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/aio.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>
76 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
77 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
79 /* Pretend that each entry is of this size in directory's i_size */
80 #define BOGO_DIRENT_SIZE 20
82 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
83 #define SHORT_SYMLINK_LEN 128
86 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
87 * inode->i_private (with i_mutex making sure that it has only one user at
88 * a time): we would prefer not to enlarge the shmem inode just for that.
91 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
92 pgoff_t start
; /* start of range currently being fallocated */
93 pgoff_t next
; /* the next page offset to be fallocated */
94 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
95 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
98 /* Flag allocation requirements to shmem_getpage */
100 SGP_READ
, /* don't exceed i_size, don't allocate page */
101 SGP_CACHE
, /* don't exceed i_size, may allocate page */
102 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
103 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
104 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
108 static unsigned long shmem_default_max_blocks(void)
110 return totalram_pages
/ 2;
113 static unsigned long shmem_default_max_inodes(void)
115 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
119 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
120 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
121 struct shmem_inode_info
*info
, pgoff_t index
);
122 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
123 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
125 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
126 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
128 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
129 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
132 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
134 return sb
->s_fs_info
;
138 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
139 * for shared memory and for shared anonymous (/dev/zero) mappings
140 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
141 * consistent with the pre-accounting of private mappings ...
143 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
145 return (flags
& VM_NORESERVE
) ?
146 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
149 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
151 if (!(flags
& VM_NORESERVE
))
152 vm_unacct_memory(VM_ACCT(size
));
155 static inline int shmem_reacct_size(unsigned long flags
,
156 loff_t oldsize
, loff_t newsize
)
158 if (!(flags
& VM_NORESERVE
)) {
159 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
160 return security_vm_enough_memory_mm(current
->mm
,
161 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
162 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
163 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
169 * ... whereas tmpfs objects are accounted incrementally as
170 * pages are allocated, in order to allow huge sparse files.
171 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
172 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
174 static inline int shmem_acct_block(unsigned long flags
)
176 return (flags
& VM_NORESERVE
) ?
177 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
180 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
182 if (flags
& VM_NORESERVE
)
183 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
186 static const struct super_operations shmem_ops
;
187 static const struct address_space_operations shmem_aops
;
188 static const struct file_operations shmem_file_operations
;
189 static const struct inode_operations shmem_inode_operations
;
190 static const struct inode_operations shmem_dir_inode_operations
;
191 static const struct inode_operations shmem_special_inode_operations
;
192 static const struct vm_operations_struct shmem_vm_ops
;
194 static LIST_HEAD(shmem_swaplist
);
195 static DEFINE_MUTEX(shmem_swaplist_mutex
);
197 static int shmem_reserve_inode(struct super_block
*sb
)
199 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
200 if (sbinfo
->max_inodes
) {
201 spin_lock(&sbinfo
->stat_lock
);
202 if (!sbinfo
->free_inodes
) {
203 spin_unlock(&sbinfo
->stat_lock
);
206 sbinfo
->free_inodes
--;
207 spin_unlock(&sbinfo
->stat_lock
);
212 static void shmem_free_inode(struct super_block
*sb
)
214 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
215 if (sbinfo
->max_inodes
) {
216 spin_lock(&sbinfo
->stat_lock
);
217 sbinfo
->free_inodes
++;
218 spin_unlock(&sbinfo
->stat_lock
);
223 * shmem_recalc_inode - recalculate the block usage of an inode
224 * @inode: inode to recalc
226 * We have to calculate the free blocks since the mm can drop
227 * undirtied hole pages behind our back.
229 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
230 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
232 * It has to be called with the spinlock held.
234 static void shmem_recalc_inode(struct inode
*inode
)
236 struct shmem_inode_info
*info
= SHMEM_I(inode
);
239 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
241 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
242 if (sbinfo
->max_blocks
)
243 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
244 info
->alloced
-= freed
;
245 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
246 shmem_unacct_blocks(info
->flags
, freed
);
251 * Replace item expected in radix tree by a new item, while holding tree lock.
253 static int shmem_radix_tree_replace(struct address_space
*mapping
,
254 pgoff_t index
, void *expected
, void *replacement
)
259 VM_BUG_ON(!expected
);
260 VM_BUG_ON(!replacement
);
261 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
264 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
265 if (item
!= expected
)
267 radix_tree_replace_slot(pslot
, replacement
);
272 * Sometimes, before we decide whether to proceed or to fail, we must check
273 * that an entry was not already brought back from swap by a racing thread.
275 * Checking page is not enough: by the time a SwapCache page is locked, it
276 * might be reused, and again be SwapCache, using the same swap as before.
278 static bool shmem_confirm_swap(struct address_space
*mapping
,
279 pgoff_t index
, swp_entry_t swap
)
284 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
286 return item
== swp_to_radix_entry(swap
);
290 * Like add_to_page_cache_locked, but error if expected item has gone.
292 static int shmem_add_to_page_cache(struct page
*page
,
293 struct address_space
*mapping
,
294 pgoff_t index
, void *expected
)
298 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
299 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
301 page_cache_get(page
);
302 page
->mapping
= mapping
;
305 spin_lock_irq(&mapping
->tree_lock
);
307 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
309 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
313 __inc_zone_page_state(page
, NR_FILE_PAGES
);
314 __inc_zone_page_state(page
, NR_SHMEM
);
315 spin_unlock_irq(&mapping
->tree_lock
);
317 page
->mapping
= NULL
;
318 spin_unlock_irq(&mapping
->tree_lock
);
319 page_cache_release(page
);
325 * Like delete_from_page_cache, but substitutes swap for page.
327 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
329 struct address_space
*mapping
= page
->mapping
;
332 spin_lock_irq(&mapping
->tree_lock
);
333 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
334 page
->mapping
= NULL
;
336 __dec_zone_page_state(page
, NR_FILE_PAGES
);
337 __dec_zone_page_state(page
, NR_SHMEM
);
338 spin_unlock_irq(&mapping
->tree_lock
);
339 page_cache_release(page
);
344 * Remove swap entry from radix tree, free the swap and its page cache.
346 static int shmem_free_swap(struct address_space
*mapping
,
347 pgoff_t index
, void *radswap
)
351 spin_lock_irq(&mapping
->tree_lock
);
352 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
353 spin_unlock_irq(&mapping
->tree_lock
);
356 free_swap_and_cache(radix_to_swp_entry(radswap
));
361 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
363 void shmem_unlock_mapping(struct address_space
*mapping
)
366 pgoff_t indices
[PAGEVEC_SIZE
];
369 pagevec_init(&pvec
, 0);
371 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
373 while (!mapping_unevictable(mapping
)) {
375 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
376 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
378 pvec
.nr
= find_get_entries(mapping
, index
,
379 PAGEVEC_SIZE
, pvec
.pages
, indices
);
382 index
= indices
[pvec
.nr
- 1] + 1;
383 pagevec_remove_exceptionals(&pvec
);
384 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
385 pagevec_release(&pvec
);
391 * Remove range of pages and swap entries from radix tree, and free them.
392 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
394 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
397 struct address_space
*mapping
= inode
->i_mapping
;
398 struct shmem_inode_info
*info
= SHMEM_I(inode
);
399 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
400 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
401 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
402 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
404 pgoff_t indices
[PAGEVEC_SIZE
];
405 long nr_swaps_freed
= 0;
410 end
= -1; /* unsigned, so actually very big */
412 pagevec_init(&pvec
, 0);
414 while (index
< end
) {
415 pvec
.nr
= find_get_entries(mapping
, index
,
416 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
417 pvec
.pages
, indices
);
420 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
421 struct page
*page
= pvec
.pages
[i
];
427 if (radix_tree_exceptional_entry(page
)) {
430 nr_swaps_freed
+= !shmem_free_swap(mapping
,
435 if (!trylock_page(page
))
437 if (!unfalloc
|| !PageUptodate(page
)) {
438 if (page
->mapping
== mapping
) {
439 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
440 truncate_inode_page(mapping
, page
);
445 pagevec_remove_exceptionals(&pvec
);
446 pagevec_release(&pvec
);
452 struct page
*page
= NULL
;
453 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
455 unsigned int top
= PAGE_CACHE_SIZE
;
460 zero_user_segment(page
, partial_start
, top
);
461 set_page_dirty(page
);
463 page_cache_release(page
);
467 struct page
*page
= NULL
;
468 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
470 zero_user_segment(page
, 0, partial_end
);
471 set_page_dirty(page
);
473 page_cache_release(page
);
480 while (index
< end
) {
483 pvec
.nr
= find_get_entries(mapping
, index
,
484 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
485 pvec
.pages
, indices
);
487 /* If all gone or hole-punch or unfalloc, we're done */
488 if (index
== start
|| end
!= -1)
490 /* But if truncating, restart to make sure all gone */
494 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
495 struct page
*page
= pvec
.pages
[i
];
501 if (radix_tree_exceptional_entry(page
)) {
504 if (shmem_free_swap(mapping
, index
, page
)) {
505 /* Swap was replaced by page: retry */
514 if (!unfalloc
|| !PageUptodate(page
)) {
515 if (page
->mapping
== mapping
) {
516 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
517 truncate_inode_page(mapping
, page
);
519 /* Page was replaced by swap: retry */
527 pagevec_remove_exceptionals(&pvec
);
528 pagevec_release(&pvec
);
532 spin_lock(&info
->lock
);
533 info
->swapped
-= nr_swaps_freed
;
534 shmem_recalc_inode(inode
);
535 spin_unlock(&info
->lock
);
538 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
540 shmem_undo_range(inode
, lstart
, lend
, false);
541 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
543 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
545 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
547 struct inode
*inode
= dentry
->d_inode
;
548 struct shmem_inode_info
*info
= SHMEM_I(inode
);
551 error
= inode_change_ok(inode
, attr
);
555 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
556 loff_t oldsize
= inode
->i_size
;
557 loff_t newsize
= attr
->ia_size
;
559 /* protected by i_mutex */
560 if ((newsize
< oldsize
&& (info
->seals
& F_SEAL_SHRINK
)) ||
561 (newsize
> oldsize
&& (info
->seals
& F_SEAL_GROW
)))
564 if (newsize
!= oldsize
) {
565 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
569 i_size_write(inode
, newsize
);
570 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
572 if (newsize
< oldsize
) {
573 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
574 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
575 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
576 /* unmap again to remove racily COWed private pages */
577 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
581 setattr_copy(inode
, attr
);
582 if (attr
->ia_valid
& ATTR_MODE
)
583 error
= posix_acl_chmod(inode
, inode
->i_mode
);
587 static void shmem_evict_inode(struct inode
*inode
)
589 struct shmem_inode_info
*info
= SHMEM_I(inode
);
591 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
592 shmem_unacct_size(info
->flags
, inode
->i_size
);
594 shmem_truncate_range(inode
, 0, (loff_t
)-1);
595 if (!list_empty(&info
->swaplist
)) {
596 mutex_lock(&shmem_swaplist_mutex
);
597 list_del_init(&info
->swaplist
);
598 mutex_unlock(&shmem_swaplist_mutex
);
601 kfree(info
->symlink
);
603 simple_xattrs_free(&info
->xattrs
);
604 WARN_ON(inode
->i_blocks
);
605 shmem_free_inode(inode
->i_sb
);
610 * If swap found in inode, free it and move page from swapcache to filecache.
