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 static 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 __inc_node_page_state(page
, NR_SHMEM_THPS
);
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
)
845 pgoff_t indices
[PAGEVEC_SIZE
];
850 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
852 while (!mapping_unevictable(mapping
)) {
854 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
855 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
857 pvec
.nr
= find_get_entries(mapping
, index
,
858 PAGEVEC_SIZE
, pvec
.pages
, indices
);
861 index
= indices
[pvec
.nr
- 1] + 1;
862 pagevec_remove_exceptionals(&pvec
);
863 check_move_unevictable_pages(&pvec
);
864 pagevec_release(&pvec
);
870 * Check whether a hole-punch or truncation needs to split a huge page,
871 * returning true if no split was required, or the split has been successful.
873 * Eviction (or truncation to 0 size) should never need to split a huge page;
874 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
875 * head, and then succeeded to trylock on tail.
877 * A split can only succeed when there are no additional references on the
878 * huge page: so the split below relies upon find_get_entries() having stopped
879 * when it found a subpage of the huge page, without getting further references.
881 static bool shmem_punch_compound(struct page
*page
, pgoff_t start
, pgoff_t end
)
883 if (!PageTransCompound(page
))
886 /* Just proceed to delete a huge page wholly within the range punched */
887 if (PageHead(page
) &&
888 page
->index
>= start
&& page
->index
+ HPAGE_PMD_NR
<= end
)
891 /* Try to split huge page, so we can truly punch the hole or truncate */
892 return split_huge_page(page
) >= 0;
896 * Remove range of pages and swap entries from page cache, and free them.
897 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
899 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
902 struct address_space
*mapping
= inode
->i_mapping
;
903 struct shmem_inode_info
*info
= SHMEM_I(inode
);
904 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
905 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
906 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
907 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
909 pgoff_t indices
[PAGEVEC_SIZE
];
910 long nr_swaps_freed
= 0;
915 end
= -1; /* unsigned, so actually very big */
919 while (index
< end
) {
920 pvec
.nr
= find_get_entries(mapping
, index
,
921 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
922 pvec
.pages
, indices
);
925 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
926 struct page
*page
= pvec
.pages
[i
];
932 if (xa_is_value(page
)) {
935 nr_swaps_freed
+= !shmem_free_swap(mapping
,
940 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
942 if (!trylock_page(page
))
945 if ((!unfalloc
|| !PageUptodate(page
)) &&
946 page_mapping(page
) == mapping
) {
947 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
948 if (shmem_punch_compound(page
, start
, end
))
949 truncate_inode_page(mapping
, page
);
953 pagevec_remove_exceptionals(&pvec
);
954 pagevec_release(&pvec
);
960 struct page
*page
= NULL
;
961 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
963 unsigned int top
= PAGE_SIZE
;
968 zero_user_segment(page
, partial_start
, top
);
969 set_page_dirty(page
);
975 struct page
*page
= NULL
;
976 shmem_getpage(inode
, end
, &page
, SGP_READ
);
978 zero_user_segment(page
, 0, partial_end
);
979 set_page_dirty(page
);
988 while (index
< end
) {
991 pvec
.nr
= find_get_entries(mapping
, index
,
992 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
993 pvec
.pages
, indices
);
995 /* If all gone or hole-punch or unfalloc, we're done */
996 if (index
== start
|| end
!= -1)
998 /* But if truncating, restart to make sure all gone */
1002 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
1003 struct page
*page
= pvec
.pages
[i
];
1009 if (xa_is_value(page
)) {
1012 if (shmem_free_swap(mapping
, index
, page
)) {
1013 /* Swap was replaced by page: retry */
1023 if (!unfalloc
|| !PageUptodate(page
)) {
1024 if (page_mapping(page
) != mapping
) {
1025 /* Page was replaced by swap: retry */
1030 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
1031 if (shmem_punch_compound(page
, start
, end
))
1032 truncate_inode_page(mapping
, page
);
1033 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1034 /* Wipe the page and don't get stuck */
1035 clear_highpage(page
);
1036 flush_dcache_page(page
);
1037 set_page_dirty(page
);
1039 round_up(start
, HPAGE_PMD_NR
))
1045 pagevec_remove_exceptionals(&pvec
);
1046 pagevec_release(&pvec
);
1050 spin_lock_irq(&info
->lock
);
1051 info
->swapped
-= nr_swaps_freed
;
1052 shmem_recalc_inode(inode
);
1053 spin_unlock_irq(&info
->lock
);
1056 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1058 shmem_undo_range(inode
, lstart
, lend
, false);
1059 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1061 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1063 static int shmem_getattr(const struct path
*path
, struct kstat
*stat
,
1064 u32 request_mask
, unsigned int query_flags
)
1066 struct inode
*inode
= path
->dentry
->d_inode
;
1067 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1068 struct shmem_sb_info
*sb_info
= SHMEM_SB(inode
->i_sb
);
1070 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1071 spin_lock_irq(&info
->lock
);
1072 shmem_recalc_inode(inode
);
1073 spin_unlock_irq(&info
->lock
);
1075 generic_fillattr(inode
, stat
);
1077 if (is_huge_enabled(sb_info
))
1078 stat
->blksize
= HPAGE_PMD_SIZE
;
1083 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1085 struct inode
*inode
= d_inode(dentry
);
1086 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1087 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1090 error
= setattr_prepare(dentry
, attr
);
1094 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1095 loff_t oldsize
= inode
->i_size
;
1096 loff_t newsize
= attr
->ia_size
;
1098 /* protected by i_mutex */
1099 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1100 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1103 if (newsize
!= oldsize
) {
1104 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1108 i_size_write(inode
, newsize
);
1109 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1111 if (newsize
<= oldsize
) {
1112 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1113 if (oldsize
> holebegin
)
1114 unmap_mapping_range(inode
->i_mapping
,
1117 shmem_truncate_range(inode
,
1118 newsize
, (loff_t
)-1);
1119 /* unmap again to remove racily COWed private pages */
1120 if (oldsize
> holebegin
)
1121 unmap_mapping_range(inode
->i_mapping
,
1125 * Part of the huge page can be beyond i_size: subject
1126 * to shrink under memory pressure.
1128 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1129 spin_lock(&sbinfo
->shrinklist_lock
);
1131 * _careful to defend against unlocked access to
1132 * ->shrink_list in shmem_unused_huge_shrink()
1134 if (list_empty_careful(&info
->shrinklist
)) {
1135 list_add_tail(&info
->shrinklist
,
1136 &sbinfo
->shrinklist
);
1137 sbinfo
->shrinklist_len
++;
1139 spin_unlock(&sbinfo
->shrinklist_lock
);
1144 setattr_copy(inode
, attr
);
1145 if (attr
->ia_valid
& ATTR_MODE
)
1146 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1150 static void shmem_evict_inode(struct inode
*inode
)
1152 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1153 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1155 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1156 shmem_unacct_size(info
->flags
, inode
->i_size
);
1158 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1159 if (!list_empty(&info
->shrinklist
)) {
1160 spin_lock(&sbinfo
->shrinklist_lock
);
1161 if (!list_empty(&info
->shrinklist
)) {
1162 list_del_init(&info
->shrinklist
);
1163 sbinfo
->shrinklist_len
--;
1165 spin_unlock(&sbinfo
->shrinklist_lock
);
1167 while (!list_empty(&info
->swaplist
)) {
1168 /* Wait while shmem_unuse() is scanning this inode... */
1169 wait_var_event(&info
->stop_eviction
,
1170 !atomic_read(&info
->stop_eviction
));
1171 mutex_lock(&shmem_swaplist_mutex
);
1172 /* ...but beware of the race if we peeked too early */
1173 if (!atomic_read(&info
->stop_eviction
))
1174 list_del_init(&info
->swaplist
);
1175 mutex_unlock(&shmem_swaplist_mutex
);
1179 simple_xattrs_free(&info
->xattrs
);
1180 WARN_ON(inode
->i_blocks
);
1181 shmem_free_inode(inode
->i_sb
);
1185 extern struct swap_info_struct
*swap_info
[];
1187 static int shmem_find_swap_entries(struct address_space
*mapping
,
1188 pgoff_t start
, unsigned int nr_entries
,
1189 struct page
**entries
, pgoff_t
*indices
,
1190 unsigned int type
, bool frontswap
)
1192 XA_STATE(xas
, &mapping
->i_pages
, start
);
1195 unsigned int ret
= 0;
1201 xas_for_each(&xas
, page
, ULONG_MAX
) {
1202 if (xas_retry(&xas
, page
))
1205 if (!xa_is_value(page
))
1208 entry
= radix_to_swp_entry(page
);
1209 if (swp_type(entry
) != type
)
1212 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1215 indices
[ret
] = xas
.xa_index
;
1216 entries
[ret
] = page
;
1218 if (need_resched()) {
1222 if (++ret
== nr_entries
)
1231 * Move the swapped pages for an inode to page cache. Returns the count
1232 * of pages swapped in, or the error in case of failure.
1234 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1240 struct address_space
*mapping
= inode
->i_mapping
;
1242 for (i
= 0; i
< pvec
.nr
; i
++) {
1243 struct page
*page
= pvec
.pages
[i
];
1245 if (!xa_is_value(page
))
1247 error
= shmem_swapin_page(inode
, indices
[i
],
1249 mapping_gfp_mask(mapping
),
1256 if (error
== -ENOMEM
)
1260 return error
? error
: ret
;
1264 * If swap found in inode, free it and move page from swapcache to filecache.
