2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
43 static struct vfsmount
*shm_mnt
;
47 * This virtual memory filesystem is heavily based on the ramfs. It
48 * extends ramfs by the ability to use swap and honor resource limits
49 * which makes it a completely usable filesystem.
52 #include <linux/xattr.h>
53 #include <linux/exportfs.h>
54 #include <linux/posix_acl.h>
55 #include <linux/posix_acl_xattr.h>
56 #include <linux/mman.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/backing-dev.h>
60 #include <linux/shmem_fs.h>
61 #include <linux/writeback.h>
62 #include <linux/blkdev.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/userfaultfd_k.h>
80 #include <linux/rmap.h>
81 #include <linux/uuid.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_rwsem making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc
{
102 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
103 pgoff_t start
; /* start of range currently being fallocated */
104 pgoff_t next
; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
109 struct shmem_options
{
110 unsigned long long blocks
;
111 unsigned long long inodes
;
112 struct mempolicy
*mpol
;
119 #define SHMEM_SEEN_BLOCKS 1
120 #define SHMEM_SEEN_INODES 2
121 #define SHMEM_SEEN_HUGE 4
122 #define SHMEM_SEEN_INUMS 8
126 static unsigned long shmem_default_max_blocks(void)
128 return totalram_pages() / 2;
131 static unsigned long shmem_default_max_inodes(void)
133 unsigned long nr_pages
= totalram_pages();
135 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
139 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
140 struct page
**pagep
, enum sgp_type sgp
,
141 gfp_t gfp
, struct vm_area_struct
*vma
,
142 vm_fault_t
*fault_type
);
143 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
144 struct page
**pagep
, enum sgp_type sgp
,
145 gfp_t gfp
, struct vm_area_struct
*vma
,
146 struct vm_fault
*vmf
, vm_fault_t
*fault_type
);
148 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
149 struct page
**pagep
, enum sgp_type sgp
)
151 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
152 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
155 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
157 return sb
->s_fs_info
;
161 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
162 * for shared memory and for shared anonymous (/dev/zero) mappings
163 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
164 * consistent with the pre-accounting of private mappings ...
166 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
168 return (flags
& VM_NORESERVE
) ?
169 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
172 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
174 if (!(flags
& VM_NORESERVE
))
175 vm_unacct_memory(VM_ACCT(size
));
178 static inline int shmem_reacct_size(unsigned long flags
,
179 loff_t oldsize
, loff_t newsize
)
181 if (!(flags
& VM_NORESERVE
)) {
182 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
183 return security_vm_enough_memory_mm(current
->mm
,
184 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
185 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
186 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
192 * ... whereas tmpfs objects are accounted incrementally as
193 * pages are allocated, in order to allow large sparse files.
194 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
195 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
197 static inline int shmem_acct_block(unsigned long flags
, long pages
)
199 if (!(flags
& VM_NORESERVE
))
202 return security_vm_enough_memory_mm(current
->mm
,
203 pages
* VM_ACCT(PAGE_SIZE
));
206 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
208 if (flags
& VM_NORESERVE
)
209 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
212 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
214 struct shmem_inode_info
*info
= SHMEM_I(inode
);
215 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
217 if (shmem_acct_block(info
->flags
, pages
))
220 if (sbinfo
->max_blocks
) {
221 if (percpu_counter_compare(&sbinfo
->used_blocks
,
222 sbinfo
->max_blocks
- pages
) > 0)
224 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
230 shmem_unacct_blocks(info
->flags
, pages
);
234 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
236 struct shmem_inode_info
*info
= SHMEM_I(inode
);
237 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
239 if (sbinfo
->max_blocks
)
240 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
241 shmem_unacct_blocks(info
->flags
, pages
);
244 static const struct super_operations shmem_ops
;
245 const struct address_space_operations shmem_aops
;
246 static const struct file_operations shmem_file_operations
;
247 static const struct inode_operations shmem_inode_operations
;
248 static const struct inode_operations shmem_dir_inode_operations
;
249 static const struct inode_operations shmem_special_inode_operations
;
250 static const struct vm_operations_struct shmem_vm_ops
;
251 static struct file_system_type shmem_fs_type
;
253 bool vma_is_shmem(struct vm_area_struct
*vma
)
255 return vma
->vm_ops
== &shmem_vm_ops
;
258 static LIST_HEAD(shmem_swaplist
);
259 static DEFINE_MUTEX(shmem_swaplist_mutex
);
262 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
263 * produces a novel ino for the newly allocated inode.
265 * It may also be called when making a hard link to permit the space needed by
266 * each dentry. However, in that case, no new inode number is needed since that
267 * internally draws from another pool of inode numbers (currently global
268 * get_next_ino()). This case is indicated by passing NULL as inop.
270 #define SHMEM_INO_BATCH 1024
271 static int shmem_reserve_inode(struct super_block
*sb
, ino_t
*inop
)
273 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
276 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
277 raw_spin_lock(&sbinfo
->stat_lock
);
278 if (sbinfo
->max_inodes
) {
279 if (!sbinfo
->free_inodes
) {
280 raw_spin_unlock(&sbinfo
->stat_lock
);
283 sbinfo
->free_inodes
--;
286 ino
= sbinfo
->next_ino
++;
287 if (unlikely(is_zero_ino(ino
)))
288 ino
= sbinfo
->next_ino
++;
289 if (unlikely(!sbinfo
->full_inums
&&
292 * Emulate get_next_ino uint wraparound for
295 if (IS_ENABLED(CONFIG_64BIT
))
296 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
297 __func__
, MINOR(sb
->s_dev
));
298 sbinfo
->next_ino
= 1;
299 ino
= sbinfo
->next_ino
++;
303 raw_spin_unlock(&sbinfo
->stat_lock
);
306 * __shmem_file_setup, one of our callers, is lock-free: it
307 * doesn't hold stat_lock in shmem_reserve_inode since
308 * max_inodes is always 0, and is called from potentially
309 * unknown contexts. As such, use a per-cpu batched allocator
310 * which doesn't require the per-sb stat_lock unless we are at
311 * the batch boundary.
313 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
314 * shmem mounts are not exposed to userspace, so we don't need
315 * to worry about things like glibc compatibility.
319 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
321 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
322 raw_spin_lock(&sbinfo
->stat_lock
);
323 ino
= sbinfo
->next_ino
;
324 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
325 raw_spin_unlock(&sbinfo
->stat_lock
);
326 if (unlikely(is_zero_ino(ino
)))
337 static void shmem_free_inode(struct super_block
*sb
)
339 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
340 if (sbinfo
->max_inodes
) {
341 raw_spin_lock(&sbinfo
->stat_lock
);
342 sbinfo
->free_inodes
++;
343 raw_spin_unlock(&sbinfo
->stat_lock
);
348 * shmem_recalc_inode - recalculate the block usage of an inode
349 * @inode: inode to recalc
351 * We have to calculate the free blocks since the mm can drop
352 * undirtied hole pages behind our back.
354 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
355 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
357 * It has to be called with the spinlock held.
359 static void shmem_recalc_inode(struct inode
*inode
)
361 struct shmem_inode_info
*info
= SHMEM_I(inode
);
364 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
366 info
->alloced
-= freed
;
367 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
368 shmem_inode_unacct_blocks(inode
, freed
);
372 bool shmem_charge(struct inode
*inode
, long pages
)
374 struct shmem_inode_info
*info
= SHMEM_I(inode
);
377 if (!shmem_inode_acct_block(inode
, pages
))
380 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
381 inode
->i_mapping
->nrpages
+= pages
;
383 spin_lock_irqsave(&info
->lock
, flags
);
384 info
->alloced
+= pages
;
385 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
386 shmem_recalc_inode(inode
);
387 spin_unlock_irqrestore(&info
->lock
, flags
);
392 void shmem_uncharge(struct inode
*inode
, long pages
)
394 struct shmem_inode_info
*info
= SHMEM_I(inode
);
397 /* nrpages adjustment done by __delete_from_page_cache() or caller */
399 spin_lock_irqsave(&info
->lock
, flags
);
400 info
->alloced
-= pages
;
401 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
402 shmem_recalc_inode(inode
);
403 spin_unlock_irqrestore(&info
->lock
, flags
);
405 shmem_inode_unacct_blocks(inode
, pages
);
409 * Replace item expected in xarray by a new item, while holding xa_lock.
411 static int shmem_replace_entry(struct address_space
*mapping
,
412 pgoff_t index
, void *expected
, void *replacement
)
414 XA_STATE(xas
, &mapping
->i_pages
, index
);
417 VM_BUG_ON(!expected
);
418 VM_BUG_ON(!replacement
);
419 item
= xas_load(&xas
);
420 if (item
!= expected
)
422 xas_store(&xas
, replacement
);
427 * Sometimes, before we decide whether to proceed or to fail, we must check
428 * that an entry was not already brought back from swap by a racing thread.
430 * Checking page is not enough: by the time a SwapCache page is locked, it
431 * might be reused, and again be SwapCache, using the same swap as before.
433 static bool shmem_confirm_swap(struct address_space
*mapping
,
434 pgoff_t index
, swp_entry_t swap
)
436 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
440 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
443 * disables huge pages for the mount;
445 * enables huge pages for the mount;
446 * SHMEM_HUGE_WITHIN_SIZE:
447 * only allocate huge pages if the page will be fully within i_size,
448 * also respect fadvise()/madvise() hints;
450 * only allocate huge pages if requested with fadvise()/madvise();
453 #define SHMEM_HUGE_NEVER 0
454 #define SHMEM_HUGE_ALWAYS 1
455 #define SHMEM_HUGE_WITHIN_SIZE 2
456 #define SHMEM_HUGE_ADVISE 3
460 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
463 * disables huge on shm_mnt and all mounts, for emergency use;
465 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
468 #define SHMEM_HUGE_DENY (-1)
469 #define SHMEM_HUGE_FORCE (-2)
471 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
472 /* ifdef here to avoid bloating shmem.o when not necessary */
474 static int shmem_huge __read_mostly
= SHMEM_HUGE_NEVER
;
476 bool shmem_is_huge(struct vm_area_struct
*vma
,
477 struct inode
*inode
, pgoff_t index
)
481 if (shmem_huge
== SHMEM_HUGE_DENY
)
483 if (vma
&& ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
484 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
)))
486 if (shmem_huge
== SHMEM_HUGE_FORCE
)
489 switch (SHMEM_SB(inode
->i_sb
)->huge
) {
490 case SHMEM_HUGE_ALWAYS
:
492 case SHMEM_HUGE_WITHIN_SIZE
:
493 index
= round_up(index
+ 1, HPAGE_PMD_NR
);
494 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
495 if (i_size
>> PAGE_SHIFT
>= index
)
498 case SHMEM_HUGE_ADVISE
:
499 if (vma
&& (vma
->vm_flags
& VM_HUGEPAGE
))
507 #if defined(CONFIG_SYSFS)
508 static int shmem_parse_huge(const char *str
)
510 if (!strcmp(str
, "never"))
511 return SHMEM_HUGE_NEVER
;
512 if (!strcmp(str
, "always"))
513 return SHMEM_HUGE_ALWAYS
;
514 if (!strcmp(str
, "within_size"))
515 return SHMEM_HUGE_WITHIN_SIZE
;
516 if (!strcmp(str
, "advise"))
517 return SHMEM_HUGE_ADVISE
;
518 if (!strcmp(str
, "deny"))
519 return SHMEM_HUGE_DENY
;
520 if (!strcmp(str
, "force"))
521 return SHMEM_HUGE_FORCE
;
526 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
527 static const char *shmem_format_huge(int huge
)
530 case SHMEM_HUGE_NEVER
:
532 case SHMEM_HUGE_ALWAYS
:
534 case SHMEM_HUGE_WITHIN_SIZE
:
535 return "within_size";
536 case SHMEM_HUGE_ADVISE
:
538 case SHMEM_HUGE_DENY
:
540 case SHMEM_HUGE_FORCE
:
549 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
550 struct shrink_control
*sc
, unsigned long nr_to_split
)
552 LIST_HEAD(list
), *pos
, *next
;
553 LIST_HEAD(to_remove
);
555 struct shmem_inode_info
*info
;
557 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
560 if (list_empty(&sbinfo
->shrinklist
))
563 spin_lock(&sbinfo
->shrinklist_lock
);
564 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
565 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
568 inode
= igrab(&info
->vfs_inode
);
570 /* inode is about to be evicted */
572 list_del_init(&info
->shrinklist
);
576 /* Check if there's anything to gain */
577 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
578 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
579 list_move(&info
->shrinklist
, &to_remove
);
583 list_move(&info
->shrinklist
, &list
);
585 sbinfo
->shrinklist_len
--;
589 spin_unlock(&sbinfo
->shrinklist_lock
);
591 list_for_each_safe(pos
, next
, &to_remove
) {
592 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
593 inode
= &info
->vfs_inode
;
594 list_del_init(&info
->shrinklist
);
598 list_for_each_safe(pos
, next
, &list
) {
601 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
602 inode
= &info
->vfs_inode
;
604 if (nr_to_split
&& split
>= nr_to_split
)
607 page
= find_get_page(inode
->i_mapping
,
608 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
612 /* No huge page at the end of the file: nothing to split */
613 if (!PageTransHuge(page
)) {
619 * Move the inode on the list back to shrinklist if we failed
620 * to lock the page at this time.
