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/pagevec.h>
63 #include <linux/percpu_counter.h>
64 #include <linux/falloc.h>
65 #include <linux/splice.h>
66 #include <linux/security.h>
67 #include <linux/swapops.h>
68 #include <linux/mempolicy.h>
69 #include <linux/namei.h>
70 #include <linux/ctype.h>
71 #include <linux/migrate.h>
72 #include <linux/highmem.h>
73 #include <linux/seq_file.h>
74 #include <linux/magic.h>
75 #include <linux/syscalls.h>
76 #include <linux/fcntl.h>
77 #include <uapi/linux/memfd.h>
78 #include <linux/userfaultfd_k.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
82 #include <linux/uaccess.h>
86 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 /* Pretend that each entry is of this size in directory's i_size */
90 #define BOGO_DIRENT_SIZE 20
92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93 #define SHORT_SYMLINK_LEN 128
96 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_rwsem making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
100 struct shmem_falloc
{
101 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
102 pgoff_t start
; /* start of range currently being fallocated */
103 pgoff_t next
; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
108 struct shmem_options
{
109 unsigned long long blocks
;
110 unsigned long long inodes
;
111 struct mempolicy
*mpol
;
118 #define SHMEM_SEEN_BLOCKS 1
119 #define SHMEM_SEEN_INODES 2
120 #define SHMEM_SEEN_HUGE 4
121 #define SHMEM_SEEN_INUMS 8
125 static unsigned long shmem_default_max_blocks(void)
127 return totalram_pages() / 2;
130 static unsigned long shmem_default_max_inodes(void)
132 unsigned long nr_pages
= totalram_pages();
134 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
138 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
139 struct page
**pagep
, enum sgp_type sgp
,
140 gfp_t gfp
, struct vm_area_struct
*vma
,
141 vm_fault_t
*fault_type
);
142 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
143 struct page
**pagep
, enum sgp_type sgp
,
144 gfp_t gfp
, struct vm_area_struct
*vma
,
145 struct vm_fault
*vmf
, vm_fault_t
*fault_type
);
147 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
148 struct page
**pagep
, enum sgp_type sgp
)
150 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
151 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
154 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
156 return sb
->s_fs_info
;
160 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
161 * for shared memory and for shared anonymous (/dev/zero) mappings
162 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
163 * consistent with the pre-accounting of private mappings ...
165 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
167 return (flags
& VM_NORESERVE
) ?
168 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
171 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
173 if (!(flags
& VM_NORESERVE
))
174 vm_unacct_memory(VM_ACCT(size
));
177 static inline int shmem_reacct_size(unsigned long flags
,
178 loff_t oldsize
, loff_t newsize
)
180 if (!(flags
& VM_NORESERVE
)) {
181 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
182 return security_vm_enough_memory_mm(current
->mm
,
183 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
184 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
185 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
191 * ... whereas tmpfs objects are accounted incrementally as
192 * pages are allocated, in order to allow large sparse files.
193 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
194 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
196 static inline int shmem_acct_block(unsigned long flags
, long pages
)
198 if (!(flags
& VM_NORESERVE
))
201 return security_vm_enough_memory_mm(current
->mm
,
202 pages
* VM_ACCT(PAGE_SIZE
));
205 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
207 if (flags
& VM_NORESERVE
)
208 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
211 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
213 struct shmem_inode_info
*info
= SHMEM_I(inode
);
214 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
216 if (shmem_acct_block(info
->flags
, pages
))
219 if (sbinfo
->max_blocks
) {
220 if (percpu_counter_compare(&sbinfo
->used_blocks
,
221 sbinfo
->max_blocks
- pages
) > 0)
223 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
229 shmem_unacct_blocks(info
->flags
, pages
);
233 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
235 struct shmem_inode_info
*info
= SHMEM_I(inode
);
236 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
238 if (sbinfo
->max_blocks
)
239 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
240 shmem_unacct_blocks(info
->flags
, pages
);
243 static const struct super_operations shmem_ops
;
244 const struct address_space_operations shmem_aops
;
245 static const struct file_operations shmem_file_operations
;
246 static const struct inode_operations shmem_inode_operations
;
247 static const struct inode_operations shmem_dir_inode_operations
;
248 static const struct inode_operations shmem_special_inode_operations
;
249 static const struct vm_operations_struct shmem_vm_ops
;
250 static struct file_system_type shmem_fs_type
;
252 bool vma_is_shmem(struct vm_area_struct
*vma
)
254 return vma
->vm_ops
== &shmem_vm_ops
;
257 static LIST_HEAD(shmem_swaplist
);
258 static DEFINE_MUTEX(shmem_swaplist_mutex
);
261 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
262 * produces a novel ino for the newly allocated inode.
264 * It may also be called when making a hard link to permit the space needed by
265 * each dentry. However, in that case, no new inode number is needed since that
266 * internally draws from another pool of inode numbers (currently global
267 * get_next_ino()). This case is indicated by passing NULL as inop.
269 #define SHMEM_INO_BATCH 1024
270 static int shmem_reserve_inode(struct super_block
*sb
, ino_t
*inop
)
272 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
275 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
276 raw_spin_lock(&sbinfo
->stat_lock
);
277 if (sbinfo
->max_inodes
) {
278 if (!sbinfo
->free_inodes
) {
279 raw_spin_unlock(&sbinfo
->stat_lock
);
282 sbinfo
->free_inodes
--;
285 ino
= sbinfo
->next_ino
++;
286 if (unlikely(is_zero_ino(ino
)))
287 ino
= sbinfo
->next_ino
++;
288 if (unlikely(!sbinfo
->full_inums
&&
291 * Emulate get_next_ino uint wraparound for
294 if (IS_ENABLED(CONFIG_64BIT
))
295 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
296 __func__
, MINOR(sb
->s_dev
));
297 sbinfo
->next_ino
= 1;
298 ino
= sbinfo
->next_ino
++;
302 raw_spin_unlock(&sbinfo
->stat_lock
);
305 * __shmem_file_setup, one of our callers, is lock-free: it
306 * doesn't hold stat_lock in shmem_reserve_inode since
307 * max_inodes is always 0, and is called from potentially
308 * unknown contexts. As such, use a per-cpu batched allocator
309 * which doesn't require the per-sb stat_lock unless we are at
310 * the batch boundary.
312 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
313 * shmem mounts are not exposed to userspace, so we don't need
314 * to worry about things like glibc compatibility.
318 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
320 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
321 raw_spin_lock(&sbinfo
->stat_lock
);
322 ino
= sbinfo
->next_ino
;
323 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
324 raw_spin_unlock(&sbinfo
->stat_lock
);
325 if (unlikely(is_zero_ino(ino
)))
336 static void shmem_free_inode(struct super_block
*sb
)
338 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
339 if (sbinfo
->max_inodes
) {
340 raw_spin_lock(&sbinfo
->stat_lock
);
341 sbinfo
->free_inodes
++;
342 raw_spin_unlock(&sbinfo
->stat_lock
);
347 * shmem_recalc_inode - recalculate the block usage of an inode
348 * @inode: inode to recalc
350 * We have to calculate the free blocks since the mm can drop
351 * undirtied hole pages behind our back.
353 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
354 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
356 * It has to be called with the spinlock held.
358 static void shmem_recalc_inode(struct inode
*inode
)
360 struct shmem_inode_info
*info
= SHMEM_I(inode
);
363 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
365 info
->alloced
-= freed
;
366 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
367 shmem_inode_unacct_blocks(inode
, freed
);
371 bool shmem_charge(struct inode
*inode
, long pages
)
373 struct shmem_inode_info
*info
= SHMEM_I(inode
);
376 if (!shmem_inode_acct_block(inode
, pages
))
379 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
380 inode
->i_mapping
->nrpages
+= pages
;
382 spin_lock_irqsave(&info
->lock
, flags
);
383 info
->alloced
+= pages
;
384 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
385 shmem_recalc_inode(inode
);
386 spin_unlock_irqrestore(&info
->lock
, flags
);
391 void shmem_uncharge(struct inode
*inode
, long pages
)
393 struct shmem_inode_info
*info
= SHMEM_I(inode
);
396 /* nrpages adjustment done by __delete_from_page_cache() or caller */
398 spin_lock_irqsave(&info
->lock
, flags
);
399 info
->alloced
-= pages
;
400 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
401 shmem_recalc_inode(inode
);
402 spin_unlock_irqrestore(&info
->lock
, flags
);
404 shmem_inode_unacct_blocks(inode
, pages
);
408 * Replace item expected in xarray by a new item, while holding xa_lock.
410 static int shmem_replace_entry(struct address_space
*mapping
,
411 pgoff_t index
, void *expected
, void *replacement
)
413 XA_STATE(xas
, &mapping
->i_pages
, index
);
416 VM_BUG_ON(!expected
);
417 VM_BUG_ON(!replacement
);
418 item
= xas_load(&xas
);
419 if (item
!= expected
)
421 xas_store(&xas
, replacement
);
426 * Sometimes, before we decide whether to proceed or to fail, we must check
427 * that an entry was not already brought back from swap by a racing thread.
429 * Checking page is not enough: by the time a SwapCache page is locked, it
430 * might be reused, and again be SwapCache, using the same swap as before.
432 static bool shmem_confirm_swap(struct address_space
*mapping
,
433 pgoff_t index
, swp_entry_t swap
)
435 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
439 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
442 * disables huge pages for the mount;
444 * enables huge pages for the mount;
445 * SHMEM_HUGE_WITHIN_SIZE:
446 * only allocate huge pages if the page will be fully within i_size,
447 * also respect fadvise()/madvise() hints;
449 * only allocate huge pages if requested with fadvise()/madvise();
452 #define SHMEM_HUGE_NEVER 0
453 #define SHMEM_HUGE_ALWAYS 1
454 #define SHMEM_HUGE_WITHIN_SIZE 2
455 #define SHMEM_HUGE_ADVISE 3
459 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
462 * disables huge on shm_mnt and all mounts, for emergency use;
464 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
467 #define SHMEM_HUGE_DENY (-1)
468 #define SHMEM_HUGE_FORCE (-2)
470 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
471 /* ifdef here to avoid bloating shmem.o when not necessary */
473 static int shmem_huge __read_mostly
= SHMEM_HUGE_NEVER
;
475 bool shmem_is_huge(struct vm_area_struct
*vma
,
476 struct inode
*inode
, pgoff_t index
)
480 if (shmem_huge
== SHMEM_HUGE_DENY
)
482 if (vma
&& ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
483 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
)))
485 if (shmem_huge
== SHMEM_HUGE_FORCE
)
488 switch (SHMEM_SB(inode
->i_sb
)->huge
) {
489 case SHMEM_HUGE_ALWAYS
:
491 case SHMEM_HUGE_WITHIN_SIZE
:
492 index
= round_up(index
+ 1, HPAGE_PMD_NR
);
493 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
494 if (i_size
>> PAGE_SHIFT
>= index
)
497 case SHMEM_HUGE_ADVISE
:
498 if (vma
&& (vma
->vm_flags
& VM_HUGEPAGE
))
506 #if defined(CONFIG_SYSFS)
507 static int shmem_parse_huge(const char *str
)
509 if (!strcmp(str
, "never"))
510 return SHMEM_HUGE_NEVER
;
511 if (!strcmp(str
, "always"))
512 return SHMEM_HUGE_ALWAYS
;
513 if (!strcmp(str
, "within_size"))
514 return SHMEM_HUGE_WITHIN_SIZE
;
515 if (!strcmp(str
, "advise"))
516 return SHMEM_HUGE_ADVISE
;
517 if (!strcmp(str
, "deny"))
518 return SHMEM_HUGE_DENY
;
519 if (!strcmp(str
, "force"))
520 return SHMEM_HUGE_FORCE
;
525 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
526 static const char *shmem_format_huge(int huge
)
529 case SHMEM_HUGE_NEVER
:
531 case SHMEM_HUGE_ALWAYS
:
533 case SHMEM_HUGE_WITHIN_SIZE
:
534 return "within_size";
535 case SHMEM_HUGE_ADVISE
:
537 case SHMEM_HUGE_DENY
:
539 case SHMEM_HUGE_FORCE
:
548 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
549 struct shrink_control
*sc
, unsigned long nr_to_split
)
551 LIST_HEAD(list
), *pos
, *next
;
552 LIST_HEAD(to_remove
);
554 struct shmem_inode_info
*info
;
556 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
557 int removed
= 0, split
= 0;
559 if (list_empty(&sbinfo
->shrinklist
))
562 spin_lock(&sbinfo
->shrinklist_lock
);
563 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
564 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
567 inode
= igrab(&info
->vfs_inode
);
569 /* inode is about to be evicted */
571 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
);
584 list_move(&info
->shrinklist
, &list
);
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 * Leave the inode on the list if we failed to lock
620 * 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 leave the inode on the list */
640 list_del_init(&info
->shrinklist
);
646 spin_lock(&sbinfo
->shrinklist_lock
);
647 list_splice_tail(&list
, &sbinfo
->shrinklist
);
648 sbinfo
->shrinklist_len
-= removed
;
649 spin_unlock(&sbinfo
->shrinklist_lock
);
654 static long shmem_unused_huge_scan(struct super_block
*sb
,
655 struct shrink_control
*sc
)
657 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
659 if (!READ_ONCE(sbinfo
->shrinklist_len
))
662 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
665 static long shmem_unused_huge_count(struct super_block
*sb
,
666 struct shrink_control
*sc
)
668 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
669 return READ_ONCE(sbinfo
->shrinklist_len
);
671 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
673 #define shmem_huge SHMEM_HUGE_DENY
675 bool shmem_is_huge(struct vm_area_struct
*vma
,
676 struct inode
*inode
, pgoff_t index
)
681 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
682 struct shrink_control
*sc
, unsigned long nr_to_split
)
686 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
689 * Like add_to_page_cache_locked, but error if expected item has gone.