1266 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1267 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1269 struct address_space
*mapping
= inode
->i_mapping
;
1271 struct pagevec pvec
;
1272 pgoff_t indices
[PAGEVEC_SIZE
];
1273 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1276 pagevec_init(&pvec
);
1278 unsigned int nr_entries
= PAGEVEC_SIZE
;
1280 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1281 nr_entries
= *fs_pages_to_unuse
;
1283 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1284 pvec
.pages
, indices
,
1291 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1295 if (frontswap_partial
) {
1296 *fs_pages_to_unuse
-= ret
;
1297 if (*fs_pages_to_unuse
== 0) {
1298 ret
= FRONTSWAP_PAGES_UNUSED
;
1303 start
= indices
[pvec
.nr
- 1];
1310 * Read all the shared memory data that resides in the swap
1311 * device 'type' back into memory, so the swap device can be
1314 int shmem_unuse(unsigned int type
, bool frontswap
,
1315 unsigned long *fs_pages_to_unuse
)
1317 struct shmem_inode_info
*info
, *next
;
1320 if (list_empty(&shmem_swaplist
))
1323 mutex_lock(&shmem_swaplist_mutex
);
1324 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1325 if (!info
->swapped
) {
1326 list_del_init(&info
->swaplist
);
1330 * Drop the swaplist mutex while searching the inode for swap;
1331 * but before doing so, make sure shmem_evict_inode() will not
1332 * remove placeholder inode from swaplist, nor let it be freed
1333 * (igrab() would protect from unlink, but not from unmount).
1335 atomic_inc(&info
->stop_eviction
);
1336 mutex_unlock(&shmem_swaplist_mutex
);
1338 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1342 mutex_lock(&shmem_swaplist_mutex
);
1343 next
= list_next_entry(info
, swaplist
);
1345 list_del_init(&info
->swaplist
);
1346 if (atomic_dec_and_test(&info
->stop_eviction
))
1347 wake_up_var(&info
->stop_eviction
);
1351 mutex_unlock(&shmem_swaplist_mutex
);
1357 * Move the page from the page cache to the swap cache.
1359 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1361 struct shmem_inode_info
*info
;
1362 struct address_space
*mapping
;
1363 struct inode
*inode
;
1367 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1368 BUG_ON(!PageLocked(page
));
1369 mapping
= page
->mapping
;
1370 index
= page
->index
;
1371 inode
= mapping
->host
;
1372 info
= SHMEM_I(inode
);
1373 if (info
->flags
& VM_LOCKED
)
1375 if (!total_swap_pages
)
1379 * Our capabilities prevent regular writeback or sync from ever calling
1380 * shmem_writepage; but a stacking filesystem might use ->writepage of
1381 * its underlying filesystem, in which case tmpfs should write out to
1382 * swap only in response to memory pressure, and not for the writeback
1385 if (!wbc
->for_reclaim
) {
1386 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1391 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1392 * value into swapfile.c, the only way we can correctly account for a
1393 * fallocated page arriving here is now to initialize it and write it.
1395 * That's okay for a page already fallocated earlier, but if we have
1396 * not yet completed the fallocation, then (a) we want to keep track
1397 * of this page in case we have to undo it, and (b) it may not be a
1398 * good idea to continue anyway, once we're pushing into swap. So
1399 * reactivate the page, and let shmem_fallocate() quit when too many.
1401 if (!PageUptodate(page
)) {
1402 if (inode
->i_private
) {
1403 struct shmem_falloc
*shmem_falloc
;
1404 spin_lock(&inode
->i_lock
);
1405 shmem_falloc
= inode
->i_private
;
1407 !shmem_falloc
->waitq
&&
1408 index
>= shmem_falloc
->start
&&
1409 index
< shmem_falloc
->next
)
1410 shmem_falloc
->nr_unswapped
++;
1412 shmem_falloc
= NULL
;
1413 spin_unlock(&inode
->i_lock
);
1417 clear_highpage(page
);
1418 flush_dcache_page(page
);
1419 SetPageUptodate(page
);
1422 swap
= get_swap_page(page
);
1427 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1428 * if it's not already there. Do it now before the page is
1429 * moved to swap cache, when its pagelock no longer protects
1430 * the inode from eviction. But don't unlock the mutex until
1431 * we've incremented swapped, because shmem_unuse_inode() will
1432 * prune a !swapped inode from the swaplist under this mutex.
1434 mutex_lock(&shmem_swaplist_mutex
);
1435 if (list_empty(&info
->swaplist
))
1436 list_add(&info
->swaplist
, &shmem_swaplist
);
1438 if (add_to_swap_cache(page
, swap
,
1439 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
,
1441 spin_lock_irq(&info
->lock
);
1442 shmem_recalc_inode(inode
);
1444 spin_unlock_irq(&info
->lock
);
1446 swap_shmem_alloc(swap
);
1447 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1449 mutex_unlock(&shmem_swaplist_mutex
);
1450 BUG_ON(page_mapped(page
));
1451 swap_writepage(page
, wbc
);
1455 mutex_unlock(&shmem_swaplist_mutex
);
1456 put_swap_page(page
, swap
);
1458 set_page_dirty(page
);
1459 if (wbc
->for_reclaim
)
1460 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1465 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1466 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1470 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1471 return; /* show nothing */
1473 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1475 seq_printf(seq
, ",mpol=%s", buffer
);
1478 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1480 struct mempolicy
*mpol
= NULL
;
1482 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1483 mpol
= sbinfo
->mpol
;
1485 spin_unlock(&sbinfo
->stat_lock
);
1489 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1490 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1493 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1497 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1499 #define vm_policy vm_private_data
1502 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1503 struct shmem_inode_info
*info
, pgoff_t index
)
1505 /* Create a pseudo vma that just contains the policy */
1506 vma_init(vma
, NULL
);
1507 /* Bias interleave by inode number to distribute better across nodes */
1508 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1509 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1512 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1514 /* Drop reference taken by mpol_shared_policy_lookup() */
1515 mpol_cond_put(vma
->vm_policy
);
1518 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1519 struct shmem_inode_info
*info
, pgoff_t index
)
1521 struct vm_area_struct pvma
;
1523 struct vm_fault vmf
;
1525 shmem_pseudo_vma_init(&pvma
, info
, index
);
1528 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1529 shmem_pseudo_vma_destroy(&pvma
);
1534 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1535 struct shmem_inode_info
*info
, pgoff_t index
)
1537 struct vm_area_struct pvma
;
1538 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1542 hindex
= round_down(index
, HPAGE_PMD_NR
);
1543 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1547 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1548 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1549 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1550 shmem_pseudo_vma_destroy(&pvma
);
1552 prep_transhuge_page(page
);
1554 count_vm_event(THP_FILE_FALLBACK
);
1558 static struct page
*shmem_alloc_page(gfp_t gfp
,
1559 struct shmem_inode_info
*info
, pgoff_t index
)
1561 struct vm_area_struct pvma
;
1564 shmem_pseudo_vma_init(&pvma
, info
, index
);
1565 page
= alloc_page_vma(gfp
, &pvma
, 0);
1566 shmem_pseudo_vma_destroy(&pvma
);
1571 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1572 struct inode
*inode
,
1573 pgoff_t index
, bool huge
)
1575 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1580 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1582 nr
= huge
? HPAGE_PMD_NR
: 1;
1584 if (!shmem_inode_acct_block(inode
, nr
))
1588 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1590 page
= shmem_alloc_page(gfp
, info
, index
);
1592 __SetPageLocked(page
);
1593 __SetPageSwapBacked(page
);
1598 shmem_inode_unacct_blocks(inode
, nr
);
1600 return ERR_PTR(err
);
1604 * When a page is moved from swapcache to shmem filecache (either by the
1605 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1606 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1607 * ignorance of the mapping it belongs to. If that mapping has special
1608 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1609 * we may need to copy to a suitable page before moving to filecache.
1611 * In a future release, this may well be extended to respect cpuset and
1612 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1613 * but for now it is a simple matter of zone.
1615 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1617 return page_zonenum(page
) > gfp_zone(gfp
);
1620 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1621 struct shmem_inode_info
*info
, pgoff_t index
)
1623 struct page
*oldpage
, *newpage
;
1624 struct address_space
*swap_mapping
;
1630 entry
.val
= page_private(oldpage
);
1631 swap_index
= swp_offset(entry
);
1632 swap_mapping
= page_mapping(oldpage
);
1635 * We have arrived here because our zones are constrained, so don't
1636 * limit chance of success by further cpuset and node constraints.
1638 gfp
&= ~GFP_CONSTRAINT_MASK
;
1639 newpage
= shmem_alloc_page(gfp
, info
, index
);
1644 copy_highpage(newpage
, oldpage
);
1645 flush_dcache_page(newpage
);
1647 __SetPageLocked(newpage
);
1648 __SetPageSwapBacked(newpage
);
1649 SetPageUptodate(newpage
);
1650 set_page_private(newpage
, entry
.val
);
1651 SetPageSwapCache(newpage
);
1654 * Our caller will very soon move newpage out of swapcache, but it's
1655 * a nice clean interface for us to replace oldpage by newpage there.
1657 xa_lock_irq(&swap_mapping
->i_pages
);
1658 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1660 mem_cgroup_migrate(oldpage
, newpage
);
1661 __inc_lruvec_page_state(newpage
, NR_FILE_PAGES
);
1662 __dec_lruvec_page_state(oldpage
, NR_FILE_PAGES
);
1664 xa_unlock_irq(&swap_mapping
->i_pages
);
1666 if (unlikely(error
)) {
1668 * Is this possible? I think not, now that our callers check
1669 * both PageSwapCache and page_private after getting page lock;
1670 * but be defensive. Reverse old to newpage for clear and free.
