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/swap.h>
37 #include <linux/uio.h>
38 #include <linux/hugetlb.h>
39 #include <linux/fs_parser.h>
40 #include <linux/swapfile.h>
43 static struct vfsmount
*shm_mnt
;
47 * This virtual memory filesystem is heavily based on the ramfs. It
48 * extends ramfs by the ability to use swap and honor resource limits
49 * which makes it a completely usable filesystem.
52 #include <linux/xattr.h>
53 #include <linux/exportfs.h>
54 #include <linux/posix_acl.h>
55 #include <linux/posix_acl_xattr.h>
56 #include <linux/mman.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/backing-dev.h>
60 #include <linux/shmem_fs.h>
61 #include <linux/writeback.h>
62 #include <linux/pagevec.h>
63 #include <linux/percpu_counter.h>
64 #include <linux/falloc.h>
65 #include <linux/splice.h>
66 #include <linux/security.h>
67 #include <linux/swapops.h>
68 #include <linux/mempolicy.h>
69 #include <linux/namei.h>
70 #include <linux/ctype.h>
71 #include <linux/migrate.h>
72 #include <linux/highmem.h>
73 #include <linux/seq_file.h>
74 #include <linux/magic.h>
75 #include <linux/syscalls.h>
76 #include <linux/fcntl.h>
77 #include <uapi/linux/memfd.h>
78 #include <linux/userfaultfd_k.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
82 #include <linux/uaccess.h>
86 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 /* Pretend that each entry is of this size in directory's i_size */
90 #define BOGO_DIRENT_SIZE 20
92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93 #define SHORT_SYMLINK_LEN 128
96 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_rwsem making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
100 struct shmem_falloc
{
101 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
102 pgoff_t start
; /* start of range currently being fallocated */
103 pgoff_t next
; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
108 struct shmem_options
{
109 unsigned long long blocks
;
110 unsigned long long inodes
;
111 struct mempolicy
*mpol
;
118 #define SHMEM_SEEN_BLOCKS 1
119 #define SHMEM_SEEN_INODES 2
120 #define SHMEM_SEEN_HUGE 4
121 #define SHMEM_SEEN_INUMS 8
125 static unsigned long shmem_default_max_blocks(void)
127 return totalram_pages() / 2;
130 static unsigned long shmem_default_max_inodes(void)
132 unsigned long nr_pages
= totalram_pages();
134 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
138 static int shmem_swapin_folio(struct inode
*inode
, pgoff_t index
,
139 struct folio
**foliop
, enum sgp_type sgp
,
140 gfp_t gfp
, struct vm_area_struct
*vma
,
141 vm_fault_t
*fault_type
);
142 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
143 struct page
**pagep
, enum sgp_type sgp
,
144 gfp_t gfp
, struct vm_area_struct
*vma
,
145 struct vm_fault
*vmf
, vm_fault_t
*fault_type
);
147 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
148 struct page
**pagep
, enum sgp_type sgp
)
150 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
151 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
154 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
156 return sb
->s_fs_info
;
160 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
161 * for shared memory and for shared anonymous (/dev/zero) mappings
162 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
163 * consistent with the pre-accounting of private mappings ...
165 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
167 return (flags
& VM_NORESERVE
) ?
168 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
171 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
173 if (!(flags
& VM_NORESERVE
))
174 vm_unacct_memory(VM_ACCT(size
));
177 static inline int shmem_reacct_size(unsigned long flags
,
178 loff_t oldsize
, loff_t newsize
)
180 if (!(flags
& VM_NORESERVE
)) {
181 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
182 return security_vm_enough_memory_mm(current
->mm
,
183 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
184 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
185 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
191 * ... whereas tmpfs objects are accounted incrementally as
192 * pages are allocated, in order to allow large sparse files.
193 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
194 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
196 static inline int shmem_acct_block(unsigned long flags
, long pages
)
198 if (!(flags
& VM_NORESERVE
))
201 return security_vm_enough_memory_mm(current
->mm
,
202 pages
* VM_ACCT(PAGE_SIZE
));
205 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
207 if (flags
& VM_NORESERVE
)
208 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
211 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
213 struct shmem_inode_info
*info
= SHMEM_I(inode
);
214 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
216 if (shmem_acct_block(info
->flags
, pages
))
219 if (sbinfo
->max_blocks
) {
220 if (percpu_counter_compare(&sbinfo
->used_blocks
,
221 sbinfo
->max_blocks
- pages
) > 0)
223 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
229 shmem_unacct_blocks(info
->flags
, pages
);
233 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
235 struct shmem_inode_info
*info
= SHMEM_I(inode
);
236 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
238 if (sbinfo
->max_blocks
)
239 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
240 shmem_unacct_blocks(info
->flags
, pages
);
243 static const struct super_operations shmem_ops
;
244 const struct address_space_operations shmem_aops
;
245 static const struct file_operations shmem_file_operations
;
246 static const struct inode_operations shmem_inode_operations
;
247 static const struct inode_operations shmem_dir_inode_operations
;
248 static const struct inode_operations shmem_special_inode_operations
;
249 static const struct vm_operations_struct shmem_vm_ops
;
250 static struct file_system_type shmem_fs_type
;
252 bool vma_is_shmem(struct vm_area_struct
*vma
)
254 return vma
->vm_ops
== &shmem_vm_ops
;
257 static LIST_HEAD(shmem_swaplist
);
258 static DEFINE_MUTEX(shmem_swaplist_mutex
);
261 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
262 * produces a novel ino for the newly allocated inode.
264 * It may also be called when making a hard link to permit the space needed by
265 * each dentry. However, in that case, no new inode number is needed since that
266 * internally draws from another pool of inode numbers (currently global
267 * get_next_ino()). This case is indicated by passing NULL as inop.
269 #define SHMEM_INO_BATCH 1024
270 static int shmem_reserve_inode(struct super_block
*sb
, ino_t
*inop
)
272 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
275 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
276 raw_spin_lock(&sbinfo
->stat_lock
);
277 if (sbinfo
->max_inodes
) {
278 if (!sbinfo
->free_inodes
) {
279 raw_spin_unlock(&sbinfo
->stat_lock
);
282 sbinfo
->free_inodes
--;
285 ino
= sbinfo
->next_ino
++;
286 if (unlikely(is_zero_ino(ino
)))
287 ino
= sbinfo
->next_ino
++;
288 if (unlikely(!sbinfo
->full_inums
&&
291 * Emulate get_next_ino uint wraparound for
294 if (IS_ENABLED(CONFIG_64BIT
))
295 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
296 __func__
, MINOR(sb
->s_dev
));
297 sbinfo
->next_ino
= 1;
298 ino
= sbinfo
->next_ino
++;
302 raw_spin_unlock(&sbinfo
->stat_lock
);
305 * __shmem_file_setup, one of our callers, is lock-free: it
306 * doesn't hold stat_lock in shmem_reserve_inode since
307 * max_inodes is always 0, and is called from potentially
308 * unknown contexts. As such, use a per-cpu batched allocator
309 * which doesn't require the per-sb stat_lock unless we are at
310 * the batch boundary.
312 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
313 * shmem mounts are not exposed to userspace, so we don't need
314 * to worry about things like glibc compatibility.
318 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
320 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
321 raw_spin_lock(&sbinfo
->stat_lock
);
322 ino
= sbinfo
->next_ino
;
323 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
324 raw_spin_unlock(&sbinfo
->stat_lock
);
325 if (unlikely(is_zero_ino(ino
)))
336 static void shmem_free_inode(struct super_block
*sb
)
338 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
339 if (sbinfo
->max_inodes
) {
340 raw_spin_lock(&sbinfo
->stat_lock
);
341 sbinfo
->free_inodes
++;
342 raw_spin_unlock(&sbinfo
->stat_lock
);
347 * shmem_recalc_inode - recalculate the block usage of an inode
348 * @inode: inode to recalc
350 * We have to calculate the free blocks since the mm can drop
351 * undirtied hole pages behind our back.
353 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
354 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
356 * It has to be called with the spinlock held.
358 static void shmem_recalc_inode(struct inode
*inode
)
360 struct shmem_inode_info
*info
= SHMEM_I(inode
);
363 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
365 info
->alloced
-= freed
;
366 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
367 shmem_inode_unacct_blocks(inode
, freed
);
371 bool shmem_charge(struct inode
*inode
, long pages
)
373 struct shmem_inode_info
*info
= SHMEM_I(inode
);
376 if (!shmem_inode_acct_block(inode
, pages
))
379 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
380 inode
->i_mapping
->nrpages
+= pages
;
382 spin_lock_irqsave(&info
->lock
, flags
);
383 info
->alloced
+= pages
;
384 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
385 shmem_recalc_inode(inode
);
386 spin_unlock_irqrestore(&info
->lock
, flags
);
391 void shmem_uncharge(struct inode
*inode
, long pages
)
393 struct shmem_inode_info
*info
= SHMEM_I(inode
);
396 /* nrpages adjustment done by __filemap_remove_folio() or caller */
398 spin_lock_irqsave(&info
->lock
, flags
);
399 info
->alloced
-= pages
;
400 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
401 shmem_recalc_inode(inode
);
402 spin_unlock_irqrestore(&info
->lock
, flags
);
404 shmem_inode_unacct_blocks(inode
, pages
);
408 * Replace item expected in xarray by a new item, while holding xa_lock.
410 static int shmem_replace_entry(struct address_space
*mapping
,
411 pgoff_t index
, void *expected
, void *replacement
)
413 XA_STATE(xas
, &mapping
->i_pages
, index
);
416 VM_BUG_ON(!expected
);
417 VM_BUG_ON(!replacement
);
418 item
= xas_load(&xas
);
419 if (item
!= expected
)
421 xas_store(&xas
, replacement
);
426 * Sometimes, before we decide whether to proceed or to fail, we must check
427 * that an entry was not already brought back from swap by a racing thread.
429 * Checking page is not enough: by the time a SwapCache page is locked, it
430 * might be reused, and again be SwapCache, using the same swap as before.
432 static bool shmem_confirm_swap(struct address_space
*mapping
,
433 pgoff_t index
, swp_entry_t swap
)
435 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
439 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
442 * disables huge pages for the mount;
444 * enables huge pages for the mount;
445 * SHMEM_HUGE_WITHIN_SIZE:
446 * only allocate huge pages if the page will be fully within i_size,
447 * also respect fadvise()/madvise() hints;
449 * only allocate huge pages if requested with fadvise()/madvise();
452 #define SHMEM_HUGE_NEVER 0
453 #define SHMEM_HUGE_ALWAYS 1
454 #define SHMEM_HUGE_WITHIN_SIZE 2
455 #define SHMEM_HUGE_ADVISE 3
459 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
462 * disables huge on shm_mnt and all mounts, for emergency use;
464 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
467 #define SHMEM_HUGE_DENY (-1)
468 #define SHMEM_HUGE_FORCE (-2)
470 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
471 /* ifdef here to avoid bloating shmem.o when not necessary */
473 static int shmem_huge __read_mostly
= SHMEM_HUGE_NEVER
;
475 bool shmem_is_huge(struct vm_area_struct
*vma
,
476 struct inode
*inode
, pgoff_t index
)
480 if (!S_ISREG(inode
->i_mode
))
482 if (shmem_huge
== SHMEM_HUGE_DENY
)
484 if (vma
&& ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
485 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
)))
487 if (shmem_huge
== SHMEM_HUGE_FORCE
)
490 switch (SHMEM_SB(inode
->i_sb
)->huge
) {
491 case SHMEM_HUGE_ALWAYS
:
493 case SHMEM_HUGE_WITHIN_SIZE
:
494 index
= round_up(index
+ 1, HPAGE_PMD_NR
);
495 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
496 if (i_size
>> PAGE_SHIFT
>= index
)
499 case SHMEM_HUGE_ADVISE
:
500 if (vma
&& (vma
->vm_flags
& VM_HUGEPAGE
))
508 #if defined(CONFIG_SYSFS)
509 static int shmem_parse_huge(const char *str
)
511 if (!strcmp(str
, "never"))
512 return SHMEM_HUGE_NEVER
;
513 if (!strcmp(str
, "always"))
514 return SHMEM_HUGE_ALWAYS
;
515 if (!strcmp(str
, "within_size"))
516 return SHMEM_HUGE_WITHIN_SIZE
;
517 if (!strcmp(str
, "advise"))
518 return SHMEM_HUGE_ADVISE
;
519 if (!strcmp(str
, "deny"))
520 return SHMEM_HUGE_DENY
;
521 if (!strcmp(str
, "force"))
522 return SHMEM_HUGE_FORCE
;
527 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
528 static const char *shmem_format_huge(int huge
)
531 case SHMEM_HUGE_NEVER
:
533 case SHMEM_HUGE_ALWAYS
:
535 case SHMEM_HUGE_WITHIN_SIZE
:
536 return "within_size";
537 case SHMEM_HUGE_ADVISE
:
539 case SHMEM_HUGE_DENY
:
541 case SHMEM_HUGE_FORCE
:
550 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
551 struct shrink_control
*sc
, unsigned long nr_to_split
)
553 LIST_HEAD(list
), *pos
, *next
;
554 LIST_HEAD(to_remove
);
556 struct shmem_inode_info
*info
;
558 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
561 if (list_empty(&sbinfo
->shrinklist
))
564 spin_lock(&sbinfo
->shrinklist_lock
);
565 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
566 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
569 inode
= igrab(&info
->vfs_inode
);
571 /* inode is about to be evicted */
573 list_del_init(&info
->shrinklist
);
577 /* Check if there's anything to gain */
578 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
579 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
580 list_move(&info
->shrinklist
, &to_remove
);
584 list_move(&info
->shrinklist
, &list
);
586 sbinfo
->shrinklist_len
--;
590 spin_unlock(&sbinfo
->shrinklist_lock
);
592 list_for_each_safe(pos
, next
, &to_remove
) {
593 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
594 inode
= &info
->vfs_inode
;
595 list_del_init(&info
->shrinklist
);
599 list_for_each_safe(pos
, next
, &list
) {
603 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
604 inode
= &info
->vfs_inode
;
606 if (nr_to_split
&& split
>= nr_to_split
)
609 index
= (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
;
610 folio
= filemap_get_folio(inode
->i_mapping
, index
);
614 /* No huge page at the end of the file: nothing to split */
615 if (!folio_test_large(folio
)) {
621 * Move the inode on the list back to shrinklist if we failed
622 * to lock the page at this time.
