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/pagemap.h>
29 #include <linux/file.h>
31 #include <linux/export.h>
32 #include <linux/swap.h>
34 static struct vfsmount
*shm_mnt
;
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
43 #include <linux/xattr.h>
44 #include <linux/exportfs.h>
45 #include <linux/posix_acl.h>
46 #include <linux/generic_acl.h>
47 #include <linux/mman.h>
48 #include <linux/string.h>
49 #include <linux/slab.h>
50 #include <linux/backing-dev.h>
51 #include <linux/shmem_fs.h>
52 #include <linux/writeback.h>
53 #include <linux/blkdev.h>
54 #include <linux/pagevec.h>
55 #include <linux/percpu_counter.h>
56 #include <linux/falloc.h>
57 #include <linux/splice.h>
58 #include <linux/security.h>
59 #include <linux/swapops.h>
60 #include <linux/mempolicy.h>
61 #include <linux/namei.h>
62 #include <linux/ctype.h>
63 #include <linux/migrate.h>
64 #include <linux/highmem.h>
65 #include <linux/seq_file.h>
66 #include <linux/magic.h>
68 #include <asm/uaccess.h>
69 #include <asm/pgtable.h>
71 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
72 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
74 /* Pretend that each entry is of this size in directory's i_size */
75 #define BOGO_DIRENT_SIZE 20
77 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
78 #define SHORT_SYMLINK_LEN 128
81 struct list_head list
; /* anchored by shmem_inode_info->xattr_list */
82 char *name
; /* xattr name */
87 /* Flag allocation requirements to shmem_getpage */
89 SGP_READ
, /* don't exceed i_size, don't allocate page */
90 SGP_CACHE
, /* don't exceed i_size, may allocate page */
91 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
92 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
93 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
97 static unsigned long shmem_default_max_blocks(void)
99 return totalram_pages
/ 2;
102 static unsigned long shmem_default_max_inodes(void)
104 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
108 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
109 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
110 struct shmem_inode_info
*info
, pgoff_t index
);
111 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
112 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
114 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
115 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
117 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
118 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
121 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
123 return sb
->s_fs_info
;
127 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
128 * for shared memory and for shared anonymous (/dev/zero) mappings
129 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
130 * consistent with the pre-accounting of private mappings ...
132 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
134 return (flags
& VM_NORESERVE
) ?
135 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
138 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
140 if (!(flags
& VM_NORESERVE
))
141 vm_unacct_memory(VM_ACCT(size
));
145 * ... whereas tmpfs objects are accounted incrementally as
146 * pages are allocated, in order to allow huge sparse files.
147 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
148 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
150 static inline int shmem_acct_block(unsigned long flags
)
152 return (flags
& VM_NORESERVE
) ?
153 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
156 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
158 if (flags
& VM_NORESERVE
)
159 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
162 static const struct super_operations shmem_ops
;
163 static const struct address_space_operations shmem_aops
;
164 static const struct file_operations shmem_file_operations
;
165 static const struct inode_operations shmem_inode_operations
;
166 static const struct inode_operations shmem_dir_inode_operations
;
167 static const struct inode_operations shmem_special_inode_operations
;
168 static const struct vm_operations_struct shmem_vm_ops
;
170 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
171 .ra_pages
= 0, /* No readahead */
172 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
175 static LIST_HEAD(shmem_swaplist
);
176 static DEFINE_MUTEX(shmem_swaplist_mutex
);
178 static int shmem_reserve_inode(struct super_block
*sb
)
180 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
181 if (sbinfo
->max_inodes
) {
182 spin_lock(&sbinfo
->stat_lock
);
183 if (!sbinfo
->free_inodes
) {
184 spin_unlock(&sbinfo
->stat_lock
);
187 sbinfo
->free_inodes
--;
188 spin_unlock(&sbinfo
->stat_lock
);
193 static void shmem_free_inode(struct super_block
*sb
)
195 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
196 if (sbinfo
->max_inodes
) {
197 spin_lock(&sbinfo
->stat_lock
);
198 sbinfo
->free_inodes
++;
199 spin_unlock(&sbinfo
->stat_lock
);
204 * shmem_recalc_inode - recalculate the block usage of an inode
205 * @inode: inode to recalc
207 * We have to calculate the free blocks since the mm can drop
208 * undirtied hole pages behind our back.
210 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
211 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
213 * It has to be called with the spinlock held.
215 static void shmem_recalc_inode(struct inode
*inode
)
217 struct shmem_inode_info
*info
= SHMEM_I(inode
);
220 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
222 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
223 if (sbinfo
->max_blocks
)
224 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
225 info
->alloced
-= freed
;
226 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
227 shmem_unacct_blocks(info
->flags
, freed
);
232 * Replace item expected in radix tree by a new item, while holding tree lock.
234 static int shmem_radix_tree_replace(struct address_space
*mapping
,
235 pgoff_t index
, void *expected
, void *replacement
)
240 VM_BUG_ON(!expected
);
241 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
243 item
= radix_tree_deref_slot_protected(pslot
,
244 &mapping
->tree_lock
);
245 if (item
!= expected
)
248 radix_tree_replace_slot(pslot
, replacement
);
250 radix_tree_delete(&mapping
->page_tree
, index
);
255 * Like add_to_page_cache_locked, but error if expected item has gone.
257 static int shmem_add_to_page_cache(struct page
*page
,
258 struct address_space
*mapping
,
259 pgoff_t index
, gfp_t gfp
, void *expected
)
263 VM_BUG_ON(!PageLocked(page
));
264 VM_BUG_ON(!PageSwapBacked(page
));
267 error
= radix_tree_preload(gfp
& GFP_RECLAIM_MASK
);
269 page_cache_get(page
);
270 page
->mapping
= mapping
;
273 spin_lock_irq(&mapping
->tree_lock
);
275 error
= radix_tree_insert(&mapping
->page_tree
,
278 error
= shmem_radix_tree_replace(mapping
, index
,
282 __inc_zone_page_state(page
, NR_FILE_PAGES
);
283 __inc_zone_page_state(page
, NR_SHMEM
);
284 spin_unlock_irq(&mapping
->tree_lock
);
286 page
->mapping
= NULL
;
287 spin_unlock_irq(&mapping
->tree_lock
);
288 page_cache_release(page
);
291 radix_tree_preload_end();
294 mem_cgroup_uncharge_cache_page(page
);
299 * Like delete_from_page_cache, but substitutes swap for page.
301 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
303 struct address_space
*mapping
= page
->mapping
;
306 spin_lock_irq(&mapping
->tree_lock
);
307 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
308 page
->mapping
= NULL
;
310 __dec_zone_page_state(page
, NR_FILE_PAGES
);
311 __dec_zone_page_state(page
, NR_SHMEM
);
312 spin_unlock_irq(&mapping
->tree_lock
);
313 page_cache_release(page
);
318 * Like find_get_pages, but collecting swap entries as well as pages.
320 static unsigned shmem_find_get_pages_and_swap(struct address_space
*mapping
,
321 pgoff_t start
, unsigned int nr_pages
,
322 struct page
**pages
, pgoff_t
*indices
)
326 unsigned int nr_found
;
330 nr_found
= radix_tree_gang_lookup_slot(&mapping
->page_tree
,
331 (void ***)pages
, indices
, start
, nr_pages
);
333 for (i
= 0; i
< nr_found
; i
++) {
336 page
= radix_tree_deref_slot((void **)pages
[i
]);
339 if (radix_tree_exception(page
)) {
340 if (radix_tree_deref_retry(page
))
343 * Otherwise, we must be storing a swap entry
344 * here as an exceptional entry: so return it
345 * without attempting to raise page count.
349 if (!page_cache_get_speculative(page
))
352 /* Has the page moved? */
353 if (unlikely(page
!= *((void **)pages
[i
]))) {
354 page_cache_release(page
);
358 indices
[ret
] = indices
[i
];
362 if (unlikely(!ret
&& nr_found
))
369 * Remove swap entry from radix tree, free the swap and its page cache.