691 static int shmem_add_to_page_cache(struct page
*page
,
692 struct address_space
*mapping
,
693 pgoff_t index
, void *expected
, gfp_t gfp
,
694 struct mm_struct
*charge_mm
)
696 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, compound_order(page
));
698 unsigned long nr
= compound_nr(page
);
701 VM_BUG_ON_PAGE(PageTail(page
), page
);
702 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
703 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
704 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
705 VM_BUG_ON(expected
&& PageTransHuge(page
));
707 page_ref_add(page
, nr
);
708 page
->mapping
= mapping
;
711 if (!PageSwapCache(page
)) {
712 error
= mem_cgroup_charge(page_folio(page
), charge_mm
, gfp
);
714 if (PageTransHuge(page
)) {
715 count_vm_event(THP_FILE_FALLBACK
);
716 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
721 cgroup_throttle_swaprate(page
, gfp
);
726 entry
= xas_find_conflict(&xas
);
727 if (entry
!= expected
)
728 xas_set_err(&xas
, -EEXIST
);
729 xas_create_range(&xas
);
733 xas_store(&xas
, page
);
738 if (PageTransHuge(page
)) {
739 count_vm_event(THP_FILE_ALLOC
);
740 __mod_lruvec_page_state(page
, NR_SHMEM_THPS
, nr
);
742 mapping
->nrpages
+= nr
;
743 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, nr
);
744 __mod_lruvec_page_state(page
, NR_SHMEM
, nr
);
746 xas_unlock_irq(&xas
);
747 } while (xas_nomem(&xas
, gfp
));
749 if (xas_error(&xas
)) {
750 error
= xas_error(&xas
);
756 page
->mapping
= NULL
;
757 page_ref_sub(page
, nr
);
762 * Like delete_from_page_cache, but substitutes swap for page.
764 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
766 struct address_space
*mapping
= page
->mapping
;
769 VM_BUG_ON_PAGE(PageCompound(page
), page
);
771 xa_lock_irq(&mapping
->i_pages
);
772 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
773 page
->mapping
= NULL
;
775 __dec_lruvec_page_state(page
, NR_FILE_PAGES
);
776 __dec_lruvec_page_state(page
, NR_SHMEM
);
777 xa_unlock_irq(&mapping
->i_pages
);
783 * Remove swap entry from page cache, free the swap and its page cache.
785 static int shmem_free_swap(struct address_space
*mapping
,
786 pgoff_t index
, void *radswap
)
790 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
793 free_swap_and_cache(radix_to_swp_entry(radswap
));
798 * Determine (in bytes) how many of the shmem object's pages mapped by the
799 * given offsets are swapped out.
801 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
802 * as long as the inode doesn't go away and racy results are not a problem.
804 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
805 pgoff_t start
, pgoff_t end
)
807 XA_STATE(xas
, &mapping
->i_pages
, start
);
809 unsigned long swapped
= 0;
812 xas_for_each(&xas
, page
, end
- 1) {
813 if (xas_retry(&xas
, page
))
815 if (xa_is_value(page
))
818 if (need_resched()) {
826 return swapped
<< PAGE_SHIFT
;
830 * Determine (in bytes) how many of the shmem object's pages mapped by the
831 * given vma is swapped out.
833 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
834 * as long as the inode doesn't go away and racy results are not a problem.
836 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
838 struct inode
*inode
= file_inode(vma
->vm_file
);
839 struct shmem_inode_info
*info
= SHMEM_I(inode
);
840 struct address_space
*mapping
= inode
->i_mapping
;
841 unsigned long swapped
;
843 /* Be careful as we don't hold info->lock */
844 swapped
= READ_ONCE(info
->swapped
);
847 * The easier cases are when the shmem object has nothing in swap, or
848 * the vma maps it whole. Then we can simply use the stats that we
854 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
855 return swapped
<< PAGE_SHIFT
;
857 /* Here comes the more involved part */
858 return shmem_partial_swap_usage(mapping
,
859 linear_page_index(vma
, vma
->vm_start
),
860 linear_page_index(vma
, vma
->vm_end
));
864 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
866 void shmem_unlock_mapping(struct address_space
*mapping
)
873 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
875 while (!mapping_unevictable(mapping
)) {
876 if (!pagevec_lookup(&pvec
, mapping
, &index
))
878 check_move_unevictable_pages(&pvec
);
879 pagevec_release(&pvec
);
885 * Check whether a hole-punch or truncation needs to split a huge page,
886 * returning true if no split was required, or the split has been successful.
888 * Eviction (or truncation to 0 size) should never need to split a huge page;
889 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
890 * head, and then succeeded to trylock on tail.
892 * A split can only succeed when there are no additional references on the
893 * huge page: so the split below relies upon find_get_entries() having stopped
894 * when it found a subpage of the huge page, without getting further references.
896 static bool shmem_punch_compound(struct page
*page
, pgoff_t start
, pgoff_t end
)
898 if (!PageTransCompound(page
))
901 /* Just proceed to delete a huge page wholly within the range punched */
902 if (PageHead(page
) &&
903 page
->index
>= start
&& page
->index
+ HPAGE_PMD_NR
<= end
)
906 /* Try to split huge page, so we can truly punch the hole or truncate */
907 return split_huge_page(page
) >= 0;
911 * Remove range of pages and swap entries from page cache, and free them.
912 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
914 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
917 struct address_space
*mapping
= inode
->i_mapping
;
918 struct shmem_inode_info
*info
= SHMEM_I(inode
);
919 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
920 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
921 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
922 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
924 pgoff_t indices
[PAGEVEC_SIZE
];
925 long nr_swaps_freed
= 0;
930 end
= -1; /* unsigned, so actually very big */
932 if (info
->fallocend
> start
&& info
->fallocend
<= end
&& !unfalloc
)
933 info
->fallocend
= start
;
937 while (index
< end
&& find_lock_entries(mapping
, index
, end
- 1,
939 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
940 struct page
*page
= pvec
.pages
[i
];
944 if (xa_is_value(page
)) {
947 nr_swaps_freed
+= !shmem_free_swap(mapping
,
951 index
+= thp_nr_pages(page
) - 1;
953 if (!unfalloc
|| !PageUptodate(page
))
954 truncate_inode_page(mapping
, page
);
957 pagevec_remove_exceptionals(&pvec
);
958 pagevec_release(&pvec
);
964 struct page
*page
= NULL
;
965 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
967 unsigned int top
= PAGE_SIZE
;
972 zero_user_segment(page
, partial_start
, top
);
973 set_page_dirty(page
);
979 struct page
*page
= NULL
;
980 shmem_getpage(inode
, end
, &page
, SGP_READ
);
982 zero_user_segment(page
, 0, partial_end
);
983 set_page_dirty(page
);
992 while (index
< end
) {
995 if (!find_get_entries(mapping
, index
, end
- 1, &pvec
,
997 /* If all gone or hole-punch or unfalloc, we're done */
998 if (index
== start
|| end
!= -1)
1000 /* But if truncating, restart to make sure all gone */
1004 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
1005 struct page
*page
= pvec
.pages
[i
];
1008 if (xa_is_value(page
)) {
1011 if (shmem_free_swap(mapping
, index
, page
)) {
1012 /* Swap was replaced by page: retry */
1022 if (!unfalloc
|| !PageUptodate(page
)) {
1023 if (page_mapping(page
) != mapping
) {
1024 /* Page was replaced by swap: retry */
1029 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
1030 if (shmem_punch_compound(page
, start
, end
))
1031 truncate_inode_page(mapping
, page
);
1032 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1033 /* Wipe the page and don't get stuck */
1034 clear_highpage(page
);
1035 flush_dcache_page(page
);
1036 set_page_dirty(page
);
1038 round_up(start
, HPAGE_PMD_NR
))
1044 pagevec_remove_exceptionals(&pvec
);
1045 pagevec_release(&pvec
);
1049 spin_lock_irq(&info
->lock
);
1050 info
->swapped
-= nr_swaps_freed
;
1051 shmem_recalc_inode(inode
);
1052 spin_unlock_irq(&info
->lock
);
1055 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1057 shmem_undo_range(inode
, lstart
, lend
, false);
1058 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1060 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1062 static int shmem_getattr(struct user_namespace
*mnt_userns
,
1063 const struct path
*path
, struct kstat
*stat
,
1064 u32 request_mask
, unsigned int query_flags
)
1066 struct inode
*inode
= path
->dentry
->d_inode
;
1067 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1069 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1070 spin_lock_irq(&info
->lock
);
1071 shmem_recalc_inode(inode
);
1072 spin_unlock_irq(&info
->lock
);
1074 generic_fillattr(&init_user_ns
, inode
, stat
);
1076 if (shmem_is_huge(NULL
, inode
, 0))
1077 stat
->blksize
= HPAGE_PMD_SIZE
;
1082 static int shmem_setattr(struct user_namespace
*mnt_userns
,
1083 struct dentry
*dentry
, struct iattr
*attr
)
1085 struct inode
*inode
= d_inode(dentry
);
1086 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1089 error
= setattr_prepare(&init_user_ns
, dentry
, attr
);
1093 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1094 loff_t oldsize
= inode
->i_size
;
1095 loff_t newsize
= attr
->ia_size
;
1097 /* protected by i_rwsem */
1098 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1099 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1102 if (newsize
!= oldsize
) {
1103 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1107 i_size_write(inode
, newsize
);
1108 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1110 if (newsize
<= oldsize
) {
1111 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1112 if (oldsize
> holebegin
)
1113 unmap_mapping_range(inode
->i_mapping
,
1116 shmem_truncate_range(inode
,
1117 newsize
, (loff_t
)-1);
1118 /* unmap again to remove racily COWed private pages */
1119 if (oldsize
> holebegin
)
1120 unmap_mapping_range(inode
->i_mapping
,
1125 setattr_copy(&init_user_ns
, inode
, attr
);
1126 if (attr
->ia_valid
& ATTR_MODE
)
1127 error
= posix_acl_chmod(&init_user_ns
, inode
, inode
->i_mode
);
1131 static void shmem_evict_inode(struct inode
*inode
)
1133 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1134 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1136 if (shmem_mapping(inode
->i_mapping
)) {
1137 shmem_unacct_size(info
->flags
, inode
->i_size
);
1139 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1140 if (!list_empty(&info
->shrinklist
)) {
1141 spin_lock(&sbinfo
->shrinklist_lock
);
1142 if (!list_empty(&info
->shrinklist
)) {
1143 list_del_init(&info
->shrinklist
);
1144 sbinfo
->shrinklist_len
--;
1146 spin_unlock(&sbinfo
->shrinklist_lock
);
1148 while (!list_empty(&info
->swaplist
)) {
1149 /* Wait while shmem_unuse() is scanning this inode... */
1150 wait_var_event(&info
->stop_eviction
,
1151 !atomic_read(&info
->stop_eviction
));
1152 mutex_lock(&shmem_swaplist_mutex
);
1153 /* ...but beware of the race if we peeked too early */
1154 if (!atomic_read(&info
->stop_eviction
))
1155 list_del_init(&info
->swaplist
);
1156 mutex_unlock(&shmem_swaplist_mutex
);
1160 simple_xattrs_free(&info
->xattrs
);
1161 WARN_ON(inode
->i_blocks
);
1162 shmem_free_inode(inode
->i_sb
);
1166 static int shmem_find_swap_entries(struct address_space
*mapping
,
1167 pgoff_t start
, unsigned int nr_entries
,
1168 struct page
**entries
, pgoff_t
*indices
,
1169 unsigned int type
, bool frontswap
)
1171 XA_STATE(xas
, &mapping
->i_pages
, start
);
1174 unsigned int ret
= 0;
1180 xas_for_each(&xas
, page
, ULONG_MAX
) {
1181 if (xas_retry(&xas
, page
))
1184 if (!xa_is_value(page
))
1187 entry
= radix_to_swp_entry(page
);
1188 if (swp_type(entry
) != type
)
1191 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1194 indices
[ret
] = xas
.xa_index
;
1195 entries
[ret
] = page
;
1197 if (need_resched()) {
1201 if (++ret
== nr_entries
)
1210 * Move the swapped pages for an inode to page cache. Returns the count
1211 * of pages swapped in, or the error in case of failure.
