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_NOHUGE
, /* like SGP_CACHE, but no huge pages */
105 SGP_HUGE
, /* like SGP_CACHE, huge pages preferred */
106 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
107 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
111 static unsigned long shmem_default_max_blocks(void)
113 return totalram_pages
/ 2;
116 static unsigned long shmem_default_max_inodes(void)
118 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
122 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
123 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
124 struct shmem_inode_info
*info
, pgoff_t index
);
125 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
126 struct page
**pagep
, enum sgp_type sgp
,
127 gfp_t gfp
, struct mm_struct
*fault_mm
, int *fault_type
);
129 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
130 struct page
**pagep
, enum sgp_type sgp
)
132 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
133 mapping_gfp_mask(inode
->i_mapping
), NULL
, NULL
);
136 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
138 return sb
->s_fs_info
;
142 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
143 * for shared memory and for shared anonymous (/dev/zero) mappings
144 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
145 * consistent with the pre-accounting of private mappings ...
147 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
149 return (flags
& VM_NORESERVE
) ?
150 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
153 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
155 if (!(flags
& VM_NORESERVE
))
156 vm_unacct_memory(VM_ACCT(size
));
159 static inline int shmem_reacct_size(unsigned long flags
,
160 loff_t oldsize
, loff_t newsize
)
162 if (!(flags
& VM_NORESERVE
)) {
163 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
164 return security_vm_enough_memory_mm(current
->mm
,
165 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
166 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
167 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
173 * ... whereas tmpfs objects are accounted incrementally as
174 * pages are allocated, in order to allow large sparse files.
175 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
176 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
178 static inline int shmem_acct_block(unsigned long flags
, long pages
)
180 if (!(flags
& VM_NORESERVE
))
183 return security_vm_enough_memory_mm(current
->mm
,
184 pages
* VM_ACCT(PAGE_SIZE
));
187 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
189 if (flags
& VM_NORESERVE
)
190 vm_unacct_memory(pages
* VM_ACCT(PAGE_SIZE
));
193 static const struct super_operations shmem_ops
;
194 static const struct address_space_operations shmem_aops
;
195 static const struct file_operations shmem_file_operations
;
196 static const struct inode_operations shmem_inode_operations
;
197 static const struct inode_operations shmem_dir_inode_operations
;
198 static const struct inode_operations shmem_special_inode_operations
;
199 static const struct vm_operations_struct shmem_vm_ops
;
201 static LIST_HEAD(shmem_swaplist
);
202 static DEFINE_MUTEX(shmem_swaplist_mutex
);
204 static int shmem_reserve_inode(struct super_block
*sb
)
206 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
207 if (sbinfo
->max_inodes
) {
208 spin_lock(&sbinfo
->stat_lock
);
209 if (!sbinfo
->free_inodes
) {
210 spin_unlock(&sbinfo
->stat_lock
);
213 sbinfo
->free_inodes
--;
214 spin_unlock(&sbinfo
->stat_lock
);
219 static void shmem_free_inode(struct super_block
*sb
)
221 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
222 if (sbinfo
->max_inodes
) {
223 spin_lock(&sbinfo
->stat_lock
);
224 sbinfo
->free_inodes
++;
225 spin_unlock(&sbinfo
->stat_lock
);
230 * shmem_recalc_inode - recalculate the block usage of an inode
231 * @inode: inode to recalc
233 * We have to calculate the free blocks since the mm can drop
234 * undirtied hole pages behind our back.
236 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
237 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
239 * It has to be called with the spinlock held.
241 static void shmem_recalc_inode(struct inode
*inode
)
243 struct shmem_inode_info
*info
= SHMEM_I(inode
);
246 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
248 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
249 if (sbinfo
->max_blocks
)
250 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
251 info
->alloced
-= freed
;
252 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
253 shmem_unacct_blocks(info
->flags
, freed
);
257 bool shmem_charge(struct inode
*inode
, long pages
)
259 struct shmem_inode_info
*info
= SHMEM_I(inode
);
260 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
263 if (shmem_acct_block(info
->flags
, pages
))
265 spin_lock_irqsave(&info
->lock
, flags
);
266 info
->alloced
+= pages
;
267 inode
->i_blocks
+= pages
* BLOCKS_PER_PAGE
;
268 shmem_recalc_inode(inode
);
269 spin_unlock_irqrestore(&info
->lock
, flags
);
270 inode
->i_mapping
->nrpages
+= pages
;
272 if (!sbinfo
->max_blocks
)
274 if (percpu_counter_compare(&sbinfo
->used_blocks
,
275 sbinfo
->max_blocks
- pages
) > 0) {
276 inode
->i_mapping
->nrpages
-= pages
;
277 spin_lock_irqsave(&info
->lock
, flags
);
278 info
->alloced
-= pages
;
279 shmem_recalc_inode(inode
);
280 spin_unlock_irqrestore(&info
->lock
, flags
);
284 percpu_counter_add(&sbinfo
->used_blocks
, pages
);
288 void shmem_uncharge(struct inode
*inode
, long pages
)
290 struct shmem_inode_info
*info
= SHMEM_I(inode
);
291 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
294 spin_lock_irqsave(&info
->lock
, flags
);
295 info
->alloced
-= pages
;
296 inode
->i_blocks
-= pages
* BLOCKS_PER_PAGE
;
297 shmem_recalc_inode(inode
);
298 spin_unlock_irqrestore(&info
->lock
, flags
);
300 if (sbinfo
->max_blocks
)
301 percpu_counter_sub(&sbinfo
->used_blocks
, pages
);
305 * Replace item expected in radix tree by a new item, while holding tree lock.
307 static int shmem_radix_tree_replace(struct address_space
*mapping
,
308 pgoff_t index
, void *expected
, void *replacement
)
313 VM_BUG_ON(!expected
);
314 VM_BUG_ON(!replacement
);
315 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
318 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
319 if (item
!= expected
)
321 radix_tree_replace_slot(pslot
, replacement
);
326 * Sometimes, before we decide whether to proceed or to fail, we must check
327 * that an entry was not already brought back from swap by a racing thread.
329 * Checking page is not enough: by the time a SwapCache page is locked, it
330 * might be reused, and again be SwapCache, using the same swap as before.
332 static bool shmem_confirm_swap(struct address_space
*mapping
,
333 pgoff_t index
, swp_entry_t swap
)
338 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
340 return item
== swp_to_radix_entry(swap
);
344 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
347 * disables huge pages for the mount;
349 * enables huge pages for the mount;
350 * SHMEM_HUGE_WITHIN_SIZE:
351 * only allocate huge pages if the page will be fully within i_size,
352 * also respect fadvise()/madvise() hints;
354 * only allocate huge pages if requested with fadvise()/madvise();
357 #define SHMEM_HUGE_NEVER 0
358 #define SHMEM_HUGE_ALWAYS 1
359 #define SHMEM_HUGE_WITHIN_SIZE 2
360 #define SHMEM_HUGE_ADVISE 3
364 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
367 * disables huge on shm_mnt and all mounts, for emergency use;
369 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
372 #define SHMEM_HUGE_DENY (-1)
373 #define SHMEM_HUGE_FORCE (-2)
375 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
376 /* ifdef here to avoid bloating shmem.o when not necessary */
378 int shmem_huge __read_mostly
;
380 static int shmem_parse_huge(const char *str
)
382 if (!strcmp(str
, "never"))
383 return SHMEM_HUGE_NEVER
;
384 if (!strcmp(str
, "always"))
385 return SHMEM_HUGE_ALWAYS
;
386 if (!strcmp(str
, "within_size"))
387 return SHMEM_HUGE_WITHIN_SIZE
;
388 if (!strcmp(str
, "advise"))
389 return SHMEM_HUGE_ADVISE
;
390 if (!strcmp(str
, "deny"))
391 return SHMEM_HUGE_DENY
;
392 if (!strcmp(str
, "force"))
393 return SHMEM_HUGE_FORCE
;
397 static const char *shmem_format_huge(int huge
)
400 case SHMEM_HUGE_NEVER
:
402 case SHMEM_HUGE_ALWAYS
:
404 case SHMEM_HUGE_WITHIN_SIZE
:
405 return "within_size";
406 case SHMEM_HUGE_ADVISE
:
408 case SHMEM_HUGE_DENY
:
410 case SHMEM_HUGE_FORCE
:
418 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
420 #define shmem_huge SHMEM_HUGE_DENY
422 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
425 * Like add_to_page_cache_locked, but error if expected item has gone.
427 static int shmem_add_to_page_cache(struct page
*page
,
428 struct address_space
*mapping
,
429 pgoff_t index
, void *expected
)
431 int error
, nr
= hpage_nr_pages(page
);
433 VM_BUG_ON_PAGE(PageTail(page
), page
);
434 VM_BUG_ON_PAGE(index
!= round_down(index
, nr
), page
);
435 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
436 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
437 VM_BUG_ON(expected
&& PageTransHuge(page
));
439 page_ref_add(page
, nr
);
440 page
->mapping
= mapping
;
443 spin_lock_irq(&mapping
->tree_lock
);
444 if (PageTransHuge(page
)) {
445 void __rcu
**results
;
450 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
,
451 &results
, &idx
, index
, 1) &&
452 idx
< index
+ HPAGE_PMD_NR
) {
457 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
458 error
= radix_tree_insert(&mapping
->page_tree
,
459 index
+ i
, page
+ i
);
462 count_vm_event(THP_FILE_ALLOC
);
464 } else if (!expected
) {
465 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
467 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
472 mapping
->nrpages
+= nr
;
473 if (PageTransHuge(page
))
474 __inc_zone_page_state(page
, NR_SHMEM_THPS
);
475 __mod_zone_page_state(page_zone(page
), NR_FILE_PAGES
, nr
);
476 __mod_zone_page_state(page_zone(page
), NR_SHMEM
, nr
);
477 spin_unlock_irq(&mapping
->tree_lock
);
479 page
->mapping
= NULL
;
480 spin_unlock_irq(&mapping
->tree_lock
);
481 page_ref_sub(page
, nr
);
487 * Like delete_from_page_cache, but substitutes swap for page.
489 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
491 struct address_space
*mapping
= page
->mapping
;
494 VM_BUG_ON_PAGE(PageCompound(page
), page
);
496 spin_lock_irq(&mapping
->tree_lock
);
497 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
498 page
->mapping
= NULL
;
500 __dec_zone_page_state(page
, NR_FILE_PAGES
);
501 __dec_zone_page_state(page
, NR_SHMEM
);
502 spin_unlock_irq(&mapping
->tree_lock
);
508 * Remove swap entry from radix tree, free the swap and its page cache.
510 static int shmem_free_swap(struct address_space
*mapping
,
511 pgoff_t index
, void *radswap
)
515 spin_lock_irq(&mapping
->tree_lock
);
516 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
517 spin_unlock_irq(&mapping
->tree_lock
);
520 free_swap_and_cache(radix_to_swp_entry(radswap
));
525 * Determine (in bytes) how many of the shmem object's pages mapped by the
526 * given offsets are swapped out.
528 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
529 * as long as the inode doesn't go away and racy results are not a problem.
531 unsigned long shmem_partial_swap_usage(struct address_space
*mapping
,
532 pgoff_t start
, pgoff_t end
)
534 struct radix_tree_iter iter
;
537 unsigned long swapped
= 0;
541 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
542 if (iter
.index
>= end
)
545 page
= radix_tree_deref_slot(slot
);
547 if (radix_tree_deref_retry(page
)) {
548 slot
= radix_tree_iter_retry(&iter
);
552 if (radix_tree_exceptional_entry(page
))
555 if (need_resched()) {
557 slot
= radix_tree_iter_next(&iter
);
563 return swapped
<< PAGE_SHIFT
;
567 * Determine (in bytes) how many of the shmem object's pages mapped by the
568 * given vma is swapped out.