1674 lru_cache_add(newpage
);
1678 ClearPageSwapCache(oldpage
);
1679 set_page_private(oldpage
, 0);
1681 unlock_page(oldpage
);
1688 * Swap in the page pointed to by *pagep.
1689 * Caller has to make sure that *pagep contains a valid swapped page.
1690 * Returns 0 and the page in pagep if success. On failure, returns the
1691 * error code and NULL in *pagep.
1693 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1694 struct page
**pagep
, enum sgp_type sgp
,
1695 gfp_t gfp
, struct vm_area_struct
*vma
,
1696 vm_fault_t
*fault_type
)
1698 struct address_space
*mapping
= inode
->i_mapping
;
1699 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1700 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1705 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1706 swap
= radix_to_swp_entry(*pagep
);
1709 /* Look it up and read it in.. */
1710 page
= lookup_swap_cache(swap
, NULL
, 0);
1712 /* Or update major stats only when swapin succeeds?? */
1714 *fault_type
|= VM_FAULT_MAJOR
;
1715 count_vm_event(PGMAJFAULT
);
1716 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1718 /* Here we actually start the io */
1719 page
= shmem_swapin(swap
, gfp
, info
, index
);
1726 /* We have to do this with page locked to prevent races */
1728 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1729 !shmem_confirm_swap(mapping
, index
, swap
)) {
1733 if (!PageUptodate(page
)) {
1737 wait_on_page_writeback(page
);
1740 * Some architectures may have to restore extra metadata to the
1741 * physical page after reading from swap.
1743 arch_swap_restore(swap
, page
);
1745 if (shmem_should_replace_page(page
, gfp
)) {
1746 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1751 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1752 swp_to_radix_entry(swap
), gfp
,
1757 spin_lock_irq(&info
->lock
);
1759 shmem_recalc_inode(inode
);
1760 spin_unlock_irq(&info
->lock
);
1762 if (sgp
== SGP_WRITE
)
1763 mark_page_accessed(page
);
1765 delete_from_swap_cache(page
);
1766 set_page_dirty(page
);
1772 if (!shmem_confirm_swap(mapping
, index
, swap
))
1784 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1786 * If we allocate a new one we do not mark it dirty. That's up to the
1787 * vm. If we swap it in we mark it dirty since we also free the swap
1788 * entry since a page cannot live in both the swap and page cache.
1790 * vmf and fault_type are only supplied by shmem_fault:
1791 * otherwise they are NULL.
1793 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1794 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1795 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1796 vm_fault_t
*fault_type
)
1798 struct address_space
*mapping
= inode
->i_mapping
;
1799 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1800 struct shmem_sb_info
*sbinfo
;
1801 struct mm_struct
*charge_mm
;
1803 enum sgp_type sgp_huge
= sgp
;
1804 pgoff_t hindex
= index
;
1809 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1811 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1814 if (sgp
<= SGP_CACHE
&&
1815 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1819 sbinfo
= SHMEM_SB(inode
->i_sb
);
1820 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1822 page
= find_lock_entry(mapping
, index
);
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 (mapping
->a_ops
!= &shmem_aops
)
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 page
= shmem_alloc_and_acct_page(gfp
, inode
, index
, true);
1893 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1899 error
= PTR_ERR(page
);
1901 if (error
!= -ENOSPC
)
1904 * Try to reclaim some space by splitting a huge page
1905 * beyond i_size on the filesystem.
1910 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1911 if (ret
== SHRINK_STOP
)
1919 if (PageTransHuge(page
))
1920 hindex
= round_down(index
, HPAGE_PMD_NR
);
1924 if (sgp
== SGP_WRITE
)
1925 __SetPageReferenced(page
);
1927 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1928 NULL
, gfp
& GFP_RECLAIM_MASK
,
1932 lru_cache_add(page
);
1934 spin_lock_irq(&info
->lock
);
1935 info
->alloced
+= compound_nr(page
);
1936 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1937 shmem_recalc_inode(inode
);
1938 spin_unlock_irq(&info
->lock
);
1941 if (PageTransHuge(page
) &&
1942 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1943 hindex
+ HPAGE_PMD_NR
- 1) {
1945 * Part of the huge page is beyond i_size: subject
1946 * to shrink under memory pressure.
1948 spin_lock(&sbinfo
->shrinklist_lock
);
1950 * _careful to defend against unlocked access to
1951 * ->shrink_list in shmem_unused_huge_shrink()
1953 if (list_empty_careful(&info
->shrinklist
)) {
1954 list_add_tail(&info
->shrinklist
,
1955 &sbinfo
->shrinklist
);
1956 sbinfo
->shrinklist_len
++;
1958 spin_unlock(&sbinfo
->shrinklist_lock
);
1962 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1964 if (sgp
== SGP_FALLOC
)
1968 * Let SGP_WRITE caller clear ends if write does not fill page;
1969 * but SGP_FALLOC on a page fallocated earlier must initialize
1970 * it now, lest undo on failure cancel our earlier guarantee.
1972 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1975 for (i
= 0; i
< compound_nr(page
); i
++) {
1976 clear_highpage(page
+ i
);
1977 flush_dcache_page(page
+ i
);
1979 SetPageUptodate(page
);
1982 /* Perhaps the file has been truncated since we checked */
1983 if (sgp
<= SGP_CACHE
&&
1984 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1986 ClearPageDirty(page
);
1987 delete_from_page_cache(page
);
1988 spin_lock_irq(&info
->lock
);
1989 shmem_recalc_inode(inode
);
1990 spin_unlock_irq(&info
->lock
);
1996 *pagep
= page
+ index
- hindex
;
2003 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
2005 if (PageTransHuge(page
)) {
2015 if (error
== -ENOSPC
&& !once
++) {
2016 spin_lock_irq(&info
->lock
);
2017 shmem_recalc_inode(inode
);
2018 spin_unlock_irq(&info
->lock
);
2021 if (error
== -EEXIST
)
2027 * This is like autoremove_wake_function, but it removes the wait queue
2028 * entry unconditionally - even if something else had already woken the
2031 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2033 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2034 list_del_init(&wait
->entry
);
2038 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2040 struct vm_area_struct
*vma
= vmf
->vma
;
2041 struct inode
*inode
= file_inode(vma
->vm_file
);
2042 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2045 vm_fault_t ret
= VM_FAULT_LOCKED
;
2048 * Trinity finds that probing a hole which tmpfs is punching can
2049 * prevent the hole-punch from ever completing: which in turn
2050 * locks writers out with its hold on i_mutex. So refrain from
2051 * faulting pages into the hole while it's being punched. Although
2052 * shmem_undo_range() does remove the additions, it may be unable to
2053 * keep up, as each new page needs its own unmap_mapping_range() call,
2054 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2056 * It does not matter if we sometimes reach this check just before the
2057 * hole-punch begins, so that one fault then races with the punch:
2058 * we just need to make racing faults a rare case.
2060 * The implementation below would be much simpler if we just used a
2061 * standard mutex or completion: but we cannot take i_mutex in fault,
2062 * and bloating every shmem inode for this unlikely case would be sad.
2064 if (unlikely(inode
->i_private
)) {
2065 struct shmem_falloc
*shmem_falloc
;
2067 spin_lock(&inode
->i_lock
);
2068 shmem_falloc
= inode
->i_private
;
2070 shmem_falloc
->waitq
&&
2071 vmf
->pgoff
>= shmem_falloc
->start
&&
2072 vmf
->pgoff
< shmem_falloc
->next
) {
2074 wait_queue_head_t
*shmem_falloc_waitq
;
2075 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2077 ret
= VM_FAULT_NOPAGE
;
2078 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2080 ret
= VM_FAULT_RETRY
;
2082 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2083 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2084 TASK_UNINTERRUPTIBLE
);
2085 spin_unlock(&inode
->i_lock
);
2089 * shmem_falloc_waitq points into the shmem_fallocate()
2090 * stack of the hole-punching task: shmem_falloc_waitq
2091 * is usually invalid by the time we reach here, but
2092 * finish_wait() does not dereference it in that case;
2093 * though i_lock needed lest racing with wake_up_all().
2095 spin_lock(&inode
->i_lock
);
2096 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2097 spin_unlock(&inode
->i_lock
);
2103 spin_unlock(&inode
->i_lock
);
2108 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2109 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2111 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2114 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2115 gfp
, vma
, vmf
, &ret
);
2117 return vmf_error(err
);
2121 unsigned long shmem_get_unmapped_area(struct file
*file
,
2122 unsigned long uaddr
, unsigned long len
,
2123 unsigned long pgoff
, unsigned long flags
)
2125 unsigned long (*get_area
)(struct file
*,
2126 unsigned long, unsigned long, unsigned long, unsigned long);
2128 unsigned long offset
;
2129 unsigned long inflated_len
;
2130 unsigned long inflated_addr
;
2131 unsigned long inflated_offset
;
2133 if (len
> TASK_SIZE
)
2136 get_area
= current
->mm
->get_unmapped_area
;
2137 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2139 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2141 if (IS_ERR_VALUE(addr
))
2143 if (addr
& ~PAGE_MASK
)
2145 if (addr
> TASK_SIZE
- len
)
2148 if (shmem_huge
== SHMEM_HUGE_DENY
)
2150 if (len
< HPAGE_PMD_SIZE
)
2152 if (flags
& MAP_FIXED
)
2155 * Our priority is to support MAP_SHARED mapped hugely;
2156 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2157 * But if caller specified an address hint and we allocated area there
2158 * successfully, respect that as before.
