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
->free_inodes
) {
283 spin_unlock(&sbinfo
->stat_lock
);
286 sbinfo
->free_inodes
--;
288 ino
= sbinfo
->next_ino
++;
289 if (unlikely(is_zero_ino(ino
)))
290 ino
= sbinfo
->next_ino
++;
291 if (unlikely(!sbinfo
->full_inums
&&
294 * Emulate get_next_ino uint wraparound for
297 if (IS_ENABLED(CONFIG_64BIT
))
298 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
299 __func__
, MINOR(sb
->s_dev
));
300 sbinfo
->next_ino
= 1;
301 ino
= sbinfo
->next_ino
++;
305 spin_unlock(&sbinfo
->stat_lock
);
308 * __shmem_file_setup, one of our callers, is lock-free: it
309 * doesn't hold stat_lock in shmem_reserve_inode since
310 * max_inodes is always 0, and is called from potentially
311 * unknown contexts. As such, use a per-cpu batched allocator
312 * which doesn't require the per-sb stat_lock unless we are at
313 * the batch boundary.
315 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
316 * shmem mounts are not exposed to userspace, so we don't need
317 * to worry about things like glibc compatibility.
320 next_ino
= per_cpu_ptr(sbinfo
->ino_batch
, get_cpu());
322 if (unlikely(ino
% SHMEM_INO_BATCH
== 0)) {
323 spin_lock(&sbinfo
->stat_lock
);
324 ino
= sbinfo
->next_ino
;
325 sbinfo
->next_ino
+= SHMEM_INO_BATCH
;
326 spin_unlock(&sbinfo
->stat_lock
);
327 if (unlikely(is_zero_ino(ino
)))
338 static void shmem_free_inode(struct super_block
*sb
)
340 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
341 if (sbinfo
->max_inodes
) {
342 spin_lock(&sbinfo
->stat_lock
);
343 sbinfo
->free_inodes
++;
344 spin_unlock(&sbinfo
->stat_lock
);
349 * shmem_recalc_inode - recalculate the block usage of an inode
350 * @inode: inode to recalc
352 * We have to calculate the free blocks since the mm can drop
353 * undirtied hole pages behind our back.
355 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
356 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
358 * It has to be called with the spinlock held.
360 static void shmem_recalc_inode(struct inode
*inode
)
362 struct shmem_inode_info
*info
= SHMEM_I(inode
);
365 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
367 info
->alloced
-= freed
;
368 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
369 shmem_inode_unacct_blocks(inode
, freed
);
373 bool shmem_charge(struct inode
*inode
, long pages
)
375 struct shmem_inode_info
*info
= SHMEM_I(inode
);
378 if (!shmem_inode_acct_block(inode
, pages
))
381 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
382 inode
->i_mapping
->nrpages
+= pages
;
384 spin_lock_irqsave(&info
->lock
, flags
);
385 info
->alloced
+= pages
;
386 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
387 shmem_recalc_inode(inode
);
388 spin_unlock_irqrestore(&info
->lock
, flags
);
393 void shmem_uncharge(struct inode
*inode
, long pages
)
395 struct shmem_inode_info
*info
= SHMEM_I(inode
);
398 /* nrpages adjustment done by __delete_from_page_cache() or caller */
400 spin_lock_irqsave(&info
->lock
, flags
);
401 info
->alloced
-= pages
;
402 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
403 shmem_recalc_inode(inode
);
404 spin_unlock_irqrestore(&info
->lock
, flags
);
406 shmem_inode_unacct_blocks(inode
, pages
);
410 * Replace item expected in xarray by a new item, while holding xa_lock.
412 static int shmem_replace_entry(struct address_space
*mapping
,
413 pgoff_t index
, void *expected
, void *replacement
)
415 XA_STATE(xas
, &mapping
->i_pages
, index
);
418 VM_BUG_ON(!expected
);
419 VM_BUG_ON(!replacement
);
420 item
= xas_load(&xas
);
421 if (item
!= expected
)
423 xas_store(&xas
, replacement
);
428 * Sometimes, before we decide whether to proceed or to fail, we must check
429 * that an entry was not already brought back from swap by a racing thread.
431 * Checking page is not enough: by the time a SwapCache page is locked, it
432 * might be reused, and again be SwapCache, using the same swap as before.
434 static bool shmem_confirm_swap(struct address_space
*mapping
,
435 pgoff_t index
, swp_entry_t swap
)
437 return xa_load(&mapping
->i_pages
, index
) == swp_to_radix_entry(swap
);
441 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
444 * disables huge pages for the mount;
446 * enables huge pages for the mount;
447 * SHMEM_HUGE_WITHIN_SIZE:
448 * only allocate huge pages if the page will be fully within i_size,
449 * also respect fadvise()/madvise() hints;
451 * only allocate huge pages if requested with fadvise()/madvise();
454 #define SHMEM_HUGE_NEVER 0
455 #define SHMEM_HUGE_ALWAYS 1
456 #define SHMEM_HUGE_WITHIN_SIZE 2
457 #define SHMEM_HUGE_ADVISE 3
461 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
464 * disables huge on shm_mnt and all mounts, for emergency use;
466 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
469 #define SHMEM_HUGE_DENY (-1)
470 #define SHMEM_HUGE_FORCE (-2)
472 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
473 /* ifdef here to avoid bloating shmem.o when not necessary */
475 static int shmem_huge __read_mostly
;
477 #if defined(CONFIG_SYSFS)
478 static int shmem_parse_huge(const char *str
)
480 if (!strcmp(str
, "never"))
481 return SHMEM_HUGE_NEVER
;
482 if (!strcmp(str
, "always"))
483 return SHMEM_HUGE_ALWAYS
;
484 if (!strcmp(str
, "within_size"))
485 return SHMEM_HUGE_WITHIN_SIZE
;
486 if (!strcmp(str
, "advise"))
487 return SHMEM_HUGE_ADVISE
;
488 if (!strcmp(str
, "deny"))
489 return SHMEM_HUGE_DENY
;
490 if (!strcmp(str
, "force"))
491 return SHMEM_HUGE_FORCE
;
496 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
497 static const char *shmem_format_huge(int huge
)
500 case SHMEM_HUGE_NEVER
:
502 case SHMEM_HUGE_ALWAYS
:
504 case SHMEM_HUGE_WITHIN_SIZE
:
505 return "within_size";
506 case SHMEM_HUGE_ADVISE
:
508 case SHMEM_HUGE_DENY
:
510 case SHMEM_HUGE_FORCE
:
519 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
520 struct shrink_control
*sc
, unsigned long nr_to_split
)
522 LIST_HEAD(list
), *pos
, *next
;
523 LIST_HEAD(to_remove
);
525 struct shmem_inode_info
*info
;
527 unsigned long batch
= sc
? sc
->nr_to_scan
: 128;
528 int removed
= 0, split
= 0;
530 if (list_empty(&sbinfo
->shrinklist
))
533 spin_lock(&sbinfo
->shrinklist_lock
);
534 list_for_each_safe(pos
, next
, &sbinfo
->shrinklist
) {
535 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
538 inode
= igrab(&info
->vfs_inode
);
540 /* inode is about to be evicted */
542 list_del_init(&info
->shrinklist
);
547 /* Check if there's anything to gain */
548 if (round_up(inode
->i_size
, PAGE_SIZE
) ==
549 round_up(inode
->i_size
, HPAGE_PMD_SIZE
)) {
550 list_move(&info
->shrinklist
, &to_remove
);
555 list_move(&info
->shrinklist
, &list
);
560 spin_unlock(&sbinfo
->shrinklist_lock
);
562 list_for_each_safe(pos
, next
, &to_remove
) {
563 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
564 inode
= &info
->vfs_inode
;
565 list_del_init(&info
->shrinklist
);
569 list_for_each_safe(pos
, next
, &list
) {
572 info
= list_entry(pos
, struct shmem_inode_info
, shrinklist
);
573 inode
= &info
->vfs_inode
;
575 if (nr_to_split
&& split
>= nr_to_split
)
578 page
= find_get_page(inode
->i_mapping
,
579 (inode
->i_size
& HPAGE_PMD_MASK
) >> PAGE_SHIFT
);
583 /* No huge page at the end of the file: nothing to split */
584 if (!PageTransHuge(page
)) {
590 * Leave the inode on the list if we failed to lock
591 * the page at this time.
593 * Waiting for the lock may lead to deadlock in the
596 if (!trylock_page(page
)) {
601 ret
= split_huge_page(page
);
605 /* If split failed leave the inode on the list */
611 list_del_init(&info
->shrinklist
);
617 spin_lock(&sbinfo
->shrinklist_lock
);
618 list_splice_tail(&list
, &sbinfo
->shrinklist
);
619 sbinfo
->shrinklist_len
-= removed
;
620 spin_unlock(&sbinfo
->shrinklist_lock
);
625 static long shmem_unused_huge_scan(struct super_block
*sb
,
626 struct shrink_control
*sc
)
628 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
630 if (!READ_ONCE(sbinfo
->shrinklist_len
))
633 return shmem_unused_huge_shrink(sbinfo
, sc
, 0);
636 static long shmem_unused_huge_count(struct super_block
*sb
,
637 struct shrink_control
*sc
)
639 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
640 return READ_ONCE(sbinfo
->shrinklist_len
);
642 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
644 #define shmem_huge SHMEM_HUGE_DENY
646 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info
*sbinfo
,
647 struct shrink_control
*sc
, unsigned long nr_to_split
)
651 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
653 static inline bool is_huge_enabled(struct shmem_sb_info
*sbinfo
)
655 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
656 (shmem_huge
== SHMEM_HUGE_FORCE
|| sbinfo
->huge
) &&
657 shmem_huge
!= SHMEM_HUGE_DENY
)
663 * Like add_to_page_cache_locked, but error if expected item has gone.
665 static int shmem_add_to_page_cache(struct page
*page
,
666 struct address_space
*mapping
,
667 pgoff_t index
, void *expected
, gfp_t gfp
,
668 struct mm_struct
*charge_mm
)
670 XA_STATE_ORDER(xas
, &mapping
->i_pages
, index
, compound_order(page
));
672 unsigned long nr
= compound_nr(page
);
675 VM_BUG_ON_PAGE(PageTail(page
), page
);
676 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
677 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
678 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
679 VM_BUG_ON(expected
&& PageTransHuge(page
));
681 page_ref_add(page
, nr
);
682 page
->mapping
= mapping
;
685 if (!PageSwapCache(page
)) {
686 error
= mem_cgroup_charge(page
, charge_mm
, gfp
);
688 if (PageTransHuge(page
)) {
689 count_vm_event(THP_FILE_FALLBACK
);
690 count_vm_event(THP_FILE_FALLBACK_CHARGE
);
695 cgroup_throttle_swaprate(page
, gfp
);
700 entry
= xas_find_conflict(&xas
);
701 if (entry
!= expected
)
702 xas_set_err(&xas
, -EEXIST
);
703 xas_create_range(&xas
);
707 xas_store(&xas
, page
);
712 if (PageTransHuge(page
)) {
713 count_vm_event(THP_FILE_ALLOC
);
714 __inc_node_page_state(page
, NR_SHMEM_THPS
);
716 mapping
->nrpages
+= nr
;
717 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, nr
);
718 __mod_lruvec_page_state(page
, NR_SHMEM
, nr
);
720 xas_unlock_irq(&xas
);
721 } while (xas_nomem(&xas
, gfp
));
723 if (xas_error(&xas
)) {
724 error
= xas_error(&xas
);
730 page
->mapping
= NULL
;
731 page_ref_sub(page
, nr
);
736 * Like delete_from_page_cache, but substitutes swap for page.
738 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
740 struct address_space
*mapping
= page
->mapping
;
743 VM_BUG_ON_PAGE(PageCompound(page
), page
);
745 xa_lock_irq(&mapping
->i_pages
);
746 error
= shmem_replace_entry(mapping
, page
->index
, page
, radswap
);
747 page
->mapping
= NULL
;
749 __dec_lruvec_page_state(page
, NR_FILE_PAGES
);
750 __dec_lruvec_page_state(page
, NR_SHMEM
);
751 xa_unlock_irq(&mapping
->i_pages
);
757 * Remove swap entry from page cache, free the swap and its page cache.
759 static int shmem_free_swap(struct address_space
*mapping
,
760 pgoff_t index
, void *radswap
)
764 old
= xa_cmpxchg_irq(&mapping
->i_pages
, index
, radswap
, NULL
, 0);
767 free_swap_and_cache(radix_to_swp_entry(radswap
));
772 * Determine (in bytes) how many of the shmem object's pages mapped by the
773 * given offsets are swapped out.
