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>
42 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
44 static struct vfsmount
*shm_mnt
;
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/blkdev.h>
64 #include <linux/pagevec.h>
65 #include <linux/percpu_counter.h>
66 #include <linux/falloc.h>
67 #include <linux/splice.h>
68 #include <linux/security.h>
69 #include <linux/swapops.h>
70 #include <linux/mempolicy.h>
71 #include <linux/namei.h>
72 #include <linux/ctype.h>
73 #include <linux/migrate.h>
74 #include <linux/highmem.h>
75 #include <linux/seq_file.h>
76 #include <linux/magic.h>
77 #include <linux/syscalls.h>
78 #include <linux/fcntl.h>
79 #include <uapi/linux/memfd.h>
80 #include <linux/userfaultfd_k.h>
81 #include <linux/rmap.h>
82 #include <linux/uuid.h>
84 #include <linux/uaccess.h>
88 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
91 /* Pretend that each entry is of this size in directory's i_size */
92 #define BOGO_DIRENT_SIZE 20
94 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
95 #define SHORT_SYMLINK_LEN 128
98 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
99 * inode->i_private (with i_mutex making sure that it has only one user at
100 * a time): we would prefer not to enlarge the shmem inode just for that.
102 struct shmem_falloc
{
103 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
104 pgoff_t start
; /* start of range currently being fallocated */
105 pgoff_t next
; /* the next page offset to be fallocated */
106 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
107 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
110 struct shmem_options
{
111 unsigned long long blocks
;
112 unsigned long long inodes
;
113 struct mempolicy
*mpol
;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
123 #define SHMEM_SEEN_INUMS 8
127 static unsigned long shmem_default_max_blocks(void)
129 return totalram_pages() / 2;
132 static unsigned long shmem_default_max_inodes(void)
134 unsigned long nr_pages
= totalram_pages();
136 return min(nr_pages
- totalhigh_pages(), nr_pages
/ 2);
140 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
141 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
142 struct shmem_inode_info
*info
, pgoff_t index
);
143 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
144 struct page
**pagep
, enum sgp_type sgp
,
145 gfp_t gfp
, struct vm_area_struct
*vma
,
146 vm_fault_t
*fault_type
);
147 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
148 struct page
**pagep
, enum sgp_type sgp
,
149 gfp_t gfp
, struct vm_area_struct
*vma
,
150 struct vm_fault
*vmf
, vm_fault_t
*fault_type
);
152 int shmem_getpage(struct inode
*inode
, pgoff_t index
,
153 struct page
**pagep
, enum sgp_type sgp
)
155 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
156 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
, NULL
);
159 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
161 return sb
->s_fs_info
;
165 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
166 * for shared memory and for shared anonymous (/dev/zero) mappings
167 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
168 * consistent with the pre-accounting of private mappings ...
170 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
172 return (flags
& VM_NORESERVE
) ?
173 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
176 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
178 if (!(flags
& VM_NORESERVE
))
179 vm_unacct_memory(VM_ACCT(size
));
182 static inline int shmem_reacct_size(unsigned long flags
,
183 loff_t oldsize
, loff_t newsize
)
185 if (!(flags
& VM_NORESERVE
)) {
186 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
187 return security_vm_enough_memory_mm(current
->mm
,
188 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
189 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
190 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
196 * ... whereas tmpfs objects are accounted incrementally as
197 * pages are allocated, in order to allow large sparse files.
198 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
199 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201 static inline int shmem_acct_block(unsigned long flags
, long pages
)
203 if (!(flags
& VM_NORESERVE
))
206 return security_vm_enough_memory_mm(current
->mm
,
207 pages
* VM_ACCT(PAGE_SIZE
));
210 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
212 if (flags
& VM_NORESERVE
)
213 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
216 static inline bool shmem_inode_acct_block(struct inode
*inode
, long pages
)
218 struct shmem_inode_info
*info
= SHMEM_I(inode
);
219 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
221 if (shmem_acct_block(info
->flags
, pages
))
224 if (sbinfo
->max_blocks
) {
225 if (percpu_counter_compare(&sbinfo
->used_blocks
,
226 sbinfo
->max_blocks
- pages
) > 0)
228 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
234 shmem_unacct_blocks(info
->flags
, pages
);
238 static inline void shmem_inode_unacct_blocks(struct inode
*inode
, long pages
)
240 struct shmem_inode_info
*info
= SHMEM_I(inode
);
241 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
243 if (sbinfo
->max_blocks
)
244 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
245 shmem_unacct_blocks(info
->flags
, pages
);
248 static const struct super_operations shmem_ops
;
249 const struct address_space_operations shmem_aops
;
250 static const struct file_operations shmem_file_operations
;
251 static const struct inode_operations shmem_inode_operations
;
252 static const struct inode_operations shmem_dir_inode_operations
;
253 static const struct inode_operations shmem_special_inode_operations
;
254 static const struct vm_operations_struct shmem_vm_ops
;
255 static struct file_system_type shmem_fs_type
;
257 bool vma_is_shmem(struct vm_area_struct
*vma
)
259 return vma
->vm_ops
== &shmem_vm_ops
;
262 static LIST_HEAD(shmem_swaplist
);
263 static DEFINE_MUTEX(shmem_swaplist_mutex
);
266 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
267 * produces a novel ino for the newly allocated inode.
269 * It may also be called when making a hard link to permit the space needed by
270 * each dentry. However, in that case, no new inode number is needed since that
271 * internally draws from another pool of inode numbers (currently global
272 * get_next_ino()). This case is indicated by passing NULL as inop.
274 #define SHMEM_INO_BATCH 1024
275 static int shmem_reserve_inode(struct super_block
*sb
, ino_t
*inop
)
277 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
280 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
281 spin_lock(&sbinfo
->stat_lock
);
282 if (sbinfo
->max_inodes
) {
283 if (!sbinfo
->free_inodes
) {
284 spin_unlock(&sbinfo
->stat_lock
);
287 sbinfo
->free_inodes
--;
290 ino
= sbinfo
->next_ino
++;
291 if (unlikely(is_zero_ino(ino
)))
292 ino
= sbinfo
->next_ino
++;
293 if (unlikely(!sbinfo
->full_inums
&&
296 * Emulate get_next_ino uint wraparound for
299 if (IS_ENABLED(CONFIG_64BIT
))
300 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
301 __func__
, MINOR(sb
->s_dev
));
302 sbinfo
->next_ino
= 1;
303 ino
= sbinfo
->next_ino
++;
307 spin_unlock(&sbinfo
->stat_lock
);
310 * __shmem_file_setup, one of our callers, is lock-free: it
311 * doesn't hold stat_lock in shmem_reserve_inode since
312 * max_inodes is always 0, and is called from potentially
313 * unknown contexts. As such, use a per-cpu batched allocator
314 * which doesn't require the per-sb stat_lock unless we are at
315 * the batch boundary.
317 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
318 * shmem mounts are not exposed to userspace, so we don't need
319 * to worry about things like glibc compatibility.
322 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
324 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
325 spin_lock(&sbinfo
->stat_lock
);
326 ino
= sbinfo
->next_ino
;
327 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
328 spin_unlock(&sbinfo
->stat_lock
);
329 if (unlikely(is_zero_ino(ino
)))
340 static void shmem_free_inode(struct super_block
*sb
)
342 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
343 if (sbinfo
->max_inodes
) {
344 spin_lock(&sbinfo
->stat_lock
);
345 sbinfo
->free_inodes
++;
346 spin_unlock(&sbinfo
->stat_lock
);
351 * shmem_recalc_inode - recalculate the block usage of an inode
352 * @inode: inode to recalc
354 * We have to calculate the free blocks since the mm can drop
355 * undirtied hole pages behind our back.
357 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
358 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
360 * It has to be called with the spinlock held.
362 static void shmem_recalc_inode(struct inode
*inode
)
364 struct shmem_inode_info
*info
= SHMEM_I(inode
);
367 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
369 info
->alloced
-= freed
;
370 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
371 shmem_inode_unacct_blocks(inode
, freed
);
375 bool shmem_charge(struct inode
*inode
, long pages
)
377 struct shmem_inode_info
*info
= SHMEM_I(inode
);
380 if (!shmem_inode_acct_block(inode
, pages
))
383 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
384 inode
->i_mapping
->nrpages
+= pages
;
386 spin_lock_irqsave(&info
->lock
, flags
);
387 info
->alloced
+= pages
;
388 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
389 shmem_recalc_inode(inode
);
390 spin_unlock_irqrestore(&info
->lock
, flags
);
395 void shmem_uncharge(struct inode
*inode
, long pages
)
397 struct shmem_inode_info
*info
= SHMEM_I(inode
);
400 /* nrpages adjustment done by __delete_from_page_cache() or caller */
402 spin_lock_irqsave(&info
->lock
, flags
);
403 info
->alloced
-= pages
;
404 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
405 shmem_recalc_inode(inode
);
406 spin_unlock_irqrestore(&info
->lock
, flags
);
408 shmem_inode_unacct_blocks(inode
, pages
);
412 * Replace item expected in xarray by a new item, while holding xa_lock.
414 static int shmem_replace_entry(struct address_space
*mapping
,
415 pgoff_t index
, void *expected
, void *replacement
)
417 XA_STATE(xas
, &mapping
->i_pages
, index
);
420 VM_BUG_ON(!expected
);
421 VM_BUG_ON(!replacement
);
422 item
= xas_load(&xas
);
423 if (item
!= expected
)
425 xas_store(&xas
, replacement
);
430 * Sometimes, before we decide whether to proceed or to fail, we must check
431 * that an entry was not already brought back from swap by a racing thread.
433 * Checking page is not enough: by the time a SwapCache page is locked, it
434 * might be reused, and again be SwapCache, using the same swap as before.
436 static bool shmem_confirm_swap(struct address_space
*mapping
,
437 pgoff_t index
, swp_entry_t swap
)
439 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
443 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
446 * disables huge pages for the mount;
448 * enables huge pages for the mount;
449 * SHMEM_HUGE_WITHIN_SIZE:
450 * only allocate huge pages if the page will be fully within i_size,
451 * also respect fadvise()/madvise() hints;
453 * only allocate huge pages if requested with fadvise()/madvise();
456 #define SHMEM_HUGE_NEVER 0
457 #define SHMEM_HUGE_ALWAYS 1
458 #define SHMEM_HUGE_WITHIN_SIZE 2
459 #define SHMEM_HUGE_ADVISE 3
463 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
466 * disables huge on shm_mnt and all mounts, for emergency use;
468 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
471 #define SHMEM_HUGE_DENY (-1)
472 #define SHMEM_HUGE_FORCE (-2)
474 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
475 /* ifdef here to avoid bloating shmem.o when not necessary */
477 static int shmem_huge __read_mostly
;
479 #if defined(CONFIG_SYSFS)
480 static int shmem_parse_huge(const char *str
)
482 if (!strcmp(str
, "never"))
483 return SHMEM_HUGE_NEVER
;
484 if (!strcmp(str
, "always"))
485 return SHMEM_HUGE_ALWAYS
;
486 if (!strcmp(str
, "within_size"))
487 return SHMEM_HUGE_WITHIN_SIZE
;
488 if (!strcmp(str
, "advise"))
489 return SHMEM_HUGE_ADVISE
;
490 if (!strcmp(str
, "deny"))
491 return SHMEM_HUGE_DENY
;
492 if (!strcmp(str
, "force"))
493 return SHMEM_HUGE_FORCE
;
498 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
499 static const char *shmem_format_huge(int huge
)
502 case SHMEM_HUGE_NEVER
:
504 case SHMEM_HUGE_ALWAYS
:
506 case SHMEM_HUGE_WITHIN_SIZE
:
507 return "within_size";
508 case SHMEM_HUGE_ADVISE
:
510 case SHMEM_HUGE_DENY
:
512 case SHMEM_HUGE_FORCE
:
521 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
522 struct shrink_control
*sc
, unsigned long nr_to_split
)
524 LIST_HEAD(list
), *pos
, *next
;
525 LIST_HEAD(to_remove
);
527 struct shmem_inode_info
*info
;
529 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
530 int removed
= 0, split
= 0;
532 if (list_empty(&sbinfo
->shrinklist
))
535 spin_lock(&sbinfo
->shrinklist_lock
);
536 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
537 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
540 inode
= igrab(&info
->vfs_inode
);
542 /* inode is about to be evicted */
544 list_del_init(&info
->shrinklist
);
549 /* Check if there's anything to gain */
550 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
551 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
552 list_move(&info
->shrinklist
, &to_remove
);
557 list_move(&info
->shrinklist
, &list
);
562 spin_unlock(&sbinfo
->shrinklist_lock
);
564 list_for_each_safe(pos
, next
, &to_remove
) {
565 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
566 inode
= &info
->vfs_inode
;
567 list_del_init(&info
->shrinklist
);
571 list_for_each_safe(pos
, next
, &list
) {
574 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
575 inode
= &info
->vfs_inode
;
577 if (nr_to_split
&& split
>= nr_to_split
)
580 page
= find_get_page(inode
->i_mapping
,
581 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
585 /* No huge page at the end of the file: nothing to split */
586 if (!PageTransHuge(page
)) {
592 * Leave the inode on the list if we failed to lock
593 * the page at this time.
