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/generic_acl.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>
70 #include <asm/uaccess.h>
71 #include <asm/pgtable.h>
73 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
74 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
79 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
80 #define SHORT_SYMLINK_LEN 128
83 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
84 * inode->i_private (with i_mutex making sure that it has only one user at
85 * a time): we would prefer not to enlarge the shmem inode just for that.
88 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
89 pgoff_t start
; /* start of range currently being fallocated */
90 pgoff_t next
; /* the next page offset to be fallocated */
91 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
92 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
95 /* Flag allocation requirements to shmem_getpage */
97 SGP_READ
, /* don't exceed i_size, don't allocate page */
98 SGP_CACHE
, /* don't exceed i_size, may allocate page */
99 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
100 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
101 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
105 static unsigned long shmem_default_max_blocks(void)
107 return totalram_pages
/ 2;
110 static unsigned long shmem_default_max_inodes(void)
112 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
116 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
117 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
118 struct shmem_inode_info
*info
, pgoff_t index
);
119 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
120 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
122 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
123 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
125 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
126 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
129 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
131 return sb
->s_fs_info
;
135 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
136 * for shared memory and for shared anonymous (/dev/zero) mappings
137 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
138 * consistent with the pre-accounting of private mappings ...
140 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
142 return (flags
& VM_NORESERVE
) ?
143 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
146 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
148 if (!(flags
& VM_NORESERVE
))
149 vm_unacct_memory(VM_ACCT(size
));
153 * ... whereas tmpfs objects are accounted incrementally as
154 * pages are allocated, in order to allow huge sparse files.
155 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
156 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
158 static inline int shmem_acct_block(unsigned long flags
)
160 return (flags
& VM_NORESERVE
) ?
161 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
164 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
166 if (flags
& VM_NORESERVE
)
167 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
170 static const struct super_operations shmem_ops
;
171 static const struct address_space_operations shmem_aops
;
172 static const struct file_operations shmem_file_operations
;
173 static const struct inode_operations shmem_inode_operations
;
174 static const struct inode_operations shmem_dir_inode_operations
;
175 static const struct inode_operations shmem_special_inode_operations
;
176 static const struct vm_operations_struct shmem_vm_ops
;
178 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
179 .ra_pages
= 0, /* No readahead */
180 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
183 static LIST_HEAD(shmem_swaplist
);
184 static DEFINE_MUTEX(shmem_swaplist_mutex
);
186 static int shmem_reserve_inode(struct super_block
*sb
)
188 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
189 if (sbinfo
->max_inodes
) {
190 spin_lock(&sbinfo
->stat_lock
);
191 if (!sbinfo
->free_inodes
) {
192 spin_unlock(&sbinfo
->stat_lock
);
195 sbinfo
->free_inodes
--;
196 spin_unlock(&sbinfo
->stat_lock
);
201 static void shmem_free_inode(struct super_block
*sb
)
203 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
204 if (sbinfo
->max_inodes
) {
205 spin_lock(&sbinfo
->stat_lock
);
206 sbinfo
->free_inodes
++;
207 spin_unlock(&sbinfo
->stat_lock
);
212 * shmem_recalc_inode - recalculate the block usage of an inode
213 * @inode: inode to recalc
215 * We have to calculate the free blocks since the mm can drop
216 * undirtied hole pages behind our back.
218 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
219 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
221 * It has to be called with the spinlock held.
223 static void shmem_recalc_inode(struct inode
*inode
)
225 struct shmem_inode_info
*info
= SHMEM_I(inode
);
228 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
230 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
231 if (sbinfo
->max_blocks
)
232 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
233 info
->alloced
-= freed
;
234 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
235 shmem_unacct_blocks(info
->flags
, freed
);
240 * Replace item expected in radix tree by a new item, while holding tree lock.
242 static int shmem_radix_tree_replace(struct address_space
*mapping
,
243 pgoff_t index
, void *expected
, void *replacement
)
248 VM_BUG_ON(!expected
);
249 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
251 item
= radix_tree_deref_slot_protected(pslot
,
252 &mapping
->tree_lock
);
253 if (item
!= expected
)
256 radix_tree_replace_slot(pslot
, replacement
);
258 radix_tree_delete(&mapping
->page_tree
, index
);
263 * Sometimes, before we decide whether to proceed or to fail, we must check
264 * that an entry was not already brought back from swap by a racing thread.
266 * Checking page is not enough: by the time a SwapCache page is locked, it
267 * might be reused, and again be SwapCache, using the same swap as before.
269 static bool shmem_confirm_swap(struct address_space
*mapping
,
270 pgoff_t index
, swp_entry_t swap
)
275 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
277 return item
== swp_to_radix_entry(swap
);
281 * Like add_to_page_cache_locked, but error if expected item has gone.
283 static int shmem_add_to_page_cache(struct page
*page
,
284 struct address_space
*mapping
,
285 pgoff_t index
, gfp_t gfp
, void *expected
)
289 VM_BUG_ON(!PageLocked(page
));
290 VM_BUG_ON(!PageSwapBacked(page
));
292 page_cache_get(page
);
293 page
->mapping
= mapping
;
296 spin_lock_irq(&mapping
->tree_lock
);
298 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
300 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
304 __inc_zone_page_state(page
, NR_FILE_PAGES
);
305 __inc_zone_page_state(page
, NR_SHMEM
);
306 spin_unlock_irq(&mapping
->tree_lock
);
308 page
->mapping
= NULL
;
309 spin_unlock_irq(&mapping
->tree_lock
);
310 page_cache_release(page
);
316 * Like delete_from_page_cache, but substitutes swap for page.
318 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
320 struct address_space
*mapping
= page
->mapping
;
323 spin_lock_irq(&mapping
->tree_lock
);
324 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
325 page
->mapping
= NULL
;
327 __dec_zone_page_state(page
, NR_FILE_PAGES
);
328 __dec_zone_page_state(page
, NR_SHMEM
);
329 spin_unlock_irq(&mapping
->tree_lock
);
330 page_cache_release(page
);
335 * Like find_get_pages, but collecting swap entries as well as pages.
337 static unsigned shmem_find_get_pages_and_swap(struct address_space
*mapping
,
338 pgoff_t start
, unsigned int nr_pages
,
339 struct page
**pages
, pgoff_t
*indices
)
342 unsigned int ret
= 0;
343 struct radix_tree_iter iter
;
350 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
353 page
= radix_tree_deref_slot(slot
);
356 if (radix_tree_exception(page
)) {
357 if (radix_tree_deref_retry(page
))
360 * Otherwise, we must be storing a swap entry
361 * here as an exceptional entry: so return it
362 * without attempting to raise page count.
366 if (!page_cache_get_speculative(page
))
369 /* Has the page moved? */
370 if (unlikely(page
!= *slot
)) {
371 page_cache_release(page
);
375 indices
[ret
] = iter
.index
;
377 if (++ret
== nr_pages
)
385 * Remove swap entry from radix tree, free the swap and its page cache.
387 static int shmem_free_swap(struct address_space
*mapping
,
388 pgoff_t index
, void *radswap
)
392 spin_lock_irq(&mapping
->tree_lock
);
393 error
= shmem_radix_tree_replace(mapping
, index
, radswap
, NULL
);
394 spin_unlock_irq(&mapping
->tree_lock
);
396 free_swap_and_cache(radix_to_swp_entry(radswap
));
401 * Pagevec may contain swap entries, so shuffle up pages before releasing.
403 static void shmem_deswap_pagevec(struct pagevec
*pvec
)
407 for (i
= 0, j
= 0; i
< pagevec_count(pvec
); i
++) {
408 struct page
*page
= pvec
->pages
[i
];
409 if (!radix_tree_exceptional_entry(page
))
410 pvec
->pages
[j
++] = page
;
416 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
418 void shmem_unlock_mapping(struct address_space
*mapping
)
421 pgoff_t indices
[PAGEVEC_SIZE
];
424 pagevec_init(&pvec
, 0);
426 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
428 while (!mapping_unevictable(mapping
)) {
430 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
431 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
433 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
434 PAGEVEC_SIZE
, pvec
.pages
, indices
);
437 index
= indices
[pvec
.nr
- 1] + 1;
438 shmem_deswap_pagevec(&pvec
);
439 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
440 pagevec_release(&pvec
);
446 * Remove range of pages and swap entries from radix tree, and free them.