570 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
571 * as long as the inode doesn't go away and racy results are not a problem.
573 unsigned long shmem_swap_usage(struct vm_area_struct
*vma
)
575 struct inode
*inode
= file_inode(vma
->vm_file
);
576 struct shmem_inode_info
*info
= SHMEM_I(inode
);
577 struct address_space
*mapping
= inode
->i_mapping
;
578 unsigned long swapped
;
580 /* Be careful as we don't hold info->lock */
581 swapped
= READ_ONCE(info
->swapped
);
584 * The easier cases are when the shmem object has nothing in swap, or
585 * the vma maps it whole. Then we can simply use the stats that we
591 if (!vma
->vm_pgoff
&& vma
->vm_end
- vma
->vm_start
>= inode
->i_size
)
592 return swapped
<< PAGE_SHIFT
;
594 /* Here comes the more involved part */
595 return shmem_partial_swap_usage(mapping
,
596 linear_page_index(vma
, vma
->vm_start
),
597 linear_page_index(vma
, vma
->vm_end
));
601 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
603 void shmem_unlock_mapping(struct address_space
*mapping
)
606 pgoff_t indices
[PAGEVEC_SIZE
];
609 pagevec_init(&pvec
, 0);
611 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
613 while (!mapping_unevictable(mapping
)) {
615 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
616 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
618 pvec
.nr
= find_get_entries(mapping
, index
,
619 PAGEVEC_SIZE
, pvec
.pages
, indices
);
622 index
= indices
[pvec
.nr
- 1] + 1;
623 pagevec_remove_exceptionals(&pvec
);
624 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
625 pagevec_release(&pvec
);
631 * Remove range of pages and swap entries from radix tree, and free them.
632 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
634 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
637 struct address_space
*mapping
= inode
->i_mapping
;
638 struct shmem_inode_info
*info
= SHMEM_I(inode
);
639 pgoff_t start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
640 pgoff_t end
= (lend
+ 1) >> PAGE_SHIFT
;
641 unsigned int partial_start
= lstart
& (PAGE_SIZE
- 1);
642 unsigned int partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
644 pgoff_t indices
[PAGEVEC_SIZE
];
645 long nr_swaps_freed
= 0;
650 end
= -1; /* unsigned, so actually very big */
652 pagevec_init(&pvec
, 0);
654 while (index
< end
) {
655 pvec
.nr
= find_get_entries(mapping
, index
,
656 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
657 pvec
.pages
, indices
);
660 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
661 struct page
*page
= pvec
.pages
[i
];
667 if (radix_tree_exceptional_entry(page
)) {
670 nr_swaps_freed
+= !shmem_free_swap(mapping
,
675 VM_BUG_ON_PAGE(page_to_pgoff(page
) != index
, page
);
677 if (!trylock_page(page
))
680 if (PageTransTail(page
)) {
681 /* Middle of THP: zero out the page */
682 clear_highpage(page
);
685 } else if (PageTransHuge(page
)) {
686 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
688 * Range ends in the middle of THP:
691 clear_highpage(page
);
695 index
+= HPAGE_PMD_NR
- 1;
696 i
+= HPAGE_PMD_NR
- 1;
699 if (!unfalloc
|| !PageUptodate(page
)) {
700 VM_BUG_ON_PAGE(PageTail(page
), page
);
701 if (page_mapping(page
) == mapping
) {
702 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
703 truncate_inode_page(mapping
, page
);
708 pagevec_remove_exceptionals(&pvec
);
709 pagevec_release(&pvec
);
715 struct page
*page
= NULL
;
716 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
);
718 unsigned int top
= PAGE_SIZE
;
723 zero_user_segment(page
, partial_start
, top
);
724 set_page_dirty(page
);
730 struct page
*page
= NULL
;
731 shmem_getpage(inode
, end
, &page
, SGP_READ
);
733 zero_user_segment(page
, 0, partial_end
);
734 set_page_dirty(page
);
743 while (index
< end
) {
746 pvec
.nr
= find_get_entries(mapping
, index
,
747 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
748 pvec
.pages
, indices
);
750 /* If all gone or hole-punch or unfalloc, we're done */
751 if (index
== start
|| end
!= -1)
753 /* But if truncating, restart to make sure all gone */
757 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
758 struct page
*page
= pvec
.pages
[i
];
764 if (radix_tree_exceptional_entry(page
)) {
767 if (shmem_free_swap(mapping
, index
, page
)) {
768 /* Swap was replaced by page: retry */
778 if (PageTransTail(page
)) {
779 /* Middle of THP: zero out the page */
780 clear_highpage(page
);
783 * Partial thp truncate due 'start' in middle
784 * of THP: don't need to look on these pages
785 * again on !pvec.nr restart.
787 if (index
!= round_down(end
, HPAGE_PMD_NR
))
790 } else if (PageTransHuge(page
)) {
791 if (index
== round_down(end
, HPAGE_PMD_NR
)) {
793 * Range ends in the middle of THP:
796 clear_highpage(page
);
800 index
+= HPAGE_PMD_NR
- 1;
801 i
+= HPAGE_PMD_NR
- 1;
804 if (!unfalloc
|| !PageUptodate(page
)) {
805 VM_BUG_ON_PAGE(PageTail(page
), page
);
806 if (page_mapping(page
) == mapping
) {
807 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
808 truncate_inode_page(mapping
, page
);
810 /* Page was replaced by swap: retry */
818 pagevec_remove_exceptionals(&pvec
);
819 pagevec_release(&pvec
);
823 spin_lock_irq(&info
->lock
);
824 info
->swapped
-= nr_swaps_freed
;
825 shmem_recalc_inode(inode
);
826 spin_unlock_irq(&info
->lock
);
829 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
831 shmem_undo_range(inode
, lstart
, lend
, false);
832 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
834 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
836 static int shmem_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
839 struct inode
*inode
= dentry
->d_inode
;
840 struct shmem_inode_info
*info
= SHMEM_I(inode
);
842 if (info
->alloced
- info
->swapped
!= inode
->i_mapping
->nrpages
) {
843 spin_lock_irq(&info
->lock
);
844 shmem_recalc_inode(inode
);
845 spin_unlock_irq(&info
->lock
);
847 generic_fillattr(inode
, stat
);
851 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
853 struct inode
*inode
= d_inode(dentry
);
854 struct shmem_inode_info
*info
= SHMEM_I(inode
);
857 error
= inode_change_ok(inode
, attr
);
861 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
862 loff_t oldsize
= inode
->i_size
;
863 loff_t newsize
= attr
->ia_size
;
865 /* protected by i_mutex */
866 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
867 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
870 if (newsize
!= oldsize
) {
871 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
875 i_size_write(inode
, newsize
);
876 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
878 if (newsize
<= oldsize
) {
879 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
880 if (oldsize
> holebegin
)
881 unmap_mapping_range(inode
->i_mapping
,
884 shmem_truncate_range(inode
,
885 newsize
, (loff_t
)-1);
886 /* unmap again to remove racily COWed private pages */
887 if (oldsize
> holebegin
)
888 unmap_mapping_range(inode
->i_mapping
,
893 setattr_copy(inode
, attr
);
894 if (attr
->ia_valid
& ATTR_MODE
)
895 error
= posix_acl_chmod(inode
, inode
->i_mode
);
899 static void shmem_evict_inode(struct inode
*inode
)
901 struct shmem_inode_info
*info
= SHMEM_I(inode
);
903 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
904 shmem_unacct_size(info
->flags
, inode
->i_size
);
906 shmem_truncate_range(inode
, 0, (loff_t
)-1);
907 if (!list_empty(&info
->swaplist
)) {
908 mutex_lock(&shmem_swaplist_mutex
);
909 list_del_init(&info
->swaplist
);
910 mutex_unlock(&shmem_swaplist_mutex
);
914 simple_xattrs_free(&info
->xattrs
);
915 WARN_ON(inode
->i_blocks
);
916 shmem_free_inode(inode
->i_sb
);
921 * If swap found in inode, free it and move page from swapcache to filecache.
923 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
924 swp_entry_t swap
, struct page
**pagep
)
926 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
932 radswap
= swp_to_radix_entry(swap
);
933 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
935 return -EAGAIN
; /* tell shmem_unuse we found nothing */
938 * Move _head_ to start search for next from here.
939 * But be careful: shmem_evict_inode checks list_empty without taking
940 * mutex, and there's an instant in list_move_tail when info->swaplist
941 * would appear empty, if it were the only one on shmem_swaplist.
943 if (shmem_swaplist
.next
!= &info
->swaplist
)
944 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
946 gfp
= mapping_gfp_mask(mapping
);
947 if (shmem_should_replace_page(*pagep
, gfp
)) {
948 mutex_unlock(&shmem_swaplist_mutex
);
949 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
950 mutex_lock(&shmem_swaplist_mutex
);
952 * We needed to drop mutex to make that restrictive page
953 * allocation, but the inode might have been freed while we
954 * dropped it: although a racing shmem_evict_inode() cannot
955 * complete without emptying the radix_tree, our page lock
956 * on this swapcache page is not enough to prevent that -
957 * free_swap_and_cache() of our swap entry will only
958 * trylock_page(), removing swap from radix_tree whatever.
960 * We must not proceed to shmem_add_to_page_cache() if the
961 * inode has been freed, but of course we cannot rely on
962 * inode or mapping or info to check that. However, we can
963 * safely check if our swap entry is still in use (and here
964 * it can't have got reused for another page): if it's still
965 * in use, then the inode cannot have been freed yet, and we
966 * can safely proceed (if it's no longer in use, that tells
967 * nothing about the inode, but we don't need to unuse swap).
969 if (!page_swapcount(*pagep
))
974 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
975 * but also to hold up shmem_evict_inode(): so inode cannot be freed
976 * beneath us (pagelock doesn't help until the page is in pagecache).
979 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
981 if (error
!= -ENOMEM
) {
983 * Truncation and eviction use free_swap_and_cache(), which
984 * only does trylock page: if we raced, best clean up here.
986 delete_from_swap_cache(*pagep
);
987 set_page_dirty(*pagep
);
989 spin_lock_irq(&info
->lock
);
991 spin_unlock_irq(&info
->lock
);
999 * Search through swapped inodes to find and replace swap by page.
1001 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
1003 struct list_head
*this, *next
;
1004 struct shmem_inode_info
*info
;
1005 struct mem_cgroup
*memcg
;
1009 * There's a faint possibility that swap page was replaced before
1010 * caller locked it: caller will come back later with the right page.
1012 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
1016 * Charge page using GFP_KERNEL while we can wait, before taking
1017 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1018 * Charged back to the user (not to caller) when swap account is used.
1020 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
,
1024 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1027 mutex_lock(&shmem_swaplist_mutex
);
1028 list_for_each_safe(this, next
, &shmem_swaplist
) {
1029 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
1031 error
= shmem_unuse_inode(info
, swap
, &page
);
1033 list_del_init(&info
->swaplist
);
1035 if (error
!= -EAGAIN
)
1037 /* found nothing in this: move on to search the next */
1039 mutex_unlock(&shmem_swaplist_mutex
);
1042 if (error
!= -ENOMEM
)
1044 mem_cgroup_cancel_charge(page
, memcg
, false);
1046 mem_cgroup_commit_charge(page
, memcg
, true, false);
1054 * Move the page from the page cache to the swap cache.
