2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/aio.h>
36 static struct vfsmount
*shm_mnt
;
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/posix_acl_xattr.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
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
));
152 static inline int shmem_reacct_size(unsigned long flags
,
153 loff_t oldsize
, loff_t newsize
)
155 if (!(flags
& VM_NORESERVE
)) {
156 if (VM_ACCT(newsize
) > VM_ACCT(oldsize
))
157 return security_vm_enough_memory_mm(current
->mm
,
158 VM_ACCT(newsize
) - VM_ACCT(oldsize
));
159 else if (VM_ACCT(newsize
) < VM_ACCT(oldsize
))
160 vm_unacct_memory(VM_ACCT(oldsize
) - VM_ACCT(newsize
));
166 * ... whereas tmpfs objects are accounted incrementally as
167 * pages are allocated, in order to allow huge sparse files.
168 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
169 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
171 static inline int shmem_acct_block(unsigned long flags
)
173 return (flags
& VM_NORESERVE
) ?
174 security_vm_enough_memory_mm(current
->mm
, VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
177 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
179 if (flags
& VM_NORESERVE
)
180 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
183 static const struct super_operations shmem_ops
;
184 static const struct address_space_operations shmem_aops
;
185 static const struct file_operations shmem_file_operations
;
186 static const struct inode_operations shmem_inode_operations
;
187 static const struct inode_operations shmem_dir_inode_operations
;
188 static const struct inode_operations shmem_special_inode_operations
;
189 static const struct vm_operations_struct shmem_vm_ops
;
191 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
192 .ra_pages
= 0, /* No readahead */
193 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
196 static LIST_HEAD(shmem_swaplist
);
197 static DEFINE_MUTEX(shmem_swaplist_mutex
);
199 static int shmem_reserve_inode(struct super_block
*sb
)
201 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
202 if (sbinfo
->max_inodes
) {
203 spin_lock(&sbinfo
->stat_lock
);
204 if (!sbinfo
->free_inodes
) {
205 spin_unlock(&sbinfo
->stat_lock
);
208 sbinfo
->free_inodes
--;
209 spin_unlock(&sbinfo
->stat_lock
);
214 static void shmem_free_inode(struct super_block
*sb
)
216 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
217 if (sbinfo
->max_inodes
) {
218 spin_lock(&sbinfo
->stat_lock
);
219 sbinfo
->free_inodes
++;
220 spin_unlock(&sbinfo
->stat_lock
);
225 * shmem_recalc_inode - recalculate the block usage of an inode
226 * @inode: inode to recalc
228 * We have to calculate the free blocks since the mm can drop
229 * undirtied hole pages behind our back.
231 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
232 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
234 * It has to be called with the spinlock held.
236 static void shmem_recalc_inode(struct inode
*inode
)
238 struct shmem_inode_info
*info
= SHMEM_I(inode
);
241 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
243 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
244 if (sbinfo
->max_blocks
)
245 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
246 info
->alloced
-= freed
;
247 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
248 shmem_unacct_blocks(info
->flags
, freed
);
253 * Replace item expected in radix tree by a new item, while holding tree lock.
255 static int shmem_radix_tree_replace(struct address_space
*mapping
,
256 pgoff_t index
, void *expected
, void *replacement
)
261 VM_BUG_ON(!expected
);
262 VM_BUG_ON(!replacement
);
263 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
266 item
= radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
);
267 if (item
!= expected
)
269 radix_tree_replace_slot(pslot
, replacement
);
274 * Sometimes, before we decide whether to proceed or to fail, we must check
275 * that an entry was not already brought back from swap by a racing thread.
277 * Checking page is not enough: by the time a SwapCache page is locked, it
278 * might be reused, and again be SwapCache, using the same swap as before.
280 static bool shmem_confirm_swap(struct address_space
*mapping
,
281 pgoff_t index
, swp_entry_t swap
)
286 item
= radix_tree_lookup(&mapping
->page_tree
, index
);
288 return item
== swp_to_radix_entry(swap
);
292 * Like add_to_page_cache_locked, but error if expected item has gone.
294 static int shmem_add_to_page_cache(struct page
*page
,
295 struct address_space
*mapping
,
296 pgoff_t index
, void *expected
)
300 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
301 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
303 page_cache_get(page
);
304 page
->mapping
= mapping
;
307 spin_lock_irq(&mapping
->tree_lock
);
309 error
= radix_tree_insert(&mapping
->page_tree
, index
, page
);
311 error
= shmem_radix_tree_replace(mapping
, index
, expected
,
315 __inc_zone_page_state(page
, NR_FILE_PAGES
);
316 __inc_zone_page_state(page
, NR_SHMEM
);
317 spin_unlock_irq(&mapping
->tree_lock
);
319 page
->mapping
= NULL
;
320 spin_unlock_irq(&mapping
->tree_lock
);
321 page_cache_release(page
);
327 * Like delete_from_page_cache, but substitutes swap for page.
329 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
331 struct address_space
*mapping
= page
->mapping
;
334 spin_lock_irq(&mapping
->tree_lock
);
335 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
336 page
->mapping
= NULL
;
338 __dec_zone_page_state(page
, NR_FILE_PAGES
);
339 __dec_zone_page_state(page
, NR_SHMEM
);
340 spin_unlock_irq(&mapping
->tree_lock
);
341 page_cache_release(page
);
346 * Remove swap entry from radix tree, free the swap and its page cache.
348 static int shmem_free_swap(struct address_space
*mapping
,
349 pgoff_t index
, void *radswap
)
353 spin_lock_irq(&mapping
->tree_lock
);
354 old
= radix_tree_delete_item(&mapping
->page_tree
, index
, radswap
);
355 spin_unlock_irq(&mapping
->tree_lock
);
358 free_swap_and_cache(radix_to_swp_entry(radswap
));
363 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
365 void shmem_unlock_mapping(struct address_space
*mapping
)
368 pgoff_t indices
[PAGEVEC_SIZE
];
371 pagevec_init(&pvec
, 0);
373 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
375 while (!mapping_unevictable(mapping
)) {
377 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
378 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
380 pvec
.nr
= find_get_entries(mapping
, index
,
381 PAGEVEC_SIZE
, pvec
.pages
, indices
);
384 index
= indices
[pvec
.nr
- 1] + 1;
385 pagevec_remove_exceptionals(&pvec
);
386 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
387 pagevec_release(&pvec
);
393 * Remove range of pages and swap entries from radix tree, and free them.
394 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
396 static void shmem_undo_range(struct inode
*inode
, loff_t lstart
, loff_t lend
,
399 struct address_space
*mapping
= inode
->i_mapping
;
400 struct shmem_inode_info
*info
= SHMEM_I(inode
);
401 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
402 pgoff_t end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
403 unsigned int partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
404 unsigned int partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
406 pgoff_t indices
[PAGEVEC_SIZE
];
407 long nr_swaps_freed
= 0;
412 end
= -1; /* unsigned, so actually very big */
414 pagevec_init(&pvec
, 0);
416 while (index
< end
) {
417 pvec
.nr
= find_get_entries(mapping
, index
,
418 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
419 pvec
.pages
, indices
);
422 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
423 struct page
*page
= pvec
.pages
[i
];
429 if (radix_tree_exceptional_entry(page
)) {
432 nr_swaps_freed
+= !shmem_free_swap(mapping
,
437 if (!trylock_page(page
))
439 if (!unfalloc
|| !PageUptodate(page
)) {
440 if (page
->mapping
== mapping
) {
441 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
442 truncate_inode_page(mapping
, page
);
447 pagevec_remove_exceptionals(&pvec
);
448 pagevec_release(&pvec
);
454 struct page
*page
= NULL
;
455 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
457 unsigned int top
= PAGE_CACHE_SIZE
;
462 zero_user_segment(page
, partial_start
, top
);
463 set_page_dirty(page
);
465 page_cache_release(page
);
469 struct page
*page
= NULL
;
470 shmem_getpage(inode
, end
, &page
, SGP_READ
, NULL
);
472 zero_user_segment(page
, 0, partial_end
);
473 set_page_dirty(page
);
475 page_cache_release(page
);
482 while (index
< end
) {
485 pvec
.nr
= find_get_entries(mapping
, index
,
486 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
487 pvec
.pages
, indices
);
489 /* If all gone or hole-punch or unfalloc, we're done */
490 if (index
== start
|| end
!= -1)
492 /* But if truncating, restart to make sure all gone */
496 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
497 struct page
*page
= pvec
.pages
[i
];
503 if (radix_tree_exceptional_entry(page
)) {
506 if (shmem_free_swap(mapping
, index
, page
)) {
507 /* Swap was replaced by page: retry */
516 if (!unfalloc
|| !PageUptodate(page
)) {
517 if (page
->mapping
== mapping
) {
518 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
519 truncate_inode_page(mapping
, page
);
521 /* Page was replaced by swap: retry */
529 pagevec_remove_exceptionals(&pvec
);
530 pagevec_release(&pvec
);
534 spin_lock(&info
->lock
);
535 info
->swapped
-= nr_swaps_freed
;
536 shmem_recalc_inode(inode
);
537 spin_unlock(&info
->lock
);
540 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
542 shmem_undo_range(inode
, lstart
, lend
, false);
543 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
545 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
547 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
549 struct inode
*inode
= dentry
->d_inode
;
552 error
= inode_change_ok(inode
, attr
);
556 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
557 loff_t oldsize
= inode
->i_size
;
558 loff_t newsize
= attr
->ia_size
;
560 if (newsize
!= oldsize
) {
561 error
= shmem_reacct_size(SHMEM_I(inode
)->flags
,
565 i_size_write(inode
, newsize
);
566 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
568 if (newsize
< oldsize
) {
569 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
570 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
571 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
572 /* unmap again to remove racily COWed private pages */
573 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
577 setattr_copy(inode
, attr
);
578 if (attr
->ia_valid
& ATTR_MODE
)
579 error
= posix_acl_chmod(inode
, inode
->i_mode
);
583 static void shmem_evict_inode(struct inode
*inode
)
585 struct shmem_inode_info
*info
= SHMEM_I(inode
);
587 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
588 shmem_unacct_size(info
->flags
, inode
->i_size
);
590 shmem_truncate_range(inode
, 0, (loff_t
)-1);
591 if (!list_empty(&info
->swaplist
)) {
592 mutex_lock(&shmem_swaplist_mutex
);
593 list_del_init(&info
->swaplist
);
594 mutex_unlock(&shmem_swaplist_mutex
);
597 kfree(info
->symlink
);
599 simple_xattrs_free(&info
->xattrs
);
600 WARN_ON(inode
->i_blocks
);
601 shmem_free_inode(inode
->i_sb
);
606 * If swap found in inode, free it and move page from swapcache to filecache.
