode
*inode
, pgoff_t index
,
1698 struct folio
*folio
, swp_entry_t swap
)
1700 struct address_space
*mapping
= inode
->i_mapping
;
1701 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1702 swp_entry_t swapin_error
;
1705 swapin_error
= make_swapin_error_entry();
1706 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
,
1707 swp_to_radix_entry(swap
),
1708 swp_to_radix_entry(swapin_error
), 0);
1709 if (old
!= swp_to_radix_entry(swap
))
1712 folio_wait_writeback(folio
);
1713 delete_from_swap_cache(folio
);
1714 spin_lock_irq(&info
->lock
);
1716 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks won't
1717 * be 0 when inode is released and thus trigger WARN_ON(inode->i_blocks) in
1718 * shmem_evict_inode.
1722 shmem_recalc_inode(inode
);
1723 spin_unlock_irq(&info
->lock
);
1728 * Swap in the folio pointed to by *foliop.
1729 * Caller has to make sure that *foliop contains a valid swapped folio.
1730 * Returns 0 and the folio in foliop if success. On failure, returns the
1731 * error code and NULL in *foliop.
1733 static int shmem_swapin_folio(struct inode
*inode
, pgoff_t index
,
1734 struct folio
**foliop
, enum sgp_type sgp
,
1735 gfp_t gfp
, struct vm_area_struct
*vma
,
1736 vm_fault_t
*fault_type
)
1738 struct address_space
*mapping
= inode
->i_mapping
;
1739 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1740 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: NULL
;
1741 struct swap_info_struct
*si
;
1742 struct folio
*folio
= NULL
;
1746 VM_BUG_ON(!*foliop
|| !xa_is_value(*foliop
));
1747 swap
= radix_to_swp_entry(*foliop
);
1750 if (is_swapin_error_entry(swap
))
1753 si
= get_swap_device(swap
);
1755 if (!shmem_confirm_swap(mapping
, index
, swap
))
1761 /* Look it up and read it in.. */
1762 folio
= swap_cache_get_folio(swap
, NULL
, 0);
1764 /* Or update major stats only when swapin succeeds?? */
1766 *fault_type
|= VM_FAULT_MAJOR
;
1767 count_vm_event(PGMAJFAULT
);
1768 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1770 /* Here we actually start the io */
1771 folio
= shmem_swapin(swap
, gfp
, info
, index
);
1778 /* We have to do this with folio locked to prevent races */
1780 if (!folio_test_swapcache(folio
) ||
1781 folio_swap_entry(folio
).val
!= swap
.val
||
1782 !shmem_confirm_swap(mapping
, index
, swap
)) {
1786 if (!folio_test_uptodate(folio
)) {
1790 folio_wait_writeback(folio
);
1793 * Some architectures may have to restore extra metadata to the
1794 * folio after reading from swap.
1796 arch_swap_restore(swap
, folio
);
1798 if (shmem_should_replace_folio(folio
, gfp
)) {
1799 error
= shmem_replace_folio(&folio
, gfp
, info
, index
);
1804 error
= shmem_add_to_page_cache(folio
, mapping
, index
,
1805 swp_to_radix_entry(swap
), gfp
,
1810 spin_lock_irq(&info
->lock
);
1812 shmem_recalc_inode(inode
);
1813 spin_unlock_irq(&info
->lock
);
1815 if (sgp
== SGP_WRITE
)
1816 folio_mark_accessed(folio
);
1818 delete_from_swap_cache(folio
);
1819 folio_mark_dirty(folio
);
1821 put_swap_device(si
);
1826 if (!shmem_confirm_swap(mapping
, index
, swap
))
1829 shmem_set_folio_swapin_error(inode
, index
, folio
, swap
);
1832 folio_unlock(folio
);
1835 put_swap_device(si
);
1841 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1843 * If we allocate a new one we do not mark it dirty. That's up to the
1844 * vm. If we swap it in we mark it dirty since we also free the swap
1845 * entry since a page cannot live in both the swap and page cache.
1847 * vma, vmf, and fault_type are only supplied by shmem_fault:
1848 * otherwise they are NULL.
1850 static int shmem_get_folio_gfp(struct inode
*inode
, pgoff_t index
,
1851 struct folio
**foliop
, enum sgp_type sgp
, gfp_t gfp
,
1852 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1853 vm_fault_t
*fault_type
)
1855 struct address_space
*mapping
= inode
->i_mapping
;
1856 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1857 struct shmem_sb_info
*sbinfo
;
1858 struct mm_struct
*charge_mm
;
1859 struct folio
*folio
;
1866 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1869 if (sgp
<= SGP_CACHE
&&
1870 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1874 sbinfo
= SHMEM_SB(inode
->i_sb
);
1875 charge_mm
= vma
? vma
->vm_mm
: NULL
;
1877 folio
= __filemap_get_folio(mapping
, index
, FGP_ENTRY
| FGP_LOCK
, 0);
1878 if (folio
&& vma
&& userfaultfd_minor(vma
)) {
1879 if (!xa_is_value(folio
)) {
1880 folio_unlock(folio
);
1883 *fault_type
= handle_userfault(vmf
, VM_UFFD_MINOR
);
1887 if (xa_is_value(folio
)) {
1888 error
= shmem_swapin_folio(inode
, index
, &folio
,
1889 sgp
, gfp
, vma
, fault_type
);
1890 if (error
== -EEXIST
)
1898 if (sgp
== SGP_WRITE
)
1899 folio_mark_accessed(folio
);
1900 if (folio_test_uptodate(folio
))
1902 /* fallocated folio */
1903 if (sgp
!= SGP_READ
)
1905 folio_unlock(folio
);
1910 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1911 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1914 if (sgp
== SGP_READ
)
1916 if (sgp
== SGP_NOALLOC
)
1920 * Fast cache lookup and swap lookup did not find it: allocate.
1923 if (vma
&& userfaultfd_missing(vma
)) {
1924 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1928 if (!shmem_is_huge(inode
, index
, false,
1929 vma
? vma
->vm_mm
: NULL
, vma
? vma
->vm_flags
: 0))
1932 huge_gfp
= vma_thp_gfp_mask(vma
);
1933 huge_gfp
= limit_gfp_mask(huge_gfp
, gfp
);
1934 folio
= shmem_alloc_and_acct_folio(huge_gfp
, inode
, index
, true);
1935 if (IS_ERR(folio
)) {
1937 folio
= shmem_alloc_and_acct_folio(gfp
, inode
, index
, false);
1939 if (IS_ERR(folio
)) {
1942 error
= PTR_ERR(folio
);
1944 if (error
!= -ENOSPC
)
1947 * Try to reclaim some space by splitting a large folio
1948 * beyond i_size on the filesystem.
1953 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1954 if (ret
== SHRINK_STOP
)
1962 hindex
= round_down(index
, folio_nr_pages(folio
));
1964 if (sgp
== SGP_WRITE
)
1965 __folio_set_referenced(folio
);
1967 error
= shmem_add_to_page_cache(folio
, mapping
, hindex
,
1968 NULL
, gfp
& GFP_RECLAIM_MASK
,
1972 folio_add_lru(folio
);
1974 spin_lock_irq(&info
->lock
);
1975 info
->alloced
+= folio_nr_pages(folio
);
1976 inode
->i_blocks
+= (blkcnt_t
)BLOCKS_PER_PAGE
<< folio_order(folio
);
1977 shmem_recalc_inode(inode
);
1978 spin_unlock_irq(&info
->lock
);
1981 if (folio_test_pmd_mappable(folio
) &&
1982 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1983 folio_next_index(folio
) - 1) {
1985 * Part of the large folio is beyond i_size: subject
1986 * to shrink under memory pressure.
1988 spin_lock(&sbinfo
->shrinklist_lock
);
1990 * _careful to defend against unlocked access to
1991 * ->shrink_list in shmem_unused_huge_shrink()
1993 if (list_empty_careful(&info
->shrinklist
)) {
1994 list_add_tail(&info
->shrinklist
,
1995 &sbinfo
->shrinklist
);
1996 sbinfo
->shrinklist_len
++;
1998 spin_unlock(&sbinfo
->shrinklist_lock
);
2002 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2004 if (sgp
== SGP_FALLOC
)
2008 * Let SGP_WRITE caller clear ends if write does not fill folio;
2009 * but SGP_FALLOC on a folio fallocated earlier must initialize
2010 * it now, lest undo on failure cancel our earlier guarantee.
