2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
31 #include <linux/fileattr.h>
33 #include <linux/random.h>
34 #include <linux/sched/signal.h>
35 #include <linux/export.h>
36 #include <linux/shmem_fs.h>
37 #include <linux/swap.h>
38 #include <linux/uio.h>
39 #include <linux/hugetlb.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
42 #include <linux/iversion.h>
45 static struct vfsmount
*shm_mnt
;
49 * This virtual memory filesystem is heavily based on the ramfs. It
50 * extends ramfs by the ability to use swap and honor resource limits
51 * which makes it a completely usable filesystem.
54 #include <linux/xattr.h>
55 #include <linux/exportfs.h>
56 #include <linux/posix_acl.h>
57 #include <linux/posix_acl_xattr.h>
58 #include <linux/mman.h>
59 #include <linux/string.h>
60 #include <linux/slab.h>
61 #include <linux/backing-dev.h>
62 #include <linux/writeback.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
81 #include <linux/quotaops.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Pretend that one inode + its dentry occupy this much memory */
94 #define BOGO_INODE_SIZE 1024
96 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
97 #define SHORT_SYMLINK_LEN 128
100 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
101 * inode->i_private (with i_rwsem making sure that it has only one user at
102 * a time): we would prefer not to enlarge the shmem inode just for that.
104 struct shmem_falloc
{
105 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
106 pgoff_t start
; /* start of range currently being fallocated */
107 pgoff_t next
; /* the next page offset to be fallocated */
108 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
109 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
112 struct shmem_options
{
113 unsigned long long blocks
;
114 unsigned long long inodes
;
115 struct mempolicy
*mpol
;
123 unsigned short quota_types
;
124 struct shmem_quota_limits qlimits
;
125 #define SHMEM_SEEN_BLOCKS 1
126 #define SHMEM_SEEN_INODES 2
127 #define SHMEM_SEEN_HUGE 4
128 #define SHMEM_SEEN_INUMS 8
129 #define SHMEM_SEEN_NOSWAP 16
130 #define SHMEM_SEEN_QUOTA 32
134 static unsigned long shmem_default_max_blocks(void)
136 return totalram_pages() / 2;
139 static unsigned long shmem_default_max_inodes(void)
141 unsigned long nr_pages
= totalram_pages();
143 return min3(nr_pages
- totalhigh_pages(), nr_pages
/ 2,
144 ULONG_MAX
/ BOGO_INODE_SIZE
);
148 static int shmem_swapin_folio(struct inode
*inode
, pgoff_t index
,
149 struct folio
**foliop
, enum sgp_type sgp
,
150 gfp_t gfp
, struct vm_area_struct
*vma
,
151 vm_fault_t
*fault_type
);
153 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
155 return sb
->s_fs_info
;
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
164 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
166 return (flags
& VM_NORESERVE
) ?
167 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
170 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
172 if (!(flags
& VM_NORESERVE
))
173 vm_unacct_memory(VM_ACCT(size
));
176 static inline int shmem_reacct_size(unsigned long flags
,
177 loff_t oldsize
, loff_t newsize
)
179 if (!(flags
& VM_NORESERVE
)) {
180 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
181 return security_vm_enough_memory_mm(current
->mm
,
182 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
183 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
184 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
190 * ... whereas tmpfs objects are accounted incrementally as
191 * pages are allocated, in order to allow large sparse files.
192 * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
193 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
195 static inline int shmem_acct_block(unsigned long flags
, long pages
)
197 if (!(flags
& VM_NORESERVE
))
200 return security_vm_enough_memory_mm(current
->mm
,
201 pages
* VM_ACCT(PAGE_SIZE
));
204 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
206 if (flags
& VM_NORESERVE
)
207 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
210 static int shmem_inode_acct_block(struct inode
*inode
, long pages
)
212 struct shmem_inode_info
*info
= SHMEM_I(inode
);
213 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
216 if (shmem_acct_block(info
->flags
, pages
))
219 might_sleep(); /* when quotas */
220 if (sbinfo
->max_blocks
) {
221 if (percpu_counter_compare(&sbinfo
->used_blocks
,
222 sbinfo
->max_blocks
- pages
) > 0)
225 err
= dquot_alloc_block_nodirty(inode
, pages
);
229 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
231 err
= dquot_alloc_block_nodirty(inode
, pages
);
239 shmem_unacct_blocks(info
->flags
, pages
);
243 static void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
245 struct shmem_inode_info
*info
= SHMEM_I(inode
);
246 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
248 might_sleep(); /* when quotas */
249 dquot_free_block_nodirty(inode
, pages
);
251 if (sbinfo
->max_blocks
)
252 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
253 shmem_unacct_blocks(info
->flags
, pages
);
256 static const struct super_operations shmem_ops
;
257 const struct address_space_operations shmem_aops
;
258 static const struct file_operations shmem_file_operations
;
259 static const struct inode_operations shmem_inode_operations
;
260 static const struct inode_operations shmem_dir_inode_operations
;
261 static const struct inode_operations shmem_special_inode_operations
;
262 static const struct vm_operations_struct shmem_vm_ops
;
263 static const struct vm_operations_struct shmem_anon_vm_ops
;
264 static struct file_system_type shmem_fs_type
;
266 bool vma_is_anon_shmem(struct vm_area_struct
*vma
)
268 return vma
->vm_ops
== &shmem_anon_vm_ops
;
271 bool vma_is_shmem(struct vm_area_struct
*vma
)
273 return vma_is_anon_shmem(vma
) || vma
->vm_ops
== &shmem_vm_ops
;
276 static LIST_HEAD(shmem_swaplist
);
277 static DEFINE_MUTEX(shmem_swaplist_mutex
);
279 #ifdef CONFIG_TMPFS_QUOTA
281 static int shmem_enable_quotas(struct super_block
*sb
,
282 unsigned short quota_types
)
286 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
| DQUOT_NOLIST_DIRTY
;
287 for (type
= 0; type
< SHMEM_MAXQUOTAS
; type
++) {
288 if (!(quota_types
& (1 << type
)))
290 err
= dquot_load_quota_sb(sb
, type
, QFMT_SHMEM
,
291 DQUOT_USAGE_ENABLED
|
292 DQUOT_LIMITS_ENABLED
);
299 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
301 for (type
--; type
>= 0; type
--)
302 dquot_quota_off(sb
, type
);
306 static void shmem_disable_quotas(struct super_block
*sb
)
310 for (type
= 0; type
< SHMEM_MAXQUOTAS
; type
++)
311 dquot_quota_off(sb
, type
);
314 static struct dquot
**shmem_get_dquots(struct inode
*inode
)
316 return SHMEM_I(inode
)->i_dquot
;
318 #endif /* CONFIG_TMPFS_QUOTA */
321 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
322 * produces a novel ino for the newly allocated inode.
324 * It may also be called when making a hard link to permit the space needed by
325 * each dentry. However, in that case, no new inode number is needed since that
326 * internally draws from another pool of inode numbers (currently global
327 * get_next_ino()). This case is indicated by passing NULL as inop.
329 #define SHMEM_INO_BATCH 1024
330 static int shmem_reserve_inode(struct super_block
*sb
, ino_t
*inop
)
332 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
335 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
336 raw_spin_lock(&sbinfo
->stat_lock
);
337 if (sbinfo
->max_inodes
) {
338 if (sbinfo
->free_ispace
< BOGO_INODE_SIZE
) {
339 raw_spin_unlock(&sbinfo
->stat_lock
);
342 sbinfo
->free_ispace
-= BOGO_INODE_SIZE
;
345 ino
= sbinfo
->next_ino
++;
346 if (unlikely(is_zero_ino(ino
)))
347 ino
= sbinfo
->next_ino
++;
348 if (unlikely(!sbinfo
->full_inums
&&
351 * Emulate get_next_ino uint wraparound for
354 if (IS_ENABLED(CONFIG_64BIT
))
355 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
356 __func__
, MINOR(sb
->s_dev
));
357 sbinfo
->next_ino
= 1;
358 ino
= sbinfo
->next_ino
++;
362 raw_spin_unlock(&sbinfo
->stat_lock
);
365 * __shmem_file_setup, one of our callers, is lock-free: it
366 * doesn't hold stat_lock in shmem_reserve_inode since
367 * max_inodes is always 0, and is called from potentially
368 * unknown contexts. As such, use a per-cpu batched allocator
369 * which doesn't require the per-sb stat_lock unless we are at
370 * the batch boundary.
372 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
373 * shmem mounts are not exposed to userspace, so we don't need
374 * to worry about things like glibc compatibility.
378 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
380 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
381 raw_spin_lock(&sbinfo
->stat_lock
);
382 ino
= sbinfo
->next_ino
;
383 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
384 raw_spin_unlock(&sbinfo
->stat_lock
);
385 if (unlikely(is_zero_ino(ino
)))
396 static void shmem_free_inode(struct super_block
*sb
, size_t freed_ispace
)
398 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
399 if (sbinfo
->max_inodes
) {
400 raw_spin_lock(&sbinfo
->stat_lock
);
401 sbinfo
->free_ispace
+= BOGO_INODE_SIZE
+ freed_ispace
;
402 raw_spin_unlock(&sbinfo
->stat_lock
);
407 * shmem_recalc_inode - recalculate the block usage of an inode
408 * @inode: inode to recalc
409 * @alloced: the change in number of pages allocated to inode
410 * @swapped: the change in number of pages swapped from inode
412 * We have to calculate the free blocks since the mm can drop
413 * undirtied hole pages behind our back.
415 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
416 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
418 static void shmem_recalc_inode(struct inode
*inode
, long alloced
, long swapped
)
420 struct shmem_inode_info
*info
= SHMEM_I(inode
);
423 spin_lock(&info
->lock
);
424 info
->alloced
+= alloced
;
425 info
->swapped
+= swapped
;
426 freed
= info
->alloced
- info
->swapped
-
427 READ_ONCE(inode
->i_mapping
->nrpages
);
429 * Special case: whereas normally shmem_recalc_inode() is called
430 * after i_mapping->nrpages has already been adjusted (up or down),
431 * shmem_writepage() has to raise swapped before nrpages is lowered -
432 * to stop a racing shmem_recalc_inode() from thinking that a page has
433 * been freed. Compensate here, to avoid the need for a followup call.
438 info
->alloced
-= freed
;
439 spin_unlock(&info
->lock
);
441 /* The quota case may block */
443 shmem_inode_unacct_blocks(inode
, freed
);
446 bool shmem_charge(struct inode
*inode
, long pages
)
448 struct address_space
*mapping
= inode
->i_mapping
;
450 if (shmem_inode_acct_block(inode
, pages
))
453 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
454 xa_lock_irq(&mapping
->i_pages
);
455 mapping
->nrpages
+= pages
;
456 xa_unlock_irq(&mapping
->i_pages
);
458 shmem_recalc_inode(inode
, pages
, 0);
462 void shmem_uncharge(struct inode
*inode
, long pages
)
464 /* pages argument is currently unused: keep it to help debugging */
465 /* nrpages adjustment done by __filemap_remove_folio() or caller */
467 shmem_recalc_inode(inode
, 0, 0);
471 * Replace item expected in xarray by a new item, while holding xa_lock.
473 static int shmem_replace_entry(struct address_space
*mapping
,
474 pgoff_t index
, void *expected
, void *replacement
)
476 XA_STATE(xas
, &mapping
->i_pages
, index
);
479 VM_BUG_ON(!expected
);
480 VM_BUG_ON(!replacement
);
481 item
= xas_load(&xas
);
482 if (item
!= expected
)
484 xas_store(&xas
, replacement
);
489 * Sometimes, before we decide whether to proceed or to fail, we must check
490 * that an entry was not already brought back from swap by a racing thread.
492 * Checking page is not enough: by the time a SwapCache page is locked, it
493 * might be reused, and again be SwapCache, using the same swap as before.
495 static bool shmem_confirm_swap(struct address_space
*mapping
,
496 pgoff_t index
, swp_entry_t swap
)
498 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
502 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
505 * disables huge pages for the mount;
507 * enables huge pages for the mount;
508 * SHMEM_HUGE_WITHIN_SIZE:
509 * only allocate huge pages if the page will be fully within i_size,
510 * also respect fadvise()/madvise() hints;
512 * only allocate huge pages if requested with fadvise()/madvise();
515 #define SHMEM_HUGE_NEVER 0
516 #define SHMEM_HUGE_ALWAYS 1
517 #define SHMEM_HUGE_WITHIN_SIZE 2
518 #define SHMEM_HUGE_ADVISE 3
522 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
525 * disables huge on shm_mnt and all mounts, for emergency use;
527 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
530 #define SHMEM_HUGE_DENY (-1)
531 #define SHMEM_HUGE_FORCE (-2)
533 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
534 /* ifdef here to avoid bloating shmem.o when not necessary */
536 static int shmem_huge __read_mostly
= SHMEM_HUGE_NEVER
;
538 bool shmem_is_huge(struct inode
*inode
, pgoff_t index
, bool shmem_huge_force
,
539 struct mm_struct
*mm
, unsigned long vm_flags
)
543 if (!S_ISREG(inode
->i_mode
))
545 if (mm
&& ((vm_flags
& VM_NOHUGEPAGE
) || test_bit(MMF_DISABLE_THP
, &mm
->flags
)))
547 if (shmem_huge
== SHMEM_HUGE_DENY
)
549 if (shmem_huge_force
|| shmem_huge
== SHMEM_HUGE_FORCE
)
552 switch (SHMEM_SB(inode
->i_sb
)->huge
) {
553 case SHMEM_HUGE_ALWAYS
:
555 case SHMEM_HUGE_WITHIN_SIZE
:
556 index
= round_up(index
+ 1, HPAGE_PMD_NR
);
557 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
558 if (i_size
>> PAGE_SHIFT
>= index
)
561 case SHMEM_HUGE_ADVISE
:
562 if (mm
&& (vm_flags
& VM_HUGEPAGE
))
570 #if defined(CONFIG_SYSFS)
571 static int shmem_parse_huge(const char *str
)
573 if (!strcmp(str
, "never"))
574 return SHMEM_HUGE_NEVER
;
575 if (!strcmp(str
, "always"))
576 return SHMEM_HUGE_ALWAYS
;
577 if (!strcmp(str
, "within_size"))
578 return SHMEM_HUGE_WITHIN_SIZE
;
579 if (!strcmp(str
, "advise"))
580 return SHMEM_HUGE_ADVISE
;
581 if (!strcmp(str
, "deny"))
582 return SHMEM_HUGE_DENY
;
583 if (!strcmp(str
, "force"))
584 return SHMEM_HUGE_FORCE
;
589 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
590 static const char *shmem_format_huge(int huge
)
593 case SHMEM_HUGE_NEVER
:
595 case SHMEM_HUGE_ALWAYS
:
597 case SHMEM_HUGE_WITHIN_SIZE
:
598 return "within_size";
599 case SHMEM_HUGE_ADVISE
:
601 case SHMEM_HUGE_DENY
:
603 case SHMEM_HUGE_FORCE
:
612 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
613 struct shrink_control
*sc
, unsigned long nr_to_split
)
615 LIST_HEAD(list
), *pos
, *next
;
616 LIST_HEAD(to_remove
);
618 struct shmem_inode_info
*info
;
620 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
623 if (list_empty(&sbinfo
->shrinklist
))
626 spin_lock(&sbinfo
->shrinklist_lock
);
627 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
628 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
631 inode
= igrab(&info
->vfs_inode
);
633 /* inode is about to be evicted */
635 list_del_init(&info
->shrinklist
);
639 /* Check if there's anything to gain */
640 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
641 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
642 list_move(&info
->shrinklist
, &to_remove
);
646 list_move(&info
->shrinklist
, &list
);
648 sbinfo
->shrinklist_len
--;
652 spin_unlock(&sbinfo
->shrinklist_lock
);
654 list_for_each_safe(pos
, next
, &to_remove
) {
655 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
656 inode
= &info
->vfs_inode
;
657 list_del_init(&info
->shrinklist
);
661 list_for_each_safe(pos
, next
, &list
) {
665 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
666 inode
= &info
->vfs_inode
;
668 if (nr_to_split
&& split
>= nr_to_split
)
671 index
= (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
;
672 folio
= filemap_get_folio(inode
->i_mapping
, index
);
676 /* No huge page at the end of the file: nothing to split */
677 if (!folio_test_large(folio
)) {
683 * Move the inode on the list back to shrinklist if we failed
684 * to lock the page at this time.