612 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
613 swp_entry_t swap
, struct page
**pagep
)
615 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
621 radswap
= swp_to_radix_entry(swap
);
622 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
624 return -EAGAIN
; /* tell shmem_unuse we found nothing */
627 * Move _head_ to start search for next from here.
628 * But be careful: shmem_evict_inode checks list_empty without taking
629 * mutex, and there's an instant in list_move_tail when info->swaplist
630 * would appear empty, if it were the only one on shmem_swaplist.
632 if (shmem_swaplist
.next
!= &info
->swaplist
)
633 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
635 gfp
= mapping_gfp_mask(mapping
);
636 if (shmem_should_replace_page(*pagep
, gfp
)) {
637 mutex_unlock(&shmem_swaplist_mutex
);
638 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
639 mutex_lock(&shmem_swaplist_mutex
);
641 * We needed to drop mutex to make that restrictive page
642 * allocation, but the inode might have been freed while we
643 * dropped it: although a racing shmem_evict_inode() cannot
644 * complete without emptying the radix_tree, our page lock
645 * on this swapcache page is not enough to prevent that -
646 * free_swap_and_cache() of our swap entry will only
647 * trylock_page(), removing swap from radix_tree whatever.
649 * We must not proceed to shmem_add_to_page_cache() if the
650 * inode has been freed, but of course we cannot rely on
651 * inode or mapping or info to check that. However, we can
652 * safely check if our swap entry is still in use (and here
653 * it can't have got reused for another page): if it's still
654 * in use, then the inode cannot have been freed yet, and we
655 * can safely proceed (if it's no longer in use, that tells
656 * nothing about the inode, but we don't need to unuse swap).
658 if (!page_swapcount(*pagep
))
663 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
664 * but also to hold up shmem_evict_inode(): so inode cannot be freed
665 * beneath us (pagelock doesn't help until the page is in pagecache).
668 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
670 if (error
!= -ENOMEM
) {
672 * Truncation and eviction use free_swap_and_cache(), which
673 * only does trylock page: if we raced, best clean up here.
675 delete_from_swap_cache(*pagep
);
676 set_page_dirty(*pagep
);
678 spin_lock(&info
->lock
);
680 spin_unlock(&info
->lock
);
688 * Search through swapped inodes to find and replace swap by page.
690 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
692 struct list_head
*this, *next
;
693 struct shmem_inode_info
*info
;
694 struct mem_cgroup
*memcg
;
698 * There's a faint possibility that swap page was replaced before
699 * caller locked it: caller will come back later with the right page.
701 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
705 * Charge page using GFP_KERNEL while we can wait, before taking
706 * the shmem_swaplist_mutex which might hold up shmem_writepage().
707 * Charged back to the user (not to caller) when swap account is used.
709 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
);
712 /* No radix_tree_preload: swap entry keeps a place for page in tree */
715 mutex_lock(&shmem_swaplist_mutex
);
716 list_for_each_safe(this, next
, &shmem_swaplist
) {
717 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
719 error
= shmem_unuse_inode(info
, swap
, &page
);
721 list_del_init(&info
->swaplist
);
723 if (error
!= -EAGAIN
)
725 /* found nothing in this: move on to search the next */
727 mutex_unlock(&shmem_swaplist_mutex
);
730 if (error
!= -ENOMEM
)
732 mem_cgroup_cancel_charge(page
, memcg
);
734 mem_cgroup_commit_charge(page
, memcg
, true);
737 page_cache_release(page
);
742 * Move the page from the page cache to the swap cache.
744 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
746 struct shmem_inode_info
*info
;
747 struct address_space
*mapping
;
752 BUG_ON(!PageLocked(page
));
753 mapping
= page
->mapping
;
755 inode
= mapping
->host
;
756 info
= SHMEM_I(inode
);
757 if (info
->flags
& VM_LOCKED
)
759 if (!total_swap_pages
)
763 * Our capabilities prevent regular writeback or sync from ever calling
764 * shmem_writepage; but a stacking filesystem might use ->writepage of
765 * its underlying filesystem, in which case tmpfs should write out to
766 * swap only in response to memory pressure, and not for the writeback
769 if (!wbc
->for_reclaim
) {
770 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
775 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
776 * value into swapfile.c, the only way we can correctly account for a
777 * fallocated page arriving here is now to initialize it and write it.
779 * That's okay for a page already fallocated earlier, but if we have
780 * not yet completed the fallocation, then (a) we want to keep track
781 * of this page in case we have to undo it, and (b) it may not be a
782 * good idea to continue anyway, once we're pushing into swap. So
783 * reactivate the page, and let shmem_fallocate() quit when too many.
785 if (!PageUptodate(page
)) {
786 if (inode
->i_private
) {
787 struct shmem_falloc
*shmem_falloc
;
788 spin_lock(&inode
->i_lock
);
789 shmem_falloc
= inode
->i_private
;
791 !shmem_falloc
->waitq
&&
792 index
>= shmem_falloc
->start
&&
793 index
< shmem_falloc
->next
)
794 shmem_falloc
->nr_unswapped
++;
797 spin_unlock(&inode
->i_lock
);
801 clear_highpage(page
);
802 flush_dcache_page(page
);
803 SetPageUptodate(page
);
806 swap
= get_swap_page();
811 * Add inode to shmem_unuse()'s list of swapped-out inodes,
812 * if it's not already there. Do it now before the page is
813 * moved to swap cache, when its pagelock no longer protects
814 * the inode from eviction. But don't unlock the mutex until
815 * we've incremented swapped, because shmem_unuse_inode() will
816 * prune a !swapped inode from the swaplist under this mutex.
818 mutex_lock(&shmem_swaplist_mutex
);
819 if (list_empty(&info
->swaplist
))
820 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
822 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
823 swap_shmem_alloc(swap
);
824 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
826 spin_lock(&info
->lock
);
828 shmem_recalc_inode(inode
);
829 spin_unlock(&info
->lock
);
831 mutex_unlock(&shmem_swaplist_mutex
);
832 BUG_ON(page_mapped(page
));
833 swap_writepage(page
, wbc
);
837 mutex_unlock(&shmem_swaplist_mutex
);
838 swapcache_free(swap
);
840 set_page_dirty(page
);
841 if (wbc
->for_reclaim
)
842 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
849 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
853 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
854 return; /* show nothing */
856 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
858 seq_printf(seq
, ",mpol=%s", buffer
);
861 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
863 struct mempolicy
*mpol
= NULL
;
865 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
868 spin_unlock(&sbinfo
->stat_lock
);
872 #endif /* CONFIG_TMPFS */
874 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
875 struct shmem_inode_info
*info
, pgoff_t index
)
877 struct vm_area_struct pvma
;
880 /* Create a pseudo vma that just contains the policy */
882 /* Bias interleave by inode number to distribute better across nodes */
883 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
885 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
887 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
889 /* Drop reference taken by mpol_shared_policy_lookup() */
890 mpol_cond_put(pvma
.vm_policy
);
895 static struct page
*shmem_alloc_page(gfp_t gfp
,
896 struct shmem_inode_info
*info
, pgoff_t index
)
898 struct vm_area_struct pvma
;
901 /* Create a pseudo vma that just contains the policy */
903 /* Bias interleave by inode number to distribute better across nodes */
904 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
906 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
908 page
= alloc_page_vma(gfp
, &pvma
, 0);
910 /* Drop reference taken by mpol_shared_policy_lookup() */
911 mpol_cond_put(pvma
.vm_policy
);
915 #else /* !CONFIG_NUMA */
917 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
920 #endif /* CONFIG_TMPFS */
922 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
923 struct shmem_inode_info
*info
, pgoff_t index
)
925 return swapin_readahead(swap
, gfp
, NULL
, 0);
928 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
929 struct shmem_inode_info
*info
, pgoff_t index
)
931 return alloc_page(gfp
);
933 #endif /* CONFIG_NUMA */
935 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
936 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
943 * When a page is moved from swapcache to shmem filecache (either by the
944 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
945 * shmem_unuse_inode()), it may have been read in earlier from swap, in
946 * ignorance of the mapping it belongs to. If that mapping has special
947 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
948 * we may need to copy to a suitable page before moving to filecache.