622 * Waiting for the lock may lead to deadlock in the
625 if (!trylock_page(page
)) {
630 ret
= split_huge_page(page
);
634 /* If split failed move the inode on the list back to shrinklist */
640 list_del_init(&info
->shrinklist
);
644 * Make sure the inode is either on the global list or deleted
645 * from any local list before iput() since it could be deleted
646 * in another thread once we put the inode (then the local list
649 spin_lock(&sbinfo
->shrinklist_lock
);
650 list_move(&info
->shrinklist
, &sbinfo
->shrinklist
);
651 sbinfo
->shrinklist_len
++;
652 spin_unlock(&sbinfo
->shrinklist_lock
);
660 static long shmem_unused_huge_scan(struct super_block
*sb
,
661 struct shrink_control
*sc
)
663 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
665 if (!READ_ONCE(sbinfo
->shrinklist_len
))
668 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
671 static long shmem_unused_huge_count(struct super_block
*sb
,
672 struct shrink_control
*sc
)
674 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
675 return READ_ONCE(sbinfo
->shrinklist_len
);
677 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
679 #define shmem_huge SHMEM_HUGE_DENY
681 bool shmem_is_huge(struct vm_area_struct
*vma
,
682 struct inode
*inode
, pgoff_t index
)
687 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
688 struct shrink_control
*sc
, unsigned long nr_to_split
)
692 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
695 * Like add_to_page_cache_locked, but error if expected item has gone.
697 static int shmem_add_to_page_cache(struct page
*page
,
698 struct address_space
*mapping
,
699 pgoff_t index
, void *expected
, gfp_t gfp
,
700 struct mm_struct
*charge_mm
)
702 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, compound_order(page
));
704 unsigned long nr
= compound_nr(page
);
707 VM_BUG_ON_PAGE(PageTail(page
), page
);
708 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
709 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
710 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
711 VM_BUG_ON(expected
&& PageTransHuge(page
));
713 page_ref_add(page
, nr
);
714 page
->mapping
= mapping
;
717 if (!PageSwapCache(page
)) {
718 error
= mem_cgroup_charge(page
, charge_mm
, gfp
);
720 if (PageTransHuge(page
)) {
721 count_vm_event(THP_FILE_FALLBACK
);
722 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
727 cgroup_throttle_swaprate(page
, gfp
);
732 entry
= xas_find_conflict(&xas
);
733 if (entry
!= expected
)
734 xas_set_err(&xas
, -EEXIST
);
735 xas_create_range(&xas
);
739 xas_store(&xas
, page
);
744 if (PageTransHuge(page
)) {
745 count_vm_event(THP_FILE_ALLOC
);
746 __mod_lruvec_page_state(page
, NR_SHMEM_THPS
, nr
);
748 mapping
->nrpages
+= nr
;
749 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, nr
);
750 __mod_lruvec_page_state(page
, NR_SHMEM
, nr
);
752 xas_unlock_irq(&xas
);
753 } while (xas_nomem(&xas
, gfp
));
755 if (xas_error(&xas
)) {
756 error
= xas_error(&xas
);
762 page
->mapping
= NULL
;
763 page_ref_sub(page
, nr
);
768 * Like delete_from_page_cache, but substitutes swap for page.
770 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
772 struct address_space
*mapping
= page
->mapping
;
775 VM_BUG_ON_PAGE(PageCompound(page
), page
);
777 xa_lock_irq(&mapping
->i_pages
);
778 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
779 page
->mapping
= NULL
;
781 __dec_lruvec_page_state(page
, NR_FILE_PAGES
);
782 __dec_lruvec_page_state(page
, NR_SHMEM
);
783 xa_unlock_irq(&mapping
->i_pages
);
789 * Remove swap entry from page cache, free the swap and its page cache.
791 static int shmem_free_swap(struct address_space
*mapping
,
792 pgoff_t index
, void *radswap
)
796 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
799 free_swap_and_cache(radix_to_swp_entry(radswap
));
804 * Determine (in bytes) how many of the shmem object's pages mapped by the
805 * given offsets are swapped out.
807 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
808 * as long as the inode doesn't go away and racy results are not a problem.
810 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
811 pgoff_t start
, pgoff_t end
)
813 XA_STATE(xas
, &mapping
->i_pages
, start
);
815 unsigned long swapped
= 0;
818 xas_for_each(&xas
, page
, end
- 1) {
819 if (xas_retry(&xas
, page
))
821 if (xa_is_value(page
))
824 if (need_resched()) {
832 return swapped
<< PAGE_SHIFT
;
836 * Determine (in bytes) how many of the shmem object's pages mapped by the
837 * given vma is swapped out.
839 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
840 * as long as the inode doesn't go away and racy results are not a problem.
842 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
844 struct inode
*inode
= file_inode(vma
->vm_file
);
845 struct shmem_inode_info
*info
= SHMEM_I(inode
);
846 struct address_space
*mapping
= inode
->i_mapping
;
847 unsigned long swapped
;
849 /* Be careful as we don't hold info->lock */
850 swapped
= READ_ONCE(info
->swapped
);
853 * The easier cases are when the shmem object has nothing in swap, or
854 * the vma maps it whole. Then we can simply use the stats that we
860 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
861 return swapped
<< PAGE_SHIFT
;
863 /* Here comes the more involved part */
864 return shmem_partial_swap_usage(mapping
,
865 linear_page_index(vma
, vma
->vm_start
),
866 linear_page_index(vma
, vma
->vm_end
));
870 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
872 void shmem_unlock_mapping(struct address_space
*mapping
)
879 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
881 while (!mapping_unevictable(mapping
)) {
882 if (!pagevec_lookup(&pvec
, mapping
, &index
))
884 check_move_unevictable_pages(&pvec
);
885 pagevec_release(&pvec
);
891 * Check whether a hole-punch or truncation needs to split a huge page,
892 * returning true if no split was required, or the split has been successful.
894 * Eviction (or truncation to 0 size) should never need to split a huge page;
895 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
896 * head, and then succeeded to trylock on tail.
898 * A split can only succeed when there are no additional references on the
899 * huge page: so the split below relies upon find_get_entries() having stopped
900 * when it found a subpage of the huge page, without getting further references.
902 static bool shmem_punch_compound(struct page
*page
, pgoff_t start
, pgoff_t end
)
904 if (!PageTransCompound(page
))
907 /* Just proceed to delete a huge page wholly within the range punched */
908 if (PageHead(page
) &&
909 page
->index
>= start
&& page
->index
+ HPAGE_PMD_NR
<= end
)
912 /* Try to split huge page, so we can truly punch the hole or truncate */
913 return split_huge_page(page
) >= 0;
917 * Remove range of pages and swap entries from page cache, and free them.
918 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
920 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
923 struct address_space
*mapping
= inode
->i_mapping
;
924 struct shmem_inode_info
*info
= SHMEM_I(inode
);
925 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
926 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
927 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
928 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
930 pgoff_t indices
[PAGEVEC_SIZE
];
931 long nr_swaps_freed
= 0;
936 end
= -1; /* unsigned, so actually very big */
938 if (info
->fallocend
> start
&& info
->fallocend
<= end
&& !unfalloc
)
939 info
->fallocend
= start
;
943 while (index
< end
&& find_lock_entries(mapping
, index
, end
- 1,
945 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
946 struct page
*page
= pvec
.pages
[i
];
950 if (xa_is_value(page
)) {
953 nr_swaps_freed
+= !shmem_free_swap(mapping
,
957 index
+= thp_nr_pages(page
) - 1;
959 if (!unfalloc
|| !PageUptodate(page
))
960 truncate_inode_page(mapping
, page
);
963 pagevec_remove_exceptionals(&pvec
);
964 pagevec_release(&pvec
);
970 struct page
*page
= NULL
;
971 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
973 unsigned int top
= PAGE_SIZE
;
978 zero_user_segment(page
, partial_start
, top
);
979 set_page_dirty(page
);
985 struct page
*page
= NULL
;
986 shmem_getpage(inode
, end
, &page
, SGP_READ
);
988 zero_user_segment(page
, 0, partial_end
);
989 set_page_dirty(page
);
998 while (index
< end
) {
1001 if (!find_get_entries(mapping
, index
, end
- 1, &pvec
,
1003 /* If all gone or hole-punch or unfalloc, we're done */
1004 if (index
== start
|| end
!= -1)
1006 /* But if truncating, restart to make sure all gone */
1010 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
1011 struct page
*page
= pvec
.pages
[i
];
1014 if (xa_is_value(page
)) {
1017 if (shmem_free_swap(mapping
, index
, page
)) {
1018 /* Swap was replaced by page: retry */
1028 if (!unfalloc
|| !PageUptodate(page
)) {
1029 if (page_mapping(page
) != mapping
) {
1030 /* Page was replaced by swap: retry */
1035 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
1036 if (shmem_punch_compound(page
, start
, end
))
1037 truncate_inode_page(mapping
, page
);
1038 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1039 /* Wipe the page and don't get stuck */
1040 clear_highpage(page
);
1041 flush_dcache_page(page
);
1042 set_page_dirty(page
);
1044 round_up(start
, HPAGE_PMD_NR
))
1050 pagevec_remove_exceptionals(&pvec
);
1051 pagevec_release(&pvec
);
1055 spin_lock_irq(&info
->lock
);
1056 info
->swapped
-= nr_swaps_freed
;
1057 shmem_recalc_inode(inode
);
1058 spin_unlock_irq(&info
->lock
);
1061 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1063 shmem_undo_range(inode
, lstart
, lend
, false);
1064 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1066 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1068 static int shmem_getattr(struct user_namespace
*mnt_userns
,
1069 const struct path
*path
, struct kstat
*stat
,
1070 u32 request_mask
, unsigned int query_flags
)
1072 struct inode
*inode
= path
->dentry
->d_inode
;
1073 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1075 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1076 spin_lock_irq(&info
->lock
);
1077 shmem_recalc_inode(inode
);
1078 spin_unlock_irq(&info
->lock
);
1080 generic_fillattr(&init_user_ns
, inode
, stat
);
1082 if (shmem_is_huge(NULL
, inode
, 0))
1083 stat
->blksize
= HPAGE_PMD_SIZE
;
1088 static int shmem_setattr(struct user_namespace
*mnt_userns
,
1089 struct dentry
*dentry
, struct iattr
*attr
)
1091 struct inode
*inode
= d_inode(dentry
);
1092 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1095 error
= setattr_prepare(&init_user_ns
, dentry
, attr
);
1099 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1100 loff_t oldsize
= inode
->i_size
;
1101 loff_t newsize
= attr
->ia_size
;
1103 /* protected by i_rwsem */
1104 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1105 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1108 if (newsize
!= oldsize
) {
1109 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1113 i_size_write(inode
, newsize
);
1114 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1116 if (newsize
<= oldsize
) {
1117 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1118 if (oldsize
> holebegin
)
1119 unmap_mapping_range(inode
->i_mapping
,
1122 shmem_truncate_range(inode
,
1123 newsize
, (loff_t
)-1);
1124 /* unmap again to remove racily COWed private pages */
1125 if (oldsize
> holebegin
)
1126 unmap_mapping_range(inode
->i_mapping
,
1131 setattr_copy(&init_user_ns
, inode
, attr
);
1132 if (attr
->ia_valid
& ATTR_MODE
)
1133 error
= posix_acl_chmod(&init_user_ns
, inode
, inode
->i_mode
);
1137 static void shmem_evict_inode(struct inode
*inode
)
1139 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1140 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1142 if (shmem_mapping(inode
->i_mapping
)) {
1143 shmem_unacct_size(info
->flags
, inode
->i_size
);
1145 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1146 if (!list_empty(&info
->shrinklist
)) {
1147 spin_lock(&sbinfo
->shrinklist_lock
);
1148 if (!list_empty(&info
->shrinklist
)) {
1149 list_del_init(&info
->shrinklist
);
1150 sbinfo
->shrinklist_len
--;
1152 spin_unlock(&sbinfo
->shrinklist_lock
);
1154 while (!list_empty(&info
->swaplist
)) {
1155 /* Wait while shmem_unuse() is scanning this inode... */
1156 wait_var_event(&info
->stop_eviction
,
1157 !atomic_read(&info
->stop_eviction
));
1158 mutex_lock(&shmem_swaplist_mutex
);
1159 /* ...but beware of the race if we peeked too early */
1160 if (!atomic_read(&info
->stop_eviction
))
1161 list_del_init(&info
->swaplist
);
1162 mutex_unlock(&shmem_swaplist_mutex
);
1166 simple_xattrs_free(&info
->xattrs
);
1167 WARN_ON(inode
->i_blocks
);
1168 shmem_free_inode(inode
->i_sb
);
1172 static int shmem_find_swap_entries(struct address_space
*mapping
,
1173 pgoff_t start
, unsigned int nr_entries
,
1174 struct page
**entries
, pgoff_t
*indices
,
1175 unsigned int type
, bool frontswap
)
1177 XA_STATE(xas
, &mapping
->i_pages
, start
);
1180 unsigned int ret
= 0;
1186 xas_for_each(&xas
, page
, ULONG_MAX
) {
1187 if (xas_retry(&xas
, page
))
1190 if (!xa_is_value(page
))
1193 entry
= radix_to_swp_entry(page
);
1194 if (swp_type(entry
) != type
)
1197 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1200 indices
[ret
] = xas
.xa_index
;
1201 entries
[ret
] = page
;
1203 if (need_resched()) {
1207 if (++ret
== nr_entries
)
1216 * Move the swapped pages for an inode to page cache. Returns the count
1217 * of pages swapped in, or the error in case of failure.