1213 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1219 struct address_space
*mapping
= inode
->i_mapping
;
1221 for (i
= 0; i
< pvec
.nr
; i
++) {
1222 struct page
*page
= pvec
.pages
[i
];
1224 if (!xa_is_value(page
))
1226 error
= shmem_swapin_page(inode
, indices
[i
],
1228 mapping_gfp_mask(mapping
),
1235 if (error
== -ENOMEM
)
1239 return error
? error
: ret
;
1243 * If swap found in inode, free it and move page from swapcache to filecache.
1245 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1246 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1248 struct address_space
*mapping
= inode
->i_mapping
;
1250 struct pagevec pvec
;
1251 pgoff_t indices
[PAGEVEC_SIZE
];
1252 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1255 pagevec_init(&pvec
);
1257 unsigned int nr_entries
= PAGEVEC_SIZE
;
1259 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1260 nr_entries
= *fs_pages_to_unuse
;
1262 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1263 pvec
.pages
, indices
,
1270 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1274 if (frontswap_partial
) {
1275 *fs_pages_to_unuse
-= ret
;
1276 if (*fs_pages_to_unuse
== 0) {
1277 ret
= FRONTSWAP_PAGES_UNUSED
;
1282 start
= indices
[pvec
.nr
- 1];
1289 * Read all the shared memory data that resides in the swap
1290 * device 'type' back into memory, so the swap device can be
1293 int shmem_unuse(unsigned int type
, bool frontswap
,
1294 unsigned long *fs_pages_to_unuse
)
1296 struct shmem_inode_info
*info
, *next
;
1299 if (list_empty(&shmem_swaplist
))
1302 mutex_lock(&shmem_swaplist_mutex
);
1303 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1304 if (!info
->swapped
) {
1305 list_del_init(&info
->swaplist
);
1309 * Drop the swaplist mutex while searching the inode for swap;
1310 * but before doing so, make sure shmem_evict_inode() will not
1311 * remove placeholder inode from swaplist, nor let it be freed
1312 * (igrab() would protect from unlink, but not from unmount).
1314 atomic_inc(&info
->stop_eviction
);
1315 mutex_unlock(&shmem_swaplist_mutex
);
1317 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1321 mutex_lock(&shmem_swaplist_mutex
);
1322 next
= list_next_entry(info
, swaplist
);
1324 list_del_init(&info
->swaplist
);
1325 if (atomic_dec_and_test(&info
->stop_eviction
))
1326 wake_up_var(&info
->stop_eviction
);
1330 mutex_unlock(&shmem_swaplist_mutex
);
1336 * Move the page from the page cache to the swap cache.
1338 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1340 struct shmem_inode_info
*info
;
1341 struct address_space
*mapping
;
1342 struct inode
*inode
;
1347 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1348 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1349 * and its shmem_writeback() needs them to be split when swapping.
1351 if (PageTransCompound(page
)) {
1352 /* Ensure the subpages are still dirty */
1354 if (split_huge_page(page
) < 0)
1356 ClearPageDirty(page
);
1359 BUG_ON(!PageLocked(page
));
1360 mapping
= page
->mapping
;
1361 index
= page
->index
;
1362 inode
= mapping
->host
;
1363 info
= SHMEM_I(inode
);
1364 if (info
->flags
& VM_LOCKED
)
1366 if (!total_swap_pages
)
1370 * Our capabilities prevent regular writeback or sync from ever calling
1371 * shmem_writepage; but a stacking filesystem might use ->writepage of
1372 * its underlying filesystem, in which case tmpfs should write out to
1373 * swap only in response to memory pressure, and not for the writeback
1376 if (!wbc
->for_reclaim
) {
1377 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1382 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1383 * value into swapfile.c, the only way we can correctly account for a
1384 * fallocated page arriving here is now to initialize it and write it.
1386 * That's okay for a page already fallocated earlier, but if we have
1387 * not yet completed the fallocation, then (a) we want to keep track
1388 * of this page in case we have to undo it, and (b) it may not be a
1389 * good idea to continue anyway, once we're pushing into swap. So
1390 * reactivate the page, and let shmem_fallocate() quit when too many.
1392 if (!PageUptodate(page
)) {
1393 if (inode
->i_private
) {
1394 struct shmem_falloc
*shmem_falloc
;
1395 spin_lock(&inode
->i_lock
);
1396 shmem_falloc
= inode
->i_private
;
1398 !shmem_falloc
->waitq
&&
1399 index
>= shmem_falloc
->start
&&
1400 index
< shmem_falloc
->next
)
1401 shmem_falloc
->nr_unswapped
++;
1403 shmem_falloc
= NULL
;
1404 spin_unlock(&inode
->i_lock
);
1408 clear_highpage(page
);
1409 flush_dcache_page(page
);
1410 SetPageUptodate(page
);
1413 swap
= get_swap_page(page
);
1418 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1419 * if it's not already there. Do it now before the page is
1420 * moved to swap cache, when its pagelock no longer protects
1421 * the inode from eviction. But don't unlock the mutex until
1422 * we've incremented swapped, because shmem_unuse_inode() will
1423 * prune a !swapped inode from the swaplist under this mutex.
1425 mutex_lock(&shmem_swaplist_mutex
);
1426 if (list_empty(&info
->swaplist
))
1427 list_add(&info
->swaplist
, &shmem_swaplist
);
1429 if (add_to_swap_cache(page
, swap
,
1430 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1432 spin_lock_irq(&info
->lock
);
1433 shmem_recalc_inode(inode
);
1435 spin_unlock_irq(&info
->lock
);
1437 swap_shmem_alloc(swap
);
1438 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1440 mutex_unlock(&shmem_swaplist_mutex
);
1441 BUG_ON(page_mapped(page
));
1442 swap_writepage(page
, wbc
);
1446 mutex_unlock(&shmem_swaplist_mutex
);
1447 put_swap_page(page
, swap
);
1449 set_page_dirty(page
);
1450 if (wbc
->for_reclaim
)
1451 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1456 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1457 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1461 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1462 return; /* show nothing */
1464 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1466 seq_printf(seq
, ",mpol=%s", buffer
);
1469 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1471 struct mempolicy
*mpol
= NULL
;
1473 raw_spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1474 mpol
= sbinfo
->mpol
;
1476 raw_spin_unlock(&sbinfo
->stat_lock
);
1480 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1481 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1484 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1488 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1490 #define vm_policy vm_private_data
1493 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1494 struct shmem_inode_info
*info
, pgoff_t index
)
1496 /* Create a pseudo vma that just contains the policy */
1497 vma_init(vma
, NULL
);
1498 /* Bias interleave by inode number to distribute better across nodes */
1499 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1500 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1503 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1505 /* Drop reference taken by mpol_shared_policy_lookup() */
1506 mpol_cond_put(vma
->vm_policy
);
1509 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1510 struct shmem_inode_info
*info
, pgoff_t index
)
1512 struct vm_area_struct pvma
;
1514 struct vm_fault vmf
= {
1518 shmem_pseudo_vma_init(&pvma
, info
, index
);
1519 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1520 shmem_pseudo_vma_destroy(&pvma
);
1526 * Make sure huge_gfp is always more limited than limit_gfp.
1527 * Some of the flags set permissions, while others set limitations.
1529 static gfp_t
limit_gfp_mask(gfp_t huge_gfp
, gfp_t limit_gfp
)
1531 gfp_t allowflags
= __GFP_IO
| __GFP_FS
| __GFP_RECLAIM
;
1532 gfp_t denyflags
= __GFP_NOWARN
| __GFP_NORETRY
;
1533 gfp_t zoneflags
= limit_gfp
& GFP_ZONEMASK
;
1534 gfp_t result
= huge_gfp
& ~(allowflags
| GFP_ZONEMASK
);
1536 /* Allow allocations only from the originally specified zones. */
1537 result
|= zoneflags
;
1540 * Minimize the result gfp by taking the union with the deny flags,
1541 * and the intersection of the allow flags.
1543 result
|= (limit_gfp
& denyflags
);
1544 result
|= (huge_gfp
& limit_gfp
) & allowflags
;
1549 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1550 struct shmem_inode_info
*info
, pgoff_t index
)
1552 struct vm_area_struct pvma
;
1553 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1557 hindex
= round_down(index
, HPAGE_PMD_NR
);
1558 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1562 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1563 page
= alloc_pages_vma(gfp
, HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(),
1565 shmem_pseudo_vma_destroy(&pvma
);
1567 prep_transhuge_page(page
);
1569 count_vm_event(THP_FILE_FALLBACK
);
1573 static struct page
*shmem_alloc_page(gfp_t gfp
,
1574 struct shmem_inode_info
*info
, pgoff_t index
)
1576 struct vm_area_struct pvma
;
1579 shmem_pseudo_vma_init(&pvma
, info
, index
);
1580 page
= alloc_page_vma(gfp
, &pvma
, 0);
1581 shmem_pseudo_vma_destroy(&pvma
);
1586 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1587 struct inode
*inode
,
1588 pgoff_t index
, bool huge
)
1590 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1595 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1597 nr
= huge
? HPAGE_PMD_NR
: 1;
1599 if (!shmem_inode_acct_block(inode
, nr
))
1603 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1605 page
= shmem_alloc_page(gfp
, info
, index
);
1607 __SetPageLocked(page
);
1608 __SetPageSwapBacked(page
);
1613 shmem_inode_unacct_blocks(inode
, nr
);
1615 return ERR_PTR(err
);
1619 * When a page is moved from swapcache to shmem filecache (either by the
1620 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1621 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1622 * ignorance of the mapping it belongs to. If that mapping has special
1623 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1624 * we may need to copy to a suitable page before moving to filecache.