775 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
776 * as long as the inode doesn't go away and racy results are not a problem.
778 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
779 pgoff_t start
, pgoff_t end
)
781 XA_STATE(xas
, &mapping
->i_pages
, start
);
783 unsigned long swapped
= 0;
786 xas_for_each(&xas
, page
, end
- 1) {
787 if (xas_retry(&xas
, page
))
789 if (xa_is_value(page
))
792 if (need_resched()) {
800 return swapped
<< PAGE_SHIFT
;
804 * Determine (in bytes) how many of the shmem object's pages mapped by the
805 * given vma is swapped out.
807 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
808 * as long as the inode doesn't go away and racy results are not a problem.
810 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
812 struct inode
*inode
= file_inode(vma
->vm_file
);
813 struct shmem_inode_info
*info
= SHMEM_I(inode
);
814 struct address_space
*mapping
= inode
->i_mapping
;
815 unsigned long swapped
;
817 /* Be careful as we don't hold info->lock */
818 swapped
= READ_ONCE(info
->swapped
);
821 * The easier cases are when the shmem object has nothing in swap, or
822 * the vma maps it whole. Then we can simply use the stats that we
828 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
829 return swapped
<< PAGE_SHIFT
;
831 /* Here comes the more involved part */
832 return shmem_partial_swap_usage(mapping
,
833 linear_page_index(vma
, vma
->vm_start
),
834 linear_page_index(vma
, vma
->vm_end
));
838 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
840 void shmem_unlock_mapping(struct address_space
*mapping
)
843 pgoff_t indices
[PAGEVEC_SIZE
];
848 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
850 while (!mapping_unevictable(mapping
)) {
852 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
853 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
855 pvec
.nr
= find_get_entries(mapping
, index
,
856 PAGEVEC_SIZE
, pvec
.pages
, indices
);
859 index
= indices
[pvec
.nr
- 1] + 1;
860 pagevec_remove_exceptionals(&pvec
);
861 check_move_unevictable_pages(&pvec
);
862 pagevec_release(&pvec
);
868 * Check whether a hole-punch or truncation needs to split a huge page,
869 * returning true if no split was required, or the split has been successful.
871 * Eviction (or truncation to 0 size) should never need to split a huge page;
872 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
873 * head, and then succeeded to trylock on tail.
875 * A split can only succeed when there are no additional references on the
876 * huge page: so the split below relies upon find_get_entries() having stopped
877 * when it found a subpage of the huge page, without getting further references.
879 static bool shmem_punch_compound(struct page
*page
, pgoff_t start
, pgoff_t end
)
881 if (!PageTransCompound(page
))
884 /* Just proceed to delete a huge page wholly within the range punched */
885 if (PageHead(page
) &&
886 page
->index
>= start
&& page
->index
+ HPAGE_PMD_NR
<= end
)
889 /* Try to split huge page, so we can truly punch the hole or truncate */
890 return split_huge_page(page
) >= 0;
894 * Remove range of pages and swap entries from page cache, and free them.
895 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
897 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
900 struct address_space
*mapping
= inode
->i_mapping
;
901 struct shmem_inode_info
*info
= SHMEM_I(inode
);
902 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
903 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
904 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
905 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
907 pgoff_t indices
[PAGEVEC_SIZE
];
908 long nr_swaps_freed
= 0;
913 end
= -1; /* unsigned, so actually very big */
917 while (index
< end
) {
918 pvec
.nr
= find_get_entries(mapping
, index
,
919 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
920 pvec
.pages
, indices
);
923 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
924 struct page
*page
= pvec
.pages
[i
];
930 if (xa_is_value(page
)) {
933 nr_swaps_freed
+= !shmem_free_swap(mapping
,
938 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
940 if (!trylock_page(page
))
943 if ((!unfalloc
|| !PageUptodate(page
)) &&
944 page_mapping(page
) == mapping
) {
945 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
946 if (shmem_punch_compound(page
, start
, end
))
947 truncate_inode_page(mapping
, page
);
951 pagevec_remove_exceptionals(&pvec
);
952 pagevec_release(&pvec
);
958 struct page
*page
= NULL
;
959 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
961 unsigned int top
= PAGE_SIZE
;
966 zero_user_segment(page
, partial_start
, top
);
967 set_page_dirty(page
);
973 struct page
*page
= NULL
;
974 shmem_getpage(inode
, end
, &page
, SGP_READ
);
976 zero_user_segment(page
, 0, partial_end
);
977 set_page_dirty(page
);
986 while (index
< end
) {
989 pvec
.nr
= find_get_entries(mapping
, index
,
990 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
991 pvec
.pages
, indices
);
993 /* If all gone or hole-punch or unfalloc, we're done */
994 if (index
== start
|| end
!= -1)
996 /* But if truncating, restart to make sure all gone */
1000 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
1001 struct page
*page
= pvec
.pages
[i
];
1007 if (xa_is_value(page
)) {
1010 if (shmem_free_swap(mapping
, index
, page
)) {
1011 /* Swap was replaced by page: retry */
1021 if (!unfalloc
|| !PageUptodate(page
)) {
1022 if (page_mapping(page
) != mapping
) {
1023 /* Page was replaced by swap: retry */
1028 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
1029 if (shmem_punch_compound(page
, start
, end
))
1030 truncate_inode_page(mapping
, page
);
1031 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1032 /* Wipe the page and don't get stuck */
1033 clear_highpage(page
);
1034 flush_dcache_page(page
);
1035 set_page_dirty(page
);
1037 round_up(start
, HPAGE_PMD_NR
))
1043 pagevec_remove_exceptionals(&pvec
);
1044 pagevec_release(&pvec
);
1048 spin_lock_irq(&info
->lock
);
1049 info
->swapped
-= nr_swaps_freed
;
1050 shmem_recalc_inode(inode
);
1051 spin_unlock_irq(&info
->lock
);
1054 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
1056 shmem_undo_range(inode
, lstart
, lend
, false);
1057 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1059 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
1061 static int shmem_getattr(const struct path
*path
, struct kstat
*stat
,
1062 u32 request_mask
, unsigned int query_flags
)
1064 struct inode
*inode
= path
->dentry
->d_inode
;
1065 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1066 struct shmem_sb_info
*sb_info
= SHMEM_SB(inode
->i_sb
);
1068 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
1069 spin_lock_irq(&info
->lock
);
1070 shmem_recalc_inode(inode
);
1071 spin_unlock_irq(&info
->lock
);
1073 generic_fillattr(inode
, stat
);
1075 if (is_huge_enabled(sb_info
))
1076 stat
->blksize
= HPAGE_PMD_SIZE
;
1081 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1083 struct inode
*inode
= d_inode(dentry
);
1084 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1085 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1088 error
= setattr_prepare(dentry
, attr
);
1092 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
1093 loff_t oldsize
= inode
->i_size
;
1094 loff_t newsize
= attr
->ia_size
;
1096 /* protected by i_mutex */
1097 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
1098 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
1101 if (newsize
!= oldsize
) {
1102 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
1106 i_size_write(inode
, newsize
);
1107 inode
->i_ctime
= inode
->i_mtime
= current_time(inode
);
1109 if (newsize
<= oldsize
) {
1110 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
1111 if (oldsize
> holebegin
)
1112 unmap_mapping_range(inode
->i_mapping
,
1115 shmem_truncate_range(inode
,
1116 newsize
, (loff_t
)-1);
1117 /* unmap again to remove racily COWed private pages */
1118 if (oldsize
> holebegin
)
1119 unmap_mapping_range(inode
->i_mapping
,
1123 * Part of the huge page can be beyond i_size: subject
1124 * to shrink under memory pressure.
1126 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1127 spin_lock(&sbinfo
->shrinklist_lock
);
1129 * _careful to defend against unlocked access to
1130 * ->shrink_list in shmem_unused_huge_shrink()
1132 if (list_empty_careful(&info
->shrinklist
)) {
1133 list_add_tail(&info
->shrinklist
,
1134 &sbinfo
->shrinklist
);
1135 sbinfo
->shrinklist_len
++;
1137 spin_unlock(&sbinfo
->shrinklist_lock
);
1142 setattr_copy(inode
, attr
);
1143 if (attr
->ia_valid
& ATTR_MODE
)
1144 error
= posix_acl_chmod(inode
, inode
->i_mode
);
1148 static void shmem_evict_inode(struct inode
*inode
)
1150 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1151 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1153 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
1154 shmem_unacct_size(info
->flags
, inode
->i_size
);
1156 shmem_truncate_range(inode
, 0, (loff_t
)-1);
1157 if (!list_empty(&info
->shrinklist
)) {
1158 spin_lock(&sbinfo
->shrinklist_lock
);
1159 if (!list_empty(&info
->shrinklist
)) {
1160 list_del_init(&info
->shrinklist
);
1161 sbinfo
->shrinklist_len
--;
1163 spin_unlock(&sbinfo
->shrinklist_lock
);
1165 while (!list_empty(&info
->swaplist
)) {
1166 /* Wait while shmem_unuse() is scanning this inode... */
1167 wait_var_event(&info
->stop_eviction
,
1168 !atomic_read(&info
->stop_eviction
));
1169 mutex_lock(&shmem_swaplist_mutex
);
1170 /* ...but beware of the race if we peeked too early */
1171 if (!atomic_read(&info
->stop_eviction
))
1172 list_del_init(&info
->swaplist
);
1173 mutex_unlock(&shmem_swaplist_mutex
);
1177 simple_xattrs_free(&info
->xattrs
);
1178 WARN_ON(inode
->i_blocks
);
1179 shmem_free_inode(inode
->i_sb
);
1183 extern struct swap_info_struct
*swap_info
[];
1185 static int shmem_find_swap_entries(struct address_space
*mapping
,
1186 pgoff_t start
, unsigned int nr_entries
,
1187 struct page
**entries
, pgoff_t
*indices
,
1188 unsigned int type
, bool frontswap
)
1190 XA_STATE(xas
, &mapping
->i_pages
, start
);
1193 unsigned int ret
= 0;
1199 xas_for_each(&xas
, page
, ULONG_MAX
) {
1200 if (xas_retry(&xas
, page
))
1203 if (!xa_is_value(page
))
1206 entry
= radix_to_swp_entry(page
);
1207 if (swp_type(entry
) != type
)
1210 !frontswap_test(swap_info
[type
], swp_offset(entry
)))
1213 indices
[ret
] = xas
.xa_index
;
1214 entries
[ret
] = page
;
1216 if (need_resched()) {
1220 if (++ret
== nr_entries
)
1229 * Move the swapped pages for an inode to page cache. Returns the count
1230 * of pages swapped in, or the error in case of failure.
1232 static int shmem_unuse_swap_entries(struct inode
*inode
, struct pagevec pvec
,
1238 struct address_space
*mapping
= inode
->i_mapping
;
1240 for (i
= 0; i
< pvec
.nr
; i
++) {
1241 struct page
*page
= pvec
.pages
[i
];
1243 if (!xa_is_value(page
))
1245 error
= shmem_swapin_page(inode
, indices
[i
],
1247 mapping_gfp_mask(mapping
),
1254 if (error
== -ENOMEM
)
1258 return error
? error
: ret
;
1262 * If swap found in inode, free it and move page from swapcache to filecache.
1264 static int shmem_unuse_inode(struct inode
*inode
, unsigned int type
,
1265 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1267 struct address_space
*mapping
= inode
->i_mapping
;
1269 struct pagevec pvec
;
1270 pgoff_t indices
[PAGEVEC_SIZE
];
1271 bool frontswap_partial
= (frontswap
&& *fs_pages_to_unuse
> 0);
1274 pagevec_init(&pvec
);
1276 unsigned int nr_entries
= PAGEVEC_SIZE
;
1278 if (frontswap_partial
&& *fs_pages_to_unuse
< PAGEVEC_SIZE
)
1279 nr_entries
= *fs_pages_to_unuse
;
1281 pvec
.nr
= shmem_find_swap_entries(mapping
, start
, nr_entries
,
1282 pvec
.pages
, indices
,
1289 ret
= shmem_unuse_swap_entries(inode
, pvec
, indices
);
1293 if (frontswap_partial
) {
1294 *fs_pages_to_unuse
-= ret
;
1295 if (*fs_pages_to_unuse
== 0) {
1296 ret
= FRONTSWAP_PAGES_UNUSED
;
1301 start
= indices
[pvec
.nr
- 1];
1308 * Read all the shared memory data that resides in the swap
1309 * device 'type' back into memory, so the swap device can be
1312 int shmem_unuse(unsigned int type
, bool frontswap
,
1313 unsigned long *fs_pages_to_unuse
)
1315 struct shmem_inode_info
*info
, *next
;
1318 if (list_empty(&shmem_swaplist
))
1321 mutex_lock(&shmem_swaplist_mutex
);
1322 list_for_each_entry_safe(info
, next
, &shmem_swaplist
, swaplist
) {
1323 if (!info
->swapped
) {
1324 list_del_init(&info
->swaplist
);
1328 * Drop the swaplist mutex while searching the inode for swap;
1329 * but before doing so, make sure shmem_evict_inode() will not
1330 * remove placeholder inode from swaplist, nor let it be freed
1331 * (igrab() would protect from unlink, but not from unmount).
