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 and shmem_writepage communicate via inode->i_private
84 * (with i_mutex making sure that it has only one user at a time):
85 * we would prefer not to enlarge the shmem inode just for that.
88 pgoff_t start
; /* start of range currently being fallocated */
89 pgoff_t next
; /* the next page offset to be fallocated */
90 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
91 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
94 /* Flag allocation requirements to shmem_getpage */
96 SGP_READ
, /* don't exceed i_size, don't allocate page */
97 SGP_CACHE
, /* don't exceed i_size, may allocate page */
98 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
99 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
100 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
104 static unsigned long shmem_default_max_blocks(void)
106 return totalram_pages
/ 2;
109 static unsigned long shmem_default_max_inodes(void)
111 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
115 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
116 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
117 struct shmem_inode_info
*info
, pgoff_t index
);
118 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
119 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
121 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
122 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
124 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
125 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
128 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
130 return sb
->s_fs_info
;
134 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
135 * for shared memory and for shared anonymous (/dev/zero) mappings
136 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
137 * consistent with the pre-accounting of private mappings ...
139 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
141 return (flags
& VM_NORESERVE
) ?
142 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
145 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
147 if (!(flags
& VM_NORESERVE
))
148 vm_unacct_memory(VM_ACCT(size
));
152 * ... whereas tmpfs objects are accounted incrementally as
153 * pages are allocated, in order to allow huge sparse files.
154 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
155 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
157 static inline int shmem_acct_block(unsigned long flags
)
159 return (flags
& VM_NORESERVE
) ?
160 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
163 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
165 if (flags
& VM_NORESERVE
)
166 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
169 static const struct super_operations shmem_ops
;
170 static const struct address_space_operations shmem_aops
;
171 static const struct file_operations shmem_file_operations
;
172 static const struct inode_operations shmem_inode_operations
;
173 static const struct inode_operations shmem_dir_inode_operations
;
174 static const struct inode_operations shmem_special_inode_operations
;
175 static const struct vm_operations_struct shmem_vm_ops
;
177 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
178 .ra_pages
= 0, /* No readahead */
179 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
182 static LIST_HEAD(shmem_swaplist
);
183 static DEFINE_MUTEX(shmem_swaplist_mutex
);
185 static int shmem_reserve_inode(struct super_block
*sb
)
187 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
188 if (sbinfo
->max_inodes
) {
189 spin_lock(&sbinfo
->stat_lock
);
190 if (!sbinfo
->free_inodes
) {
191 spin_unlock(&sbinfo
->stat_lock
);
194 sbinfo
->free_inodes
--;
195 spin_unlock(&sbinfo
->stat_lock
);
200 static void shmem_free_inode(struct super_block
*sb
)
202 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
203 if (sbinfo
->max_inodes
) {
204 spin_lock(&sbinfo
->stat_lock
);
205 sbinfo
->free_inodes
++;
206 spin_unlock(&sbinfo
->stat_lock
);
211 * shmem_recalc_inode - recalculate the block usage of an inode
212 * @inode: inode to recalc
214 * We have to calculate the free blocks since the mm can drop
215 * undirtied hole pages behind our back.
217 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
218 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
220 * It has to be called with the spinlock held.
222 static void shmem_recalc_inode(struct inode
*inode
)
224 struct shmem_inode_info
*info
= SHMEM_I(inode
);
227 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
229 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
230 if (sbinfo
->max_blocks
)
231 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
232 info
->alloced
-= freed
;
233 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
234 shmem_unacct_blocks(info
->flags
, freed
);
239 * Replace item expected in radix tree by a new item, while holding tree lock.
241 static int shmem_radix_tree_replace(struct address_space
*mapping
,
242 pgoff_t index
, void *expected
, void *replacement
)
247 VM_BUG_ON(!expected
);
248 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
250 item
= radix_tree_deref_slot_protected(pslot
,
251 &mapping
->tree_lock
);
252 if (item
!= expected
)
255 radix_tree_replace_slot(pslot
, replacement
);
257 radix_tree_delete(&mapping
->page_tree
, index
);
262 * Sometimes, before we decide whether to proceed or to fail, we must check
263 * that an entry was not already brought back from swap by a racing thread.
265 * Checking page is not enough: by the time a SwapCache page is locked, it
266 * might be reused, and again be SwapCache, using the same swap as before.
268 static bool shmem_confirm_swap(struct address_space
*mapping
,
269 pgoff_t index
, swp_entry_t swap
)
274 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
276 return item
== swp_to_radix_entry(swap
);
280 * Like add_to_page_cache_locked, but error if expected item has gone.
282 static int shmem_add_to_page_cache(struct page
*page
,
283 struct address_space
*mapping
,
284 pgoff_t index
, gfp_t gfp
, void *expected
)
288 VM_BUG_ON(!PageLocked(page
));
289 VM_BUG_ON(!PageSwapBacked(page
));
291 page_cache_get(page
);
292 page
->mapping
= mapping
;
295 spin_lock_irq(&mapping
->tree_lock
);
297 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
299 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
303 __inc_zone_page_state(page
, NR_FILE_PAGES
);
304 __inc_zone_page_state(page
, NR_SHMEM
);
305 spin_unlock_irq(&mapping
->tree_lock
);
307 page
->mapping
= NULL
;
308 spin_unlock_irq(&mapping
->tree_lock
);
309 page_cache_release(page
);
315 * Like delete_from_page_cache, but substitutes swap for page.
317 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
319 struct address_space
*mapping
= page
->mapping
;
322 spin_lock_irq(&mapping
->tree_lock
);
323 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
324 page
->mapping
= NULL
;
326 __dec_zone_page_state(page
, NR_FILE_PAGES
);
327 __dec_zone_page_state(page
, NR_SHMEM
);
328 spin_unlock_irq(&mapping
->tree_lock
);
329 page_cache_release(page
);
334 * Like find_get_pages, but collecting swap entries as well as pages.
336 static unsigned shmem_find_get_pages_and_swap(struct address_space
*mapping
,
337 pgoff_t start
, unsigned int nr_pages
,
338 struct page
**pages
, pgoff_t
*indices
)
341 unsigned int ret
= 0;
342 struct radix_tree_iter iter
;
349 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
352 page
= radix_tree_deref_slot(slot
);
355 if (radix_tree_exception(page
)) {
356 if (radix_tree_deref_retry(page
))
359 * Otherwise, we must be storing a swap entry
360 * here as an exceptional entry: so return it
361 * without attempting to raise page count.
365 if (!page_cache_get_speculative(page
))
368 /* Has the page moved? */
369 if (unlikely(page
!= *slot
)) {
370 page_cache_release(page
);
374 indices
[ret
] = iter
.index
;
376 if (++ret
== nr_pages
)
384 * Remove swap entry from radix tree, free the swap and its page cache.
386 static int shmem_free_swap(struct address_space
*mapping
,
387 pgoff_t index
, void *radswap
)
391 spin_lock_irq(&mapping
->tree_lock
);
392 error
= shmem_radix_tree_replace(mapping
, index
, radswap
, NULL
);
393 spin_unlock_irq(&mapping
->tree_lock
);
395 free_swap_and_cache(radix_to_swp_entry(radswap
));
400 * Pagevec may contain swap entries, so shuffle up pages before releasing.
402 static void shmem_deswap_pagevec(struct pagevec
*pvec
)
406 for (i
= 0, j
= 0; i
< pagevec_count(pvec
); i
++) {
407 struct page
*page
= pvec
->pages
[i
];
408 if (!radix_tree_exceptional_entry(page
))
409 pvec
->pages
[j
++] = page
;
415 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
417 void shmem_unlock_mapping(struct address_space
*mapping
)
420 pgoff_t indices
[PAGEVEC_SIZE
];
423 pagevec_init(&pvec
, 0);
425 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
427 while (!mapping_unevictable(mapping
)) {
429 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
430 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
432 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
433 PAGEVEC_SIZE
, pvec
.pages
, indices
);
436 index
= indices
[pvec
.nr
- 1] + 1;
437 shmem_deswap_pagevec(&pvec
);
438 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
439 pagevec_release(&pvec
);
445 * Remove range of pages and swap entries from radix tree, and free them.