608 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
609 swp_entry_t swap
, struct page
**pagep
)
611 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
617 radswap
= swp_to_radix_entry(swap
);
618 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
620 return -EAGAIN
; /* tell shmem_unuse we found nothing */
623 * Move _head_ to start search for next from here.
624 * But be careful: shmem_evict_inode checks list_empty without taking
625 * mutex, and there's an instant in list_move_tail when info->swaplist
626 * would appear empty, if it were the only one on shmem_swaplist.
628 if (shmem_swaplist
.next
!= &info
->swaplist
)
629 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
631 gfp
= mapping_gfp_mask(mapping
);
632 if (shmem_should_replace_page(*pagep
, gfp
)) {
633 mutex_unlock(&shmem_swaplist_mutex
);
634 error
= shmem_replace_page(pagep
, gfp
, info
, index
);
635 mutex_lock(&shmem_swaplist_mutex
);
637 * We needed to drop mutex to make that restrictive page
638 * allocation, but the inode might have been freed while we
639 * dropped it: although a racing shmem_evict_inode() cannot
640 * complete without emptying the radix_tree, our page lock
641 * on this swapcache page is not enough to prevent that -
642 * free_swap_and_cache() of our swap entry will only
643 * trylock_page(), removing swap from radix_tree whatever.
645 * We must not proceed to shmem_add_to_page_cache() if the
646 * inode has been freed, but of course we cannot rely on
647 * inode or mapping or info to check that. However, we can
648 * safely check if our swap entry is still in use (and here
649 * it can't have got reused for another page): if it's still
650 * in use, then the inode cannot have been freed yet, and we
651 * can safely proceed (if it's no longer in use, that tells
652 * nothing about the inode, but we don't need to unuse swap).
654 if (!page_swapcount(*pagep
))
659 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
660 * but also to hold up shmem_evict_inode(): so inode cannot be freed
661 * beneath us (pagelock doesn't help until the page is in pagecache).
664 error
= shmem_add_to_page_cache(*pagep
, mapping
, index
,
666 if (error
!= -ENOMEM
) {
668 * Truncation and eviction use free_swap_and_cache(), which
669 * only does trylock page: if we raced, best clean up here.
671 delete_from_swap_cache(*pagep
);
672 set_page_dirty(*pagep
);
674 spin_lock(&info
->lock
);
676 spin_unlock(&info
->lock
);
684 * Search through swapped inodes to find and replace swap by page.
686 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
688 struct list_head
*this, *next
;
689 struct shmem_inode_info
*info
;
690 struct mem_cgroup
*memcg
;
694 * There's a faint possibility that swap page was replaced before
695 * caller locked it: caller will come back later with the right page.
697 if (unlikely(!PageSwapCache(page
) || page_private(page
) != swap
.val
))
701 * Charge page using GFP_KERNEL while we can wait, before taking
702 * the shmem_swaplist_mutex which might hold up shmem_writepage().
703 * Charged back to the user (not to caller) when swap account is used.
705 error
= mem_cgroup_try_charge(page
, current
->mm
, GFP_KERNEL
, &memcg
);
708 /* No radix_tree_preload: swap entry keeps a place for page in tree */
711 mutex_lock(&shmem_swaplist_mutex
);
712 list_for_each_safe(this, next
, &shmem_swaplist
) {
713 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
715 error
= shmem_unuse_inode(info
, swap
, &page
);
717 list_del_init(&info
->swaplist
);
719 if (error
!= -EAGAIN
)
721 /* found nothing in this: move on to search the next */
723 mutex_unlock(&shmem_swaplist_mutex
);
726 if (error
!= -ENOMEM
)
728 mem_cgroup_cancel_charge(page
, memcg
);
730 mem_cgroup_commit_charge(page
, memcg
, true);
733 page_cache_release(page
);
738 * Move the page from the page cache to the swap cache.
740 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
742 struct shmem_inode_info
*info
;
743 struct address_space
*mapping
;
748 BUG_ON(!PageLocked(page
));
749 mapping
= page
->mapping
;
751 inode
= mapping
->host
;
752 info
= SHMEM_I(inode
);
753 if (info
->flags
& VM_LOCKED
)
755 if (!total_swap_pages
)
759 * shmem_backing_dev_info's capabilities prevent regular writeback or
760 * sync from ever calling shmem_writepage; but a stacking filesystem
761 * might use ->writepage of its underlying filesystem, in which case
762 * tmpfs should write out to swap only in response to memory pressure,
763 * and not for the writeback threads or sync.
765 if (!wbc
->for_reclaim
) {
766 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
771 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
772 * value into swapfile.c, the only way we can correctly account for a
773 * fallocated page arriving here is now to initialize it and write it.
775 * That's okay for a page already fallocated earlier, but if we have
776 * not yet completed the fallocation, then (a) we want to keep track
777 * of this page in case we have to undo it, and (b) it may not be a
778 * good idea to continue anyway, once we're pushing into swap. So
779 * reactivate the page, and let shmem_fallocate() quit when too many.
781 if (!PageUptodate(page
)) {
782 if (inode
->i_private
) {
783 struct shmem_falloc
*shmem_falloc
;
784 spin_lock(&inode
->i_lock
);
785 shmem_falloc
= inode
->i_private
;
787 !shmem_falloc
->waitq
&&
788 index
>= shmem_falloc
->start
&&
789 index
< shmem_falloc
->next
)
790 shmem_falloc
->nr_unswapped
++;
793 spin_unlock(&inode
->i_lock
);
797 clear_highpage(page
);
798 flush_dcache_page(page
);
799 SetPageUptodate(page
);
802 swap
= get_swap_page();
807 * Add inode to shmem_unuse()'s list of swapped-out inodes,
808 * if it's not already there. Do it now before the page is
809 * moved to swap cache, when its pagelock no longer protects
810 * the inode from eviction. But don't unlock the mutex until
811 * we've incremented swapped, because shmem_unuse_inode() will
812 * prune a !swapped inode from the swaplist under this mutex.
