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 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_rwsem making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc
{
102 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
103 pgoff_t start
; /* start of range currently being fallocated */
104 pgoff_t next
; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
109 struct shmem_options
{
110 unsigned long long blocks
;
111 unsigned long long inodes
;
112 struct mempolicy
*mpol
;
120 unsigned short quota_types
;
121 struct shmem_quota_limits qlimits
;
122 #define SHMEM_SEEN_BLOCKS 1
123 #define SHMEM_SEEN_INODES 2
124 #define SHMEM_SEEN_HUGE 4
125 #define SHMEM_SEEN_INUMS 8
126 #define SHMEM_SEEN_NOSWAP 16
127 #define SHMEM_SEEN_QUOTA 32
131 static unsigned long shmem_default_max_blocks(void)
133 return totalram_pages() / 2;
136 static unsigned long shmem_default_max_inodes(void)
138 unsigned long nr_pages
= totalram_pages();
140 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
144 static int shmem_swapin_folio(struct inode
*inode
, pgoff_t index
,
145 struct folio
**foliop
, enum sgp_type sgp
,
146 gfp_t gfp
, struct vm_area_struct
*vma
,
147 vm_fault_t
*fault_type
);
149 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
151 return sb
->s_fs_info
;
155 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
156 * for shared memory and for shared anonymous (/dev/zero) mappings
157 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
158 * consistent with the pre-accounting of private mappings ...
160 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
162 return (flags
& VM_NORESERVE
) ?
163 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
166 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
168 if (!(flags
& VM_NORESERVE
))
169 vm_unacct_memory(VM_ACCT(size
));
172 static inline int shmem_reacct_size(unsigned long flags
,
173 loff_t oldsize
, loff_t newsize
)
175 if (!(flags
& VM_NORESERVE
)) {
176 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
177 return security_vm_enough_memory_mm(current
->mm
,
178 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
179 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
180 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
186 * ... whereas tmpfs objects are accounted incrementally as
187 * pages are allocated, in order to allow large sparse files.
188 * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
189 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
191 static inline int shmem_acct_block(unsigned long flags
, long pages
)
193 if (!(flags
& VM_NORESERVE
))
196 return security_vm_enough_memory_mm(current
->mm
,
197 pages
* VM_ACCT(PAGE_SIZE
));
200 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
202 if (flags
& VM_NORESERVE
)
203 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
206 static int shmem_inode_acct_block(struct inode
*inode
, long pages
)
208 struct shmem_inode_info
*info
= SHMEM_I(inode
);
209 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
212 if (shmem_acct_block(info
->flags
, pages
))
215 might_sleep(); /* when quotas */
216 if (sbinfo
->max_blocks
) {
217 if (percpu_counter_compare(&sbinfo
->used_blocks
,
218 sbinfo
->max_blocks
- pages
) > 0)
221 err
= dquot_alloc_block_nodirty(inode
, pages
);
225 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
227 err
= dquot_alloc_block_nodirty(inode
, pages
);
235 shmem_unacct_blocks(info
->flags
, pages
);
239 static void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
241 struct shmem_inode_info
*info
= SHMEM_I(inode
);
242 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
244 might_sleep(); /* when quotas */
245 dquot_free_block_nodirty(inode
, pages
);
247 if (sbinfo
->max_blocks
)
248 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
249 shmem_unacct_blocks(info
->flags
, pages
);
252 static const struct super_operations shmem_ops
;
253 const struct address_space_operations shmem_aops
;
254 static const struct file_operations shmem_file_operations
;
255 static const struct inode_operations shmem_inode_operations
;
256 static const struct inode_operations shmem_dir_inode_operations
;
257 static const struct inode_operations shmem_special_inode_operations
;
258 static const struct vm_operations_struct shmem_vm_ops
;
259 static const struct vm_operations_struct shmem_anon_vm_ops
;
260 static struct file_system_type shmem_fs_type
;
262 bool vma_is_anon_shmem(struct vm_area_struct
*vma
)
264 return vma
->vm_ops
== &shmem_anon_vm_ops
;
267 bool vma_is_shmem(struct vm_area_struct
*vma
)
269 return vma_is_anon_shmem(vma
) || vma
->vm_ops
== &shmem_vm_ops
;
272 static LIST_HEAD(shmem_swaplist
);
273 static DEFINE_MUTEX(shmem_swaplist_mutex
);
275 #ifdef CONFIG_TMPFS_QUOTA
277 static int shmem_enable_quotas(struct super_block
*sb
,
278 unsigned short quota_types
)
282 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
| DQUOT_NOLIST_DIRTY
;
283 for (type
= 0; type
< SHMEM_MAXQUOTAS
; type
++) {
284 if (!(quota_types
& (1 << type
)))
286 err
= dquot_load_quota_sb(sb
, type
, QFMT_SHMEM
,
287 DQUOT_USAGE_ENABLED
|
288 DQUOT_LIMITS_ENABLED
);
295 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
297 for (type
--; type
>= 0; type
--)
298 dquot_quota_off(sb
, type
);
302 static void shmem_disable_quotas(struct super_block
*sb
)
306 for (type
= 0; type
< SHMEM_MAXQUOTAS
; type
++)
307 dquot_quota_off(sb
, type
);
310 static struct dquot
**shmem_get_dquots(struct inode
*inode
)
312 return SHMEM_I(inode
)->i_dquot
;
314 #endif /* CONFIG_TMPFS_QUOTA */
317 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
318 * produces a novel ino for the newly allocated inode.
320 * It may also be called when making a hard link to permit the space needed by
321 * each dentry. However, in that case, no new inode number is needed since that
322 * internally draws from another pool of inode numbers (currently global
323 * get_next_ino()). This case is indicated by passing NULL as inop.
325 #define SHMEM_INO_BATCH 1024
326 static int shmem_reserve_inode(struct super_block
*sb
, ino_t
*inop
)
328 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
331 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
332 raw_spin_lock(&sbinfo
->stat_lock
);
333 if (sbinfo
->max_inodes
) {
334 if (!sbinfo
->free_inodes
) {
335 raw_spin_unlock(&sbinfo
->stat_lock
);
338 sbinfo
->free_inodes
--;
341 ino
= sbinfo
->next_ino
++;
342 if (unlikely(is_zero_ino(ino
)))
343 ino
= sbinfo
->next_ino
++;
344 if (unlikely(!sbinfo
->full_inums
&&
347 * Emulate get_next_ino uint wraparound for
350 if (IS_ENABLED(CONFIG_64BIT
))
351 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
352 __func__
, MINOR(sb
->s_dev
));
353 sbinfo
->next_ino
= 1;
354 ino
= sbinfo
->next_ino
++;
358 raw_spin_unlock(&sbinfo
->stat_lock
);
361 * __shmem_file_setup, one of our callers, is lock-free: it
362 * doesn't hold stat_lock in shmem_reserve_inode since
363 * max_inodes is always 0, and is called from potentially
364 * unknown contexts. As such, use a per-cpu batched allocator
365 * which doesn't require the per-sb stat_lock unless we are at
366 * the batch boundary.
368 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
369 * shmem mounts are not exposed to userspace, so we don't need
370 * to worry about things like glibc compatibility.
374 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
376 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
377 raw_spin_lock(&sbinfo
->stat_lock
);
378 ino
= sbinfo
->next_ino
;
379 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
380 raw_spin_unlock(&sbinfo
->stat_lock
);
381 if (unlikely(is_zero_ino(ino
)))
392 static void shmem_free_inode(struct super_block
*sb
)
394 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
395 if (sbinfo
->max_inodes
) {
396 raw_spin_lock(&sbinfo
->stat_lock
);
397 sbinfo
->free_inodes
++;
398 raw_spin_unlock(&sbinfo
->stat_lock
);
403 * shmem_recalc_inode - recalculate the block usage of an inode
404 * @inode: inode to recalc
405 * @alloced: the change in number of pages allocated to inode
406 * @swapped: the change in number of pages swapped from inode
408 * We have to calculate the free blocks since the mm can drop
409 * undirtied hole pages behind our back.
411 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
412 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
414 static void shmem_recalc_inode(struct inode
*inode
, long alloced
, long swapped
)
416 struct shmem_inode_info
*info
= SHMEM_I(inode
);
419 spin_lock(&info
->lock
);
420 info
->alloced
+= alloced
;
421 info
->swapped
+= swapped
;
422 freed
= info
->alloced
- info
->swapped
-
423 READ_ONCE(inode
->i_mapping
->nrpages
);
425 * Special case: whereas normally shmem_recalc_inode() is called
426 * after i_mapping->nrpages has already been adjusted (up or down),
427 * shmem_writepage() has to raise swapped before nrpages is lowered -
428 * to stop a racing shmem_recalc_inode() from thinking that a page has
429 * been freed. Compensate here, to avoid the need for a followup call.
434 info
->alloced
-= freed
;
435 spin_unlock(&info
->lock
);
437 /* The quota case may block */
439 shmem_inode_unacct_blocks(inode
, freed
);
442 bool shmem_charge(struct inode
*inode
, long pages
)
444 struct address_space
*mapping
= inode
->i_mapping
;
446 if (shmem_inode_acct_block(inode
, pages
))
449 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
450 xa_lock_irq(&mapping
->i_pages
);
451 mapping
->nrpages
+= pages
;
452 xa_unlock_irq(&mapping
->i_pages
);
454 shmem_recalc_inode(inode
, pages
, 0);
458 void shmem_uncharge(struct inode
*inode
, long pages
)
460 /* pages argument is currently unused: keep it to help debugging */
461 /* nrpages adjustment done by __filemap_remove_folio() or caller */
463 shmem_recalc_inode(inode
, 0, 0);
467 * Replace item expected in xarray by a new item, while holding xa_lock.
469 static int shmem_replace_entry(struct address_space
*mapping
,
470 pgoff_t index
, void *expected
, void *replacement
)
472 XA_STATE(xas
, &mapping
->i_pages
, index
);
475 VM_BUG_ON(!expected
);
476 VM_BUG_ON(!replacement
);
477 item
= xas_load(&xas
);
478 if (item
!= expected
)
480 xas_store(&xas
, replacement
);
485 * Sometimes, before we decide whether to proceed or to fail, we must check
486 * that an entry was not already brought back from swap by a racing thread.
488 * Checking page is not enough: by the time a SwapCache page is locked, it
489 * might be reused, and again be SwapCache, using the same swap as before.
491 static bool shmem_confirm_swap(struct address_space
*mapping
,
492 pgoff_t index
, swp_entry_t swap
)
494 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
498 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
501 * disables huge pages for the mount;
503 * enables huge pages for the mount;
504 * SHMEM_HUGE_WITHIN_SIZE:
505 * only allocate huge pages if the page will be fully within i_size,
506 * also respect fadvise()/madvise() hints;
508 * only allocate huge pages if requested with fadvise()/madvise();
511 #define SHMEM_HUGE_NEVER 0
512 #define SHMEM_HUGE_ALWAYS 1
513 #define SHMEM_HUGE_WITHIN_SIZE 2
514 #define SHMEM_HUGE_ADVISE 3
518 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
521 * disables huge on shm_mnt and all mounts, for emergency use;
523 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
526 #define SHMEM_HUGE_DENY (-1)
527 #define SHMEM_HUGE_FORCE (-2)
529 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
530 /* ifdef here to avoid bloating shmem.o when not necessary */
532 static int shmem_huge __read_mostly
= SHMEM_HUGE_NEVER
;
534 bool shmem_is_huge(struct inode
*inode
, pgoff_t index
, bool shmem_huge_force
,
535 struct mm_struct
*mm
, unsigned long vm_flags
)
539 if (!S_ISREG(inode
->i_mode
))
541 if (mm
&& ((vm_flags
& VM_NOHUGEPAGE
) || test_bit(MMF_DISABLE_THP
, &mm
->flags
)))
543 if (shmem_huge
== SHMEM_HUGE_DENY
)
545 if (shmem_huge_force
|| shmem_huge
== SHMEM_HUGE_FORCE
)
548 switch (SHMEM_SB(inode
->i_sb
)->huge
) {
549 case SHMEM_HUGE_ALWAYS
:
551 case SHMEM_HUGE_WITHIN_SIZE
:
552 index
= round_up(index
+ 1, HPAGE_PMD_NR
);
553 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
554 if (i_size
>> PAGE_SHIFT
>= index
)
557 case SHMEM_HUGE_ADVISE
:
558 if (mm
&& (vm_flags
& VM_HUGEPAGE
))
566 #if defined(CONFIG_SYSFS)
567 static int shmem_parse_huge(const char *str
)
569 if (!strcmp(str
, "never"))
570 return SHMEM_HUGE_NEVER
;
571 if (!strcmp(str
, "always"))
572 return SHMEM_HUGE_ALWAYS
;
573 if (!strcmp(str
, "within_size"))
574 return SHMEM_HUGE_WITHIN_SIZE
;
575 if (!strcmp(str
, "advise"))
576 return SHMEM_HUGE_ADVISE
;
577 if (!strcmp(str
, "deny"))
578 return SHMEM_HUGE_DENY
;
579 if (!strcmp(str
, "force"))
580 return SHMEM_HUGE_FORCE
;
585 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
586 static const char *shmem_format_huge(int huge
)
589 case SHMEM_HUGE_NEVER
:
591 case SHMEM_HUGE_ALWAYS
:
593 case SHMEM_HUGE_WITHIN_SIZE
:
594 return "within_size";
595 case SHMEM_HUGE_ADVISE
:
597 case SHMEM_HUGE_DENY
:
599 case SHMEM_HUGE_FORCE
:
608 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
609 struct shrink_control
*sc
, unsigned long nr_to_split
)
611 LIST_HEAD(list
), *pos
, *next
;
612 LIST_HEAD(to_remove
);
614 struct shmem_inode_info
*info
;
616 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
619 if (list_empty(&sbinfo
->shrinklist
))
622 spin_lock(&sbinfo
->shrinklist_lock
);
623 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
624 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
627 inode
= igrab(&info
->vfs_inode
);
629 /* inode is about to be evicted */
631 list_del_init(&info
->shrinklist
);
635 /* Check if there's anything to gain */
636 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
637 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
638 list_move(&info
->shrinklist
, &to_remove
);
642 list_move(&info
->shrinklist
, &list
);
644 sbinfo
->shrinklist_len
--;
648 spin_unlock(&sbinfo
->shrinklist_lock
);
650 list_for_each_safe(pos
, next
, &to_remove
) {
651 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
652 inode
= &info
->vfs_inode
;
653 list_del_init(&info
->shrinklist
);
657 list_for_each_safe(pos
, next
, &list
) {
661 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
662 inode
= &info
->vfs_inode
;
664 if (nr_to_split
&& split
>= nr_to_split
)
667 index
= (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
;
668 folio
= filemap_get_folio(inode
->i_mapping
, index
);
672 /* No huge page at the end of the file: nothing to split */
673 if (!folio_test_large(folio
)) {
679 * Move the inode on the list back to shrinklist if we failed
680 * to lock the page at this time.