446 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
448 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
451 struct address_space
*mapping
= inode
->i_mapping
;
452 struct shmem_inode_info
*info
= SHMEM_I(inode
);
453 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
454 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
455 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
456 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
458 pgoff_t indices
[PAGEVEC_SIZE
];
459 long nr_swaps_freed
= 0;
464 end
= -1; /* unsigned, so actually very big */
466 pagevec_init(&pvec
, 0);
468 while (index
< end
) {
469 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
470 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
471 pvec
.pages
, indices
);
474 mem_cgroup_uncharge_start();
475 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
476 struct page
*page
= pvec
.pages
[i
];
482 if (radix_tree_exceptional_entry(page
)) {
485 nr_swaps_freed
+= !shmem_free_swap(mapping
,
490 if (!trylock_page(page
))
492 if (!unfalloc
|| !PageUptodate(page
)) {
493 if (page
->mapping
== mapping
) {
494 VM_BUG_ON(PageWriteback(page
));
495 truncate_inode_page(mapping
, page
);
500 shmem_deswap_pagevec(&pvec
);
501 pagevec_release(&pvec
);
502 mem_cgroup_uncharge_end();
508 struct page
*page
= NULL
;
509 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
511 unsigned int top
= PAGE_CACHE_SIZE
;
516 zero_user_segment(page
, partial_start
, top
);
517 set_page_dirty(page
);
519 page_cache_release(page
);
523 struct page
*page
= NULL
;
524 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
526 zero_user_segment(page
, 0, partial_end
);
527 set_page_dirty(page
);
529 page_cache_release(page
);
538 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
539 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
540 pvec
.pages
, indices
);
542 if (index
== start
|| unfalloc
)
547 if ((index
== start
|| unfalloc
) && indices
[0] >= end
) {
548 shmem_deswap_pagevec(&pvec
);
549 pagevec_release(&pvec
);
552 mem_cgroup_uncharge_start();
553 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
554 struct page
*page
= pvec
.pages
[i
];
560 if (radix_tree_exceptional_entry(page
)) {
563 nr_swaps_freed
+= !shmem_free_swap(mapping
,
569 if (!unfalloc
|| !PageUptodate(page
)) {
570 if (page
->mapping
== mapping
) {
571 VM_BUG_ON(PageWriteback(page
));
572 truncate_inode_page(mapping
, page
);
577 shmem_deswap_pagevec(&pvec
);
578 pagevec_release(&pvec
);
579 mem_cgroup_uncharge_end();
583 spin_lock(&info
->lock
);
584 info
->swapped
-= nr_swaps_freed
;
585 shmem_recalc_inode(inode
);
586 spin_unlock(&info
->lock
);
589 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
591 shmem_undo_range(inode
, lstart
, lend
, false);
592 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
594 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
596 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
598 struct inode
*inode
= dentry
->d_inode
;
601 error
= inode_change_ok(inode
, attr
);
605 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
606 loff_t oldsize
= inode
->i_size
;
607 loff_t newsize
= attr
->ia_size
;
609 if (newsize
!= oldsize
) {
610 i_size_write(inode
, newsize
);
611 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
613 if (newsize
< oldsize
) {
614 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
615 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
616 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
617 /* unmap again to remove racily COWed private pages */
618 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
622 setattr_copy(inode
, attr
);
623 #ifdef CONFIG_TMPFS_POSIX_ACL
624 if (attr
->ia_valid
& ATTR_MODE
)
625 error
= generic_acl_chmod(inode
);
630 static void shmem_evict_inode(struct inode
*inode
)
632 struct shmem_inode_info
*info
= SHMEM_I(inode
);
634 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
635 shmem_unacct_size(info
->flags
, inode
->i_size
);
637 shmem_truncate_range(inode
, 0, (loff_t
)-1);
638 if (!list_empty(&info
->swaplist
)) {
639 mutex_lock(&shmem_swaplist_mutex
);
640 list_del_init(&info
->swaplist
);
641 mutex_unlock(&shmem_swaplist_mutex
);
644 kfree(info
->symlink
);
646 simple_xattrs_free(&info
->xattrs
);
647 WARN_ON(inode
->i_blocks
);
648 shmem_free_inode(inode
->i_sb
);
653 * If swap found in inode, free it and move page from swapcache to filecache.
655 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
656 swp_entry_t swap
, struct page
**pagep
)
658 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
664 radswap
= swp_to_radix_entry(swap
);
665 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
670 * Move _head_ to start search for next from here.
671 * But be careful: shmem_evict_inode checks list_empty without taking
672 * mutex, and there's an instant in list_move_tail when info->swaplist
673 * would appear empty, if it were the only one on shmem_swaplist.
675 if (shmem_swaplist
.next
!= &info
->swaplist
)
676 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
678 gfp
= mapping_gfp_mask(mapping
);
679 if (shmem_should_replace_page(*pagep
, gfp
)) {
680 mutex_unlock(&shmem_swaplist_mutex
);
681 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
682 mutex_lock(&shmem_swaplist_mutex
);
684 * We needed to drop mutex to make that restrictive page
685 * allocation, but the inode might have been freed while we
686 * dropped it: although a racing shmem_evict_inode() cannot
687 * complete without emptying the radix_tree, our page lock
688 * on this swapcache page is not enough to prevent that -
689 * free_swap_and_cache() of our swap entry will only
690 * trylock_page(), removing swap from radix_tree whatever.
692 * We must not proceed to shmem_add_to_page_cache() if the
693 * inode has been freed, but of course we cannot rely on
694 * inode or mapping or info to check that. However, we can
695 * safely check if our swap entry is still in use (and here
696 * it can't have got reused for another page): if it's still
697 * in use, then the inode cannot have been freed yet, and we
698 * can safely proceed (if it's no longer in use, that tells
699 * nothing about the inode, but we don't need to unuse swap).
701 if (!page_swapcount(*pagep
))
706 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
707 * but also to hold up shmem_evict_inode(): so inode cannot be freed
708 * beneath us (pagelock doesn't help until the page is in pagecache).
711 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
712 GFP_NOWAIT
, radswap
);
713 if (error
!= -ENOMEM
) {
715 * Truncation and eviction use free_swap_and_cache(), which
716 * only does trylock page: if we raced, best clean up here.
718 delete_from_swap_cache(*pagep
);
719 set_page_dirty(*pagep
);
721 spin_lock(&info
->lock
);
723 spin_unlock(&info
->lock
);
726 error
= 1; /* not an error, but entry was found */
732 * Search through swapped inodes to find and replace swap by page.
734 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
736 struct list_head
*this, *next
;
737 struct shmem_inode_info
*info
;
742 * There's a faint possibility that swap page was replaced before
743 * caller locked it: caller will come back later with the right page.
745 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
749 * Charge page using GFP_KERNEL while we can wait, before taking
750 * the shmem_swaplist_mutex which might hold up shmem_writepage().
751 * Charged back to the user (not to caller) when swap account is used.
753 error
= mem_cgroup_cache_charge(page
, current
->mm
, GFP_KERNEL
);
756 /* No radix_tree_preload: swap entry keeps a place for page in tree */
758 mutex_lock(&shmem_swaplist_mutex
);
759 list_for_each_safe(this, next
, &shmem_swaplist
) {
760 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
762 found
= shmem_unuse_inode(info
, swap
, &page
);
764 list_del_init(&info
->swaplist
);
769 mutex_unlock(&shmem_swaplist_mutex
);
775 page_cache_release(page
);
780 * Move the page from the page cache to the swap cache.
782 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
784 struct shmem_inode_info
*info
;
785 struct address_space
*mapping
;
790 BUG_ON(!PageLocked(page
));
791 mapping
= page
->mapping
;
793 inode
= mapping
->host
;
794 info
= SHMEM_I(inode
);
795 if (info
->flags
& VM_LOCKED
)
797 if (!total_swap_pages
)
801 * shmem_backing_dev_info's capabilities prevent regular writeback or
802 * sync from ever calling shmem_writepage; but a stacking filesystem
803 * might use ->writepage of its underlying filesystem, in which case
804 * tmpfs should write out to swap only in response to memory pressure,
805 * and not for the writeback threads or sync.