371 static int shmem_free_swap(struct address_space
*mapping
,
372 pgoff_t index
, void *radswap
)
376 spin_lock_irq(&mapping
->tree_lock
);
377 error
= shmem_radix_tree_replace(mapping
, index
, radswap
, NULL
);
378 spin_unlock_irq(&mapping
->tree_lock
);
380 free_swap_and_cache(radix_to_swp_entry(radswap
));
385 * Pagevec may contain swap entries, so shuffle up pages before releasing.
387 static void shmem_deswap_pagevec(struct pagevec
*pvec
)
391 for (i
= 0, j
= 0; i
< pagevec_count(pvec
); i
++) {
392 struct page
*page
= pvec
->pages
[i
];
393 if (!radix_tree_exceptional_entry(page
))
394 pvec
->pages
[j
++] = page
;
400 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
402 void shmem_unlock_mapping(struct address_space
*mapping
)
405 pgoff_t indices
[PAGEVEC_SIZE
];
408 pagevec_init(&pvec
, 0);
410 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
412 while (!mapping_unevictable(mapping
)) {
414 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
415 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
417 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
418 PAGEVEC_SIZE
, pvec
.pages
, indices
);
421 index
= indices
[pvec
.nr
- 1] + 1;
422 shmem_deswap_pagevec(&pvec
);
423 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
424 pagevec_release(&pvec
);
430 * Remove range of pages and swap entries from radix tree, and free them.
431 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
433 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
436 struct address_space
*mapping
= inode
->i_mapping
;
437 struct shmem_inode_info
*info
= SHMEM_I(inode
);
438 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
439 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
440 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
441 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
443 pgoff_t indices
[PAGEVEC_SIZE
];
444 long nr_swaps_freed
= 0;
449 end
= -1; /* unsigned, so actually very big */
451 pagevec_init(&pvec
, 0);
453 while (index
< end
) {
454 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
455 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
456 pvec
.pages
, indices
);
459 mem_cgroup_uncharge_start();
460 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
461 struct page
*page
= pvec
.pages
[i
];
467 if (radix_tree_exceptional_entry(page
)) {
470 nr_swaps_freed
+= !shmem_free_swap(mapping
,
475 if (!trylock_page(page
))
477 if (!unfalloc
|| !PageUptodate(page
)) {
478 if (page
->mapping
== mapping
) {
479 VM_BUG_ON(PageWriteback(page
));
480 truncate_inode_page(mapping
, page
);
485 shmem_deswap_pagevec(&pvec
);
486 pagevec_release(&pvec
);
487 mem_cgroup_uncharge_end();
493 struct page
*page
= NULL
;
494 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
496 unsigned int top
= PAGE_CACHE_SIZE
;
501 zero_user_segment(page
, partial_start
, top
);
502 set_page_dirty(page
);
504 page_cache_release(page
);
508 struct page
*page
= NULL
;
509 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
511 zero_user_segment(page
, 0, partial_end
);
512 set_page_dirty(page
);
514 page_cache_release(page
);
523 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
524 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
525 pvec
.pages
, indices
);
527 if (index
== start
|| unfalloc
)
532 if ((index
== start
|| unfalloc
) && indices
[0] >= end
) {
533 shmem_deswap_pagevec(&pvec
);
534 pagevec_release(&pvec
);
537 mem_cgroup_uncharge_start();
538 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
539 struct page
*page
= pvec
.pages
[i
];
545 if (radix_tree_exceptional_entry(page
)) {
548 nr_swaps_freed
+= !shmem_free_swap(mapping
,
554 if (!unfalloc
|| !PageUptodate(page
)) {
555 if (page
->mapping
== mapping
) {
556 VM_BUG_ON(PageWriteback(page
));
557 truncate_inode_page(mapping
, page
);
562 shmem_deswap_pagevec(&pvec
);
563 pagevec_release(&pvec
);
564 mem_cgroup_uncharge_end();
568 spin_lock(&info
->lock
);
569 info
->swapped
-= nr_swaps_freed
;
570 shmem_recalc_inode(inode
);
571 spin_unlock(&info
->lock
);
574 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
576 shmem_undo_range(inode
, lstart
, lend
, false);
577 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
579 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
581 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
583 struct inode
*inode
= dentry
->d_inode
;
586 error
= inode_change_ok(inode
, attr
);
590 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
591 loff_t oldsize
= inode
->i_size
;
592 loff_t newsize
= attr
->ia_size
;
594 if (newsize
!= oldsize
) {
595 i_size_write(inode
, newsize
);
596 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
598 if (newsize
< oldsize
) {
599 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
600 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
601 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
602 /* unmap again to remove racily COWed private pages */
603 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
607 setattr_copy(inode
, attr
);
608 #ifdef CONFIG_TMPFS_POSIX_ACL
609 if (attr
->ia_valid
& ATTR_MODE
)
610 error
= generic_acl_chmod(inode
);
615 static void shmem_evict_inode(struct inode
*inode
)
617 struct shmem_inode_info
*info
= SHMEM_I(inode
);
618 struct shmem_xattr
*xattr
, *nxattr
;
620 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
621 shmem_unacct_size(info
->flags
, inode
->i_size
);
623 shmem_truncate_range(inode
, 0, (loff_t
)-1);
624 if (!list_empty(&info
->swaplist
)) {
625 mutex_lock(&shmem_swaplist_mutex
);
626 list_del_init(&info
->swaplist
);
627 mutex_unlock(&shmem_swaplist_mutex
);
630 kfree(info
->symlink
);
632 list_for_each_entry_safe(xattr
, nxattr
, &info
->xattr_list
, list
) {
636 BUG_ON(inode
->i_blocks
);
637 shmem_free_inode(inode
->i_sb
);
642 * If swap found in inode, free it and move page from swapcache to filecache.
644 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
645 swp_entry_t swap
, struct page
**pagep
)
647 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
653 radswap
= swp_to_radix_entry(swap
);
654 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
659 * Move _head_ to start search for next from here.
660 * But be careful: shmem_evict_inode checks list_empty without taking
661 * mutex, and there's an instant in list_move_tail when info->swaplist
662 * would appear empty, if it were the only one on shmem_swaplist.
664 if (shmem_swaplist
.next
!= &info
->swaplist
)
665 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
667 gfp
= mapping_gfp_mask(mapping
);
668 if (shmem_should_replace_page(*pagep
, gfp
)) {
669 mutex_unlock(&shmem_swaplist_mutex
);
670 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
671 mutex_lock(&shmem_swaplist_mutex
);
673 * We needed to drop mutex to make that restrictive page
674 * allocation; but the inode might already be freed by now,
675 * and we cannot refer to inode or mapping or info to check.
676 * However, we do hold page lock on the PageSwapCache page,
677 * so can check if that still has our reference remaining.
679 if (!page_swapcount(*pagep
))
684 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
685 * but also to hold up shmem_evict_inode(): so inode cannot be freed
686 * beneath us (pagelock doesn't help until the page is in pagecache).
689 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
690 GFP_NOWAIT
, radswap
);
691 if (error
!= -ENOMEM
) {
693 * Truncation and eviction use free_swap_and_cache(), which
694 * only does trylock page: if we raced, best clean up here.
696 delete_from_swap_cache(*pagep
);
697 set_page_dirty(*pagep
);
699 spin_lock(&info
->lock
);
701 spin_unlock(&info
->lock
);
704 error
= 1; /* not an error, but entry was found */
710 * Search through swapped inodes to find and replace swap by page.
712 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
714 struct list_head
*this, *next
;
715 struct shmem_inode_info
*info
;
720 * There's a faint possibility that swap page was replaced before
721 * caller locked it: it will come back later with the right page.
723 if (unlikely(!PageSwapCache(page
)))
727 * Charge page using GFP_KERNEL while we can wait, before taking
728 * the shmem_swaplist_mutex which might hold up shmem_writepage().
729 * Charged back to the user (not to caller) when swap account is used.