1626 * In a future release, this may well be extended to respect cpuset and
1627 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1628 * but for now it is a simple matter of zone.
1630 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1632 return page_zonenum(page
) > gfp_zone(gfp
);
1635 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1636 struct shmem_inode_info
*info
, pgoff_t index
)
1638 struct page
*oldpage
, *newpage
;
1639 struct folio
*old
, *new;
1640 struct address_space
*swap_mapping
;
1646 entry
.val
= page_private(oldpage
);
1647 swap_index
= swp_offset(entry
);
1648 swap_mapping
= page_mapping(oldpage
);
1651 * We have arrived here because our zones are constrained, so don't
1652 * limit chance of success by further cpuset and node constraints.
1654 gfp
&= ~GFP_CONSTRAINT_MASK
;
1655 newpage
= shmem_alloc_page(gfp
, info
, index
);
1660 copy_highpage(newpage
, oldpage
);
1661 flush_dcache_page(newpage
);
1663 __SetPageLocked(newpage
);
1664 __SetPageSwapBacked(newpage
);
1665 SetPageUptodate(newpage
);
1666 set_page_private(newpage
, entry
.val
);
1667 SetPageSwapCache(newpage
);
1670 * Our caller will very soon move newpage out of swapcache, but it's
1671 * a nice clean interface for us to replace oldpage by newpage there.
1673 xa_lock_irq(&swap_mapping
->i_pages
);
1674 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1676 old
= page_folio(oldpage
);
1677 new = page_folio(newpage
);
1678 mem_cgroup_migrate(old
, new);
1679 __inc_lruvec_page_state(newpage
, NR_FILE_PAGES
);
1680 __dec_lruvec_page_state(oldpage
, NR_FILE_PAGES
);
1682 xa_unlock_irq(&swap_mapping
->i_pages
);
1684 if (unlikely(error
)) {
1686 * Is this possible? I think not, now that our callers check
1687 * both PageSwapCache and page_private after getting page lock;
1688 * but be defensive. Reverse old to newpage for clear and free.
1692 lru_cache_add(newpage
);
1696 ClearPageSwapCache(oldpage
);
1697 set_page_private(oldpage
, 0);
1699 unlock_page(oldpage
);
1706 * Swap in the page pointed to by *pagep.
1707 * Caller has to make sure that *pagep contains a valid swapped page.
1708 * Returns 0 and the page in pagep if success. On failure, returns the
1709 * error code and NULL in *pagep.
1711 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1712 struct page
**pagep
, enum sgp_type sgp
,
1713 gfp_t gfp
, struct vm_area_struct
*vma
,
1714 vm_fault_t
*fault_type
)
1716 struct address_space
*mapping
= inode
->i_mapping
;
1717 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1718 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: NULL
;
1723 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1724 swap
= radix_to_swp_entry(*pagep
);
1727 /* Look it up and read it in.. */
1728 page
= lookup_swap_cache(swap
, NULL
, 0);
1730 /* Or update major stats only when swapin succeeds?? */
1732 *fault_type
|= VM_FAULT_MAJOR
;
1733 count_vm_event(PGMAJFAULT
);
1734 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1736 /* Here we actually start the io */
1737 page
= shmem_swapin(swap
, gfp
, info
, index
);
1744 /* We have to do this with page locked to prevent races */
1746 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1747 !shmem_confirm_swap(mapping
, index
, swap
)) {
1751 if (!PageUptodate(page
)) {
1755 wait_on_page_writeback(page
);
1758 * Some architectures may have to restore extra metadata to the
1759 * physical page after reading from swap.
1761 arch_swap_restore(swap
, page
);
1763 if (shmem_should_replace_page(page
, gfp
)) {
1764 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1769 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1770 swp_to_radix_entry(swap
), gfp
,
1775 spin_lock_irq(&info
->lock
);
1777 shmem_recalc_inode(inode
);
1778 spin_unlock_irq(&info
->lock
);
1780 if (sgp
== SGP_WRITE
)
1781 mark_page_accessed(page
);
1783 delete_from_swap_cache(page
);
1784 set_page_dirty(page
);
1790 if (!shmem_confirm_swap(mapping
, index
, swap
))
1802 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1804 * If we allocate a new one we do not mark it dirty. That's up to the
1805 * vm. If we swap it in we mark it dirty since we also free the swap
1806 * entry since a page cannot live in both the swap and page cache.
1808 * vma, vmf, and fault_type are only supplied by shmem_fault:
1809 * otherwise they are NULL.
1811 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1812 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1813 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1814 vm_fault_t
*fault_type
)
1816 struct address_space
*mapping
= inode
->i_mapping
;
1817 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1818 struct shmem_sb_info
*sbinfo
;
1819 struct mm_struct
*charge_mm
;
1821 pgoff_t hindex
= index
;
1827 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1830 if (sgp
<= SGP_CACHE
&&
1831 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1835 sbinfo
= SHMEM_SB(inode
->i_sb
);
1836 charge_mm
= vma
? vma
->vm_mm
: NULL
;
1838 page
= pagecache_get_page(mapping
, index
,
1839 FGP_ENTRY
| FGP_HEAD
| FGP_LOCK
, 0);
1841 if (page
&& vma
&& userfaultfd_minor(vma
)) {
1842 if (!xa_is_value(page
)) {
1846 *fault_type
= handle_userfault(vmf
, VM_UFFD_MINOR
);
1850 if (xa_is_value(page
)) {
1851 error
= shmem_swapin_page(inode
, index
, &page
,
1852 sgp
, gfp
, vma
, fault_type
);
1853 if (error
== -EEXIST
)
1861 hindex
= page
->index
;
1862 if (sgp
== SGP_WRITE
)
1863 mark_page_accessed(page
);
1864 if (PageUptodate(page
))
1866 /* fallocated page */
1867 if (sgp
!= SGP_READ
)
1874 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1875 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1878 if (sgp
== SGP_READ
)
1880 if (sgp
== SGP_NOALLOC
)
1884 * Fast cache lookup and swap lookup did not find it: allocate.
1887 if (vma
&& userfaultfd_missing(vma
)) {
1888 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1892 /* Never use a huge page for shmem_symlink() */
1893 if (S_ISLNK(inode
->i_mode
))
1895 if (!shmem_is_huge(vma
, inode
, index
))
1898 huge_gfp
= vma_thp_gfp_mask(vma
);
1899 huge_gfp
= limit_gfp_mask(huge_gfp
, gfp
);
1900 page
= shmem_alloc_and_acct_page(huge_gfp
, inode
, index
, true);
1903 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1909 error
= PTR_ERR(page
);
1911 if (error
!= -ENOSPC
)
1914 * Try to reclaim some space by splitting a huge page
1915 * beyond i_size on the filesystem.
1920 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1921 if (ret
== SHRINK_STOP
)
1929 if (PageTransHuge(page
))
1930 hindex
= round_down(index
, HPAGE_PMD_NR
);
1934 if (sgp
== SGP_WRITE
)
1935 __SetPageReferenced(page
);
1937 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1938 NULL
, gfp
& GFP_RECLAIM_MASK
,
1942 lru_cache_add(page
);
1944 spin_lock_irq(&info
->lock
);
1945 info
->alloced
+= compound_nr(page
);
1946 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1947 shmem_recalc_inode(inode
);
1948 spin_unlock_irq(&info
->lock
);
1951 if (PageTransHuge(page
) &&
1952 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1953 hindex
+ HPAGE_PMD_NR
- 1) {
1955 * Part of the huge page is beyond i_size: subject
1956 * to shrink under memory pressure.
1958 spin_lock(&sbinfo
->shrinklist_lock
);
1960 * _careful to defend against unlocked access to
1961 * ->shrink_list in shmem_unused_huge_shrink()
1963 if (list_empty_careful(&info
->shrinklist
)) {
1964 list_add_tail(&info
->shrinklist
,
1965 &sbinfo
->shrinklist
);
1966 sbinfo
->shrinklist_len
++;
1968 spin_unlock(&sbinfo
->shrinklist_lock
);
1972 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1974 if (sgp
== SGP_FALLOC
)
1978 * Let SGP_WRITE caller clear ends if write does not fill page;
1979 * but SGP_FALLOC on a page fallocated earlier must initialize
1980 * it now, lest undo on failure cancel our earlier guarantee.
1982 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1985 for (i
= 0; i
< compound_nr(page
); i
++) {
1986 clear_highpage(page
+ i
);
1987 flush_dcache_page(page
+ i
);
1989 SetPageUptodate(page
);
1992 /* Perhaps the file has been truncated since we checked */
1993 if (sgp
<= SGP_CACHE
&&
1994 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1996 ClearPageDirty(page
);
1997 delete_from_page_cache(page
);
1998 spin_lock_irq(&info
->lock
);
1999 shmem_recalc_inode(inode
);
2000 spin_unlock_irq(&info
->lock
);
2006 *pagep
= page
+ index
- hindex
;
2013 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
2015 if (PageTransHuge(page
)) {
2025 if (error
== -ENOSPC
&& !once
++) {
2026 spin_lock_irq(&info
->lock
);
2027 shmem_recalc_inode(inode
);
2028 spin_unlock_irq(&info
->lock
);
2031 if (error
== -EEXIST
)
2037 * This is like autoremove_wake_function, but it removes the wait queue
2038 * entry unconditionally - even if something else had already woken the
2041 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2043 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2044 list_del_init(&wait
->entry
);
2048 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2050 struct vm_area_struct
*vma
= vmf
->vma
;
2051 struct inode
*inode
= file_inode(vma
->vm_file
);
2052 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2054 vm_fault_t ret
= VM_FAULT_LOCKED
;
2057 * Trinity finds that probing a hole which tmpfs is punching can
2058 * prevent the hole-punch from ever completing: which in turn
2059 * locks writers out with its hold on i_rwsem. So refrain from
2060 * faulting pages into the hole while it's being punched. Although
2061 * shmem_undo_range() does remove the additions, it may be unable to
2062 * keep up, as each new page needs its own unmap_mapping_range() call,
2063 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2065 * It does not matter if we sometimes reach this check just before the
2066 * hole-punch begins, so that one fault then races with the punch:
2067 * we just need to make racing faults a rare case.
2069 * The implementation below would be much simpler if we just used a
2070 * standard mutex or completion: but we cannot take i_rwsem in fault,
2071 * and bloating every shmem inode for this unlikely case would be sad.
2073 if (unlikely(inode
->i_private
)) {
2074 struct shmem_falloc
*shmem_falloc
;
2076 spin_lock(&inode
->i_lock
);
2077 shmem_falloc
= inode
->i_private
;
2079 shmem_falloc
->waitq
&&
2080 vmf
->pgoff
>= shmem_falloc
->start
&&
2081 vmf
->pgoff
< shmem_falloc
->next
) {
2083 wait_queue_head_t
*shmem_falloc_waitq
;
2084 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2086 ret
= VM_FAULT_NOPAGE
;
2087 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2089 ret
= VM_FAULT_RETRY
;
2091 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2092 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2093 TASK_UNINTERRUPTIBLE
);
2094 spin_unlock(&inode
->i_lock
);
2098 * shmem_falloc_waitq points into the shmem_fallocate()
2099 * stack of the hole-punching task: shmem_falloc_waitq
2100 * is usually invalid by the time we reach here, but
2101 * finish_wait() does not dereference it in that case;
2102 * though i_lock needed lest racing with wake_up_all().