1333 atomic_inc(&info
->stop_eviction
);
1334 mutex_unlock(&shmem_swaplist_mutex
);
1336 error
= shmem_unuse_inode(&info
->vfs_inode
, type
, frontswap
,
1340 mutex_lock(&shmem_swaplist_mutex
);
1341 next
= list_next_entry(info
, swaplist
);
1343 list_del_init(&info
->swaplist
);
1344 if (atomic_dec_and_test(&info
->stop_eviction
))
1345 wake_up_var(&info
->stop_eviction
);
1349 mutex_unlock(&shmem_swaplist_mutex
);
1355 * Move the page from the page cache to the swap cache.
1357 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1359 struct shmem_inode_info
*info
;
1360 struct address_space
*mapping
;
1361 struct inode
*inode
;
1365 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1366 BUG_ON(!PageLocked(page
));
1367 mapping
= page
->mapping
;
1368 index
= page
->index
;
1369 inode
= mapping
->host
;
1370 info
= SHMEM_I(inode
);
1371 if (info
->flags
& VM_LOCKED
)
1373 if (!total_swap_pages
)
1377 * Our capabilities prevent regular writeback or sync from ever calling
1378 * shmem_writepage; but a stacking filesystem might use ->writepage of
1379 * its underlying filesystem, in which case tmpfs should write out to
1380 * swap only in response to memory pressure, and not for the writeback
1383 if (!wbc
->for_reclaim
) {
1384 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1389 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1390 * value into swapfile.c, the only way we can correctly account for a
1391 * fallocated page arriving here is now to initialize it and write it.
1393 * That's okay for a page already fallocated earlier, but if we have
1394 * not yet completed the fallocation, then (a) we want to keep track
1395 * of this page in case we have to undo it, and (b) it may not be a
1396 * good idea to continue anyway, once we're pushing into swap. So
1397 * reactivate the page, and let shmem_fallocate() quit when too many.
1399 if (!PageUptodate(page
)) {
1400 if (inode
->i_private
) {
1401 struct shmem_falloc
*shmem_falloc
;
1402 spin_lock(&inode
->i_lock
);
1403 shmem_falloc
= inode
->i_private
;
1405 !shmem_falloc
->waitq
&&
1406 index
>= shmem_falloc
->start
&&
1407 index
< shmem_falloc
->next
)
1408 shmem_falloc
->nr_unswapped
++;
1410 shmem_falloc
= NULL
;
1411 spin_unlock(&inode
->i_lock
);
1415 clear_highpage(page
);
1416 flush_dcache_page(page
);
1417 SetPageUptodate(page
);
1420 swap
= get_swap_page(page
);
1425 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1426 * if it's not already there. Do it now before the page is
1427 * moved to swap cache, when its pagelock no longer protects
1428 * the inode from eviction. But don't unlock the mutex until
1429 * we've incremented swapped, because shmem_unuse_inode() will
1430 * prune a !swapped inode from the swaplist under this mutex.
1432 mutex_lock(&shmem_swaplist_mutex
);
1433 if (list_empty(&info
->swaplist
))
1434 list_add(&info
->swaplist
, &shmem_swaplist
);
1436 if (add_to_swap_cache(page
, swap
,
1437 __GFP_HIGH
| __GFP_NOMEMALLOC
| __GFP_NOWARN
) == 0) {
1438 spin_lock_irq(&info
->lock
);
1439 shmem_recalc_inode(inode
);
1441 spin_unlock_irq(&info
->lock
);
1443 swap_shmem_alloc(swap
);
1444 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1446 mutex_unlock(&shmem_swaplist_mutex
);
1447 BUG_ON(page_mapped(page
));
1448 swap_writepage(page
, wbc
);
1452 mutex_unlock(&shmem_swaplist_mutex
);
1453 put_swap_page(page
, swap
);
1455 set_page_dirty(page
);
1456 if (wbc
->for_reclaim
)
1457 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1462 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1463 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1467 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1468 return; /* show nothing */
1470 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1472 seq_printf(seq
, ",mpol=%s", buffer
);
1475 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1477 struct mempolicy
*mpol
= NULL
;
1479 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1480 mpol
= sbinfo
->mpol
;
1482 spin_unlock(&sbinfo
->stat_lock
);
1486 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1487 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1490 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1494 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1496 #define vm_policy vm_private_data
1499 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1500 struct shmem_inode_info
*info
, pgoff_t index
)
1502 /* Create a pseudo vma that just contains the policy */
1503 vma_init(vma
, NULL
);
1504 /* Bias interleave by inode number to distribute better across nodes */
1505 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1506 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1509 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1511 /* Drop reference taken by mpol_shared_policy_lookup() */
1512 mpol_cond_put(vma
->vm_policy
);
1515 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1516 struct shmem_inode_info
*info
, pgoff_t index
)
1518 struct vm_area_struct pvma
;
1520 struct vm_fault vmf
;
1522 shmem_pseudo_vma_init(&pvma
, info
, index
);
1525 page
= swap_cluster_readahead(swap
, gfp
, &vmf
);
1526 shmem_pseudo_vma_destroy(&pvma
);
1531 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1532 struct shmem_inode_info
*info
, pgoff_t index
)
1534 struct vm_area_struct pvma
;
1535 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
1539 hindex
= round_down(index
, HPAGE_PMD_NR
);
1540 if (xa_find(&mapping
->i_pages
, &hindex
, hindex
+ HPAGE_PMD_NR
- 1,
1544 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1545 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1546 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1547 shmem_pseudo_vma_destroy(&pvma
);
1549 prep_transhuge_page(page
);
1551 count_vm_event(THP_FILE_FALLBACK
);
1555 static struct page
*shmem_alloc_page(gfp_t gfp
,
1556 struct shmem_inode_info
*info
, pgoff_t index
)
1558 struct vm_area_struct pvma
;
1561 shmem_pseudo_vma_init(&pvma
, info
, index
);
1562 page
= alloc_page_vma(gfp
, &pvma
, 0);
1563 shmem_pseudo_vma_destroy(&pvma
);
1568 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1569 struct inode
*inode
,
1570 pgoff_t index
, bool huge
)
1572 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1577 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1579 nr
= huge
? HPAGE_PMD_NR
: 1;
1581 if (!shmem_inode_acct_block(inode
, nr
))
1585 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1587 page
= shmem_alloc_page(gfp
, info
, index
);
1589 __SetPageLocked(page
);
1590 __SetPageSwapBacked(page
);
1595 shmem_inode_unacct_blocks(inode
, nr
);
1597 return ERR_PTR(err
);
1601 * When a page is moved from swapcache to shmem filecache (either by the
1602 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1603 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1604 * ignorance of the mapping it belongs to. If that mapping has special
1605 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1606 * we may need to copy to a suitable page before moving to filecache.
1608 * In a future release, this may well be extended to respect cpuset and
1609 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1610 * but for now it is a simple matter of zone.
1612 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1614 return page_zonenum(page
) > gfp_zone(gfp
);
1617 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1618 struct shmem_inode_info
*info
, pgoff_t index
)
1620 struct page
*oldpage
, *newpage
;
1621 struct address_space
*swap_mapping
;
1627 entry
.val
= page_private(oldpage
);
1628 swap_index
= swp_offset(entry
);
1629 swap_mapping
= page_mapping(oldpage
);
1632 * We have arrived here because our zones are constrained, so don't
1633 * limit chance of success by further cpuset and node constraints.
1635 gfp
&= ~GFP_CONSTRAINT_MASK
;
1636 newpage
= shmem_alloc_page(gfp
, info
, index
);
1641 copy_highpage(newpage
, oldpage
);
1642 flush_dcache_page(newpage
);
1644 __SetPageLocked(newpage
);
1645 __SetPageSwapBacked(newpage
);
1646 SetPageUptodate(newpage
);
1647 set_page_private(newpage
, entry
.val
);
1648 SetPageSwapCache(newpage
);
1651 * Our caller will very soon move newpage out of swapcache, but it's
1652 * a nice clean interface for us to replace oldpage by newpage there.
1654 xa_lock_irq(&swap_mapping
->i_pages
);
1655 error
= shmem_replace_entry(swap_mapping
, swap_index
, oldpage
, newpage
);
1657 mem_cgroup_migrate(oldpage
, newpage
);
1658 __inc_lruvec_page_state(newpage
, NR_FILE_PAGES
);
1659 __dec_lruvec_page_state(oldpage
, NR_FILE_PAGES
);
1661 xa_unlock_irq(&swap_mapping
->i_pages
);
1663 if (unlikely(error
)) {
1665 * Is this possible? I think not, now that our callers check
1666 * both PageSwapCache and page_private after getting page lock;
1667 * but be defensive. Reverse old to newpage for clear and free.
1671 lru_cache_add(newpage
);
1675 ClearPageSwapCache(oldpage
);
1676 set_page_private(oldpage
, 0);
1678 unlock_page(oldpage
);
1685 * Swap in the page pointed to by *pagep.
1686 * Caller has to make sure that *pagep contains a valid swapped page.
1687 * Returns 0 and the page in pagep if success. On failure, returns the
1688 * the error code and NULL in *pagep.
1690 static int shmem_swapin_page(struct inode
*inode
, pgoff_t index
,
1691 struct page
**pagep
, enum sgp_type sgp
,
1692 gfp_t gfp
, struct vm_area_struct
*vma
,
1693 vm_fault_t
*fault_type
)
1695 struct address_space
*mapping
= inode
->i_mapping
;
1696 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1697 struct mm_struct
*charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1702 VM_BUG_ON(!*pagep
|| !xa_is_value(*pagep
));
1703 swap
= radix_to_swp_entry(*pagep
);
1706 /* Look it up and read it in.. */
1707 page
= lookup_swap_cache(swap
, NULL
, 0);
1709 /* Or update major stats only when swapin succeeds?? */
1711 *fault_type
|= VM_FAULT_MAJOR
;
1712 count_vm_event(PGMAJFAULT
);
1713 count_memcg_event_mm(charge_mm
, PGMAJFAULT
);
1715 /* Here we actually start the io */
1716 page
= shmem_swapin(swap
, gfp
, info
, index
);
1723 /* We have to do this with page locked to prevent races */
1725 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1726 !shmem_confirm_swap(mapping
, index
, swap
)) {
1730 if (!PageUptodate(page
)) {
1734 wait_on_page_writeback(page
);
1736 if (shmem_should_replace_page(page
, gfp
)) {
1737 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1742 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1743 swp_to_radix_entry(swap
), gfp
,
1748 spin_lock_irq(&info
->lock
);
1750 shmem_recalc_inode(inode
);
1751 spin_unlock_irq(&info
->lock
);
1753 if (sgp
== SGP_WRITE
)
1754 mark_page_accessed(page
);
1756 delete_from_swap_cache(page
);
1757 set_page_dirty(page
);
1763 if (!shmem_confirm_swap(mapping
, index
, swap
))
1775 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1777 * If we allocate a new one we do not mark it dirty. That's up to the
1778 * vm. If we swap it in we mark it dirty since we also free the swap
1779 * entry since a page cannot live in both the swap and page cache.
1781 * vmf and fault_type are only supplied by shmem_fault:
1782 * otherwise they are NULL.
