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/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
36 static struct vfsmount
*shm_mnt
;
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/posix_acl_xattr.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
69 #include <linux/syscalls.h>
70 #include <linux/fcntl.h>
71 #include <uapi/linux/memfd.h>
73 #include <asm/uaccess.h>
74 #include <asm/pgtable.h>
78 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
79 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
81 /* Pretend that each entry is of this size in directory's i_size */
82 #define BOGO_DIRENT_SIZE 20
84 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
85 #define SHORT_SYMLINK_LEN 128
88 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
89 * inode->i_private (with i_mutex making sure that it has only one user at
90 * a time): we would prefer not to enlarge the shmem inode just for that.
93 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
94 pgoff_t start
; /* start of range currently being fallocated */
95 pgoff_t next
; /* the next page offset to be fallocated */
96 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
97 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
100 /* Flag allocation requirements to shmem_getpage */
102 SGP_READ
, /* don't exceed i_size, don't allocate page */
103 SGP_CACHE
, /* don't exceed i_size, may allocate page */
104 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
105 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
109 static unsigned long shmem_default_max_blocks(void)
111 return totalram_pages
/ 2;
114 static unsigned long shmem_default_max_inodes(void)
116 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
120 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
121 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
122 struct shmem_inode_info
*info
, pgoff_t index
);
123 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
124 struct page
**pagep
, enum sgp_type sgp
,
125 gfp_t gfp
, struct mm_struct
*fault_mm
, int *fault_type
);
127 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
128 struct page
**pagep
, enum sgp_type sgp
)
130 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
131 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
);
134 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
136 return sb
->s_fs_info
;
140 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
141 * for shared memory and for shared anonymous (/dev/zero) mappings
142 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
143 * consistent with the pre-accounting of private mappings ...
145 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
147 return (flags
& VM_NORESERVE
) ?
148 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
151 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
153 if (!(flags
& VM_NORESERVE
))
154 vm_unacct_memory(VM_ACCT(size
));
157 static inline int shmem_reacct_size(unsigned long flags
,
158 loff_t oldsize
, loff_t newsize
)
160 if (!(flags
& VM_NORESERVE
)) {
161 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
162 return security_vm_enough_memory_mm(current
->mm
,
163 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
164 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
165 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
171 * ... whereas tmpfs objects are accounted incrementally as
172 * pages are allocated, in order to allow large sparse files.
173 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
174 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
176 static inline int shmem_acct_block(unsigned long flags
)
178 return (flags
& VM_NORESERVE
) ?
179 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_SIZE
)) : 0;
182 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
184 if (flags
& VM_NORESERVE
)
185 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
188 static const struct super_operations shmem_ops
;
189 static const struct address_space_operations shmem_aops
;
190 static const struct file_operations shmem_file_operations
;
191 static const struct inode_operations shmem_inode_operations
;
192 static const struct inode_operations shmem_dir_inode_operations
;
193 static const struct inode_operations shmem_special_inode_operations
;
194 static const struct vm_operations_struct shmem_vm_ops
;
196 static LIST_HEAD(shmem_swaplist
);
197 static DEFINE_MUTEX(shmem_swaplist_mutex
);
199 static int shmem_reserve_inode(struct super_block
*sb
)
201 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
202 if (sbinfo
->max_inodes
) {
203 spin_lock(&sbinfo
->stat_lock
);
204 if (!sbinfo
->free_inodes
) {
205 spin_unlock(&sbinfo
->stat_lock
);
208 sbinfo
->free_inodes
--;
209 spin_unlock(&sbinfo
->stat_lock
);
214 static void shmem_free_inode(struct super_block
*sb
)
216 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
217 if (sbinfo
->max_inodes
) {
218 spin_lock(&sbinfo
->stat_lock
);
219 sbinfo
->free_inodes
++;
220 spin_unlock(&sbinfo
->stat_lock
);
225 * shmem_recalc_inode - recalculate the block usage of an inode
226 * @inode: inode to recalc
228 * We have to calculate the free blocks since the mm can drop
229 * undirtied hole pages behind our back.
231 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
232 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
234 * It has to be called with the spinlock held.
236 static void shmem_recalc_inode(struct inode
*inode
)
238 struct shmem_inode_info
*info
= SHMEM_I(inode
);
241 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
243 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
244 if (sbinfo
->max_blocks
)
245 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
246 info
->alloced
-= freed
;
247 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
248 shmem_unacct_blocks(info
->flags
, freed
);
253 * Replace item expected in radix tree by a new item, while holding tree lock.
255 static int shmem_radix_tree_replace(struct address_space
*mapping
,
256 pgoff_t index
, void *expected
, void *replacement
)
261 VM_BUG_ON(!expected
);
262 VM_BUG_ON(!replacement
);
263 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
266 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
267 if (item
!= expected
)
269 radix_tree_replace_slot(pslot
, replacement
);
274 * Sometimes, before we decide whether to proceed or to fail, we must check
275 * that an entry was not already brought back from swap by a racing thread.
277 * Checking page is not enough: by the time a SwapCache page is locked, it
278 * might be reused, and again be SwapCache, using the same swap as before.
280 static bool shmem_confirm_swap(struct address_space
*mapping
,
281 pgoff_t index
, swp_entry_t swap
)
286 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
288 return item
== swp_to_radix_entry(swap
);
292 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
295 * disables huge pages for the mount;
297 * enables huge pages for the mount;
298 * SHMEM_HUGE_WITHIN_SIZE:
299 * only allocate huge pages if the page will be fully within i_size,
300 * also respect fadvise()/madvise() hints;
302 * only allocate huge pages if requested with fadvise()/madvise();
305 #define SHMEM_HUGE_NEVER 0
306 #define SHMEM_HUGE_ALWAYS 1
307 #define SHMEM_HUGE_WITHIN_SIZE 2
308 #define SHMEM_HUGE_ADVISE 3
312 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
315 * disables huge on shm_mnt and all mounts, for emergency use;
317 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
320 #define SHMEM_HUGE_DENY (-1)
321 #define SHMEM_HUGE_FORCE (-2)
323 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
324 /* ifdef here to avoid bloating shmem.o when not necessary */
326 int shmem_huge __read_mostly
;
328 static int shmem_parse_huge(const char *str
)
330 if (!strcmp(str
, "never"))
331 return SHMEM_HUGE_NEVER
;
332 if (!strcmp(str
, "always"))
333 return SHMEM_HUGE_ALWAYS
;
334 if (!strcmp(str
, "within_size"))
335 return SHMEM_HUGE_WITHIN_SIZE
;
336 if (!strcmp(str
, "advise"))
337 return SHMEM_HUGE_ADVISE
;
338 if (!strcmp(str
, "deny"))
339 return SHMEM_HUGE_DENY
;
340 if (!strcmp(str
, "force"))
341 return SHMEM_HUGE_FORCE
;
345 static const char *shmem_format_huge(int huge
)
348 case SHMEM_HUGE_NEVER
:
350 case SHMEM_HUGE_ALWAYS
:
352 case SHMEM_HUGE_WITHIN_SIZE
:
353 return "within_size";
354 case SHMEM_HUGE_ADVISE
:
356 case SHMEM_HUGE_DENY
:
358 case SHMEM_HUGE_FORCE
:
366 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
368 #define shmem_huge SHMEM_HUGE_DENY
370 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
373 * Like add_to_page_cache_locked, but error if expected item has gone.
375 static int shmem_add_to_page_cache(struct page
*page
,
376 struct address_space
*mapping
,
377 pgoff_t index
, void *expected
)
381 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
382 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
385 page
->mapping
= mapping
;
388 spin_lock_irq(&mapping
->tree_lock
);
390 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
392 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
396 __inc_zone_page_state(page
, NR_FILE_PAGES
);
397 __inc_zone_page_state(page
, NR_SHMEM
);
398 spin_unlock_irq(&mapping
->tree_lock
);
400 page
->mapping
= NULL
;
401 spin_unlock_irq(&mapping
->tree_lock
);
408 * Like delete_from_page_cache, but substitutes swap for page.
410 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
412 struct address_space
*mapping
= page
->mapping
;
415 spin_lock_irq(&mapping
->tree_lock
);
416 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
417 page
->mapping
= NULL
;
419 __dec_zone_page_state(page
, NR_FILE_PAGES
);
420 __dec_zone_page_state(page
, NR_SHMEM
);
421 spin_unlock_irq(&mapping
->tree_lock
);
427 * Remove swap entry from radix tree, free the swap and its page cache.
429 static int shmem_free_swap(struct address_space
*mapping
,
430 pgoff_t index
, void *radswap
)
434 spin_lock_irq(&mapping
->tree_lock
);
435 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
436 spin_unlock_irq(&mapping
->tree_lock
);
439 free_swap_and_cache(radix_to_swp_entry(radswap
));
444 * Determine (in bytes) how many of the shmem object's pages mapped by the
445 * given offsets are swapped out.
447 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
448 * as long as the inode doesn't go away and racy results are not a problem.
450 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
451 pgoff_t start
, pgoff_t end
)
453 struct radix_tree_iter iter
;
456 unsigned long swapped
= 0;
460 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
461 if (iter
.index
>= end
)
464 page
= radix_tree_deref_slot(slot
);
466 if (radix_tree_deref_retry(page
)) {
467 slot
= radix_tree_iter_retry(&iter
);
471 if (radix_tree_exceptional_entry(page
))
474 if (need_resched()) {
476 slot
= radix_tree_iter_next(&iter
);
482 return swapped
<< PAGE_SHIFT
;
486 * Determine (in bytes) how many of the shmem object's pages mapped by the
487 * given vma is swapped out.
489 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
490 * as long as the inode doesn't go away and racy results are not a problem.