682 * Waiting for the lock may lead to deadlock in the
685 if (!folio_trylock(folio
)) {
690 ret
= split_folio(folio
);
694 /* If split failed move the inode on the list back to shrinklist */
700 list_del_init(&info
->shrinklist
);
704 * Make sure the inode is either on the global list or deleted
705 * from any local list before iput() since it could be deleted
706 * in another thread once we put the inode (then the local list
709 spin_lock(&sbinfo
->shrinklist_lock
);
710 list_move(&info
->shrinklist
, &sbinfo
->shrinklist
);
711 sbinfo
->shrinklist_len
++;
712 spin_unlock(&sbinfo
->shrinklist_lock
);
720 static long shmem_unused_huge_scan(struct super_block
*sb
,
721 struct shrink_control
*sc
)
723 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
725 if (!READ_ONCE(sbinfo
->shrinklist_len
))
728 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
731 static long shmem_unused_huge_count(struct super_block
*sb
,
732 struct shrink_control
*sc
)
734 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
735 return READ_ONCE(sbinfo
->shrinklist_len
);
737 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
739 #define shmem_huge SHMEM_HUGE_DENY
741 bool shmem_is_huge(struct inode
*inode
, pgoff_t index
, bool shmem_huge_force
,
742 struct mm_struct
*mm
, unsigned long vm_flags
)
747 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
748 struct shrink_control
*sc
, unsigned long nr_to_split
)
752 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
755 * Like filemap_add_folio, but error if expected item has gone.
757 static int shmem_add_to_page_cache(struct folio
*folio
,
758 struct address_space
*mapping
,
759 pgoff_t index
, void *expected
, gfp_t gfp
,
760 struct mm_struct
*charge_mm
)
762 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, folio_order(folio
));
763 long nr
= folio_nr_pages(folio
);
766 VM_BUG_ON_FOLIO(index
!= round_down(index
, nr
), folio
);
767 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
768 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio
), folio
);
769 VM_BUG_ON(expected
&& folio_test_large(folio
));
771 folio_ref_add(folio
, nr
);
772 folio
->mapping
= mapping
;
773 folio
->index
= index
;
775 if (!folio_test_swapcache(folio
)) {
776 error
= mem_cgroup_charge(folio
, charge_mm
, gfp
);
778 if (folio_test_pmd_mappable(folio
)) {
779 count_vm_event(THP_FILE_FALLBACK
);
780 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
785 folio_throttle_swaprate(folio
, gfp
);
789 if (expected
!= xas_find_conflict(&xas
)) {
790 xas_set_err(&xas
, -EEXIST
);
793 if (expected
&& xas_find_conflict(&xas
)) {
794 xas_set_err(&xas
, -EEXIST
);
797 xas_store(&xas
, folio
);
800 if (folio_test_pmd_mappable(folio
)) {
801 count_vm_event(THP_FILE_ALLOC
);
802 __lruvec_stat_mod_folio(folio
, NR_SHMEM_THPS
, nr
);
804 mapping
->nrpages
+= nr
;
805 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, nr
);
806 __lruvec_stat_mod_folio(folio
, NR_SHMEM
, nr
);
808 xas_unlock_irq(&xas
);
809 } while (xas_nomem(&xas
, gfp
));
811 if (xas_error(&xas
)) {
812 error
= xas_error(&xas
);
818 folio
->mapping
= NULL
;
819 folio_ref_sub(folio
, nr
);
824 * Like delete_from_page_cache, but substitutes swap for @folio.
826 static void shmem_delete_from_page_cache(struct folio
*folio
, void *radswap
)
828 struct address_space
*mapping
= folio
->mapping
;
829 long nr
= folio_nr_pages(folio
);
832 xa_lock_irq(&mapping
->i_pages
);
833 error
= shmem_replace_entry(mapping
, folio
->index
, folio
, radswap
);
834 folio
->mapping
= NULL
;
835 mapping
->nrpages
-= nr
;
836 __lruvec_stat_mod_folio(folio
, NR_FILE_PAGES
, -nr
);
837 __lruvec_stat_mod_folio(folio
, NR_SHMEM
, -nr
);
838 xa_unlock_irq(&mapping
->i_pages
);
844 * Remove swap entry from page cache, free the swap and its page cache.
846 static int shmem_free_swap(struct address_space
*mapping
,
847 pgoff_t index
, void *radswap
)
851 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
854 free_swap_and_cache(radix_to_swp_entry(radswap
));
859 * Determine (in bytes) how many of the shmem object's pages mapped by the
860 * given offsets are swapped out.
862 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
863 * as long as the inode doesn't go away and racy results are not a problem.
865 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
866 pgoff_t start
, pgoff_t end
)
868 XA_STATE(xas
, &mapping
->i_pages
, start
);
870 unsigned long swapped
= 0;
873 xas_for_each(&xas
, page
, end
- 1) {
874 if (xas_retry(&xas
, page
))
876 if (xa_is_value(page
))
879 if (need_resched()) {
887 return swapped
<< PAGE_SHIFT
;
891 * Determine (in bytes) how many of the shmem object's pages mapped by the
892 * given vma is swapped out.
894 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
895 * as long as the inode doesn't go away and racy results are not a problem.
897 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
899 struct inode
*inode
= file_inode(vma
->vm_file
);
900 struct shmem_inode_info
*info
= SHMEM_I(inode
);
901 struct address_space
*mapping
= inode
->i_mapping
;
902 unsigned long swapped
;
904 /* Be careful as we don't hold info->lock */
905 swapped
= READ_ONCE(info
->swapped
);
908 * The easier cases are when the shmem object has nothing in swap, or
909 * the vma maps it whole. Then we can simply use the stats that we
915 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
916 return swapped
<< PAGE_SHIFT
;
918 /* Here comes the more involved part */
919 return shmem_partial_swap_usage(mapping
, vma
->vm_pgoff
,
920 vma
->vm_pgoff
+ vma_pages(vma
));
924 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
926 void shmem_unlock_mapping(struct address_space
*mapping
)
928 struct folio_batch fbatch
;
931 folio_batch_init(&fbatch
);
933 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
935 while (!mapping_unevictable(mapping
) &&
936 filemap_get_folios(mapping
, &index
, ~0UL, &fbatch
)) {
937 check_move_unevictable_folios(&fbatch
);
938 folio_batch_release(&fbatch
);
943 static struct folio
*shmem_get_partial_folio(struct inode
*inode
, pgoff_t index
)
948 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
949 * beyond i_size, and reports fallocated folios as holes.
951 folio
= filemap_get_entry(inode
->i_mapping
, index
);
954 if (!xa_is_value(folio
)) {
956 if (folio
->mapping
== inode
->i_mapping
)
958 /* The folio has been swapped out */
963 * But read a folio back from swap if any of it is within i_size
964 * (although in some cases this is just a waste of time).
967 shmem_get_folio(inode
, index
, &folio
, SGP_READ
);
972 * Remove range of pages and swap entries from page cache, and free them.
973 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
975 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
978 struct address_space
*mapping
= inode
->i_mapping
;
979 struct shmem_inode_info
*info
= SHMEM_I(inode
);
980 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
981 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
982 struct folio_batch fbatch
;
983 pgoff_t indices
[PAGEVEC_SIZE
];
986 long nr_swaps_freed
= 0;
991 end
= -1; /* unsigned, so actually very big */
993 if (info
->fallocend
> start
&& info
->fallocend
<= end
&& !unfalloc
)
994 info
->fallocend
= start
;
996 folio_batch_init(&fbatch
);
998 while (index
< end
&& find_lock_entries(mapping
, &index
, end
- 1,
1000 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
1001 folio
= fbatch
.folios
[i
];
1003 if (xa_is_value(folio
)) {
1006 nr_swaps_freed
+= !shmem_free_swap(mapping
,
1011 if (!unfalloc
|| !folio_test_uptodate(folio
))
1012 truncate_inode_folio(mapping
, folio
);
1013 folio_unlock(folio
);
1015 folio_batch_remove_exceptionals(&fbatch
);
1016 folio_batch_release(&fbatch
);
1021 * When undoing a failed fallocate, we want none of the partial folio
1022 * zeroing and splitting below, but shall want to truncate the whole
1023 * folio when !uptodate indicates that it was added by this fallocate,
1024 * even when [lstart, lend] covers only a part of the folio.
1029 same_folio
= (lstart
>> PAGE_SHIFT
) == (lend
>> PAGE_SHIFT
);
1030 folio
= shmem_get_partial_folio(inode
, lstart
>> PAGE_SHIFT
);
1032 same_folio
= lend
< folio_pos(folio
) + folio_size(folio
);
1033 folio_mark_dirty(folio
);
1034 if (!truncate_inode_partial_folio(folio
, lstart
, lend
)) {
1035 start
= folio
->index
+ folio_nr_pages(folio
);
1039 folio_unlock(folio
);
1045 folio
= shmem_get_partial_folio(inode
, lend
>> PAGE_SHIFT
);
1047 folio_mark_dirty(folio
);
1048 if (!truncate_inode_partial_folio(folio
, lstart
, lend
))
1050 folio_unlock(folio
);
1057 while (index
< end
) {
1060 if (!find_get_entries(mapping
, &index
, end
- 1, &fbatch
,
1062 /* If all gone or hole-punch or unfalloc, we're done */
1063 if (index
== start
|| end
!= -1)
1065 /* But if truncating, restart to make sure all gone */
1069 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
1070 folio
= fbatch
.folios
[i
];
1072 if (xa_is_value(folio
)) {
1075 if (shmem_free_swap(mapping
, indices
[i
], folio
)) {
1076 /* Swap was replaced by page: retry */
1086 if (!unfalloc
|| !folio_test_uptodate(folio
)) {
1087 if (folio_mapping(folio
) != mapping
) {
1088 /* Page was replaced by swap: retry */
1089 folio_unlock(folio
);
1093 VM_BUG_ON_FOLIO(folio_test_writeback(folio
),
1095 truncate_inode_folio(mapping
, folio
);
1097 folio_unlock(folio
);
1099 folio_batch_remove_exceptionals(&fbatch
);
1100 folio_batch_release(&fbatch
);
1103 shmem_recalc_inode(inode
, 0, -nr_swaps_freed
);
1106 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1108 shmem_undo_range(inode
, lstart
, lend
, false);
1109 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1110 inode_inc_iversion(inode
);
1112 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1114 static int shmem_getattr(struct mnt_idmap
*idmap
,
1115 const struct path
*path
, struct kstat
*stat
,
1116 u32 request_mask
, unsigned int query_flags
)
1118 struct inode
*inode
= path
->dentry
->d_inode
;
1119 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1121 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
)
1122 shmem_recalc_inode(inode
, 0, 0);
1124 if (info
->fsflags
& FS_APPEND_FL
)
1125 stat
->attributes
|= STATX_ATTR_APPEND
;
1126 if (info
->fsflags
& FS_IMMUTABLE_FL
)
1127 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
1128 if (info
->fsflags
& FS_NODUMP_FL
)
1129 stat
->attributes
|= STATX_ATTR_NODUMP
;
1130 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
1131 STATX_ATTR_IMMUTABLE
|
1133 generic_fillattr(idmap
, inode
, stat
);
1135 if (shmem_is_huge(inode
, 0, false, NULL
, 0))
1136 stat
->blksize
= HPAGE_PMD_SIZE
;
1138 if (request_mask
& STATX_BTIME
) {
1139 stat
->result_mask
|= STATX_BTIME
;
1140 stat
->btime
.tv_sec
= info
->i_crtime
.tv_sec
;
1141 stat
->btime
.tv_nsec
= info
->i_crtime
.tv_nsec
;
1147 static int shmem_setattr(struct mnt_idmap
*idmap
,
1148 struct dentry
*dentry
, struct iattr
*attr
)
1150 struct inode
*inode
= d_inode(dentry
);
1151 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1153 bool update_mtime
= false;
1154 bool update_ctime
= true;
1156 error
= setattr_prepare(idmap
, dentry
, attr
);
1160 if ((info
->seals
& F_SEAL_EXEC
) && (attr
->ia_valid
& ATTR_MODE
)) {
1161 if ((inode
->i_mode
^ attr
->ia_mode
) & 0111) {
1166 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1167 loff_t oldsize
= inode
->i_size
;
1168 loff_t newsize
= attr
->ia_size
;
1170 /* protected by i_rwsem */
1171 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1172 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1175 if (newsize
!= oldsize
) {
1176 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1180 i_size_write(inode
, newsize
);
1181 update_mtime
= true;
1183 update_ctime
= false;
1185 if (newsize
<= oldsize
) {
1186 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1187 if (oldsize
> holebegin
)
1188 unmap_mapping_range(inode
->i_mapping
,
1191 shmem_truncate_range(inode
,
1192 newsize
, (loff_t
)-1);
1193 /* unmap again to remove racily COWed private pages */
1194 if (oldsize
> holebegin
)
1195 unmap_mapping_range(inode
->i_mapping
,
1200 if (is_quota_modification(idmap
, inode
, attr
)) {
1201 error
= dquot_initialize(inode
);
1206 /* Transfer quota accounting */
1207 if (i_uid_needs_update(idmap
, attr
, inode
) ||
1208 i_gid_needs_update(idmap
, attr
, inode
)) {
1209 error
= dquot_transfer(idmap
, inode
, attr
);
1215 setattr_copy(idmap
, inode
, attr
);
1216 if (attr
->ia_valid
& ATTR_MODE
)
1217 error
= posix_acl_chmod(idmap
, dentry
, inode
->i_mode
);
1218 if (!error
&& update_ctime
) {
1219 inode
->i_ctime
= current_time(inode
);
1221 inode
->i_mtime
= inode
->i_ctime
;
1222 inode_inc_iversion(inode
);
1227 static void shmem_evict_inode(struct inode
*inode
)
1229 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1230 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1232 if (shmem_mapping(inode
->i_mapping
)) {
1233 shmem_unacct_size(info
->flags
, inode
->i_size
);
1235 mapping_set_exiting(inode
->i_mapping
);
1236 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1237 if (!list_empty(&info
->shrinklist
)) {
1238 spin_lock(&sbinfo
->shrinklist_lock
);
1239 if (!list_empty(&info
->shrinklist
)) {
1240 list_del_init(&info
->shrinklist
);
1241 sbinfo
->shrinklist_len
--;
1243 spin_unlock(&sbinfo
->shrinklist_lock
);
1245 while (!list_empty(&info
->swaplist
)) {
1246 /* Wait while shmem_unuse() is scanning this inode... */
1247 wait_var_event(&info
->stop_eviction
,
1248 !atomic_read(&info
->stop_eviction
));
1249 mutex_lock(&shmem_swaplist_mutex
);
1250 /* ...but beware of the race if we peeked too early */
1251 if (!atomic_read(&info
->stop_eviction
))
1252 list_del_init(&info
->swaplist
);
1253 mutex_unlock(&shmem_swaplist_mutex
);
1257 simple_xattrs_free(&info
->xattrs
);
1258 WARN_ON(inode
->i_blocks
);
1259 shmem_free_inode(inode
->i_sb
);
1261 #ifdef CONFIG_TMPFS_QUOTA
1262 dquot_free_inode(inode
);
1267 static int shmem_find_swap_entries(struct address_space
*mapping
,
1268 pgoff_t start
, struct folio_batch
*fbatch
,
1269 pgoff_t
*indices
, unsigned int type
)
1271 XA_STATE(xas
, &mapping
->i_pages
, start
);
1272 struct folio
*folio
;
1276 xas_for_each(&xas
, folio
, ULONG_MAX
) {
1277 if (xas_retry(&xas
, folio
))
1280 if (!xa_is_value(folio
))
1283 entry
= radix_to_swp_entry(folio
);
1285 * swapin error entries can be found in the mapping. But they're
1286 * deliberately ignored here as we've done everything we can do.