950 * In a future release, this may well be extended to respect cpuset and
951 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
952 * but for now it is a simple matter of zone.
954 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
956 return page_zonenum(page
) > gfp_zone(gfp
);
959 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
960 struct shmem_inode_info
*info
, pgoff_t index
)
962 struct page
*oldpage
, *newpage
;
963 struct address_space
*swap_mapping
;
968 swap_index
= page_private(oldpage
);
969 swap_mapping
= page_mapping(oldpage
);
972 * We have arrived here because our zones are constrained, so don't
973 * limit chance of success by further cpuset and node constraints.
975 gfp
&= ~GFP_CONSTRAINT_MASK
;
976 newpage
= shmem_alloc_page(gfp
, info
, index
);
980 page_cache_get(newpage
);
981 copy_highpage(newpage
, oldpage
);
982 flush_dcache_page(newpage
);
984 __set_page_locked(newpage
);
985 SetPageUptodate(newpage
);
986 SetPageSwapBacked(newpage
);
987 set_page_private(newpage
, swap_index
);
988 SetPageSwapCache(newpage
);
991 * Our caller will very soon move newpage out of swapcache, but it's
992 * a nice clean interface for us to replace oldpage by newpage there.
994 spin_lock_irq(&swap_mapping
->tree_lock
);
995 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
998 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
999 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1001 spin_unlock_irq(&swap_mapping
->tree_lock
);
1003 if (unlikely(error
)) {
1005 * Is this possible? I think not, now that our callers check
1006 * both PageSwapCache and page_private after getting page lock;
1007 * but be defensive. Reverse old to newpage for clear and free.
1011 mem_cgroup_migrate(oldpage
, newpage
, true);
1012 lru_cache_add_anon(newpage
);
1016 ClearPageSwapCache(oldpage
);
1017 set_page_private(oldpage
, 0);
1019 unlock_page(oldpage
);
1020 page_cache_release(oldpage
);
1021 page_cache_release(oldpage
);
1026 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1028 * If we allocate a new one we do not mark it dirty. That's up to the
1029 * vm. If we swap it in we mark it dirty since we also free the swap
1030 * entry since a page cannot live in both the swap and page cache
1032 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1033 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1035 struct address_space
*mapping
= inode
->i_mapping
;
1036 struct shmem_inode_info
*info
;
1037 struct shmem_sb_info
*sbinfo
;
1038 struct mem_cgroup
*memcg
;
1045 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1049 page
= find_lock_entry(mapping
, index
);
1050 if (radix_tree_exceptional_entry(page
)) {
1051 swap
= radix_to_swp_entry(page
);
1055 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1056 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1061 if (page
&& sgp
== SGP_WRITE
)
1062 mark_page_accessed(page
);
1064 /* fallocated page? */
1065 if (page
&& !PageUptodate(page
)) {
1066 if (sgp
!= SGP_READ
)
1069 page_cache_release(page
);
1072 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1078 * Fast cache lookup did not find it:
1079 * bring it back from swap or allocate.
1081 info
= SHMEM_I(inode
);
1082 sbinfo
= SHMEM_SB(inode
->i_sb
);
1085 /* Look it up and read it in.. */
1086 page
= lookup_swap_cache(swap
);
1088 /* here we actually do the io */
1090 *fault_type
|= VM_FAULT_MAJOR
;
1091 page
= shmem_swapin(swap
, gfp
, info
, index
);
1098 /* We have to do this with page locked to prevent races */
1100 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1101 !shmem_confirm_swap(mapping
, index
, swap
)) {
1102 error
= -EEXIST
; /* try again */
1105 if (!PageUptodate(page
)) {
1109 wait_on_page_writeback(page
);
1111 if (shmem_should_replace_page(page
, gfp
)) {
1112 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1117 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1119 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1120 swp_to_radix_entry(swap
));
1122 * We already confirmed swap under page lock, and make
1123 * no memory allocation here, so usually no possibility
1124 * of error; but free_swap_and_cache() only trylocks a
1125 * page, so it is just possible that the entry has been
1126 * truncated or holepunched since swap was confirmed.
1127 * shmem_undo_range() will have done some of the
1128 * unaccounting, now delete_from_swap_cache() will do
1130 * Reset swap.val? No, leave it so "failed" goes back to
1131 * "repeat": reading a hole and writing should succeed.
1134 mem_cgroup_cancel_charge(page
, memcg
);
1135 delete_from_swap_cache(page
);
1141 mem_cgroup_commit_charge(page
, memcg
, true);
1143 spin_lock(&info
->lock
);
1145 shmem_recalc_inode(inode
);
1146 spin_unlock(&info
->lock
);
1148 if (sgp
== SGP_WRITE
)
1149 mark_page_accessed(page
);
1151 delete_from_swap_cache(page
);
1152 set_page_dirty(page
);
1156 if (shmem_acct_block(info
->flags
)) {
1160 if (sbinfo
->max_blocks
) {
1161 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1162 sbinfo
->max_blocks
) >= 0) {
1166 percpu_counter_inc(&sbinfo
->used_blocks
);
1169 page
= shmem_alloc_page(gfp
, info
, index
);
1175 __SetPageSwapBacked(page
);
1176 __set_page_locked(page
);
1177 if (sgp
== SGP_WRITE
)
1178 __SetPageReferenced(page
);
1180 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1183 error
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
1185 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1187 radix_tree_preload_end();
1190 mem_cgroup_cancel_charge(page
, memcg
);
1193 mem_cgroup_commit_charge(page
, memcg
, false);
1194 lru_cache_add_anon(page
);
1196 spin_lock(&info
->lock
);
1198 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1199 shmem_recalc_inode(inode
);
1200 spin_unlock(&info
->lock
);
1204 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1206 if (sgp
== SGP_FALLOC
)
1210 * Let SGP_WRITE caller clear ends if write does not fill page;
1211 * but SGP_FALLOC on a page fallocated earlier must initialize
1212 * it now, lest undo on failure cancel our earlier guarantee.
1214 if (sgp
!= SGP_WRITE
) {
1215 clear_highpage(page
);
1216 flush_dcache_page(page
);
1217 SetPageUptodate(page
);
1219 if (sgp
== SGP_DIRTY
)
1220 set_page_dirty(page
);
1223 /* Perhaps the file has been truncated since we checked */
1224 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1225 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1239 info
= SHMEM_I(inode
);
1240 ClearPageDirty(page
);
1241 delete_from_page_cache(page
);
1242 spin_lock(&info
->lock
);
1244 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1245 spin_unlock(&info
->lock
);
1247 sbinfo
= SHMEM_SB(inode
->i_sb
);
1248 if (sbinfo
->max_blocks
)
1249 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1251 shmem_unacct_blocks(info
->flags
, 1);
1253 if (swap
.val
&& error
!= -EINVAL
&&
1254 !shmem_confirm_swap(mapping
, index
, swap
))
1259 page_cache_release(page
);
1261 if (error
== -ENOSPC
&& !once
++) {
1262 info
= SHMEM_I(inode
);
1263 spin_lock(&info
->lock
);
1264 shmem_recalc_inode(inode
);
1265 spin_unlock(&info
->lock
);
1268 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1273 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1275 struct inode
*inode
= file_inode(vma
->vm_file
);
1277 int ret
= VM_FAULT_LOCKED
;
1280 * Trinity finds that probing a hole which tmpfs is punching can
1281 * prevent the hole-punch from ever completing: which in turn
1282 * locks writers out with its hold on i_mutex. So refrain from
1283 * faulting pages into the hole while it's being punched. Although
1284 * shmem_undo_range() does remove the additions, it may be unable to
1285 * keep up, as each new page needs its own unmap_mapping_range() call,
1286 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1288 * It does not matter if we sometimes reach this check just before the
1289 * hole-punch begins, so that one fault then races with the punch:
1290 * we just need to make racing faults a rare case.
1292 * The implementation below would be much simpler if we just used a
1293 * standard mutex or completion: but we cannot take i_mutex in fault,
1294 * and bloating every shmem inode for this unlikely case would be sad.
1296 if (unlikely(inode
->i_private
)) {
1297 struct shmem_falloc
*shmem_falloc
;
1299 spin_lock(&inode
->i_lock
);
1300 shmem_falloc
= inode
->i_private
;
1302 shmem_falloc
->waitq
&&
1303 vmf
->pgoff
>= shmem_falloc
->start
&&
1304 vmf
->pgoff
< shmem_falloc
->next
) {
1305 wait_queue_head_t
*shmem_falloc_waitq
;
1306 DEFINE_WAIT(shmem_fault_wait
);
1308 ret
= VM_FAULT_NOPAGE
;
1309 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1310 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1311 /* It's polite to up mmap_sem if we can */
1312 up_read(&vma
->vm_mm
->mmap_sem
);
1313 ret
= VM_FAULT_RETRY
;
1316 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1317 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1318 TASK_UNINTERRUPTIBLE
);
1319 spin_unlock(&inode
->i_lock
);
1323 * shmem_falloc_waitq points into the shmem_fallocate()
1324 * stack of the hole-punching task: shmem_falloc_waitq
1325 * is usually invalid by the time we reach here, but
1326 * finish_wait() does not dereference it in that case;
1327 * though i_lock needed lest racing with wake_up_all().