2104 spin_lock(&inode
->i_lock
);
2105 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2106 spin_unlock(&inode
->i_lock
);
2112 spin_unlock(&inode
->i_lock
);
2115 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
,
2116 gfp
, vma
, vmf
, &ret
);
2118 return vmf_error(err
);
2122 unsigned long shmem_get_unmapped_area(struct file
*file
,
2123 unsigned long uaddr
, unsigned long len
,
2124 unsigned long pgoff
, unsigned long flags
)
2126 unsigned long (*get_area
)(struct file
*,
2127 unsigned long, unsigned long, unsigned long, unsigned long);
2129 unsigned long offset
;
2130 unsigned long inflated_len
;
2131 unsigned long inflated_addr
;
2132 unsigned long inflated_offset
;
2134 if (len
> TASK_SIZE
)
2137 get_area
= current
->mm
->get_unmapped_area
;
2138 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2140 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2142 if (IS_ERR_VALUE(addr
))
2144 if (addr
& ~PAGE_MASK
)
2146 if (addr
> TASK_SIZE
- len
)
2149 if (shmem_huge
== SHMEM_HUGE_DENY
)
2151 if (len
< HPAGE_PMD_SIZE
)
2153 if (flags
& MAP_FIXED
)
2156 * Our priority is to support MAP_SHARED mapped hugely;
2157 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2158 * But if caller specified an address hint and we allocated area there
2159 * successfully, respect that as before.
2164 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2165 struct super_block
*sb
;
2168 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2169 sb
= file_inode(file
)->i_sb
;
2172 * Called directly from mm/mmap.c, or drivers/char/mem.c
2173 * for "/dev/zero", to create a shared anonymous object.
2175 if (IS_ERR(shm_mnt
))
2177 sb
= shm_mnt
->mnt_sb
;
2179 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2183 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2184 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2186 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2189 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2190 if (inflated_len
> TASK_SIZE
)
2192 if (inflated_len
< len
)
2195 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2196 if (IS_ERR_VALUE(inflated_addr
))
2198 if (inflated_addr
& ~PAGE_MASK
)
2201 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2202 inflated_addr
+= offset
- inflated_offset
;
2203 if (inflated_offset
> offset
)
2204 inflated_addr
+= HPAGE_PMD_SIZE
;
2206 if (inflated_addr
> TASK_SIZE
- len
)
2208 return inflated_addr
;
2212 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2214 struct inode
*inode
= file_inode(vma
->vm_file
);
2215 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2218 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2221 struct inode
*inode
= file_inode(vma
->vm_file
);
2224 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2225 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2229 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
2231 struct inode
*inode
= file_inode(file
);
2232 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2233 int retval
= -ENOMEM
;
2236 * What serializes the accesses to info->flags?
2237 * ipc_lock_object() when called from shmctl_do_lock(),
2238 * no serialization needed when called from shm_destroy().
2240 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2241 if (!user_shm_lock(inode
->i_size
, ucounts
))
2243 info
->flags
|= VM_LOCKED
;
2244 mapping_set_unevictable(file
->f_mapping
);
2246 if (!lock
&& (info
->flags
& VM_LOCKED
) && ucounts
) {
2247 user_shm_unlock(inode
->i_size
, ucounts
);
2248 info
->flags
&= ~VM_LOCKED
;
2249 mapping_clear_unevictable(file
->f_mapping
);
2257 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2259 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2262 ret
= seal_check_future_write(info
->seals
, vma
);
2266 /* arm64 - allow memory tagging on RAM-based files */
2267 vma
->vm_flags
|= VM_MTE_ALLOWED
;
2269 file_accessed(file
);
2270 vma
->vm_ops
= &shmem_vm_ops
;
2271 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
2272 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2273 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2274 khugepaged_enter(vma
, vma
->vm_flags
);
2279 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2280 umode_t mode
, dev_t dev
, unsigned long flags
)
2282 struct inode
*inode
;
2283 struct shmem_inode_info
*info
;
2284 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2287 if (shmem_reserve_inode(sb
, &ino
))
2290 inode
= new_inode(sb
);
2293 inode_init_owner(&init_user_ns
, inode
, dir
, mode
);
2294 inode
->i_blocks
= 0;
2295 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2296 inode
->i_generation
= prandom_u32();
2297 info
= SHMEM_I(inode
);
2298 memset(info
, 0, (char *)inode
- (char *)info
);
2299 spin_lock_init(&info
->lock
);
2300 atomic_set(&info
->stop_eviction
, 0);
2301 info
->seals
= F_SEAL_SEAL
;
2302 info
->flags
= flags
& VM_NORESERVE
;
2303 INIT_LIST_HEAD(&info
->shrinklist
);
2304 INIT_LIST_HEAD(&info
->swaplist
);
2305 simple_xattrs_init(&info
->xattrs
);
2306 cache_no_acl(inode
);
2308 switch (mode
& S_IFMT
) {
2310 inode
->i_op
= &shmem_special_inode_operations
;
2311 init_special_inode(inode
, mode
, dev
);
2314 inode
->i_mapping
->a_ops
= &shmem_aops
;
2315 inode
->i_op
= &shmem_inode_operations
;
2316 inode
->i_fop
= &shmem_file_operations
;
2317 mpol_shared_policy_init(&info
->policy
,
2318 shmem_get_sbmpol(sbinfo
));
2322 /* Some things misbehave if size == 0 on a directory */
2323 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2324 inode
->i_op
= &shmem_dir_inode_operations
;
2325 inode
->i_fop
= &simple_dir_operations
;
2329 * Must not load anything in the rbtree,
2330 * mpol_free_shared_policy will not be called.
2332 mpol_shared_policy_init(&info
->policy
, NULL
);
2336 lockdep_annotate_inode_mutex_key(inode
);
2338 shmem_free_inode(sb
);
2342 #ifdef CONFIG_USERFAULTFD
2343 int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2345 struct vm_area_struct
*dst_vma
,
2346 unsigned long dst_addr
,
2347 unsigned long src_addr
,
2349 struct page
**pagep
)
2351 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2352 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2353 struct address_space
*mapping
= inode
->i_mapping
;
2354 gfp_t gfp
= mapping_gfp_mask(mapping
);
2355 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2361 if (!shmem_inode_acct_block(inode
, 1)) {
2363 * We may have got a page, returned -ENOENT triggering a retry,
2364 * and now we find ourselves with -ENOMEM. Release the page, to
2365 * avoid a BUG_ON in our caller.
2367 if (unlikely(*pagep
)) {
2376 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2378 goto out_unacct_blocks
;
2380 if (!zeropage
) { /* COPY */
2381 page_kaddr
= kmap_atomic(page
);
2382 ret
= copy_from_user(page_kaddr
,
2383 (const void __user
*)src_addr
,
2385 kunmap_atomic(page_kaddr
);
2387 /* fallback to copy_from_user outside mmap_lock */
2388 if (unlikely(ret
)) {
2391 /* don't free the page */
2392 goto out_unacct_blocks
;
2394 } else { /* ZEROPAGE */
2395 clear_highpage(page
);
2402 VM_BUG_ON(PageLocked(page
));
2403 VM_BUG_ON(PageSwapBacked(page
));
2404 __SetPageLocked(page
);
2405 __SetPageSwapBacked(page
);
2406 __SetPageUptodate(page
);
2409 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2410 if (unlikely(pgoff
>= max_off
))
2413 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2414 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2418 ret
= mfill_atomic_install_pte(dst_mm
, dst_pmd
, dst_vma
, dst_addr
,
2421 goto out_delete_from_cache
;
2423 spin_lock_irq(&info
->lock
);
2425 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2426 shmem_recalc_inode(inode
);
2427 spin_unlock_irq(&info
->lock
);
2432 out_delete_from_cache
:
2433 delete_from_page_cache(page
);
2438 shmem_inode_unacct_blocks(inode
, 1);
2441 #endif /* CONFIG_USERFAULTFD */
2444 static const struct inode_operations shmem_symlink_inode_operations
;
2445 static const struct inode_operations shmem_short_symlink_operations
;
2447 #ifdef CONFIG_TMPFS_XATTR
2448 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2450 #define shmem_initxattrs NULL
2454 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2455 loff_t pos
, unsigned len
, unsigned flags
,
2456 struct page
**pagep
, void **fsdata
)
2458 struct inode
*inode
= mapping
->host
;
2459 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2460 pgoff_t index
= pos
>> PAGE_SHIFT
;
2462 /* i_rwsem is held by caller */
2463 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2464 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2465 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2467 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2471 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2475 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2476 loff_t pos
, unsigned len
, unsigned copied
,
2477 struct page
*page
, void *fsdata
)
2479 struct inode
*inode
= mapping
->host
;
2481 if (pos
+ copied
> inode
->i_size
)
2482 i_size_write(inode
, pos
+ copied
);
2484 if (!PageUptodate(page
)) {
2485 struct page
*head
= compound_head(page
);
2486 if (PageTransCompound(page
)) {
2489 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2490 if (head
+ i
== page
)
2492 clear_highpage(head
+ i
);
2493 flush_dcache_page(head
+ i
);
2496 if (copied
< PAGE_SIZE
) {
2497 unsigned from
= pos
& (PAGE_SIZE
- 1);
2498 zero_user_segments(page
, 0, from
,
2499 from
+ copied
, PAGE_SIZE
);
2501 SetPageUptodate(head
);
2503 set_page_dirty(page
);
2510 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2512 struct file
*file
= iocb
->ki_filp
;
2513 struct inode
*inode
= file_inode(file
);
2514 struct address_space
*mapping
= inode
->i_mapping
;
2516 unsigned long offset
;
2517 enum sgp_type sgp
= SGP_READ
;
2520 loff_t
*ppos
= &iocb
->ki_pos
;
2523 * Might this read be for a stacking filesystem? Then when reading
2524 * holes of a sparse file, we actually need to allocate those pages,
2525 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2527 if (!iter_is_iovec(to
))
2530 index
= *ppos
>> PAGE_SHIFT
;
2531 offset
= *ppos
& ~PAGE_MASK
;
2534 struct page
*page
= NULL
;
2536 unsigned long nr
, ret
;
2537 loff_t i_size
= i_size_read(inode
);
2539 end_index
= i_size
>> PAGE_SHIFT
;
2540 if (index
> end_index
)
2542 if (index
== end_index
) {
2543 nr
= i_size
& ~PAGE_MASK
;
2548 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2550 if (error
== -EINVAL
)
2555 if (sgp
== SGP_CACHE
)
2556 set_page_dirty(page
);
2561 * We must evaluate after, since reads (unlike writes)
2562 * are called without i_rwsem protection against truncate
2565 i_size
= i_size_read(inode
);
2566 end_index
= i_size
>> PAGE_SHIFT
;
2567 if (index
== end_index
) {
2568 nr
= i_size
& ~PAGE_MASK
;
2579 * If users can be writing to this page using arbitrary
2580 * virtual addresses, take care about potential aliasing
2581 * before reading the page on the kernel side.
2583 if (mapping_writably_mapped(mapping
))
2584 flush_dcache_page(page
);
2586 * Mark the page accessed if we read the beginning.
2589 mark_page_accessed(page
);
2591 page
= ZERO_PAGE(0);
2596 * Ok, we have the page, and it's up-to-date, so
2597 * now we can copy it to user space...