2012 if (sgp
!= SGP_WRITE
&& !folio_test_uptodate(folio
)) {
2013 long i
, n
= folio_nr_pages(folio
);
2015 for (i
= 0; i
< n
; i
++)
2016 clear_highpage(folio_page(folio
, i
));
2017 flush_dcache_folio(folio
);
2018 folio_mark_uptodate(folio
);
2021 /* Perhaps the file has been truncated since we checked */
2022 if (sgp
<= SGP_CACHE
&&
2023 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
2025 folio_clear_dirty(folio
);
2026 filemap_remove_folio(folio
);
2027 spin_lock_irq(&info
->lock
);
2028 shmem_recalc_inode(inode
);
2029 spin_unlock_irq(&info
->lock
);
2042 shmem_inode_unacct_blocks(inode
, folio_nr_pages(folio
));
2044 if (folio_test_large(folio
)) {
2045 folio_unlock(folio
);
2051 folio_unlock(folio
);
2054 if (error
== -ENOSPC
&& !once
++) {
2055 spin_lock_irq(&info
->lock
);
2056 shmem_recalc_inode(inode
);
2057 spin_unlock_irq(&info
->lock
);
2060 if (error
== -EEXIST
)
2065 int shmem_get_folio(struct inode
*inode
, pgoff_t index
, struct folio
**foliop
,
2068 return shmem_get_folio_gfp(inode
, index
, foliop
, sgp
,
2069 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
2073 * This is like autoremove_wake_function, but it removes the wait queue
2074 * entry unconditionally - even if something else had already woken the
2077 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2079 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2080 list_del_init(&wait
->entry
);
2084 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2086 struct vm_area_struct
*vma
= vmf
->vma
;
2087 struct inode
*inode
= file_inode(vma
->vm_file
);
2088 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2089 struct folio
*folio
= NULL
;
2091 vm_fault_t ret
= VM_FAULT_LOCKED
;
2094 * Trinity finds that probing a hole which tmpfs is punching can
2095 * prevent the hole-punch from ever completing: which in turn
2096 * locks writers out with its hold on i_rwsem. So refrain from
2097 * faulting pages into the hole while it's being punched. Although
2098 * shmem_undo_range() does remove the additions, it may be unable to
2099 * keep up, as each new page needs its own unmap_mapping_range() call,
2100 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2102 * It does not matter if we sometimes reach this check just before the
2103 * hole-punch begins, so that one fault then races with the punch:
2104 * we just need to make racing faults a rare case.
2106 * The implementation below would be much simpler if we just used a
2107 * standard mutex or completion: but we cannot take i_rwsem in fault,
2108 * and bloating every shmem inode for this unlikely case would be sad.
2110 if (unlikely(inode
->i_private
)) {
2111 struct shmem_falloc
*shmem_falloc
;
2113 spin_lock(&inode
->i_lock
);
2114 shmem_falloc
= inode
->i_private
;
2116 shmem_falloc
->waitq
&&
2117 vmf
->pgoff
>= shmem_falloc
->start
&&
2118 vmf
->pgoff
< shmem_falloc
->next
) {
2120 wait_queue_head_t
*shmem_falloc_waitq
;
2121 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2123 ret
= VM_FAULT_NOPAGE
;
2124 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2126 ret
= VM_FAULT_RETRY
;
2128 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2129 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2130 TASK_UNINTERRUPTIBLE
);
2131 spin_unlock(&inode
->i_lock
);
2135 * shmem_falloc_waitq points into the shmem_fallocate()
2136 * stack of the hole-punching task: shmem_falloc_waitq
2137 * is usually invalid by the time we reach here, but
2138 * finish_wait() does not dereference it in that case;
2139 * though i_lock needed lest racing with wake_up_all().
2141 spin_lock(&inode
->i_lock
);
2142 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2143 spin_unlock(&inode
->i_lock
);
2149 spin_unlock(&inode
->i_lock
);
2152 err
= shmem_get_folio_gfp(inode
, vmf
->pgoff
, &folio
, SGP_CACHE
,
2153 gfp
, vma
, vmf
, &ret
);
2155 return vmf_error(err
);
2157 vmf
->page
= folio_file_page(folio
, vmf
->pgoff
);
2161 unsigned long shmem_get_unmapped_area(struct file
*file
,
2162 unsigned long uaddr
, unsigned long len
,
2163 unsigned long pgoff
, unsigned long flags
)
2165 unsigned long (*get_area
)(struct file
*,
2166 unsigned long, unsigned long, unsigned long, unsigned long);
2168 unsigned long offset
;
2169 unsigned long inflated_len
;
2170 unsigned long inflated_addr
;
2171 unsigned long inflated_offset
;
2173 if (len
> TASK_SIZE
)
2176 get_area
= current
->mm
->get_unmapped_area
;
2177 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2179 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2181 if (IS_ERR_VALUE(addr
))
2183 if (addr
& ~PAGE_MASK
)
2185 if (addr
> TASK_SIZE
- len
)
2188 if (shmem_huge
== SHMEM_HUGE_DENY
)
2190 if (len
< HPAGE_PMD_SIZE
)
2192 if (flags
& MAP_FIXED
)
2195 * Our priority is to support MAP_SHARED mapped hugely;
2196 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2197 * But if caller specified an address hint and we allocated area there
2198 * successfully, respect that as before.
2203 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2204 struct super_block
*sb
;
2207 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2208 sb
= file_inode(file
)->i_sb
;
2211 * Called directly from mm/mmap.c, or drivers/char/mem.c
2212 * for "/dev/zero", to create a shared anonymous object.
2214 if (IS_ERR(shm_mnt
))
2216 sb
= shm_mnt
->mnt_sb
;
2218 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2222 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2223 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2225 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2228 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2229 if (inflated_len
> TASK_SIZE
)
2231 if (inflated_len
< len
)
2234 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2235 if (IS_ERR_VALUE(inflated_addr
))
2237 if (inflated_addr
& ~PAGE_MASK
)
2240 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2241 inflated_addr
+= offset
- inflated_offset
;
2242 if (inflated_offset
> offset
)
2243 inflated_addr
+= HPAGE_PMD_SIZE
;
2245 if (inflated_addr
> TASK_SIZE
- len
)
2247 return inflated_addr
;
2251 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2253 struct inode
*inode
= file_inode(vma
->vm_file
);
2254 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2257 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2260 struct inode
*inode
= file_inode(vma
->vm_file
);
2263 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2264 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2268 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
2270 struct inode
*inode
= file_inode(file
);
2271 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2272 int retval
= -ENOMEM
;
2275 * What serializes the accesses to info->flags?
2276 * ipc_lock_object() when called from shmctl_do_lock(),
2277 * no serialization needed when called from shm_destroy().
2279 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2280 if (!user_shm_lock(inode
->i_size
, ucounts
))
2282 info
->flags
|= VM_LOCKED
;
2283 mapping_set_unevictable(file
->f_mapping
);
2285 if (!lock
&& (info
->flags
& VM_LOCKED
) && ucounts
) {
2286 user_shm_unlock(inode
->i_size
, ucounts
);
2287 info
->flags
&= ~VM_LOCKED
;
2288 mapping_clear_unevictable(file
->f_mapping
);
2296 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2298 struct inode
*inode
= file_inode(file
);
2299 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2302 ret
= seal_check_future_write(info
->seals
, vma
);
2306 /* arm64 - allow memory tagging on RAM-based files */
2307 vm_flags_set(vma
, VM_MTE_ALLOWED
);
2309 file_accessed(file
);
2310 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2312 vma
->vm_ops
= &shmem_vm_ops
;
2314 vma
->vm_ops
= &shmem_anon_vm_ops
;
2318 #ifdef CONFIG_TMPFS_XATTR
2319 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2322 * chattr's fsflags are unrelated to extended attributes,
2323 * but tmpfs has chosen to enable them under the same config option.
2325 static void shmem_set_inode_flags(struct inode
*inode
, unsigned int fsflags
)
2327 unsigned int i_flags
= 0;
2329 if (fsflags
& FS_NOATIME_FL
)
2330 i_flags
|= S_NOATIME
;
2331 if (fsflags
& FS_APPEND_FL
)
2332 i_flags
|= S_APPEND
;
2333 if (fsflags
& FS_IMMUTABLE_FL
)
2334 i_flags
|= S_IMMUTABLE
;
2336 * But FS_NODUMP_FL does not require any action in i_flags.