2163 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2164 struct super_block
*sb
;
2167 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2168 sb
= file_inode(file
)->i_sb
;
2171 * Called directly from mm/mmap.c, or drivers/char/mem.c
2172 * for "/dev/zero", to create a shared anonymous object.
2174 if (IS_ERR(shm_mnt
))
2176 sb
= shm_mnt
->mnt_sb
;
2178 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2182 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2183 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2185 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2188 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2189 if (inflated_len
> TASK_SIZE
)
2191 if (inflated_len
< len
)
2194 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2195 if (IS_ERR_VALUE(inflated_addr
))
2197 if (inflated_addr
& ~PAGE_MASK
)
2200 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2201 inflated_addr
+= offset
- inflated_offset
;
2202 if (inflated_offset
> offset
)
2203 inflated_addr
+= HPAGE_PMD_SIZE
;
2205 if (inflated_addr
> TASK_SIZE
- len
)
2207 return inflated_addr
;
2211 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2213 struct inode
*inode
= file_inode(vma
->vm_file
);
2214 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2217 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2220 struct inode
*inode
= file_inode(vma
->vm_file
);
2223 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2224 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2228 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2230 struct inode
*inode
= file_inode(file
);
2231 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2232 int retval
= -ENOMEM
;
2235 * What serializes the accesses to info->flags?
2236 * ipc_lock_object() when called from shmctl_do_lock(),
2237 * no serialization needed when called from shm_destroy().
2239 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2240 if (!user_shm_lock(inode
->i_size
, user
))
2242 info
->flags
|= VM_LOCKED
;
2243 mapping_set_unevictable(file
->f_mapping
);
2245 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2246 user_shm_unlock(inode
->i_size
, user
);
2247 info
->flags
&= ~VM_LOCKED
;
2248 mapping_clear_unevictable(file
->f_mapping
);
2256 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2258 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2260 if (info
->seals
& F_SEAL_FUTURE_WRITE
) {
2262 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2263 * "future write" seal active.
2265 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_WRITE
))
2269 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2270 * MAP_SHARED and read-only, take care to not allow mprotect to
2271 * revert protections on such mappings. Do this only for shared
2272 * mappings. For private mappings, don't need to mask
2273 * VM_MAYWRITE as we still want them to be COW-writable.
2275 if (vma
->vm_flags
& VM_SHARED
)
2276 vma
->vm_flags
&= ~(VM_MAYWRITE
);
2279 /* arm64 - allow memory tagging on RAM-based files */
2280 vma
->vm_flags
|= VM_MTE_ALLOWED
;
2282 file_accessed(file
);
2283 vma
->vm_ops
= &shmem_vm_ops
;
2284 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
2285 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2286 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2287 khugepaged_enter(vma
, vma
->vm_flags
);
2292 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2293 umode_t mode
, dev_t dev
, unsigned long flags
)
2295 struct inode
*inode
;
2296 struct shmem_inode_info
*info
;
2297 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2300 if (shmem_reserve_inode(sb
, &ino
))
2303 inode
= new_inode(sb
);
2306 inode_init_owner(inode
, dir
, mode
);
2307 inode
->i_blocks
= 0;
2308 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2309 inode
->i_generation
= prandom_u32();
2310 info
= SHMEM_I(inode
);
2311 memset(info
, 0, (char *)inode
- (char *)info
);
2312 spin_lock_init(&info
->lock
);
2313 atomic_set(&info
->stop_eviction
, 0);
2314 info
->seals
= F_SEAL_SEAL
;
2315 info
->flags
= flags
& VM_NORESERVE
;
2316 INIT_LIST_HEAD(&info
->shrinklist
);
2317 INIT_LIST_HEAD(&info
->swaplist
);
2318 simple_xattrs_init(&info
->xattrs
);
2319 cache_no_acl(inode
);
2321 switch (mode
& S_IFMT
) {
2323 inode
->i_op
= &shmem_special_inode_operations
;
2324 init_special_inode(inode
, mode
, dev
);
2327 inode
->i_mapping
->a_ops
= &shmem_aops
;
2328 inode
->i_op
= &shmem_inode_operations
;
2329 inode
->i_fop
= &shmem_file_operations
;
2330 mpol_shared_policy_init(&info
->policy
,
2331 shmem_get_sbmpol(sbinfo
));
2335 /* Some things misbehave if size == 0 on a directory */
2336 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2337 inode
->i_op
= &shmem_dir_inode_operations
;
2338 inode
->i_fop
= &simple_dir_operations
;
2342 * Must not load anything in the rbtree,
2343 * mpol_free_shared_policy will not be called.
2345 mpol_shared_policy_init(&info
->policy
, NULL
);
2349 lockdep_annotate_inode_mutex_key(inode
);
2351 shmem_free_inode(sb
);
2355 bool shmem_mapping(struct address_space
*mapping
)
2357 return mapping
->a_ops
== &shmem_aops
;
2360 static int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2362 struct vm_area_struct
*dst_vma
,
2363 unsigned long dst_addr
,
2364 unsigned long src_addr
,
2366 struct page
**pagep
)
2368 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2369 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2370 struct address_space
*mapping
= inode
->i_mapping
;
2371 gfp_t gfp
= mapping_gfp_mask(mapping
);
2372 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2376 pte_t _dst_pte
, *dst_pte
;
2378 pgoff_t offset
, max_off
;
2381 if (!shmem_inode_acct_block(inode
, 1))
2385 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2387 goto out_unacct_blocks
;
2389 if (!zeropage
) { /* mcopy_atomic */
2390 page_kaddr
= kmap_atomic(page
);
2391 ret
= copy_from_user(page_kaddr
,
2392 (const void __user
*)src_addr
,
2394 kunmap_atomic(page_kaddr
);
2396 /* fallback to copy_from_user outside mmap_lock */
2397 if (unlikely(ret
)) {
2399 shmem_inode_unacct_blocks(inode
, 1);
2400 /* don't free the page */
2403 } else { /* mfill_zeropage_atomic */
2404 clear_highpage(page
);
2411 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2412 __SetPageLocked(page
);
2413 __SetPageSwapBacked(page
);
2414 __SetPageUptodate(page
);
2417 offset
= linear_page_index(dst_vma
, dst_addr
);
2418 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2419 if (unlikely(offset
>= max_off
))
2422 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2423 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2427 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2428 if (dst_vma
->vm_flags
& VM_WRITE
)
2429 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2432 * We don't set the pte dirty if the vma has no
2433 * VM_WRITE permission, so mark the page dirty or it
2434 * could be freed from under us. We could do it
2435 * unconditionally before unlock_page(), but doing it
2436 * only if VM_WRITE is not set is faster.
2438 set_page_dirty(page
);
2441 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2444 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2445 if (unlikely(offset
>= max_off
))
2446 goto out_release_unlock
;
2449 if (!pte_none(*dst_pte
))
2450 goto out_release_unlock
;
2452 lru_cache_add(page
);
2454 spin_lock_irq(&info
->lock
);
2456 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2457 shmem_recalc_inode(inode
);
2458 spin_unlock_irq(&info
->lock
);
2460 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2461 page_add_file_rmap(page
, false);
2462 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2464 /* No need to invalidate - it was non-present before */
2465 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2466 pte_unmap_unlock(dst_pte
, ptl
);
2472 pte_unmap_unlock(dst_pte
, ptl
);
2473 ClearPageDirty(page
);
2474 delete_from_page_cache(page
);
2479 shmem_inode_unacct_blocks(inode
, 1);
2483 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2485 struct vm_area_struct
*dst_vma
,
2486 unsigned long dst_addr
,
2487 unsigned long src_addr
,
2488 struct page
**pagep
)
2490 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2491 dst_addr
, src_addr
, false, pagep
);
2494 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2496 struct vm_area_struct
*dst_vma
,
2497 unsigned long dst_addr
)
2499 struct page
*page
= NULL
;
2501 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2502 dst_addr
, 0, true, &page
);
2506 static const struct inode_operations shmem_symlink_inode_operations
;
2507 static const struct inode_operations shmem_short_symlink_operations
;
2509 #ifdef CONFIG_TMPFS_XATTR
2510 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2512 #define shmem_initxattrs NULL
2516 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2517 loff_t pos
, unsigned len
, unsigned flags
,
2518 struct page
**pagep
, void **fsdata
)
2520 struct inode
*inode
= mapping
->host
;
2521 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2522 pgoff_t index
= pos
>> PAGE_SHIFT
;
2524 /* i_mutex is held by caller */
2525 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2526 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2527 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2529 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2533 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2537 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2538 loff_t pos
, unsigned len
, unsigned copied
,
2539 struct page
*page
, void *fsdata
)
2541 struct inode
*inode
= mapping
->host
;
2543 if (pos
+ copied
> inode
->i_size
)
2544 i_size_write(inode
, pos
+ copied
);
2546 if (!PageUptodate(page
)) {
2547 struct page
*head
= compound_head(page
);
2548 if (PageTransCompound(page
)) {
2551 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2552 if (head
+ i
== page
)
2554 clear_highpage(head
+ i
);
2555 flush_dcache_page(head
+ i
);
2558 if (copied
< PAGE_SIZE
) {
2559 unsigned from
= pos
& (PAGE_SIZE
- 1);
2560 zero_user_segments(page
, 0, from
,
2561 from
+ copied
, PAGE_SIZE
);
2563 SetPageUptodate(head
);
2565 set_page_dirty(page
);
2572 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2574 struct file
*file
= iocb
->ki_filp
;
2575 struct inode
*inode
= file_inode(file
);
2576 struct address_space
*mapping
= inode
->i_mapping
;
2578 unsigned long offset
;
2579 enum sgp_type sgp
= SGP_READ
;
2582 loff_t
*ppos
= &iocb
->ki_pos
;
2585 * Might this read be for a stacking filesystem? Then when reading
2586 * holes of a sparse file, we actually need to allocate those pages,
2587 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2589 if (!iter_is_iovec(to
))
2592 index
= *ppos
>> PAGE_SHIFT
;
2593 offset
= *ppos
& ~PAGE_MASK
;
2596 struct page
*page
= NULL
;
2598 unsigned long nr
, ret
;
2599 loff_t i_size
= i_size_read(inode
);
2601 end_index
= i_size
>> PAGE_SHIFT
;
2602 if (index
> end_index
)
2604 if (index
== end_index
) {
2605 nr
= i_size
& ~PAGE_MASK
;
2610 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2612 if (error
== -EINVAL
)
2617 if (sgp
== SGP_CACHE
)
2618 set_page_dirty(page
);
2623 * We must evaluate after, since reads (unlike writes)
2624 * are called without i_mutex protection against truncate
2627 i_size
= i_size_read(inode
);
2628 end_index
= i_size
>> PAGE_SHIFT
;
2629 if (index
== end_index
) {
2630 nr
= i_size
& ~PAGE_MASK
;
2641 * If users can be writing to this page using arbitrary
2642 * virtual addresses, take care about potential aliasing
2643 * before reading the page on the kernel side.