814 mutex_lock(&shmem_swaplist_mutex
);
815 if (list_empty(&info
->swaplist
))
816 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
818 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
819 swap_shmem_alloc(swap
);
820 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
822 spin_lock(&info
->lock
);
824 shmem_recalc_inode(inode
);
825 spin_unlock(&info
->lock
);
827 mutex_unlock(&shmem_swaplist_mutex
);
828 BUG_ON(page_mapped(page
));
829 swap_writepage(page
, wbc
);
833 mutex_unlock(&shmem_swaplist_mutex
);
834 swapcache_free(swap
);
836 set_page_dirty(page
);
837 if (wbc
->for_reclaim
)
838 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
845 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
849 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
850 return; /* show nothing */
852 mpol_to_str(buffer
, sizeof(buffer
), mpol
);
854 seq_printf(seq
, ",mpol=%s", buffer
);
857 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
859 struct mempolicy
*mpol
= NULL
;
861 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
864 spin_unlock(&sbinfo
->stat_lock
);
868 #endif /* CONFIG_TMPFS */
870 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
871 struct shmem_inode_info
*info
, pgoff_t index
)
873 struct vm_area_struct pvma
;
876 /* Create a pseudo vma that just contains the policy */
878 /* Bias interleave by inode number to distribute better across nodes */
879 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
881 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
883 page
= swapin_readahead(swap
, gfp
, &pvma
, 0);
885 /* Drop reference taken by mpol_shared_policy_lookup() */
886 mpol_cond_put(pvma
.vm_policy
);
891 static struct page
*shmem_alloc_page(gfp_t gfp
,
892 struct shmem_inode_info
*info
, pgoff_t index
)
894 struct vm_area_struct pvma
;
897 /* Create a pseudo vma that just contains the policy */
899 /* Bias interleave by inode number to distribute better across nodes */
900 pvma
.vm_pgoff
= index
+ info
->vfs_inode
.i_ino
;
902 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
904 page
= alloc_page_vma(gfp
, &pvma
, 0);
906 /* Drop reference taken by mpol_shared_policy_lookup() */
907 mpol_cond_put(pvma
.vm_policy
);
911 #else /* !CONFIG_NUMA */
913 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
916 #endif /* CONFIG_TMPFS */
918 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
919 struct shmem_inode_info
*info
, pgoff_t index
)
921 return swapin_readahead(swap
, gfp
, NULL
, 0);
924 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
925 struct shmem_inode_info
*info
, pgoff_t index
)
927 return alloc_page(gfp
);
929 #endif /* CONFIG_NUMA */
931 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
932 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
939 * When a page is moved from swapcache to shmem filecache (either by the
940 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
941 * shmem_unuse_inode()), it may have been read in earlier from swap, in
942 * ignorance of the mapping it belongs to. If that mapping has special
943 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
944 * we may need to copy to a suitable page before moving to filecache.
946 * In a future release, this may well be extended to respect cpuset and
947 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
948 * but for now it is a simple matter of zone.
950 static bool shmem_should_replace_page(struct page
*page
, gfp_t gfp
)
952 return page_zonenum(page
) > gfp_zone(gfp
);
955 static int shmem_replace_page(struct page
**pagep
, gfp_t gfp
,
956 struct shmem_inode_info
*info
, pgoff_t index
)
958 struct page
*oldpage
, *newpage
;
959 struct address_space
*swap_mapping
;
964 swap_index
= page_private(oldpage
);
965 swap_mapping
= page_mapping(oldpage
);
968 * We have arrived here because our zones are constrained, so don't
969 * limit chance of success by further cpuset and node constraints.
971 gfp
&= ~GFP_CONSTRAINT_MASK
;
972 newpage
= shmem_alloc_page(gfp
, info
, index
);
976 page_cache_get(newpage
);
977 copy_highpage(newpage
, oldpage
);
978 flush_dcache_page(newpage
);
980 __set_page_locked(newpage
);
981 SetPageUptodate(newpage
);
982 SetPageSwapBacked(newpage
);
983 set_page_private(newpage
, swap_index
);
984 SetPageSwapCache(newpage
);
987 * Our caller will very soon move newpage out of swapcache, but it's
988 * a nice clean interface for us to replace oldpage by newpage there.
990 spin_lock_irq(&swap_mapping
->tree_lock
);
991 error
= shmem_radix_tree_replace(swap_mapping
, swap_index
, oldpage
,
994 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
995 __dec_zone_page_state(oldpage
, NR_FILE_PAGES
);
997 spin_unlock_irq(&swap_mapping
->tree_lock
);
999 if (unlikely(error
)) {
1001 * Is this possible? I think not, now that our callers check
1002 * both PageSwapCache and page_private after getting page lock;
1003 * but be defensive. Reverse old to newpage for clear and free.
1007 mem_cgroup_migrate(oldpage
, newpage
, false);
1008 lru_cache_add_anon(newpage
);
1012 ClearPageSwapCache(oldpage
);
1013 set_page_private(oldpage
, 0);
1015 unlock_page(oldpage
);
1016 page_cache_release(oldpage
);
1017 page_cache_release(oldpage
);
1022 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1024 * If we allocate a new one we do not mark it dirty. That's up to the
1025 * vm. If we swap it in we mark it dirty since we also free the swap
1026 * entry since a page cannot live in both the swap and page cache
1028 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
1029 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
1031 struct address_space
*mapping
= inode
->i_mapping
;
1032 struct shmem_inode_info
*info
;
1033 struct shmem_sb_info
*sbinfo
;
1034 struct mem_cgroup
*memcg
;
1041 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
1045 page
= find_lock_entry(mapping
, index
);
1046 if (radix_tree_exceptional_entry(page
)) {
1047 swap
= radix_to_swp_entry(page
);
1051 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1052 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1057 if (page
&& sgp
== SGP_WRITE
)
1058 mark_page_accessed(page
);
1060 /* fallocated page? */
1061 if (page
&& !PageUptodate(page
)) {
1062 if (sgp
!= SGP_READ
)
1065 page_cache_release(page
);
1068 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
1074 * Fast cache lookup did not find it:
1075 * bring it back from swap or allocate.
1077 info
= SHMEM_I(inode
);
1078 sbinfo
= SHMEM_SB(inode
->i_sb
);
1081 /* Look it up and read it in.. */
1082 page
= lookup_swap_cache(swap
);
1084 /* here we actually do the io */
1086 *fault_type
|= VM_FAULT_MAJOR
;
1087 page
= shmem_swapin(swap
, gfp
, info
, index
);
1094 /* We have to do this with page locked to prevent races */
1096 if (!PageSwapCache(page
) || page_private(page
) != swap
.val
||
1097 !shmem_confirm_swap(mapping
, index
, swap
)) {
1098 error
= -EEXIST
; /* try again */
1101 if (!PageUptodate(page
)) {
1105 wait_on_page_writeback(page
);
1107 if (shmem_should_replace_page(page
, gfp
)) {
1108 error
= shmem_replace_page(&page
, gfp
, info
, index
);
1113 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1115 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1116 swp_to_radix_entry(swap
));
1118 * We already confirmed swap under page lock, and make
1119 * no memory allocation here, so usually no possibility
1120 * of error; but free_swap_and_cache() only trylocks a
1121 * page, so it is just possible that the entry has been
1122 * truncated or holepunched since swap was confirmed.
1123 * shmem_undo_range() will have done some of the
1124 * unaccounting, now delete_from_swap_cache() will do
1125 * the rest (including mem_cgroup_uncharge_swapcache).
1126 * Reset swap.val? No, leave it so "failed" goes back to
1127 * "repeat": reading a hole and writing should succeed.
1130 mem_cgroup_cancel_charge(page
, memcg
);
1131 delete_from_swap_cache(page
);
1137 mem_cgroup_commit_charge(page
, memcg
, true);
1139 spin_lock(&info
->lock
);
1141 shmem_recalc_inode(inode
);
1142 spin_unlock(&info
->lock
);
1144 if (sgp
== SGP_WRITE
)
1145 mark_page_accessed(page
);
1147 delete_from_swap_cache(page
);
1148 set_page_dirty(page
);
1152 if (shmem_acct_block(info
->flags
)) {
1156 if (sbinfo
->max_blocks
) {
1157 if (percpu_counter_compare(&sbinfo
->used_blocks
,
1158 sbinfo
->max_blocks
) >= 0) {
1162 percpu_counter_inc(&sbinfo
->used_blocks
);
1165 page
= shmem_alloc_page(gfp
, info
, index
);
1171 __SetPageSwapBacked(page
);
1172 __set_page_locked(page
);
1173 if (sgp
== SGP_WRITE
)
1174 __SetPageReferenced(page
);
1176 error
= mem_cgroup_try_charge(page
, current
->mm
, gfp
, &memcg
);
1179 error
= radix_tree_maybe_preload(gfp
& GFP_RECLAIM_MASK
);
1181 error
= shmem_add_to_page_cache(page
, mapping
, index
,
1183 radix_tree_preload_end();
1186 mem_cgroup_cancel_charge(page
, memcg
);
1189 mem_cgroup_commit_charge(page
, memcg
, false);
1190 lru_cache_add_anon(page
);
1192 spin_lock(&info
->lock
);
1194 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
1195 shmem_recalc_inode(inode
);
1196 spin_unlock(&info
->lock
);
1200 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1202 if (sgp
== SGP_FALLOC
)
1206 * Let SGP_WRITE caller clear ends if write does not fill page;
1207 * but SGP_FALLOC on a page fallocated earlier must initialize
1208 * it now, lest undo on failure cancel our earlier guarantee.