2338 inode_set_flags(inode
, i_flags
, S_NOATIME
| S_APPEND
| S_IMMUTABLE
);
2341 static void shmem_set_inode_flags(struct inode
*inode
, unsigned int fsflags
)
2344 #define shmem_initxattrs NULL
2347 static struct inode
*shmem_get_inode(struct mnt_idmap
*idmap
, struct super_block
*sb
,
2348 struct inode
*dir
, umode_t mode
, dev_t dev
,
2349 unsigned long flags
)
2351 struct inode
*inode
;
2352 struct shmem_inode_info
*info
;
2353 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2356 if (shmem_reserve_inode(sb
, &ino
))
2359 inode
= new_inode(sb
);
2362 inode_init_owner(idmap
, inode
, dir
, mode
);
2363 inode
->i_blocks
= 0;
2364 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2365 inode
->i_generation
= get_random_u32();
2366 info
= SHMEM_I(inode
);
2367 memset(info
, 0, (char *)inode
- (char *)info
);
2368 spin_lock_init(&info
->lock
);
2369 atomic_set(&info
->stop_eviction
, 0);
2370 info
->seals
= F_SEAL_SEAL
;
2371 info
->flags
= flags
& VM_NORESERVE
;
2372 info
->i_crtime
= inode
->i_mtime
;
2373 info
->fsflags
= (dir
== NULL
) ? 0 :
2374 SHMEM_I(dir
)->fsflags
& SHMEM_FL_INHERITED
;
2376 shmem_set_inode_flags(inode
, info
->fsflags
);
2377 INIT_LIST_HEAD(&info
->shrinklist
);
2378 INIT_LIST_HEAD(&info
->swaplist
);
2379 simple_xattrs_init(&info
->xattrs
);
2380 cache_no_acl(inode
);
2381 mapping_set_large_folios(inode
->i_mapping
);
2383 switch (mode
& S_IFMT
) {
2385 inode
->i_op
= &shmem_special_inode_operations
;
2386 init_special_inode(inode
, mode
, dev
);
2389 inode
->i_mapping
->a_ops
= &shmem_aops
;
2390 inode
->i_op
= &shmem_inode_operations
;
2391 inode
->i_fop
= &shmem_file_operations
;
2392 mpol_shared_policy_init(&info
->policy
,
2393 shmem_get_sbmpol(sbinfo
));
2397 /* Some things misbehave if size == 0 on a directory */
2398 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2399 inode
->i_op
= &shmem_dir_inode_operations
;
2400 inode
->i_fop
= &simple_dir_operations
;
2404 * Must not load anything in the rbtree,
2405 * mpol_free_shared_policy will not be called.
2407 mpol_shared_policy_init(&info
->policy
, NULL
);
2411 lockdep_annotate_inode_mutex_key(inode
);
2413 shmem_free_inode(sb
);
2417 #ifdef CONFIG_USERFAULTFD
2418 int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2420 struct vm_area_struct
*dst_vma
,
2421 unsigned long dst_addr
,
2422 unsigned long src_addr
,
2423 bool zeropage
, bool wp_copy
,
2424 struct page
**pagep
)
2426 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2427 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2428 struct address_space
*mapping
= inode
->i_mapping
;
2429 gfp_t gfp
= mapping_gfp_mask(mapping
);
2430 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2432 struct folio
*folio
;
2436 if (!shmem_inode_acct_block(inode
, 1)) {
2438 * We may have got a page, returned -ENOENT triggering a retry,
2439 * and now we find ourselves with -ENOMEM. Release the page, to
2440 * avoid a BUG_ON in our caller.
2442 if (unlikely(*pagep
)) {
2451 folio
= shmem_alloc_folio(gfp
, info
, pgoff
);
2453 goto out_unacct_blocks
;
2455 if (!zeropage
) { /* COPY */
2456 page_kaddr
= kmap_local_folio(folio
, 0);
2458 * The read mmap_lock is held here. Despite the
2459 * mmap_lock being read recursive a deadlock is still
2460 * possible if a writer has taken a lock. For example:
2462 * process A thread 1 takes read lock on own mmap_lock
2463 * process A thread 2 calls mmap, blocks taking write lock
2464 * process B thread 1 takes page fault, read lock on own mmap lock
2465 * process B thread 2 calls mmap, blocks taking write lock
2466 * process A thread 1 blocks taking read lock on process B
2467 * process B thread 1 blocks taking read lock on process A
2469 * Disable page faults to prevent potential deadlock
2470 * and retry the copy outside the mmap_lock.
2472 pagefault_disable();
2473 ret
= copy_from_user(page_kaddr
,
2474 (const void __user
*)src_addr
,
2477 kunmap_local(page_kaddr
);
2479 /* fallback to copy_from_user outside mmap_lock */
2480 if (unlikely(ret
)) {
2481 *pagep
= &folio
->page
;
2483 /* don't free the page */
2484 goto out_unacct_blocks
;
2487 flush_dcache_folio(folio
);
2488 } else { /* ZEROPAGE */
2489 clear_user_highpage(&folio
->page
, dst_addr
);
2492 folio
= page_folio(*pagep
);
2493 VM_BUG_ON_FOLIO(folio_test_large(folio
), folio
);
2497 VM_BUG_ON(folio_test_locked(folio
));
2498 VM_BUG_ON(folio_test_swapbacked(folio
));
2499 __folio_set_locked(folio
);
2500 __folio_set_swapbacked(folio
);
2501 __folio_mark_uptodate(folio
);
2504 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2505 if (unlikely(pgoff
>= max_off
))
2508 ret
= shmem_add_to_page_cache(folio
, mapping
, pgoff
, NULL
,
2509 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2513 ret
= mfill_atomic_install_pte(dst_mm
, dst_pmd
, dst_vma
, dst_addr
,
2514 &folio
->page
, true, wp_copy
);
2516 goto out_delete_from_cache
;
2518 spin_lock_irq(&info
->lock
);
2520 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2521 shmem_recalc_inode(inode
);
2522 spin_unlock_irq(&info
->lock
);
2524 folio_unlock(folio
);
2526 out_delete_from_cache
:
2527 filemap_remove_folio(folio
);
2529 folio_unlock(folio
);
2532 shmem_inode_unacct_blocks(inode
, 1);
2535 #endif /* CONFIG_USERFAULTFD */
2538 static const struct inode_operations shmem_symlink_inode_operations
;
2539 static const struct inode_operations shmem_short_symlink_operations
;
2542 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2543 loff_t pos
, unsigned len
,
2544 struct page
**pagep
, void **fsdata
)
2546 struct inode
*inode
= mapping
->host
;
2547 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2548 pgoff_t index
= pos
>> PAGE_SHIFT
;
2549 struct folio
*folio
;
2552 /* i_rwsem is held by caller */
2553 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2554 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2555 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2557 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2561 ret
= shmem_get_folio(inode
, index
, &folio
, SGP_WRITE
);
2566 *pagep
= folio_file_page(folio
, index
);
2567 if (PageHWPoison(*pagep
)) {
2568 folio_unlock(folio
);
2578 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2579 loff_t pos
, unsigned len
, unsigned copied
,
2580 struct page
*page
, void *fsdata
)
2582 struct folio
*folio
= page_folio(page
);
2583 struct inode
*inode
= mapping
->host
;
2585 if (pos
+ copied
> inode
->i_size
)
2586 i_size_write(inode
, pos
+ copied
);
2588 if (!folio_test_uptodate(folio
)) {
2589 if (copied
< folio_size(folio
)) {
2590 size_t from
= offset_in_folio(folio
, pos
);
2591 folio_zero_segments(folio
, 0, from
,
2592 from
+ copied
, folio_size(folio
));
2594 folio_mark_uptodate(folio
);
2596 folio_mark_dirty(folio
);
2597 folio_unlock(folio
);
2603 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2605 struct file
*file
= iocb
->ki_filp
;
2606 struct inode
*inode
= file_inode(file
);
2607 struct address_space
*mapping
= inode
->i_mapping
;
2609 unsigned long offset
;
2612 loff_t
*ppos
= &iocb
->ki_pos
;
2614 index
= *ppos
>> PAGE_SHIFT
;
2615 offset
= *ppos
& ~PAGE_MASK
;
2618 struct folio
*folio
= NULL
;
2619 struct page
*page
= NULL
;
2621 unsigned long nr
, ret
;
2622 loff_t i_size
= i_size_read(inode
);
2624 end_index
= i_size
>> PAGE_SHIFT
;
2625 if (index
> end_index
)
2627 if (index
== end_index
) {
2628 nr
= i_size
& ~PAGE_MASK
;
2633 error
= shmem_get_folio(inode
, index
, &folio
, SGP_READ
);
2635 if (error
== -EINVAL
)
2640 folio_unlock(folio
);
2642 page
= folio_file_page(folio
, index
);
2643 if (PageHWPoison(page
)) {
2651 * We must evaluate after, since reads (unlike writes)
2652 * are called without i_rwsem protection against truncate
2655 i_size
= i_size_read(inode
);
2656 end_index
= i_size
>> PAGE_SHIFT
;
2657 if (index
== end_index
) {
2658 nr
= i_size
& ~PAGE_MASK
;
2669 * If users can be writing to this page using arbitrary
2670 * virtual addresses, take care about potential aliasing
2671 * before reading the page on the kernel side.