2645 if (mapping_writably_mapped(mapping
))
2646 flush_dcache_page(page
);
2648 * Mark the page accessed if we read the beginning.
2651 mark_page_accessed(page
);
2653 page
= ZERO_PAGE(0);
2658 * Ok, we have the page, and it's up-to-date, so
2659 * now we can copy it to user space...
2661 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2664 index
+= offset
>> PAGE_SHIFT
;
2665 offset
&= ~PAGE_MASK
;
2668 if (!iov_iter_count(to
))
2677 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2678 file_accessed(file
);
2679 return retval
? retval
: error
;
2683 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2685 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2686 pgoff_t index
, pgoff_t end
, int whence
)
2689 struct pagevec pvec
;
2690 pgoff_t indices
[PAGEVEC_SIZE
];
2694 pagevec_init(&pvec
);
2695 pvec
.nr
= 1; /* start small: we may be there already */
2697 pvec
.nr
= find_get_entries(mapping
, index
,
2698 pvec
.nr
, pvec
.pages
, indices
);
2700 if (whence
== SEEK_DATA
)
2704 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2705 if (index
< indices
[i
]) {
2706 if (whence
== SEEK_HOLE
) {
2712 page
= pvec
.pages
[i
];
2713 if (page
&& !xa_is_value(page
)) {
2714 if (!PageUptodate(page
))
2718 (page
&& whence
== SEEK_DATA
) ||
2719 (!page
&& whence
== SEEK_HOLE
)) {
2724 pagevec_remove_exceptionals(&pvec
);
2725 pagevec_release(&pvec
);
2726 pvec
.nr
= PAGEVEC_SIZE
;
2732 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2734 struct address_space
*mapping
= file
->f_mapping
;
2735 struct inode
*inode
= mapping
->host
;
2739 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2740 return generic_file_llseek_size(file
, offset
, whence
,
2741 MAX_LFS_FILESIZE
, i_size_read(inode
));
2743 /* We're holding i_mutex so we can access i_size directly */
2745 if (offset
< 0 || offset
>= inode
->i_size
)
2748 start
= offset
>> PAGE_SHIFT
;
2749 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2750 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2751 new_offset
<<= PAGE_SHIFT
;
2752 if (new_offset
> offset
) {
2753 if (new_offset
< inode
->i_size
)
2754 offset
= new_offset
;
2755 else if (whence
== SEEK_DATA
)
2758 offset
= inode
->i_size
;
2763 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2764 inode_unlock(inode
);
2768 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2771 struct inode
*inode
= file_inode(file
);
2772 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2773 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2774 struct shmem_falloc shmem_falloc
;
2775 pgoff_t start
, index
, end
;
2778 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2783 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2784 struct address_space
*mapping
= file
->f_mapping
;
2785 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2786 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2787 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2789 /* protected by i_mutex */
2790 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2795 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2796 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2797 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2798 spin_lock(&inode
->i_lock
);
2799 inode
->i_private
= &shmem_falloc
;
2800 spin_unlock(&inode
->i_lock
);
2802 if ((u64
)unmap_end
> (u64
)unmap_start
)
2803 unmap_mapping_range(mapping
, unmap_start
,
2804 1 + unmap_end
- unmap_start
, 0);
2805 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2806 /* No need to unmap again: hole-punching leaves COWed pages */
2808 spin_lock(&inode
->i_lock
);
2809 inode
->i_private
= NULL
;
2810 wake_up_all(&shmem_falloc_waitq
);
2811 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2812 spin_unlock(&inode
->i_lock
);
2817 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2818 error
= inode_newsize_ok(inode
, offset
+ len
);
2822 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2827 start
= offset
>> PAGE_SHIFT
;
2828 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2829 /* Try to avoid a swapstorm if len is impossible to satisfy */
2830 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2835 shmem_falloc
.waitq
= NULL
;
2836 shmem_falloc
.start
= start
;
2837 shmem_falloc
.next
= start
;
2838 shmem_falloc
.nr_falloced
= 0;
2839 shmem_falloc
.nr_unswapped
= 0;
2840 spin_lock(&inode
->i_lock
);
2841 inode
->i_private
= &shmem_falloc
;
2842 spin_unlock(&inode
->i_lock
);
2844 for (index
= start
; index
< end
; index
++) {
2848 * Good, the fallocate(2) manpage permits EINTR: we may have
2849 * been interrupted because we are using up too much memory.
2851 if (signal_pending(current
))
2853 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2856 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2858 /* Remove the !PageUptodate pages we added */
2859 if (index
> start
) {
2860 shmem_undo_range(inode
,
2861 (loff_t
)start
<< PAGE_SHIFT
,
2862 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2868 * Inform shmem_writepage() how far we have reached.
2869 * No need for lock or barrier: we have the page lock.
2871 shmem_falloc
.next
++;
2872 if (!PageUptodate(page
))
2873 shmem_falloc
.nr_falloced
++;
2876 * If !PageUptodate, leave it that way so that freeable pages
2877 * can be recognized if we need to rollback on error later.
2878 * But set_page_dirty so that memory pressure will swap rather
2879 * than free the pages we are allocating (and SGP_CACHE pages
2880 * might still be clean: we now need to mark those dirty too).
2882 set_page_dirty(page
);
2888 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2889 i_size_write(inode
, offset
+ len
);
2890 inode
->i_ctime
= current_time(inode
);
2892 spin_lock(&inode
->i_lock
);
2893 inode
->i_private
= NULL
;
2894 spin_unlock(&inode
->i_lock
);
2896 inode_unlock(inode
);
2900 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2902 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2904 buf
->f_type
= TMPFS_MAGIC
;
2905 buf
->f_bsize
= PAGE_SIZE
;
2906 buf
->f_namelen
= NAME_MAX
;
2907 if (sbinfo
->max_blocks
) {
2908 buf
->f_blocks
= sbinfo
->max_blocks
;
2910 buf
->f_bfree
= sbinfo
->max_blocks
-
2911 percpu_counter_sum(&sbinfo
->used_blocks
);
2913 if (sbinfo
->max_inodes
) {
2914 buf
->f_files
= sbinfo
->max_inodes
;
2915 buf
->f_ffree
= sbinfo
->free_inodes
;
2917 /* else leave those fields 0 like simple_statfs */
2922 * File creation. Allocate an inode, and we're done..
2925 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2927 struct inode
*inode
;
2928 int error
= -ENOSPC
;
2930 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2932 error
= simple_acl_create(dir
, inode
);
2935 error
= security_inode_init_security(inode
, dir
,
2937 shmem_initxattrs
, NULL
);
2938 if (error
&& error
!= -EOPNOTSUPP
)
2942 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2943 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2944 d_instantiate(dentry
, inode
);
2945 dget(dentry
); /* Extra count - pin the dentry in core */
2954 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2956 struct inode
*inode
;
2957 int error
= -ENOSPC
;
2959 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2961 error
= security_inode_init_security(inode
, dir
,
2963 shmem_initxattrs
, NULL
);
2964 if (error
&& error
!= -EOPNOTSUPP
)
2966 error
= simple_acl_create(dir
, inode
);
2969 d_tmpfile(dentry
, inode
);
2977 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2981 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2987 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2990 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2996 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2998 struct inode
*inode
= d_inode(old_dentry
);
3002 * No ordinary (disk based) filesystem counts links as inodes;
3003 * but each new link needs a new dentry, pinning lowmem, and
3004 * tmpfs dentries cannot be pruned until they are unlinked.
3005 * But if an O_TMPFILE file is linked into the tmpfs, the
3006 * first link must skip that, to get the accounting right.