595 * Waiting for the lock may lead to deadlock in the
598 if (!trylock_page(page
)) {
603 ret
= split_huge_page(page
);
607 /* If split failed leave the inode on the list */
613 list_del_init(&info
->shrinklist
);
619 spin_lock(&sbinfo
->shrinklist_lock
);
620 list_splice_tail(&list
, &sbinfo
->shrinklist
);
621 sbinfo
->shrinklist_len
-= removed
;
622 spin_unlock(&sbinfo
->shrinklist_lock
);
627 static long shmem_unused_huge_scan(struct super_block
*sb
,
628 struct shrink_control
*sc
)
630 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
632 if (!READ_ONCE(sbinfo
->shrinklist_len
))
635 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
638 static long shmem_unused_huge_count(struct super_block
*sb
,
639 struct shrink_control
*sc
)
641 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
642 return READ_ONCE(sbinfo
->shrinklist_len
);
644 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
646 #define shmem_huge SHMEM_HUGE_DENY
648 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
649 struct shrink_control
*sc
, unsigned long nr_to_split
)
653 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
655 static inline bool is_huge_enabled(struct shmem_sb_info
*sbinfo
)
657 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
658 (shmem_huge
== SHMEM_HUGE_FORCE
|| sbinfo
->huge
) &&
659 shmem_huge
!= SHMEM_HUGE_DENY
)
665 * Like add_to_page_cache_locked, but error if expected item has gone.
667 static int shmem_add_to_page_cache(struct page
*page
,
668 struct address_space
*mapping
,
669 pgoff_t index
, void *expected
, gfp_t gfp
,
670 struct mm_struct
*charge_mm
)
672 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, compound_order(page
));
674 unsigned long nr
= compound_nr(page
);
677 VM_BUG_ON_PAGE(PageTail(page
), page
);
678 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
679 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
680 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
681 VM_BUG_ON(expected
&& PageTransHuge(page
));
683 page_ref_add(page
, nr
);
684 page
->mapping
= mapping
;
687 if (!PageSwapCache(page
)) {
688 error
= mem_cgroup_charge(page
, charge_mm
, gfp
);
690 if (PageTransHuge(page
)) {
691 count_vm_event(THP_FILE_FALLBACK
);
692 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
697 cgroup_throttle_swaprate(page
, gfp
);
702 entry
= xas_find_conflict(&xas
);
703 if (entry
!= expected
)
704 xas_set_err(&xas
, -EEXIST
);
705 xas_create_range(&xas
);
709 xas_store(&xas
, page
);
714 if (PageTransHuge(page
)) {
715 count_vm_event(THP_FILE_ALLOC
);
716 __mod_lruvec_page_state(page
, NR_SHMEM_THPS
, nr
);
718 mapping
->nrpages
+= nr
;
719 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, nr
);
720 __mod_lruvec_page_state(page
, NR_SHMEM
, nr
);
722 xas_unlock_irq(&xas
);
723 } while (xas_nomem(&xas
, gfp
));
725 if (xas_error(&xas
)) {
726 error
= xas_error(&xas
);
732 page
->mapping
= NULL
;
733 page_ref_sub(page
, nr
);
738 * Like delete_from_page_cache, but substitutes swap for page.
740 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
742 struct address_space
*mapping
= page
->mapping
;
745 VM_BUG_ON_PAGE(PageCompound(page
), page
);
747 xa_lock_irq(&mapping
->i_pages
);
748 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
749 page
->mapping
= NULL
;
751 __dec_lruvec_page_state(page
, NR_FILE_PAGES
);
752 __dec_lruvec_page_state(page
, NR_SHMEM
);
753 xa_unlock_irq(&mapping
->i_pages
);
759 * Remove swap entry from page cache, free the swap and its page cache.
761 static int shmem_free_swap(struct address_space
*mapping
,
762 pgoff_t index
, void *radswap
)
766 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
769 free_swap_and_cache(radix_to_swp_entry(radswap
));
774 * Determine (in bytes) how many of the shmem object's pages mapped by the
775 * given offsets are swapped out.
777 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
778 * as long as the inode doesn't go away and racy results are not a problem.
780 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
781 pgoff_t start
, pgoff_t end
)
783 XA_STATE(xas
, &mapping
->i_pages
, start
);
785 unsigned long swapped
= 0;
788 xas_for_each(&xas
, page
, end
- 1) {
789 if (xas_retry(&xas
, page
))
791 if (xa_is_value(page
))
794 if (need_resched()) {
802 return swapped
<< PAGE_SHIFT
;
806 * Determine (in bytes) how many of the shmem object's pages mapped by the
807 * given vma is swapped out.
809 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
810 * as long as the inode doesn't go away and racy results are not a problem.
812 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
814 struct inode
*inode
= file_inode(vma
->vm_file
);
815 struct shmem_inode_info
*info
= SHMEM_I(inode
);
816 struct address_space
*mapping
= inode
->i_mapping
;
817 unsigned long swapped
;
819 /* Be careful as we don't hold info->lock */
820 swapped
= READ_ONCE(info
->swapped
);
823 * The easier cases are when the shmem object has nothing in swap, or
824 * the vma maps it whole. Then we can simply use the stats that we
830 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
831 return swapped
<< PAGE_SHIFT
;
833 /* Here comes the more involved part */
834 return shmem_partial_swap_usage(mapping
,
835 linear_page_index(vma
, vma
->vm_start
),
836 linear_page_index(vma
, vma
->vm_end
));
840 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
842 void shmem_unlock_mapping(struct address_space
*mapping
)
849 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
851 while (!mapping_unevictable(mapping
)) {
852 if (!pagevec_lookup(&pvec
, mapping
, &index
))
854 check_move_unevictable_pages(&pvec
);
855 pagevec_release(&pvec
);
861 * Check whether a hole-punch or truncation needs to split a huge page,
862 * returning true if no split was required, or the split has been successful.
864 * Eviction (or truncation to 0 size) should never need to split a huge page;
865 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
866 * head, and then succeeded to trylock on tail.
868 * A split can only succeed when there are no additional references on the
869 * huge page: so the split below relies upon find_get_entries() having stopped
870 * when it found a subpage of the huge page, without getting further references.
872 static bool shmem_punch_compound(struct page
*page
, pgoff_t start
, pgoff_t end
)
874 if (!PageTransCompound(page
))
877 /* Just proceed to delete a huge page wholly within the range punched */
878 if (PageHead(page
) &&
879 page
->index
>= start
&& page
->index
+ HPAGE_PMD_NR
<= end
)
882 /* Try to split huge page, so we can truly punch the hole or truncate */
883 return split_huge_page(page
) >= 0;
887 * Remove range of pages and swap entries from page cache, and free them.
888 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
890 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
893 struct address_space
*mapping
= inode
->i_mapping
;
894 struct shmem_inode_info
*info
= SHMEM_I(inode
);
895 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
896 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
897 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
898 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
900 pgoff_t indices
[PAGEVEC_SIZE
];
901 long nr_swaps_freed
= 0;
906 end
= -1; /* unsigned, so actually very big */
910 while (index
< end
&& find_lock_entries(mapping
, index
, end
- 1,
912 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
913 struct page
*page
= pvec
.pages
[i
];
917 if (xa_is_value(page
)) {
920 nr_swaps_freed
+= !shmem_free_swap(mapping
,
924 index
+= thp_nr_pages(page
) - 1;
926 if (!unfalloc
|| !PageUptodate(page
))
927 truncate_inode_page(mapping
, page
);
930 pagevec_remove_exceptionals(&pvec
);
931 pagevec_release(&pvec
);
937 struct page
*page
= NULL
;
938 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
940 unsigned int top
= PAGE_SIZE
;
945 zero_user_segment(page
, partial_start
, top
);
946 set_page_dirty(page
);
952 struct page
*page
= NULL
;
953 shmem_getpage(inode
, end
, &page
, SGP_READ
);
955 zero_user_segment(page
, 0, partial_end
);
956 set_page_dirty(page
);
965 while (index
< end
) {
968 if (!find_get_entries(mapping
, index
, end
- 1, &pvec
,
970 /* If all gone or hole-punch or unfalloc, we're done */
971 if (index
== start
|| end
!= -1)
973 /* But if truncating, restart to make sure all gone */
977 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
978 struct page
*page
= pvec
.pages
[i
];
981 if (xa_is_value(page
)) {
984 if (shmem_free_swap(mapping
, index
, page
)) {
985 /* Swap was replaced by page: retry */
995 if (!unfalloc
|| !PageUptodate(page
)) {
996 if (page_mapping(page
) != mapping
) {
997 /* Page was replaced by swap: retry */
1002 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
1003 if (shmem_punch_compound(page
, start
, end
))
1004 truncate_inode_page(mapping
, page
);
1005 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1006 /* Wipe the page and don't get stuck */
1007 clear_highpage(page
);
1008 flush_dcache_page(page
);
1009 set_page_dirty(page
);
1011 round_up(start
, HPAGE_PMD_NR
))
1017 pagevec_remove_exceptionals(&pvec
);
1018 pagevec_release(&pvec
);
1022 spin_lock_irq(&info
->lock
);
1023 info
->swapped
-= nr_swaps_freed
;
1024 shmem_recalc_inode(inode
);
1025 spin_unlock_irq(&info
->lock
);
1028 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1030 shmem_undo_range(inode
, lstart
, lend
, false);
1031 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1033 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1035 static int shmem_getattr(struct user_namespace
*mnt_userns
,
1036 const struct path
*path
, struct kstat
*stat
,
1037 u32 request_mask
, unsigned int query_flags
)
1039 struct inode
*inode
= path
->dentry
->d_inode
;
1040 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1041 struct shmem_sb_info
*sb_info
= SHMEM_SB(inode
->i_sb
);
1043 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1044 spin_lock_irq(&info
->lock
);
1045 shmem_recalc_inode(inode
);
1046 spin_unlock_irq(&info
->lock
);
1048 generic_fillattr(&init_user_ns
, inode
, stat
);
1050 if (is_huge_enabled(sb_info
))
1051 stat
->blksize
= HPAGE_PMD_SIZE
;
1056 static int shmem_setattr(struct user_namespace
*mnt_userns
,
1057 struct dentry
*dentry
, struct iattr
*attr
)
1059 struct inode
*inode
= d_inode(dentry
);
1060 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1061 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1064 error
= setattr_prepare(&init_user_ns
, dentry
, attr
);
1068 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1069 loff_t oldsize
= inode
->i_size
;
1070 loff_t newsize
= attr
->ia_size
;
1072 /* protected by i_mutex */
1073 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1074 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1077 if (newsize
!= oldsize
) {
1078 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1082 i_size_write(inode
, newsize
);
1083 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1085 if (newsize
<= oldsize
) {
1086 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1087 if (oldsize
> holebegin
)
1088 unmap_mapping_range(inode
->i_mapping
,
1091 shmem_truncate_range(inode
,
1092 newsize
, (loff_t
)-1);
1093 /* unmap again to remove racily COWed private pages */
1094 if (oldsize
> holebegin
)
1095 unmap_mapping_range(inode
->i_mapping
,
1099 * Part of the huge page can be beyond i_size: subject
1100 * to shrink under memory pressure.