686 * Waiting for the lock may lead to deadlock in the
689 if (!folio_trylock(folio
)) {
694 ret
= split_folio(folio
);
698 /* If split failed move the inode on the list back to shrinklist */
704 list_del_init(&info
->shrinklist
);
708 * Make sure the inode is either on the global list or deleted
709 * from any local list before iput() since it could be deleted
710 * in another thread once we put the inode (then the local list
713 spin_lock(&sbinfo
->shrinklist_lock
);
714 list_move(&info
->shrinklist
, &sbinfo
->shrinklist
);
715 sbinfo
->shrinklist_len
++;
716 spin_unlock(&sbinfo
->shrinklist_lock
);
724 static long shmem_unused_huge_scan(struct super_block
*sb
,
725 struct shrink_control
*sc
)
727 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
729 if (!READ_ONCE(sbinfo
->shrinklist_len
))
732 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
735 static long shmem_unused_huge_count(struct super_block
*sb
,
736 struct shrink_control
*sc
)
738 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
739 return READ_ONCE(sbinfo
->shrinklist_len
);
741 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
743 #define shmem_huge SHMEM_HUGE_DENY
745 bool shmem_is_huge(struct inode
*inode
, pgoff_t index
, bool shmem_huge_force
,
746 struct mm_struct
*mm
, unsigned long vm_flags
)
751 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
752 struct shrink_control
*sc
, unsigned long nr_to_split
)
756 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
759 * Like filemap_add_folio, but error if expected item has gone.
761 static int shmem_add_to_page_cache(struct folio
*folio
,
762 struct address_space
*mapping
,
763 pgoff_t index
, void *expected
, gfp_t gfp
,
764 struct mm_struct
*charge_mm
)
766 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, folio_order(folio
));
767 long nr
= folio_nr_pages(folio
);
770 VM_BUG_ON_FOLIO(index
!= round_down(index
, nr
), folio
);
771 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
772 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio
), folio
);
773 VM_BUG_ON(expected
&& folio_test_large(folio
));
775 folio_ref_add(folio
, nr
);
776 folio
->mapping
= mapping
;
777 folio
->index
= index
;
779 if (!folio_test_swapcache(folio
)) {
780 error
= mem_cgroup_charge(folio
, charge_mm
, gfp
);
782 if (folio_test_pmd_mappable(folio
)) {
783 count_vm_event(THP_FILE_FALLBACK
);
784 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
789 folio_throttle_swaprate(folio
, gfp
);
793 if (expected
!= xas_find_conflict(&xas
)) {
794 xas_set_err(&xas
, -EEXIST
);
797 if (expected
&& xas_find_conflict(&xas
)) {
798 xas_set_err(&xas
, -EEXIST
);
801 xas_store(&xas
, folio
);
804 if (folio_test_pmd_mappable(folio
)) {
805 count_vm_event(THP_FILE_ALLOC
);
806 __lruvec_stat_mod_folio(folio
, NR_SHMEM_THPS
, nr
);
808 mapping
->nrpages
+= nr
;
809 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, nr
);
810 __lruvec_stat_mod_folio(folio
, NR_SHMEM
, nr
);
812 xas_unlock_irq(&xas
);
813 } while (xas_nomem(&xas
, gfp
));
815 if (xas_error(&xas
)) {
816 error
= xas_error(&xas
);
822 folio
->mapping
= NULL
;
823 folio_ref_sub(folio
, nr
);
828 * Like delete_from_page_cache, but substitutes swap for @folio.
830 static void shmem_delete_from_page_cache(struct folio
*folio
, void *radswap
)
832 struct address_space
*mapping
= folio
->mapping
;
833 long nr
= folio_nr_pages(folio
);
836 xa_lock_irq(&mapping
->i_pages
);
837 error
= shmem_replace_entry(mapping
, folio
->index
, folio
, radswap
);
838 folio
->mapping
= NULL
;
839 mapping
->nrpages
-= nr
;
840 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, -nr
);
841 __lruvec_stat_mod_folio(folio
, NR_SHMEM
, -nr
);
842 xa_unlock_irq(&mapping
->i_pages
);
848 * Remove swap entry from page cache, free the swap and its page cache.
850 static int shmem_free_swap(struct address_space
*mapping
,
851 pgoff_t index
, void *radswap
)
855 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
858 free_swap_and_cache(radix_to_swp_entry(radswap
));
863 * Determine (in bytes) how many of the shmem object's pages mapped by the
864 * given offsets are swapped out.
866 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
867 * as long as the inode doesn't go away and racy results are not a problem.
869 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
870 pgoff_t start
, pgoff_t end
)
872 XA_STATE(xas
, &mapping
->i_pages
, start
);
874 unsigned long swapped
= 0;
875 unsigned long max
= end
- 1;
878 xas_for_each(&xas
, page
, max
) {
879 if (xas_retry(&xas
, page
))
881 if (xa_is_value(page
))
883 if (xas
.xa_index
== max
)
885 if (need_resched()) {
893 return swapped
<< PAGE_SHIFT
;
897 * Determine (in bytes) how many of the shmem object's pages mapped by the
898 * given vma is swapped out.
900 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
901 * as long as the inode doesn't go away and racy results are not a problem.
903 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
905 struct inode
*inode
= file_inode(vma
->vm_file
);
906 struct shmem_inode_info
*info
= SHMEM_I(inode
);
907 struct address_space
*mapping
= inode
->i_mapping
;
908 unsigned long swapped
;
910 /* Be careful as we don't hold info->lock */
911 swapped
= READ_ONCE(info
->swapped
);
914 * The easier cases are when the shmem object has nothing in swap, or
915 * the vma maps it whole. Then we can simply use the stats that we
921 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
922 return swapped
<< PAGE_SHIFT
;
924 /* Here comes the more involved part */
925 return shmem_partial_swap_usage(mapping
, vma
->vm_pgoff
,
926 vma
->vm_pgoff
+ vma_pages(vma
));
930 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
932 void shmem_unlock_mapping(struct address_space
*mapping
)
934 struct folio_batch fbatch
;
937 folio_batch_init(&fbatch
);
939 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
941 while (!mapping_unevictable(mapping
) &&
942 filemap_get_folios(mapping
, &index
, ~0UL, &fbatch
)) {
943 check_move_unevictable_folios(&fbatch
);
944 folio_batch_release(&fbatch
);
949 static struct folio
*shmem_get_partial_folio(struct inode
*inode
, pgoff_t index
)
954 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
955 * beyond i_size, and reports fallocated folios as holes.
957 folio
= filemap_get_entry(inode
->i_mapping
, index
);
960 if (!xa_is_value(folio
)) {
962 if (folio
->mapping
== inode
->i_mapping
)
964 /* The folio has been swapped out */
969 * But read a folio back from swap if any of it is within i_size
970 * (although in some cases this is just a waste of time).
973 shmem_get_folio(inode
, index
, &folio
, SGP_READ
);
978 * Remove range of pages and swap entries from page cache, and free them.
979 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
981 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
984 struct address_space
*mapping
= inode
->i_mapping
;
985 struct shmem_inode_info
*info
= SHMEM_I(inode
);
986 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
987 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
988 struct folio_batch fbatch
;
989 pgoff_t indices
[PAGEVEC_SIZE
];
992 long nr_swaps_freed
= 0;
997 end
= -1; /* unsigned, so actually very big */
999 if (info
->fallocend
> start
&& info
->fallocend
<= end
&& !unfalloc
)
1000 info
->fallocend
= start
;
1002 folio_batch_init(&fbatch
);
1004 while (index
< end
&& find_lock_entries(mapping
, &index
, end
- 1,
1005 &fbatch
, indices
)) {
1006 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
1007 folio
= fbatch
.folios
[i
];
1009 if (xa_is_value(folio
)) {
1012 nr_swaps_freed
+= !shmem_free_swap(mapping
,
1017 if (!unfalloc
|| !folio_test_uptodate(folio
))
1018 truncate_inode_folio(mapping
, folio
);
1019 folio_unlock(folio
);
1021 folio_batch_remove_exceptionals(&fbatch
);
1022 folio_batch_release(&fbatch
);
1027 * When undoing a failed fallocate, we want none of the partial folio
1028 * zeroing and splitting below, but shall want to truncate the whole
1029 * folio when !uptodate indicates that it was added by this fallocate,
1030 * even when [lstart, lend] covers only a part of the folio.
1035 same_folio
= (lstart
>> PAGE_SHIFT
) == (lend
>> PAGE_SHIFT
);
1036 folio
= shmem_get_partial_folio(inode
, lstart
>> PAGE_SHIFT
);
1038 same_folio
= lend
< folio_pos(folio
) + folio_size(folio
);
1039 folio_mark_dirty(folio
);
1040 if (!truncate_inode_partial_folio(folio
, lstart
, lend
)) {
1041 start
= folio_next_index(folio
);
1045 folio_unlock(folio
);
1051 folio
= shmem_get_partial_folio(inode
, lend
>> PAGE_SHIFT
);
1053 folio_mark_dirty(folio
);
1054 if (!truncate_inode_partial_folio(folio
, lstart
, lend
))
1056 folio_unlock(folio
);
1063 while (index
< end
) {
1066 if (!find_get_entries(mapping
, &index
, end
- 1, &fbatch
,
1068 /* If all gone or hole-punch or unfalloc, we're done */
1069 if (index
== start
|| end
!= -1)
1071 /* But if truncating, restart to make sure all gone */
1075 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
1076 folio
= fbatch
.folios
[i
];
1078 if (xa_is_value(folio
)) {
1081 if (shmem_free_swap(mapping
, indices
[i
], folio
)) {
1082 /* Swap was replaced by page: retry */
1092 if (!unfalloc
|| !folio_test_uptodate(folio
)) {
1093 if (folio_mapping(folio
) != mapping
) {
1094 /* Page was replaced by swap: retry */
1095 folio_unlock(folio
);
1099 VM_BUG_ON_FOLIO(folio_test_writeback(folio
),
1101 truncate_inode_folio(mapping
, folio
);
1103 folio_unlock(folio
);
1105 folio_batch_remove_exceptionals(&fbatch
);
1106 folio_batch_release(&fbatch
);
1109 shmem_recalc_inode(inode
, 0, -nr_swaps_freed
);
1112 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1114 shmem_undo_range(inode
, lstart
, lend
, false);
1115 inode
->i_mtime
= inode_set_ctime_current(inode
);
1116 inode_inc_iversion(inode
);
1118 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1120 static int shmem_getattr(struct mnt_idmap
*idmap
,
1121 const struct path
*path
, struct kstat
*stat
,
1122 u32 request_mask
, unsigned int query_flags
)
1124 struct inode
*inode
= path
->dentry
->d_inode
;
1125 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1127 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
)
1128 shmem_recalc_inode(inode
, 0, 0);
1130 if (info
->fsflags
& FS_APPEND_FL
)
1131 stat
->attributes
|= STATX_ATTR_APPEND
;
1132 if (info
->fsflags
& FS_IMMUTABLE_FL
)
1133 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
1134 if (info
->fsflags
& FS_NODUMP_FL
)
1135 stat
->attributes
|= STATX_ATTR_NODUMP
;
1136 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
1137 STATX_ATTR_IMMUTABLE
|
1139 generic_fillattr(idmap
, request_mask
, inode
, stat
);
1141 if (shmem_is_huge(inode
, 0, false, NULL
, 0))
1142 stat
->blksize
= HPAGE_PMD_SIZE
;
1144 if (request_mask
& STATX_BTIME
) {
1145 stat
->result_mask
|= STATX_BTIME
;
1146 stat
->btime
.tv_sec
= info
->i_crtime
.tv_sec
;
1147 stat
->btime
.tv_nsec
= info
->i_crtime
.tv_nsec
;
1153 static int shmem_setattr(struct mnt_idmap
*idmap
,
1154 struct dentry
*dentry
, struct iattr
*attr
)
1156 struct inode
*inode
= d_inode(dentry
);
1157 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1159 bool update_mtime
= false;
1160 bool update_ctime
= true;
1162 error
= setattr_prepare(idmap
, dentry
, attr
);
1166 if ((info
->seals
& F_SEAL_EXEC
) && (attr
->ia_valid
& ATTR_MODE
)) {
1167 if ((inode
->i_mode
^ attr
->ia_mode
) & 0111) {
1172 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1173 loff_t oldsize
= inode
->i_size
;
1174 loff_t newsize
= attr
->ia_size
;
1176 /* protected by i_rwsem */
1177 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1178 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1181 if (newsize
!= oldsize
) {
1182 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1186 i_size_write(inode
, newsize
);
1187 update_mtime
= true;
1189 update_ctime
= false;
1191 if (newsize
<= oldsize
) {
1192 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1193 if (oldsize
> holebegin
)
1194 unmap_mapping_range(inode
->i_mapping
,
1197 shmem_truncate_range(inode
,
1198 newsize
, (loff_t
)-1);
1199 /* unmap again to remove racily COWed private pages */
1200 if (oldsize
> holebegin
)
1201 unmap_mapping_range(inode
->i_mapping
,
1206 if (is_quota_modification(idmap
, inode
, attr
)) {
1207 error
= dquot_initialize(inode
);
1212 /* Transfer quota accounting */
1213 if (i_uid_needs_update(idmap
, attr
, inode
) ||
1214 i_gid_needs_update(idmap
, attr
, inode
)) {
1215 error
= dquot_transfer(idmap
, inode
, attr
);
1221 setattr_copy(idmap
, inode
, attr
);
1222 if (attr
->ia_valid
& ATTR_MODE
)
1223 error
= posix_acl_chmod(idmap
, dentry
, inode
->i_mode
);
1224 if (!error
&& update_ctime
) {
1225 inode_set_ctime_current(inode
);
1227 inode
->i_mtime
= inode_get_ctime(inode
);
1228 inode_inc_iversion(inode
);
1233 static void shmem_evict_inode(struct inode
*inode
)
1235 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1236 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1239 if (shmem_mapping(inode
->i_mapping
)) {
1240 shmem_unacct_size(info
->flags
, inode
->i_size
);
1242 mapping_set_exiting(inode
->i_mapping
);
1243 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1244 if (!list_empty(&info
->shrinklist
)) {
1245 spin_lock(&sbinfo
->shrinklist_lock
);
1246 if (!list_empty(&info
->shrinklist
)) {
1247 list_del_init(&info
->shrinklist
);
1248 sbinfo
->shrinklist_len
--;
1250 spin_unlock(&sbinfo
->shrinklist_lock
);
1252 while (!list_empty(&info
->swaplist
)) {
1253 /* Wait while shmem_unuse() is scanning this inode... */
1254 wait_var_event(&info
->stop_eviction
,
1255 !atomic_read(&info
->stop_eviction
));
1256 mutex_lock(&shmem_swaplist_mutex
);
1257 /* ...but beware of the race if we peeked too early */
1258 if (!atomic_read(&info
->stop_eviction
))
1259 list_del_init(&info
->swaplist
);
1260 mutex_unlock(&shmem_swaplist_mutex
);
1264 simple_xattrs_free(&info
->xattrs
, sbinfo
->max_inodes
? &freed
: NULL
);
1265 shmem_free_inode(inode
->i_sb
, freed
);
1266 WARN_ON(inode
->i_blocks
);
1268 #ifdef CONFIG_TMPFS_QUOTA
1269 dquot_free_inode(inode
);
1274 static int shmem_find_swap_entries(struct address_space
*mapping
,
1275 pgoff_t start
, struct folio_batch
*fbatch
,
1276 pgoff_t
*indices
, unsigned int type
)
1278 XA_STATE(xas
, &mapping
->i_pages
, start
);
1279 struct folio
*folio
;
1283 xas_for_each(&xas
, folio
, ULONG_MAX
) {
1284 if (xas_retry(&xas
, folio
))
1287 if (!xa_is_value(folio
))
1290 entry
= radix_to_swp_entry(folio
);
1292 * swapin error entries can be found in the mapping. But they're
1293 * deliberately ignored here as we've done everything we can do.