492 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
494 struct inode
*inode
= file_inode(vma
->vm_file
);
495 struct shmem_inode_info
*info
= SHMEM_I(inode
);
496 struct address_space
*mapping
= inode
->i_mapping
;
497 unsigned long swapped
;
499 /* Be careful as we don't hold info->lock */
500 swapped
= READ_ONCE(info
->swapped
);
503 * The easier cases are when the shmem object has nothing in swap, or
504 * the vma maps it whole. Then we can simply use the stats that we
510 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
511 return swapped
<< PAGE_SHIFT
;
513 /* Here comes the more involved part */
514 return shmem_partial_swap_usage(mapping
,
515 linear_page_index(vma
, vma
->vm_start
),
516 linear_page_index(vma
, vma
->vm_end
));
520 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
522 void shmem_unlock_mapping(struct address_space
*mapping
)
525 pgoff_t indices
[PAGEVEC_SIZE
];
528 pagevec_init(&pvec
, 0);
530 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
532 while (!mapping_unevictable(mapping
)) {
534 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
535 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
537 pvec
.nr
= find_get_entries(mapping
, index
,
538 PAGEVEC_SIZE
, pvec
.pages
, indices
);
541 index
= indices
[pvec
.nr
- 1] + 1;
542 pagevec_remove_exceptionals(&pvec
);
543 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
544 pagevec_release(&pvec
);
550 * Remove range of pages and swap entries from radix tree, and free them.
551 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
553 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
556 struct address_space
*mapping
= inode
->i_mapping
;
557 struct shmem_inode_info
*info
= SHMEM_I(inode
);
558 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
559 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
560 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
561 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
563 pgoff_t indices
[PAGEVEC_SIZE
];
564 long nr_swaps_freed
= 0;
569 end
= -1; /* unsigned, so actually very big */
571 pagevec_init(&pvec
, 0);
573 while (index
< end
) {
574 pvec
.nr
= find_get_entries(mapping
, index
,
575 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
576 pvec
.pages
, indices
);
579 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
580 struct page
*page
= pvec
.pages
[i
];
586 if (radix_tree_exceptional_entry(page
)) {
589 nr_swaps_freed
+= !shmem_free_swap(mapping
,
594 if (!trylock_page(page
))
596 if (!unfalloc
|| !PageUptodate(page
)) {
597 if (page
->mapping
== mapping
) {
598 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
599 truncate_inode_page(mapping
, page
);
604 pagevec_remove_exceptionals(&pvec
);
605 pagevec_release(&pvec
);
611 struct page
*page
= NULL
;
612 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
614 unsigned int top
= PAGE_SIZE
;
619 zero_user_segment(page
, partial_start
, top
);
620 set_page_dirty(page
);
626 struct page
*page
= NULL
;
627 shmem_getpage(inode
, end
, &page
, SGP_READ
);
629 zero_user_segment(page
, 0, partial_end
);
630 set_page_dirty(page
);
639 while (index
< end
) {
642 pvec
.nr
= find_get_entries(mapping
, index
,
643 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
644 pvec
.pages
, indices
);
646 /* If all gone or hole-punch or unfalloc, we're done */
647 if (index
== start
|| end
!= -1)
649 /* But if truncating, restart to make sure all gone */
653 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
654 struct page
*page
= pvec
.pages
[i
];
660 if (radix_tree_exceptional_entry(page
)) {
663 if (shmem_free_swap(mapping
, index
, page
)) {
664 /* Swap was replaced by page: retry */
673 if (!unfalloc
|| !PageUptodate(page
)) {
674 if (page
->mapping
== mapping
) {
675 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
676 truncate_inode_page(mapping
, page
);
678 /* Page was replaced by swap: retry */
686 pagevec_remove_exceptionals(&pvec
);
687 pagevec_release(&pvec
);
691 spin_lock(&info
->lock
);
692 info
->swapped
-= nr_swaps_freed
;
693 shmem_recalc_inode(inode
);
694 spin_unlock(&info
->lock
);
697 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
699 shmem_undo_range(inode
, lstart
, lend
, false);
700 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
702 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
704 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
707 struct inode
*inode
= dentry
->d_inode
;
708 struct shmem_inode_info
*info
= SHMEM_I(inode
);
710 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
711 spin_lock(&info
->lock
);
712 shmem_recalc_inode(inode
);
713 spin_unlock(&info
->lock
);
715 generic_fillattr(inode
, stat
);
719 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
721 struct inode
*inode
= d_inode(dentry
);
722 struct shmem_inode_info
*info
= SHMEM_I(inode
);
725 error
= inode_change_ok(inode
, attr
);
729 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
730 loff_t oldsize
= inode
->i_size
;
731 loff_t newsize
= attr
->ia_size
;
733 /* protected by i_mutex */
734 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
735 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
738 if (newsize
!= oldsize
) {
739 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
743 i_size_write(inode
, newsize
);
744 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
746 if (newsize
<= oldsize
) {
747 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
748 if (oldsize
> holebegin
)
749 unmap_mapping_range(inode
->i_mapping
,
752 shmem_truncate_range(inode
,
753 newsize
, (loff_t
)-1);
754 /* unmap again to remove racily COWed private pages */
755 if (oldsize
> holebegin
)
756 unmap_mapping_range(inode
->i_mapping
,
761 setattr_copy(inode
, attr
);
762 if (attr
->ia_valid
& ATTR_MODE
)
763 error
= posix_acl_chmod(inode
, inode
->i_mode
);
767 static void shmem_evict_inode(struct inode
*inode
)
769 struct shmem_inode_info
*info
= SHMEM_I(inode
);
771 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
772 shmem_unacct_size(info
->flags
, inode
->i_size
);
774 shmem_truncate_range(inode
, 0, (loff_t
)-1);
775 if (!list_empty(&info
->swaplist
)) {
776 mutex_lock(&shmem_swaplist_mutex
);
777 list_del_init(&info
->swaplist
);
778 mutex_unlock(&shmem_swaplist_mutex
);
782 simple_xattrs_free(&info
->xattrs
);
783 WARN_ON(inode
->i_blocks
);
784 shmem_free_inode(inode
->i_sb
);
789 * If swap found in inode, free it and move page from swapcache to filecache.
791 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
792 swp_entry_t swap
, struct page
**pagep
)
794 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
800 radswap
= swp_to_radix_entry(swap
);
801 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
803 return -EAGAIN
; /* tell shmem_unuse we found nothing */
806 * Move _head_ to start search for next from here.
807 * But be careful: shmem_evict_inode checks list_empty without taking
808 * mutex, and there's an instant in list_move_tail when info->swaplist
809 * would appear empty, if it were the only one on shmem_swaplist.
811 if (shmem_swaplist
.next
!= &info
->swaplist
)
812 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
814 gfp
= mapping_gfp_mask(mapping
);
815 if (shmem_should_replace_page(*pagep
, gfp
)) {
816 mutex_unlock(&shmem_swaplist_mutex
);
817 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
818 mutex_lock(&shmem_swaplist_mutex
);
820 * We needed to drop mutex to make that restrictive page
821 * allocation, but the inode might have been freed while we
822 * dropped it: although a racing shmem_evict_inode() cannot
823 * complete without emptying the radix_tree, our page lock
824 * on this swapcache page is not enough to prevent that -
825 * free_swap_and_cache() of our swap entry will only
826 * trylock_page(), removing swap from radix_tree whatever.
828 * We must not proceed to shmem_add_to_page_cache() if the
829 * inode has been freed, but of course we cannot rely on
830 * inode or mapping or info to check that. However, we can
831 * safely check if our swap entry is still in use (and here
832 * it can't have got reused for another page): if it's still
833 * in use, then the inode cannot have been freed yet, and we
834 * can safely proceed (if it's no longer in use, that tells
835 * nothing about the inode, but we don't need to unuse swap).
837 if (!page_swapcount(*pagep
))
842 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
843 * but also to hold up shmem_evict_inode(): so inode cannot be freed
844 * beneath us (pagelock doesn't help until the page is in pagecache).
847 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
849 if (error
!= -ENOMEM
) {
851 * Truncation and eviction use free_swap_and_cache(), which
852 * only does trylock page: if we raced, best clean up here.
854 delete_from_swap_cache(*pagep
);
855 set_page_dirty(*pagep
);
857 spin_lock(&info
->lock
);
859 spin_unlock(&info
->lock
);
867 * Search through swapped inodes to find and replace swap by page.
869 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
871 struct list_head
*this, *next
;
872 struct shmem_inode_info
*info
;
873 struct mem_cgroup
*memcg
;
877 * There's a faint possibility that swap page was replaced before
878 * caller locked it: caller will come back later with the right page.
880 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
884 * Charge page using GFP_KERNEL while we can wait, before taking
885 * the shmem_swaplist_mutex which might hold up shmem_writepage().
886 * Charged back to the user (not to caller) when swap account is used.
888 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
892 /* No radix_tree_preload: swap entry keeps a place for page in tree */
895 mutex_lock(&shmem_swaplist_mutex
);
896 list_for_each_safe(this, next
, &shmem_swaplist
) {
897 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
899 error
= shmem_unuse_inode(info
, swap
, &page
);
901 list_del_init(&info
->swaplist
);
903 if (error
!= -EAGAIN
)
905 /* found nothing in this: move on to search the next */
907 mutex_unlock(&shmem_swaplist_mutex
);
910 if (error
!= -ENOMEM
)
912 mem_cgroup_cancel_charge(page
, memcg
, false);
914 mem_cgroup_commit_charge(page
, memcg
, true, false);
922 * Move the page from the page cache to the swap cache.
924 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
926 struct shmem_inode_info
*info
;
927 struct address_space
*mapping
;
932 BUG_ON(!PageLocked(page
));
933 mapping
= page
->mapping
;
935 inode
= mapping
->host
;
936 info
= SHMEM_I(inode
);
937 if (info
->flags
& VM_LOCKED
)
939 if (!total_swap_pages
)
943 * Our capabilities prevent regular writeback or sync from ever calling
944 * shmem_writepage; but a stacking filesystem might use ->writepage of
945 * its underlying filesystem, in which case tmpfs should write out to
946 * swap only in response to memory pressure, and not for the writeback
949 if (!wbc
->for_reclaim
) {
950 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
955 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
956 * value into swapfile.c, the only way we can correctly account for a
957 * fallocated page arriving here is now to initialize it and write it.
959 * That's okay for a page already fallocated earlier, but if we have
960 * not yet completed the fallocation, then (a) we want to keep track
961 * of this page in case we have to undo it, and (b) it may not be a
962 * good idea to continue anyway, once we're pushing into swap. So
963 * reactivate the page, and let shmem_fallocate() quit when too many.