1219 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1225 struct address_space
*mapping
= inode
->i_mapping
;
1227 for (i
= 0; i
< pvec
.nr
; i
++) {
1228 struct page
*page
= pvec
.pages
[i
];
1230 if (!xa_is_value(page
))
1232 error
= shmem_swapin_page(inode
, indices
[i
],
1234 mapping_gfp_mask(mapping
),
1241 if (error
== -ENOMEM
)
1245 return error
? error
: ret
;
1249 * If swap found in inode, free it and move page from swapcache to filecache.
1251 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1252 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1254 struct address_space
*mapping
= inode
->i_mapping
;
1256 struct pagevec pvec
;
1257 pgoff_t indices
[PAGEVEC_SIZE
];
1258 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1261 pagevec_init(&pvec
);
1263 unsigned int nr_entries
= PAGEVEC_SIZE
;
1265 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1266 nr_entries
= *fs_pages_to_unuse
;
1268 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1269 pvec
.pages
, indices
,
1276 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1280 if (frontswap_partial
) {
1281 *fs_pages_to_unuse
-= ret
;
1282 if (*fs_pages_to_unuse
== 0) {
1283 ret
= FRONTSWAP_PAGES_UNUSED
;
1288 start
= indices
[pvec
.nr
- 1];
1295 * Read all the shared memory data that resides in the swap
1296 * device 'type' back into memory, so the swap device can be
1299 int shmem_unuse(unsigned int type
, bool frontswap
,
1300 unsigned long *fs_pages_to_unuse
)
1302 struct shmem_inode_info
*info
, *next
;
1305 if (list_empty(&shmem_swaplist
))
1308 mutex_lock(&shmem_swaplist_mutex
);
1309 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1310 if (!info
->swapped
) {
1311 list_del_init(&info
->swaplist
);
1315 * Drop the swaplist mutex while searching the inode for swap;
1316 * but before doing so, make sure shmem_evict_inode() will not
1317 * remove placeholder inode from swaplist, nor let it be freed
1318 * (igrab() would protect from unlink, but not from unmount).
1320 atomic_inc(&info
->stop_eviction
);
1321 mutex_unlock(&shmem_swaplist_mutex
);
1323 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1327 mutex_lock(&shmem_swaplist_mutex
);
1328 next
= list_next_entry(info
, swaplist
);
1330 list_del_init(&info
->swaplist
);
1331 if (atomic_dec_and_test(&info
->stop_eviction
))
1332 wake_up_var(&info
->stop_eviction
);
1336 mutex_unlock(&shmem_swaplist_mutex
);
1342 * Move the page from the page cache to the swap cache.
1344 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1346 struct shmem_inode_info
*info
;
1347 struct address_space
*mapping
;
1348 struct inode
*inode
;
1353 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1354 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1355 * and its shmem_writeback() needs them to be split when swapping.
1357 if (PageTransCompound(page
)) {
1358 /* Ensure the subpages are still dirty */
1360 if (split_huge_page(page
) < 0)
1362 ClearPageDirty(page
);
1365 BUG_ON(!PageLocked(page
));
1366 mapping
= page
->mapping
;
1367 index
= page
->index
;
1368 inode
= mapping
->host
;
1369 info
= SHMEM_I(inode
);
1370 if (info
->flags
& VM_LOCKED
)
1372 if (!total_swap_pages
)
1376 * Our capabilities prevent regular writeback or sync from ever calling
1377 * shmem_writepage; but a stacking filesystem might use ->writepage of
1378 * its underlying filesystem, in which case tmpfs should write out to
1379 * swap only in response to memory pressure, and not for the writeback
1382 if (!wbc
->for_reclaim
) {
1383 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1388 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1389 * value into swapfile.c, the only way we can correctly account for a
1390 * fallocated page arriving here is now to initialize it and write it.
1392 * That's okay for a page already fallocated earlier, but if we have
1393 * not yet completed the fallocation, then (a) we want to keep track
1394 * of this page in case we have to undo it, and (b) it may not be a
1395 * good idea to continue anyway, once we're pushing into swap. So
1396 * reactivate the page, and let shmem_fallocate() quit when too many.
1398 if (!PageUptodate(page
)) {
1399 if (inode
->i_private
) {
1400 struct shmem_falloc
*shmem_falloc
;
1401 spin_lock(&inode
->i_lock
);
1402 shmem_falloc
= inode
->i_private
;
1404 !shmem_falloc
->waitq
&&
1405 index
>= shmem_falloc
->start
&&
1406 index
< shmem_falloc
->next
)
1407 shmem_falloc
->nr_unswapped
++;
1409 shmem_falloc
= NULL
;
1410 spin_unlock(&inode
->i_lock
);
1414 clear_highpage(page
);
1415 flush_dcache_page(page
);
1416 SetPageUptodate(page
);
1419 swap
= get_swap_page(page
);
1424 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1425 * if it's not already there. Do it now before the page is
1426 * moved to swap cache, when its pagelock no longer protects
1427 * the inode from eviction. But don't unlock the mutex until
1428 * we've incremented swapped, because shmem_unuse_inode() will
1429 * prune a !swapped inode from the swaplist under this mutex.
1431 mutex_lock(&shmem_swaplist_mutex
);
1432 if (list_empty(&info
->swaplist
))
1433 list_add(&info
->swaplist
, &shmem_swaplist
);
1435 if (add_to_swap_cache(page
, swap
,
1436 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1438 spin_lock_irq(&info
->lock
);
1439 shmem_recalc_inode(inode
);
1441 spin_unlock_irq(&info
->lock
);
1443 swap_shmem_alloc(swap
);
1444 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1446 mutex_unlock(&shmem_swaplist_mutex
);
1447 BUG_ON(page_mapped(page
));
1448 swap_writepage(page
, wbc
);
1452 mutex_unlock(&shmem_swaplist_mutex
);
1453 put_swap_page(page
, swap
);
1455 set_page_dirty(page
);
1456 if (wbc
->for_reclaim
)
1457 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1462 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1463 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1467 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1468 return; /* show nothing */
1470 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1472 seq_printf(seq
, ",mpol=%s", buffer
);
1475 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1477 struct mempolicy
*mpol
= NULL
;
1479 raw_spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1480 mpol
= sbinfo
->mpol
;
1482 raw_spin_unlock(&sbinfo
->stat_lock
);
1486 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1487 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1490 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1494 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1496 #define vm_policy vm_private_data
1499 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1500 struct shmem_inode_info
*info
, pgoff_t index
)
1502 /* Create a pseudo vma that just contains the policy */
1503 vma_init(vma
, NULL
);
1504 /* Bias interleave by inode number to distribute better across nodes */
1505 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1506 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1509 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1511 /* Drop reference taken by mpol_shared_policy_lookup() */
1512 mpol_cond_put(vma
->vm_policy
);
1515 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1516 struct shmem_inode_info
*info
, pgoff_t index
)
1518 struct vm_area_struct pvma
;
1520 struct vm_fault vmf
= {
1524 shmem_pseudo_vma_init(&pvma
, info
, index
);
1525 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1526 shmem_pseudo_vma_destroy(&pvma
);
1532 * Make sure huge_gfp is always more limited than limit_gfp.
1533 * Some of the flags set permissions, while others set limitations.
1535 static gfp_t
limit_gfp_mask(gfp_t huge_gfp
, gfp_t limit_gfp
)
1537 gfp_t allowflags
= __GFP_IO
| __GFP_FS
| __GFP_RECLAIM
;
1538 gfp_t denyflags
= __GFP_NOWARN
| __GFP_NORETRY
;
1539 gfp_t zoneflags
= limit_gfp
& GFP_ZONEMASK
;
1540 gfp_t result
= huge_gfp
& ~(allowflags
| GFP_ZONEMASK
);
1542 /* Allow allocations only from the originally specified zones. */
1543 result
|= zoneflags
;
1546 * Minimize the result gfp by taking the union with the deny flags,
1547 * and the intersection of the allow flags.
1549 result
|= (limit_gfp
& denyflags
);
1550 result
|= (huge_gfp
& limit_gfp
) & allowflags
;
1555 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1556 struct shmem_inode_info
*info
, pgoff_t index
)
1558 struct vm_area_struct pvma
;
1559 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1563 hindex
= round_down(index
, HPAGE_PMD_NR
);
1564 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1568 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1569 page
= alloc_pages_vma(gfp
, HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(),
1571 shmem_pseudo_vma_destroy(&pvma
);
1573 prep_transhuge_page(page
);
1575 count_vm_event(THP_FILE_FALLBACK
);
1579 static struct page
*shmem_alloc_page(gfp_t gfp
,
1580 struct shmem_inode_info
*info
, pgoff_t index
)
1582 struct vm_area_struct pvma
;
1585 shmem_pseudo_vma_init(&pvma
, info
, index
);
1586 page
= alloc_page_vma(gfp
, &pvma
, 0);
1587 shmem_pseudo_vma_destroy(&pvma
);
1592 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1593 struct inode
*inode
,
1594 pgoff_t index
, bool huge
)
1596 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1601 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1603 nr
= huge
? HPAGE_PMD_NR
: 1;
1605 if (!shmem_inode_acct_block(inode
, nr
))
1609 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1611 page
= shmem_alloc_page(gfp
, info
, index
);
1613 __SetPageLocked(page
);
1614 __SetPageSwapBacked(page
);
1619 shmem_inode_unacct_blocks(inode
, nr
);
1621 return ERR_PTR(err
);
1625 * When a page is moved from swapcache to shmem filecache (either by the
1626 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1627 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1628 * ignorance of the mapping it belongs to. If that mapping has special
1629 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1630 * we may need to copy to a suitable page before moving to filecache.
