2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/aio.h>
36 static struct vfsmount
*shm_mnt
;
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/generic_acl.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
70 #include <asm/uaccess.h>
71 #include <asm/pgtable.h>
73 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
74 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
79 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
80 #define SHORT_SYMLINK_LEN 128
83 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
84 * inode->i_private (with i_mutex making sure that it has only one user at
85 * a time): we would prefer not to enlarge the shmem inode just for that.
88 wait_queue_head_t
*waitq
; /* faults into hole wait for punch to end */
89 pgoff_t start
; /* start of range currently being fallocated */
90 pgoff_t next
; /* the next page offset to be fallocated */
91 pgoff_t nr_falloced
; /* how many new pages have been fallocated */
92 pgoff_t nr_unswapped
; /* how often writepage refused to swap out */
95 /* Flag allocation requirements to shmem_getpage */
97 SGP_READ
, /* don't exceed i_size, don't allocate page */
98 SGP_CACHE
, /* don't exceed i_size, may allocate page */
99 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
100 SGP_WRITE
, /* may exceed i_size, may allocate !Uptodate page */
101 SGP_FALLOC
, /* like SGP_WRITE, but make existing page Uptodate */
105 static unsigned long shmem_default_max_blocks(void)
107 return totalram_pages
/ 2;
110 static unsigned long shmem_default_max_inodes(void)
112 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
116 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
);
117 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
118 struct shmem_inode_info
*info
, pgoff_t index
);
119 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
120 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
122 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
123 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
125 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
126 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
129 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
131 return sb
->s_fs_info
;
135 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
136 * for shared memory and for shared anonymous (/dev/zero) mappings
137 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
138 * consistent with the pre-accounting of private mappings ...
140 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
142 return (flags
& VM_NORESERVE
) ?
143 0 : security_vm_enough_memory_mm(current
->mm
, VM_ACCT(size
));
146 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
148 if (!(flags
& VM_NORESERVE
))
149 vm_unacct_memory(VM_ACCT(size
));
153 * ... whereas tmpfs objects are accounted incrementally as
154 * pages are allocated, in order to allow huge sparse files.
155 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
156 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
158 static inline int shmem_acct_block(unsigned long flags
)
160 return (flags
& VM_NORESERVE
) ?
161 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
164 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
166 if (flags
& VM_NORESERVE
)
167 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
170 static const struct super_operations shmem_ops
;
171 static const struct address_space_operations shmem_aops
;
172 static const struct file_operations shmem_file_operations
;
173 static const struct inode_operations shmem_inode_operations
;
174 static const struct inode_operations shmem_dir_inode_operations
;
175 static const struct inode_operations shmem_special_inode_operations
;
176 static const struct vm_operations_struct shmem_vm_ops
;
178 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
179 .ra_pages
= 0, /* No readahead */
180 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
183 static LIST_HEAD(shmem_swaplist
);
184 static DEFINE_MUTEX(shmem_swaplist_mutex
);
186 static int shmem_reserve_inode(struct super_block
*sb
)
188 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
189 if (sbinfo
->max_inodes
) {
190 spin_lock(&sbinfo
->stat_lock
);
191 if (!sbinfo
->free_inodes
) {
192 spin_unlock(&sbinfo
->stat_lock
);
195 sbinfo
->free_inodes
--;
196 spin_unlock(&sbinfo
->stat_lock
);
201 static void shmem_free_inode(struct super_block
*sb
)
203 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
204 if (sbinfo
->max_inodes
) {
205 spin_lock(&sbinfo
->stat_lock
);
206 sbinfo
->free_inodes
++;
207 spin_unlock(&sbinfo
->stat_lock
);
212 * shmem_recalc_inode - recalculate the block usage of an inode
213 * @inode: inode to recalc
215 * We have to calculate the free blocks since the mm can drop
216 * undirtied hole pages behind our back.
218 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
219 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
221 * It has to be called with the spinlock held.
223 static void shmem_recalc_inode(struct inode
*inode
)
225 struct shmem_inode_info
*info
= SHMEM_I(inode
);
228 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
230 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
231 if (sbinfo
->max_blocks
)
232 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
233 info
->alloced
-= freed
;
234 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
235 shmem_unacct_blocks(info
->flags
, freed
);
240 * Replace item expected in radix tree by a new item, while holding tree lock.
242 static int shmem_radix_tree_replace(struct address_space
*mapping
,
243 pgoff_t index
, void *expected
, void *replacement
)
248 VM_BUG_ON(!expected
);
249 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
251 item
= radix_tree_deref_slot_protected(pslot
,
252 &mapping
->tree_lock
);
253 if (item
!= expected
)
256 radix_tree_replace_slot(pslot
, replacement
);
258 radix_tree_delete(&mapping
->page_tree
, index
);
263 * Sometimes, before we decide whether to proceed or to fail, we must check
264 * that an entry was not already brought back from swap by a racing thread.
266 * Checking page is not enough: by the time a SwapCache page is locked, it
267 * might be reused, and again be SwapCache, using the same swap as before.
269 static bool shmem_confirm_swap(struct address_space
*mapping
,
270 pgoff_t index
, swp_entry_t swap
)
275 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
277 return item
== swp_to_radix_entry(swap
);
281 * Like add_to_page_cache_locked, but error if expected item has gone.
283 static int shmem_add_to_page_cache(struct page
*page
,
284 struct address_space
*mapping
,
285 pgoff_t index
, gfp_t gfp
, void *expected
)
289 VM_BUG_ON(!PageLocked(page
));
290 VM_BUG_ON(!PageSwapBacked(page
));
292 page_cache_get(page
);
293 page
->mapping
= mapping
;
296 spin_lock_irq(&mapping
->tree_lock
);
298 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
300 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
304 __inc_zone_page_state(page
, NR_FILE_PAGES
);
305 __inc_zone_page_state(page
, NR_SHMEM
);
306 spin_unlock_irq(&mapping
->tree_lock
);
308 page
->mapping
= NULL
;
309 spin_unlock_irq(&mapping
->tree_lock
);
310 page_cache_release(page
);
316 * Like delete_from_page_cache, but substitutes swap for page.
318 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
320 struct address_space
*mapping
= page
->mapping
;
323 spin_lock_irq(&mapping
->tree_lock
);
324 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
325 page
->mapping
= NULL
;
327 __dec_zone_page_state(page
, NR_FILE_PAGES
);
328 __dec_zone_page_state(page
, NR_SHMEM
);
329 spin_unlock_irq(&mapping
->tree_lock
);
330 page_cache_release(page
);
335 * Like find_get_pages, but collecting swap entries as well as pages.
337 static unsigned shmem_find_get_pages_and_swap(struct address_space
*mapping
,
338 pgoff_t start
, unsigned int nr_pages
,
339 struct page
**pages
, pgoff_t
*indices
)
342 unsigned int ret
= 0;
343 struct radix_tree_iter iter
;
350 radix_tree_for_each_slot(slot
, &mapping
->page_tree
, &iter
, start
) {
353 page
= radix_tree_deref_slot(slot
);
356 if (radix_tree_exception(page
)) {
357 if (radix_tree_deref_retry(page
))
360 * Otherwise, we must be storing a swap entry
361 * here as an exceptional entry: so return it
362 * without attempting to raise page count.
366 if (!page_cache_get_speculative(page
))
369 /* Has the page moved? */
370 if (unlikely(page
!= *slot
)) {
371 page_cache_release(page
);
375 indices
[ret
] = iter
.index
;
377 if (++ret
== nr_pages
)
385 * Remove swap entry from radix tree, free the swap and its page cache.
387 static int shmem_free_swap(struct address_space
*mapping
,
388 pgoff_t index
, void *radswap
)
392 spin_lock_irq(&mapping
->tree_lock
);
393 error
= shmem_radix_tree_replace(mapping
, index
, radswap
, NULL
);
394 spin_unlock_irq(&mapping
->tree_lock
);
396 free_swap_and_cache(radix_to_swp_entry(radswap
));
401 * Pagevec may contain swap entries, so shuffle up pages before releasing.
403 static void shmem_deswap_pagevec(struct pagevec
*pvec
)
407 for (i
= 0, j
= 0; i
< pagevec_count(pvec
); i
++) {
408 struct page
*page
= pvec
->pages
[i
];
409 if (!radix_tree_exceptional_entry(page
))
410 pvec
->pages
[j
++] = page
;
416 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
418 void shmem_unlock_mapping(struct address_space
*mapping
)
421 pgoff_t indices
[PAGEVEC_SIZE
];
424 pagevec_init(&pvec
, 0);
426 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
428 while (!mapping_unevictable(mapping
)) {
430 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
431 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
433 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
434 PAGEVEC_SIZE
, pvec
.pages
, indices
);
437 index
= indices
[pvec
.nr
- 1] + 1;
438 shmem_deswap_pagevec(&pvec
);
439 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
440 pagevec_release(&pvec
);
446 * Remove range of pages and swap entries from radix tree, and free them.