447 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
449 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
452 struct address_space
*mapping
= inode
->i_mapping
;
453 struct shmem_inode_info
*info
= SHMEM_I(inode
);
454 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
455 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
456 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
457 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
459 pgoff_t indices
[PAGEVEC_SIZE
];
460 long nr_swaps_freed
= 0;
465 end
= -1; /* unsigned, so actually very big */
467 pagevec_init(&pvec
, 0);
469 while (index
< end
) {
470 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
471 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
472 pvec
.pages
, indices
);
475 mem_cgroup_uncharge_start();
476 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
477 struct page
*page
= pvec
.pages
[i
];
483 if (radix_tree_exceptional_entry(page
)) {
486 if (shmem_free_swap(mapping
, index
, page
)) {
487 /* Swap was replaced by page: retry */
495 if (!trylock_page(page
))
497 if (!unfalloc
|| !PageUptodate(page
)) {
498 if (page
->mapping
== mapping
) {
499 VM_BUG_ON(PageWriteback(page
));
500 truncate_inode_page(mapping
, page
);
502 /* Page was replaced by swap: retry */
510 shmem_deswap_pagevec(&pvec
);
511 pagevec_release(&pvec
);
512 mem_cgroup_uncharge_end();
518 struct page
*page
= NULL
;
519 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
521 unsigned int top
= PAGE_CACHE_SIZE
;
526 zero_user_segment(page
, partial_start
, top
);
527 set_page_dirty(page
);
529 page_cache_release(page
);
533 struct page
*page
= NULL
;
534 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
536 zero_user_segment(page
, 0, partial_end
);
537 set_page_dirty(page
);
539 page_cache_release(page
);
546 while (index
< end
) {
548 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
549 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
550 pvec
.pages
, indices
);
552 /* If all gone or hole-punch or unfalloc, we're done */
553 if (index
== start
|| end
!= -1)
555 /* But if truncating, restart to make sure all gone */
559 mem_cgroup_uncharge_start();
560 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
561 struct page
*page
= pvec
.pages
[i
];
567 if (radix_tree_exceptional_entry(page
)) {
570 if (shmem_free_swap(mapping
, index
, page
)) {
571 /* Swap was replaced by page: retry */
580 if (!unfalloc
|| !PageUptodate(page
)) {
581 if (page
->mapping
== mapping
) {
582 VM_BUG_ON(PageWriteback(page
));
583 truncate_inode_page(mapping
, page
);
585 /* Page was replaced by swap: retry */
593 shmem_deswap_pagevec(&pvec
);
594 pagevec_release(&pvec
);
595 mem_cgroup_uncharge_end();
599 spin_lock(&info
->lock
);
600 info
->swapped
-= nr_swaps_freed
;
601 shmem_recalc_inode(inode
);
602 spin_unlock(&info
->lock
);
605 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
607 shmem_undo_range(inode
, lstart
, lend
, false);
608 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
610 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
612 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
614 struct inode
*inode
= dentry
->d_inode
;
617 error
= inode_change_ok(inode
, attr
);
621 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
622 loff_t oldsize
= inode
->i_size
;
623 loff_t newsize
= attr
->ia_size
;
625 if (newsize
!= oldsize
) {
626 i_size_write(inode
, newsize
);
627 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
629 if (newsize
< oldsize
) {
630 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
631 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
632 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
633 /* unmap again to remove racily COWed private pages */
634 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
638 setattr_copy(inode
, attr
);
639 #ifdef CONFIG_TMPFS_POSIX_ACL
640 if (attr
->ia_valid
& ATTR_MODE
)
641 error
= generic_acl_chmod(inode
);
646 static void shmem_evict_inode(struct inode
*inode
)
648 struct shmem_inode_info
*info
= SHMEM_I(inode
);
650 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
651 shmem_unacct_size(info
->flags
, inode
->i_size
);
653 shmem_truncate_range(inode
, 0, (loff_t
)-1);
654 if (!list_empty(&info
->swaplist
)) {
655 mutex_lock(&shmem_swaplist_mutex
);
656 list_del_init(&info
->swaplist
);
657 mutex_unlock(&shmem_swaplist_mutex
);
660 kfree(info
->symlink
);
662 simple_xattrs_free(&info
->xattrs
);
663 WARN_ON(inode
->i_blocks
);
664 shmem_free_inode(inode
->i_sb
);
669 * If swap found in inode, free it and move page from swapcache to filecache.
671 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
672 swp_entry_t swap
, struct page
**pagep
)
674 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
680 radswap
= swp_to_radix_entry(swap
);
681 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
686 * Move _head_ to start search for next from here.
687 * But be careful: shmem_evict_inode checks list_empty without taking
688 * mutex, and there's an instant in list_move_tail when info->swaplist
689 * would appear empty, if it were the only one on shmem_swaplist.
691 if (shmem_swaplist
.next
!= &info
->swaplist
)
692 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
694 gfp
= mapping_gfp_mask(mapping
);
695 if (shmem_should_replace_page(*pagep
, gfp
)) {
696 mutex_unlock(&shmem_swaplist_mutex
);
697 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
698 mutex_lock(&shmem_swaplist_mutex
);
700 * We needed to drop mutex to make that restrictive page
701 * allocation, but the inode might have been freed while we
702 * dropped it: although a racing shmem_evict_inode() cannot
703 * complete without emptying the radix_tree, our page lock
704 * on this swapcache page is not enough to prevent that -
705 * free_swap_and_cache() of our swap entry will only
706 * trylock_page(), removing swap from radix_tree whatever.
708 * We must not proceed to shmem_add_to_page_cache() if the
709 * inode has been freed, but of course we cannot rely on
710 * inode or mapping or info to check that. However, we can
711 * safely check if our swap entry is still in use (and here
712 * it can't have got reused for another page): if it's still
713 * in use, then the inode cannot have been freed yet, and we
714 * can safely proceed (if it's no longer in use, that tells
715 * nothing about the inode, but we don't need to unuse swap).
717 if (!page_swapcount(*pagep
))
722 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
723 * but also to hold up shmem_evict_inode(): so inode cannot be freed
724 * beneath us (pagelock doesn't help until the page is in pagecache).
727 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
728 GFP_NOWAIT
, radswap
);
729 if (error
!= -ENOMEM
) {
731 * Truncation and eviction use free_swap_and_cache(), which
732 * only does trylock page: if we raced, best clean up here.
734 delete_from_swap_cache(*pagep
);
735 set_page_dirty(*pagep
);
737 spin_lock(&info
->lock
);
739 spin_unlock(&info
->lock
);
742 error
= 1; /* not an error, but entry was found */
748 * Search through swapped inodes to find and replace swap by page.
750 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
752 struct list_head
*this, *next
;
753 struct shmem_inode_info
*info
;
758 * There's a faint possibility that swap page was replaced before
759 * caller locked it: caller will come back later with the right page.
761 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
765 * Charge page using GFP_KERNEL while we can wait, before taking
766 * the shmem_swaplist_mutex which might hold up shmem_writepage().
767 * Charged back to the user (not to caller) when swap account is used.
769 error
= mem_cgroup_cache_charge(page
, current
->mm
, GFP_KERNEL
);
772 /* No radix_tree_preload: swap entry keeps a place for page in tree */
774 mutex_lock(&shmem_swaplist_mutex
);
775 list_for_each_safe(this, next
, &shmem_swaplist
) {
776 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
778 found
= shmem_unuse_inode(info
, swap
, &page
);
780 list_del_init(&info
->swaplist
);
785 mutex_unlock(&shmem_swaplist_mutex
);
791 page_cache_release(page
);
796 * Move the page from the page cache to the swap cache.
798 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
800 struct shmem_inode_info
*info
;
801 struct address_space
*mapping
;
806 BUG_ON(!PageLocked(page
));
807 mapping
= page
->mapping
;
809 inode
= mapping
->host
;
810 info
= SHMEM_I(inode
);
811 if (info
->flags
& VM_LOCKED
)
813 if (!total_swap_pages
)
817 * shmem_backing_dev_info's capabilities prevent regular writeback or
818 * sync from ever calling shmem_writepage; but a stacking filesystem
819 * might use ->writepage of its underlying filesystem, in which case
820 * tmpfs should write out to swap only in response to memory pressure,
821 * and not for the writeback threads or sync.
823 if (!wbc
->for_reclaim
) {
824 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
829 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
830 * value into swapfile.c, the only way we can correctly account for a
831 * fallocated page arriving here is now to initialize it and write it.
833 * That's okay for a page already fallocated earlier, but if we have
834 * not yet completed the fallocation, then (a) we want to keep track
835 * of this page in case we have to undo it, and (b) it may not be a
836 * good idea to continue anyway, once we're pushing into swap. So
837 * reactivate the page, and let shmem_fallocate() quit when too many.