731 error
= mem_cgroup_cache_charge(page
, current
->mm
, GFP_KERNEL
);
734 /* No radix_tree_preload: swap entry keeps a place for page in tree */
736 mutex_lock(&shmem_swaplist_mutex
);
737 list_for_each_safe(this, next
, &shmem_swaplist
) {
738 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
740 found
= shmem_unuse_inode(info
, swap
, &page
);
742 list_del_init(&info
->swaplist
);
747 mutex_unlock(&shmem_swaplist_mutex
);
753 page_cache_release(page
);
758 * Move the page from the page cache to the swap cache.
760 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
762 struct shmem_inode_info
*info
;
763 struct address_space
*mapping
;
768 BUG_ON(!PageLocked(page
));
769 mapping
= page
->mapping
;
771 inode
= mapping
->host
;
772 info
= SHMEM_I(inode
);
773 if (info
->flags
& VM_LOCKED
)
775 if (!total_swap_pages
)
779 * shmem_backing_dev_info's capabilities prevent regular writeback or
780 * sync from ever calling shmem_writepage; but a stacking filesystem
781 * might use ->writepage of its underlying filesystem, in which case
782 * tmpfs should write out to swap only in response to memory pressure,
783 * and not for the writeback threads or sync.
785 if (!wbc
->for_reclaim
) {
786 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
791 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
792 * value into swapfile.c, the only way we can correctly account for a
793 * fallocated page arriving here is now to initialize it and write it.
795 if (!PageUptodate(page
)) {
796 clear_highpage(page
);
797 flush_dcache_page(page
);
798 SetPageUptodate(page
);
801 swap
= get_swap_page();
806 * Add inode to shmem_unuse()'s list of swapped-out inodes,
807 * if it's not already there. Do it now before the page is
808 * moved to swap cache, when its pagelock no longer protects
809 * the inode from eviction. But don't unlock the mutex until
810 * we've incremented swapped, because shmem_unuse_inode() will
811 * prune a !swapped inode from the swaplist under this mutex.
813 mutex_lock(&shmem_swaplist_mutex
);
814 if (list_empty(&info
->swaplist
))
815 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
817 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
818 swap_shmem_alloc(swap
);
819 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
821 spin_lock(&info
->lock
);
823 shmem_recalc_inode(inode
);
824 spin_unlock(&info
->lock
);
826 mutex_unlock(&shmem_swaplist_mutex
);
827 BUG_ON(page_mapped(page
));
828 swap_writepage(page
, wbc
);
832 mutex_unlock(&shmem_swaplist_mutex
);
833 swapcache_free(swap
, NULL
);
835 set_page_dirty(page
);
836 if (wbc
->for_reclaim
)
837 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
844 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
848 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
849 return; /* show nothing */
851 mpol_to_str(buffer
, sizeof(buffer
), mpol
, 1);
853 seq_printf(seq
, ",mpol=%s", buffer
);
856 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
858 struct mempolicy
*mpol
= NULL
;
860 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
863 spin_unlock(&sbinfo
->stat_lock
);
867 #endif /* CONFIG_TMPFS */
869 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
870 struct shmem_inode_info
*info
, pgoff_t index
)
872 struct mempolicy mpol
, *spol
;
873 struct vm_area_struct pvma
;
875 spol
= mpol_cond_copy(&mpol
,
876 mpol_shared_policy_lookup(&info
->policy
, index
));
878 /* Create a pseudo vma that just contains the policy */
880 pvma
.vm_pgoff
= index
;
882 pvma
.vm_policy
= spol
;
883 return swapin_readahead(swap
, gfp
, &pvma
, 0);
886 static struct page
*shmem_alloc_page(gfp_t gfp
,
887 struct shmem_inode_info
*info
, pgoff_t index
)
889 struct vm_area_struct pvma
;
891 /* Create a pseudo vma that just contains the policy */
893 pvma
.vm_pgoff
= index
;
895 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
898 * alloc_page_vma() will drop the shared policy reference
900 return alloc_page_vma(gfp
, &pvma
, 0);
902 #else /* !CONFIG_NUMA */
904 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
907 #endif /* CONFIG_TMPFS */
909 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
910 struct shmem_inode_info
*info
, pgoff_t index
)
912 return swapin_readahead(swap
, gfp
, NULL
, 0);
915 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
916 struct shmem_inode_info
*info
, pgoff_t index
)
918 return alloc_page(gfp
);
920 #endif /* CONFIG_NUMA */
922 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
923 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
930 * When a page is moved from swapcache to shmem filecache (either by the
931 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
932 * shmem_unuse_inode()), it may have been read in earlier from swap, in
933 * ignorance of the mapping it belongs to. If that mapping has special
934 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
935 * we may need to copy to a suitable page before moving to filecache.
937 * In a future release, this may well be extended to respect cpuset and
938 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
939 * but for now it is a simple matter of zone.
941 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
943 return page_zonenum(page
) > gfp_zone(gfp
);
946 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
947 struct shmem_inode_info
*info
, pgoff_t index
)
949 struct page
*oldpage
, *newpage
;
950 struct address_space
*swap_mapping
;
955 swap_index
= page_private(oldpage
);
956 swap_mapping
= page_mapping(oldpage
);
959 * We have arrived here because our zones are constrained, so don't
960 * limit chance of success by further cpuset and node constraints.
962 gfp
&= ~GFP_CONSTRAINT_MASK
;
963 newpage
= shmem_alloc_page(gfp
, info
, index
);
966 VM_BUG_ON(shmem_should_replace_page(newpage
, gfp
));
969 page_cache_get(newpage
);
970 copy_highpage(newpage
, oldpage
);
972 VM_BUG_ON(!PageLocked(oldpage
));
973 __set_page_locked(newpage
);
974 VM_BUG_ON(!PageUptodate(oldpage
));
975 SetPageUptodate(newpage
);
976 VM_BUG_ON(!PageSwapBacked(oldpage
));
977 SetPageSwapBacked(newpage
);
978 VM_BUG_ON(!swap_index
);
979 set_page_private(newpage
, swap_index
);
980 VM_BUG_ON(!PageSwapCache(oldpage
));
981 SetPageSwapCache(newpage
);
984 * Our caller will very soon move newpage out of swapcache, but it's
985 * a nice clean interface for us to replace oldpage by newpage there.
987 spin_lock_irq(&swap_mapping
->tree_lock
);
988 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
990 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
991 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
992 spin_unlock_irq(&swap_mapping
->tree_lock
);
995 mem_cgroup_replace_page_cache(oldpage
, newpage
);
996 lru_cache_add_anon(newpage
);
998 ClearPageSwapCache(oldpage
);
999 set_page_private(oldpage
, 0);
1001 unlock_page(oldpage
);
1002 page_cache_release(oldpage
);
1003 page_cache_release(oldpage
);
1008 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1010 * If we allocate a new one we do not mark it dirty. That's up to the
1011 * vm. If we swap it in we mark it dirty since we also free the swap
1012 * entry since a page cannot live in both the swap and page cache
1014 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1015 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1017 struct address_space
*mapping
= inode
->i_mapping
;
1018 struct shmem_inode_info
*info
;
1019 struct shmem_sb_info
*sbinfo
;
1026 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1030 page
= find_lock_page(mapping
, index
);
1031 if (radix_tree_exceptional_entry(page
)) {
1032 swap
= radix_to_swp_entry(page
);
1036 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1037 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1042 /* fallocated page? */
1043 if (page
&& !PageUptodate(page
)) {
1044 if (sgp
!= SGP_READ
)
1047 page_cache_release(page
);
1050 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1056 * Fast cache lookup did not find it:
1057 * bring it back from swap or allocate.