965 if (!PageUptodate(page
)) {
966 if (inode
->i_private
) {
967 struct shmem_falloc
*shmem_falloc
;
968 spin_lock(&inode
->i_lock
);
969 shmem_falloc
= inode
->i_private
;
971 !shmem_falloc
->waitq
&&
972 index
>= shmem_falloc
->start
&&
973 index
< shmem_falloc
->next
)
974 shmem_falloc
->nr_unswapped
++;
977 spin_unlock(&inode
->i_lock
);
981 clear_highpage(page
);
982 flush_dcache_page(page
);
983 SetPageUptodate(page
);
986 swap
= get_swap_page();
990 if (mem_cgroup_try_charge_swap(page
, swap
))
994 * Add inode to shmem_unuse()'s list of swapped-out inodes,
995 * if it's not already there. Do it now before the page is
996 * moved to swap cache, when its pagelock no longer protects
997 * the inode from eviction. But don't unlock the mutex until
998 * we've incremented swapped, because shmem_unuse_inode() will
999 * prune a !swapped inode from the swaplist under this mutex.
1001 mutex_lock(&shmem_swaplist_mutex
);
1002 if (list_empty(&info
->swaplist
))
1003 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1005 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1006 spin_lock(&info
->lock
);
1007 shmem_recalc_inode(inode
);
1009 spin_unlock(&info
->lock
);
1011 swap_shmem_alloc(swap
);
1012 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1014 mutex_unlock(&shmem_swaplist_mutex
);
1015 BUG_ON(page_mapped(page
));
1016 swap_writepage(page
, wbc
);
1020 mutex_unlock(&shmem_swaplist_mutex
);
1022 swapcache_free(swap
);
1024 set_page_dirty(page
);
1025 if (wbc
->for_reclaim
)
1026 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1031 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1032 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1036 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1037 return; /* show nothing */
1039 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1041 seq_printf(seq
, ",mpol=%s", buffer
);
1044 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1046 struct mempolicy
*mpol
= NULL
;
1048 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1049 mpol
= sbinfo
->mpol
;
1051 spin_unlock(&sbinfo
->stat_lock
);
1055 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1056 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1059 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1063 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1065 #define vm_policy vm_private_data
1068 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1069 struct shmem_inode_info
*info
, pgoff_t index
)
1071 struct vm_area_struct pvma
;
1074 /* Create a pseudo vma that just contains the policy */
1076 /* Bias interleave by inode number to distribute better across nodes */
1077 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1079 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1081 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1083 /* Drop reference taken by mpol_shared_policy_lookup() */
1084 mpol_cond_put(pvma
.vm_policy
);
1089 static struct page
*shmem_alloc_page(gfp_t gfp
,
1090 struct shmem_inode_info
*info
, pgoff_t index
)
1092 struct vm_area_struct pvma
;
1095 /* Create a pseudo vma that just contains the policy */
1097 /* Bias interleave by inode number to distribute better across nodes */
1098 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1100 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1102 page
= alloc_pages_vma(gfp
, 0, &pvma
, 0, numa_node_id(), false);
1104 __SetPageLocked(page
);
1105 __SetPageSwapBacked(page
);
1108 /* Drop reference taken by mpol_shared_policy_lookup() */
1109 mpol_cond_put(pvma
.vm_policy
);
1115 * When a page is moved from swapcache to shmem filecache (either by the
1116 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1117 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1118 * ignorance of the mapping it belongs to. If that mapping has special
1119 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1120 * we may need to copy to a suitable page before moving to filecache.
1122 * In a future release, this may well be extended to respect cpuset and
1123 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1124 * but for now it is a simple matter of zone.
1126 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1128 return page_zonenum(page
) > gfp_zone(gfp
);
1131 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1132 struct shmem_inode_info
*info
, pgoff_t index
)
1134 struct page
*oldpage
, *newpage
;
1135 struct address_space
*swap_mapping
;
1140 swap_index
= page_private(oldpage
);
1141 swap_mapping
= page_mapping(oldpage
);
1144 * We have arrived here because our zones are constrained, so don't
1145 * limit chance of success by further cpuset and node constraints.
1147 gfp
&= ~GFP_CONSTRAINT_MASK
;
1148 newpage
= shmem_alloc_page(gfp
, info
, index
);
1153 copy_highpage(newpage
, oldpage
);
1154 flush_dcache_page(newpage
);
1156 SetPageUptodate(newpage
);
1157 set_page_private(newpage
, swap_index
);
1158 SetPageSwapCache(newpage
);
1161 * Our caller will very soon move newpage out of swapcache, but it's
1162 * a nice clean interface for us to replace oldpage by newpage there.
1164 spin_lock_irq(&swap_mapping
->tree_lock
);
1165 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1168 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1169 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1171 spin_unlock_irq(&swap_mapping
->tree_lock
);
1173 if (unlikely(error
)) {
1175 * Is this possible? I think not, now that our callers check
1176 * both PageSwapCache and page_private after getting page lock;
1177 * but be defensive. Reverse old to newpage for clear and free.
1181 mem_cgroup_migrate(oldpage
, newpage
);
1182 lru_cache_add_anon(newpage
);
1186 ClearPageSwapCache(oldpage
);
1187 set_page_private(oldpage
, 0);
1189 unlock_page(oldpage
);
1196 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1198 * If we allocate a new one we do not mark it dirty. That's up to the
1199 * vm. If we swap it in we mark it dirty since we also free the swap
1200 * entry since a page cannot live in both the swap and page cache.
1202 * fault_mm and fault_type are only supplied by shmem_fault:
1203 * otherwise they are NULL.
1205 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1206 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1207 struct mm_struct
*fault_mm
, int *fault_type
)
1209 struct address_space
*mapping
= inode
->i_mapping
;
1210 struct shmem_inode_info
*info
;
1211 struct shmem_sb_info
*sbinfo
;
1212 struct mm_struct
*charge_mm
;
1213 struct mem_cgroup
*memcg
;
1220 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1224 page
= find_lock_entry(mapping
, index
);
1225 if (radix_tree_exceptional_entry(page
)) {
1226 swap
= radix_to_swp_entry(page
);
1230 if (sgp
<= SGP_CACHE
&&
1231 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1236 if (page
&& sgp
== SGP_WRITE
)
1237 mark_page_accessed(page
);
1239 /* fallocated page? */
1240 if (page
&& !PageUptodate(page
)) {
1241 if (sgp
!= SGP_READ
)
1247 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1253 * Fast cache lookup did not find it:
1254 * bring it back from swap or allocate.
1256 info
= SHMEM_I(inode
);
1257 sbinfo
= SHMEM_SB(inode
->i_sb
);
1258 charge_mm
= fault_mm
? : current
->mm
;
1261 /* Look it up and read it in.. */
1262 page
= lookup_swap_cache(swap
);
1264 /* Or update major stats only when swapin succeeds?? */
1266 *fault_type
|= VM_FAULT_MAJOR
;
1267 count_vm_event(PGMAJFAULT
);
1268 mem_cgroup_count_vm_event(fault_mm
, PGMAJFAULT
);
1270 /* Here we actually start the io */
1271 page
= shmem_swapin(swap
, gfp
, info
, index
);
1278 /* We have to do this with page locked to prevent races */
1280 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1281 !shmem_confirm_swap(mapping
, index
, swap
)) {
1282 error
= -EEXIST
; /* try again */
1285 if (!PageUptodate(page
)) {
1289 wait_on_page_writeback(page
);
1291 if (shmem_should_replace_page(page
, gfp
)) {
1292 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1297 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1300 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1301 swp_to_radix_entry(swap
));
1303 * We already confirmed swap under page lock, and make
1304 * no memory allocation here, so usually no possibility
1305 * of error; but free_swap_and_cache() only trylocks a
1306 * page, so it is just possible that the entry has been
1307 * truncated or holepunched since swap was confirmed.
1308 * shmem_undo_range() will have done some of the
1309 * unaccounting, now delete_from_swap_cache() will do
1311 * Reset swap.val? No, leave it so "failed" goes back to
1312 * "repeat": reading a hole and writing should succeed.
1315 mem_cgroup_cancel_charge(page
, memcg
, false);
1316 delete_from_swap_cache(page
);
1322 mem_cgroup_commit_charge(page
, memcg
, true, false);
1324 spin_lock(&info
->lock
);
1326 shmem_recalc_inode(inode
);
1327 spin_unlock(&info
->lock
);
1329 if (sgp
== SGP_WRITE
)
1330 mark_page_accessed(page
);
1332 delete_from_swap_cache(page
);
1333 set_page_dirty(page
);
1337 if (shmem_acct_block(info
->flags
)) {
1341 if (sbinfo
->max_blocks
) {
1342 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1343 sbinfo
->max_blocks
) >= 0) {
1347 percpu_counter_inc(&sbinfo
->used_blocks
);
1350 page
= shmem_alloc_page(gfp
, info
, index
);
1355 if (sgp
== SGP_WRITE
)
1356 __SetPageReferenced(page
);
1358 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1362 error
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
1364 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1366 radix_tree_preload_end();
1369 mem_cgroup_cancel_charge(page
, memcg
, false);
1372 mem_cgroup_commit_charge(page
, memcg
, false, false);
1373 lru_cache_add_anon(page
);
1375 spin_lock(&info
->lock
);
1377 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1378 shmem_recalc_inode(inode
);
1379 spin_unlock(&info
->lock
);
1383 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1385 if (sgp
== SGP_FALLOC
)
1389 * Let SGP_WRITE caller clear ends if write does not fill page;
1390 * but SGP_FALLOC on a page fallocated earlier must initialize
1391 * it now, lest undo on failure cancel our earlier guarantee.