624 * Waiting for the lock may lead to deadlock in the
627 if (!folio_trylock(folio
)) {
632 ret
= split_huge_page(&folio
->page
);
636 /* If split failed move the inode on the list back to shrinklist */
642 list_del_init(&info
->shrinklist
);
646 * Make sure the inode is either on the global list or deleted
647 * from any local list before iput() since it could be deleted
648 * in another thread once we put the inode (then the local list
651 spin_lock(&sbinfo
->shrinklist_lock
);
652 list_move(&info
->shrinklist
, &sbinfo
->shrinklist
);
653 sbinfo
->shrinklist_len
++;
654 spin_unlock(&sbinfo
->shrinklist_lock
);
662 static long shmem_unused_huge_scan(struct super_block
*sb
,
663 struct shrink_control
*sc
)
665 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
667 if (!READ_ONCE(sbinfo
->shrinklist_len
))
670 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
673 static long shmem_unused_huge_count(struct super_block
*sb
,
674 struct shrink_control
*sc
)
676 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
677 return READ_ONCE(sbinfo
->shrinklist_len
);
679 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
681 #define shmem_huge SHMEM_HUGE_DENY
683 bool shmem_is_huge(struct vm_area_struct
*vma
,
684 struct inode
*inode
, pgoff_t index
)
689 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
690 struct shrink_control
*sc
, unsigned long nr_to_split
)
694 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
697 * Like filemap_add_folio, but error if expected item has gone.
699 static int shmem_add_to_page_cache(struct folio
*folio
,
700 struct address_space
*mapping
,
701 pgoff_t index
, void *expected
, gfp_t gfp
,
702 struct mm_struct
*charge_mm
)
704 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, folio_order(folio
));
705 long nr
= folio_nr_pages(folio
);
708 VM_BUG_ON_FOLIO(index
!= round_down(index
, nr
), folio
);
709 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
710 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio
), folio
);
711 VM_BUG_ON(expected
&& folio_test_large(folio
));
713 folio_ref_add(folio
, nr
);
714 folio
->mapping
= mapping
;
715 folio
->index
= index
;
717 if (!folio_test_swapcache(folio
)) {
718 error
= mem_cgroup_charge(folio
, charge_mm
, gfp
);
720 if (folio_test_pmd_mappable(folio
)) {
721 count_vm_event(THP_FILE_FALLBACK
);
722 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
727 folio_throttle_swaprate(folio
, gfp
);
731 if (expected
!= xas_find_conflict(&xas
)) {
732 xas_set_err(&xas
, -EEXIST
);
735 if (expected
&& xas_find_conflict(&xas
)) {
736 xas_set_err(&xas
, -EEXIST
);
739 xas_store(&xas
, folio
);
742 if (folio_test_pmd_mappable(folio
)) {
743 count_vm_event(THP_FILE_ALLOC
);
744 __lruvec_stat_mod_folio(folio
, NR_SHMEM_THPS
, nr
);
746 mapping
->nrpages
+= nr
;
747 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, nr
);
748 __lruvec_stat_mod_folio(folio
, NR_SHMEM
, nr
);
750 xas_unlock_irq(&xas
);
751 } while (xas_nomem(&xas
, gfp
));
753 if (xas_error(&xas
)) {
754 error
= xas_error(&xas
);
760 folio
->mapping
= NULL
;
761 folio_ref_sub(folio
, nr
);
766 * Like delete_from_page_cache, but substitutes swap for page.
768 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
770 struct address_space
*mapping
= page
->mapping
;
773 VM_BUG_ON_PAGE(PageCompound(page
), page
);
775 xa_lock_irq(&mapping
->i_pages
);
776 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
777 page
->mapping
= NULL
;
779 __dec_lruvec_page_state(page
, NR_FILE_PAGES
);
780 __dec_lruvec_page_state(page
, NR_SHMEM
);
781 xa_unlock_irq(&mapping
->i_pages
);
787 * Remove swap entry from page cache, free the swap and its page cache.
789 static int shmem_free_swap(struct address_space
*mapping
,
790 pgoff_t index
, void *radswap
)
794 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
797 free_swap_and_cache(radix_to_swp_entry(radswap
));
802 * Determine (in bytes) how many of the shmem object's pages mapped by the
803 * given offsets are swapped out.
805 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
806 * as long as the inode doesn't go away and racy results are not a problem.
808 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
809 pgoff_t start
, pgoff_t end
)
811 XA_STATE(xas
, &mapping
->i_pages
, start
);
813 unsigned long swapped
= 0;
816 xas_for_each(&xas
, page
, end
- 1) {
817 if (xas_retry(&xas
, page
))
819 if (xa_is_value(page
))
822 if (need_resched()) {
830 return swapped
<< PAGE_SHIFT
;
834 * Determine (in bytes) how many of the shmem object's pages mapped by the
835 * given vma is swapped out.
837 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
838 * as long as the inode doesn't go away and racy results are not a problem.
840 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
842 struct inode
*inode
= file_inode(vma
->vm_file
);
843 struct shmem_inode_info
*info
= SHMEM_I(inode
);
844 struct address_space
*mapping
= inode
->i_mapping
;
845 unsigned long swapped
;
847 /* Be careful as we don't hold info->lock */
848 swapped
= READ_ONCE(info
->swapped
);
851 * The easier cases are when the shmem object has nothing in swap, or
852 * the vma maps it whole. Then we can simply use the stats that we
858 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
859 return swapped
<< PAGE_SHIFT
;
861 /* Here comes the more involved part */
862 return shmem_partial_swap_usage(mapping
, vma
->vm_pgoff
,
863 vma
->vm_pgoff
+ vma_pages(vma
));
867 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
869 void shmem_unlock_mapping(struct address_space
*mapping
)
871 struct folio_batch fbatch
;
874 folio_batch_init(&fbatch
);
876 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
878 while (!mapping_unevictable(mapping
) &&
879 filemap_get_folios(mapping
, &index
, ~0UL, &fbatch
)) {
880 check_move_unevictable_folios(&fbatch
);
881 folio_batch_release(&fbatch
);
886 static struct folio
*shmem_get_partial_folio(struct inode
*inode
, pgoff_t index
)
892 * At first avoid shmem_getpage(,,,SGP_READ): that fails
893 * beyond i_size, and reports fallocated pages as holes.
895 folio
= __filemap_get_folio(inode
->i_mapping
, index
,
896 FGP_ENTRY
| FGP_LOCK
, 0);
897 if (!xa_is_value(folio
))
900 * But read a page back from swap if any of it is within i_size
901 * (although in some cases this is just a waste of time).
904 shmem_getpage(inode
, index
, &page
, SGP_READ
);
905 return page
? page_folio(page
) : NULL
;
909 * Remove range of pages and swap entries from page cache, and free them.
910 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
912 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
915 struct address_space
*mapping
= inode
->i_mapping
;
916 struct shmem_inode_info
*info
= SHMEM_I(inode
);
917 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
918 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
919 struct folio_batch fbatch
;
920 pgoff_t indices
[PAGEVEC_SIZE
];
923 long nr_swaps_freed
= 0;
928 end
= -1; /* unsigned, so actually very big */
930 if (info
->fallocend
> start
&& info
->fallocend
<= end
&& !unfalloc
)
931 info
->fallocend
= start
;
933 folio_batch_init(&fbatch
);
935 while (index
< end
&& find_lock_entries(mapping
, index
, end
- 1,
937 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
938 folio
= fbatch
.folios
[i
];
942 if (xa_is_value(folio
)) {
945 nr_swaps_freed
+= !shmem_free_swap(mapping
,
949 index
+= folio_nr_pages(folio
) - 1;
951 if (!unfalloc
|| !folio_test_uptodate(folio
))
952 truncate_inode_folio(mapping
, folio
);
955 folio_batch_remove_exceptionals(&fbatch
);
956 folio_batch_release(&fbatch
);
961 same_folio
= (lstart
>> PAGE_SHIFT
) == (lend
>> PAGE_SHIFT
);
962 folio
= shmem_get_partial_folio(inode
, lstart
>> PAGE_SHIFT
);
964 same_folio
= lend
< folio_pos(folio
) + folio_size(folio
);
965 folio_mark_dirty(folio
);
966 if (!truncate_inode_partial_folio(folio
, lstart
, lend
)) {
967 start
= folio
->index
+ folio_nr_pages(folio
);
977 folio
= shmem_get_partial_folio(inode
, lend
>> PAGE_SHIFT
);
979 folio_mark_dirty(folio
);
980 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
];
1003 if (xa_is_value(folio
)) {
1006 if (shmem_free_swap(mapping
, index
, folio
)) {
1007 /* Swap was replaced by page: retry */
1017 if (!unfalloc
|| !folio_test_uptodate(folio
)) {
1018 if (folio_mapping(folio
) != mapping
) {
1019 /* Page was replaced by swap: retry */
1020 folio_unlock(folio
);
1024 VM_BUG_ON_FOLIO(folio_test_writeback(folio
),
1026 truncate_inode_folio(mapping
, folio
);
1028 index
= folio
->index
+ folio_nr_pages(folio
) - 1;
1029 folio_unlock(folio
);
1031 folio_batch_remove_exceptionals(&fbatch
);
1032 folio_batch_release(&fbatch
);
1036 spin_lock_irq(&info
->lock
);
1037 info
->swapped
-= nr_swaps_freed
;
1038 shmem_recalc_inode(inode
);
1039 spin_unlock_irq(&info
->lock
);
1042 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1044 shmem_undo_range(inode
, lstart
, lend
, false);
1045 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1047 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1049 static int shmem_getattr(struct user_namespace
*mnt_userns
,
1050 const struct path
*path
, struct kstat
*stat
,
1051 u32 request_mask
, unsigned int query_flags
)
1053 struct inode
*inode
= path
->dentry
->d_inode
;
1054 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1056 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1057 spin_lock_irq(&info
->lock
);
1058 shmem_recalc_inode(inode
);
1059 spin_unlock_irq(&info
->lock
);
1061 if (info
->fsflags
& FS_APPEND_FL
)
1062 stat
->attributes
|= STATX_ATTR_APPEND
;
1063 if (info
->fsflags
& FS_IMMUTABLE_FL
)
1064 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
1065 if (info
->fsflags
& FS_NODUMP_FL
)
1066 stat
->attributes
|= STATX_ATTR_NODUMP
;
1067 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
1068 STATX_ATTR_IMMUTABLE
|
1070 generic_fillattr(&init_user_ns
, inode
, stat
);
1072 if (shmem_is_huge(NULL
, inode
, 0))
1073 stat
->blksize
= HPAGE_PMD_SIZE
;
1075 if (request_mask
& STATX_BTIME
) {
1076 stat
->result_mask
|= STATX_BTIME
;
1077 stat
->btime
.tv_sec
= info
->i_crtime
.tv_sec
;
1078 stat
->btime
.tv_nsec
= info
->i_crtime
.tv_nsec
;
1084 static int shmem_setattr(struct user_namespace
*mnt_userns
,
1085 struct dentry
*dentry
, struct iattr
*attr
)
1087 struct inode
*inode
= d_inode(dentry
);
1088 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1091 error
= setattr_prepare(&init_user_ns
, dentry
, attr
);
1095 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1096 loff_t oldsize
= inode
->i_size
;
1097 loff_t newsize
= attr
->ia_size
;
1099 /* protected by i_rwsem */
1100 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1101 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1104 if (newsize
!= oldsize
) {
1105 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1109 i_size_write(inode
, newsize
);
1110 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1112 if (newsize
<= oldsize
) {
1113 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1114 if (oldsize
> holebegin
)
1115 unmap_mapping_range(inode
->i_mapping
,
1118 shmem_truncate_range(inode
,
1119 newsize
, (loff_t
)-1);
1120 /* unmap again to remove racily COWed private pages */
1121 if (oldsize
> holebegin
)
1122 unmap_mapping_range(inode
->i_mapping
,
1127 setattr_copy(&init_user_ns
, inode
, attr
);
1128 if (attr
->ia_valid
& ATTR_MODE
)
1129 error
= posix_acl_chmod(&init_user_ns
, inode
, inode
->i_mode
);
1133 static void shmem_evict_inode(struct inode
*inode
)
1135 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1136 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1138 if (shmem_mapping(inode
->i_mapping
)) {
1139 shmem_unacct_size(info
->flags
, inode
->i_size
);
1141 mapping_set_exiting(inode
->i_mapping
);
1142 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1143 if (!list_empty(&info
->shrinklist
)) {
1144 spin_lock(&sbinfo
->shrinklist_lock
);
1145 if (!list_empty(&info
->shrinklist
)) {
1146 list_del_init(&info
->shrinklist
);
1147 sbinfo
->shrinklist_len
--;
1149 spin_unlock(&sbinfo
->shrinklist_lock
);
1151 while (!list_empty(&info
->swaplist
)) {
1152 /* Wait while shmem_unuse() is scanning this inode... */
1153 wait_var_event(&info
->stop_eviction
,
1154 !atomic_read(&info
->stop_eviction
));
1155 mutex_lock(&shmem_swaplist_mutex
);
1156 /* ...but beware of the race if we peeked too early */
1157 if (!atomic_read(&info
->stop_eviction
))
1158 list_del_init(&info
->swaplist
);
1159 mutex_unlock(&shmem_swaplist_mutex
);
1163 simple_xattrs_free(&info
->xattrs
);
1164 WARN_ON(inode
->i_blocks
);
1165 shmem_free_inode(inode
->i_sb
);
1169 static int shmem_find_swap_entries(struct address_space
*mapping
,
1170 pgoff_t start
, struct folio_batch
*fbatch
,
1171 pgoff_t
*indices
, unsigned int type
)
1173 XA_STATE(xas
, &mapping
->i_pages
, start
);
1174 struct folio
*folio
;
1178 xas_for_each(&xas
, folio
, ULONG_MAX
) {
1179 if (xas_retry(&xas
, folio
))
1182 if (!xa_is_value(folio
))
1185 entry
= radix_to_swp_entry(folio
);
1187 * swapin error entries can be found in the mapping. But they're
1188 * deliberately ignored here as we've done everything we can do.