807 if (!wbc
->for_reclaim
) {
808 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
813 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
814 * value into swapfile.c, the only way we can correctly account for a
815 * fallocated page arriving here is now to initialize it and write it.
817 * That's okay for a page already fallocated earlier, but if we have
818 * not yet completed the fallocation, then (a) we want to keep track
819 * of this page in case we have to undo it, and (b) it may not be a
820 * good idea to continue anyway, once we're pushing into swap. So
821 * reactivate the page, and let shmem_fallocate() quit when too many.
823 if (!PageUptodate(page
)) {
824 if (inode
->i_private
) {
825 struct shmem_falloc
*shmem_falloc
;
826 spin_lock(&inode
->i_lock
);
827 shmem_falloc
= inode
->i_private
;
829 index
>= shmem_falloc
->start
&&
830 index
< shmem_falloc
->next
)
831 shmem_falloc
->nr_unswapped
++;
834 spin_unlock(&inode
->i_lock
);
838 clear_highpage(page
);
839 flush_dcache_page(page
);
840 SetPageUptodate(page
);
843 swap
= get_swap_page();
848 * Add inode to shmem_unuse()'s list of swapped-out inodes,
849 * if it's not already there. Do it now before the page is
850 * moved to swap cache, when its pagelock no longer protects
851 * the inode from eviction. But don't unlock the mutex until
852 * we've incremented swapped, because shmem_unuse_inode() will
853 * prune a !swapped inode from the swaplist under this mutex.
855 mutex_lock(&shmem_swaplist_mutex
);
856 if (list_empty(&info
->swaplist
))
857 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
859 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
860 swap_shmem_alloc(swap
);
861 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
863 spin_lock(&info
->lock
);
865 shmem_recalc_inode(inode
);
866 spin_unlock(&info
->lock
);
868 mutex_unlock(&shmem_swaplist_mutex
);
869 BUG_ON(page_mapped(page
));
870 swap_writepage(page
, wbc
);
874 mutex_unlock(&shmem_swaplist_mutex
);
875 swapcache_free(swap
, NULL
);
877 set_page_dirty(page
);
878 if (wbc
->for_reclaim
)
879 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
886 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
890 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
891 return; /* show nothing */
893 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
895 seq_printf(seq
, ",mpol=%s", buffer
);
898 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
900 struct mempolicy
*mpol
= NULL
;
902 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
905 spin_unlock(&sbinfo
->stat_lock
);
909 #endif /* CONFIG_TMPFS */
911 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
912 struct shmem_inode_info
*info
, pgoff_t index
)
914 struct vm_area_struct pvma
;
917 /* Create a pseudo vma that just contains the policy */
919 /* Bias interleave by inode number to distribute better across nodes */
920 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
922 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
924 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
926 /* Drop reference taken by mpol_shared_policy_lookup() */
927 mpol_cond_put(pvma
.vm_policy
);
932 static struct page
*shmem_alloc_page(gfp_t gfp
,
933 struct shmem_inode_info
*info
, pgoff_t index
)
935 struct vm_area_struct pvma
;
938 /* Create a pseudo vma that just contains the policy */
940 /* Bias interleave by inode number to distribute better across nodes */
941 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
943 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
945 page
= alloc_page_vma(gfp
, &pvma
, 0);
947 /* Drop reference taken by mpol_shared_policy_lookup() */
948 mpol_cond_put(pvma
.vm_policy
);
952 #else /* !CONFIG_NUMA */
954 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
957 #endif /* CONFIG_TMPFS */
959 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
960 struct shmem_inode_info
*info
, pgoff_t index
)
962 return swapin_readahead(swap
, gfp
, NULL
, 0);
965 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
966 struct shmem_inode_info
*info
, pgoff_t index
)
968 return alloc_page(gfp
);
970 #endif /* CONFIG_NUMA */
972 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
973 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
980 * When a page is moved from swapcache to shmem filecache (either by the
981 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
982 * shmem_unuse_inode()), it may have been read in earlier from swap, in
983 * ignorance of the mapping it belongs to. If that mapping has special
984 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
985 * we may need to copy to a suitable page before moving to filecache.
987 * In a future release, this may well be extended to respect cpuset and
988 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
989 * but for now it is a simple matter of zone.
991 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
993 return page_zonenum(page
) > gfp_zone(gfp
);
996 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
997 struct shmem_inode_info
*info
, pgoff_t index
)
999 struct page
*oldpage
, *newpage
;
1000 struct address_space
*swap_mapping
;
1005 swap_index
= page_private(oldpage
);
1006 swap_mapping
= page_mapping(oldpage
);
1009 * We have arrived here because our zones are constrained, so don't
1010 * limit chance of success by further cpuset and node constraints.
1012 gfp
&= ~GFP_CONSTRAINT_MASK
;
1013 newpage
= shmem_alloc_page(gfp
, info
, index
);
1017 page_cache_get(newpage
);
1018 copy_highpage(newpage
, oldpage
);
1019 flush_dcache_page(newpage
);
1021 __set_page_locked(newpage
);
1022 SetPageUptodate(newpage
);
1023 SetPageSwapBacked(newpage
);
1024 set_page_private(newpage
, swap_index
);
1025 SetPageSwapCache(newpage
);
1028 * Our caller will very soon move newpage out of swapcache, but it's
1029 * a nice clean interface for us to replace oldpage by newpage there.
1031 spin_lock_irq(&swap_mapping
->tree_lock
);
1032 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1035 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1036 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1038 spin_unlock_irq(&swap_mapping
->tree_lock
);
1040 if (unlikely(error
)) {
1042 * Is this possible? I think not, now that our callers check
1043 * both PageSwapCache and page_private after getting page lock;
1044 * but be defensive. Reverse old to newpage for clear and free.
1048 mem_cgroup_replace_page_cache(oldpage
, newpage
);
1049 lru_cache_add_anon(newpage
);
1053 ClearPageSwapCache(oldpage
);
1054 set_page_private(oldpage
, 0);
1056 unlock_page(oldpage
);
1057 page_cache_release(oldpage
);
1058 page_cache_release(oldpage
);
1063 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1065 * If we allocate a new one we do not mark it dirty. That's up to the
1066 * vm. If we swap it in we mark it dirty since we also free the swap
1067 * entry since a page cannot live in both the swap and page cache
1069 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1070 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1072 struct address_space
*mapping
= inode
->i_mapping
;
1073 struct shmem_inode_info
*info
;
1074 struct shmem_sb_info
*sbinfo
;
1081 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1085 page
= find_lock_page(mapping
, index
);
1086 if (radix_tree_exceptional_entry(page
)) {
1087 swap
= radix_to_swp_entry(page
);
1091 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1092 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1097 /* fallocated page? */
1098 if (page
&& !PageUptodate(page
)) {
1099 if (sgp
!= SGP_READ
)
1102 page_cache_release(page
);
1105 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1111 * Fast cache lookup did not find it:
1112 * bring it back from swap or allocate.
1114 info
= SHMEM_I(inode
);
1115 sbinfo
= SHMEM_SB(inode
->i_sb
);
1118 /* Look it up and read it in.. */
1119 page
= lookup_swap_cache(swap
);
1121 /* here we actually do the io */
1123 *fault_type
|= VM_FAULT_MAJOR
;
1124 page
= shmem_swapin(swap
, gfp
, info
, index
);
1131 /* We have to do this with page locked to prevent races */
1133 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1134 !shmem_confirm_swap(mapping
, index
, swap
)) {
1135 error
= -EEXIST
; /* try again */
1138 if (!PageUptodate(page
)) {
1142 wait_on_page_writeback(page
);
1144 if (shmem_should_replace_page(page
, gfp
)) {
1145 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1150 error
= mem_cgroup_cache_charge(page
, current
->mm
,
1151 gfp
& GFP_RECLAIM_MASK
);
1153 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1154 gfp
, swp_to_radix_entry(swap
));
1156 * We already confirmed swap under page lock, and make
1157 * no memory allocation here, so usually no possibility
1158 * of error; but free_swap_and_cache() only trylocks a
1159 * page, so it is just possible that the entry has been
1160 * truncated or holepunched since swap was confirmed.