839 if (!PageUptodate(page
)) {
840 if (inode
->i_private
) {
841 struct shmem_falloc
*shmem_falloc
;
842 spin_lock(&inode
->i_lock
);
843 shmem_falloc
= inode
->i_private
;
845 !shmem_falloc
->waitq
&&
846 index
>= shmem_falloc
->start
&&
847 index
< shmem_falloc
->next
)
848 shmem_falloc
->nr_unswapped
++;
851 spin_unlock(&inode
->i_lock
);
855 clear_highpage(page
);
856 flush_dcache_page(page
);
857 SetPageUptodate(page
);
860 swap
= get_swap_page();
865 * Add inode to shmem_unuse()'s list of swapped-out inodes,
866 * if it's not already there. Do it now before the page is
867 * moved to swap cache, when its pagelock no longer protects
868 * the inode from eviction. But don't unlock the mutex until
869 * we've incremented swapped, because shmem_unuse_inode() will
870 * prune a !swapped inode from the swaplist under this mutex.
872 mutex_lock(&shmem_swaplist_mutex
);
873 if (list_empty(&info
->swaplist
))
874 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
876 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
877 swap_shmem_alloc(swap
);
878 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
880 spin_lock(&info
->lock
);
882 shmem_recalc_inode(inode
);
883 spin_unlock(&info
->lock
);
885 mutex_unlock(&shmem_swaplist_mutex
);
886 BUG_ON(page_mapped(page
));
887 swap_writepage(page
, wbc
);
891 mutex_unlock(&shmem_swaplist_mutex
);
892 swapcache_free(swap
, NULL
);
894 set_page_dirty(page
);
895 if (wbc
->for_reclaim
)
896 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
903 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
907 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
908 return; /* show nothing */
910 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
912 seq_printf(seq
, ",mpol=%s", buffer
);
915 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
917 struct mempolicy
*mpol
= NULL
;
919 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
922 spin_unlock(&sbinfo
->stat_lock
);
926 #endif /* CONFIG_TMPFS */
928 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
929 struct shmem_inode_info
*info
, pgoff_t index
)
931 struct vm_area_struct pvma
;
934 /* Create a pseudo vma that just contains the policy */
936 /* Bias interleave by inode number to distribute better across nodes */
937 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
939 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
941 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
943 /* Drop reference taken by mpol_shared_policy_lookup() */
944 mpol_cond_put(pvma
.vm_policy
);
949 static struct page
*shmem_alloc_page(gfp_t gfp
,
950 struct shmem_inode_info
*info
, pgoff_t index
)
952 struct vm_area_struct pvma
;
955 /* Create a pseudo vma that just contains the policy */
957 /* Bias interleave by inode number to distribute better across nodes */
958 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
960 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
962 page
= alloc_page_vma(gfp
, &pvma
, 0);
963 if (page
&& is_cma_pageblock(page
)) {
965 page
= alloc_pages(gfp
& ~__GFP_MOVABLE
, 0);
968 /* Drop reference taken by mpol_shared_policy_lookup() */
969 mpol_cond_put(pvma
.vm_policy
);
973 #else /* !CONFIG_NUMA */
975 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
978 #endif /* CONFIG_TMPFS */
980 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
981 struct shmem_inode_info
*info
, pgoff_t index
)
983 return swapin_readahead(swap
, gfp
, NULL
, 0);
986 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
987 struct shmem_inode_info
*info
, pgoff_t index
)
989 return alloc_page(gfp
);
991 #endif /* CONFIG_NUMA */
993 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
994 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
1001 * When a page is moved from swapcache to shmem filecache (either by the
1002 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1003 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1004 * ignorance of the mapping it belongs to. If that mapping has special
1005 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1006 * we may need to copy to a suitable page before moving to filecache.
1008 * In a future release, this may well be extended to respect cpuset and
1009 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1010 * but for now it is a simple matter of zone.
1012 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1014 return page_zonenum(page
) > gfp_zone(gfp
);
1017 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1018 struct shmem_inode_info
*info
, pgoff_t index
)
1020 struct page
*oldpage
, *newpage
;
1021 struct address_space
*swap_mapping
;
1026 swap_index
= page_private(oldpage
);
1027 swap_mapping
= page_mapping(oldpage
);
1030 * We have arrived here because our zones are constrained, so don't
1031 * limit chance of success by further cpuset and node constraints.
1033 gfp
&= ~GFP_CONSTRAINT_MASK
;
1034 newpage
= shmem_alloc_page(gfp
, info
, index
);
1038 page_cache_get(newpage
);
1039 copy_highpage(newpage
, oldpage
);
1040 flush_dcache_page(newpage
);
1042 __set_page_locked(newpage
);
1043 SetPageUptodate(newpage
);
1044 SetPageSwapBacked(newpage
);
1045 set_page_private(newpage
, swap_index
);
1046 SetPageSwapCache(newpage
);
1049 * Our caller will very soon move newpage out of swapcache, but it's
1050 * a nice clean interface for us to replace oldpage by newpage there.
1052 spin_lock_irq(&swap_mapping
->tree_lock
);
1053 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1056 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1057 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1059 spin_unlock_irq(&swap_mapping
->tree_lock
);
1061 if (unlikely(error
)) {
1063 * Is this possible? I think not, now that our callers check
1064 * both PageSwapCache and page_private after getting page lock;
1065 * but be defensive. Reverse old to newpage for clear and free.
1069 mem_cgroup_replace_page_cache(oldpage
, newpage
);
1070 lru_cache_add_anon(newpage
);
1074 ClearPageSwapCache(oldpage
);
1075 set_page_private(oldpage
, 0);
1077 unlock_page(oldpage
);
1078 page_cache_release(oldpage
);
1079 page_cache_release(oldpage
);
1084 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1086 * If we allocate a new one we do not mark it dirty. That's up to the
1087 * vm. If we swap it in we mark it dirty since we also free the swap
1088 * entry since a page cannot live in both the swap and page cache
1090 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1091 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1093 struct address_space
*mapping
= inode
->i_mapping
;
1094 struct shmem_inode_info
*info
;
1095 struct shmem_sb_info
*sbinfo
;
1102 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1106 page
= find_lock_page(mapping
, index
);
1107 if (radix_tree_exceptional_entry(page
)) {
1108 swap
= radix_to_swp_entry(page
);
1112 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1113 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1118 /* fallocated page? */
1119 if (page
&& !PageUptodate(page
)) {
1120 if (sgp
!= SGP_READ
)
1123 page_cache_release(page
);
1126 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1132 * Fast cache lookup did not find it:
1133 * bring it back from swap or allocate.
1135 info
= SHMEM_I(inode
);
1136 sbinfo
= SHMEM_SB(inode
->i_sb
);
1139 /* Look it up and read it in.. */
1140 page
= lookup_swap_cache(swap
);
1142 /* here we actually do the io */
1144 *fault_type
|= VM_FAULT_MAJOR
;
1145 page
= shmem_swapin(swap
, gfp
, info
, index
);
1152 /* We have to do this with page locked to prevent races */
1154 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1155 !shmem_confirm_swap(mapping
, index
, swap
)) {
1156 error
= -EEXIST
; /* try again */
1159 if (!PageUptodate(page
)) {
1163 wait_on_page_writeback(page
);
1165 if (shmem_should_replace_page(page
, gfp
)) {
1166 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1171 error
= mem_cgroup_cache_charge(page
, current
->mm
,
1172 gfp
& GFP_RECLAIM_MASK
);
1174 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1175 gfp
, swp_to_radix_entry(swap
));
1177 * We already confirmed swap under page lock, and make
1178 * no memory allocation here, so usually no possibility
1179 * of error; but free_swap_and_cache() only trylocks a
1180 * page, so it is just possible that the entry has been
1181 * truncated or holepunched since swap was confirmed.
1182 * shmem_undo_range() will have done some of the
1183 * unaccounting, now delete_from_swap_cache() will do
1184 * the rest (including mem_cgroup_uncharge_swapcache).
1185 * Reset swap.val? No, leave it so "failed" goes back to
1186 * "repeat": reading a hole and writing should succeed.