1190 if (swp_type(entry
) != type
)
1193 indices
[folio_batch_count(fbatch
)] = xas
.xa_index
;
1194 if (!folio_batch_add(fbatch
, folio
))
1197 if (need_resched()) {
1204 return xas
.xa_index
;
1208 * Move the swapped pages for an inode to page cache. Returns the count
1209 * of pages swapped in, or the error in case of failure.
1211 static int shmem_unuse_swap_entries(struct inode
*inode
,
1212 struct folio_batch
*fbatch
, pgoff_t
*indices
)
1217 struct address_space
*mapping
= inode
->i_mapping
;
1219 for (i
= 0; i
< folio_batch_count(fbatch
); i
++) {
1220 struct folio
*folio
= fbatch
->folios
[i
];
1222 if (!xa_is_value(folio
))
1224 error
= shmem_swapin_folio(inode
, indices
[i
],
1226 mapping_gfp_mask(mapping
),
1229 folio_unlock(folio
);
1233 if (error
== -ENOMEM
)
1237 return error
? error
: ret
;
1241 * If swap found in inode, free it and move page from swapcache to filecache.
1243 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
)
1245 struct address_space
*mapping
= inode
->i_mapping
;
1247 struct folio_batch fbatch
;
1248 pgoff_t indices
[PAGEVEC_SIZE
];
1252 folio_batch_init(&fbatch
);
1253 shmem_find_swap_entries(mapping
, start
, &fbatch
, indices
, type
);
1254 if (folio_batch_count(&fbatch
) == 0) {
1259 ret
= shmem_unuse_swap_entries(inode
, &fbatch
, indices
);
1263 start
= indices
[folio_batch_count(&fbatch
) - 1];
1270 * Read all the shared memory data that resides in the swap
1271 * device 'type' back into memory, so the swap device can be
1274 int shmem_unuse(unsigned int type
)
1276 struct shmem_inode_info
*info
, *next
;
1279 if (list_empty(&shmem_swaplist
))
1282 mutex_lock(&shmem_swaplist_mutex
);
1283 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1284 if (!info
->swapped
) {
1285 list_del_init(&info
->swaplist
);
1289 * Drop the swaplist mutex while searching the inode for swap;
1290 * but before doing so, make sure shmem_evict_inode() will not
1291 * remove placeholder inode from swaplist, nor let it be freed
1292 * (igrab() would protect from unlink, but not from unmount).
1294 atomic_inc(&info
->stop_eviction
);
1295 mutex_unlock(&shmem_swaplist_mutex
);
1297 error
= shmem_unuse_inode(&info
->vfs_inode
, type
);
1300 mutex_lock(&shmem_swaplist_mutex
);
1301 next
= list_next_entry(info
, swaplist
);
1303 list_del_init(&info
->swaplist
);
1304 if (atomic_dec_and_test(&info
->stop_eviction
))
1305 wake_up_var(&info
->stop_eviction
);
1309 mutex_unlock(&shmem_swaplist_mutex
);
1315 * Move the page from the page cache to the swap cache.
1317 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1319 struct folio
*folio
= page_folio(page
);
1320 struct shmem_inode_info
*info
;
1321 struct address_space
*mapping
;
1322 struct inode
*inode
;
1327 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1328 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1329 * and its shmem_writeback() needs them to be split when swapping.
1331 if (PageTransCompound(page
)) {
1332 /* Ensure the subpages are still dirty */
1334 if (split_huge_page(page
) < 0)
1336 ClearPageDirty(page
);
1339 BUG_ON(!PageLocked(page
));
1340 mapping
= page
->mapping
;
1341 index
= page
->index
;
1342 inode
= mapping
->host
;
1343 info
= SHMEM_I(inode
);
1344 if (info
->flags
& VM_LOCKED
)
1346 if (!total_swap_pages
)
1350 * Our capabilities prevent regular writeback or sync from ever calling
1351 * shmem_writepage; but a stacking filesystem might use ->writepage of
1352 * its underlying filesystem, in which case tmpfs should write out to
1353 * swap only in response to memory pressure, and not for the writeback
1356 if (!wbc
->for_reclaim
) {
1357 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1362 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1363 * value into swapfile.c, the only way we can correctly account for a
1364 * fallocated page arriving here is now to initialize it and write it.
1366 * That's okay for a page already fallocated earlier, but if we have
1367 * not yet completed the fallocation, then (a) we want to keep track
1368 * of this page in case we have to undo it, and (b) it may not be a
1369 * good idea to continue anyway, once we're pushing into swap. So
1370 * reactivate the page, and let shmem_fallocate() quit when too many.
1372 if (!PageUptodate(page
)) {
1373 if (inode
->i_private
) {
1374 struct shmem_falloc
*shmem_falloc
;
1375 spin_lock(&inode
->i_lock
);
1376 shmem_falloc
= inode
->i_private
;
1378 !shmem_falloc
->waitq
&&
1379 index
>= shmem_falloc
->start
&&
1380 index
< shmem_falloc
->next
)
1381 shmem_falloc
->nr_unswapped
++;
1383 shmem_falloc
= NULL
;
1384 spin_unlock(&inode
->i_lock
);
1388 clear_highpage(page
);
1389 flush_dcache_page(page
);
1390 SetPageUptodate(page
);
1393 swap
= folio_alloc_swap(folio
);
1398 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1399 * if it's not already there. Do it now before the page is
1400 * moved to swap cache, when its pagelock no longer protects
1401 * the inode from eviction. But don't unlock the mutex until
1402 * we've incremented swapped, because shmem_unuse_inode() will
1403 * prune a !swapped inode from the swaplist under this mutex.
1405 mutex_lock(&shmem_swaplist_mutex
);
1406 if (list_empty(&info
->swaplist
))
1407 list_add(&info
->swaplist
, &shmem_swaplist
);
1409 if (add_to_swap_cache(page
, swap
,
1410 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1412 spin_lock_irq(&info
->lock
);
1413 shmem_recalc_inode(inode
);
1415 spin_unlock_irq(&info
->lock
);
1417 swap_shmem_alloc(swap
);
1418 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1420 mutex_unlock(&shmem_swaplist_mutex
);
1421 BUG_ON(page_mapped(page
));
1422 swap_writepage(page
, wbc
);
1426 mutex_unlock(&shmem_swaplist_mutex
);
1427 put_swap_page(page
, swap
);
1429 set_page_dirty(page
);
1430 if (wbc
->for_reclaim
)
1431 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1436 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1437 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1441 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1442 return; /* show nothing */
1444 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1446 seq_printf(seq
, ",mpol=%s", buffer
);
1449 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1451 struct mempolicy
*mpol
= NULL
;
1453 raw_spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1454 mpol
= sbinfo
->mpol
;
1456 raw_spin_unlock(&sbinfo
->stat_lock
);
1460 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1461 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1464 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1468 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1470 #define vm_policy vm_private_data
1473 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1474 struct shmem_inode_info
*info
, pgoff_t index
)
1476 /* Create a pseudo vma that just contains the policy */
1477 vma_init(vma
, NULL
);
1478 /* Bias interleave by inode number to distribute better across nodes */
1479 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1480 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1483 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1485 /* Drop reference taken by mpol_shared_policy_lookup() */
1486 mpol_cond_put(vma
->vm_policy
);
1489 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1490 struct shmem_inode_info
*info
, pgoff_t index
)
1492 struct vm_area_struct pvma
;
1494 struct vm_fault vmf
= {
1498 shmem_pseudo_vma_init(&pvma
, info
, index
);
1499 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1500 shmem_pseudo_vma_destroy(&pvma
);
1506 * Make sure huge_gfp is always more limited than limit_gfp.
1507 * Some of the flags set permissions, while others set limitations.
1509 static gfp_t
limit_gfp_mask(gfp_t huge_gfp
, gfp_t limit_gfp
)
1511 gfp_t allowflags
= __GFP_IO
| __GFP_FS
| __GFP_RECLAIM
;
1512 gfp_t denyflags
= __GFP_NOWARN
| __GFP_NORETRY
;
1513 gfp_t zoneflags
= limit_gfp
& GFP_ZONEMASK
;
1514 gfp_t result
= huge_gfp
& ~(allowflags
| GFP_ZONEMASK
);
1516 /* Allow allocations only from the originally specified zones. */
1517 result
|= zoneflags
;
1520 * Minimize the result gfp by taking the union with the deny flags,
1521 * and the intersection of the allow flags.
1523 result
|= (limit_gfp
& denyflags
);
1524 result
|= (huge_gfp
& limit_gfp
) & allowflags
;
1529 static struct folio
*shmem_alloc_hugefolio(gfp_t gfp
,
1530 struct shmem_inode_info
*info
, pgoff_t index
)
1532 struct vm_area_struct pvma
;
1533 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1535 struct folio
*folio
;
1537 hindex
= round_down(index
, HPAGE_PMD_NR
);
1538 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1542 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1543 folio
= vma_alloc_folio(gfp
, HPAGE_PMD_ORDER
, &pvma
, 0, true);
1544 shmem_pseudo_vma_destroy(&pvma
);
1546 count_vm_event(THP_FILE_FALLBACK
);
1550 static struct folio
*shmem_alloc_folio(gfp_t gfp
,
1551 struct shmem_inode_info
*info
, pgoff_t index
)
1553 struct vm_area_struct pvma
;
1554 struct folio
*folio
;
1556 shmem_pseudo_vma_init(&pvma
, info
, index
);
1557 folio
= vma_alloc_folio(gfp
, 0, &pvma
, 0, false);
1558 shmem_pseudo_vma_destroy(&pvma
);
1563 static struct page
*shmem_alloc_page(gfp_t gfp
,
1564 struct shmem_inode_info
*info
, pgoff_t index
)
1566 return &shmem_alloc_folio(gfp
, info
, index
)->page
;
1569 static struct folio
*shmem_alloc_and_acct_folio(gfp_t gfp
, struct inode
*inode
,
1570 pgoff_t index
, bool huge
)
1572 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1573 struct folio
*folio
;
1577 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1579 nr
= huge
? HPAGE_PMD_NR
: 1;
1581 if (!shmem_inode_acct_block(inode
, nr
))
1585 folio
= shmem_alloc_hugefolio(gfp
, info
, index
);
1587 folio
= shmem_alloc_folio(gfp
, info
, index
);
1589 __folio_set_locked(folio
);
1590 __folio_set_swapbacked(folio
);
1595 shmem_inode_unacct_blocks(inode
, nr
);
1597 return ERR_PTR(err
);
1601 * When a page is moved from swapcache to shmem filecache (either by the
1602 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1603 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1604 * ignorance of the mapping it belongs to. If that mapping has special
1605 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1606 * we may need to copy to a suitable page before moving to filecache.