1102 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1103 spin_lock(&sbinfo
->shrinklist_lock
);
1105 * _careful to defend against unlocked access to
1106 * ->shrink_list in shmem_unused_huge_shrink()
1108 if (list_empty_careful(&info
->shrinklist
)) {
1109 list_add_tail(&info
->shrinklist
,
1110 &sbinfo
->shrinklist
);
1111 sbinfo
->shrinklist_len
++;
1113 spin_unlock(&sbinfo
->shrinklist_lock
);
1118 setattr_copy(&init_user_ns
, inode
, attr
);
1119 if (attr
->ia_valid
& ATTR_MODE
)
1120 error
= posix_acl_chmod(&init_user_ns
, inode
, inode
->i_mode
);
1124 static void shmem_evict_inode(struct inode
*inode
)
1126 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1127 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1129 if (shmem_mapping(inode
->i_mapping
)) {
1130 shmem_unacct_size(info
->flags
, inode
->i_size
);
1132 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1133 if (!list_empty(&info
->shrinklist
)) {
1134 spin_lock(&sbinfo
->shrinklist_lock
);
1135 if (!list_empty(&info
->shrinklist
)) {
1136 list_del_init(&info
->shrinklist
);
1137 sbinfo
->shrinklist_len
--;
1139 spin_unlock(&sbinfo
->shrinklist_lock
);
1141 while (!list_empty(&info
->swaplist
)) {
1142 /* Wait while shmem_unuse() is scanning this inode... */
1143 wait_var_event(&info
->stop_eviction
,
1144 !atomic_read(&info
->stop_eviction
));
1145 mutex_lock(&shmem_swaplist_mutex
);
1146 /* ...but beware of the race if we peeked too early */
1147 if (!atomic_read(&info
->stop_eviction
))
1148 list_del_init(&info
->swaplist
);
1149 mutex_unlock(&shmem_swaplist_mutex
);
1153 simple_xattrs_free(&info
->xattrs
);
1154 WARN_ON(inode
->i_blocks
);
1155 shmem_free_inode(inode
->i_sb
);
1159 extern struct swap_info_struct
*swap_info
[];
1161 static int shmem_find_swap_entries(struct address_space
*mapping
,
1162 pgoff_t start
, unsigned int nr_entries
,
1163 struct page
**entries
, pgoff_t
*indices
,
1164 unsigned int type
, bool frontswap
)
1166 XA_STATE(xas
, &mapping
->i_pages
, start
);
1169 unsigned int ret
= 0;
1175 xas_for_each(&xas
, page
, ULONG_MAX
) {
1176 if (xas_retry(&xas
, page
))
1179 if (!xa_is_value(page
))
1182 entry
= radix_to_swp_entry(page
);
1183 if (swp_type(entry
) != type
)
1186 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1189 indices
[ret
] = xas
.xa_index
;
1190 entries
[ret
] = page
;
1192 if (need_resched()) {
1196 if (++ret
== nr_entries
)
1205 * Move the swapped pages for an inode to page cache. Returns the count
1206 * of pages swapped in, or the error in case of failure.
1208 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1214 struct address_space
*mapping
= inode
->i_mapping
;
1216 for (i
= 0; i
< pvec
.nr
; i
++) {
1217 struct page
*page
= pvec
.pages
[i
];
1219 if (!xa_is_value(page
))
1221 error
= shmem_swapin_page(inode
, indices
[i
],
1223 mapping_gfp_mask(mapping
),
1230 if (error
== -ENOMEM
)
1234 return error
? error
: ret
;
1238 * If swap found in inode, free it and move page from swapcache to filecache.
1240 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1241 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1243 struct address_space
*mapping
= inode
->i_mapping
;
1245 struct pagevec pvec
;
1246 pgoff_t indices
[PAGEVEC_SIZE
];
1247 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1250 pagevec_init(&pvec
);
1252 unsigned int nr_entries
= PAGEVEC_SIZE
;
1254 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1255 nr_entries
= *fs_pages_to_unuse
;
1257 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1258 pvec
.pages
, indices
,
1265 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1269 if (frontswap_partial
) {
1270 *fs_pages_to_unuse
-= ret
;
1271 if (*fs_pages_to_unuse
== 0) {
1272 ret
= FRONTSWAP_PAGES_UNUSED
;
1277 start
= indices
[pvec
.nr
- 1];
1284 * Read all the shared memory data that resides in the swap
1285 * device 'type' back into memory, so the swap device can be
1288 int shmem_unuse(unsigned int type
, bool frontswap
,
1289 unsigned long *fs_pages_to_unuse
)
1291 struct shmem_inode_info
*info
, *next
;
1294 if (list_empty(&shmem_swaplist
))
1297 mutex_lock(&shmem_swaplist_mutex
);
1298 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1299 if (!info
->swapped
) {
1300 list_del_init(&info
->swaplist
);
1304 * Drop the swaplist mutex while searching the inode for swap;
1305 * but before doing so, make sure shmem_evict_inode() will not
1306 * remove placeholder inode from swaplist, nor let it be freed
1307 * (igrab() would protect from unlink, but not from unmount).
1309 atomic_inc(&info
->stop_eviction
);
1310 mutex_unlock(&shmem_swaplist_mutex
);
1312 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1316 mutex_lock(&shmem_swaplist_mutex
);
1317 next
= list_next_entry(info
, swaplist
);
1319 list_del_init(&info
->swaplist
);
1320 if (atomic_dec_and_test(&info
->stop_eviction
))
1321 wake_up_var(&info
->stop_eviction
);
1325 mutex_unlock(&shmem_swaplist_mutex
);
1331 * Move the page from the page cache to the swap cache.
1333 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1335 struct shmem_inode_info
*info
;
1336 struct address_space
*mapping
;
1337 struct inode
*inode
;
1341 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1342 BUG_ON(!PageLocked(page
));
1343 mapping
= page
->mapping
;
1344 index
= page
->index
;
1345 inode
= mapping
->host
;
1346 info
= SHMEM_I(inode
);
1347 if (info
->flags
& VM_LOCKED
)
1349 if (!total_swap_pages
)
1353 * Our capabilities prevent regular writeback or sync from ever calling
1354 * shmem_writepage; but a stacking filesystem might use ->writepage of
1355 * its underlying filesystem, in which case tmpfs should write out to
1356 * swap only in response to memory pressure, and not for the writeback
1359 if (!wbc
->for_reclaim
) {
1360 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1365 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1366 * value into swapfile.c, the only way we can correctly account for a
1367 * fallocated page arriving here is now to initialize it and write it.
1369 * That's okay for a page already fallocated earlier, but if we have
1370 * not yet completed the fallocation, then (a) we want to keep track
1371 * of this page in case we have to undo it, and (b) it may not be a
1372 * good idea to continue anyway, once we're pushing into swap. So
1373 * reactivate the page, and let shmem_fallocate() quit when too many.
1375 if (!PageUptodate(page
)) {
1376 if (inode
->i_private
) {
1377 struct shmem_falloc
*shmem_falloc
;
1378 spin_lock(&inode
->i_lock
);
1379 shmem_falloc
= inode
->i_private
;
1381 !shmem_falloc
->waitq
&&
1382 index
>= shmem_falloc
->start
&&
1383 index
< shmem_falloc
->next
)
1384 shmem_falloc
->nr_unswapped
++;
1386 shmem_falloc
= NULL
;
1387 spin_unlock(&inode
->i_lock
);
1391 clear_highpage(page
);
1392 flush_dcache_page(page
);
1393 SetPageUptodate(page
);
1396 swap
= get_swap_page(page
);
1401 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1402 * if it's not already there. Do it now before the page is
1403 * moved to swap cache, when its pagelock no longer protects
1404 * the inode from eviction. But don't unlock the mutex until
1405 * we've incremented swapped, because shmem_unuse_inode() will
1406 * prune a !swapped inode from the swaplist under this mutex.
1408 mutex_lock(&shmem_swaplist_mutex
);
1409 if (list_empty(&info
->swaplist
))
1410 list_add(&info
->swaplist
, &shmem_swaplist
);
1412 if (add_to_swap_cache(page
, swap
,
1413 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1415 spin_lock_irq(&info
->lock
);
1416 shmem_recalc_inode(inode
);
1418 spin_unlock_irq(&info
->lock
);
1420 swap_shmem_alloc(swap
);
1421 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1423 mutex_unlock(&shmem_swaplist_mutex
);
1424 BUG_ON(page_mapped(page
));
1425 swap_writepage(page
, wbc
);
1429 mutex_unlock(&shmem_swaplist_mutex
);
1430 put_swap_page(page
, swap
);
1432 set_page_dirty(page
);
1433 if (wbc
->for_reclaim
)
1434 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1439 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1440 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1444 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1445 return; /* show nothing */
1447 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1449 seq_printf(seq
, ",mpol=%s", buffer
);
1452 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1454 struct mempolicy
*mpol
= NULL
;
1456 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1457 mpol
= sbinfo
->mpol
;
1459 spin_unlock(&sbinfo
->stat_lock
);
1463 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1464 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1467 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1471 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1473 #define vm_policy vm_private_data
1476 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1477 struct shmem_inode_info
*info
, pgoff_t index
)
1479 /* Create a pseudo vma that just contains the policy */
1480 vma_init(vma
, NULL
);
1481 /* Bias interleave by inode number to distribute better across nodes */
1482 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1483 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1486 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1488 /* Drop reference taken by mpol_shared_policy_lookup() */
1489 mpol_cond_put(vma
->vm_policy
);
1492 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1493 struct shmem_inode_info
*info
, pgoff_t index
)
1495 struct vm_area_struct pvma
;
1497 struct vm_fault vmf
= {
1501 shmem_pseudo_vma_init(&pvma
, info
, index
);
1502 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1503 shmem_pseudo_vma_destroy(&pvma
);
1509 * Make sure huge_gfp is always more limited than limit_gfp.
1510 * Some of the flags set permissions, while others set limitations.
1512 static gfp_t
limit_gfp_mask(gfp_t huge_gfp
, gfp_t limit_gfp
)
1514 gfp_t allowflags
= __GFP_IO
| __GFP_FS
| __GFP_RECLAIM
;
1515 gfp_t denyflags
= __GFP_NOWARN
| __GFP_NORETRY
;
1516 gfp_t zoneflags
= limit_gfp
& GFP_ZONEMASK
;
1517 gfp_t result
= huge_gfp
& ~(allowflags
| GFP_ZONEMASK
);
1519 /* Allow allocations only from the originally specified zones. */
1520 result
|= zoneflags
;
1523 * Minimize the result gfp by taking the union with the deny flags,
1524 * and the intersection of the allow flags.
1526 result
|= (limit_gfp
& denyflags
);
1527 result
|= (huge_gfp
& limit_gfp
) & allowflags
;
1532 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1533 struct shmem_inode_info
*info
, pgoff_t index
)
1535 struct vm_area_struct pvma
;
1536 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1540 hindex
= round_down(index
, HPAGE_PMD_NR
);
1541 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1545 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1546 page
= alloc_pages_vma(gfp
, HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(),
1548 shmem_pseudo_vma_destroy(&pvma
);
1550 prep_transhuge_page(page
);
1552 count_vm_event(THP_FILE_FALLBACK
);
1556 static struct page
*shmem_alloc_page(gfp_t gfp
,
1557 struct shmem_inode_info
*info
, pgoff_t index
)
1559 struct vm_area_struct pvma
;
1562 shmem_pseudo_vma_init(&pvma
, info
, index
);
1563 page
= alloc_page_vma(gfp
, &pvma
, 0);
1564 shmem_pseudo_vma_destroy(&pvma
);
1569 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1570 struct inode
*inode
,
1571 pgoff_t index
, bool huge
)
1573 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1578 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1580 nr
= huge
? HPAGE_PMD_NR
: 1;
1582 if (!shmem_inode_acct_block(inode
, nr
))
1586 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1588 page
= shmem_alloc_page(gfp
, info
, index
);
1590 __SetPageLocked(page
);
1591 __SetPageSwapBacked(page
);
1596 shmem_inode_unacct_blocks(inode
, nr
);
1598 return ERR_PTR(err
);
1602 * When a page is moved from swapcache to shmem filecache (either by the
1603 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1604 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1605 * ignorance of the mapping it belongs to. If that mapping has special
1606 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1607 * we may need to copy to a suitable page before moving to filecache.