1161 * shmem_undo_range() will have done some of the
1162 * unaccounting, now delete_from_swap_cache() will do
1163 * the rest (including mem_cgroup_uncharge_swapcache).
1164 * Reset swap.val? No, leave it so "failed" goes back to
1165 * "repeat": reading a hole and writing should succeed.
1168 delete_from_swap_cache(page
);
1173 spin_lock(&info
->lock
);
1175 shmem_recalc_inode(inode
);
1176 spin_unlock(&info
->lock
);
1178 delete_from_swap_cache(page
);
1179 set_page_dirty(page
);
1183 if (shmem_acct_block(info
->flags
)) {
1187 if (sbinfo
->max_blocks
) {
1188 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1189 sbinfo
->max_blocks
) >= 0) {
1193 percpu_counter_inc(&sbinfo
->used_blocks
);
1196 page
= shmem_alloc_page(gfp
, info
, index
);
1202 SetPageSwapBacked(page
);
1203 __set_page_locked(page
);
1204 error
= mem_cgroup_cache_charge(page
, current
->mm
,
1205 gfp
& GFP_RECLAIM_MASK
);
1208 error
= radix_tree_preload(gfp
& GFP_RECLAIM_MASK
);
1210 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1212 radix_tree_preload_end();
1215 mem_cgroup_uncharge_cache_page(page
);
1218 lru_cache_add_anon(page
);
1220 spin_lock(&info
->lock
);
1222 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1223 shmem_recalc_inode(inode
);
1224 spin_unlock(&info
->lock
);
1228 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1230 if (sgp
== SGP_FALLOC
)
1234 * Let SGP_WRITE caller clear ends if write does not fill page;
1235 * but SGP_FALLOC on a page fallocated earlier must initialize
1236 * it now, lest undo on failure cancel our earlier guarantee.
1238 if (sgp
!= SGP_WRITE
) {
1239 clear_highpage(page
);
1240 flush_dcache_page(page
);
1241 SetPageUptodate(page
);
1243 if (sgp
== SGP_DIRTY
)
1244 set_page_dirty(page
);
1247 /* Perhaps the file has been truncated since we checked */
1248 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1249 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1263 info
= SHMEM_I(inode
);
1264 ClearPageDirty(page
);
1265 delete_from_page_cache(page
);
1266 spin_lock(&info
->lock
);
1268 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1269 spin_unlock(&info
->lock
);
1271 sbinfo
= SHMEM_SB(inode
->i_sb
);
1272 if (sbinfo
->max_blocks
)
1273 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1275 shmem_unacct_blocks(info
->flags
, 1);
1277 if (swap
.val
&& error
!= -EINVAL
&&
1278 !shmem_confirm_swap(mapping
, index
, swap
))
1283 page_cache_release(page
);
1285 if (error
== -ENOSPC
&& !once
++) {
1286 info
= SHMEM_I(inode
);
1287 spin_lock(&info
->lock
);
1288 shmem_recalc_inode(inode
);
1289 spin_unlock(&info
->lock
);
1292 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1297 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1299 struct inode
*inode
= file_inode(vma
->vm_file
);
1301 int ret
= VM_FAULT_LOCKED
;
1303 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1305 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1307 if (ret
& VM_FAULT_MAJOR
) {
1308 count_vm_event(PGMAJFAULT
);
1309 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1315 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1317 struct inode
*inode
= file_inode(vma
->vm_file
);
1318 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1321 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1324 struct inode
*inode
= file_inode(vma
->vm_file
);
1327 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1328 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1332 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1334 struct inode
*inode
= file_inode(file
);
1335 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1336 int retval
= -ENOMEM
;
1338 spin_lock(&info
->lock
);
1339 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1340 if (!user_shm_lock(inode
->i_size
, user
))
1342 info
->flags
|= VM_LOCKED
;
1343 mapping_set_unevictable(file
->f_mapping
);
1345 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1346 user_shm_unlock(inode
->i_size
, user
);
1347 info
->flags
&= ~VM_LOCKED
;
1348 mapping_clear_unevictable(file
->f_mapping
);
1353 spin_unlock(&info
->lock
);
1357 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1359 file_accessed(file
);
1360 vma
->vm_ops
= &shmem_vm_ops
;
1364 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1365 umode_t mode
, dev_t dev
, unsigned long flags
)
1367 struct inode
*inode
;
1368 struct shmem_inode_info
*info
;
1369 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1371 if (shmem_reserve_inode(sb
))
1374 inode
= new_inode(sb
);
1376 inode
->i_ino
= get_next_ino();
1377 inode_init_owner(inode
, dir
, mode
);
1378 inode
->i_blocks
= 0;
1379 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1380 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1381 inode
->i_generation
= get_seconds();
1382 info
= SHMEM_I(inode
);
1383 memset(info
, 0, (char *)inode
- (char *)info
);
1384 spin_lock_init(&info
->lock
);
1385 info
->flags
= flags
& VM_NORESERVE
;
1386 INIT_LIST_HEAD(&info
->swaplist
);
1387 simple_xattrs_init(&info
->xattrs
);
1388 cache_no_acl(inode
);
1390 switch (mode
& S_IFMT
) {
1392 inode
->i_op
= &shmem_special_inode_operations
;
1393 init_special_inode(inode
, mode
, dev
);
1396 inode
->i_mapping
->a_ops
= &shmem_aops
;
1397 inode
->i_op
= &shmem_inode_operations
;
1398 inode
->i_fop
= &shmem_file_operations
;
1399 mpol_shared_policy_init(&info
->policy
,
1400 shmem_get_sbmpol(sbinfo
));
1404 /* Some things misbehave if size == 0 on a directory */
1405 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1406 inode
->i_op
= &shmem_dir_inode_operations
;
1407 inode
->i_fop
= &simple_dir_operations
;
1411 * Must not load anything in the rbtree,
1412 * mpol_free_shared_policy will not be called.
1414 mpol_shared_policy_init(&info
->policy
, NULL
);
1418 shmem_free_inode(sb
);
1423 static const struct inode_operations shmem_symlink_inode_operations
;
1424 static const struct inode_operations shmem_short_symlink_operations
;
1426 #ifdef CONFIG_TMPFS_XATTR
1427 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1429 #define shmem_initxattrs NULL
1433 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1434 loff_t pos
, unsigned len
, unsigned flags
,
1435 struct page
**pagep
, void **fsdata
)
1437 struct inode
*inode
= mapping
->host
;
1438 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1439 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1443 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1444 loff_t pos
, unsigned len
, unsigned copied
,
1445 struct page
*page
, void *fsdata
)
1447 struct inode
*inode
= mapping
->host
;
1449 if (pos
+ copied
> inode
->i_size
)
1450 i_size_write(inode
, pos
+ copied
);
1452 if (!PageUptodate(page
)) {
1453 if (copied
< PAGE_CACHE_SIZE
) {
1454 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1455 zero_user_segments(page
, 0, from
,
1456 from
+ copied
, PAGE_CACHE_SIZE
);
1458 SetPageUptodate(page
);
1460 set_page_dirty(page
);
1462 page_cache_release(page
);
1467 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1469 struct inode
*inode
= file_inode(filp
);
1470 struct address_space
*mapping
= inode
->i_mapping
;
1472 unsigned long offset
;
1473 enum sgp_type sgp
= SGP_READ
;
1476 * Might this read be for a stacking filesystem? Then when reading
1477 * holes of a sparse file, we actually need to allocate those pages,
1478 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1480 if (segment_eq(get_fs(), KERNEL_DS
))
1483 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1484 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1487 struct page
*page
= NULL
;
1489 unsigned long nr
, ret
;
1490 loff_t i_size
= i_size_read(inode
);
1492 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1493 if (index
> end_index
)
1495 if (index
== end_index
) {
1496 nr
= i_size
& ~PAGE_CACHE_MASK
;
1501 desc
->error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1503 if (desc
->error
== -EINVAL
)
1511 * We must evaluate after, since reads (unlike writes)
1512 * are called without i_mutex protection against truncate
1514 nr
= PAGE_CACHE_SIZE
;
1515 i_size
= i_size_read(inode
);
1516 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1517 if (index
== end_index
) {
1518 nr
= i_size
& ~PAGE_CACHE_MASK
;
1521 page_cache_release(page
);
1529 * If users can be writing to this page using arbitrary
1530 * virtual addresses, take care about potential aliasing
1531 * before reading the page on the kernel side.