1189 delete_from_swap_cache(page
);
1194 spin_lock(&info
->lock
);
1196 shmem_recalc_inode(inode
);
1197 spin_unlock(&info
->lock
);
1199 delete_from_swap_cache(page
);
1200 set_page_dirty(page
);
1204 if (shmem_acct_block(info
->flags
)) {
1208 if (sbinfo
->max_blocks
) {
1209 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1210 sbinfo
->max_blocks
) >= 0) {
1214 percpu_counter_inc(&sbinfo
->used_blocks
);
1217 page
= shmem_alloc_page(gfp
, info
, index
);
1223 SetPageSwapBacked(page
);
1224 __set_page_locked(page
);
1225 error
= mem_cgroup_cache_charge(page
, current
->mm
,
1226 gfp
& GFP_RECLAIM_MASK
);
1229 error
= radix_tree_preload(gfp
& GFP_RECLAIM_MASK
);
1231 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1233 radix_tree_preload_end();
1236 mem_cgroup_uncharge_cache_page(page
);
1239 lru_cache_add_anon(page
);
1241 spin_lock(&info
->lock
);
1243 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1244 shmem_recalc_inode(inode
);
1245 spin_unlock(&info
->lock
);
1249 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1251 if (sgp
== SGP_FALLOC
)
1255 * Let SGP_WRITE caller clear ends if write does not fill page;
1256 * but SGP_FALLOC on a page fallocated earlier must initialize
1257 * it now, lest undo on failure cancel our earlier guarantee.
1259 if (sgp
!= SGP_WRITE
) {
1260 clear_highpage(page
);
1261 flush_dcache_page(page
);
1262 SetPageUptodate(page
);
1264 if (sgp
== SGP_DIRTY
)
1265 set_page_dirty(page
);
1268 /* Perhaps the file has been truncated since we checked */
1269 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1270 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1284 info
= SHMEM_I(inode
);
1285 ClearPageDirty(page
);
1286 delete_from_page_cache(page
);
1287 spin_lock(&info
->lock
);
1289 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1290 spin_unlock(&info
->lock
);
1292 sbinfo
= SHMEM_SB(inode
->i_sb
);
1293 if (sbinfo
->max_blocks
)
1294 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1296 shmem_unacct_blocks(info
->flags
, 1);
1298 if (swap
.val
&& error
!= -EINVAL
&&
1299 !shmem_confirm_swap(mapping
, index
, swap
))
1304 page_cache_release(page
);
1306 if (error
== -ENOSPC
&& !once
++) {
1307 info
= SHMEM_I(inode
);
1308 spin_lock(&info
->lock
);
1309 shmem_recalc_inode(inode
);
1310 spin_unlock(&info
->lock
);
1313 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1318 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1320 struct inode
*inode
= file_inode(vma
->vm_file
);
1322 int ret
= VM_FAULT_LOCKED
;
1325 * Trinity finds that probing a hole which tmpfs is punching can
1326 * prevent the hole-punch from ever completing: which in turn
1327 * locks writers out with its hold on i_mutex. So refrain from
1328 * faulting pages into the hole while it's being punched. Although
1329 * shmem_undo_range() does remove the additions, it may be unable to
1330 * keep up, as each new page needs its own unmap_mapping_range() call,
1331 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1333 * It does not matter if we sometimes reach this check just before the
1334 * hole-punch begins, so that one fault then races with the punch:
1335 * we just need to make racing faults a rare case.
1337 * The implementation below would be much simpler if we just used a
1338 * standard mutex or completion: but we cannot take i_mutex in fault,
1339 * and bloating every shmem inode for this unlikely case would be sad.
1341 if (unlikely(inode
->i_private
)) {
1342 struct shmem_falloc
*shmem_falloc
;
1344 spin_lock(&inode
->i_lock
);
1345 shmem_falloc
= inode
->i_private
;
1347 shmem_falloc
->waitq
&&
1348 vmf
->pgoff
>= shmem_falloc
->start
&&
1349 vmf
->pgoff
< shmem_falloc
->next
) {
1350 wait_queue_head_t
*shmem_falloc_waitq
;
1351 DEFINE_WAIT(shmem_fault_wait
);
1353 ret
= VM_FAULT_NOPAGE
;
1354 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1355 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1356 /* It's polite to up mmap_sem if we can */
1357 up_read(&vma
->vm_mm
->mmap_sem
);
1358 ret
= VM_FAULT_RETRY
;
1361 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1362 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1363 TASK_UNINTERRUPTIBLE
);
1364 spin_unlock(&inode
->i_lock
);
1368 * shmem_falloc_waitq points into the shmem_fallocate()
1369 * stack of the hole-punching task: shmem_falloc_waitq
1370 * is usually invalid by the time we reach here, but
1371 * finish_wait() does not dereference it in that case;
1372 * though i_lock needed lest racing with wake_up_all().
1374 spin_lock(&inode
->i_lock
);
1375 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1376 spin_unlock(&inode
->i_lock
);
1379 spin_unlock(&inode
->i_lock
);
1382 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1384 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1386 if (ret
& VM_FAULT_MAJOR
) {
1387 count_vm_event(PGMAJFAULT
);
1388 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1394 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1396 struct inode
*inode
= file_inode(vma
->vm_file
);
1397 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1400 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1403 struct inode
*inode
= file_inode(vma
->vm_file
);
1406 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1407 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1411 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1413 struct inode
*inode
= file_inode(file
);
1414 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1415 int retval
= -ENOMEM
;
1417 spin_lock(&info
->lock
);
1418 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1419 if (!user_shm_lock(inode
->i_size
, user
))
1421 info
->flags
|= VM_LOCKED
;
1422 mapping_set_unevictable(file
->f_mapping
);
1424 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1425 user_shm_unlock(inode
->i_size
, user
);
1426 info
->flags
&= ~VM_LOCKED
;
1427 mapping_clear_unevictable(file
->f_mapping
);
1432 spin_unlock(&info
->lock
);
1436 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1438 file_accessed(file
);
1439 vma
->vm_ops
= &shmem_vm_ops
;
1443 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1444 umode_t mode
, dev_t dev
, unsigned long flags
)
1446 struct inode
*inode
;
1447 struct shmem_inode_info
*info
;
1448 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1450 if (shmem_reserve_inode(sb
))
1453 inode
= new_inode(sb
);
1455 inode
->i_ino
= get_next_ino();
1456 inode_init_owner(inode
, dir
, mode
);
1457 inode
->i_blocks
= 0;
1458 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1459 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1460 inode
->i_generation
= get_seconds();
1461 info
= SHMEM_I(inode
);
1462 memset(info
, 0, (char *)inode
- (char *)info
);
1463 spin_lock_init(&info
->lock
);
1464 info
->flags
= flags
& VM_NORESERVE
;
1465 INIT_LIST_HEAD(&info
->swaplist
);
1466 simple_xattrs_init(&info
->xattrs
);
1467 cache_no_acl(inode
);
1469 switch (mode
& S_IFMT
) {
1471 inode
->i_op
= &shmem_special_inode_operations
;
1472 init_special_inode(inode
, mode
, dev
);
1475 inode
->i_mapping
->a_ops
= &shmem_aops
;
1476 inode
->i_op
= &shmem_inode_operations
;
1477 inode
->i_fop
= &shmem_file_operations
;
1478 mpol_shared_policy_init(&info
->policy
,
1479 shmem_get_sbmpol(sbinfo
));
1483 /* Some things misbehave if size == 0 on a directory */
1484 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1485 inode
->i_op
= &shmem_dir_inode_operations
;
1486 inode
->i_fop
= &simple_dir_operations
;
1490 * Must not load anything in the rbtree,
1491 * mpol_free_shared_policy will not be called.