1632 * In a future release, this may well be extended to respect cpuset and
1633 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1634 * but for now it is a simple matter of zone.
1636 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1638 return page_zonenum(page
) > gfp_zone(gfp
);
1641 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1642 struct shmem_inode_info
*info
, pgoff_t index
)
1644 struct page
*oldpage
, *newpage
;
1645 struct address_space
*swap_mapping
;
1651 entry
.val
= page_private(oldpage
);
1652 swap_index
= swp_offset(entry
);
1653 swap_mapping
= page_mapping(oldpage
);
1656 * We have arrived here because our zones are constrained, so don't
1657 * limit chance of success by further cpuset and node constraints.
1659 gfp
&= ~GFP_CONSTRAINT_MASK
;
1660 newpage
= shmem_alloc_page(gfp
, info
, index
);
1665 copy_highpage(newpage
, oldpage
);
1666 flush_dcache_page(newpage
);
1668 __SetPageLocked(newpage
);
1669 __SetPageSwapBacked(newpage
);
1670 SetPageUptodate(newpage
);
1671 set_page_private(newpage
, entry
.val
);
1672 SetPageSwapCache(newpage
);
1675 * Our caller will very soon move newpage out of swapcache, but it's
1676 * a nice clean interface for us to replace oldpage by newpage there.
1678 xa_lock_irq(&swap_mapping
->i_pages
);
1679 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1681 mem_cgroup_migrate(oldpage
, newpage
);
1682 __inc_lruvec_page_state(newpage
, NR_FILE_PAGES
);
1683 __dec_lruvec_page_state(oldpage
, NR_FILE_PAGES
);
1685 xa_unlock_irq(&swap_mapping
->i_pages
);
1687 if (unlikely(error
)) {
1689 * Is this possible? I think not, now that our callers check
1690 * both PageSwapCache and page_private after getting page lock;
1691 * but be defensive. Reverse old to newpage for clear and free.
1695 lru_cache_add(newpage
);
1699 ClearPageSwapCache(oldpage
);
1700 set_page_private(oldpage
, 0);
1702 unlock_page(oldpage
);
1709 * Swap in the page pointed to by *pagep.
1710 * Caller has to make sure that *pagep contains a valid swapped page.
1711 * Returns 0 and the page in pagep if success. On failure, returns the
1712 * error code and NULL in *pagep.
1714 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1715 struct page
**pagep
, enum sgp_type sgp
,
1716 gfp_t gfp
, struct vm_area_struct
*vma
,
1717 vm_fault_t
*fault_type
)
1719 struct address_space
*mapping
= inode
->i_mapping
;
1720 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1721 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: NULL
;
1726 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1727 swap
= radix_to_swp_entry(*pagep
);
1730 /* Look it up and read it in.. */
1731 page
= lookup_swap_cache(swap
, NULL
, 0);
1733 /* Or update major stats only when swapin succeeds?? */
1735 *fault_type
|= VM_FAULT_MAJOR
;
1736 count_vm_event(PGMAJFAULT
);
1737 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1739 /* Here we actually start the io */
1740 page
= shmem_swapin(swap
, gfp
, info
, index
);
1747 /* We have to do this with page locked to prevent races */
1749 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1750 !shmem_confirm_swap(mapping
, index
, swap
)) {
1754 if (!PageUptodate(page
)) {
1758 wait_on_page_writeback(page
);
1761 * Some architectures may have to restore extra metadata to the
1762 * physical page after reading from swap.
1764 arch_swap_restore(swap
, page
);
1766 if (shmem_should_replace_page(page
, gfp
)) {
1767 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1772 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1773 swp_to_radix_entry(swap
), gfp
,
1778 spin_lock_irq(&info
->lock
);
1780 shmem_recalc_inode(inode
);
1781 spin_unlock_irq(&info
->lock
);
1783 if (sgp
== SGP_WRITE
)
1784 mark_page_accessed(page
);
1786 delete_from_swap_cache(page
);
1787 set_page_dirty(page
);
1793 if (!shmem_confirm_swap(mapping
, index
, swap
))
1805 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1807 * If we allocate a new one we do not mark it dirty. That's up to the
1808 * vm. If we swap it in we mark it dirty since we also free the swap
1809 * entry since a page cannot live in both the swap and page cache.
1811 * vma, vmf, and fault_type are only supplied by shmem_fault:
1812 * otherwise they are NULL.
1814 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1815 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1816 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1817 vm_fault_t
*fault_type
)
1819 struct address_space
*mapping
= inode
->i_mapping
;
1820 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1821 struct shmem_sb_info
*sbinfo
;
1822 struct mm_struct
*charge_mm
;
1824 pgoff_t hindex
= index
;
1830 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1833 if (sgp
<= SGP_CACHE
&&
1834 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1838 sbinfo
= SHMEM_SB(inode
->i_sb
);
1839 charge_mm
= vma
? vma
->vm_mm
: NULL
;
1841 page
= pagecache_get_page(mapping
, index
,
1842 FGP_ENTRY
| FGP_HEAD
| FGP_LOCK
, 0);
1844 if (page
&& vma
&& userfaultfd_minor(vma
)) {
1845 if (!xa_is_value(page
)) {
1849 *fault_type
= handle_userfault(vmf
, VM_UFFD_MINOR
);
1853 if (xa_is_value(page
)) {
1854 error
= shmem_swapin_page(inode
, index
, &page
,
1855 sgp
, gfp
, vma
, fault_type
);
1856 if (error
== -EEXIST
)
1864 hindex
= page
->index
;
1865 if (sgp
== SGP_WRITE
)
1866 mark_page_accessed(page
);
1867 if (PageUptodate(page
))
1869 /* fallocated page */
1870 if (sgp
!= SGP_READ
)
1877 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1878 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1881 if (sgp
== SGP_READ
)
1883 if (sgp
== SGP_NOALLOC
)
1887 * Fast cache lookup and swap lookup did not find it: allocate.
1890 if (vma
&& userfaultfd_missing(vma
)) {
1891 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1895 /* Never use a huge page for shmem_symlink() */
1896 if (S_ISLNK(inode
->i_mode
))
1898 if (!shmem_is_huge(vma
, inode
, index
))
1901 huge_gfp
= vma_thp_gfp_mask(vma
);
1902 huge_gfp
= limit_gfp_mask(huge_gfp
, gfp
);
1903 page
= shmem_alloc_and_acct_page(huge_gfp
, inode
, index
, true);
1906 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1912 error
= PTR_ERR(page
);
1914 if (error
!= -ENOSPC
)
1917 * Try to reclaim some space by splitting a huge page
1918 * beyond i_size on the filesystem.
1923 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1924 if (ret
== SHRINK_STOP
)
1932 if (PageTransHuge(page
))
1933 hindex
= round_down(index
, HPAGE_PMD_NR
);
1937 if (sgp
== SGP_WRITE
)
1938 __SetPageReferenced(page
);
1940 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1941 NULL
, gfp
& GFP_RECLAIM_MASK
,
1945 lru_cache_add(page
);
1947 spin_lock_irq(&info
->lock
);
1948 info
->alloced
+= compound_nr(page
);
1949 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1950 shmem_recalc_inode(inode
);
1951 spin_unlock_irq(&info
->lock
);
1954 if (PageTransHuge(page
) &&
1955 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1956 hindex
+ HPAGE_PMD_NR
- 1) {
1958 * Part of the huge page is beyond i_size: subject
1959 * to shrink under memory pressure.
1961 spin_lock(&sbinfo
->shrinklist_lock
);
1963 * _careful to defend against unlocked access to
1964 * ->shrink_list in shmem_unused_huge_shrink()
1966 if (list_empty_careful(&info
->shrinklist
)) {
1967 list_add_tail(&info
->shrinklist
,
1968 &sbinfo
->shrinklist
);
1969 sbinfo
->shrinklist_len
++;
1971 spin_unlock(&sbinfo
->shrinklist_lock
);
1975 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1977 if (sgp
== SGP_FALLOC
)
1981 * Let SGP_WRITE caller clear ends if write does not fill page;
1982 * but SGP_FALLOC on a page fallocated earlier must initialize
1983 * it now, lest undo on failure cancel our earlier guarantee.
1985 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1988 for (i
= 0; i
< compound_nr(page
); i
++) {
1989 clear_highpage(page
+ i
);
1990 flush_dcache_page(page
+ i
);
1992 SetPageUptodate(page
);
1995 /* Perhaps the file has been truncated since we checked */
1996 if (sgp
<= SGP_CACHE
&&
1997 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1999 ClearPageDirty(page
);
2000 delete_from_page_cache(page
);
2001 spin_lock_irq(&info
->lock
);
2002 shmem_recalc_inode(inode
);
2003 spin_unlock_irq(&info
->lock
);
2009 *pagep
= page
+ index
- hindex
;
2016 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
2018 if (PageTransHuge(page
)) {
2028 if (error
== -ENOSPC
&& !once
++) {
2029 spin_lock_irq(&info
->lock
);
2030 shmem_recalc_inode(inode
);
2031 spin_unlock_irq(&info
->lock
);
2034 if (error
== -EEXIST
)
2040 * This is like autoremove_wake_function, but it removes the wait queue
2041 * entry unconditionally - even if something else had already woken the
2044 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2046 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2047 list_del_init(&wait
->entry
);
2051 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2053 struct vm_area_struct
*vma
= vmf
->vma
;
2054 struct inode
*inode
= file_inode(vma
->vm_file
);
2055 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2057 vm_fault_t ret
= VM_FAULT_LOCKED
;
2060 * Trinity finds that probing a hole which tmpfs is punching can
2061 * prevent the hole-punch from ever completing: which in turn
2062 * locks writers out with its hold on i_rwsem. So refrain from
2063 * faulting pages into the hole while it's being punched. Although
2064 * shmem_undo_range() does remove the additions, it may be unable to
2065 * keep up, as each new page needs its own unmap_mapping_range() call,
2066 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2068 * It does not matter if we sometimes reach this check just before the
2069 * hole-punch begins, so that one fault then races with the punch:
2070 * we just need to make racing faults a rare case.
2072 * The implementation below would be much simpler if we just used a
2073 * standard mutex or completion: but we cannot take i_rwsem in fault,
2074 * and bloating every shmem inode for this unlikely case would be sad.
2076 if (unlikely(inode
->i_private
)) {
2077 struct shmem_falloc
*shmem_falloc
;
2079 spin_lock(&inode
->i_lock
);
2080 shmem_falloc
= inode
->i_private
;
2082 shmem_falloc
->waitq
&&
2083 vmf
->pgoff
>= shmem_falloc
->start
&&
2084 vmf
->pgoff
< shmem_falloc
->next
) {
2086 wait_queue_head_t
*shmem_falloc_waitq
;
2087 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2089 ret
= VM_FAULT_NOPAGE
;
2090 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2092 ret
= VM_FAULT_RETRY
;
2094 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2095 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2096 TASK_UNINTERRUPTIBLE
);
2097 spin_unlock(&inode
->i_lock
);
2101 * shmem_falloc_waitq points into the shmem_fallocate()
2102 * stack of the hole-punching task: shmem_falloc_waitq
2103 * is usually invalid by the time we reach here, but
2104 * finish_wait() does not dereference it in that case;
2105 * though i_lock needed lest racing with wake_up_all().