1533 if (mapping_writably_mapped(mapping
))
1534 flush_dcache_page(page
);
1536 * Mark the page accessed if we read the beginning.
1539 mark_page_accessed(page
);
1541 page
= ZERO_PAGE(0);
1542 page_cache_get(page
);
1546 * Ok, we have the page, and it's up-to-date, so
1547 * now we can copy it to user space...
1549 * The actor routine returns how many bytes were actually used..
1550 * NOTE! This may not be the same as how much of a user buffer
1551 * we filled up (we may be padding etc), so we can only update
1552 * "pos" here (the actor routine has to update the user buffer
1553 * pointers and the remaining count).
1555 ret
= actor(desc
, page
, offset
, nr
);
1557 index
+= offset
>> PAGE_CACHE_SHIFT
;
1558 offset
&= ~PAGE_CACHE_MASK
;
1560 page_cache_release(page
);
1561 if (ret
!= nr
|| !desc
->count
)
1567 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1568 file_accessed(filp
);
1571 static ssize_t
shmem_file_aio_read(struct kiocb
*iocb
,
1572 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
1574 struct file
*filp
= iocb
->ki_filp
;
1578 loff_t
*ppos
= &iocb
->ki_pos
;
1580 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1584 for (seg
= 0; seg
< nr_segs
; seg
++) {
1585 read_descriptor_t desc
;
1588 desc
.arg
.buf
= iov
[seg
].iov_base
;
1589 desc
.count
= iov
[seg
].iov_len
;
1590 if (desc
.count
== 0)
1593 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1594 retval
+= desc
.written
;
1596 retval
= retval
?: desc
.error
;
1605 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1606 struct pipe_inode_info
*pipe
, size_t len
,
1609 struct address_space
*mapping
= in
->f_mapping
;
1610 struct inode
*inode
= mapping
->host
;
1611 unsigned int loff
, nr_pages
, req_pages
;
1612 struct page
*pages
[PIPE_DEF_BUFFERS
];
1613 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1615 pgoff_t index
, end_index
;
1618 struct splice_pipe_desc spd
= {
1621 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1623 .ops
= &page_cache_pipe_buf_ops
,
1624 .spd_release
= spd_release_page
,
1627 isize
= i_size_read(inode
);
1628 if (unlikely(*ppos
>= isize
))
1631 left
= isize
- *ppos
;
1632 if (unlikely(left
< len
))
1635 if (splice_grow_spd(pipe
, &spd
))
1638 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1639 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1640 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1641 nr_pages
= min(req_pages
, pipe
->buffers
);
1643 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1644 nr_pages
, spd
.pages
);
1645 index
+= spd
.nr_pages
;
1648 while (spd
.nr_pages
< nr_pages
) {
1649 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1653 spd
.pages
[spd
.nr_pages
++] = page
;
1657 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1658 nr_pages
= spd
.nr_pages
;
1661 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1662 unsigned int this_len
;
1667 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1668 page
= spd
.pages
[page_nr
];
1670 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1671 error
= shmem_getpage(inode
, index
, &page
,
1676 page_cache_release(spd
.pages
[page_nr
]);
1677 spd
.pages
[page_nr
] = page
;
1680 isize
= i_size_read(inode
);
1681 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1682 if (unlikely(!isize
|| index
> end_index
))
1685 if (end_index
== index
) {
1688 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1692 this_len
= min(this_len
, plen
- loff
);
1696 spd
.partial
[page_nr
].offset
= loff
;
1697 spd
.partial
[page_nr
].len
= this_len
;
1704 while (page_nr
< nr_pages
)
1705 page_cache_release(spd
.pages
[page_nr
++]);
1708 error
= splice_to_pipe(pipe
, &spd
);
1710 splice_shrink_spd(&spd
);
1720 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1722 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1723 pgoff_t index
, pgoff_t end
, int whence
)
1726 struct pagevec pvec
;
1727 pgoff_t indices
[PAGEVEC_SIZE
];
1731 pagevec_init(&pvec
, 0);
1732 pvec
.nr
= 1; /* start small: we may be there already */
1734 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
1735 pvec
.nr
, pvec
.pages
, indices
);
1737 if (whence
== SEEK_DATA
)
1741 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1742 if (index
< indices
[i
]) {
1743 if (whence
== SEEK_HOLE
) {
1749 page
= pvec
.pages
[i
];
1750 if (page
&& !radix_tree_exceptional_entry(page
)) {
1751 if (!PageUptodate(page
))
1755 (page
&& whence
== SEEK_DATA
) ||
1756 (!page
&& whence
== SEEK_HOLE
)) {
1761 shmem_deswap_pagevec(&pvec
);
1762 pagevec_release(&pvec
);
1763 pvec
.nr
= PAGEVEC_SIZE
;
1769 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1771 struct address_space
*mapping
= file
->f_mapping
;
1772 struct inode
*inode
= mapping
->host
;
1776 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1777 return generic_file_llseek_size(file
, offset
, whence
,
1778 MAX_LFS_FILESIZE
, i_size_read(inode
));
1779 mutex_lock(&inode
->i_mutex
);
1780 /* We're holding i_mutex so we can access i_size directly */
1784 else if (offset
>= inode
->i_size
)
1787 start
= offset
>> PAGE_CACHE_SHIFT
;
1788 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1789 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1790 new_offset
<<= PAGE_CACHE_SHIFT
;
1791 if (new_offset
> offset
) {
1792 if (new_offset
< inode
->i_size
)
1793 offset
= new_offset
;
1794 else if (whence
== SEEK_DATA
)
1797 offset
= inode
->i_size
;
1801 if (offset
>= 0 && offset
!= file
->f_pos
) {
1802 file
->f_pos
= offset
;
1803 file
->f_version
= 0;
1805 mutex_unlock(&inode
->i_mutex
);
1809 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
1812 struct inode
*inode
= file_inode(file
);
1813 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1814 struct shmem_falloc shmem_falloc
;
1815 pgoff_t start
, index
, end
;
1818 mutex_lock(&inode
->i_mutex
);
1820 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1821 struct address_space
*mapping
= file
->f_mapping
;
1822 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
1823 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
1825 if ((u64
)unmap_end
> (u64
)unmap_start
)
1826 unmap_mapping_range(mapping
, unmap_start
,
1827 1 + unmap_end
- unmap_start
, 0);
1828 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
1829 /* No need to unmap again: hole-punching leaves COWed pages */
1834 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1835 error
= inode_newsize_ok(inode
, offset
+ len
);
1839 start
= offset
>> PAGE_CACHE_SHIFT
;
1840 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1841 /* Try to avoid a swapstorm if len is impossible to satisfy */
1842 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
1847 shmem_falloc
.start
= start
;
1848 shmem_falloc
.next
= start
;
1849 shmem_falloc
.nr_falloced
= 0;
1850 shmem_falloc
.nr_unswapped
= 0;
1851 spin_lock(&inode
->i_lock
);
1852 inode
->i_private
= &shmem_falloc
;
1853 spin_unlock(&inode
->i_lock
);
1855 for (index
= start
; index
< end
; index
++) {
1859 * Good, the fallocate(2) manpage permits EINTR: we may have
1860 * been interrupted because we are using up too much memory.
1862 if (signal_pending(current
))
1864 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
1867 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
1870 /* Remove the !PageUptodate pages we added */
1871 shmem_undo_range(inode
,
1872 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
1873 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
1878 * Inform shmem_writepage() how far we have reached.
1879 * No need for lock or barrier: we have the page lock.
1881 shmem_falloc
.next
++;
1882 if (!PageUptodate(page
))
1883 shmem_falloc
.nr_falloced
++;
1886 * If !PageUptodate, leave it that way so that freeable pages
1887 * can be recognized if we need to rollback on error later.