1295 if (swp_type(entry
) != type
)
1298 indices
[folio_batch_count(fbatch
)] = xas
.xa_index
;
1299 if (!folio_batch_add(fbatch
, folio
))
1302 if (need_resched()) {
1309 return xas
.xa_index
;
1313 * Move the swapped pages for an inode to page cache. Returns the count
1314 * of pages swapped in, or the error in case of failure.
1316 static int shmem_unuse_swap_entries(struct inode
*inode
,
1317 struct folio_batch
*fbatch
, pgoff_t
*indices
)
1322 struct address_space
*mapping
= inode
->i_mapping
;
1324 for (i
= 0; i
< folio_batch_count(fbatch
); i
++) {
1325 struct folio
*folio
= fbatch
->folios
[i
];
1327 if (!xa_is_value(folio
))
1329 error
= shmem_swapin_folio(inode
, indices
[i
],
1331 mapping_gfp_mask(mapping
),
1334 folio_unlock(folio
);
1338 if (error
== -ENOMEM
)
1342 return error
? error
: ret
;
1346 * If swap found in inode, free it and move page from swapcache to filecache.
1348 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
)
1350 struct address_space
*mapping
= inode
->i_mapping
;
1352 struct folio_batch fbatch
;
1353 pgoff_t indices
[PAGEVEC_SIZE
];
1357 folio_batch_init(&fbatch
);
1358 shmem_find_swap_entries(mapping
, start
, &fbatch
, indices
, type
);
1359 if (folio_batch_count(&fbatch
) == 0) {
1364 ret
= shmem_unuse_swap_entries(inode
, &fbatch
, indices
);
1368 start
= indices
[folio_batch_count(&fbatch
) - 1];
1375 * Read all the shared memory data that resides in the swap
1376 * device 'type' back into memory, so the swap device can be
1379 int shmem_unuse(unsigned int type
)
1381 struct shmem_inode_info
*info
, *next
;
1384 if (list_empty(&shmem_swaplist
))
1387 mutex_lock(&shmem_swaplist_mutex
);
1388 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1389 if (!info
->swapped
) {
1390 list_del_init(&info
->swaplist
);
1394 * Drop the swaplist mutex while searching the inode for swap;
1395 * but before doing so, make sure shmem_evict_inode() will not
1396 * remove placeholder inode from swaplist, nor let it be freed
1397 * (igrab() would protect from unlink, but not from unmount).
1399 atomic_inc(&info
->stop_eviction
);
1400 mutex_unlock(&shmem_swaplist_mutex
);
1402 error
= shmem_unuse_inode(&info
->vfs_inode
, type
);
1405 mutex_lock(&shmem_swaplist_mutex
);
1406 next
= list_next_entry(info
, swaplist
);
1408 list_del_init(&info
->swaplist
);
1409 if (atomic_dec_and_test(&info
->stop_eviction
))
1410 wake_up_var(&info
->stop_eviction
);
1414 mutex_unlock(&shmem_swaplist_mutex
);
1420 * Move the page from the page cache to the swap cache.
1422 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1424 struct folio
*folio
= page_folio(page
);
1425 struct address_space
*mapping
= folio
->mapping
;
1426 struct inode
*inode
= mapping
->host
;
1427 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1428 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1433 * Our capabilities prevent regular writeback or sync from ever calling
1434 * shmem_writepage; but a stacking filesystem might use ->writepage of
1435 * its underlying filesystem, in which case tmpfs should write out to
1436 * swap only in response to memory pressure, and not for the writeback
1439 if (WARN_ON_ONCE(!wbc
->for_reclaim
))
1442 if (WARN_ON_ONCE((info
->flags
& VM_LOCKED
) || sbinfo
->noswap
))
1445 if (!total_swap_pages
)
1449 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1450 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1451 * and its shmem_writeback() needs them to be split when swapping.
1453 if (folio_test_large(folio
)) {
1454 /* Ensure the subpages are still dirty */
1455 folio_test_set_dirty(folio
);
1456 if (split_huge_page(page
) < 0)
1458 folio
= page_folio(page
);
1459 folio_clear_dirty(folio
);
1462 index
= folio
->index
;
1465 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1466 * value into swapfile.c, the only way we can correctly account for a
1467 * fallocated folio arriving here is now to initialize it and write it.
1469 * That's okay for a folio already fallocated earlier, but if we have
1470 * not yet completed the fallocation, then (a) we want to keep track
1471 * of this folio in case we have to undo it, and (b) it may not be a
1472 * good idea to continue anyway, once we're pushing into swap. So
1473 * reactivate the folio, and let shmem_fallocate() quit when too many.
1475 if (!folio_test_uptodate(folio
)) {
1476 if (inode
->i_private
) {
1477 struct shmem_falloc
*shmem_falloc
;
1478 spin_lock(&inode
->i_lock
);
1479 shmem_falloc
= inode
->i_private
;
1481 !shmem_falloc
->waitq
&&
1482 index
>= shmem_falloc
->start
&&
1483 index
< shmem_falloc
->next
)
1484 shmem_falloc
->nr_unswapped
++;
1486 shmem_falloc
= NULL
;
1487 spin_unlock(&inode
->i_lock
);
1491 folio_zero_range(folio
, 0, folio_size(folio
));
1492 flush_dcache_folio(folio
);
1493 folio_mark_uptodate(folio
);
1496 swap
= folio_alloc_swap(folio
);
1501 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1502 * if it's not already there. Do it now before the folio is
1503 * moved to swap cache, when its pagelock no longer protects
1504 * the inode from eviction. But don't unlock the mutex until
1505 * we've incremented swapped, because shmem_unuse_inode() will
1506 * prune a !swapped inode from the swaplist under this mutex.
1508 mutex_lock(&shmem_swaplist_mutex
);
1509 if (list_empty(&info
->swaplist
))
1510 list_add(&info
->swaplist
, &shmem_swaplist
);
1512 if (add_to_swap_cache(folio
, swap
,
1513 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1515 shmem_recalc_inode(inode
, 0, 1);
1516 swap_shmem_alloc(swap
);
1517 shmem_delete_from_page_cache(folio
, swp_to_radix_entry(swap
));
1519 mutex_unlock(&shmem_swaplist_mutex
);
1520 BUG_ON(folio_mapped(folio
));
1521 swap_writepage(&folio
->page
, wbc
);
1525 mutex_unlock(&shmem_swaplist_mutex
);
1526 put_swap_folio(folio
, swap
);
1528 folio_mark_dirty(folio
);
1529 if (wbc
->for_reclaim
)
1530 return AOP_WRITEPAGE_ACTIVATE
; /* Return with folio locked */
1531 folio_unlock(folio
);
1535 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1536 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1540 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1541 return; /* show nothing */
1543 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1545 seq_printf(seq
, ",mpol=%s", buffer
);
1548 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1550 struct mempolicy
*mpol
= NULL
;
1552 raw_spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1553 mpol
= sbinfo
->mpol
;
1555 raw_spin_unlock(&sbinfo
->stat_lock
);
1559 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1560 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1563 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1567 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1569 #define vm_policy vm_private_data
1572 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1573 struct shmem_inode_info
*info
, pgoff_t index
)
1575 /* Create a pseudo vma that just contains the policy */
1576 vma_init(vma
, NULL
);
1577 /* Bias interleave by inode number to distribute better across nodes */
1578 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1579 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1582 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1584 /* Drop reference taken by mpol_shared_policy_lookup() */
1585 mpol_cond_put(vma
->vm_policy
);
1588 static struct folio
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1589 struct shmem_inode_info
*info
, pgoff_t index
)
1591 struct vm_area_struct pvma
;
1593 struct vm_fault vmf
= {
1597 shmem_pseudo_vma_init(&pvma
, info
, index
);
1598 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1599 shmem_pseudo_vma_destroy(&pvma
);
1603 return page_folio(page
);
1607 * Make sure huge_gfp is always more limited than limit_gfp.
1608 * Some of the flags set permissions, while others set limitations.
1610 static gfp_t
limit_gfp_mask(gfp_t huge_gfp
, gfp_t limit_gfp
)
1612 gfp_t allowflags
= __GFP_IO
| __GFP_FS
| __GFP_RECLAIM
;
1613 gfp_t denyflags
= __GFP_NOWARN
| __GFP_NORETRY
;
1614 gfp_t zoneflags
= limit_gfp
& GFP_ZONEMASK
;
1615 gfp_t result
= huge_gfp
& ~(allowflags
| GFP_ZONEMASK
);
1617 /* Allow allocations only from the originally specified zones. */
1618 result
|= zoneflags
;
1621 * Minimize the result gfp by taking the union with the deny flags,
1622 * and the intersection of the allow flags.
1624 result
|= (limit_gfp
& denyflags
);
1625 result
|= (huge_gfp
& limit_gfp
) & allowflags
;
1630 static struct folio
*shmem_alloc_hugefolio(gfp_t gfp
,
1631 struct shmem_inode_info
*info
, pgoff_t index
)
1633 struct vm_area_struct pvma
;
1634 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1636 struct folio
*folio
;
1638 hindex
= round_down(index
, HPAGE_PMD_NR
);
1639 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1643 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1644 folio
= vma_alloc_folio(gfp
, HPAGE_PMD_ORDER
, &pvma
, 0, true);
1645 shmem_pseudo_vma_destroy(&pvma
);
1647 count_vm_event(THP_FILE_FALLBACK
);
1651 static struct folio
*shmem_alloc_folio(gfp_t gfp
,
1652 struct shmem_inode_info
*info
, pgoff_t index
)
1654 struct vm_area_struct pvma
;
1655 struct folio
*folio
;
1657 shmem_pseudo_vma_init(&pvma
, info
, index
);
1658 folio
= vma_alloc_folio(gfp
, 0, &pvma
, 0, false);
1659 shmem_pseudo_vma_destroy(&pvma
);
1664 static struct folio
*shmem_alloc_and_acct_folio(gfp_t gfp
, struct inode
*inode
,
1665 pgoff_t index
, bool huge
)
1667 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1668 struct folio
*folio
;
1672 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1674 nr
= huge
? HPAGE_PMD_NR
: 1;
1676 err
= shmem_inode_acct_block(inode
, nr
);
1681 folio
= shmem_alloc_hugefolio(gfp
, info
, index
);
1683 folio
= shmem_alloc_folio(gfp
, info
, index
);
1685 __folio_set_locked(folio
);
1686 __folio_set_swapbacked(folio
);
1691 shmem_inode_unacct_blocks(inode
, nr
);
1693 return ERR_PTR(err
);
1697 * When a page is moved from swapcache to shmem filecache (either by the
1698 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1699 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1700 * ignorance of the mapping it belongs to. If that mapping has special
1701 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1702 * we may need to copy to a suitable page before moving to filecache.
1704 * In a future release, this may well be extended to respect cpuset and
1705 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1706 * but for now it is a simple matter of zone.
1708 static bool shmem_should_replace_folio(struct folio
*folio
, gfp_t gfp
)
1710 return folio_zonenum(folio
) > gfp_zone(gfp
);
1713 static int shmem_replace_folio(struct folio
**foliop
, gfp_t gfp
,
1714 struct shmem_inode_info
*info
, pgoff_t index
)
1716 struct folio
*old
, *new;
1717 struct address_space
*swap_mapping
;
1724 swap_index
= swp_offset(entry
);
1725 swap_mapping
= swap_address_space(entry
);
1728 * We have arrived here because our zones are constrained, so don't
1729 * limit chance of success by further cpuset and node constraints.
1731 gfp
&= ~GFP_CONSTRAINT_MASK
;
1732 VM_BUG_ON_FOLIO(folio_test_large(old
), old
);
1733 new = shmem_alloc_folio(gfp
, info
, index
);
1738 folio_copy(new, old
);
1739 flush_dcache_folio(new);
1741 __folio_set_locked(new);
1742 __folio_set_swapbacked(new);
1743 folio_mark_uptodate(new);
1745 folio_set_swapcache(new);
1748 * Our caller will very soon move newpage out of swapcache, but it's
1749 * a nice clean interface for us to replace oldpage by newpage there.
1751 xa_lock_irq(&swap_mapping
->i_pages
);
1752 error
= shmem_replace_entry(swap_mapping
, swap_index
, old
, new);
1754 mem_cgroup_migrate(old
, new);
1755 __lruvec_stat_mod_folio(new, NR_FILE_PAGES
, 1);
1756 __lruvec_stat_mod_folio(new, NR_SHMEM
, 1);
1757 __lruvec_stat_mod_folio(old
, NR_FILE_PAGES
, -1);
1758 __lruvec_stat_mod_folio(old
, NR_SHMEM
, -1);
1760 xa_unlock_irq(&swap_mapping
->i_pages
);
1762 if (unlikely(error
)) {
1764 * Is this possible? I think not, now that our callers check
1765 * both PageSwapCache and page_private after getting page lock;
1766 * but be defensive. Reverse old to newpage for clear and free.
1774 folio_clear_swapcache(old
);
1775 old
->private = NULL
;
1778 folio_put_refs(old
, 2);
1782 static void shmem_set_folio_swapin_error(struct inode
*inode
, pgoff_t index
,
1783 struct folio
*folio
, swp_entry_t swap
)
1785 struct address_space
*mapping
= inode
->i_mapping
;
1786 swp_entry_t swapin_error
;
1789 swapin_error
= make_poisoned_swp_entry();
1790 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
,
1791 swp_to_radix_entry(swap
),
1792 swp_to_radix_entry(swapin_error
), 0);
1793 if (old
!= swp_to_radix_entry(swap
))
1796 folio_wait_writeback(folio
);
1797 delete_from_swap_cache(folio
);
1799 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
1800 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
1801 * in shmem_evict_inode().
1803 shmem_recalc_inode(inode
, -1, -1);
1808 * Swap in the folio pointed to by *foliop.
1809 * Caller has to make sure that *foliop contains a valid swapped folio.
1810 * Returns 0 and the folio in foliop if success. On failure, returns the
1811 * error code and NULL in *foliop.
1813 static int shmem_swapin_folio(struct inode
*inode
, pgoff_t index
,
1814 struct folio
**foliop
, enum sgp_type sgp
,
1815 gfp_t gfp
, struct vm_area_struct
*vma
,
1816 vm_fault_t
*fault_type
)
1818 struct address_space
*mapping
= inode
->i_mapping
;
1819 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1820 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: NULL
;
1821 struct swap_info_struct
*si
;
1822 struct folio
*folio
= NULL
;
1826 VM_BUG_ON(!*foliop
|| !xa_is_value(*foliop
));
1827 swap
= radix_to_swp_entry(*foliop
);
1830 if (is_poisoned_swp_entry(swap
))
1833 si
= get_swap_device(swap
);
1835 if (!shmem_confirm_swap(mapping
, index
, swap
))
1841 /* Look it up and read it in.. */
1842 folio
= swap_cache_get_folio(swap
, NULL
, 0);
1844 /* Or update major stats only when swapin succeeds?? */
1846 *fault_type
|= VM_FAULT_MAJOR
;
1847 count_vm_event(PGMAJFAULT
);
1848 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1850 /* Here we actually start the io */
1851 folio
= shmem_swapin(swap
, gfp
, info
, index
);
1858 /* We have to do this with folio locked to prevent races */
1860 if (!folio_test_swapcache(folio
) ||
1861 folio
->swap
.val
!= swap
.val
||
1862 !shmem_confirm_swap(mapping
, index
, swap
)) {
1866 if (!folio_test_uptodate(folio
)) {
1870 folio_wait_writeback(folio
);
1873 * Some architectures may have to restore extra metadata to the
1874 * folio after reading from swap.