1784 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1785 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1786 struct vm_area_struct
*vma
, struct vm_fault
*vmf
,
1787 vm_fault_t
*fault_type
)
1789 struct address_space
*mapping
= inode
->i_mapping
;
1790 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1791 struct shmem_sb_info
*sbinfo
;
1792 struct mm_struct
*charge_mm
;
1794 enum sgp_type sgp_huge
= sgp
;
1795 pgoff_t hindex
= index
;
1800 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1802 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1805 if (sgp
<= SGP_CACHE
&&
1806 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1810 sbinfo
= SHMEM_SB(inode
->i_sb
);
1811 charge_mm
= vma
? vma
->vm_mm
: current
->mm
;
1813 page
= find_lock_entry(mapping
, index
);
1814 if (xa_is_value(page
)) {
1815 error
= shmem_swapin_page(inode
, index
, &page
,
1816 sgp
, gfp
, vma
, fault_type
);
1817 if (error
== -EEXIST
)
1824 if (page
&& sgp
== SGP_WRITE
)
1825 mark_page_accessed(page
);
1827 /* fallocated page? */
1828 if (page
&& !PageUptodate(page
)) {
1829 if (sgp
!= SGP_READ
)
1835 if (page
|| sgp
== SGP_READ
) {
1841 * Fast cache lookup did not find it:
1842 * bring it back from swap or allocate.
1845 if (vma
&& userfaultfd_missing(vma
)) {
1846 *fault_type
= handle_userfault(vmf
, VM_UFFD_MISSING
);
1850 /* shmem_symlink() */
1851 if (mapping
->a_ops
!= &shmem_aops
)
1853 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1855 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1857 switch (sbinfo
->huge
) {
1858 case SHMEM_HUGE_NEVER
:
1860 case SHMEM_HUGE_WITHIN_SIZE
: {
1864 off
= round_up(index
, HPAGE_PMD_NR
);
1865 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1866 if (i_size
>= HPAGE_PMD_SIZE
&&
1867 i_size
>> PAGE_SHIFT
>= off
)
1872 case SHMEM_HUGE_ADVISE
:
1873 if (sgp_huge
== SGP_HUGE
)
1875 /* TODO: implement fadvise() hints */
1880 page
= shmem_alloc_and_acct_page(gfp
, inode
, index
, true);
1883 page
= shmem_alloc_and_acct_page(gfp
, inode
,
1889 error
= PTR_ERR(page
);
1891 if (error
!= -ENOSPC
)
1894 * Try to reclaim some space by splitting a huge page
1895 * beyond i_size on the filesystem.
1900 ret
= shmem_unused_huge_shrink(sbinfo
, NULL
, 1);
1901 if (ret
== SHRINK_STOP
)
1909 if (PageTransHuge(page
))
1910 hindex
= round_down(index
, HPAGE_PMD_NR
);
1914 if (sgp
== SGP_WRITE
)
1915 __SetPageReferenced(page
);
1917 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1918 NULL
, gfp
& GFP_RECLAIM_MASK
,
1922 lru_cache_add(page
);
1924 spin_lock_irq(&info
->lock
);
1925 info
->alloced
+= compound_nr(page
);
1926 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1927 shmem_recalc_inode(inode
);
1928 spin_unlock_irq(&info
->lock
);
1931 if (PageTransHuge(page
) &&
1932 DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
) <
1933 hindex
+ HPAGE_PMD_NR
- 1) {
1935 * Part of the huge page is beyond i_size: subject
1936 * to shrink under memory pressure.
1938 spin_lock(&sbinfo
->shrinklist_lock
);
1940 * _careful to defend against unlocked access to
1941 * ->shrink_list in shmem_unused_huge_shrink()
1943 if (list_empty_careful(&info
->shrinklist
)) {
1944 list_add_tail(&info
->shrinklist
,
1945 &sbinfo
->shrinklist
);
1946 sbinfo
->shrinklist_len
++;
1948 spin_unlock(&sbinfo
->shrinklist_lock
);
1952 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1954 if (sgp
== SGP_FALLOC
)
1958 * Let SGP_WRITE caller clear ends if write does not fill page;
1959 * but SGP_FALLOC on a page fallocated earlier must initialize
1960 * it now, lest undo on failure cancel our earlier guarantee.
1962 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1963 struct page
*head
= compound_head(page
);
1966 for (i
= 0; i
< compound_nr(head
); i
++) {
1967 clear_highpage(head
+ i
);
1968 flush_dcache_page(head
+ i
);
1970 SetPageUptodate(head
);
1973 /* Perhaps the file has been truncated since we checked */
1974 if (sgp
<= SGP_CACHE
&&
1975 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1977 ClearPageDirty(page
);
1978 delete_from_page_cache(page
);
1979 spin_lock_irq(&info
->lock
);
1980 shmem_recalc_inode(inode
);
1981 spin_unlock_irq(&info
->lock
);
1986 *pagep
= page
+ index
- hindex
;
1993 shmem_inode_unacct_blocks(inode
, compound_nr(page
));
1995 if (PageTransHuge(page
)) {
2005 if (error
== -ENOSPC
&& !once
++) {
2006 spin_lock_irq(&info
->lock
);
2007 shmem_recalc_inode(inode
);
2008 spin_unlock_irq(&info
->lock
);
2011 if (error
== -EEXIST
)
2017 * This is like autoremove_wake_function, but it removes the wait queue
2018 * entry unconditionally - even if something else had already woken the
2021 static int synchronous_wake_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2023 int ret
= default_wake_function(wait
, mode
, sync
, key
);
2024 list_del_init(&wait
->entry
);
2028 static vm_fault_t
shmem_fault(struct vm_fault
*vmf
)
2030 struct vm_area_struct
*vma
= vmf
->vma
;
2031 struct inode
*inode
= file_inode(vma
->vm_file
);
2032 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
2035 vm_fault_t ret
= VM_FAULT_LOCKED
;
2038 * Trinity finds that probing a hole which tmpfs is punching can
2039 * prevent the hole-punch from ever completing: which in turn
2040 * locks writers out with its hold on i_mutex. So refrain from
2041 * faulting pages into the hole while it's being punched. Although
2042 * shmem_undo_range() does remove the additions, it may be unable to
2043 * keep up, as each new page needs its own unmap_mapping_range() call,
2044 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2046 * It does not matter if we sometimes reach this check just before the
2047 * hole-punch begins, so that one fault then races with the punch:
2048 * we just need to make racing faults a rare case.
2050 * The implementation below would be much simpler if we just used a
2051 * standard mutex or completion: but we cannot take i_mutex in fault,
2052 * and bloating every shmem inode for this unlikely case would be sad.
2054 if (unlikely(inode
->i_private
)) {
2055 struct shmem_falloc
*shmem_falloc
;
2057 spin_lock(&inode
->i_lock
);
2058 shmem_falloc
= inode
->i_private
;
2060 shmem_falloc
->waitq
&&
2061 vmf
->pgoff
>= shmem_falloc
->start
&&
2062 vmf
->pgoff
< shmem_falloc
->next
) {
2064 wait_queue_head_t
*shmem_falloc_waitq
;
2065 DEFINE_WAIT_FUNC(shmem_fault_wait
, synchronous_wake_function
);
2067 ret
= VM_FAULT_NOPAGE
;
2068 fpin
= maybe_unlock_mmap_for_io(vmf
, NULL
);
2070 ret
= VM_FAULT_RETRY
;
2072 shmem_falloc_waitq
= shmem_falloc
->waitq
;
2073 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
2074 TASK_UNINTERRUPTIBLE
);
2075 spin_unlock(&inode
->i_lock
);
2079 * shmem_falloc_waitq points into the shmem_fallocate()
2080 * stack of the hole-punching task: shmem_falloc_waitq
2081 * is usually invalid by the time we reach here, but
2082 * finish_wait() does not dereference it in that case;
2083 * though i_lock needed lest racing with wake_up_all().
2085 spin_lock(&inode
->i_lock
);
2086 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
2087 spin_unlock(&inode
->i_lock
);
2093 spin_unlock(&inode
->i_lock
);
2098 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
2099 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
2101 else if (vma
->vm_flags
& VM_HUGEPAGE
)
2104 err
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
2105 gfp
, vma
, vmf
, &ret
);
2107 return vmf_error(err
);
2111 unsigned long shmem_get_unmapped_area(struct file
*file
,
2112 unsigned long uaddr
, unsigned long len
,
2113 unsigned long pgoff
, unsigned long flags
)
2115 unsigned long (*get_area
)(struct file
*,
2116 unsigned long, unsigned long, unsigned long, unsigned long);
2118 unsigned long offset
;
2119 unsigned long inflated_len
;
2120 unsigned long inflated_addr
;
2121 unsigned long inflated_offset
;
2123 if (len
> TASK_SIZE
)
2126 get_area
= current
->mm
->get_unmapped_area
;
2127 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
2129 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
2131 if (IS_ERR_VALUE(addr
))
2133 if (addr
& ~PAGE_MASK
)
2135 if (addr
> TASK_SIZE
- len
)
2138 if (shmem_huge
== SHMEM_HUGE_DENY
)
2140 if (len
< HPAGE_PMD_SIZE
)
2142 if (flags
& MAP_FIXED
)
2145 * Our priority is to support MAP_SHARED mapped hugely;
2146 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2147 * But if caller specified an address hint and we allocated area there
2148 * successfully, respect that as before.
2153 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
2154 struct super_block
*sb
;
2157 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
2158 sb
= file_inode(file
)->i_sb
;
2161 * Called directly from mm/mmap.c, or drivers/char/mem.c
2162 * for "/dev/zero", to create a shared anonymous object.
2164 if (IS_ERR(shm_mnt
))
2166 sb
= shm_mnt
->mnt_sb
;
2168 if (SHMEM_SB(sb
)->huge
== SHMEM_HUGE_NEVER
)
2172 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
2173 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
2175 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
2178 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
2179 if (inflated_len
> TASK_SIZE
)
2181 if (inflated_len
< len
)
2184 inflated_addr
= get_area(NULL
, uaddr
, inflated_len
, 0, flags
);
2185 if (IS_ERR_VALUE(inflated_addr
))
2187 if (inflated_addr
& ~PAGE_MASK
)
2190 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
2191 inflated_addr
+= offset
- inflated_offset
;
2192 if (inflated_offset
> offset
)
2193 inflated_addr
+= HPAGE_PMD_SIZE
;
2195 if (inflated_addr
> TASK_SIZE
- len
)
2197 return inflated_addr
;
2201 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
2203 struct inode
*inode
= file_inode(vma
->vm_file
);
2204 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
2207 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
2210 struct inode
*inode
= file_inode(vma
->vm_file
);
2213 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2214 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
2218 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2220 struct inode
*inode
= file_inode(file
);
2221 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2222 int retval
= -ENOMEM
;
2225 * What serializes the accesses to info->flags?
2226 * ipc_lock_object() when called from shmctl_do_lock(),
2227 * no serialization needed when called from shm_destroy().
2229 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
2230 if (!user_shm_lock(inode
->i_size
, user
))
2232 info
->flags
|= VM_LOCKED
;
2233 mapping_set_unevictable(file
->f_mapping
);
2235 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
2236 user_shm_unlock(inode
->i_size
, user
);
2237 info
->flags
&= ~VM_LOCKED
;
2238 mapping_clear_unevictable(file
->f_mapping
);
2246 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2248 struct shmem_inode_info
*info
= SHMEM_I(file_inode(file
));
2250 if (info
->seals
& F_SEAL_FUTURE_WRITE
) {
2252 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2253 * "future write" seal active.
2255 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_WRITE
))
2259 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2260 * MAP_SHARED and read-only, take care to not allow mprotect to
2261 * revert protections on such mappings. Do this only for shared
2262 * mappings. For private mappings, don't need to mask
2263 * VM_MAYWRITE as we still want them to be COW-writable.
2265 if (vma
->vm_flags
& VM_SHARED
)
2266 vma
->vm_flags
&= ~(VM_MAYWRITE
);
2269 file_accessed(file
);
2270 vma
->vm_ops
= &shmem_vm_ops
;
2271 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
2272 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
2273 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
2274 khugepaged_enter(vma
, vma
->vm_flags
);
2279 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
2280 umode_t mode
, dev_t dev
, unsigned long flags
)
2282 struct inode
*inode
;
2283 struct shmem_inode_info
*info
;
2284 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2287 if (shmem_reserve_inode(sb
, &ino
))
2290 inode
= new_inode(sb
);
2293 inode_init_owner(inode
, dir
, mode
);
2294 inode
->i_blocks
= 0;
2295 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2296 inode
->i_generation
= prandom_u32();
2297 info
= SHMEM_I(inode
);
2298 memset(info
, 0, (char *)inode
- (char *)info
);
2299 spin_lock_init(&info
->lock
);
2300 atomic_set(&info
->stop_eviction
, 0);
2301 info
->seals
= F_SEAL_SEAL
;
2302 info
->flags
= flags
& VM_NORESERVE
;
2303 INIT_LIST_HEAD(&info
->shrinklist
);
2304 INIT_LIST_HEAD(&info
->swaplist
);
2305 simple_xattrs_init(&info
->xattrs
);
2306 cache_no_acl(inode
);
2308 switch (mode
& S_IFMT
) {
2310 inode
->i_op
= &shmem_special_inode_operations
;
2311 init_special_inode(inode
, mode
, dev
);
2314 inode
->i_mapping
->a_ops
= &shmem_aops
;
2315 inode
->i_op
= &shmem_inode_operations
;
2316 inode
->i_fop
= &shmem_file_operations
;
2317 mpol_shared_policy_init(&info
->policy
,
2318 shmem_get_sbmpol(sbinfo
));
2322 /* Some things misbehave if size == 0 on a directory */
2323 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
2324 inode
->i_op
= &shmem_dir_inode_operations
;
2325 inode
->i_fop
= &simple_dir_operations
;
2329 * Must not load anything in the rbtree,
2330 * mpol_free_shared_policy will not be called.