1056 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
1058 struct shmem_inode_info
*info
;
1059 struct address_space
*mapping
;
1060 struct inode
*inode
;
1064 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1065 BUG_ON(!PageLocked(page
));
1066 mapping
= page
->mapping
;
1067 index
= page
->index
;
1068 inode
= mapping
->host
;
1069 info
= SHMEM_I(inode
);
1070 if (info
->flags
& VM_LOCKED
)
1072 if (!total_swap_pages
)
1076 * Our capabilities prevent regular writeback or sync from ever calling
1077 * shmem_writepage; but a stacking filesystem might use ->writepage of
1078 * its underlying filesystem, in which case tmpfs should write out to
1079 * swap only in response to memory pressure, and not for the writeback
1082 if (!wbc
->for_reclaim
) {
1083 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1088 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1089 * value into swapfile.c, the only way we can correctly account for a
1090 * fallocated page arriving here is now to initialize it and write it.
1092 * That's okay for a page already fallocated earlier, but if we have
1093 * not yet completed the fallocation, then (a) we want to keep track
1094 * of this page in case we have to undo it, and (b) it may not be a
1095 * good idea to continue anyway, once we're pushing into swap. So
1096 * reactivate the page, and let shmem_fallocate() quit when too many.
1098 if (!PageUptodate(page
)) {
1099 if (inode
->i_private
) {
1100 struct shmem_falloc
*shmem_falloc
;
1101 spin_lock(&inode
->i_lock
);
1102 shmem_falloc
= inode
->i_private
;
1104 !shmem_falloc
->waitq
&&
1105 index
>= shmem_falloc
->start
&&
1106 index
< shmem_falloc
->next
)
1107 shmem_falloc
->nr_unswapped
++;
1109 shmem_falloc
= NULL
;
1110 spin_unlock(&inode
->i_lock
);
1114 clear_highpage(page
);
1115 flush_dcache_page(page
);
1116 SetPageUptodate(page
);
1119 swap
= get_swap_page();
1123 if (mem_cgroup_try_charge_swap(page
, swap
))
1127 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1128 * if it's not already there. Do it now before the page is
1129 * moved to swap cache, when its pagelock no longer protects
1130 * the inode from eviction. But don't unlock the mutex until
1131 * we've incremented swapped, because shmem_unuse_inode() will
1132 * prune a !swapped inode from the swaplist under this mutex.
1134 mutex_lock(&shmem_swaplist_mutex
);
1135 if (list_empty(&info
->swaplist
))
1136 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
1138 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
1139 spin_lock_irq(&info
->lock
);
1140 shmem_recalc_inode(inode
);
1142 spin_unlock_irq(&info
->lock
);
1144 swap_shmem_alloc(swap
);
1145 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
1147 mutex_unlock(&shmem_swaplist_mutex
);
1148 BUG_ON(page_mapped(page
));
1149 swap_writepage(page
, wbc
);
1153 mutex_unlock(&shmem_swaplist_mutex
);
1155 swapcache_free(swap
);
1157 set_page_dirty(page
);
1158 if (wbc
->for_reclaim
)
1159 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
1164 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1165 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1169 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
1170 return; /* show nothing */
1172 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
1174 seq_printf(seq
, ",mpol=%s", buffer
);
1177 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1179 struct mempolicy
*mpol
= NULL
;
1181 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
1182 mpol
= sbinfo
->mpol
;
1184 spin_unlock(&sbinfo
->stat_lock
);
1188 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1189 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
1192 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1196 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1198 #define vm_policy vm_private_data
1201 static void shmem_pseudo_vma_init(struct vm_area_struct
*vma
,
1202 struct shmem_inode_info
*info
, pgoff_t index
)
1204 /* Create a pseudo vma that just contains the policy */
1206 /* Bias interleave by inode number to distribute better across nodes */
1207 vma
->vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
1209 vma
->vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
1212 static void shmem_pseudo_vma_destroy(struct vm_area_struct
*vma
)
1214 /* Drop reference taken by mpol_shared_policy_lookup() */
1215 mpol_cond_put(vma
->vm_policy
);
1218 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
1219 struct shmem_inode_info
*info
, pgoff_t index
)
1221 struct vm_area_struct pvma
;
1224 shmem_pseudo_vma_init(&pvma
, info
, index
);
1225 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
1226 shmem_pseudo_vma_destroy(&pvma
);
1231 static struct page
*shmem_alloc_hugepage(gfp_t gfp
,
1232 struct shmem_inode_info
*info
, pgoff_t index
)
1234 struct vm_area_struct pvma
;
1235 struct inode
*inode
= &info
->vfs_inode
;
1236 struct address_space
*mapping
= inode
->i_mapping
;
1237 pgoff_t idx
, hindex
= round_down(index
, HPAGE_PMD_NR
);
1238 void __rcu
**results
;
1241 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1245 if (radix_tree_gang_lookup_slot(&mapping
->page_tree
, &results
, &idx
,
1246 hindex
, 1) && idx
< hindex
+ HPAGE_PMD_NR
) {
1252 shmem_pseudo_vma_init(&pvma
, info
, hindex
);
1253 page
= alloc_pages_vma(gfp
| __GFP_COMP
| __GFP_NORETRY
| __GFP_NOWARN
,
1254 HPAGE_PMD_ORDER
, &pvma
, 0, numa_node_id(), true);
1255 shmem_pseudo_vma_destroy(&pvma
);
1257 prep_transhuge_page(page
);
1261 static struct page
*shmem_alloc_page(gfp_t gfp
,
1262 struct shmem_inode_info
*info
, pgoff_t index
)
1264 struct vm_area_struct pvma
;
1267 shmem_pseudo_vma_init(&pvma
, info
, index
);
1268 page
= alloc_page_vma(gfp
, &pvma
, 0);
1269 shmem_pseudo_vma_destroy(&pvma
);
1274 static struct page
*shmem_alloc_and_acct_page(gfp_t gfp
,
1275 struct shmem_inode_info
*info
, struct shmem_sb_info
*sbinfo
,
1276 pgoff_t index
, bool huge
)
1282 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1284 nr
= huge
? HPAGE_PMD_NR
: 1;
1286 if (shmem_acct_block(info
->flags
, nr
))
1288 if (sbinfo
->max_blocks
) {
1289 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1290 sbinfo
->max_blocks
- nr
) > 0)
1292 percpu_counter_add(&sbinfo
->used_blocks
, nr
);
1296 page
= shmem_alloc_hugepage(gfp
, info
, index
);
1298 page
= shmem_alloc_page(gfp
, info
, index
);
1300 __SetPageLocked(page
);
1301 __SetPageSwapBacked(page
);
1306 if (sbinfo
->max_blocks
)
1307 percpu_counter_add(&sbinfo
->used_blocks
, -nr
);
1309 shmem_unacct_blocks(info
->flags
, nr
);
1311 return ERR_PTR(err
);
1315 * When a page is moved from swapcache to shmem filecache (either by the
1316 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1317 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1318 * ignorance of the mapping it belongs to. If that mapping has special
1319 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1320 * we may need to copy to a suitable page before moving to filecache.
1322 * In a future release, this may well be extended to respect cpuset and
1323 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1324 * but for now it is a simple matter of zone.
1326 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1328 return page_zonenum(page
) > gfp_zone(gfp
);
1331 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1332 struct shmem_inode_info
*info
, pgoff_t index
)
1334 struct page
*oldpage
, *newpage
;
1335 struct address_space
*swap_mapping
;
1340 swap_index
= page_private(oldpage
);
1341 swap_mapping
= page_mapping(oldpage
);
1344 * We have arrived here because our zones are constrained, so don't
1345 * limit chance of success by further cpuset and node constraints.
1347 gfp
&= ~GFP_CONSTRAINT_MASK
;
1348 newpage
= shmem_alloc_page(gfp
, info
, index
);
1353 copy_highpage(newpage
, oldpage
);
1354 flush_dcache_page(newpage
);
1356 SetPageUptodate(newpage
);
1357 set_page_private(newpage
, swap_index
);
1358 SetPageSwapCache(newpage
);
1361 * Our caller will very soon move newpage out of swapcache, but it's
1362 * a nice clean interface for us to replace oldpage by newpage there.
1364 spin_lock_irq(&swap_mapping
->tree_lock
);
1365 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1368 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1369 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1371 spin_unlock_irq(&swap_mapping
->tree_lock
);
1373 if (unlikely(error
)) {
1375 * Is this possible? I think not, now that our callers check
1376 * both PageSwapCache and page_private after getting page lock;
1377 * but be defensive. Reverse old to newpage for clear and free.
1381 mem_cgroup_migrate(oldpage
, newpage
);
1382 lru_cache_add_anon(newpage
);
1386 ClearPageSwapCache(oldpage
);
1387 set_page_private(oldpage
, 0);
1389 unlock_page(oldpage
);
1396 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1398 * If we allocate a new one we do not mark it dirty. That's up to the
1399 * vm. If we swap it in we mark it dirty since we also free the swap
1400 * entry since a page cannot live in both the swap and page cache.
1402 * fault_mm and fault_type are only supplied by shmem_fault:
1403 * otherwise they are NULL.
1405 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1406 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
,
1407 struct mm_struct
*fault_mm
, int *fault_type
)
1409 struct address_space
*mapping
= inode
->i_mapping
;
1410 struct shmem_inode_info
*info
;
1411 struct shmem_sb_info
*sbinfo
;
1412 struct mm_struct
*charge_mm
;
1413 struct mem_cgroup
*memcg
;
1416 enum sgp_type sgp_huge
= sgp
;
1417 pgoff_t hindex
= index
;
1422 if (index
> (MAX_LFS_FILESIZE
>> PAGE_SHIFT
))
1424 if (sgp
== SGP_NOHUGE
|| sgp
== SGP_HUGE
)
1428 page
= find_lock_entry(mapping
, index
);
1429 if (radix_tree_exceptional_entry(page
)) {
1430 swap
= radix_to_swp_entry(page
);
1434 if (sgp
<= SGP_CACHE
&&
1435 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1440 if (page
&& sgp
== SGP_WRITE
)
1441 mark_page_accessed(page
);
1443 /* fallocated page? */
1444 if (page
&& !PageUptodate(page
)) {
1445 if (sgp
!= SGP_READ
)
1451 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1457 * Fast cache lookup did not find it:
1458 * bring it back from swap or allocate.
1460 info
= SHMEM_I(inode
);
1461 sbinfo
= SHMEM_SB(inode
->i_sb
);
1462 charge_mm
= fault_mm
? : current
->mm
;
1465 /* Look it up and read it in.. */
1466 page
= lookup_swap_cache(swap
);
1468 /* Or update major stats only when swapin succeeds?? */
1470 *fault_type
|= VM_FAULT_MAJOR
;
1471 count_vm_event(PGMAJFAULT
);
1472 mem_cgroup_count_vm_event(fault_mm
, PGMAJFAULT
);
1474 /* Here we actually start the io */
1475 page
= shmem_swapin(swap
, gfp
, info
, index
);
1482 /* We have to do this with page locked to prevent races */
1484 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1485 !shmem_confirm_swap(mapping
, index
, swap
)) {
1486 error
= -EEXIST
; /* try again */
1489 if (!PageUptodate(page
)) {
1493 wait_on_page_writeback(page
);
1495 if (shmem_should_replace_page(page
, gfp
)) {
1496 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1501 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1504 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1505 swp_to_radix_entry(swap
));
1507 * We already confirmed swap under page lock, and make
1508 * no memory allocation here, so usually no possibility
1509 * of error; but free_swap_and_cache() only trylocks a
1510 * page, so it is just possible that the entry has been
1511 * truncated or holepunched since swap was confirmed.
1512 * shmem_undo_range() will have done some of the
1513 * unaccounting, now delete_from_swap_cache() will do
1515 * Reset swap.val? No, leave it so "failed" goes back to
1516 * "repeat": reading a hole and writing should succeed.