1609 * In a future release, this may well be extended to respect cpuset and
1610 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1611 * but for now it is a simple matter of zone.
1613 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1615 return page_zonenum(page
) > gfp_zone(gfp
);
1618 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1619 struct shmem_inode_info
*info
, pgoff_t index
)
1621 struct page
*oldpage
, *newpage
;
1622 struct address_space
*swap_mapping
;
1628 entry
.val
= page_private(oldpage
);
1629 swap_index
= swp_offset(entry
);
1630 swap_mapping
= page_mapping(oldpage
);
1633 * We have arrived here because our zones are constrained, so don't
1634 * limit chance of success by further cpuset and node constraints.
1636 gfp
&= ~GFP_CONSTRAINT_MASK
;
1637 newpage
= shmem_alloc_page(gfp
, info
, index
);
1642 copy_highpage(newpage
, oldpage
);
1643 flush_dcache_page(newpage
);
1645 __SetPageLocked(newpage
);
1646 __SetPageSwapBacked(newpage
);
1647 SetPageUptodate(newpage
);
1648 set_page_private(newpage
, entry
.val
);
1649 SetPageSwapCache(newpage
);
1652 * Our caller will very soon move newpage out of swapcache, but it's
1653 * a nice clean interface for us to replace oldpage by newpage there.
1655 xa_lock_irq(&swap_mapping
->i_pages
);
1656 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1658 mem_cgroup_migrate(oldpage
, newpage
);
1659 __inc_lruvec_page_state(newpage
, NR_FILE_PAGES
);
1660 __dec_lruvec_page_state(oldpage
, NR_FILE_PAGES
);
1662 xa_unlock_irq(&swap_mapping
->i_pages
);
1664 if (unlikely(error
)) {
1666 * Is this possible? I think not, now that our callers check
1667 * both PageSwapCache and page_private after getting page lock;
1668 * but be defensive. Reverse old to newpage for clear and free.
1672 lru_cache_add(newpage
);
1676 ClearPageSwapCache(oldpage
);
1677 set_page_private(oldpage
, 0);
1679 unlock_page(oldpage
);
1686 * Swap in the page pointed to by *pagep.
1687 * Caller has to make sure that *pagep contains a valid swapped page.
1688 * Returns 0 and the page in pagep if success. On failure, returns the
1689 * error code and NULL in *pagep.
1691 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1692 struct page
**pagep
, enum sgp_type sgp
,
1693 gfp_t gfp
, struct vm_area_struct
*vma
,
1694 vm_fault_t
*fault_type
)
1696 struct address_space
*mapping
= inode
->i_mapping
;
1697 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1698 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1703 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1704 swap
= radix_to_swp_entry(*pagep
);
1707 /* Look it up and read it in.. */
1708 page
= lookup_swap_cache(swap
, NULL
, 0);
1710 /* Or update major stats only when swapin succeeds?? */
1712 *fault_type
|= VM_FAULT_MAJOR
;
1713 count_vm_event(PGMAJFAULT
);
1714 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1716 /* Here we actually start the io */
1717 page
= shmem_swapin(swap
, gfp
, info
, index
);
1724 /* We have to do this with page locked to prevent races */
1726 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1727 !shmem_confirm_swap(mapping
, index
, swap
)) {
1731 if (!PageUptodate(page
)) {
1735 wait_on_page_writeback(page
);
1738 * Some architectures may have to restore extra metadata to the
1739 * physical page after reading from swap.
1741 arch_swap_restore(swap
, page
);
1743 if (shmem_should_replace_page(page
, gfp
)) {
1744 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1749 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1750 swp_to_radix_entry(swap
), gfp
,
1755 spin_lock_irq(&info
->lock
);
1757 shmem_recalc_inode(inode
);
1758 spin_unlock_irq(&info
->lock
);
1760 if (sgp
== SGP_WRITE
)
1761 mark_page_accessed(page
);
1763 delete_from_swap_cache(page
);
1764 set_page_dirty(page
);
1770 if (!shmem_confirm_swap(mapping
, index
, swap
))
1782 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1784 * If we allocate a new one we do not mark it dirty. That's up to the
1785 * vm. If we swap it in we mark it dirty since we also free the swap
1786 * entry since a page cannot live in both the swap and page cache.
1788 * vmf and fault_type are only supplied by shmem_fault:
1789 * otherwise they are NULL.
1791 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1792 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1793 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1794 vm_fault_t
*fault_type
)
1796 struct address_space
*mapping
= inode
->i_mapping
;
1797 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1798 struct shmem_sb_info
*sbinfo
;
1799 struct mm_struct
*charge_mm
;
1801 enum sgp_type sgp_huge
= sgp
;
1802 pgoff_t hindex
= index
;
1808 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1810 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1813 if (sgp
<= SGP_CACHE
&&
1814 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1818 sbinfo
= SHMEM_SB(inode
->i_sb
);
1819 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1821 page
= pagecache_get_page(mapping
, index
,
1822 FGP_ENTRY
| FGP_HEAD
| FGP_LOCK
, 0);
1823 if (xa_is_value(page
)) {
1824 error
= shmem_swapin_page(inode
, index
, &page
,
1825 sgp
, gfp
, vma
, fault_type
);
1826 if (error
== -EEXIST
)
1834 hindex
= page
->index
;
1835 if (page
&& sgp
== SGP_WRITE
)
1836 mark_page_accessed(page
);
1838 /* fallocated page? */
1839 if (page
&& !PageUptodate(page
)) {
1840 if (sgp
!= SGP_READ
)
1847 if (page
|| sgp
== SGP_READ
)
1851 * Fast cache lookup did not find it:
1852 * bring it back from swap or allocate.
1855 if (vma
&& userfaultfd_missing(vma
)) {
1856 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1860 /* shmem_symlink() */
1861 if (!shmem_mapping(mapping
))
1863 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1865 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1867 switch (sbinfo
->huge
) {
1868 case SHMEM_HUGE_NEVER
:
1870 case SHMEM_HUGE_WITHIN_SIZE
: {
1874 off
= round_up(index
, HPAGE_PMD_NR
);
1875 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1876 if (i_size
>= HPAGE_PMD_SIZE
&&
1877 i_size
>> PAGE_SHIFT
>= off
)
1882 case SHMEM_HUGE_ADVISE
:
1883 if (sgp_huge
== SGP_HUGE
)
1885 /* TODO: implement fadvise() hints */
1890 huge_gfp
= vma_thp_gfp_mask(vma
);
1891 huge_gfp
= limit_gfp_mask(huge_gfp
, gfp
);
1892 page
= shmem_alloc_and_acct_page(huge_gfp
, inode
, index
, true);
1895 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1901 error
= PTR_ERR(page
);
1903 if (error
!= -ENOSPC
)
1906 * Try to reclaim some space by splitting a huge page
1907 * beyond i_size on the filesystem.
1912 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1913 if (ret
== SHRINK_STOP
)
1921 if (PageTransHuge(page
))
1922 hindex
= round_down(index
, HPAGE_PMD_NR
);
1926 if (sgp
== SGP_WRITE
)
1927 __SetPageReferenced(page
);
1929 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1930 NULL
, gfp
& GFP_RECLAIM_MASK
,
1934 lru_cache_add(page
);
1936 spin_lock_irq(&info
->lock
);
1937 info
->alloced
+= compound_nr(page
);
1938 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1939 shmem_recalc_inode(inode
);
1940 spin_unlock_irq(&info
->lock
);
1943 if (PageTransHuge(page
) &&
1944 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1945 hindex
+ HPAGE_PMD_NR
- 1) {
1947 * Part of the huge page is beyond i_size: subject
1948 * to shrink under memory pressure.
1950 spin_lock(&sbinfo
->shrinklist_lock
);
1952 * _careful to defend against unlocked access to
1953 * ->shrink_list in shmem_unused_huge_shrink()
1955 if (list_empty_careful(&info
->shrinklist
)) {
1956 list_add_tail(&info
->shrinklist
,
1957 &sbinfo
->shrinklist
);
1958 sbinfo
->shrinklist_len
++;
1960 spin_unlock(&sbinfo
->shrinklist_lock
);
1964 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1966 if (sgp
== SGP_FALLOC
)
1970 * Let SGP_WRITE caller clear ends if write does not fill page;
1971 * but SGP_FALLOC on a page fallocated earlier must initialize
1972 * it now, lest undo on failure cancel our earlier guarantee.
1974 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1977 for (i
= 0; i
< compound_nr(page
); i
++) {
1978 clear_highpage(page
+ i
);
1979 flush_dcache_page(page
+ i
);
1981 SetPageUptodate(page
);
1984 /* Perhaps the file has been truncated since we checked */
1985 if (sgp
<= SGP_CACHE
&&
1986 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1988 ClearPageDirty(page
);
1989 delete_from_page_cache(page
);
1990 spin_lock_irq(&info
->lock
);
1991 shmem_recalc_inode(inode
);
1992 spin_unlock_irq(&info
->lock
);
1998 *pagep
= page
+ index
- hindex
;
2005 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
2007 if (PageTransHuge(page
)) {
2017 if (error
== -ENOSPC
&& !once
++) {
2018 spin_lock_irq(&info
->lock
);
2019 shmem_recalc_inode(inode
);
2020 spin_unlock_irq(&info
->lock
);
2023 if (error
== -EEXIST
)
2029 * This is like autoremove_wake_function, but it removes the wait queue
2030 * entry unconditionally - even if something else had already woken the
2033 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2035 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2036 list_del_init(&wait
->entry
);
2040 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2042 struct vm_area_struct
*vma
= vmf
->vma
;
2043 struct inode
*inode
= file_inode(vma
->vm_file
);
2044 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2047 vm_fault_t ret
= VM_FAULT_LOCKED
;
2050 * Trinity finds that probing a hole which tmpfs is punching can
2051 * prevent the hole-punch from ever completing: which in turn
2052 * locks writers out with its hold on i_mutex. So refrain from
2053 * faulting pages into the hole while it's being punched. Although
2054 * shmem_undo_range() does remove the additions, it may be unable to
2055 * keep up, as each new page needs its own unmap_mapping_range() call,
2056 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2058 * It does not matter if we sometimes reach this check just before the
2059 * hole-punch begins, so that one fault then races with the punch:
2060 * we just need to make racing faults a rare case.
2062 * The implementation below would be much simpler if we just used a
2063 * standard mutex or completion: but we cannot take i_mutex in fault,
2064 * and bloating every shmem inode for this unlikely case would be sad.
2066 if (unlikely(inode
->i_private
)) {
2067 struct shmem_falloc
*shmem_falloc
;
2069 spin_lock(&inode
->i_lock
);
2070 shmem_falloc
= inode
->i_private
;
2072 shmem_falloc
->waitq
&&
2073 vmf
->pgoff
>= shmem_falloc
->start
&&
2074 vmf
->pgoff
< shmem_falloc
->next
) {
2076 wait_queue_head_t
*shmem_falloc_waitq
;
2077 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2079 ret
= VM_FAULT_NOPAGE
;
2080 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2082 ret
= VM_FAULT_RETRY
;
2084 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2085 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2086 TASK_UNINTERRUPTIBLE
);
2087 spin_unlock(&inode
->i_lock
);
2091 * shmem_falloc_waitq points into the shmem_fallocate()
2092 * stack of the hole-punching task: shmem_falloc_waitq
2093 * is usually invalid by the time we reach here, but
2094 * finish_wait() does not dereference it in that case;
2095 * though i_lock needed lest racing with wake_up_all().