447 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
449 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
452 struct address_space
*mapping
= inode
->i_mapping
;
453 struct shmem_inode_info
*info
= SHMEM_I(inode
);
454 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
455 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
456 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
457 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
459 pgoff_t indices
[PAGEVEC_SIZE
];
460 long nr_swaps_freed
= 0;
465 end
= -1; /* unsigned, so actually very big */
467 pagevec_init(&pvec
, 0);
469 while (index
< end
) {
470 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
471 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
472 pvec
.pages
, indices
);
475 mem_cgroup_uncharge_start();
476 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
477 struct page
*page
= pvec
.pages
[i
];
483 if (radix_tree_exceptional_entry(page
)) {
486 nr_swaps_freed
+= !shmem_free_swap(mapping
,
491 if (!trylock_page(page
))
493 if (!unfalloc
|| !PageUptodate(page
)) {
494 if (page
->mapping
== mapping
) {
495 VM_BUG_ON(PageWriteback(page
));
496 truncate_inode_page(mapping
, page
);
501 shmem_deswap_pagevec(&pvec
);
502 pagevec_release(&pvec
);
503 mem_cgroup_uncharge_end();
509 struct page
*page
= NULL
;
510 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
512 unsigned int top
= PAGE_CACHE_SIZE
;
517 zero_user_segment(page
, partial_start
, top
);
518 set_page_dirty(page
);
520 page_cache_release(page
);
524 struct page
*page
= NULL
;
525 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
527 zero_user_segment(page
, 0, partial_end
);
528 set_page_dirty(page
);
530 page_cache_release(page
);
537 while (index
< end
) {
539 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
540 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
541 pvec
.pages
, indices
);
543 /* If all gone or hole-punch or unfalloc, we're done */
544 if (index
== start
|| end
!= -1)
546 /* But if truncating, restart to make sure all gone */
550 mem_cgroup_uncharge_start();
551 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
552 struct page
*page
= pvec
.pages
[i
];
558 if (radix_tree_exceptional_entry(page
)) {
561 if (shmem_free_swap(mapping
, index
, page
)) {
562 /* Swap was replaced by page: retry */
571 if (!unfalloc
|| !PageUptodate(page
)) {
572 if (page
->mapping
== mapping
) {
573 VM_BUG_ON(PageWriteback(page
));
574 truncate_inode_page(mapping
, page
);
576 /* Page was replaced by swap: retry */
584 shmem_deswap_pagevec(&pvec
);
585 pagevec_release(&pvec
);
586 mem_cgroup_uncharge_end();
590 spin_lock(&info
->lock
);
591 info
->swapped
-= nr_swaps_freed
;
592 shmem_recalc_inode(inode
);
593 spin_unlock(&info
->lock
);
596 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
598 shmem_undo_range(inode
, lstart
, lend
, false);
599 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
601 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
603 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
605 struct inode
*inode
= dentry
->d_inode
;
608 error
= inode_change_ok(inode
, attr
);
612 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
613 loff_t oldsize
= inode
->i_size
;
614 loff_t newsize
= attr
->ia_size
;
616 if (newsize
!= oldsize
) {
617 i_size_write(inode
, newsize
);
618 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
620 if (newsize
< oldsize
) {
621 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
622 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
623 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
624 /* unmap again to remove racily COWed private pages */
625 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
629 setattr_copy(inode
, attr
);
630 #ifdef CONFIG_TMPFS_POSIX_ACL
631 if (attr
->ia_valid
& ATTR_MODE
)
632 error
= generic_acl_chmod(inode
);
637 static void shmem_evict_inode(struct inode
*inode
)
639 struct shmem_inode_info
*info
= SHMEM_I(inode
);
641 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
642 shmem_unacct_size(info
->flags
, inode
->i_size
);
644 shmem_truncate_range(inode
, 0, (loff_t
)-1);
645 if (!list_empty(&info
->swaplist
)) {
646 mutex_lock(&shmem_swaplist_mutex
);
647 list_del_init(&info
->swaplist
);
648 mutex_unlock(&shmem_swaplist_mutex
);
651 kfree(info
->symlink
);
653 simple_xattrs_free(&info
->xattrs
);
654 WARN_ON(inode
->i_blocks
);
655 shmem_free_inode(inode
->i_sb
);
660 * If swap found in inode, free it and move page from swapcache to filecache.
662 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
663 swp_entry_t swap
, struct page
**pagep
)
665 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
671 radswap
= swp_to_radix_entry(swap
);
672 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
677 * Move _head_ to start search for next from here.
678 * But be careful: shmem_evict_inode checks list_empty without taking
679 * mutex, and there's an instant in list_move_tail when info->swaplist
680 * would appear empty, if it were the only one on shmem_swaplist.
682 if (shmem_swaplist
.next
!= &info
->swaplist
)
683 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
685 gfp
= mapping_gfp_mask(mapping
);
686 if (shmem_should_replace_page(*pagep
, gfp
)) {
687 mutex_unlock(&shmem_swaplist_mutex
);
688 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
689 mutex_lock(&shmem_swaplist_mutex
);
691 * We needed to drop mutex to make that restrictive page
692 * allocation, but the inode might have been freed while we
693 * dropped it: although a racing shmem_evict_inode() cannot
694 * complete without emptying the radix_tree, our page lock
695 * on this swapcache page is not enough to prevent that -
696 * free_swap_and_cache() of our swap entry will only
697 * trylock_page(), removing swap from radix_tree whatever.
699 * We must not proceed to shmem_add_to_page_cache() if the
700 * inode has been freed, but of course we cannot rely on
701 * inode or mapping or info to check that. However, we can
702 * safely check if our swap entry is still in use (and here
703 * it can't have got reused for another page): if it's still
704 * in use, then the inode cannot have been freed yet, and we
705 * can safely proceed (if it's no longer in use, that tells
706 * nothing about the inode, but we don't need to unuse swap).
708 if (!page_swapcount(*pagep
))
713 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
714 * but also to hold up shmem_evict_inode(): so inode cannot be freed
715 * beneath us (pagelock doesn't help until the page is in pagecache).
718 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
719 GFP_NOWAIT
, radswap
);
720 if (error
!= -ENOMEM
) {
722 * Truncation and eviction use free_swap_and_cache(), which
723 * only does trylock page: if we raced, best clean up here.
725 delete_from_swap_cache(*pagep
);
726 set_page_dirty(*pagep
);
728 spin_lock(&info
->lock
);
730 spin_unlock(&info
->lock
);
733 error
= 1; /* not an error, but entry was found */
739 * Search through swapped inodes to find and replace swap by page.
741 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
743 struct list_head
*this, *next
;
744 struct shmem_inode_info
*info
;
749 * There's a faint possibility that swap page was replaced before
750 * caller locked it: caller will come back later with the right page.
752 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
756 * Charge page using GFP_KERNEL while we can wait, before taking
757 * the shmem_swaplist_mutex which might hold up shmem_writepage().
758 * Charged back to the user (not to caller) when swap account is used.
760 error
= mem_cgroup_cache_charge(page
, current
->mm
, GFP_KERNEL
);
763 /* No radix_tree_preload: swap entry keeps a place for page in tree */
765 mutex_lock(&shmem_swaplist_mutex
);
766 list_for_each_safe(this, next
, &shmem_swaplist
) {
767 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
769 found
= shmem_unuse_inode(info
, swap
, &page
);
771 list_del_init(&info
->swaplist
);
776 mutex_unlock(&shmem_swaplist_mutex
);
782 page_cache_release(page
);
787 * Move the page from the page cache to the swap cache.
789 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
791 struct shmem_inode_info
*info
;
792 struct address_space
*mapping
;
797 BUG_ON(!PageLocked(page
));
798 mapping
= page
->mapping
;
800 inode
= mapping
->host
;
801 info
= SHMEM_I(inode
);
802 if (info
->flags
& VM_LOCKED
)
804 if (!total_swap_pages
)
808 * shmem_backing_dev_info's capabilities prevent regular writeback or
809 * sync from ever calling shmem_writepage; but a stacking filesystem
810 * might use ->writepage of its underlying filesystem, in which case
811 * tmpfs should write out to swap only in response to memory pressure,
812 * and not for the writeback threads or sync.