1519 mem_cgroup_cancel_charge(page
, memcg
, false);
1520 delete_from_swap_cache(page
);
1526 mem_cgroup_commit_charge(page
, memcg
, true, false);
1528 spin_lock_irq(&info
->lock
);
1530 shmem_recalc_inode(inode
);
1531 spin_unlock_irq(&info
->lock
);
1533 if (sgp
== SGP_WRITE
)
1534 mark_page_accessed(page
);
1536 delete_from_swap_cache(page
);
1537 set_page_dirty(page
);
1541 /* shmem_symlink() */
1542 if (mapping
->a_ops
!= &shmem_aops
)
1544 if (shmem_huge
== SHMEM_HUGE_DENY
|| sgp_huge
== SGP_NOHUGE
)
1546 if (shmem_huge
== SHMEM_HUGE_FORCE
)
1548 switch (sbinfo
->huge
) {
1551 case SHMEM_HUGE_NEVER
:
1553 case SHMEM_HUGE_WITHIN_SIZE
:
1554 off
= round_up(index
, HPAGE_PMD_NR
);
1555 i_size
= round_up(i_size_read(inode
), PAGE_SIZE
);
1556 if (i_size
>= HPAGE_PMD_SIZE
&&
1557 i_size
>> PAGE_SHIFT
>= off
)
1560 case SHMEM_HUGE_ADVISE
:
1561 if (sgp_huge
== SGP_HUGE
)
1563 /* TODO: implement fadvise() hints */
1568 page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1571 alloc_nohuge
: page
= shmem_alloc_and_acct_page(gfp
, info
, sbinfo
,
1575 error
= PTR_ERR(page
);
1580 if (PageTransHuge(page
))
1581 hindex
= round_down(index
, HPAGE_PMD_NR
);
1585 if (sgp
== SGP_WRITE
)
1586 __SetPageReferenced(page
);
1588 error
= mem_cgroup_try_charge(page
, charge_mm
, gfp
, &memcg
,
1589 PageTransHuge(page
));
1592 error
= radix_tree_maybe_preload_order(gfp
& GFP_RECLAIM_MASK
,
1593 compound_order(page
));
1595 error
= shmem_add_to_page_cache(page
, mapping
, hindex
,
1597 radix_tree_preload_end();
1600 mem_cgroup_cancel_charge(page
, memcg
,
1601 PageTransHuge(page
));
1604 mem_cgroup_commit_charge(page
, memcg
, false,
1605 PageTransHuge(page
));
1606 lru_cache_add_anon(page
);
1608 spin_lock_irq(&info
->lock
);
1609 info
->alloced
+= 1 << compound_order(page
);
1610 inode
->i_blocks
+= BLOCKS_PER_PAGE
<< compound_order(page
);
1611 shmem_recalc_inode(inode
);
1612 spin_unlock_irq(&info
->lock
);
1616 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1618 if (sgp
== SGP_FALLOC
)
1622 * Let SGP_WRITE caller clear ends if write does not fill page;
1623 * but SGP_FALLOC on a page fallocated earlier must initialize
1624 * it now, lest undo on failure cancel our earlier guarantee.
1626 if (sgp
!= SGP_WRITE
&& !PageUptodate(page
)) {
1627 struct page
*head
= compound_head(page
);
1630 for (i
= 0; i
< (1 << compound_order(head
)); i
++) {
1631 clear_highpage(head
+ i
);
1632 flush_dcache_page(head
+ i
);
1634 SetPageUptodate(head
);
1638 /* Perhaps the file has been truncated since we checked */
1639 if (sgp
<= SGP_CACHE
&&
1640 ((loff_t
)index
<< PAGE_SHIFT
) >= i_size_read(inode
)) {
1642 ClearPageDirty(page
);
1643 delete_from_page_cache(page
);
1644 spin_lock_irq(&info
->lock
);
1645 shmem_recalc_inode(inode
);
1646 spin_unlock_irq(&info
->lock
);
1651 *pagep
= page
+ index
- hindex
;
1658 if (sbinfo
->max_blocks
)
1659 percpu_counter_sub(&sbinfo
->used_blocks
,
1660 1 << compound_order(page
));
1661 shmem_unacct_blocks(info
->flags
, 1 << compound_order(page
));
1663 if (PageTransHuge(page
)) {
1669 if (swap
.val
&& !shmem_confirm_swap(mapping
, index
, swap
))
1676 if (error
== -ENOSPC
&& !once
++) {
1677 info
= SHMEM_I(inode
);
1678 spin_lock_irq(&info
->lock
);
1679 shmem_recalc_inode(inode
);
1680 spin_unlock_irq(&info
->lock
);
1683 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1688 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1690 struct inode
*inode
= file_inode(vma
->vm_file
);
1691 gfp_t gfp
= mapping_gfp_mask(inode
->i_mapping
);
1694 int ret
= VM_FAULT_LOCKED
;
1697 * Trinity finds that probing a hole which tmpfs is punching can
1698 * prevent the hole-punch from ever completing: which in turn
1699 * locks writers out with its hold on i_mutex. So refrain from
1700 * faulting pages into the hole while it's being punched. Although
1701 * shmem_undo_range() does remove the additions, it may be unable to
1702 * keep up, as each new page needs its own unmap_mapping_range() call,
1703 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1705 * It does not matter if we sometimes reach this check just before the
1706 * hole-punch begins, so that one fault then races with the punch:
1707 * we just need to make racing faults a rare case.
1709 * The implementation below would be much simpler if we just used a
1710 * standard mutex or completion: but we cannot take i_mutex in fault,
1711 * and bloating every shmem inode for this unlikely case would be sad.
1713 if (unlikely(inode
->i_private
)) {
1714 struct shmem_falloc
*shmem_falloc
;
1716 spin_lock(&inode
->i_lock
);
1717 shmem_falloc
= inode
->i_private
;
1719 shmem_falloc
->waitq
&&
1720 vmf
->pgoff
>= shmem_falloc
->start
&&
1721 vmf
->pgoff
< shmem_falloc
->next
) {
1722 wait_queue_head_t
*shmem_falloc_waitq
;
1723 DEFINE_WAIT(shmem_fault_wait
);
1725 ret
= VM_FAULT_NOPAGE
;
1726 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1727 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1728 /* It's polite to up mmap_sem if we can */
1729 up_read(&vma
->vm_mm
->mmap_sem
);
1730 ret
= VM_FAULT_RETRY
;
1733 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1734 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1735 TASK_UNINTERRUPTIBLE
);
1736 spin_unlock(&inode
->i_lock
);
1740 * shmem_falloc_waitq points into the shmem_fallocate()
1741 * stack of the hole-punching task: shmem_falloc_waitq
1742 * is usually invalid by the time we reach here, but
1743 * finish_wait() does not dereference it in that case;
1744 * though i_lock needed lest racing with wake_up_all().
1746 spin_lock(&inode
->i_lock
);
1747 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1748 spin_unlock(&inode
->i_lock
);
1751 spin_unlock(&inode
->i_lock
);
1755 if (vma
->vm_flags
& VM_HUGEPAGE
)
1757 else if (vma
->vm_flags
& VM_NOHUGEPAGE
)
1760 error
= shmem_getpage_gfp(inode
, vmf
->pgoff
, &vmf
->page
, sgp
,
1761 gfp
, vma
->vm_mm
, &ret
);
1763 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1767 unsigned long shmem_get_unmapped_area(struct file
*file
,
1768 unsigned long uaddr
, unsigned long len
,
1769 unsigned long pgoff
, unsigned long flags
)
1771 unsigned long (*get_area
)(struct file
*,
1772 unsigned long, unsigned long, unsigned long, unsigned long);
1774 unsigned long offset
;
1775 unsigned long inflated_len
;
1776 unsigned long inflated_addr
;
1777 unsigned long inflated_offset
;
1779 if (len
> TASK_SIZE
)
1782 get_area
= current
->mm
->get_unmapped_area
;
1783 addr
= get_area(file
, uaddr
, len
, pgoff
, flags
);
1785 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1787 if (IS_ERR_VALUE(addr
))
1789 if (addr
& ~PAGE_MASK
)
1791 if (addr
> TASK_SIZE
- len
)
1794 if (shmem_huge
== SHMEM_HUGE_DENY
)
1796 if (len
< HPAGE_PMD_SIZE
)
1798 if (flags
& MAP_FIXED
)
1801 * Our priority is to support MAP_SHARED mapped hugely;
1802 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1803 * But if caller specified an address hint, respect that as before.
1808 if (shmem_huge
!= SHMEM_HUGE_FORCE
) {
1809 struct super_block
*sb
;
1812 VM_BUG_ON(file
->f_op
!= &shmem_file_operations
);
1813 sb
= file_inode(file
)->i_sb
;
1816 * Called directly from mm/mmap.c, or drivers/char/mem.c
1817 * for "/dev/zero", to create a shared anonymous object.
1819 if (IS_ERR(shm_mnt
))
1821 sb
= shm_mnt
->mnt_sb
;
1823 if (SHMEM_SB(sb
)->huge
!= SHMEM_HUGE_NEVER
)
1827 offset
= (pgoff
<< PAGE_SHIFT
) & (HPAGE_PMD_SIZE
-1);
1828 if (offset
&& offset
+ len
< 2 * HPAGE_PMD_SIZE
)
1830 if ((addr
& (HPAGE_PMD_SIZE
-1)) == offset
)
1833 inflated_len
= len
+ HPAGE_PMD_SIZE
- PAGE_SIZE
;
1834 if (inflated_len
> TASK_SIZE
)
1836 if (inflated_len
< len
)
1839 inflated_addr
= get_area(NULL
, 0, inflated_len
, 0, flags
);
1840 if (IS_ERR_VALUE(inflated_addr
))
1842 if (inflated_addr
& ~PAGE_MASK
)
1845 inflated_offset
= inflated_addr
& (HPAGE_PMD_SIZE
-1);
1846 inflated_addr
+= offset
- inflated_offset
;
1847 if (inflated_offset
> offset
)
1848 inflated_addr
+= HPAGE_PMD_SIZE
;
1850 if (inflated_addr
> TASK_SIZE
- len
)
1852 return inflated_addr
;
1856 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1858 struct inode
*inode
= file_inode(vma
->vm_file
);
1859 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1862 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1865 struct inode
*inode
= file_inode(vma
->vm_file
);
1868 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1869 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1873 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1875 struct inode
*inode
= file_inode(file
);
1876 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1877 int retval
= -ENOMEM
;
1879 spin_lock_irq(&info
->lock
);
1880 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1881 if (!user_shm_lock(inode
->i_size
, user
))
1883 info
->flags
|= VM_LOCKED
;
1884 mapping_set_unevictable(file
->f_mapping
);
1886 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1887 user_shm_unlock(inode
->i_size
, user
);
1888 info
->flags
&= ~VM_LOCKED
;
1889 mapping_clear_unevictable(file
->f_mapping
);
1894 spin_unlock_irq(&info
->lock
);
1898 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1900 file_accessed(file
);
1901 vma
->vm_ops
= &shmem_vm_ops
;
1905 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1906 umode_t mode
, dev_t dev
, unsigned long flags
)
1908 struct inode
*inode
;
1909 struct shmem_inode_info
*info
;
1910 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1912 if (shmem_reserve_inode(sb
))
1915 inode
= new_inode(sb
);
1917 inode
->i_ino
= get_next_ino();
1918 inode_init_owner(inode
, dir
, mode
);
1919 inode
->i_blocks
= 0;
1920 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1921 inode
->i_generation
= get_seconds();
1922 info
= SHMEM_I(inode
);
1923 memset(info
, 0, (char *)inode
- (char *)info
);
1924 spin_lock_init(&info
->lock
);
1925 info
->seals
= F_SEAL_SEAL
;
1926 info
->flags
= flags
& VM_NORESERVE
;
1927 INIT_LIST_HEAD(&info
->swaplist
);
1928 simple_xattrs_init(&info
->xattrs
);
1929 cache_no_acl(inode
);
1931 switch (mode
& S_IFMT
) {
1933 inode
->i_op
= &shmem_special_inode_operations
;
1934 init_special_inode(inode
, mode
, dev
);
1937 inode
->i_mapping
->a_ops
= &shmem_aops
;
1938 inode
->i_op
= &shmem_inode_operations
;
1939 inode
->i_fop
= &shmem_file_operations
;
1940 mpol_shared_policy_init(&info
->policy
,
1941 shmem_get_sbmpol(sbinfo
));
1945 /* Some things misbehave if size == 0 on a directory */
1946 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1947 inode
->i_op
= &shmem_dir_inode_operations
;
1948 inode
->i_fop
= &simple_dir_operations
;
1952 * Must not load anything in the rbtree,
1953 * mpol_free_shared_policy will not be called.