1888 * But set_page_dirty so that memory pressure will swap rather
1889 * than free the pages we are allocating (and SGP_CACHE pages
1890 * might still be clean: we now need to mark those dirty too).
1892 set_page_dirty(page
);
1894 page_cache_release(page
);
1898 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
1899 i_size_write(inode
, offset
+ len
);
1900 inode
->i_ctime
= CURRENT_TIME
;
1902 spin_lock(&inode
->i_lock
);
1903 inode
->i_private
= NULL
;
1904 spin_unlock(&inode
->i_lock
);
1906 mutex_unlock(&inode
->i_mutex
);
1910 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1912 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
1914 buf
->f_type
= TMPFS_MAGIC
;
1915 buf
->f_bsize
= PAGE_CACHE_SIZE
;
1916 buf
->f_namelen
= NAME_MAX
;
1917 if (sbinfo
->max_blocks
) {
1918 buf
->f_blocks
= sbinfo
->max_blocks
;
1920 buf
->f_bfree
= sbinfo
->max_blocks
-
1921 percpu_counter_sum(&sbinfo
->used_blocks
);
1923 if (sbinfo
->max_inodes
) {
1924 buf
->f_files
= sbinfo
->max_inodes
;
1925 buf
->f_ffree
= sbinfo
->free_inodes
;
1927 /* else leave those fields 0 like simple_statfs */
1932 * File creation. Allocate an inode, and we're done..
1935 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
1937 struct inode
*inode
;
1938 int error
= -ENOSPC
;
1940 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
1942 error
= security_inode_init_security(inode
, dir
,
1944 shmem_initxattrs
, NULL
);
1946 if (error
!= -EOPNOTSUPP
) {
1951 #ifdef CONFIG_TMPFS_POSIX_ACL
1952 error
= generic_acl_init(inode
, dir
);
1960 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1961 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1962 d_instantiate(dentry
, inode
);
1963 dget(dentry
); /* Extra count - pin the dentry in core */
1968 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1972 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
1978 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
1981 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
1987 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1989 struct inode
*inode
= old_dentry
->d_inode
;
1993 * No ordinary (disk based) filesystem counts links as inodes;
1994 * but each new link needs a new dentry, pinning lowmem, and
1995 * tmpfs dentries cannot be pruned until they are unlinked.
1997 ret
= shmem_reserve_inode(inode
->i_sb
);
2001 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2002 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2004 ihold(inode
); /* New dentry reference */
2005 dget(dentry
); /* Extra pinning count for the created dentry */
2006 d_instantiate(dentry
, inode
);
2011 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2013 struct inode
*inode
= dentry
->d_inode
;
2015 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2016 shmem_free_inode(inode
->i_sb
);
2018 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2019 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2021 dput(dentry
); /* Undo the count from "create" - this does all the work */
2025 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2027 if (!simple_empty(dentry
))
2030 drop_nlink(dentry
->d_inode
);
2032 return shmem_unlink(dir
, dentry
);
2036 * The VFS layer already does all the dentry stuff for rename,
2037 * we just have to decrement the usage count for the target if
2038 * it exists so that the VFS layer correctly free's it when it
2041 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2043 struct inode
*inode
= old_dentry
->d_inode
;
2044 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2046 if (!simple_empty(new_dentry
))
2049 if (new_dentry
->d_inode
) {
2050 (void) shmem_unlink(new_dir
, new_dentry
);
2052 drop_nlink(old_dir
);
2053 } else if (they_are_dirs
) {
2054 drop_nlink(old_dir
);
2058 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2059 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2060 old_dir
->i_ctime
= old_dir
->i_mtime
=
2061 new_dir
->i_ctime
= new_dir
->i_mtime
=
2062 inode
->i_ctime
= CURRENT_TIME
;
2066 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2070 struct inode
*inode
;
2073 struct shmem_inode_info
*info
;
2075 len
= strlen(symname
) + 1;
2076 if (len
> PAGE_CACHE_SIZE
)
2077 return -ENAMETOOLONG
;
2079 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2083 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2084 shmem_initxattrs
, NULL
);
2086 if (error
!= -EOPNOTSUPP
) {
2093 info
= SHMEM_I(inode
);
2094 inode
->i_size
= len
-1;
2095 if (len
<= SHORT_SYMLINK_LEN
) {
2096 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2097 if (!info
->symlink
) {
2101 inode
->i_op
= &shmem_short_symlink_operations
;
2103 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2108 inode
->i_mapping
->a_ops
= &shmem_aops
;
2109 inode
->i_op
= &shmem_symlink_inode_operations
;
2110 kaddr
= kmap_atomic(page
);
2111 memcpy(kaddr
, symname
, len
);
2112 kunmap_atomic(kaddr
);
2113 SetPageUptodate(page
);
2114 set_page_dirty(page
);
2116 page_cache_release(page
);
2118 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2119 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2120 d_instantiate(dentry
, inode
);
2125 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
2127 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
2131 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
2133 struct page
*page
= NULL
;
2134 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
2135 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
2141 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2143 if (!IS_ERR(nd_get_link(nd
))) {
2144 struct page
*page
= cookie
;
2146 mark_page_accessed(page
);
2147 page_cache_release(page
);
2151 #ifdef CONFIG_TMPFS_XATTR
2153 * Superblocks without xattr inode operations may get some security.* xattr
2154 * support from the LSM "for free". As soon as we have any other xattrs
2155 * like ACLs, we also need to implement the security.* handlers at
2156 * filesystem level, though.
2160 * Callback for security_inode_init_security() for acquiring xattrs.
2162 static int shmem_initxattrs(struct inode
*inode
,
2163 const struct xattr
*xattr_array
,
2166 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2167 const struct xattr
*xattr
;
2168 struct simple_xattr
*new_xattr
;
2171 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2172 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2176 len
= strlen(xattr
->name
) + 1;
2177 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2179 if (!new_xattr
->name
) {
2184 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2185 XATTR_SECURITY_PREFIX_LEN
);
2186 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2189 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2195 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2196 #ifdef CONFIG_TMPFS_POSIX_ACL
2197 &generic_acl_access_handler
,
2198 &generic_acl_default_handler
,
2203 static int shmem_xattr_validate(const char *name
)
2205 struct { const char *prefix
; size_t len
; } arr
[] = {
2206 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2207 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2211 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2212 size_t preflen
= arr
[i
].len
;
2213 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2222 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2223 void *buffer
, size_t size
)
2225 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2229 * If this is a request for a synthetic attribute in the system.*
2230 * namespace use the generic infrastructure to resolve a handler
2231 * for it via sb->s_xattr.
2233 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2234 return generic_getxattr(dentry
, name
, buffer
, size
);
2236 err
= shmem_xattr_validate(name
);
2240 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2243 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2244 const void *value
, size_t size
, int flags
)
2246 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2250 * If this is a request for a synthetic attribute in the system.*
2251 * namespace use the generic infrastructure to resolve a handler
2252 * for it via sb->s_xattr.
2254 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2255 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2257 err
= shmem_xattr_validate(name
);
2261 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2264 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2266 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2270 * If this is a request for a synthetic attribute in the system.*
2271 * namespace use the generic infrastructure to resolve a handler
2272 * for it via sb->s_xattr.