1329 spin_lock(&inode
->i_lock
);
1330 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1331 spin_unlock(&inode
->i_lock
);
1334 spin_unlock(&inode
->i_lock
);
1337 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1339 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1341 if (ret
& VM_FAULT_MAJOR
) {
1342 count_vm_event(PGMAJFAULT
);
1343 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1349 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1351 struct inode
*inode
= file_inode(vma
->vm_file
);
1352 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1355 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1358 struct inode
*inode
= file_inode(vma
->vm_file
);
1361 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1362 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1366 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1368 struct inode
*inode
= file_inode(file
);
1369 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1370 int retval
= -ENOMEM
;
1372 spin_lock(&info
->lock
);
1373 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1374 if (!user_shm_lock(inode
->i_size
, user
))
1376 info
->flags
|= VM_LOCKED
;
1377 mapping_set_unevictable(file
->f_mapping
);
1379 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1380 user_shm_unlock(inode
->i_size
, user
);
1381 info
->flags
&= ~VM_LOCKED
;
1382 mapping_clear_unevictable(file
->f_mapping
);
1387 spin_unlock(&info
->lock
);
1391 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1393 file_accessed(file
);
1394 vma
->vm_ops
= &shmem_vm_ops
;
1398 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1399 umode_t mode
, dev_t dev
, unsigned long flags
)
1401 struct inode
*inode
;
1402 struct shmem_inode_info
*info
;
1403 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1405 if (shmem_reserve_inode(sb
))
1408 inode
= new_inode(sb
);
1410 inode
->i_ino
= get_next_ino();
1411 inode_init_owner(inode
, dir
, mode
);
1412 inode
->i_blocks
= 0;
1413 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1414 inode
->i_generation
= get_seconds();
1415 info
= SHMEM_I(inode
);
1416 memset(info
, 0, (char *)inode
- (char *)info
);
1417 spin_lock_init(&info
->lock
);
1418 info
->seals
= F_SEAL_SEAL
;
1419 info
->flags
= flags
& VM_NORESERVE
;
1420 INIT_LIST_HEAD(&info
->swaplist
);
1421 simple_xattrs_init(&info
->xattrs
);
1422 cache_no_acl(inode
);
1424 switch (mode
& S_IFMT
) {
1426 inode
->i_op
= &shmem_special_inode_operations
;
1427 init_special_inode(inode
, mode
, dev
);
1430 inode
->i_mapping
->a_ops
= &shmem_aops
;
1431 inode
->i_op
= &shmem_inode_operations
;
1432 inode
->i_fop
= &shmem_file_operations
;
1433 mpol_shared_policy_init(&info
->policy
,
1434 shmem_get_sbmpol(sbinfo
));
1438 /* Some things misbehave if size == 0 on a directory */
1439 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1440 inode
->i_op
= &shmem_dir_inode_operations
;
1441 inode
->i_fop
= &simple_dir_operations
;
1445 * Must not load anything in the rbtree,
1446 * mpol_free_shared_policy will not be called.
1448 mpol_shared_policy_init(&info
->policy
, NULL
);
1452 shmem_free_inode(sb
);
1456 bool shmem_mapping(struct address_space
*mapping
)
1458 return mapping
->host
->i_sb
->s_op
== &shmem_ops
;
1462 static const struct inode_operations shmem_symlink_inode_operations
;
1463 static const struct inode_operations shmem_short_symlink_operations
;
1465 #ifdef CONFIG_TMPFS_XATTR
1466 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1468 #define shmem_initxattrs NULL
1472 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1473 loff_t pos
, unsigned len
, unsigned flags
,
1474 struct page
**pagep
, void **fsdata
)
1476 struct inode
*inode
= mapping
->host
;
1477 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1478 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1480 /* i_mutex is held by caller */
1481 if (unlikely(info
->seals
)) {
1482 if (info
->seals
& F_SEAL_WRITE
)
1484 if ((info
->seals
& F_SEAL_GROW
) && pos
+ len
> inode
->i_size
)
1488 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1492 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1493 loff_t pos
, unsigned len
, unsigned copied
,
1494 struct page
*page
, void *fsdata
)
1496 struct inode
*inode
= mapping
->host
;
1498 if (pos
+ copied
> inode
->i_size
)
1499 i_size_write(inode
, pos
+ copied
);
1501 if (!PageUptodate(page
)) {
1502 if (copied
< PAGE_CACHE_SIZE
) {
1503 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1504 zero_user_segments(page
, 0, from
,
1505 from
+ copied
, PAGE_CACHE_SIZE
);
1507 SetPageUptodate(page
);
1509 set_page_dirty(page
);
1511 page_cache_release(page
);
1516 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1518 struct file
*file
= iocb
->ki_filp
;
1519 struct inode
*inode
= file_inode(file
);
1520 struct address_space
*mapping
= inode
->i_mapping
;
1522 unsigned long offset
;
1523 enum sgp_type sgp
= SGP_READ
;
1526 loff_t
*ppos
= &iocb
->ki_pos
;
1529 * Might this read be for a stacking filesystem? Then when reading
1530 * holes of a sparse file, we actually need to allocate those pages,
1531 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1533 if (!iter_is_iovec(to
))
1536 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1537 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1540 struct page
*page
= NULL
;
1542 unsigned long nr
, ret
;
1543 loff_t i_size
= i_size_read(inode
);
1545 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1546 if (index
> end_index
)
1548 if (index
== end_index
) {
1549 nr
= i_size
& ~PAGE_CACHE_MASK
;
1554 error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1556 if (error
== -EINVAL
)
1564 * We must evaluate after, since reads (unlike writes)
1565 * are called without i_mutex protection against truncate
1567 nr
= PAGE_CACHE_SIZE
;
1568 i_size
= i_size_read(inode
);
1569 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1570 if (index
== end_index
) {
1571 nr
= i_size
& ~PAGE_CACHE_MASK
;
1574 page_cache_release(page
);
1582 * If users can be writing to this page using arbitrary
1583 * virtual addresses, take care about potential aliasing
1584 * before reading the page on the kernel side.
1586 if (mapping_writably_mapped(mapping
))
1587 flush_dcache_page(page
);
1589 * Mark the page accessed if we read the beginning.
1592 mark_page_accessed(page
);
1594 page
= ZERO_PAGE(0);
1595 page_cache_get(page
);
1599 * Ok, we have the page, and it's up-to-date, so
1600 * now we can copy it to user space...
1602 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
1605 index
+= offset
>> PAGE_CACHE_SHIFT
;
1606 offset
&= ~PAGE_CACHE_MASK
;
1608 page_cache_release(page
);
1609 if (!iov_iter_count(to
))
1618 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1619 file_accessed(file
);
1620 return retval
? retval
: error
;
1623 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1624 struct pipe_inode_info
*pipe
, size_t len
,
1627 struct address_space
*mapping
= in
->f_mapping
;
1628 struct inode
*inode
= mapping
->host
;
1629 unsigned int loff
, nr_pages
, req_pages
;
1630 struct page
*pages
[PIPE_DEF_BUFFERS
];
1631 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1633 pgoff_t index
, end_index
;
1636 struct splice_pipe_desc spd
= {
1639 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1641 .ops
= &page_cache_pipe_buf_ops
,
1642 .spd_release
= spd_release_page
,
1645 isize
= i_size_read(inode
);
1646 if (unlikely(*ppos
>= isize
))
1649 left
= isize
- *ppos
;
1650 if (unlikely(left
< len
))
1653 if (splice_grow_spd(pipe
, &spd
))
1656 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1657 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1658 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1659 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
1661 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1662 nr_pages
, spd
.pages
);
1663 index
+= spd
.nr_pages
;
1666 while (spd
.nr_pages
< nr_pages
) {
1667 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1671 spd
.pages
[spd
.nr_pages
++] = page
;
1675 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1676 nr_pages
= spd
.nr_pages
;
1679 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1680 unsigned int this_len
;
1685 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1686 page
= spd
.pages
[page_nr
];
1688 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1689 error
= shmem_getpage(inode
, index
, &page
,
1694 page_cache_release(spd
.pages
[page_nr
]);
1695 spd
.pages
[page_nr
] = page
;
1698 isize
= i_size_read(inode
);
1699 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1700 if (unlikely(!isize
|| index
> end_index
))
1703 if (end_index
== index
) {
1706 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1710 this_len
= min(this_len
, plen
- loff
);
1714 spd
.partial
[page_nr
].offset
= loff
;
1715 spd
.partial
[page_nr
].len
= this_len
;
1722 while (page_nr
< nr_pages
)
1723 page_cache_release(spd
.pages
[page_nr
++]);
1726 error
= splice_to_pipe(pipe
, &spd
);
1728 splice_shrink_spd(&spd
);
1738 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1740 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1741 pgoff_t index
, pgoff_t end
, int whence
)
1744 struct pagevec pvec
;
1745 pgoff_t indices
[PAGEVEC_SIZE
];
1749 pagevec_init(&pvec
, 0);
1750 pvec
.nr
= 1; /* start small: we may be there already */
1752 pvec
.nr
= find_get_entries(mapping
, index
,
1753 pvec
.nr
, pvec
.pages
, indices
);
1755 if (whence
== SEEK_DATA
)
1759 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1760 if (index
< indices
[i
]) {
1761 if (whence
== SEEK_HOLE
) {
1767 page
= pvec
.pages
[i
];
1768 if (page
&& !radix_tree_exceptional_entry(page
)) {
1769 if (!PageUptodate(page
))
1773 (page
&& whence
== SEEK_DATA
) ||
1774 (!page
&& whence
== SEEK_HOLE
)) {
1779 pagevec_remove_exceptionals(&pvec
);
1780 pagevec_release(&pvec
);
1781 pvec
.nr
= PAGEVEC_SIZE
;
1787 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1789 struct address_space
*mapping
= file
->f_mapping
;
1790 struct inode
*inode
= mapping
->host
;
1794 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1795 return generic_file_llseek_size(file
, offset
, whence
,
1796 MAX_LFS_FILESIZE
, i_size_read(inode
));
1797 mutex_lock(&inode
->i_mutex
);
1798 /* We're holding i_mutex so we can access i_size directly */
1802 else if (offset
>= inode
->i_size
)
1805 start
= offset
>> PAGE_CACHE_SHIFT
;
1806 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1807 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1808 new_offset
<<= PAGE_CACHE_SHIFT
;
1809 if (new_offset
> offset
) {
1810 if (new_offset
< inode
->i_size
)
1811 offset
= new_offset
;
1812 else if (whence
== SEEK_DATA
)
1815 offset
= inode
->i_size
;
1820 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
1821 mutex_unlock(&inode
->i_mutex
);
1826 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1827 * so reuse a tag which we firmly believe is never set or cleared on shmem.