2332 mpol_shared_policy_init(&info
->policy
, NULL
);
2336 lockdep_annotate_inode_mutex_key(inode
);
2338 shmem_free_inode(sb
);
2342 bool shmem_mapping(struct address_space
*mapping
)
2344 return mapping
->a_ops
== &shmem_aops
;
2347 static int shmem_mfill_atomic_pte(struct mm_struct
*dst_mm
,
2349 struct vm_area_struct
*dst_vma
,
2350 unsigned long dst_addr
,
2351 unsigned long src_addr
,
2353 struct page
**pagep
)
2355 struct inode
*inode
= file_inode(dst_vma
->vm_file
);
2356 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2357 struct address_space
*mapping
= inode
->i_mapping
;
2358 gfp_t gfp
= mapping_gfp_mask(mapping
);
2359 pgoff_t pgoff
= linear_page_index(dst_vma
, dst_addr
);
2363 pte_t _dst_pte
, *dst_pte
;
2365 pgoff_t offset
, max_off
;
2368 if (!shmem_inode_acct_block(inode
, 1))
2372 page
= shmem_alloc_page(gfp
, info
, pgoff
);
2374 goto out_unacct_blocks
;
2376 if (!zeropage
) { /* mcopy_atomic */
2377 page_kaddr
= kmap_atomic(page
);
2378 ret
= copy_from_user(page_kaddr
,
2379 (const void __user
*)src_addr
,
2381 kunmap_atomic(page_kaddr
);
2383 /* fallback to copy_from_user outside mmap_lock */
2384 if (unlikely(ret
)) {
2386 shmem_inode_unacct_blocks(inode
, 1);
2387 /* don't free the page */
2390 } else { /* mfill_zeropage_atomic */
2391 clear_highpage(page
);
2398 VM_BUG_ON(PageLocked(page
) || PageSwapBacked(page
));
2399 __SetPageLocked(page
);
2400 __SetPageSwapBacked(page
);
2401 __SetPageUptodate(page
);
2404 offset
= linear_page_index(dst_vma
, dst_addr
);
2405 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2406 if (unlikely(offset
>= max_off
))
2409 ret
= shmem_add_to_page_cache(page
, mapping
, pgoff
, NULL
,
2410 gfp
& GFP_RECLAIM_MASK
, dst_mm
);
2414 _dst_pte
= mk_pte(page
, dst_vma
->vm_page_prot
);
2415 if (dst_vma
->vm_flags
& VM_WRITE
)
2416 _dst_pte
= pte_mkwrite(pte_mkdirty(_dst_pte
));
2419 * We don't set the pte dirty if the vma has no
2420 * VM_WRITE permission, so mark the page dirty or it
2421 * could be freed from under us. We could do it
2422 * unconditionally before unlock_page(), but doing it
2423 * only if VM_WRITE is not set is faster.
2425 set_page_dirty(page
);
2428 dst_pte
= pte_offset_map_lock(dst_mm
, dst_pmd
, dst_addr
, &ptl
);
2431 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2432 if (unlikely(offset
>= max_off
))
2433 goto out_release_unlock
;
2436 if (!pte_none(*dst_pte
))
2437 goto out_release_unlock
;
2439 lru_cache_add(page
);
2441 spin_lock_irq(&info
->lock
);
2443 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
2444 shmem_recalc_inode(inode
);
2445 spin_unlock_irq(&info
->lock
);
2447 inc_mm_counter(dst_mm
, mm_counter_file(page
));
2448 page_add_file_rmap(page
, false);
2449 set_pte_at(dst_mm
, dst_addr
, dst_pte
, _dst_pte
);
2451 /* No need to invalidate - it was non-present before */
2452 update_mmu_cache(dst_vma
, dst_addr
, dst_pte
);
2453 pte_unmap_unlock(dst_pte
, ptl
);
2459 pte_unmap_unlock(dst_pte
, ptl
);
2460 ClearPageDirty(page
);
2461 delete_from_page_cache(page
);
2466 shmem_inode_unacct_blocks(inode
, 1);
2470 int shmem_mcopy_atomic_pte(struct mm_struct
*dst_mm
,
2472 struct vm_area_struct
*dst_vma
,
2473 unsigned long dst_addr
,
2474 unsigned long src_addr
,
2475 struct page
**pagep
)
2477 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2478 dst_addr
, src_addr
, false, pagep
);
2481 int shmem_mfill_zeropage_pte(struct mm_struct
*dst_mm
,
2483 struct vm_area_struct
*dst_vma
,
2484 unsigned long dst_addr
)
2486 struct page
*page
= NULL
;
2488 return shmem_mfill_atomic_pte(dst_mm
, dst_pmd
, dst_vma
,
2489 dst_addr
, 0, true, &page
);
2493 static const struct inode_operations shmem_symlink_inode_operations
;
2494 static const struct inode_operations shmem_short_symlink_operations
;
2496 #ifdef CONFIG_TMPFS_XATTR
2497 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
2499 #define shmem_initxattrs NULL
2503 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
2504 loff_t pos
, unsigned len
, unsigned flags
,
2505 struct page
**pagep
, void **fsdata
)
2507 struct inode
*inode
= mapping
->host
;
2508 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2509 pgoff_t index
= pos
>> PAGE_SHIFT
;
2511 /* i_mutex is held by caller */
2512 if (unlikely(info
->seals
& (F_SEAL_GROW
|
2513 F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))) {
2514 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
))
2516 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
2520 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2524 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2525 loff_t pos
, unsigned len
, unsigned copied
,
2526 struct page
*page
, void *fsdata
)
2528 struct inode
*inode
= mapping
->host
;
2530 if (pos
+ copied
> inode
->i_size
)
2531 i_size_write(inode
, pos
+ copied
);
2533 if (!PageUptodate(page
)) {
2534 struct page
*head
= compound_head(page
);
2535 if (PageTransCompound(page
)) {
2538 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2539 if (head
+ i
== page
)
2541 clear_highpage(head
+ i
);
2542 flush_dcache_page(head
+ i
);
2545 if (copied
< PAGE_SIZE
) {
2546 unsigned from
= pos
& (PAGE_SIZE
- 1);
2547 zero_user_segments(page
, 0, from
,
2548 from
+ copied
, PAGE_SIZE
);
2550 SetPageUptodate(head
);
2552 set_page_dirty(page
);
2559 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2561 struct file
*file
= iocb
->ki_filp
;
2562 struct inode
*inode
= file_inode(file
);
2563 struct address_space
*mapping
= inode
->i_mapping
;
2565 unsigned long offset
;
2566 enum sgp_type sgp
= SGP_READ
;
2569 loff_t
*ppos
= &iocb
->ki_pos
;
2572 * Might this read be for a stacking filesystem? Then when reading
2573 * holes of a sparse file, we actually need to allocate those pages,
2574 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2576 if (!iter_is_iovec(to
))
2579 index
= *ppos
>> PAGE_SHIFT
;
2580 offset
= *ppos
& ~PAGE_MASK
;
2583 struct page
*page
= NULL
;
2585 unsigned long nr
, ret
;
2586 loff_t i_size
= i_size_read(inode
);
2588 end_index
= i_size
>> PAGE_SHIFT
;
2589 if (index
> end_index
)
2591 if (index
== end_index
) {
2592 nr
= i_size
& ~PAGE_MASK
;
2597 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2599 if (error
== -EINVAL
)
2604 if (sgp
== SGP_CACHE
)
2605 set_page_dirty(page
);
2610 * We must evaluate after, since reads (unlike writes)
2611 * are called without i_mutex protection against truncate
2614 i_size
= i_size_read(inode
);
2615 end_index
= i_size
>> PAGE_SHIFT
;
2616 if (index
== end_index
) {
2617 nr
= i_size
& ~PAGE_MASK
;
2628 * If users can be writing to this page using arbitrary
2629 * virtual addresses, take care about potential aliasing
2630 * before reading the page on the kernel side.
2632 if (mapping_writably_mapped(mapping
))
2633 flush_dcache_page(page
);
2635 * Mark the page accessed if we read the beginning.
2638 mark_page_accessed(page
);
2640 page
= ZERO_PAGE(0);
2645 * Ok, we have the page, and it's up-to-date, so
2646 * now we can copy it to user space...
2648 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2651 index
+= offset
>> PAGE_SHIFT
;
2652 offset
&= ~PAGE_MASK
;
2655 if (!iov_iter_count(to
))
2664 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2665 file_accessed(file
);
2666 return retval
? retval
: error
;
2670 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2672 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2673 pgoff_t index
, pgoff_t end
, int whence
)
2676 struct pagevec pvec
;
2677 pgoff_t indices
[PAGEVEC_SIZE
];
2681 pagevec_init(&pvec
);
2682 pvec
.nr
= 1; /* start small: we may be there already */
2684 pvec
.nr
= find_get_entries(mapping
, index
,
2685 pvec
.nr
, pvec
.pages
, indices
);
2687 if (whence
== SEEK_DATA
)
2691 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2692 if (index
< indices
[i
]) {
2693 if (whence
== SEEK_HOLE
) {
2699 page
= pvec
.pages
[i
];
2700 if (page
&& !xa_is_value(page
)) {
2701 if (!PageUptodate(page
))
2705 (page
&& whence
== SEEK_DATA
) ||
2706 (!page
&& whence
== SEEK_HOLE
)) {
2711 pagevec_remove_exceptionals(&pvec
);
2712 pagevec_release(&pvec
);
2713 pvec
.nr
= PAGEVEC_SIZE
;
2719 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2721 struct address_space
*mapping
= file
->f_mapping
;
2722 struct inode
*inode
= mapping
->host
;
2726 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2727 return generic_file_llseek_size(file
, offset
, whence
,
2728 MAX_LFS_FILESIZE
, i_size_read(inode
));
2730 /* We're holding i_mutex so we can access i_size directly */
2732 if (offset
< 0 || offset
>= inode
->i_size
)
2735 start
= offset
>> PAGE_SHIFT
;
2736 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2737 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2738 new_offset
<<= PAGE_SHIFT
;
2739 if (new_offset
> offset
) {
2740 if (new_offset
< inode
->i_size
)
2741 offset
= new_offset
;
2742 else if (whence
== SEEK_DATA
)
2745 offset
= inode
->i_size
;
2750 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2751 inode_unlock(inode
);
2755 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2758 struct inode
*inode
= file_inode(file
);
2759 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2760 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2761 struct shmem_falloc shmem_falloc
;
2762 pgoff_t start
, index
, end
;
2765 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2770 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2771 struct address_space
*mapping
= file
->f_mapping
;
2772 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2773 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2774 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2776 /* protected by i_mutex */
2777 if (info
->seals
& (F_SEAL_WRITE
| F_SEAL_FUTURE_WRITE
)) {
2782 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2783 shmem_falloc
.start
= (u64
)unmap_start
>> PAGE_SHIFT
;
2784 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2785 spin_lock(&inode
->i_lock
);
2786 inode
->i_private
= &shmem_falloc
;
2787 spin_unlock(&inode
->i_lock
);
2789 if ((u64
)unmap_end
> (u64
)unmap_start
)
2790 unmap_mapping_range(mapping
, unmap_start
,
2791 1 + unmap_end
- unmap_start
, 0);
2792 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2793 /* No need to unmap again: hole-punching leaves COWed pages */
2795 spin_lock(&inode
->i_lock
);
2796 inode
->i_private
= NULL
;
2797 wake_up_all(&shmem_falloc_waitq
);
2798 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq
.head
));
2799 spin_unlock(&inode
->i_lock
);
2804 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2805 error
= inode_newsize_ok(inode
, offset
+ len
);
2809 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2814 start
= offset
>> PAGE_SHIFT
;
2815 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2816 /* Try to avoid a swapstorm if len is impossible to satisfy */
2817 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2822 shmem_falloc
.waitq
= NULL
;
2823 shmem_falloc
.start
= start
;
2824 shmem_falloc
.next
= start
;
2825 shmem_falloc
.nr_falloced
= 0;
2826 shmem_falloc
.nr_unswapped
= 0;
2827 spin_lock(&inode
->i_lock
);
2828 inode
->i_private
= &shmem_falloc
;
2829 spin_unlock(&inode
->i_lock
);
2831 for (index
= start
; index
< end
; index
++) {
2835 * Good, the fallocate(2) manpage permits EINTR: we may have
2836 * been interrupted because we are using up too much memory.