2107 spin_lock(&inode
->i_lock
);
2108 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2109 spin_unlock(&inode
->i_lock
);
2115 spin_unlock(&inode
->i_lock
);
2118 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
,
2119 gfp
, vma
, vmf
, &ret
);
2121 return vmf_error(err
);
2125 unsigned long shmem_get_unmapped_area(struct file
*file
,
2126 unsigned long uaddr
, unsigned long len
,
2127 unsigned long pgoff
, unsigned long flags
)
2129 unsigned long (*get_area
)(struct file
*,
2130 unsigned long, unsigned long, unsigned long, unsigned long);
2132 unsigned long offset
;
2133 unsigned long inflated_len
;
2134 unsigned long inflated_addr
;
2135 unsigned long inflated_offset
;
2137 if (len
> TASK_SIZE
)
2140 get_area
= current
->mm
->get_unmapped_area
;
2141 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2143 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2145 if (IS_ERR_VALUE(addr
))
2147 if (addr
& ~PAGE_MASK
)
2149 if (addr
> TASK_SIZE
- len
)
2152 if (shmem_huge
== SHMEM_HUGE_DENY
)
2154 if (len
< HPAGE_PMD_SIZE
)
2156 if (flags
& MAP_FIXED
)
2159 * Our priority is to support MAP_SHARED mapped hugely;
2160 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2161 * But if caller specified an address hint and we allocated area there
2162 * successfully, respect that as before.
2167 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2168 struct super_block
*sb
;
2171 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2172 sb
= file_inode(file
)->i_sb
;
2175 * Called directly from mm/mmap.c, or drivers/char/mem.c
2176 * for "/dev/zero", to create a shared anonymous object.
2178 if (IS_ERR(shm_mnt
))
2180 sb
= shm_mnt
->mnt_sb
;
2182 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2186 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2187 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2189 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2192 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2193 if (inflated_len
> TASK_SIZE
)
2195 if (inflated_len
< len
)
2198 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2199 if (IS_ERR_VALUE(inflated_addr
))
2201 if (inflated_addr
& ~PAGE_MASK
)
2204 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2205 inflated_addr
+= offset
- inflated_offset
;
2206 if (inflated_offset
> offset
)
2207 inflated_addr
+= HPAGE_PMD_SIZE
;
2209 if (inflated_addr
> TASK_SIZE
- len
)
2211 return inflated_addr
;
2215 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2217 struct inode
*inode
= file_inode(vma
->vm_file
);
2218 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2221 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2224 struct inode
*inode
= file_inode(vma
->vm_file
);
2227 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2228 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2232 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
2234 struct inode
*inode
= file_inode(file
);
2235 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2236 int retval
= -ENOMEM
;
2239 * What serializes the accesses to info->flags?
2240 * ipc_lock_object() when called from shmctl_do_lock(),
2241 * no serialization needed when called from shm_destroy().
2243 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2244 if (!user_shm_lock(inode
->i_size
, ucounts
))
2246 info
->flags
|= VM_LOCKED
;
2247 mapping_set_unevictable(file
->f_mapping
);
2249 if (!lock
&& (info
->flags
& VM_LOCKED
) && ucounts
) {
2250 user_shm_unlock(inode
->i_size
, ucounts
);
2251 info
->flags
&= ~VM_LOCKED
;
2252 mapping_clear_unevictable(file
->f_mapping
);
2260 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2262 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2265 ret
= seal_check_future_write(info
->seals
, vma
);
2269 /* arm64 - allow memory tagging on RAM-based files */
2270 vma
->vm_flags
|= VM_MTE_ALLOWED
;
2272 file_accessed(file
);
2273 vma
->vm_ops
= &shmem_vm_ops
;
2274 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
2275 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2276 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2277 khugepaged_enter(vma
, vma
->vm_flags
);
2282 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2283 umode_t mode
, dev_t dev
, unsigned long flags
)
2285 struct inode
*inode
;
2286 struct shmem_inode_info
*info
;
2287 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2290 if (shmem_reserve_inode(sb
, &ino
))
2293 inode
= new_inode(sb
);
2296 inode_init_owner(&init_user_ns
, inode
, dir
, mode
);
2297 inode
->i_blocks
= 0;
2298 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2299 inode
->i_generation
= prandom_u32();
2300 info
= SHMEM_I(inode
);
2301 memset(info
, 0, (char *)inode
- (char *)info
);
2302 spin_lock_init(&info
->lock
);
2303 atomic_set(&info
->stop_eviction
, 0);
2304 info
->seals
= F_SEAL_SEAL
;
2305 info
->flags
= flags
& VM_NORESERVE
;
2306 INIT_LIST_HEAD(&info
->shrinklist
);
2307 INIT_LIST_HEAD(&info
->swaplist
);
2308 simple_xattrs_init(&info
->xattrs
);
2309 cache_no_acl(inode
);
2311 switch (mode
& S_IFMT
) {
2313 inode
->i_op
= &shmem_special_inode_operations
;
2314 init_special_inode(inode
, mode
, dev
);
2317 inode
->i_mapping
->a_ops
= &shmem_aops
;
2318 inode
->i_op
= &shmem_inode_operations
;
2319 inode
->i_fop
= &shmem_file_operations
;
2320 mpol_shared_policy_init(&info
->policy
,
2321 shmem_get_sbmpol(sbinfo
));
2325 /* Some things misbehave if size == 0 on a directory */
2326 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2327 inode
->i_op
= &shmem_dir_inode_operations
;
2328 inode
->i_fop
= &simple_dir_operations
;
2332 * Must not load anything in the rbtree,
2333 * mpol_free_shared_policy will not be called.
2335 mpol_shared_policy_init(&info
->policy
, NULL
);
2339 lockdep_annotate_inode_mutex_key(inode
);
2341 shmem_free_inode(sb
);
2345 #ifdef CONFIG_USERFAULTFD
2346 int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2348 struct vm_area_struct
*dst_vma
,
2349 unsigned long dst_addr
,
2350 unsigned long src_addr
,
2352 struct page
**pagep
)
2354 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2355 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2356 struct address_space
*mapping
= inode
->i_mapping
;
2357 gfp_t gfp
= mapping_gfp_mask(mapping
);
2358 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2364 if (!shmem_inode_acct_block(inode
, 1)) {
2366 * We may have got a page, returned -ENOENT triggering a retry,
2367 * and now we find ourselves with -ENOMEM. Release the page, to
2368 * avoid a BUG_ON in our caller.
2370 if (unlikely(*pagep
)) {
2379 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2381 goto out_unacct_blocks
;
2383 if (!zeropage
) { /* COPY */
2384 page_kaddr
= kmap_atomic(page
);
2385 ret
= copy_from_user(page_kaddr
,
2386 (const void __user
*)src_addr
,
2388 kunmap_atomic(page_kaddr
);
2390 /* fallback to copy_from_user outside mmap_lock */
2391 if (unlikely(ret
)) {
2394 /* don't free the page */
2395 goto out_unacct_blocks
;
2397 } else { /* ZEROPAGE */
2398 clear_highpage(page
);
2405 VM_BUG_ON(PageLocked(page
));
2406 VM_BUG_ON(PageSwapBacked(page
));
2407 __SetPageLocked(page
);
2408 __SetPageSwapBacked(page
);
2409 __SetPageUptodate(page
);
2412 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2413 if (unlikely(pgoff
>= max_off
))
2416 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2417 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2421 ret
= mfill_atomic_install_pte(dst_mm
, dst_pmd
, dst_vma
, dst_addr
,
2424 goto out_delete_from_cache
;
2426 spin_lock_irq(&info
->lock
);
2428 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2429 shmem_recalc_inode(inode
);
2430 spin_unlock_irq(&info
->lock
);
2435 out_delete_from_cache
:
2436 delete_from_page_cache(page
);
2441 shmem_inode_unacct_blocks(inode
, 1);
2444 #endif /* CONFIG_USERFAULTFD */
2447 static const struct inode_operations shmem_symlink_inode_operations
;
2448 static const struct inode_operations shmem_short_symlink_operations
;
2450 #ifdef CONFIG_TMPFS_XATTR
2451 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2453 #define shmem_initxattrs NULL
2457 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2458 loff_t pos
, unsigned len
, unsigned flags
,
2459 struct page
**pagep
, void **fsdata
)
2461 struct inode
*inode
= mapping
->host
;
2462 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2463 pgoff_t index
= pos
>> PAGE_SHIFT
;
2465 /* i_rwsem is held by caller */
2466 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2467 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2468 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2470 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2474 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2478 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2479 loff_t pos
, unsigned len
, unsigned copied
,
2480 struct page
*page
, void *fsdata
)
2482 struct inode
*inode
= mapping
->host
;
2484 if (pos
+ copied
> inode
->i_size
)
2485 i_size_write(inode
, pos
+ copied
);
2487 if (!PageUptodate(page
)) {
2488 struct page
*head
= compound_head(page
);
2489 if (PageTransCompound(page
)) {
2492 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2493 if (head
+ i
== page
)
2495 clear_highpage(head
+ i
);
2496 flush_dcache_page(head
+ i
);
2499 if (copied
< PAGE_SIZE
) {
2500 unsigned from
= pos
& (PAGE_SIZE
- 1);
2501 zero_user_segments(page
, 0, from
,
2502 from
+ copied
, PAGE_SIZE
);
2504 SetPageUptodate(head
);
2506 set_page_dirty(page
);
2513 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2515 struct file
*file
= iocb
->ki_filp
;
2516 struct inode
*inode
= file_inode(file
);
2517 struct address_space
*mapping
= inode
->i_mapping
;
2519 unsigned long offset
;
2520 enum sgp_type sgp
= SGP_READ
;
2523 loff_t
*ppos
= &iocb
->ki_pos
;
2526 * Might this read be for a stacking filesystem? Then when reading
2527 * holes of a sparse file, we actually need to allocate those pages,
2528 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2530 if (!iter_is_iovec(to
))
2533 index
= *ppos
>> PAGE_SHIFT
;
2534 offset
= *ppos
& ~PAGE_MASK
;
2537 struct page
*page
= NULL
;
2539 unsigned long nr
, ret
;
2540 loff_t i_size
= i_size_read(inode
);
2542 end_index
= i_size
>> PAGE_SHIFT
;
2543 if (index
> end_index
)
2545 if (index
== end_index
) {
2546 nr
= i_size
& ~PAGE_MASK
;
2551 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2553 if (error
== -EINVAL
)
2558 if (sgp
== SGP_CACHE
)
2559 set_page_dirty(page
);
2564 * We must evaluate after, since reads (unlike writes)
2565 * are called without i_rwsem protection against truncate
2568 i_size
= i_size_read(inode
);
2569 end_index
= i_size
>> PAGE_SHIFT
;
2570 if (index
== end_index
) {
2571 nr
= i_size
& ~PAGE_MASK
;
2582 * If users can be writing to this page using arbitrary
2583 * virtual addresses, take care about potential aliasing
2584 * before reading the page on the kernel side.
2586 if (mapping_writably_mapped(mapping
))
2587 flush_dcache_page(page
);
2589 * Mark the page accessed if we read the beginning.
2592 mark_page_accessed(page
);
2594 page
= ZERO_PAGE(0);
2599 * Ok, we have the page, and it's up-to-date, so
2600 * now we can copy it to user space...