1955 mpol_shared_policy_init(&info
->policy
, NULL
);
1959 shmem_free_inode(sb
);
1963 bool shmem_mapping(struct address_space
*mapping
)
1968 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
1972 static const struct inode_operations shmem_symlink_inode_operations
;
1973 static const struct inode_operations shmem_short_symlink_operations
;
1975 #ifdef CONFIG_TMPFS_XATTR
1976 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1978 #define shmem_initxattrs NULL
1982 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1983 loff_t pos
, unsigned len
, unsigned flags
,
1984 struct page
**pagep
, void **fsdata
)
1986 struct inode
*inode
= mapping
->host
;
1987 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1988 pgoff_t index
= pos
>> PAGE_SHIFT
;
1990 /* i_mutex is held by caller */
1991 if (unlikely(info
->seals
)) {
1992 if (info
->seals
& F_SEAL_WRITE
)
1994 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
1998 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
);
2002 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
2003 loff_t pos
, unsigned len
, unsigned copied
,
2004 struct page
*page
, void *fsdata
)
2006 struct inode
*inode
= mapping
->host
;
2008 if (pos
+ copied
> inode
->i_size
)
2009 i_size_write(inode
, pos
+ copied
);
2011 if (!PageUptodate(page
)) {
2012 struct page
*head
= compound_head(page
);
2013 if (PageTransCompound(page
)) {
2016 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
2017 if (head
+ i
== page
)
2019 clear_highpage(head
+ i
);
2020 flush_dcache_page(head
+ i
);
2023 if (copied
< PAGE_SIZE
) {
2024 unsigned from
= pos
& (PAGE_SIZE
- 1);
2025 zero_user_segments(page
, 0, from
,
2026 from
+ copied
, PAGE_SIZE
);
2028 SetPageUptodate(head
);
2030 set_page_dirty(page
);
2037 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2039 struct file
*file
= iocb
->ki_filp
;
2040 struct inode
*inode
= file_inode(file
);
2041 struct address_space
*mapping
= inode
->i_mapping
;
2043 unsigned long offset
;
2044 enum sgp_type sgp
= SGP_READ
;
2047 loff_t
*ppos
= &iocb
->ki_pos
;
2050 * Might this read be for a stacking filesystem? Then when reading
2051 * holes of a sparse file, we actually need to allocate those pages,
2052 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2054 if (!iter_is_iovec(to
))
2057 index
= *ppos
>> PAGE_SHIFT
;
2058 offset
= *ppos
& ~PAGE_MASK
;
2061 struct page
*page
= NULL
;
2063 unsigned long nr
, ret
;
2064 loff_t i_size
= i_size_read(inode
);
2066 end_index
= i_size
>> PAGE_SHIFT
;
2067 if (index
> end_index
)
2069 if (index
== end_index
) {
2070 nr
= i_size
& ~PAGE_MASK
;
2075 error
= shmem_getpage(inode
, index
, &page
, sgp
);
2077 if (error
== -EINVAL
)
2082 if (sgp
== SGP_CACHE
)
2083 set_page_dirty(page
);
2088 * We must evaluate after, since reads (unlike writes)
2089 * are called without i_mutex protection against truncate
2092 i_size
= i_size_read(inode
);
2093 end_index
= i_size
>> PAGE_SHIFT
;
2094 if (index
== end_index
) {
2095 nr
= i_size
& ~PAGE_MASK
;
2106 * If users can be writing to this page using arbitrary
2107 * virtual addresses, take care about potential aliasing
2108 * before reading the page on the kernel side.
2110 if (mapping_writably_mapped(mapping
))
2111 flush_dcache_page(page
);
2113 * Mark the page accessed if we read the beginning.
2116 mark_page_accessed(page
);
2118 page
= ZERO_PAGE(0);
2123 * Ok, we have the page, and it's up-to-date, so
2124 * now we can copy it to user space...
2126 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
2129 index
+= offset
>> PAGE_SHIFT
;
2130 offset
&= ~PAGE_MASK
;
2133 if (!iov_iter_count(to
))
2142 *ppos
= ((loff_t
) index
<< PAGE_SHIFT
) + offset
;
2143 file_accessed(file
);
2144 return retval
? retval
: error
;
2147 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
2148 struct pipe_inode_info
*pipe
, size_t len
,
2151 struct address_space
*mapping
= in
->f_mapping
;
2152 struct inode
*inode
= mapping
->host
;
2153 unsigned int loff
, nr_pages
, req_pages
;
2154 struct page
*pages
[PIPE_DEF_BUFFERS
];
2155 struct partial_page partial
[PIPE_DEF_BUFFERS
];
2157 pgoff_t index
, end_index
;
2160 struct splice_pipe_desc spd
= {
2163 .nr_pages_max
= PIPE_DEF_BUFFERS
,
2165 .ops
= &page_cache_pipe_buf_ops
,
2166 .spd_release
= spd_release_page
,
2169 isize
= i_size_read(inode
);
2170 if (unlikely(*ppos
>= isize
))
2173 left
= isize
- *ppos
;
2174 if (unlikely(left
< len
))
2177 if (splice_grow_spd(pipe
, &spd
))
2180 index
= *ppos
>> PAGE_SHIFT
;
2181 loff
= *ppos
& ~PAGE_MASK
;
2182 req_pages
= (len
+ loff
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2183 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
2185 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
2186 nr_pages
, spd
.pages
);
2187 index
+= spd
.nr_pages
;
2190 while (spd
.nr_pages
< nr_pages
) {
2191 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
);
2195 spd
.pages
[spd
.nr_pages
++] = page
;
2199 index
= *ppos
>> PAGE_SHIFT
;
2200 nr_pages
= spd
.nr_pages
;
2203 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
2204 unsigned int this_len
;
2209 this_len
= min_t(unsigned long, len
, PAGE_SIZE
- loff
);
2210 page
= spd
.pages
[page_nr
];
2212 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
2213 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
);
2217 put_page(spd
.pages
[page_nr
]);
2218 spd
.pages
[page_nr
] = page
;
2221 isize
= i_size_read(inode
);
2222 end_index
= (isize
- 1) >> PAGE_SHIFT
;
2223 if (unlikely(!isize
|| index
> end_index
))
2226 if (end_index
== index
) {
2229 plen
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2233 this_len
= min(this_len
, plen
- loff
);
2237 spd
.partial
[page_nr
].offset
= loff
;
2238 spd
.partial
[page_nr
].len
= this_len
;
2245 while (page_nr
< nr_pages
)
2246 put_page(spd
.pages
[page_nr
++]);
2249 error
= splice_to_pipe(pipe
, &spd
);
2251 splice_shrink_spd(&spd
);
2261 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2263 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
2264 pgoff_t index
, pgoff_t end
, int whence
)
2267 struct pagevec pvec
;
2268 pgoff_t indices
[PAGEVEC_SIZE
];
2272 pagevec_init(&pvec
, 0);
2273 pvec
.nr
= 1; /* start small: we may be there already */
2275 pvec
.nr
= find_get_entries(mapping
, index
,
2276 pvec
.nr
, pvec
.pages
, indices
);
2278 if (whence
== SEEK_DATA
)
2282 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
2283 if (index
< indices
[i
]) {
2284 if (whence
== SEEK_HOLE
) {
2290 page
= pvec
.pages
[i
];
2291 if (page
&& !radix_tree_exceptional_entry(page
)) {
2292 if (!PageUptodate(page
))
2296 (page
&& whence
== SEEK_DATA
) ||
2297 (!page
&& whence
== SEEK_HOLE
)) {
2302 pagevec_remove_exceptionals(&pvec
);
2303 pagevec_release(&pvec
);
2304 pvec
.nr
= PAGEVEC_SIZE
;
2310 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2312 struct address_space
*mapping
= file
->f_mapping
;
2313 struct inode
*inode
= mapping
->host
;
2317 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
2318 return generic_file_llseek_size(file
, offset
, whence
,
2319 MAX_LFS_FILESIZE
, i_size_read(inode
));
2321 /* We're holding i_mutex so we can access i_size directly */
2325 else if (offset
>= inode
->i_size
)
2328 start
= offset
>> PAGE_SHIFT
;
2329 end
= (inode
->i_size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2330 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
2331 new_offset
<<= PAGE_SHIFT
;
2332 if (new_offset
> offset
) {
2333 if (new_offset
< inode
->i_size
)
2334 offset
= new_offset
;
2335 else if (whence
== SEEK_DATA
)
2338 offset
= inode
->i_size
;
2343 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
2344 inode_unlock(inode
);
2349 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2350 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2352 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2353 #define LAST_SCAN 4 /* about 150ms max */
2355 static void shmem_tag_pins(struct address_space
*mapping
)
2357 struct radix_tree_iter iter
;
2366 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
2367 page
= radix_tree_deref_slot(slot
);
2368 if (!page
|| radix_tree_exception(page
)) {
2369 if (radix_tree_deref_retry(page
)) {
2370 slot
= radix_tree_iter_retry(&iter
);
2373 } else if (page_count(page
) - page_mapcount(page
) > 1) {
2374 spin_lock_irq(&mapping
->tree_lock
);
2375 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
2377 spin_unlock_irq(&mapping
->tree_lock
);
2380 if (need_resched()) {
2382 slot
= radix_tree_iter_next(&iter
);
2389 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2390 * via get_user_pages(), drivers might have some pending I/O without any active
2391 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2392 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2393 * them to be dropped.
2394 * The caller must guarantee that no new user will acquire writable references
2395 * to those pages to avoid races.
2397 static int shmem_wait_for_pins(struct address_space
*mapping
)
2399 struct radix_tree_iter iter
;
2405 shmem_tag_pins(mapping
);
2408 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
2409 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
2413 lru_add_drain_all();
2414 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
2419 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
2420 start
, SHMEM_TAG_PINNED
) {
2422 page
= radix_tree_deref_slot(slot
);
2423 if (radix_tree_exception(page
)) {
2424 if (radix_tree_deref_retry(page
)) {
2425 slot
= radix_tree_iter_retry(&iter
);
2433 page_count(page
) - page_mapcount(page
) != 1) {
2434 if (scan
< LAST_SCAN
)
2435 goto continue_resched
;
2438 * On the last scan, we clean up all those tags
2439 * we inserted; but make a note that we still
2440 * found pages pinned.
2445 spin_lock_irq(&mapping
->tree_lock
);
2446 radix_tree_tag_clear(&mapping
->page_tree
,
2447 iter
.index
, SHMEM_TAG_PINNED
);
2448 spin_unlock_irq(&mapping
->tree_lock
);
2450 if (need_resched()) {
2452 slot
= radix_tree_iter_next(&iter
);
2461 #define F_ALL_SEALS (F_SEAL_SEAL | \
2466 int shmem_add_seals(struct file
*file
, unsigned int seals
)
2468 struct inode
*inode
= file_inode(file
);
2469 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2474 * Sealing allows multiple parties to share a shmem-file but restrict
2475 * access to a specific subset of file operations. Seals can only be
2476 * added, but never removed. This way, mutually untrusted parties can
2477 * share common memory regions with a well-defined policy. A malicious
2478 * peer can thus never perform unwanted operations on a shared object.