2838 if (signal_pending(current
))
2840 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2843 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2845 /* Remove the !PageUptodate pages we added */
2846 if (index
> start
) {
2847 shmem_undo_range(inode
,
2848 (loff_t
)start
<< PAGE_SHIFT
,
2849 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2855 * Inform shmem_writepage() how far we have reached.
2856 * No need for lock or barrier: we have the page lock.
2858 shmem_falloc
.next
++;
2859 if (!PageUptodate(page
))
2860 shmem_falloc
.nr_falloced
++;
2863 * If !PageUptodate, leave it that way so that freeable pages
2864 * can be recognized if we need to rollback on error later.
2865 * But set_page_dirty so that memory pressure will swap rather
2866 * than free the pages we are allocating (and SGP_CACHE pages
2867 * might still be clean: we now need to mark those dirty too).
2869 set_page_dirty(page
);
2875 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2876 i_size_write(inode
, offset
+ len
);
2877 inode
->i_ctime
= current_time(inode
);
2879 spin_lock(&inode
->i_lock
);
2880 inode
->i_private
= NULL
;
2881 spin_unlock(&inode
->i_lock
);
2883 inode_unlock(inode
);
2887 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2889 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2891 buf
->f_type
= TMPFS_MAGIC
;
2892 buf
->f_bsize
= PAGE_SIZE
;
2893 buf
->f_namelen
= NAME_MAX
;
2894 if (sbinfo
->max_blocks
) {
2895 buf
->f_blocks
= sbinfo
->max_blocks
;
2897 buf
->f_bfree
= sbinfo
->max_blocks
-
2898 percpu_counter_sum(&sbinfo
->used_blocks
);
2900 if (sbinfo
->max_inodes
) {
2901 buf
->f_files
= sbinfo
->max_inodes
;
2902 buf
->f_ffree
= sbinfo
->free_inodes
;
2904 /* else leave those fields 0 like simple_statfs */
2909 * File creation. Allocate an inode, and we're done..
2912 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2914 struct inode
*inode
;
2915 int error
= -ENOSPC
;
2917 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2919 error
= simple_acl_create(dir
, inode
);
2922 error
= security_inode_init_security(inode
, dir
,
2924 shmem_initxattrs
, NULL
);
2925 if (error
&& error
!= -EOPNOTSUPP
)
2929 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2930 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
2931 d_instantiate(dentry
, inode
);
2932 dget(dentry
); /* Extra count - pin the dentry in core */
2941 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2943 struct inode
*inode
;
2944 int error
= -ENOSPC
;
2946 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2948 error
= security_inode_init_security(inode
, dir
,
2950 shmem_initxattrs
, NULL
);
2951 if (error
&& error
!= -EOPNOTSUPP
)
2953 error
= simple_acl_create(dir
, inode
);
2956 d_tmpfile(dentry
, inode
);
2964 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2968 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2974 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2977 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2983 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2985 struct inode
*inode
= d_inode(old_dentry
);
2989 * No ordinary (disk based) filesystem counts links as inodes;
2990 * but each new link needs a new dentry, pinning lowmem, and
2991 * tmpfs dentries cannot be pruned until they are unlinked.
2992 * But if an O_TMPFILE file is linked into the tmpfs, the
2993 * first link must skip that, to get the accounting right.
2995 if (inode
->i_nlink
) {
2996 ret
= shmem_reserve_inode(inode
->i_sb
, NULL
);
3001 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3002 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3004 ihold(inode
); /* New dentry reference */
3005 dget(dentry
); /* Extra pinning count for the created dentry */
3006 d_instantiate(dentry
, inode
);
3011 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
3013 struct inode
*inode
= d_inode(dentry
);
3015 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
3016 shmem_free_inode(inode
->i_sb
);
3018 dir
->i_size
-= BOGO_DIRENT_SIZE
;
3019 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= current_time(inode
);
3021 dput(dentry
); /* Undo the count from "create" - this does all the work */
3025 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3027 if (!simple_empty(dentry
))
3030 drop_nlink(d_inode(dentry
));
3032 return shmem_unlink(dir
, dentry
);
3035 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
3037 bool old_is_dir
= d_is_dir(old_dentry
);
3038 bool new_is_dir
= d_is_dir(new_dentry
);
3040 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
3042 drop_nlink(old_dir
);
3045 drop_nlink(new_dir
);
3049 old_dir
->i_ctime
= old_dir
->i_mtime
=
3050 new_dir
->i_ctime
= new_dir
->i_mtime
=
3051 d_inode(old_dentry
)->i_ctime
=
3052 d_inode(new_dentry
)->i_ctime
= current_time(old_dir
);
3057 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
3059 struct dentry
*whiteout
;
3062 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
3066 error
= shmem_mknod(old_dir
, whiteout
,
3067 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
3073 * Cheat and hash the whiteout while the old dentry is still in
3074 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3076 * d_lookup() will consistently find one of them at this point,
3077 * not sure which one, but that isn't even important.
3084 * The VFS layer already does all the dentry stuff for rename,
3085 * we just have to decrement the usage count for the target if
3086 * it exists so that the VFS layer correctly free's it when it
3089 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
3091 struct inode
*inode
= d_inode(old_dentry
);
3092 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
3094 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
3097 if (flags
& RENAME_EXCHANGE
)
3098 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
3100 if (!simple_empty(new_dentry
))
3103 if (flags
& RENAME_WHITEOUT
) {
3106 error
= shmem_whiteout(old_dir
, old_dentry
);
3111 if (d_really_is_positive(new_dentry
)) {
3112 (void) shmem_unlink(new_dir
, new_dentry
);
3113 if (they_are_dirs
) {
3114 drop_nlink(d_inode(new_dentry
));
3115 drop_nlink(old_dir
);
3117 } else if (they_are_dirs
) {
3118 drop_nlink(old_dir
);
3122 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
3123 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
3124 old_dir
->i_ctime
= old_dir
->i_mtime
=
3125 new_dir
->i_ctime
= new_dir
->i_mtime
=
3126 inode
->i_ctime
= current_time(old_dir
);
3130 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
3134 struct inode
*inode
;
3137 len
= strlen(symname
) + 1;
3138 if (len
> PAGE_SIZE
)
3139 return -ENAMETOOLONG
;
3141 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
| 0777, 0,
3146 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
3147 shmem_initxattrs
, NULL
);
3148 if (error
&& error
!= -EOPNOTSUPP
) {
3153 inode
->i_size
= len
-1;
3154 if (len
<= SHORT_SYMLINK_LEN
) {
3155 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
3156 if (!inode
->i_link
) {
3160 inode
->i_op
= &shmem_short_symlink_operations
;
3162 inode_nohighmem(inode
);
3163 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
3168 inode
->i_mapping
->a_ops
= &shmem_aops
;
3169 inode
->i_op
= &shmem_symlink_inode_operations
;
3170 memcpy(page_address(page
), symname
, len
);
3171 SetPageUptodate(page
);
3172 set_page_dirty(page
);
3176 dir
->i_size
+= BOGO_DIRENT_SIZE
;
3177 dir
->i_ctime
= dir
->i_mtime
= current_time(dir
);
3178 d_instantiate(dentry
, inode
);
3183 static void shmem_put_link(void *arg
)
3185 mark_page_accessed(arg
);
3189 static const char *shmem_get_link(struct dentry
*dentry
,
3190 struct inode
*inode
,
3191 struct delayed_call
*done
)
3193 struct page
*page
= NULL
;
3196 page
= find_get_page(inode
->i_mapping
, 0);
3198 return ERR_PTR(-ECHILD
);
3199 if (!PageUptodate(page
)) {
3201 return ERR_PTR(-ECHILD
);
3204 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3206 return ERR_PTR(error
);
3209 set_delayed_call(done
, shmem_put_link
, page
);
3210 return page_address(page
);
3213 #ifdef CONFIG_TMPFS_XATTR
3215 * Superblocks without xattr inode operations may get some security.* xattr
3216 * support from the LSM "for free". As soon as we have any other xattrs
3217 * like ACLs, we also need to implement the security.* handlers at
3218 * filesystem level, though.
3222 * Callback for security_inode_init_security() for acquiring xattrs.
3224 static int shmem_initxattrs(struct inode
*inode
,
3225 const struct xattr
*xattr_array
,
3228 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3229 const struct xattr
*xattr
;
3230 struct simple_xattr
*new_xattr
;
3233 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3234 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3238 len
= strlen(xattr
->name
) + 1;
3239 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3241 if (!new_xattr
->name
) {
3246 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3247 XATTR_SECURITY_PREFIX_LEN
);
3248 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3251 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3257 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3258 struct dentry
*unused
, struct inode
*inode
,
3259 const char *name
, void *buffer
, size_t size
)
3261 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3263 name
= xattr_full_name(handler
, name
);
3264 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3267 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3268 struct dentry
*unused
, struct inode
*inode
,
3269 const char *name
, const void *value
,
3270 size_t size
, int flags
)
3272 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3274 name
= xattr_full_name(handler
, name
);
3275 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
, NULL
);
3278 static const struct xattr_handler shmem_security_xattr_handler
= {
3279 .prefix
= XATTR_SECURITY_PREFIX
,
3280 .get
= shmem_xattr_handler_get
,
3281 .set
= shmem_xattr_handler_set
,
3284 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3285 .prefix
= XATTR_TRUSTED_PREFIX
,
3286 .get
= shmem_xattr_handler_get
,
3287 .set
= shmem_xattr_handler_set
,
3290 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3291 #ifdef CONFIG_TMPFS_POSIX_ACL
3292 &posix_acl_access_xattr_handler
,
3293 &posix_acl_default_xattr_handler
,
3295 &shmem_security_xattr_handler
,
3296 &shmem_trusted_xattr_handler
,
3300 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3302 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3303 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3305 #endif /* CONFIG_TMPFS_XATTR */
3307 static const struct inode_operations shmem_short_symlink_operations
= {
3308 .get_link
= simple_get_link
,
3309 #ifdef CONFIG_TMPFS_XATTR
3310 .listxattr
= shmem_listxattr
,
3314 static const struct inode_operations shmem_symlink_inode_operations
= {
3315 .get_link
= shmem_get_link
,
3316 #ifdef CONFIG_TMPFS_XATTR
3317 .listxattr
= shmem_listxattr
,
3321 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3323 return ERR_PTR(-ESTALE
);
3326 static int shmem_match(struct inode
*ino
, void *vfh
)
3330 inum
= (inum
<< 32) | fh
[1];
3331 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3334 /* Find any alias of inode, but prefer a hashed alias */
3335 static struct dentry
*shmem_find_alias(struct inode
*inode
)
3337 struct dentry
*alias
= d_find_alias(inode
);
3339 return alias
?: d_find_any_alias(inode
);
3343 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3344 struct fid
*fid
, int fh_len
, int fh_type
)
3346 struct inode
*inode
;
3347 struct dentry
*dentry
= NULL
;
3354 inum
= (inum
<< 32) | fid
->raw
[1];
3356 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3357 shmem_match
, fid
->raw