2673 if (mapping_writably_mapped(mapping
))
2674 flush_dcache_page(page
);
2676 * Mark the page accessed if we read the beginning.
2679 folio_mark_accessed(folio
);
2681 * Ok, we have the page, and it's up-to-date, so
2682 * now we can copy it to user space...
2684 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2687 } else if (user_backed_iter(to
)) {
2689 * Copy to user tends to be so well optimized, but
2690 * clear_user() not so much, that it is noticeably
2691 * faster to copy the zero page instead of clearing.
2693 ret
= copy_page_to_iter(ZERO_PAGE(0), offset
, nr
, to
);
2696 * But submitting the same page twice in a row to
2697 * splice() - or others? - can result in confusion:
2698 * so don't attempt that optimization on pipes etc.
2700 ret
= iov_iter_zero(nr
, to
);
2705 index
+= offset
>> PAGE_SHIFT
;
2706 offset
&= ~PAGE_MASK
;
2708 if (!iov_iter_count(to
))
2717 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2718 file_accessed(file
);
2719 return retval
? retval
: error
;
2722 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2724 struct address_space
*mapping
= file
->f_mapping
;
2725 struct inode
*inode
= mapping
->host
;
2727 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2728 return generic_file_llseek_size(file
, offset
, whence
,
2729 MAX_LFS_FILESIZE
, i_size_read(inode
));
2734 /* We're holding i_rwsem so we can access i_size directly */
2735 offset
= mapping_seek_hole_data(mapping
, offset
, inode
->i_size
, whence
);
2737 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2738 inode_unlock(inode
);
2742 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2745 struct inode
*inode
= file_inode(file
);
2746 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2747 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2748 struct shmem_falloc shmem_falloc
;
2749 pgoff_t start
, index
, end
, undo_fallocend
;
2752 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2757 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2758 struct address_space
*mapping
= file
->f_mapping
;
2759 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2760 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2761 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2763 /* protected by i_rwsem */
2764 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2769 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2770 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2771 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2772 spin_lock(&inode
->i_lock
);
2773 inode
->i_private
= &shmem_falloc
;
2774 spin_unlock(&inode
->i_lock
);
2776 if ((u64
)unmap_end
> (u64
)unmap_start
)
2777 unmap_mapping_range(mapping
, unmap_start
,
2778 1 + unmap_end
- unmap_start
, 0);
2779 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2780 /* No need to unmap again: hole-punching leaves COWed pages */
2782 spin_lock(&inode
->i_lock
);
2783 inode
->i_private
= NULL
;
2784 wake_up_all(&shmem_falloc_waitq
);
2785 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2786 spin_unlock(&inode
->i_lock
);
2791 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2792 error
= inode_newsize_ok(inode
, offset
+ len
);
2796 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2801 start
= offset
>> PAGE_SHIFT
;
2802 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2803 /* Try to avoid a swapstorm if len is impossible to satisfy */
2804 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2809 shmem_falloc
.waitq
= NULL
;
2810 shmem_falloc
.start
= start
;
2811 shmem_falloc
.next
= start
;
2812 shmem_falloc
.nr_falloced
= 0;
2813 shmem_falloc
.nr_unswapped
= 0;
2814 spin_lock(&inode
->i_lock
);
2815 inode
->i_private
= &shmem_falloc
;
2816 spin_unlock(&inode
->i_lock
);
2819 * info->fallocend is only relevant when huge pages might be
2820 * involved: to prevent split_huge_page() freeing fallocated
2821 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2823 undo_fallocend
= info
->fallocend
;
2824 if (info
->fallocend
< end
)
2825 info
->fallocend
= end
;
2827 for (index
= start
; index
< end
; ) {
2828 struct folio
*folio
;
2831 * Good, the fallocate(2) manpage permits EINTR: we may have
2832 * been interrupted because we are using up too much memory.
2834 if (signal_pending(current
))
2836 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2839 error
= shmem_get_folio(inode
, index
, &folio
,
2842 info
->fallocend
= undo_fallocend
;
2843 /* Remove the !uptodate folios we added */
2844 if (index
> start
) {
2845 shmem_undo_range(inode
,
2846 (loff_t
)start
<< PAGE_SHIFT
,
2847 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2853 * Here is a more important optimization than it appears:
2854 * a second SGP_FALLOC on the same large folio will clear it,
2855 * making it uptodate and un-undoable if we fail later.
2857 index
= folio_next_index(folio
);
2858 /* Beware 32-bit wraparound */
2863 * Inform shmem_writepage() how far we have reached.
2864 * No need for lock or barrier: we have the page lock.
2866 if (!folio_test_uptodate(folio
))
2867 shmem_falloc
.nr_falloced
+= index
- shmem_falloc
.next
;
2868 shmem_falloc
.next
= index
;
2871 * If !uptodate, leave it that way so that freeable folios
2872 * can be recognized if we need to rollback on error later.
2873 * But mark it dirty so that memory pressure will swap rather
2874 * than free the folios we are allocating (and SGP_CACHE folios
2875 * might still be clean: we now need to mark those dirty too).
2877 folio_mark_dirty(folio
);
2878 folio_unlock(folio
);
2883 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2884 i_size_write(inode
, offset
+ len
);
2886 spin_lock(&inode
->i_lock
);
2887 inode
->i_private
= NULL
;
2888 spin_unlock(&inode
->i_lock
);
2891 file_modified(file
);
2892 inode_unlock(inode
);
2896 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2898 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2900 buf
->f_type
= TMPFS_MAGIC
;
2901 buf
->f_bsize
= PAGE_SIZE
;
2902 buf
->f_namelen
= NAME_MAX
;
2903 if (sbinfo
->max_blocks
) {
2904 buf
->f_blocks
= sbinfo
->max_blocks
;
2906 buf
->f_bfree
= sbinfo
->max_blocks
-
2907 percpu_counter_sum(&sbinfo
->used_blocks
);
2909 if (sbinfo
->max_inodes
) {
2910 buf
->f_files
= sbinfo
->max_inodes
;
2911 buf
->f_ffree
= sbinfo
->free_inodes
;
2913 /* else leave those fields 0 like simple_statfs */
2915 buf
->f_fsid
= uuid_to_fsid(dentry
->d_sb
->s_uuid
.b
);
2921 * File creation. Allocate an inode, and we're done..
2924 shmem_mknod(struct mnt_idmap
*idmap
, struct inode
*dir
,
2925 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2927 struct inode
*inode
;
2928 int error
= -ENOSPC
;
2930 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2932 error
= simple_acl_create(dir
, inode
);
2935 error
= security_inode_init_security(inode
, dir
,
2937 shmem_initxattrs
, NULL
);
2938 if (error
&& error
!= -EOPNOTSUPP
)
2942 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2943 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2944 inode_inc_iversion(dir
);
2945 d_instantiate(dentry
, inode
);
2946 dget(dentry
); /* Extra count - pin the dentry in core */
2955 shmem_tmpfile(struct mnt_idmap
*idmap
, struct inode
*dir
,
2956 struct file
*file
, umode_t mode
)
2958 struct inode
*inode
;
2959 int error
= -ENOSPC
;
2961 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2963 error
= security_inode_init_security(inode
, dir
,
2965 shmem_initxattrs
, NULL
);
2966 if (error
&& error
!= -EOPNOTSUPP
)
2968 error
= simple_acl_create(dir
, inode
);
2971 d_tmpfile(file
, inode
);
2973 return finish_open_simple(file
, error
);
2979 static int shmem_mkdir(struct mnt_idmap
*idmap
, struct inode
*dir
,
2980 struct dentry
*dentry
, umode_t mode
)
2984 error
= shmem_mknod(idmap
, dir
, dentry
, mode
| S_IFDIR
, 0);
2991 static int shmem_create(struct mnt_idmap
*idmap
, struct inode
*dir
,
2992 struct dentry
*dentry
, umode_t mode
, bool excl
)
2994 return shmem_mknod(idmap
, dir
, dentry
, mode
| S_IFREG
, 0);
3000 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
3002 struct inode
*inode
= d_inode(old_dentry
);
3006 * No ordinary (disk based) filesystem counts links as inodes;
3007 * but each new link needs a new dentry, pinning lowmem, and
3008 * tmpfs dentries cannot be pruned until they are unlinked.