1288 if (swp_type(entry
) != type
)
1291 indices
[folio_batch_count(fbatch
)] = xas
.xa_index
;
1292 if (!folio_batch_add(fbatch
, folio
))
1295 if (need_resched()) {
1302 return xas
.xa_index
;
1306 * Move the swapped pages for an inode to page cache. Returns the count
1307 * of pages swapped in, or the error in case of failure.
1309 static int shmem_unuse_swap_entries(struct inode
*inode
,
1310 struct folio_batch
*fbatch
, pgoff_t
*indices
)
1315 struct address_space
*mapping
= inode
->i_mapping
;
1317 for (i
= 0; i
< folio_batch_count(fbatch
); i
++) {
1318 struct folio
*folio
= fbatch
->folios
[i
];
1320 if (!xa_is_value(folio
))
1322 error
= shmem_swapin_folio(inode
, indices
[i
],
1324 mapping_gfp_mask(mapping
),
1327 folio_unlock(folio
);
1331 if (error
== -ENOMEM
)
1335 return error
? error
: ret
;
1339 * If swap found in inode, free it and move page from swapcache to filecache.
1341 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
)
1343 struct address_space
*mapping
= inode
->i_mapping
;
1345 struct folio_batch fbatch
;
1346 pgoff_t indices
[PAGEVEC_SIZE
];
1350 folio_batch_init(&fbatch
);
1351 shmem_find_swap_entries(mapping
, start
, &fbatch
, indices
, type
);
1352 if (folio_batch_count(&fbatch
) == 0) {
1357 ret
= shmem_unuse_swap_entries(inode
, &fbatch
, indices
);
1361 start
= indices
[folio_batch_count(&fbatch
) - 1];
1368 * Read all the shared memory data that resides in the swap
1369 * device 'type' back into memory, so the swap device can be
1372 int shmem_unuse(unsigned int type
)
1374 struct shmem_inode_info
*info
, *next
;
1377 if (list_empty(&shmem_swaplist
))
1380 mutex_lock(&shmem_swaplist_mutex
);
1381 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1382 if (!info
->swapped
) {
1383 list_del_init(&info
->swaplist
);
1387 * Drop the swaplist mutex while searching the inode for swap;
1388 * but before doing so, make sure shmem_evict_inode() will not
1389 * remove placeholder inode from swaplist, nor let it be freed
1390 * (igrab() would protect from unlink, but not from unmount).
1392 atomic_inc(&info
->stop_eviction
);
1393 mutex_unlock(&shmem_swaplist_mutex
);
1395 error
= shmem_unuse_inode(&info
->vfs_inode
, type
);
1398 mutex_lock(&shmem_swaplist_mutex
);
1399 next
= list_next_entry(info
, swaplist
);
1401 list_del_init(&info
->swaplist
);
1402 if (atomic_dec_and_test(&info
->stop_eviction
))
1403 wake_up_var(&info
->stop_eviction
);
1407 mutex_unlock(&shmem_swaplist_mutex
);
1413 * Move the page from the page cache to the swap cache.
1415 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1417 struct folio
*folio
= page_folio(page
);
1418 struct address_space
*mapping
= folio
->mapping
;
1419 struct inode
*inode
= mapping
->host
;
1420 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1421 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1426 * Our capabilities prevent regular writeback or sync from ever calling
1427 * shmem_writepage; but a stacking filesystem might use ->writepage of
1428 * its underlying filesystem, in which case tmpfs should write out to
1429 * swap only in response to memory pressure, and not for the writeback
1432 if (WARN_ON_ONCE(!wbc
->for_reclaim
))
1435 if (WARN_ON_ONCE((info
->flags
& VM_LOCKED
) || sbinfo
->noswap
))
1438 if (!total_swap_pages
)
1442 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1443 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1444 * and its shmem_writeback() needs them to be split when swapping.
1446 if (folio_test_large(folio
)) {
1447 /* Ensure the subpages are still dirty */
1448 folio_test_set_dirty(folio
);
1449 if (split_huge_page(page
) < 0)
1451 folio
= page_folio(page
);
1452 folio_clear_dirty(folio
);
1455 index
= folio
->index
;
1458 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1459 * value into swapfile.c, the only way we can correctly account for a
1460 * fallocated folio arriving here is now to initialize it and write it.
1462 * That's okay for a folio already fallocated earlier, but if we have
1463 * not yet completed the fallocation, then (a) we want to keep track
1464 * of this folio in case we have to undo it, and (b) it may not be a
1465 * good idea to continue anyway, once we're pushing into swap. So
1466 * reactivate the folio, and let shmem_fallocate() quit when too many.
1468 if (!folio_test_uptodate(folio
)) {
1469 if (inode
->i_private
) {
1470 struct shmem_falloc
*shmem_falloc
;
1471 spin_lock(&inode
->i_lock
);
1472 shmem_falloc
= inode
->i_private
;
1474 !shmem_falloc
->waitq
&&
1475 index
>= shmem_falloc
->start
&&
1476 index
< shmem_falloc
->next
)
1477 shmem_falloc
->nr_unswapped
++;
1479 shmem_falloc
= NULL
;
1480 spin_unlock(&inode
->i_lock
);
1484 folio_zero_range(folio
, 0, folio_size(folio
));
1485 flush_dcache_folio(folio
);
1486 folio_mark_uptodate(folio
);
1489 swap
= folio_alloc_swap(folio
);
1494 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1495 * if it's not already there. Do it now before the folio is
1496 * moved to swap cache, when its pagelock no longer protects
1497 * the inode from eviction. But don't unlock the mutex until
1498 * we've incremented swapped, because shmem_unuse_inode() will
1499 * prune a !swapped inode from the swaplist under this mutex.
1501 mutex_lock(&shmem_swaplist_mutex
);
1502 if (list_empty(&info
->swaplist
))
1503 list_add(&info
->swaplist
, &shmem_swaplist
);
1505 if (add_to_swap_cache(folio
, swap
,
1506 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1508 shmem_recalc_inode(inode
, 0, 1);
1509 swap_shmem_alloc(swap
);
1510 shmem_delete_from_page_cache(folio
, swp_to_radix_entry(swap
));
1512 mutex_unlock(&shmem_swaplist_mutex
);
1513 BUG_ON(folio_mapped(folio
));
1514 swap_writepage(&folio
->page
, wbc
);
1518 mutex_unlock(&shmem_swaplist_mutex
);
1519 put_swap_folio(folio
, swap
);
1521 folio_mark_dirty(folio
);
1522 if (wbc
->for_reclaim
)
1523 return AOP_WRITEPAGE_ACTIVATE
; /* Return with folio locked */
1524 folio_unlock(folio
);
1528 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1529 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1533 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1534 return; /* show nothing */
1536 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1538 seq_printf(seq
, ",mpol=%s", buffer
);
1541 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1543 struct mempolicy
*mpol
= NULL
;
1545 raw_spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1546 mpol
= sbinfo
->mpol
;
1548 raw_spin_unlock(&sbinfo
->stat_lock
);
1552 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1553 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1556 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1560 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1562 #define vm_policy vm_private_data
1565 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1566 struct shmem_inode_info
*info
, pgoff_t index
)
1568 /* Create a pseudo vma that just contains the policy */
1569 vma_init(vma
, NULL
);
1570 /* Bias interleave by inode number to distribute better across nodes */
1571 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1572 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1575 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1577 /* Drop reference taken by mpol_shared_policy_lookup() */
1578 mpol_cond_put(vma
->vm_policy
);
1581 static struct folio
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1582 struct shmem_inode_info
*info
, pgoff_t index
)
1584 struct vm_area_struct pvma
;
1586 struct vm_fault vmf
= {
1590 shmem_pseudo_vma_init(&pvma
, info
, index
);
1591 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1592 shmem_pseudo_vma_destroy(&pvma
);
1596 return page_folio(page
);
1600 * Make sure huge_gfp is always more limited than limit_gfp.
1601 * Some of the flags set permissions, while others set limitations.
1603 static gfp_t
limit_gfp_mask(gfp_t huge_gfp
, gfp_t limit_gfp
)
1605 gfp_t allowflags
= __GFP_IO
| __GFP_FS
| __GFP_RECLAIM
;
1606 gfp_t denyflags
= __GFP_NOWARN
| __GFP_NORETRY
;
1607 gfp_t zoneflags
= limit_gfp
& GFP_ZONEMASK
;
1608 gfp_t result
= huge_gfp
& ~(allowflags
| GFP_ZONEMASK
);
1610 /* Allow allocations only from the originally specified zones. */
1611 result
|= zoneflags
;
1614 * Minimize the result gfp by taking the union with the deny flags,
1615 * and the intersection of the allow flags.
1617 result
|= (limit_gfp
& denyflags
);
1618 result
|= (huge_gfp
& limit_gfp
) & allowflags
;
1623 static struct folio
*shmem_alloc_hugefolio(gfp_t gfp
,
1624 struct shmem_inode_info
*info
, pgoff_t index
)
1626 struct vm_area_struct pvma
;
1627 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1629 struct folio
*folio
;
1631 hindex
= round_down(index
, HPAGE_PMD_NR
);
1632 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1636 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1637 folio
= vma_alloc_folio(gfp
, HPAGE_PMD_ORDER
, &pvma
, 0, true);
1638 shmem_pseudo_vma_destroy(&pvma
);
1640 count_vm_event(THP_FILE_FALLBACK
);
1644 static struct folio
*shmem_alloc_folio(gfp_t gfp
,
1645 struct shmem_inode_info
*info
, pgoff_t index
)
1647 struct vm_area_struct pvma
;
1648 struct folio
*folio
;
1650 shmem_pseudo_vma_init(&pvma
, info
, index
);
1651 folio
= vma_alloc_folio(gfp
, 0, &pvma
, 0, false);
1652 shmem_pseudo_vma_destroy(&pvma
);
1657 static struct folio
*shmem_alloc_and_acct_folio(gfp_t gfp
, struct inode
*inode
,
1658 pgoff_t index
, bool huge
)
1660 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1661 struct folio
*folio
;
1665 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1667 nr
= huge
? HPAGE_PMD_NR
: 1;
1669 err
= shmem_inode_acct_block(inode
, nr
);
1674 folio
= shmem_alloc_hugefolio(gfp
, info
, index
);
1676 folio
= shmem_alloc_folio(gfp
, info
, index
);
1678 __folio_set_locked(folio
);
1679 __folio_set_swapbacked(folio
);
1684 shmem_inode_unacct_blocks(inode
, nr
);
1686 return ERR_PTR(err
);
1690 * When a page is moved from swapcache to shmem filecache (either by the
1691 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1692 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1693 * ignorance of the mapping it belongs to. If that mapping has special
1694 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1695 * we may need to copy to a suitable page before moving to filecache.
1697 * In a future release, this may well be extended to respect cpuset and
1698 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1699 * but for now it is a simple matter of zone.
1701 static bool shmem_should_replace_folio(struct folio
*folio
, gfp_t gfp
)
1703 return folio_zonenum(folio
) > gfp_zone(gfp
);
1706 static int shmem_replace_folio(struct folio
**foliop
, gfp_t gfp
,
1707 struct shmem_inode_info
*info
, pgoff_t index
)
1709 struct folio
*old
, *new;
1710 struct address_space
*swap_mapping
;
1716 entry
= folio_swap_entry(old
);
1717 swap_index
= swp_offset(entry
);
1718 swap_mapping
= swap_address_space(entry
);
1721 * We have arrived here because our zones are constrained, so don't
1722 * limit chance of success by further cpuset and node constraints.
1724 gfp
&= ~GFP_CONSTRAINT_MASK
;
1725 VM_BUG_ON_FOLIO(folio_test_large(old
), old
);
1726 new = shmem_alloc_folio(gfp
, info
, index
);
1731 folio_copy(new, old
);
1732 flush_dcache_folio(new);
1734 __folio_set_locked(new);
1735 __folio_set_swapbacked(new);
1736 folio_mark_uptodate(new);
1737 folio_set_swap_entry(new, entry
);
1738 folio_set_swapcache(new);
1741 * Our caller will very soon move newpage out of swapcache, but it's
1742 * a nice clean interface for us to replace oldpage by newpage there.
1744 xa_lock_irq(&swap_mapping
->i_pages
);
1745 error
= shmem_replace_entry(swap_mapping
, swap_index
, old
, new);
1747 mem_cgroup_migrate(old
, new);
1748 __lruvec_stat_mod_folio(new, NR_FILE_PAGES
, 1);
1749 __lruvec_stat_mod_folio(new, NR_SHMEM
, 1);
1750 __lruvec_stat_mod_folio(old
, NR_FILE_PAGES
, -1);
1751 __lruvec_stat_mod_folio(old
, NR_SHMEM
, -1);
1753 xa_unlock_irq(&swap_mapping
->i_pages
);
1755 if (unlikely(error
)) {
1757 * Is this possible? I think not, now that our callers check
1758 * both PageSwapCache and page_private after getting page lock;
1759 * but be defensive. Reverse old to newpage for clear and free.
1767 folio_clear_swapcache(old
);
1768 old
->private = NULL
;
1771 folio_put_refs(old
, 2);
1775 static void shmem_set_folio_swapin_error(struct inode
*inode
, pgoff_t index
,
1776 struct folio
*folio
, swp_entry_t swap
)
1778 struct address_space
*mapping
= inode
->i_mapping
;
1779 swp_entry_t swapin_error
;
1782 swapin_error
= make_swapin_error_entry();
1783 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
,
1784 swp_to_radix_entry(swap
),
1785 swp_to_radix_entry(swapin_error
), 0);
1786 if (old
!= swp_to_radix_entry(swap
))
1789 folio_wait_writeback(folio
);
1790 delete_from_swap_cache(folio
);
1792 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
1793 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
1794 * in shmem_evict_inode().
1796 shmem_recalc_inode(inode
, -1, -1);
1801 * Swap in the folio pointed to by *foliop.
1802 * Caller has to make sure that *foliop contains a valid swapped folio.
1803 * Returns 0 and the folio in foliop if success. On failure, returns the
1804 * error code and NULL in *foliop.
1806 static int shmem_swapin_folio(struct inode
*inode
, pgoff_t index
,
1807 struct folio
**foliop
, enum sgp_type sgp
,
1808 gfp_t gfp
, struct vm_area_struct
*vma
,
1809 vm_fault_t
*fault_type
)
1811 struct address_space
*mapping
= inode
->i_mapping
;
1812 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1813 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: NULL
;
1814 struct swap_info_struct
*si
;
1815 struct folio
*folio
= NULL
;
1819 VM_BUG_ON(!*foliop
|| !xa_is_value(*foliop
));
1820 swap
= radix_to_swp_entry(*foliop
);
1823 if (is_swapin_error_entry(swap
))
1826 si
= get_swap_device(swap
);
1828 if (!shmem_confirm_swap(mapping
, index
, swap
))
1834 /* Look it up and read it in.. */
1835 folio
= swap_cache_get_folio(swap
, NULL
, 0);
1837 /* Or update major stats only when swapin succeeds?? */
1839 *fault_type
|= VM_FAULT_MAJOR
;
1840 count_vm_event(PGMAJFAULT
);
1841 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1843 /* Here we actually start the io */
1844 folio
= shmem_swapin(swap
, gfp
, info
, index
);
1851 /* We have to do this with folio locked to prevent races */
1853 if (!folio_test_swapcache(folio
) ||
1854 folio_swap_entry(folio
).val
!= swap
.val
||
1855 !shmem_confirm_swap(mapping
, index
, swap
)) {
1859 if (!folio_test_uptodate(folio
)) {
1863 folio_wait_writeback(folio
);
1866 * Some architectures may have to restore extra metadata to the
1867 * folio after reading from swap.