1829 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
1830 #define LAST_SCAN 4 /* about 150ms max */
1832 static void shmem_tag_pins(struct address_space
*mapping
)
1834 struct radix_tree_iter iter
;
1844 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
1845 page
= radix_tree_deref_slot(slot
);
1846 if (!page
|| radix_tree_exception(page
)) {
1847 if (radix_tree_deref_retry(page
))
1849 } else if (page_count(page
) - page_mapcount(page
) > 1) {
1850 spin_lock_irq(&mapping
->tree_lock
);
1851 radix_tree_tag_set(&mapping
->page_tree
, iter
.index
,
1853 spin_unlock_irq(&mapping
->tree_lock
);
1856 if (need_resched()) {
1858 start
= iter
.index
+ 1;
1866 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1867 * via get_user_pages(), drivers might have some pending I/O without any active
1868 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1869 * and see whether it has an elevated ref-count. If so, we tag them and wait for
1870 * them to be dropped.
1871 * The caller must guarantee that no new user will acquire writable references
1872 * to those pages to avoid races.
1874 static int shmem_wait_for_pins(struct address_space
*mapping
)
1876 struct radix_tree_iter iter
;
1882 shmem_tag_pins(mapping
);
1885 for (scan
= 0; scan
<= LAST_SCAN
; scan
++) {
1886 if (!radix_tree_tagged(&mapping
->page_tree
, SHMEM_TAG_PINNED
))
1890 lru_add_drain_all();
1891 else if (schedule_timeout_killable((HZ
<< scan
) / 200))
1897 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
,
1898 start
, SHMEM_TAG_PINNED
) {
1900 page
= radix_tree_deref_slot(slot
);
1901 if (radix_tree_exception(page
)) {
1902 if (radix_tree_deref_retry(page
))
1909 page_count(page
) - page_mapcount(page
) != 1) {
1910 if (scan
< LAST_SCAN
)
1911 goto continue_resched
;
1914 * On the last scan, we clean up all those tags
1915 * we inserted; but make a note that we still
1916 * found pages pinned.
1921 spin_lock_irq(&mapping
->tree_lock
);
1922 radix_tree_tag_clear(&mapping
->page_tree
,
1923 iter
.index
, SHMEM_TAG_PINNED
);
1924 spin_unlock_irq(&mapping
->tree_lock
);
1926 if (need_resched()) {
1928 start
= iter
.index
+ 1;
1938 #define F_ALL_SEALS (F_SEAL_SEAL | \
1943 int shmem_add_seals(struct file
*file
, unsigned int seals
)
1945 struct inode
*inode
= file_inode(file
);
1946 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1951 * Sealing allows multiple parties to share a shmem-file but restrict
1952 * access to a specific subset of file operations. Seals can only be
1953 * added, but never removed. This way, mutually untrusted parties can
1954 * share common memory regions with a well-defined policy. A malicious
1955 * peer can thus never perform unwanted operations on a shared object.
1957 * Seals are only supported on special shmem-files and always affect
1958 * the whole underlying inode. Once a seal is set, it may prevent some
1959 * kinds of access to the file. Currently, the following seals are
1961 * SEAL_SEAL: Prevent further seals from being set on this file
1962 * SEAL_SHRINK: Prevent the file from shrinking
1963 * SEAL_GROW: Prevent the file from growing
1964 * SEAL_WRITE: Prevent write access to the file
1966 * As we don't require any trust relationship between two parties, we
1967 * must prevent seals from being removed. Therefore, sealing a file
1968 * only adds a given set of seals to the file, it never touches
1969 * existing seals. Furthermore, the "setting seals"-operation can be
1970 * sealed itself, which basically prevents any further seal from being
1973 * Semantics of sealing are only defined on volatile files. Only
1974 * anonymous shmem files support sealing. More importantly, seals are
1975 * never written to disk. Therefore, there's no plan to support it on
1979 if (file
->f_op
!= &shmem_file_operations
)
1981 if (!(file
->f_mode
& FMODE_WRITE
))
1983 if (seals
& ~(unsigned int)F_ALL_SEALS
)
1986 mutex_lock(&inode
->i_mutex
);
1988 if (info
->seals
& F_SEAL_SEAL
) {
1993 if ((seals
& F_SEAL_WRITE
) && !(info
->seals
& F_SEAL_WRITE
)) {
1994 error
= mapping_deny_writable(file
->f_mapping
);
1998 error
= shmem_wait_for_pins(file
->f_mapping
);
2000 mapping_allow_writable(file
->f_mapping
);
2005 info
->seals
|= seals
;
2009 mutex_unlock(&inode
->i_mutex
);
2012 EXPORT_SYMBOL_GPL(shmem_add_seals
);
2014 int shmem_get_seals(struct file
*file
)
2016 if (file
->f_op
!= &shmem_file_operations
)
2019 return SHMEM_I(file_inode(file
))->seals
;
2021 EXPORT_SYMBOL_GPL(shmem_get_seals
);
2023 long shmem_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
2029 /* disallow upper 32bit */
2033 error
= shmem_add_seals(file
, arg
);
2036 error
= shmem_get_seals(file
);
2046 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
2049 struct inode
*inode
= file_inode(file
);
2050 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
2051 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2052 struct shmem_falloc shmem_falloc
;
2053 pgoff_t start
, index
, end
;
2056 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2059 mutex_lock(&inode
->i_mutex
);
2061 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
2062 struct address_space
*mapping
= file
->f_mapping
;
2063 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
2064 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
2065 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
2067 /* protected by i_mutex */
2068 if (info
->seals
& F_SEAL_WRITE
) {
2073 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
2074 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
2075 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
2076 spin_lock(&inode
->i_lock
);
2077 inode
->i_private
= &shmem_falloc
;
2078 spin_unlock(&inode
->i_lock
);
2080 if ((u64
)unmap_end
> (u64
)unmap_start
)
2081 unmap_mapping_range(mapping
, unmap_start
,
2082 1 + unmap_end
- unmap_start
, 0);
2083 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
2084 /* No need to unmap again: hole-punching leaves COWed pages */
2086 spin_lock(&inode
->i_lock
);
2087 inode
->i_private
= NULL
;
2088 wake_up_all(&shmem_falloc_waitq
);
2089 spin_unlock(&inode
->i_lock
);
2094 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2095 error
= inode_newsize_ok(inode
, offset
+ len
);
2099 if ((info
->seals
& F_SEAL_GROW
) && offset
+ len
> inode
->i_size
) {
2104 start
= offset
>> PAGE_CACHE_SHIFT
;
2105 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
2106 /* Try to avoid a swapstorm if len is impossible to satisfy */
2107 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
2112 shmem_falloc
.waitq
= NULL
;
2113 shmem_falloc
.start
= start
;
2114 shmem_falloc
.next
= start
;
2115 shmem_falloc
.nr_falloced
= 0;
2116 shmem_falloc
.nr_unswapped
= 0;
2117 spin_lock(&inode
->i_lock
);
2118 inode
->i_private
= &shmem_falloc
;
2119 spin_unlock(&inode
->i_lock
);
2121 for (index
= start
; index
< end
; index
++) {
2125 * Good, the fallocate(2) manpage permits EINTR: we may have
2126 * been interrupted because we are using up too much memory.
2128 if (signal_pending(current
))
2130 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
2133 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
2136 /* Remove the !PageUptodate pages we added */
2137 shmem_undo_range(inode
,
2138 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
2139 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
2144 * Inform shmem_writepage() how far we have reached.
2145 * No need for lock or barrier: we have the page lock.
2147 shmem_falloc
.next
++;
2148 if (!PageUptodate(page
))
2149 shmem_falloc
.nr_falloced
++;
2152 * If !PageUptodate, leave it that way so that freeable pages
2153 * can be recognized if we need to rollback on error later.