814 if (!wbc
->for_reclaim
) {
815 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
820 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
821 * value into swapfile.c, the only way we can correctly account for a
822 * fallocated page arriving here is now to initialize it and write it.
824 * That's okay for a page already fallocated earlier, but if we have
825 * not yet completed the fallocation, then (a) we want to keep track
826 * of this page in case we have to undo it, and (b) it may not be a
827 * good idea to continue anyway, once we're pushing into swap. So
828 * reactivate the page, and let shmem_fallocate() quit when too many.
830 if (!PageUptodate(page
)) {
831 if (inode
->i_private
) {
832 struct shmem_falloc
*shmem_falloc
;
833 spin_lock(&inode
->i_lock
);
834 shmem_falloc
= inode
->i_private
;
836 !shmem_falloc
->waitq
&&
837 index
>= shmem_falloc
->start
&&
838 index
< shmem_falloc
->next
)
839 shmem_falloc
->nr_unswapped
++;
842 spin_unlock(&inode
->i_lock
);
846 clear_highpage(page
);
847 flush_dcache_page(page
);
848 SetPageUptodate(page
);
852 swap
= get_swap_page();
854 swap
= get_swap_page_by_memcg(page
);
860 * Add inode to shmem_unuse()'s list of swapped-out inodes,
861 * if it's not already there. Do it now before the page is
862 * moved to swap cache, when its pagelock no longer protects
863 * the inode from eviction. But don't unlock the mutex until
864 * we've incremented swapped, because shmem_unuse_inode() will
865 * prune a !swapped inode from the swaplist under this mutex.
867 mutex_lock(&shmem_swaplist_mutex
);
868 if (list_empty(&info
->swaplist
))
869 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
871 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
872 swap_shmem_alloc(swap
);
873 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
875 spin_lock(&info
->lock
);
877 shmem_recalc_inode(inode
);
878 spin_unlock(&info
->lock
);
880 mutex_unlock(&shmem_swaplist_mutex
);
881 BUG_ON(page_mapped(page
));
882 swap_writepage(page
, wbc
);
886 mutex_unlock(&shmem_swaplist_mutex
);
887 swapcache_free(swap
, NULL
);
889 set_page_dirty(page
);
890 if (wbc
->for_reclaim
)
891 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
898 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
902 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
903 return; /* show nothing */
905 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
907 seq_printf(seq
, ",mpol=%s", buffer
);
910 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
912 struct mempolicy
*mpol
= NULL
;
914 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
917 spin_unlock(&sbinfo
->stat_lock
);
921 #endif /* CONFIG_TMPFS */
923 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
924 struct shmem_inode_info
*info
, pgoff_t index
)
926 struct vm_area_struct pvma
;
929 /* Create a pseudo vma that just contains the policy */
931 /* Bias interleave by inode number to distribute better across nodes */
932 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
934 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
936 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
938 /* Drop reference taken by mpol_shared_policy_lookup() */
939 mpol_cond_put(pvma
.vm_policy
);
944 static struct page
*shmem_alloc_page(gfp_t gfp
,
945 struct shmem_inode_info
*info
, pgoff_t index
)
947 struct vm_area_struct pvma
;
950 /* Create a pseudo vma that just contains the policy */
952 /* Bias interleave by inode number to distribute better across nodes */
953 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
955 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
957 page
= alloc_page_vma(gfp
, &pvma
, 0);
959 /* Drop reference taken by mpol_shared_policy_lookup() */
960 mpol_cond_put(pvma
.vm_policy
);
964 #else /* !CONFIG_NUMA */
966 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
969 #endif /* CONFIG_TMPFS */
971 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
972 struct shmem_inode_info
*info
, pgoff_t index
)
974 return swapin_readahead(swap
, gfp
, NULL
, 0);
977 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
978 struct shmem_inode_info
*info
, pgoff_t index
)
980 return alloc_page(gfp
);
982 #endif /* CONFIG_NUMA */
984 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
985 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
992 * When a page is moved from swapcache to shmem filecache (either by the
993 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
994 * shmem_unuse_inode()), it may have been read in earlier from swap, in
995 * ignorance of the mapping it belongs to. If that mapping has special
996 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
997 * we may need to copy to a suitable page before moving to filecache.
999 * In a future release, this may well be extended to respect cpuset and
1000 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1001 * but for now it is a simple matter of zone.
1003 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
1005 return page_zonenum(page
) > gfp_zone(gfp
);
1008 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
1009 struct shmem_inode_info
*info
, pgoff_t index
)
1011 struct page
*oldpage
, *newpage
;
1012 struct address_space
*swap_mapping
;
1017 swap_index
= page_private(oldpage
);
1018 swap_mapping
= page_mapping(oldpage
);
1021 * We have arrived here because our zones are constrained, so don't
1022 * limit chance of success by further cpuset and node constraints.
1024 gfp
&= ~GFP_CONSTRAINT_MASK
;
1025 newpage
= shmem_alloc_page(gfp
, info
, index
);
1029 page_cache_get(newpage
);
1030 copy_highpage(newpage
, oldpage
);
1031 flush_dcache_page(newpage
);
1033 __set_page_locked(newpage
);
1034 SetPageUptodate(newpage
);
1035 SetPageSwapBacked(newpage
);
1036 set_page_private(newpage
, swap_index
);
1037 SetPageSwapCache(newpage
);
1040 * Our caller will very soon move newpage out of swapcache, but it's
1041 * a nice clean interface for us to replace oldpage by newpage there.
1043 spin_lock_irq(&swap_mapping
->tree_lock
);
1044 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
1047 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
1048 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
1050 spin_unlock_irq(&swap_mapping
->tree_lock
);
1052 if (unlikely(error
)) {
1054 * Is this possible? I think not, now that our callers check
1055 * both PageSwapCache and page_private after getting page lock;
1056 * but be defensive. Reverse old to newpage for clear and free.
1060 mem_cgroup_replace_page_cache(oldpage
, newpage
);
1061 lru_cache_add_anon(newpage
);
1065 ClearPageSwapCache(oldpage
);
1066 set_page_private(oldpage
, 0);
1068 unlock_page(oldpage
);
1069 page_cache_release(oldpage
);
1070 page_cache_release(oldpage
);
1075 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1077 * If we allocate a new one we do not mark it dirty. That's up to the
1078 * vm. If we swap it in we mark it dirty since we also free the swap
1079 * entry since a page cannot live in both the swap and page cache
1081 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1082 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1084 struct address_space
*mapping
= inode
->i_mapping
;
1085 struct shmem_inode_info
*info
;
1086 struct shmem_sb_info
*sbinfo
;
1093 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1097 page
= find_lock_page(mapping
, index
);
1098 if (radix_tree_exceptional_entry(page
)) {
1099 swap
= radix_to_swp_entry(page
);
1103 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1104 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1109 /* fallocated page? */
1110 if (page
&& !PageUptodate(page
)) {
1111 if (sgp
!= SGP_READ
)
1114 page_cache_release(page
);
1117 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1123 * Fast cache lookup did not find it:
1124 * bring it back from swap or allocate.
1126 info
= SHMEM_I(inode
);
1127 sbinfo
= SHMEM_SB(inode
->i_sb
);
1130 /* Look it up and read it in.. */
1131 page
= lookup_swap_cache(swap
);
1133 /* here we actually do the io */
1135 *fault_type
|= VM_FAULT_MAJOR
;
1136 page
= shmem_swapin(swap
, gfp
, info
, index
);
1143 /* We have to do this with page locked to prevent races */
1145 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1146 !shmem_confirm_swap(mapping
, index
, swap
)) {
1147 error
= -EEXIST
; /* try again */
1150 if (!PageUptodate(page
)) {
1154 wait_on_page_writeback(page
);
1156 if (shmem_should_replace_page(page
, gfp
)) {
1157 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1162 error
= mem_cgroup_cache_charge(page
, current
->mm
,
1163 gfp
& GFP_RECLAIM_MASK
);
1165 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1166 gfp
, swp_to_radix_entry(swap
));
1168 * We already confirmed swap under page lock, and make
1169 * no memory allocation here, so usually no possibility
1170 * of error; but free_swap_and_cache() only trylocks a
1171 * page, so it is just possible that the entry has been
1172 * truncated or holepunched since swap was confirmed.
1173 * shmem_undo_range() will have done some of the
1174 * unaccounting, now delete_from_swap_cache() will do
1175 * the rest (including mem_cgroup_uncharge_swapcache).
1176 * Reset swap.val? No, leave it so "failed" goes back to
1177 * "repeat": reading a hole and writing should succeed.