1393 if (sgp
!= SGP_WRITE
) {
1394 clear_highpage(page
);
1395 flush_dcache_page(page
);
1396 SetPageUptodate(page
);
1400 /* Perhaps the file has been truncated since we checked */
1401 if (sgp
<= SGP_CACHE
&&
1402 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1404 ClearPageDirty(page
);
1405 delete_from_page_cache(page
);
1406 spin_lock(&info
->lock
);
1407 shmem_recalc_inode(inode
);
1408 spin_unlock(&info
->lock
);
1420 if (sbinfo
->max_blocks
)
1421 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1423 shmem_unacct_blocks(info
->flags
, 1);
1425 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1432 if (error
== -ENOSPC
&& !once
++) {
1433 info
= SHMEM_I(inode
);
1434 spin_lock(&info
->lock
);
1435 shmem_recalc_inode(inode
);
1436 spin_unlock(&info
->lock
);
1439 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1444 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1446 struct inode
*inode
= file_inode(vma
->vm_file
);
1447 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1449 int ret
= VM_FAULT_LOCKED
;
1452 * Trinity finds that probing a hole which tmpfs is punching can
1453 * prevent the hole-punch from ever completing: which in turn
1454 * locks writers out with its hold on i_mutex. So refrain from
1455 * faulting pages into the hole while it's being punched. Although
1456 * shmem_undo_range() does remove the additions, it may be unable to
1457 * keep up, as each new page needs its own unmap_mapping_range() call,
1458 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1460 * It does not matter if we sometimes reach this check just before the
1461 * hole-punch begins, so that one fault then races with the punch:
1462 * we just need to make racing faults a rare case.
1464 * The implementation below would be much simpler if we just used a
1465 * standard mutex or completion: but we cannot take i_mutex in fault,
1466 * and bloating every shmem inode for this unlikely case would be sad.
1468 if (unlikely(inode
->i_private
)) {
1469 struct shmem_falloc
*shmem_falloc
;
1471 spin_lock(&inode
->i_lock
);
1472 shmem_falloc
= inode
->i_private
;
1474 shmem_falloc
->waitq
&&
1475 vmf
->pgoff
>= shmem_falloc
->start
&&
1476 vmf
->pgoff
< shmem_falloc
->next
) {
1477 wait_queue_head_t
*shmem_falloc_waitq
;
1478 DEFINE_WAIT(shmem_fault_wait
);
1480 ret
= VM_FAULT_NOPAGE
;
1481 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1482 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1483 /* It's polite to up mmap_sem if we can */
1484 up_read(&vma
->vm_mm
->mmap_sem
);
1485 ret
= VM_FAULT_RETRY
;
1488 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1489 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1490 TASK_UNINTERRUPTIBLE
);
1491 spin_unlock(&inode
->i_lock
);
1495 * shmem_falloc_waitq points into the shmem_fallocate()
1496 * stack of the hole-punching task: shmem_falloc_waitq
1497 * is usually invalid by the time we reach here, but
1498 * finish_wait() does not dereference it in that case;
1499 * though i_lock needed lest racing with wake_up_all().
1501 spin_lock(&inode
->i_lock
);
1502 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1503 spin_unlock(&inode
->i_lock
);
1506 spin_unlock(&inode
->i_lock
);
1509 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
,
1510 gfp
, vma
->vm_mm
, &ret
);
1512 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1516 unsigned long shmem_get_unmapped_area(struct file
*file
,
1517 unsigned long uaddr
, unsigned long len
,
1518 unsigned long pgoff
, unsigned long flags
)
1520 unsigned long (*get_area
)(struct file
*,
1521 unsigned long, unsigned long, unsigned long, unsigned long);
1523 unsigned long offset
;
1524 unsigned long inflated_len
;
1525 unsigned long inflated_addr
;
1526 unsigned long inflated_offset
;
1528 if (len
> TASK_SIZE
)
1531 get_area
= current
->mm
->get_unmapped_area
;
1532 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
1534 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1536 if (IS_ERR_VALUE(addr
))
1538 if (addr
& ~PAGE_MASK
)
1540 if (addr
> TASK_SIZE
- len
)
1543 if (shmem_huge
== SHMEM_HUGE_DENY
)
1545 if (len
< HPAGE_PMD_SIZE
)
1547 if (flags
& MAP_FIXED
)
1550 * Our priority is to support MAP_SHARED mapped hugely;
1551 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1552 * But if caller specified an address hint, respect that as before.
1557 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
1558 struct super_block
*sb
;
1561 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
1562 sb
= file_inode(file
)->i_sb
;
1565 * Called directly from mm/mmap.c, or drivers/char/mem.c
1566 * for "/dev/zero", to create a shared anonymous object.
1568 if (IS_ERR(shm_mnt
))
1570 sb
= shm_mnt
->mnt_sb
;
1572 if (SHMEM_SB(sb
)->huge
!= SHMEM_HUGE_NEVER
)
1576 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
1577 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
1579 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
1582 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
1583 if (inflated_len
> TASK_SIZE
)
1585 if (inflated_len
< len
)
1588 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
1589 if (IS_ERR_VALUE(inflated_addr
))
1591 if (inflated_addr
& ~PAGE_MASK
)
1594 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
1595 inflated_addr
+= offset
- inflated_offset
;
1596 if (inflated_offset
> offset
)
1597 inflated_addr
+= HPAGE_PMD_SIZE
;
1599 if (inflated_addr
> TASK_SIZE
- len
)
1601 return inflated_addr
;
1605 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1607 struct inode
*inode
= file_inode(vma
->vm_file
);
1608 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1611 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1614 struct inode
*inode
= file_inode(vma
->vm_file
);
1617 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1618 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1622 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1624 struct inode
*inode
= file_inode(file
);
1625 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1626 int retval
= -ENOMEM
;
1628 spin_lock(&info
->lock
);
1629 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1630 if (!user_shm_lock(inode
->i_size
, user
))
1632 info
->flags
|= VM_LOCKED
;
1633 mapping_set_unevictable(file
->f_mapping
);
1635 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1636 user_shm_unlock(inode
->i_size
, user
);
1637 info
->flags
&= ~VM_LOCKED
;
1638 mapping_clear_unevictable(file
->f_mapping
);
1643 spin_unlock(&info
->lock
);
1647 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1649 file_accessed(file
);
1650 vma
->vm_ops
= &shmem_vm_ops
;
1654 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1655 umode_t mode
, dev_t dev
, unsigned long flags
)
1657 struct inode
*inode
;
1658 struct shmem_inode_info
*info
;
1659 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1661 if (shmem_reserve_inode(sb
))
1664 inode
= new_inode(sb
);
1666 inode
->i_ino
= get_next_ino();
1667 inode_init_owner(inode
, dir
, mode
);
1668 inode
->i_blocks
= 0;
1669 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1670 inode
->i_generation
= get_seconds();
1671 info
= SHMEM_I(inode
);
1672 memset(info
, 0, (char *)inode
- (char *)info
);
1673 spin_lock_init(&info
->lock
);
1674 info
->seals
= F_SEAL_SEAL
;
1675 info
->flags
= flags
& VM_NORESERVE
;
1676 INIT_LIST_HEAD(&info
->swaplist
);
1677 simple_xattrs_init(&info
->xattrs
);
1678 cache_no_acl(inode
);
1680 switch (mode
& S_IFMT
) {
1682 inode
->i_op
= &shmem_special_inode_operations
;
1683 init_special_inode(inode
, mode
, dev
);
1686 inode
->i_mapping
->a_ops
= &shmem_aops
;
1687 inode
->i_op
= &shmem_inode_operations
;
1688 inode
->i_fop
= &shmem_file_operations
;
1689 mpol_shared_policy_init(&info
->policy
,
1690 shmem_get_sbmpol(sbinfo
));
1694 /* Some things misbehave if size == 0 on a directory */
1695 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1696 inode
->i_op
= &shmem_dir_inode_operations
;
1697 inode
->i_fop
= &simple_dir_operations
;
1701 * Must not load anything in the rbtree,
1702 * mpol_free_shared_policy will not be called.
1704 mpol_shared_policy_init(&info
->policy
, NULL
);
1708 shmem_free_inode(sb
);
1712 bool shmem_mapping(struct address_space
*mapping
)
1717 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
1721 static const struct inode_operations shmem_symlink_inode_operations
;
1722 static const struct inode_operations shmem_short_symlink_operations
;
1724 #ifdef CONFIG_TMPFS_XATTR
1725 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1727 #define shmem_initxattrs NULL
1731 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1732 loff_t pos
, unsigned len
, unsigned flags
,
1733 struct page
**pagep
, void **fsdata
)
1735 struct inode
*inode
= mapping
->host
;
1736 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1737 pgoff_t index
= pos
>> PAGE_SHIFT
;
1739 /* i_mutex is held by caller */
1740 if (unlikely(info
->seals
)) {
1741 if (info
->seals
& F_SEAL_WRITE
)
1743 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
1747 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
1751 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1752 loff_t pos
, unsigned len
, unsigned copied
,
1753 struct page
*page
, void *fsdata
)
1755 struct inode
*inode
= mapping
->host
;
1757 if (pos
+ copied
> inode
->i_size
)
1758 i_size_write(inode
, pos
+ copied
);
1760 if (!PageUptodate(page
)) {
1761 if (copied
< PAGE_SIZE
) {
1762 unsigned from
= pos
& (PAGE_SIZE
- 1);
1763 zero_user_segments(page
, 0, from
,
1764 from
+ copied
, PAGE_SIZE
);
1766 SetPageUptodate(page
);
1768 set_page_dirty(page
);
1775 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1777 struct file
*file
= iocb
->ki_filp
;
1778 struct inode
*inode
= file_inode(file
);
1779 struct address_space
*mapping
= inode
->i_mapping
;
1781 unsigned long offset
;
1782 enum sgp_type sgp
= SGP_READ
;
1785 loff_t
*ppos
= &iocb
->ki_pos
;
1788 * Might this read be for a stacking filesystem? Then when reading
1789 * holes of a sparse file, we actually need to allocate those pages,
1790 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1792 if (!iter_is_iovec(to
))
1795 index
= *ppos
>> PAGE_SHIFT
;
1796 offset
= *ppos
& ~PAGE_MASK
;
1799 struct page
*page
= NULL
;
1801 unsigned long nr
, ret
;
1802 loff_t i_size
= i_size_read(inode
);
1804 end_index
= i_size
>> PAGE_SHIFT
;
1805 if (index
> end_index
)
1807 if (index
== end_index
) {
1808 nr
= i_size
& ~PAGE_MASK
;
1813 error
= shmem_getpage(inode
, index
, &page
, sgp
);
1815 if (error
== -EINVAL
)
1820 if (sgp
== SGP_CACHE
)
1821 set_page_dirty(page
);
1826 * We must evaluate after, since reads (unlike writes)
1827 * are called without i_mutex protection against truncate
1830 i_size
= i_size_read(inode
);
1831 end_index
= i_size
>> PAGE_SHIFT
;
1832 if (index
== end_index
) {
1833 nr
= i_size
& ~PAGE_MASK
;
1844 * If users can be writing to this page using arbitrary
1845 * virtual addresses, take care about potential aliasing
1846 * before reading the page on the kernel side.