2097 spin_lock(&inode
->i_lock
);
2098 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2099 spin_unlock(&inode
->i_lock
);
2105 spin_unlock(&inode
->i_lock
);
2110 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2111 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2113 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2116 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2117 gfp
, vma
, vmf
, &ret
);
2119 return vmf_error(err
);
2123 unsigned long shmem_get_unmapped_area(struct file
*file
,
2124 unsigned long uaddr
, unsigned long len
,
2125 unsigned long pgoff
, unsigned long flags
)
2127 unsigned long (*get_area
)(struct file
*,
2128 unsigned long, unsigned long, unsigned long, unsigned long);
2130 unsigned long offset
;
2131 unsigned long inflated_len
;
2132 unsigned long inflated_addr
;
2133 unsigned long inflated_offset
;
2135 if (len
> TASK_SIZE
)
2138 get_area
= current
->mm
->get_unmapped_area
;
2139 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2141 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2143 if (IS_ERR_VALUE(addr
))
2145 if (addr
& ~PAGE_MASK
)
2147 if (addr
> TASK_SIZE
- len
)
2150 if (shmem_huge
== SHMEM_HUGE_DENY
)
2152 if (len
< HPAGE_PMD_SIZE
)
2154 if (flags
& MAP_FIXED
)
2157 * Our priority is to support MAP_SHARED mapped hugely;
2158 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2159 * But if caller specified an address hint and we allocated area there
2160 * successfully, respect that as before.
2165 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2166 struct super_block
*sb
;
2169 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2170 sb
= file_inode(file
)->i_sb
;
2173 * Called directly from mm/mmap.c, or drivers/char/mem.c
2174 * for "/dev/zero", to create a shared anonymous object.
2176 if (IS_ERR(shm_mnt
))
2178 sb
= shm_mnt
->mnt_sb
;
2180 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2184 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2185 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2187 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2190 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2191 if (inflated_len
> TASK_SIZE
)
2193 if (inflated_len
< len
)
2196 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2197 if (IS_ERR_VALUE(inflated_addr
))
2199 if (inflated_addr
& ~PAGE_MASK
)
2202 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2203 inflated_addr
+= offset
- inflated_offset
;
2204 if (inflated_offset
> offset
)
2205 inflated_addr
+= HPAGE_PMD_SIZE
;
2207 if (inflated_addr
> TASK_SIZE
- len
)
2209 return inflated_addr
;
2213 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2215 struct inode
*inode
= file_inode(vma
->vm_file
);
2216 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2219 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2222 struct inode
*inode
= file_inode(vma
->vm_file
);
2225 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2226 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2230 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2232 struct inode
*inode
= file_inode(file
);
2233 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2234 int retval
= -ENOMEM
;
2237 * What serializes the accesses to info->flags?
2238 * ipc_lock_object() when called from shmctl_do_lock(),
2239 * no serialization needed when called from shm_destroy().
2241 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2242 if (!user_shm_lock(inode
->i_size
, user
))
2244 info
->flags
|= VM_LOCKED
;
2245 mapping_set_unevictable(file
->f_mapping
);
2247 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2248 user_shm_unlock(inode
->i_size
, user
);
2249 info
->flags
&= ~VM_LOCKED
;
2250 mapping_clear_unevictable(file
->f_mapping
);
2258 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2260 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2263 ret
= seal_check_future_write(info
->seals
, vma
);
2267 /* arm64 - allow memory tagging on RAM-based files */
2268 vma
->vm_flags
|= VM_MTE_ALLOWED
;
2270 file_accessed(file
);
2271 vma
->vm_ops
= &shmem_vm_ops
;
2272 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
2273 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2274 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2275 khugepaged_enter(vma
, vma
->vm_flags
);
2280 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2281 umode_t mode
, dev_t dev
, unsigned long flags
)
2283 struct inode
*inode
;
2284 struct shmem_inode_info
*info
;
2285 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2288 if (shmem_reserve_inode(sb
, &ino
))
2291 inode
= new_inode(sb
);
2294 inode_init_owner(&init_user_ns
, inode
, dir
, mode
);
2295 inode
->i_blocks
= 0;
2296 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2297 inode
->i_generation
= prandom_u32();
2298 info
= SHMEM_I(inode
);
2299 memset(info
, 0, (char *)inode
- (char *)info
);
2300 spin_lock_init(&info
->lock
);
2301 atomic_set(&info
->stop_eviction
, 0);
2302 info
->seals
= F_SEAL_SEAL
;
2303 info
->flags
= flags
& VM_NORESERVE
;
2304 INIT_LIST_HEAD(&info
->shrinklist
);
2305 INIT_LIST_HEAD(&info
->swaplist
);
2306 simple_xattrs_init(&info
->xattrs
);
2307 cache_no_acl(inode
);
2309 switch (mode
& S_IFMT
) {
2311 inode
->i_op
= &shmem_special_inode_operations
;
2312 init_special_inode(inode
, mode
, dev
);
2315 inode
->i_mapping
->a_ops
= &shmem_aops
;
2316 inode
->i_op
= &shmem_inode_operations
;
2317 inode
->i_fop
= &shmem_file_operations
;
2318 mpol_shared_policy_init(&info
->policy
,
2319 shmem_get_sbmpol(sbinfo
));
2323 /* Some things misbehave if size == 0 on a directory */
2324 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2325 inode
->i_op
= &shmem_dir_inode_operations
;
2326 inode
->i_fop
= &simple_dir_operations
;
2330 * Must not load anything in the rbtree,
2331 * mpol_free_shared_policy will not be called.
2333 mpol_shared_policy_init(&info
->policy
, NULL
);
2337 lockdep_annotate_inode_mutex_key(inode
);
2339 shmem_free_inode(sb
);
2343 static 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
);
2359 pte_t _dst_pte
, *dst_pte
;
2361 pgoff_t offset
, max_off
;
2364 if (!shmem_inode_acct_block(inode
, 1)) {
2366 * We may have got a page, returned -ENOENT triggering a retry,
2367 * and now we find ourselves with -ENOMEM. Release the page, to
2368 * avoid a BUG_ON in our caller.
2370 if (unlikely(*pagep
)) {
2378 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2380 goto out_unacct_blocks
;
2382 if (!zeropage
) { /* mcopy_atomic */
2383 page_kaddr
= kmap_atomic(page
);
2384 ret
= copy_from_user(page_kaddr
,
2385 (const void __user
*)src_addr
,
2387 kunmap_atomic(page_kaddr
);
2389 /* fallback to copy_from_user outside mmap_lock */
2390 if (unlikely(ret
)) {
2392 shmem_inode_unacct_blocks(inode
, 1);
2393 /* don't free the page */
2396 } else { /* mfill_zeropage_atomic */
2397 clear_highpage(page
);
2404 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2405 __SetPageLocked(page
);
2406 __SetPageSwapBacked(page
);
2407 __SetPageUptodate(page
);
2410 offset
= linear_page_index(dst_vma
, dst_addr
);
2411 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2412 if (unlikely(offset
>= max_off
))
2415 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2416 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2420 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2421 if (dst_vma
->vm_flags
& VM_WRITE
)
2422 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2425 * We don't set the pte dirty if the vma has no
2426 * VM_WRITE permission, so mark the page dirty or it
2427 * could be freed from under us. We could do it
2428 * unconditionally before unlock_page(), but doing it
2429 * only if VM_WRITE is not set is faster.
2431 set_page_dirty(page
);
2434 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2437 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2438 if (unlikely(offset
>= max_off
))
2439 goto out_release_unlock
;
2442 if (!pte_none(*dst_pte
))
2443 goto out_release_unlock
;
2445 lru_cache_add(page
);
2447 spin_lock_irq(&info
->lock
);
2449 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2450 shmem_recalc_inode(inode
);
2451 spin_unlock_irq(&info
->lock
);
2453 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2454 page_add_file_rmap(page
, false);
2455 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2457 /* No need to invalidate - it was non-present before */
2458 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2459 pte_unmap_unlock(dst_pte
, ptl
);
2465 pte_unmap_unlock(dst_pte
, ptl
);
2466 ClearPageDirty(page
);
2467 delete_from_page_cache(page
);
2472 shmem_inode_unacct_blocks(inode
, 1);
2476 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2478 struct vm_area_struct
*dst_vma
,
2479 unsigned long dst_addr
,
2480 unsigned long src_addr
,
2481 struct page
**pagep
)
2483 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2484 dst_addr
, src_addr
, false, pagep
);
2487 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2489 struct vm_area_struct
*dst_vma
,
2490 unsigned long dst_addr
)
2492 struct page
*page
= NULL
;
2494 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2495 dst_addr
, 0, true, &page
);
2499 static const struct inode_operations shmem_symlink_inode_operations
;
2500 static const struct inode_operations shmem_short_symlink_operations
;
2502 #ifdef CONFIG_TMPFS_XATTR
2503 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2505 #define shmem_initxattrs NULL
2509 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2510 loff_t pos
, unsigned len
, unsigned flags
,
2511 struct page
**pagep
, void **fsdata
)
2513 struct inode
*inode
= mapping
->host
;
2514 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2515 pgoff_t index
= pos
>> PAGE_SHIFT
;
2517 /* i_mutex is held by caller */
2518 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2519 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2520 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2522 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2526 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2530 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2531 loff_t pos
, unsigned len
, unsigned copied
,
2532 struct page
*page
, void *fsdata
)
2534 struct inode
*inode
= mapping
->host
;
2536 if (pos
+ copied
> inode
->i_size
)
2537 i_size_write(inode
, pos
+ copied
);
2539 if (!PageUptodate(page
)) {
2540 struct page
*head
= compound_head(page
);
2541 if (PageTransCompound(page
)) {
2544 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2545 if (head
+ i
== page
)
2547 clear_highpage(head
+ i
);
2548 flush_dcache_page(head
+ i
);
2551 if (copied
< PAGE_SIZE
) {
2552 unsigned from
= pos
& (PAGE_SIZE
- 1);
2553 zero_user_segments(page
, 0, from
,
2554 from
+ copied
, PAGE_SIZE
);
2556 SetPageUptodate(head
);
2558 set_page_dirty(page
);
2565 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2567 struct file
*file
= iocb
->ki_filp
;
2568 struct inode
*inode
= file_inode(file
);
2569 struct address_space
*mapping
= inode
->i_mapping
;
2571 unsigned long offset
;
2572 enum sgp_type sgp
= SGP_READ
;
2575 loff_t
*ppos
= &iocb
->ki_pos
;
2578 * Might this read be for a stacking filesystem? Then when reading
2579 * holes of a sparse file, we actually need to allocate those pages,
2580 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2582 if (!iter_is_iovec(to
))
2585 index
= *ppos
>> PAGE_SHIFT
;
2586 offset
= *ppos
& ~PAGE_MASK
;
2589 struct page
*page
= NULL
;
2591 unsigned long nr
, ret
;
2592 loff_t i_size
= i_size_read(inode
);
2594 end_index
= i_size
>> PAGE_SHIFT
;
2595 if (index
> end_index
)
2597 if (index
== end_index
) {
2598 nr
= i_size
& ~PAGE_MASK
;
2603 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2605 if (error
== -EINVAL
)
2610 if (sgp
== SGP_CACHE
)
2611 set_page_dirty(page
);
2616 * We must evaluate after, since reads (unlike writes)
2617 * are called without i_mutex protection against truncate
2620 i_size
= i_size_read(inode
);
2621 end_index
= i_size
>> PAGE_SHIFT
;
2622 if (index
== end_index
) {
2623 nr
= i_size
& ~PAGE_MASK
;
2634 * If users can be writing to this page using arbitrary
2635 * virtual addresses, take care about potential aliasing
2636 * before reading the page on the kernel side.
2638 if (mapping_writably_mapped(mapping
))
2639 flush_dcache_page(page
);
2641 * Mark the page accessed if we read the beginning.
2644 mark_page_accessed(page
);
2646 page
= ZERO_PAGE(0);
2651 * Ok, we have the page, and it's up-to-date, so
2652 * now we can copy it to user space...