1608 * In a future release, this may well be extended to respect cpuset and
1609 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1610 * but for now it is a simple matter of zone.
1612 static bool shmem_should_replace_folio(struct folio
*folio
, gfp_t gfp
)
1614 return folio_zonenum(folio
) > gfp_zone(gfp
);
1617 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1618 struct shmem_inode_info
*info
, pgoff_t index
)
1620 struct page
*oldpage
, *newpage
;
1621 struct folio
*old
, *new;
1622 struct address_space
*swap_mapping
;
1628 entry
.val
= page_private(oldpage
);
1629 swap_index
= swp_offset(entry
);
1630 swap_mapping
= page_mapping(oldpage
);
1633 * We have arrived here because our zones are constrained, so don't
1634 * limit chance of success by further cpuset and node constraints.
1636 gfp
&= ~GFP_CONSTRAINT_MASK
;
1637 newpage
= shmem_alloc_page(gfp
, info
, index
);
1642 copy_highpage(newpage
, oldpage
);
1643 flush_dcache_page(newpage
);
1645 __SetPageLocked(newpage
);
1646 __SetPageSwapBacked(newpage
);
1647 SetPageUptodate(newpage
);
1648 set_page_private(newpage
, entry
.val
);
1649 SetPageSwapCache(newpage
);
1652 * Our caller will very soon move newpage out of swapcache, but it's
1653 * a nice clean interface for us to replace oldpage by newpage there.
1655 xa_lock_irq(&swap_mapping
->i_pages
);
1656 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1658 old
= page_folio(oldpage
);
1659 new = page_folio(newpage
);
1660 mem_cgroup_migrate(old
, new);
1661 __inc_lruvec_page_state(newpage
, NR_FILE_PAGES
);
1662 __inc_lruvec_page_state(newpage
, NR_SHMEM
);
1663 __dec_lruvec_page_state(oldpage
, NR_FILE_PAGES
);
1664 __dec_lruvec_page_state(oldpage
, NR_SHMEM
);
1666 xa_unlock_irq(&swap_mapping
->i_pages
);
1668 if (unlikely(error
)) {
1670 * Is this possible? I think not, now that our callers check
1671 * both PageSwapCache and page_private after getting page lock;
1672 * but be defensive. Reverse old to newpage for clear and free.
1676 lru_cache_add(newpage
);
1680 ClearPageSwapCache(oldpage
);
1681 set_page_private(oldpage
, 0);
1683 unlock_page(oldpage
);
1689 static void shmem_set_folio_swapin_error(struct inode
*inode
, pgoff_t index
,
1690 struct folio
*folio
, swp_entry_t swap
)
1692 struct address_space
*mapping
= inode
->i_mapping
;
1693 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1694 swp_entry_t swapin_error
;
1697 swapin_error
= make_swapin_error_entry(&folio
->page
);
1698 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
,
1699 swp_to_radix_entry(swap
),
1700 swp_to_radix_entry(swapin_error
), 0);
1701 if (old
!= swp_to_radix_entry(swap
))
1704 folio_wait_writeback(folio
);
1705 delete_from_swap_cache(folio
);
1706 spin_lock_irq(&info
->lock
);
1708 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks won't
1709 * be 0 when inode is released and thus trigger WARN_ON(inode->i_blocks) in
1710 * shmem_evict_inode.
1714 shmem_recalc_inode(inode
);
1715 spin_unlock_irq(&info
->lock
);
1720 * Swap in the folio pointed to by *foliop.
1721 * Caller has to make sure that *foliop contains a valid swapped folio.
1722 * Returns 0 and the folio in foliop if success. On failure, returns the
1723 * error code and NULL in *foliop.
1725 static int shmem_swapin_folio(struct inode
*inode
, pgoff_t index
,
1726 struct folio
**foliop
, enum sgp_type sgp
,
1727 gfp_t gfp
, struct vm_area_struct
*vma
,
1728 vm_fault_t
*fault_type
)
1730 struct address_space
*mapping
= inode
->i_mapping
;
1731 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1732 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: NULL
;
1734 struct folio
*folio
= NULL
;
1738 VM_BUG_ON(!*foliop
|| !xa_is_value(*foliop
));
1739 swap
= radix_to_swp_entry(*foliop
);
1742 if (is_swapin_error_entry(swap
))
1745 /* Look it up and read it in.. */
1746 page
= lookup_swap_cache(swap
, NULL
, 0);
1748 /* Or update major stats only when swapin succeeds?? */
1750 *fault_type
|= VM_FAULT_MAJOR
;
1751 count_vm_event(PGMAJFAULT
);
1752 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1754 /* Here we actually start the io */
1755 page
= shmem_swapin(swap
, gfp
, info
, index
);
1761 folio
= page_folio(page
);
1763 /* We have to do this with folio locked to prevent races */
1765 if (!folio_test_swapcache(folio
) ||
1766 folio_swap_entry(folio
).val
!= swap
.val
||
1767 !shmem_confirm_swap(mapping
, index
, swap
)) {
1771 if (!folio_test_uptodate(folio
)) {
1775 folio_wait_writeback(folio
);
1778 * Some architectures may have to restore extra metadata to the
1779 * folio after reading from swap.
1781 arch_swap_restore(swap
, folio
);
1783 if (shmem_should_replace_folio(folio
, gfp
)) {
1784 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1785 folio
= page_folio(page
);
1790 error
= shmem_add_to_page_cache(folio
, mapping
, index
,
1791 swp_to_radix_entry(swap
), gfp
,
1796 spin_lock_irq(&info
->lock
);
1798 shmem_recalc_inode(inode
);
1799 spin_unlock_irq(&info
->lock
);
1801 if (sgp
== SGP_WRITE
)
1802 folio_mark_accessed(folio
);
1804 delete_from_swap_cache(folio
);
1805 folio_mark_dirty(folio
);
1811 if (!shmem_confirm_swap(mapping
, index
, swap
))
1814 shmem_set_folio_swapin_error(inode
, index
, folio
, swap
);
1817 folio_unlock(folio
);
1825 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1827 * If we allocate a new one we do not mark it dirty. That's up to the
1828 * vm. If we swap it in we mark it dirty since we also free the swap
1829 * entry since a page cannot live in both the swap and page cache.
1831 * vma, vmf, and fault_type are only supplied by shmem_fault:
1832 * otherwise they are NULL.
1834 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1835 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1836 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1837 vm_fault_t
*fault_type
)
1839 struct address_space
*mapping
= inode
->i_mapping
;
1840 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1841 struct shmem_sb_info
*sbinfo
;
1842 struct mm_struct
*charge_mm
;
1843 struct folio
*folio
;
1844 pgoff_t hindex
= index
;
1850 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1853 if (sgp
<= SGP_CACHE
&&
1854 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1858 sbinfo
= SHMEM_SB(inode
->i_sb
);
1859 charge_mm
= vma
? vma
->vm_mm
: NULL
;
1861 folio
= __filemap_get_folio(mapping
, index
, FGP_ENTRY
| FGP_LOCK
, 0);
1862 if (folio
&& vma
&& userfaultfd_minor(vma
)) {
1863 if (!xa_is_value(folio
)) {
1864 folio_unlock(folio
);
1867 *fault_type
= handle_userfault(vmf
, VM_UFFD_MINOR
);
1871 if (xa_is_value(folio
)) {
1872 error
= shmem_swapin_folio(inode
, index
, &folio
,
1873 sgp
, gfp
, vma
, fault_type
);
1874 if (error
== -EEXIST
)
1877 *pagep
= &folio
->page
;
1882 hindex
= folio
->index
;
1883 if (sgp
== SGP_WRITE
)
1884 folio_mark_accessed(folio
);
1885 if (folio_test_uptodate(folio
))
1887 /* fallocated page */
1888 if (sgp
!= SGP_READ
)
1890 folio_unlock(folio
);
1895 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1896 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1899 if (sgp
== SGP_READ
)
1901 if (sgp
== SGP_NOALLOC
)
1905 * Fast cache lookup and swap lookup did not find it: allocate.
1908 if (vma
&& userfaultfd_missing(vma
)) {
1909 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1913 if (!shmem_is_huge(vma
, inode
, index
))
1916 huge_gfp
= vma_thp_gfp_mask(vma
);
1917 huge_gfp
= limit_gfp_mask(huge_gfp
, gfp
);
1918 folio
= shmem_alloc_and_acct_folio(huge_gfp
, inode
, index
, true);
1919 if (IS_ERR(folio
)) {
1921 folio
= shmem_alloc_and_acct_folio(gfp
, inode
, index
, false);
1923 if (IS_ERR(folio
)) {
1926 error
= PTR_ERR(folio
);
1928 if (error
!= -ENOSPC
)
1931 * Try to reclaim some space by splitting a huge page
1932 * beyond i_size on the filesystem.
1937 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1938 if (ret
== SHRINK_STOP
)
1946 hindex
= round_down(index
, folio_nr_pages(folio
));
1948 if (sgp
== SGP_WRITE
)
1949 __folio_set_referenced(folio
);
1951 error
= shmem_add_to_page_cache(folio
, mapping
, hindex
,
1952 NULL
, gfp
& GFP_RECLAIM_MASK
,
1956 folio_add_lru(folio
);
1958 spin_lock_irq(&info
->lock
);
1959 info
->alloced
+= folio_nr_pages(folio
);
1960 inode
->i_blocks
+= (blkcnt_t
)BLOCKS_PER_PAGE
<< folio_order(folio
);
1961 shmem_recalc_inode(inode
);
1962 spin_unlock_irq(&info
->lock
);
1965 if (folio_test_pmd_mappable(folio
) &&
1966 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1967 hindex
+ HPAGE_PMD_NR
- 1) {
1969 * Part of the huge page is beyond i_size: subject
1970 * to shrink under memory pressure.
1972 spin_lock(&sbinfo
->shrinklist_lock
);
1974 * _careful to defend against unlocked access to
1975 * ->shrink_list in shmem_unused_huge_shrink()
1977 if (list_empty_careful(&info
->shrinklist
)) {
1978 list_add_tail(&info
->shrinklist
,
1979 &sbinfo
->shrinklist
);
1980 sbinfo
->shrinklist_len
++;
1982 spin_unlock(&sbinfo
->shrinklist_lock
);
1986 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1988 if (sgp
== SGP_FALLOC
)
1992 * Let SGP_WRITE caller clear ends if write does not fill page;
1993 * but SGP_FALLOC on a page fallocated earlier must initialize
1994 * it now, lest undo on failure cancel our earlier guarantee.
1996 if (sgp
!= SGP_WRITE
&& !folio_test_uptodate(folio
)) {
1997 long i
, n
= folio_nr_pages(folio
);
1999 for (i
= 0; i
< n
; i
++)
2000 clear_highpage(folio_page(folio
, i
));
2001 flush_dcache_folio(folio
);
2002 folio_mark_uptodate(folio
);
2005 /* Perhaps the file has been truncated since we checked */
2006 if (sgp
<= SGP_CACHE
&&
2007 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
2009 folio_clear_dirty(folio
);
2010 filemap_remove_folio(folio
);
2011 spin_lock_irq(&info
->lock
);
2012 shmem_recalc_inode(inode
);
2013 spin_unlock_irq(&info
->lock
);
2019 *pagep
= folio_page(folio
, index
- hindex
);
2026 shmem_inode_unacct_blocks(inode
, folio_nr_pages(folio
));
2028 if (folio_test_large(folio
)) {
2029 folio_unlock(folio
);
2035 folio_unlock(folio
);
2038 if (error
== -ENOSPC
&& !once
++) {
2039 spin_lock_irq(&info
->lock
);
2040 shmem_recalc_inode(inode
);
2041 spin_unlock_irq(&info
->lock
);
2044 if (error
== -EEXIST
)
2050 * This is like autoremove_wake_function, but it removes the wait queue
2051 * entry unconditionally - even if something else had already woken the
2054 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2056 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2057 list_del_init(&wait
->entry
);
2061 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2063 struct vm_area_struct
*vma
= vmf
->vma
;
2064 struct inode
*inode
= file_inode(vma
->vm_file
);
2065 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2067 vm_fault_t ret
= VM_FAULT_LOCKED
;
2070 * Trinity finds that probing a hole which tmpfs is punching can
2071 * prevent the hole-punch from ever completing: which in turn
2072 * locks writers out with its hold on i_rwsem. So refrain from
2073 * faulting pages into the hole while it's being punched. Although
2074 * shmem_undo_range() does remove the additions, it may be unable to
2075 * keep up, as each new page needs its own unmap_mapping_range() call,
2076 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2078 * It does not matter if we sometimes reach this check just before the
2079 * hole-punch begins, so that one fault then races with the punch:
2080 * we just need to make racing faults a rare case.