2274 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2275 return generic_removexattr(dentry
, name
);
2277 err
= shmem_xattr_validate(name
);
2281 return simple_xattr_remove(&info
->xattrs
, name
);
2284 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2286 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2287 return simple_xattr_list(&info
->xattrs
, buffer
, size
);
2289 #endif /* CONFIG_TMPFS_XATTR */
2291 static const struct inode_operations shmem_short_symlink_operations
= {
2292 .readlink
= generic_readlink
,
2293 .follow_link
= shmem_follow_short_symlink
,
2294 #ifdef CONFIG_TMPFS_XATTR
2295 .setxattr
= shmem_setxattr
,
2296 .getxattr
= shmem_getxattr
,
2297 .listxattr
= shmem_listxattr
,
2298 .removexattr
= shmem_removexattr
,
2302 static const struct inode_operations shmem_symlink_inode_operations
= {
2303 .readlink
= generic_readlink
,
2304 .follow_link
= shmem_follow_link
,
2305 .put_link
= shmem_put_link
,
2306 #ifdef CONFIG_TMPFS_XATTR
2307 .setxattr
= shmem_setxattr
,
2308 .getxattr
= shmem_getxattr
,
2309 .listxattr
= shmem_listxattr
,
2310 .removexattr
= shmem_removexattr
,
2314 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2316 return ERR_PTR(-ESTALE
);
2319 static int shmem_match(struct inode
*ino
, void *vfh
)
2323 inum
= (inum
<< 32) | fh
[1];
2324 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2327 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2328 struct fid
*fid
, int fh_len
, int fh_type
)
2330 struct inode
*inode
;
2331 struct dentry
*dentry
= NULL
;
2338 inum
= (inum
<< 32) | fid
->raw
[1];
2340 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2341 shmem_match
, fid
->raw
);
2343 dentry
= d_find_alias(inode
);
2350 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2351 struct inode
*parent
)
2355 return FILEID_INVALID
;
2358 if (inode_unhashed(inode
)) {
2359 /* Unfortunately insert_inode_hash is not idempotent,
2360 * so as we hash inodes here rather than at creation
2361 * time, we need a lock to ensure we only try
2364 static DEFINE_SPINLOCK(lock
);
2366 if (inode_unhashed(inode
))
2367 __insert_inode_hash(inode
,
2368 inode
->i_ino
+ inode
->i_generation
);
2372 fh
[0] = inode
->i_generation
;
2373 fh
[1] = inode
->i_ino
;
2374 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2380 static const struct export_operations shmem_export_ops
= {
2381 .get_parent
= shmem_get_parent
,
2382 .encode_fh
= shmem_encode_fh
,
2383 .fh_to_dentry
= shmem_fh_to_dentry
,
2386 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2389 char *this_char
, *value
, *rest
;
2390 struct mempolicy
*mpol
= NULL
;
2394 while (options
!= NULL
) {
2395 this_char
= options
;
2398 * NUL-terminate this option: unfortunately,
2399 * mount options form a comma-separated list,
2400 * but mpol's nodelist may also contain commas.
2402 options
= strchr(options
, ',');
2403 if (options
== NULL
)
2406 if (!isdigit(*options
)) {
2413 if ((value
= strchr(this_char
,'=')) != NULL
) {
2417 "tmpfs: No value for mount option '%s'\n",
2422 if (!strcmp(this_char
,"size")) {
2423 unsigned long long size
;
2424 size
= memparse(value
,&rest
);
2426 size
<<= PAGE_SHIFT
;
2427 size
*= totalram_pages
;
2433 sbinfo
->max_blocks
=
2434 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2435 } else if (!strcmp(this_char
,"nr_blocks")) {
2436 sbinfo
->max_blocks
= memparse(value
, &rest
);
2439 } else if (!strcmp(this_char
,"nr_inodes")) {
2440 sbinfo
->max_inodes
= memparse(value
, &rest
);
2443 } else if (!strcmp(this_char
,"mode")) {
2446 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2449 } else if (!strcmp(this_char
,"uid")) {
2452 uid
= simple_strtoul(value
, &rest
, 0);
2455 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2456 if (!uid_valid(sbinfo
->uid
))
2458 } else if (!strcmp(this_char
,"gid")) {
2461 gid
= simple_strtoul(value
, &rest
, 0);
2464 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2465 if (!gid_valid(sbinfo
->gid
))
2467 } else if (!strcmp(this_char
,"mpol")) {
2470 if (mpol_parse_str(value
, &mpol
))
2473 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2478 sbinfo
->mpol
= mpol
;
2482 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2490 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2492 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2493 struct shmem_sb_info config
= *sbinfo
;
2494 unsigned long inodes
;
2495 int error
= -EINVAL
;
2498 if (shmem_parse_options(data
, &config
, true))
2501 spin_lock(&sbinfo
->stat_lock
);
2502 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2503 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2505 if (config
.max_inodes
< inodes
)
2508 * Those tests disallow limited->unlimited while any are in use;
2509 * but we must separately disallow unlimited->limited, because
2510 * in that case we have no record of how much is already in use.
2512 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2514 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2518 sbinfo
->max_blocks
= config
.max_blocks
;
2519 sbinfo
->max_inodes
= config
.max_inodes
;
2520 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2523 * Preserve previous mempolicy unless mpol remount option was specified.
2526 mpol_put(sbinfo
->mpol
);
2527 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2530 spin_unlock(&sbinfo
->stat_lock
);
2534 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2536 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2538 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2539 seq_printf(seq
, ",size=%luk",
2540 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2541 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2542 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2543 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2544 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2545 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2546 seq_printf(seq
, ",uid=%u",
2547 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2548 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2549 seq_printf(seq
, ",gid=%u",
2550 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2551 shmem_show_mpol(seq
, sbinfo
->mpol
);
2554 #endif /* CONFIG_TMPFS */
2556 static void shmem_put_super(struct super_block
*sb
)
2558 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2560 percpu_counter_destroy(&sbinfo
->used_blocks
);
2561 mpol_put(sbinfo
->mpol
);
2563 sb
->s_fs_info
= NULL
;
2566 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2568 struct inode
*inode
;
2569 struct shmem_sb_info
*sbinfo
;
2572 /* Round up to L1_CACHE_BYTES to resist false sharing */
2573 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2574 L1_CACHE_BYTES
), GFP_KERNEL
);
2578 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2579 sbinfo
->uid
= current_fsuid();
2580 sbinfo
->gid
= current_fsgid();
2581 sb
->s_fs_info
= sbinfo
;
2585 * Per default we only allow half of the physical ram per
2586 * tmpfs instance, limiting inodes to one per page of lowmem;
2587 * but the internal instance is left unlimited.