1180 delete_from_swap_cache(page
);
1185 spin_lock(&info
->lock
);
1187 shmem_recalc_inode(inode
);
1188 spin_unlock(&info
->lock
);
1190 delete_from_swap_cache(page
);
1191 set_page_dirty(page
);
1195 if (shmem_acct_block(info
->flags
)) {
1199 if (sbinfo
->max_blocks
) {
1200 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1201 sbinfo
->max_blocks
) >= 0) {
1205 percpu_counter_inc(&sbinfo
->used_blocks
);
1208 page
= shmem_alloc_page(gfp
, info
, index
);
1214 SetPageSwapBacked(page
);
1215 __set_page_locked(page
);
1216 error
= mem_cgroup_cache_charge(page
, current
->mm
,
1217 gfp
& GFP_RECLAIM_MASK
);
1220 error
= radix_tree_preload(gfp
& GFP_RECLAIM_MASK
);
1222 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1224 radix_tree_preload_end();
1227 mem_cgroup_uncharge_cache_page(page
);
1230 lru_cache_add_anon(page
);
1232 spin_lock(&info
->lock
);
1234 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1235 shmem_recalc_inode(inode
);
1236 spin_unlock(&info
->lock
);
1240 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1242 if (sgp
== SGP_FALLOC
)
1246 * Let SGP_WRITE caller clear ends if write does not fill page;
1247 * but SGP_FALLOC on a page fallocated earlier must initialize
1248 * it now, lest undo on failure cancel our earlier guarantee.
1250 if (sgp
!= SGP_WRITE
) {
1251 clear_highpage(page
);
1252 flush_dcache_page(page
);
1253 SetPageUptodate(page
);
1255 if (sgp
== SGP_DIRTY
)
1256 set_page_dirty(page
);
1259 /* Perhaps the file has been truncated since we checked */
1260 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1261 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1275 info
= SHMEM_I(inode
);
1276 ClearPageDirty(page
);
1277 delete_from_page_cache(page
);
1278 spin_lock(&info
->lock
);
1280 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1281 spin_unlock(&info
->lock
);
1283 sbinfo
= SHMEM_SB(inode
->i_sb
);
1284 if (sbinfo
->max_blocks
)
1285 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1287 shmem_unacct_blocks(info
->flags
, 1);
1289 if (swap
.val
&& error
!= -EINVAL
&&
1290 !shmem_confirm_swap(mapping
, index
, swap
))
1295 page_cache_release(page
);
1297 if (error
== -ENOSPC
&& !once
++) {
1298 info
= SHMEM_I(inode
);
1299 spin_lock(&info
->lock
);
1300 shmem_recalc_inode(inode
);
1301 spin_unlock(&info
->lock
);
1304 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1309 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1311 struct inode
*inode
= file_inode(vma
->vm_file
);
1313 int ret
= VM_FAULT_LOCKED
;
1316 * Trinity finds that probing a hole which tmpfs is punching can
1317 * prevent the hole-punch from ever completing: which in turn
1318 * locks writers out with its hold on i_mutex. So refrain from
1319 * faulting pages into the hole while it's being punched. Although
1320 * shmem_undo_range() does remove the additions, it may be unable to
1321 * keep up, as each new page needs its own unmap_mapping_range() call,
1322 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1324 * It does not matter if we sometimes reach this check just before the
1325 * hole-punch begins, so that one fault then races with the punch:
1326 * we just need to make racing faults a rare case.
1328 * The implementation below would be much simpler if we just used a
1329 * standard mutex or completion: but we cannot take i_mutex in fault,
1330 * and bloating every shmem inode for this unlikely case would be sad.
1332 if (unlikely(inode
->i_private
)) {
1333 struct shmem_falloc
*shmem_falloc
;
1335 spin_lock(&inode
->i_lock
);
1336 shmem_falloc
= inode
->i_private
;
1338 shmem_falloc
->waitq
&&
1339 vmf
->pgoff
>= shmem_falloc
->start
&&
1340 vmf
->pgoff
< shmem_falloc
->next
) {
1341 wait_queue_head_t
*shmem_falloc_waitq
;
1342 DEFINE_WAIT(shmem_fault_wait
);
1344 ret
= VM_FAULT_NOPAGE
;
1345 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1346 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1347 /* It's polite to up mmap_sem if we can */
1348 up_read(&vma
->vm_mm
->mmap_sem
);
1349 ret
= VM_FAULT_RETRY
;
1352 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1353 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1354 TASK_UNINTERRUPTIBLE
);
1355 spin_unlock(&inode
->i_lock
);
1359 * shmem_falloc_waitq points into the shmem_fallocate()
1360 * stack of the hole-punching task: shmem_falloc_waitq
1361 * is usually invalid by the time we reach here, but
1362 * finish_wait() does not dereference it in that case;
1363 * though i_lock needed lest racing with wake_up_all().
1365 spin_lock(&inode
->i_lock
);
1366 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1367 spin_unlock(&inode
->i_lock
);
1370 spin_unlock(&inode
->i_lock
);
1373 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1375 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1377 if (ret
& VM_FAULT_MAJOR
) {
1378 count_vm_event(PGMAJFAULT
);
1379 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1385 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1387 struct inode
*inode
= file_inode(vma
->vm_file
);
1388 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1391 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1394 struct inode
*inode
= file_inode(vma
->vm_file
);
1397 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1398 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1402 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1404 struct inode
*inode
= file_inode(file
);
1405 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1406 int retval
= -ENOMEM
;
1408 spin_lock(&info
->lock
);
1409 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1410 if (!user_shm_lock(inode
->i_size
, user
))
1412 info
->flags
|= VM_LOCKED
;
1413 mapping_set_unevictable(file
->f_mapping
);
1415 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1416 user_shm_unlock(inode
->i_size
, user
);
1417 info
->flags
&= ~VM_LOCKED
;
1418 mapping_clear_unevictable(file
->f_mapping
);
1423 spin_unlock(&info
->lock
);
1427 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1429 file_accessed(file
);
1430 vma
->vm_ops
= &shmem_vm_ops
;
1434 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1435 umode_t mode
, dev_t dev
, unsigned long flags
, int atomic_copy
)
1437 struct inode
*inode
;
1438 struct shmem_inode_info
*info
;
1439 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1441 if (shmem_reserve_inode(sb
))
1444 inode
= new_inode(sb
);
1446 /* We don't let shmem use __GFP_SLOWHIGHMEM */
1447 mapping_set_gfp_mask(inode
->i_mapping
, GFP_HIGHUSER_MOVABLE
);
1448 inode
->i_ino
= get_next_ino();
1449 inode_init_owner(inode
, dir
, mode
);
1450 inode
->i_blocks
= 0;
1451 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1452 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1453 inode
->i_generation
= get_seconds();
1454 info
= SHMEM_I(inode
);
1455 memset(info
, 0, (char *)inode
- (char *)info
);
1456 spin_lock_init(&info
->lock
);
1457 info
->flags
= flags
& VM_NORESERVE
;
1459 inode
->i_flags
|= S_ATOMIC_COPY
;
1460 INIT_LIST_HEAD(&info
->swaplist
);
1461 simple_xattrs_init(&info
->xattrs
);
1462 cache_no_acl(inode
);
1464 switch (mode
& S_IFMT
) {
1466 inode
->i_op
= &shmem_special_inode_operations
;
1467 init_special_inode(inode
, mode
, dev
);
1470 inode
->i_mapping
->a_ops
= &shmem_aops
;
1471 inode
->i_op
= &shmem_inode_operations
;
1472 inode
->i_fop
= &shmem_file_operations
;
1473 mpol_shared_policy_init(&info
->policy
,
1474 shmem_get_sbmpol(sbinfo
));
1478 /* Some things misbehave if size == 0 on a directory */
1479 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1480 inode
->i_op
= &shmem_dir_inode_operations
;
1481 inode
->i_fop
= &simple_dir_operations
;
1485 * Must not load anything in the rbtree,
1486 * mpol_free_shared_policy will not be called.