1876 arch_swap_restore(swap
, folio
);
1878 if (shmem_should_replace_folio(folio
, gfp
)) {
1879 error
= shmem_replace_folio(&folio
, gfp
, info
, index
);
1884 error
= shmem_add_to_page_cache(folio
, mapping
, index
,
1885 swp_to_radix_entry(swap
), gfp
,
1890 shmem_recalc_inode(inode
, 0, -1);
1892 if (sgp
== SGP_WRITE
)
1893 folio_mark_accessed(folio
);
1895 delete_from_swap_cache(folio
);
1896 folio_mark_dirty(folio
);
1898 put_swap_device(si
);
1903 if (!shmem_confirm_swap(mapping
, index
, swap
))
1906 shmem_set_folio_swapin_error(inode
, index
, folio
, swap
);
1909 folio_unlock(folio
);
1912 put_swap_device(si
);
1918 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1920 * If we allocate a new one we do not mark it dirty. That's up to the
1921 * vm. If we swap it in we mark it dirty since we also free the swap
1922 * entry since a page cannot live in both the swap and page cache.
1924 * vma, vmf, and fault_type are only supplied by shmem_fault:
1925 * otherwise they are NULL.
1927 static int shmem_get_folio_gfp(struct inode
*inode
, pgoff_t index
,
1928 struct folio
**foliop
, enum sgp_type sgp
, gfp_t gfp
,
1929 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1930 vm_fault_t
*fault_type
)
1932 struct address_space
*mapping
= inode
->i_mapping
;
1933 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1934 struct shmem_sb_info
*sbinfo
;
1935 struct mm_struct
*charge_mm
;
1936 struct folio
*folio
;
1943 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1946 if (sgp
<= SGP_CACHE
&&
1947 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1951 sbinfo
= SHMEM_SB(inode
->i_sb
);
1952 charge_mm
= vma
? vma
->vm_mm
: NULL
;
1954 folio
= filemap_get_entry(mapping
, index
);
1955 if (folio
&& vma
&& userfaultfd_minor(vma
)) {
1956 if (!xa_is_value(folio
))
1958 *fault_type
= handle_userfault(vmf
, VM_UFFD_MINOR
);
1962 if (xa_is_value(folio
)) {
1963 error
= shmem_swapin_folio(inode
, index
, &folio
,
1964 sgp
, gfp
, vma
, fault_type
);
1965 if (error
== -EEXIST
)
1975 /* Has the folio been truncated or swapped out? */
1976 if (unlikely(folio
->mapping
!= mapping
)) {
1977 folio_unlock(folio
);
1981 if (sgp
== SGP_WRITE
)
1982 folio_mark_accessed(folio
);
1983 if (folio_test_uptodate(folio
))
1985 /* fallocated folio */
1986 if (sgp
!= SGP_READ
)
1988 folio_unlock(folio
);
1993 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1994 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1997 if (sgp
== SGP_READ
)
1999 if (sgp
== SGP_NOALLOC
)
2003 * Fast cache lookup and swap lookup did not find it: allocate.
2006 if (vma
&& userfaultfd_missing(vma
)) {
2007 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
2011 if (!shmem_is_huge(inode
, index
, false,
2012 vma
? vma
->vm_mm
: NULL
, vma
? vma
->vm_flags
: 0))
2015 huge_gfp
= vma_thp_gfp_mask(vma
);
2016 huge_gfp
= limit_gfp_mask(huge_gfp
, gfp
);
2017 folio
= shmem_alloc_and_acct_folio(huge_gfp
, inode
, index
, true);
2018 if (IS_ERR(folio
)) {
2020 folio
= shmem_alloc_and_acct_folio(gfp
, inode
, index
, false);
2022 if (IS_ERR(folio
)) {
2025 error
= PTR_ERR(folio
);
2027 if (error
!= -ENOSPC
)
2030 * Try to reclaim some space by splitting a large folio
2031 * beyond i_size on the filesystem.
2036 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
2037 if (ret
== SHRINK_STOP
)
2045 hindex
= round_down(index
, folio_nr_pages(folio
));
2047 if (sgp
== SGP_WRITE
)
2048 __folio_set_referenced(folio
);
2050 error
= shmem_add_to_page_cache(folio
, mapping
, hindex
,
2051 NULL
, gfp
& GFP_RECLAIM_MASK
,
2056 folio_add_lru(folio
);
2057 shmem_recalc_inode(inode
, folio_nr_pages(folio
), 0);
2060 if (folio_test_pmd_mappable(folio
) &&
2061 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
2062 folio_next_index(folio
) - 1) {
2064 * Part of the large folio is beyond i_size: subject
2065 * to shrink under memory pressure.
2067 spin_lock(&sbinfo
->shrinklist_lock
);
2069 * _careful to defend against unlocked access to
2070 * ->shrink_list in shmem_unused_huge_shrink()
2072 if (list_empty_careful(&info
->shrinklist
)) {
2073 list_add_tail(&info
->shrinklist
,
2074 &sbinfo
->shrinklist
);
2075 sbinfo
->shrinklist_len
++;
2077 spin_unlock(&sbinfo
->shrinklist_lock
);
2081 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2083 if (sgp
== SGP_FALLOC
)
2087 * Let SGP_WRITE caller clear ends if write does not fill folio;
2088 * but SGP_FALLOC on a folio fallocated earlier must initialize
2089 * it now, lest undo on failure cancel our earlier guarantee.
2091 if (sgp
!= SGP_WRITE
&& !folio_test_uptodate(folio
)) {
2092 long i
, n
= folio_nr_pages(folio
);
2094 for (i
= 0; i
< n
; i
++)
2095 clear_highpage(folio_page(folio
, i
));
2096 flush_dcache_folio(folio
);
2097 folio_mark_uptodate(folio
);
2100 /* Perhaps the file has been truncated since we checked */
2101 if (sgp
<= SGP_CACHE
&&
2102 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
2104 folio_clear_dirty(folio
);
2105 filemap_remove_folio(folio
);
2106 shmem_recalc_inode(inode
, 0, 0);
2119 shmem_inode_unacct_blocks(inode
, folio_nr_pages(folio
));
2121 if (folio_test_large(folio
)) {
2122 folio_unlock(folio
);
2128 folio_unlock(folio
);
2131 if (error
== -ENOSPC
&& !once
++) {
2132 shmem_recalc_inode(inode
, 0, 0);
2135 if (error
== -EEXIST
)
2140 int shmem_get_folio(struct inode
*inode
, pgoff_t index
, struct folio
**foliop
,
2143 return shmem_get_folio_gfp(inode
, index
, foliop
, sgp
,
2144 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
2148 * This is like autoremove_wake_function, but it removes the wait queue
2149 * entry unconditionally - even if something else had already woken the
2152 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2154 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2155 list_del_init(&wait
->entry
);
2159 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2161 struct vm_area_struct
*vma
= vmf
->vma
;
2162 struct inode
*inode
= file_inode(vma
->vm_file
);
2163 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2164 struct folio
*folio
= NULL
;
2166 vm_fault_t ret
= VM_FAULT_LOCKED
;
2169 * Trinity finds that probing a hole which tmpfs is punching can
2170 * prevent the hole-punch from ever completing: which in turn
2171 * locks writers out with its hold on i_rwsem. So refrain from
2172 * faulting pages into the hole while it's being punched. Although
2173 * shmem_undo_range() does remove the additions, it may be unable to
2174 * keep up, as each new page needs its own unmap_mapping_range() call,
2175 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2177 * It does not matter if we sometimes reach this check just before the
2178 * hole-punch begins, so that one fault then races with the punch:
2179 * we just need to make racing faults a rare case.
2181 * The implementation below would be much simpler if we just used a
2182 * standard mutex or completion: but we cannot take i_rwsem in fault,
2183 * and bloating every shmem inode for this unlikely case would be sad.
2185 if (unlikely(inode
->i_private
)) {
2186 struct shmem_falloc
*shmem_falloc
;
2188 spin_lock(&inode
->i_lock
);
2189 shmem_falloc
= inode
->i_private
;
2191 shmem_falloc
->waitq
&&
2192 vmf
->pgoff
>= shmem_falloc
->start
&&
2193 vmf
->pgoff
< shmem_falloc
->next
) {
2195 wait_queue_head_t
*shmem_falloc_waitq
;
2196 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2198 ret
= VM_FAULT_NOPAGE
;
2199 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2201 ret
= VM_FAULT_RETRY
;
2203 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2204 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2205 TASK_UNINTERRUPTIBLE
);
2206 spin_unlock(&inode
->i_lock
);
2210 * shmem_falloc_waitq points into the shmem_fallocate()
2211 * stack of the hole-punching task: shmem_falloc_waitq
2212 * is usually invalid by the time we reach here, but
2213 * finish_wait() does not dereference it in that case;
2214 * though i_lock needed lest racing with wake_up_all().
2216 spin_lock(&inode
->i_lock
);
2217 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2218 spin_unlock(&inode
->i_lock
);
2224 spin_unlock(&inode
->i_lock
);
2227 err
= shmem_get_folio_gfp(inode
, vmf
->pgoff
, &folio
, SGP_CACHE
,
2228 gfp
, vma
, vmf
, &ret
);
2230 return vmf_error(err
);
2232 vmf
->page
= folio_file_page(folio
, vmf
->pgoff
);
2236 unsigned long shmem_get_unmapped_area(struct file
*file
,
2237 unsigned long uaddr
, unsigned long len
,
2238 unsigned long pgoff
, unsigned long flags
)
2240 unsigned long (*get_area
)(struct file
*,
2241 unsigned long, unsigned long, unsigned long, unsigned long);
2243 unsigned long offset
;
2244 unsigned long inflated_len
;
2245 unsigned long inflated_addr
;
2246 unsigned long inflated_offset
;
2248 if (len
> TASK_SIZE
)
2251 get_area
= current
->mm
->get_unmapped_area
;
2252 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2254 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2256 if (IS_ERR_VALUE(addr
))
2258 if (addr
& ~PAGE_MASK
)
2260 if (addr
> TASK_SIZE
- len
)
2263 if (shmem_huge
== SHMEM_HUGE_DENY
)
2265 if (len
< HPAGE_PMD_SIZE
)
2267 if (flags
& MAP_FIXED
)
2270 * Our priority is to support MAP_SHARED mapped hugely;
2271 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2272 * But if caller specified an address hint and we allocated area there
2273 * successfully, respect that as before.
2278 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2279 struct super_block
*sb
;
2282 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2283 sb
= file_inode(file
)->i_sb
;
2286 * Called directly from mm/mmap.c, or drivers/char/mem.c
2287 * for "/dev/zero", to create a shared anonymous object.
2289 if (IS_ERR(shm_mnt
))
2291 sb
= shm_mnt
->mnt_sb
;
2293 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2297 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2298 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2300 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2303 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2304 if (inflated_len
> TASK_SIZE
)
2306 if (inflated_len
< len
)
2309 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2310 if (IS_ERR_VALUE(inflated_addr
))
2312 if (inflated_addr
& ~PAGE_MASK
)
2315 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2316 inflated_addr
+= offset
- inflated_offset
;
2317 if (inflated_offset
> offset
)
2318 inflated_addr
+= HPAGE_PMD_SIZE
;
2320 if (inflated_addr
> TASK_SIZE
- len
)
2322 return inflated_addr
;
2326 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2328 struct inode
*inode
= file_inode(vma
->vm_file
);
2329 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2332 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2335 struct inode
*inode
= file_inode(vma
->vm_file
);
2338 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2339 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2343 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
2345 struct inode
*inode
= file_inode(file
);
2346 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2347 int retval
= -ENOMEM
;
2350 * What serializes the accesses to info->flags?
2351 * ipc_lock_object() when called from shmctl_do_lock(),
2352 * no serialization needed when called from shm_destroy().
2354 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2355 if (!user_shm_lock(inode
->i_size
, ucounts
))
2357 info
->flags
|= VM_LOCKED
;
2358 mapping_set_unevictable(file
->f_mapping
);
2360 if (!lock
&& (info
->flags
& VM_LOCKED
) && ucounts
) {
2361 user_shm_unlock(inode
->i_size
, ucounts
);
2362 info
->flags
&= ~VM_LOCKED
;
2363 mapping_clear_unevictable(file
->f_mapping
);
2371 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2373 struct inode
*inode
= file_inode(file
);
2374 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2377 ret
= seal_check_future_write(info
->seals
, vma
);
2381 /* arm64 - allow memory tagging on RAM-based files */
2382 vm_flags_set(vma
, VM_MTE_ALLOWED
);
2384 file_accessed(file
);
2385 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2387 vma
->vm_ops
= &shmem_vm_ops
;
2389 vma
->vm_ops
= &shmem_anon_vm_ops
;
2393 static int shmem_file_open(struct inode
*inode
, struct file
*file
)
2395 file
->f_mode
|= FMODE_CAN_ODIRECT
;
2396 return generic_file_open(inode
, file
);
2399 #ifdef CONFIG_TMPFS_XATTR
2400 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2403 * chattr's fsflags are unrelated to extended attributes,
2404 * but tmpfs has chosen to enable them under the same config option.
2406 static void shmem_set_inode_flags(struct inode
*inode
, unsigned int fsflags
)
2408 unsigned int i_flags
= 0;
2410 if (fsflags
& FS_NOATIME_FL
)
2411 i_flags
|= S_NOATIME
;
2412 if (fsflags
& FS_APPEND_FL
)
2413 i_flags
|= S_APPEND
;
2414 if (fsflags
& FS_IMMUTABLE_FL
)
2415 i_flags
|= S_IMMUTABLE
;
2417 * But FS_NODUMP_FL does not require any action in i_flags.
2419 inode_set_flags(inode
, i_flags
, S_NOATIME
| S_APPEND
| S_IMMUTABLE
);
2422 static void shmem_set_inode_flags(struct inode
*inode
, unsigned int fsflags
)
2425 #define shmem_initxattrs NULL
2428 static struct offset_ctx
*shmem_get_offset_ctx(struct inode
*inode
)
2430 return &SHMEM_I(inode
)->dir_offsets
;
2433 static struct inode
*__shmem_get_inode(struct mnt_idmap
*idmap
,
2434 struct super_block
*sb
,
2435 struct inode
*dir
, umode_t mode
,
2436 dev_t dev
, unsigned long flags
)
2438 struct inode
*inode
;
2439 struct shmem_inode_info
*info
;
2440 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2444 err
= shmem_reserve_inode(sb
, &ino
);
2446 return ERR_PTR(err
);
2449 inode
= new_inode(sb
);
2451 shmem_free_inode(sb
, 0);
2452 return ERR_PTR(-ENOSPC
);
2456 inode_init_owner(idmap
, inode
, dir
, mode
);
2457 inode
->i_blocks
= 0;
2458 inode
->i_atime
= inode
->i_mtime
= inode_set_ctime_current(inode
);
2459 inode
->i_generation
= get_random_u32();
2460 info
= SHMEM_I(inode
);
2461 memset(info
, 0, (char *)inode
- (char *)info
);
2462 spin_lock_init(&info
->lock
);
2463 atomic_set(&info
->stop_eviction
, 0);
2464 info
->seals
= F_SEAL_SEAL
;
2465 info
->flags
= flags
& VM_NORESERVE
;
2466 info
->i_crtime
= inode
->i_mtime
;
2467 info
->fsflags
= (dir
== NULL
) ? 0 :
2468 SHMEM_I(dir
)->fsflags
& SHMEM_FL_INHERITED
;
2470 shmem_set_inode_flags(inode
, info
->fsflags
);
2471 INIT_LIST_HEAD(&info
->shrinklist
);
2472 INIT_LIST_HEAD(&info
->swaplist
);
2473 INIT_LIST_HEAD(&info
->swaplist
);
2475 mapping_set_unevictable(inode
->i_mapping
);
2476 simple_xattrs_init(&info
->xattrs
);
2477 cache_no_acl(inode
);
2478 mapping_set_large_folios(inode
->i_mapping
);
2480 switch (mode
& S_IFMT
) {
2482 inode
->i_op
= &shmem_special_inode_operations
;
2483 init_special_inode(inode
, mode
, dev
);
2486 inode
->i_mapping
->a_ops
= &shmem_aops
;
2487 inode
->i_op
= &shmem_inode_operations
;
2488 inode
->i_fop
= &shmem_file_operations
;
2489 mpol_shared_policy_init(&info
->policy
,
2490 shmem_get_sbmpol(sbinfo
));
2494 /* Some things misbehave if size == 0 on a directory */
2495 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2496 inode
->i_op
= &shmem_dir_inode_operations
;
2497 inode
->i_fop
= &simple_offset_dir_operations
;
2498 simple_offset_init(shmem_get_offset_ctx(inode
));
2502 * Must not load anything in the rbtree,
2503 * mpol_free_shared_policy will not be called.