2599 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2602 index
+= offset
>> PAGE_SHIFT
;
2603 offset
&= ~PAGE_MASK
;
2606 if (!iov_iter_count(to
))
2615 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2616 file_accessed(file
);
2617 return retval
? retval
: error
;
2620 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2622 struct address_space
*mapping
= file
->f_mapping
;
2623 struct inode
*inode
= mapping
->host
;
2625 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2626 return generic_file_llseek_size(file
, offset
, whence
,
2627 MAX_LFS_FILESIZE
, i_size_read(inode
));
2632 /* We're holding i_rwsem so we can access i_size directly */
2633 offset
= mapping_seek_hole_data(mapping
, offset
, inode
->i_size
, whence
);
2635 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2636 inode_unlock(inode
);
2640 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2643 struct inode
*inode
= file_inode(file
);
2644 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2645 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2646 struct shmem_falloc shmem_falloc
;
2647 pgoff_t start
, index
, end
, undo_fallocend
;
2650 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2655 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2656 struct address_space
*mapping
= file
->f_mapping
;
2657 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2658 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2659 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2661 /* protected by i_rwsem */
2662 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2667 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2668 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2669 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2670 spin_lock(&inode
->i_lock
);
2671 inode
->i_private
= &shmem_falloc
;
2672 spin_unlock(&inode
->i_lock
);
2674 if ((u64
)unmap_end
> (u64
)unmap_start
)
2675 unmap_mapping_range(mapping
, unmap_start
,
2676 1 + unmap_end
- unmap_start
, 0);
2677 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2678 /* No need to unmap again: hole-punching leaves COWed pages */
2680 spin_lock(&inode
->i_lock
);
2681 inode
->i_private
= NULL
;
2682 wake_up_all(&shmem_falloc_waitq
);
2683 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2684 spin_unlock(&inode
->i_lock
);
2689 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2690 error
= inode_newsize_ok(inode
, offset
+ len
);
2694 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2699 start
= offset
>> PAGE_SHIFT
;
2700 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2701 /* Try to avoid a swapstorm if len is impossible to satisfy */
2702 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2707 shmem_falloc
.waitq
= NULL
;
2708 shmem_falloc
.start
= start
;
2709 shmem_falloc
.next
= start
;
2710 shmem_falloc
.nr_falloced
= 0;
2711 shmem_falloc
.nr_unswapped
= 0;
2712 spin_lock(&inode
->i_lock
);
2713 inode
->i_private
= &shmem_falloc
;
2714 spin_unlock(&inode
->i_lock
);
2717 * info->fallocend is only relevant when huge pages might be
2718 * involved: to prevent split_huge_page() freeing fallocated
2719 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2721 undo_fallocend
= info
->fallocend
;
2722 if (info
->fallocend
< end
)
2723 info
->fallocend
= end
;
2725 for (index
= start
; index
< end
; ) {
2729 * Good, the fallocate(2) manpage permits EINTR: we may have
2730 * been interrupted because we are using up too much memory.
2732 if (signal_pending(current
))
2734 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2737 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2739 info
->fallocend
= undo_fallocend
;
2740 /* Remove the !PageUptodate pages we added */
2741 if (index
> start
) {
2742 shmem_undo_range(inode
,
2743 (loff_t
)start
<< PAGE_SHIFT
,
2744 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2751 * Here is a more important optimization than it appears:
2752 * a second SGP_FALLOC on the same huge page will clear it,
2753 * making it PageUptodate and un-undoable if we fail later.
2755 if (PageTransCompound(page
)) {
2756 index
= round_up(index
, HPAGE_PMD_NR
);
2757 /* Beware 32-bit wraparound */
2763 * Inform shmem_writepage() how far we have reached.
2764 * No need for lock or barrier: we have the page lock.
2766 if (!PageUptodate(page
))
2767 shmem_falloc
.nr_falloced
+= index
- shmem_falloc
.next
;
2768 shmem_falloc
.next
= index
;
2771 * If !PageUptodate, leave it that way so that freeable pages
2772 * can be recognized if we need to rollback on error later.
2773 * But set_page_dirty so that memory pressure will swap rather
2774 * than free the pages we are allocating (and SGP_CACHE pages
2775 * might still be clean: we now need to mark those dirty too).
2777 set_page_dirty(page
);
2783 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2784 i_size_write(inode
, offset
+ len
);
2785 inode
->i_ctime
= current_time(inode
);
2787 spin_lock(&inode
->i_lock
);
2788 inode
->i_private
= NULL
;
2789 spin_unlock(&inode
->i_lock
);
2791 inode_unlock(inode
);
2795 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2797 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2799 buf
->f_type
= TMPFS_MAGIC
;
2800 buf
->f_bsize
= PAGE_SIZE
;
2801 buf
->f_namelen
= NAME_MAX
;
2802 if (sbinfo
->max_blocks
) {
2803 buf
->f_blocks
= sbinfo
->max_blocks
;
2805 buf
->f_bfree
= sbinfo
->max_blocks
-
2806 percpu_counter_sum(&sbinfo
->used_blocks
);
2808 if (sbinfo
->max_inodes
) {
2809 buf
->f_files
= sbinfo
->max_inodes
;
2810 buf
->f_ffree
= sbinfo
->free_inodes
;
2812 /* else leave those fields 0 like simple_statfs */
2814 buf
->f_fsid
= uuid_to_fsid(dentry
->d_sb
->s_uuid
.b
);
2820 * File creation. Allocate an inode, and we're done..
2823 shmem_mknod(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2824 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2826 struct inode
*inode
;
2827 int error
= -ENOSPC
;
2829 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2831 error
= simple_acl_create(dir
, inode
);
2834 error
= security_inode_init_security(inode
, dir
,
2836 shmem_initxattrs
, NULL
);
2837 if (error
&& error
!= -EOPNOTSUPP
)
2841 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2842 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2843 d_instantiate(dentry
, inode
);
2844 dget(dentry
); /* Extra count - pin the dentry in core */
2853 shmem_tmpfile(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2854 struct dentry
*dentry
, umode_t mode
)
2856 struct inode
*inode
;
2857 int error
= -ENOSPC
;
2859 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2861 error
= security_inode_init_security(inode
, dir
,
2863 shmem_initxattrs
, NULL
);
2864 if (error
&& error
!= -EOPNOTSUPP
)
2866 error
= simple_acl_create(dir
, inode
);
2869 d_tmpfile(dentry
, inode
);
2877 static int shmem_mkdir(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2878 struct dentry
*dentry
, umode_t mode
)
2882 if ((error
= shmem_mknod(&init_user_ns
, dir
, dentry
,
2883 mode
| S_IFDIR
, 0)))
2889 static int shmem_create(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2890 struct dentry
*dentry
, umode_t mode
, bool excl
)
2892 return shmem_mknod(&init_user_ns
, dir
, dentry
, mode
| S_IFREG
, 0);
2898 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2900 struct inode
*inode
= d_inode(old_dentry
);
2904 * No ordinary (disk based) filesystem counts links as inodes;
2905 * but each new link needs a new dentry, pinning lowmem, and
2906 * tmpfs dentries cannot be pruned until they are unlinked.
2907 * But if an O_TMPFILE file is linked into the tmpfs, the
2908 * first link must skip that, to get the accounting right.
2910 if (inode
->i_nlink
) {
2911 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
2916 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2917 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2919 ihold(inode
); /* New dentry reference */
2920 dget(dentry
); /* Extra pinning count for the created dentry */
2921 d_instantiate(dentry
, inode
);
2926 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2928 struct inode
*inode
= d_inode(dentry
);
2930 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2931 shmem_free_inode(inode
->i_sb
);
2933 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2934 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2936 dput(dentry
); /* Undo the count from "create" - this does all the work */
2940 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2942 if (!simple_empty(dentry
))
2945 drop_nlink(d_inode(dentry
));
2947 return shmem_unlink(dir
, dentry
);
2950 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2952 bool old_is_dir
= d_is_dir(old_dentry
);
2953 bool new_is_dir
= d_is_dir(new_dentry
);
2955 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2957 drop_nlink(old_dir
);
2960 drop_nlink(new_dir
);
2964 old_dir
->i_ctime
= old_dir
->i_mtime
=
2965 new_dir
->i_ctime
= new_dir
->i_mtime
=
2966 d_inode(old_dentry
)->i_ctime
=
2967 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
2972 static int shmem_whiteout(struct user_namespace
*mnt_userns
,
2973 struct inode
*old_dir
, struct dentry
*old_dentry
)
2975 struct dentry
*whiteout
;
2978 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2982 error
= shmem_mknod(&init_user_ns
, old_dir
, whiteout
,
2983 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2989 * Cheat and hash the whiteout while the old dentry is still in
2990 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2992 * d_lookup() will consistently find one of them at this point,
2993 * not sure which one, but that isn't even important.
3000 * The VFS layer already does all the dentry stuff for rename,
3001 * we just have to decrement the usage count for the target if
3002 * it exists so that the VFS layer correctly free's it when it
3005 static int shmem_rename2(struct user_namespace
*mnt_userns
,
3006 struct inode
*old_dir
, struct dentry
*old_dentry
,
3007 struct inode
*new_dir
, struct dentry
*new_dentry
,
3010 struct inode
*inode
= d_inode(old_dentry
);
3011 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3013 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3016 if (flags
& RENAME_EXCHANGE
)
3017 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3019 if (!simple_empty(new_dentry
))
3022 if (flags
& RENAME_WHITEOUT
) {
3025 error
= shmem_whiteout(&init_user_ns
, old_dir
, old_dentry
);
3030 if (d_really_is_positive(new_dentry
)) {
3031 (void) shmem_unlink(new_dir
, new_dentry
);
3032 if (they_are_dirs
) {
3033 drop_nlink(d_inode(new_dentry
));
3034 drop_nlink(old_dir
);
3036 } else if (they_are_dirs
) {
3037 drop_nlink(old_dir
);
3041 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3042 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3043 old_dir
->i_ctime
= old_dir
->i_mtime
=
3044 new_dir
->i_ctime
= new_dir
->i_mtime
=
3045 inode
->i_ctime
= current_time(old_dir
);
3049 static int shmem_symlink(struct user_namespace
*mnt_userns
, struct inode
*dir
,
3050 struct dentry
*dentry
, const char *symname
)
3054 struct inode
*inode
;
3057 len
= strlen(symname
) + 1;
3058 if (len
> PAGE_SIZE
)
3059 return -ENAMETOOLONG
;
3061 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3066 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3067 shmem_initxattrs
, NULL
);
3068 if (error
&& error
!= -EOPNOTSUPP
) {
3073 inode
->i_size
= len
-1;
3074 if (len
<= SHORT_SYMLINK_LEN
) {
3075 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3076 if (!inode
->i_link
) {
3080 inode
->i_op
= &shmem_short_symlink_operations
;
3082 inode_nohighmem(inode
);
3083 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3088 inode
->i_mapping
->a_ops
= &shmem_aops
;
3089 inode
->i_op
= &shmem_symlink_inode_operations
;
3090 memcpy(page_address(page
), symname
, len
);
3091 SetPageUptodate(page
);
3092 set_page_dirty(page
);
3096 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3097 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3098 d_instantiate(dentry
, inode
);
3103 static void shmem_put_link(void *arg
)
3105 mark_page_accessed(arg
);
3109 static const char *shmem_get_link(struct dentry
*dentry
,
3110 struct inode
*inode
,
3111 struct delayed_call
*done
)
3113 struct page
*page
= NULL
;
3116 page
= find_get_page(inode
->i_mapping
, 0);
3118 return ERR_PTR(-ECHILD
);
3119 if (!PageUptodate(page
)) {
3121 return ERR_PTR(-ECHILD
);
3124 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3126 return ERR_PTR(error
);
3129 set_delayed_call(done
, shmem_put_link
, page
);
3130 return page_address(page
);
3133 #ifdef CONFIG_TMPFS_XATTR
3135 * Superblocks without xattr inode operations may get some security.* xattr
3136 * support from the LSM "for free". As soon as we have any other xattrs
3137 * like ACLs, we also need to implement the security.* handlers at
3138 * filesystem level, though.
3142 * Callback for security_inode_init_security() for acquiring xattrs.