1210 if (sgp
!= SGP_WRITE
) {
1211 clear_highpage(page
);
1212 flush_dcache_page(page
);
1213 SetPageUptodate(page
);
1215 if (sgp
== SGP_DIRTY
)
1216 set_page_dirty(page
);
1219 /* Perhaps the file has been truncated since we checked */
1220 if (sgp
!= SGP_WRITE
&& sgp
!= SGP_FALLOC
&&
1221 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1235 info
= SHMEM_I(inode
);
1236 ClearPageDirty(page
);
1237 delete_from_page_cache(page
);
1238 spin_lock(&info
->lock
);
1240 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1241 spin_unlock(&info
->lock
);
1243 sbinfo
= SHMEM_SB(inode
->i_sb
);
1244 if (sbinfo
->max_blocks
)
1245 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1247 shmem_unacct_blocks(info
->flags
, 1);
1249 if (swap
.val
&& error
!= -EINVAL
&&
1250 !shmem_confirm_swap(mapping
, index
, swap
))
1255 page_cache_release(page
);
1257 if (error
== -ENOSPC
&& !once
++) {
1258 info
= SHMEM_I(inode
);
1259 spin_lock(&info
->lock
);
1260 shmem_recalc_inode(inode
);
1261 spin_unlock(&info
->lock
);
1264 if (error
== -EEXIST
) /* from above or from radix_tree_insert */
1269 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1271 struct inode
*inode
= file_inode(vma
->vm_file
);
1273 int ret
= VM_FAULT_LOCKED
;
1276 * Trinity finds that probing a hole which tmpfs is punching can
1277 * prevent the hole-punch from ever completing: which in turn
1278 * locks writers out with its hold on i_mutex. So refrain from
1279 * faulting pages into the hole while it's being punched. Although
1280 * shmem_undo_range() does remove the additions, it may be unable to
1281 * keep up, as each new page needs its own unmap_mapping_range() call,
1282 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1284 * It does not matter if we sometimes reach this check just before the
1285 * hole-punch begins, so that one fault then races with the punch:
1286 * we just need to make racing faults a rare case.
1288 * The implementation below would be much simpler if we just used a
1289 * standard mutex or completion: but we cannot take i_mutex in fault,
1290 * and bloating every shmem inode for this unlikely case would be sad.
1292 if (unlikely(inode
->i_private
)) {
1293 struct shmem_falloc
*shmem_falloc
;
1295 spin_lock(&inode
->i_lock
);
1296 shmem_falloc
= inode
->i_private
;
1298 shmem_falloc
->waitq
&&
1299 vmf
->pgoff
>= shmem_falloc
->start
&&
1300 vmf
->pgoff
< shmem_falloc
->next
) {
1301 wait_queue_head_t
*shmem_falloc_waitq
;
1302 DEFINE_WAIT(shmem_fault_wait
);
1304 ret
= VM_FAULT_NOPAGE
;
1305 if ((vmf
->flags
& FAULT_FLAG_ALLOW_RETRY
) &&
1306 !(vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)) {
1307 /* It's polite to up mmap_sem if we can */
1308 up_read(&vma
->vm_mm
->mmap_sem
);
1309 ret
= VM_FAULT_RETRY
;
1312 shmem_falloc_waitq
= shmem_falloc
->waitq
;
1313 prepare_to_wait(shmem_falloc_waitq
, &shmem_fault_wait
,
1314 TASK_UNINTERRUPTIBLE
);
1315 spin_unlock(&inode
->i_lock
);
1319 * shmem_falloc_waitq points into the shmem_fallocate()
1320 * stack of the hole-punching task: shmem_falloc_waitq
1321 * is usually invalid by the time we reach here, but
1322 * finish_wait() does not dereference it in that case;
1323 * though i_lock needed lest racing with wake_up_all().
1325 spin_lock(&inode
->i_lock
);
1326 finish_wait(shmem_falloc_waitq
, &shmem_fault_wait
);
1327 spin_unlock(&inode
->i_lock
);
1330 spin_unlock(&inode
->i_lock
);
1333 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1335 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1337 if (ret
& VM_FAULT_MAJOR
) {
1338 count_vm_event(PGMAJFAULT
);
1339 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1345 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1347 struct inode
*inode
= file_inode(vma
->vm_file
);
1348 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1351 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1354 struct inode
*inode
= file_inode(vma
->vm_file
);
1357 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1358 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1362 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1364 struct inode
*inode
= file_inode(file
);
1365 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1366 int retval
= -ENOMEM
;
1368 spin_lock(&info
->lock
);
1369 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1370 if (!user_shm_lock(inode
->i_size
, user
))
1372 info
->flags
|= VM_LOCKED
;
1373 mapping_set_unevictable(file
->f_mapping
);
1375 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1376 user_shm_unlock(inode
->i_size
, user
);
1377 info
->flags
&= ~VM_LOCKED
;
1378 mapping_clear_unevictable(file
->f_mapping
);
1383 spin_unlock(&info
->lock
);
1387 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1389 file_accessed(file
);
1390 vma
->vm_ops
= &shmem_vm_ops
;
1394 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1395 umode_t mode
, dev_t dev
, unsigned long flags
)
1397 struct inode
*inode
;
1398 struct shmem_inode_info
*info
;
1399 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1401 if (shmem_reserve_inode(sb
))
1404 inode
= new_inode(sb
);
1406 inode
->i_ino
= get_next_ino();
1407 inode_init_owner(inode
, dir
, mode
);
1408 inode
->i_blocks
= 0;
1409 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1410 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1411 inode
->i_generation
= get_seconds();
1412 info
= SHMEM_I(inode
);
1413 memset(info
, 0, (char *)inode
- (char *)info
);
1414 spin_lock_init(&info
->lock
);
1415 info
->flags
= flags
& VM_NORESERVE
;
1416 INIT_LIST_HEAD(&info
->swaplist
);
1417 simple_xattrs_init(&info
->xattrs
);
1418 cache_no_acl(inode
);
1420 switch (mode
& S_IFMT
) {
1422 inode
->i_op
= &shmem_special_inode_operations
;
1423 init_special_inode(inode
, mode
, dev
);
1426 inode
->i_mapping
->a_ops
= &shmem_aops
;
1427 inode
->i_op
= &shmem_inode_operations
;
1428 inode
->i_fop
= &shmem_file_operations
;
1429 mpol_shared_policy_init(&info
->policy
,
1430 shmem_get_sbmpol(sbinfo
));
1434 /* Some things misbehave if size == 0 on a directory */
1435 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1436 inode
->i_op
= &shmem_dir_inode_operations
;
1437 inode
->i_fop
= &simple_dir_operations
;
1441 * Must not load anything in the rbtree,
1442 * mpol_free_shared_policy will not be called.
1444 mpol_shared_policy_init(&info
->policy
, NULL
);
1448 shmem_free_inode(sb
);
1452 bool shmem_mapping(struct address_space
*mapping
)
1454 return mapping
->backing_dev_info
== &shmem_backing_dev_info
;
1458 static const struct inode_operations shmem_symlink_inode_operations
;
1459 static const struct inode_operations shmem_short_symlink_operations
;
1461 #ifdef CONFIG_TMPFS_XATTR
1462 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1464 #define shmem_initxattrs NULL
1468 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1469 loff_t pos
, unsigned len
, unsigned flags
,
1470 struct page
**pagep
, void **fsdata
)
1472 struct inode
*inode
= mapping
->host
;
1473 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1474 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1478 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1479 loff_t pos
, unsigned len
, unsigned copied
,
1480 struct page
*page
, void *fsdata
)
1482 struct inode
*inode
= mapping
->host
;
1484 if (pos
+ copied
> inode
->i_size
)
1485 i_size_write(inode
, pos
+ copied
);
1487 if (!PageUptodate(page
)) {
1488 if (copied
< PAGE_CACHE_SIZE
) {
1489 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1490 zero_user_segments(page
, 0, from
,
1491 from
+ copied
, PAGE_CACHE_SIZE
);
1493 SetPageUptodate(page
);
1495 set_page_dirty(page
);
1497 page_cache_release(page
);
1502 static ssize_t
shmem_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1504 struct file
*file
= iocb
->ki_filp
;
1505 struct inode
*inode
= file_inode(file
);
1506 struct address_space
*mapping
= inode
->i_mapping
;
1508 unsigned long offset
;
1509 enum sgp_type sgp
= SGP_READ
;
1512 loff_t
*ppos
= &iocb
->ki_pos
;
1515 * Might this read be for a stacking filesystem? Then when reading
1516 * holes of a sparse file, we actually need to allocate those pages,
1517 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1519 if (segment_eq(get_fs(), KERNEL_DS
))
1522 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1523 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1526 struct page
*page
= NULL
;
1528 unsigned long nr
, ret
;
1529 loff_t i_size
= i_size_read(inode
);
1531 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1532 if (index
> end_index
)
1534 if (index
== end_index
) {
1535 nr
= i_size
& ~PAGE_CACHE_MASK
;
1540 error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1542 if (error
== -EINVAL
)
1550 * We must evaluate after, since reads (unlike writes)
1551 * are called without i_mutex protection against truncate
1553 nr
= PAGE_CACHE_SIZE
;
1554 i_size
= i_size_read(inode
);
1555 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1556 if (index
== end_index
) {
1557 nr
= i_size
& ~PAGE_CACHE_MASK
;
1560 page_cache_release(page
);
1568 * If users can be writing to this page using arbitrary
1569 * virtual addresses, take care about potential aliasing
1570 * before reading the page on the kernel side.