2602 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2605 index
+= offset
>> PAGE_SHIFT
;
2606 offset
&= ~PAGE_MASK
;
2609 if (!iov_iter_count(to
))
2618 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2619 file_accessed(file
);
2620 return retval
? retval
: error
;
2623 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2625 struct address_space
*mapping
= file
->f_mapping
;
2626 struct inode
*inode
= mapping
->host
;
2628 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2629 return generic_file_llseek_size(file
, offset
, whence
,
2630 MAX_LFS_FILESIZE
, i_size_read(inode
));
2635 /* We're holding i_rwsem so we can access i_size directly */
2636 offset
= mapping_seek_hole_data(mapping
, offset
, inode
->i_size
, whence
);
2638 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2639 inode_unlock(inode
);
2643 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2646 struct inode
*inode
= file_inode(file
);
2647 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2648 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2649 struct shmem_falloc shmem_falloc
;
2650 pgoff_t start
, index
, end
, undo_fallocend
;
2653 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2658 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2659 struct address_space
*mapping
= file
->f_mapping
;
2660 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2661 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2662 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2664 /* protected by i_rwsem */
2665 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2670 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2671 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2672 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2673 spin_lock(&inode
->i_lock
);
2674 inode
->i_private
= &shmem_falloc
;
2675 spin_unlock(&inode
->i_lock
);
2677 if ((u64
)unmap_end
> (u64
)unmap_start
)
2678 unmap_mapping_range(mapping
, unmap_start
,
2679 1 + unmap_end
- unmap_start
, 0);
2680 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2681 /* No need to unmap again: hole-punching leaves COWed pages */
2683 spin_lock(&inode
->i_lock
);
2684 inode
->i_private
= NULL
;
2685 wake_up_all(&shmem_falloc_waitq
);
2686 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2687 spin_unlock(&inode
->i_lock
);
2692 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2693 error
= inode_newsize_ok(inode
, offset
+ len
);
2697 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2702 start
= offset
>> PAGE_SHIFT
;
2703 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2704 /* Try to avoid a swapstorm if len is impossible to satisfy */
2705 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2710 shmem_falloc
.waitq
= NULL
;
2711 shmem_falloc
.start
= start
;
2712 shmem_falloc
.next
= start
;
2713 shmem_falloc
.nr_falloced
= 0;
2714 shmem_falloc
.nr_unswapped
= 0;
2715 spin_lock(&inode
->i_lock
);
2716 inode
->i_private
= &shmem_falloc
;
2717 spin_unlock(&inode
->i_lock
);
2720 * info->fallocend is only relevant when huge pages might be
2721 * involved: to prevent split_huge_page() freeing fallocated
2722 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2724 undo_fallocend
= info
->fallocend
;
2725 if (info
->fallocend
< end
)
2726 info
->fallocend
= end
;
2728 for (index
= start
; index
< end
; ) {
2732 * Good, the fallocate(2) manpage permits EINTR: we may have
2733 * been interrupted because we are using up too much memory.
2735 if (signal_pending(current
))
2737 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2740 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2742 info
->fallocend
= undo_fallocend
;
2743 /* Remove the !PageUptodate pages we added */
2744 if (index
> start
) {
2745 shmem_undo_range(inode
,
2746 (loff_t
)start
<< PAGE_SHIFT
,
2747 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2754 * Here is a more important optimization than it appears:
2755 * a second SGP_FALLOC on the same huge page will clear it,
2756 * making it PageUptodate and un-undoable if we fail later.
2758 if (PageTransCompound(page
)) {
2759 index
= round_up(index
, HPAGE_PMD_NR
);
2760 /* Beware 32-bit wraparound */
2766 * Inform shmem_writepage() how far we have reached.
2767 * No need for lock or barrier: we have the page lock.
2769 if (!PageUptodate(page
))
2770 shmem_falloc
.nr_falloced
+= index
- shmem_falloc
.next
;
2771 shmem_falloc
.next
= index
;
2774 * If !PageUptodate, leave it that way so that freeable pages
2775 * can be recognized if we need to rollback on error later.
2776 * But set_page_dirty so that memory pressure will swap rather
2777 * than free the pages we are allocating (and SGP_CACHE pages
2778 * might still be clean: we now need to mark those dirty too).
2780 set_page_dirty(page
);
2786 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2787 i_size_write(inode
, offset
+ len
);
2788 inode
->i_ctime
= current_time(inode
);
2790 spin_lock(&inode
->i_lock
);
2791 inode
->i_private
= NULL
;
2792 spin_unlock(&inode
->i_lock
);
2794 inode_unlock(inode
);
2798 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2800 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2802 buf
->f_type
= TMPFS_MAGIC
;
2803 buf
->f_bsize
= PAGE_SIZE
;
2804 buf
->f_namelen
= NAME_MAX
;
2805 if (sbinfo
->max_blocks
) {
2806 buf
->f_blocks
= sbinfo
->max_blocks
;
2808 buf
->f_bfree
= sbinfo
->max_blocks
-
2809 percpu_counter_sum(&sbinfo
->used_blocks
);
2811 if (sbinfo
->max_inodes
) {
2812 buf
->f_files
= sbinfo
->max_inodes
;
2813 buf
->f_ffree
= sbinfo
->free_inodes
;
2815 /* else leave those fields 0 like simple_statfs */
2817 buf
->f_fsid
= uuid_to_fsid(dentry
->d_sb
->s_uuid
.b
);
2823 * File creation. Allocate an inode, and we're done..
2826 shmem_mknod(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2827 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2829 struct inode
*inode
;
2830 int error
= -ENOSPC
;
2832 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2834 error
= simple_acl_create(dir
, inode
);
2837 error
= security_inode_init_security(inode
, dir
,
2839 shmem_initxattrs
, NULL
);
2840 if (error
&& error
!= -EOPNOTSUPP
)
2844 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2845 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2846 d_instantiate(dentry
, inode
);
2847 dget(dentry
); /* Extra count - pin the dentry in core */
2856 shmem_tmpfile(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2857 struct dentry
*dentry
, umode_t mode
)
2859 struct inode
*inode
;
2860 int error
= -ENOSPC
;
2862 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2864 error
= security_inode_init_security(inode
, dir
,
2866 shmem_initxattrs
, NULL
);
2867 if (error
&& error
!= -EOPNOTSUPP
)
2869 error
= simple_acl_create(dir
, inode
);
2872 d_tmpfile(dentry
, inode
);
2880 static int shmem_mkdir(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2881 struct dentry
*dentry
, umode_t mode
)
2885 if ((error
= shmem_mknod(&init_user_ns
, dir
, dentry
,
2886 mode
| S_IFDIR
, 0)))
2892 static int shmem_create(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2893 struct dentry
*dentry
, umode_t mode
, bool excl
)
2895 return shmem_mknod(&init_user_ns
, dir
, dentry
, mode
| S_IFREG
, 0);
2901 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2903 struct inode
*inode
= d_inode(old_dentry
);
2907 * No ordinary (disk based) filesystem counts links as inodes;
2908 * but each new link needs a new dentry, pinning lowmem, and
2909 * tmpfs dentries cannot be pruned until they are unlinked.
2910 * But if an O_TMPFILE file is linked into the tmpfs, the
2911 * first link must skip that, to get the accounting right.
2913 if (inode
->i_nlink
) {
2914 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
2919 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2920 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2922 ihold(inode
); /* New dentry reference */
2923 dget(dentry
); /* Extra pinning count for the created dentry */
2924 d_instantiate(dentry
, inode
);
2929 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2931 struct inode
*inode
= d_inode(dentry
);
2933 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2934 shmem_free_inode(inode
->i_sb
);
2936 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2937 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2939 dput(dentry
); /* Undo the count from "create" - this does all the work */
2943 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2945 if (!simple_empty(dentry
))
2948 drop_nlink(d_inode(dentry
));
2950 return shmem_unlink(dir
, dentry
);
2953 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2955 bool old_is_dir
= d_is_dir(old_dentry
);
2956 bool new_is_dir
= d_is_dir(new_dentry
);
2958 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2960 drop_nlink(old_dir
);
2963 drop_nlink(new_dir
);
2967 old_dir
->i_ctime
= old_dir
->i_mtime
=
2968 new_dir
->i_ctime
= new_dir
->i_mtime
=
2969 d_inode(old_dentry
)->i_ctime
=
2970 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
2975 static int shmem_whiteout(struct user_namespace
*mnt_userns
,
2976 struct inode
*old_dir
, struct dentry
*old_dentry
)
2978 struct dentry
*whiteout
;
2981 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2985 error
= shmem_mknod(&init_user_ns
, old_dir
, whiteout
,
2986 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2992 * Cheat and hash the whiteout while the old dentry is still in
2993 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2995 * d_lookup() will consistently find one of them at this point,
2996 * not sure which one, but that isn't even important.
3003 * The VFS layer already does all the dentry stuff for rename,
3004 * we just have to decrement the usage count for the target if
3005 * it exists so that the VFS layer correctly free's it when it
3008 static int shmem_rename2(struct user_namespace
*mnt_userns
,
3009 struct inode
*old_dir
, struct dentry
*old_dentry
,
3010 struct inode
*new_dir
, struct dentry
*new_dentry
,
3013 struct inode
*inode
= d_inode(old_dentry
);
3014 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3016 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3019 if (flags
& RENAME_EXCHANGE
)
3020 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3022 if (!simple_empty(new_dentry
))
3025 if (flags
& RENAME_WHITEOUT
) {
3028 error
= shmem_whiteout(&init_user_ns
, old_dir
, old_dentry
);
3033 if (d_really_is_positive(new_dentry
)) {
3034 (void) shmem_unlink(new_dir
, new_dentry
);
3035 if (they_are_dirs
) {
3036 drop_nlink(d_inode(new_dentry
));
3037 drop_nlink(old_dir
);
3039 } else if (they_are_dirs
) {
3040 drop_nlink(old_dir
);
3044 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3045 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3046 old_dir
->i_ctime
= old_dir
->i_mtime
=
3047 new_dir
->i_ctime
= new_dir
->i_mtime
=
3048 inode
->i_ctime
= current_time(old_dir
);
3052 static int shmem_symlink(struct user_namespace
*mnt_userns
, struct inode
*dir
,
3053 struct dentry
*dentry
, const char *symname
)
3057 struct inode
*inode
;
3060 len
= strlen(symname
) + 1;
3061 if (len
> PAGE_SIZE
)
3062 return -ENAMETOOLONG
;
3064 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3069 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3070 shmem_initxattrs
, NULL
);
3071 if (error
&& error
!= -EOPNOTSUPP
) {
3076 inode
->i_size
= len
-1;
3077 if (len
<= SHORT_SYMLINK_LEN
) {
3078 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3079 if (!inode
->i_link
) {
3083 inode
->i_op
= &shmem_short_symlink_operations
;
3085 inode_nohighmem(inode
);
3086 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3091 inode
->i_mapping
->a_ops
= &shmem_aops
;
3092 inode
->i_op
= &shmem_symlink_inode_operations
;
3093 memcpy(page_address(page
), symname
, len
);
3094 SetPageUptodate(page
);
3095 set_page_dirty(page
);
3099 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3100 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3101 d_instantiate(dentry
, inode
);
3106 static void shmem_put_link(void *arg
)
3108 mark_page_accessed(arg
);
3112 static const char *shmem_get_link(struct dentry
*dentry
,
3113 struct inode
*inode
,
3114 struct delayed_call
*done
)
3116 struct page
*page
= NULL
;
3119 page
= find_get_page(inode
->i_mapping
, 0);
3121 return ERR_PTR(-ECHILD
);
3122 if (!PageUptodate(page
)) {
3124 return ERR_PTR(-ECHILD
);
3127 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3129 return ERR_PTR(error
);
3132 set_delayed_call(done
, shmem_put_link
, page
);
3133 return page_address(page
);
3136 #ifdef CONFIG_TMPFS_XATTR
3138 * Superblocks without xattr inode operations may get some security.* xattr
3139 * support from the LSM "for free". As soon as we have any other xattrs
3140 * like ACLs, we also need to implement the security.* handlers at
3141 * filesystem level, though.
3145 * Callback for security_inode_init_security() for acquiring xattrs.