1059 info
= SHMEM_I(inode
);
1060 sbinfo
= SHMEM_SB(inode
->i_sb
);
1063 /* Look it up and read it in.. */
1064 page
= lookup_swap_cache(swap
);
1066 /* here we actually do the io */
1068 *fault_type
|= VM_FAULT_MAJOR
;
1069 page
= shmem_swapin(swap
, gfp
, info
, index
);
1076 /* We have to do this with page locked to prevent races */
1078 if (!PageSwapCache(page
) || page
->mapping
) {
1079 error
= -EEXIST
; /* try again */
1082 if (!PageUptodate(page
)) {
1086 wait_on_page_writeback(page
);
1088 if (shmem_should_replace_page(page
, gfp
)) {
1089 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1094 error
= mem_cgroup_cache_charge(page
, current
->mm
,
1095 gfp
& GFP_RECLAIM_MASK
);
1097 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1098 gfp
, swp_to_radix_entry(swap
));
1102 spin_lock(&info
->lock
);
1104 shmem_recalc_inode(inode
);
1105 spin_unlock(&info
->lock
);
1107 delete_from_swap_cache(page
);
1108 set_page_dirty(page
);
1112 if (shmem_acct_block(info
->flags
)) {
1116 if (sbinfo
->max_blocks
) {
1117 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1118 sbinfo
->max_blocks
) >= 0) {
1122 percpu_counter_inc(&sbinfo
->used_blocks
);
1125 page
= shmem_alloc_page(gfp
, info
, index
);
1131 SetPageSwapBacked(page
);
1132 __set_page_locked(page
);
1133 error
= mem_cgroup_cache_charge(page
, current
->mm
,
1134 gfp
& GFP_RECLAIM_MASK
);
1136 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1140 lru_cache_add_anon(page
);
1142 spin_lock(&info
->lock
);
1144 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1145 shmem_recalc_inode(inode
);
1146 spin_unlock(&info
->lock
);
1150 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1152 if (sgp
== SGP_FALLOC
)
1156 * Let SGP_WRITE caller clear ends if write does not fill page;
1157 * but SGP_FALLOC on a page fallocated earlier must initialize
1158 * it now, lest undo on failure cancel our earlier guarantee.
1160 if (sgp
!= SGP_WRITE
) {
1161 clear_highpage(page
);
1162 flush_dcache_page(page
);
1163 SetPageUptodate(page
);
1165 if (sgp
== SGP_DIRTY
)
1166 set_page_dirty(page
);
1169 /* Perhaps the file has been truncated since we checked */
1170 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1171 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1185 info
= SHMEM_I(inode
);
1186 ClearPageDirty(page
);
1187 delete_from_page_cache(page
);
1188 spin_lock(&info
->lock
);
1190 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1191 spin_unlock(&info
->lock
);
1193 sbinfo
= SHMEM_SB(inode
->i_sb
);
1194 if (sbinfo
->max_blocks
)
1195 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1197 shmem_unacct_blocks(info
->flags
, 1);
1199 if (swap
.val
&& error
!= -EINVAL
) {
1200 struct page
*test
= find_get_page(mapping
, index
);
1201 if (test
&& !radix_tree_exceptional_entry(test
))
1202 page_cache_release(test
);
1203 /* Have another try if the entry has changed */
1204 if (test
!= swp_to_radix_entry(swap
))
1209 page_cache_release(page
);
1211 if (error
== -ENOSPC
&& !once
++) {
1212 info
= SHMEM_I(inode
);
1213 spin_lock(&info
->lock
);
1214 shmem_recalc_inode(inode
);
1215 spin_unlock(&info
->lock
);
1218 if (error
== -EEXIST
)
1223 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1225 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1227 int ret
= VM_FAULT_LOCKED
;
1229 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1231 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1233 if (ret
& VM_FAULT_MAJOR
) {
1234 count_vm_event(PGMAJFAULT
);
1235 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1241 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1243 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1244 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1247 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1250 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1253 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1254 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1258 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1260 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1261 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1262 int retval
= -ENOMEM
;
1264 spin_lock(&info
->lock
);
1265 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1266 if (!user_shm_lock(inode
->i_size
, user
))
1268 info
->flags
|= VM_LOCKED
;
1269 mapping_set_unevictable(file
->f_mapping
);
1271 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1272 user_shm_unlock(inode
->i_size
, user
);
1273 info
->flags
&= ~VM_LOCKED
;
1274 mapping_clear_unevictable(file
->f_mapping
);
1279 spin_unlock(&info
->lock
);
1283 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1285 file_accessed(file
);
1286 vma
->vm_ops
= &shmem_vm_ops
;
1287 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1291 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1292 umode_t mode
, dev_t dev
, unsigned long flags
)
1294 struct inode
*inode
;
1295 struct shmem_inode_info
*info
;
1296 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1298 if (shmem_reserve_inode(sb
))
1301 inode
= new_inode(sb
);
1303 inode
->i_ino
= get_next_ino();
1304 inode_init_owner(inode
, dir
, mode
);
1305 inode
->i_blocks
= 0;
1306 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1307 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1308 inode
->i_generation
= get_seconds();
1309 info
= SHMEM_I(inode
);
1310 memset(info
, 0, (char *)inode
- (char *)info
);
1311 spin_lock_init(&info
->lock
);
1312 info
->flags
= flags
& VM_NORESERVE
;
1313 INIT_LIST_HEAD(&info
->swaplist
);
1314 INIT_LIST_HEAD(&info
->xattr_list
);
1315 cache_no_acl(inode
);
1317 switch (mode
& S_IFMT
) {
1319 inode
->i_op
= &shmem_special_inode_operations
;
1320 init_special_inode(inode
, mode
, dev
);
1323 inode
->i_mapping
->a_ops
= &shmem_aops
;
1324 inode
->i_op
= &shmem_inode_operations
;
1325 inode
->i_fop
= &shmem_file_operations
;
1326 mpol_shared_policy_init(&info
->policy
,
1327 shmem_get_sbmpol(sbinfo
));
1331 /* Some things misbehave if size == 0 on a directory */
1332 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1333 inode
->i_op
= &shmem_dir_inode_operations
;
1334 inode
->i_fop
= &simple_dir_operations
;
1338 * Must not load anything in the rbtree,
1339 * mpol_free_shared_policy will not be called.
1341 mpol_shared_policy_init(&info
->policy
, NULL
);
1345 shmem_free_inode(sb
);
1350 static const struct inode_operations shmem_symlink_inode_operations
;
1351 static const struct inode_operations shmem_short_symlink_operations
;
1353 #ifdef CONFIG_TMPFS_XATTR
1354 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1356 #define shmem_initxattrs NULL
1360 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1361 loff_t pos
, unsigned len
, unsigned flags
,
1362 struct page
**pagep
, void **fsdata
)
1364 struct inode
*inode
= mapping
->host
;
1365 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1366 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1370 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1371 loff_t pos
, unsigned len
, unsigned copied
,
1372 struct page
*page
, void *fsdata
)
1374 struct inode
*inode
= mapping
->host
;
1376 if (pos
+ copied
> inode
->i_size
)
1377 i_size_write(inode
, pos
+ copied
);
1379 if (!PageUptodate(page
)) {
1380 if (copied
< PAGE_CACHE_SIZE
) {
1381 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1382 zero_user_segments(page
, 0, from
,
1383 from
+ copied
, PAGE_CACHE_SIZE
);
1385 SetPageUptodate(page
);
1387 set_page_dirty(page
);
1389 page_cache_release(page
);
1394 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1396 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
1397 struct address_space
*mapping
= inode
->i_mapping
;
1399 unsigned long offset
;
1400 enum sgp_type sgp
= SGP_READ
;
1403 * Might this read be for a stacking filesystem? Then when reading
1404 * holes of a sparse file, we actually need to allocate those pages,
1405 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1407 if (segment_eq(get_fs(), KERNEL_DS
))
1410 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1411 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1414 struct page
*page
= NULL
;
1416 unsigned long nr
, ret
;
1417 loff_t i_size
= i_size_read(inode
);
1419 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1420 if (index
> end_index
)
1422 if (index
== end_index
) {
1423 nr
= i_size
& ~PAGE_CACHE_MASK
;
1428 desc
->error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1430 if (desc
->error
== -EINVAL
)
1438 * We must evaluate after, since reads (unlike writes)
1439 * are called without i_mutex protection against truncate
1441 nr
= PAGE_CACHE_SIZE
;
1442 i_size
= i_size_read(inode
);
1443 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1444 if (index
== end_index
) {
1445 nr
= i_size
& ~PAGE_CACHE_MASK
;
1448 page_cache_release(page
);
1456 * If users can be writing to this page using arbitrary
1457 * virtual addresses, take care about potential aliasing
1458 * before reading the page on the kernel side.