1869 arch_swap_restore(swap
, folio
);
1871 if (shmem_should_replace_folio(folio
, gfp
)) {
1872 error
= shmem_replace_folio(&folio
, gfp
, info
, index
);
1877 error
= shmem_add_to_page_cache(folio
, mapping
, index
,
1878 swp_to_radix_entry(swap
), gfp
,
1883 shmem_recalc_inode(inode
, 0, -1);
1885 if (sgp
== SGP_WRITE
)
1886 folio_mark_accessed(folio
);
1888 delete_from_swap_cache(folio
);
1889 folio_mark_dirty(folio
);
1891 put_swap_device(si
);
1896 if (!shmem_confirm_swap(mapping
, index
, swap
))
1899 shmem_set_folio_swapin_error(inode
, index
, folio
, swap
);
1902 folio_unlock(folio
);
1905 put_swap_device(si
);
1911 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1913 * If we allocate a new one we do not mark it dirty. That's up to the
1914 * vm. If we swap it in we mark it dirty since we also free the swap
1915 * entry since a page cannot live in both the swap and page cache.
1917 * vma, vmf, and fault_type are only supplied by shmem_fault:
1918 * otherwise they are NULL.
1920 static int shmem_get_folio_gfp(struct inode
*inode
, pgoff_t index
,
1921 struct folio
**foliop
, enum sgp_type sgp
, gfp_t gfp
,
1922 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1923 vm_fault_t
*fault_type
)
1925 struct address_space
*mapping
= inode
->i_mapping
;
1926 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1927 struct shmem_sb_info
*sbinfo
;
1928 struct mm_struct
*charge_mm
;
1929 struct folio
*folio
;
1936 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1939 if (sgp
<= SGP_CACHE
&&
1940 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1944 sbinfo
= SHMEM_SB(inode
->i_sb
);
1945 charge_mm
= vma
? vma
->vm_mm
: NULL
;
1947 folio
= filemap_get_entry(mapping
, index
);
1948 if (folio
&& vma
&& userfaultfd_minor(vma
)) {
1949 if (!xa_is_value(folio
))
1951 *fault_type
= handle_userfault(vmf
, VM_UFFD_MINOR
);
1955 if (xa_is_value(folio
)) {
1956 error
= shmem_swapin_folio(inode
, index
, &folio
,
1957 sgp
, gfp
, vma
, fault_type
);
1958 if (error
== -EEXIST
)
1968 /* Has the folio been truncated or swapped out? */
1969 if (unlikely(folio
->mapping
!= mapping
)) {
1970 folio_unlock(folio
);
1974 if (sgp
== SGP_WRITE
)
1975 folio_mark_accessed(folio
);
1976 if (folio_test_uptodate(folio
))
1978 /* fallocated folio */
1979 if (sgp
!= SGP_READ
)
1981 folio_unlock(folio
);
1986 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1987 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1990 if (sgp
== SGP_READ
)
1992 if (sgp
== SGP_NOALLOC
)
1996 * Fast cache lookup and swap lookup did not find it: allocate.
1999 if (vma
&& userfaultfd_missing(vma
)) {
2000 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
2004 if (!shmem_is_huge(inode
, index
, false,
2005 vma
? vma
->vm_mm
: NULL
, vma
? vma
->vm_flags
: 0))
2008 huge_gfp
= vma_thp_gfp_mask(vma
);
2009 huge_gfp
= limit_gfp_mask(huge_gfp
, gfp
);
2010 folio
= shmem_alloc_and_acct_folio(huge_gfp
, inode
, index
, true);
2011 if (IS_ERR(folio
)) {
2013 folio
= shmem_alloc_and_acct_folio(gfp
, inode
, index
, false);
2015 if (IS_ERR(folio
)) {
2018 error
= PTR_ERR(folio
);
2020 if (error
!= -ENOSPC
)
2023 * Try to reclaim some space by splitting a large folio
2024 * beyond i_size on the filesystem.
2029 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
2030 if (ret
== SHRINK_STOP
)
2038 hindex
= round_down(index
, folio_nr_pages(folio
));
2040 if (sgp
== SGP_WRITE
)
2041 __folio_set_referenced(folio
);
2043 error
= shmem_add_to_page_cache(folio
, mapping
, hindex
,
2044 NULL
, gfp
& GFP_RECLAIM_MASK
,
2049 folio_add_lru(folio
);
2050 shmem_recalc_inode(inode
, folio_nr_pages(folio
), 0);
2053 if (folio_test_pmd_mappable(folio
) &&
2054 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
2055 folio_next_index(folio
) - 1) {
2057 * Part of the large folio is beyond i_size: subject
2058 * to shrink under memory pressure.
2060 spin_lock(&sbinfo
->shrinklist_lock
);
2062 * _careful to defend against unlocked access to
2063 * ->shrink_list in shmem_unused_huge_shrink()
2065 if (list_empty_careful(&info
->shrinklist
)) {
2066 list_add_tail(&info
->shrinklist
,
2067 &sbinfo
->shrinklist
);
2068 sbinfo
->shrinklist_len
++;
2070 spin_unlock(&sbinfo
->shrinklist_lock
);
2074 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2076 if (sgp
== SGP_FALLOC
)
2080 * Let SGP_WRITE caller clear ends if write does not fill folio;
2081 * but SGP_FALLOC on a folio fallocated earlier must initialize
2082 * it now, lest undo on failure cancel our earlier guarantee.
2084 if (sgp
!= SGP_WRITE
&& !folio_test_uptodate(folio
)) {
2085 long i
, n
= folio_nr_pages(folio
);
2087 for (i
= 0; i
< n
; i
++)
2088 clear_highpage(folio_page(folio
, i
));
2089 flush_dcache_folio(folio
);
2090 folio_mark_uptodate(folio
);
2093 /* Perhaps the file has been truncated since we checked */
2094 if (sgp
<= SGP_CACHE
&&
2095 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
2097 folio_clear_dirty(folio
);
2098 filemap_remove_folio(folio
);
2099 shmem_recalc_inode(inode
, 0, 0);
2112 shmem_inode_unacct_blocks(inode
, folio_nr_pages(folio
));
2114 if (folio_test_large(folio
)) {
2115 folio_unlock(folio
);
2121 folio_unlock(folio
);
2124 if (error
== -ENOSPC
&& !once
++) {
2125 shmem_recalc_inode(inode
, 0, 0);
2128 if (error
== -EEXIST
)
2133 int shmem_get_folio(struct inode
*inode
, pgoff_t index
, struct folio
**foliop
,
2136 return shmem_get_folio_gfp(inode
, index
, foliop
, sgp
,
2137 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
2141 * This is like autoremove_wake_function, but it removes the wait queue
2142 * entry unconditionally - even if something else had already woken the
2145 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2147 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2148 list_del_init(&wait
->entry
);
2152 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2154 struct vm_area_struct
*vma
= vmf
->vma
;
2155 struct inode
*inode
= file_inode(vma
->vm_file
);
2156 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2157 struct folio
*folio
= NULL
;
2159 vm_fault_t ret
= VM_FAULT_LOCKED
;
2162 * Trinity finds that probing a hole which tmpfs is punching can
2163 * prevent the hole-punch from ever completing: which in turn
2164 * locks writers out with its hold on i_rwsem. So refrain from
2165 * faulting pages into the hole while it's being punched. Although
2166 * shmem_undo_range() does remove the additions, it may be unable to
2167 * keep up, as each new page needs its own unmap_mapping_range() call,
2168 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2170 * It does not matter if we sometimes reach this check just before the
2171 * hole-punch begins, so that one fault then races with the punch:
2172 * we just need to make racing faults a rare case.
2174 * The implementation below would be much simpler if we just used a
2175 * standard mutex or completion: but we cannot take i_rwsem in fault,
2176 * and bloating every shmem inode for this unlikely case would be sad.
2178 if (unlikely(inode
->i_private
)) {
2179 struct shmem_falloc
*shmem_falloc
;
2181 spin_lock(&inode
->i_lock
);
2182 shmem_falloc
= inode
->i_private
;
2184 shmem_falloc
->waitq
&&
2185 vmf
->pgoff
>= shmem_falloc
->start
&&
2186 vmf
->pgoff
< shmem_falloc
->next
) {
2188 wait_queue_head_t
*shmem_falloc_waitq
;
2189 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2191 ret
= VM_FAULT_NOPAGE
;
2192 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2194 ret
= VM_FAULT_RETRY
;
2196 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2197 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2198 TASK_UNINTERRUPTIBLE
);
2199 spin_unlock(&inode
->i_lock
);
2203 * shmem_falloc_waitq points into the shmem_fallocate()
2204 * stack of the hole-punching task: shmem_falloc_waitq
2205 * is usually invalid by the time we reach here, but
2206 * finish_wait() does not dereference it in that case;
2207 * though i_lock needed lest racing with wake_up_all().
2209 spin_lock(&inode
->i_lock
);
2210 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2211 spin_unlock(&inode
->i_lock
);
2217 spin_unlock(&inode
->i_lock
);
2220 err
= shmem_get_folio_gfp(inode
, vmf
->pgoff
, &folio
, SGP_CACHE
,
2221 gfp
, vma
, vmf
, &ret
);
2223 return vmf_error(err
);
2225 vmf
->page
= folio_file_page(folio
, vmf
->pgoff
);
2229 unsigned long shmem_get_unmapped_area(struct file
*file
,
2230 unsigned long uaddr
, unsigned long len
,
2231 unsigned long pgoff
, unsigned long flags
)
2233 unsigned long (*get_area
)(struct file
*,
2234 unsigned long, unsigned long, unsigned long, unsigned long);
2236 unsigned long offset
;
2237 unsigned long inflated_len
;
2238 unsigned long inflated_addr
;
2239 unsigned long inflated_offset
;
2241 if (len
> TASK_SIZE
)
2244 get_area
= current
->mm
->get_unmapped_area
;
2245 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2247 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2249 if (IS_ERR_VALUE(addr
))
2251 if (addr
& ~PAGE_MASK
)
2253 if (addr
> TASK_SIZE
- len
)
2256 if (shmem_huge
== SHMEM_HUGE_DENY
)
2258 if (len
< HPAGE_PMD_SIZE
)
2260 if (flags
& MAP_FIXED
)
2263 * Our priority is to support MAP_SHARED mapped hugely;
2264 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2265 * But if caller specified an address hint and we allocated area there
2266 * successfully, respect that as before.
2271 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2272 struct super_block
*sb
;
2275 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2276 sb
= file_inode(file
)->i_sb
;
2279 * Called directly from mm/mmap.c, or drivers/char/mem.c
2280 * for "/dev/zero", to create a shared anonymous object.
2282 if (IS_ERR(shm_mnt
))
2284 sb
= shm_mnt
->mnt_sb
;
2286 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2290 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2291 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2293 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2296 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2297 if (inflated_len
> TASK_SIZE
)
2299 if (inflated_len
< len
)
2302 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2303 if (IS_ERR_VALUE(inflated_addr
))
2305 if (inflated_addr
& ~PAGE_MASK
)
2308 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2309 inflated_addr
+= offset
- inflated_offset
;
2310 if (inflated_offset
> offset
)
2311 inflated_addr
+= HPAGE_PMD_SIZE
;
2313 if (inflated_addr
> TASK_SIZE
- len
)
2315 return inflated_addr
;
2319 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2321 struct inode
*inode
= file_inode(vma
->vm_file
);
2322 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2325 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2328 struct inode
*inode
= file_inode(vma
->vm_file
);
2331 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2332 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2336 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
2338 struct inode
*inode
= file_inode(file
);
2339 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2340 int retval
= -ENOMEM
;
2343 * What serializes the accesses to info->flags?
2344 * ipc_lock_object() when called from shmctl_do_lock(),
2345 * no serialization needed when called from shm_destroy().
2347 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2348 if (!user_shm_lock(inode
->i_size
, ucounts
))
2350 info
->flags
|= VM_LOCKED
;
2351 mapping_set_unevictable(file
->f_mapping
);
2353 if (!lock
&& (info
->flags
& VM_LOCKED
) && ucounts
) {
2354 user_shm_unlock(inode
->i_size
, ucounts
);
2355 info
->flags
&= ~VM_LOCKED
;
2356 mapping_clear_unevictable(file
->f_mapping
);
2364 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2366 struct inode
*inode
= file_inode(file
);
2367 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2370 ret
= seal_check_future_write(info
->seals
, vma
);
2374 /* arm64 - allow memory tagging on RAM-based files */
2375 vm_flags_set(vma
, VM_MTE_ALLOWED
);
2377 file_accessed(file
);
2378 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2380 vma
->vm_ops
= &shmem_vm_ops
;
2382 vma
->vm_ops
= &shmem_anon_vm_ops
;
2386 #ifdef CONFIG_TMPFS_XATTR
2387 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2390 * chattr's fsflags are unrelated to extended attributes,
2391 * but tmpfs has chosen to enable them under the same config option.
2393 static void shmem_set_inode_flags(struct inode
*inode
, unsigned int fsflags
)
2395 unsigned int i_flags
= 0;
2397 if (fsflags
& FS_NOATIME_FL
)
2398 i_flags
|= S_NOATIME
;
2399 if (fsflags
& FS_APPEND_FL
)
2400 i_flags
|= S_APPEND
;
2401 if (fsflags
& FS_IMMUTABLE_FL
)
2402 i_flags
|= S_IMMUTABLE
;
2404 * But FS_NODUMP_FL does not require any action in i_flags.