2154 * But set_page_dirty so that memory pressure will swap rather
2155 * than free the pages we are allocating (and SGP_CACHE pages
2156 * might still be clean: we now need to mark those dirty too).
2158 set_page_dirty(page
);
2160 page_cache_release(page
);
2164 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
2165 i_size_write(inode
, offset
+ len
);
2166 inode
->i_ctime
= CURRENT_TIME
;
2168 spin_lock(&inode
->i_lock
);
2169 inode
->i_private
= NULL
;
2170 spin_unlock(&inode
->i_lock
);
2172 mutex_unlock(&inode
->i_mutex
);
2176 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2178 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2180 buf
->f_type
= TMPFS_MAGIC
;
2181 buf
->f_bsize
= PAGE_CACHE_SIZE
;
2182 buf
->f_namelen
= NAME_MAX
;
2183 if (sbinfo
->max_blocks
) {
2184 buf
->f_blocks
= sbinfo
->max_blocks
;
2186 buf
->f_bfree
= sbinfo
->max_blocks
-
2187 percpu_counter_sum(&sbinfo
->used_blocks
);
2189 if (sbinfo
->max_inodes
) {
2190 buf
->f_files
= sbinfo
->max_inodes
;
2191 buf
->f_ffree
= sbinfo
->free_inodes
;
2193 /* else leave those fields 0 like simple_statfs */
2198 * File creation. Allocate an inode, and we're done..
2201 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2203 struct inode
*inode
;
2204 int error
= -ENOSPC
;
2206 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2208 error
= simple_acl_create(dir
, inode
);
2211 error
= security_inode_init_security(inode
, dir
,
2213 shmem_initxattrs
, NULL
);
2214 if (error
&& error
!= -EOPNOTSUPP
)
2218 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2219 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2220 d_instantiate(dentry
, inode
);
2221 dget(dentry
); /* Extra count - pin the dentry in core */
2230 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2232 struct inode
*inode
;
2233 int error
= -ENOSPC
;
2235 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2237 error
= security_inode_init_security(inode
, dir
,
2239 shmem_initxattrs
, NULL
);
2240 if (error
&& error
!= -EOPNOTSUPP
)
2242 error
= simple_acl_create(dir
, inode
);
2245 d_tmpfile(dentry
, inode
);
2253 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2257 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2263 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2266 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2272 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2274 struct inode
*inode
= old_dentry
->d_inode
;
2278 * No ordinary (disk based) filesystem counts links as inodes;
2279 * but each new link needs a new dentry, pinning lowmem, and
2280 * tmpfs dentries cannot be pruned until they are unlinked.
2282 ret
= shmem_reserve_inode(inode
->i_sb
);
2286 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2287 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2289 ihold(inode
); /* New dentry reference */
2290 dget(dentry
); /* Extra pinning count for the created dentry */
2291 d_instantiate(dentry
, inode
);
2296 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2298 struct inode
*inode
= dentry
->d_inode
;
2300 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2301 shmem_free_inode(inode
->i_sb
);
2303 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2304 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2306 dput(dentry
); /* Undo the count from "create" - this does all the work */
2310 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2312 if (!simple_empty(dentry
))
2315 drop_nlink(dentry
->d_inode
);
2317 return shmem_unlink(dir
, dentry
);
2320 static int shmem_exchange(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2322 bool old_is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
2323 bool new_is_dir
= S_ISDIR(new_dentry
->d_inode
->i_mode
);
2325 if (old_dir
!= new_dir
&& old_is_dir
!= new_is_dir
) {
2327 drop_nlink(old_dir
);
2330 drop_nlink(new_dir
);
2334 old_dir
->i_ctime
= old_dir
->i_mtime
=
2335 new_dir
->i_ctime
= new_dir
->i_mtime
=
2336 old_dentry
->d_inode
->i_ctime
=
2337 new_dentry
->d_inode
->i_ctime
= CURRENT_TIME
;
2342 static int shmem_whiteout(struct inode
*old_dir
, struct dentry
*old_dentry
)
2344 struct dentry
*whiteout
;
2347 whiteout
= d_alloc(old_dentry
->d_parent
, &old_dentry
->d_name
);
2351 error
= shmem_mknod(old_dir
, whiteout
,
2352 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
2358 * Cheat and hash the whiteout while the old dentry is still in
2359 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2361 * d_lookup() will consistently find one of them at this point,
2362 * not sure which one, but that isn't even important.
2369 * The VFS layer already does all the dentry stuff for rename,
2370 * we just have to decrement the usage count for the target if
2371 * it exists so that the VFS layer correctly free's it when it
2374 static int shmem_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
, unsigned int flags
)
2376 struct inode
*inode
= old_dentry
->d_inode
;
2377 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2379 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
2382 if (flags
& RENAME_EXCHANGE
)
2383 return shmem_exchange(old_dir
, old_dentry
, new_dir
, new_dentry
);
2385 if (!simple_empty(new_dentry
))
2388 if (flags
& RENAME_WHITEOUT
) {
2391 error
= shmem_whiteout(old_dir
, old_dentry
);
2396 if (new_dentry
->d_inode
) {
2397 (void) shmem_unlink(new_dir
, new_dentry
);
2398 if (they_are_dirs
) {
2399 drop_nlink(new_dentry
->d_inode
);
2400 drop_nlink(old_dir
);
2402 } else if (they_are_dirs
) {
2403 drop_nlink(old_dir
);
2407 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2408 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2409 old_dir
->i_ctime
= old_dir
->i_mtime
=
2410 new_dir
->i_ctime
= new_dir
->i_mtime
=
2411 inode
->i_ctime
= CURRENT_TIME
;
2415 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2419 struct inode
*inode
;
2422 struct shmem_inode_info
*info
;
2424 len
= strlen(symname
) + 1;
2425 if (len
> PAGE_CACHE_SIZE
)
2426 return -ENAMETOOLONG
;
2428 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2432 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2433 shmem_initxattrs
, NULL
);
2435 if (error
!= -EOPNOTSUPP
) {
2442 info
= SHMEM_I(inode
);
2443 inode
->i_size
= len
-1;
2444 if (len
<= SHORT_SYMLINK_LEN
) {
2445 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2446 if (!info
->symlink
) {
2450 inode
->i_op
= &shmem_short_symlink_operations
;
2452 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2457 inode
->i_mapping
->a_ops
= &shmem_aops
;
2458 inode
->i_op
= &shmem_symlink_inode_operations
;
2459 kaddr
= kmap_atomic(page
);
2460 memcpy(kaddr
, symname
, len
);
2461 kunmap_atomic(kaddr
);
2462 SetPageUptodate(page
);
2463 set_page_dirty(page
);
2465 page_cache_release(page
);
2467 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2468 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2469 d_instantiate(dentry
, inode
);
2474 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
2476 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
2480 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
2482 struct page
*page
= NULL
;
2483 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
2484 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
2490 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2492 if (!IS_ERR(nd_get_link(nd
))) {
2493 struct page
*page
= cookie
;
2495 mark_page_accessed(page
);
2496 page_cache_release(page
);
2500 #ifdef CONFIG_TMPFS_XATTR
2502 * Superblocks without xattr inode operations may get some security.* xattr
2503 * support from the LSM "for free". As soon as we have any other xattrs
2504 * like ACLs, we also need to implement the security.* handlers at
2505 * filesystem level, though.
2509 * Callback for security_inode_init_security() for acquiring xattrs.
2511 static int shmem_initxattrs(struct inode
*inode
,
2512 const struct xattr
*xattr_array
,
2515 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2516 const struct xattr
*xattr
;
2517 struct simple_xattr
*new_xattr
;
2520 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2521 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2525 len
= strlen(xattr
->name
) + 1;
2526 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2528 if (!new_xattr
->name
) {
2533 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2534 XATTR_SECURITY_PREFIX_LEN
);
2535 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2538 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2544 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2545 #ifdef CONFIG_TMPFS_POSIX_ACL
2546 &posix_acl_access_xattr_handler
,
2547 &posix_acl_default_xattr_handler
,
2552 static int shmem_xattr_validate(const char *name
)
2554 struct { const char *prefix
; size_t len
; } arr
[] = {
2555 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2556 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2560 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2561 size_t preflen
= arr
[i
].len
;
2562 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2571 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2572 void *buffer
, size_t size
)
2574 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2578 * If this is a request for a synthetic attribute in the system.*
2579 * namespace use the generic infrastructure to resolve a handler
2580 * for it via sb->s_xattr.
2582 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2583 return generic_getxattr(dentry
, name
, buffer
, size
);
2585 err
= shmem_xattr_validate(name
);
2589 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2592 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2593 const void *value
, size_t size
, int flags
)
2595 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2599 * If this is a request for a synthetic attribute in the system.*
2600 * namespace use the generic infrastructure to resolve a handler
2601 * for it via sb->s_xattr.
2603 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2604 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2606 err
= shmem_xattr_validate(name
);
2610 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2613 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2615 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2619 * If this is a request for a synthetic attribute in the system.*
2620 * namespace use the generic infrastructure to resolve a handler
2621 * for it via sb->s_xattr.