2505 mpol_shared_policy_init(&info
->policy
, NULL
);
2509 lockdep_annotate_inode_mutex_key(inode
);
2513 #ifdef CONFIG_TMPFS_QUOTA
2514 static struct inode
*shmem_get_inode(struct mnt_idmap
*idmap
,
2515 struct super_block
*sb
, struct inode
*dir
,
2516 umode_t mode
, dev_t dev
, unsigned long flags
)
2519 struct inode
*inode
;
2521 inode
= __shmem_get_inode(idmap
, sb
, dir
, mode
, dev
, flags
);
2525 err
= dquot_initialize(inode
);
2529 err
= dquot_alloc_inode(inode
);
2537 inode
->i_flags
|= S_NOQUOTA
;
2539 return ERR_PTR(err
);
2542 static inline struct inode
*shmem_get_inode(struct mnt_idmap
*idmap
,
2543 struct super_block
*sb
, struct inode
*dir
,
2544 umode_t mode
, dev_t dev
, unsigned long flags
)
2546 return __shmem_get_inode(idmap
, sb
, dir
, mode
, dev
, flags
);
2548 #endif /* CONFIG_TMPFS_QUOTA */
2550 #ifdef CONFIG_USERFAULTFD
2551 int shmem_mfill_atomic_pte(pmd_t
*dst_pmd
,
2552 struct vm_area_struct
*dst_vma
,
2553 unsigned long dst_addr
,
2554 unsigned long src_addr
,
2556 struct folio
**foliop
)
2558 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2559 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2560 struct address_space
*mapping
= inode
->i_mapping
;
2561 gfp_t gfp
= mapping_gfp_mask(mapping
);
2562 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2564 struct folio
*folio
;
2568 if (shmem_inode_acct_block(inode
, 1)) {
2570 * We may have got a page, returned -ENOENT triggering a retry,
2571 * and now we find ourselves with -ENOMEM. Release the page, to
2572 * avoid a BUG_ON in our caller.
2574 if (unlikely(*foliop
)) {
2583 folio
= shmem_alloc_folio(gfp
, info
, pgoff
);
2585 goto out_unacct_blocks
;
2587 if (uffd_flags_mode_is(flags
, MFILL_ATOMIC_COPY
)) {
2588 page_kaddr
= kmap_local_folio(folio
, 0);
2590 * The read mmap_lock is held here. Despite the
2591 * mmap_lock being read recursive a deadlock is still
2592 * possible if a writer has taken a lock. For example:
2594 * process A thread 1 takes read lock on own mmap_lock
2595 * process A thread 2 calls mmap, blocks taking write lock
2596 * process B thread 1 takes page fault, read lock on own mmap lock
2597 * process B thread 2 calls mmap, blocks taking write lock
2598 * process A thread 1 blocks taking read lock on process B
2599 * process B thread 1 blocks taking read lock on process A
2601 * Disable page faults to prevent potential deadlock
2602 * and retry the copy outside the mmap_lock.
2604 pagefault_disable();
2605 ret
= copy_from_user(page_kaddr
,
2606 (const void __user
*)src_addr
,
2609 kunmap_local(page_kaddr
);
2611 /* fallback to copy_from_user outside mmap_lock */
2612 if (unlikely(ret
)) {
2615 /* don't free the page */
2616 goto out_unacct_blocks
;
2619 flush_dcache_folio(folio
);
2620 } else { /* ZEROPAGE */
2621 clear_user_highpage(&folio
->page
, dst_addr
);
2625 VM_BUG_ON_FOLIO(folio_test_large(folio
), folio
);
2629 VM_BUG_ON(folio_test_locked(folio
));
2630 VM_BUG_ON(folio_test_swapbacked(folio
));
2631 __folio_set_locked(folio
);
2632 __folio_set_swapbacked(folio
);
2633 __folio_mark_uptodate(folio
);
2636 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2637 if (unlikely(pgoff
>= max_off
))
2640 ret
= shmem_add_to_page_cache(folio
, mapping
, pgoff
, NULL
,
2641 gfp
& GFP_RECLAIM_MASK
, dst_vma
->vm_mm
);
2645 ret
= mfill_atomic_install_pte(dst_pmd
, dst_vma
, dst_addr
,
2646 &folio
->page
, true, flags
);
2648 goto out_delete_from_cache
;
2650 shmem_recalc_inode(inode
, 1, 0);
2651 folio_unlock(folio
);
2653 out_delete_from_cache
:
2654 filemap_remove_folio(folio
);
2656 folio_unlock(folio
);
2659 shmem_inode_unacct_blocks(inode
, 1);
2662 #endif /* CONFIG_USERFAULTFD */
2665 static const struct inode_operations shmem_symlink_inode_operations
;
2666 static const struct inode_operations shmem_short_symlink_operations
;
2669 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2670 loff_t pos
, unsigned len
,
2671 struct page
**pagep
, void **fsdata
)
2673 struct inode
*inode
= mapping
->host
;
2674 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2675 pgoff_t index
= pos
>> PAGE_SHIFT
;
2676 struct folio
*folio
;
2679 /* i_rwsem is held by caller */
2680 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2681 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2682 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2684 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2688 ret
= shmem_get_folio(inode
, index
, &folio
, SGP_WRITE
);
2693 *pagep
= folio_file_page(folio
, index
);
2694 if (PageHWPoison(*pagep
)) {
2695 folio_unlock(folio
);
2705 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2706 loff_t pos
, unsigned len
, unsigned copied
,
2707 struct page
*page
, void *fsdata
)
2709 struct folio
*folio
= page_folio(page
);
2710 struct inode
*inode
= mapping
->host
;
2712 if (pos
+ copied
> inode
->i_size
)
2713 i_size_write(inode
, pos
+ copied
);
2715 if (!folio_test_uptodate(folio
)) {
2716 if (copied
< folio_size(folio
)) {
2717 size_t from
= offset_in_folio(folio
, pos
);
2718 folio_zero_segments(folio
, 0, from
,
2719 from
+ copied
, folio_size(folio
));
2721 folio_mark_uptodate(folio
);
2723 folio_mark_dirty(folio
);
2724 folio_unlock(folio
);
2730 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2732 struct file
*file
= iocb
->ki_filp
;
2733 struct inode
*inode
= file_inode(file
);
2734 struct address_space
*mapping
= inode
->i_mapping
;
2736 unsigned long offset
;
2739 loff_t
*ppos
= &iocb
->ki_pos
;
2741 index
= *ppos
>> PAGE_SHIFT
;
2742 offset
= *ppos
& ~PAGE_MASK
;
2745 struct folio
*folio
= NULL
;
2746 struct page
*page
= NULL
;
2748 unsigned long nr
, ret
;
2749 loff_t i_size
= i_size_read(inode
);
2751 end_index
= i_size
>> PAGE_SHIFT
;
2752 if (index
> end_index
)
2754 if (index
== end_index
) {
2755 nr
= i_size
& ~PAGE_MASK
;
2760 error
= shmem_get_folio(inode
, index
, &folio
, SGP_READ
);
2762 if (error
== -EINVAL
)
2767 folio_unlock(folio
);
2769 page
= folio_file_page(folio
, index
);
2770 if (PageHWPoison(page
)) {
2778 * We must evaluate after, since reads (unlike writes)
2779 * are called without i_rwsem protection against truncate
2782 i_size
= i_size_read(inode
);
2783 end_index
= i_size
>> PAGE_SHIFT
;
2784 if (index
== end_index
) {
2785 nr
= i_size
& ~PAGE_MASK
;
2796 * If users can be writing to this page using arbitrary
2797 * virtual addresses, take care about potential aliasing
2798 * before reading the page on the kernel side.
2800 if (mapping_writably_mapped(mapping
))
2801 flush_dcache_page(page
);
2803 * Mark the page accessed if we read the beginning.
2806 folio_mark_accessed(folio
);
2808 * Ok, we have the page, and it's up-to-date, so
2809 * now we can copy it to user space...
2811 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2814 } else if (user_backed_iter(to
)) {
2816 * Copy to user tends to be so well optimized, but
2817 * clear_user() not so much, that it is noticeably
2818 * faster to copy the zero page instead of clearing.
2820 ret
= copy_page_to_iter(ZERO_PAGE(0), offset
, nr
, to
);
2823 * But submitting the same page twice in a row to
2824 * splice() - or others? - can result in confusion:
2825 * so don't attempt that optimization on pipes etc.
2827 ret
= iov_iter_zero(nr
, to
);
2832 index
+= offset
>> PAGE_SHIFT
;
2833 offset
&= ~PAGE_MASK
;
2835 if (!iov_iter_count(to
))
2844 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2845 file_accessed(file
);
2846 return retval
? retval
: error
;
2849 static ssize_t
shmem_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2851 struct file
*file
= iocb
->ki_filp
;
2852 struct inode
*inode
= file
->f_mapping
->host
;
2856 ret
= generic_write_checks(iocb
, from
);
2859 ret
= file_remove_privs(file
);
2862 ret
= file_update_time(file
);
2865 ret
= generic_perform_write(iocb
, from
);
2867 inode_unlock(inode
);
2871 static bool zero_pipe_buf_get(struct pipe_inode_info
*pipe
,
2872 struct pipe_buffer
*buf
)
2877 static void zero_pipe_buf_release(struct pipe_inode_info
*pipe
,
2878 struct pipe_buffer
*buf
)
2882 static bool zero_pipe_buf_try_steal(struct pipe_inode_info
*pipe
,
2883 struct pipe_buffer
*buf
)
2888 static const struct pipe_buf_operations zero_pipe_buf_ops
= {
2889 .release
= zero_pipe_buf_release
,
2890 .try_steal
= zero_pipe_buf_try_steal
,
2891 .get
= zero_pipe_buf_get
,
2894 static size_t splice_zeropage_into_pipe(struct pipe_inode_info
*pipe
,
2895 loff_t fpos
, size_t size
)
2897 size_t offset
= fpos
& ~PAGE_MASK
;
2899 size
= min_t(size_t, size
, PAGE_SIZE
- offset
);
2901 if (!pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
)) {
2902 struct pipe_buffer
*buf
= pipe_head_buf(pipe
);
2904 *buf
= (struct pipe_buffer
) {
2905 .ops
= &zero_pipe_buf_ops
,
2906 .page
= ZERO_PAGE(0),
2916 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
2917 struct pipe_inode_info
*pipe
,
2918 size_t len
, unsigned int flags
)
2920 struct inode
*inode
= file_inode(in
);
2921 struct address_space
*mapping
= inode
->i_mapping
;
2922 struct folio
*folio
= NULL
;
2923 size_t total_spliced
= 0, used
, npages
, n
, part
;
2927 /* Work out how much data we can actually add into the pipe */
2928 used
= pipe_occupancy(pipe
->head
, pipe
->tail
);
2929 npages
= max_t(ssize_t
, pipe
->max_usage
- used
, 0);
2930 len
= min_t(size_t, len
, npages
* PAGE_SIZE
);
2933 if (*ppos
>= i_size_read(inode
))
2936 error
= shmem_get_folio(inode
, *ppos
/ PAGE_SIZE
, &folio
,
2939 if (error
== -EINVAL
)
2944 folio_unlock(folio
);
2946 if (folio_test_hwpoison(folio
) ||
2947 (folio_test_large(folio
) &&
2948 folio_test_has_hwpoisoned(folio
))) {
2955 * i_size must be checked after we know the pages are Uptodate.
2957 * Checking i_size after the check allows us to calculate
2958 * the correct value for "nr", which means the zero-filled
2959 * part of the page is not copied back to userspace (unless
2960 * another truncate extends the file - this is desired though).
2962 isize
= i_size_read(inode
);
2963 if (unlikely(*ppos
>= isize
))
2965 part
= min_t(loff_t
, isize
- *ppos
, len
);
2969 * If users can be writing to this page using arbitrary
2970 * virtual addresses, take care about potential aliasing
2971 * before reading the page on the kernel side.
2973 if (mapping_writably_mapped(mapping
))
2974 flush_dcache_folio(folio
);
2975 folio_mark_accessed(folio
);
2977 * Ok, we have the page, and it's up-to-date, so we can
2978 * now splice it into the pipe.
2980 n
= splice_folio_into_pipe(pipe
, folio
, *ppos
, part
);
2984 n
= splice_zeropage_into_pipe(pipe
, *ppos
, part
);
2992 in
->f_ra
.prev_pos
= *ppos
;
2993 if (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
3003 return total_spliced
? total_spliced
: error
;
3006 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
3008 struct address_space
*mapping
= file
->f_mapping
;
3009 struct inode
*inode
= mapping
->host
;
3011 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
3012 return generic_file_llseek_size(file
, offset
, whence
,
3013 MAX_LFS_FILESIZE
, i_size_read(inode
));
3018 /* We're holding i_rwsem so we can access i_size directly */
3019 offset
= mapping_seek_hole_data(mapping
, offset
, inode
->i_size
, whence
);
3021 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
3022 inode_unlock(inode
);
3026 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
3029 struct inode
*inode
= file_inode(file
);
3030 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3031 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3032 struct shmem_falloc shmem_falloc
;
3033 pgoff_t start
, index
, end
, undo_fallocend
;
3036 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
3041 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
3042 struct address_space
*mapping
= file
->f_mapping
;
3043 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
3044 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
3045 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
3047 /* protected by i_rwsem */
3048 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
3053 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
3054 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
3055 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
3056 spin_lock(&inode
->i_lock
);
3057 inode
->i_private
= &shmem_falloc
;
3058 spin_unlock(&inode
->i_lock
);
3060 if ((u64
)unmap_end
> (u64
)unmap_start
)
3061 unmap_mapping_range(mapping
, unmap_start
,
3062 1 + unmap_end
- unmap_start
, 0);
3063 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
3064 /* No need to unmap again: hole-punching leaves COWed pages */
3066 spin_lock(&inode
->i_lock
);
3067 inode
->i_private
= NULL
;
3068 wake_up_all(&shmem_falloc_waitq
);
3069 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
3070 spin_unlock(&inode
->i_lock
);
3075 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3076 error
= inode_newsize_ok(inode
, offset
+ len
);
3080 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
3085 start
= offset
>> PAGE_SHIFT
;
3086 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
3087 /* Try to avoid a swapstorm if len is impossible to satisfy */
3088 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
3093 shmem_falloc
.waitq
= NULL
;
3094 shmem_falloc
.start
= start
;
3095 shmem_falloc
.next
= start
;
3096 shmem_falloc
.nr_falloced
= 0;
3097 shmem_falloc
.nr_unswapped
= 0;
3098 spin_lock(&inode
->i_lock
);
3099 inode
->i_private
= &shmem_falloc
;
3100 spin_unlock(&inode
->i_lock
);
3103 * info->fallocend is only relevant when huge pages might be
3104 * involved: to prevent split_huge_page() freeing fallocated
3105 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3107 undo_fallocend
= info
->fallocend
;
3108 if (info
->fallocend
< end
)
3109 info
->fallocend
= end
;
3111 for (index
= start
; index
< end
; ) {
3112 struct folio
*folio
;
3115 * Good, the fallocate(2) manpage permits EINTR: we may have
3116 * been interrupted because we are using up too much memory.