2406 inode_set_flags(inode
, i_flags
, S_NOATIME
| S_APPEND
| S_IMMUTABLE
);
2409 static void shmem_set_inode_flags(struct inode
*inode
, unsigned int fsflags
)
2412 #define shmem_initxattrs NULL
2415 static struct offset_ctx
*shmem_get_offset_ctx(struct inode
*inode
)
2417 return &SHMEM_I(inode
)->dir_offsets
;
2420 static struct inode
*__shmem_get_inode(struct mnt_idmap
*idmap
,
2421 struct super_block
*sb
,
2422 struct inode
*dir
, umode_t mode
,
2423 dev_t dev
, unsigned long flags
)
2425 struct inode
*inode
;
2426 struct shmem_inode_info
*info
;
2427 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2431 err
= shmem_reserve_inode(sb
, &ino
);
2433 return ERR_PTR(err
);
2436 inode
= new_inode(sb
);
2439 shmem_free_inode(sb
);
2440 return ERR_PTR(-ENOSPC
);
2444 inode_init_owner(idmap
, inode
, dir
, mode
);
2445 inode
->i_blocks
= 0;
2446 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2447 inode
->i_generation
= get_random_u32();
2448 info
= SHMEM_I(inode
);
2449 memset(info
, 0, (char *)inode
- (char *)info
);
2450 spin_lock_init(&info
->lock
);
2451 atomic_set(&info
->stop_eviction
, 0);
2452 info
->seals
= F_SEAL_SEAL
;
2453 info
->flags
= flags
& VM_NORESERVE
;
2454 info
->i_crtime
= inode
->i_mtime
;
2455 info
->fsflags
= (dir
== NULL
) ? 0 :
2456 SHMEM_I(dir
)->fsflags
& SHMEM_FL_INHERITED
;
2458 shmem_set_inode_flags(inode
, info
->fsflags
);
2459 INIT_LIST_HEAD(&info
->shrinklist
);
2460 INIT_LIST_HEAD(&info
->swaplist
);
2461 INIT_LIST_HEAD(&info
->swaplist
);
2463 mapping_set_unevictable(inode
->i_mapping
);
2464 simple_xattrs_init(&info
->xattrs
);
2465 cache_no_acl(inode
);
2466 mapping_set_large_folios(inode
->i_mapping
);
2468 switch (mode
& S_IFMT
) {
2470 inode
->i_op
= &shmem_special_inode_operations
;
2471 init_special_inode(inode
, mode
, dev
);
2474 inode
->i_mapping
->a_ops
= &shmem_aops
;
2475 inode
->i_op
= &shmem_inode_operations
;
2476 inode
->i_fop
= &shmem_file_operations
;
2477 mpol_shared_policy_init(&info
->policy
,
2478 shmem_get_sbmpol(sbinfo
));
2482 /* Some things misbehave if size == 0 on a directory */
2483 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2484 inode
->i_op
= &shmem_dir_inode_operations
;
2485 inode
->i_fop
= &simple_offset_dir_operations
;
2486 simple_offset_init(shmem_get_offset_ctx(inode
));
2490 * Must not load anything in the rbtree,
2491 * mpol_free_shared_policy will not be called.
2493 mpol_shared_policy_init(&info
->policy
, NULL
);
2497 lockdep_annotate_inode_mutex_key(inode
);
2501 #ifdef CONFIG_TMPFS_QUOTA
2502 static struct inode
*shmem_get_inode(struct mnt_idmap
*idmap
,
2503 struct super_block
*sb
, struct inode
*dir
,
2504 umode_t mode
, dev_t dev
, unsigned long flags
)
2507 struct inode
*inode
;
2509 inode
= __shmem_get_inode(idmap
, sb
, dir
, mode
, dev
, flags
);
2513 err
= dquot_initialize(inode
);
2517 err
= dquot_alloc_inode(inode
);
2525 inode
->i_flags
|= S_NOQUOTA
;
2527 return ERR_PTR(err
);
2530 static inline struct inode
*shmem_get_inode(struct mnt_idmap
*idmap
,
2531 struct super_block
*sb
, struct inode
*dir
,
2532 umode_t mode
, dev_t dev
, unsigned long flags
)
2534 return __shmem_get_inode(idmap
, sb
, dir
, mode
, dev
, flags
);
2536 #endif /* CONFIG_TMPFS_QUOTA */
2538 #ifdef CONFIG_USERFAULTFD
2539 int shmem_mfill_atomic_pte(pmd_t
*dst_pmd
,
2540 struct vm_area_struct
*dst_vma
,
2541 unsigned long dst_addr
,
2542 unsigned long src_addr
,
2544 struct folio
**foliop
)
2546 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2547 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2548 struct address_space
*mapping
= inode
->i_mapping
;
2549 gfp_t gfp
= mapping_gfp_mask(mapping
);
2550 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2552 struct folio
*folio
;
2556 if (shmem_inode_acct_block(inode
, 1)) {
2558 * We may have got a page, returned -ENOENT triggering a retry,
2559 * and now we find ourselves with -ENOMEM. Release the page, to
2560 * avoid a BUG_ON in our caller.
2562 if (unlikely(*foliop
)) {
2571 folio
= shmem_alloc_folio(gfp
, info
, pgoff
);
2573 goto out_unacct_blocks
;
2575 if (uffd_flags_mode_is(flags
, MFILL_ATOMIC_COPY
)) {
2576 page_kaddr
= kmap_local_folio(folio
, 0);
2578 * The read mmap_lock is held here. Despite the
2579 * mmap_lock being read recursive a deadlock is still
2580 * possible if a writer has taken a lock. For example:
2582 * process A thread 1 takes read lock on own mmap_lock
2583 * process A thread 2 calls mmap, blocks taking write lock
2584 * process B thread 1 takes page fault, read lock on own mmap lock
2585 * process B thread 2 calls mmap, blocks taking write lock
2586 * process A thread 1 blocks taking read lock on process B
2587 * process B thread 1 blocks taking read lock on process A
2589 * Disable page faults to prevent potential deadlock
2590 * and retry the copy outside the mmap_lock.
2592 pagefault_disable();
2593 ret
= copy_from_user(page_kaddr
,
2594 (const void __user
*)src_addr
,
2597 kunmap_local(page_kaddr
);
2599 /* fallback to copy_from_user outside mmap_lock */
2600 if (unlikely(ret
)) {
2603 /* don't free the page */
2604 goto out_unacct_blocks
;
2607 flush_dcache_folio(folio
);
2608 } else { /* ZEROPAGE */
2609 clear_user_highpage(&folio
->page
, dst_addr
);
2613 VM_BUG_ON_FOLIO(folio_test_large(folio
), folio
);
2617 VM_BUG_ON(folio_test_locked(folio
));
2618 VM_BUG_ON(folio_test_swapbacked(folio
));
2619 __folio_set_locked(folio
);
2620 __folio_set_swapbacked(folio
);
2621 __folio_mark_uptodate(folio
);
2624 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2625 if (unlikely(pgoff
>= max_off
))
2628 ret
= shmem_add_to_page_cache(folio
, mapping
, pgoff
, NULL
,
2629 gfp
& GFP_RECLAIM_MASK
, dst_vma
->vm_mm
);
2633 ret
= mfill_atomic_install_pte(dst_pmd
, dst_vma
, dst_addr
,
2634 &folio
->page
, true, flags
);
2636 goto out_delete_from_cache
;
2638 shmem_recalc_inode(inode
, 1, 0);
2639 folio_unlock(folio
);
2641 out_delete_from_cache
:
2642 filemap_remove_folio(folio
);
2644 folio_unlock(folio
);
2647 shmem_inode_unacct_blocks(inode
, 1);
2650 #endif /* CONFIG_USERFAULTFD */
2653 static const struct inode_operations shmem_symlink_inode_operations
;
2654 static const struct inode_operations shmem_short_symlink_operations
;
2657 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2658 loff_t pos
, unsigned len
,
2659 struct page
**pagep
, void **fsdata
)
2661 struct inode
*inode
= mapping
->host
;
2662 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2663 pgoff_t index
= pos
>> PAGE_SHIFT
;
2664 struct folio
*folio
;
2667 /* i_rwsem is held by caller */
2668 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2669 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2670 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2672 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2676 ret
= shmem_get_folio(inode
, index
, &folio
, SGP_WRITE
);
2681 *pagep
= folio_file_page(folio
, index
);
2682 if (PageHWPoison(*pagep
)) {
2683 folio_unlock(folio
);
2693 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2694 loff_t pos
, unsigned len
, unsigned copied
,
2695 struct page
*page
, void *fsdata
)
2697 struct folio
*folio
= page_folio(page
);
2698 struct inode
*inode
= mapping
->host
;
2700 if (pos
+ copied
> inode
->i_size
)
2701 i_size_write(inode
, pos
+ copied
);
2703 if (!folio_test_uptodate(folio
)) {
2704 if (copied
< folio_size(folio
)) {
2705 size_t from
= offset_in_folio(folio
, pos
);
2706 folio_zero_segments(folio
, 0, from
,
2707 from
+ copied
, folio_size(folio
));
2709 folio_mark_uptodate(folio
);
2711 folio_mark_dirty(folio
);
2712 folio_unlock(folio
);
2718 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2720 struct file
*file
= iocb
->ki_filp
;
2721 struct inode
*inode
= file_inode(file
);
2722 struct address_space
*mapping
= inode
->i_mapping
;
2724 unsigned long offset
;
2727 loff_t
*ppos
= &iocb
->ki_pos
;
2729 index
= *ppos
>> PAGE_SHIFT
;
2730 offset
= *ppos
& ~PAGE_MASK
;
2733 struct folio
*folio
= NULL
;
2734 struct page
*page
= NULL
;
2736 unsigned long nr
, ret
;
2737 loff_t i_size
= i_size_read(inode
);
2739 end_index
= i_size
>> PAGE_SHIFT
;
2740 if (index
> end_index
)
2742 if (index
== end_index
) {
2743 nr
= i_size
& ~PAGE_MASK
;
2748 error
= shmem_get_folio(inode
, index
, &folio
, SGP_READ
);
2750 if (error
== -EINVAL
)
2755 folio_unlock(folio
);
2757 page
= folio_file_page(folio
, index
);
2758 if (PageHWPoison(page
)) {
2766 * We must evaluate after, since reads (unlike writes)
2767 * are called without i_rwsem protection against truncate
2770 i_size
= i_size_read(inode
);
2771 end_index
= i_size
>> PAGE_SHIFT
;
2772 if (index
== end_index
) {
2773 nr
= i_size
& ~PAGE_MASK
;
2784 * If users can be writing to this page using arbitrary
2785 * virtual addresses, take care about potential aliasing
2786 * before reading the page on the kernel side.
2788 if (mapping_writably_mapped(mapping
))
2789 flush_dcache_page(page
);
2791 * Mark the page accessed if we read the beginning.
2794 folio_mark_accessed(folio
);
2796 * Ok, we have the page, and it's up-to-date, so
2797 * now we can copy it to user space...
2799 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2802 } else if (user_backed_iter(to
)) {
2804 * Copy to user tends to be so well optimized, but
2805 * clear_user() not so much, that it is noticeably
2806 * faster to copy the zero page instead of clearing.
2808 ret
= copy_page_to_iter(ZERO_PAGE(0), offset
, nr
, to
);
2811 * But submitting the same page twice in a row to
2812 * splice() - or others? - can result in confusion:
2813 * so don't attempt that optimization on pipes etc.
2815 ret
= iov_iter_zero(nr
, to
);
2820 index
+= offset
>> PAGE_SHIFT
;
2821 offset
&= ~PAGE_MASK
;
2823 if (!iov_iter_count(to
))
2832 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2833 file_accessed(file
);
2834 return retval
? retval
: error
;
2837 static bool zero_pipe_buf_get(struct pipe_inode_info
*pipe
,
2838 struct pipe_buffer
*buf
)
2843 static void zero_pipe_buf_release(struct pipe_inode_info
*pipe
,
2844 struct pipe_buffer
*buf
)
2848 static bool zero_pipe_buf_try_steal(struct pipe_inode_info
*pipe
,
2849 struct pipe_buffer
*buf
)
2854 static const struct pipe_buf_operations zero_pipe_buf_ops
= {
2855 .release
= zero_pipe_buf_release
,
2856 .try_steal
= zero_pipe_buf_try_steal
,
2857 .get
= zero_pipe_buf_get
,
2860 static size_t splice_zeropage_into_pipe(struct pipe_inode_info
*pipe
,
2861 loff_t fpos
, size_t size
)
2863 size_t offset
= fpos
& ~PAGE_MASK
;
2865 size
= min_t(size_t, size
, PAGE_SIZE
- offset
);
2867 if (!pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
)) {
2868 struct pipe_buffer
*buf
= pipe_head_buf(pipe
);
2870 *buf
= (struct pipe_buffer
) {
2871 .ops
= &zero_pipe_buf_ops
,
2872 .page
= ZERO_PAGE(0),
2882 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
2883 struct pipe_inode_info
*pipe
,
2884 size_t len
, unsigned int flags
)
2886 struct inode
*inode
= file_inode(in
);
2887 struct address_space
*mapping
= inode
->i_mapping
;
2888 struct folio
*folio
= NULL
;
2889 size_t total_spliced
= 0, used
, npages
, n
, part
;
2893 /* Work out how much data we can actually add into the pipe */
2894 used
= pipe_occupancy(pipe
->head
, pipe
->tail
);
2895 npages
= max_t(ssize_t
, pipe
->max_usage
- used
, 0);
2896 len
= min_t(size_t, len
, npages
* PAGE_SIZE
);
2899 if (*ppos
>= i_size_read(inode
))
2902 error
= shmem_get_folio(inode
, *ppos
/ PAGE_SIZE
, &folio
,
2905 if (error
== -EINVAL
)
2910 folio_unlock(folio
);
2912 if (folio_test_hwpoison(folio
) ||
2913 (folio_test_large(folio
) &&
2914 folio_test_has_hwpoisoned(folio
))) {
2921 * i_size must be checked after we know the pages are Uptodate.
2923 * Checking i_size after the check allows us to calculate
2924 * the correct value for "nr", which means the zero-filled
2925 * part of the page is not copied back to userspace (unless
2926 * another truncate extends the file - this is desired though).
2928 isize
= i_size_read(inode
);
2929 if (unlikely(*ppos
>= isize
))
2931 part
= min_t(loff_t
, isize
- *ppos
, len
);
2935 * If users can be writing to this page using arbitrary
2936 * virtual addresses, take care about potential aliasing
2937 * before reading the page on the kernel side.
2939 if (mapping_writably_mapped(mapping
))
2940 flush_dcache_folio(folio
);
2941 folio_mark_accessed(folio
);
2943 * Ok, we have the page, and it's up-to-date, so we can
2944 * now splice it into the pipe.
2946 n
= splice_folio_into_pipe(pipe
, folio
, *ppos
, part
);
2950 n
= splice_zeropage_into_pipe(pipe
, *ppos
, part
);
2958 in
->f_ra
.prev_pos
= *ppos
;
2959 if (pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
))
2969 return total_spliced
? total_spliced
: error
;
2972 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2974 struct address_space
*mapping
= file
->f_mapping
;
2975 struct inode
*inode
= mapping
->host
;
2977 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2978 return generic_file_llseek_size(file
, offset
, whence
,
2979 MAX_LFS_FILESIZE
, i_size_read(inode
));
2984 /* We're holding i_rwsem so we can access i_size directly */
2985 offset
= mapping_seek_hole_data(mapping
, offset
, inode
->i_size
, whence
);
2987 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2988 inode_unlock(inode
);
2992 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2995 struct inode
*inode
= file_inode(file
);
2996 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2997 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2998 struct shmem_falloc shmem_falloc
;
2999 pgoff_t start
, index
, end
, undo_fallocend
;
3002 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
3007 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
3008 struct address_space
*mapping
= file
->f_mapping
;
3009 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
3010 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
3011 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
3013 /* protected by i_rwsem */
3014 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
3019 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
3020 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
3021 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
3022 spin_lock(&inode
->i_lock
);
3023 inode
->i_private
= &shmem_falloc
;
3024 spin_unlock(&inode
->i_lock
);
3026 if ((u64
)unmap_end
> (u64
)unmap_start
)
3027 unmap_mapping_range(mapping
, unmap_start
,
3028 1 + unmap_end
- unmap_start
, 0);
3029 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
3030 /* No need to unmap again: hole-punching leaves COWed pages */
3032 spin_lock(&inode
->i_lock
);
3033 inode
->i_private
= NULL
;
3034 wake_up_all(&shmem_falloc_waitq
);
3035 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
3036 spin_unlock(&inode
->i_lock
);
3041 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3042 error
= inode_newsize_ok(inode
, offset
+ len
);
3046 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
3051 start
= offset
>> PAGE_SHIFT
;
3052 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
3053 /* Try to avoid a swapstorm if len is impossible to satisfy */
3054 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
3059 shmem_falloc
.waitq
= NULL
;
3060 shmem_falloc
.start
= start
;
3061 shmem_falloc
.next
= start
;
3062 shmem_falloc
.nr_falloced
= 0;
3063 shmem_falloc
.nr_unswapped
= 0;
3064 spin_lock(&inode
->i_lock
);
3065 inode
->i_private
= &shmem_falloc
;
3066 spin_unlock(&inode
->i_lock
);
3069 * info->fallocend is only relevant when huge pages might be
3070 * involved: to prevent split_huge_page() freeing fallocated
3071 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3073 undo_fallocend
= info
->fallocend
;
3074 if (info
->fallocend
< end
)
3075 info
->fallocend
= end
;
3077 for (index
= start
; index
< end
; ) {
3078 struct folio
*folio
;
3081 * Good, the fallocate(2) manpage permits EINTR: we may have
3082 * been interrupted because we are using up too much memory.