1493 mpol_shared_policy_init(&info
->policy
, NULL
);
1497 shmem_free_inode(sb
);
1502 static const struct inode_operations shmem_symlink_inode_operations
;
1503 static const struct inode_operations shmem_short_symlink_operations
;
1505 #ifdef CONFIG_TMPFS_XATTR
1506 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1508 #define shmem_initxattrs NULL
1512 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1513 loff_t pos
, unsigned len
, unsigned flags
,
1514 struct page
**pagep
, void **fsdata
)
1516 struct inode
*inode
= mapping
->host
;
1517 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1518 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1522 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1523 loff_t pos
, unsigned len
, unsigned copied
,
1524 struct page
*page
, void *fsdata
)
1526 struct inode
*inode
= mapping
->host
;
1528 if (pos
+ copied
> inode
->i_size
)
1529 i_size_write(inode
, pos
+ copied
);
1531 if (!PageUptodate(page
)) {
1532 if (copied
< PAGE_CACHE_SIZE
) {
1533 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1534 zero_user_segments(page
, 0, from
,
1535 from
+ copied
, PAGE_CACHE_SIZE
);
1537 SetPageUptodate(page
);
1539 set_page_dirty(page
);
1541 page_cache_release(page
);
1546 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1548 struct inode
*inode
= file_inode(filp
);
1549 struct address_space
*mapping
= inode
->i_mapping
;
1551 unsigned long offset
;
1552 enum sgp_type sgp
= SGP_READ
;
1555 * Might this read be for a stacking filesystem? Then when reading
1556 * holes of a sparse file, we actually need to allocate those pages,
1557 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1559 if (segment_eq(get_fs(), KERNEL_DS
))
1562 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1563 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1566 struct page
*page
= NULL
;
1568 unsigned long nr
, ret
;
1569 loff_t i_size
= i_size_read(inode
);
1571 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1572 if (index
> end_index
)
1574 if (index
== end_index
) {
1575 nr
= i_size
& ~PAGE_CACHE_MASK
;
1580 desc
->error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1582 if (desc
->error
== -EINVAL
)
1590 * We must evaluate after, since reads (unlike writes)
1591 * are called without i_mutex protection against truncate
1593 nr
= PAGE_CACHE_SIZE
;
1594 i_size
= i_size_read(inode
);
1595 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1596 if (index
== end_index
) {
1597 nr
= i_size
& ~PAGE_CACHE_MASK
;
1600 page_cache_release(page
);
1608 * If users can be writing to this page using arbitrary
1609 * virtual addresses, take care about potential aliasing
1610 * before reading the page on the kernel side.
1612 if (mapping_writably_mapped(mapping
))
1613 flush_dcache_page(page
);
1615 * Mark the page accessed if we read the beginning.
1618 mark_page_accessed(page
);
1620 page
= ZERO_PAGE(0);
1621 page_cache_get(page
);
1625 * Ok, we have the page, and it's up-to-date, so
1626 * now we can copy it to user space...
1628 * The actor routine returns how many bytes were actually used..
1629 * NOTE! This may not be the same as how much of a user buffer
1630 * we filled up (we may be padding etc), so we can only update
1631 * "pos" here (the actor routine has to update the user buffer
1632 * pointers and the remaining count).
1634 ret
= actor(desc
, page
, offset
, nr
);
1636 index
+= offset
>> PAGE_CACHE_SHIFT
;
1637 offset
&= ~PAGE_CACHE_MASK
;
1639 page_cache_release(page
);
1640 if (ret
!= nr
|| !desc
->count
)
1646 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1647 file_accessed(filp
);
1650 static ssize_t
shmem_file_aio_read(struct kiocb
*iocb
,
1651 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
1653 struct file
*filp
= iocb
->ki_filp
;
1657 loff_t
*ppos
= &iocb
->ki_pos
;
1659 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1663 for (seg
= 0; seg
< nr_segs
; seg
++) {
1664 read_descriptor_t desc
;
1667 desc
.arg
.buf
= iov
[seg
].iov_base
;
1668 desc
.count
= iov
[seg
].iov_len
;
1669 if (desc
.count
== 0)
1672 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1673 retval
+= desc
.written
;
1675 retval
= retval
?: desc
.error
;
1684 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1685 struct pipe_inode_info
*pipe
, size_t len
,
1688 struct address_space
*mapping
= in
->f_mapping
;
1689 struct inode
*inode
= mapping
->host
;
1690 unsigned int loff
, nr_pages
, req_pages
;
1691 struct page
*pages
[PIPE_DEF_BUFFERS
];
1692 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1694 pgoff_t index
, end_index
;
1697 struct splice_pipe_desc spd
= {
1700 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1702 .ops
= &page_cache_pipe_buf_ops
,
1703 .spd_release
= spd_release_page
,
1706 isize
= i_size_read(inode
);
1707 if (unlikely(*ppos
>= isize
))
1710 left
= isize
- *ppos
;
1711 if (unlikely(left
< len
))
1714 if (splice_grow_spd(pipe
, &spd
))
1717 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1718 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1719 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1720 nr_pages
= min(req_pages
, pipe
->buffers
);
1722 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1723 nr_pages
, spd
.pages
);
1724 index
+= spd
.nr_pages
;
1727 while (spd
.nr_pages
< nr_pages
) {
1728 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1732 spd
.pages
[spd
.nr_pages
++] = page
;
1736 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1737 nr_pages
= spd
.nr_pages
;
1740 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1741 unsigned int this_len
;
1746 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1747 page
= spd
.pages
[page_nr
];
1749 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1750 error
= shmem_getpage(inode
, index
, &page
,
1755 page_cache_release(spd
.pages
[page_nr
]);
1756 spd
.pages
[page_nr
] = page
;
1759 isize
= i_size_read(inode
);
1760 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1761 if (unlikely(!isize
|| index
> end_index
))
1764 if (end_index
== index
) {
1767 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1771 this_len
= min(this_len
, plen
- loff
);
1775 spd
.partial
[page_nr
].offset
= loff
;
1776 spd
.partial
[page_nr
].len
= this_len
;
1783 while (page_nr
< nr_pages
)
1784 page_cache_release(spd
.pages
[page_nr
++]);
1787 error
= splice_to_pipe(pipe
, &spd
);
1789 splice_shrink_spd(&spd
);
1799 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1801 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1802 pgoff_t index
, pgoff_t end
, int whence
)
1805 struct pagevec pvec
;
1806 pgoff_t indices
[PAGEVEC_SIZE
];
1810 pagevec_init(&pvec
, 0);
1811 pvec
.nr
= 1; /* start small: we may be there already */
1813 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
1814 pvec
.nr
, pvec
.pages
, indices
);
1816 if (whence
== SEEK_DATA
)
1820 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1821 if (index
< indices
[i
]) {
1822 if (whence
== SEEK_HOLE
) {
1828 page
= pvec
.pages
[i
];
1829 if (page
&& !radix_tree_exceptional_entry(page
)) {
1830 if (!PageUptodate(page
))
1834 (page
&& whence
== SEEK_DATA
) ||
1835 (!page
&& whence
== SEEK_HOLE
)) {
1840 shmem_deswap_pagevec(&pvec
);
1841 pagevec_release(&pvec
);
1842 pvec
.nr
= PAGEVEC_SIZE
;
1848 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1850 struct address_space
*mapping
= file
->f_mapping
;
1851 struct inode
*inode
= mapping
->host
;
1855 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1856 return generic_file_llseek_size(file
, offset
, whence
,
1857 MAX_LFS_FILESIZE
, i_size_read(inode
));
1858 mutex_lock(&inode
->i_mutex
);
1859 /* We're holding i_mutex so we can access i_size directly */
1863 else if (offset
>= inode
->i_size
)
1866 start
= offset
>> PAGE_CACHE_SHIFT
;
1867 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1868 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1869 new_offset
<<= PAGE_CACHE_SHIFT
;
1870 if (new_offset
> offset
) {
1871 if (new_offset
< inode
->i_size
)
1872 offset
= new_offset
;
1873 else if (whence
== SEEK_DATA
)
1876 offset
= inode
->i_size
;
1880 if (offset
>= 0 && offset
!= file
->f_pos
) {
1881 file
->f_pos
= offset
;
1882 file
->f_version
= 0;
1884 mutex_unlock(&inode
->i_mutex
);
1888 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
1891 struct inode
*inode
= file_inode(file
);
1892 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1893 struct shmem_falloc shmem_falloc
;
1894 pgoff_t start
, index
, end
;
1897 mutex_lock(&inode
->i_mutex
);
1899 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1900 struct address_space
*mapping
= file
->f_mapping
;
1901 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
1902 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
1903 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
1905 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
1906 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
1907 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
1908 spin_lock(&inode
->i_lock
);
1909 inode
->i_private
= &shmem_falloc
;
1910 spin_unlock(&inode
->i_lock
);
1912 if ((u64
)unmap_end
> (u64
)unmap_start
)
1913 unmap_mapping_range(mapping
, unmap_start
,
1914 1 + unmap_end
- unmap_start
, 0);
1915 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
1916 /* No need to unmap again: hole-punching leaves COWed pages */
1918 spin_lock(&inode
->i_lock
);
1919 inode
->i_private
= NULL
;
1920 wake_up_all(&shmem_falloc_waitq
);
1921 spin_unlock(&inode
->i_lock
);
1926 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1927 error
= inode_newsize_ok(inode
, offset
+ len
);
1931 start
= offset
>> PAGE_CACHE_SHIFT
;
1932 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1933 /* Try to avoid a swapstorm if len is impossible to satisfy */
1934 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
1939 shmem_falloc
.waitq
= NULL
;
1940 shmem_falloc
.start
= start
;
1941 shmem_falloc
.next
= start
;
1942 shmem_falloc
.nr_falloced
= 0;
1943 shmem_falloc
.nr_unswapped
= 0;
1944 spin_lock(&inode
->i_lock
);
1945 inode
->i_private
= &shmem_falloc
;
1946 spin_unlock(&inode
->i_lock
);
1948 for (index
= start
; index
< end
; index
++) {
1952 * Good, the fallocate(2) manpage permits EINTR: we may have
1953 * been interrupted because we are using up too much memory.