2082 * The implementation below would be much simpler if we just used a
2083 * standard mutex or completion: but we cannot take i_rwsem in fault,
2084 * and bloating every shmem inode for this unlikely case would be sad.
2086 if (unlikely(inode
->i_private
)) {
2087 struct shmem_falloc
*shmem_falloc
;
2089 spin_lock(&inode
->i_lock
);
2090 shmem_falloc
= inode
->i_private
;
2092 shmem_falloc
->waitq
&&
2093 vmf
->pgoff
>= shmem_falloc
->start
&&
2094 vmf
->pgoff
< shmem_falloc
->next
) {
2096 wait_queue_head_t
*shmem_falloc_waitq
;
2097 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2099 ret
= VM_FAULT_NOPAGE
;
2100 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2102 ret
= VM_FAULT_RETRY
;
2104 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2105 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2106 TASK_UNINTERRUPTIBLE
);
2107 spin_unlock(&inode
->i_lock
);
2111 * shmem_falloc_waitq points into the shmem_fallocate()
2112 * stack of the hole-punching task: shmem_falloc_waitq
2113 * is usually invalid by the time we reach here, but
2114 * finish_wait() does not dereference it in that case;
2115 * though i_lock needed lest racing with wake_up_all().
2117 spin_lock(&inode
->i_lock
);
2118 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2119 spin_unlock(&inode
->i_lock
);
2125 spin_unlock(&inode
->i_lock
);
2128 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
,
2129 gfp
, vma
, vmf
, &ret
);
2131 return vmf_error(err
);
2135 unsigned long shmem_get_unmapped_area(struct file
*file
,
2136 unsigned long uaddr
, unsigned long len
,
2137 unsigned long pgoff
, unsigned long flags
)
2139 unsigned long (*get_area
)(struct file
*,
2140 unsigned long, unsigned long, unsigned long, unsigned long);
2142 unsigned long offset
;
2143 unsigned long inflated_len
;
2144 unsigned long inflated_addr
;
2145 unsigned long inflated_offset
;
2147 if (len
> TASK_SIZE
)
2150 get_area
= current
->mm
->get_unmapped_area
;
2151 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2153 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2155 if (IS_ERR_VALUE(addr
))
2157 if (addr
& ~PAGE_MASK
)
2159 if (addr
> TASK_SIZE
- len
)
2162 if (shmem_huge
== SHMEM_HUGE_DENY
)
2164 if (len
< HPAGE_PMD_SIZE
)
2166 if (flags
& MAP_FIXED
)
2169 * Our priority is to support MAP_SHARED mapped hugely;
2170 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2171 * But if caller specified an address hint and we allocated area there
2172 * successfully, respect that as before.
2177 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2178 struct super_block
*sb
;
2181 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2182 sb
= file_inode(file
)->i_sb
;
2185 * Called directly from mm/mmap.c, or drivers/char/mem.c
2186 * for "/dev/zero", to create a shared anonymous object.
2188 if (IS_ERR(shm_mnt
))
2190 sb
= shm_mnt
->mnt_sb
;
2192 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2196 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2197 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2199 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2202 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2203 if (inflated_len
> TASK_SIZE
)
2205 if (inflated_len
< len
)
2208 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2209 if (IS_ERR_VALUE(inflated_addr
))
2211 if (inflated_addr
& ~PAGE_MASK
)
2214 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2215 inflated_addr
+= offset
- inflated_offset
;
2216 if (inflated_offset
> offset
)
2217 inflated_addr
+= HPAGE_PMD_SIZE
;
2219 if (inflated_addr
> TASK_SIZE
- len
)
2221 return inflated_addr
;
2225 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2227 struct inode
*inode
= file_inode(vma
->vm_file
);
2228 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2231 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2234 struct inode
*inode
= file_inode(vma
->vm_file
);
2237 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2238 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2242 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
2244 struct inode
*inode
= file_inode(file
);
2245 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2246 int retval
= -ENOMEM
;
2249 * What serializes the accesses to info->flags?
2250 * ipc_lock_object() when called from shmctl_do_lock(),
2251 * no serialization needed when called from shm_destroy().
2253 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2254 if (!user_shm_lock(inode
->i_size
, ucounts
))
2256 info
->flags
|= VM_LOCKED
;
2257 mapping_set_unevictable(file
->f_mapping
);
2259 if (!lock
&& (info
->flags
& VM_LOCKED
) && ucounts
) {
2260 user_shm_unlock(inode
->i_size
, ucounts
);
2261 info
->flags
&= ~VM_LOCKED
;
2262 mapping_clear_unevictable(file
->f_mapping
);
2270 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2272 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2275 ret
= seal_check_future_write(info
->seals
, vma
);
2279 /* arm64 - allow memory tagging on RAM-based files */
2280 vma
->vm_flags
|= VM_MTE_ALLOWED
;
2282 file_accessed(file
);
2283 vma
->vm_ops
= &shmem_vm_ops
;
2287 #ifdef CONFIG_TMPFS_XATTR
2288 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2291 * chattr's fsflags are unrelated to extended attributes,
2292 * but tmpfs has chosen to enable them under the same config option.
2294 static void shmem_set_inode_flags(struct inode
*inode
, unsigned int fsflags
)
2296 unsigned int i_flags
= 0;
2298 if (fsflags
& FS_NOATIME_FL
)
2299 i_flags
|= S_NOATIME
;
2300 if (fsflags
& FS_APPEND_FL
)
2301 i_flags
|= S_APPEND
;
2302 if (fsflags
& FS_IMMUTABLE_FL
)
2303 i_flags
|= S_IMMUTABLE
;
2305 * But FS_NODUMP_FL does not require any action in i_flags.
2307 inode_set_flags(inode
, i_flags
, S_NOATIME
| S_APPEND
| S_IMMUTABLE
);
2310 static void shmem_set_inode_flags(struct inode
*inode
, unsigned int fsflags
)
2313 #define shmem_initxattrs NULL
2316 static struct inode
*shmem_get_inode(struct super_block
*sb
, struct inode
*dir
,
2317 umode_t mode
, dev_t dev
, unsigned long flags
)
2319 struct inode
*inode
;
2320 struct shmem_inode_info
*info
;
2321 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2324 if (shmem_reserve_inode(sb
, &ino
))
2327 inode
= new_inode(sb
);
2330 inode_init_owner(&init_user_ns
, inode
, dir
, mode
);
2331 inode
->i_blocks
= 0;
2332 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2333 inode
->i_generation
= prandom_u32();
2334 info
= SHMEM_I(inode
);
2335 memset(info
, 0, (char *)inode
- (char *)info
);
2336 spin_lock_init(&info
->lock
);
2337 atomic_set(&info
->stop_eviction
, 0);
2338 info
->seals
= F_SEAL_SEAL
;
2339 info
->flags
= flags
& VM_NORESERVE
;
2340 info
->i_crtime
= inode
->i_mtime
;
2341 info
->fsflags
= (dir
== NULL
) ? 0 :
2342 SHMEM_I(dir
)->fsflags
& SHMEM_FL_INHERITED
;
2344 shmem_set_inode_flags(inode
, info
->fsflags
);
2345 INIT_LIST_HEAD(&info
->shrinklist
);
2346 INIT_LIST_HEAD(&info
->swaplist
);
2347 simple_xattrs_init(&info
->xattrs
);
2348 cache_no_acl(inode
);
2349 mapping_set_large_folios(inode
->i_mapping
);
2351 switch (mode
& S_IFMT
) {
2353 inode
->i_op
= &shmem_special_inode_operations
;
2354 init_special_inode(inode
, mode
, dev
);
2357 inode
->i_mapping
->a_ops
= &shmem_aops
;
2358 inode
->i_op
= &shmem_inode_operations
;
2359 inode
->i_fop
= &shmem_file_operations
;
2360 mpol_shared_policy_init(&info
->policy
,
2361 shmem_get_sbmpol(sbinfo
));
2365 /* Some things misbehave if size == 0 on a directory */
2366 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2367 inode
->i_op
= &shmem_dir_inode_operations
;
2368 inode
->i_fop
= &simple_dir_operations
;
2372 * Must not load anything in the rbtree,
2373 * mpol_free_shared_policy will not be called.
2375 mpol_shared_policy_init(&info
->policy
, NULL
);
2379 lockdep_annotate_inode_mutex_key(inode
);
2381 shmem_free_inode(sb
);
2385 #ifdef CONFIG_USERFAULTFD
2386 int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2388 struct vm_area_struct
*dst_vma
,
2389 unsigned long dst_addr
,
2390 unsigned long src_addr
,
2391 bool zeropage
, bool wp_copy
,
2392 struct page
**pagep
)
2394 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2395 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2396 struct address_space
*mapping
= inode
->i_mapping
;
2397 gfp_t gfp
= mapping_gfp_mask(mapping
);
2398 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2400 struct folio
*folio
;
2405 if (!shmem_inode_acct_block(inode
, 1)) {
2407 * We may have got a page, returned -ENOENT triggering a retry,
2408 * and now we find ourselves with -ENOMEM. Release the page, to
2409 * avoid a BUG_ON in our caller.
2411 if (unlikely(*pagep
)) {
2420 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2422 goto out_unacct_blocks
;
2424 if (!zeropage
) { /* COPY */
2425 page_kaddr
= kmap_atomic(page
);
2426 ret
= copy_from_user(page_kaddr
,
2427 (const void __user
*)src_addr
,
2429 kunmap_atomic(page_kaddr
);
2431 /* fallback to copy_from_user outside mmap_lock */
2432 if (unlikely(ret
)) {
2435 /* don't free the page */
2436 goto out_unacct_blocks
;
2439 flush_dcache_page(page
);
2440 } else { /* ZEROPAGE */
2441 clear_user_highpage(page
, dst_addr
);
2448 VM_BUG_ON(PageLocked(page
));
2449 VM_BUG_ON(PageSwapBacked(page
));
2450 __SetPageLocked(page
);
2451 __SetPageSwapBacked(page
);
2452 __SetPageUptodate(page
);
2455 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2456 if (unlikely(pgoff
>= max_off
))
2459 folio
= page_folio(page
);
2460 ret
= shmem_add_to_page_cache(folio
, mapping
, pgoff
, NULL
,
2461 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2465 ret
= mfill_atomic_install_pte(dst_mm
, dst_pmd
, dst_vma
, dst_addr
,
2466 page
, true, wp_copy
);
2468 goto out_delete_from_cache
;
2470 spin_lock_irq(&info
->lock
);
2472 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2473 shmem_recalc_inode(inode
);
2474 spin_unlock_irq(&info
->lock
);
2478 out_delete_from_cache
:
2479 delete_from_page_cache(page
);
2484 shmem_inode_unacct_blocks(inode
, 1);
2487 #endif /* CONFIG_USERFAULTFD */
2490 static const struct inode_operations shmem_symlink_inode_operations
;
2491 static const struct inode_operations shmem_short_symlink_operations
;
2494 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2495 loff_t pos
, unsigned len
,
2496 struct page
**pagep
, void **fsdata
)
2498 struct inode
*inode
= mapping
->host
;
2499 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2500 pgoff_t index
= pos
>> PAGE_SHIFT
;
2503 /* i_rwsem is held by caller */
2504 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2505 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2506 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2508 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2512 ret
= shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2517 if (PageHWPoison(*pagep
)) {
2518 unlock_page(*pagep
);
2528 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2529 loff_t pos
, unsigned len
, unsigned copied
,
2530 struct page
*page
, void *fsdata
)
2532 struct inode
*inode
= mapping
->host
;
2534 if (pos
+ copied
> inode
->i_size
)
2535 i_size_write(inode
, pos
+ copied
);
2537 if (!PageUptodate(page
)) {
2538 struct page
*head
= compound_head(page
);
2539 if (PageTransCompound(page
)) {
2542 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2543 if (head
+ i
== page
)
2545 clear_highpage(head
+ i
);
2546 flush_dcache_page(head
+ i
);
2549 if (copied
< PAGE_SIZE
) {
2550 unsigned from
= pos
& (PAGE_SIZE
- 1);
2551 zero_user_segments(page
, 0, from
,
2552 from
+ copied
, PAGE_SIZE
);
2554 SetPageUptodate(head
);
2556 set_page_dirty(page
);
2563 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2565 struct file
*file
= iocb
->ki_filp
;
2566 struct inode
*inode
= file_inode(file
);
2567 struct address_space
*mapping
= inode
->i_mapping
;
2569 unsigned long offset
;
2572 loff_t
*ppos
= &iocb
->ki_pos
;
2574 index
= *ppos
>> PAGE_SHIFT
;
2575 offset
= *ppos
& ~PAGE_MASK
;
2578 struct page
*page
= NULL
;
2580 unsigned long nr
, ret
;
2581 loff_t i_size
= i_size_read(inode
);
2583 end_index
= i_size
>> PAGE_SHIFT
;
2584 if (index
> end_index
)
2586 if (index
== end_index
) {
2587 nr
= i_size
& ~PAGE_MASK
;
2592 error
= shmem_getpage(inode
, index
, &page
, SGP_READ
);
2594 if (error
== -EINVAL
)
2601 if (PageHWPoison(page
)) {
2609 * We must evaluate after, since reads (unlike writes)
2610 * are called without i_rwsem protection against truncate
2613 i_size
= i_size_read(inode
);
2614 end_index
= i_size
>> PAGE_SHIFT
;
2615 if (index
== end_index
) {
2616 nr
= i_size
& ~PAGE_MASK
;
2627 * If users can be writing to this page using arbitrary
2628 * virtual addresses, take care about potential aliasing
2629 * before reading the page on the kernel side.