3084 if (signal_pending(current
))
3086 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
3089 error
= shmem_get_folio(inode
, index
, &folio
,
3092 info
->fallocend
= undo_fallocend
;
3093 /* Remove the !uptodate folios we added */
3094 if (index
> start
) {
3095 shmem_undo_range(inode
,
3096 (loff_t
)start
<< PAGE_SHIFT
,
3097 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
3103 * Here is a more important optimization than it appears:
3104 * a second SGP_FALLOC on the same large folio will clear it,
3105 * making it uptodate and un-undoable if we fail later.
3107 index
= folio_next_index(folio
);
3108 /* Beware 32-bit wraparound */
3113 * Inform shmem_writepage() how far we have reached.
3114 * No need for lock or barrier: we have the page lock.
3116 if (!folio_test_uptodate(folio
))
3117 shmem_falloc
.nr_falloced
+= index
- shmem_falloc
.next
;
3118 shmem_falloc
.next
= index
;
3121 * If !uptodate, leave it that way so that freeable folios
3122 * can be recognized if we need to rollback on error later.
3123 * But mark it dirty so that memory pressure will swap rather
3124 * than free the folios we are allocating (and SGP_CACHE folios
3125 * might still be clean: we now need to mark those dirty too).
3127 folio_mark_dirty(folio
);
3128 folio_unlock(folio
);
3133 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
3134 i_size_write(inode
, offset
+ len
);
3136 spin_lock(&inode
->i_lock
);
3137 inode
->i_private
= NULL
;
3138 spin_unlock(&inode
->i_lock
);
3141 file_modified(file
);
3142 inode_unlock(inode
);
3146 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
3148 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
3150 buf
->f_type
= TMPFS_MAGIC
;
3151 buf
->f_bsize
= PAGE_SIZE
;
3152 buf
->f_namelen
= NAME_MAX
;
3153 if (sbinfo
->max_blocks
) {
3154 buf
->f_blocks
= sbinfo
->max_blocks
;
3156 buf
->f_bfree
= sbinfo
->max_blocks
-
3157 percpu_counter_sum(&sbinfo
->used_blocks
);
3159 if (sbinfo
->max_inodes
) {
3160 buf
->f_files
= sbinfo
->max_inodes
;
3161 buf
->f_ffree
= sbinfo
->free_inodes
;
3163 /* else leave those fields 0 like simple_statfs */
3165 buf
->f_fsid
= uuid_to_fsid(dentry
->d_sb
->s_uuid
.b
);
3171 * File creation. Allocate an inode, and we're done..
3174 shmem_mknod(struct mnt_idmap
*idmap
, struct inode
*dir
,
3175 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3177 struct inode
*inode
;
3180 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
3183 return PTR_ERR(inode
);
3185 error
= simple_acl_create(dir
, inode
);
3188 error
= security_inode_init_security(inode
, dir
,
3190 shmem_initxattrs
, NULL
);
3191 if (error
&& error
!= -EOPNOTSUPP
)
3194 error
= simple_offset_add(shmem_get_offset_ctx(dir
), dentry
);
3198 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3199 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3200 inode_inc_iversion(dir
);
3201 d_instantiate(dentry
, inode
);
3202 dget(dentry
); /* Extra count - pin the dentry in core */
3211 shmem_tmpfile(struct mnt_idmap
*idmap
, struct inode
*dir
,
3212 struct file
*file
, umode_t mode
)
3214 struct inode
*inode
;
3217 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
3219 if (IS_ERR(inode
)) {
3220 error
= PTR_ERR(inode
);
3224 error
= security_inode_init_security(inode
, dir
,
3226 shmem_initxattrs
, NULL
);
3227 if (error
&& error
!= -EOPNOTSUPP
)
3229 error
= simple_acl_create(dir
, inode
);
3232 d_tmpfile(file
, inode
);
3235 return finish_open_simple(file
, error
);
3241 static int shmem_mkdir(struct mnt_idmap
*idmap
, struct inode
*dir
,
3242 struct dentry
*dentry
, umode_t mode
)
3246 error
= shmem_mknod(idmap
, dir
, dentry
, mode
| S_IFDIR
, 0);
3253 static int shmem_create(struct mnt_idmap
*idmap
, struct inode
*dir
,
3254 struct dentry
*dentry
, umode_t mode
, bool excl
)
3256 return shmem_mknod(idmap
, dir
, dentry
, mode
| S_IFREG
, 0);
3262 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
3264 struct inode
*inode
= d_inode(old_dentry
);
3268 * No ordinary (disk based) filesystem counts links as inodes;
3269 * but each new link needs a new dentry, pinning lowmem, and
3270 * tmpfs dentries cannot be pruned until they are unlinked.
3271 * But if an O_TMPFILE file is linked into the tmpfs, the
3272 * first link must skip that, to get the accounting right.
3274 if (inode
->i_nlink
) {
3275 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
3280 ret
= simple_offset_add(shmem_get_offset_ctx(dir
), dentry
);
3283 shmem_free_inode(inode
->i_sb
);
3287 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3288 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3289 inode_inc_iversion(dir
);
3291 ihold(inode
); /* New dentry reference */
3292 dget(dentry
); /* Extra pinning count for the created dentry */
3293 d_instantiate(dentry
, inode
);
3298 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3300 struct inode
*inode
= d_inode(dentry
);
3302 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3303 shmem_free_inode(inode
->i_sb
);
3305 simple_offset_remove(shmem_get_offset_ctx(dir
), dentry
);
3307 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3308 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3309 inode_inc_iversion(dir
);
3311 dput(dentry
); /* Undo the count from "create" - this does all the work */
3315 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3317 if (!simple_empty(dentry
))
3320 drop_nlink(d_inode(dentry
));
3322 return shmem_unlink(dir
, dentry
);
3325 static int shmem_whiteout(struct mnt_idmap
*idmap
,
3326 struct inode
*old_dir
, struct dentry
*old_dentry
)
3328 struct dentry
*whiteout
;
3331 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3335 error
= shmem_mknod(idmap
, old_dir
, whiteout
,
3336 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3342 * Cheat and hash the whiteout while the old dentry is still in
3343 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3345 * d_lookup() will consistently find one of them at this point,
3346 * not sure which one, but that isn't even important.
3353 * The VFS layer already does all the dentry stuff for rename,
3354 * we just have to decrement the usage count for the target if
3355 * it exists so that the VFS layer correctly free's it when it
3358 static int shmem_rename2(struct mnt_idmap
*idmap
,
3359 struct inode
*old_dir
, struct dentry
*old_dentry
,
3360 struct inode
*new_dir
, struct dentry
*new_dentry
,
3363 struct inode
*inode
= d_inode(old_dentry
);
3364 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3367 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3370 if (flags
& RENAME_EXCHANGE
)
3371 return simple_offset_rename_exchange(old_dir
, old_dentry
,
3372 new_dir
, new_dentry
);
3374 if (!simple_empty(new_dentry
))
3377 if (flags
& RENAME_WHITEOUT
) {
3378 error
= shmem_whiteout(idmap
, old_dir
, old_dentry
);
3383 simple_offset_remove(shmem_get_offset_ctx(old_dir
), old_dentry
);
3384 error
= simple_offset_add(shmem_get_offset_ctx(new_dir
), old_dentry
);
3388 if (d_really_is_positive(new_dentry
)) {
3389 (void) shmem_unlink(new_dir
, new_dentry
);
3390 if (they_are_dirs
) {
3391 drop_nlink(d_inode(new_dentry
));
3392 drop_nlink(old_dir
);
3394 } else if (they_are_dirs
) {
3395 drop_nlink(old_dir
);
3399 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3400 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3401 old_dir
->i_ctime
= old_dir
->i_mtime
=
3402 new_dir
->i_ctime
= new_dir
->i_mtime
=
3403 inode
->i_ctime
= current_time(old_dir
);
3404 inode_inc_iversion(old_dir
);
3405 inode_inc_iversion(new_dir
);
3409 static int shmem_symlink(struct mnt_idmap
*idmap
, struct inode
*dir
,
3410 struct dentry
*dentry
, const char *symname
)
3414 struct inode
*inode
;
3415 struct folio
*folio
;
3417 len
= strlen(symname
) + 1;
3418 if (len
> PAGE_SIZE
)
3419 return -ENAMETOOLONG
;
3421 inode
= shmem_get_inode(idmap
, dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3425 return PTR_ERR(inode
);
3427 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3428 shmem_initxattrs
, NULL
);
3429 if (error
&& error
!= -EOPNOTSUPP
)
3432 error
= simple_offset_add(shmem_get_offset_ctx(dir
), dentry
);
3436 inode
->i_size
= len
-1;
3437 if (len
<= SHORT_SYMLINK_LEN
) {
3438 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3439 if (!inode
->i_link
) {
3441 goto out_remove_offset
;
3443 inode
->i_op
= &shmem_short_symlink_operations
;
3445 inode_nohighmem(inode
);
3446 error
= shmem_get_folio(inode
, 0, &folio
, SGP_WRITE
);
3448 goto out_remove_offset
;
3449 inode
->i_mapping
->a_ops
= &shmem_aops
;
3450 inode
->i_op
= &shmem_symlink_inode_operations
;
3451 memcpy(folio_address(folio
), symname
, len
);
3452 folio_mark_uptodate(folio
);
3453 folio_mark_dirty(folio
);
3454 folio_unlock(folio
);
3457 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3458 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3459 inode_inc_iversion(dir
);
3460 d_instantiate(dentry
, inode
);
3465 simple_offset_remove(shmem_get_offset_ctx(dir
), dentry
);
3471 static void shmem_put_link(void *arg
)
3473 folio_mark_accessed(arg
);
3477 static const char *shmem_get_link(struct dentry
*dentry
,
3478 struct inode
*inode
,
3479 struct delayed_call
*done
)
3481 struct folio
*folio
= NULL
;
3485 folio
= filemap_get_folio(inode
->i_mapping
, 0);
3487 return ERR_PTR(-ECHILD
);
3488 if (PageHWPoison(folio_page(folio
, 0)) ||
3489 !folio_test_uptodate(folio
)) {
3491 return ERR_PTR(-ECHILD
);
3494 error
= shmem_get_folio(inode
, 0, &folio
, SGP_READ
);
3496 return ERR_PTR(error
);
3498 return ERR_PTR(-ECHILD
);
3499 if (PageHWPoison(folio_page(folio
, 0))) {
3500 folio_unlock(folio
);
3502 return ERR_PTR(-ECHILD
);
3504 folio_unlock(folio
);
3506 set_delayed_call(done
, shmem_put_link
, folio
);
3507 return folio_address(folio
);
3510 #ifdef CONFIG_TMPFS_XATTR
3512 static int shmem_fileattr_get(struct dentry
*dentry
, struct fileattr
*fa
)
3514 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3516 fileattr_fill_flags(fa
, info
->fsflags
& SHMEM_FL_USER_VISIBLE
);
3521 static int shmem_fileattr_set(struct mnt_idmap
*idmap
,
3522 struct dentry
*dentry
, struct fileattr
*fa
)
3524 struct inode
*inode
= d_inode(dentry
);
3525 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3527 if (fileattr_has_fsx(fa
))
3529 if (fa
->flags
& ~SHMEM_FL_USER_MODIFIABLE
)
3532 info
->fsflags
= (info
->fsflags
& ~SHMEM_FL_USER_MODIFIABLE
) |
3533 (fa
->flags
& SHMEM_FL_USER_MODIFIABLE
);
3535 shmem_set_inode_flags(inode
, info
->fsflags
);
3536 inode
->i_ctime
= current_time(inode
);
3537 inode_inc_iversion(inode
);
3542 * Superblocks without xattr inode operations may get some security.* xattr
3543 * support from the LSM "for free". As soon as we have any other xattrs
3544 * like ACLs, we also need to implement the security.* handlers at
3545 * filesystem level, though.
3549 * Callback for security_inode_init_security() for acquiring xattrs.