1955 if (signal_pending(current
))
1957 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
1960 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
1963 /* Remove the !PageUptodate pages we added */
1964 if (index
> start
) {
1965 shmem_undo_range(inode
,
1966 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
1967 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) - 1, true);
1973 * Inform shmem_writepage() how far we have reached.
1974 * No need for lock or barrier: we have the page lock.
1976 shmem_falloc
.next
++;
1977 if (!PageUptodate(page
))
1978 shmem_falloc
.nr_falloced
++;
1981 * If !PageUptodate, leave it that way so that freeable pages
1982 * can be recognized if we need to rollback on error later.
1983 * But set_page_dirty so that memory pressure will swap rather
1984 * than free the pages we are allocating (and SGP_CACHE pages
1985 * might still be clean: we now need to mark those dirty too).
1987 set_page_dirty(page
);
1989 page_cache_release(page
);
1993 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
1994 i_size_write(inode
, offset
+ len
);
1995 inode
->i_ctime
= CURRENT_TIME
;
1997 spin_lock(&inode
->i_lock
);
1998 inode
->i_private
= NULL
;
1999 spin_unlock(&inode
->i_lock
);
2001 mutex_unlock(&inode
->i_mutex
);
2005 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2007 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2009 buf
->f_type
= TMPFS_MAGIC
;
2010 buf
->f_bsize
= PAGE_CACHE_SIZE
;
2011 buf
->f_namelen
= NAME_MAX
;
2012 if (sbinfo
->max_blocks
) {
2013 buf
->f_blocks
= sbinfo
->max_blocks
;
2015 buf
->f_bfree
= sbinfo
->max_blocks
-
2016 percpu_counter_sum(&sbinfo
->used_blocks
);
2018 if (sbinfo
->max_inodes
) {
2019 buf
->f_files
= sbinfo
->max_inodes
;
2020 buf
->f_ffree
= sbinfo
->free_inodes
;
2022 /* else leave those fields 0 like simple_statfs */
2027 * File creation. Allocate an inode, and we're done..
2030 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2032 struct inode
*inode
;
2033 int error
= -ENOSPC
;
2035 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
2037 error
= security_inode_init_security(inode
, dir
,
2039 shmem_initxattrs
, NULL
);
2041 if (error
!= -EOPNOTSUPP
) {
2046 #ifdef CONFIG_TMPFS_POSIX_ACL
2047 error
= generic_acl_init(inode
, dir
);
2055 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2056 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2057 d_instantiate(dentry
, inode
);
2058 dget(dentry
); /* Extra count - pin the dentry in core */
2064 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2066 struct inode
*inode
;
2067 int error
= -ENOSPC
;
2069 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
2071 error
= security_inode_init_security(inode
, dir
,
2073 shmem_initxattrs
, NULL
);
2075 if (error
!= -EOPNOTSUPP
) {
2080 #ifdef CONFIG_TMPFS_POSIX_ACL
2081 error
= generic_acl_init(inode
, dir
);
2089 d_tmpfile(dentry
, inode
);
2094 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2098 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2104 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2107 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2113 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2115 struct inode
*inode
= old_dentry
->d_inode
;
2119 * No ordinary (disk based) filesystem counts links as inodes;
2120 * but each new link needs a new dentry, pinning lowmem, and
2121 * tmpfs dentries cannot be pruned until they are unlinked.
2123 ret
= shmem_reserve_inode(inode
->i_sb
);
2127 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2128 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2130 ihold(inode
); /* New dentry reference */
2131 dget(dentry
); /* Extra pinning count for the created dentry */
2132 d_instantiate(dentry
, inode
);
2137 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2139 struct inode
*inode
= dentry
->d_inode
;
2141 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2142 shmem_free_inode(inode
->i_sb
);
2144 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2145 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2147 dput(dentry
); /* Undo the count from "create" - this does all the work */
2151 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2153 if (!simple_empty(dentry
))
2156 drop_nlink(dentry
->d_inode
);
2158 return shmem_unlink(dir
, dentry
);
2162 * The VFS layer already does all the dentry stuff for rename,
2163 * we just have to decrement the usage count for the target if
2164 * it exists so that the VFS layer correctly free's it when it
2167 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2169 struct inode
*inode
= old_dentry
->d_inode
;
2170 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2172 if (!simple_empty(new_dentry
))
2175 if (new_dentry
->d_inode
) {
2176 (void) shmem_unlink(new_dir
, new_dentry
);
2177 if (they_are_dirs
) {
2178 drop_nlink(new_dentry
->d_inode
);
2179 drop_nlink(old_dir
);
2181 } else if (they_are_dirs
) {
2182 drop_nlink(old_dir
);
2186 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2187 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2188 old_dir
->i_ctime
= old_dir
->i_mtime
=
2189 new_dir
->i_ctime
= new_dir
->i_mtime
=
2190 inode
->i_ctime
= CURRENT_TIME
;
2194 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2198 struct inode
*inode
;
2201 struct shmem_inode_info
*info
;
2203 len
= strlen(symname
) + 1;
2204 if (len
> PAGE_CACHE_SIZE
)
2205 return -ENAMETOOLONG
;
2207 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2211 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2212 shmem_initxattrs
, NULL
);
2214 if (error
!= -EOPNOTSUPP
) {
2221 info
= SHMEM_I(inode
);
2222 inode
->i_size
= len
-1;
2223 if (len
<= SHORT_SYMLINK_LEN
) {
2224 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2225 if (!info
->symlink
) {
2229 inode
->i_op
= &shmem_short_symlink_operations
;
2231 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2236 inode
->i_mapping
->a_ops
= &shmem_aops
;
2237 inode
->i_op
= &shmem_symlink_inode_operations
;
2238 kaddr
= kmap_atomic(page
);
2239 memcpy(kaddr
, symname
, len
);
2240 kunmap_atomic(kaddr
);
2241 SetPageUptodate(page
);
2242 set_page_dirty(page
);
2244 page_cache_release(page
);
2246 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2247 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2248 d_instantiate(dentry
, inode
);
2253 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
2255 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
2259 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
2261 struct page
*page
= NULL
;
2262 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
2263 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
2269 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2271 if (!IS_ERR(nd_get_link(nd
))) {
2272 struct page
*page
= cookie
;
2274 mark_page_accessed(page
);
2275 page_cache_release(page
);
2279 #ifdef CONFIG_TMPFS_XATTR
2281 * Superblocks without xattr inode operations may get some security.* xattr
2282 * support from the LSM "for free". As soon as we have any other xattrs
2283 * like ACLs, we also need to implement the security.* handlers at
2284 * filesystem level, though.
2288 * Callback for security_inode_init_security() for acquiring xattrs.
2290 static int shmem_initxattrs(struct inode
*inode
,
2291 const struct xattr
*xattr_array
,
2294 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2295 const struct xattr
*xattr
;
2296 struct simple_xattr
*new_xattr
;
2299 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2300 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2304 len
= strlen(xattr
->name
) + 1;
2305 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2307 if (!new_xattr
->name
) {
2312 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2313 XATTR_SECURITY_PREFIX_LEN
);
2314 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2317 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2323 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2324 #ifdef CONFIG_TMPFS_POSIX_ACL
2325 &generic_acl_access_handler
,
2326 &generic_acl_default_handler
,
2331 static int shmem_xattr_validate(const char *name
)
2333 struct { const char *prefix
; size_t len
; } arr
[] = {
2334 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2335 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2339 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2340 size_t preflen
= arr
[i
].len
;
2341 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2350 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2351 void *buffer
, size_t size
)
2353 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2357 * If this is a request for a synthetic attribute in the system.*
2358 * namespace use the generic infrastructure to resolve a handler
2359 * for it via sb->s_xattr.
2361 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2362 return generic_getxattr(dentry
, name
, buffer
, size
);
2364 err
= shmem_xattr_validate(name
);
2368 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2371 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2372 const void *value
, size_t size
, int flags
)
2374 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2378 * If this is a request for a synthetic attribute in the system.*
2379 * namespace use the generic infrastructure to resolve a handler
2380 * for it via sb->s_xattr.