3118 if (signal_pending(current
))
3120 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
3123 error
= shmem_get_folio(inode
, index
, &folio
,
3126 info
->fallocend
= undo_fallocend
;
3127 /* Remove the !uptodate folios we added */
3128 if (index
> start
) {
3129 shmem_undo_range(inode
,
3130 (loff_t
)start
<< PAGE_SHIFT
,
3131 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
3137 * Here is a more important optimization than it appears:
3138 * a second SGP_FALLOC on the same large folio will clear it,
3139 * making it uptodate and un-undoable if we fail later.
3141 index
= folio_next_index(folio
);
3142 /* Beware 32-bit wraparound */
3147 * Inform shmem_writepage() how far we have reached.
3148 * No need for lock or barrier: we have the page lock.
3150 if (!folio_test_uptodate(folio
))
3151 shmem_falloc
.nr_falloced
+= index
- shmem_falloc
.next
;
3152 shmem_falloc
.next
= index
;
3155 * If !uptodate, leave it that way so that freeable folios
3156 * can be recognized if we need to rollback on error later.
3157 * But mark it dirty so that memory pressure will swap rather
3158 * than free the folios we are allocating (and SGP_CACHE folios
3159 * might still be clean: we now need to mark those dirty too).
3161 folio_mark_dirty(folio
);
3162 folio_unlock(folio
);
3167 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
3168 i_size_write(inode
, offset
+ len
);
3170 spin_lock(&inode
->i_lock
);
3171 inode
->i_private
= NULL
;
3172 spin_unlock(&inode
->i_lock
);
3175 file_modified(file
);
3176 inode_unlock(inode
);
3180 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
3182 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
3184 buf
->f_type
= TMPFS_MAGIC
;
3185 buf
->f_bsize
= PAGE_SIZE
;
3186 buf
->f_namelen
= NAME_MAX
;
3187 if (sbinfo
->max_blocks
) {
3188 buf
->f_blocks
= sbinfo
->max_blocks
;
3190 buf
->f_bfree
= sbinfo
->max_blocks
-
3191 percpu_counter_sum(&sbinfo
->used_blocks
);
3193 if (sbinfo
->max_inodes
) {
3194 buf
->f_files
= sbinfo
->max_inodes
;
3195 buf
->f_ffree
= sbinfo
->free_ispace
/ BOGO_INODE_SIZE
;
3197 /* else leave those fields 0 like simple_statfs */
3199 buf
->f_fsid
= uuid_to_fsid(dentry
->d_sb
->s_uuid
.b
);
3205 * File creation. Allocate an inode, and we're done..
3208 shmem_mknod(struct mnt_idmap
*idmap
, struct inode
*dir
,
3209 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3211 struct inode
*inode
;
3214 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
3216 return PTR_ERR(inode
);
3218 error
= simple_acl_create(dir
, inode
);
3221 error
= security_inode_init_security(inode
, dir
,
3223 shmem_initxattrs
, NULL
);
3224 if (error
&& error
!= -EOPNOTSUPP
)
3227 error
= simple_offset_add(shmem_get_offset_ctx(dir
), dentry
);
3231 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3232 dir
->i_mtime
= inode_set_ctime_current(dir
);
3233 inode_inc_iversion(dir
);
3234 d_instantiate(dentry
, inode
);
3235 dget(dentry
); /* Extra count - pin the dentry in core */
3244 shmem_tmpfile(struct mnt_idmap
*idmap
, struct inode
*dir
,
3245 struct file
*file
, umode_t mode
)
3247 struct inode
*inode
;
3250 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
3252 if (IS_ERR(inode
)) {
3253 error
= PTR_ERR(inode
);
3257 error
= security_inode_init_security(inode
, dir
,
3259 shmem_initxattrs
, NULL
);
3260 if (error
&& error
!= -EOPNOTSUPP
)
3262 error
= simple_acl_create(dir
, inode
);
3265 d_tmpfile(file
, inode
);
3268 return finish_open_simple(file
, error
);
3274 static int shmem_mkdir(struct mnt_idmap
*idmap
, struct inode
*dir
,
3275 struct dentry
*dentry
, umode_t mode
)
3279 error
= shmem_mknod(idmap
, dir
, dentry
, mode
| S_IFDIR
, 0);
3286 static int shmem_create(struct mnt_idmap
*idmap
, struct inode
*dir
,
3287 struct dentry
*dentry
, umode_t mode
, bool excl
)
3289 return shmem_mknod(idmap
, dir
, dentry
, mode
| S_IFREG
, 0);
3295 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
3297 struct inode
*inode
= d_inode(old_dentry
);
3301 * No ordinary (disk based) filesystem counts links as inodes;
3302 * but each new link needs a new dentry, pinning lowmem, and
3303 * tmpfs dentries cannot be pruned until they are unlinked.
3304 * But if an O_TMPFILE file is linked into the tmpfs, the
3305 * first link must skip that, to get the accounting right.
3307 if (inode
->i_nlink
) {
3308 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
3313 ret
= simple_offset_add(shmem_get_offset_ctx(dir
), dentry
);
3316 shmem_free_inode(inode
->i_sb
, 0);
3320 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3321 dir
->i_mtime
= inode_set_ctime_to_ts(dir
,
3322 inode_set_ctime_current(inode
));
3323 inode_inc_iversion(dir
);
3325 ihold(inode
); /* New dentry reference */
3326 dget(dentry
); /* Extra pinning count for the created dentry */
3327 d_instantiate(dentry
, inode
);
3332 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3334 struct inode
*inode
= d_inode(dentry
);
3336 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3337 shmem_free_inode(inode
->i_sb
, 0);
3339 simple_offset_remove(shmem_get_offset_ctx(dir
), dentry
);
3341 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3342 dir
->i_mtime
= inode_set_ctime_to_ts(dir
,
3343 inode_set_ctime_current(inode
));
3344 inode_inc_iversion(dir
);
3346 dput(dentry
); /* Undo the count from "create" - this does all the work */
3350 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3352 if (!simple_empty(dentry
))
3355 drop_nlink(d_inode(dentry
));
3357 return shmem_unlink(dir
, dentry
);
3360 static int shmem_whiteout(struct mnt_idmap
*idmap
,
3361 struct inode
*old_dir
, struct dentry
*old_dentry
)
3363 struct dentry
*whiteout
;
3366 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3370 error
= shmem_mknod(idmap
, old_dir
, whiteout
,
3371 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3377 * Cheat and hash the whiteout while the old dentry is still in
3378 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3380 * d_lookup() will consistently find one of them at this point,
3381 * not sure which one, but that isn't even important.
3388 * The VFS layer already does all the dentry stuff for rename,
3389 * we just have to decrement the usage count for the target if
3390 * it exists so that the VFS layer correctly free's it when it
3393 static int shmem_rename2(struct mnt_idmap
*idmap
,
3394 struct inode
*old_dir
, struct dentry
*old_dentry
,
3395 struct inode
*new_dir
, struct dentry
*new_dentry
,
3398 struct inode
*inode
= d_inode(old_dentry
);
3399 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3402 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3405 if (flags
& RENAME_EXCHANGE
)
3406 return simple_offset_rename_exchange(old_dir
, old_dentry
,
3407 new_dir
, new_dentry
);
3409 if (!simple_empty(new_dentry
))
3412 if (flags
& RENAME_WHITEOUT
) {
3413 error
= shmem_whiteout(idmap
, old_dir
, old_dentry
);
3418 simple_offset_remove(shmem_get_offset_ctx(old_dir
), old_dentry
);
3419 error
= simple_offset_add(shmem_get_offset_ctx(new_dir
), old_dentry
);
3423 if (d_really_is_positive(new_dentry
)) {
3424 (void) shmem_unlink(new_dir
, new_dentry
);
3425 if (they_are_dirs
) {
3426 drop_nlink(d_inode(new_dentry
));
3427 drop_nlink(old_dir
);
3429 } else if (they_are_dirs
) {
3430 drop_nlink(old_dir
);
3434 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3435 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3436 simple_rename_timestamp(old_dir
, old_dentry
, new_dir
, new_dentry
);
3437 inode_inc_iversion(old_dir
);
3438 inode_inc_iversion(new_dir
);
3442 static int shmem_symlink(struct mnt_idmap
*idmap
, struct inode
*dir
,
3443 struct dentry
*dentry
, const char *symname
)
3447 struct inode
*inode
;
3448 struct folio
*folio
;
3450 len
= strlen(symname
) + 1;
3451 if (len
> PAGE_SIZE
)
3452 return -ENAMETOOLONG
;
3454 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3458 return PTR_ERR(inode
);
3460 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3461 shmem_initxattrs
, NULL
);
3462 if (error
&& error
!= -EOPNOTSUPP
)
3465 error
= simple_offset_add(shmem_get_offset_ctx(dir
), dentry
);
3469 inode
->i_size
= len
-1;
3470 if (len
<= SHORT_SYMLINK_LEN
) {
3471 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3472 if (!inode
->i_link
) {
3474 goto out_remove_offset
;
3476 inode
->i_op
= &shmem_short_symlink_operations
;
3478 inode_nohighmem(inode
);
3479 error
= shmem_get_folio(inode
, 0, &folio
, SGP_WRITE
);
3481 goto out_remove_offset
;
3482 inode
->i_mapping
->a_ops
= &shmem_aops
;
3483 inode
->i_op
= &shmem_symlink_inode_operations
;
3484 memcpy(folio_address(folio
), symname
, len
);
3485 folio_mark_uptodate(folio
);
3486 folio_mark_dirty(folio
);
3487 folio_unlock(folio
);
3490 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3491 dir
->i_mtime
= inode_set_ctime_current(dir
);
3492 inode_inc_iversion(dir
);
3493 d_instantiate(dentry
, inode
);
3498 simple_offset_remove(shmem_get_offset_ctx(dir
), dentry
);
3504 static void shmem_put_link(void *arg
)
3506 folio_mark_accessed(arg
);
3510 static const char *shmem_get_link(struct dentry
*dentry
,
3511 struct inode
*inode
,
3512 struct delayed_call
*done
)
3514 struct folio
*folio
= NULL
;
3518 folio
= filemap_get_folio(inode
->i_mapping
, 0);
3520 return ERR_PTR(-ECHILD
);
3521 if (PageHWPoison(folio_page(folio
, 0)) ||
3522 !folio_test_uptodate(folio
)) {
3524 return ERR_PTR(-ECHILD
);
3527 error
= shmem_get_folio(inode
, 0, &folio
, SGP_READ
);
3529 return ERR_PTR(error
);
3531 return ERR_PTR(-ECHILD
);
3532 if (PageHWPoison(folio_page(folio
, 0))) {
3533 folio_unlock(folio
);
3535 return ERR_PTR(-ECHILD
);
3537 folio_unlock(folio
);
3539 set_delayed_call(done
, shmem_put_link
, folio
);
3540 return folio_address(folio
);
3543 #ifdef CONFIG_TMPFS_XATTR
3545 static int shmem_fileattr_get(struct dentry
*dentry
, struct fileattr
*fa
)
3547 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3549 fileattr_fill_flags(fa
, info
->fsflags
& SHMEM_FL_USER_VISIBLE
);
3554 static int shmem_fileattr_set(struct mnt_idmap
*idmap
,
3555 struct dentry
*dentry
, struct fileattr
*fa
)
3557 struct inode
*inode
= d_inode(dentry
);
3558 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3560 if (fileattr_has_fsx(fa
))
3562 if (fa
->flags
& ~SHMEM_FL_USER_MODIFIABLE
)
3565 info
->fsflags
= (info
->fsflags
& ~SHMEM_FL_USER_MODIFIABLE
) |
3566 (fa
->flags
& SHMEM_FL_USER_MODIFIABLE
);
3568 shmem_set_inode_flags(inode
, info
->fsflags
);
3569 inode_set_ctime_current(inode
);
3570 inode_inc_iversion(inode
);
3575 * Superblocks without xattr inode operations may get some security.* xattr
3576 * support from the LSM "for free". As soon as we have any other xattrs
3577 * like ACLs, we also need to implement the security.* handlers at
3578 * filesystem level, though.
3582 * Callback for security_inode_init_security() for acquiring xattrs.