1572 if (mapping_writably_mapped(mapping
))
1573 flush_dcache_page(page
);
1575 * Mark the page accessed if we read the beginning.
1578 mark_page_accessed(page
);
1580 page
= ZERO_PAGE(0);
1581 page_cache_get(page
);
1585 * Ok, we have the page, and it's up-to-date, so
1586 * now we can copy it to user space...
1588 ret
= copy_page_to_iter(page
, offset
, nr
, to
);
1591 index
+= offset
>> PAGE_CACHE_SHIFT
;
1592 offset
&= ~PAGE_CACHE_MASK
;
1594 page_cache_release(page
);
1595 if (!iov_iter_count(to
))
1604 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1605 file_accessed(file
);
1606 return retval
? retval
: error
;
1609 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1610 struct pipe_inode_info
*pipe
, size_t len
,
1613 struct address_space
*mapping
= in
->f_mapping
;
1614 struct inode
*inode
= mapping
->host
;
1615 unsigned int loff
, nr_pages
, req_pages
;
1616 struct page
*pages
[PIPE_DEF_BUFFERS
];
1617 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1619 pgoff_t index
, end_index
;
1622 struct splice_pipe_desc spd
= {
1625 .nr_pages_max
= PIPE_DEF_BUFFERS
,
1627 .ops
= &page_cache_pipe_buf_ops
,
1628 .spd_release
= spd_release_page
,
1631 isize
= i_size_read(inode
);
1632 if (unlikely(*ppos
>= isize
))
1635 left
= isize
- *ppos
;
1636 if (unlikely(left
< len
))
1639 if (splice_grow_spd(pipe
, &spd
))
1642 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1643 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1644 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1645 nr_pages
= min(req_pages
, spd
.nr_pages_max
);
1647 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1648 nr_pages
, spd
.pages
);
1649 index
+= spd
.nr_pages
;
1652 while (spd
.nr_pages
< nr_pages
) {
1653 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1657 spd
.pages
[spd
.nr_pages
++] = page
;
1661 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1662 nr_pages
= spd
.nr_pages
;
1665 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1666 unsigned int this_len
;
1671 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1672 page
= spd
.pages
[page_nr
];
1674 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1675 error
= shmem_getpage(inode
, index
, &page
,
1680 page_cache_release(spd
.pages
[page_nr
]);
1681 spd
.pages
[page_nr
] = page
;
1684 isize
= i_size_read(inode
);
1685 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1686 if (unlikely(!isize
|| index
> end_index
))
1689 if (end_index
== index
) {
1692 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1696 this_len
= min(this_len
, plen
- loff
);
1700 spd
.partial
[page_nr
].offset
= loff
;
1701 spd
.partial
[page_nr
].len
= this_len
;
1708 while (page_nr
< nr_pages
)
1709 page_cache_release(spd
.pages
[page_nr
++]);
1712 error
= splice_to_pipe(pipe
, &spd
);
1714 splice_shrink_spd(&spd
);
1724 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1726 static pgoff_t
shmem_seek_hole_data(struct address_space
*mapping
,
1727 pgoff_t index
, pgoff_t end
, int whence
)
1730 struct pagevec pvec
;
1731 pgoff_t indices
[PAGEVEC_SIZE
];
1735 pagevec_init(&pvec
, 0);
1736 pvec
.nr
= 1; /* start small: we may be there already */
1738 pvec
.nr
= find_get_entries(mapping
, index
,
1739 pvec
.nr
, pvec
.pages
, indices
);
1741 if (whence
== SEEK_DATA
)
1745 for (i
= 0; i
< pvec
.nr
; i
++, index
++) {
1746 if (index
< indices
[i
]) {
1747 if (whence
== SEEK_HOLE
) {
1753 page
= pvec
.pages
[i
];
1754 if (page
&& !radix_tree_exceptional_entry(page
)) {
1755 if (!PageUptodate(page
))
1759 (page
&& whence
== SEEK_DATA
) ||
1760 (!page
&& whence
== SEEK_HOLE
)) {
1765 pagevec_remove_exceptionals(&pvec
);
1766 pagevec_release(&pvec
);
1767 pvec
.nr
= PAGEVEC_SIZE
;
1773 static loff_t
shmem_file_llseek(struct file
*file
, loff_t offset
, int whence
)
1775 struct address_space
*mapping
= file
->f_mapping
;
1776 struct inode
*inode
= mapping
->host
;
1780 if (whence
!= SEEK_DATA
&& whence
!= SEEK_HOLE
)
1781 return generic_file_llseek_size(file
, offset
, whence
,
1782 MAX_LFS_FILESIZE
, i_size_read(inode
));
1783 mutex_lock(&inode
->i_mutex
);
1784 /* We're holding i_mutex so we can access i_size directly */
1788 else if (offset
>= inode
->i_size
)
1791 start
= offset
>> PAGE_CACHE_SHIFT
;
1792 end
= (inode
->i_size
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1793 new_offset
= shmem_seek_hole_data(mapping
, start
, end
, whence
);
1794 new_offset
<<= PAGE_CACHE_SHIFT
;
1795 if (new_offset
> offset
) {
1796 if (new_offset
< inode
->i_size
)
1797 offset
= new_offset
;
1798 else if (whence
== SEEK_DATA
)
1801 offset
= inode
->i_size
;
1806 offset
= vfs_setpos(file
, offset
, MAX_LFS_FILESIZE
);
1807 mutex_unlock(&inode
->i_mutex
);
1811 static long shmem_fallocate(struct file
*file
, int mode
, loff_t offset
,
1814 struct inode
*inode
= file_inode(file
);
1815 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
1816 struct shmem_falloc shmem_falloc
;
1817 pgoff_t start
, index
, end
;
1820 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
1823 mutex_lock(&inode
->i_mutex
);
1825 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1826 struct address_space
*mapping
= file
->f_mapping
;
1827 loff_t unmap_start
= round_up(offset
, PAGE_SIZE
);
1828 loff_t unmap_end
= round_down(offset
+ len
, PAGE_SIZE
) - 1;
1829 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq
);
1831 shmem_falloc
.waitq
= &shmem_falloc_waitq
;
1832 shmem_falloc
.start
= unmap_start
>> PAGE_SHIFT
;
1833 shmem_falloc
.next
= (unmap_end
+ 1) >> PAGE_SHIFT
;
1834 spin_lock(&inode
->i_lock
);
1835 inode
->i_private
= &shmem_falloc
;
1836 spin_unlock(&inode
->i_lock
);
1838 if ((u64
)unmap_end
> (u64
)unmap_start
)
1839 unmap_mapping_range(mapping
, unmap_start
,
1840 1 + unmap_end
- unmap_start
, 0);
1841 shmem_truncate_range(inode
, offset
, offset
+ len
- 1);
1842 /* No need to unmap again: hole-punching leaves COWed pages */
1844 spin_lock(&inode
->i_lock
);
1845 inode
->i_private
= NULL
;
1846 wake_up_all(&shmem_falloc_waitq
);
1847 spin_unlock(&inode
->i_lock
);
1852 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1853 error
= inode_newsize_ok(inode
, offset
+ len
);
1857 start
= offset
>> PAGE_CACHE_SHIFT
;
1858 end
= (offset
+ len
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1859 /* Try to avoid a swapstorm if len is impossible to satisfy */
1860 if (sbinfo
->max_blocks
&& end
- start
> sbinfo
->max_blocks
) {
1865 shmem_falloc
.waitq
= NULL
;
1866 shmem_falloc
.start
= start
;
1867 shmem_falloc
.next
= start
;
1868 shmem_falloc
.nr_falloced
= 0;
1869 shmem_falloc
.nr_unswapped
= 0;
1870 spin_lock(&inode
->i_lock
);
1871 inode
->i_private
= &shmem_falloc
;
1872 spin_unlock(&inode
->i_lock
);
1874 for (index
= start
; index
< end
; index
++) {
1878 * Good, the fallocate(2) manpage permits EINTR: we may have
1879 * been interrupted because we are using up too much memory.
1881 if (signal_pending(current
))
1883 else if (shmem_falloc
.nr_unswapped
> shmem_falloc
.nr_falloced
)
1886 error
= shmem_getpage(inode
, index
, &page
, SGP_FALLOC
,
1889 /* Remove the !PageUptodate pages we added */
1890 shmem_undo_range(inode
,
1891 (loff_t
)start
<< PAGE_CACHE_SHIFT
,
1892 (loff_t
)index
<< PAGE_CACHE_SHIFT
, true);
1897 * Inform shmem_writepage() how far we have reached.
1898 * No need for lock or barrier: we have the page lock.
1900 shmem_falloc
.next
++;
1901 if (!PageUptodate(page
))
1902 shmem_falloc
.nr_falloced
++;
1905 * If !PageUptodate, leave it that way so that freeable pages
1906 * can be recognized if we need to rollback on error later.