3147 static int shmem_initxattrs(struct inode
*inode
,
3148 const struct xattr
*xattr_array
,
3151 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3152 const struct xattr
*xattr
;
3153 struct simple_xattr
*new_xattr
;
3156 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3157 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3161 len
= strlen(xattr
->name
) + 1;
3162 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3164 if (!new_xattr
->name
) {
3169 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3170 XATTR_SECURITY_PREFIX_LEN
);
3171 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3174 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3180 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3181 struct dentry
*unused
, struct inode
*inode
,
3182 const char *name
, void *buffer
, size_t size
)
3184 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3186 name
= xattr_full_name(handler
, name
);
3187 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3190 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3191 struct user_namespace
*mnt_userns
,
3192 struct dentry
*unused
, struct inode
*inode
,
3193 const char *name
, const void *value
,
3194 size_t size
, int flags
)
3196 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3198 name
= xattr_full_name(handler
, name
);
3199 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3202 static const struct xattr_handler shmem_security_xattr_handler
= {
3203 .prefix
= XATTR_SECURITY_PREFIX
,
3204 .get
= shmem_xattr_handler_get
,
3205 .set
= shmem_xattr_handler_set
,
3208 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3209 .prefix
= XATTR_TRUSTED_PREFIX
,
3210 .get
= shmem_xattr_handler_get
,
3211 .set
= shmem_xattr_handler_set
,
3214 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3215 #ifdef CONFIG_TMPFS_POSIX_ACL
3216 &posix_acl_access_xattr_handler
,
3217 &posix_acl_default_xattr_handler
,
3219 &shmem_security_xattr_handler
,
3220 &shmem_trusted_xattr_handler
,
3224 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3226 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3227 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3229 #endif /* CONFIG_TMPFS_XATTR */
3231 static const struct inode_operations shmem_short_symlink_operations
= {
3232 .get_link
= simple_get_link
,
3233 #ifdef CONFIG_TMPFS_XATTR
3234 .listxattr
= shmem_listxattr
,
3238 static const struct inode_operations shmem_symlink_inode_operations
= {
3239 .get_link
= shmem_get_link
,
3240 #ifdef CONFIG_TMPFS_XATTR
3241 .listxattr
= shmem_listxattr
,
3245 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3247 return ERR_PTR(-ESTALE
);
3250 static int shmem_match(struct inode
*ino
, void *vfh
)
3254 inum
= (inum
<< 32) | fh
[1];
3255 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3258 /* Find any alias of inode, but prefer a hashed alias */
3259 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3261 struct dentry
*alias
= d_find_alias(inode
);
3263 return alias
?: d_find_any_alias(inode
);
3267 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3268 struct fid
*fid
, int fh_len
, int fh_type
)
3270 struct inode
*inode
;
3271 struct dentry
*dentry
= NULL
;
3278 inum
= (inum
<< 32) | fid
->raw
[1];
3280 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3281 shmem_match
, fid
->raw
);
3283 dentry
= shmem_find_alias(inode
);
3290 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3291 struct inode
*parent
)
3295 return FILEID_INVALID
;
3298 if (inode_unhashed(inode
)) {
3299 /* Unfortunately insert_inode_hash is not idempotent,
3300 * so as we hash inodes here rather than at creation
3301 * time, we need a lock to ensure we only try
3304 static DEFINE_SPINLOCK(lock
);
3306 if (inode_unhashed(inode
))
3307 __insert_inode_hash(inode
,
3308 inode
->i_ino
+ inode
->i_generation
);
3312 fh
[0] = inode
->i_generation
;
3313 fh
[1] = inode
->i_ino
;
3314 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3320 static const struct export_operations shmem_export_ops
= {
3321 .get_parent
= shmem_get_parent
,
3322 .encode_fh
= shmem_encode_fh
,
3323 .fh_to_dentry
= shmem_fh_to_dentry
,
3339 static const struct constant_table shmem_param_enums_huge
[] = {
3340 {"never", SHMEM_HUGE_NEVER
},
3341 {"always", SHMEM_HUGE_ALWAYS
},
3342 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3343 {"advise", SHMEM_HUGE_ADVISE
},
3347 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3348 fsparam_u32 ("gid", Opt_gid
),
3349 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3350 fsparam_u32oct("mode", Opt_mode
),
3351 fsparam_string("mpol", Opt_mpol
),
3352 fsparam_string("nr_blocks", Opt_nr_blocks
),
3353 fsparam_string("nr_inodes", Opt_nr_inodes
),
3354 fsparam_string("size", Opt_size
),
3355 fsparam_u32 ("uid", Opt_uid
),
3356 fsparam_flag ("inode32", Opt_inode32
),
3357 fsparam_flag ("inode64", Opt_inode64
),
3361 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3363 struct shmem_options
*ctx
= fc
->fs_private
;
3364 struct fs_parse_result result
;
3365 unsigned long long size
;
3369 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3375 size
= memparse(param
->string
, &rest
);
3377 size
<<= PAGE_SHIFT
;
3378 size
*= totalram_pages();
3384 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3385 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3388 ctx
->blocks
= memparse(param
->string
, &rest
);
3391 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3394 ctx
->inodes
= memparse(param
->string
, &rest
);
3397 ctx
->seen
|= SHMEM_SEEN_INODES
;
3400 ctx
->mode
= result
.uint_32
& 07777;
3403 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3404 if (!uid_valid(ctx
->uid
))
3408 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3409 if (!gid_valid(ctx
->gid
))
3413 ctx
->huge
= result
.uint_32
;
3414 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3415 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3416 has_transparent_hugepage()))
3417 goto unsupported_parameter
;
3418 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3421 if (IS_ENABLED(CONFIG_NUMA
)) {
3422 mpol_put(ctx
->mpol
);
3424 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3428 goto unsupported_parameter
;
3430 ctx
->full_inums
= false;
3431 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3434 if (sizeof(ino_t
) < 8) {
3436 "Cannot use inode64 with <64bit inums in kernel\n");
3438 ctx
->full_inums
= true;
3439 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3444 unsupported_parameter
:
3445 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3447 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3450 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3452 char *options
= data
;
3455 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3460 while (options
!= NULL
) {
3461 char *this_char
= options
;
3464 * NUL-terminate this option: unfortunately,
3465 * mount options form a comma-separated list,
3466 * but mpol's nodelist may also contain commas.
3468 options
= strchr(options
, ',');
3469 if (options
== NULL
)
3472 if (!isdigit(*options
)) {
3478 char *value
= strchr(this_char
, '=');
3484 len
= strlen(value
);
3486 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3495 * Reconfigure a shmem filesystem.
3497 * Note that we disallow change from limited->unlimited blocks/inodes while any
3498 * are in use; but we must separately disallow unlimited->limited, because in
3499 * that case we have no record of how much is already in use.
3501 static int shmem_reconfigure(struct fs_context
*fc
)
3503 struct shmem_options
*ctx
= fc
->fs_private
;
3504 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3505 unsigned long inodes
;
3506 struct mempolicy
*mpol
= NULL
;
3509 raw_spin_lock(&sbinfo
->stat_lock
);
3510 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3511 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3512 if (!sbinfo
->max_blocks
) {
3513 err
= "Cannot retroactively limit size";
3516 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3518 err
= "Too small a size for current use";
3522 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3523 if (!sbinfo
->max_inodes
) {
3524 err
= "Cannot retroactively limit inodes";
3527 if (ctx
->inodes
< inodes
) {
3528 err
= "Too few inodes for current use";
3533 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
3534 sbinfo
->next_ino
> UINT_MAX
) {
3535 err
= "Current inum too high to switch to 32-bit inums";
3539 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3540 sbinfo
->huge
= ctx
->huge
;
3541 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
3542 sbinfo
->full_inums
= ctx
->full_inums
;
3543 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3544 sbinfo
->max_blocks
= ctx
->blocks
;
3545 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3546 sbinfo
->max_inodes
= ctx
->inodes
;
3547 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3551 * Preserve previous mempolicy unless mpol remount option was specified.
3554 mpol
= sbinfo
->mpol
;
3555 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3558 raw_spin_unlock(&sbinfo
->stat_lock
);
3562 raw_spin_unlock(&sbinfo
->stat_lock
);
3563 return invalfc(fc
, "%s", err
);
3566 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3568 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3570 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3571 seq_printf(seq
, ",size=%luk",
3572 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3573 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3574 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3575 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3576 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3577 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3578 seq_printf(seq
, ",uid=%u",
3579 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3580 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3581 seq_printf(seq
, ",gid=%u",
3582 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3585 * Showing inode{64,32} might be useful even if it's the system default,
3586 * since then people don't have to resort to checking both here and
3587 * /proc/config.gz to confirm 64-bit inums were successfully applied
3588 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3590 * We hide it when inode64 isn't the default and we are using 32-bit
3591 * inodes, since that probably just means the feature isn't even under
3596 * +-----------------+-----------------+
3597 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3598 * +------------------+-----------------+-----------------+
3599 * | full_inums=true | show | show |
3600 * | full_inums=false | show | hide |
3601 * +------------------+-----------------+-----------------+
3604 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
3605 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
3606 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3607 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3609 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3611 shmem_show_mpol(seq
, sbinfo
->mpol
);
3615 #endif /* CONFIG_TMPFS */
3617 static void shmem_put_super(struct super_block
*sb
)
3619 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3621 free_percpu(sbinfo
->ino_batch
);
3622 percpu_counter_destroy(&sbinfo
->used_blocks
);
3623 mpol_put(sbinfo
->mpol
);
3625 sb
->s_fs_info
= NULL
;
3628 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3630 struct shmem_options
*ctx
= fc
->fs_private
;
3631 struct inode
*inode
;
3632 struct shmem_sb_info
*sbinfo
;
3634 /* Round up to L1_CACHE_BYTES to resist false sharing */
3635 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3636 L1_CACHE_BYTES
), GFP_KERNEL
);
3640 sb
->s_fs_info
= sbinfo
;
3644 * Per default we only allow half of the physical ram per
3645 * tmpfs instance, limiting inodes to one per page of lowmem;
3646 * but the internal instance is left unlimited.