1488 mpol_shared_policy_init(&info
->policy
, NULL
);
1492 shmem_free_inode(sb
);
1497 static const struct inode_operations shmem_symlink_inode_operations
;
1498 static const struct inode_operations shmem_short_symlink_operations
;
1500 #ifdef CONFIG_TMPFS_XATTR
1501 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1503 #define shmem_initxattrs NULL
1507 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1508 loff_t pos
, unsigned len
, unsigned flags
,
1509 struct page
**pagep
, void **fsdata
)
1511 struct inode
*inode
= mapping
->host
;
1512 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1513 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1517 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1518 loff_t pos
, unsigned len
, unsigned copied
,
1519 struct page
*page
, void *fsdata
)
1521 struct inode
*inode
= mapping
->host
;
1523 if (pos
+ copied
> inode
->i_size
)
1524 i_size_write(inode
, pos
+ copied
);
1526 if (!PageUptodate(page
)) {
1527 if (copied
< PAGE_CACHE_SIZE
) {
1528 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1529 zero_user_segments(page
, 0, from
,
1530 from
+ copied
, PAGE_CACHE_SIZE
);
1532 SetPageUptodate(page
);
1534 set_page_dirty(page
);
1536 page_cache_release(page
);
1541 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1543 struct inode
*inode
= file_inode(filp
);
1544 struct address_space
*mapping
= inode
->i_mapping
;
1546 unsigned long offset
;
1547 enum sgp_type sgp
= SGP_READ
;
1550 * Might this read be for a stacking filesystem? Then when reading
1551 * holes of a sparse file, we actually need to allocate those pages,
1552 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1554 if (segment_eq(get_fs(), KERNEL_DS
))
1557 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1558 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1561 struct page
*page
= NULL
;
1563 unsigned long nr
, ret
;
1564 loff_t i_size
= i_size_read(inode
);
1566 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1567 if (index
> end_index
)
1569 if (index
== end_index
) {
1570 nr
= i_size
& ~PAGE_CACHE_MASK
;
1575 desc
->error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1577 if (desc
->error
== -EINVAL
)
1585 * We must evaluate after, since reads (unlike writes)
1586 * are called without i_mutex protection against truncate
1588 nr
= PAGE_CACHE_SIZE
;
1589 i_size
= i_size_read(inode
);
1590 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1591 if (index
== end_index
) {
1592 nr
= i_size
& ~PAGE_CACHE_MASK
;
1595 page_cache_release(page
);
1603 * If users can be writing to this page using arbitrary
1604 * virtual addresses, take care about potential aliasing
1605 * before reading the page on the kernel side.
1607 if (mapping_writably_mapped(mapping
))
1608 flush_dcache_page(page
);
1610 * Mark the page accessed if we read the beginning.
1613 mark_page_accessed(page
);
1615 page
= ZERO_PAGE(0);
1616 page_cache_get(page
);
1620 * Ok, we have the page, and it's up-to-date, so
1621 * now we can copy it to user space...
1623 * The actor routine returns how many bytes were actually used..
1624 * NOTE! This may not be the same as how much of a user buffer
1625 * we filled up (we may be padding etc), so we can only update
1626 * "pos" here (the actor routine has to update the user buffer
1627 * pointers and the remaining count).
1629 ret
= actor(desc
, page
, offset
, nr
);
1631 index
+= offset
>> PAGE_CACHE_SHIFT
;
1632 offset
&= ~PAGE_CACHE_MASK
;
1634 page_cache_release(page
);
1635 if (ret
!= nr
|| !desc
->count
)
1641 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1642 file_accessed(filp
);
1645 static ssize_t
shmem_file_aio_read(struct kiocb
*iocb
,
1646 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
1648 struct file
*filp
= iocb
->ki_filp
;
1652 loff_t
*ppos
= &iocb
->ki_pos
;
1654 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1658 for (seg
= 0; seg
< nr_segs
; seg
++) {
1659 read_descriptor_t desc
;
1662 desc
.arg
.buf
= iov
[seg
].iov_base
;
1663 desc
.count
= iov
[seg
].iov_len
;
1664 if (desc
.count
== 0)
1667 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1668 retval
+= desc
.written
;
1670 retval
= retval
?: desc
.error
;
1679 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1680 struct pipe_inode_info
*pipe
, size_t len
,
1683 struct address_space
*mapping
= in
->f_mapping
;
1684 struct inode
*inode
= mapping
->host
;
1685 unsigned int loff
, nr_pages
, req_pages
;
1686 struct page
*pages
[PIPE_DEF_BUFFERS
];
1687 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1689 pgoff_t index
, end_index
;
1692 struct splice_pipe_desc spd
= {
1695 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1697 .ops
= &page_cache_pipe_buf_ops
,
1698 .spd_release
= spd_release_page
,
1701 isize
= i_size_read(inode
);
1702 if (unlikely(*ppos
>= isize
))
1705 left
= isize
- *ppos
;
1706 if (unlikely(left
< len
))
1709 if (splice_grow_spd(pipe
, &spd
))
1712 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1713 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1714 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1715 nr_pages
= min(req_pages
, pipe
->buffers
);
1717 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1718 nr_pages
, spd
.pages
);
1719 index
+= spd
.nr_pages
;
1722 while (spd
.nr_pages
< nr_pages
) {
1723 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1727 spd
.pages
[spd
.nr_pages
++] = page
;
1731 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1732 nr_pages
= spd
.nr_pages
;
1735 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1736 unsigned int this_len
;
1741 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1742 page
= spd
.pages
[page_nr
];
1744 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1745 error
= shmem_getpage(inode
, index
, &page
,
1750 page_cache_release(spd
.pages
[page_nr
]);
1751 spd
.pages
[page_nr
] = page
;
1754 isize
= i_size_read(inode
);
1755 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1756 if (unlikely(!isize
|| index
> end_index
))
1759 if (end_index
== index
) {
1762 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1766 this_len
= min(this_len
, plen
- loff
);
1770 spd
.partial
[page_nr
].offset
= loff
;
1771 spd
.partial
[page_nr
].len
= this_len
;
1778 while (page_nr
< nr_pages
)
1779 page_cache_release(spd
.pages
[page_nr
++]);
1782 error
= splice_to_pipe(pipe
, &spd
);
1784 splice_shrink_spd(&spd
);
1794 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1796 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1797 pgoff_t index
, pgoff_t end
, int whence
)
1800 struct pagevec pvec
;
1801 pgoff_t indices
[PAGEVEC_SIZE
];
1805 pagevec_init(&pvec
, 0);
1806 pvec
.nr
= 1; /* start small: we may be there already */
1808 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
1809 pvec
.nr
, pvec
.pages
, indices
);
1811 if (whence
== SEEK_DATA
)
1815 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1816 if (index
< indices
[i
]) {
1817 if (whence
== SEEK_HOLE
) {
1823 page
= pvec
.pages
[i
];
1824 if (page
&& !radix_tree_exceptional_entry(page
)) {
1825 if (!PageUptodate(page
))
1829 (page
&& whence
== SEEK_DATA
) ||
1830 (!page
&& whence
== SEEK_HOLE
)) {
1835 shmem_deswap_pagevec(&pvec
);
1836 pagevec_release(&pvec
);
1837 pvec
.nr
= PAGEVEC_SIZE
;
1843 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1845 struct address_space
*mapping
= file
->f_mapping
;
1846 struct inode
*inode
= mapping
->host
;
1850 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1851 return generic_file_llseek_size(file
, offset
, whence
,
1852 MAX_LFS_FILESIZE
, i_size_read(inode
));
1853 mutex_lock(&inode
->i_mutex
);
1854 /* We're holding i_mutex so we can access i_size directly */
1858 else if (offset
>= inode
->i_size
)
1861 start
= offset
>> PAGE_CACHE_SHIFT
;
1862 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1863 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1864 new_offset
<<= PAGE_CACHE_SHIFT
;
1865 if (new_offset
> offset
) {
1866 if (new_offset
< inode
->i_size
)
1867 offset
= new_offset
;
1868 else if (whence
== SEEK_DATA
)
1871 offset
= inode
->i_size
;
1875 if (offset
>= 0 && offset
!= file
->f_pos
) {
1876 file
->f_pos
= offset
;
1877 file
->f_version
= 0;
1879 mutex_unlock(&inode
->i_mutex
);
1883 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
1886 struct inode
*inode
= file_inode(file
);
1887 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1888 struct shmem_falloc shmem_falloc
;
1889 pgoff_t start
, index
, end
;
1892 //To avoid nested lock
1893 if (!mutex_trylock(&inode
->i_mutex
))
1896 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1897 struct address_space
*mapping
= file
->f_mapping
;
1898 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
1899 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
1900 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
1902 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
1903 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
1904 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
1905 spin_lock(&inode
->i_lock
);
1906 inode
->i_private
= &shmem_falloc
;
1907 spin_unlock(&inode
->i_lock
);
1909 if ((u64
)unmap_end
> (u64
)unmap_start
)
1910 unmap_mapping_range(mapping
, unmap_start
,
1911 1 + unmap_end
- unmap_start
, 0);
1912 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
1913 /* No need to unmap again: hole-punching leaves COWed pages */
1915 spin_lock(&inode
->i_lock
);
1916 inode
->i_private
= NULL
;
1917 wake_up_all(&shmem_falloc_waitq
);
1918 spin_unlock(&inode
->i_lock
);
1923 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1924 error
= inode_newsize_ok(inode
, offset
+ len
);
1928 start
= offset
>> PAGE_CACHE_SHIFT
;
1929 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1930 /* Try to avoid a swapstorm if len is impossible to satisfy */
1931 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
1936 shmem_falloc
.waitq
= NULL
;
1937 shmem_falloc
.start
= start
;
1938 shmem_falloc
.next
= start
;
1939 shmem_falloc
.nr_falloced
= 0;
1940 shmem_falloc
.nr_unswapped
= 0;
1941 spin_lock(&inode
->i_lock
);
1942 inode
->i_private
= &shmem_falloc
;
1943 spin_unlock(&inode
->i_lock
);
1945 for (index
= start
; index
< end
; index
++) {
1949 * Good, the fallocate(2) manpage permits EINTR: we may have
1950 * been interrupted because we are using up too much memory.