3144 static int shmem_initxattrs(struct inode
*inode
,
3145 const struct xattr
*xattr_array
,
3148 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3149 const struct xattr
*xattr
;
3150 struct simple_xattr
*new_xattr
;
3153 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3154 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3158 len
= strlen(xattr
->name
) + 1;
3159 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3161 if (!new_xattr
->name
) {
3166 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3167 XATTR_SECURITY_PREFIX_LEN
);
3168 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3171 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3177 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3178 struct dentry
*unused
, struct inode
*inode
,
3179 const char *name
, void *buffer
, size_t size
)
3181 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3183 name
= xattr_full_name(handler
, name
);
3184 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3187 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3188 struct user_namespace
*mnt_userns
,
3189 struct dentry
*unused
, struct inode
*inode
,
3190 const char *name
, const void *value
,
3191 size_t size
, int flags
)
3193 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3195 name
= xattr_full_name(handler
, name
);
3196 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3199 static const struct xattr_handler shmem_security_xattr_handler
= {
3200 .prefix
= XATTR_SECURITY_PREFIX
,
3201 .get
= shmem_xattr_handler_get
,
3202 .set
= shmem_xattr_handler_set
,
3205 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3206 .prefix
= XATTR_TRUSTED_PREFIX
,
3207 .get
= shmem_xattr_handler_get
,
3208 .set
= shmem_xattr_handler_set
,
3211 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3212 #ifdef CONFIG_TMPFS_POSIX_ACL
3213 &posix_acl_access_xattr_handler
,
3214 &posix_acl_default_xattr_handler
,
3216 &shmem_security_xattr_handler
,
3217 &shmem_trusted_xattr_handler
,
3221 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3223 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3224 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3226 #endif /* CONFIG_TMPFS_XATTR */
3228 static const struct inode_operations shmem_short_symlink_operations
= {
3229 .get_link
= simple_get_link
,
3230 #ifdef CONFIG_TMPFS_XATTR
3231 .listxattr
= shmem_listxattr
,
3235 static const struct inode_operations shmem_symlink_inode_operations
= {
3236 .get_link
= shmem_get_link
,
3237 #ifdef CONFIG_TMPFS_XATTR
3238 .listxattr
= shmem_listxattr
,
3242 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3244 return ERR_PTR(-ESTALE
);
3247 static int shmem_match(struct inode
*ino
, void *vfh
)
3251 inum
= (inum
<< 32) | fh
[1];
3252 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3255 /* Find any alias of inode, but prefer a hashed alias */
3256 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3258 struct dentry
*alias
= d_find_alias(inode
);
3260 return alias
?: d_find_any_alias(inode
);
3264 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3265 struct fid
*fid
, int fh_len
, int fh_type
)
3267 struct inode
*inode
;
3268 struct dentry
*dentry
= NULL
;
3275 inum
= (inum
<< 32) | fid
->raw
[1];
3277 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3278 shmem_match
, fid
->raw
);
3280 dentry
= shmem_find_alias(inode
);
3287 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3288 struct inode
*parent
)
3292 return FILEID_INVALID
;
3295 if (inode_unhashed(inode
)) {
3296 /* Unfortunately insert_inode_hash is not idempotent,
3297 * so as we hash inodes here rather than at creation
3298 * time, we need a lock to ensure we only try
3301 static DEFINE_SPINLOCK(lock
);
3303 if (inode_unhashed(inode
))
3304 __insert_inode_hash(inode
,
3305 inode
->i_ino
+ inode
->i_generation
);
3309 fh
[0] = inode
->i_generation
;
3310 fh
[1] = inode
->i_ino
;
3311 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3317 static const struct export_operations shmem_export_ops
= {
3318 .get_parent
= shmem_get_parent
,
3319 .encode_fh
= shmem_encode_fh
,
3320 .fh_to_dentry
= shmem_fh_to_dentry
,
3336 static const struct constant_table shmem_param_enums_huge
[] = {
3337 {"never", SHMEM_HUGE_NEVER
},
3338 {"always", SHMEM_HUGE_ALWAYS
},
3339 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3340 {"advise", SHMEM_HUGE_ADVISE
},
3344 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3345 fsparam_u32 ("gid", Opt_gid
),
3346 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3347 fsparam_u32oct("mode", Opt_mode
),
3348 fsparam_string("mpol", Opt_mpol
),
3349 fsparam_string("nr_blocks", Opt_nr_blocks
),
3350 fsparam_string("nr_inodes", Opt_nr_inodes
),
3351 fsparam_string("size", Opt_size
),
3352 fsparam_u32 ("uid", Opt_uid
),
3353 fsparam_flag ("inode32", Opt_inode32
),
3354 fsparam_flag ("inode64", Opt_inode64
),
3358 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3360 struct shmem_options
*ctx
= fc
->fs_private
;
3361 struct fs_parse_result result
;
3362 unsigned long long size
;
3366 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3372 size
= memparse(param
->string
, &rest
);
3374 size
<<= PAGE_SHIFT
;
3375 size
*= totalram_pages();
3381 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3382 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3385 ctx
->blocks
= memparse(param
->string
, &rest
);
3388 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3391 ctx
->inodes
= memparse(param
->string
, &rest
);
3394 ctx
->seen
|= SHMEM_SEEN_INODES
;
3397 ctx
->mode
= result
.uint_32
& 07777;
3400 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3401 if (!uid_valid(ctx
->uid
))
3405 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3406 if (!gid_valid(ctx
->gid
))
3410 ctx
->huge
= result
.uint_32
;
3411 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3412 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3413 has_transparent_hugepage()))
3414 goto unsupported_parameter
;
3415 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3418 if (IS_ENABLED(CONFIG_NUMA
)) {
3419 mpol_put(ctx
->mpol
);
3421 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3425 goto unsupported_parameter
;
3427 ctx
->full_inums
= false;
3428 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3431 if (sizeof(ino_t
) < 8) {
3433 "Cannot use inode64 with <64bit inums in kernel\n");
3435 ctx
->full_inums
= true;
3436 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3441 unsupported_parameter
:
3442 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3444 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3447 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3449 char *options
= data
;
3452 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3457 while (options
!= NULL
) {
3458 char *this_char
= options
;
3461 * NUL-terminate this option: unfortunately,
3462 * mount options form a comma-separated list,
3463 * but mpol's nodelist may also contain commas.
3465 options
= strchr(options
, ',');
3466 if (options
== NULL
)
3469 if (!isdigit(*options
)) {
3475 char *value
= strchr(this_char
, '=');
3481 len
= strlen(value
);
3483 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3492 * Reconfigure a shmem filesystem.
3494 * Note that we disallow change from limited->unlimited blocks/inodes while any
3495 * are in use; but we must separately disallow unlimited->limited, because in
3496 * that case we have no record of how much is already in use.
3498 static int shmem_reconfigure(struct fs_context
*fc
)
3500 struct shmem_options
*ctx
= fc
->fs_private
;
3501 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3502 unsigned long inodes
;
3503 struct mempolicy
*mpol
= NULL
;
3506 raw_spin_lock(&sbinfo
->stat_lock
);
3507 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3508 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3509 if (!sbinfo
->max_blocks
) {
3510 err
= "Cannot retroactively limit size";
3513 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3515 err
= "Too small a size for current use";
3519 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3520 if (!sbinfo
->max_inodes
) {
3521 err
= "Cannot retroactively limit inodes";
3524 if (ctx
->inodes
< inodes
) {
3525 err
= "Too few inodes for current use";
3530 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
3531 sbinfo
->next_ino
> UINT_MAX
) {
3532 err
= "Current inum too high to switch to 32-bit inums";
3536 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3537 sbinfo
->huge
= ctx
->huge
;
3538 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
3539 sbinfo
->full_inums
= ctx
->full_inums
;
3540 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3541 sbinfo
->max_blocks
= ctx
->blocks
;
3542 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3543 sbinfo
->max_inodes
= ctx
->inodes
;
3544 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3548 * Preserve previous mempolicy unless mpol remount option was specified.
3551 mpol
= sbinfo
->mpol
;
3552 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3555 raw_spin_unlock(&sbinfo
->stat_lock
);
3559 raw_spin_unlock(&sbinfo
->stat_lock
);
3560 return invalfc(fc
, "%s", err
);
3563 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3565 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3567 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3568 seq_printf(seq
, ",size=%luk",
3569 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3570 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3571 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3572 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3573 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3574 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3575 seq_printf(seq
, ",uid=%u",
3576 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3577 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3578 seq_printf(seq
, ",gid=%u",
3579 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3582 * Showing inode{64,32} might be useful even if it's the system default,
3583 * since then people don't have to resort to checking both here and
3584 * /proc/config.gz to confirm 64-bit inums were successfully applied
3585 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3587 * We hide it when inode64 isn't the default and we are using 32-bit
3588 * inodes, since that probably just means the feature isn't even under
3593 * +-----------------+-----------------+
3594 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3595 * +------------------+-----------------+-----------------+
3596 * | full_inums=true | show | show |
3597 * | full_inums=false | show | hide |
3598 * +------------------+-----------------+-----------------+
3601 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
3602 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
3603 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3604 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3606 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3608 shmem_show_mpol(seq
, sbinfo
->mpol
);
3612 #endif /* CONFIG_TMPFS */
3614 static void shmem_put_super(struct super_block
*sb
)
3616 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3618 free_percpu(sbinfo
->ino_batch
);
3619 percpu_counter_destroy(&sbinfo
->used_blocks
);
3620 mpol_put(sbinfo
->mpol
);
3622 sb
->s_fs_info
= NULL
;
3625 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3627 struct shmem_options
*ctx
= fc
->fs_private
;
3628 struct inode
*inode
;
3629 struct shmem_sb_info
*sbinfo
;
3631 /* Round up to L1_CACHE_BYTES to resist false sharing */
3632 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3633 L1_CACHE_BYTES
), GFP_KERNEL
);
3637 sb
->s_fs_info
= sbinfo
;
3641 * Per default we only allow half of the physical ram per
3642 * tmpfs instance, limiting inodes to one per page of lowmem;
3643 * but the internal instance is left unlimited.