1460 if (mapping_writably_mapped(mapping
))
1461 flush_dcache_page(page
);
1463 * Mark the page accessed if we read the beginning.
1466 mark_page_accessed(page
);
1468 page
= ZERO_PAGE(0);
1469 page_cache_get(page
);
1473 * Ok, we have the page, and it's up-to-date, so
1474 * now we can copy it to user space...
1476 * The actor routine returns how many bytes were actually used..
1477 * NOTE! This may not be the same as how much of a user buffer
1478 * we filled up (we may be padding etc), so we can only update
1479 * "pos" here (the actor routine has to update the user buffer
1480 * pointers and the remaining count).
1482 ret
= actor(desc
, page
, offset
, nr
);
1484 index
+= offset
>> PAGE_CACHE_SHIFT
;
1485 offset
&= ~PAGE_CACHE_MASK
;
1487 page_cache_release(page
);
1488 if (ret
!= nr
|| !desc
->count
)
1494 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1495 file_accessed(filp
);
1498 static ssize_t
shmem_file_aio_read(struct kiocb
*iocb
,
1499 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
1501 struct file
*filp
= iocb
->ki_filp
;
1505 loff_t
*ppos
= &iocb
->ki_pos
;
1507 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1511 for (seg
= 0; seg
< nr_segs
; seg
++) {
1512 read_descriptor_t desc
;
1515 desc
.arg
.buf
= iov
[seg
].iov_base
;
1516 desc
.count
= iov
[seg
].iov_len
;
1517 if (desc
.count
== 0)
1520 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1521 retval
+= desc
.written
;
1523 retval
= retval
?: desc
.error
;
1532 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1533 struct pipe_inode_info
*pipe
, size_t len
,
1536 struct address_space
*mapping
= in
->f_mapping
;
1537 struct inode
*inode
= mapping
->host
;
1538 unsigned int loff
, nr_pages
, req_pages
;
1539 struct page
*pages
[PIPE_DEF_BUFFERS
];
1540 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1542 pgoff_t index
, end_index
;
1545 struct splice_pipe_desc spd
= {
1549 .ops
= &page_cache_pipe_buf_ops
,
1550 .spd_release
= spd_release_page
,
1553 isize
= i_size_read(inode
);
1554 if (unlikely(*ppos
>= isize
))
1557 left
= isize
- *ppos
;
1558 if (unlikely(left
< len
))
1561 if (splice_grow_spd(pipe
, &spd
))
1564 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1565 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1566 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1567 nr_pages
= min(req_pages
, pipe
->buffers
);
1569 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1570 nr_pages
, spd
.pages
);
1571 index
+= spd
.nr_pages
;
1574 while (spd
.nr_pages
< nr_pages
) {
1575 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1579 spd
.pages
[spd
.nr_pages
++] = page
;
1583 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1584 nr_pages
= spd
.nr_pages
;
1587 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1588 unsigned int this_len
;
1593 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1594 page
= spd
.pages
[page_nr
];
1596 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1597 error
= shmem_getpage(inode
, index
, &page
,
1602 page_cache_release(spd
.pages
[page_nr
]);
1603 spd
.pages
[page_nr
] = page
;
1606 isize
= i_size_read(inode
);
1607 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1608 if (unlikely(!isize
|| index
> end_index
))
1611 if (end_index
== index
) {
1614 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1618 this_len
= min(this_len
, plen
- loff
);
1622 spd
.partial
[page_nr
].offset
= loff
;
1623 spd
.partial
[page_nr
].len
= this_len
;
1630 while (page_nr
< nr_pages
)
1631 page_cache_release(spd
.pages
[page_nr
++]);
1634 error
= splice_to_pipe(pipe
, &spd
);
1636 splice_shrink_spd(pipe
, &spd
);
1645 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
1648 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1649 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1650 pgoff_t start
, index
, end
;
1653 mutex_lock(&inode
->i_mutex
);
1655 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1656 struct address_space
*mapping
= file
->f_mapping
;
1657 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
1658 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
1660 if ((u64
)unmap_end
> (u64
)unmap_start
)
1661 unmap_mapping_range(mapping
, unmap_start
,
1662 1 + unmap_end
- unmap_start
, 0);
1663 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
1664 /* No need to unmap again: hole-punching leaves COWed pages */
1669 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1670 error
= inode_newsize_ok(inode
, offset
+ len
);
1674 start
= offset
>> PAGE_CACHE_SHIFT
;
1675 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1676 /* Try to avoid a swapstorm if len is impossible to satisfy */
1677 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
1682 for (index
= start
; index
< end
; index
++) {
1686 * Good, the fallocate(2) manpage permits EINTR: we may have
1687 * been interrupted because we are using up too much memory.
1689 if (signal_pending(current
))
1692 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
1695 /* Remove the !PageUptodate pages we added */
1696 shmem_undo_range(inode
,
1697 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
1698 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
1703 * If !PageUptodate, leave it that way so that freeable pages
1704 * can be recognized if we need to rollback on error later.
1705 * But set_page_dirty so that memory pressure will swap rather
1706 * than free the pages we are allocating (and SGP_CACHE pages
1707 * might still be clean: we now need to mark those dirty too).
1709 set_page_dirty(page
);
1711 page_cache_release(page
);
1715 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
1716 i_size_write(inode
, offset
+ len
);
1718 inode
->i_ctime
= CURRENT_TIME
;
1720 mutex_unlock(&inode
->i_mutex
);
1724 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1726 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
1728 buf
->f_type
= TMPFS_MAGIC
;
1729 buf
->f_bsize
= PAGE_CACHE_SIZE
;
1730 buf
->f_namelen
= NAME_MAX
;
1731 if (sbinfo
->max_blocks
) {
1732 buf
->f_blocks
= sbinfo
->max_blocks
;
1734 buf
->f_bfree
= sbinfo
->max_blocks
-
1735 percpu_counter_sum(&sbinfo
->used_blocks
);
1737 if (sbinfo
->max_inodes
) {
1738 buf
->f_files
= sbinfo
->max_inodes
;
1739 buf
->f_ffree
= sbinfo
->free_inodes
;
1741 /* else leave those fields 0 like simple_statfs */
1746 * File creation. Allocate an inode, and we're done..
1749 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
1751 struct inode
*inode
;
1752 int error
= -ENOSPC
;
1754 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
1756 error
= security_inode_init_security(inode
, dir
,
1758 shmem_initxattrs
, NULL
);
1760 if (error
!= -EOPNOTSUPP
) {
1765 #ifdef CONFIG_TMPFS_POSIX_ACL
1766 error
= generic_acl_init(inode
, dir
);
1774 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1775 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1776 d_instantiate(dentry
, inode
);
1777 dget(dentry
); /* Extra count - pin the dentry in core */
1782 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1786 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
1792 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
1793 struct nameidata
*nd
)
1795 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
1801 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1803 struct inode
*inode
= old_dentry
->d_inode
;
1807 * No ordinary (disk based) filesystem counts links as inodes;
1808 * but each new link needs a new dentry, pinning lowmem, and
1809 * tmpfs dentries cannot be pruned until they are unlinked.