1952 if (signal_pending(current
))
1954 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
1957 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
1960 /* Remove the !PageUptodate pages we added */
1961 shmem_undo_range(inode
,
1962 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
1963 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
1968 * Inform shmem_writepage() how far we have reached.
1969 * No need for lock or barrier: we have the page lock.
1971 shmem_falloc
.next
++;
1972 if (!PageUptodate(page
))
1973 shmem_falloc
.nr_falloced
++;
1976 * If !PageUptodate, leave it that way so that freeable pages
1977 * can be recognized if we need to rollback on error later.
1978 * But set_page_dirty so that memory pressure will swap rather
1979 * than free the pages we are allocating (and SGP_CACHE pages
1980 * might still be clean: we now need to mark those dirty too).
1982 set_page_dirty(page
);
1984 page_cache_release(page
);
1988 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
1989 i_size_write(inode
, offset
+ len
);
1990 inode
->i_ctime
= CURRENT_TIME
;
1992 spin_lock(&inode
->i_lock
);
1993 inode
->i_private
= NULL
;
1994 spin_unlock(&inode
->i_lock
);
1996 mutex_unlock(&inode
->i_mutex
);
2000 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
2002 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
2004 buf
->f_type
= TMPFS_MAGIC
;
2005 buf
->f_bsize
= PAGE_CACHE_SIZE
;
2006 buf
->f_namelen
= NAME_MAX
;
2007 if (sbinfo
->max_blocks
) {
2008 buf
->f_blocks
= sbinfo
->max_blocks
;
2010 buf
->f_bfree
= sbinfo
->max_blocks
-
2011 percpu_counter_sum(&sbinfo
->used_blocks
);
2013 if (sbinfo
->max_inodes
) {
2014 buf
->f_files
= sbinfo
->max_inodes
;
2015 buf
->f_ffree
= sbinfo
->free_inodes
;
2017 /* else leave those fields 0 like simple_statfs */
2022 * File creation. Allocate an inode, and we're done..
2025 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
2027 struct inode
*inode
;
2028 int error
= -ENOSPC
;
2030 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
, 0);
2032 error
= security_inode_init_security(inode
, dir
,
2034 shmem_initxattrs
, NULL
);
2036 if (error
!= -EOPNOTSUPP
) {
2041 #ifdef CONFIG_TMPFS_POSIX_ACL
2042 error
= generic_acl_init(inode
, dir
);
2050 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2051 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2052 d_instantiate(dentry
, inode
);
2053 dget(dentry
); /* Extra count - pin the dentry in core */
2058 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2062 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2068 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2071 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2077 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2079 struct inode
*inode
= old_dentry
->d_inode
;
2083 * No ordinary (disk based) filesystem counts links as inodes;
2084 * but each new link needs a new dentry, pinning lowmem, and
2085 * tmpfs dentries cannot be pruned until they are unlinked.
2087 ret
= shmem_reserve_inode(inode
->i_sb
);
2091 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2092 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2094 ihold(inode
); /* New dentry reference */
2095 dget(dentry
); /* Extra pinning count for the created dentry */
2096 d_instantiate(dentry
, inode
);
2101 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2103 struct inode
*inode
= dentry
->d_inode
;
2105 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2106 shmem_free_inode(inode
->i_sb
);
2108 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2109 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2111 dput(dentry
); /* Undo the count from "create" - this does all the work */
2115 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2117 if (!simple_empty(dentry
))
2120 drop_nlink(dentry
->d_inode
);
2122 return shmem_unlink(dir
, dentry
);
2126 * The VFS layer already does all the dentry stuff for rename,
2127 * we just have to decrement the usage count for the target if
2128 * it exists so that the VFS layer correctly free's it when it
2131 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2133 struct inode
*inode
= old_dentry
->d_inode
;
2134 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2136 if (!simple_empty(new_dentry
))
2139 if (new_dentry
->d_inode
) {
2140 (void) shmem_unlink(new_dir
, new_dentry
);
2141 if (they_are_dirs
) {
2142 drop_nlink(new_dentry
->d_inode
);
2143 drop_nlink(old_dir
);
2145 } else if (they_are_dirs
) {
2146 drop_nlink(old_dir
);
2150 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2151 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2152 old_dir
->i_ctime
= old_dir
->i_mtime
=
2153 new_dir
->i_ctime
= new_dir
->i_mtime
=
2154 inode
->i_ctime
= CURRENT_TIME
;
2158 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2162 struct inode
*inode
;
2165 struct shmem_inode_info
*info
;
2167 len
= strlen(symname
) + 1;
2168 if (len
> PAGE_CACHE_SIZE
)
2169 return -ENAMETOOLONG
;
2171 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
, 0);
2175 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2176 shmem_initxattrs
, NULL
);
2178 if (error
!= -EOPNOTSUPP
) {
2185 info
= SHMEM_I(inode
);
2186 inode
->i_size
= len
-1;
2187 if (len
<= SHORT_SYMLINK_LEN
) {
2188 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2189 if (!info
->symlink
) {
2193 inode
->i_op
= &shmem_short_symlink_operations
;
2195 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2200 inode
->i_mapping
->a_ops
= &shmem_aops
;
2201 inode
->i_op
= &shmem_symlink_inode_operations
;
2202 kaddr
= kmap_atomic(page
);
2203 memcpy(kaddr
, symname
, len
);
2204 kunmap_atomic(kaddr
);
2205 SetPageUptodate(page
);
2206 set_page_dirty(page
);
2208 page_cache_release(page
);
2210 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2211 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2212 d_instantiate(dentry
, inode
);
2217 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
2219 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
2223 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
2225 struct page
*page
= NULL
;
2226 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
2227 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
2233 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2235 if (!IS_ERR(nd_get_link(nd
))) {
2236 struct page
*page
= cookie
;
2238 mark_page_accessed(page
);
2239 page_cache_release(page
);
2243 #ifdef CONFIG_TMPFS_XATTR
2245 * Superblocks without xattr inode operations may get some security.* xattr
2246 * support from the LSM "for free". As soon as we have any other xattrs
2247 * like ACLs, we also need to implement the security.* handlers at
2248 * filesystem level, though.
2252 * Callback for security_inode_init_security() for acquiring xattrs.
2254 static int shmem_initxattrs(struct inode
*inode
,
2255 const struct xattr
*xattr_array
,
2258 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2259 const struct xattr
*xattr
;
2260 struct simple_xattr
*new_xattr
;
2263 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2264 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2268 len
= strlen(xattr
->name
) + 1;
2269 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2271 if (!new_xattr
->name
) {
2276 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2277 XATTR_SECURITY_PREFIX_LEN
);
2278 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2281 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2287 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2288 #ifdef CONFIG_TMPFS_POSIX_ACL
2289 &generic_acl_access_handler
,
2290 &generic_acl_default_handler
,
2295 static int shmem_xattr_validate(const char *name
)
2297 struct { const char *prefix
; size_t len
; } arr
[] = {
2298 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2299 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2303 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2304 size_t preflen
= arr
[i
].len
;
2305 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2314 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2315 void *buffer
, size_t size
)
2317 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2321 * If this is a request for a synthetic attribute in the system.*
2322 * namespace use the generic infrastructure to resolve a handler
2323 * for it via sb->s_xattr.
2325 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2326 return generic_getxattr(dentry
, name
, buffer
, size
);
2328 err
= shmem_xattr_validate(name
);
2332 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2335 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2336 const void *value
, size_t size
, int flags
)
2338 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2342 * If this is a request for a synthetic attribute in the system.*
2343 * namespace use the generic infrastructure to resolve a handler
2344 * for it via sb->s_xattr.