2631 if (mapping_writably_mapped(mapping
))
2632 flush_dcache_page(page
);
2634 * Mark the page accessed if we read the beginning.
2637 mark_page_accessed(page
);
2639 * Ok, we have the page, and it's up-to-date, so
2640 * now we can copy it to user space...
2642 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2645 } else if (user_backed_iter(to
)) {
2647 * Copy to user tends to be so well optimized, but
2648 * clear_user() not so much, that it is noticeably
2649 * faster to copy the zero page instead of clearing.
2651 ret
= copy_page_to_iter(ZERO_PAGE(0), offset
, nr
, to
);
2654 * But submitting the same page twice in a row to
2655 * splice() - or others? - can result in confusion:
2656 * so don't attempt that optimization on pipes etc.
2658 ret
= iov_iter_zero(nr
, to
);
2663 index
+= offset
>> PAGE_SHIFT
;
2664 offset
&= ~PAGE_MASK
;
2666 if (!iov_iter_count(to
))
2675 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2676 file_accessed(file
);
2677 return retval
? retval
: error
;
2680 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2682 struct address_space
*mapping
= file
->f_mapping
;
2683 struct inode
*inode
= mapping
->host
;
2685 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2686 return generic_file_llseek_size(file
, offset
, whence
,
2687 MAX_LFS_FILESIZE
, i_size_read(inode
));
2692 /* We're holding i_rwsem so we can access i_size directly */
2693 offset
= mapping_seek_hole_data(mapping
, offset
, inode
->i_size
, whence
);
2695 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2696 inode_unlock(inode
);
2700 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2703 struct inode
*inode
= file_inode(file
);
2704 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2705 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2706 struct shmem_falloc shmem_falloc
;
2707 pgoff_t start
, index
, end
, undo_fallocend
;
2710 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2715 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2716 struct address_space
*mapping
= file
->f_mapping
;
2717 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2718 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2719 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2721 /* protected by i_rwsem */
2722 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2727 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2728 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2729 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2730 spin_lock(&inode
->i_lock
);
2731 inode
->i_private
= &shmem_falloc
;
2732 spin_unlock(&inode
->i_lock
);
2734 if ((u64
)unmap_end
> (u64
)unmap_start
)
2735 unmap_mapping_range(mapping
, unmap_start
,
2736 1 + unmap_end
- unmap_start
, 0);
2737 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2738 /* No need to unmap again: hole-punching leaves COWed pages */
2740 spin_lock(&inode
->i_lock
);
2741 inode
->i_private
= NULL
;
2742 wake_up_all(&shmem_falloc_waitq
);
2743 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2744 spin_unlock(&inode
->i_lock
);
2749 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2750 error
= inode_newsize_ok(inode
, offset
+ len
);
2754 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2759 start
= offset
>> PAGE_SHIFT
;
2760 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2761 /* Try to avoid a swapstorm if len is impossible to satisfy */
2762 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2767 shmem_falloc
.waitq
= NULL
;
2768 shmem_falloc
.start
= start
;
2769 shmem_falloc
.next
= start
;
2770 shmem_falloc
.nr_falloced
= 0;
2771 shmem_falloc
.nr_unswapped
= 0;
2772 spin_lock(&inode
->i_lock
);
2773 inode
->i_private
= &shmem_falloc
;
2774 spin_unlock(&inode
->i_lock
);
2777 * info->fallocend is only relevant when huge pages might be
2778 * involved: to prevent split_huge_page() freeing fallocated
2779 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2781 undo_fallocend
= info
->fallocend
;
2782 if (info
->fallocend
< end
)
2783 info
->fallocend
= end
;
2785 for (index
= start
; index
< end
; ) {
2789 * Good, the fallocate(2) manpage permits EINTR: we may have
2790 * been interrupted because we are using up too much memory.
2792 if (signal_pending(current
))
2794 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2797 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2799 info
->fallocend
= undo_fallocend
;
2800 /* Remove the !PageUptodate pages we added */
2801 if (index
> start
) {
2802 shmem_undo_range(inode
,
2803 (loff_t
)start
<< PAGE_SHIFT
,
2804 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2811 * Here is a more important optimization than it appears:
2812 * a second SGP_FALLOC on the same huge page will clear it,
2813 * making it PageUptodate and un-undoable if we fail later.
2815 if (PageTransCompound(page
)) {
2816 index
= round_up(index
, HPAGE_PMD_NR
);
2817 /* Beware 32-bit wraparound */
2823 * Inform shmem_writepage() how far we have reached.
2824 * No need for lock or barrier: we have the page lock.
2826 if (!PageUptodate(page
))
2827 shmem_falloc
.nr_falloced
+= index
- shmem_falloc
.next
;
2828 shmem_falloc
.next
= index
;
2831 * If !PageUptodate, leave it that way so that freeable pages
2832 * can be recognized if we need to rollback on error later.
2833 * But set_page_dirty so that memory pressure will swap rather
2834 * than free the pages we are allocating (and SGP_CACHE pages
2835 * might still be clean: we now need to mark those dirty too).
2837 set_page_dirty(page
);
2843 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2844 i_size_write(inode
, offset
+ len
);
2846 spin_lock(&inode
->i_lock
);
2847 inode
->i_private
= NULL
;
2848 spin_unlock(&inode
->i_lock
);
2851 file_modified(file
);
2852 inode_unlock(inode
);
2856 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2858 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2860 buf
->f_type
= TMPFS_MAGIC
;
2861 buf
->f_bsize
= PAGE_SIZE
;
2862 buf
->f_namelen
= NAME_MAX
;
2863 if (sbinfo
->max_blocks
) {
2864 buf
->f_blocks
= sbinfo
->max_blocks
;
2866 buf
->f_bfree
= sbinfo
->max_blocks
-
2867 percpu_counter_sum(&sbinfo
->used_blocks
);
2869 if (sbinfo
->max_inodes
) {
2870 buf
->f_files
= sbinfo
->max_inodes
;
2871 buf
->f_ffree
= sbinfo
->free_inodes
;
2873 /* else leave those fields 0 like simple_statfs */
2875 buf
->f_fsid
= uuid_to_fsid(dentry
->d_sb
->s_uuid
.b
);
2881 * File creation. Allocate an inode, and we're done..
2884 shmem_mknod(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2885 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2887 struct inode
*inode
;
2888 int error
= -ENOSPC
;
2890 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2892 error
= simple_acl_create(dir
, inode
);
2895 error
= security_inode_init_security(inode
, dir
,
2897 shmem_initxattrs
, NULL
);
2898 if (error
&& error
!= -EOPNOTSUPP
)
2902 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2903 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2904 d_instantiate(dentry
, inode
);
2905 dget(dentry
); /* Extra count - pin the dentry in core */
2914 shmem_tmpfile(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2915 struct dentry
*dentry
, umode_t mode
)
2917 struct inode
*inode
;
2918 int error
= -ENOSPC
;
2920 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2922 error
= security_inode_init_security(inode
, dir
,
2924 shmem_initxattrs
, NULL
);
2925 if (error
&& error
!= -EOPNOTSUPP
)
2927 error
= simple_acl_create(dir
, inode
);
2930 d_tmpfile(dentry
, inode
);
2938 static int shmem_mkdir(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2939 struct dentry
*dentry
, umode_t mode
)
2943 if ((error
= shmem_mknod(&init_user_ns
, dir
, dentry
,
2944 mode
| S_IFDIR
, 0)))
2950 static int shmem_create(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2951 struct dentry
*dentry
, umode_t mode
, bool excl
)
2953 return shmem_mknod(&init_user_ns
, dir
, dentry
, mode
| S_IFREG
, 0);
2959 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2961 struct inode
*inode
= d_inode(old_dentry
);
2965 * No ordinary (disk based) filesystem counts links as inodes;
2966 * but each new link needs a new dentry, pinning lowmem, and
2967 * tmpfs dentries cannot be pruned until they are unlinked.
2968 * But if an O_TMPFILE file is linked into the tmpfs, the
2969 * first link must skip that, to get the accounting right.
2971 if (inode
->i_nlink
) {
2972 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
2977 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2978 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2980 ihold(inode
); /* New dentry reference */
2981 dget(dentry
); /* Extra pinning count for the created dentry */
2982 d_instantiate(dentry
, inode
);
2987 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2989 struct inode
*inode
= d_inode(dentry
);
2991 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2992 shmem_free_inode(inode
->i_sb
);
2994 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2995 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2997 dput(dentry
); /* Undo the count from "create" - this does all the work */
3001 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3003 if (!simple_empty(dentry
))
3006 drop_nlink(d_inode(dentry
));
3008 return shmem_unlink(dir
, dentry
);
3011 static int shmem_whiteout(struct user_namespace
*mnt_userns
,
3012 struct inode
*old_dir
, struct dentry
*old_dentry
)
3014 struct dentry
*whiteout
;
3017 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3021 error
= shmem_mknod(&init_user_ns
, old_dir
, whiteout
,
3022 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3028 * Cheat and hash the whiteout while the old dentry is still in
3029 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3031 * d_lookup() will consistently find one of them at this point,
3032 * not sure which one, but that isn't even important.
3039 * The VFS layer already does all the dentry stuff for rename,
3040 * we just have to decrement the usage count for the target if
3041 * it exists so that the VFS layer correctly free's it when it
3044 static int shmem_rename2(struct user_namespace
*mnt_userns
,
3045 struct inode
*old_dir
, struct dentry
*old_dentry
,
3046 struct inode
*new_dir
, struct dentry
*new_dentry
,
3049 struct inode
*inode
= d_inode(old_dentry
);
3050 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3052 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3055 if (flags
& RENAME_EXCHANGE
)
3056 return simple_rename_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3058 if (!simple_empty(new_dentry
))
3061 if (flags
& RENAME_WHITEOUT
) {
3064 error
= shmem_whiteout(&init_user_ns
, old_dir
, old_dentry
);
3069 if (d_really_is_positive(new_dentry
)) {
3070 (void) shmem_unlink(new_dir
, new_dentry
);
3071 if (they_are_dirs
) {
3072 drop_nlink(d_inode(new_dentry
));
3073 drop_nlink(old_dir
);
3075 } else if (they_are_dirs
) {
3076 drop_nlink(old_dir
);
3080 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3081 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3082 old_dir
->i_ctime
= old_dir
->i_mtime
=
3083 new_dir
->i_ctime
= new_dir
->i_mtime
=
3084 inode
->i_ctime
= current_time(old_dir
);
3088 static int shmem_symlink(struct user_namespace
*mnt_userns
, struct inode
*dir
,
3089 struct dentry
*dentry
, const char *symname
)
3093 struct inode
*inode
;
3096 len
= strlen(symname
) + 1;
3097 if (len
> PAGE_SIZE
)
3098 return -ENAMETOOLONG
;
3100 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3105 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3106 shmem_initxattrs
, NULL
);
3107 if (error
&& error
!= -EOPNOTSUPP
) {
3112 inode
->i_size
= len
-1;
3113 if (len
<= SHORT_SYMLINK_LEN
) {
3114 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3115 if (!inode
->i_link
) {
3119 inode
->i_op
= &shmem_short_symlink_operations
;
3121 inode_nohighmem(inode
);
3122 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3127 inode
->i_mapping
->a_ops
= &shmem_aops
;
3128 inode
->i_op
= &shmem_symlink_inode_operations
;
3129 memcpy(page_address(page
), symname
, len
);
3130 SetPageUptodate(page
);
3131 set_page_dirty(page
);
3135 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3136 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3137 d_instantiate(dentry
, inode
);
3142 static void shmem_put_link(void *arg
)
3144 mark_page_accessed(arg
);
3148 static const char *shmem_get_link(struct dentry
*dentry
,
3149 struct inode
*inode
,
3150 struct delayed_call
*done
)
3152 struct page
*page
= NULL
;
3155 page
= find_get_page(inode
->i_mapping
, 0);
3157 return ERR_PTR(-ECHILD
);
3158 if (PageHWPoison(page
) ||
3159 !PageUptodate(page
)) {
3161 return ERR_PTR(-ECHILD
);
3164 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3166 return ERR_PTR(error
);
3168 return ERR_PTR(-ECHILD
);
3169 if (PageHWPoison(page
)) {
3172 return ERR_PTR(-ECHILD
);
3176 set_delayed_call(done
, shmem_put_link
, page
);
3177 return page_address(page
);
3180 #ifdef CONFIG_TMPFS_XATTR
3182 static int shmem_fileattr_get(struct dentry
*dentry
, struct fileattr
*fa
)
3184 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3186 fileattr_fill_flags(fa
, info
->fsflags
& SHMEM_FL_USER_VISIBLE
);
3191 static int shmem_fileattr_set(struct user_namespace
*mnt_userns
,
3192 struct dentry
*dentry
, struct fileattr
*fa
)
3194 struct inode
*inode
= d_inode(dentry
);
3195 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3197 if (fileattr_has_fsx(fa
))
3199 if (fa
->flags
& ~SHMEM_FL_USER_MODIFIABLE
)
3202 info
->fsflags
= (info
->fsflags
& ~SHMEM_FL_USER_MODIFIABLE
) |
3203 (fa
->flags
& SHMEM_FL_USER_MODIFIABLE
);
3205 shmem_set_inode_flags(inode
, info
->fsflags
);
3206 inode
->i_ctime
= current_time(inode
);
3211 * Superblocks without xattr inode operations may get some security.* xattr
3212 * support from the LSM "for free". As soon as we have any other xattrs
3213 * like ACLs, we also need to implement the security.* handlers at
3214 * filesystem level, though.