1811 ret
= shmem_reserve_inode(inode
->i_sb
);
1815 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1816 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1818 ihold(inode
); /* New dentry reference */
1819 dget(dentry
); /* Extra pinning count for the created dentry */
1820 d_instantiate(dentry
, inode
);
1825 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
1827 struct inode
*inode
= dentry
->d_inode
;
1829 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
1830 shmem_free_inode(inode
->i_sb
);
1832 dir
->i_size
-= BOGO_DIRENT_SIZE
;
1833 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1835 dput(dentry
); /* Undo the count from "create" - this does all the work */
1839 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1841 if (!simple_empty(dentry
))
1844 drop_nlink(dentry
->d_inode
);
1846 return shmem_unlink(dir
, dentry
);
1850 * The VFS layer already does all the dentry stuff for rename,
1851 * we just have to decrement the usage count for the target if
1852 * it exists so that the VFS layer correctly free's it when it
1855 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
1857 struct inode
*inode
= old_dentry
->d_inode
;
1858 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
1860 if (!simple_empty(new_dentry
))
1863 if (new_dentry
->d_inode
) {
1864 (void) shmem_unlink(new_dir
, new_dentry
);
1866 drop_nlink(old_dir
);
1867 } else if (they_are_dirs
) {
1868 drop_nlink(old_dir
);
1872 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
1873 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
1874 old_dir
->i_ctime
= old_dir
->i_mtime
=
1875 new_dir
->i_ctime
= new_dir
->i_mtime
=
1876 inode
->i_ctime
= CURRENT_TIME
;
1880 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1884 struct inode
*inode
;
1887 struct shmem_inode_info
*info
;
1889 len
= strlen(symname
) + 1;
1890 if (len
> PAGE_CACHE_SIZE
)
1891 return -ENAMETOOLONG
;
1893 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
1897 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
1898 shmem_initxattrs
, NULL
);
1900 if (error
!= -EOPNOTSUPP
) {
1907 info
= SHMEM_I(inode
);
1908 inode
->i_size
= len
-1;
1909 if (len
<= SHORT_SYMLINK_LEN
) {
1910 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
1911 if (!info
->symlink
) {
1915 inode
->i_op
= &shmem_short_symlink_operations
;
1917 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
1922 inode
->i_mapping
->a_ops
= &shmem_aops
;
1923 inode
->i_op
= &shmem_symlink_inode_operations
;
1924 kaddr
= kmap_atomic(page
);
1925 memcpy(kaddr
, symname
, len
);
1926 kunmap_atomic(kaddr
);
1927 SetPageUptodate(page
);
1928 set_page_dirty(page
);
1930 page_cache_release(page
);
1932 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1933 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1934 d_instantiate(dentry
, inode
);
1939 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
1941 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
1945 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1947 struct page
*page
= NULL
;
1948 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
1949 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
1955 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
1957 if (!IS_ERR(nd_get_link(nd
))) {
1958 struct page
*page
= cookie
;
1960 mark_page_accessed(page
);
1961 page_cache_release(page
);
1965 #ifdef CONFIG_TMPFS_XATTR
1967 * Superblocks without xattr inode operations may get some security.* xattr
1968 * support from the LSM "for free". As soon as we have any other xattrs
1969 * like ACLs, we also need to implement the security.* handlers at
1970 * filesystem level, though.
1974 * Allocate new xattr and copy in the value; but leave the name to callers.
1976 static struct shmem_xattr
*shmem_xattr_alloc(const void *value
, size_t size
)
1978 struct shmem_xattr
*new_xattr
;
1982 len
= sizeof(*new_xattr
) + size
;
1983 if (len
<= sizeof(*new_xattr
))
1986 new_xattr
= kmalloc(len
, GFP_KERNEL
);
1990 new_xattr
->size
= size
;
1991 memcpy(new_xattr
->value
, value
, size
);
1996 * Callback for security_inode_init_security() for acquiring xattrs.
1998 static int shmem_initxattrs(struct inode
*inode
,
1999 const struct xattr
*xattr_array
,
2002 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2003 const struct xattr
*xattr
;
2004 struct shmem_xattr
*new_xattr
;
2007 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2008 new_xattr
= shmem_xattr_alloc(xattr
->value
, xattr
->value_len
);
2012 len
= strlen(xattr
->name
) + 1;
2013 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2015 if (!new_xattr
->name
) {
2020 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2021 XATTR_SECURITY_PREFIX_LEN
);
2022 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2025 spin_lock(&info
->lock
);
2026 list_add(&new_xattr
->list
, &info
->xattr_list
);
2027 spin_unlock(&info
->lock
);
2033 static int shmem_xattr_get(struct dentry
*dentry
, const char *name
,
2034 void *buffer
, size_t size
)
2036 struct shmem_inode_info
*info
;
2037 struct shmem_xattr
*xattr
;
2040 info
= SHMEM_I(dentry
->d_inode
);
2042 spin_lock(&info
->lock
);
2043 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
2044 if (strcmp(name
, xattr
->name
))
2049 if (size
< xattr
->size
)
2052 memcpy(buffer
, xattr
->value
, xattr
->size
);
2056 spin_unlock(&info
->lock
);
2060 static int shmem_xattr_set(struct inode
*inode
, const char *name
,
2061 const void *value
, size_t size
, int flags
)
2063 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2064 struct shmem_xattr
*xattr
;
2065 struct shmem_xattr
*new_xattr
= NULL
;
2068 /* value == NULL means remove */
2070 new_xattr
= shmem_xattr_alloc(value
, size
);
2074 new_xattr
->name
= kstrdup(name
, GFP_KERNEL
);
2075 if (!new_xattr
->name
) {
2081 spin_lock(&info
->lock
);
2082 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
2083 if (!strcmp(name
, xattr
->name
)) {
2084 if (flags
& XATTR_CREATE
) {
2087 } else if (new_xattr
) {
2088 list_replace(&xattr
->list
, &new_xattr
->list
);
2090 list_del(&xattr
->list
);
2095 if (flags
& XATTR_REPLACE
) {
2099 list_add(&new_xattr
->list
, &info
->xattr_list
);
2103 spin_unlock(&info
->lock
);
2110 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2111 #ifdef CONFIG_TMPFS_POSIX_ACL
2112 &generic_acl_access_handler
,
2113 &generic_acl_default_handler
,
2118 static int shmem_xattr_validate(const char *name
)
2120 struct { const char *prefix
; size_t len
; } arr
[] = {
2121 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2122 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2126 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2127 size_t preflen
= arr
[i
].len
;
2128 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2137 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2138 void *buffer
, size_t size
)
2143 * If this is a request for a synthetic attribute in the system.*
2144 * namespace use the generic infrastructure to resolve a handler
2145 * for it via sb->s_xattr.
2147 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2148 return generic_getxattr(dentry
, name
, buffer
, size
);
2150 err
= shmem_xattr_validate(name
);
2154 return shmem_xattr_get(dentry
, name
, buffer
, size
);
2157 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2158 const void *value
, size_t size
, int flags
)
2163 * If this is a request for a synthetic attribute in the system.*
2164 * namespace use the generic infrastructure to resolve a handler
2165 * for it via sb->s_xattr.
2167 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2168 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2170 err
= shmem_xattr_validate(name
);
2175 value
= ""; /* empty EA, do not remove */
2177 return shmem_xattr_set(dentry
->d_inode
, name
, value
, size
, flags
);
2181 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2186 * If this is a request for a synthetic attribute in the system.*
2187 * namespace use the generic infrastructure to resolve a handler
2188 * for it via sb->s_xattr.