2346 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2347 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2349 err
= shmem_xattr_validate(name
);
2353 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2356 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2358 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2362 * If this is a request for a synthetic attribute in the system.*
2363 * namespace use the generic infrastructure to resolve a handler
2364 * for it via sb->s_xattr.
2366 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2367 return generic_removexattr(dentry
, name
);
2369 err
= shmem_xattr_validate(name
);
2373 return simple_xattr_remove(&info
->xattrs
, name
);
2376 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2378 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2379 return simple_xattr_list(&info
->xattrs
, buffer
, size
);
2381 #endif /* CONFIG_TMPFS_XATTR */
2383 static const struct inode_operations shmem_short_symlink_operations
= {
2384 .readlink
= generic_readlink
,
2385 .follow_link
= shmem_follow_short_symlink
,
2386 #ifdef CONFIG_TMPFS_XATTR
2387 .setxattr
= shmem_setxattr
,
2388 .getxattr
= shmem_getxattr
,
2389 .listxattr
= shmem_listxattr
,
2390 .removexattr
= shmem_removexattr
,
2394 static const struct inode_operations shmem_symlink_inode_operations
= {
2395 .readlink
= generic_readlink
,
2396 .follow_link
= shmem_follow_link
,
2397 .put_link
= shmem_put_link
,
2398 #ifdef CONFIG_TMPFS_XATTR
2399 .setxattr
= shmem_setxattr
,
2400 .getxattr
= shmem_getxattr
,
2401 .listxattr
= shmem_listxattr
,
2402 .removexattr
= shmem_removexattr
,
2406 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2408 return ERR_PTR(-ESTALE
);
2411 static int shmem_match(struct inode
*ino
, void *vfh
)
2415 inum
= (inum
<< 32) | fh
[1];
2416 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2419 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2420 struct fid
*fid
, int fh_len
, int fh_type
)
2422 struct inode
*inode
;
2423 struct dentry
*dentry
= NULL
;
2430 inum
= (inum
<< 32) | fid
->raw
[1];
2432 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2433 shmem_match
, fid
->raw
);
2435 dentry
= d_find_alias(inode
);
2442 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2443 struct inode
*parent
)
2447 return FILEID_INVALID
;
2450 if (inode_unhashed(inode
)) {
2451 /* Unfortunately insert_inode_hash is not idempotent,
2452 * so as we hash inodes here rather than at creation
2453 * time, we need a lock to ensure we only try
2456 static DEFINE_SPINLOCK(lock
);
2458 if (inode_unhashed(inode
))
2459 __insert_inode_hash(inode
,
2460 inode
->i_ino
+ inode
->i_generation
);
2464 fh
[0] = inode
->i_generation
;
2465 fh
[1] = inode
->i_ino
;
2466 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2472 static const struct export_operations shmem_export_ops
= {
2473 .get_parent
= shmem_get_parent
,
2474 .encode_fh
= shmem_encode_fh
,
2475 .fh_to_dentry
= shmem_fh_to_dentry
,
2478 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2481 char *this_char
, *value
, *rest
;
2482 struct mempolicy
*mpol
= NULL
;
2486 while (options
!= NULL
) {
2487 this_char
= options
;
2490 * NUL-terminate this option: unfortunately,
2491 * mount options form a comma-separated list,
2492 * but mpol's nodelist may also contain commas.
2494 options
= strchr(options
, ',');
2495 if (options
== NULL
)
2498 if (!isdigit(*options
)) {
2505 if ((value
= strchr(this_char
,'=')) != NULL
) {
2509 "tmpfs: No value for mount option '%s'\n",
2514 if (!strcmp(this_char
,"size")) {
2515 unsigned long long size
;
2516 size
= memparse(value
,&rest
);
2518 size
<<= PAGE_SHIFT
;
2519 size
*= totalram_pages
;
2525 sbinfo
->max_blocks
=
2526 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2527 } else if (!strcmp(this_char
,"nr_blocks")) {
2528 sbinfo
->max_blocks
= memparse(value
, &rest
);
2531 } else if (!strcmp(this_char
,"nr_inodes")) {
2532 sbinfo
->max_inodes
= memparse(value
, &rest
);
2535 } else if (!strcmp(this_char
,"mode")) {
2538 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2541 } else if (!strcmp(this_char
,"uid")) {
2544 uid
= simple_strtoul(value
, &rest
, 0);
2547 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2548 if (!uid_valid(sbinfo
->uid
))
2550 } else if (!strcmp(this_char
,"gid")) {
2553 gid
= simple_strtoul(value
, &rest
, 0);
2556 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2557 if (!gid_valid(sbinfo
->gid
))
2559 } else if (!strcmp(this_char
,"mpol")) {
2562 if (mpol_parse_str(value
, &mpol
))
2565 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2570 sbinfo
->mpol
= mpol
;
2574 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2582 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2584 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2585 struct shmem_sb_info config
= *sbinfo
;
2586 unsigned long inodes
;
2587 int error
= -EINVAL
;
2590 if (shmem_parse_options(data
, &config
, true))
2593 spin_lock(&sbinfo
->stat_lock
);
2594 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2595 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2597 if (config
.max_inodes
< inodes
)
2600 * Those tests disallow limited->unlimited while any are in use;
2601 * but we must separately disallow unlimited->limited, because
2602 * in that case we have no record of how much is already in use.
2604 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2606 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2610 sbinfo
->max_blocks
= config
.max_blocks
;
2611 sbinfo
->max_inodes
= config
.max_inodes
;
2612 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2615 * Preserve previous mempolicy unless mpol remount option was specified.
2618 mpol_put(sbinfo
->mpol
);
2619 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2622 spin_unlock(&sbinfo
->stat_lock
);
2626 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2628 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2630 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2631 seq_printf(seq
, ",size=%luk",
2632 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2633 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2634 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2635 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2636 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2637 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2638 seq_printf(seq
, ",uid=%u",
2639 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2640 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2641 seq_printf(seq
, ",gid=%u",
2642 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2643 shmem_show_mpol(seq
, sbinfo
->mpol
);
2646 #endif /* CONFIG_TMPFS */
2648 static void shmem_put_super(struct super_block
*sb
)
2650 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2652 percpu_counter_destroy(&sbinfo
->used_blocks
);
2653 mpol_put(sbinfo
->mpol
);
2655 sb
->s_fs_info
= NULL
;
2658 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2660 struct inode
*inode
;
2661 struct shmem_sb_info
*sbinfo
;
2664 /* Round up to L1_CACHE_BYTES to resist false sharing */
2665 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2666 L1_CACHE_BYTES
), GFP_KERNEL
);
2670 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2671 sbinfo
->uid
= current_fsuid();
2672 sbinfo
->gid
= current_fsgid();
2673 sb
->s_fs_info
= sbinfo
;
2677 * Per default we only allow half of the physical ram per
2678 * tmpfs instance, limiting inodes to one per page of lowmem;
2679 * but the internal instance is left unlimited.