2480 * Seals are only supported on special shmem-files and always affect
2481 * the whole underlying inode. Once a seal is set, it may prevent some
2482 * kinds of access to the file. Currently, the following seals are
2484 * SEAL_SEAL: Prevent further seals from being set on this file
2485 * SEAL_SHRINK: Prevent the file from shrinking
2486 * SEAL_GROW: Prevent the file from growing
2487 * SEAL_WRITE: Prevent write access to the file
2489 * As we don't require any trust relationship between two parties, we
2490 * must prevent seals from being removed. Therefore, sealing a file
2491 * only adds a given set of seals to the file, it never touches
2492 * existing seals. Furthermore, the "setting seals"-operation can be
2493 * sealed itself, which basically prevents any further seal from being
2496 * Semantics of sealing are only defined on volatile files. Only
2497 * anonymous shmem files support sealing. More importantly, seals are
2498 * never written to disk. Therefore, there's no plan to support it on
2502 if (file
->f_op
!= &shmem_file_operations
)
2504 if (!(file
->f_mode
& FMODE_WRITE
))
2506 if (seals
& ~(unsigned int)F_ALL_SEALS
)
2511 if (info
->seals
& F_SEAL_SEAL
) {
2516 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
2517 error
= mapping_deny_writable(file
->f_mapping
);
2521 error
= shmem_wait_for_pins(file
->f_mapping
);
2523 mapping_allow_writable(file
->f_mapping
);
2528 info
->seals
|= seals
;
2532 inode_unlock(inode
);
2535 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2537 int shmem_get_seals(struct file
*file
)
2539 if (file
->f_op
!= &shmem_file_operations
)
2542 return SHMEM_I(file_inode(file
))->seals
;
2544 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2546 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2552 /* disallow upper 32bit */
2556 error
= shmem_add_seals(file
, arg
);
2559 error
= shmem_get_seals(file
);
2569 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2572 struct inode
*inode
= file_inode(file
);
2573 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2574 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2575 struct shmem_falloc shmem_falloc
;
2576 pgoff_t start
, index
, end
;
2579 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2584 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2585 struct address_space
*mapping
= file
->f_mapping
;
2586 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2587 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2588 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2590 /* protected by i_mutex */
2591 if (info
->seals
& F_SEAL_WRITE
) {
2596 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2597 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2598 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2599 spin_lock(&inode
->i_lock
);
2600 inode
->i_private
= &shmem_falloc
;
2601 spin_unlock(&inode
->i_lock
);
2603 if ((u64
)unmap_end
> (u64
)unmap_start
)
2604 unmap_mapping_range(mapping
, unmap_start
,
2605 1 + unmap_end
- unmap_start
, 0);
2606 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2607 /* No need to unmap again: hole-punching leaves COWed pages */
2609 spin_lock(&inode
->i_lock
);
2610 inode
->i_private
= NULL
;
2611 wake_up_all(&shmem_falloc_waitq
);
2612 spin_unlock(&inode
->i_lock
);
2617 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2618 error
= inode_newsize_ok(inode
, offset
+ len
);
2622 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2627 start
= offset
>> PAGE_SHIFT
;
2628 end
= (offset
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2629 /* Try to avoid a swapstorm if len is impossible to satisfy */
2630 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2635 shmem_falloc
.waitq
= NULL
;
2636 shmem_falloc
.start
= start
;
2637 shmem_falloc
.next
= start
;
2638 shmem_falloc
.nr_falloced
= 0;
2639 shmem_falloc
.nr_unswapped
= 0;
2640 spin_lock(&inode
->i_lock
);
2641 inode
->i_private
= &shmem_falloc
;
2642 spin_unlock(&inode
->i_lock
);
2644 for (index
= start
; index
< end
; index
++) {
2648 * Good, the fallocate(2) manpage permits EINTR: we may have
2649 * been interrupted because we are using up too much memory.
2651 if (signal_pending(current
))
2653 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2656 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
);
2658 /* Remove the !PageUptodate pages we added */
2659 if (index
> start
) {
2660 shmem_undo_range(inode
,
2661 (loff_t
)start
<< PAGE_SHIFT
,
2662 ((loff_t
)index
<< PAGE_SHIFT
) - 1, true);
2668 * Inform shmem_writepage() how far we have reached.
2669 * No need for lock or barrier: we have the page lock.
2671 shmem_falloc
.next
++;
2672 if (!PageUptodate(page
))
2673 shmem_falloc
.nr_falloced
++;
2676 * If !PageUptodate, leave it that way so that freeable pages
2677 * can be recognized if we need to rollback on error later.
2678 * But set_page_dirty so that memory pressure will swap rather
2679 * than free the pages we are allocating (and SGP_CACHE pages
2680 * might still be clean: we now need to mark those dirty too).
2682 set_page_dirty(page
);
2688 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2689 i_size_write(inode
, offset
+ len
);
2690 inode
->i_ctime
= CURRENT_TIME
;
2692 spin_lock(&inode
->i_lock
);
2693 inode
->i_private
= NULL
;
2694 spin_unlock(&inode
->i_lock
);
2696 inode_unlock(inode
);
2700 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2702 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2704 buf
->f_type
= TMPFS_MAGIC
;
2705 buf
->f_bsize
= PAGE_SIZE
;
2706 buf
->f_namelen
= NAME_MAX
;
2707 if (sbinfo
->max_blocks
) {
2708 buf
->f_blocks
= sbinfo
->max_blocks
;
2710 buf
->f_bfree
= sbinfo
->max_blocks
-
2711 percpu_counter_sum(&sbinfo
->used_blocks
);
2713 if (sbinfo
->max_inodes
) {
2714 buf
->f_files
= sbinfo
->max_inodes
;
2715 buf
->f_ffree
= sbinfo
->free_inodes
;
2717 /* else leave those fields 0 like simple_statfs */
2722 * File creation. Allocate an inode, and we're done..
2725 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2727 struct inode
*inode
;
2728 int error
= -ENOSPC
;
2730 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2732 error
= simple_acl_create(dir
, inode
);
2735 error
= security_inode_init_security(inode
, dir
,
2737 shmem_initxattrs
, NULL
);
2738 if (error
&& error
!= -EOPNOTSUPP
)
2742 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2743 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2744 d_instantiate(dentry
, inode
);
2745 dget(dentry
); /* Extra count - pin the dentry in core */
2754 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2756 struct inode
*inode
;
2757 int error
= -ENOSPC
;
2759 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2761 error
= security_inode_init_security(inode
, dir
,
2763 shmem_initxattrs
, NULL
);
2764 if (error
&& error
!= -EOPNOTSUPP
)
2766 error
= simple_acl_create(dir
, inode
);
2769 d_tmpfile(dentry
, inode
);
2777 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2781 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2787 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2790 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2796 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2798 struct inode
*inode
= d_inode(old_dentry
);
2802 * No ordinary (disk based) filesystem counts links as inodes;
2803 * but each new link needs a new dentry, pinning lowmem, and
2804 * tmpfs dentries cannot be pruned until they are unlinked.
2806 ret
= shmem_reserve_inode(inode
->i_sb
);
2810 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2811 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2813 ihold(inode
); /* New dentry reference */
2814 dget(dentry
); /* Extra pinning count for the created dentry */
2815 d_instantiate(dentry
, inode
);
2820 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2822 struct inode
*inode
= d_inode(dentry
);
2824 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2825 shmem_free_inode(inode
->i_sb
);
2827 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2828 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2830 dput(dentry
); /* Undo the count from "create" - this does all the work */
2834 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2836 if (!simple_empty(dentry
))
2839 drop_nlink(d_inode(dentry
));
2841 return shmem_unlink(dir
, dentry
);
2844 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2846 bool old_is_dir
= d_is_dir(old_dentry
);
2847 bool new_is_dir
= d_is_dir(new_dentry
);
2849 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2851 drop_nlink(old_dir
);
2854 drop_nlink(new_dir
);
2858 old_dir
->i_ctime
= old_dir
->i_mtime
=
2859 new_dir
->i_ctime
= new_dir
->i_mtime
=
2860 d_inode(old_dentry
)->i_ctime
=
2861 d_inode(new_dentry
)->i_ctime
= CURRENT_TIME
;
2866 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2868 struct dentry
*whiteout
;
2871 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2875 error
= shmem_mknod(old_dir
, whiteout
,
2876 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2882 * Cheat and hash the whiteout while the old dentry is still in
2883 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2885 * d_lookup() will consistently find one of them at this point,
2886 * not sure which one, but that isn't even important.
2893 * The VFS layer already does all the dentry stuff for rename,
2894 * we just have to decrement the usage count for the target if
2895 * it exists so that the VFS layer correctly free's it when it
2898 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2900 struct inode
*inode
= d_inode(old_dentry
);
2901 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2903 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2906 if (flags
& RENAME_EXCHANGE
)
2907 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2909 if (!simple_empty(new_dentry
))
2912 if (flags
& RENAME_WHITEOUT
) {
2915 error
= shmem_whiteout(old_dir
, old_dentry
);
2920 if (d_really_is_positive(new_dentry
)) {
2921 (void) shmem_unlink(new_dir
, new_dentry
);
2922 if (they_are_dirs
) {
2923 drop_nlink(d_inode(new_dentry
));
2924 drop_nlink(old_dir
);
2926 } else if (they_are_dirs
) {
2927 drop_nlink(old_dir
);
2931 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2932 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2933 old_dir
->i_ctime
= old_dir
->i_mtime
=
2934 new_dir
->i_ctime
= new_dir
->i_mtime
=
2935 inode
->i_ctime
= CURRENT_TIME
;
2939 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2943 struct inode
*inode
;
2945 struct shmem_inode_info
*info
;
2947 len
= strlen(symname
) + 1;
2948 if (len
> PAGE_SIZE
)
2949 return -ENAMETOOLONG
;
2951 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2955 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2956 shmem_initxattrs
, NULL
);
2958 if (error
!= -EOPNOTSUPP
) {
2965 info
= SHMEM_I(inode
);
2966 inode
->i_size
= len
-1;
2967 if (len
<= SHORT_SYMLINK_LEN
) {
2968 inode
->i_link
= kmemdup(symname
, len
, GFP_KERNEL
);
2969 if (!inode
->i_link
) {
2973 inode
->i_op
= &shmem_short_symlink_operations
;
2975 inode_nohighmem(inode
);
2976 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
);
2981 inode
->i_mapping
->a_ops
= &shmem_aops
;
2982 inode
->i_op
= &shmem_symlink_inode_operations
;
2983 memcpy(page_address(page
), symname
, len
);
2984 SetPageUptodate(page
);
2985 set_page_dirty(page
);
2989 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2990 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2991 d_instantiate(dentry
, inode
);
2996 static void shmem_put_link(void *arg
)
2998 mark_page_accessed(arg
);
3002 static const char *shmem_get_link(struct dentry
*dentry
,
3003 struct inode
*inode
,
3004 struct delayed_call
*done
)
3006 struct page
*page
= NULL
;
3009 page
= find_get_page(inode
->i_mapping
, 0);
3011 return ERR_PTR(-ECHILD
);
3012 if (!PageUptodate(page
)) {
3014 return ERR_PTR(-ECHILD
);
3017 error
= shmem_getpage(inode
, 0, &page
, SGP_READ
);
3019 return ERR_PTR(error
);
3022 set_delayed_call(done
, shmem_put_link
, page
);
3023 return page_address(page
);
3026 #ifdef CONFIG_TMPFS_XATTR
3028 * Superblocks without xattr inode operations may get some security.* xattr
3029 * support from the LSM "for free". As soon as we have any other xattrs
3030 * like ACLs, we also need to implement the security.* handlers at
3031 * filesystem level, though.