1848 if (mapping_writably_mapped(mapping
))
1849 flush_dcache_page(page
);
1851 * Mark the page accessed if we read the beginning.
1854 mark_page_accessed(page
);
1856 page
= ZERO_PAGE(0);
1861 * Ok, we have the page, and it's up-to-date, so
1862 * now we can copy it to user space...
1864 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
1867 index
+= offset
>> PAGE_SHIFT
;
1868 offset
&= ~PAGE_MASK
;
1871 if (!iov_iter_count(to
))
1880 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
1881 file_accessed(file
);
1882 return retval
? retval
: error
;
1885 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1886 struct pipe_inode_info
*pipe
, size_t len
,
1889 struct address_space
*mapping
= in
->f_mapping
;
1890 struct inode
*inode
= mapping
->host
;
1891 unsigned int loff
, nr_pages
, req_pages
;
1892 struct page
*pages
[PIPE_DEF_BUFFERS
];
1893 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1895 pgoff_t index
, end_index
;
1898 struct splice_pipe_desc spd
= {
1901 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1903 .ops
= &page_cache_pipe_buf_ops
,
1904 .spd_release
= spd_release_page
,
1907 isize
= i_size_read(inode
);
1908 if (unlikely(*ppos
>= isize
))
1911 left
= isize
- *ppos
;
1912 if (unlikely(left
< len
))
1915 if (splice_grow_spd(pipe
, &spd
))
1918 index
= *ppos
>> PAGE_SHIFT
;
1919 loff
= *ppos
& ~PAGE_MASK
;
1920 req_pages
= (len
+ loff
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1921 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
1923 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1924 nr_pages
, spd
.pages
);
1925 index
+= spd
.nr_pages
;
1928 while (spd
.nr_pages
< nr_pages
) {
1929 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
);
1933 spd
.pages
[spd
.nr_pages
++] = page
;
1937 index
= *ppos
>> PAGE_SHIFT
;
1938 nr_pages
= spd
.nr_pages
;
1941 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1942 unsigned int this_len
;
1947 this_len
= min_t(unsigned long, len
, PAGE_SIZE
- loff
);
1948 page
= spd
.pages
[page_nr
];
1950 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1951 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
);
1955 put_page(spd
.pages
[page_nr
]);
1956 spd
.pages
[page_nr
] = page
;
1959 isize
= i_size_read(inode
);
1960 end_index
= (isize
- 1) >> PAGE_SHIFT
;
1961 if (unlikely(!isize
|| index
> end_index
))
1964 if (end_index
== index
) {
1967 plen
= ((isize
- 1) & ~PAGE_MASK
) + 1;
1971 this_len
= min(this_len
, plen
- loff
);
1975 spd
.partial
[page_nr
].offset
= loff
;
1976 spd
.partial
[page_nr
].len
= this_len
;
1983 while (page_nr
< nr_pages
)
1984 put_page(spd
.pages
[page_nr
++]);
1987 error
= splice_to_pipe(pipe
, &spd
);
1989 splice_shrink_spd(&spd
);
1999 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2001 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2002 pgoff_t index
, pgoff_t end
, int whence
)
2005 struct pagevec pvec
;
2006 pgoff_t indices
[PAGEVEC_SIZE
];
2010 pagevec_init(&pvec
, 0);
2011 pvec
.nr
= 1; /* start small: we may be there already */
2013 pvec
.nr
= find_get_entries(mapping
, index
,
2014 pvec
.nr
, pvec
.pages
, indices
);
2016 if (whence
== SEEK_DATA
)
2020 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2021 if (index
< indices
[i
]) {
2022 if (whence
== SEEK_HOLE
) {
2028 page
= pvec
.pages
[i
];
2029 if (page
&& !radix_tree_exceptional_entry(page
)) {
2030 if (!PageUptodate(page
))
2034 (page
&& whence
== SEEK_DATA
) ||
2035 (!page
&& whence
== SEEK_HOLE
)) {
2040 pagevec_remove_exceptionals(&pvec
);
2041 pagevec_release(&pvec
);
2042 pvec
.nr
= PAGEVEC_SIZE
;
2048 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2050 struct address_space
*mapping
= file
->f_mapping
;
2051 struct inode
*inode
= mapping
->host
;
2055 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2056 return generic_file_llseek_size(file
, offset
, whence
,
2057 MAX_LFS_FILESIZE
, i_size_read(inode
));
2059 /* We're holding i_mutex so we can access i_size directly */
2063 else if (offset
>= inode
->i_size
)
2066 start
= offset
>> PAGE_SHIFT
;
2067 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2068 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2069 new_offset
<<= PAGE_SHIFT
;
2070 if (new_offset
> offset
) {
2071 if (new_offset
< inode
->i_size
)
2072 offset
= new_offset
;
2073 else if (whence
== SEEK_DATA
)
2076 offset
= inode
->i_size
;
2081 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2082 inode_unlock(inode
);
2087 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2088 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2090 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2091 #define LAST_SCAN 4 /* about 150ms max */
2093 static void shmem_tag_pins(struct address_space
*mapping
)
2095 struct radix_tree_iter iter
;
2104 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2105 page
= radix_tree_deref_slot(slot
);
2106 if (!page
|| radix_tree_exception(page
)) {
2107 if (radix_tree_deref_retry(page
)) {
2108 slot
= radix_tree_iter_retry(&iter
);
2111 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2112 spin_lock_irq(&mapping
->tree_lock
);
2113 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2115 spin_unlock_irq(&mapping
->tree_lock
);
2118 if (need_resched()) {
2120 slot
= radix_tree_iter_next(&iter
);
2127 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2128 * via get_user_pages(), drivers might have some pending I/O without any active
2129 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2130 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2131 * them to be dropped.
2132 * The caller must guarantee that no new user will acquire writable references
2133 * to those pages to avoid races.
2135 static int shmem_wait_for_pins(struct address_space
*mapping
)
2137 struct radix_tree_iter iter
;
2143 shmem_tag_pins(mapping
);
2146 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2147 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2151 lru_add_drain_all();
2152 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2157 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2158 start
, SHMEM_TAG_PINNED
) {
2160 page
= radix_tree_deref_slot(slot
);
2161 if (radix_tree_exception(page
)) {
2162 if (radix_tree_deref_retry(page
)) {
2163 slot
= radix_tree_iter_retry(&iter
);
2171 page_count(page
) - page_mapcount(page
) != 1) {
2172 if (scan
< LAST_SCAN
)
2173 goto continue_resched
;
2176 * On the last scan, we clean up all those tags
2177 * we inserted; but make a note that we still
2178 * found pages pinned.
2183 spin_lock_irq(&mapping
->tree_lock
);
2184 radix_tree_tag_clear(&mapping
->page_tree
,
2185 iter
.index
, SHMEM_TAG_PINNED
);
2186 spin_unlock_irq(&mapping
->tree_lock
);
2188 if (need_resched()) {
2190 slot
= radix_tree_iter_next(&iter
);
2199 #define F_ALL_SEALS (F_SEAL_SEAL | \
2204 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2206 struct inode
*inode
= file_inode(file
);
2207 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2212 * Sealing allows multiple parties to share a shmem-file but restrict
2213 * access to a specific subset of file operations. Seals can only be
2214 * added, but never removed. This way, mutually untrusted parties can
2215 * share common memory regions with a well-defined policy. A malicious
2216 * peer can thus never perform unwanted operations on a shared object.
2218 * Seals are only supported on special shmem-files and always affect
2219 * the whole underlying inode. Once a seal is set, it may prevent some
2220 * kinds of access to the file. Currently, the following seals are
2222 * SEAL_SEAL: Prevent further seals from being set on this file
2223 * SEAL_SHRINK: Prevent the file from shrinking
2224 * SEAL_GROW: Prevent the file from growing
2225 * SEAL_WRITE: Prevent write access to the file
2227 * As we don't require any trust relationship between two parties, we
2228 * must prevent seals from being removed. Therefore, sealing a file
2229 * only adds a given set of seals to the file, it never touches
2230 * existing seals. Furthermore, the "setting seals"-operation can be
2231 * sealed itself, which basically prevents any further seal from being
2234 * Semantics of sealing are only defined on volatile files. Only
2235 * anonymous shmem files support sealing. More importantly, seals are
2236 * never written to disk. Therefore, there's no plan to support it on
2240 if (file
->f_op
!= &shmem_file_operations
)
2242 if (!(file
->f_mode
& FMODE_WRITE
))
2244 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2249 if (info
->seals
& F_SEAL_SEAL
) {
2254 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2255 error
= mapping_deny_writable(file
->f_mapping
);
2259 error
= shmem_wait_for_pins(file
->f_mapping
);
2261 mapping_allow_writable(file
->f_mapping
);
2266 info
->seals
|= seals
;
2270 inode_unlock(inode
);
2273 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2275 int shmem_get_seals(struct file
*file
)
2277 if (file
->f_op
!= &shmem_file_operations
)
2280 return SHMEM_I(file_inode(file
))->seals
;
2282 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2284 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2290 /* disallow upper 32bit */
2294 error
= shmem_add_seals(file
, arg
);
2297 error
= shmem_get_seals(file
);
2307 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2310 struct inode
*inode
= file_inode(file
);
2311 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2312 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2313 struct shmem_falloc shmem_falloc
;
2314 pgoff_t start
, index
, end
;
2317 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2322 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2323 struct address_space
*mapping
= file
->f_mapping
;
2324 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2325 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2326 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2328 /* protected by i_mutex */
2329 if (info
->seals
& F_SEAL_WRITE
) {
2334 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2335 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2336 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2337 spin_lock(&inode
->i_lock
);
2338 inode
->i_private
= &shmem_falloc
;
2339 spin_unlock(&inode
->i_lock
);
2341 if ((u64
)unmap_end
> (u64
)unmap_start
)
2342 unmap_mapping_range(mapping
, unmap_start
,
2343 1 + unmap_end
- unmap_start
, 0);
2344 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2345 /* No need to unmap again: hole-punching leaves COWed pages */
2347 spin_lock(&inode
->i_lock
);
2348 inode
->i_private
= NULL
;
2349 wake_up_all(&shmem_falloc_waitq
);
2350 spin_unlock(&inode
->i_lock
);
2355 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2356 error
= inode_newsize_ok(inode
, offset
+ len
);
2360 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2365 start
= offset
>> PAGE_SHIFT
;
2366 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2367 /* Try to avoid a swapstorm if len is impossible to satisfy */
2368 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2373 shmem_falloc
.waitq
= NULL
;
2374 shmem_falloc
.start
= start
;
2375 shmem_falloc
.next
= start
;
2376 shmem_falloc
.nr_falloced
= 0;
2377 shmem_falloc
.nr_unswapped
= 0;
2378 spin_lock(&inode
->i_lock
);
2379 inode
->i_private
= &shmem_falloc
;
2380 spin_unlock(&inode
->i_lock
);
2382 for (index
= start
; index
< end
; index
++) {
2386 * Good, the fallocate(2) manpage permits EINTR: we may have
2387 * been interrupted because we are using up too much memory.