3551 static int shmem_initxattrs(struct inode
*inode
,
3552 const struct xattr
*xattr_array
,
3555 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3556 const struct xattr
*xattr
;
3557 struct simple_xattr
*new_xattr
;
3560 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3561 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3565 len
= strlen(xattr
->name
) + 1;
3566 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3568 if (!new_xattr
->name
) {
3573 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3574 XATTR_SECURITY_PREFIX_LEN
);
3575 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3578 simple_xattr_add(&info
->xattrs
, new_xattr
);
3584 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3585 struct dentry
*unused
, struct inode
*inode
,
3586 const char *name
, void *buffer
, size_t size
)
3588 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3590 name
= xattr_full_name(handler
, name
);
3591 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3594 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3595 struct mnt_idmap
*idmap
,
3596 struct dentry
*unused
, struct inode
*inode
,
3597 const char *name
, const void *value
,
3598 size_t size
, int flags
)
3600 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3603 name
= xattr_full_name(handler
, name
);
3604 err
= simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3606 inode
->i_ctime
= current_time(inode
);
3607 inode_inc_iversion(inode
);
3612 static const struct xattr_handler shmem_security_xattr_handler
= {
3613 .prefix
= XATTR_SECURITY_PREFIX
,
3614 .get
= shmem_xattr_handler_get
,
3615 .set
= shmem_xattr_handler_set
,
3618 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3619 .prefix
= XATTR_TRUSTED_PREFIX
,
3620 .get
= shmem_xattr_handler_get
,
3621 .set
= shmem_xattr_handler_set
,
3624 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3625 &shmem_security_xattr_handler
,
3626 &shmem_trusted_xattr_handler
,
3630 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3632 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3633 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3635 #endif /* CONFIG_TMPFS_XATTR */
3637 static const struct inode_operations shmem_short_symlink_operations
= {
3638 .getattr
= shmem_getattr
,
3639 .setattr
= shmem_setattr
,
3640 .get_link
= simple_get_link
,
3641 #ifdef CONFIG_TMPFS_XATTR
3642 .listxattr
= shmem_listxattr
,
3646 static const struct inode_operations shmem_symlink_inode_operations
= {
3647 .getattr
= shmem_getattr
,
3648 .setattr
= shmem_setattr
,
3649 .get_link
= shmem_get_link
,
3650 #ifdef CONFIG_TMPFS_XATTR
3651 .listxattr
= shmem_listxattr
,
3655 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3657 return ERR_PTR(-ESTALE
);
3660 static int shmem_match(struct inode
*ino
, void *vfh
)
3664 inum
= (inum
<< 32) | fh
[1];
3665 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3668 /* Find any alias of inode, but prefer a hashed alias */
3669 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3671 struct dentry
*alias
= d_find_alias(inode
);
3673 return alias
?: d_find_any_alias(inode
);
3677 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3678 struct fid
*fid
, int fh_len
, int fh_type
)
3680 struct inode
*inode
;
3681 struct dentry
*dentry
= NULL
;
3688 inum
= (inum
<< 32) | fid
->raw
[1];
3690 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3691 shmem_match
, fid
->raw
);
3693 dentry
= shmem_find_alias(inode
);
3700 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3701 struct inode
*parent
)
3705 return FILEID_INVALID
;
3708 if (inode_unhashed(inode
)) {
3709 /* Unfortunately insert_inode_hash is not idempotent,
3710 * so as we hash inodes here rather than at creation
3711 * time, we need a lock to ensure we only try
3714 static DEFINE_SPINLOCK(lock
);
3716 if (inode_unhashed(inode
))
3717 __insert_inode_hash(inode
,
3718 inode
->i_ino
+ inode
->i_generation
);
3722 fh
[0] = inode
->i_generation
;
3723 fh
[1] = inode
->i_ino
;
3724 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3730 static const struct export_operations shmem_export_ops
= {
3731 .get_parent
= shmem_get_parent
,
3732 .encode_fh
= shmem_encode_fh
,
3733 .fh_to_dentry
= shmem_fh_to_dentry
,
3751 Opt_usrquota_block_hardlimit
,
3752 Opt_usrquota_inode_hardlimit
,
3753 Opt_grpquota_block_hardlimit
,
3754 Opt_grpquota_inode_hardlimit
,
3757 static const struct constant_table shmem_param_enums_huge
[] = {
3758 {"never", SHMEM_HUGE_NEVER
},
3759 {"always", SHMEM_HUGE_ALWAYS
},
3760 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3761 {"advise", SHMEM_HUGE_ADVISE
},
3765 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3766 fsparam_u32 ("gid", Opt_gid
),
3767 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3768 fsparam_u32oct("mode", Opt_mode
),
3769 fsparam_string("mpol", Opt_mpol
),
3770 fsparam_string("nr_blocks", Opt_nr_blocks
),
3771 fsparam_string("nr_inodes", Opt_nr_inodes
),
3772 fsparam_string("size", Opt_size
),
3773 fsparam_u32 ("uid", Opt_uid
),
3774 fsparam_flag ("inode32", Opt_inode32
),
3775 fsparam_flag ("inode64", Opt_inode64
),
3776 fsparam_flag ("noswap", Opt_noswap
),
3777 #ifdef CONFIG_TMPFS_QUOTA
3778 fsparam_flag ("quota", Opt_quota
),
3779 fsparam_flag ("usrquota", Opt_usrquota
),
3780 fsparam_flag ("grpquota", Opt_grpquota
),
3781 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit
),
3782 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit
),
3783 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit
),
3784 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit
),
3789 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3791 struct shmem_options
*ctx
= fc
->fs_private
;
3792 struct fs_parse_result result
;
3793 unsigned long long size
;
3799 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3805 size
= memparse(param
->string
, &rest
);
3807 size
<<= PAGE_SHIFT
;
3808 size
*= totalram_pages();
3814 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3815 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3818 ctx
->blocks
= memparse(param
->string
, &rest
);
3819 if (*rest
|| ctx
->blocks
> S64_MAX
)
3821 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3824 ctx
->inodes
= memparse(param
->string
, &rest
);
3827 ctx
->seen
|= SHMEM_SEEN_INODES
;
3830 ctx
->mode
= result
.uint_32
& 07777;
3833 kuid
= make_kuid(current_user_ns(), result
.uint_32
);
3834 if (!uid_valid(kuid
))
3838 * The requested uid must be representable in the
3839 * filesystem's idmapping.
3841 if (!kuid_has_mapping(fc
->user_ns
, kuid
))
3847 kgid
= make_kgid(current_user_ns(), result
.uint_32
);
3848 if (!gid_valid(kgid
))
3852 * The requested gid must be representable in the
3853 * filesystem's idmapping.
3855 if (!kgid_has_mapping(fc
->user_ns
, kgid
))
3861 ctx
->huge
= result
.uint_32
;
3862 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3863 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3864 has_transparent_hugepage()))
3865 goto unsupported_parameter
;
3866 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3869 if (IS_ENABLED(CONFIG_NUMA
)) {
3870 mpol_put(ctx
->mpol
);
3872 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3876 goto unsupported_parameter
;
3878 ctx
->full_inums
= false;
3879 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3882 if (sizeof(ino_t
) < 8) {
3884 "Cannot use inode64 with <64bit inums in kernel\n");
3886 ctx
->full_inums
= true;
3887 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3890 if ((fc
->user_ns
!= &init_user_ns
) || !capable(CAP_SYS_ADMIN
)) {
3892 "Turning off swap in unprivileged tmpfs mounts unsupported");
3895 ctx
->seen
|= SHMEM_SEEN_NOSWAP
;
3898 if (fc
->user_ns
!= &init_user_ns
)
3899 return invalfc(fc
, "Quotas in unprivileged tmpfs mounts are unsupported");
3900 ctx
->seen
|= SHMEM_SEEN_QUOTA
;
3901 ctx
->quota_types
|= (QTYPE_MASK_USR
| QTYPE_MASK_GRP
);
3904 if (fc
->user_ns
!= &init_user_ns
)
3905 return invalfc(fc
, "Quotas in unprivileged tmpfs mounts are unsupported");
3906 ctx
->seen
|= SHMEM_SEEN_QUOTA
;
3907 ctx
->quota_types
|= QTYPE_MASK_USR
;
3910 if (fc
->user_ns
!= &init_user_ns
)
3911 return invalfc(fc
, "Quotas in unprivileged tmpfs mounts are unsupported");
3912 ctx
->seen
|= SHMEM_SEEN_QUOTA
;
3913 ctx
->quota_types
|= QTYPE_MASK_GRP
;
3915 case Opt_usrquota_block_hardlimit
:
3916 size
= memparse(param
->string
, &rest
);
3919 if (size
> SHMEM_QUOTA_MAX_SPC_LIMIT
)
3921 "User quota block hardlimit too large.");
3922 ctx
->qlimits
.usrquota_bhardlimit
= size
;
3924 case Opt_grpquota_block_hardlimit
:
3925 size
= memparse(param
->string
, &rest
);
3928 if (size
> SHMEM_QUOTA_MAX_SPC_LIMIT
)
3930 "Group quota block hardlimit too large.");
3931 ctx
->qlimits
.grpquota_bhardlimit
= size
;
3933 case Opt_usrquota_inode_hardlimit
:
3934 size
= memparse(param
->string
, &rest
);
3937 if (size
> SHMEM_QUOTA_MAX_INO_LIMIT
)
3939 "User quota inode hardlimit too large.");
3940 ctx
->qlimits
.usrquota_ihardlimit
= size
;
3942 case Opt_grpquota_inode_hardlimit
:
3943 size
= memparse(param
->string
, &rest
);
3946 if (size
> SHMEM_QUOTA_MAX_INO_LIMIT
)
3948 "Group quota inode hardlimit too large.");
3949 ctx
->qlimits
.grpquota_ihardlimit
= size
;
3954 unsupported_parameter
:
3955 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3957 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3960 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3962 char *options
= data
;
3965 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3970 while (options
!= NULL
) {
3971 char *this_char
= options
;
3974 * NUL-terminate this option: unfortunately,
3975 * mount options form a comma-separated list,
3976 * but mpol's nodelist may also contain commas.
3978 options
= strchr(options
, ',');
3979 if (options
== NULL
)
3982 if (!isdigit(*options
)) {
3988 char *value
= strchr(this_char
, '=');
3994 len
= strlen(value
);
3996 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
4005 * Reconfigure a shmem filesystem.
4007 * Note that we disallow change from limited->unlimited blocks/inodes while any
4008 * are in use; but we must separately disallow unlimited->limited, because in
4009 * that case we have no record of how much is already in use.
4011 static int shmem_reconfigure(struct fs_context
*fc
)
4013 struct shmem_options
*ctx
= fc
->fs_private
;
4014 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
4015 unsigned long inodes
;
4016 struct mempolicy
*mpol
= NULL
;
4019 raw_spin_lock(&sbinfo
->stat_lock
);
4020 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
4022 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
4023 if (!sbinfo
->max_blocks
) {
4024 err
= "Cannot retroactively limit size";
4027 if (percpu_counter_compare(&sbinfo
->used_blocks
,
4029 err
= "Too small a size for current use";
4033 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
4034 if (!sbinfo
->max_inodes
) {
4035 err
= "Cannot retroactively limit inodes";
4038 if (ctx
->inodes
< inodes
) {
4039 err
= "Too few inodes for current use";
4044 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
4045 sbinfo
->next_ino
> UINT_MAX
) {
4046 err
= "Current inum too high to switch to 32-bit inums";
4049 if ((ctx
->seen
& SHMEM_SEEN_NOSWAP
) && ctx
->noswap
&& !sbinfo
->noswap
) {
4050 err
= "Cannot disable swap on remount";
4053 if (!(ctx
->seen
& SHMEM_SEEN_NOSWAP
) && !ctx
->noswap
&& sbinfo
->noswap
) {
4054 err
= "Cannot enable swap on remount if it was disabled on first mount";
4058 if (ctx
->seen
& SHMEM_SEEN_QUOTA
&&
4059 !sb_any_quota_loaded(fc
->root
->d_sb
)) {
4060 err
= "Cannot enable quota on remount";
4064 #ifdef CONFIG_TMPFS_QUOTA
4065 #define CHANGED_LIMIT(name) \
4066 (ctx->qlimits.name## hardlimit && \
4067 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4069 if (CHANGED_LIMIT(usrquota_b
) || CHANGED_LIMIT(usrquota_i
) ||
4070 CHANGED_LIMIT(grpquota_b
) || CHANGED_LIMIT(grpquota_i
)) {
4071 err
= "Cannot change global quota limit on remount";
4074 #endif /* CONFIG_TMPFS_QUOTA */
4076 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
4077 sbinfo
->huge
= ctx
->huge
;
4078 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
4079 sbinfo
->full_inums
= ctx
->full_inums
;
4080 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
4081 sbinfo
->max_blocks
= ctx
->blocks
;
4082 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
4083 sbinfo
->max_inodes
= ctx
->inodes
;
4084 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
4088 * Preserve previous mempolicy unless mpol remount option was specified.
4091 mpol
= sbinfo
->mpol
;
4092 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
4097 sbinfo
->noswap
= true;
4099 raw_spin_unlock(&sbinfo
->stat_lock
);
4103 raw_spin_unlock(&sbinfo
->stat_lock
);
4104 return invalfc(fc
, "%s", err
);
4107 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
4109 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
4110 struct mempolicy
*mpol
;
4112 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
4113 seq_printf(seq
, ",size=%luk",
4114 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
4115 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
4116 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
4117 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
4118 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
4119 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
4120 seq_printf(seq
, ",uid=%u",
4121 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
4122 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
4123 seq_printf(seq
, ",gid=%u",
4124 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
4127 * Showing inode{64,32} might be useful even if it's the system default,
4128 * since then people don't have to resort to checking both here and
4129 * /proc/config.gz to confirm 64-bit inums were successfully applied
4130 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4132 * We hide it when inode64 isn't the default and we are using 32-bit
4133 * inodes, since that probably just means the feature isn't even under
4138 * +-----------------+-----------------+
4139 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4140 * +------------------+-----------------+-----------------+
4141 * | full_inums=true | show | show |
4142 * | full_inums=false | show | hide |
4143 * +------------------+-----------------+-----------------+
4146 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
4147 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
4148 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4149 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4151 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
4153 mpol
= shmem_get_sbmpol(sbinfo
);
4154 shmem_show_mpol(seq
, mpol
);
4157 seq_printf(seq
, ",noswap");
4161 #endif /* CONFIG_TMPFS */
4163 static void shmem_put_super(struct super_block
*sb
)
4165 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
4167 #ifdef CONFIG_TMPFS_QUOTA
4168 shmem_disable_quotas(sb
);
4170 free_percpu(sbinfo
->ino_batch
);
4171 percpu_counter_destroy(&sbinfo
->used_blocks
);
4172 mpol_put(sbinfo
->mpol
);
4174 sb
->s_fs_info
= NULL
;
4177 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
4179 struct shmem_options
*ctx
= fc
->fs_private
;
4180 struct inode
*inode
;
4181 struct shmem_sb_info
*sbinfo
;
4182 int error
= -ENOMEM
;
4184 /* Round up to L1_CACHE_BYTES to resist false sharing */
4185 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
4186 L1_CACHE_BYTES
), GFP_KERNEL
);
4190 sb
->s_fs_info
= sbinfo
;
4194 * Per default we only allow half of the physical ram per
4195 * tmpfs instance, limiting inodes to one per page of lowmem;
4196 * but the internal instance is left unlimited.