3009 * But if an O_TMPFILE file is linked into the tmpfs, the
3010 * first link must skip that, to get the accounting right.
3012 if (inode
->i_nlink
) {
3013 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
3018 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3019 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3020 inode_inc_iversion(dir
);
3022 ihold(inode
); /* New dentry reference */
3023 dget(dentry
); /* Extra pinning count for the created dentry */
3024 d_instantiate(dentry
, inode
);
3029 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3031 struct inode
*inode
= d_inode(dentry
);
3033 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3034 shmem_free_inode(inode
->i_sb
);
3036 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3037 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3038 inode_inc_iversion(dir
);
3040 dput(dentry
); /* Undo the count from "create" - this does all the work */
3044 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3046 if (!simple_empty(dentry
))
3049 drop_nlink(d_inode(dentry
));
3051 return shmem_unlink(dir
, dentry
);
3054 static int shmem_whiteout(struct mnt_idmap
*idmap
,
3055 struct inode
*old_dir
, struct dentry
*old_dentry
)
3057 struct dentry
*whiteout
;
3060 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3064 error
= shmem_mknod(idmap
, old_dir
, whiteout
,
3065 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3071 * Cheat and hash the whiteout while the old dentry is still in
3072 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3074 * d_lookup() will consistently find one of them at this point,
3075 * not sure which one, but that isn't even important.
3082 * The VFS layer already does all the dentry stuff for rename,
3083 * we just have to decrement the usage count for the target if
3084 * it exists so that the VFS layer correctly free's it when it
3087 static int shmem_rename2(struct mnt_idmap
*idmap
,
3088 struct inode
*old_dir
, struct dentry
*old_dentry
,
3089 struct inode
*new_dir
, struct dentry
*new_dentry
,
3092 struct inode
*inode
= d_inode(old_dentry
);
3093 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3095 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3098 if (flags
& RENAME_EXCHANGE
)
3099 return simple_rename_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3101 if (!simple_empty(new_dentry
))
3104 if (flags
& RENAME_WHITEOUT
) {
3107 error
= shmem_whiteout(idmap
, old_dir
, old_dentry
);
3112 if (d_really_is_positive(new_dentry
)) {
3113 (void) shmem_unlink(new_dir
, new_dentry
);
3114 if (they_are_dirs
) {
3115 drop_nlink(d_inode(new_dentry
));
3116 drop_nlink(old_dir
);
3118 } else if (they_are_dirs
) {
3119 drop_nlink(old_dir
);
3123 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3124 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3125 old_dir
->i_ctime
= old_dir
->i_mtime
=
3126 new_dir
->i_ctime
= new_dir
->i_mtime
=
3127 inode
->i_ctime
= current_time(old_dir
);
3128 inode_inc_iversion(old_dir
);
3129 inode_inc_iversion(new_dir
);
3133 static int shmem_symlink(struct mnt_idmap
*idmap
, struct inode
*dir
,
3134 struct dentry
*dentry
, const char *symname
)
3138 struct inode
*inode
;
3139 struct folio
*folio
;
3141 len
= strlen(symname
) + 1;
3142 if (len
> PAGE_SIZE
)
3143 return -ENAMETOOLONG
;
3145 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3150 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3151 shmem_initxattrs
, NULL
);
3152 if (error
&& error
!= -EOPNOTSUPP
) {
3157 inode
->i_size
= len
-1;
3158 if (len
<= SHORT_SYMLINK_LEN
) {
3159 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3160 if (!inode
->i_link
) {
3164 inode
->i_op
= &shmem_short_symlink_operations
;
3166 inode_nohighmem(inode
);
3167 error
= shmem_get_folio(inode
, 0, &folio
, SGP_WRITE
);
3172 inode
->i_mapping
->a_ops
= &shmem_aops
;
3173 inode
->i_op
= &shmem_symlink_inode_operations
;
3174 memcpy(folio_address(folio
), symname
, len
);
3175 folio_mark_uptodate(folio
);
3176 folio_mark_dirty(folio
);
3177 folio_unlock(folio
);
3180 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3181 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3182 inode_inc_iversion(dir
);
3183 d_instantiate(dentry
, inode
);
3188 static void shmem_put_link(void *arg
)
3190 folio_mark_accessed(arg
);
3194 static const char *shmem_get_link(struct dentry
*dentry
,
3195 struct inode
*inode
,
3196 struct delayed_call
*done
)
3198 struct folio
*folio
= NULL
;
3202 folio
= filemap_get_folio(inode
->i_mapping
, 0);
3204 return ERR_PTR(-ECHILD
);
3205 if (PageHWPoison(folio_page(folio
, 0)) ||
3206 !folio_test_uptodate(folio
)) {
3208 return ERR_PTR(-ECHILD
);
3211 error
= shmem_get_folio(inode
, 0, &folio
, SGP_READ
);
3213 return ERR_PTR(error
);
3215 return ERR_PTR(-ECHILD
);
3216 if (PageHWPoison(folio_page(folio
, 0))) {
3217 folio_unlock(folio
);
3219 return ERR_PTR(-ECHILD
);
3221 folio_unlock(folio
);
3223 set_delayed_call(done
, shmem_put_link
, folio
);
3224 return folio_address(folio
);
3227 #ifdef CONFIG_TMPFS_XATTR
3229 static int shmem_fileattr_get(struct dentry
*dentry
, struct fileattr
*fa
)
3231 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3233 fileattr_fill_flags(fa
, info
->fsflags
& SHMEM_FL_USER_VISIBLE
);
3238 static int shmem_fileattr_set(struct mnt_idmap
*idmap
,
3239 struct dentry
*dentry
, struct fileattr
*fa
)
3241 struct inode
*inode
= d_inode(dentry
);
3242 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3244 if (fileattr_has_fsx(fa
))
3246 if (fa
->flags
& ~SHMEM_FL_USER_MODIFIABLE
)
3249 info
->fsflags
= (info
->fsflags
& ~SHMEM_FL_USER_MODIFIABLE
) |
3250 (fa
->flags
& SHMEM_FL_USER_MODIFIABLE
);
3252 shmem_set_inode_flags(inode
, info
->fsflags
);
3253 inode
->i_ctime
= current_time(inode
);
3254 inode_inc_iversion(inode
);
3259 * Superblocks without xattr inode operations may get some security.* xattr
3260 * support from the LSM "for free". As soon as we have any other xattrs
3261 * like ACLs, we also need to implement the security.* handlers at
3262 * filesystem level, though.
3266 * Callback for security_inode_init_security() for acquiring xattrs.