2623 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2624 return generic_removexattr(dentry
, name
);
2626 err
= shmem_xattr_validate(name
);
2630 return simple_xattr_remove(&info
->xattrs
, name
);
2633 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2635 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2636 return simple_xattr_list(&info
->xattrs
, buffer
, size
);
2638 #endif /* CONFIG_TMPFS_XATTR */
2640 static const struct inode_operations shmem_short_symlink_operations
= {
2641 .readlink
= generic_readlink
,
2642 .follow_link
= shmem_follow_short_symlink
,
2643 #ifdef CONFIG_TMPFS_XATTR
2644 .setxattr
= shmem_setxattr
,
2645 .getxattr
= shmem_getxattr
,
2646 .listxattr
= shmem_listxattr
,
2647 .removexattr
= shmem_removexattr
,
2651 static const struct inode_operations shmem_symlink_inode_operations
= {
2652 .readlink
= generic_readlink
,
2653 .follow_link
= shmem_follow_link
,
2654 .put_link
= shmem_put_link
,
2655 #ifdef CONFIG_TMPFS_XATTR
2656 .setxattr
= shmem_setxattr
,
2657 .getxattr
= shmem_getxattr
,
2658 .listxattr
= shmem_listxattr
,
2659 .removexattr
= shmem_removexattr
,
2663 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2665 return ERR_PTR(-ESTALE
);
2668 static int shmem_match(struct inode
*ino
, void *vfh
)
2672 inum
= (inum
<< 32) | fh
[1];
2673 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2676 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2677 struct fid
*fid
, int fh_len
, int fh_type
)
2679 struct inode
*inode
;
2680 struct dentry
*dentry
= NULL
;
2687 inum
= (inum
<< 32) | fid
->raw
[1];
2689 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2690 shmem_match
, fid
->raw
);
2692 dentry
= d_find_alias(inode
);
2699 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2700 struct inode
*parent
)
2704 return FILEID_INVALID
;
2707 if (inode_unhashed(inode
)) {
2708 /* Unfortunately insert_inode_hash is not idempotent,
2709 * so as we hash inodes here rather than at creation
2710 * time, we need a lock to ensure we only try
2713 static DEFINE_SPINLOCK(lock
);
2715 if (inode_unhashed(inode
))
2716 __insert_inode_hash(inode
,
2717 inode
->i_ino
+ inode
->i_generation
);
2721 fh
[0] = inode
->i_generation
;
2722 fh
[1] = inode
->i_ino
;
2723 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2729 static const struct export_operations shmem_export_ops
= {
2730 .get_parent
= shmem_get_parent
,
2731 .encode_fh
= shmem_encode_fh
,
2732 .fh_to_dentry
= shmem_fh_to_dentry
,
2735 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2738 char *this_char
, *value
, *rest
;
2739 struct mempolicy
*mpol
= NULL
;
2743 while (options
!= NULL
) {
2744 this_char
= options
;
2747 * NUL-terminate this option: unfortunately,
2748 * mount options form a comma-separated list,
2749 * but mpol's nodelist may also contain commas.
2751 options
= strchr(options
, ',');
2752 if (options
== NULL
)
2755 if (!isdigit(*options
)) {
2762 if ((value
= strchr(this_char
,'=')) != NULL
) {
2766 "tmpfs: No value for mount option '%s'\n",
2771 if (!strcmp(this_char
,"size")) {
2772 unsigned long long size
;
2773 size
= memparse(value
,&rest
);
2775 size
<<= PAGE_SHIFT
;
2776 size
*= totalram_pages
;
2782 sbinfo
->max_blocks
=
2783 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2784 } else if (!strcmp(this_char
,"nr_blocks")) {
2785 sbinfo
->max_blocks
= memparse(value
, &rest
);
2788 } else if (!strcmp(this_char
,"nr_inodes")) {
2789 sbinfo
->max_inodes
= memparse(value
, &rest
);
2792 } else if (!strcmp(this_char
,"mode")) {
2795 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2798 } else if (!strcmp(this_char
,"uid")) {
2801 uid
= simple_strtoul(value
, &rest
, 0);
2804 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2805 if (!uid_valid(sbinfo
->uid
))
2807 } else if (!strcmp(this_char
,"gid")) {
2810 gid
= simple_strtoul(value
, &rest
, 0);
2813 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2814 if (!gid_valid(sbinfo
->gid
))
2816 } else if (!strcmp(this_char
,"mpol")) {
2819 if (mpol_parse_str(value
, &mpol
))
2822 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2827 sbinfo
->mpol
= mpol
;
2831 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2839 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2841 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2842 struct shmem_sb_info config
= *sbinfo
;
2843 unsigned long inodes
;
2844 int error
= -EINVAL
;
2847 if (shmem_parse_options(data
, &config
, true))
2850 spin_lock(&sbinfo
->stat_lock
);
2851 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2852 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2854 if (config
.max_inodes
< inodes
)
2857 * Those tests disallow limited->unlimited while any are in use;
2858 * but we must separately disallow unlimited->limited, because
2859 * in that case we have no record of how much is already in use.
2861 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2863 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2867 sbinfo
->max_blocks
= config
.max_blocks
;
2868 sbinfo
->max_inodes
= config
.max_inodes
;
2869 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2872 * Preserve previous mempolicy unless mpol remount option was specified.
2875 mpol_put(sbinfo
->mpol
);
2876 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2879 spin_unlock(&sbinfo
->stat_lock
);
2883 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2885 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2887 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2888 seq_printf(seq
, ",size=%luk",
2889 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2890 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2891 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2892 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2893 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2894 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2895 seq_printf(seq
, ",uid=%u",
2896 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2897 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2898 seq_printf(seq
, ",gid=%u",
2899 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2900 shmem_show_mpol(seq
, sbinfo
->mpol
);
2904 #define MFD_NAME_PREFIX "memfd:"
2905 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2906 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2908 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2910 SYSCALL_DEFINE2(memfd_create
,
2911 const char __user
*, uname
,
2912 unsigned int, flags
)
2914 struct shmem_inode_info
*info
;
2920 if (flags
& ~(unsigned int)MFD_ALL_FLAGS
)
2923 /* length includes terminating zero */
2924 len
= strnlen_user(uname
, MFD_NAME_MAX_LEN
+ 1);
2927 if (len
> MFD_NAME_MAX_LEN
+ 1)
2930 name
= kmalloc(len
+ MFD_NAME_PREFIX_LEN
, GFP_TEMPORARY
);
2934 strcpy(name
, MFD_NAME_PREFIX
);
2935 if (copy_from_user(&name
[MFD_NAME_PREFIX_LEN
], uname
, len
)) {
2940 /* terminating-zero may have changed after strnlen_user() returned */
2941 if (name
[len
+ MFD_NAME_PREFIX_LEN
- 1]) {
2946 fd
= get_unused_fd_flags((flags
& MFD_CLOEXEC
) ? O_CLOEXEC
: 0);
2952 file
= shmem_file_setup(name
, 0, VM_NORESERVE
);
2954 error
= PTR_ERR(file
);
2957 info
= SHMEM_I(file_inode(file
));
2958 file
->f_mode
|= FMODE_LSEEK
| FMODE_PREAD
| FMODE_PWRITE
;
2959 file
->f_flags
|= O_RDWR
| O_LARGEFILE
;
2960 if (flags
& MFD_ALLOW_SEALING
)
2961 info
->seals
&= ~F_SEAL_SEAL
;
2963 fd_install(fd
, file
);
2974 #endif /* CONFIG_TMPFS */
2976 static void shmem_put_super(struct super_block
*sb
)
2978 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2980 percpu_counter_destroy(&sbinfo
->used_blocks
);
2981 mpol_put(sbinfo
->mpol
);
2983 sb
->s_fs_info
= NULL
;
2986 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2988 struct inode
*inode
;
2989 struct shmem_sb_info
*sbinfo
;
2992 /* Round up to L1_CACHE_BYTES to resist false sharing */
2993 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2994 L1_CACHE_BYTES
), GFP_KERNEL
);
2998 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2999 sbinfo
->uid
= current_fsuid();
3000 sbinfo
->gid
= current_fsgid();
3001 sb
->s_fs_info
= sbinfo
;
3005 * Per default we only allow half of the physical ram per
3006 * tmpfs instance, limiting inodes to one per page of lowmem;
3007 * but the internal instance is left unlimited.