3218 * Callback for security_inode_init_security() for acquiring xattrs.
3220 static int shmem_initxattrs(struct inode
*inode
,
3221 const struct xattr
*xattr_array
,
3224 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3225 const struct xattr
*xattr
;
3226 struct simple_xattr
*new_xattr
;
3229 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3230 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3234 len
= strlen(xattr
->name
) + 1;
3235 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3237 if (!new_xattr
->name
) {
3242 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3243 XATTR_SECURITY_PREFIX_LEN
);
3244 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3247 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3253 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3254 struct dentry
*unused
, struct inode
*inode
,
3255 const char *name
, void *buffer
, size_t size
)
3257 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3259 name
= xattr_full_name(handler
, name
);
3260 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3263 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3264 struct user_namespace
*mnt_userns
,
3265 struct dentry
*unused
, struct inode
*inode
,
3266 const char *name
, const void *value
,
3267 size_t size
, int flags
)
3269 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3271 name
= xattr_full_name(handler
, name
);
3272 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3275 static const struct xattr_handler shmem_security_xattr_handler
= {
3276 .prefix
= XATTR_SECURITY_PREFIX
,
3277 .get
= shmem_xattr_handler_get
,
3278 .set
= shmem_xattr_handler_set
,
3281 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3282 .prefix
= XATTR_TRUSTED_PREFIX
,
3283 .get
= shmem_xattr_handler_get
,
3284 .set
= shmem_xattr_handler_set
,
3287 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3288 #ifdef CONFIG_TMPFS_POSIX_ACL
3289 &posix_acl_access_xattr_handler
,
3290 &posix_acl_default_xattr_handler
,
3292 &shmem_security_xattr_handler
,
3293 &shmem_trusted_xattr_handler
,
3297 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3299 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3300 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3302 #endif /* CONFIG_TMPFS_XATTR */
3304 static const struct inode_operations shmem_short_symlink_operations
= {
3305 .getattr
= shmem_getattr
,
3306 .get_link
= simple_get_link
,
3307 #ifdef CONFIG_TMPFS_XATTR
3308 .listxattr
= shmem_listxattr
,
3312 static const struct inode_operations shmem_symlink_inode_operations
= {
3313 .getattr
= shmem_getattr
,
3314 .get_link
= shmem_get_link
,
3315 #ifdef CONFIG_TMPFS_XATTR
3316 .listxattr
= shmem_listxattr
,
3320 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3322 return ERR_PTR(-ESTALE
);
3325 static int shmem_match(struct inode
*ino
, void *vfh
)
3329 inum
= (inum
<< 32) | fh
[1];
3330 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3333 /* Find any alias of inode, but prefer a hashed alias */
3334 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3336 struct dentry
*alias
= d_find_alias(inode
);
3338 return alias
?: d_find_any_alias(inode
);
3342 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3343 struct fid
*fid
, int fh_len
, int fh_type
)
3345 struct inode
*inode
;
3346 struct dentry
*dentry
= NULL
;
3353 inum
= (inum
<< 32) | fid
->raw
[1];
3355 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3356 shmem_match
, fid
->raw
);
3358 dentry
= shmem_find_alias(inode
);
3365 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3366 struct inode
*parent
)
3370 return FILEID_INVALID
;
3373 if (inode_unhashed(inode
)) {
3374 /* Unfortunately insert_inode_hash is not idempotent,
3375 * so as we hash inodes here rather than at creation
3376 * time, we need a lock to ensure we only try
3379 static DEFINE_SPINLOCK(lock
);
3381 if (inode_unhashed(inode
))
3382 __insert_inode_hash(inode
,
3383 inode
->i_ino
+ inode
->i_generation
);
3387 fh
[0] = inode
->i_generation
;
3388 fh
[1] = inode
->i_ino
;
3389 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3395 static const struct export_operations shmem_export_ops
= {
3396 .get_parent
= shmem_get_parent
,
3397 .encode_fh
= shmem_encode_fh
,
3398 .fh_to_dentry
= shmem_fh_to_dentry
,
3414 static const struct constant_table shmem_param_enums_huge
[] = {
3415 {"never", SHMEM_HUGE_NEVER
},
3416 {"always", SHMEM_HUGE_ALWAYS
},
3417 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3418 {"advise", SHMEM_HUGE_ADVISE
},
3422 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3423 fsparam_u32 ("gid", Opt_gid
),
3424 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3425 fsparam_u32oct("mode", Opt_mode
),
3426 fsparam_string("mpol", Opt_mpol
),
3427 fsparam_string("nr_blocks", Opt_nr_blocks
),
3428 fsparam_string("nr_inodes", Opt_nr_inodes
),
3429 fsparam_string("size", Opt_size
),
3430 fsparam_u32 ("uid", Opt_uid
),
3431 fsparam_flag ("inode32", Opt_inode32
),
3432 fsparam_flag ("inode64", Opt_inode64
),
3436 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3438 struct shmem_options
*ctx
= fc
->fs_private
;
3439 struct fs_parse_result result
;
3440 unsigned long long size
;
3444 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3450 size
= memparse(param
->string
, &rest
);
3452 size
<<= PAGE_SHIFT
;
3453 size
*= totalram_pages();
3459 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3460 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3463 ctx
->blocks
= memparse(param
->string
, &rest
);
3464 if (*rest
|| ctx
->blocks
> S64_MAX
)
3466 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3469 ctx
->inodes
= memparse(param
->string
, &rest
);
3472 ctx
->seen
|= SHMEM_SEEN_INODES
;
3475 ctx
->mode
= result
.uint_32
& 07777;
3478 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3479 if (!uid_valid(ctx
->uid
))
3483 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3484 if (!gid_valid(ctx
->gid
))
3488 ctx
->huge
= result
.uint_32
;
3489 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3490 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3491 has_transparent_hugepage()))
3492 goto unsupported_parameter
;
3493 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3496 if (IS_ENABLED(CONFIG_NUMA
)) {
3497 mpol_put(ctx
->mpol
);
3499 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3503 goto unsupported_parameter
;
3505 ctx
->full_inums
= false;
3506 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3509 if (sizeof(ino_t
) < 8) {
3511 "Cannot use inode64 with <64bit inums in kernel\n");
3513 ctx
->full_inums
= true;
3514 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3519 unsupported_parameter
:
3520 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3522 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3525 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3527 char *options
= data
;
3530 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3535 while (options
!= NULL
) {
3536 char *this_char
= options
;
3539 * NUL-terminate this option: unfortunately,
3540 * mount options form a comma-separated list,
3541 * but mpol's nodelist may also contain commas.
3543 options
= strchr(options
, ',');
3544 if (options
== NULL
)
3547 if (!isdigit(*options
)) {
3553 char *value
= strchr(this_char
, '=');
3559 len
= strlen(value
);
3561 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3570 * Reconfigure a shmem filesystem.
3572 * Note that we disallow change from limited->unlimited blocks/inodes while any
3573 * are in use; but we must separately disallow unlimited->limited, because in
3574 * that case we have no record of how much is already in use.
3576 static int shmem_reconfigure(struct fs_context
*fc
)
3578 struct shmem_options
*ctx
= fc
->fs_private
;
3579 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3580 unsigned long inodes
;
3581 struct mempolicy
*mpol
= NULL
;
3584 raw_spin_lock(&sbinfo
->stat_lock
);
3585 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3587 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3588 if (!sbinfo
->max_blocks
) {
3589 err
= "Cannot retroactively limit size";
3592 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3594 err
= "Too small a size for current use";
3598 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3599 if (!sbinfo
->max_inodes
) {
3600 err
= "Cannot retroactively limit inodes";
3603 if (ctx
->inodes
< inodes
) {
3604 err
= "Too few inodes for current use";
3609 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
3610 sbinfo
->next_ino
> UINT_MAX
) {
3611 err
= "Current inum too high to switch to 32-bit inums";
3615 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3616 sbinfo
->huge
= ctx
->huge
;
3617 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
3618 sbinfo
->full_inums
= ctx
->full_inums
;
3619 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3620 sbinfo
->max_blocks
= ctx
->blocks
;
3621 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3622 sbinfo
->max_inodes
= ctx
->inodes
;
3623 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3627 * Preserve previous mempolicy unless mpol remount option was specified.
3630 mpol
= sbinfo
->mpol
;
3631 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3634 raw_spin_unlock(&sbinfo
->stat_lock
);
3638 raw_spin_unlock(&sbinfo
->stat_lock
);
3639 return invalfc(fc
, "%s", err
);
3642 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3644 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3646 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3647 seq_printf(seq
, ",size=%luk",
3648 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3649 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3650 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3651 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3652 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3653 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3654 seq_printf(seq
, ",uid=%u",
3655 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3656 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3657 seq_printf(seq
, ",gid=%u",
3658 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3661 * Showing inode{64,32} might be useful even if it's the system default,
3662 * since then people don't have to resort to checking both here and
3663 * /proc/config.gz to confirm 64-bit inums were successfully applied
3664 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3666 * We hide it when inode64 isn't the default and we are using 32-bit
3667 * inodes, since that probably just means the feature isn't even under
3672 * +-----------------+-----------------+
3673 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3674 * +------------------+-----------------+-----------------+
3675 * | full_inums=true | show | show |
3676 * | full_inums=false | show | hide |
3677 * +------------------+-----------------+-----------------+
3680 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
3681 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
3682 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3683 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3685 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3687 shmem_show_mpol(seq
, sbinfo
->mpol
);
3691 #endif /* CONFIG_TMPFS */
3693 static void shmem_put_super(struct super_block
*sb
)
3695 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3697 free_percpu(sbinfo
->ino_batch
);
3698 percpu_counter_destroy(&sbinfo
->used_blocks
);
3699 mpol_put(sbinfo
->mpol
);
3701 sb
->s_fs_info
= NULL
;
3704 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3706 struct shmem_options
*ctx
= fc
->fs_private
;
3707 struct inode
*inode
;
3708 struct shmem_sb_info
*sbinfo
;
3710 /* Round up to L1_CACHE_BYTES to resist false sharing */
3711 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3712 L1_CACHE_BYTES
), GFP_KERNEL
);
3716 sb
->s_fs_info
= sbinfo
;
3720 * Per default we only allow half of the physical ram per
3721 * tmpfs instance, limiting inodes to one per page of lowmem;
3722 * but the internal instance is left unlimited.