3268 static int shmem_initxattrs(struct inode
*inode
,
3269 const struct xattr
*xattr_array
,
3272 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3273 const struct xattr
*xattr
;
3274 struct simple_xattr
*new_xattr
;
3277 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3278 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3282 len
= strlen(xattr
->name
) + 1;
3283 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3285 if (!new_xattr
->name
) {
3290 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3291 XATTR_SECURITY_PREFIX_LEN
);
3292 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3295 simple_xattr_add(&info
->xattrs
, new_xattr
);
3301 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3302 struct dentry
*unused
, struct inode
*inode
,
3303 const char *name
, void *buffer
, size_t size
)
3305 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3307 name
= xattr_full_name(handler
, name
);
3308 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3311 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3312 struct mnt_idmap
*idmap
,
3313 struct dentry
*unused
, struct inode
*inode
,
3314 const char *name
, const void *value
,
3315 size_t size
, int flags
)
3317 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3320 name
= xattr_full_name(handler
, name
);
3321 err
= simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3323 inode
->i_ctime
= current_time(inode
);
3324 inode_inc_iversion(inode
);
3329 static const struct xattr_handler shmem_security_xattr_handler
= {
3330 .prefix
= XATTR_SECURITY_PREFIX
,
3331 .get
= shmem_xattr_handler_get
,
3332 .set
= shmem_xattr_handler_set
,
3335 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3336 .prefix
= XATTR_TRUSTED_PREFIX
,
3337 .get
= shmem_xattr_handler_get
,
3338 .set
= shmem_xattr_handler_set
,
3341 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3342 #ifdef CONFIG_TMPFS_POSIX_ACL
3343 &posix_acl_access_xattr_handler
,
3344 &posix_acl_default_xattr_handler
,
3346 &shmem_security_xattr_handler
,
3347 &shmem_trusted_xattr_handler
,
3351 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3353 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3354 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3356 #endif /* CONFIG_TMPFS_XATTR */
3358 static const struct inode_operations shmem_short_symlink_operations
= {
3359 .getattr
= shmem_getattr
,
3360 .get_link
= simple_get_link
,
3361 #ifdef CONFIG_TMPFS_XATTR
3362 .listxattr
= shmem_listxattr
,
3366 static const struct inode_operations shmem_symlink_inode_operations
= {
3367 .getattr
= shmem_getattr
,
3368 .get_link
= shmem_get_link
,
3369 #ifdef CONFIG_TMPFS_XATTR
3370 .listxattr
= shmem_listxattr
,
3374 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3376 return ERR_PTR(-ESTALE
);
3379 static int shmem_match(struct inode
*ino
, void *vfh
)
3383 inum
= (inum
<< 32) | fh
[1];
3384 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3387 /* Find any alias of inode, but prefer a hashed alias */
3388 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3390 struct dentry
*alias
= d_find_alias(inode
);
3392 return alias
?: d_find_any_alias(inode
);
3396 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3397 struct fid
*fid
, int fh_len
, int fh_type
)
3399 struct inode
*inode
;
3400 struct dentry
*dentry
= NULL
;
3407 inum
= (inum
<< 32) | fid
->raw
[1];
3409 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3410 shmem_match
, fid
->raw
);
3412 dentry
= shmem_find_alias(inode
);
3419 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3420 struct inode
*parent
)
3424 return FILEID_INVALID
;
3427 if (inode_unhashed(inode
)) {
3428 /* Unfortunately insert_inode_hash is not idempotent,
3429 * so as we hash inodes here rather than at creation
3430 * time, we need a lock to ensure we only try
3433 static DEFINE_SPINLOCK(lock
);
3435 if (inode_unhashed(inode
))
3436 __insert_inode_hash(inode
,
3437 inode
->i_ino
+ inode
->i_generation
);
3441 fh
[0] = inode
->i_generation
;
3442 fh
[1] = inode
->i_ino
;
3443 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3449 static const struct export_operations shmem_export_ops
= {
3450 .get_parent
= shmem_get_parent
,
3451 .encode_fh
= shmem_encode_fh
,
3452 .fh_to_dentry
= shmem_fh_to_dentry
,
3468 static const struct constant_table shmem_param_enums_huge
[] = {
3469 {"never", SHMEM_HUGE_NEVER
},
3470 {"always", SHMEM_HUGE_ALWAYS
},
3471 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3472 {"advise", SHMEM_HUGE_ADVISE
},
3476 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3477 fsparam_u32 ("gid", Opt_gid
),
3478 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3479 fsparam_u32oct("mode", Opt_mode
),
3480 fsparam_string("mpol", Opt_mpol
),
3481 fsparam_string("nr_blocks", Opt_nr_blocks
),
3482 fsparam_string("nr_inodes", Opt_nr_inodes
),
3483 fsparam_string("size", Opt_size
),
3484 fsparam_u32 ("uid", Opt_uid
),
3485 fsparam_flag ("inode32", Opt_inode32
),
3486 fsparam_flag ("inode64", Opt_inode64
),
3490 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3492 struct shmem_options
*ctx
= fc
->fs_private
;
3493 struct fs_parse_result result
;
3494 unsigned long long size
;
3498 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3504 size
= memparse(param
->string
, &rest
);
3506 size
<<= PAGE_SHIFT
;
3507 size
*= totalram_pages();
3513 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3514 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3517 ctx
->blocks
= memparse(param
->string
, &rest
);
3518 if (*rest
|| ctx
->blocks
> S64_MAX
)
3520 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3523 ctx
->inodes
= memparse(param
->string
, &rest
);
3526 ctx
->seen
|= SHMEM_SEEN_INODES
;
3529 ctx
->mode
= result
.uint_32
& 07777;
3532 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3533 if (!uid_valid(ctx
->uid
))
3537 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3538 if (!gid_valid(ctx
->gid
))
3542 ctx
->huge
= result
.uint_32
;
3543 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3544 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3545 has_transparent_hugepage()))
3546 goto unsupported_parameter
;
3547 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3550 if (IS_ENABLED(CONFIG_NUMA
)) {
3551 mpol_put(ctx
->mpol
);
3553 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3557 goto unsupported_parameter
;
3559 ctx
->full_inums
= false;
3560 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3563 if (sizeof(ino_t
) < 8) {
3565 "Cannot use inode64 with <64bit inums in kernel\n");
3567 ctx
->full_inums
= true;
3568 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3573 unsupported_parameter
:
3574 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3576 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3579 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3581 char *options
= data
;
3584 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3589 while (options
!= NULL
) {
3590 char *this_char
= options
;
3593 * NUL-terminate this option: unfortunately,
3594 * mount options form a comma-separated list,
3595 * but mpol's nodelist may also contain commas.
3597 options
= strchr(options
, ',');
3598 if (options
== NULL
)
3601 if (!isdigit(*options
)) {
3607 char *value
= strchr(this_char
, '=');
3613 len
= strlen(value
);
3615 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3624 * Reconfigure a shmem filesystem.
3626 * Note that we disallow change from limited->unlimited blocks/inodes while any
3627 * are in use; but we must separately disallow unlimited->limited, because in
3628 * that case we have no record of how much is already in use.
3630 static int shmem_reconfigure(struct fs_context
*fc
)
3632 struct shmem_options
*ctx
= fc
->fs_private
;
3633 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3634 unsigned long inodes
;
3635 struct mempolicy
*mpol
= NULL
;
3638 raw_spin_lock(&sbinfo
->stat_lock
);
3639 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3641 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3642 if (!sbinfo
->max_blocks
) {
3643 err
= "Cannot retroactively limit size";
3646 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3648 err
= "Too small a size for current use";
3652 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3653 if (!sbinfo
->max_inodes
) {
3654 err
= "Cannot retroactively limit inodes";
3657 if (ctx
->inodes
< inodes
) {
3658 err
= "Too few inodes for current use";
3663 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
3664 sbinfo
->next_ino
> UINT_MAX
) {
3665 err
= "Current inum too high to switch to 32-bit inums";
3669 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3670 sbinfo
->huge
= ctx
->huge
;
3671 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
3672 sbinfo
->full_inums
= ctx
->full_inums
;
3673 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3674 sbinfo
->max_blocks
= ctx
->blocks
;
3675 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3676 sbinfo
->max_inodes
= ctx
->inodes
;
3677 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3681 * Preserve previous mempolicy unless mpol remount option was specified.
3684 mpol
= sbinfo
->mpol
;
3685 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3688 raw_spin_unlock(&sbinfo
->stat_lock
);
3692 raw_spin_unlock(&sbinfo
->stat_lock
);
3693 return invalfc(fc
, "%s", err
);
3696 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3698 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3700 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3701 seq_printf(seq
, ",size=%luk",
3702 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3703 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3704 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3705 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3706 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3707 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3708 seq_printf(seq
, ",uid=%u",
3709 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3710 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3711 seq_printf(seq
, ",gid=%u",
3712 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3715 * Showing inode{64,32} might be useful even if it's the system default,
3716 * since then people don't have to resort to checking both here and
3717 * /proc/config.gz to confirm 64-bit inums were successfully applied
3718 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3720 * We hide it when inode64 isn't the default and we are using 32-bit
3721 * inodes, since that probably just means the feature isn't even under
3726 * +-----------------+-----------------+
3727 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3728 * +------------------+-----------------+-----------------+
3729 * | full_inums=true | show | show |
3730 * | full_inums=false | show | hide |
3731 * +------------------+-----------------+-----------------+
3734 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
3735 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
3736 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3737 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3739 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3741 shmem_show_mpol(seq
, sbinfo
->mpol
);
3745 #endif /* CONFIG_TMPFS */
3747 static void shmem_put_super(struct super_block
*sb
)
3749 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3751 free_percpu(sbinfo
->ino_batch
);
3752 percpu_counter_destroy(&sbinfo
->used_blocks
);
3753 mpol_put(sbinfo
->mpol
);
3755 sb
->s_fs_info
= NULL
;
3758 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3760 struct shmem_options
*ctx
= fc
->fs_private
;
3761 struct inode
*inode
;
3762 struct shmem_sb_info
*sbinfo
;
3764 /* Round up to L1_CACHE_BYTES to resist false sharing */
3765 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3766 L1_CACHE_BYTES
), GFP_KERNEL
);
3770 sb
->s_fs_info
= sbinfo
;
3774 * Per default we only allow half of the physical ram per
3775 * tmpfs instance, limiting inodes to one per page of lowmem;
3776 * but the internal instance is left unlimited.