2389 if (signal_pending(current
))
2391 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2394 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2396 /* Remove the !PageUptodate pages we added */
2397 if (index
> start
) {
2398 shmem_undo_range(inode
,
2399 (loff_t
)start
<< PAGE_SHIFT
,
2400 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2406 * Inform shmem_writepage() how far we have reached.
2407 * No need for lock or barrier: we have the page lock.
2409 shmem_falloc
.next
++;
2410 if (!PageUptodate(page
))
2411 shmem_falloc
.nr_falloced
++;
2414 * If !PageUptodate, leave it that way so that freeable pages
2415 * can be recognized if we need to rollback on error later.
2416 * But set_page_dirty so that memory pressure will swap rather
2417 * than free the pages we are allocating (and SGP_CACHE pages
2418 * might still be clean: we now need to mark those dirty too).
2420 set_page_dirty(page
);
2426 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2427 i_size_write(inode
, offset
+ len
);
2428 inode
->i_ctime
= CURRENT_TIME
;
2430 spin_lock(&inode
->i_lock
);
2431 inode
->i_private
= NULL
;
2432 spin_unlock(&inode
->i_lock
);
2434 inode_unlock(inode
);
2438 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2440 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2442 buf
->f_type
= TMPFS_MAGIC
;
2443 buf
->f_bsize
= PAGE_SIZE
;
2444 buf
->f_namelen
= NAME_MAX
;
2445 if (sbinfo
->max_blocks
) {
2446 buf
->f_blocks
= sbinfo
->max_blocks
;
2448 buf
->f_bfree
= sbinfo
->max_blocks
-
2449 percpu_counter_sum(&sbinfo
->used_blocks
);
2451 if (sbinfo
->max_inodes
) {
2452 buf
->f_files
= sbinfo
->max_inodes
;
2453 buf
->f_ffree
= sbinfo
->free_inodes
;
2455 /* else leave those fields 0 like simple_statfs */
2460 * File creation. Allocate an inode, and we're done..
2463 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2465 struct inode
*inode
;
2466 int error
= -ENOSPC
;
2468 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2470 error
= simple_acl_create(dir
, inode
);
2473 error
= security_inode_init_security(inode
, dir
,
2475 shmem_initxattrs
, NULL
);
2476 if (error
&& error
!= -EOPNOTSUPP
)
2480 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2481 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2482 d_instantiate(dentry
, inode
);
2483 dget(dentry
); /* Extra count - pin the dentry in core */
2492 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2494 struct inode
*inode
;
2495 int error
= -ENOSPC
;
2497 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2499 error
= security_inode_init_security(inode
, dir
,
2501 shmem_initxattrs
, NULL
);
2502 if (error
&& error
!= -EOPNOTSUPP
)
2504 error
= simple_acl_create(dir
, inode
);
2507 d_tmpfile(dentry
, inode
);
2515 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2519 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2525 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2528 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2534 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2536 struct inode
*inode
= d_inode(old_dentry
);
2540 * No ordinary (disk based) filesystem counts links as inodes;
2541 * but each new link needs a new dentry, pinning lowmem, and
2542 * tmpfs dentries cannot be pruned until they are unlinked.
2544 ret
= shmem_reserve_inode(inode
->i_sb
);
2548 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2549 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2551 ihold(inode
); /* New dentry reference */
2552 dget(dentry
); /* Extra pinning count for the created dentry */
2553 d_instantiate(dentry
, inode
);
2558 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2560 struct inode
*inode
= d_inode(dentry
);
2562 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2563 shmem_free_inode(inode
->i_sb
);
2565 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2566 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2568 dput(dentry
); /* Undo the count from "create" - this does all the work */
2572 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2574 if (!simple_empty(dentry
))
2577 drop_nlink(d_inode(dentry
));
2579 return shmem_unlink(dir
, dentry
);
2582 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2584 bool old_is_dir
= d_is_dir(old_dentry
);
2585 bool new_is_dir
= d_is_dir(new_dentry
);
2587 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2589 drop_nlink(old_dir
);
2592 drop_nlink(new_dir
);
2596 old_dir
->i_ctime
= old_dir
->i_mtime
=
2597 new_dir
->i_ctime
= new_dir
->i_mtime
=
2598 d_inode(old_dentry
)->i_ctime
=
2599 d_inode(new_dentry
)->i_ctime
= CURRENT_TIME
;
2604 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2606 struct dentry
*whiteout
;
2609 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2613 error
= shmem_mknod(old_dir
, whiteout
,
2614 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2620 * Cheat and hash the whiteout while the old dentry is still in
2621 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2623 * d_lookup() will consistently find one of them at this point,
2624 * not sure which one, but that isn't even important.
2631 * The VFS layer already does all the dentry stuff for rename,
2632 * we just have to decrement the usage count for the target if
2633 * it exists so that the VFS layer correctly free's it when it
2636 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2638 struct inode
*inode
= d_inode(old_dentry
);
2639 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2641 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2644 if (flags
& RENAME_EXCHANGE
)
2645 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2647 if (!simple_empty(new_dentry
))
2650 if (flags
& RENAME_WHITEOUT
) {
2653 error
= shmem_whiteout(old_dir
, old_dentry
);
2658 if (d_really_is_positive(new_dentry
)) {
2659 (void) shmem_unlink(new_dir
, new_dentry
);
2660 if (they_are_dirs
) {
2661 drop_nlink(d_inode(new_dentry
));
2662 drop_nlink(old_dir
);
2664 } else if (they_are_dirs
) {
2665 drop_nlink(old_dir
);
2669 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2670 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2671 old_dir
->i_ctime
= old_dir
->i_mtime
=
2672 new_dir
->i_ctime
= new_dir
->i_mtime
=
2673 inode
->i_ctime
= CURRENT_TIME
;
2677 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2681 struct inode
*inode
;
2683 struct shmem_inode_info
*info
;
2685 len
= strlen(symname
) + 1;
2686 if (len
> PAGE_SIZE
)
2687 return -ENAMETOOLONG
;
2689 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2693 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2694 shmem_initxattrs
, NULL
);
2696 if (error
!= -EOPNOTSUPP
) {
2703 info
= SHMEM_I(inode
);
2704 inode
->i_size
= len
-1;
2705 if (len
<= SHORT_SYMLINK_LEN
) {
2706 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
2707 if (!inode
->i_link
) {
2711 inode
->i_op
= &shmem_short_symlink_operations
;
2713 inode_nohighmem(inode
);
2714 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
2719 inode
->i_mapping
->a_ops
= &shmem_aops
;
2720 inode
->i_op
= &shmem_symlink_inode_operations
;
2721 memcpy(page_address(page
), symname
, len
);
2722 SetPageUptodate(page
);
2723 set_page_dirty(page
);
2727 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2728 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2729 d_instantiate(dentry
, inode
);
2734 static void shmem_put_link(void *arg
)
2736 mark_page_accessed(arg
);
2740 static const char *shmem_get_link(struct dentry
*dentry
,
2741 struct inode
*inode
,
2742 struct delayed_call
*done
)
2744 struct page
*page
= NULL
;
2747 page
= find_get_page(inode
->i_mapping
, 0);
2749 return ERR_PTR(-ECHILD
);
2750 if (!PageUptodate(page
)) {
2752 return ERR_PTR(-ECHILD
);
2755 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
2757 return ERR_PTR(error
);
2760 set_delayed_call(done
, shmem_put_link
, page
);
2761 return page_address(page
);
2764 #ifdef CONFIG_TMPFS_XATTR
2766 * Superblocks without xattr inode operations may get some security.* xattr
2767 * support from the LSM "for free". As soon as we have any other xattrs
2768 * like ACLs, we also need to implement the security.* handlers at
2769 * filesystem level, though.
2773 * Callback for security_inode_init_security() for acquiring xattrs.