3584 static int shmem_initxattrs(struct inode
*inode
,
3585 const struct xattr
*xattr_array
,
3588 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3589 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3590 const struct xattr
*xattr
;
3591 struct simple_xattr
*new_xattr
;
3595 if (sbinfo
->max_inodes
) {
3596 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3597 ispace
+= simple_xattr_space(xattr
->name
,
3598 xattr
->value_len
+ XATTR_SECURITY_PREFIX_LEN
);
3601 raw_spin_lock(&sbinfo
->stat_lock
);
3602 if (sbinfo
->free_ispace
< ispace
)
3605 sbinfo
->free_ispace
-= ispace
;
3606 raw_spin_unlock(&sbinfo
->stat_lock
);
3612 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3613 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3617 len
= strlen(xattr
->name
) + 1;
3618 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3619 GFP_KERNEL_ACCOUNT
);
3620 if (!new_xattr
->name
) {
3625 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3626 XATTR_SECURITY_PREFIX_LEN
);
3627 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3630 simple_xattr_add(&info
->xattrs
, new_xattr
);
3633 if (xattr
->name
!= NULL
) {
3635 raw_spin_lock(&sbinfo
->stat_lock
);
3636 sbinfo
->free_ispace
+= ispace
;
3637 raw_spin_unlock(&sbinfo
->stat_lock
);
3639 simple_xattrs_free(&info
->xattrs
, NULL
);
3646 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3647 struct dentry
*unused
, struct inode
*inode
,
3648 const char *name
, void *buffer
, size_t size
)
3650 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3652 name
= xattr_full_name(handler
, name
);
3653 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3656 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3657 struct mnt_idmap
*idmap
,
3658 struct dentry
*unused
, struct inode
*inode
,
3659 const char *name
, const void *value
,
3660 size_t size
, int flags
)
3662 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3663 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
3664 struct simple_xattr
*old_xattr
;
3667 name
= xattr_full_name(handler
, name
);
3668 if (value
&& sbinfo
->max_inodes
) {
3669 ispace
= simple_xattr_space(name
, size
);
3670 raw_spin_lock(&sbinfo
->stat_lock
);
3671 if (sbinfo
->free_ispace
< ispace
)
3674 sbinfo
->free_ispace
-= ispace
;
3675 raw_spin_unlock(&sbinfo
->stat_lock
);
3680 old_xattr
= simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3681 if (!IS_ERR(old_xattr
)) {
3683 if (old_xattr
&& sbinfo
->max_inodes
)
3684 ispace
= simple_xattr_space(old_xattr
->name
,
3686 simple_xattr_free(old_xattr
);
3688 inode_set_ctime_current(inode
);
3689 inode_inc_iversion(inode
);
3692 raw_spin_lock(&sbinfo
->stat_lock
);
3693 sbinfo
->free_ispace
+= ispace
;
3694 raw_spin_unlock(&sbinfo
->stat_lock
);
3696 return PTR_ERR(old_xattr
);
3699 static const struct xattr_handler shmem_security_xattr_handler
= {
3700 .prefix
= XATTR_SECURITY_PREFIX
,
3701 .get
= shmem_xattr_handler_get
,
3702 .set
= shmem_xattr_handler_set
,
3705 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3706 .prefix
= XATTR_TRUSTED_PREFIX
,
3707 .get
= shmem_xattr_handler_get
,
3708 .set
= shmem_xattr_handler_set
,
3711 static const struct xattr_handler shmem_user_xattr_handler
= {
3712 .prefix
= XATTR_USER_PREFIX
,
3713 .get
= shmem_xattr_handler_get
,
3714 .set
= shmem_xattr_handler_set
,
3717 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3718 &shmem_security_xattr_handler
,
3719 &shmem_trusted_xattr_handler
,
3720 &shmem_user_xattr_handler
,
3724 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3726 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3727 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3729 #endif /* CONFIG_TMPFS_XATTR */
3731 static const struct inode_operations shmem_short_symlink_operations
= {
3732 .getattr
= shmem_getattr
,
3733 .setattr
= shmem_setattr
,
3734 .get_link
= simple_get_link
,
3735 #ifdef CONFIG_TMPFS_XATTR
3736 .listxattr
= shmem_listxattr
,
3740 static const struct inode_operations shmem_symlink_inode_operations
= {
3741 .getattr
= shmem_getattr
,
3742 .setattr
= shmem_setattr
,
3743 .get_link
= shmem_get_link
,
3744 #ifdef CONFIG_TMPFS_XATTR
3745 .listxattr
= shmem_listxattr
,
3749 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3751 return ERR_PTR(-ESTALE
);
3754 static int shmem_match(struct inode
*ino
, void *vfh
)
3758 inum
= (inum
<< 32) | fh
[1];
3759 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3762 /* Find any alias of inode, but prefer a hashed alias */
3763 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3765 struct dentry
*alias
= d_find_alias(inode
);
3767 return alias
?: d_find_any_alias(inode
);
3771 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3772 struct fid
*fid
, int fh_len
, int fh_type
)
3774 struct inode
*inode
;
3775 struct dentry
*dentry
= NULL
;
3782 inum
= (inum
<< 32) | fid
->raw
[1];
3784 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3785 shmem_match
, fid
->raw
);
3787 dentry
= shmem_find_alias(inode
);
3794 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3795 struct inode
*parent
)
3799 return FILEID_INVALID
;
3802 if (inode_unhashed(inode
)) {
3803 /* Unfortunately insert_inode_hash is not idempotent,
3804 * so as we hash inodes here rather than at creation
3805 * time, we need a lock to ensure we only try
3808 static DEFINE_SPINLOCK(lock
);
3810 if (inode_unhashed(inode
))
3811 __insert_inode_hash(inode
,
3812 inode
->i_ino
+ inode
->i_generation
);
3816 fh
[0] = inode
->i_generation
;
3817 fh
[1] = inode
->i_ino
;
3818 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3824 static const struct export_operations shmem_export_ops
= {
3825 .get_parent
= shmem_get_parent
,
3826 .encode_fh
= shmem_encode_fh
,
3827 .fh_to_dentry
= shmem_fh_to_dentry
,
3845 Opt_usrquota_block_hardlimit
,
3846 Opt_usrquota_inode_hardlimit
,
3847 Opt_grpquota_block_hardlimit
,
3848 Opt_grpquota_inode_hardlimit
,
3851 static const struct constant_table shmem_param_enums_huge
[] = {
3852 {"never", SHMEM_HUGE_NEVER
},
3853 {"always", SHMEM_HUGE_ALWAYS
},
3854 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3855 {"advise", SHMEM_HUGE_ADVISE
},
3859 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3860 fsparam_u32 ("gid", Opt_gid
),
3861 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3862 fsparam_u32oct("mode", Opt_mode
),
3863 fsparam_string("mpol", Opt_mpol
),
3864 fsparam_string("nr_blocks", Opt_nr_blocks
),
3865 fsparam_string("nr_inodes", Opt_nr_inodes
),
3866 fsparam_string("size", Opt_size
),
3867 fsparam_u32 ("uid", Opt_uid
),
3868 fsparam_flag ("inode32", Opt_inode32
),
3869 fsparam_flag ("inode64", Opt_inode64
),
3870 fsparam_flag ("noswap", Opt_noswap
),
3871 #ifdef CONFIG_TMPFS_QUOTA
3872 fsparam_flag ("quota", Opt_quota
),
3873 fsparam_flag ("usrquota", Opt_usrquota
),
3874 fsparam_flag ("grpquota", Opt_grpquota
),
3875 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit
),
3876 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit
),
3877 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit
),
3878 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit
),
3883 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3885 struct shmem_options
*ctx
= fc
->fs_private
;
3886 struct fs_parse_result result
;
3887 unsigned long long size
;
3893 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3899 size
= memparse(param
->string
, &rest
);
3901 size
<<= PAGE_SHIFT
;
3902 size
*= totalram_pages();
3908 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3909 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3912 ctx
->blocks
= memparse(param
->string
, &rest
);
3913 if (*rest
|| ctx
->blocks
> LONG_MAX
)
3915 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3918 ctx
->inodes
= memparse(param
->string
, &rest
);
3919 if (*rest
|| ctx
->inodes
> ULONG_MAX
/ BOGO_INODE_SIZE
)
3921 ctx
->seen
|= SHMEM_SEEN_INODES
;
3924 ctx
->mode
= result
.uint_32
& 07777;
3927 kuid
= make_kuid(current_user_ns(), result
.uint_32
);
3928 if (!uid_valid(kuid
))
3932 * The requested uid must be representable in the
3933 * filesystem's idmapping.
3935 if (!kuid_has_mapping(fc
->user_ns
, kuid
))
3941 kgid
= make_kgid(current_user_ns(), result
.uint_32
);
3942 if (!gid_valid(kgid
))
3946 * The requested gid must be representable in the
3947 * filesystem's idmapping.
3949 if (!kgid_has_mapping(fc
->user_ns
, kgid
))
3955 ctx
->huge
= result
.uint_32
;
3956 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3957 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3958 has_transparent_hugepage()))
3959 goto unsupported_parameter
;
3960 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3963 if (IS_ENABLED(CONFIG_NUMA
)) {
3964 mpol_put(ctx
->mpol
);
3966 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3970 goto unsupported_parameter
;
3972 ctx
->full_inums
= false;
3973 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3976 if (sizeof(ino_t
) < 8) {
3978 "Cannot use inode64 with <64bit inums in kernel\n");
3980 ctx
->full_inums
= true;
3981 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3984 if ((fc
->user_ns
!= &init_user_ns
) || !capable(CAP_SYS_ADMIN
)) {
3986 "Turning off swap in unprivileged tmpfs mounts unsupported");
3989 ctx
->seen
|= SHMEM_SEEN_NOSWAP
;
3992 if (fc
->user_ns
!= &init_user_ns
)
3993 return invalfc(fc
, "Quotas in unprivileged tmpfs mounts are unsupported");
3994 ctx
->seen
|= SHMEM_SEEN_QUOTA
;
3995 ctx
->quota_types
|= (QTYPE_MASK_USR
| QTYPE_MASK_GRP
);
3998 if (fc
->user_ns
!= &init_user_ns
)
3999 return invalfc(fc
, "Quotas in unprivileged tmpfs mounts are unsupported");
4000 ctx
->seen
|= SHMEM_SEEN_QUOTA
;
4001 ctx
->quota_types
|= QTYPE_MASK_USR
;
4004 if (fc
->user_ns
!= &init_user_ns
)
4005 return invalfc(fc
, "Quotas in unprivileged tmpfs mounts are unsupported");
4006 ctx
->seen
|= SHMEM_SEEN_QUOTA
;
4007 ctx
->quota_types
|= QTYPE_MASK_GRP
;
4009 case Opt_usrquota_block_hardlimit
:
4010 size
= memparse(param
->string
, &rest
);
4013 if (size
> SHMEM_QUOTA_MAX_SPC_LIMIT
)
4015 "User quota block hardlimit too large.");
4016 ctx
->qlimits
.usrquota_bhardlimit
= size
;
4018 case Opt_grpquota_block_hardlimit
:
4019 size
= memparse(param
->string
, &rest
);
4022 if (size
> SHMEM_QUOTA_MAX_SPC_LIMIT
)
4024 "Group quota block hardlimit too large.");
4025 ctx
->qlimits
.grpquota_bhardlimit
= size
;
4027 case Opt_usrquota_inode_hardlimit
:
4028 size
= memparse(param
->string
, &rest
);
4031 if (size
> SHMEM_QUOTA_MAX_INO_LIMIT
)
4033 "User quota inode hardlimit too large.");
4034 ctx
->qlimits
.usrquota_ihardlimit
= size
;
4036 case Opt_grpquota_inode_hardlimit
:
4037 size
= memparse(param
->string
, &rest
);
4040 if (size
> SHMEM_QUOTA_MAX_INO_LIMIT
)
4042 "Group quota inode hardlimit too large.");
4043 ctx
->qlimits
.grpquota_ihardlimit
= size
;
4048 unsupported_parameter
:
4049 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
4051 return invalfc(fc
, "Bad value for '%s'", param
->key
);
4054 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
4056 char *options
= data
;
4059 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
4064 while (options
!= NULL
) {
4065 char *this_char
= options
;
4068 * NUL-terminate this option: unfortunately,
4069 * mount options form a comma-separated list,
4070 * but mpol's nodelist may also contain commas.
4072 options
= strchr(options
, ',');
4073 if (options
== NULL
)
4076 if (!isdigit(*options
)) {
4082 char *value
= strchr(this_char
, '=');
4088 len
= strlen(value
);
4090 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
4099 * Reconfigure a shmem filesystem.
4101 static int shmem_reconfigure(struct fs_context
*fc
)
4103 struct shmem_options
*ctx
= fc
->fs_private
;
4104 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
4105 unsigned long used_isp
;
4106 struct mempolicy
*mpol
= NULL
;
4109 raw_spin_lock(&sbinfo
->stat_lock
);
4110 used_isp
= sbinfo
->max_inodes
* BOGO_INODE_SIZE
- sbinfo
->free_ispace
;
4112 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
4113 if (!sbinfo
->max_blocks
) {
4114 err
= "Cannot retroactively limit size";
4117 if (percpu_counter_compare(&sbinfo
->used_blocks
,
4119 err
= "Too small a size for current use";
4123 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
4124 if (!sbinfo
->max_inodes
) {
4125 err
= "Cannot retroactively limit inodes";
4128 if (ctx
->inodes
* BOGO_INODE_SIZE
< used_isp
) {
4129 err
= "Too few inodes for current use";
4134 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
4135 sbinfo
->next_ino
> UINT_MAX
) {
4136 err
= "Current inum too high to switch to 32-bit inums";
4139 if ((ctx
->seen
& SHMEM_SEEN_NOSWAP
) && ctx
->noswap
&& !sbinfo
->noswap
) {
4140 err
= "Cannot disable swap on remount";
4143 if (!(ctx
->seen
& SHMEM_SEEN_NOSWAP
) && !ctx
->noswap
&& sbinfo
->noswap
) {
4144 err
= "Cannot enable swap on remount if it was disabled on first mount";
4148 if (ctx
->seen
& SHMEM_SEEN_QUOTA
&&
4149 !sb_any_quota_loaded(fc
->root
->d_sb
)) {
4150 err
= "Cannot enable quota on remount";
4154 #ifdef CONFIG_TMPFS_QUOTA
4155 #define CHANGED_LIMIT(name) \
4156 (ctx->qlimits.name## hardlimit && \
4157 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4159 if (CHANGED_LIMIT(usrquota_b
) || CHANGED_LIMIT(usrquota_i
) ||
4160 CHANGED_LIMIT(grpquota_b
) || CHANGED_LIMIT(grpquota_i
)) {
4161 err
= "Cannot change global quota limit on remount";
4164 #endif /* CONFIG_TMPFS_QUOTA */
4166 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
4167 sbinfo
->huge
= ctx
->huge
;
4168 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
4169 sbinfo
->full_inums
= ctx
->full_inums
;
4170 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
4171 sbinfo
->max_blocks
= ctx
->blocks
;
4172 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
4173 sbinfo
->max_inodes
= ctx
->inodes
;
4174 sbinfo
->free_ispace
= ctx
->inodes
* BOGO_INODE_SIZE
- used_isp
;
4178 * Preserve previous mempolicy unless mpol remount option was specified.
4181 mpol
= sbinfo
->mpol
;
4182 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
4187 sbinfo
->noswap
= true;
4189 raw_spin_unlock(&sbinfo
->stat_lock
);
4193 raw_spin_unlock(&sbinfo
->stat_lock
);
4194 return invalfc(fc
, "%s", err
);
4197 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
4199 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
4200 struct mempolicy
*mpol
;
4202 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
4203 seq_printf(seq
, ",size=%luk", K(sbinfo
->max_blocks
));
4204 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
4205 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
4206 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
4207 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
4208 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
4209 seq_printf(seq
, ",uid=%u",
4210 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
4211 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
4212 seq_printf(seq
, ",gid=%u",
4213 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
4216 * Showing inode{64,32} might be useful even if it's the system default,
4217 * since then people don't have to resort to checking both here and
4218 * /proc/config.gz to confirm 64-bit inums were successfully applied
4219 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4221 * We hide it when inode64 isn't the default and we are using 32-bit
4222 * inodes, since that probably just means the feature isn't even under
4227 * +-----------------+-----------------+
4228 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4229 * +------------------+-----------------+-----------------+
4230 * | full_inums=true | show | show |
4231 * | full_inums=false | show | hide |
4232 * +------------------+-----------------+-----------------+
4235 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
4236 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
4237 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4238 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4240 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
4242 mpol
= shmem_get_sbmpol(sbinfo
);
4243 shmem_show_mpol(seq
, mpol
);
4246 seq_printf(seq
, ",noswap");
4250 #endif /* CONFIG_TMPFS */
4252 static void shmem_put_super(struct super_block
*sb
)
4254 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
4256 #ifdef CONFIG_TMPFS_QUOTA
4257 shmem_disable_quotas(sb
);
4259 free_percpu(sbinfo
->ino_batch
);
4260 percpu_counter_destroy(&sbinfo
->used_blocks
);
4261 mpol_put(sbinfo
->mpol
);
4263 sb
->s_fs_info
= NULL
;
4266 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
4268 struct shmem_options
*ctx
= fc
->fs_private
;
4269 struct inode
*inode
;
4270 struct shmem_sb_info
*sbinfo
;
4271 int error
= -ENOMEM
;
4273 /* Round up to L1_CACHE_BYTES to resist false sharing */
4274 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
4275 L1_CACHE_BYTES
), GFP_KERNEL
);
4279 sb
->s_fs_info
= sbinfo
;
4283 * Per default we only allow half of the physical ram per
4284 * tmpfs instance, limiting inodes to one per page of lowmem;
4285 * but the internal instance is left unlimited.