3009 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
3010 sbinfo
->max_blocks
= shmem_default_max_blocks();
3011 sbinfo
->max_inodes
= shmem_default_max_inodes();
3012 if (shmem_parse_options(data
, sbinfo
, false)) {
3017 sb
->s_flags
|= MS_NOUSER
;
3019 sb
->s_export_op
= &shmem_export_ops
;
3020 sb
->s_flags
|= MS_NOSEC
;
3022 sb
->s_flags
|= MS_NOUSER
;
3025 spin_lock_init(&sbinfo
->stat_lock
);
3026 if (percpu_counter_init(&sbinfo
->used_blocks
, 0, GFP_KERNEL
))
3028 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
3030 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
3031 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
3032 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
3033 sb
->s_magic
= TMPFS_MAGIC
;
3034 sb
->s_op
= &shmem_ops
;
3035 sb
->s_time_gran
= 1;
3036 #ifdef CONFIG_TMPFS_XATTR
3037 sb
->s_xattr
= shmem_xattr_handlers
;
3039 #ifdef CONFIG_TMPFS_POSIX_ACL
3040 sb
->s_flags
|= MS_POSIXACL
;
3043 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
3046 inode
->i_uid
= sbinfo
->uid
;
3047 inode
->i_gid
= sbinfo
->gid
;
3048 sb
->s_root
= d_make_root(inode
);
3054 shmem_put_super(sb
);
3058 static struct kmem_cache
*shmem_inode_cachep
;
3060 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
3062 struct shmem_inode_info
*info
;
3063 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
3066 return &info
->vfs_inode
;
3069 static void shmem_destroy_callback(struct rcu_head
*head
)
3071 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
3072 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
3075 static void shmem_destroy_inode(struct inode
*inode
)
3077 if (S_ISREG(inode
->i_mode
))
3078 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
3079 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
3082 static void shmem_init_inode(void *foo
)
3084 struct shmem_inode_info
*info
= foo
;
3085 inode_init_once(&info
->vfs_inode
);
3088 static int shmem_init_inodecache(void)
3090 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
3091 sizeof(struct shmem_inode_info
),
3092 0, SLAB_PANIC
, shmem_init_inode
);
3096 static void shmem_destroy_inodecache(void)
3098 kmem_cache_destroy(shmem_inode_cachep
);
3101 static const struct address_space_operations shmem_aops
= {
3102 .writepage
= shmem_writepage
,
3103 .set_page_dirty
= __set_page_dirty_no_writeback
,
3105 .write_begin
= shmem_write_begin
,
3106 .write_end
= shmem_write_end
,
3108 #ifdef CONFIG_MIGRATION
3109 .migratepage
= migrate_page
,
3111 .error_remove_page
= generic_error_remove_page
,
3114 static const struct file_operations shmem_file_operations
= {
3117 .llseek
= shmem_file_llseek
,
3118 .read
= new_sync_read
,
3119 .write
= new_sync_write
,
3120 .read_iter
= shmem_file_read_iter
,
3121 .write_iter
= generic_file_write_iter
,
3122 .fsync
= noop_fsync
,
3123 .splice_read
= shmem_file_splice_read
,
3124 .splice_write
= iter_file_splice_write
,
3125 .fallocate
= shmem_fallocate
,
3129 static const struct inode_operations shmem_inode_operations
= {
3130 .setattr
= shmem_setattr
,
3131 #ifdef CONFIG_TMPFS_XATTR
3132 .setxattr
= shmem_setxattr
,
3133 .getxattr
= shmem_getxattr
,
3134 .listxattr
= shmem_listxattr
,
3135 .removexattr
= shmem_removexattr
,
3136 .set_acl
= simple_set_acl
,
3140 static const struct inode_operations shmem_dir_inode_operations
= {
3142 .create
= shmem_create
,
3143 .lookup
= simple_lookup
,
3145 .unlink
= shmem_unlink
,
3146 .symlink
= shmem_symlink
,
3147 .mkdir
= shmem_mkdir
,
3148 .rmdir
= shmem_rmdir
,
3149 .mknod
= shmem_mknod
,
3150 .rename2
= shmem_rename2
,
3151 .tmpfile
= shmem_tmpfile
,
3153 #ifdef CONFIG_TMPFS_XATTR
3154 .setxattr
= shmem_setxattr
,
3155 .getxattr
= shmem_getxattr
,
3156 .listxattr
= shmem_listxattr
,
3157 .removexattr
= shmem_removexattr
,
3159 #ifdef CONFIG_TMPFS_POSIX_ACL
3160 .setattr
= shmem_setattr
,
3161 .set_acl
= simple_set_acl
,
3165 static const struct inode_operations shmem_special_inode_operations
= {
3166 #ifdef CONFIG_TMPFS_XATTR
3167 .setxattr
= shmem_setxattr
,
3168 .getxattr
= shmem_getxattr
,
3169 .listxattr
= shmem_listxattr
,
3170 .removexattr
= shmem_removexattr
,
3172 #ifdef CONFIG_TMPFS_POSIX_ACL
3173 .setattr
= shmem_setattr
,
3174 .set_acl
= simple_set_acl
,
3178 static const struct super_operations shmem_ops
= {
3179 .alloc_inode
= shmem_alloc_inode
,
3180 .destroy_inode
= shmem_destroy_inode
,
3182 .statfs
= shmem_statfs
,
3183 .remount_fs
= shmem_remount_fs
,
3184 .show_options
= shmem_show_options
,
3186 .evict_inode
= shmem_evict_inode
,
3187 .drop_inode
= generic_delete_inode
,
3188 .put_super
= shmem_put_super
,
3191 static const struct vm_operations_struct shmem_vm_ops
= {
3192 .fault
= shmem_fault
,
3193 .map_pages
= filemap_map_pages
,
3195 .set_policy
= shmem_set_policy
,
3196 .get_policy
= shmem_get_policy
,
3200 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
3201 int flags
, const char *dev_name
, void *data
)
3203 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
3206 static struct file_system_type shmem_fs_type
= {
3207 .owner
= THIS_MODULE
,
3209 .mount
= shmem_mount
,
3210 .kill_sb
= kill_litter_super
,
3211 .fs_flags
= FS_USERNS_MOUNT
,
3214 int __init
shmem_init(void)
3218 /* If rootfs called this, don't re-init */
3219 if (shmem_inode_cachep
)
3222 error
= shmem_init_inodecache();
3226 error
= register_filesystem(&shmem_fs_type
);
3228 printk(KERN_ERR
"Could not register tmpfs\n");
3232 shm_mnt
= kern_mount(&shmem_fs_type
);
3233 if (IS_ERR(shm_mnt
)) {
3234 error
= PTR_ERR(shm_mnt
);
3235 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
3241 unregister_filesystem(&shmem_fs_type
);
3243 shmem_destroy_inodecache();
3245 shm_mnt
= ERR_PTR(error
);
3249 #else /* !CONFIG_SHMEM */
3252 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3254 * This is intended for small system where the benefits of the full
3255 * shmem code (swap-backed and resource-limited) are outweighed by
3256 * their complexity. On systems without swap this code should be
3257 * effectively equivalent, but much lighter weight.
3260 static struct file_system_type shmem_fs_type
= {
3262 .mount
= ramfs_mount
,
3263 .kill_sb
= kill_litter_super
,
3264 .fs_flags
= FS_USERNS_MOUNT
,
3267 int __init
shmem_init(void)
3269 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
3271 shm_mnt
= kern_mount(&shmem_fs_type
);
3272 BUG_ON(IS_ERR(shm_mnt
));
3277 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
3282 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
3287 void shmem_unlock_mapping(struct address_space
*mapping
)
3291 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
3293 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
3295 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3297 #define shmem_vm_ops generic_file_vm_ops
3298 #define shmem_file_operations ramfs_file_operations
3299 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3300 #define shmem_acct_size(flags, size) 0
3301 #define shmem_unacct_size(flags, size) do {} while (0)
3303 #endif /* CONFIG_SHMEM */
3307 static struct dentry_operations anon_ops
= {
3308 .d_dname
= simple_dname
3311 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
3312 unsigned long flags
, unsigned int i_flags
)
3315 struct inode
*inode
;
3317 struct super_block
*sb
;
3320 if (IS_ERR(shm_mnt
))
3321 return ERR_CAST(shm_mnt
);
3323 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3324 return ERR_PTR(-EINVAL
);
3326 if (shmem_acct_size(flags
, size
))
3327 return ERR_PTR(-ENOMEM
);
3329 res
= ERR_PTR(-ENOMEM
);
3331 this.len
= strlen(name
);
3332 this.hash
= 0; /* will go */
3333 sb
= shm_mnt
->mnt_sb
;
3334 path
.mnt
= mntget(shm_mnt
);
3335 path
.dentry
= d_alloc_pseudo(sb
, &this);
3338 d_set_d_op(path
.dentry
, &anon_ops
);
3340 res
= ERR_PTR(-ENOSPC
);
3341 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3345 inode
->i_flags
|= i_flags
;
3346 d_instantiate(path
.dentry
, inode
);
3347 inode
->i_size
= size
;
3348 clear_nlink(inode
); /* It is unlinked */
3349 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3353 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3354 &shmem_file_operations
);
3361 shmem_unacct_size(flags
, size
);
3368 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3369 * kernel internal. There will be NO LSM permission checks against the
3370 * underlying inode. So users of this interface must do LSM checks at a
3371 * higher layer. The one user is the big_key implementation. LSM checks
3372 * are provided at the key level rather than the inode level.
3373 * @name: name for dentry (to be seen in /proc/<pid>/maps
3374 * @size: size to be set for the file
3375 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3377 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3379 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3383 * shmem_file_setup - get an unlinked file living in tmpfs
3384 * @name: name for dentry (to be seen in /proc/<pid>/maps
3385 * @size: size to be set for the file
3386 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3388 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3390 return __shmem_file_setup(name
, size
, flags
, 0);
3392 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3395 * shmem_zero_setup - setup a shared anonymous mapping
3396 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3398 int shmem_zero_setup(struct vm_area_struct
*vma
)
3401 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3403 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
3405 return PTR_ERR(file
);
3409 vma
->vm_file
= file
;
3410 vma
->vm_ops
= &shmem_vm_ops
;
3415 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3416 * @mapping: the page's address_space
3417 * @index: the page index
3418 * @gfp: the page allocator flags to use if allocating
3420 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3421 * with any new page allocations done using the specified allocation flags.
3422 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3423 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3424 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3426 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3427 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3429 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3430 pgoff_t index
, gfp_t gfp
)
3433 struct inode
*inode
= mapping
->host
;
3437 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3438 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3440 page
= ERR_PTR(error
);
3446 * The tiny !SHMEM case uses ramfs without swap
3448 return read_cache_page_gfp(mapping
, index
, gfp
);
3451 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);