3778 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3779 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3780 ctx
->blocks
= shmem_default_max_blocks();
3781 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3782 ctx
->inodes
= shmem_default_max_inodes();
3783 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
3784 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
3786 sb
->s_flags
|= SB_NOUSER
;
3788 sb
->s_export_op
= &shmem_export_ops
;
3789 sb
->s_flags
|= SB_NOSEC
| SB_I_VERSION
;
3791 sb
->s_flags
|= SB_NOUSER
;
3793 sbinfo
->max_blocks
= ctx
->blocks
;
3794 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3795 if (sb
->s_flags
& SB_KERNMOUNT
) {
3796 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
3797 if (!sbinfo
->ino_batch
)
3800 sbinfo
->uid
= ctx
->uid
;
3801 sbinfo
->gid
= ctx
->gid
;
3802 sbinfo
->full_inums
= ctx
->full_inums
;
3803 sbinfo
->mode
= ctx
->mode
;
3804 sbinfo
->huge
= ctx
->huge
;
3805 sbinfo
->mpol
= ctx
->mpol
;
3808 raw_spin_lock_init(&sbinfo
->stat_lock
);
3809 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3811 spin_lock_init(&sbinfo
->shrinklist_lock
);
3812 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3814 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3815 sb
->s_blocksize
= PAGE_SIZE
;
3816 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3817 sb
->s_magic
= TMPFS_MAGIC
;
3818 sb
->s_op
= &shmem_ops
;
3819 sb
->s_time_gran
= 1;
3820 #ifdef CONFIG_TMPFS_XATTR
3821 sb
->s_xattr
= shmem_xattr_handlers
;
3823 #ifdef CONFIG_TMPFS_POSIX_ACL
3824 sb
->s_flags
|= SB_POSIXACL
;
3826 uuid_gen(&sb
->s_uuid
);
3828 inode
= shmem_get_inode(&nop_mnt_idmap
, sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0,
3832 inode
->i_uid
= sbinfo
->uid
;
3833 inode
->i_gid
= sbinfo
->gid
;
3834 sb
->s_root
= d_make_root(inode
);
3840 shmem_put_super(sb
);
3844 static int shmem_get_tree(struct fs_context
*fc
)
3846 return get_tree_nodev(fc
, shmem_fill_super
);
3849 static void shmem_free_fc(struct fs_context
*fc
)
3851 struct shmem_options
*ctx
= fc
->fs_private
;
3854 mpol_put(ctx
->mpol
);
3859 static const struct fs_context_operations shmem_fs_context_ops
= {
3860 .free
= shmem_free_fc
,
3861 .get_tree
= shmem_get_tree
,
3863 .parse_monolithic
= shmem_parse_options
,
3864 .parse_param
= shmem_parse_one
,
3865 .reconfigure
= shmem_reconfigure
,
3869 static struct kmem_cache
*shmem_inode_cachep
;
3871 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3873 struct shmem_inode_info
*info
;
3874 info
= alloc_inode_sb(sb
, shmem_inode_cachep
, GFP_KERNEL
);
3877 return &info
->vfs_inode
;
3880 static void shmem_free_in_core_inode(struct inode
*inode
)
3882 if (S_ISLNK(inode
->i_mode
))
3883 kfree(inode
->i_link
);
3884 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3887 static void shmem_destroy_inode(struct inode
*inode
)
3889 if (S_ISREG(inode
->i_mode
))
3890 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3893 static void shmem_init_inode(void *foo
)
3895 struct shmem_inode_info
*info
= foo
;
3896 inode_init_once(&info
->vfs_inode
);
3899 static void shmem_init_inodecache(void)
3901 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3902 sizeof(struct shmem_inode_info
),
3903 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3906 static void shmem_destroy_inodecache(void)
3908 kmem_cache_destroy(shmem_inode_cachep
);
3911 /* Keep the page in page cache instead of truncating it */
3912 static int shmem_error_remove_page(struct address_space
*mapping
,
3918 const struct address_space_operations shmem_aops
= {
3919 .writepage
= shmem_writepage
,
3920 .dirty_folio
= noop_dirty_folio
,
3922 .write_begin
= shmem_write_begin
,
3923 .write_end
= shmem_write_end
,
3925 #ifdef CONFIG_MIGRATION
3926 .migrate_folio
= migrate_folio
,
3928 .error_remove_page
= shmem_error_remove_page
,
3930 EXPORT_SYMBOL(shmem_aops
);
3932 static const struct file_operations shmem_file_operations
= {
3934 .open
= generic_file_open
,
3935 .get_unmapped_area
= shmem_get_unmapped_area
,
3937 .llseek
= shmem_file_llseek
,
3938 .read_iter
= shmem_file_read_iter
,
3939 .write_iter
= generic_file_write_iter
,
3940 .fsync
= noop_fsync
,
3941 .splice_read
= generic_file_splice_read
,
3942 .splice_write
= iter_file_splice_write
,
3943 .fallocate
= shmem_fallocate
,
3947 static const struct inode_operations shmem_inode_operations
= {
3948 .getattr
= shmem_getattr
,
3949 .setattr
= shmem_setattr
,
3950 #ifdef CONFIG_TMPFS_XATTR
3951 .listxattr
= shmem_listxattr
,
3952 .set_acl
= simple_set_acl
,
3953 .fileattr_get
= shmem_fileattr_get
,
3954 .fileattr_set
= shmem_fileattr_set
,
3958 static const struct inode_operations shmem_dir_inode_operations
= {
3960 .getattr
= shmem_getattr
,
3961 .create
= shmem_create
,
3962 .lookup
= simple_lookup
,
3964 .unlink
= shmem_unlink
,
3965 .symlink
= shmem_symlink
,
3966 .mkdir
= shmem_mkdir
,
3967 .rmdir
= shmem_rmdir
,
3968 .mknod
= shmem_mknod
,
3969 .rename
= shmem_rename2
,
3970 .tmpfile
= shmem_tmpfile
,
3972 #ifdef CONFIG_TMPFS_XATTR
3973 .listxattr
= shmem_listxattr
,
3974 .fileattr_get
= shmem_fileattr_get
,
3975 .fileattr_set
= shmem_fileattr_set
,
3977 #ifdef CONFIG_TMPFS_POSIX_ACL
3978 .setattr
= shmem_setattr
,
3979 .set_acl
= simple_set_acl
,
3983 static const struct inode_operations shmem_special_inode_operations
= {
3984 .getattr
= shmem_getattr
,
3985 #ifdef CONFIG_TMPFS_XATTR
3986 .listxattr
= shmem_listxattr
,
3988 #ifdef CONFIG_TMPFS_POSIX_ACL
3989 .setattr
= shmem_setattr
,
3990 .set_acl
= simple_set_acl
,
3994 static const struct super_operations shmem_ops
= {
3995 .alloc_inode
= shmem_alloc_inode
,
3996 .free_inode
= shmem_free_in_core_inode
,
3997 .destroy_inode
= shmem_destroy_inode
,
3999 .statfs
= shmem_statfs
,
4000 .show_options
= shmem_show_options
,
4002 .evict_inode
= shmem_evict_inode
,
4003 .drop_inode
= generic_delete_inode
,
4004 .put_super
= shmem_put_super
,
4005 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4006 .nr_cached_objects
= shmem_unused_huge_count
,
4007 .free_cached_objects
= shmem_unused_huge_scan
,
4011 static const struct vm_operations_struct shmem_vm_ops
= {
4012 .fault
= shmem_fault
,
4013 .map_pages
= filemap_map_pages
,
4015 .set_policy
= shmem_set_policy
,
4016 .get_policy
= shmem_get_policy
,
4020 static const struct vm_operations_struct shmem_anon_vm_ops
= {
4021 .fault
= shmem_fault
,
4022 .map_pages
= filemap_map_pages
,
4024 .set_policy
= shmem_set_policy
,
4025 .get_policy
= shmem_get_policy
,
4029 int shmem_init_fs_context(struct fs_context
*fc
)
4031 struct shmem_options
*ctx
;
4033 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
4037 ctx
->mode
= 0777 | S_ISVTX
;
4038 ctx
->uid
= current_fsuid();
4039 ctx
->gid
= current_fsgid();
4041 fc
->fs_private
= ctx
;
4042 fc
->ops
= &shmem_fs_context_ops
;
4046 static struct file_system_type shmem_fs_type
= {
4047 .owner
= THIS_MODULE
,
4049 .init_fs_context
= shmem_init_fs_context
,
4051 .parameters
= shmem_fs_parameters
,
4053 .kill_sb
= kill_litter_super
,
4055 .fs_flags
= FS_USERNS_MOUNT
| FS_ALLOW_IDMAP
,
4057 .fs_flags
= FS_USERNS_MOUNT
,
4061 void __init
shmem_init(void)
4065 shmem_init_inodecache();
4067 error
= register_filesystem(&shmem_fs_type
);
4069 pr_err("Could not register tmpfs\n");
4073 shm_mnt
= kern_mount(&shmem_fs_type
);
4074 if (IS_ERR(shm_mnt
)) {
4075 error
= PTR_ERR(shm_mnt
);
4076 pr_err("Could not kern_mount tmpfs\n");
4080 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4081 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
4082 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4084 shmem_huge