);
3359 dentry
= shmem_find_alias(inode
);
3366 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3367 struct inode
*parent
)
3371 return FILEID_INVALID
;
3374 if (inode_unhashed(inode
)) {
3375 /* Unfortunately insert_inode_hash is not idempotent,
3376 * so as we hash inodes here rather than at creation
3377 * time, we need a lock to ensure we only try
3380 static DEFINE_SPINLOCK(lock
);
3382 if (inode_unhashed(inode
))
3383 __insert_inode_hash(inode
,
3384 inode
->i_ino
+ inode
->i_generation
);
3388 fh
[0] = inode
->i_generation
;
3389 fh
[1] = inode
->i_ino
;
3390 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3396 static const struct export_operations shmem_export_ops
= {
3397 .get_parent
= shmem_get_parent
,
3398 .encode_fh
= shmem_encode_fh
,
3399 .fh_to_dentry
= shmem_fh_to_dentry
,
3415 static const struct constant_table shmem_param_enums_huge
[] = {
3416 {"never", SHMEM_HUGE_NEVER
},
3417 {"always", SHMEM_HUGE_ALWAYS
},
3418 {"within_size", SHMEM_HUGE_WITHIN_SIZE
},
3419 {"advise", SHMEM_HUGE_ADVISE
},
3423 const struct fs_parameter_spec shmem_fs_parameters
[] = {
3424 fsparam_u32 ("gid", Opt_gid
),
3425 fsparam_enum ("huge", Opt_huge
, shmem_param_enums_huge
),
3426 fsparam_u32oct("mode", Opt_mode
),
3427 fsparam_string("mpol", Opt_mpol
),
3428 fsparam_string("nr_blocks", Opt_nr_blocks
),
3429 fsparam_string("nr_inodes", Opt_nr_inodes
),
3430 fsparam_string("size", Opt_size
),
3431 fsparam_u32 ("uid", Opt_uid
),
3432 fsparam_flag ("inode32", Opt_inode32
),
3433 fsparam_flag ("inode64", Opt_inode64
),
3437 static int shmem_parse_one(struct fs_context
*fc
, struct fs_parameter
*param
)
3439 struct shmem_options
*ctx
= fc
->fs_private
;
3440 struct fs_parse_result result
;
3441 unsigned long long size
;
3445 opt
= fs_parse(fc
, shmem_fs_parameters
, param
, &result
);
3451 size
= memparse(param
->string
, &rest
);
3453 size
<<= PAGE_SHIFT
;
3454 size
*= totalram_pages();
3460 ctx
->blocks
= DIV_ROUND_UP(size
, PAGE_SIZE
);
3461 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3464 ctx
->blocks
= memparse(param
->string
, &rest
);
3467 ctx
->seen
|= SHMEM_SEEN_BLOCKS
;
3470 ctx
->inodes
= memparse(param
->string
, &rest
);
3473 ctx
->seen
|= SHMEM_SEEN_INODES
;
3476 ctx
->mode
= result
.uint_32
& 07777;
3479 ctx
->uid
= make_kuid(current_user_ns(), result
.uint_32
);
3480 if (!uid_valid(ctx
->uid
))
3484 ctx
->gid
= make_kgid(current_user_ns(), result
.uint_32
);
3485 if (!gid_valid(ctx
->gid
))
3489 ctx
->huge
= result
.uint_32
;
3490 if (ctx
->huge
!= SHMEM_HUGE_NEVER
&&
3491 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
3492 has_transparent_hugepage()))
3493 goto unsupported_parameter
;
3494 ctx
->seen
|= SHMEM_SEEN_HUGE
;
3497 if (IS_ENABLED(CONFIG_NUMA
)) {
3498 mpol_put(ctx
->mpol
);
3500 if (mpol_parse_str(param
->string
, &ctx
->mpol
))
3504 goto unsupported_parameter
;
3506 ctx
->full_inums
= false;
3507 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3510 if (sizeof(ino_t
) < 8) {
3512 "Cannot use inode64 with <64bit inums in kernel\n");
3514 ctx
->full_inums
= true;
3515 ctx
->seen
|= SHMEM_SEEN_INUMS
;
3520 unsupported_parameter
:
3521 return invalfc(fc
, "Unsupported parameter '%s'", param
->key
);
3523 return invalfc(fc
, "Bad value for '%s'", param
->key
);
3526 static int shmem_parse_options(struct fs_context
*fc
, void *data
)
3528 char *options
= data
;
3531 int err
= security_sb_eat_lsm_opts(options
, &fc
->security
);
3536 while (options
!= NULL
) {
3537 char *this_char
= options
;
3540 * NUL-terminate this option: unfortunately,
3541 * mount options form a comma-separated list,
3542 * but mpol's nodelist may also contain commas.
3544 options
= strchr(options
, ',');
3545 if (options
== NULL
)
3548 if (!isdigit(*options
)) {
3554 char *value
= strchr(this_char
,'=');
3560 len
= strlen(value
);
3562 err
= vfs_parse_fs_string(fc
, this_char
, value
, len
);
3571 * Reconfigure a shmem filesystem.
3573 * Note that we disallow change from limited->unlimited blocks/inodes while any
3574 * are in use; but we must separately disallow unlimited->limited, because in
3575 * that case we have no record of how much is already in use.
3577 static int shmem_reconfigure(struct fs_context
*fc
)
3579 struct shmem_options
*ctx
= fc
->fs_private
;
3580 struct shmem_sb_info
*sbinfo
= SHMEM_SB(fc
->root
->d_sb
);
3581 unsigned long inodes
;
3584 spin_lock(&sbinfo
->stat_lock
);
3585 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3586 if ((ctx
->seen
& SHMEM_SEEN_BLOCKS
) && ctx
->blocks
) {
3587 if (!sbinfo
->max_blocks
) {
3588 err
= "Cannot retroactively limit size";
3591 if (percpu_counter_compare(&sbinfo
->used_blocks
,
3593 err
= "Too small a size for current use";
3597 if ((ctx
->seen
& SHMEM_SEEN_INODES
) && ctx
->inodes
) {
3598 if (!sbinfo
->max_inodes
) {
3599 err
= "Cannot retroactively limit inodes";
3602 if (ctx
->inodes
< inodes
) {
3603 err
= "Too few inodes for current use";
3608 if ((ctx
->seen
& SHMEM_SEEN_INUMS
) && !ctx
->full_inums
&&
3609 sbinfo
->next_ino
> UINT_MAX
) {
3610 err
= "Current inum too high to switch to 32-bit inums";
3614 if (ctx
->seen
& SHMEM_SEEN_HUGE
)
3615 sbinfo
->huge
= ctx
->huge
;
3616 if (ctx
->seen
& SHMEM_SEEN_INUMS
)
3617 sbinfo
->full_inums
= ctx
->full_inums
;
3618 if (ctx
->seen
& SHMEM_SEEN_BLOCKS
)
3619 sbinfo
->max_blocks
= ctx
->blocks
;
3620 if (ctx
->seen
& SHMEM_SEEN_INODES
) {
3621 sbinfo
->max_inodes
= ctx
->inodes
;
3622 sbinfo
->free_inodes
= ctx
->inodes
- inodes
;
3626 * Preserve previous mempolicy unless mpol remount option was specified.
3629 mpol_put(sbinfo
->mpol
);
3630 sbinfo
->mpol
= ctx
->mpol
; /* transfers initial ref */
3633 spin_unlock(&sbinfo
->stat_lock
);
3636 spin_unlock(&sbinfo
->stat_lock
);
3637 return invalfc(fc
, "%s", err
);
3640 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3642 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3644 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3645 seq_printf(seq
, ",size=%luk",
3646 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3647 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3648 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3649 if (sbinfo
->mode
!= (0777 | S_ISVTX
))
3650 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3651 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3652 seq_printf(seq
, ",uid=%u",
3653 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3654 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3655 seq_printf(seq
, ",gid=%u",
3656 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3659 * Showing inode{64,32} might be useful even if it's the system default,
3660 * since then people don't have to resort to checking both here and
3661 * /proc/config.gz to confirm 64-bit inums were successfully applied
3662 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3664 * We hide it when inode64 isn't the default and we are using 32-bit
3665 * inodes, since that probably just means the feature isn't even under
3670 * +-----------------+-----------------+
3671 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3672 * +------------------+-----------------+-----------------+
3673 * | full_inums=true | show | show |
3674 * | full_inums=false | show | hide |
3675 * +------------------+-----------------+-----------------+
3678 if (IS_ENABLED(CONFIG_TMPFS_INODE64
) || sbinfo
->full_inums
)
3679 seq_printf(seq
, ",inode%d", (sbinfo
->full_inums
? 64 : 32));
3680 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3681 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3683 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3685 shmem_show_mpol(seq
, sbinfo
->mpol
);
3689 #endif /* CONFIG_TMPFS */
3691 static void shmem_put_super(struct super_block
*sb
)
3693 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3695 free_percpu(sbinfo
->ino_batch
);
3696 percpu_counter_destroy(&sbinfo
->used_blocks
);
3697 mpol_put(sbinfo
->mpol
);
3699 sb
->s_fs_info
= NULL
;
3702 static int shmem_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
3704 struct shmem_options
*ctx
= fc
->fs_private
;
3705 struct inode
*inode
;
3706 struct shmem_sb_info
*sbinfo
;
3709 /* Round up to L1_CACHE_BYTES to resist false sharing */
3710 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3711 L1_CACHE_BYTES
), GFP_KERNEL
);
3715 sb
->s_fs_info
= sbinfo
;
3719 * Per default we only allow half of the physical ram per
3720 * tmpfs instance, limiting inodes to one per page of lowmem;
3721 * but the internal instance is left unlimited.
3723 if (!(sb
->s_flags
& SB_KERNMOUNT
)) {
3724 if (!(ctx
->seen
& SHMEM_SEEN_BLOCKS
))
3725 ctx
->blocks
= shmem_default_max_blocks();
3726 if (!(ctx
->seen
& SHMEM_SEEN_INODES
))
3727 ctx
->inodes
= shmem_default_max_inodes();
3728 if (!(ctx
->seen
& SHMEM_SEEN_INUMS
))
3729 ctx
->full_inums
= IS_ENABLED(CONFIG_TMPFS_INODE64
);
3731 sb
->s_flags
|= SB_NOUSER
;
3733 sb
->s_export_op
= &shmem_export_ops
;
3734 sb
->s_flags
|= SB_NOSEC
;
3736 sb
->s_flags
|= SB_NOUSER
;
3738 sbinfo
->max_blocks
= ctx
->blocks
;
3739 sbinfo
->free_inodes
= sbinfo
->max_inodes
= ctx
->inodes
;
3740 if (sb
->s_flags
& SB_KERNMOUNT
) {
3741 sbinfo
->ino_batch
= alloc_percpu(ino_t
);
3742 if (!sbinfo
->ino_batch
)
3745 sbinfo
->uid
= ctx
->uid
;
3746 sbinfo
->gid
= ctx
->gid
;
3747 sbinfo
->full_inums
= ctx
->full_inums
;
3748 sbinfo
->mode
= ctx
->mode
;
3749 sbinfo
->huge
= ctx
->huge
;
3750 sbinfo
->mpol
= ctx
->mpol
;
3753 spin_lock_init(&sbinfo
->stat_lock
);
3754 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3756 spin_lock_init(&sbinfo
->shrinklist_lock
);
3757 INIT_LIST_HEAD(&sbinfo
->shrinklist
);
3759 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3760 sb
->s_blocksize
= PAGE_SIZE
;
3761 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3762 sb
->s_magic
= TMPFS_MAGIC
;
3763 sb
->s_op
= &shmem_ops
;
3764 sb
->s_time_gran
= 1;
3765 #ifdef CONFIG_TMPFS_XATTR
3766 sb
->s_xattr
= shmem_xattr_handlers
;
3768 #ifdef CONFIG_TMPFS_POSIX_ACL
3769 sb
->s_flags
|= SB_POSIXACL
;
3771 uuid_gen(&sb
->s_uuid
);
3773 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3776 inode
->i_uid
= sbinfo
->uid
;
3777 inode
->i_gid
= sbinfo
->gid
;
3778 sb
->s_root
= d_make_root(inode
);
3784 shmem_put_super(sb
);
3788 static int shmem_get_tree(struct fs_context
*fc
)
3790 return get_tree_nodev(fc
, shmem_fill_super
);
3793 static void shmem_free_fc(struct fs_context
*fc
)
3795 struct shmem_options
*ctx
= fc
->fs_private
;
3798 mpol_put(ctx
->mpol
);
3803 static const struct fs_context_operations shmem_fs_context_ops
= {
3804 .free
= shmem_free_fc
,
3805 .get_tree
= shmem_get_tree
,
3807 .parse_monolithic
= shmem_parse_options
,
3808 .parse_param
= shmem_parse_one
,
3809 .reconfigure
= shmem_reconfigure
,
3813 static struct kmem_cache
*shmem_inode_cachep
;
3815 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3817 struct shmem_inode_info
*info
;
3818 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3821 return &info
->vfs_inode
;
3824 static void shmem_free_in_core_inode(struct inode
*inode
)
3826 if (S_ISLNK(inode
->i_mode
))
3827 kfree(inode
->i_link
);
3828 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3831 static void shmem_destroy_inode(struct inode
*inode
)
3833 if (S_ISREG(inode
->i_mode
))
3834 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3837 static void shmem_init_inode(void *foo
)
3839 struct shmem_inode_info
*info
= foo
;
3840 inode_init_once(&info
->vfs_inode
);
3843 static void shmem_init_inodecache(void)
3845 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3846 sizeof(struct shmem_inode_info
),
3847 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3850 static void shmem_destroy_inodecache(void)