2382 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2383 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2385 err
= shmem_xattr_validate(name
);
2389 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2392 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2394 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2398 * If this is a request for a synthetic attribute in the system.*
2399 * namespace use the generic infrastructure to resolve a handler
2400 * for it via sb->s_xattr.
2402 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2403 return generic_removexattr(dentry
, name
);
2405 err
= shmem_xattr_validate(name
);
2409 return simple_xattr_remove(&info
->xattrs
, name
);
2412 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2414 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2415 return simple_xattr_list(&info
->xattrs
, buffer
, size
);
2417 #endif /* CONFIG_TMPFS_XATTR */
2419 static const struct inode_operations shmem_short_symlink_operations
= {
2420 .readlink
= generic_readlink
,
2421 .follow_link
= shmem_follow_short_symlink
,
2422 #ifdef CONFIG_TMPFS_XATTR
2423 .setxattr
= shmem_setxattr
,
2424 .getxattr
= shmem_getxattr
,
2425 .listxattr
= shmem_listxattr
,
2426 .removexattr
= shmem_removexattr
,
2430 static const struct inode_operations shmem_symlink_inode_operations
= {
2431 .readlink
= generic_readlink
,
2432 .follow_link
= shmem_follow_link
,
2433 .put_link
= shmem_put_link
,
2434 #ifdef CONFIG_TMPFS_XATTR
2435 .setxattr
= shmem_setxattr
,
2436 .getxattr
= shmem_getxattr
,
2437 .listxattr
= shmem_listxattr
,
2438 .removexattr
= shmem_removexattr
,
2442 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2444 return ERR_PTR(-ESTALE
);
2447 static int shmem_match(struct inode
*ino
, void *vfh
)
2451 inum
= (inum
<< 32) | fh
[1];
2452 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2455 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2456 struct fid
*fid
, int fh_len
, int fh_type
)
2458 struct inode
*inode
;
2459 struct dentry
*dentry
= NULL
;
2466 inum
= (inum
<< 32) | fid
->raw
[1];
2468 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2469 shmem_match
, fid
->raw
);
2471 dentry
= d_find_alias(inode
);
2478 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2479 struct inode
*parent
)
2483 return FILEID_INVALID
;
2486 if (inode_unhashed(inode
)) {
2487 /* Unfortunately insert_inode_hash is not idempotent,
2488 * so as we hash inodes here rather than at creation
2489 * time, we need a lock to ensure we only try
2492 static DEFINE_SPINLOCK(lock
);
2494 if (inode_unhashed(inode
))
2495 __insert_inode_hash(inode
,
2496 inode
->i_ino
+ inode
->i_generation
);
2500 fh
[0] = inode
->i_generation
;
2501 fh
[1] = inode
->i_ino
;
2502 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2508 static const struct export_operations shmem_export_ops
= {
2509 .get_parent
= shmem_get_parent
,
2510 .encode_fh
= shmem_encode_fh
,
2511 .fh_to_dentry
= shmem_fh_to_dentry
,
2514 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2517 char *this_char
, *value
, *rest
;
2518 struct mempolicy
*mpol
= NULL
;
2522 while (options
!= NULL
) {
2523 this_char
= options
;
2526 * NUL-terminate this option: unfortunately,
2527 * mount options form a comma-separated list,
2528 * but mpol's nodelist may also contain commas.
2530 options
= strchr(options
, ',');
2531 if (options
== NULL
)
2534 if (!isdigit(*options
)) {
2541 if ((value
= strchr(this_char
,'=')) != NULL
) {
2545 "tmpfs: No value for mount option '%s'\n",
2550 if (!strcmp(this_char
,"size")) {
2551 unsigned long long size
;
2552 size
= memparse(value
,&rest
);
2554 size
<<= PAGE_SHIFT
;
2555 size
*= totalram_pages
;
2561 sbinfo
->max_blocks
=
2562 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2563 } else if (!strcmp(this_char
,"nr_blocks")) {
2564 sbinfo
->max_blocks
= memparse(value
, &rest
);
2567 } else if (!strcmp(this_char
,"nr_inodes")) {
2568 sbinfo
->max_inodes
= memparse(value
, &rest
);
2571 } else if (!strcmp(this_char
,"mode")) {
2574 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2577 } else if (!strcmp(this_char
,"uid")) {
2580 uid
= simple_strtoul(value
, &rest
, 0);
2583 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2584 if (!uid_valid(sbinfo
->uid
))
2586 } else if (!strcmp(this_char
,"gid")) {
2589 gid
= simple_strtoul(value
, &rest
, 0);
2592 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2593 if (!gid_valid(sbinfo
->gid
))
2595 } else if (!strcmp(this_char
,"mpol")) {
2598 if (mpol_parse_str(value
, &mpol
))
2601 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2606 sbinfo
->mpol
= mpol
;
2610 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2618 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2620 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2621 struct shmem_sb_info config
= *sbinfo
;
2622 unsigned long inodes
;
2623 int error
= -EINVAL
;
2626 if (shmem_parse_options(data
, &config
, true))
2629 spin_lock(&sbinfo
->stat_lock
);
2630 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2631 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2633 if (config
.max_inodes
< inodes
)
2636 * Those tests disallow limited->unlimited while any are in use;
2637 * but we must separately disallow unlimited->limited, because
2638 * in that case we have no record of how much is already in use.
2640 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2642 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2646 sbinfo
->max_blocks
= config
.max_blocks
;
2647 sbinfo
->max_inodes
= config
.max_inodes
;
2648 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2651 * Preserve previous mempolicy unless mpol remount option was specified.
2654 mpol_put(sbinfo
->mpol
);
2655 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2658 spin_unlock(&sbinfo
->stat_lock
);
2662 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2664 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2666 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2667 seq_printf(seq
, ",size=%luk",
2668 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2669 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2670 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2671 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2672 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2673 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2674 seq_printf(seq
, ",uid=%u",
2675 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2676 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2677 seq_printf(seq
, ",gid=%u",
2678 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2679 shmem_show_mpol(seq
, sbinfo
->mpol
);
2682 #endif /* CONFIG_TMPFS */
2684 static void shmem_put_super(struct super_block
*sb
)
2686 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2688 percpu_counter_destroy(&sbinfo
->used_blocks
);
2689 mpol_put(sbinfo
->mpol
);
2691 sb
->s_fs_info
= NULL
;
2694 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2696 struct inode
*inode
;
2697 struct shmem_sb_info
*sbinfo
;
2700 /* Round up to L1_CACHE_BYTES to resist false sharing */
2701 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2702 L1_CACHE_BYTES
), GFP_KERNEL
);
2706 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2707 sbinfo
->uid
= current_fsuid();
2708 sbinfo
->gid
= current_fsgid();
2709 sb
->s_fs_info
= sbinfo
;
2713 * Per default we only allow half of the physical ram per
2714 * tmpfs instance, limiting inodes to one per page of lowmem;
2715 * but the internal instance is left unlimited.