3035 * Callback for security_inode_init_security() for acquiring xattrs.
3037 static int shmem_initxattrs(struct inode
*inode
,
3038 const struct xattr
*xattr_array
,
3041 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3042 const struct xattr
*xattr
;
3043 struct simple_xattr
*new_xattr
;
3046 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
3047 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
3051 len
= strlen(xattr
->name
) + 1;
3052 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
3054 if (!new_xattr
->name
) {
3059 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
3060 XATTR_SECURITY_PREFIX_LEN
);
3061 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
3064 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
3070 static int shmem_xattr_handler_get(const struct xattr_handler
*handler
,
3071 struct dentry
*unused
, struct inode
*inode
,
3072 const char *name
, void *buffer
, size_t size
)
3074 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3076 name
= xattr_full_name(handler
, name
);
3077 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
3080 static int shmem_xattr_handler_set(const struct xattr_handler
*handler
,
3081 struct dentry
*unused
, struct inode
*inode
,
3082 const char *name
, const void *value
,
3083 size_t size
, int flags
)
3085 struct shmem_inode_info
*info
= SHMEM_I(inode
);
3087 name
= xattr_full_name(handler
, name
);
3088 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
3091 static const struct xattr_handler shmem_security_xattr_handler
= {
3092 .prefix
= XATTR_SECURITY_PREFIX
,
3093 .get
= shmem_xattr_handler_get
,
3094 .set
= shmem_xattr_handler_set
,
3097 static const struct xattr_handler shmem_trusted_xattr_handler
= {
3098 .prefix
= XATTR_TRUSTED_PREFIX
,
3099 .get
= shmem_xattr_handler_get
,
3100 .set
= shmem_xattr_handler_set
,
3103 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
3104 #ifdef CONFIG_TMPFS_POSIX_ACL
3105 &posix_acl_access_xattr_handler
,
3106 &posix_acl_default_xattr_handler
,
3108 &shmem_security_xattr_handler
,
3109 &shmem_trusted_xattr_handler
,
3113 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
3115 struct shmem_inode_info
*info
= SHMEM_I(d_inode(dentry
));
3116 return simple_xattr_list(d_inode(dentry
), &info
->xattrs
, buffer
, size
);
3118 #endif /* CONFIG_TMPFS_XATTR */
3120 static const struct inode_operations shmem_short_symlink_operations
= {
3121 .readlink
= generic_readlink
,
3122 .get_link
= simple_get_link
,
3123 #ifdef CONFIG_TMPFS_XATTR
3124 .setxattr
= generic_setxattr
,
3125 .getxattr
= generic_getxattr
,
3126 .listxattr
= shmem_listxattr
,
3127 .removexattr
= generic_removexattr
,
3131 static const struct inode_operations shmem_symlink_inode_operations
= {
3132 .readlink
= generic_readlink
,
3133 .get_link
= shmem_get_link
,
3134 #ifdef CONFIG_TMPFS_XATTR
3135 .setxattr
= generic_setxattr
,
3136 .getxattr
= generic_getxattr
,
3137 .listxattr
= shmem_listxattr
,
3138 .removexattr
= generic_removexattr
,
3142 static struct dentry
*shmem_get_parent(struct dentry
*child
)
3144 return ERR_PTR(-ESTALE
);
3147 static int shmem_match(struct inode
*ino
, void *vfh
)
3151 inum
= (inum
<< 32) | fh
[1];
3152 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
3155 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
3156 struct fid
*fid
, int fh_len
, int fh_type
)
3158 struct inode
*inode
;
3159 struct dentry
*dentry
= NULL
;
3166 inum
= (inum
<< 32) | fid
->raw
[1];
3168 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
3169 shmem_match
, fid
->raw
);
3171 dentry
= d_find_alias(inode
);
3178 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
3179 struct inode
*parent
)
3183 return FILEID_INVALID
;
3186 if (inode_unhashed(inode
)) {
3187 /* Unfortunately insert_inode_hash is not idempotent,
3188 * so as we hash inodes here rather than at creation
3189 * time, we need a lock to ensure we only try
3192 static DEFINE_SPINLOCK(lock
);
3194 if (inode_unhashed(inode
))
3195 __insert_inode_hash(inode
,
3196 inode
->i_ino
+ inode
->i_generation
);
3200 fh
[0] = inode
->i_generation
;
3201 fh
[1] = inode
->i_ino
;
3202 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
3208 static const struct export_operations shmem_export_ops
= {
3209 .get_parent
= shmem_get_parent
,
3210 .encode_fh
= shmem_encode_fh
,
3211 .fh_to_dentry
= shmem_fh_to_dentry
,
3214 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
3217 char *this_char
, *value
, *rest
;
3218 struct mempolicy
*mpol
= NULL
;
3222 while (options
!= NULL
) {
3223 this_char
= options
;
3226 * NUL-terminate this option: unfortunately,
3227 * mount options form a comma-separated list,
3228 * but mpol's nodelist may also contain commas.
3230 options
= strchr(options
, ',');
3231 if (options
== NULL
)
3234 if (!isdigit(*options
)) {
3241 if ((value
= strchr(this_char
,'=')) != NULL
) {
3244 pr_err("tmpfs: No value for mount option '%s'\n",
3249 if (!strcmp(this_char
,"size")) {
3250 unsigned long long size
;
3251 size
= memparse(value
,&rest
);
3253 size
<<= PAGE_SHIFT
;
3254 size
*= totalram_pages
;
3260 sbinfo
->max_blocks
=
3261 DIV_ROUND_UP(size
, PAGE_SIZE
);
3262 } else if (!strcmp(this_char
,"nr_blocks")) {
3263 sbinfo
->max_blocks
= memparse(value
, &rest
);
3266 } else if (!strcmp(this_char
,"nr_inodes")) {
3267 sbinfo
->max_inodes
= memparse(value
, &rest
);
3270 } else if (!strcmp(this_char
,"mode")) {
3273 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
3276 } else if (!strcmp(this_char
,"uid")) {
3279 uid
= simple_strtoul(value
, &rest
, 0);
3282 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
3283 if (!uid_valid(sbinfo
->uid
))
3285 } else if (!strcmp(this_char
,"gid")) {
3288 gid
= simple_strtoul(value
, &rest
, 0);
3291 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
3292 if (!gid_valid(sbinfo
->gid
))
3294 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3295 } else if (!strcmp(this_char
, "huge")) {
3297 huge
= shmem_parse_huge(value
);
3300 if (!has_transparent_hugepage() &&
3301 huge
!= SHMEM_HUGE_NEVER
)
3303 sbinfo
->huge
= huge
;
3306 } else if (!strcmp(this_char
,"mpol")) {
3309 if (mpol_parse_str(value
, &mpol
))
3313 pr_err("tmpfs: Bad mount option %s\n", this_char
);
3317 sbinfo
->mpol
= mpol
;
3321 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3329 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
3331 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3332 struct shmem_sb_info config
= *sbinfo
;
3333 unsigned long inodes
;
3334 int error
= -EINVAL
;
3337 if (shmem_parse_options(data
, &config
, true))
3340 spin_lock(&sbinfo
->stat_lock
);
3341 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
3342 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
3344 if (config
.max_inodes
< inodes
)
3347 * Those tests disallow limited->unlimited while any are in use;
3348 * but we must separately disallow unlimited->limited, because
3349 * in that case we have no record of how much is already in use.
3351 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
3353 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
3357 sbinfo
->huge
= config
.huge
;
3358 sbinfo
->max_blocks
= config
.max_blocks
;
3359 sbinfo
->max_inodes
= config
.max_inodes
;
3360 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
3363 * Preserve previous mempolicy unless mpol remount option was specified.
3366 mpol_put(sbinfo
->mpol
);
3367 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
3370 spin_unlock(&sbinfo
->stat_lock
);
3374 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
3376 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
3378 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
3379 seq_printf(seq
, ",size=%luk",
3380 sbinfo
->max_blocks
<< (PAGE_SHIFT
- 10));
3381 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
3382 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
3383 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
3384 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
3385 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
3386 seq_printf(seq
, ",uid=%u",
3387 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
3388 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
3389 seq_printf(seq
, ",gid=%u",
3390 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
3391 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3392 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3394 seq_printf(seq
, ",huge=%s", shmem_format_huge(sbinfo
->huge
));
3396 shmem_show_mpol(seq
, sbinfo
->mpol
);
3400 #define MFD_NAME_PREFIX "memfd:"
3401 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3402 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3404 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3406 SYSCALL_DEFINE2(memfd_create
,
3407 const char __user
*, uname
,
3408 unsigned int, flags
)
3410 struct shmem_inode_info
*info
;
3416 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
3419 /* length includes terminating zero */
3420 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
3423 if (len
> MFD_NAME_MAX_LEN
+ 1)
3426 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
3430 strcpy(name
, MFD_NAME_PREFIX
);
3431 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
3436 /* terminating-zero may have changed after strnlen_user() returned */
3437 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
3442 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
3448 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
3450 error
= PTR_ERR(file
);
3453 info
= SHMEM_I(file_inode(file
));
3454 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
3455 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
3456 if (flags
& MFD_ALLOW_SEALING
)
3457 info
->seals
&= ~F_SEAL_SEAL
;
3459 fd_install(fd
, file
);
3470 #endif /* CONFIG_TMPFS */
3472 static void shmem_put_super(struct super_block
*sb
)
3474 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
3476 percpu_counter_destroy(&sbinfo
->used_blocks
);
3477 mpol_put(sbinfo
->mpol
);
3479 sb
->s_fs_info
= NULL
;
3482 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
3484 struct inode
*inode
;
3485 struct shmem_sb_info
*sbinfo
;
3488 /* Round up to L1_CACHE_BYTES to resist false sharing */
3489 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
3490 L1_CACHE_BYTES
), GFP_KERNEL
);
3494 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
3495 sbinfo
->uid
= current_fsuid();
3496 sbinfo
->gid
= current_fsgid();
3497 sb
->s_fs_info
= sbinfo
;
3501 * Per default we only allow half of the physical ram per
3502 * tmpfs instance, limiting inodes to one per page of lowmem;
3503 * but the internal instance is left unlimited.