2654 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2657 index
+= offset
>> PAGE_SHIFT
;
2658 offset
&= ~PAGE_MASK
;
2661 if (!iov_iter_count(to
))
2670 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2671 file_accessed(file
);
2672 return retval
? retval
: error
;
2675 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2677 struct address_space
*mapping
= file
->f_mapping
;
2678 struct inode
*inode
= mapping
->host
;
2680 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2681 return generic_file_llseek_size(file
, offset
, whence
,
2682 MAX_LFS_FILESIZE
, i_size_read(inode
));
2687 /* We're holding i_mutex so we can access i_size directly */
2688 offset
= mapping_seek_hole_data(mapping
, offset
, inode
->i_size
, whence
);
2690 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2691 inode_unlock(inode
);
2695 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2698 struct inode
*inode
= file_inode(file
);
2699 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2700 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2701 struct shmem_falloc shmem_falloc
;
2702 pgoff_t start
, index
, end
;
2705 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2710 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2711 struct address_space
*mapping
= file
->f_mapping
;
2712 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2713 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2714 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2716 /* protected by i_mutex */
2717 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2722 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2723 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2724 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2725 spin_lock(&inode
->i_lock
);
2726 inode
->i_private
= &shmem_falloc
;
2727 spin_unlock(&inode
->i_lock
);
2729 if ((u64
)unmap_end
> (u64
)unmap_start
)
2730 unmap_mapping_range(mapping
, unmap_start
,
2731 1 + unmap_end
- unmap_start
, 0);
2732 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2733 /* No need to unmap again: hole-punching leaves COWed pages */
2735 spin_lock(&inode
->i_lock
);
2736 inode
->i_private
= NULL
;
2737 wake_up_all(&shmem_falloc_waitq
);
2738 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2739 spin_unlock(&inode
->i_lock
);
2744 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2745 error
= inode_newsize_ok(inode
, offset
+ len
);
2749 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2754 start
= offset
>> PAGE_SHIFT
;
2755 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2756 /* Try to avoid a swapstorm if len is impossible to satisfy */
2757 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2762 shmem_falloc
.waitq
= NULL
;
2763 shmem_falloc
.start
= start
;
2764 shmem_falloc
.next
= start
;
2765 shmem_falloc
.nr_falloced
= 0;
2766 shmem_falloc
.nr_unswapped
= 0;
2767 spin_lock(&inode
->i_lock
);
2768 inode
->i_private
= &shmem_falloc
;
2769 spin_unlock(&inode
->i_lock
);
2771 for (index
= start
; index
< end
; index
++) {
2775 * Good, the fallocate(2) manpage permits EINTR: we may have
2776 * been interrupted because we are using up too much memory.
2778 if (signal_pending(current
))
2780 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2783 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2785 /* Remove the !PageUptodate pages we added */
2786 if (index
> start
) {
2787 shmem_undo_range(inode
,
2788 (loff_t
)start
<< PAGE_SHIFT
,
2789 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2795 * Inform shmem_writepage() how far we have reached.
2796 * No need for lock or barrier: we have the page lock.
2798 shmem_falloc
.next
++;
2799 if (!PageUptodate(page
))
2800 shmem_falloc
.nr_falloced
++;
2803 * If !PageUptodate, leave it that way so that freeable pages
2804 * can be recognized if we need to rollback on error later.
2805 * But set_page_dirty so that memory pressure will swap rather
2806 * than free the pages we are allocating (and SGP_CACHE pages
2807 * might still be clean: we now need to mark those dirty too).
2809 set_page_dirty(page
);
2815 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2816 i_size_write(inode
, offset
+ len
);
2817 inode
->i_ctime
= current_time(inode
);
2819 spin_lock(&inode
->i_lock
);
2820 inode
->i_private
= NULL
;
2821 spin_unlock(&inode
->i_lock
);
2823 inode_unlock(inode
);
2827 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2829 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2831 buf
->f_type
= TMPFS_MAGIC
;
2832 buf
->f_bsize
= PAGE_SIZE
;
2833 buf
->f_namelen
= NAME_MAX
;
2834 if (sbinfo
->max_blocks
) {
2835 buf
->f_blocks
= sbinfo
->max_blocks
;
2837 buf
->f_bfree
= sbinfo
->max_blocks
-
2838 percpu_counter_sum(&sbinfo
->used_blocks
);
2840 if (sbinfo
->max_inodes
) {
2841 buf
->f_files
= sbinfo
->max_inodes
;
2842 buf
->f_ffree
= sbinfo
->free_inodes
;
2844 /* else leave those fields 0 like simple_statfs */
2846 buf
->f_fsid
= uuid_to_fsid(dentry
->d_sb
->s_uuid
.b
);
2852 * File creation. Allocate an inode, and we're done..
2855 shmem_mknod(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2856 struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2858 struct inode
*inode
;
2859 int error
= -ENOSPC
;
2861 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2863 error
= simple_acl_create(dir
, inode
);
2866 error
= security_inode_init_security(inode
, dir
,
2868 shmem_initxattrs
, NULL
);
2869 if (error
&& error
!= -EOPNOTSUPP
)
2873 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2874 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2875 d_instantiate(dentry
, inode
);
2876 dget(dentry
); /* Extra count - pin the dentry in core */
2885 shmem_tmpfile(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2886 struct dentry
*dentry
, umode_t mode
)
2888 struct inode
*inode
;
2889 int error
= -ENOSPC
;
2891 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2893 error
= security_inode_init_security(inode
, dir
,
2895 shmem_initxattrs
, NULL
);
2896 if (error
&& error
!= -EOPNOTSUPP
)
2898 error
= simple_acl_create(dir
, inode
);
2901 d_tmpfile(dentry
, inode
);
2909 static int shmem_mkdir(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2910 struct dentry
*dentry
, umode_t mode
)
2914 if ((error
= shmem_mknod(&init_user_ns
, dir
, dentry
,
2915 mode
| S_IFDIR
, 0)))
2921 static int shmem_create(struct user_namespace
*mnt_userns
, struct inode
*dir
,
2922 struct dentry
*dentry
, umode_t mode
, bool excl
)
2924 return shmem_mknod(&init_user_ns
, dir
, dentry
, mode
| S_IFREG
, 0);
2930 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2932 struct inode
*inode
= d_inode(old_dentry
);
2936 * No ordinary (disk based) filesystem counts links as inodes;
2937 * but each new link needs a new dentry, pinning lowmem, and
2938 * tmpfs dentries cannot be pruned until they are unlinked.
2939 * But if an O_TMPFILE file is linked into the tmpfs, the
2940 * first link must skip that, to get the accounting right.
2942 if (inode
->i_nlink
) {
2943 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
2948 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2949 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2951 ihold(inode
); /* New dentry reference */
2952 dget(dentry
); /* Extra pinning count for the created dentry */
2953 d_instantiate(dentry
, inode
);
2958 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2960 struct inode
*inode
= d_inode(dentry
);
2962 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2963 shmem_free_inode(inode
->i_sb
);
2965 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2966 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
2968 dput(dentry
); /* Undo the count from "create" - this does all the work */
2972 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2974 if (!simple_empty(dentry
))
2977 drop_nlink(d_inode(dentry
));
2979 return shmem_unlink(dir
, dentry
);
2982 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2984 bool old_is_dir
= d_is_dir(old_dentry
);
2985 bool new_is_dir
= d_is_dir(new_dentry
);
2987 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2989 drop_nlink(old_dir
);
2992 drop_nlink(new_dir
);
2996 old_dir
->i_ctime
= old_dir
->i_mtime
=
2997 new_dir
->i_ctime
= new_dir
->i_mtime
=
2998 d_inode(old_dentry
)->i_ctime
=
2999 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3004 static int shmem_whiteout(struct user_namespace
*mnt_userns
,
3005 struct inode
*old_dir
, struct dentry
*old_dentry
)
3007 struct dentry
*whiteout
;
3010 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3014 error
= shmem_mknod(&init_user_ns
, old_dir
, whiteout
,
3015 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3021 * Cheat and hash the whiteout while the old dentry is still in
3022 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3024 * d_lookup() will consistently find one of them at this point,
3025 * not sure which one, but that isn't even important.
3032 * The VFS layer already does all the dentry stuff for rename,
3033 * we just have to decrement the usage count for the target if
3034 * it exists so that the VFS layer correctly free's it when it
3037 static int shmem_rename2(struct user_namespace
*mnt_userns
,
3038 struct inode
*old_dir
, struct dentry
*old_dentry
,
3039 struct inode
*new_dir
, struct dentry
*new_dentry
,
3042 struct inode
*inode
= d_inode(old_dentry
);
3043 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3045 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3048 if (flags
& RENAME_EXCHANGE
)
3049 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3051 if (!simple_empty(new_dentry
))
3054 if (flags
& RENAME_WHITEOUT
) {
3057 error
= shmem_whiteout(&init_user_ns
, old_dir
, old_dentry
);
3062 if (d_really_is_positive(new_dentry
)) {
3063 (void) shmem_unlink(new_dir
, new_dentry
);
3064 if (they_are_dirs
) {
3065 drop_nlink(d_inode(new_dentry
));
3066 drop_nlink(old_dir
);
3068 } else if (they_are_dirs
) {
3069 drop_nlink(old_dir
);
3073 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3074 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3075 old_dir
->i_ctime
= old_dir
->i_mtime
=
3076 new_dir
->i_ctime
= new_dir
->i_mtime
=
3077 inode
->i_ctime
= current_time(old_dir
);
3081 static int shmem_symlink(struct user_namespace
*mnt_userns
, struct inode
*dir
,
3082 struct dentry
*dentry
, const char *symname
)
3086 struct inode
*inode
;
3089 len
= strlen(symname
) + 1;
3090 if (len
> PAGE_SIZE
)
3091 return -ENAMETOOLONG
;
3093 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3098 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3099 shmem_initxattrs
, NULL
);
3100 if (error
&& error
!= -EOPNOTSUPP
) {
3105 inode
->i_size
= len
-1;
3106 if (len
<= SHORT_SYMLINK_LEN
) {
3107 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3108 if (!inode
->i_link
) {
3112 inode
->i_op
= &shmem_short_symlink_operations
;
3114 inode_nohighmem(inode
);
3115 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3120 inode
->i_mapping
->a_ops
= &shmem_aops
;
3121 inode
->i_op
= &shmem_symlink_inode_operations
;
3122 memcpy(page_address(page
), symname
, len
);
3123 SetPageUptodate(page
);
3124 set_page_dirty(page
);
3128 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3129 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3130 d_instantiate(dentry
, inode
);
3135 static void shmem_put_link(void *arg
)
3137 mark_page_accessed(arg
);
3141 static const char *shmem_get_link(struct dentry
*dentry
,
3142 struct inode
*inode
,
3143 struct delayed_call
*done
)
3145 struct page
*page
= NULL
;
3148 page
= find_get_page(inode
->i_mapping
, 0);
3150 return ERR_PTR(-ECHILD
);
3151 if (!PageUptodate(page
)) {
3153 return ERR_PTR(-ECHILD
);
3156 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3158 return ERR_PTR(error
);
3161 set_delayed_call(done
, shmem_put_link
, page
);
3162 return page_address(page
);
3165 #ifdef CONFIG_TMPFS_XATTR
3167 * Superblocks without xattr inode operations may get some security.* xattr
3168 * support from the LSM "for free". As soon as we have any other xattrs
3169 * like ACLs, we also need to implement the security.* handlers at
3170 * filesystem level, though.
3174 * Callback for security_inode_init_security() for acquiring xattrs.