2717 if (!(sb
->s_flags
& MS_NOUSER
)) {
2718 sbinfo
->max_blocks
= shmem_default_max_blocks();
2719 sbinfo
->max_inodes
= shmem_default_max_inodes();
2720 if (shmem_parse_options(data
, sbinfo
, false)) {
2725 sb
->s_export_op
= &shmem_export_ops
;
2726 sb
->s_flags
|= MS_NOSEC
;
2728 sb
->s_flags
|= MS_NOUSER
;
2731 spin_lock_init(&sbinfo
->stat_lock
);
2732 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
2734 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2736 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
2737 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2738 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2739 sb
->s_magic
= TMPFS_MAGIC
;
2740 sb
->s_op
= &shmem_ops
;
2741 sb
->s_time_gran
= 1;
2742 #ifdef CONFIG_TMPFS_XATTR
2743 sb
->s_xattr
= shmem_xattr_handlers
;
2745 #ifdef CONFIG_TMPFS_POSIX_ACL
2746 sb
->s_flags
|= MS_POSIXACL
;
2749 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
2752 inode
->i_uid
= sbinfo
->uid
;
2753 inode
->i_gid
= sbinfo
->gid
;
2754 sb
->s_root
= d_make_root(inode
);
2760 shmem_put_super(sb
);
2764 static struct kmem_cache
*shmem_inode_cachep
;
2766 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2768 struct shmem_inode_info
*info
;
2769 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2772 return &info
->vfs_inode
;
2775 static void shmem_destroy_callback(struct rcu_head
*head
)
2777 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
2778 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2781 static void shmem_destroy_inode(struct inode
*inode
)
2783 if (S_ISREG(inode
->i_mode
))
2784 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2785 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
2788 static void shmem_init_inode(void *foo
)
2790 struct shmem_inode_info
*info
= foo
;
2791 inode_init_once(&info
->vfs_inode
);
2794 static int shmem_init_inodecache(void)
2796 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2797 sizeof(struct shmem_inode_info
),
2798 0, SLAB_PANIC
, shmem_init_inode
);
2802 static void shmem_destroy_inodecache(void)
2804 kmem_cache_destroy(shmem_inode_cachep
);
2807 static const struct address_space_operations shmem_aops
= {
2808 .writepage
= shmem_writepage
,
2809 .set_page_dirty
= __set_page_dirty_no_writeback
,
2811 .write_begin
= shmem_write_begin
,
2812 .write_end
= shmem_write_end
,
2814 .migratepage
= migrate_page
,
2815 .error_remove_page
= generic_error_remove_page
,
2818 static const struct file_operations shmem_file_operations
= {
2821 .llseek
= shmem_file_llseek
,
2822 .read
= do_sync_read
,
2823 .write
= do_sync_write
,
2824 .aio_read
= shmem_file_aio_read
,
2825 .aio_write
= generic_file_aio_write
,
2826 .fsync
= noop_fsync
,
2827 .splice_read
= shmem_file_splice_read
,
2828 .splice_write
= generic_file_splice_write
,
2829 .fallocate
= shmem_fallocate
,
2833 static const struct inode_operations shmem_inode_operations
= {
2834 .setattr
= shmem_setattr
,
2835 #ifdef CONFIG_TMPFS_XATTR
2836 .setxattr
= shmem_setxattr
,
2837 .getxattr
= shmem_getxattr
,
2838 .listxattr
= shmem_listxattr
,
2839 .removexattr
= shmem_removexattr
,
2843 static const struct inode_operations shmem_dir_inode_operations
= {
2845 .create
= shmem_create
,
2846 .lookup
= simple_lookup
,
2848 .unlink
= shmem_unlink
,
2849 .symlink
= shmem_symlink
,
2850 .mkdir
= shmem_mkdir
,
2851 .rmdir
= shmem_rmdir
,
2852 .mknod
= shmem_mknod
,
2853 .rename
= shmem_rename
,
2854 .tmpfile
= shmem_tmpfile
,
2856 #ifdef CONFIG_TMPFS_XATTR
2857 .setxattr
= shmem_setxattr
,
2858 .getxattr
= shmem_getxattr
,
2859 .listxattr
= shmem_listxattr
,
2860 .removexattr
= shmem_removexattr
,
2862 #ifdef CONFIG_TMPFS_POSIX_ACL
2863 .setattr
= shmem_setattr
,
2867 static const struct inode_operations shmem_special_inode_operations
= {
2868 #ifdef CONFIG_TMPFS_XATTR
2869 .setxattr
= shmem_setxattr
,
2870 .getxattr
= shmem_getxattr
,
2871 .listxattr
= shmem_listxattr
,
2872 .removexattr
= shmem_removexattr
,
2874 #ifdef CONFIG_TMPFS_POSIX_ACL
2875 .setattr
= shmem_setattr
,
2879 static const struct super_operations shmem_ops
= {
2880 .alloc_inode
= shmem_alloc_inode
,
2881 .destroy_inode
= shmem_destroy_inode
,
2883 .statfs
= shmem_statfs
,
2884 .remount_fs
= shmem_remount_fs
,
2885 .show_options
= shmem_show_options
,
2887 .evict_inode
= shmem_evict_inode
,
2888 .drop_inode
= generic_delete_inode
,
2889 .put_super
= shmem_put_super
,
2892 static const struct vm_operations_struct shmem_vm_ops
= {
2893 .fault
= shmem_fault
,
2895 .set_policy
= shmem_set_policy
,
2896 .get_policy
= shmem_get_policy
,
2898 .remap_pages
= generic_file_remap_pages
,
2901 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
2902 int flags
, const char *dev_name
, void *data
)
2904 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
2907 static struct file_system_type shmem_fs_type
= {
2908 .owner
= THIS_MODULE
,
2910 .mount
= shmem_mount
,
2911 .kill_sb
= kill_litter_super
,
2912 .fs_flags
= FS_USERNS_MOUNT
,
2915 int __init
shmem_init(void)
2919 error
= bdi_init(&shmem_backing_dev_info
);
2923 error
= shmem_init_inodecache();
2927 error
= register_filesystem(&shmem_fs_type
);
2929 printk(KERN_ERR
"Could not register tmpfs\n");
2933 shm_mnt
= vfs_kern_mount(&shmem_fs_type
, MS_NOUSER
,
2934 shmem_fs_type
.name
, NULL
);
2935 if (IS_ERR(shm_mnt
)) {
2936 error
= PTR_ERR(shm_mnt
);
2937 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2943 unregister_filesystem(&shmem_fs_type
);
2945 shmem_destroy_inodecache();
2947 bdi_destroy(&shmem_backing_dev_info
);
2949 shm_mnt
= ERR_PTR(error
);
2953 #else /* !CONFIG_SHMEM */
2956 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2958 * This is intended for small system where the benefits of the full
2959 * shmem code (swap-backed and resource-limited) are outweighed by
2960 * their complexity. On systems without swap this code should be
2961 * effectively equivalent, but much lighter weight.
2964 static struct file_system_type shmem_fs_type
= {
2966 .mount
= ramfs_mount
,
2967 .kill_sb
= kill_litter_super
,
2968 .fs_flags
= FS_USERNS_MOUNT
,
2971 int __init
shmem_init(void)
2973 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
2975 shm_mnt
= kern_mount(&shmem_fs_type
);
2976 BUG_ON(IS_ERR(shm_mnt
));
2981 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
2986 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2991 void shmem_unlock_mapping(struct address_space
*mapping
)
2995 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
2997 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
2999 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
3001 #define shmem_vm_ops generic_file_vm_ops
3002 #define shmem_file_operations ramfs_file_operations
3003 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3004 #define shmem_acct_size(flags, size) 0
3005 #define shmem_unacct_size(flags, size) do {} while (0)
3007 #endif /* CONFIG_SHMEM */
3011 static struct dentry_operations anon_ops
= {
3012 .d_dname
= simple_dname
3016 * shmem_file_setup - get an unlinked file living in tmpfs
3017 * @name: name for dentry (to be seen in /proc/<pid>/maps
3018 * @size: size to be set for the file
3019 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3021 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3024 struct inode
*inode
;
3026 struct super_block
*sb
;
3029 if (IS_ERR(shm_mnt
))
3030 return ERR_CAST(shm_mnt
);
3032 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
3033 return ERR_PTR(-EINVAL
);
3035 if (shmem_acct_size(flags
, size
))
3036 return ERR_PTR(-ENOMEM
);
3038 res
= ERR_PTR(-ENOMEM
);
3040 this.len
= strlen(name
);
3041 this.hash
= 0; /* will go */
3042 sb
= shm_mnt
->mnt_sb
;
3043 path
.dentry
= d_alloc_pseudo(sb
, &this);
3046 d_set_d_op(path
.dentry
, &anon_ops
);
3047 path
.mnt
= mntget(shm_mnt
);
3049 res
= ERR_PTR(-ENOSPC
);
3050 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
3054 d_instantiate(path
.dentry
, inode
);
3055 inode
->i_size
= size
;
3056 clear_nlink(inode
); /* It is unlinked */
3057 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3061 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3062 &shmem_file_operations
);
3071 shmem_unacct_size(flags
, size
);
3074 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3076 void shmem_set_file(struct vm_area_struct
*vma
, struct file
*file
)
3080 vma
->vm_file
= file
;
3081 vma
->vm_ops
= &shmem_vm_ops
;
3085 * shmem_zero_setup - setup a shared anonymous mapping
3086 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3088 int shmem_zero_setup(struct vm_area_struct
*vma
)
3091 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3093 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
3095 return PTR_ERR(file
);
3097 shmem_set_file(vma
, file
);
3102 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3103 * @mapping: the page's address_space
3104 * @index: the page index
3105 * @gfp: the page allocator flags to use if allocating
3107 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3108 * with any new page allocations done using the specified allocation flags.
3109 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3110 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3111 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3113 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3114 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3116 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3117 pgoff_t index
, gfp_t gfp
)
3120 struct inode
*inode
= mapping
->host
;
3124 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3125 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3127 page
= ERR_PTR(error
);
3133 * The tiny !SHMEM case uses ramfs without swap
3135 return read_cache_page_gfp(mapping
, index
, gfp
);
3138 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);