3724 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3725 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3726 ctx
->blocks
= shmem_default_max_blocks();
3727 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3728 ctx
->inodes
= shmem_default_max_inodes();
3729 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
3730 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
3732 sb
->s_flags
|= SB_NOUSER
;
3734 sb
->s_export_op
= &shmem_export_ops
;
3735 sb
->s_flags
|= SB_NOSEC
;
3737 sb
->s_flags
|= SB_NOUSER
;
3739 sbinfo
->max_blocks
= ctx
->blocks
;
3740 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3741 if (sb
->s_flags
& SB_KERNMOUNT
) {
3742 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
3743 if (!sbinfo
->ino_batch
)
3746 sbinfo
->uid
= ctx
->uid
;
3747 sbinfo
->gid
= ctx
->gid
;
3748 sbinfo
->full_inums
= ctx
->full_inums
;
3749 sbinfo
->mode
= ctx
->mode
;
3750 sbinfo
->huge
= ctx
->huge
;
3751 sbinfo
->mpol
= ctx
->mpol
;
3754 raw_spin_lock_init(&sbinfo
->stat_lock
);
3755 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3757 spin_lock_init(&sbinfo
->shrinklist_lock
);
3758 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3760 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3761 sb
->s_blocksize
= PAGE_SIZE
;
3762 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3763 sb
->s_magic
= TMPFS_MAGIC
;
3764 sb
->s_op
= &shmem_ops
;
3765 sb
->s_time_gran
= 1;
3766 #ifdef CONFIG_TMPFS_XATTR
3767 sb
->s_xattr
= shmem_xattr_handlers
;
3769 #ifdef CONFIG_TMPFS_POSIX_ACL
3770 sb
->s_flags
|= SB_POSIXACL
;
3772 uuid_gen(&sb
->s_uuid
);
3774 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3777 inode
->i_uid
= sbinfo
->uid
;
3778 inode
->i_gid
= sbinfo
->gid
;
3779 sb
->s_root
= d_make_root(inode
);
3785 shmem_put_super(sb
);
3789 static int shmem_get_tree(struct fs_context
*fc
)
3791 return get_tree_nodev(fc
, shmem_fill_super
);
3794 static void shmem_free_fc(struct fs_context
*fc
)
3796 struct shmem_options
*ctx
= fc
->fs_private
;
3799 mpol_put(ctx
->mpol
);
3804 static const struct fs_context_operations shmem_fs_context_ops
= {
3805 .free
= shmem_free_fc
,
3806 .get_tree
= shmem_get_tree
,
3808 .parse_monolithic
= shmem_parse_options
,
3809 .parse_param
= shmem_parse_one
,
3810 .reconfigure
= shmem_reconfigure
,
3814 static struct kmem_cache
*shmem_inode_cachep
;
3816 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3818 struct shmem_inode_info
*info
;
3819 info
= alloc_inode_sb(sb
, shmem_inode_cachep
, GFP_KERNEL
);
3822 return &info
->vfs_inode
;
3825 static void shmem_free_in_core_inode(struct inode
*inode
)
3827 if (S_ISLNK(inode
->i_mode
))
3828 kfree(inode
->i_link
);
3829 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3832 static void shmem_destroy_inode(struct inode
*inode
)
3834 if (S_ISREG(inode
->i_mode
))
3835 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3838 static void shmem_init_inode(void *foo
)
3840 struct shmem_inode_info
*info
= foo
;
3841 inode_init_once(&info
->vfs_inode
);
3844 static void shmem_init_inodecache(void)
3846 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3847 sizeof(struct shmem_inode_info
),
3848 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3851 static void shmem_destroy_inodecache(void)
3853 kmem_cache_destroy(shmem_inode_cachep
);
3856 /* Keep the page in page cache instead of truncating it */
3857 static int shmem_error_remove_page(struct address_space
*mapping
,
3863 const struct address_space_operations shmem_aops
= {
3864 .writepage
= shmem_writepage
,
3865 .dirty_folio
= noop_dirty_folio
,
3867 .write_begin
= shmem_write_begin
,
3868 .write_end
= shmem_write_end
,
3870 #ifdef CONFIG_MIGRATION
3871 .migrate_folio
= migrate_folio
,
3873 .error_remove_page
= shmem_error_remove_page
,
3875 EXPORT_SYMBOL(shmem_aops
);
3877 static const struct file_operations shmem_file_operations
= {
3879 .get_unmapped_area
= shmem_get_unmapped_area
,
3881 .llseek
= shmem_file_llseek
,
3882 .read_iter
= shmem_file_read_iter
,
3883 .write_iter
= generic_file_write_iter
,
3884 .fsync
= noop_fsync
,
3885 .splice_read
= generic_file_splice_read
,
3886 .splice_write
= iter_file_splice_write
,
3887 .fallocate
= shmem_fallocate
,
3891 static const struct inode_operations shmem_inode_operations
= {
3892 .getattr
= shmem_getattr
,
3893 .setattr
= shmem_setattr
,
3894 #ifdef CONFIG_TMPFS_XATTR
3895 .listxattr
= shmem_listxattr
,
3896 .set_acl
= simple_set_acl
,
3897 .fileattr_get
= shmem_fileattr_get
,
3898 .fileattr_set
= shmem_fileattr_set
,
3902 static const struct inode_operations shmem_dir_inode_operations
= {
3904 .getattr
= shmem_getattr
,
3905 .create
= shmem_create
,
3906 .lookup
= simple_lookup
,
3908 .unlink
= shmem_unlink
,
3909 .symlink
= shmem_symlink
,
3910 .mkdir
= shmem_mkdir
,
3911 .rmdir
= shmem_rmdir
,
3912 .mknod
= shmem_mknod
,
3913 .rename
= shmem_rename2
,
3914 .tmpfile
= shmem_tmpfile
,
3916 #ifdef CONFIG_TMPFS_XATTR
3917 .listxattr
= shmem_listxattr
,
3918 .fileattr_get
= shmem_fileattr_get
,
3919 .fileattr_set
= shmem_fileattr_set
,
3921 #ifdef CONFIG_TMPFS_POSIX_ACL
3922 .setattr
= shmem_setattr
,
3923 .set_acl
= simple_set_acl
,
3927 static const struct inode_operations shmem_special_inode_operations
= {
3928 .getattr
= shmem_getattr
,
3929 #ifdef CONFIG_TMPFS_XATTR
3930 .listxattr
= shmem_listxattr
,
3932 #ifdef CONFIG_TMPFS_POSIX_ACL
3933 .setattr
= shmem_setattr
,
3934 .set_acl
= simple_set_acl
,
3938 static const struct super_operations shmem_ops
= {
3939 .alloc_inode
= shmem_alloc_inode
,
3940 .free_inode
= shmem_free_in_core_inode
,
3941 .destroy_inode
= shmem_destroy_inode
,
3943 .statfs
= shmem_statfs
,
3944 .show_options
= shmem_show_options
,
3946 .evict_inode
= shmem_evict_inode
,
3947 .drop_inode
= generic_delete_inode
,
3948 .put_super
= shmem_put_super
,
3949 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3950 .nr_cached_objects
= shmem_unused_huge_count
,
3951 .free_cached_objects
= shmem_unused_huge_scan
,
3955 static const struct vm_operations_struct shmem_vm_ops
= {
3956 .fault
= shmem_fault
,
3957 .map_pages
= filemap_map_pages
,
3959 .set_policy
= shmem_set_policy
,
3960 .get_policy
= shmem_get_policy
,
3964 int shmem_init_fs_context(struct fs_context
*fc
)
3966 struct shmem_options
*ctx
;
3968 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3972 ctx
->mode
= 0777 | S_ISVTX
;
3973 ctx
->uid
= current_fsuid();
3974 ctx
->gid
= current_fsgid();
3976 fc
->fs_private
= ctx
;
3977 fc
->ops
= &shmem_fs_context_ops
;
3981 static struct file_system_type shmem_fs_type
= {
3982 .owner
= THIS_MODULE
,
3984 .init_fs_context
= shmem_init_fs_context
,
3986 .parameters
= shmem_fs_parameters
,
3988 .kill_sb
= kill_litter_super
,
3989 .fs_flags
= FS_USERNS_MOUNT
,
3992 void __init
shmem_init(void)
3996 shmem_init_inodecache();
3998 error
= register_filesystem(&shmem_fs_type
);
4000 pr_err("Could not register tmpfs\n");
4004 shm_mnt
= kern_mount(&shmem_fs_type
);
4005 if (IS_ERR(shm_mnt
)) {
4006 error
= PTR_ERR(shm_mnt
);
4007 pr_err("Could not kern_mount tmpfs\n");
4011 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4012 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
4013 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4015 shmem_huge
= SHMEM_HUGE_NEVER
; /* just in case it was patched */
4020 unregister_filesystem(&shmem_fs_type
);
4022 shmem_destroy_inodecache();
4023 shm_mnt
= ERR_PTR(error
);
4026 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4027 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4028 struct kobj_attribute
*attr
, char *buf
)
4030 static const int values
[] = {
4032 SHMEM_HUGE_WITHIN_SIZE
,
4041 for (i
= 0; i
< ARRAY_SIZE(values
); i
++) {
4042 len
+= sysfs_emit_at(buf
, len
,
4043 shmem_huge
== values
[i
] ? "%s[%s]" : "%s%s",
4045 shmem_format_huge(values
[i
]));
4048 len
+= sysfs_emit_at(buf
, len
, "\n");
4053 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4054 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4059 if (count
+ 1 > sizeof(tmp
))
4061 memcpy(tmp
, buf
, count
);
4063 if (count
&& tmp
[count
- 1] == '\n')
4064 tmp
[count
- 1] = '\0';
4066 huge
= shmem_parse_huge(tmp
);
4067 if (huge
== -EINVAL
)
4069 if (!has_transparent_hugepage() &&
4070 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4074 if (shmem_huge
> SHMEM_HUGE_DENY
)
4075 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4079 struct kobj_attribute shmem_enabled_attr
= __ATTR_RW(shmem_enabled
);
4080 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4082 #else /* !CONFIG_SHMEM */
4085 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4087 * This is intended for small system where the benefits of the full
4088 * shmem code (swap-backed and resource-limited) are outweighed by
4089 * their complexity. On systems without swap this code should be
4090 * effectively equivalent, but much lighter weight.
4093 static struct file_system_type shmem_fs_type
= {
4095 .init_fs_context
= ramfs_init_fs_context
,
4096 .parameters
= ramfs_fs_parameters
,
4097 .kill_sb
= kill_litter_super
,
4098 .fs_flags
= FS_USERNS_MOUNT
,
4101 void __init
shmem_init(void)
4103 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4105 shm_mnt
= kern_mount(&shmem_fs_type
);
4106 BUG_ON(IS_ERR(shm_mnt
));
4109 int shmem_unuse(unsigned int type
)
4114 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
4119 void shmem_unlock_mapping(struct address_space
*mapping
)
4124 unsigned long shmem_get_unmapped_area(struct file
*file
,
4125 unsigned long addr
, unsigned long len
,
4126 unsigned long pgoff
, unsigned long flags
)
4128 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4132 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4134 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4136 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4138 #define shmem_vm_ops generic_file_vm_ops
4139 #define shmem_file_operations ramfs_file_operations
4140 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4141 #define shmem_acct_size(flags, size) 0
4142 #define shmem_unacct_size(flags, size) do {} while (0)
4144 #endif /* CONFIG_SHMEM */
4148 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4149 unsigned long flags
, unsigned int i_flags
)
4151 struct inode
*inode
;
4155 return ERR_CAST(mnt
);
4157 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4158 return ERR_PTR(-EINVAL
);
4160 if (shmem_acct_size(flags
, size
))
4161 return ERR_PTR(-ENOMEM
);
4163 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4165 if (unlikely(!inode
)) {
4166 shmem_unacct_size(flags
, size
);
4167 return ERR_PTR(-ENOSPC
);
4169 inode
->i_flags
|= i_flags
;
4170 inode
->i_size
= size
;
4171 clear_nlink(inode
); /* It is unlinked */
4172 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4174 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4175 &shmem_file_operations
);
4182 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4183 * kernel internal. There will be NO LSM permission checks against the
4184 * underlying inode. So users of this interface must do LSM checks at a
4185 * higher layer. The users are the big_key and shm implementations. LSM
4186 * checks are provided at the key or shm level rather than the inode.
4187 * @name: name for dentry (to be seen in /proc/<pid>/maps
4188 * @size: size to be set for the file
4189 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4191 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4193 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4197 * shmem_file_setup - get an unlinked file living in tmpfs
4198 * @name: name for dentry (to be seen in /proc/<pid>/maps
4199 * @size: size to be set for the file
4200 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4202 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4204 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4206 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4209 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4210 * @mnt: the tmpfs mount where the file will be created
4211 * @name: name for dentry (to be seen in /proc/<pid>/maps
4212 * @size: size to be set for the file
4213 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4215 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4216 loff_t size
, unsigned long flags
)
4218 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4220 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4223 * shmem_zero_setup - setup a shared anonymous mapping
4224 * @vma: the vma to be mmapped is prepared by do_mmap
4226 int shmem_zero_setup(struct vm_area_struct
*vma
)
4229 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4232 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4233 * between XFS directory reading and selinux: since this file is only
4234 * accessible to the user through its mapping, use S_PRIVATE flag to
4235 * bypass file security, in the same way as shmem_kernel_file_setup().
4237 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4239 return PTR_ERR(file
);
4243 vma
->vm_file
= file
;
4244 vma
->vm_ops
= &shmem_vm_ops
;
4250 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4251 * @mapping: the page's address_space
4252 * @index: the page index
4253 * @gfp: the page allocator flags to use if allocating
4255 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4256 * with any new page allocations done using the specified allocation flags.
4257 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4258 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4259 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4261 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4262 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4264 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4265 pgoff_t index
, gfp_t gfp
)
4268 struct inode
*inode
= mapping
->host
;
4272 BUG_ON(!shmem_mapping(mapping
));
4273 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4274 gfp
, NULL
, NULL
, NULL
);
4276 return ERR_PTR(error
);
4279 if (PageHWPoison(page
)) {
4281 return ERR_PTR(-EIO
);
4287 * The tiny !SHMEM case uses ramfs without swap
4289 return read_cache_page_gfp(mapping
, index
, gfp
);
4292 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);