4198 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
4199 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
4200 ctx
->blocks
= shmem_default_max_blocks();
4201 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
4202 ctx
->inodes
= shmem_default_max_inodes();
4203 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
4204 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
4205 sbinfo
->noswap
= ctx
->noswap
;
4207 sb
->s_flags
|= SB_NOUSER
;
4209 sb
->s_export_op
= &shmem_export_ops
;
4210 sb
->s_flags
|= SB_NOSEC
| SB_I_VERSION
;
4212 sb
->s_flags
|= SB_NOUSER
;
4214 sbinfo
->max_blocks
= ctx
->blocks
;
4215 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
4216 if (sb
->s_flags
& SB_KERNMOUNT
) {
4217 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
4218 if (!sbinfo
->ino_batch
)
4221 sbinfo
->uid
= ctx
->uid
;
4222 sbinfo
->gid
= ctx
->gid
;
4223 sbinfo
->full_inums
= ctx
->full_inums
;
4224 sbinfo
->mode
= ctx
->mode
;
4225 sbinfo
->huge
= ctx
->huge
;
4226 sbinfo
->mpol
= ctx
->mpol
;
4229 raw_spin_lock_init(&sbinfo
->stat_lock
);
4230 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
4232 spin_lock_init(&sbinfo
->shrinklist_lock
);
4233 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
4235 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
4236 sb
->s_blocksize
= PAGE_SIZE
;
4237 sb
->s_blocksize_bits
= PAGE_SHIFT
;
4238 sb
->s_magic
= TMPFS_MAGIC
;
4239 sb
->s_op
= &shmem_ops
;
4240 sb
->s_time_gran
= 1;
4241 #ifdef CONFIG_TMPFS_XATTR
4242 sb
->s_xattr
= shmem_xattr_handlers
;
4244 #ifdef CONFIG_TMPFS_POSIX_ACL
4245 sb
->s_flags
|= SB_POSIXACL
;
4247 uuid_gen(&sb
->s_uuid
);
4249 #ifdef CONFIG_TMPFS_QUOTA
4250 if (ctx
->seen
& SHMEM_SEEN_QUOTA
) {
4251 sb
->dq_op
= &shmem_quota_operations
;
4252 sb
->s_qcop
= &dquot_quotactl_sysfile_ops
;
4253 sb
->s_quota_types
= QTYPE_MASK_USR
| QTYPE_MASK_GRP
;
4255 /* Copy the default limits from ctx into sbinfo */
4256 memcpy(&sbinfo
->qlimits
, &ctx
->qlimits
,
4257 sizeof(struct shmem_quota_limits
));
4259 if (shmem_enable_quotas(sb
, ctx
->quota_types
))
4262 #endif /* CONFIG_TMPFS_QUOTA */
4264 inode
= shmem_get_inode(&nop_mnt_idmap
, sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0,
4266 if (IS_ERR(inode
)) {
4267 error
= PTR_ERR(inode
);
4270 inode
->i_uid
= sbinfo
->uid
;
4271 inode
->i_gid
= sbinfo
->gid
;
4272 sb
->s_root
= d_make_root(inode
);
4278 shmem_put_super(sb
);
4282 static int shmem_get_tree(struct fs_context
*fc
)
4284 return get_tree_nodev(fc
, shmem_fill_super
);
4287 static void shmem_free_fc(struct fs_context
*fc
)
4289 struct shmem_options
*ctx
= fc
->fs_private
;
4292 mpol_put(ctx
->mpol
);
4297 static const struct fs_context_operations shmem_fs_context_ops
= {
4298 .free
= shmem_free_fc
,
4299 .get_tree
= shmem_get_tree
,
4301 .parse_monolithic
= shmem_parse_options
,
4302 .parse_param
= shmem_parse_one
,
4303 .reconfigure
= shmem_reconfigure
,
4307 static struct kmem_cache
*shmem_inode_cachep
;
4309 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
4311 struct shmem_inode_info
*info
;
4312 info
= alloc_inode_sb(sb
, shmem_inode_cachep
, GFP_KERNEL
);
4315 return &info
->vfs_inode
;
4318 static void shmem_free_in_core_inode(struct inode
*inode
)
4320 if (S_ISLNK(inode
->i_mode
))
4321 kfree(inode
->i_link
);
4322 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
4325 static void shmem_destroy_inode(struct inode
*inode
)
4327 if (S_ISREG(inode
->i_mode
))
4328 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
4329 if (S_ISDIR(inode
->i_mode
))
4330 simple_offset_destroy(shmem_get_offset_ctx(inode
));
4333 static void shmem_init_inode(void *foo
)
4335 struct shmem_inode_info
*info
= foo
;
4336 inode_init_once(&info
->vfs_inode
);
4339 static void shmem_init_inodecache(void)
4341 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
4342 sizeof(struct shmem_inode_info
),
4343 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
4346 static void shmem_destroy_inodecache(void)
4348 kmem_cache_destroy(shmem_inode_cachep
);
4351 /* Keep the page in page cache instead of truncating it */
4352 static int shmem_error_remove_page(struct address_space
*mapping
,
4358 const struct address_space_operations shmem_aops
= {
4359 .writepage
= shmem_writepage
,
4360 .dirty_folio
= noop_dirty_folio
,
4362 .write_begin
= shmem_write_begin
,
4363 .write_end
= shmem_write_end
,
4365 #ifdef CONFIG_MIGRATION
4366 .migrate_folio
= migrate_folio
,
4368 .error_remove_page
= shmem_error_remove_page
,
4370 EXPORT_SYMBOL(shmem_aops
);
4372 static const struct file_operations shmem_file_operations
= {
4374 .open
= generic_file_open
,
4375 .get_unmapped_area
= shmem_get_unmapped_area
,
4377 .llseek
= shmem_file_llseek
,
4378 .read_iter
= shmem_file_read_iter
,
4379 .write_iter
= generic_file_write_iter
,
4380 .fsync
= noop_fsync
,
4381 .splice_read
= shmem_file_splice_read
,
4382 .splice_write
= iter_file_splice_write
,
4383 .fallocate
= shmem_fallocate
,
4387 static const struct inode_operations shmem_inode_operations
= {
4388 .getattr
= shmem_getattr
,
4389 .setattr
= shmem_setattr
,
4390 #ifdef CONFIG_TMPFS_XATTR
4391 .listxattr
= shmem_listxattr
,
4392 .set_acl
= simple_set_acl
,
4393 .fileattr_get
= shmem_fileattr_get
,
4394 .fileattr_set
= shmem_fileattr_set
,
4398 static const struct inode_operations shmem_dir_inode_operations
= {
4400 .getattr
= shmem_getattr
,
4401 .create
= shmem_create
,
4402 .lookup
= simple_lookup
,
4404 .unlink
= shmem_unlink
,
4405 .symlink
= shmem_symlink
,
4406 .mkdir
= shmem_mkdir
,
4407 .rmdir
= shmem_rmdir
,
4408 .mknod
= shmem_mknod
,
4409 .rename
= shmem_rename2
,
4410 .tmpfile
= shmem_tmpfile
,
4411 .get_offset_ctx
= shmem_get_offset_ctx
,
4413 #ifdef CONFIG_TMPFS_XATTR
4414 .listxattr
= shmem_listxattr
,
4415 .fileattr_get
= shmem_fileattr_get
,
4416 .fileattr_set
= shmem_fileattr_set
,
4418 #ifdef CONFIG_TMPFS_POSIX_ACL
4419 .setattr
= shmem_setattr
,
4420 .set_acl
= simple_set_acl
,
4424 static const struct inode_operations shmem_special_inode_operations
= {
4425 .getattr
= shmem_getattr
,
4426 #ifdef CONFIG_TMPFS_XATTR
4427 .listxattr
= shmem_listxattr
,
4429 #ifdef CONFIG_TMPFS_POSIX_ACL
4430 .setattr
= shmem_setattr
,
4431 .set_acl
= simple_set_acl
,
4435 static const struct super_operations shmem_ops
= {
4436 .alloc_inode
= shmem_alloc_inode
,
4437 .free_inode
= shmem_free_in_core_inode
,
4438 .destroy_inode
= shmem_destroy_inode
,
4440 .statfs
= shmem_statfs
,
4441 .show_options
= shmem_show_options
,
4443 #ifdef CONFIG_TMPFS_QUOTA
4444 .get_dquots
= shmem_get_dquots
,
4446 .evict_inode
= shmem_evict_inode
,
4447 .drop_inode
= generic_delete_inode
,
4448 .put_super
= shmem_put_super
,
4449 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4450 .nr_cached_objects
= shmem_unused_huge_count
,
4451 .free_cached_objects
= shmem_unused_huge_scan
,
4455 static const struct vm_operations_struct shmem_vm_ops
= {
4456 .fault
= shmem_fault
,
4457 .map_pages
= filemap_map_pages
,
4459 .set_policy
= shmem_set_policy
,
4460 .get_policy
= shmem_get_policy
,
4464 static const struct vm_operations_struct shmem_anon_vm_ops
= {
4465 .fault
= shmem_fault
,
4466 .map_pages
= filemap_map_pages
,
4468 .set_policy
= shmem_set_policy
,
4469 .get_policy
= shmem_get_policy
,
4473 int shmem_init_fs_context(struct fs_context
*fc
)
4475 struct shmem_options
*ctx
;
4477 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
4481 ctx
->mode
= 0777 | S_ISVTX
;
4482 ctx
->uid
= current_fsuid();
4483 ctx
->gid
= current_fsgid();
4485 fc
->fs_private
= ctx
;
4486 fc
->ops
= &shmem_fs_context_ops
;
4490 static struct file_system_type shmem_fs_type
= {
4491 .owner
= THIS_MODULE
,
4493 .init_fs_context
= shmem_init_fs_context
,
4495 .parameters
= shmem_fs_parameters
,
4497 .kill_sb
= kill_litter_super
,
4499 .fs_flags
= FS_USERNS_MOUNT
| FS_ALLOW_IDMAP
,
4501 .fs_flags
= FS_USERNS_MOUNT
,
4505 void __init
shmem_init(void)
4509 shmem_init_inodecache();
4511 #ifdef CONFIG_TMPFS_QUOTA
4512 error
= register_quota_format(&shmem_quota_format
);
4514 pr_err("Could not register quota format\n");
4519 error
= register_filesystem(&shmem_fs_type
);
4521 pr_err("Could not register tmpfs\n");
4525 shm_mnt
= kern_mount(&shmem_fs_type
);
4526 if (IS_ERR(shm_mnt
)) {
4527 error
= PTR_ERR(shm_mnt
);
4528 pr_err("Could not kern_mount tmpfs\n");
4532 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4533 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
4534 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4536 shmem_huge
= SHMEM_HUGE_NEVER
; /* just in case it was patched */
4541 unregister_filesystem(&shmem_fs_type
);
4543 #ifdef CONFIG_TMPFS_QUOTA
4544 unregister_quota_format(&shmem_quota_format
);
4547 shmem_destroy_inodecache();
4548 shm_mnt
= ERR_PTR(error
);
4551 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4552 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4553 struct kobj_attribute
*attr
, char *buf
)
4555 static const int values
[] = {
4557 SHMEM_HUGE_WITHIN_SIZE
,
4566 for (i
= 0; i
< ARRAY_SIZE(values
); i
++) {
4567 len
+= sysfs_emit_at(buf
, len
,
4568 shmem_huge
== values
[i
] ? "%s[%s]" : "%s%s",
4570 shmem_format_huge(values
[i
]));
4573 len
+= sysfs_emit_at(buf
, len
, "\n");
4578 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4579 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4584 if (count
+ 1 > sizeof(tmp
))
4586 memcpy(tmp
, buf
, count
);
4588 if (count
&& tmp
[count
- 1] == '\n')
4589 tmp
[count
- 1] = '\0';
4591 huge
= shmem_parse_huge(tmp
);
4592 if (huge
== -EINVAL
)
4594 if (!has_transparent_hugepage() &&
4595 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4599 if (shmem_huge
> SHMEM_HUGE_DENY
)
4600 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4604 struct kobj_attribute shmem_enabled_attr
= __ATTR_RW(shmem_enabled
);
4605 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4607 #else /* !CONFIG_SHMEM */
4610 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4612 * This is intended for small system where the benefits of the full
4613 * shmem code (swap-backed and resource-limited) are outweighed by
4614 * their complexity. On systems without swap this code should be
4615 * effectively equivalent, but much lighter weight.
4618 static struct file_system_type shmem_fs_type
= {
4620 .init_fs_context
= ramfs_init_fs_context
,
4621 .parameters
= ramfs_fs_parameters
,
4622 .kill_sb
= ramfs_kill_sb
,
4623 .fs_flags
= FS_USERNS_MOUNT
,
4626 void __init
shmem_init(void)
4628 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4630 shm_mnt
= kern_mount(&shmem_fs_type
);
4631 BUG_ON(IS_ERR(shm_mnt
));
4634 int shmem_unuse(unsigned int type
)
4639 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
4644 void shmem_unlock_mapping(struct address_space
*mapping
)
4649 unsigned long shmem_get_unmapped_area(struct file
*file
,
4650 unsigned long addr
, unsigned long len
,
4651 unsigned long pgoff
, unsigned long flags
)
4653 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4657 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4659 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4661 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4663 #define shmem_vm_ops generic_file_vm_ops
4664 #define shmem_anon_vm_ops generic_file_vm_ops
4665 #define shmem_file_operations ramfs_file_operations
4666 #define shmem_acct_size(flags, size) 0
4667 #define shmem_unacct_size(flags, size) do {} while (0)
4669 static inline struct inode
*shmem_get_inode(struct mnt_idmap
*idmap
, struct super_block
*sb
, struct inode
*dir
,
4670 umode_t mode
, dev_t dev
, unsigned long flags
)
4672 struct inode
*inode
= ramfs_get_inode(sb
, dir
, mode
, dev
);
4673 return inode
? inode
: ERR_PTR(-ENOSPC
);
4676 #endif /* CONFIG_SHMEM */
4680 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4681 unsigned long flags
, unsigned int i_flags
)
4683 struct inode
*inode
;
4687 return ERR_CAST(mnt
);
4689 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4690 return ERR_PTR(-EINVAL
);
4692 if (shmem_acct_size(flags
, size
))
4693 return ERR_PTR(-ENOMEM
);
4695 if (is_idmapped_mnt(mnt
))
4696 return ERR_PTR(-EINVAL
);
4698 inode
= shmem_get_inode(&nop_mnt_idmap
, mnt
->mnt_sb
, NULL
,
4699 S_IFREG
| S_IRWXUGO
, 0, flags
);
4701 if (IS_ERR(inode
)) {
4702 shmem_unacct_size(flags
, size
);
4703 return ERR_CAST(inode
);
4705 inode
->i_flags
|= i_flags
;
4706 inode
->i_size
= size
;
4707 clear_nlink(inode
); /* It is unlinked */
4708 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4710 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4711 &shmem_file_operations
);
4718 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4719 * kernel internal. There will be NO LSM permission checks against the
4720 * underlying inode. So users of this interface must do LSM checks at a
4721 * higher layer. The users are the big_key and shm implementations. LSM
4722 * checks are provided at the key or shm level rather than the inode.
4723 * @name: name for dentry (to be seen in /proc/<pid>/maps
4724 * @size: size to be set for the file
4725 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4727 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4729 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4733 * shmem_file_setup - get an unlinked file living in tmpfs
4734 * @name: name for dentry (to be seen in /proc/<pid>/maps
4735 * @size: size to be set for the file
4736 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4738 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4740 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4742 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4745 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4746 * @mnt: the tmpfs mount where the file will be created
4747 * @name: name for dentry (to be seen in /proc/<pid>/maps
4748 * @size: size to be set for the file
4749 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4751 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4752 loff_t size
, unsigned long flags
)
4754 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4756 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4759 * shmem_zero_setup - setup a shared anonymous mapping
4760 * @vma: the vma to be mmapped is prepared by do_mmap
4762 int shmem_zero_setup(struct vm_area_struct
*vma
)
4765 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4768 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4769 * between XFS directory reading and selinux: since this file is only
4770 * accessible to the user through its mapping, use S_PRIVATE flag to
4771 * bypass file security, in the same way as shmem_kernel_file_setup().
4773 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4775 return PTR_ERR(file
);
4779 vma
->vm_file
= file
;
4780 vma
->vm_ops
= &shmem_anon_vm_ops
;
4786 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4787 * @mapping: the folio's address_space
4788 * @index: the folio index
4789 * @gfp: the page allocator flags to use if allocating
4791 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4792 * with any new page allocations done using the specified allocation flags.
4793 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4794 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4795 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4797 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4798 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4800 struct folio
*shmem_read_folio_gfp(struct address_space
*mapping
,
4801 pgoff_t index
, gfp_t gfp
)
4804 struct inode
*inode
= mapping
->host
;
4805 struct folio
*folio
;
4808 BUG_ON(!shmem_mapping(mapping
));
4809 error
= shmem_get_folio_gfp(inode
, index
, &folio
, SGP_CACHE
,
4810 gfp
, NULL
, NULL
, NULL
);
4812 return ERR_PTR(error
);
4814 folio_unlock(folio
);
4818 * The tiny !SHMEM case uses ramfs without swap
4820 return mapping_read_folio_gfp(mapping
, index
, gfp
);
4823 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp
);
4825 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4826 pgoff_t index
, gfp_t gfp
)
4828 struct folio
*folio
= shmem_read_folio_gfp(mapping
, index
, gfp
);
4832 return &folio
->page
;
4834 page
= folio_file_page(folio
, index
);
4835 if (PageHWPoison(page
)) {
4837 return ERR_PTR(-EIO
);
4842 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);