3176 static int shmem_initxattrs(struct inode
*inode
,
3177 const struct xattr
*xattr_array
,
3180 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3181 const struct xattr
*xattr
;
3182 struct simple_xattr
*new_xattr
;
3185 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3186 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3190 len
= strlen(xattr
->name
) + 1;
3191 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3193 if (!new_xattr
->name
) {
3198 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3199 XATTR_SECURITY_PREFIX_LEN
);
3200 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3203 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3209 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3210 struct dentry
*unused
, struct inode
*inode
,
3211 const char *name
, void *buffer
, size_t size
)
3213 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3215 name
= xattr_full_name(handler
, name
);
3216 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3219 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3220 struct user_namespace
*mnt_userns
,
3221 struct dentry
*unused
, struct inode
*inode
,
3222 const char *name
, const void *value
,
3223 size_t size
, int flags
)
3225 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3227 name
= xattr_full_name(handler
, name
);
3228 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3231 static const struct xattr_handler shmem_security_xattr_handler
= {
3232 .prefix
= XATTR_SECURITY_PREFIX
,
3233 .get
= shmem_xattr_handler_get
,
3234 .set
= shmem_xattr_handler_set
,
3237 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3238 .prefix
= XATTR_TRUSTED_PREFIX
,
3239 .get
= shmem_xattr_handler_get
,
3240 .set
= shmem_xattr_handler_set
,
3243 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3244 #ifdef CONFIG_TMPFS_POSIX_ACL
3245 &posix_acl_access_xattr_handler
,
3246 &posix_acl_default_xattr_handler
,
3248 &shmem_security_xattr_handler
,
3249 &shmem_trusted_xattr_handler
,
3253 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3255 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3256 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3258 #endif /* CONFIG_TMPFS_XATTR */
3260 static const struct inode_operations shmem_short_symlink_operations
= {
3261 .get_link
= simple_get_link
,
3262 #ifdef CONFIG_TMPFS_XATTR
3263 .listxattr
= shmem_listxattr
,
3267 static const struct inode_operations shmem_symlink_inode_operations
= {
3268 .get_link
= shmem_get_link
,
3269 #ifdef CONFIG_TMPFS_XATTR
3270 .listxattr
= shmem_listxattr
,
3274 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3276 return ERR_PTR(-ESTALE
);
3279 static int shmem_match(struct inode
*ino
, void *vfh
)
3283 inum
= (inum
<< 32) | fh
[1];
3284 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3287 /* Find any alias of inode, but prefer a hashed alias */
3288 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3290 struct dentry
*alias
= d_find_alias(inode
);
3292 return alias
?: d_find_any_alias(inode
);
3296 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3297 struct fid
*fid
, int fh_len
, int fh_type
)
3299 struct inode
*inode
;
3300 struct dentry
*dentry
= NULL
;
3307 inum
= (inum
<< 32) | fid
->raw
[1];
3309 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3310 shmem_match
, fid
->raw
);
3312 dentry
= shmem_find_alias(inode
);
3319 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3320 struct inode
*parent
)
3324 return FILEID_INVALID
;
3327 if (inode_unhashed(inode
)) {
3328 /* Unfortunately insert_inode_hash is not idempotent,
3329 * so as we hash inodes here rather than at creation
3330 * time, we need a lock to ensure we only try
3333 static DEFINE_SPINLOCK(lock
);
3335 if (inode_unhashed(inode
))
3336 __insert_inode_hash(inode
,
3337 inode
->i_ino
+ inode
->i_generation
);
3341 fh
[0] = inode
->i_generation
;
3342 fh
[1] = inode
->i_ino
;
3343 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3349 static const struct export_operations shmem_export_ops
= {
3350 .get_parent
= shmem_get_parent
,
3351 .encode_fh
= shmem_encode_fh
,
3352 .fh_to_dentry
= shmem_fh_to_dentry
,
3368 static const struct constant_table shmem_param_enums_huge
[] = {
3369 {"never", SHMEM_HUGE_NEVER
},
3370 {"always", SHMEM_HUGE_ALWAYS
},
3371 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3372 {"advise", SHMEM_HUGE_ADVISE
},
3376 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3377 fsparam_u32 ("gid", Opt_gid
),
3378 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3379 fsparam_u32oct("mode", Opt_mode
),
3380 fsparam_string("mpol", Opt_mpol
),
3381 fsparam_string("nr_blocks", Opt_nr_blocks
),
3382 fsparam_string("nr_inodes", Opt_nr_inodes
),
3383 fsparam_string("size", Opt_size
),
3384 fsparam_u32 ("uid", Opt_uid
),
3385 fsparam_flag ("inode32", Opt_inode32
),
3386 fsparam_flag ("inode64", Opt_inode64
),
3390 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3392 struct shmem_options
*ctx
= fc
->fs_private
;
3393 struct fs_parse_result result
;
3394 unsigned long long size
;
3398 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3404 size
= memparse(param
->string
, &rest
);
3406 size
<<= PAGE_SHIFT
;
3407 size
*= totalram_pages();
3413 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3414 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3417 ctx
->blocks
= memparse(param
->string
, &rest
);
3420 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3423 ctx
->inodes
= memparse(param
->string
, &rest
);
3426 ctx
->seen
|= SHMEM_SEEN_INODES
;
3429 ctx
->mode
= result
.uint_32
& 07777;
3432 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3433 if (!uid_valid(ctx
->uid
))
3437 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3438 if (!gid_valid(ctx
->gid
))
3442 ctx
->huge
= result
.uint_32
;
3443 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3444 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3445 has_transparent_hugepage()))
3446 goto unsupported_parameter
;
3447 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3450 if (IS_ENABLED(CONFIG_NUMA
)) {
3451 mpol_put(ctx
->mpol
);
3453 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3457 goto unsupported_parameter
;
3459 ctx
->full_inums
= false;
3460 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3463 if (sizeof(ino_t
) < 8) {
3465 "Cannot use inode64 with <64bit inums in kernel\n");
3467 ctx
->full_inums
= true;
3468 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3473 unsupported_parameter
:
3474 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3476 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3479 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3481 char *options
= data
;
3484 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3489 while (options
!= NULL
) {
3490 char *this_char
= options
;
3493 * NUL-terminate this option: unfortunately,
3494 * mount options form a comma-separated list,
3495 * but mpol's nodelist may also contain commas.
3497 options
= strchr(options
, ',');
3498 if (options
== NULL
)
3501 if (!isdigit(*options
)) {
3507 char *value
= strchr(this_char
, '=');
3513 len
= strlen(value
);
3515 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3524 * Reconfigure a shmem filesystem.
3526 * Note that we disallow change from limited->unlimited blocks/inodes while any
3527 * are in use; but we must separately disallow unlimited->limited, because in
3528 * that case we have no record of how much is already in use.
3530 static int shmem_reconfigure(struct fs_context
*fc
)
3532 struct shmem_options
*ctx
= fc
->fs_private
;
3533 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3534 unsigned long inodes
;
3537 spin_lock(&sbinfo
->stat_lock
);
3538 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3539 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3540 if (!sbinfo
->max_blocks
) {
3541 err
= "Cannot retroactively limit size";
3544 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3546 err
= "Too small a size for current use";
3550 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3551 if (!sbinfo
->max_inodes
) {
3552 err
= "Cannot retroactively limit inodes";
3555 if (ctx
->inodes
< inodes
) {
3556 err
= "Too few inodes for current use";
3561 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
3562 sbinfo
->next_ino
> UINT_MAX
) {
3563 err
= "Current inum too high to switch to 32-bit inums";
3567 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3568 sbinfo
->huge
= ctx
->huge
;
3569 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
3570 sbinfo
->full_inums
= ctx
->full_inums
;
3571 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3572 sbinfo
->max_blocks
= ctx
->blocks
;
3573 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3574 sbinfo
->max_inodes
= ctx
->inodes
;
3575 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3579 * Preserve previous mempolicy unless mpol remount option was specified.
3582 mpol_put(sbinfo
->mpol
);
3583 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3586 spin_unlock(&sbinfo
->stat_lock
);
3589 spin_unlock(&sbinfo
->stat_lock
);
3590 return invalfc(fc
, "%s", err
);
3593 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3595 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3597 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3598 seq_printf(seq
, ",size=%luk",
3599 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3600 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3601 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3602 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3603 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3604 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3605 seq_printf(seq
, ",uid=%u",
3606 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3607 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3608 seq_printf(seq
, ",gid=%u",
3609 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3612 * Showing inode{64,32} might be useful even if it's the system default,
3613 * since then people don't have to resort to checking both here and
3614 * /proc/config.gz to confirm 64-bit inums were successfully applied
3615 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3617 * We hide it when inode64 isn't the default and we are using 32-bit
3618 * inodes, since that probably just means the feature isn't even under
3623 * +-----------------+-----------------+
3624 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3625 * +------------------+-----------------+-----------------+
3626 * | full_inums=true | show | show |
3627 * | full_inums=false | show | hide |
3628 * +------------------+-----------------+-----------------+
3631 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
3632 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
3633 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3634 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3636 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3638 shmem_show_mpol(seq
, sbinfo
->mpol
);
3642 #endif /* CONFIG_TMPFS */
3644 static void shmem_put_super(struct super_block
*sb
)
3646 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3648 free_percpu(sbinfo
->ino_batch
);
3649 percpu_counter_destroy(&sbinfo
->used_blocks
);
3650 mpol_put(sbinfo
->mpol
);
3652 sb
->s_fs_info
= NULL
;
3655 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3657 struct shmem_options
*ctx
= fc
->fs_private
;
3658 struct inode
*inode
;
3659 struct shmem_sb_info
*sbinfo
;
3662 /* Round up to L1_CACHE_BYTES to resist false sharing */
3663 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3664 L1_CACHE_BYTES
), GFP_KERNEL
);
3668 sb
->s_fs_info
= sbinfo
;
3672 * Per default we only allow half of the physical ram per
3673 * tmpfs instance, limiting inodes to one per page of lowmem;
3674 * but the internal instance is left unlimited.