2775 static int shmem_initxattrs(struct inode
*inode
,
2776 const struct xattr
*xattr_array
,
2779 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2780 const struct xattr
*xattr
;
2781 struct simple_xattr
*new_xattr
;
2784 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2785 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2789 len
= strlen(xattr
->name
) + 1;
2790 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2792 if (!new_xattr
->name
) {
2797 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2798 XATTR_SECURITY_PREFIX_LEN
);
2799 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2802 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2808 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
2809 struct dentry
*unused
, struct inode
*inode
,
2810 const char *name
, void *buffer
, size_t size
)
2812 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2814 name
= xattr_full_name(handler
, name
);
2815 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2818 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
2819 struct dentry
*unused
, struct inode
*inode
,
2820 const char *name
, const void *value
,
2821 size_t size
, int flags
)
2823 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2825 name
= xattr_full_name(handler
, name
);
2826 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2829 static const struct xattr_handler shmem_security_xattr_handler
= {
2830 .prefix
= XATTR_SECURITY_PREFIX
,
2831 .get
= shmem_xattr_handler_get
,
2832 .set
= shmem_xattr_handler_set
,
2835 static const struct xattr_handler shmem_trusted_xattr_handler
= {
2836 .prefix
= XATTR_TRUSTED_PREFIX
,
2837 .get
= shmem_xattr_handler_get
,
2838 .set
= shmem_xattr_handler_set
,
2841 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2842 #ifdef CONFIG_TMPFS_POSIX_ACL
2843 &posix_acl_access_xattr_handler
,
2844 &posix_acl_default_xattr_handler
,
2846 &shmem_security_xattr_handler
,
2847 &shmem_trusted_xattr_handler
,
2851 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2853 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
2854 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
2856 #endif /* CONFIG_TMPFS_XATTR */
2858 static const struct inode_operations shmem_short_symlink_operations
= {
2859 .readlink
= generic_readlink
,
2860 .get_link
= simple_get_link
,
2861 #ifdef CONFIG_TMPFS_XATTR
2862 .setxattr
= generic_setxattr
,
2863 .getxattr
= generic_getxattr
,
2864 .listxattr
= shmem_listxattr
,
2865 .removexattr
= generic_removexattr
,
2869 static const struct inode_operations shmem_symlink_inode_operations
= {
2870 .readlink
= generic_readlink
,
2871 .get_link
= shmem_get_link
,
2872 #ifdef CONFIG_TMPFS_XATTR
2873 .setxattr
= generic_setxattr
,
2874 .getxattr
= generic_getxattr
,
2875 .listxattr
= shmem_listxattr
,
2876 .removexattr
= generic_removexattr
,
2880 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2882 return ERR_PTR(-ESTALE
);
2885 static int shmem_match(struct inode
*ino
, void *vfh
)
2889 inum
= (inum
<< 32) | fh
[1];
2890 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2893 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2894 struct fid
*fid
, int fh_len
, int fh_type
)
2896 struct inode
*inode
;
2897 struct dentry
*dentry
= NULL
;
2904 inum
= (inum
<< 32) | fid
->raw
[1];
2906 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2907 shmem_match
, fid
->raw
);
2909 dentry
= d_find_alias(inode
);
2916 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2917 struct inode
*parent
)
2921 return FILEID_INVALID
;
2924 if (inode_unhashed(inode
)) {
2925 /* Unfortunately insert_inode_hash is not idempotent,
2926 * so as we hash inodes here rather than at creation
2927 * time, we need a lock to ensure we only try
2930 static DEFINE_SPINLOCK(lock
);
2932 if (inode_unhashed(inode
))
2933 __insert_inode_hash(inode
,
2934 inode
->i_ino
+ inode
->i_generation
);
2938 fh
[0] = inode
->i_generation
;
2939 fh
[1] = inode
->i_ino
;
2940 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2946 static const struct export_operations shmem_export_ops
= {
2947 .get_parent
= shmem_get_parent
,
2948 .encode_fh
= shmem_encode_fh
,
2949 .fh_to_dentry
= shmem_fh_to_dentry
,
2952 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2955 char *this_char
, *value
, *rest
;
2956 struct mempolicy
*mpol
= NULL
;
2960 while (options
!= NULL
) {
2961 this_char
= options
;
2964 * NUL-terminate this option: unfortunately,
2965 * mount options form a comma-separated list,
2966 * but mpol's nodelist may also contain commas.
2968 options
= strchr(options
, ',');
2969 if (options
== NULL
)
2972 if (!isdigit(*options
)) {
2979 if ((value
= strchr(this_char
,'=')) != NULL
) {
2982 pr_err("tmpfs: No value for mount option '%s'\n",
2987 if (!strcmp(this_char
,"size")) {
2988 unsigned long long size
;
2989 size
= memparse(value
,&rest
);
2991 size
<<= PAGE_SHIFT
;
2992 size
*= totalram_pages
;
2998 sbinfo
->max_blocks
=
2999 DIV_ROUND_UP(size
, PAGE_SIZE
);
3000 } else if (!strcmp(this_char
,"nr_blocks")) {
3001 sbinfo
->max_blocks
= memparse(value
, &rest
);
3004 } else if (!strcmp(this_char
,"nr_inodes")) {
3005 sbinfo
->max_inodes
= memparse(value
, &rest
);
3008 } else if (!strcmp(this_char
,"mode")) {
3011 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3014 } else if (!strcmp(this_char
,"uid")) {
3017 uid
= simple_strtoul(value
, &rest
, 0);
3020 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3021 if (!uid_valid(sbinfo
->uid
))
3023 } else if (!strcmp(this_char
,"gid")) {
3026 gid
= simple_strtoul(value
, &rest
, 0);
3029 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3030 if (!gid_valid(sbinfo
->gid
))
3032 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3033 } else if (!strcmp(this_char
, "huge")) {
3035 huge
= shmem_parse_huge(value
);
3038 if (!has_transparent_hugepage() &&
3039 huge
!= SHMEM_HUGE_NEVER
)
3041 sbinfo
->huge
= huge
;
3044 } else if (!strcmp(this_char
,"mpol")) {
3047 if (mpol_parse_str(value
, &mpol
))
3051 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3055 sbinfo
->mpol
= mpol
;
3059 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3067 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3069 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3070 struct shmem_sb_info config
= *sbinfo
;
3071 unsigned long inodes
;
3072 int error
= -EINVAL
;
3075 if (shmem_parse_options(data
, &config
, true))
3078 spin_lock(&sbinfo
->stat_lock
);
3079 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3080 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3082 if (config
.max_inodes
< inodes
)
3085 * Those tests disallow limited->unlimited while any are in use;
3086 * but we must separately disallow unlimited->limited, because
3087 * in that case we have no record of how much is already in use.
3089 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3091 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3095 sbinfo
->huge
= config
.huge
;
3096 sbinfo
->max_blocks
= config
.max_blocks
;
3097 sbinfo
->max_inodes
= config
.max_inodes
;
3098 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3101 * Preserve previous mempolicy unless mpol remount option was specified.
3104 mpol_put(sbinfo
->mpol
);
3105 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3108 spin_unlock(&sbinfo
->stat_lock
);
3112 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3114 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3116 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3117 seq_printf(seq
, ",size=%luk",
3118 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3119 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3120 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3121 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3122 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3123 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3124 seq_printf(seq
, ",uid=%u",
3125 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3126 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3127 seq_printf(seq
, ",gid=%u",
3128 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3129 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3130 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3132 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3134 shmem_show_mpol(seq
, sbinfo
->mpol
);
3138 #define MFD_NAME_PREFIX "memfd:"
3139 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3140 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3142 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3144 SYSCALL_DEFINE2(memfd_create
,
3145 const char __user
*, uname
,
3146 unsigned int, flags
)
3148 struct shmem_inode_info
*info
;
3154 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3157 /* length includes terminating zero */
3158 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3161 if (len
> MFD_NAME_MAX_LEN
+ 1)
3164 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
3168 strcpy(name
, MFD_NAME_PREFIX
);
3169 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3174 /* terminating-zero may have changed after strnlen_user() returned */
3175 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3180 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3186 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3188 error
= PTR_ERR(file
);
3191 info
= SHMEM_I(file_inode(file
));
3192 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3193 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3194 if (flags
& MFD_ALLOW_SEALING
)
3195 info
->seals
&= ~F_SEAL_SEAL
;
3197 fd_install(fd
, file
);
3208 #endif /* CONFIG_TMPFS */
3210 static void shmem_put_super(struct super_block
*sb
)
3212 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3214 percpu_counter_destroy(&sbinfo
->used_blocks
);
3215 mpol_put(sbinfo
->mpol
);
3217 sb
->s_fs_info
= NULL
;
3220 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3222 struct inode
*inode
;
3223 struct shmem_sb_info
*sbinfo
;
3226 /* Round up to L1_CACHE_BYTES to resist false sharing */
3227 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3228 L1_CACHE_BYTES
), GFP_KERNEL
);
3232 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3233 sbinfo
->uid
= current_fsuid();
3234 sbinfo
->gid
= current_fsgid();
3235 sb
->s_fs_info
= sbinfo
;
3239 * Per default we only allow half of the physical ram per
3240 * tmpfs instance, limiting inodes to one per page of lowmem;
3241 * but the internal instance is left unlimited.