2190 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2191 return generic_removexattr(dentry
, name
);
2193 err
= shmem_xattr_validate(name
);
2197 return shmem_xattr_set(dentry
->d_inode
, name
, NULL
, 0, XATTR_REPLACE
);
2200 static bool xattr_is_trusted(const char *name
)
2202 return !strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
);
2205 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2207 bool trusted
= capable(CAP_SYS_ADMIN
);
2208 struct shmem_xattr
*xattr
;
2209 struct shmem_inode_info
*info
;
2212 info
= SHMEM_I(dentry
->d_inode
);
2214 spin_lock(&info
->lock
);
2215 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
2218 /* skip "trusted." attributes for unprivileged callers */
2219 if (!trusted
&& xattr_is_trusted(xattr
->name
))
2222 len
= strlen(xattr
->name
) + 1;
2229 memcpy(buffer
, xattr
->name
, len
);
2233 spin_unlock(&info
->lock
);
2237 #endif /* CONFIG_TMPFS_XATTR */
2239 static const struct inode_operations shmem_short_symlink_operations
= {
2240 .readlink
= generic_readlink
,
2241 .follow_link
= shmem_follow_short_symlink
,
2242 #ifdef CONFIG_TMPFS_XATTR
2243 .setxattr
= shmem_setxattr
,
2244 .getxattr
= shmem_getxattr
,
2245 .listxattr
= shmem_listxattr
,
2246 .removexattr
= shmem_removexattr
,
2250 static const struct inode_operations shmem_symlink_inode_operations
= {
2251 .readlink
= generic_readlink
,
2252 .follow_link
= shmem_follow_link
,
2253 .put_link
= shmem_put_link
,
2254 #ifdef CONFIG_TMPFS_XATTR
2255 .setxattr
= shmem_setxattr
,
2256 .getxattr
= shmem_getxattr
,
2257 .listxattr
= shmem_listxattr
,
2258 .removexattr
= shmem_removexattr
,
2262 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2264 return ERR_PTR(-ESTALE
);
2267 static int shmem_match(struct inode
*ino
, void *vfh
)
2271 inum
= (inum
<< 32) | fh
[1];
2272 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2275 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2276 struct fid
*fid
, int fh_len
, int fh_type
)
2278 struct inode
*inode
;
2279 struct dentry
*dentry
= NULL
;
2280 u64 inum
= fid
->raw
[2];
2281 inum
= (inum
<< 32) | fid
->raw
[1];
2286 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2287 shmem_match
, fid
->raw
);
2289 dentry
= d_find_alias(inode
);
2296 static int shmem_encode_fh(struct dentry
*dentry
, __u32
*fh
, int *len
,
2299 struct inode
*inode
= dentry
->d_inode
;
2306 if (inode_unhashed(inode
)) {
2307 /* Unfortunately insert_inode_hash is not idempotent,
2308 * so as we hash inodes here rather than at creation
2309 * time, we need a lock to ensure we only try
2312 static DEFINE_SPINLOCK(lock
);
2314 if (inode_unhashed(inode
))
2315 __insert_inode_hash(inode
,
2316 inode
->i_ino
+ inode
->i_generation
);
2320 fh
[0] = inode
->i_generation
;
2321 fh
[1] = inode
->i_ino
;
2322 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2328 static const struct export_operations shmem_export_ops
= {
2329 .get_parent
= shmem_get_parent
,
2330 .encode_fh
= shmem_encode_fh
,
2331 .fh_to_dentry
= shmem_fh_to_dentry
,
2334 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2337 char *this_char
, *value
, *rest
;
2341 while (options
!= NULL
) {
2342 this_char
= options
;
2345 * NUL-terminate this option: unfortunately,
2346 * mount options form a comma-separated list,
2347 * but mpol's nodelist may also contain commas.
2349 options
= strchr(options
, ',');
2350 if (options
== NULL
)
2353 if (!isdigit(*options
)) {
2360 if ((value
= strchr(this_char
,'=')) != NULL
) {
2364 "tmpfs: No value for mount option '%s'\n",
2369 if (!strcmp(this_char
,"size")) {
2370 unsigned long long size
;
2371 size
= memparse(value
,&rest
);
2373 size
<<= PAGE_SHIFT
;
2374 size
*= totalram_pages
;
2380 sbinfo
->max_blocks
=
2381 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2382 } else if (!strcmp(this_char
,"nr_blocks")) {
2383 sbinfo
->max_blocks
= memparse(value
, &rest
);
2386 } else if (!strcmp(this_char
,"nr_inodes")) {
2387 sbinfo
->max_inodes
= memparse(value
, &rest
);
2390 } else if (!strcmp(this_char
,"mode")) {
2393 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2396 } else if (!strcmp(this_char
,"uid")) {
2399 uid
= simple_strtoul(value
, &rest
, 0);
2402 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2403 if (!uid_valid(sbinfo
->uid
))
2405 } else if (!strcmp(this_char
,"gid")) {
2408 gid
= simple_strtoul(value
, &rest
, 0);
2411 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2412 if (!gid_valid(sbinfo
->gid
))
2414 } else if (!strcmp(this_char
,"mpol")) {
2415 if (mpol_parse_str(value
, &sbinfo
->mpol
, 1))
2418 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2426 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2432 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2434 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2435 struct shmem_sb_info config
= *sbinfo
;
2436 unsigned long inodes
;
2437 int error
= -EINVAL
;
2439 if (shmem_parse_options(data
, &config
, true))
2442 spin_lock(&sbinfo
->stat_lock
);
2443 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2444 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2446 if (config
.max_inodes
< inodes
)
2449 * Those tests disallow limited->unlimited while any are in use;
2450 * but we must separately disallow unlimited->limited, because
2451 * in that case we have no record of how much is already in use.
2453 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2455 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2459 sbinfo
->max_blocks
= config
.max_blocks
;
2460 sbinfo
->max_inodes
= config
.max_inodes
;
2461 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2463 mpol_put(sbinfo
->mpol
);
2464 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2466 spin_unlock(&sbinfo
->stat_lock
);
2470 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2472 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2474 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2475 seq_printf(seq
, ",size=%luk",
2476 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2477 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2478 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2479 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2480 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2481 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2482 seq_printf(seq
, ",uid=%u",
2483 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2484 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2485 seq_printf(seq
, ",gid=%u",
2486 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2487 shmem_show_mpol(seq
, sbinfo
->mpol
);
2490 #endif /* CONFIG_TMPFS */
2492 static void shmem_put_super(struct super_block
*sb
)
2494 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2496 percpu_counter_destroy(&sbinfo
->used_blocks
);
2498 sb
->s_fs_info
= NULL
;
2501 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2503 struct inode
*inode
;
2504 struct shmem_sb_info
*sbinfo
;
2507 /* Round up to L1_CACHE_BYTES to resist false sharing */
2508 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2509 L1_CACHE_BYTES
), GFP_KERNEL
);
2513 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2514 sbinfo
->uid
= current_fsuid();
2515 sbinfo
->gid
= current_fsgid();
2516 sb
->s_fs_info
= sbinfo
;
2520 * Per default we only allow half of the physical ram per
2521 * tmpfs instance, limiting inodes to one per page of lowmem;
2522 * but the internal instance is left unlimited.