3852 kmem_cache_destroy(shmem_inode_cachep
);
3855 static const struct address_space_operations shmem_aops
= {
3856 .writepage
= shmem_writepage
,
3857 .set_page_dirty
= __set_page_dirty_no_writeback
,
3859 .write_begin
= shmem_write_begin
,
3860 .write_end
= shmem_write_end
,
3862 #ifdef CONFIG_MIGRATION
3863 .migratepage
= migrate_page
,
3865 .error_remove_page
= generic_error_remove_page
,
3868 static const struct file_operations shmem_file_operations
= {
3870 .get_unmapped_area
= shmem_get_unmapped_area
,
3872 .llseek
= shmem_file_llseek
,
3873 .read_iter
= shmem_file_read_iter
,
3874 .write_iter
= generic_file_write_iter
,
3875 .fsync
= noop_fsync
,
3876 .splice_read
= generic_file_splice_read
,
3877 .splice_write
= iter_file_splice_write
,
3878 .fallocate
= shmem_fallocate
,
3882 static const struct inode_operations shmem_inode_operations
= {
3883 .getattr
= shmem_getattr
,
3884 .setattr
= shmem_setattr
,
3885 #ifdef CONFIG_TMPFS_XATTR
3886 .listxattr
= shmem_listxattr
,
3887 .set_acl
= simple_set_acl
,
3891 static const struct inode_operations shmem_dir_inode_operations
= {
3893 .create
= shmem_create
,
3894 .lookup
= simple_lookup
,
3896 .unlink
= shmem_unlink
,
3897 .symlink
= shmem_symlink
,
3898 .mkdir
= shmem_mkdir
,
3899 .rmdir
= shmem_rmdir
,
3900 .mknod
= shmem_mknod
,
3901 .rename
= shmem_rename2
,
3902 .tmpfile
= shmem_tmpfile
,
3904 #ifdef CONFIG_TMPFS_XATTR
3905 .listxattr
= shmem_listxattr
,
3907 #ifdef CONFIG_TMPFS_POSIX_ACL
3908 .setattr
= shmem_setattr
,
3909 .set_acl
= simple_set_acl
,
3913 static const struct inode_operations shmem_special_inode_operations
= {
3914 #ifdef CONFIG_TMPFS_XATTR
3915 .listxattr
= shmem_listxattr
,
3917 #ifdef CONFIG_TMPFS_POSIX_ACL
3918 .setattr
= shmem_setattr
,
3919 .set_acl
= simple_set_acl
,
3923 static const struct super_operations shmem_ops
= {
3924 .alloc_inode
= shmem_alloc_inode
,
3925 .free_inode
= shmem_free_in_core_inode
,
3926 .destroy_inode
= shmem_destroy_inode
,
3928 .statfs
= shmem_statfs
,
3929 .show_options
= shmem_show_options
,
3931 .evict_inode
= shmem_evict_inode
,
3932 .drop_inode
= generic_delete_inode
,
3933 .put_super
= shmem_put_super
,
3934 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3935 .nr_cached_objects
= shmem_unused_huge_count
,
3936 .free_cached_objects
= shmem_unused_huge_scan
,
3940 static const struct vm_operations_struct shmem_vm_ops
= {
3941 .fault
= shmem_fault
,
3942 .map_pages
= filemap_map_pages
,
3944 .set_policy
= shmem_set_policy
,
3945 .get_policy
= shmem_get_policy
,
3949 int shmem_init_fs_context(struct fs_context
*fc
)
3951 struct shmem_options
*ctx
;
3953 ctx
= kzalloc(sizeof(struct shmem_options
), GFP_KERNEL
);
3957 ctx
->mode
= 0777 | S_ISVTX
;
3958 ctx
->uid
= current_fsuid();
3959 ctx
->gid
= current_fsgid();
3961 fc
->fs_private
= ctx
;
3962 fc
->ops
= &shmem_fs_context_ops
;
3966 static struct file_system_type shmem_fs_type
= {
3967 .owner
= THIS_MODULE
,
3969 .init_fs_context
= shmem_init_fs_context
,
3971 .parameters
= shmem_fs_parameters
,
3973 .kill_sb
= kill_litter_super
,
3974 .fs_flags
= FS_USERNS_MOUNT
,
3977 int __init
shmem_init(void)
3981 shmem_init_inodecache();
3983 error
= register_filesystem(&shmem_fs_type
);
3985 pr_err("Could not register tmpfs\n");
3989 shm_mnt
= kern_mount(&shmem_fs_type
);
3990 if (IS_ERR(shm_mnt
)) {
3991 error
= PTR_ERR(shm_mnt
);
3992 pr_err("Could not kern_mount tmpfs\n");
3996 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3997 if (has_transparent_hugepage() && shmem_huge
> SHMEM_HUGE_DENY
)
3998 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4000 shmem_huge
= 0; /* just in case it was patched */
4005 unregister_filesystem(&shmem_fs_type
);
4007 shmem_destroy_inodecache();
4008 shm_mnt
= ERR_PTR(error
);
4012 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4013 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
4014 struct kobj_attribute
*attr
, char *buf
)
4016 static const int values
[] = {
4018 SHMEM_HUGE_WITHIN_SIZE
,
4026 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
4027 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
4029 count
+= sprintf(buf
+ count
, fmt
,
4030 shmem_format_huge(values
[i
]));
4032 buf
[count
- 1] = '\n';
4036 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
4037 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
4042 if (count
+ 1 > sizeof(tmp
))
4044 memcpy(tmp
, buf
, count
);
4046 if (count
&& tmp
[count
- 1] == '\n')
4047 tmp
[count
- 1] = '\0';
4049 huge
= shmem_parse_huge(tmp
);
4050 if (huge
== -EINVAL
)
4052 if (!has_transparent_hugepage() &&
4053 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
4057 if (shmem_huge
> SHMEM_HUGE_DENY
)
4058 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
4062 struct kobj_attribute shmem_enabled_attr
=
4063 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
4064 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4066 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4067 bool shmem_huge_enabled(struct vm_area_struct
*vma
)
4069 struct inode
*inode
= file_inode(vma
->vm_file
);
4070 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
4074 if ((vma
->vm_flags
& VM_NOHUGEPAGE
) ||
4075 test_bit(MMF_DISABLE_THP
, &vma
->vm_mm
->flags
))
4077 if (shmem_huge
== SHMEM_HUGE_FORCE
)
4079 if (shmem_huge
== SHMEM_HUGE_DENY
)
4081 switch (sbinfo
->huge
) {
4082 case SHMEM_HUGE_NEVER
:
4084 case SHMEM_HUGE_ALWAYS
:
4086 case SHMEM_HUGE_WITHIN_SIZE
:
4087 off
= round_up(vma
->vm_pgoff
, HPAGE_PMD_NR
);
4088 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
4089 if (i_size
>= HPAGE_PMD_SIZE
&&
4090 i_size
>> PAGE_SHIFT
>= off
)
4093 case SHMEM_HUGE_ADVISE
:
4094 /* TODO: implement fadvise() hints */
4095 return (vma
->vm_flags
& VM_HUGEPAGE
);
4101 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4103 #else /* !CONFIG_SHMEM */
4106 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4108 * This is intended for small system where the benefits of the full
4109 * shmem code (swap-backed and resource-limited) are outweighed by
4110 * their complexity. On systems without swap this code should be
4111 * effectively equivalent, but much lighter weight.
4114 static struct file_system_type shmem_fs_type
= {
4116 .init_fs_context
= ramfs_init_fs_context
,
4117 .parameters
= ramfs_fs_parameters
,
4118 .kill_sb
= kill_litter_super
,
4119 .fs_flags
= FS_USERNS_MOUNT
,
4122 int __init
shmem_init(void)
4124 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
4126 shm_mnt
= kern_mount(&shmem_fs_type
);
4127 BUG_ON(IS_ERR(shm_mnt
));
4132 int shmem_unuse(unsigned int type
, bool frontswap
,
4133 unsigned long *fs_pages_to_unuse
)
4138 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
4143 void shmem_unlock_mapping(struct address_space
*mapping
)
4148 unsigned long shmem_get_unmapped_area(struct file
*file
,
4149 unsigned long addr
, unsigned long len
,
4150 unsigned long pgoff
, unsigned long flags
)
4152 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
4156 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
4158 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
4160 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
4162 #define shmem_vm_ops generic_file_vm_ops
4163 #define shmem_file_operations ramfs_file_operations
4164 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4165 #define shmem_acct_size(flags, size) 0
4166 #define shmem_unacct_size(flags, size) do {} while (0)
4168 #endif /* CONFIG_SHMEM */
4172 static struct file
*__shmem_file_setup(struct vfsmount
*mnt
, const char *name
, loff_t size
,
4173 unsigned long flags
, unsigned int i_flags
)
4175 struct inode
*inode
;
4179 return ERR_CAST(mnt
);
4181 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
4182 return ERR_PTR(-EINVAL
);
4184 if (shmem_acct_size(flags
, size
))
4185 return ERR_PTR(-ENOMEM
);
4187 inode
= shmem_get_inode(mnt
->mnt_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0,
4189 if (unlikely(!inode
)) {
4190 shmem_unacct_size(flags
, size
);
4191 return ERR_PTR(-ENOSPC
);
4193 inode
->i_flags
|= i_flags
;
4194 inode
->i_size
= size
;
4195 clear_nlink(inode
); /* It is unlinked */
4196 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
4198 res
= alloc_file_pseudo(inode
, mnt
, name
, O_RDWR
,
4199 &shmem_file_operations
);
4206 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4207 * kernel internal. There will be NO LSM permission checks against the
4208 * underlying inode. So users of this interface must do LSM checks at a
4209 * higher layer. The users are the big_key and shm implementations. LSM
4210 * checks are provided at the key or shm level rather than the inode.
4211 * @name: name for dentry (to be seen in /proc/<pid>/maps
4212 * @size: size to be set for the file
4213 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4215 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4217 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, S_PRIVATE
);
4221 * shmem_file_setup - get an unlinked file living in tmpfs
4222 * @name: name for dentry (to be seen in /proc/<pid>/maps
4223 * @size: size to be set for the file
4224 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4226 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
4228 return __shmem_file_setup(shm_mnt
, name
, size
, flags
, 0);
4230 EXPORT_SYMBOL_GPL(shmem_file_setup
);
4233 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4234 * @mnt: the tmpfs mount where the file will be created
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_with_mnt(struct vfsmount
*mnt
, const char *name
,
4240 loff_t size
, unsigned long flags
)
4242 return __shmem_file_setup(mnt
, name
, size
, flags
, 0);
4244 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt
);
4247 * shmem_zero_setup - setup a shared anonymous mapping
4248 * @vma: the vma to be mmapped is prepared by do_mmap
4250 int shmem_zero_setup(struct vm_area_struct
*vma
)
4253 loff_t size
= vma
->vm_end
- vma
->vm_start
;
4256 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4257 * between XFS directory reading and selinux: since this file is only
4258 * accessible to the user through its mapping, use S_PRIVATE flag to
4259 * bypass file security, in the same way as shmem_kernel_file_setup().
4261 file
= shmem_kernel_file_setup("dev/zero", size
, vma
->vm_flags
);
4263 return PTR_ERR(file
);
4267 vma
->vm_file
= file
;
4268 vma
->vm_ops
= &shmem_vm_ops
;
4270 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
) &&
4271 ((vma
->vm_start
+ ~HPAGE_PMD_MASK
) & HPAGE_PMD_MASK
) <
4272 (vma
->vm_end
& HPAGE_PMD_MASK
)) {
4273 khugepaged_enter(vma
, vma
->vm_flags
);
4280 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4281 * @mapping: the page's address_space
4282 * @index: the page index
4283 * @gfp: the page allocator flags to use if allocating
4285 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4286 * with any new page allocations done using the specified allocation flags.
4287 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4288 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4289 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4291 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4292 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4294 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4295 pgoff_t index
, gfp_t gfp
)
4298 struct inode
*inode
= mapping
->host
;
4302 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4303 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4304 gfp
, NULL
, NULL
, NULL
);
4306 page
= ERR_PTR(error
);
4312 * The tiny !SHMEM case uses ramfs without swap
4314 return read_cache_page_gfp(mapping
, index
, gfp
);
4317 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);