3676 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3677 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3678 ctx
->blocks
= shmem_default_max_blocks();
3679 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3680 ctx
->inodes
= shmem_default_max_inodes();
3681 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
3682 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
3684 sb
->s_flags
|= SB_NOUSER
;
3686 sb
->s_export_op
= &shmem_export_ops
;
3687 sb
->s_flags
|= SB_NOSEC
;
3689 sb
->s_flags
|= SB_NOUSER
;
3691 sbinfo
->max_blocks
= ctx
->blocks
;
3692 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3693 if (sb
->s_flags
& SB_KERNMOUNT
) {
3694 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
3695 if (!sbinfo
->ino_batch
)
3698 sbinfo
->uid
= ctx
->uid
;
3699 sbinfo
->gid
= ctx
->gid
;
3700 sbinfo
->full_inums
= ctx
->full_inums
;
3701 sbinfo
->mode
= ctx
->mode
;
3702 sbinfo
->huge
= ctx
->huge
;
3703 sbinfo
->mpol
= ctx
->mpol
;
3706 spin_lock_init(&sbinfo
->stat_lock
);
3707 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3709 spin_lock_init(&sbinfo
->shrinklist_lock
);
3710 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3712 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3713 sb
->s_blocksize
= PAGE_SIZE
;
3714 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3715 sb
->s_magic
= TMPFS_MAGIC
;
3716 sb
->s_op
= &shmem_ops
;
3717 sb
->s_time_gran
= 1;
3718 #ifdef CONFIG_TMPFS_XATTR
3719 sb
->s_xattr
= shmem_xattr_handlers
;
3721 #ifdef CONFIG_TMPFS_POSIX_ACL
3722 sb
->s_flags
|= SB_POSIXACL
;
3724 uuid_gen(&sb
->s_uuid
);
3726 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3729 inode
->i_uid
= sbinfo
->uid
;
3730 inode
->i_gid
= sbinfo
->gid
;
3731 sb
->s_root
= d_make_root(inode
);
3737 shmem_put_super(sb
);
3741 static int shmem_get_tree(struct fs_context
*fc
)
3743 return get_tree_nodev(fc
, shmem_fill_super
);
3746 static void shmem_free_fc(struct fs_context
*fc
)
3748 struct shmem_options
*ctx
= fc
->fs_private
;
3751 mpol_put(ctx
->mpol
);
3756 static const struct fs_context_operations shmem_fs_context_ops
= {
3757 .free
= shmem_free_fc
,
3758 .get_tree
= shmem_get_tree
,
3760 .parse_monolithic
= shmem_parse_options
,
3761 .parse_param
= shmem_parse_one
,
3762 .reconfigure
= shmem_reconfigure
,
3766 static struct kmem_cache
*shmem_inode_cachep
;
3768 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3770 struct shmem_inode_info
*info
;
3771 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3774 return &info
->vfs_inode
;
3777 static void shmem_free_in_core_inode(struct inode
*inode
)
3779 if (S_ISLNK(inode
->i_mode
))
3780 kfree(inode
->i_link
);
3781 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3784 static void shmem_destroy_inode(struct inode
*inode
)
3786 if (S_ISREG(inode
->i_mode
))
3787 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3790 static void shmem_init_inode(void *foo
)
3792 struct shmem_inode_info
*info
= foo
;
3793 inode_init_once(&info
->vfs_inode
);
3796 static void shmem_init_inodecache(void)
3798 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3799 sizeof(struct shmem_inode_info
),
3800 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3803 static void shmem_destroy_inodecache(void)
3805 kmem_cache_destroy(shmem_inode_cachep
);
3808 const struct address_space_operations shmem_aops
= {
3809 .writepage
= shmem_writepage
,
3810 .set_page_dirty
= __set_page_dirty_no_writeback
,
3812 .write_begin
= shmem_write_begin
,
3813 .write_end
= shmem_write_end
,
3815 #ifdef CONFIG_MIGRATION
3816 .migratepage
= migrate_page
,
3818 .error_remove_page
= generic_error_remove_page
,
3820 EXPORT_SYMBOL(shmem_aops
);
3822 static const struct file_operations shmem_file_operations
= {
3824 .get_unmapped_area
= shmem_get_unmapped_area
,
3826 .llseek
= shmem_file_llseek
,
3827 .read_iter
= shmem_file_read_iter
,
3828 .write_iter
= generic_file_write_iter
,
3829 .fsync
= noop_fsync
,
3830 .splice_read
= generic_file_splice_read
,
3831 .splice_write
= iter_file_splice_write
,
3832 .fallocate
= shmem_fallocate
,
3836 static const struct inode_operations shmem_inode_operations
= {
3837 .getattr
= shmem_getattr
,
3838 .setattr
= shmem_setattr
,
3839 #ifdef CONFIG_TMPFS_XATTR
3840 .listxattr
= shmem_listxattr
,
3841 .set_acl
= simple_set_acl
,
3845 static const struct inode_operations shmem_dir_inode_operations
= {
3847 .create
= shmem_create
,
3848 .lookup
= simple_lookup
,
3850 .unlink
= shmem_unlink
,
3851 .symlink
= shmem_symlink
,
3852 .mkdir
= shmem_mkdir
,
3853 .rmdir
= shmem_rmdir
,
3854 .mknod
= shmem_mknod
,
3855 .rename
= shmem_rename2
,
3856 .tmpfile
= shmem_tmpfile
,
3858 #ifdef CONFIG_TMPFS_XATTR
3859 .listxattr
= shmem_listxattr
,
3861 #ifdef CONFIG_TMPFS_POSIX_ACL
3862 .setattr
= shmem_setattr
,
3863 .set_acl
= simple_set_acl
,
3867 static const struct inode_operations shmem_special_inode_operations
= {
3868 #ifdef CONFIG_TMPFS_XATTR
3869 .listxattr
= shmem_listxattr
,
3871 #ifdef CONFIG_TMPFS_POSIX_ACL
3872 .setattr
= shmem_setattr
,
3873 .set_acl
= simple_set_acl
,
3877 static const struct super_operations shmem_ops
= {
3878 .alloc_inode
= shmem_alloc_inode
,
3879 .free_inode
= shmem_free_in_core_inode
,
3880 .destroy_inode
= shmem_destroy_inode
,
3882 .statfs
= shmem_statfs
,
3883 .show_options
= shmem_show_options
,
3885 .evict_inode
= shmem_evict_inode
,
3886 .drop_inode
= generic_delete_inode
,
3887 .put_super
= shmem_put_super
,
3888 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3889 .nr_cached_objects
= shmem_unused_huge_count
,
3890 .free_cached_objects
= shmem_unused_huge_scan
,
3894 static const struct vm_operations_struct shmem_vm_ops
= {
3895 .fault
= shmem_fault
,
3896 .map_pages
= filemap_map_pages
,
3898 .set_policy
= shmem_set_policy
,
3899 .get_policy
= shmem_get_policy
,
3903 int shmem_init_fs_context(struct fs_context
*fc
)
3905 struct shmem_options
*ctx
;
3907 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3911 ctx
->mode
= 0777 | S_ISVTX
;
3912 ctx
->uid
= current_fsuid();
3913 ctx
->gid
= current_fsgid();
3915 fc
->fs_private
= ctx
;
3916 fc
->ops
= &shmem_fs_context_ops
;
3920 static struct file_system_type shmem_fs_type
= {
3921 .owner
= THIS_MODULE
,
3923 .init_fs_context
= shmem_init_fs_context
,
3925 .parameters
= shmem_fs_parameters
,
3927 .kill_sb
= kill_litter_super
,
3928 .fs_flags
= FS_USERNS_MOUNT
| FS_THP_SUPPORT
,
3931 int __init
shmem_init(void)
3935 shmem_init_inodecache();
3937 error
= register_filesystem(&shmem_fs_type
);
3939 pr_err("Could not register tmpfs\n");
3943 shm_mnt
= kern_mount(&shmem_fs_type
);
3944 if (IS_ERR(shm_mnt
)) {
3945 error
= PTR_ERR(shm_mnt
);
3946 pr_err("Could not kern_mount tmpfs\n");
3950 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3951 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3952 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3954 shmem_huge
= 0; /* just in case it was patched */
3959 unregister_filesystem(&shmem_fs_type
);
3961 shmem_destroy_inodecache();
3962 shm_mnt
= ERR_PTR(error
);
3966 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3967 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3968 struct kobj_attribute
*attr
, char *buf
)
3970 static const int values
[] = {
3972 SHMEM_HUGE_WITHIN_SIZE
,
3981 for (i
= 0; i
< ARRAY_SIZE(values
); i
++) {
3982 len
+= sysfs_emit_at(buf
, len
,
3983 shmem_huge
== values
[i
] ? "%s[%s]" : "%s%s",
3985 shmem_format_huge(values
[i
]));
3988 len
+= sysfs_emit_at(buf
, len
, "\n");
3993 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3994 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3999 if (count
+ 1 > sizeof(tmp
))
4001 memcpy(tmp
, buf
, count
);
4003 if (count
&& tmp
[count
- 1] == '\n')
4004 tmp
[count
- 1] = '\0';
4006 huge
= shmem_parse_huge(tmp
);
4007 if (huge
== -EINVAL
)
4009 if (!has_transparent_hugepage() &&
4010 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4014 if (shmem_huge
> SHMEM_HUGE_DENY
)
4015 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4019 struct kobj_attribute shmem_enabled_attr
=
4020 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4021 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4023 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4024 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4026 struct inode
*inode
= file_inode(vma
->vm_file
);
4027 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4031 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
4032 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
4034 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4036 if (shmem_huge
== SHMEM_HUGE_DENY
)
4038 switch (sbinfo
->huge
) {
4039 case SHMEM_HUGE_NEVER
:
4041 case SHMEM_HUGE_ALWAYS
:
4043 case SHMEM_HUGE_WITHIN_SIZE
:
4044 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4045 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4046 if (i_size
>= HPAGE_PMD_SIZE
&&
4047 i_size
>> PAGE_SHIFT
>= off
)
4050 case SHMEM_HUGE_ADVISE
:
4051 /* TODO: implement fadvise() hints */
4052 return (vma
->vm_flags
& VM_HUGEPAGE
);
4058 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4060 #else /* !CONFIG_SHMEM */
4063 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4065 * This is intended for small system where the benefits of the full
4066 * shmem code (swap-backed and resource-limited) are outweighed by
4067 * their complexity. On systems without swap this code should be
4068 * effectively equivalent, but much lighter weight.
4071 static struct file_system_type shmem_fs_type
= {
4073 .init_fs_context
= ramfs_init_fs_context
,
4074 .parameters
= ramfs_fs_parameters
,
4075 .kill_sb
= kill_litter_super
,
4076 .fs_flags
= FS_USERNS_MOUNT
,
4079 int __init
shmem_init(void)
4081 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4083 shm_mnt
= kern_mount(&shmem_fs_type
);
4084 BUG_ON(IS_ERR(shm_mnt
));
4089 int shmem_unuse(unsigned int type
, bool frontswap
,
4090 unsigned long *fs_pages_to_unuse
)
4095 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4100 void shmem_unlock_mapping(struct address_space
*mapping
)
4105 unsigned long shmem_get_unmapped_area(struct file
*file
,
4106 unsigned long addr
, unsigned long len
,
4107 unsigned long pgoff
, unsigned long flags
)
4109 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4113 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4115 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4117 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4119 #define shmem_vm_ops generic_file_vm_ops
4120 #define shmem_file_operations ramfs_file_operations
4121 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4122 #define shmem_acct_size(flags, size) 0
4123 #define shmem_unacct_size(flags, size) do {} while (0)
4125 #endif /* CONFIG_SHMEM */
4129 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4130 unsigned long flags
, unsigned int i_flags
)
4132 struct inode
*inode
;
4136 return ERR_CAST(mnt
);
4138 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4139 return ERR_PTR(-EINVAL
);
4141 if (shmem_acct_size(flags
, size
))
4142 return ERR_PTR(-ENOMEM
);
4144 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4146 if (unlikely(!inode
)) {
4147 shmem_unacct_size(flags
, size
);
4148 return ERR_PTR(-ENOSPC
);
4150 inode
->i_flags
|= i_flags
;
4151 inode
->i_size
= size
;
4152 clear_nlink(inode
); /* It is unlinked */
4153 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4155 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4156 &shmem_file_operations
);
4163 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4164 * kernel internal. There will be NO LSM permission checks against the
4165 * underlying inode. So users of this interface must do LSM checks at a
4166 * higher layer. The users are the big_key and shm implementations. LSM
4167 * checks are provided at the key or shm level rather than the inode.
4168 * @name: name for dentry (to be seen in /proc/<pid>/maps
4169 * @size: size to be set for the file
4170 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4172 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4174 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4178 * shmem_file_setup - get an unlinked file living in tmpfs
4179 * @name: name for dentry (to be seen in /proc/<pid>/maps
4180 * @size: size to be set for the file
4181 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4183 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4185 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4187 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4190 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4191 * @mnt: the tmpfs mount where the file will be created
4192 * @name: name for dentry (to be seen in /proc/<pid>/maps
4193 * @size: size to be set for the file
4194 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4196 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4197 loff_t size
, unsigned long flags
)
4199 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4201 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4204 * shmem_zero_setup - setup a shared anonymous mapping
4205 * @vma: the vma to be mmapped is prepared by do_mmap
4207 int shmem_zero_setup(struct vm_area_struct
*vma
)
4210 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4213 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4214 * between XFS directory reading and selinux: since this file is only
4215 * accessible to the user through its mapping, use S_PRIVATE flag to
4216 * bypass file security, in the same way as shmem_kernel_file_setup().
4218 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4220 return PTR_ERR(file
);
4224 vma
->vm_file
= file
;
4225 vma
->vm_ops
= &shmem_vm_ops
;
4227 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
4228 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4229 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4230 khugepaged_enter(vma
, vma
->vm_flags
);
4235 EXPORT_SYMBOL_GPL(shmem_zero_setup
);
4238 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4239 * @mapping: the page's address_space
4240 * @index: the page index
4241 * @gfp: the page allocator flags to use if allocating
4243 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4244 * with any new page allocations done using the specified allocation flags.
4245 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4246 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4247 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4249 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4250 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4252 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4253 pgoff_t index
, gfp_t gfp
)
4256 struct inode
*inode
= mapping
->host
;
4260 BUG_ON(!shmem_mapping(mapping
));
4261 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4262 gfp
, NULL
, NULL
, NULL
);
4264 page
= ERR_PTR(error
);
4270 * The tiny !SHMEM case uses ramfs without swap
4272 return read_cache_page_gfp(mapping
, index
, gfp
);
4275 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);