3645 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3646 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3647 ctx
->blocks
= shmem_default_max_blocks();
3648 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3649 ctx
->inodes
= shmem_default_max_inodes();
3650 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
3651 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
3653 sb
->s_flags
|= SB_NOUSER
;
3655 sb
->s_export_op
= &shmem_export_ops
;
3656 sb
->s_flags
|= SB_NOSEC
;
3658 sb
->s_flags
|= SB_NOUSER
;
3660 sbinfo
->max_blocks
= ctx
->blocks
;
3661 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3662 if (sb
->s_flags
& SB_KERNMOUNT
) {
3663 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
3664 if (!sbinfo
->ino_batch
)
3667 sbinfo
->uid
= ctx
->uid
;
3668 sbinfo
->gid
= ctx
->gid
;
3669 sbinfo
->full_inums
= ctx
->full_inums
;
3670 sbinfo
->mode
= ctx
->mode
;
3671 sbinfo
->huge
= ctx
->huge
;
3672 sbinfo
->mpol
= ctx
->mpol
;
3675 raw_spin_lock_init(&sbinfo
->stat_lock
);
3676 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3678 spin_lock_init(&sbinfo
->shrinklist_lock
);
3679 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3681 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3682 sb
->s_blocksize
= PAGE_SIZE
;
3683 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3684 sb
->s_magic
= TMPFS_MAGIC
;
3685 sb
->s_op
= &shmem_ops
;
3686 sb
->s_time_gran
= 1;
3687 #ifdef CONFIG_TMPFS_XATTR
3688 sb
->s_xattr
= shmem_xattr_handlers
;
3690 #ifdef CONFIG_TMPFS_POSIX_ACL
3691 sb
->s_flags
|= SB_POSIXACL
;
3693 uuid_gen(&sb
->s_uuid
);
3695 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3698 inode
->i_uid
= sbinfo
->uid
;
3699 inode
->i_gid
= sbinfo
->gid
;
3700 sb
->s_root
= d_make_root(inode
);
3706 shmem_put_super(sb
);
3710 static int shmem_get_tree(struct fs_context
*fc
)
3712 return get_tree_nodev(fc
, shmem_fill_super
);
3715 static void shmem_free_fc(struct fs_context
*fc
)
3717 struct shmem_options
*ctx
= fc
->fs_private
;
3720 mpol_put(ctx
->mpol
);
3725 static const struct fs_context_operations shmem_fs_context_ops
= {
3726 .free
= shmem_free_fc
,
3727 .get_tree
= shmem_get_tree
,
3729 .parse_monolithic
= shmem_parse_options
,
3730 .parse_param
= shmem_parse_one
,
3731 .reconfigure
= shmem_reconfigure
,
3735 static struct kmem_cache
*shmem_inode_cachep
;
3737 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3739 struct shmem_inode_info
*info
;
3740 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3743 return &info
->vfs_inode
;
3746 static void shmem_free_in_core_inode(struct inode
*inode
)
3748 if (S_ISLNK(inode
->i_mode
))
3749 kfree(inode
->i_link
);
3750 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3753 static void shmem_destroy_inode(struct inode
*inode
)
3755 if (S_ISREG(inode
->i_mode
))
3756 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3759 static void shmem_init_inode(void *foo
)
3761 struct shmem_inode_info
*info
= foo
;
3762 inode_init_once(&info
->vfs_inode
);
3765 static void shmem_init_inodecache(void)
3767 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3768 sizeof(struct shmem_inode_info
),
3769 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3772 static void shmem_destroy_inodecache(void)
3774 kmem_cache_destroy(shmem_inode_cachep
);
3777 const struct address_space_operations shmem_aops
= {
3778 .writepage
= shmem_writepage
,
3779 .set_page_dirty
= __set_page_dirty_no_writeback
,
3781 .write_begin
= shmem_write_begin
,
3782 .write_end
= shmem_write_end
,
3784 #ifdef CONFIG_MIGRATION
3785 .migratepage
= migrate_page
,
3787 .error_remove_page
= generic_error_remove_page
,
3789 EXPORT_SYMBOL(shmem_aops
);
3791 static const struct file_operations shmem_file_operations
= {
3793 .get_unmapped_area
= shmem_get_unmapped_area
,
3795 .llseek
= shmem_file_llseek
,
3796 .read_iter
= shmem_file_read_iter
,
3797 .write_iter
= generic_file_write_iter
,
3798 .fsync
= noop_fsync
,
3799 .splice_read
= generic_file_splice_read
,
3800 .splice_write
= iter_file_splice_write
,
3801 .fallocate
= shmem_fallocate
,
3805 static const struct inode_operations shmem_inode_operations
= {
3806 .getattr
= shmem_getattr
,
3807 .setattr
= shmem_setattr
,
3808 #ifdef CONFIG_TMPFS_XATTR
3809 .listxattr
= shmem_listxattr
,
3810 .set_acl
= simple_set_acl
,
3814 static const struct inode_operations shmem_dir_inode_operations
= {
3816 .create
= shmem_create
,
3817 .lookup
= simple_lookup
,
3819 .unlink
= shmem_unlink
,
3820 .symlink
= shmem_symlink
,
3821 .mkdir
= shmem_mkdir
,
3822 .rmdir
= shmem_rmdir
,
3823 .mknod
= shmem_mknod
,
3824 .rename
= shmem_rename2
,
3825 .tmpfile
= shmem_tmpfile
,
3827 #ifdef CONFIG_TMPFS_XATTR
3828 .listxattr
= shmem_listxattr
,
3830 #ifdef CONFIG_TMPFS_POSIX_ACL
3831 .setattr
= shmem_setattr
,
3832 .set_acl
= simple_set_acl
,
3836 static const struct inode_operations shmem_special_inode_operations
= {
3837 #ifdef CONFIG_TMPFS_XATTR
3838 .listxattr
= shmem_listxattr
,
3840 #ifdef CONFIG_TMPFS_POSIX_ACL
3841 .setattr
= shmem_setattr
,
3842 .set_acl
= simple_set_acl
,
3846 static const struct super_operations shmem_ops
= {
3847 .alloc_inode
= shmem_alloc_inode
,
3848 .free_inode
= shmem_free_in_core_inode
,
3849 .destroy_inode
= shmem_destroy_inode
,
3851 .statfs
= shmem_statfs
,
3852 .show_options
= shmem_show_options
,
3854 .evict_inode
= shmem_evict_inode
,
3855 .drop_inode
= generic_delete_inode
,
3856 .put_super
= shmem_put_super
,
3857 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3858 .nr_cached_objects
= shmem_unused_huge_count
,
3859 .free_cached_objects
= shmem_unused_huge_scan
,
3863 static const struct vm_operations_struct shmem_vm_ops
= {
3864 .fault
= shmem_fault
,
3865 .map_pages
= filemap_map_pages
,
3867 .set_policy
= shmem_set_policy
,
3868 .get_policy
= shmem_get_policy
,
3872 int shmem_init_fs_context(struct fs_context
*fc
)
3874 struct shmem_options
*ctx
;
3876 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3880 ctx
->mode
= 0777 | S_ISVTX
;
3881 ctx
->uid
= current_fsuid();
3882 ctx
->gid
= current_fsgid();
3884 fc
->fs_private
= ctx
;
3885 fc
->ops
= &shmem_fs_context_ops
;
3889 static struct file_system_type shmem_fs_type
= {
3890 .owner
= THIS_MODULE
,
3892 .init_fs_context
= shmem_init_fs_context
,
3894 .parameters
= shmem_fs_parameters
,
3896 .kill_sb
= kill_litter_super
,
3897 .fs_flags
= FS_USERNS_MOUNT
| FS_THP_SUPPORT
,
3900 int __init
shmem_init(void)
3904 shmem_init_inodecache();
3906 error
= register_filesystem(&shmem_fs_type
);
3908 pr_err("Could not register tmpfs\n");
3912 shm_mnt
= kern_mount(&shmem_fs_type
);
3913 if (IS_ERR(shm_mnt
)) {
3914 error
= PTR_ERR(shm_mnt
);
3915 pr_err("Could not kern_mount tmpfs\n");
3919 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3920 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3921 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3923 shmem_huge
= SHMEM_HUGE_NEVER
; /* just in case it was patched */
3928 unregister_filesystem(&shmem_fs_type
);
3930 shmem_destroy_inodecache();
3931 shm_mnt
= ERR_PTR(error
);
3935 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3936 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3937 struct kobj_attribute
*attr
, char *buf
)
3939 static const int values
[] = {
3941 SHMEM_HUGE_WITHIN_SIZE
,
3950 for (i
= 0; i
< ARRAY_SIZE(values
); i
++) {
3951 len
+= sysfs_emit_at(buf
, len
,
3952 shmem_huge
== values
[i
] ? "%s[%s]" : "%s%s",
3954 shmem_format_huge(values
[i
]));
3957 len
+= sysfs_emit_at(buf
, len
, "\n");
3962 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3963 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3968 if (count
+ 1 > sizeof(tmp
))
3970 memcpy(tmp
, buf
, count
);
3972 if (count
&& tmp
[count
- 1] == '\n')
3973 tmp
[count
- 1] = '\0';
3975 huge
= shmem_parse_huge(tmp
);
3976 if (huge
== -EINVAL
)
3978 if (!has_transparent_hugepage() &&
3979 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3983 if (shmem_huge
> SHMEM_HUGE_DENY
)
3984 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3988 struct kobj_attribute shmem_enabled_attr
=
3989 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3990 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3992 #else /* !CONFIG_SHMEM */
3995 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3997 * This is intended for small system where the benefits of the full
3998 * shmem code (swap-backed and resource-limited) are outweighed by
3999 * their complexity. On systems without swap this code should be
4000 * effectively equivalent, but much lighter weight.
4003 static struct file_system_type shmem_fs_type
= {
4005 .init_fs_context
= ramfs_init_fs_context
,
4006 .parameters
= ramfs_fs_parameters
,
4007 .kill_sb
= kill_litter_super
,
4008 .fs_flags
= FS_USERNS_MOUNT
,
4011 int __init
shmem_init(void)
4013 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4015 shm_mnt
= kern_mount(&shmem_fs_type
);
4016 BUG_ON(IS_ERR(shm_mnt
));
4021 int shmem_unuse(unsigned int type
, bool frontswap
,
4022 unsigned long *fs_pages_to_unuse
)
4027 int shmem_lock(struct file
*file
, int lock
, struct ucounts
*ucounts
)
4032 void shmem_unlock_mapping(struct address_space
*mapping
)
4037 unsigned long shmem_get_unmapped_area(struct file
*file
,
4038 unsigned long addr
, unsigned long len
,
4039 unsigned long pgoff
, unsigned long flags
)
4041 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4045 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4047 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4049 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4051 #define shmem_vm_ops generic_file_vm_ops
4052 #define shmem_file_operations ramfs_file_operations
4053 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4054 #define shmem_acct_size(flags, size) 0
4055 #define shmem_unacct_size(flags, size) do {} while (0)
4057 #endif /* CONFIG_SHMEM */
4061 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4062 unsigned long flags
, unsigned int i_flags
)
4064 struct inode
*inode
;
4068 return ERR_CAST(mnt
);
4070 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4071 return ERR_PTR(-EINVAL
);
4073 if (shmem_acct_size(flags
, size
))
4074 return ERR_PTR(-ENOMEM
);
4076 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4078 if (unlikely(!inode
)) {
4079 shmem_unacct_size(flags
, size
);
4080 return ERR_PTR(-ENOSPC
);
4082 inode
->i_flags
|= i_flags
;
4083 inode
->i_size
= size
;
4084 clear_nlink(inode
); /* It is unlinked */
4085 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4087 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4088 &shmem_file_operations
);
4095 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4096 * kernel internal. There will be NO LSM permission checks against the
4097 * underlying inode. So users of this interface must do LSM checks at a
4098 * higher layer. The users are the big_key and shm implementations. LSM
4099 * checks are provided at the key or shm level rather than the inode.
4100 * @name: name for dentry (to be seen in /proc/<pid>/maps
4101 * @size: size to be set for the file
4102 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4104 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4106 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4110 * shmem_file_setup - get an unlinked file living in tmpfs
4111 * @name: name for dentry (to be seen in /proc/<pid>/maps
4112 * @size: size to be set for the file
4113 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4115 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4117 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4119 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4122 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4123 * @mnt: the tmpfs mount where the file will be created
4124 * @name: name for dentry (to be seen in /proc/<pid>/maps
4125 * @size: size to be set for the file
4126 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4128 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4129 loff_t size
, unsigned long flags
)
4131 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4133 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4136 * shmem_zero_setup - setup a shared anonymous mapping
4137 * @vma: the vma to be mmapped is prepared by do_mmap
4139 int shmem_zero_setup(struct vm_area_struct
*vma
)
4142 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4145 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4146 * between XFS directory reading and selinux: since this file is only
4147 * accessible to the user through its mapping, use S_PRIVATE flag to
4148 * bypass file security, in the same way as shmem_kernel_file_setup().
4150 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4152 return PTR_ERR(file
);
4156 vma
->vm_file
= file
;
4157 vma
->vm_ops
= &shmem_vm_ops
;
4159 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
4160 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4161 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4162 khugepaged_enter(vma
, vma
->vm_flags
);
4169 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4170 * @mapping: the page's address_space
4171 * @index: the page index
4172 * @gfp: the page allocator flags to use if allocating
4174 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4175 * with any new page allocations done using the specified allocation flags.
4176 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4177 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4178 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4180 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4181 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4183 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4184 pgoff_t index
, gfp_t gfp
)
4187 struct inode
*inode
= mapping
->host
;
4191 BUG_ON(!shmem_mapping(mapping
));
4192 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4193 gfp
, NULL
, NULL
, NULL
);
4195 page
= ERR_PTR(error
);
4201 * The tiny !SHMEM case uses ramfs without swap
4203 return read_cache_page_gfp(mapping
, index
, gfp
);
4206 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);