2681 if (!(sb
->s_flags
& MS_NOUSER
)) {
2682 sbinfo
->max_blocks
= shmem_default_max_blocks();
2683 sbinfo
->max_inodes
= shmem_default_max_inodes();
2684 if (shmem_parse_options(data
, sbinfo
, false)) {
2689 sb
->s_export_op
= &shmem_export_ops
;
2690 sb
->s_flags
|= MS_NOSEC
;
2692 sb
->s_flags
|= MS_NOUSER
;
2695 spin_lock_init(&sbinfo
->stat_lock
);
2696 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
2698 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2700 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
2701 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2702 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2703 sb
->s_magic
= TMPFS_MAGIC
;
2704 sb
->s_op
= &shmem_ops
;
2705 sb
->s_time_gran
= 1;
2706 #ifdef CONFIG_TMPFS_XATTR
2707 sb
->s_xattr
= shmem_xattr_handlers
;
2709 #ifdef CONFIG_TMPFS_POSIX_ACL
2710 sb
->s_flags
|= MS_POSIXACL
;
2713 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
, 0);
2716 inode
->i_uid
= sbinfo
->uid
;
2717 inode
->i_gid
= sbinfo
->gid
;
2718 sb
->s_root
= d_make_root(inode
);
2724 shmem_put_super(sb
);
2728 static struct kmem_cache
*shmem_inode_cachep
;
2730 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2732 struct shmem_inode_info
*info
;
2733 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2736 return &info
->vfs_inode
;
2739 static void shmem_destroy_callback(struct rcu_head
*head
)
2741 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
2742 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2745 static void shmem_destroy_inode(struct inode
*inode
)
2747 if (S_ISREG(inode
->i_mode
))
2748 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2749 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
2752 static void shmem_init_inode(void *foo
)
2754 struct shmem_inode_info
*info
= foo
;
2755 inode_init_once(&info
->vfs_inode
);
2758 static int shmem_init_inodecache(void)
2760 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2761 sizeof(struct shmem_inode_info
),
2762 0, SLAB_PANIC
, shmem_init_inode
);
2766 static void shmem_destroy_inodecache(void)
2768 kmem_cache_destroy(shmem_inode_cachep
);
2771 static const struct address_space_operations shmem_aops
= {
2772 .writepage
= shmem_writepage
,
2773 .set_page_dirty
= __set_page_dirty_no_writeback
,
2775 .write_begin
= shmem_write_begin
,
2776 .write_end
= shmem_write_end
,
2778 .migratepage
= migrate_page
,
2779 .error_remove_page
= generic_error_remove_page
,
2782 static const struct file_operations shmem_file_operations
= {
2785 .llseek
= shmem_file_llseek
,
2786 .read
= do_sync_read
,
2787 .write
= do_sync_write
,
2788 .aio_read
= shmem_file_aio_read
,
2789 .aio_write
= generic_file_aio_write
,
2790 .fsync
= noop_fsync
,
2791 .splice_read
= shmem_file_splice_read
,
2792 .splice_write
= generic_file_splice_write
,
2793 .fallocate
= shmem_fallocate
,
2797 static const struct inode_operations shmem_inode_operations
= {
2798 .setattr
= shmem_setattr
,
2799 #ifdef CONFIG_TMPFS_XATTR
2800 .setxattr
= shmem_setxattr
,
2801 .getxattr
= shmem_getxattr
,
2802 .listxattr
= shmem_listxattr
,
2803 .removexattr
= shmem_removexattr
,
2807 static const struct inode_operations shmem_dir_inode_operations
= {
2809 .create
= shmem_create
,
2810 .lookup
= simple_lookup
,
2812 .unlink
= shmem_unlink
,
2813 .symlink
= shmem_symlink
,
2814 .mkdir
= shmem_mkdir
,
2815 .rmdir
= shmem_rmdir
,
2816 .mknod
= shmem_mknod
,
2817 .rename
= shmem_rename
,
2819 #ifdef CONFIG_TMPFS_XATTR
2820 .setxattr
= shmem_setxattr
,
2821 .getxattr
= shmem_getxattr
,
2822 .listxattr
= shmem_listxattr
,
2823 .removexattr
= shmem_removexattr
,
2825 #ifdef CONFIG_TMPFS_POSIX_ACL
2826 .setattr
= shmem_setattr
,
2830 static const struct inode_operations shmem_special_inode_operations
= {
2831 #ifdef CONFIG_TMPFS_XATTR
2832 .setxattr
= shmem_setxattr
,
2833 .getxattr
= shmem_getxattr
,
2834 .listxattr
= shmem_listxattr
,
2835 .removexattr
= shmem_removexattr
,
2837 #ifdef CONFIG_TMPFS_POSIX_ACL
2838 .setattr
= shmem_setattr
,
2842 static const struct super_operations shmem_ops
= {
2843 .alloc_inode
= shmem_alloc_inode
,
2844 .destroy_inode
= shmem_destroy_inode
,
2846 .statfs
= shmem_statfs
,
2847 .remount_fs
= shmem_remount_fs
,
2848 .show_options
= shmem_show_options
,
2850 .evict_inode
= shmem_evict_inode
,
2851 .drop_inode
= generic_delete_inode
,
2852 .put_super
= shmem_put_super
,
2855 static const struct vm_operations_struct shmem_vm_ops
= {
2856 .fault
= shmem_fault
,
2858 .set_policy
= shmem_set_policy
,
2859 .get_policy
= shmem_get_policy
,
2861 .remap_pages
= generic_file_remap_pages
,
2864 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
2865 int flags
, const char *dev_name
, void *data
)
2867 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
2870 static struct file_system_type shmem_fs_type
= {
2871 .owner
= THIS_MODULE
,
2873 .mount
= shmem_mount
,
2874 .kill_sb
= kill_litter_super
,
2875 .fs_flags
= FS_USERNS_MOUNT
,
2878 int __init
shmem_init(void)
2882 error
= bdi_init(&shmem_backing_dev_info
);
2886 error
= shmem_init_inodecache();
2890 error
= register_filesystem(&shmem_fs_type
);
2892 printk(KERN_ERR
"Could not register tmpfs\n");
2896 shm_mnt
= vfs_kern_mount(&shmem_fs_type
, MS_NOUSER
,
2897 shmem_fs_type
.name
, NULL
);
2898 if (IS_ERR(shm_mnt
)) {
2899 error
= PTR_ERR(shm_mnt
);
2900 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2906 unregister_filesystem(&shmem_fs_type
);
2908 shmem_destroy_inodecache();
2910 bdi_destroy(&shmem_backing_dev_info
);
2912 shm_mnt
= ERR_PTR(error
);
2916 #else /* !CONFIG_SHMEM */
2919 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2921 * This is intended for small system where the benefits of the full
2922 * shmem code (swap-backed and resource-limited) are outweighed by
2923 * their complexity. On systems without swap this code should be
2924 * effectively equivalent, but much lighter weight.
2927 static struct file_system_type shmem_fs_type
= {
2929 .mount
= ramfs_mount
,
2930 .kill_sb
= kill_litter_super
,
2931 .fs_flags
= FS_USERNS_MOUNT
,
2934 int __init
shmem_init(void)
2936 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
2938 shm_mnt
= kern_mount(&shmem_fs_type
);
2939 BUG_ON(IS_ERR(shm_mnt
));
2944 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
2949 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2954 void shmem_unlock_mapping(struct address_space
*mapping
)
2958 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
2960 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
2962 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
2964 #define shmem_vm_ops generic_file_vm_ops
2965 #define shmem_file_operations ramfs_file_operations
2966 #define shmem_get_inode(sb, dir, mode, dev, flags, atomic_copy) ramfs_get_inode(sb, dir, mode, dev)
2967 #define shmem_acct_size(flags, size) 0
2968 #define shmem_unacct_size(flags, size) do {} while (0)
2970 #endif /* CONFIG_SHMEM */
2974 static struct dentry_operations anon_ops
= {
2975 .d_dname
= simple_dname
2979 * shmem_file_setup - get an unlinked file living in tmpfs
2980 * @name: name for dentry (to be seen in /proc/<pid>/maps
2981 * @size: size to be set for the file
2982 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2983 * @atomic_copy: Atomically copy the area when hibernating?
2985 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
,
2989 struct inode
*inode
;
2991 struct super_block
*sb
;
2994 if (IS_ERR(shm_mnt
))
2995 return ERR_CAST(shm_mnt
);
2997 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
2998 return ERR_PTR(-EINVAL
);
3000 if (shmem_acct_size(flags
, size
))
3001 return ERR_PTR(-ENOMEM
);
3003 res
= ERR_PTR(-ENOMEM
);
3005 this.len
= strlen(name
);
3006 this.hash
= 0; /* will go */
3007 sb
= shm_mnt
->mnt_sb
;
3008 path
.dentry
= d_alloc_pseudo(sb
, &this);
3011 d_set_d_op(path
.dentry
, &anon_ops
);
3012 path
.mnt
= mntget(shm_mnt
);
3014 res
= ERR_PTR(-ENOSPC
);
3015 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
, atomic_copy
);
3019 d_instantiate(path
.dentry
, inode
);
3020 inode
->i_size
= size
;
3021 clear_nlink(inode
); /* It is unlinked */
3022 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
3026 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
3027 &shmem_file_operations
);
3036 shmem_unacct_size(flags
, size
);
3039 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3041 void shmem_set_file(struct vm_area_struct
*vma
, struct file
*file
)
3045 vma
->vm_file
= file
;
3046 vma
->vm_ops
= &shmem_vm_ops
;
3050 * shmem_zero_setup - setup a shared anonymous mapping
3051 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3053 int shmem_zero_setup(struct vm_area_struct
*vma
)
3056 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3058 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
, 0);
3060 return PTR_ERR(file
);
3062 shmem_set_file(vma
, file
);
3067 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3068 * @mapping: the page's address_space
3069 * @index: the page index
3070 * @gfp: the page allocator flags to use if allocating
3072 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3073 * with any new page allocations done using the specified allocation flags.
3074 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3075 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3076 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3078 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3079 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3081 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3082 pgoff_t index
, gfp_t gfp
)
3085 struct inode
*inode
= mapping
->host
;
3089 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3090 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3092 page
= ERR_PTR(error
);
3098 * The tiny !SHMEM case uses ramfs without swap
3100 return read_cache_page_gfp(mapping
, index
, gfp
);
3103 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);