1907 * But set_page_dirty so that memory pressure will swap rather
1908 * than free the pages we are allocating (and SGP_CACHE pages
1909 * might still be clean: we now need to mark those dirty too).
1911 set_page_dirty(page
);
1913 page_cache_release(page
);
1917 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
)
1918 i_size_write(inode
, offset
+ len
);
1919 inode
->i_ctime
= CURRENT_TIME
;
1921 spin_lock(&inode
->i_lock
);
1922 inode
->i_private
= NULL
;
1923 spin_unlock(&inode
->i_lock
);
1925 mutex_unlock(&inode
->i_mutex
);
1929 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1931 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
1933 buf
->f_type
= TMPFS_MAGIC
;
1934 buf
->f_bsize
= PAGE_CACHE_SIZE
;
1935 buf
->f_namelen
= NAME_MAX
;
1936 if (sbinfo
->max_blocks
) {
1937 buf
->f_blocks
= sbinfo
->max_blocks
;
1939 buf
->f_bfree
= sbinfo
->max_blocks
-
1940 percpu_counter_sum(&sbinfo
->used_blocks
);
1942 if (sbinfo
->max_inodes
) {
1943 buf
->f_files
= sbinfo
->max_inodes
;
1944 buf
->f_ffree
= sbinfo
->free_inodes
;
1946 /* else leave those fields 0 like simple_statfs */
1951 * File creation. Allocate an inode, and we're done..
1954 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
1956 struct inode
*inode
;
1957 int error
= -ENOSPC
;
1959 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
1961 error
= simple_acl_create(dir
, inode
);
1964 error
= security_inode_init_security(inode
, dir
,
1966 shmem_initxattrs
, NULL
);
1967 if (error
&& error
!= -EOPNOTSUPP
)
1971 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1972 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1973 d_instantiate(dentry
, inode
);
1974 dget(dentry
); /* Extra count - pin the dentry in core */
1983 shmem_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1985 struct inode
*inode
;
1986 int error
= -ENOSPC
;
1988 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, 0, VM_NORESERVE
);
1990 error
= security_inode_init_security(inode
, dir
,
1992 shmem_initxattrs
, NULL
);
1993 if (error
&& error
!= -EOPNOTSUPP
)
1995 error
= simple_acl_create(dir
, inode
);
1998 d_tmpfile(dentry
, inode
);
2006 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
2010 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
2016 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2019 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
2025 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
2027 struct inode
*inode
= old_dentry
->d_inode
;
2031 * No ordinary (disk based) filesystem counts links as inodes;
2032 * but each new link needs a new dentry, pinning lowmem, and
2033 * tmpfs dentries cannot be pruned until they are unlinked.
2035 ret
= shmem_reserve_inode(inode
->i_sb
);
2039 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2040 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2042 ihold(inode
); /* New dentry reference */
2043 dget(dentry
); /* Extra pinning count for the created dentry */
2044 d_instantiate(dentry
, inode
);
2049 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
2051 struct inode
*inode
= dentry
->d_inode
;
2053 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
2054 shmem_free_inode(inode
->i_sb
);
2056 dir
->i_size
-= BOGO_DIRENT_SIZE
;
2057 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2059 dput(dentry
); /* Undo the count from "create" - this does all the work */
2063 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2065 if (!simple_empty(dentry
))
2068 drop_nlink(dentry
->d_inode
);
2070 return shmem_unlink(dir
, dentry
);
2074 * The VFS layer already does all the dentry stuff for rename,
2075 * we just have to decrement the usage count for the target if
2076 * it exists so that the VFS layer correctly free's it when it
2079 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
2081 struct inode
*inode
= old_dentry
->d_inode
;
2082 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
2084 if (!simple_empty(new_dentry
))
2087 if (new_dentry
->d_inode
) {
2088 (void) shmem_unlink(new_dir
, new_dentry
);
2090 drop_nlink(old_dir
);
2091 } else if (they_are_dirs
) {
2092 drop_nlink(old_dir
);
2096 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
2097 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
2098 old_dir
->i_ctime
= old_dir
->i_mtime
=
2099 new_dir
->i_ctime
= new_dir
->i_mtime
=
2100 inode
->i_ctime
= CURRENT_TIME
;
2104 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
2108 struct inode
*inode
;
2111 struct shmem_inode_info
*info
;
2113 len
= strlen(symname
) + 1;
2114 if (len
> PAGE_CACHE_SIZE
)
2115 return -ENAMETOOLONG
;
2117 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
2121 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
2122 shmem_initxattrs
, NULL
);
2124 if (error
!= -EOPNOTSUPP
) {
2131 info
= SHMEM_I(inode
);
2132 inode
->i_size
= len
-1;
2133 if (len
<= SHORT_SYMLINK_LEN
) {
2134 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
2135 if (!info
->symlink
) {
2139 inode
->i_op
= &shmem_short_symlink_operations
;
2141 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
2146 inode
->i_mapping
->a_ops
= &shmem_aops
;
2147 inode
->i_op
= &shmem_symlink_inode_operations
;
2148 kaddr
= kmap_atomic(page
);
2149 memcpy(kaddr
, symname
, len
);
2150 kunmap_atomic(kaddr
);
2151 SetPageUptodate(page
);
2152 set_page_dirty(page
);
2154 page_cache_release(page
);
2156 dir
->i_size
+= BOGO_DIRENT_SIZE
;
2157 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
2158 d_instantiate(dentry
, inode
);
2163 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
2165 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
2169 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
2171 struct page
*page
= NULL
;
2172 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
2173 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
2179 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
2181 if (!IS_ERR(nd_get_link(nd
))) {
2182 struct page
*page
= cookie
;
2184 mark_page_accessed(page
);
2185 page_cache_release(page
);
2189 #ifdef CONFIG_TMPFS_XATTR
2191 * Superblocks without xattr inode operations may get some security.* xattr
2192 * support from the LSM "for free". As soon as we have any other xattrs
2193 * like ACLs, we also need to implement the security.* handlers at
2194 * filesystem level, though.
2198 * Callback for security_inode_init_security() for acquiring xattrs.
2200 static int shmem_initxattrs(struct inode
*inode
,
2201 const struct xattr
*xattr_array
,
2204 struct shmem_inode_info
*info
= SHMEM_I(inode
);
2205 const struct xattr
*xattr
;
2206 struct simple_xattr
*new_xattr
;
2209 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
2210 new_xattr
= simple_xattr_alloc(xattr
->value
, xattr
->value_len
);
2214 len
= strlen(xattr
->name
) + 1;
2215 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
2217 if (!new_xattr
->name
) {
2222 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
2223 XATTR_SECURITY_PREFIX_LEN
);
2224 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
2227 simple_xattr_list_add(&info
->xattrs
, new_xattr
);
2233 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
2234 #ifdef CONFIG_TMPFS_POSIX_ACL
2235 &posix_acl_access_xattr_handler
,
2236 &posix_acl_default_xattr_handler
,
2241 static int shmem_xattr_validate(const char *name
)
2243 struct { const char *prefix
; size_t len
; } arr
[] = {
2244 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
2245 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
2249 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
2250 size_t preflen
= arr
[i
].len
;
2251 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
2260 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
2261 void *buffer
, size_t size
)
2263 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2267 * If this is a request for a synthetic attribute in the system.*
2268 * namespace use the generic infrastructure to resolve a handler
2269 * for it via sb->s_xattr.
2271 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2272 return generic_getxattr(dentry
, name
, buffer
, size
);
2274 err
= shmem_xattr_validate(name
);
2278 return simple_xattr_get(&info
->xattrs
, name
, buffer
, size
);
2281 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
2282 const void *value
, size_t size
, int flags
)
2284 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2288 * If this is a request for a synthetic attribute in the system.*
2289 * namespace use the generic infrastructure to resolve a handler
2290 * for it via sb->s_xattr.
2292 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2293 return generic_setxattr(dentry
, name
, value
, size
, flags
);
2295 err
= shmem_xattr_validate(name
);
2299 return simple_xattr_set(&info
->xattrs
, name
, value
, size
, flags
);
2302 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
2304 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2308 * If this is a request for a synthetic attribute in the system.*
2309 * namespace use the generic infrastructure to resolve a handler
2310 * for it via sb->s_xattr.