3008 if (inode
->i_nlink
) {
3009 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
3014 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3015 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3017 ihold(inode
); /* New dentry reference */
3018 dget(dentry
); /* Extra pinning count for the created dentry */
3019 d_instantiate(dentry
, inode
);
3024 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3026 struct inode
*inode
= d_inode(dentry
);
3028 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3029 shmem_free_inode(inode
->i_sb
);
3031 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3032 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3034 dput(dentry
); /* Undo the count from "create" - this does all the work */
3038 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3040 if (!simple_empty(dentry
))
3043 drop_nlink(d_inode(dentry
));
3045 return shmem_unlink(dir
, dentry
);
3048 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3050 bool old_is_dir
= d_is_dir(old_dentry
);
3051 bool new_is_dir
= d_is_dir(new_dentry
);
3053 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3055 drop_nlink(old_dir
);
3058 drop_nlink(new_dir
);
3062 old_dir
->i_ctime
= old_dir
->i_mtime
=
3063 new_dir
->i_ctime
= new_dir
->i_mtime
=
3064 d_inode(old_dentry
)->i_ctime
=
3065 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3070 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3072 struct dentry
*whiteout
;
3075 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3079 error
= shmem_mknod(old_dir
, whiteout
,
3080 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3086 * Cheat and hash the whiteout while the old dentry is still in
3087 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3089 * d_lookup() will consistently find one of them at this point,
3090 * not sure which one, but that isn't even important.
3097 * The VFS layer already does all the dentry stuff for rename,
3098 * we just have to decrement the usage count for the target if
3099 * it exists so that the VFS layer correctly free's it when it
3102 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3104 struct inode
*inode
= d_inode(old_dentry
);
3105 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3107 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3110 if (flags
& RENAME_EXCHANGE
)
3111 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3113 if (!simple_empty(new_dentry
))
3116 if (flags
& RENAME_WHITEOUT
) {
3119 error
= shmem_whiteout(old_dir
, old_dentry
);
3124 if (d_really_is_positive(new_dentry
)) {
3125 (void) shmem_unlink(new_dir
, new_dentry
);
3126 if (they_are_dirs
) {
3127 drop_nlink(d_inode(new_dentry
));
3128 drop_nlink(old_dir
);
3130 } else if (they_are_dirs
) {
3131 drop_nlink(old_dir
);
3135 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3136 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3137 old_dir
->i_ctime
= old_dir
->i_mtime
=
3138 new_dir
->i_ctime
= new_dir
->i_mtime
=
3139 inode
->i_ctime
= current_time(old_dir
);
3143 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3147 struct inode
*inode
;
3150 len
= strlen(symname
) + 1;
3151 if (len
> PAGE_SIZE
)
3152 return -ENAMETOOLONG
;
3154 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3159 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3160 shmem_initxattrs
, NULL
);
3161 if (error
&& error
!= -EOPNOTSUPP
) {
3166 inode
->i_size
= len
-1;
3167 if (len
<= SHORT_SYMLINK_LEN
) {
3168 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3169 if (!inode
->i_link
) {
3173 inode
->i_op
= &shmem_short_symlink_operations
;
3175 inode_nohighmem(inode
);
3176 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3181 inode
->i_mapping
->a_ops
= &shmem_aops
;
3182 inode
->i_op
= &shmem_symlink_inode_operations
;
3183 memcpy(page_address(page
), symname
, len
);
3184 SetPageUptodate(page
);
3185 set_page_dirty(page
);
3189 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3190 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3191 d_instantiate(dentry
, inode
);
3196 static void shmem_put_link(void *arg
)
3198 mark_page_accessed(arg
);
3202 static const char *shmem_get_link(struct dentry
*dentry
,
3203 struct inode
*inode
,
3204 struct delayed_call
*done
)
3206 struct page
*page
= NULL
;
3209 page
= find_get_page(inode
->i_mapping
, 0);
3211 return ERR_PTR(-ECHILD
);
3212 if (!PageUptodate(page
)) {
3214 return ERR_PTR(-ECHILD
);
3217 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3219 return ERR_PTR(error
);
3222 set_delayed_call(done
, shmem_put_link
, page
);
3223 return page_address(page
);
3226 #ifdef CONFIG_TMPFS_XATTR
3228 * Superblocks without xattr inode operations may get some security.* xattr
3229 * support from the LSM "for free". As soon as we have any other xattrs
3230 * like ACLs, we also need to implement the security.* handlers at
3231 * filesystem level, though.
3235 * Callback for security_inode_init_security() for acquiring xattrs.
3237 static int shmem_initxattrs(struct inode
*inode
,
3238 const struct xattr
*xattr_array
,
3241 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3242 const struct xattr
*xattr
;
3243 struct simple_xattr
*new_xattr
;
3246 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3247 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3251 len
= strlen(xattr
->name
) + 1;
3252 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3254 if (!new_xattr
->name
) {
3259 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3260 XATTR_SECURITY_PREFIX_LEN
);
3261 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3264 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3270 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3271 struct dentry
*unused
, struct inode
*inode
,
3272 const char *name
, void *buffer
, size_t size
)
3274 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3276 name
= xattr_full_name(handler
, name
);
3277 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3280 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3281 struct dentry
*unused
, struct inode
*inode
,
3282 const char *name
, const void *value
,
3283 size_t size
, int flags
)
3285 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3287 name
= xattr_full_name(handler
, name
);
3288 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3291 static const struct xattr_handler shmem_security_xattr_handler
= {
3292 .prefix
= XATTR_SECURITY_PREFIX
,
3293 .get
= shmem_xattr_handler_get
,
3294 .set
= shmem_xattr_handler_set
,
3297 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3298 .prefix
= XATTR_TRUSTED_PREFIX
,
3299 .get
= shmem_xattr_handler_get
,
3300 .set
= shmem_xattr_handler_set
,
3303 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3304 #ifdef CONFIG_TMPFS_POSIX_ACL
3305 &posix_acl_access_xattr_handler
,
3306 &posix_acl_default_xattr_handler
,
3308 &shmem_security_xattr_handler
,
3309 &shmem_trusted_xattr_handler
,
3313 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3315 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3316 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3318 #endif /* CONFIG_TMPFS_XATTR */
3320 static const struct inode_operations shmem_short_symlink_operations
= {
3321 .get_link
= simple_get_link
,
3322 #ifdef CONFIG_TMPFS_XATTR
3323 .listxattr
= shmem_listxattr
,
3327 static const struct inode_operations shmem_symlink_inode_operations
= {
3328 .get_link
= shmem_get_link
,
3329 #ifdef CONFIG_TMPFS_XATTR
3330 .listxattr
= shmem_listxattr
,
3334 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3336 return ERR_PTR(-ESTALE
);
3339 static int shmem_match(struct inode
*ino
, void *vfh
)
3343 inum
= (inum
<< 32) | fh
[1];
3344 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3347 /* Find any alias of inode, but prefer a hashed alias */
3348 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3350 struct dentry
*alias
= d_find_alias(inode
);
3352 return alias
?: d_find_any_alias(inode
);
3356 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3357 struct fid
*fid
, int fh_len
, int fh_type
)
3359 struct inode
*inode
;
3360 struct dentry
*dentry
= NULL
;
3367 inum
= (inum
<< 32) | fid
->raw
[1];
3369 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3370 shmem_match
, fid
->raw
);
3372 dentry
= shmem_find_alias(inode
);
3379 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3380 struct inode
*parent
)
3384 return FILEID_INVALID
;
3387 if (inode_unhashed(inode
)) {
3388 /* Unfortunately insert_inode_hash is not idempotent,
3389 * so as we hash inodes here rather than at creation
3390 * time, we need a lock to ensure we only try
3393 static DEFINE_SPINLOCK(lock
);
3395 if (inode_unhashed(inode
))
3396 __insert_inode_hash(inode
,
3397 inode
->i_ino
+ inode
->i_generation
);
3401 fh
[0] = inode
->i_generation
;
3402 fh
[1] = inode
->i_ino
;
3403 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3409 static const struct export_operations shmem_export_ops
= {
3410 .get_parent
= shmem_get_parent
,
3411 .encode_fh
= shmem_encode_fh
,
3412 .fh_to_dentry
= shmem_fh_to_dentry
,
3428 static const struct constant_table shmem_param_enums_huge
[] = {
3429 {"never", SHMEM_HUGE_NEVER
},
3430 {"always", SHMEM_HUGE_ALWAYS
},
3431 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3432 {"advise", SHMEM_HUGE_ADVISE
},
3436 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3437 fsparam_u32 ("gid", Opt_gid
),
3438 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3439 fsparam_u32oct("mode", Opt_mode
),
3440 fsparam_string("mpol", Opt_mpol
),
3441 fsparam_string("nr_blocks", Opt_nr_blocks
),
3442 fsparam_string("nr_inodes", Opt_nr_inodes
),
3443 fsparam_string("size", Opt_size
),
3444 fsparam_u32 ("uid", Opt_uid
),
3445 fsparam_flag ("inode32", Opt_inode32
),
3446 fsparam_flag ("inode64", Opt_inode64
),
3450 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3452 struct shmem_options
*ctx
= fc
->fs_private
;
3453 struct fs_parse_result result
;
3454 unsigned long long size
;
3458 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3464 size
= memparse(param
->string
, &rest
);
3466 size
<<= PAGE_SHIFT
;
3467 size
*= totalram_pages();
3473 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3474 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3477 ctx
->blocks
= memparse(param
->string
, &rest
);
3480 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3483 ctx
->inodes
= memparse(param
->string
, &rest
);
3486 ctx
->seen
|= SHMEM_SEEN_INODES
;
3489 ctx
->mode
= result
.uint_32
& 07777;
3492 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3493 if (!uid_valid(ctx
->uid
))
3497 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3498 if (!gid_valid(ctx
->gid
))
3502 ctx
->huge
= result
.uint_32
;
3503 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3504 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3505 has_transparent_hugepage()))
3506 goto unsupported_parameter
;
3507 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3510 if (IS_ENABLED(CONFIG_NUMA
)) {
3511 mpol_put(ctx
->mpol
);
3513 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3517 goto unsupported_parameter
;
3519 ctx
->full_inums
= false;
3520 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3523 if (sizeof(ino_t
) < 8) {
3525 "Cannot use inode64 with <64bit inums in kernel\n");
3527 ctx
->full_inums
= true;
3528 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3533 unsupported_parameter
:
3534 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3536 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3539 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3541 char *options
= data
;
3544 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3549 while (options
!= NULL
) {
3550 char *this_char
= options
;
3553 * NUL-terminate this option: unfortunately,
3554 * mount options form a comma-separated list,
3555 * but mpol's nodelist may also contain commas.