3505 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3506 sbinfo
->max_blocks
= shmem_default_max_blocks();
3507 sbinfo
->max_inodes
= shmem_default_max_inodes();
3508 if (shmem_parse_options(data
, sbinfo
, false)) {
3513 sb
->s_flags
|= MS_NOUSER
;
3515 sb
->s_export_op
= &shmem_export_ops
;
3516 sb
->s_flags
|= MS_NOSEC
;
3518 sb
->s_flags
|= MS_NOUSER
;
3521 spin_lock_init(&sbinfo
->stat_lock
);
3522 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3524 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3526 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3527 sb
->s_blocksize
= PAGE_SIZE
;
3528 sb
->s_blocksize_bits
= PAGE_SHIFT
;
3529 sb
->s_magic
= TMPFS_MAGIC
;
3530 sb
->s_op
= &shmem_ops
;
3531 sb
->s_time_gran
= 1;
3532 #ifdef CONFIG_TMPFS_XATTR
3533 sb
->s_xattr
= shmem_xattr_handlers
;
3535 #ifdef CONFIG_TMPFS_POSIX_ACL
3536 sb
->s_flags
|= MS_POSIXACL
;
3539 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3542 inode
->i_uid
= sbinfo
->uid
;
3543 inode
->i_gid
= sbinfo
->gid
;
3544 sb
->s_root
= d_make_root(inode
);
3550 shmem_put_super(sb
);
3554 static struct kmem_cache
*shmem_inode_cachep
;
3556 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3558 struct shmem_inode_info
*info
;
3559 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3562 return &info
->vfs_inode
;
3565 static void shmem_destroy_callback(struct rcu_head
*head
)
3567 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3568 if (S_ISLNK(inode
->i_mode
))
3569 kfree(inode
->i_link
);
3570 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3573 static void shmem_destroy_inode(struct inode
*inode
)
3575 if (S_ISREG(inode
->i_mode
))
3576 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3577 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3580 static void shmem_init_inode(void *foo
)
3582 struct shmem_inode_info
*info
= foo
;
3583 inode_init_once(&info
->vfs_inode
);
3586 static int shmem_init_inodecache(void)
3588 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3589 sizeof(struct shmem_inode_info
),
3590 0, SLAB_PANIC
|SLAB_ACCOUNT
, shmem_init_inode
);
3594 static void shmem_destroy_inodecache(void)
3596 kmem_cache_destroy(shmem_inode_cachep
);
3599 static const struct address_space_operations shmem_aops
= {
3600 .writepage
= shmem_writepage
,
3601 .set_page_dirty
= __set_page_dirty_no_writeback
,
3603 .write_begin
= shmem_write_begin
,
3604 .write_end
= shmem_write_end
,
3606 #ifdef CONFIG_MIGRATION
3607 .migratepage
= migrate_page
,
3609 .error_remove_page
= generic_error_remove_page
,
3612 static const struct file_operations shmem_file_operations
= {
3614 .get_unmapped_area
= shmem_get_unmapped_area
,
3616 .llseek
= shmem_file_llseek
,
3617 .read_iter
= shmem_file_read_iter
,
3618 .write_iter
= generic_file_write_iter
,
3619 .fsync
= noop_fsync
,
3620 .splice_read
= shmem_file_splice_read
,
3621 .splice_write
= iter_file_splice_write
,
3622 .fallocate
= shmem_fallocate
,
3626 static const struct inode_operations shmem_inode_operations
= {
3627 .getattr
= shmem_getattr
,
3628 .setattr
= shmem_setattr
,
3629 #ifdef CONFIG_TMPFS_XATTR
3630 .setxattr
= generic_setxattr
,
3631 .getxattr
= generic_getxattr
,
3632 .listxattr
= shmem_listxattr
,
3633 .removexattr
= generic_removexattr
,
3634 .set_acl
= simple_set_acl
,
3638 static const struct inode_operations shmem_dir_inode_operations
= {
3640 .create
= shmem_create
,
3641 .lookup
= simple_lookup
,
3643 .unlink
= shmem_unlink
,
3644 .symlink
= shmem_symlink
,
3645 .mkdir
= shmem_mkdir
,
3646 .rmdir
= shmem_rmdir
,
3647 .mknod
= shmem_mknod
,
3648 .rename2
= shmem_rename2
,
3649 .tmpfile
= shmem_tmpfile
,
3651 #ifdef CONFIG_TMPFS_XATTR
3652 .setxattr
= generic_setxattr
,
3653 .getxattr
= generic_getxattr
,
3654 .listxattr
= shmem_listxattr
,
3655 .removexattr
= generic_removexattr
,
3657 #ifdef CONFIG_TMPFS_POSIX_ACL
3658 .setattr
= shmem_setattr
,
3659 .set_acl
= simple_set_acl
,
3663 static const struct inode_operations shmem_special_inode_operations
= {
3664 #ifdef CONFIG_TMPFS_XATTR
3665 .setxattr
= generic_setxattr
,
3666 .getxattr
= generic_getxattr
,
3667 .listxattr
= shmem_listxattr
,
3668 .removexattr
= generic_removexattr
,
3670 #ifdef CONFIG_TMPFS_POSIX_ACL
3671 .setattr
= shmem_setattr
,
3672 .set_acl
= simple_set_acl
,
3676 static const struct super_operations shmem_ops
= {
3677 .alloc_inode
= shmem_alloc_inode
,
3678 .destroy_inode
= shmem_destroy_inode
,
3680 .statfs
= shmem_statfs
,
3681 .remount_fs
= shmem_remount_fs
,
3682 .show_options
= shmem_show_options
,
3684 .evict_inode
= shmem_evict_inode
,
3685 .drop_inode
= generic_delete_inode
,
3686 .put_super
= shmem_put_super
,
3689 static const struct vm_operations_struct shmem_vm_ops
= {
3690 .fault
= shmem_fault
,
3691 .map_pages
= filemap_map_pages
,
3693 .set_policy
= shmem_set_policy
,
3694 .get_policy
= shmem_get_policy
,
3698 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3699 int flags
, const char *dev_name
, void *data
)
3701 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3704 static struct file_system_type shmem_fs_type
= {
3705 .owner
= THIS_MODULE
,
3707 .mount
= shmem_mount
,
3708 .kill_sb
= kill_litter_super
,
3709 .fs_flags
= FS_USERNS_MOUNT
,
3712 int __init
shmem_init(void)
3716 /* If rootfs called this, don't re-init */
3717 if (shmem_inode_cachep
)
3720 error
= shmem_init_inodecache();
3724 error
= register_filesystem(&shmem_fs_type
);
3726 pr_err("Could not register tmpfs\n");
3730 shm_mnt
= kern_mount(&shmem_fs_type
);
3731 if (IS_ERR(shm_mnt
)) {
3732 error
= PTR_ERR(shm_mnt
);
3733 pr_err("Could not kern_mount tmpfs\n");
3737 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3738 if (has_transparent_hugepage() && shmem_huge
< SHMEM_HUGE_DENY
)
3739 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3741 shmem_huge
= 0; /* just in case it was patched */
3746 unregister_filesystem(&shmem_fs_type
);
3748 shmem_destroy_inodecache();
3750 shm_mnt
= ERR_PTR(error
);
3754 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3755 static ssize_t
shmem_enabled_show(struct kobject
*kobj
,
3756 struct kobj_attribute
*attr
, char *buf
)
3760 SHMEM_HUGE_WITHIN_SIZE
,
3768 for (i
= 0, count
= 0; i
< ARRAY_SIZE(values
); i
++) {
3769 const char *fmt
= shmem_huge
== values
[i
] ? "[%s] " : "%s ";
3771 count
+= sprintf(buf
+ count
, fmt
,
3772 shmem_format_huge(values
[i
]));
3774 buf
[count
- 1] = '\n';
3778 static ssize_t
shmem_enabled_store(struct kobject
*kobj
,
3779 struct kobj_attribute
*attr
, const char *buf
, size_t count
)
3784 if (count
+ 1 > sizeof(tmp
))
3786 memcpy(tmp
, buf
, count
);
3788 if (count
&& tmp
[count
- 1] == '\n')
3789 tmp
[count
- 1] = '\0';
3791 huge
= shmem_parse_huge(tmp
);
3792 if (huge
== -EINVAL
)
3794 if (!has_transparent_hugepage() &&
3795 huge
!= SHMEM_HUGE_NEVER
&& huge
!= SHMEM_HUGE_DENY
)
3799 if (shmem_huge
< SHMEM_HUGE_DENY
)
3800 SHMEM_SB(shm_mnt
->mnt_sb
)->huge
= shmem_huge
;
3804 struct kobj_attribute shmem_enabled_attr
=
3805 __ATTR(shmem_enabled
, 0644, shmem_enabled_show
, shmem_enabled_store
);
3806 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3808 #else /* !CONFIG_SHMEM */
3811 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3813 * This is intended for small system where the benefits of the full
3814 * shmem code (swap-backed and resource-limited) are outweighed by
3815 * their complexity. On systems without swap this code should be
3816 * effectively equivalent, but much lighter weight.
3819 static struct file_system_type shmem_fs_type
= {
3821 .mount
= ramfs_mount
,
3822 .kill_sb
= kill_litter_super
,
3823 .fs_flags
= FS_USERNS_MOUNT
,
3826 int __init
shmem_init(void)
3828 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3830 shm_mnt
= kern_mount(&shmem_fs_type
);
3831 BUG_ON(IS_ERR(shm_mnt
));
3836 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3841 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3846 void shmem_unlock_mapping(struct address_space
*mapping
)
3851 unsigned long shmem_get_unmapped_area(struct file
*file
,
3852 unsigned long addr
, unsigned long len
,
3853 unsigned long pgoff
, unsigned long flags
)
3855 return current
->mm
->get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
3859 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3861 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3863 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3865 #define shmem_vm_ops generic_file_vm_ops
3866 #define shmem_file_operations ramfs_file_operations
3867 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3868 #define shmem_acct_size(flags, size) 0
3869 #define shmem_unacct_size(flags, size) do {} while (0)
3871 #endif /* CONFIG_SHMEM */
3875 static struct dentry_operations anon_ops
= {
3876 .d_dname
= simple_dname
3879 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3880 unsigned long flags
, unsigned int i_flags
)
3883 struct inode
*inode
;
3885 struct super_block
*sb
;
3888 if (IS_ERR(shm_mnt
))
3889 return ERR_CAST(shm_mnt
);
3891 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3892 return ERR_PTR(-EINVAL
);
3894 if (shmem_acct_size(flags
, size
))
3895 return ERR_PTR(-ENOMEM
);
3897 res
= ERR_PTR(-ENOMEM
);
3899 this.len
= strlen(name
);
3900 this.hash
= 0; /* will go */
3901 sb
= shm_mnt
->mnt_sb
;
3902 path
.mnt
= mntget(shm_mnt
);
3903 path
.dentry
= d_alloc_pseudo(sb
, &this);
3906 d_set_d_op(path
.dentry
, &anon_ops
);
3908 res
= ERR_PTR(-ENOSPC
);
3909 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3913 inode
->i_flags
|= i_flags
;
3914 d_instantiate(path
.dentry
, inode
);
3915 inode
->i_size
= size
;
3916 clear_nlink(inode
); /* It is unlinked */
3917 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3921 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3922 &shmem_file_operations
);
3929 shmem_unacct_size(flags
, size
);
3936 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3937 * kernel internal. There will be NO LSM permission checks against the
3938 * underlying inode. So users of this interface must do LSM checks at a
3939 * higher layer. The users are the big_key and shm implementations. LSM
3940 * checks are provided at the key or shm level rather than the inode.
3941 * @name: name for dentry (to be seen in /proc/<pid>/maps
3942 * @size: size to be set for the file
3943 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3945 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3947 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3951 * shmem_file_setup - get an unlinked file living in tmpfs
3952 * @name: name for dentry (to be seen in /proc/<pid>/maps
3953 * @size: size to be set for the file
3954 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3956 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3958 return __shmem_file_setup(name
, size
, flags
, 0);
3960 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3963 * shmem_zero_setup - setup a shared anonymous mapping
3964 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3966 int shmem_zero_setup(struct vm_area_struct
*vma
)
3969 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3972 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3973 * between XFS directory reading and selinux: since this file is only
3974 * accessible to the user through its mapping, use S_PRIVATE flag to
3975 * bypass file security, in the same way as shmem_kernel_file_setup().
3977 file
= __shmem_file_setup("dev/zero", size
, vma
->vm_flags
, S_PRIVATE
);
3979 return PTR_ERR(file
);
3983 vma
->vm_file
= file
;
3984 vma
->vm_ops
= &shmem_vm_ops
;
3989 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3990 * @mapping: the page's address_space
3991 * @index: the page index
3992 * @gfp: the page allocator flags to use if allocating
3994 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3995 * with any new page allocations done using the specified allocation flags.
3996 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3997 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3998 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4000 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4001 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4003 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
4004 pgoff_t index
, gfp_t gfp
)
4007 struct inode
*inode
= mapping
->host
;
4011 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
4012 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
,
4015 page
= ERR_PTR(error
);
4021 * The tiny !SHMEM case uses ramfs without swap
4023 return read_cache_page_gfp(mapping
, index
, gfp
);
4026 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);