3243 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3244 sbinfo
->max_blocks
= shmem_default_max_blocks();
3245 sbinfo
->max_inodes
= shmem_default_max_inodes();
3246 if (shmem_parse_options(data
, sbinfo
, false)) {
3251 sb
->s_flags
|= MS_NOUSER
;
3253 sb
->s_export_op
= &shmem_export_ops
;
3254 sb
->s_flags
|= MS_NOSEC
;
3256 sb
->s_flags
|= MS_NOUSER
;
3259 spin_lock_init(&sbinfo
->stat_lock
);
3260 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3262 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3264 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3265 sb
->s_blocksize
= PAGE_SIZE
;
3266 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3267 sb
->s_magic
= TMPFS_MAGIC
;
3268 sb
->s_op
= &shmem_ops
;
3269 sb
->s_time_gran
= 1;
3270 #ifdef CONFIG_TMPFS_XATTR
3271 sb
->s_xattr
= shmem_xattr_handlers
;
3273 #ifdef CONFIG_TMPFS_POSIX_ACL
3274 sb
->s_flags
|= MS_POSIXACL
;
3277 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3280 inode
->i_uid
= sbinfo
->uid
;
3281 inode
->i_gid
= sbinfo
->gid
;
3282 sb
->s_root
= d_make_root(inode
);
3288 shmem_put_super(sb
);
3292 static struct kmem_cache
*shmem_inode_cachep
;
3294 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3296 struct shmem_inode_info
*info
;
3297 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3300 return &info
->vfs_inode
;
3303 static void shmem_destroy_callback(struct rcu_head
*head
)
3305 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3306 if (S_ISLNK(inode
->i_mode
))
3307 kfree(inode
->i_link
);
3308 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3311 static void shmem_destroy_inode(struct inode
*inode
)
3313 if (S_ISREG(inode
->i_mode
))
3314 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3315 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3318 static void shmem_init_inode(void *foo
)
3320 struct shmem_inode_info
*info
= foo
;
3321 inode_init_once(&info
->vfs_inode
);
3324 static int shmem_init_inodecache(void)
3326 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3327 sizeof(struct shmem_inode_info
),
3328 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3332 static void shmem_destroy_inodecache(void)
3334 kmem_cache_destroy(shmem_inode_cachep
);
3337 static const struct address_space_operations shmem_aops
= {
3338 .writepage
= shmem_writepage
,
3339 .set_page_dirty
= __set_page_dirty_no_writeback
,
3341 .write_begin
= shmem_write_begin
,
3342 .write_end
= shmem_write_end
,
3344 #ifdef CONFIG_MIGRATION
3345 .migratepage
= migrate_page
,
3347 .error_remove_page
= generic_error_remove_page
,
3350 static const struct file_operations shmem_file_operations
= {
3352 .get_unmapped_area
= shmem_get_unmapped_area
,
3354 .llseek
= shmem_file_llseek
,
3355 .read_iter
= shmem_file_read_iter
,
3356 .write_iter
= generic_file_write_iter
,
3357 .fsync
= noop_fsync
,
3358 .splice_read
= shmem_file_splice_read
,
3359 .splice_write
= iter_file_splice_write
,
3360 .fallocate
= shmem_fallocate
,
3364 static const struct inode_operations shmem_inode_operations
= {
3365 .getattr
= shmem_getattr
,
3366 .setattr
= shmem_setattr
,
3367 #ifdef CONFIG_TMPFS_XATTR
3368 .setxattr
= generic_setxattr
,
3369 .getxattr
= generic_getxattr
,
3370 .listxattr
= shmem_listxattr
,
3371 .removexattr
= generic_removexattr
,
3372 .set_acl
= simple_set_acl
,
3376 static const struct inode_operations shmem_dir_inode_operations
= {
3378 .create
= shmem_create
,
3379 .lookup
= simple_lookup
,
3381 .unlink
= shmem_unlink
,
3382 .symlink
= shmem_symlink
,
3383 .mkdir
= shmem_mkdir
,
3384 .rmdir
= shmem_rmdir
,
3385 .mknod
= shmem_mknod
,
3386 .rename2
= shmem_rename2
,
3387 .tmpfile
= shmem_tmpfile
,
3389 #ifdef CONFIG_TMPFS_XATTR
3390 .setxattr
= generic_setxattr
,
3391 .getxattr
= generic_getxattr
,
3392 .listxattr
= shmem_listxattr
,
3393 .removexattr
= generic_removexattr
,
3395 #ifdef CONFIG_TMPFS_POSIX_ACL
3396 .setattr
= shmem_setattr
,
3397 .set_acl
= simple_set_acl
,
3401 static const struct inode_operations shmem_special_inode_operations
= {
3402 #ifdef CONFIG_TMPFS_XATTR
3403 .setxattr
= generic_setxattr
,
3404 .getxattr
= generic_getxattr
,
3405 .listxattr
= shmem_listxattr
,
3406 .removexattr
= generic_removexattr
,
3408 #ifdef CONFIG_TMPFS_POSIX_ACL
3409 .setattr
= shmem_setattr
,
3410 .set_acl
= simple_set_acl
,
3414 static const struct super_operations shmem_ops
= {
3415 .alloc_inode
= shmem_alloc_inode
,
3416 .destroy_inode
= shmem_destroy_inode
,
3418 .statfs
= shmem_statfs
,
3419 .remount_fs
= shmem_remount_fs
,
3420 .show_options
= shmem_show_options
,
3422 .evict_inode
= shmem_evict_inode
,
3423 .drop_inode
= generic_delete_inode
,
3424 .put_super
= shmem_put_super
,
3427 static const struct vm_operations_struct shmem_vm_ops
= {
3428 .fault
= shmem_fault
,
3429 .map_pages
= filemap_map_pages
,
3431 .set_policy
= shmem_set_policy
,
3432 .get_policy
= shmem_get_policy
,
3436 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3437 int flags
, const char *dev_name
, void *data
)
3439 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3442 static struct file_system_type shmem_fs_type
= {
3443 .owner
= THIS_MODULE
,
3445 .mount
= shmem_mount
,
3446 .kill_sb
= kill_litter_super
,
3447 .fs_flags
= FS_USERNS_MOUNT
,
3450 int __init
shmem_init(void)
3454 /* If rootfs called this, don't re-init */
3455 if (shmem_inode_cachep
)
3458 error
= shmem_init_inodecache();
3462 error
= register_filesystem(&shmem_fs_type
);
3464 pr_err("Could not register tmpfs\n");
3468 shm_mnt
= kern_mount(&shmem_fs_type
);
3469 if (IS_ERR(shm_mnt
)) {
3470 error
= PTR_ERR(shm_mnt
);
3471 pr_err("Could not kern_mount tmpfs\n");
3475 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3476 if (has_transparent_hugepage() && shmem_huge
< SHMEM_HUGE_DENY
)
3477 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3479 shmem_huge
= 0; /* just in case it was patched */
3484 unregister_filesystem(&shmem_fs_type
);
3486 shmem_destroy_inodecache();
3488 shm_mnt
= ERR_PTR(error
);
3492 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3493 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3494 struct kobj_attribute
*attr
, char *buf
)
3498 SHMEM_HUGE_WITHIN_SIZE
,
3506 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3507 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3509 count
+= sprintf(buf
+ count
, fmt
,
3510 shmem_format_huge(values
[i
]));
3512 buf
[count
- 1] = '\n';
3516 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3517 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3522 if (count
+ 1 > sizeof(tmp
))
3524 memcpy(tmp
, buf
, count
);
3526 if (count
&& tmp
[count
- 1] == '\n')
3527 tmp
[count
- 1] = '\0';
3529 huge
= shmem_parse_huge(tmp
);
3530 if (huge
== -EINVAL
)
3532 if (!has_transparent_hugepage() &&
3533 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3537 if (shmem_huge
< SHMEM_HUGE_DENY
)
3538 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3542 struct kobj_attribute shmem_enabled_attr
=
3543 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3544 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3546 #else /* !CONFIG_SHMEM */
3549 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3551 * This is intended for small system where the benefits of the full
3552 * shmem code (swap-backed and resource-limited) are outweighed by
3553 * their complexity. On systems without swap this code should be
3554 * effectively equivalent, but much lighter weight.
3557 static struct file_system_type shmem_fs_type
= {
3559 .mount
= ramfs_mount
,
3560 .kill_sb
= kill_litter_super
,
3561 .fs_flags
= FS_USERNS_MOUNT
,
3564 int __init
shmem_init(void)
3566 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3568 shm_mnt
= kern_mount(&shmem_fs_type
);
3569 BUG_ON(IS_ERR(shm_mnt
));
3574 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3579 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3584 void shmem_unlock_mapping(struct address_space
*mapping
)
3589 unsigned long shmem_get_unmapped_area(struct file
*file
,
3590 unsigned long addr
, unsigned long len
,
3591 unsigned long pgoff
, unsigned long flags
)
3593 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
3597 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3599 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3601 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3603 #define shmem_vm_ops generic_file_vm_ops
3604 #define shmem_file_operations ramfs_file_operations
3605 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3606 #define shmem_acct_size(flags, size) 0
3607 #define shmem_unacct_size(flags, size) do {} while (0)
3609 #endif /* CONFIG_SHMEM */
3613 static struct dentry_operations anon_ops
= {
3614 .d_dname
= simple_dname
3617 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3618 unsigned long flags
, unsigned int i_flags
)
3621 struct inode
*inode
;
3623 struct super_block
*sb
;
3626 if (IS_ERR(shm_mnt
))
3627 return ERR_CAST(shm_mnt
);
3629 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3630 return ERR_PTR(-EINVAL
);
3632 if (shmem_acct_size(flags
, size
))
3633 return ERR_PTR(-ENOMEM
);
3635 res
= ERR_PTR(-ENOMEM
);
3637 this.len
= strlen(name
);
3638 this.hash
= 0; /* will go */
3639 sb
= shm_mnt
->mnt_sb
;
3640 path
.mnt
= mntget(shm_mnt
);
3641 path
.dentry
= d_alloc_pseudo(sb
, &this);
3644 d_set_d_op(path
.dentry
, &anon_ops
);
3646 res
= ERR_PTR(-ENOSPC
);
3647 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3651 inode
->i_flags
|= i_flags
;
3652 d_instantiate(path
.dentry
, inode
);
3653 inode
->i_size
= size
;
3654 clear_nlink(inode
); /* It is unlinked */
3655 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3659 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3660 &shmem_file_operations
);
3667 shmem_unacct_size(flags
, size
);
3674 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3675 * kernel internal. There will be NO LSM permission checks against the
3676 * underlying inode. So users of this interface must do LSM checks at a
3677 * higher layer. The users are the big_key and shm implementations. LSM
3678 * checks are provided at the key or shm level rather than the inode.
3679 * @name: name for dentry (to be seen in /proc/<pid>/maps
3680 * @size: size to be set for the file
3681 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3683 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3685 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3689 * shmem_file_setup - get an unlinked file living in tmpfs
3690 * @name: name for dentry (to be seen in /proc/<pid>/maps
3691 * @size: size to be set for the file
3692 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3694 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3696 return __shmem_file_setup(name
, size
, flags
, 0);
3698 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3701 * shmem_zero_setup - setup a shared anonymous mapping
3702 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3704 int shmem_zero_setup(struct vm_area_struct
*vma
)
3707 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3710 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3711 * between XFS directory reading and selinux: since this file is only
3712 * accessible to the user through its mapping, use S_PRIVATE flag to
3713 * bypass file security, in the same way as shmem_kernel_file_setup().
3715 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
3717 return PTR_ERR(file
);
3721 vma
->vm_file
= file
;
3722 vma
->vm_ops
= &shmem_vm_ops
;
3727 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3728 * @mapping: the page's address_space
3729 * @index: the page index
3730 * @gfp: the page allocator flags to use if allocating
3732 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3733 * with any new page allocations done using the specified allocation flags.
3734 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3735 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3736 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3738 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3739 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3741 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3742 pgoff_t index
, gfp_t gfp
)
3745 struct inode
*inode
= mapping
->host
;
3749 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3750 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
3753 page
= ERR_PTR(error
);
3759 * The tiny !SHMEM case uses ramfs without swap
3761 return read_cache_page_gfp(mapping
, index
, gfp
);
3764 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);