2589 if (!(sb
->s_flags
& MS_NOUSER
)) {
2590 sbinfo
->max_blocks
= shmem_default_max_blocks();
2591 sbinfo
->max_inodes
= shmem_default_max_inodes();
2592 if (shmem_parse_options(data
, sbinfo
, false)) {
2597 sb
->s_export_op
= &shmem_export_ops
;
2598 sb
->s_flags
|= MS_NOSEC
;
2600 sb
->s_flags
|= MS_NOUSER
;
2603 spin_lock_init(&sbinfo
->stat_lock
);
2604 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
2606 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2608 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
2609 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2610 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2611 sb
->s_magic
= TMPFS_MAGIC
;
2612 sb
->s_op
= &shmem_ops
;
2613 sb
->s_time_gran
= 1;
2614 #ifdef CONFIG_TMPFS_XATTR
2615 sb
->s_xattr
= shmem_xattr_handlers
;
2617 #ifdef CONFIG_TMPFS_POSIX_ACL
2618 sb
->s_flags
|= MS_POSIXACL
;
2621 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
2624 inode
->i_uid
= sbinfo
->uid
;
2625 inode
->i_gid
= sbinfo
->gid
;
2626 sb
->s_root
= d_make_root(inode
);
2632 shmem_put_super(sb
);
2636 static struct kmem_cache
*shmem_inode_cachep
;
2638 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2640 struct shmem_inode_info
*info
;
2641 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2644 return &info
->vfs_inode
;
2647 static void shmem_destroy_callback(struct rcu_head
*head
)
2649 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
2650 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2653 static void shmem_destroy_inode(struct inode
*inode
)
2655 if (S_ISREG(inode
->i_mode
))
2656 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2657 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
2660 static void shmem_init_inode(void *foo
)
2662 struct shmem_inode_info
*info
= foo
;
2663 inode_init_once(&info
->vfs_inode
);
2666 static int shmem_init_inodecache(void)
2668 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2669 sizeof(struct shmem_inode_info
),
2670 0, SLAB_PANIC
, shmem_init_inode
);
2674 static void shmem_destroy_inodecache(void)
2676 kmem_cache_destroy(shmem_inode_cachep
);
2679 static const struct address_space_operations shmem_aops
= {
2680 .writepage
= shmem_writepage
,
2681 .set_page_dirty
= __set_page_dirty_no_writeback
,
2683 .write_begin
= shmem_write_begin
,
2684 .write_end
= shmem_write_end
,
2686 .migratepage
= migrate_page
,
2687 .error_remove_page
= generic_error_remove_page
,
2690 static const struct file_operations shmem_file_operations
= {
2693 .llseek
= shmem_file_llseek
,
2694 .read
= do_sync_read
,
2695 .write
= do_sync_write
,
2696 .aio_read
= shmem_file_aio_read
,
2697 .aio_write
= generic_file_aio_write
,
2698 .fsync
= noop_fsync
,
2699 .splice_read
= shmem_file_splice_read
,
2700 .splice_write
= generic_file_splice_write
,
2701 .fallocate
= shmem_fallocate
,
2705 static const struct inode_operations shmem_inode_operations
= {
2706 .setattr
= shmem_setattr
,
2707 #ifdef CONFIG_TMPFS_XATTR
2708 .setxattr
= shmem_setxattr
,
2709 .getxattr
= shmem_getxattr
,
2710 .listxattr
= shmem_listxattr
,
2711 .removexattr
= shmem_removexattr
,
2715 static const struct inode_operations shmem_dir_inode_operations
= {
2717 .create
= shmem_create
,
2718 .lookup
= simple_lookup
,
2720 .unlink
= shmem_unlink
,
2721 .symlink
= shmem_symlink
,
2722 .mkdir
= shmem_mkdir
,
2723 .rmdir
= shmem_rmdir
,
2724 .mknod
= shmem_mknod
,
2725 .rename
= shmem_rename
,
2727 #ifdef CONFIG_TMPFS_XATTR
2728 .setxattr
= shmem_setxattr
,
2729 .getxattr
= shmem_getxattr
,
2730 .listxattr
= shmem_listxattr
,
2731 .removexattr
= shmem_removexattr
,
2733 #ifdef CONFIG_TMPFS_POSIX_ACL
2734 .setattr
= shmem_setattr
,
2738 static const struct inode_operations shmem_special_inode_operations
= {
2739 #ifdef CONFIG_TMPFS_XATTR
2740 .setxattr
= shmem_setxattr
,
2741 .getxattr
= shmem_getxattr
,
2742 .listxattr
= shmem_listxattr
,
2743 .removexattr
= shmem_removexattr
,
2745 #ifdef CONFIG_TMPFS_POSIX_ACL
2746 .setattr
= shmem_setattr
,
2750 static const struct super_operations shmem_ops
= {
2751 .alloc_inode
= shmem_alloc_inode
,
2752 .destroy_inode
= shmem_destroy_inode
,
2754 .statfs
= shmem_statfs
,
2755 .remount_fs
= shmem_remount_fs
,
2756 .show_options
= shmem_show_options
,
2758 .evict_inode
= shmem_evict_inode
,
2759 .drop_inode
= generic_delete_inode
,
2760 .put_super
= shmem_put_super
,
2763 static const struct vm_operations_struct shmem_vm_ops
= {
2764 .fault
= shmem_fault
,
2766 .set_policy
= shmem_set_policy
,
2767 .get_policy
= shmem_get_policy
,
2769 .remap_pages
= generic_file_remap_pages
,
2772 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
2773 int flags
, const char *dev_name
, void *data
)
2775 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
2778 static struct file_system_type shmem_fs_type
= {
2779 .owner
= THIS_MODULE
,
2781 .mount
= shmem_mount
,
2782 .kill_sb
= kill_litter_super
,
2783 .fs_flags
= FS_USERNS_MOUNT
,
2786 int __init
shmem_init(void)
2790 error
= bdi_init(&shmem_backing_dev_info
);
2794 error
= shmem_init_inodecache();
2798 error
= register_filesystem(&shmem_fs_type
);
2800 printk(KERN_ERR
"Could not register tmpfs\n");
2804 shm_mnt
= vfs_kern_mount(&shmem_fs_type
, MS_NOUSER
,
2805 shmem_fs_type
.name
, NULL
);
2806 if (IS_ERR(shm_mnt
)) {
2807 error
= PTR_ERR(shm_mnt
);
2808 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2814 unregister_filesystem(&shmem_fs_type
);
2816 shmem_destroy_inodecache();
2818 bdi_destroy(&shmem_backing_dev_info
);
2820 shm_mnt
= ERR_PTR(error
);
2824 #else /* !CONFIG_SHMEM */
2827 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2829 * This is intended for small system where the benefits of the full
2830 * shmem code (swap-backed and resource-limited) are outweighed by
2831 * their complexity. On systems without swap this code should be
2832 * effectively equivalent, but much lighter weight.
2835 static struct file_system_type shmem_fs_type
= {
2837 .mount
= ramfs_mount
,
2838 .kill_sb
= kill_litter_super
,
2839 .fs_flags
= FS_USERNS_MOUNT
,
2842 int __init
shmem_init(void)
2844 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
2846 shm_mnt
= kern_mount(&shmem_fs_type
);
2847 BUG_ON(IS_ERR(shm_mnt
));
2852 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
2857 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2862 void shmem_unlock_mapping(struct address_space
*mapping
)
2866 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
2868 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
2870 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
2872 #define shmem_vm_ops generic_file_vm_ops
2873 #define shmem_file_operations ramfs_file_operations
2874 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2875 #define shmem_acct_size(flags, size) 0
2876 #define shmem_unacct_size(flags, size) do {} while (0)
2878 #endif /* CONFIG_SHMEM */
2882 static char *shmem_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
2884 return dynamic_dname(dentry
, buffer
, buflen
, "/%s (deleted)",
2885 dentry
->d_name
.name
);
2888 static struct dentry_operations anon_ops
= {
2889 .d_dname
= shmem_dname
2893 * shmem_file_setup - get an unlinked file living in tmpfs
2894 * @name: name for dentry (to be seen in /proc/<pid>/maps
2895 * @size: size to be set for the file
2896 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2898 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
2901 struct inode
*inode
;
2903 struct super_block
*sb
;
2906 if (IS_ERR(shm_mnt
))
2907 return ERR_CAST(shm_mnt
);
2909 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
2910 return ERR_PTR(-EINVAL
);
2912 if (shmem_acct_size(flags
, size
))
2913 return ERR_PTR(-ENOMEM
);
2915 res
= ERR_PTR(-ENOMEM
);
2917 this.len
= strlen(name
);
2918 this.hash
= 0; /* will go */
2919 sb
= shm_mnt
->mnt_sb
;
2920 path
.dentry
= d_alloc_pseudo(sb
, &this);
2923 d_set_d_op(path
.dentry
, &anon_ops
);
2924 path
.mnt
= mntget(shm_mnt
);
2926 res
= ERR_PTR(-ENOSPC
);
2927 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
2931 d_instantiate(path
.dentry
, inode
);
2932 inode
->i_size
= size
;
2933 clear_nlink(inode
); /* It is unlinked */
2934 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
2938 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
2939 &shmem_file_operations
);
2948 shmem_unacct_size(flags
, size
);
2951 EXPORT_SYMBOL_GPL(shmem_file_setup
);
2954 * shmem_zero_setup - setup a shared anonymous mapping
2955 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2957 int shmem_zero_setup(struct vm_area_struct
*vma
)
2960 loff_t size
= vma
->vm_end
- vma
->vm_start
;
2962 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
2964 return PTR_ERR(file
);
2968 vma
->vm_file
= file
;
2969 vma
->vm_ops
= &shmem_vm_ops
;
2974 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2975 * @mapping: the page's address_space
2976 * @index: the page index
2977 * @gfp: the page allocator flags to use if allocating
2979 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2980 * with any new page allocations done using the specified allocation flags.
2981 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2982 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2983 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2985 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2986 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2988 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
2989 pgoff_t index
, gfp_t gfp
)
2992 struct inode
*inode
= mapping
->host
;
2996 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
2997 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
2999 page
= ERR_PTR(error
);
3005 * The tiny !SHMEM case uses ramfs without swap
3007 return read_cache_page_gfp(mapping
, index
, gfp
);
3010 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);