2312 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
2313 return generic_removexattr(dentry
, name
);
2315 err
= shmem_xattr_validate(name
);
2319 return simple_xattr_remove(&info
->xattrs
, name
);
2322 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
2324 struct shmem_inode_info
*info
= SHMEM_I(dentry
->d_inode
);
2325 return simple_xattr_list(&info
->xattrs
, buffer
, size
);
2327 #endif /* CONFIG_TMPFS_XATTR */
2329 static const struct inode_operations shmem_short_symlink_operations
= {
2330 .readlink
= generic_readlink
,
2331 .follow_link
= shmem_follow_short_symlink
,
2332 #ifdef CONFIG_TMPFS_XATTR
2333 .setxattr
= shmem_setxattr
,
2334 .getxattr
= shmem_getxattr
,
2335 .listxattr
= shmem_listxattr
,
2336 .removexattr
= shmem_removexattr
,
2340 static const struct inode_operations shmem_symlink_inode_operations
= {
2341 .readlink
= generic_readlink
,
2342 .follow_link
= shmem_follow_link
,
2343 .put_link
= shmem_put_link
,
2344 #ifdef CONFIG_TMPFS_XATTR
2345 .setxattr
= shmem_setxattr
,
2346 .getxattr
= shmem_getxattr
,
2347 .listxattr
= shmem_listxattr
,
2348 .removexattr
= shmem_removexattr
,
2352 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2354 return ERR_PTR(-ESTALE
);
2357 static int shmem_match(struct inode
*ino
, void *vfh
)
2361 inum
= (inum
<< 32) | fh
[1];
2362 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2365 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2366 struct fid
*fid
, int fh_len
, int fh_type
)
2368 struct inode
*inode
;
2369 struct dentry
*dentry
= NULL
;
2376 inum
= (inum
<< 32) | fid
->raw
[1];
2378 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2379 shmem_match
, fid
->raw
);
2381 dentry
= d_find_alias(inode
);
2388 static int shmem_encode_fh(struct inode
*inode
, __u32
*fh
, int *len
,
2389 struct inode
*parent
)
2393 return FILEID_INVALID
;
2396 if (inode_unhashed(inode
)) {
2397 /* Unfortunately insert_inode_hash is not idempotent,
2398 * so as we hash inodes here rather than at creation
2399 * time, we need a lock to ensure we only try
2402 static DEFINE_SPINLOCK(lock
);
2404 if (inode_unhashed(inode
))
2405 __insert_inode_hash(inode
,
2406 inode
->i_ino
+ inode
->i_generation
);
2410 fh
[0] = inode
->i_generation
;
2411 fh
[1] = inode
->i_ino
;
2412 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2418 static const struct export_operations shmem_export_ops
= {
2419 .get_parent
= shmem_get_parent
,
2420 .encode_fh
= shmem_encode_fh
,
2421 .fh_to_dentry
= shmem_fh_to_dentry
,
2424 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2427 char *this_char
, *value
, *rest
;
2428 struct mempolicy
*mpol
= NULL
;
2432 while (options
!= NULL
) {
2433 this_char
= options
;
2436 * NUL-terminate this option: unfortunately,
2437 * mount options form a comma-separated list,
2438 * but mpol's nodelist may also contain commas.
2440 options
= strchr(options
, ',');
2441 if (options
== NULL
)
2444 if (!isdigit(*options
)) {
2451 if ((value
= strchr(this_char
,'=')) != NULL
) {
2455 "tmpfs: No value for mount option '%s'\n",
2460 if (!strcmp(this_char
,"size")) {
2461 unsigned long long size
;
2462 size
= memparse(value
,&rest
);
2464 size
<<= PAGE_SHIFT
;
2465 size
*= totalram_pages
;
2471 sbinfo
->max_blocks
=
2472 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2473 } else if (!strcmp(this_char
,"nr_blocks")) {
2474 sbinfo
->max_blocks
= memparse(value
, &rest
);
2477 } else if (!strcmp(this_char
,"nr_inodes")) {
2478 sbinfo
->max_inodes
= memparse(value
, &rest
);
2481 } else if (!strcmp(this_char
,"mode")) {
2484 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2487 } else if (!strcmp(this_char
,"uid")) {
2490 uid
= simple_strtoul(value
, &rest
, 0);
2493 sbinfo
->uid
= make_kuid(current_user_ns(), uid
);
2494 if (!uid_valid(sbinfo
->uid
))
2496 } else if (!strcmp(this_char
,"gid")) {
2499 gid
= simple_strtoul(value
, &rest
, 0);
2502 sbinfo
->gid
= make_kgid(current_user_ns(), gid
);
2503 if (!gid_valid(sbinfo
->gid
))
2505 } else if (!strcmp(this_char
,"mpol")) {
2508 if (mpol_parse_str(value
, &mpol
))
2511 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2516 sbinfo
->mpol
= mpol
;
2520 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2528 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2530 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2531 struct shmem_sb_info config
= *sbinfo
;
2532 unsigned long inodes
;
2533 int error
= -EINVAL
;
2536 if (shmem_parse_options(data
, &config
, true))
2539 spin_lock(&sbinfo
->stat_lock
);
2540 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2541 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2543 if (config
.max_inodes
< inodes
)
2546 * Those tests disallow limited->unlimited while any are in use;
2547 * but we must separately disallow unlimited->limited, because
2548 * in that case we have no record of how much is already in use.
2550 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2552 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2556 sbinfo
->max_blocks
= config
.max_blocks
;
2557 sbinfo
->max_inodes
= config
.max_inodes
;
2558 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2561 * Preserve previous mempolicy unless mpol remount option was specified.
2564 mpol_put(sbinfo
->mpol
);
2565 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2568 spin_unlock(&sbinfo
->stat_lock
);
2572 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2574 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2576 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2577 seq_printf(seq
, ",size=%luk",
2578 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2579 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2580 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2581 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2582 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2583 if (!uid_eq(sbinfo
->uid
, GLOBAL_ROOT_UID
))
2584 seq_printf(seq
, ",uid=%u",
2585 from_kuid_munged(&init_user_ns
, sbinfo
->uid
));
2586 if (!gid_eq(sbinfo
->gid
, GLOBAL_ROOT_GID
))
2587 seq_printf(seq
, ",gid=%u",
2588 from_kgid_munged(&init_user_ns
, sbinfo
->gid
));
2589 shmem_show_mpol(seq
, sbinfo
->mpol
);
2592 #endif /* CONFIG_TMPFS */
2594 static void shmem_put_super(struct super_block
*sb
)
2596 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2598 percpu_counter_destroy(&sbinfo
->used_blocks
);
2599 mpol_put(sbinfo
->mpol
);
2601 sb
->s_fs_info
= NULL
;
2604 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2606 struct inode
*inode
;
2607 struct shmem_sb_info
*sbinfo
;
2610 /* Round up to L1_CACHE_BYTES to resist false sharing */
2611 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2612 L1_CACHE_BYTES
), GFP_KERNEL
);
2616 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2617 sbinfo
->uid
= current_fsuid();
2618 sbinfo
->gid
= current_fsgid();
2619 sb
->s_fs_info
= sbinfo
;
2623 * Per default we only allow half of the physical ram per
2624 * tmpfs instance, limiting inodes to one per page of lowmem;
2625 * but the internal instance is left unlimited.