2524 if (!(sb
->s_flags
& MS_NOUSER
)) {
2525 sbinfo
->max_blocks
= shmem_default_max_blocks();
2526 sbinfo
->max_inodes
= shmem_default_max_inodes();
2527 if (shmem_parse_options(data
, sbinfo
, false)) {
2532 sb
->s_export_op
= &shmem_export_ops
;
2533 sb
->s_flags
|= MS_NOSEC
;
2535 sb
->s_flags
|= MS_NOUSER
;
2538 spin_lock_init(&sbinfo
->stat_lock
);
2539 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
2541 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2543 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
2544 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2545 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2546 sb
->s_magic
= TMPFS_MAGIC
;
2547 sb
->s_op
= &shmem_ops
;
2548 sb
->s_time_gran
= 1;
2549 #ifdef CONFIG_TMPFS_XATTR
2550 sb
->s_xattr
= shmem_xattr_handlers
;
2552 #ifdef CONFIG_TMPFS_POSIX_ACL
2553 sb
->s_flags
|= MS_POSIXACL
;
2556 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
2559 inode
->i_uid
= sbinfo
->uid
;
2560 inode
->i_gid
= sbinfo
->gid
;
2561 sb
->s_root
= d_make_root(inode
);
2567 shmem_put_super(sb
);
2571 static struct kmem_cache
*shmem_inode_cachep
;
2573 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2575 struct shmem_inode_info
*info
;
2576 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2579 return &info
->vfs_inode
;
2582 static void shmem_destroy_callback(struct rcu_head
*head
)
2584 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
2585 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2588 static void shmem_destroy_inode(struct inode
*inode
)
2590 if (S_ISREG(inode
->i_mode
))
2591 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2592 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
2595 static void shmem_init_inode(void *foo
)
2597 struct shmem_inode_info
*info
= foo
;
2598 inode_init_once(&info
->vfs_inode
);
2601 static int shmem_init_inodecache(void)
2603 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2604 sizeof(struct shmem_inode_info
),
2605 0, SLAB_PANIC
, shmem_init_inode
);
2609 static void shmem_destroy_inodecache(void)
2611 kmem_cache_destroy(shmem_inode_cachep
);
2614 static const struct address_space_operations shmem_aops
= {
2615 .writepage
= shmem_writepage
,
2616 .set_page_dirty
= __set_page_dirty_no_writeback
,
2618 .write_begin
= shmem_write_begin
,
2619 .write_end
= shmem_write_end
,
2621 .migratepage
= migrate_page
,
2622 .error_remove_page
= generic_error_remove_page
,
2625 static const struct file_operations shmem_file_operations
= {
2628 .llseek
= generic_file_llseek
,
2629 .read
= do_sync_read
,
2630 .write
= do_sync_write
,
2631 .aio_read
= shmem_file_aio_read
,
2632 .aio_write
= generic_file_aio_write
,
2633 .fsync
= noop_fsync
,
2634 .splice_read
= shmem_file_splice_read
,
2635 .splice_write
= generic_file_splice_write
,
2636 .fallocate
= shmem_fallocate
,
2640 static const struct inode_operations shmem_inode_operations
= {
2641 .setattr
= shmem_setattr
,
2642 #ifdef CONFIG_TMPFS_XATTR
2643 .setxattr
= shmem_setxattr
,
2644 .getxattr
= shmem_getxattr
,
2645 .listxattr
= shmem_listxattr
,
2646 .removexattr
= shmem_removexattr
,
2650 static const struct inode_operations shmem_dir_inode_operations
= {
2652 .create
= shmem_create
,
2653 .lookup
= simple_lookup
,
2655 .unlink
= shmem_unlink
,
2656 .symlink
= shmem_symlink
,
2657 .mkdir
= shmem_mkdir
,
2658 .rmdir
= shmem_rmdir
,
2659 .mknod
= shmem_mknod
,
2660 .rename
= shmem_rename
,
2662 #ifdef CONFIG_TMPFS_XATTR
2663 .setxattr
= shmem_setxattr
,
2664 .getxattr
= shmem_getxattr
,
2665 .listxattr
= shmem_listxattr
,
2666 .removexattr
= shmem_removexattr
,
2668 #ifdef CONFIG_TMPFS_POSIX_ACL
2669 .setattr
= shmem_setattr
,
2673 static const struct inode_operations shmem_special_inode_operations
= {
2674 #ifdef CONFIG_TMPFS_XATTR
2675 .setxattr
= shmem_setxattr
,
2676 .getxattr
= shmem_getxattr
,
2677 .listxattr
= shmem_listxattr
,
2678 .removexattr
= shmem_removexattr
,
2680 #ifdef CONFIG_TMPFS_POSIX_ACL
2681 .setattr
= shmem_setattr
,
2685 static const struct super_operations shmem_ops
= {
2686 .alloc_inode
= shmem_alloc_inode
,
2687 .destroy_inode
= shmem_destroy_inode
,
2689 .statfs
= shmem_statfs
,
2690 .remount_fs
= shmem_remount_fs
,
2691 .show_options
= shmem_show_options
,
2693 .evict_inode
= shmem_evict_inode
,
2694 .drop_inode
= generic_delete_inode
,
2695 .put_super
= shmem_put_super
,
2698 static const struct vm_operations_struct shmem_vm_ops
= {
2699 .fault
= shmem_fault
,
2701 .set_policy
= shmem_set_policy
,
2702 .get_policy
= shmem_get_policy
,
2706 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
2707 int flags
, const char *dev_name
, void *data
)
2709 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
2712 static struct file_system_type shmem_fs_type
= {
2713 .owner
= THIS_MODULE
,
2715 .mount
= shmem_mount
,
2716 .kill_sb
= kill_litter_super
,
2719 int __init
shmem_init(void)
2723 error
= bdi_init(&shmem_backing_dev_info
);
2727 error
= shmem_init_inodecache();
2731 error
= register_filesystem(&shmem_fs_type
);
2733 printk(KERN_ERR
"Could not register tmpfs\n");
2737 shm_mnt
= vfs_kern_mount(&shmem_fs_type
, MS_NOUSER
,
2738 shmem_fs_type
.name
, NULL
);
2739 if (IS_ERR(shm_mnt
)) {
2740 error
= PTR_ERR(shm_mnt
);
2741 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2747 unregister_filesystem(&shmem_fs_type
);
2749 shmem_destroy_inodecache();
2751 bdi_destroy(&shmem_backing_dev_info
);
2753 shm_mnt
= ERR_PTR(error
);
2757 #else /* !CONFIG_SHMEM */
2760 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2762 * This is intended for small system where the benefits of the full
2763 * shmem code (swap-backed and resource-limited) are outweighed by
2764 * their complexity. On systems without swap this code should be
2765 * effectively equivalent, but much lighter weight.
2768 #include <linux/ramfs.h>
2770 static struct file_system_type shmem_fs_type
= {
2772 .mount
= ramfs_mount
,
2773 .kill_sb
= kill_litter_super
,
2776 int __init
shmem_init(void)
2778 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
2780 shm_mnt
= kern_mount(&shmem_fs_type
);
2781 BUG_ON(IS_ERR(shm_mnt
));
2786 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
2791 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2796 void shmem_unlock_mapping(struct address_space
*mapping
)
2800 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
2802 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
2804 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
2806 #define shmem_vm_ops generic_file_vm_ops
2807 #define shmem_file_operations ramfs_file_operations
2808 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2809 #define shmem_acct_size(flags, size) 0
2810 #define shmem_unacct_size(flags, size) do {} while (0)
2812 #endif /* CONFIG_SHMEM */
2817 * shmem_file_setup - get an unlinked file living in tmpfs
2818 * @name: name for dentry (to be seen in /proc/<pid>/maps
2819 * @size: size to be set for the file
2820 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2822 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
2826 struct inode
*inode
;
2828 struct dentry
*root
;
2831 if (IS_ERR(shm_mnt
))
2832 return (void *)shm_mnt
;
2834 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
2835 return ERR_PTR(-EINVAL
);
2837 if (shmem_acct_size(flags
, size
))
2838 return ERR_PTR(-ENOMEM
);
2842 this.len
= strlen(name
);
2843 this.hash
= 0; /* will go */
2844 root
= shm_mnt
->mnt_root
;
2845 path
.dentry
= d_alloc(root
, &this);
2848 path
.mnt
= mntget(shm_mnt
);
2851 inode
= shmem_get_inode(root
->d_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
2855 d_instantiate(path
.dentry
, inode
);
2856 inode
->i_size
= size
;
2857 clear_nlink(inode
); /* It is unlinked */
2859 error
= ramfs_nommu_expand_for_mapping(inode
, size
);
2865 file
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
2866 &shmem_file_operations
);
2875 shmem_unacct_size(flags
, size
);
2876 return ERR_PTR(error
);
2878 EXPORT_SYMBOL_GPL(shmem_file_setup
);
2881 * shmem_zero_setup - setup a shared anonymous mapping
2882 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2884 int shmem_zero_setup(struct vm_area_struct
*vma
)
2887 loff_t size
= vma
->vm_end
- vma
->vm_start
;
2889 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
2891 return PTR_ERR(file
);
2895 vma
->vm_file
= file
;
2896 vma
->vm_ops
= &shmem_vm_ops
;
2897 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
2902 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2903 * @mapping: the page's address_space
2904 * @index: the page index
2905 * @gfp: the page allocator flags to use if allocating
2907 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2908 * with any new page allocations done using the specified allocation flags.
2909 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2910 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2911 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2913 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2914 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2916 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
2917 pgoff_t index
, gfp_t gfp
)
2920 struct inode
*inode
= mapping
->host
;
2924 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
2925 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
2927 page
= ERR_PTR(error
);
2933 * The tiny !SHMEM case uses ramfs without swap
2935 return read_cache_page_gfp(mapping
, index
, gfp
);
2938 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);