3557 options
= strchr(options
, ',');
3558 if (options
== NULL
)
3561 if (!isdigit(*options
)) {
3567 char *value
= strchr(this_char
,'=');
3573 len
= strlen(value
);
3575 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3584 * Reconfigure a shmem filesystem.
3586 * Note that we disallow change from limited->unlimited blocks/inodes while any
3587 * are in use; but we must separately disallow unlimited->limited, because in
3588 * that case we have no record of how much is already in use.
3590 static int shmem_reconfigure(struct fs_context
*fc
)
3592 struct shmem_options
*ctx
= fc
->fs_private
;
3593 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3594 unsigned long inodes
;
3597 spin_lock(&sbinfo
->stat_lock
);
3598 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3599 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3600 if (!sbinfo
->max_blocks
) {
3601 err
= "Cannot retroactively limit size";
3604 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3606 err
= "Too small a size for current use";
3610 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3611 if (!sbinfo
->max_inodes
) {
3612 err
= "Cannot retroactively limit inodes";
3615 if (ctx
->inodes
< inodes
) {
3616 err
= "Too few inodes for current use";
3621 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
3622 sbinfo
->next_ino
> UINT_MAX
) {
3623 err
= "Current inum too high to switch to 32-bit inums";
3627 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3628 sbinfo
->huge
= ctx
->huge
;
3629 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
3630 sbinfo
->full_inums
= ctx
->full_inums
;
3631 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3632 sbinfo
->max_blocks
= ctx
->blocks
;
3633 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3634 sbinfo
->max_inodes
= ctx
->inodes
;
3635 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3639 * Preserve previous mempolicy unless mpol remount option was specified.
3642 mpol_put(sbinfo
->mpol
);
3643 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3646 spin_unlock(&sbinfo
->stat_lock
);
3649 spin_unlock(&sbinfo
->stat_lock
);
3650 return invalfc(fc
, "%s", err
);
3653 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3655 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3657 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3658 seq_printf(seq
, ",size=%luk",
3659 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3660 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3661 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3662 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3663 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3664 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3665 seq_printf(seq
, ",uid=%u",
3666 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3667 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3668 seq_printf(seq
, ",gid=%u",
3669 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3672 * Showing inode{64,32} might be useful even if it's the system default,
3673 * since then people don't have to resort to checking both here and
3674 * /proc/config.gz to confirm 64-bit inums were successfully applied
3675 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3677 * We hide it when inode64 isn't the default and we are using 32-bit
3678 * inodes, since that probably just means the feature isn't even under
3683 * +-----------------+-----------------+
3684 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3685 * +------------------+-----------------+-----------------+
3686 * | full_inums=true | show | show |
3687 * | full_inums=false | show | hide |
3688 * +------------------+-----------------+-----------------+
3691 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
3692 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
3693 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3694 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3696 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3698 shmem_show_mpol(seq
, sbinfo
->mpol
);
3702 #endif /* CONFIG_TMPFS */
3704 static void shmem_put_super(struct super_block
*sb
)
3706 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3708 free_percpu(sbinfo
->ino_batch
);
3709 percpu_counter_destroy(&sbinfo
->used_blocks
);
3710 mpol_put(sbinfo
->mpol
);
3712 sb
->s_fs_info
= NULL
;
3715 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3717 struct shmem_options
*ctx
= fc
->fs_private
;
3718 struct inode
*inode
;
3719 struct shmem_sb_info
*sbinfo
;
3722 /* Round up to L1_CACHE_BYTES to resist false sharing */
3723 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3724 L1_CACHE_BYTES
), GFP_KERNEL
);
3728 sb
->s_fs_info
= sbinfo
;
3732 * Per default we only allow half of the physical ram per
3733 * tmpfs instance, limiting inodes to one per page of lowmem;
3734 * but the internal instance is left unlimited.
3736 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3737 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3738 ctx
->blocks
= shmem_default_max_blocks();
3739 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3740 ctx
->inodes
= shmem_default_max_inodes();
3741 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
3742 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
3744 sb
->s_flags
|= SB_NOUSER
;
3746 sb
->s_export_op
= &shmem_export_ops
;
3747 sb
->s_flags
|= SB_NOSEC
;
3749 sb
->s_flags
|= SB_NOUSER
;
3751 sbinfo
->max_blocks
= ctx
->blocks
;
3752 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3753 if (sb
->s_flags
& SB_KERNMOUNT
) {
3754 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
3755 if (!sbinfo
->ino_batch
)
3758 sbinfo
->uid
= ctx
->uid
;
3759 sbinfo
->gid
= ctx
->gid
;
3760 sbinfo
->full_inums
= ctx
->full_inums
;
3761 sbinfo
->mode
= ctx
->mode
;
3762 sbinfo
->huge
= ctx
->huge
;
3763 sbinfo
->mpol
= ctx
->mpol
;
3766 spin_lock_init(&sbinfo
->stat_lock
);
3767 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3769 spin_lock_init(&sbinfo
->shrinklist_lock
);
3770 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3772 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3773 sb
->s_blocksize
= PAGE_SIZE
;
3774 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3775 sb
->s_magic
= TMPFS_MAGIC
;
3776 sb
->s_op
= &shmem_ops
;
3777 sb
->s_time_gran
= 1;
3778 #ifdef CONFIG_TMPFS_XATTR
3779 sb
->s_xattr
= shmem_xattr_handlers
;
3781 #ifdef CONFIG_TMPFS_POSIX_ACL
3782 sb
->s_flags
|= SB_POSIXACL
;
3784 uuid_gen(&sb
->s_uuid
);
3786 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3789 inode
->i_uid
= sbinfo
->uid
;
3790 inode
->i_gid
= sbinfo
->gid
;
3791 sb
->s_root
= d_make_root(inode
);
3797 shmem_put_super(sb
);
3801 static int shmem_get_tree(struct fs_context
*fc
)
3803 return get_tree_nodev(fc
, shmem_fill_super
);
3806 static void shmem_free_fc(struct fs_context
*fc
)
3808 struct shmem_options
*ctx
= fc
->fs_private
;
3811 mpol_put(ctx
->mpol
);
3816 static const struct fs_context_operations shmem_fs_context_ops
= {
3817 .free
= shmem_free_fc
,
3818 .get_tree
= shmem_get_tree
,
3820 .parse_monolithic
= shmem_parse_options
,
3821 .parse_param
= shmem_parse_one
,
3822 .reconfigure
= shmem_reconfigure
,
3826 static struct kmem_cache
*shmem_inode_cachep
;
3828 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3830 struct shmem_inode_info
*info
;
3831 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3834 return &info
->vfs_inode
;
3837 static void shmem_free_in_core_inode(struct inode
*inode
)
3839 if (S_ISLNK(inode
->i_mode
))
3840 kfree(inode
->i_link
);
3841 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3844 static void shmem_destroy_inode(struct inode
*inode
)
3846 if (S_ISREG(inode
->i_mode
))
3847 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3850 static void shmem_init_inode(void *foo
)
3852 struct shmem_inode_info
*info
= foo
;
3853 inode_init_once(&info
->vfs_inode
);
3856 static void shmem_init_inodecache(void)
3858 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3859 sizeof(struct shmem_inode_info
),
3860 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3863 static void shmem_destroy_inodecache(void)
3865 kmem_cache_destroy(shmem_inode_cachep
);
3868 static const struct address_space_operations shmem_aops
= {
3869 .writepage
= shmem_writepage
,
3870 .set_page_dirty
= __set_page_dirty_no_writeback
,
3872 .write_begin
= shmem_write_begin
,
3873 .write_end
= shmem_write_end
,
3875 #ifdef CONFIG_MIGRATION
3876 .migratepage
= migrate_page
,
3878 .error_remove_page
= generic_error_remove_page
,
3881 static const struct file_operations shmem_file_operations
= {
3883 .get_unmapped_area
= shmem_get_unmapped_area
,
3885 .llseek
= shmem_file_llseek
,
3886 .read_iter
= shmem_file_read_iter
,
3887 .write_iter
= generic_file_write_iter
,
3888 .fsync
= noop_fsync
,
3889 .splice_read
= generic_file_splice_read
,
3890 .splice_write
= iter_file_splice_write
,
3891 .fallocate
= shmem_fallocate
,
3895 static const struct inode_operations shmem_inode_operations
= {
3896 .getattr
= shmem_getattr
,
3897 .setattr
= shmem_setattr
,
3898 #ifdef CONFIG_TMPFS_XATTR
3899 .listxattr
= shmem_listxattr
,
3900 .set_acl
= simple_set_acl
,
3904 static const struct inode_operations shmem_dir_inode_operations
= {
3906 .create
= shmem_create
,
3907 .lookup
= simple_lookup
,
3909 .unlink
= shmem_unlink
,
3910 .symlink
= shmem_symlink
,
3911 .mkdir
= shmem_mkdir
,
3912 .rmdir
= shmem_rmdir
,
3913 .mknod
= shmem_mknod
,
3914 .rename
= shmem_rename2
,
3915 .tmpfile
= shmem_tmpfile
,
3917 #ifdef CONFIG_TMPFS_XATTR
3918 .listxattr
= shmem_listxattr
,
3920 #ifdef CONFIG_TMPFS_POSIX_ACL
3921 .setattr
= shmem_setattr
,
3922 .set_acl
= simple_set_acl
,
3926 static const struct inode_operations shmem_special_inode_operations
= {
3927 #ifdef CONFIG_TMPFS_XATTR
3928 .listxattr
= shmem_listxattr
,
3930 #ifdef CONFIG_TMPFS_POSIX_ACL
3931 .setattr
= shmem_setattr
,
3932 .set_acl
= simple_set_acl
,
3936 static const struct super_operations shmem_ops