3648 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3649 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3650 ctx
->blocks
= shmem_default_max_blocks();
3651 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3652 ctx
->inodes
= shmem_default_max_inodes();
3653 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
3654 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
3656 sb
->s_flags
|= SB_NOUSER
;
3658 sb
->s_export_op
= &shmem_export_ops
;
3659 sb
->s_flags
|= SB_NOSEC
;
3661 sb
->s_flags
|= SB_NOUSER
;
3663 sbinfo
->max_blocks
= ctx
->blocks
;
3664 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3665 if (sb
->s_flags
& SB_KERNMOUNT
) {
3666 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
3667 if (!sbinfo
->ino_batch
)
3670 sbinfo
->uid
= ctx
->uid
;
3671 sbinfo
->gid
= ctx
->gid
;
3672 sbinfo
->full_inums
= ctx
->full_inums
;
3673 sbinfo
->mode
= ctx
->mode
;
3674 sbinfo
->huge
= ctx
->huge
;
3675 sbinfo
->mpol
= ctx
->mpol
;
3678 raw_spin_lock_init(&sbinfo
->stat_lock
);
3679 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3681 spin_lock_init(&sbinfo
->shrinklist_lock
);
3682 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3684 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3685 sb
->s_blocksize
= PAGE_SIZE
;
3686 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3687 sb
->s_magic
= TMPFS_MAGIC
;
3688 sb
->s_op
= &shmem_ops
;
3689 sb
->s_time_gran
= 1;
3690 #ifdef CONFIG_TMPFS_XATTR
3691 sb
->s_xattr
= shmem_xattr_handlers
;
3693 #ifdef CONFIG_TMPFS_POSIX_ACL
3694 sb
->s_flags
|= SB_POSIXACL
;
3696 uuid_gen(&sb
->s_uuid
);
3698 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3701 inode
->i_uid
= sbinfo
->uid
;
3702 inode
->i_gid
= sbinfo
->gid
;
3703 sb
->s_root
= d_make_root(inode
);
3709 shmem_put_super(sb
);
3713 static int shmem_get_tree(struct fs_context
*fc
)
3715 return get_tree_nodev(fc
, shmem_fill_super
);
3718 static void shmem_free_fc(struct fs_context
*fc
)
3720 struct shmem_options
*ctx
= fc
->fs_private
;
3723 mpol_put(ctx
->mpol
);
3728 static const struct fs_context_operations shmem_fs_context_ops
= {
3729 .free
= shmem_free_fc
,
3730 .get_tree
= shmem_get_tree
,
3732 .parse_monolithic
= shmem_parse_options
,
3733 .parse_param
= shmem_parse_one
,
3734 .reconfigure
= shmem_reconfigure
,
3738 static struct kmem_cache
*shmem_inode_cachep
;
3740 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3742 struct shmem_inode_info
*info
;
3743 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3746 return &info
->vfs_inode
;
3749 static void shmem_free_in_core_inode(struct inode
*inode
)
3751 if (S_ISLNK(inode
->i_mode
))
3752 kfree(inode
->i_link
);
3753 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3756 static void shmem_destroy_inode(struct inode
*inode
)
3758 if (S_ISREG(inode
->i_mode
))
3759 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3762 static void shmem_init_inode(void *foo
)
3764 struct shmem_inode_info
*info
= foo
;
3765 inode_init_once(&info
->vfs_inode
);
3768 static void shmem_init_inodecache(void)
3770 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3771 sizeof(struct shmem_inode_info
),
3772 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3775 static void shmem_destroy_inodecache(void)
3777 kmem_cache_destroy(shmem_inode_cachep
);
3780 const struct address_space_operations shmem_aops
= {
3781 .writepage
= shmem_writepage
,
3782 .set_page_dirty
= __set_page_dirty_no_writeback
,
3784 .write_begin
= shmem_write_begin
,
3785 .write_end
= shmem_write_end
,
3787 #ifdef CONFIG_MIGRATION
3788 .migratepage
= migrate_page
,
3790 .error_remove_page
= generic_error_remove_page
,
3792 EXPORT_SYMBOL(shmem_aops
);
3794 static const struct file_operations shmem_file_operations
= {
3796 .get_unmapped_area
= shmem_get_unmapped_area
,
3798 .llseek
= shmem_file_llseek
,
3799 .read_iter
= shmem_file_read_iter
,
3800 .write_iter
= generic_file_write_iter
,
3801 .fsync
= noop_fsync
,
3802 .splice_read
= generic_file_splice_read
,
3803 .splice_write
= iter_file_splice_write
,
3804 .fallocate
= shmem_fallocate
,
3808 static const struct inode_operations shmem_inode_operations
= {
3809 .getattr
= shmem_getattr
,
3810 .setattr
= shmem_setattr
,
3811 #ifdef CONFIG_TMPFS_XATTR
3812 .listxattr
= shmem_listxattr
,
3813 .set_acl
= simple_set_acl
,
3817 static const struct inode_operations shmem_dir_inode_operations
= {
3819 .create
= shmem_create
,
3820 .lookup
= simple_lookup
,
3822 .unlink
= shmem_unlink
,
3823 .symlink
= shmem_symlink
,
3824 .mkdir
= shmem_mkdir
,
3825 .rmdir
= shmem_rmdir
,
3826 .mknod
= shmem_mknod
,
3827 .rename
= shmem_rename2
,
3828 .tmpfile
= shmem_tmpfile
,
3830 #ifdef CONFIG_TMPFS_XATTR
3831 .listxattr
= shmem_listxattr
,
3833 #ifdef CONFIG_TMPFS_POSIX_ACL
3834 .setattr
= shmem_setattr
,
3835 .set_acl
= simple_set_acl
,
3839 static const struct inode_operations shmem_special_inode_operations
= {
3840 #ifdef CONFIG_TMPFS_XATTR
3841 .listxattr
= shmem_listxattr
,
3843 #ifdef CONFIG_TMPFS_POSIX_ACL
3844 .setattr
= shmem_setattr
,
3845 .set_acl
= simple_set_acl
,
3849 static const struct super_operations shmem_ops
= {
3850 .alloc_inode
= shmem_alloc_inode
,
3851 .free_inode
= shmem_free_in_core_inode
,
3852 .destroy_inode
= shmem_destroy_inode
,
3854 .statfs
= shmem_statfs
,
3855 .show_options
= shmem_show_options
,
3857 .evict_inode
= shmem_evict_inode
,
3858 .drop_inode
= generic_delete_inode
,
3859 .put_super
= shmem_put_super
,
3860 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3861 .nr_cached_objects
= shmem_unused_huge_count
,
3862 .free_cached_objects
= shmem_unused_huge_scan
,
3866 static const struct vm_operations_struct shmem_vm_ops
= {
3867 .fault
= shmem_fault
,
3868 .map_pages
= filemap_map_pages
,
3870 .set_policy
= shmem_set_policy
,
3871 .get_policy
= shmem_get_policy
,
3875 int shmem_init_fs_context(struct fs_context
*fc
)
3877 struct shmem_options
*ctx
;
3879 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3883 ctx
->mode
= 0777 | S_ISVTX
;
3884 ctx
->uid
= current_fsuid();
3885 ctx
->gid
= current_fsgid();
3887 fc
->fs_private
= ctx
;
3888 fc
->ops
= &shmem_fs_context_ops
;
3892 static struct file_system_type shmem_fs_type
= {
3893 .owner
= THIS_MODULE
,
3895 .init_fs_context
= shmem_init_fs_context
,
3897 .parameters
= shmem_fs_parameters
,
3899 .kill_sb
= kill_litter_super
,
3900 .fs_flags
= FS_USERNS_MOUNT
| FS_THP_SUPPORT
,
3903 int __init
shmem_init(void)
3907 shmem_init_inodecache();
3909 error
= register_filesystem(&shmem_fs_type
);
3911 pr_err("Could not register tmpfs\n");
3915 shm_mnt
= kern_mount(&shmem_fs_type
);
3916 if (IS_ERR(shm_mnt
)) {
3917 error
= PTR_ERR(shm_mnt
);
3918 pr_err("Could not kern_mount tmpfs\n");
3922 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3923 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3924 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3926 shmem_huge
= SHMEM_HUGE_NEVER
; /* just in case it was patched */
3931 unregister_filesystem(&shmem_fs_type
);
3933 shmem_destroy_inodecache();
3934 shm_mnt
= ERR_PTR(error
);
3938 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3939 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3940 struct kobj_attribute
*attr
, char *buf
)
3942 static const int values
[] = {
3944 SHMEM_HUGE_WITHIN_SIZE
,
3953 for (i
= 0; i
< ARRAY_SIZE(values
); i
++) {
3954 len
+= sysfs_emit_at(buf
, len
,
3955 shmem_huge
== values
[i
] ? "%s[%s]" : "%s%s",
3957 shmem_format_huge(values
[i
]));
3960 len
+= sysfs_emit_at(buf
, len
, "\n");
3965 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3966 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3971 if (count
+ 1 > sizeof(tmp
))
3973 memcpy(tmp
, buf
, count
);
3975 if (count
&& tmp
[count
- 1] == '\n')
3976 tmp
[count
- 1] = '\0';
3978 huge
= shmem_parse_huge(tmp
);
3979 if (huge
== -EINVAL
)
3981 if (!has_transparent_hugepage() &&
3982 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3986 if (shmem_huge
> SHMEM_HUGE_DENY
)
3987 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3991 struct kobj_attribute shmem_enabled_attr
=
3992 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3993 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3995 #else /* !CONFIG_SHMEM */
3998 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4000 * This is intended for small system where the benefits of the full
4001 * shmem code (swap-backed and resource-limited) are outweighed by
4002 * their complexity. On systems without swap this code should be
4003 * effectively equivalent, but much lighter weight.
4006 static struct file_system_type shmem_fs_type
= {
4008 .init_fs_context
= ramfs_init_fs_context
,
4009 .parameters
= ramfs_fs_parameters
,
4010 .kill_sb
= kill_litter_super
,
4011 .fs_flags
= FS_USERNS_MOUNT
,
4014 int __init
shmem_init(void)
4016 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4018 shm_mnt
= kern_mount(&shmem_fs_type
);
4019 BUG_ON(IS_ERR(shm_mnt
));
4024 int shmem_unuse(unsigned int type
, bool frontswap
,
4025 unsigned long *fs_pages_to_unuse
)
4030 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
4035 void shmem_unlock_mapping(struct address_space
*mapping
)
4040 unsigned long shmem_get_unmapped_area(struct file
*file
,
4041 unsigned long addr
, unsigned long len
,
4042 unsigned long pgoff
, unsigned long flags
)
4044 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4048 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4050 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4052 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4054 #define shmem_vm_ops generic_file_vm_ops
4055 #define shmem_file_operations ramfs_file_operations
4056 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4057 #define shmem_acct_size(flags, size) 0
4058 #define shmem_unacct_size(flags, size) do {} while (0)
4060 #endif /* CONFIG_SHMEM */
4064 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4065 unsigned long flags
, unsigned int i_flags
)
4067 struct inode
*inode
;
4071 return ERR_CAST(mnt
);
4073 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4074 return ERR_PTR(-EINVAL
);
4076 if (shmem_acct_size(flags
, size
))
4077 return ERR_PTR(-ENOMEM
);
4079 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4081 if (unlikely(!inode
)) {
4082 shmem_unacct_size(flags
, size
);
4083 return ERR_PTR(-ENOSPC
);
4085 inode
->i_flags
|= i_flags
;
4086 inode
->i_size
= size
;
4087 clear_nlink(inode
); /* It is unlinked */
4088 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4090 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4091 &shmem_file_operations
);
4098 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4099 * kernel internal. There will be NO LSM permission checks against the
4100 * underlying inode. So users of this interface must do LSM checks at a
4101 * higher layer. The users are the big_key and shm implementations. LSM
4102 * checks are provided at the key or shm level rather than the inode.
4103 * @name: name for dentry (to be seen in /proc/<pid>/maps
4104 * @size: size to be set for the file
4105 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4107 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4109 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4113 * shmem_file_setup - get an unlinked file living in tmpfs
4114 * @name: name for dentry (to be seen in /proc/<pid>/maps
4115 * @size: size to be set for the file
4116 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4118 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4120 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4122 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4125 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4126 * @mnt: the tmpfs mount where the file will be created
4127 * @name: name for dentry (to be seen in /proc/<pid>/maps
4128 * @size: size to be set for the file
4129 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4131 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4132 loff_t size
, unsigned long flags
)
4134 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4136 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4139 * shmem_zero_setup - setup a shared anonymous mapping
4140 * @vma: the vma to be mmapped is prepared by do_mmap
4142 int shmem_zero_setup(struct vm_area_struct
*vma
)
4145 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4148 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4149 * between XFS directory reading and selinux: since this file is only
4150 * accessible to the user through its mapping, use S_PRIVATE flag to
4151 * bypass file security, in the same way as shmem_kernel_file_setup().
4153 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4155 return PTR_ERR(file
);
4159 vma
->vm_file
= file
;
4160 vma
->vm_ops
= &shmem_vm_ops
;
4162 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
4163 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4164 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4165 khugepaged_enter(vma
, vma
->vm_flags
);
4170 EXPORT_SYMBOL_GPL(shmem_zero_setup
);
4173 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4174 * @mapping: the page's address_space
4175 * @index: the page index
4176 * @gfp: the page allocator flags to use if allocating
4178 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4179 * with any new page allocations done using the specified allocation flags.
4180 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4181 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4182 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4184 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4185 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4187 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4188 pgoff_t index
, gfp_t gfp
)
4191 struct inode
*inode
= mapping
->host
;
4195 BUG_ON(!shmem_mapping(mapping
));
4196 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4197 gfp
, NULL
, NULL
, NULL
);
4199 page
= ERR_PTR(error
);
4205 * The tiny !SHMEM case uses ramfs without swap
4207 return read_cache_page_gfp(mapping
, index
, gfp
);
4210 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);