4287 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
4288 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
4289 ctx
->blocks
= shmem_default_max_blocks();
4290 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
4291 ctx
->inodes
= shmem_default_max_inodes();
4292 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
4293 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
4294 sbinfo
->noswap
= ctx
->noswap
;
4296 sb
->s_flags
|= SB_NOUSER
;
4298 sb
->s_export_op
= &shmem_export_ops
;
4299 sb
->s_flags
|= SB_NOSEC
| SB_I_VERSION
;
4301 sb
->s_flags
|= SB_NOUSER
;
4303 sbinfo
->max_blocks
= ctx
->blocks
;
4304 sbinfo
->max_inodes
= ctx
->inodes
;
4305 sbinfo
->free_ispace
= sbinfo
->max_inodes
* BOGO_INODE_SIZE
;
4306 if (sb
->s_flags
& SB_KERNMOUNT
) {
4307 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
4308 if (!sbinfo
->ino_batch
)
4311 sbinfo
->uid
= ctx
->uid
;
4312 sbinfo
->gid
= ctx
->gid
;
4313 sbinfo
->full_inums
= ctx
->full_inums
;
4314 sbinfo
->mode
= ctx
->mode
;
4315 sbinfo
->huge
= ctx
->huge
;
4316 sbinfo
->mpol
= ctx
->mpol
;
4319 raw_spin_lock_init(&sbinfo
->stat_lock
);
4320 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
4322 spin_lock_init(&sbinfo
->shrinklist_lock
);
4323 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
4325 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
4326 sb
->s_blocksize
= PAGE_SIZE
;
4327 sb
->s_blocksize_bits
= PAGE_SHIFT
;
4328 sb
->s_magic
= TMPFS_MAGIC
;
4329 sb
->s_op
= &shmem_ops
;
4330 sb
->s_time_gran
= 1;
4331 #ifdef CONFIG_TMPFS_XATTR
4332 sb
->s_xattr
= shmem_xattr_handlers
;
4334 #ifdef CONFIG_TMPFS_POSIX_ACL
4335 sb
->s_flags
|= SB_POSIXACL
;
4337 uuid_gen(&sb
->s_uuid
);
4339 #ifdef CONFIG_TMPFS_QUOTA
4340 if (ctx
->seen
& SHMEM_SEEN_QUOTA
) {
4341 sb
->dq_op
= &shmem_quota_operations
;
4342 sb
->s_qcop
= &dquot_quotactl_sysfile_ops
;
4343 sb
->s_quota_types
= QTYPE_MASK_USR
| QTYPE_MASK_GRP
;
4345 /* Copy the default limits from ctx into sbinfo */
4346 memcpy(&sbinfo
->qlimits
, &ctx
->qlimits
,
4347 sizeof(struct shmem_quota_limits
));
4349 if (shmem_enable_quotas(sb
, ctx
->quota_types
))
4352 #endif /* CONFIG_TMPFS_QUOTA */
4354 inode
= shmem_get_inode(&nop_mnt_idmap
, sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0,
4356 if (IS_ERR(inode
)) {
4357 error
= PTR_ERR(inode
);
4360 inode
->i_uid
= sbinfo
->uid
;
4361 inode
->i_gid
= sbinfo
->gid
;
4362 sb
->s_root
= d_make_root(inode
);
4368 shmem_put_super(sb
);
4372 static int shmem_get_tree(struct fs_context
*fc
)
4374 return get_tree_nodev(fc
, shmem_fill_super
);
4377 static void shmem_free_fc(struct fs_context
*fc
)
4379 struct shmem_options
*ctx
= fc
->fs_private
;
4382 mpol_put(ctx
->mpol
);
4387 static const struct fs_context_operations shmem_fs_context_ops
= {
4388 .free
= shmem_free_fc
,
4389 .get_tree
= shmem_get_tree
,
4391 .parse_monolithic
= shmem_parse_options
,
4392 .parse_param
= shmem_parse_one
,
4393 .reconfigure
= shmem_reconfigure
,
4397 static struct kmem_cache
*shmem_inode_cachep
;
4399 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
4401 struct shmem_inode_info
*info
;
4402 info
= alloc_inode_sb(sb
, shmem_inode_cachep
, GFP_KERNEL
);
4405 return &info
->vfs_inode
;
4408 static void shmem_free_in_core_inode(struct inode
*inode
)
4410 if (S_ISLNK(inode
->i_mode
))
4411 kfree(inode
->i_link
);
4412 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
4415 static void shmem_destroy_inode(struct inode
*inode
)
4417 if (S_ISREG(inode
->i_mode
))
4418 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
4419 if (S_ISDIR(inode
->i_mode
))
4420 simple_offset_destroy(shmem_get_offset_ctx(inode
));
4423 static void shmem_init_inode(void *foo
)
4425 struct shmem_inode_info
*info
= foo
;
4426 inode_init_once(&info
->vfs_inode
);
4429 static void shmem_init_inodecache(void)
4431 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
4432 sizeof(struct shmem_inode_info
),
4433 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
4436 static void shmem_destroy_inodecache(void)
4438 kmem_cache_destroy(shmem_inode_cachep
);
4441 /* Keep the page in page cache instead of truncating it */
4442 static int shmem_error_remove_page(struct address_space
*mapping
,
4448 const struct address_space_operations shmem_aops
= {
4449 .writepage
= shmem_writepage
,
4450 .dirty_folio
= noop_dirty_folio
,
4452 .write_begin
= shmem_write_begin
,
4453 .write_end
= shmem_write_end
,
4455 #ifdef CONFIG_MIGRATION
4456 .migrate_folio
= migrate_folio
,
4458 .error_remove_page
= shmem_error_remove_page
,
4460 EXPORT_SYMBOL(shmem_aops
);
4462 static const struct file_operations shmem_file_operations
= {
4464 .open
= shmem_file_open
,
4465 .get_unmapped_area
= shmem_get_unmapped_area
,
4467 .llseek
= shmem_file_llseek
,
4468 .read_iter
= shmem_file_read_iter
,
4469 .write_iter
= shmem_file_write_iter
,
4470 .fsync
= noop_fsync
,
4471 .splice_read
= shmem_file_splice_read
,
4472 .splice_write
= iter_file_splice_write
,
4473 .fallocate
= shmem_fallocate
,
4477 static const struct inode_operations shmem_inode_operations
= {
4478 .getattr
= shmem_getattr
,
4479 .setattr
= shmem_setattr
,
4480 #ifdef CONFIG_TMPFS_XATTR
4481 .listxattr
= shmem_listxattr
,
4482 .set_acl
= simple_set_acl
,
4483 .fileattr_get
= shmem_fileattr_get
,
4484 .fileattr_set
= shmem_fileattr_set
,
4488 static const struct inode_operations shmem_dir_inode_operations
= {
4490 .getattr
= shmem_getattr
,
4491 .create
= shmem_create
,
4492 .lookup
= simple_lookup
,
4494 .unlink
= shmem_unlink
,
4495 .symlink
= shmem_symlink
,
4496 .mkdir
= shmem_mkdir
,
4497 .rmdir
= shmem_rmdir
,
4498 .mknod
= shmem_mknod
,
4499 .rename
= shmem_rename2
,
4500 .tmpfile
= shmem_tmpfile
,
4501 .get_offset_ctx
= shmem_get_offset_ctx
,
4503 #ifdef CONFIG_TMPFS_XATTR
4504 .listxattr
= shmem_listxattr
,
4505 .fileattr_get
= shmem_fileattr_get
,
4506 .fileattr_set
= shmem_fileattr_set
,
4508 #ifdef CONFIG_TMPFS_POSIX_ACL
4509 .setattr
= shmem_setattr
,
4510 .set_acl
= simple_set_acl
,
4514 static const struct inode_operations shmem_special_inode_operations
= {
4515 .getattr
= shmem_getattr
,
4516 #ifdef CONFIG_TMPFS_XATTR
4517 .listxattr
= shmem_listxattr
,
4519 #ifdef CONFIG_TMPFS_POSIX_ACL
4520 .setattr
= shmem_setattr
,
4521 .set_acl
= simple_set_acl
,
4525 static const struct super_operations shmem_ops
= {
4526 .alloc_inode
= shmem_alloc_inode
,
4527 .free_inode
= shmem_free_in_core_inode
,
4528 .destroy_inode
= shmem_destroy_inode
,
4530 .statfs
= shmem_statfs
,
4531 .show_options
= shmem_show_options
,
4533 #ifdef CONFIG_TMPFS_QUOTA
4534 .get_dquots
= shmem_get_dquots
,
4536 .evict_inode
= shmem_evict_inode
,
4537 .drop_inode
= generic_delete_inode
,
4538 .put_super
= shmem_put_super
,
4539 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4540 .nr_cached_objects
= shmem_unused_huge_count
,
4541 .free_cached_objects
= shmem_unused_huge_scan
,
4545 static const struct vm_operations_struct shmem_vm_ops
= {
4546 .fault
= shmem_fault
,
4547 .map_pages
= filemap_map_pages
,
4549 .set_policy
= shmem_set_policy
,
4550 .get_policy
= shmem_get_policy
,
4554 static const struct vm_operations_struct shmem_anon_vm_ops
= {
4555 .fault
= shmem_fault
,
4556 .map_pages
= filemap_map_pages
,
4558 .set_policy
= shmem_set_policy
,
4559 .get_policy
= shmem_get_policy
,
4563 int shmem_init_fs_context(struct fs_context
*fc
)
4565 struct shmem_options
*ctx
;
4567 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
4571 ctx
->mode
= 0777 | S_ISVTX
;
4572 ctx
->uid
= current_fsuid();
4573 ctx
->gid
= current_fsgid();
4575 fc
->fs_private
= ctx
;
4576 fc
->ops
= &shmem_fs_context_ops
;
4580 static struct file_system_type shmem_fs_type
= {
4581 .owner
= THIS_MODULE
,
4583 .init_fs_context
= shmem_init_fs_context
,
4585 .parameters
= shmem_fs_parameters
,
4587 .kill_sb
= kill_litter_super
,
4589 .fs_flags
= FS_USERNS_MOUNT
| FS_ALLOW_IDMAP
| FS_MGTIME
,
4591 .fs_flags
= FS_USERNS_MOUNT
,
4595 void __init
shmem_init(void)
4599 shmem_init_inodecache();
4601 #ifdef CONFIG_TMPFS_QUOTA
4602 error
= register_quota_format(&shmem_quota_format
);
4604 pr_err("Could not register quota format\n");
4609 error
= register_filesystem(&shmem_fs_type
);
4611 pr_err("Could not register tmpfs\n");
4615 shm_mnt
= kern_mount(&shmem_fs_type
);
4616 if (IS_ERR(shm_mnt
)) {
4617 error
= PTR_ERR(shm_mnt
);
4618 pr_err("Could not kern_mount tmpfs\n");
4622 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4623 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
4624 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4626 shmem_huge
= SHMEM_HUGE_NEVER
; /* just in case it was patched */
4631 unregister_filesystem(&shmem_fs_type
);
4633 #ifdef CONFIG_TMPFS_QUOTA
4634 unregister_quota_format(&shmem_quota_format
);
4637 shmem_destroy_inodecache();
4638 shm_mnt
= ERR_PTR(error
);
4641 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4642 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4643 struct kobj_attribute
*attr
, char *buf
)
4645 static const int values
[] = {
4647 SHMEM_HUGE_WITHIN_SIZE
,
4656 for (i
= 0; i
< ARRAY_SIZE(values
); i
++) {
4657 len
+= sysfs_emit_at(buf
, len
,
4658 shmem_huge
== values
[i
] ? "%s[%s]" : "%s%s",
4660 shmem_format_huge(values
[i
]));
4663 len
+= sysfs_emit_at(buf
, len
, "\n");
4668 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4669 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4674 if (count
+ 1 > sizeof(tmp
))
4676 memcpy(tmp
, buf
, count
);
4678 if (count
&& tmp
[count
- 1] == '\n')
4679 tmp
[count
- 1] = '\0';
4681 huge
= shmem_parse_huge(tmp
);
4682 if (huge
== -EINVAL
)
4684 if (!has_transparent_hugepage() &&
4685 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4689 if (shmem_huge
> SHMEM_HUGE_DENY
)
4690 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4694 struct kobj_attribute shmem_enabled_attr
= __ATTR_RW(shmem_enabled
);
4695 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4697 #else /* !CONFIG_SHMEM */
4700 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4702 * This is intended for small system where the benefits of the full
4703 * shmem code (swap-backed and resource-limited) are outweighed by
4704 * their complexity. On systems without swap this code should be
4705 * effectively equivalent, but much lighter weight.
4708 static struct file_system_type shmem_fs_type
= {
4710 .init_fs_context
= ramfs_init_fs_context
,
4711 .parameters
= ramfs_fs_parameters
,
4712 .kill_sb
= ramfs_kill_sb
,
4713 .fs_flags
= FS_USERNS_MOUNT
,
4716 void __init
shmem_init(void)
4718 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4720 shm_mnt
= kern_mount(&shmem_fs_type
);
4721 BUG_ON(IS_ERR(shm_mnt
));
4724 int shmem_unuse(unsigned int type
)
4729 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
4734 void shmem_unlock_mapping(struct address_space
*mapping
)
4739 unsigned long shmem_get_unmapped_area(struct file
*file
,
4740 unsigned long addr
, unsigned long len
,
4741 unsigned long pgoff
, unsigned long flags
)
4743 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4747 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4749 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4751 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4753 #define shmem_vm_ops generic_file_vm_ops
4754 #define shmem_anon_vm_ops generic_file_vm_ops
4755 #define shmem_file_operations ramfs_file_operations
4756 #define shmem_acct_size(flags, size) 0
4757 #define shmem_unacct_size(flags, size) do {} while (0)
4759 static inline struct inode
*shmem_get_inode(struct mnt_idmap
*idmap
, struct super_block
*sb
, struct inode
*dir
,
4760 umode_t mode
, dev_t dev
, unsigned long flags
)
4762 struct inode
*inode
= ramfs_get_inode(sb
, dir
, mode
, dev
);
4763 return inode
? inode
: ERR_PTR(-ENOSPC
);
4766 #endif /* CONFIG_SHMEM */
4770 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4771 unsigned long flags
, unsigned int i_flags
)
4773 struct inode
*inode
;
4777 return ERR_CAST(mnt
);
4779 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4780 return ERR_PTR(-EINVAL
);
4782 if (shmem_acct_size(flags
, size
))
4783 return ERR_PTR(-ENOMEM
);
4785 if (is_idmapped_mnt(mnt
))
4786 return ERR_PTR(-EINVAL
);
4788 inode
= shmem_get_inode(&nop_mnt_idmap
, mnt
->mnt_sb
, NULL
,
4789 S_IFREG
| S_IRWXUGO
, 0, flags
);
4791 if (IS_ERR(inode
)) {
4792 shmem_unacct_size(flags
, size
);
4793 return ERR_CAST(inode
);
4795 inode
->i_flags
|= i_flags
;
4796 inode
->i_size
= size
;
4797 clear_nlink(inode
); /* It is unlinked */
4798 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4800 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4801 &shmem_file_operations
);
4808 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4809 * kernel internal. There will be NO LSM permission checks against the
4810 * underlying inode. So users of this interface must do LSM checks at a
4811 * higher layer. The users are the big_key and shm implementations. LSM
4812 * checks are provided at the key or shm level rather than the inode.
4813 * @name: name for dentry (to be seen in /proc/<pid>/maps
4814 * @size: size to be set for the file
4815 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4817 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4819 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4823 * shmem_file_setup - get an unlinked file living in tmpfs
4824 * @name: name for dentry (to be seen in /proc/<pid>/maps
4825 * @size: size to be set for the file
4826 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4828 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4830 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4832 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4835 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4836 * @mnt: the tmpfs mount where the file will be created
4837 * @name: name for dentry (to be seen in /proc/<pid>/maps
4838 * @size: size to be set for the file
4839 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4841 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4842 loff_t size
, unsigned long flags
)
4844 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4846 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4849 * shmem_zero_setup - setup a shared anonymous mapping
4850 * @vma: the vma to be mmapped is prepared by do_mmap
4852 int shmem_zero_setup(struct vm_area_struct
*vma
)
4855 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4858 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4859 * between XFS directory reading and selinux: since this file is only
4860 * accessible to the user through its mapping, use S_PRIVATE flag to
4861 * bypass file security, in the same way as shmem_kernel_file_setup().
4863 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4865 return PTR_ERR(file
);
4869 vma
->vm_file
= file
;
4870 vma
->vm_ops
= &shmem_anon_vm_ops
;
4876 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4877 * @mapping: the folio's address_space
4878 * @index: the folio index
4879 * @gfp: the page allocator flags to use if allocating
4881 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4882 * with any new page allocations done using the specified allocation flags.
4883 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4884 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4885 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4887 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4888 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4890 struct folio
*shmem_read_folio_gfp(struct address_space
*mapping
,
4891 pgoff_t index
, gfp_t gfp
)
4894 struct inode
*inode
= mapping
->host
;
4895 struct folio
*folio
;
4898 BUG_ON(!shmem_mapping(mapping
));
4899 error
= shmem_get_folio_gfp(inode
, index
, &folio
, SGP_CACHE
,
4900 gfp
, NULL
, NULL
, NULL
);
4902 return ERR_PTR(error
);
4904 folio_unlock(folio
);
4908 * The tiny !SHMEM case uses ramfs without swap
4910 return mapping_read_folio_gfp(mapping
, index
, gfp
);
4913 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp
);
4915 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4916 pgoff_t index
, gfp_t gfp
)
4918 struct folio
*folio
= shmem_read_folio_gfp(mapping
, index
, gfp
);
4922 return &folio
->page
;
4924 page
= folio_file_page(folio
, index
);
4925 if (PageHWPoison(page
)) {
4927 return ERR_PTR(-EIO
);
4932 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);