= {
3937 .alloc_inode
= shmem_alloc_inode
,
3938 .free_inode
= shmem_free_in_core_inode
,
3939 .destroy_inode
= shmem_destroy_inode
,
3941 .statfs
= shmem_statfs
,
3942 .show_options
= shmem_show_options
,
3944 .evict_inode
= shmem_evict_inode
,
3945 .drop_inode
= generic_delete_inode
,
3946 .put_super
= shmem_put_super
,
3947 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3948 .nr_cached_objects
= shmem_unused_huge_count
,
3949 .free_cached_objects
= shmem_unused_huge_scan
,
3953 static const struct vm_operations_struct shmem_vm_ops
= {
3954 .fault
= shmem_fault
,
3955 .map_pages
= filemap_map_pages
,
3957 .set_policy
= shmem_set_policy
,
3958 .get_policy
= shmem_get_policy
,
3962 int shmem_init_fs_context(struct fs_context
*fc
)
3964 struct shmem_options
*ctx
;
3966 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3970 ctx
->mode
= 0777 | S_ISVTX
;
3971 ctx
->uid
= current_fsuid();
3972 ctx
->gid
= current_fsgid();
3974 fc
->fs_private
= ctx
;
3975 fc
->ops
= &shmem_fs_context_ops
;
3979 static struct file_system_type shmem_fs_type
= {
3980 .owner
= THIS_MODULE
,
3982 .init_fs_context
= shmem_init_fs_context
,
3984 .parameters
= shmem_fs_parameters
,
3986 .kill_sb
= kill_litter_super
,
3987 .fs_flags
= FS_USERNS_MOUNT
| FS_THP_SUPPORT
,
3990 int __init
shmem_init(void)
3994 shmem_init_inodecache();
3996 error
= register_filesystem(&shmem_fs_type
);
3998 pr_err("Could not register tmpfs\n");
4002 shm_mnt
= kern_mount(&shmem_fs_type
);
4003 if (IS_ERR(shm_mnt
)) {
4004 error
= PTR_ERR(shm_mnt
);
4005 pr_err("Could not kern_mount tmpfs\n");
4009 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4010 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
4011 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4013 shmem_huge
= 0; /* just in case it was patched */
4018 unregister_filesystem(&shmem_fs_type
);
4020 shmem_destroy_inodecache();
4021 shm_mnt
= ERR_PTR(error
);
4025 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4026 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4027 struct kobj_attribute
*attr
, char *buf
)
4029 static const int values
[] = {
4031 SHMEM_HUGE_WITHIN_SIZE
,
4039 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
4040 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
4042 count
+= sprintf(buf
+ count
, fmt
,
4043 shmem_format_huge(values
[i
]));
4045 buf
[count
- 1] = '\n';
4049 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4050 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4055 if (count
+ 1 > sizeof(tmp
))
4057 memcpy(tmp
, buf
, count
);
4059 if (count
&& tmp
[count
- 1] == '\n')
4060 tmp
[count
- 1] = '\0';
4062 huge
= shmem_parse_huge(tmp
);
4063 if (huge
== -EINVAL
)
4065 if (!has_transparent_hugepage() &&
4066 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4070 if (shmem_huge
> SHMEM_HUGE_DENY
)
4071 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4075 struct kobj_attribute shmem_enabled_attr
=
4076 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4077 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4079 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4080 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4082 struct inode
*inode
= file_inode(vma
->vm_file
);
4083 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4087 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
4088 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
4090 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4092 if (shmem_huge
== SHMEM_HUGE_DENY
)
4094 switch (sbinfo
->huge
) {
4095 case SHMEM_HUGE_NEVER
:
4097 case SHMEM_HUGE_ALWAYS
:
4099 case SHMEM_HUGE_WITHIN_SIZE
:
4100 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4101 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4102 if (i_size
>= HPAGE_PMD_SIZE
&&
4103 i_size
>> PAGE_SHIFT
>= off
)
4106 case SHMEM_HUGE_ADVISE
:
4107 /* TODO: implement fadvise() hints */
4108 return (vma
->vm_flags
& VM_HUGEPAGE
);
4114 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4116 #else /* !CONFIG_SHMEM */
4119 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4121 * This is intended for small system where the benefits of the full
4122 * shmem code (swap-backed and resource-limited) are outweighed by
4123 * their complexity. On systems without swap this code should be
4124 * effectively equivalent, but much lighter weight.
4127 static struct file_system_type shmem_fs_type
= {
4129 .init_fs_context
= ramfs_init_fs_context
,
4130 .parameters
= ramfs_fs_parameters
,
4131 .kill_sb
= kill_litter_super
,
4132 .fs_flags
= FS_USERNS_MOUNT
,
4135 int __init
shmem_init(void)
4137 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4139 shm_mnt
= kern_mount(&shmem_fs_type
);
4140 BUG_ON(IS_ERR(shm_mnt
));
4145 int shmem_unuse(unsigned int type
, bool frontswap
,
4146 unsigned long *fs_pages_to_unuse
)
4151 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4156 void shmem_unlock_mapping(struct address_space
*mapping
)
4161 unsigned long shmem_get_unmapped_area(struct file
*file
,
4162 unsigned long addr
, unsigned long len
,
4163 unsigned long pgoff
, unsigned long flags
)
4165 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4169 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4171 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4173 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4175 #define shmem_vm_ops generic_file_vm_ops
4176 #define shmem_file_operations ramfs_file_operations
4177 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4178 #define shmem_acct_size(flags, size) 0
4179 #define shmem_unacct_size(flags, size) do {} while (0)
4181 #endif /* CONFIG_SHMEM */
4185 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4186 unsigned long flags
, unsigned int i_flags
)
4188 struct inode
*inode
;
4192 return ERR_CAST(mnt
);
4194 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4195 return ERR_PTR(-EINVAL
);
4197 if (shmem_acct_size(flags
, size
))
4198 return ERR_PTR(-ENOMEM
);
4200 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4202 if (unlikely(!inode
)) {
4203 shmem_unacct_size(flags
, size
);
4204 return ERR_PTR(-ENOSPC
);
4206 inode
->i_flags
|= i_flags
;
4207 inode
->i_size
= size
;
4208 clear_nlink(inode
); /* It is unlinked */
4209 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4211 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4212 &shmem_file_operations
);
4219 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4220 * kernel internal. There will be NO LSM permission checks against the
4221 * underlying inode. So users of this interface must do LSM checks at a
4222 * higher layer. The users are the big_key and shm implementations. LSM
4223 * checks are provided at the key or shm level rather than the inode.
4224 * @name: name for dentry (to be seen in /proc/<pid>/maps
4225 * @size: size to be set for the file
4226 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4228 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4230 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4234 * shmem_file_setup - get an unlinked file living in tmpfs
4235 * @name: name for dentry (to be seen in /proc/<pid>/maps
4236 * @size: size to be set for the file
4237 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4239 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4241 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4243 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4246 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4247 * @mnt: the tmpfs mount where the file will be created
4248 * @name: name for dentry (to be seen in /proc/<pid>/maps
4249 * @size: size to be set for the file
4250 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4252 struct file
*shmem_file_setup_with_mnt(struct vfsmount
*mnt
, const char *name
,
4253 loff_t size
, unsigned long flags
)
4255 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4257 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4260 * shmem_zero_setup - setup a shared anonymous mapping
4261 * @vma: the vma to be mmapped is prepared by do_mmap
4263 int shmem_zero_setup(struct vm_area_struct
*vma
)
4266 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4269 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4270 * between XFS directory reading and selinux: since this file is only
4271 * accessible to the user through its mapping, use S_PRIVATE flag to
4272 * bypass file security, in the same way as shmem_kernel_file_setup().
4274 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4276 return PTR_ERR(file
);
4280 vma
->vm_file
= file
;
4281 vma
->vm_ops
= &shmem_vm_ops
;
4283 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
4284 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4285 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4286 khugepaged_enter(vma
, vma
->vm_flags
);
4293 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4294 * @mapping: the page's address_space
4295 * @index: the page index
4296 * @gfp: the page allocator flags to use if allocating
4298 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4299 * with any new page allocations done using the specified allocation flags.
4300 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4301 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4302 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4304 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4305 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4307 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4308 pgoff_t index
, gfp_t gfp
)
4311 struct inode
*inode
= mapping
->host
;
4315 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4316 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4317 gfp
, NULL
, NULL
, NULL
);
4319 page
= ERR_PTR(error
);
4325 * The tiny !SHMEM case uses ramfs without swap
4327 return read_cache_page_gfp(mapping
, index
, gfp
);
4330 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);