= SHMEM_HUGE_NEVER
; /* just in case it was patched */
4089 unregister_filesystem(&shmem_fs_type
);
4091 shmem_destroy_inodecache();
4092 shm_mnt
= ERR_PTR(error
);
4095 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4096 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4097 struct kobj_attribute
*attr
, char *buf
)
4099 static const int values
[] = {
4101 SHMEM_HUGE_WITHIN_SIZE
,
4110 for (i
= 0; i
< ARRAY_SIZE(values
); i
++) {
4111 len
+= sysfs_emit_at(buf
, len
,
4112 shmem_huge
== values
[i
] ? "%s[%s]" : "%s%s",
4114 shmem_format_huge(values
[i
]));
4117 len
+= sysfs_emit_at(buf
, len
, "\n");
4122 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4123 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4128 if (count
+ 1 > sizeof(tmp
))
4130 memcpy(tmp
, buf
, count
);
4132 if (count
&& tmp
[count
- 1] == '\n')
4133 tmp
[count
- 1] = '\0';
4135 huge
= shmem_parse_huge(tmp
);
4136 if (huge
== -EINVAL
)
4138 if (!has_transparent_hugepage() &&
4139 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4143 if (shmem_huge
> SHMEM_HUGE_DENY
)
4144 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4148 struct kobj_attribute shmem_enabled_attr
= __ATTR_RW(shmem_enabled
);
4149 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4151 #else /* !CONFIG_SHMEM */
4154 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4156 * This is intended for small system where the benefits of the full
4157 * shmem code (swap-backed and resource-limited) are outweighed by
4158 * their complexity. On systems without swap this code should be
4159 * effectively equivalent, but much lighter weight.
4162 static struct file_system_type shmem_fs_type
= {
4164 .init_fs_context
= ramfs_init_fs_context
,
4165 .parameters
= ramfs_fs_parameters
,
4166 .kill_sb
= kill_litter_super
,
4167 .fs_flags
= FS_USERNS_MOUNT
,
4170 void __init
shmem_init(void)
4172 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4174 shm_mnt
= kern_mount(&shmem_fs_type
);
4175 BUG_ON(IS_ERR(shm_mnt
));
4178 int shmem_unuse(unsigned int type
)
4183 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
4188 void shmem_unlock_mapping(struct address_space
*mapping
)
4193 unsigned long shmem_get_unmapped_area(struct file
*file
,
4194 unsigned long addr
, unsigned long len
,
4195 unsigned long pgoff
, unsigned long flags
)
4197 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4201 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4203 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4205 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4207 #define shmem_vm_ops generic_file_vm_ops
4208 #define shmem_anon_vm_ops generic_file_vm_ops
4209 #define shmem_file_operations ramfs_file_operations
4210 #define shmem_get_inode(idmap, sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4211 #define shmem_acct_size(flags, size) 0
4212 #define shmem_unacct_size(flags, size) do {} while (0)
4214 #endif /* CONFIG_SHMEM */
4218 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4219 unsigned long flags
, unsigned int i_flags
)
4221 struct inode
*inode
;
4225 return ERR_CAST(mnt
);
4227 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4228 return ERR_PTR(-EINVAL
);
4230 if (shmem_acct_size(flags
, size
))
4231 return ERR_PTR(-ENOMEM
);
4233 if (is_idmapped_mnt(mnt
))
4234 return ERR_PTR(-EINVAL
);
4236 inode
= shmem_get_inode(&nop_mnt_idmap
, mnt
->mnt_sb
, NULL
,
4237 S_IFREG
| S_IRWXUGO
, 0, flags
);
4238 if (unlikely(!inode
)) {
4239 shmem_unacct_size(flags
, size
);
4240 return ERR_PTR(-ENOSPC
);
4242 inode
->i_flags
|= i_flags
;
4243 inode
->i_size
= size
;
4244 clear_nlink(inode
); /* It is unlinked */
4245 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4247 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4248 &shmem_file_operations
);
4255 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4256 * kernel internal. There will be NO LSM permission checks against the
4257 * underlying inode. So users of this interface must do LSM checks at a
4258 * higher layer. The users are the big_key and shm implementations. LSM
4259 * checks are provided at the key or shm level rather than the inode.
4260 * @name: name for dentry (to be seen in /proc/<pid>/maps
4261 * @size: size to be set for the file
4262 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4264 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4266 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4270 * shmem_file_setup - get an unlinked file living in tmpfs
4271 * @name: name for dentry (to be seen in /proc/<pid>/maps
4272 * @size: size to be set for the file
4273 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4275 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4277 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4279 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4282 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4283 * @mnt: the tmpfs mount where the file will be created
4284 * @name: name for dentry (to be seen in /proc/<pid>/maps
4285 * @size: size to be set for the file
4286 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4288 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4289 loff_t size
, unsigned long flags
)
4291 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4293 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4296 * shmem_zero_setup - setup a shared anonymous mapping
4297 * @vma: the vma to be mmapped is prepared by do_mmap
4299 int shmem_zero_setup(struct vm_area_struct
*vma
)
4302 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4305 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4306 * between XFS directory reading and selinux: since this file is only
4307 * accessible to the user through its mapping, use S_PRIVATE flag to
4308 * bypass file security, in the same way as shmem_kernel_file_setup().
4310 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4312 return PTR_ERR(file
);
4316 vma
->vm_file
= file
;
4317 vma
->vm_ops
= &shmem_anon_vm_ops
;
4323 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4324 * @mapping: the folio's address_space
4325 * @index: the folio index
4326 * @gfp: the page allocator flags to use if allocating
4328 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4329 * with any new page allocations done using the specified allocation flags.
4330 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4331 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4332 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4334 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4335 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4337 struct folio
*shmem_read_folio_gfp(struct address_space
*mapping
,
4338 pgoff_t index
, gfp_t gfp
)
4341 struct inode
*inode
= mapping
->host
;
4342 struct folio
*folio
;
4345 BUG_ON(!shmem_mapping(mapping
));
4346 error
= shmem_get_folio_gfp(inode
, index
, &folio
, SGP_CACHE
,
4347 gfp
, NULL
, NULL
, NULL
);
4349 return ERR_PTR(error
);
4351 folio_unlock(folio
);
4355 * The tiny !SHMEM case uses ramfs without swap
4357 return mapping_read_folio_gfp(mapping
, index
, gfp
);
4360 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp
);
4362 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4363 pgoff_t index
, gfp_t gfp
)
4365 struct folio
*folio
= shmem_read_folio_gfp(mapping
, index
, gfp
);
4369 return &folio
->page
;
4371 page
= folio_file_page(folio
, index
);
4372 if (PageHWPoison(page
)) {
4374 return ERR_PTR(-EIO
);
4379 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);