2627 if (!(sb
->s_flags
& MS_KERNMOUNT
)) {
2628 sbinfo
->max_blocks
= shmem_default_max_blocks();
2629 sbinfo
->max_inodes
= shmem_default_max_inodes();
2630 if (shmem_parse_options(data
, sbinfo
, false)) {
2635 sb
->s_flags
|= MS_NOUSER
;
2637 sb
->s_export_op
= &shmem_export_ops
;
2638 sb
->s_flags
|= MS_NOSEC
;
2640 sb
->s_flags
|= MS_NOUSER
;
2643 spin_lock_init(&sbinfo
->stat_lock
);
2644 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
2646 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2648 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
2649 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2650 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2651 sb
->s_magic
= TMPFS_MAGIC
;
2652 sb
->s_op
= &shmem_ops
;
2653 sb
->s_time_gran
= 1;
2654 #ifdef CONFIG_TMPFS_XATTR
2655 sb
->s_xattr
= shmem_xattr_handlers
;
2657 #ifdef CONFIG_TMPFS_POSIX_ACL
2658 sb
->s_flags
|= MS_POSIXACL
;
2661 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
2664 inode
->i_uid
= sbinfo
->uid
;
2665 inode
->i_gid
= sbinfo
->gid
;
2666 sb
->s_root
= d_make_root(inode
);
2672 shmem_put_super(sb
);
2676 static struct kmem_cache
*shmem_inode_cachep
;
2678 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2680 struct shmem_inode_info
*info
;
2681 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2684 return &info
->vfs_inode
;
2687 static void shmem_destroy_callback(struct rcu_head
*head
)
2689 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
2690 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2693 static void shmem_destroy_inode(struct inode
*inode
)
2695 if (S_ISREG(inode
->i_mode
))
2696 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2697 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
2700 static void shmem_init_inode(void *foo
)
2702 struct shmem_inode_info
*info
= foo
;
2703 inode_init_once(&info
->vfs_inode
);
2706 static int shmem_init_inodecache(void)
2708 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2709 sizeof(struct shmem_inode_info
),
2710 0, SLAB_PANIC
, shmem_init_inode
);
2714 static void shmem_destroy_inodecache(void)
2716 kmem_cache_destroy(shmem_inode_cachep
);
2719 static const struct address_space_operations shmem_aops
= {
2720 .writepage
= shmem_writepage
,
2721 .set_page_dirty
= __set_page_dirty_no_writeback
,
2723 .write_begin
= shmem_write_begin
,
2724 .write_end
= shmem_write_end
,
2726 .migratepage
= migrate_page
,
2727 .error_remove_page
= generic_error_remove_page
,
2730 static const struct file_operations shmem_file_operations
= {
2733 .llseek
= shmem_file_llseek
,
2734 .read
= new_sync_read
,
2735 .write
= new_sync_write
,
2736 .read_iter
= shmem_file_read_iter
,
2737 .write_iter
= generic_file_write_iter
,
2738 .fsync
= noop_fsync
,
2739 .splice_read
= shmem_file_splice_read
,
2740 .splice_write
= iter_file_splice_write
,
2741 .fallocate
= shmem_fallocate
,
2745 static const struct inode_operations shmem_inode_operations
= {
2746 .setattr
= shmem_setattr
,
2747 #ifdef CONFIG_TMPFS_XATTR
2748 .setxattr
= shmem_setxattr
,
2749 .getxattr
= shmem_getxattr
,
2750 .listxattr
= shmem_listxattr
,
2751 .removexattr
= shmem_removexattr
,
2752 .set_acl
= simple_set_acl
,
2756 static const struct inode_operations shmem_dir_inode_operations
= {
2758 .create
= shmem_create
,
2759 .lookup
= simple_lookup
,
2761 .unlink
= shmem_unlink
,
2762 .symlink
= shmem_symlink
,
2763 .mkdir
= shmem_mkdir
,
2764 .rmdir
= shmem_rmdir
,
2765 .mknod
= shmem_mknod
,
2766 .rename
= shmem_rename
,
2767 .tmpfile
= shmem_tmpfile
,
2769 #ifdef CONFIG_TMPFS_XATTR
2770 .setxattr
= shmem_setxattr
,
2771 .getxattr
= shmem_getxattr
,
2772 .listxattr
= shmem_listxattr
,
2773 .removexattr
= shmem_removexattr
,
2775 #ifdef CONFIG_TMPFS_POSIX_ACL
2776 .setattr
= shmem_setattr
,
2777 .set_acl
= simple_set_acl
,
2781 static const struct inode_operations shmem_special_inode_operations
= {
2782 #ifdef CONFIG_TMPFS_XATTR
2783 .setxattr
= shmem_setxattr
,
2784 .getxattr
= shmem_getxattr
,
2785 .listxattr
= shmem_listxattr
,
2786 .removexattr
= shmem_removexattr
,
2788 #ifdef CONFIG_TMPFS_POSIX_ACL
2789 .setattr
= shmem_setattr
,
2790 .set_acl
= simple_set_acl
,
2794 static const struct super_operations shmem_ops
= {
2795 .alloc_inode
= shmem_alloc_inode
,
2796 .destroy_inode
= shmem_destroy_inode
,
2798 .statfs
= shmem_statfs
,
2799 .remount_fs
= shmem_remount_fs
,
2800 .show_options
= shmem_show_options
,
2802 .evict_inode
= shmem_evict_inode
,
2803 .drop_inode
= generic_delete_inode
,
2804 .put_super
= shmem_put_super
,
2807 static const struct vm_operations_struct shmem_vm_ops
= {
2808 .fault
= shmem_fault
,
2809 .map_pages
= filemap_map_pages
,
2811 .set_policy
= shmem_set_policy
,
2812 .get_policy
= shmem_get_policy
,
2814 .remap_pages
= generic_file_remap_pages
,
2817 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
2818 int flags
, const char *dev_name
, void *data
)
2820 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
2823 static struct file_system_type shmem_fs_type
= {
2824 .owner
= THIS_MODULE
,
2826 .mount
= shmem_mount
,
2827 .kill_sb
= kill_litter_super
,
2828 .fs_flags
= FS_USERNS_MOUNT
,
2831 int __init
shmem_init(void)
2835 /* If rootfs called this, don't re-init */
2836 if (shmem_inode_cachep
)
2839 error
= bdi_init(&shmem_backing_dev_info
);
2843 error
= shmem_init_inodecache();
2847 error
= register_filesystem(&shmem_fs_type
);
2849 printk(KERN_ERR
"Could not register tmpfs\n");
2853 shm_mnt
= kern_mount(&shmem_fs_type
);
2854 if (IS_ERR(shm_mnt
)) {
2855 error
= PTR_ERR(shm_mnt
);
2856 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2862 unregister_filesystem(&shmem_fs_type
);
2864 shmem_destroy_inodecache();
2866 bdi_destroy(&shmem_backing_dev_info
);
2868 shm_mnt
= ERR_PTR(error
);
2872 #else /* !CONFIG_SHMEM */
2875 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2877 * This is intended for small system where the benefits of the full
2878 * shmem code (swap-backed and resource-limited) are outweighed by
2879 * their complexity. On systems without swap this code should be
2880 * effectively equivalent, but much lighter weight.
2883 static struct file_system_type shmem_fs_type
= {
2885 .mount
= ramfs_mount
,
2886 .kill_sb
= kill_litter_super
,
2887 .fs_flags
= FS_USERNS_MOUNT
,
2890 int __init
shmem_init(void)
2892 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
2894 shm_mnt
= kern_mount(&shmem_fs_type
);
2895 BUG_ON(IS_ERR(shm_mnt
));
2900 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
2905 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2910 void shmem_unlock_mapping(struct address_space
*mapping
)
2914 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
2916 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
2918 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
2920 #define shmem_vm_ops generic_file_vm_ops
2921 #define shmem_file_operations ramfs_file_operations
2922 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2923 #define shmem_acct_size(flags, size) 0
2924 #define shmem_unacct_size(flags, size) do {} while (0)
2926 #endif /* CONFIG_SHMEM */
2930 static struct dentry_operations anon_ops
= {
2931 .d_dname
= simple_dname
2934 static struct file
*__shmem_file_setup(const char *name
, loff_t size
,
2935 unsigned long flags
, unsigned int i_flags
)
2938 struct inode
*inode
;
2940 struct super_block
*sb
;
2943 if (IS_ERR(shm_mnt
))
2944 return ERR_CAST(shm_mnt
);
2946 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
2947 return ERR_PTR(-EINVAL
);
2949 if (shmem_acct_size(flags
, size
))
2950 return ERR_PTR(-ENOMEM
);
2952 res
= ERR_PTR(-ENOMEM
);
2954 this.len
= strlen(name
);
2955 this.hash
= 0; /* will go */
2956 sb
= shm_mnt
->mnt_sb
;
2957 path
.mnt
= mntget(shm_mnt
);
2958 path
.dentry
= d_alloc_pseudo(sb
, &this);
2961 d_set_d_op(path
.dentry
, &anon_ops
);
2963 res
= ERR_PTR(-ENOSPC
);
2964 inode
= shmem_get_inode(sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
2968 inode
->i_flags
|= i_flags
;
2969 d_instantiate(path
.dentry
, inode
);
2970 inode
->i_size
= size
;
2971 clear_nlink(inode
); /* It is unlinked */
2972 res
= ERR_PTR(ramfs_nommu_expand_for_mapping(inode
, size
));
2976 res
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
2977 &shmem_file_operations
);
2984 shmem_unacct_size(flags
, size
);
2991 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2992 * kernel internal. There will be NO LSM permission checks against the
2993 * underlying inode. So users of this interface must do LSM checks at a
2994 * higher layer. The one user is the big_key implementation. LSM checks
2995 * are provided at the key level rather than the inode level.
2996 * @name: name for dentry (to be seen in /proc/<pid>/maps
2997 * @size: size to be set for the file
2998 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3000 struct file
*shmem_kernel_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3002 return __shmem_file_setup(name
, size
, flags
, S_PRIVATE
);
3006 * shmem_file_setup - get an unlinked file living in tmpfs
3007 * @name: name for dentry (to be seen in /proc/<pid>/maps
3008 * @size: size to be set for the file
3009 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3011 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
3013 return __shmem_file_setup(name
, size
, flags
, 0);
3015 EXPORT_SYMBOL_GPL(shmem_file_setup
);
3018 * shmem_zero_setup - setup a shared anonymous mapping
3019 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3021 int shmem_zero_setup(struct vm_area_struct
*vma
)
3024 loff_t size
= vma
->vm_end
- vma
->vm_start
;
3026 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
3028 return PTR_ERR(file
);
3032 vma
->vm_file
= file
;
3033 vma
->vm_ops
= &shmem_vm_ops
;
3038 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3039 * @mapping: the page's address_space
3040 * @index: the page index
3041 * @gfp: the page allocator flags to use if allocating
3043 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3044 * with any new page allocations done using the specified allocation flags.
3045 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3046 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3047 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3049 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3050 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3052 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
3053 pgoff_t index
, gfp_t gfp
)
3056 struct inode
*inode
= mapping
->host
;
3060 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
3061 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
3063 page
= ERR_PTR(error
);
3069 * The tiny !SHMEM case uses ramfs without swap
3071 return read_cache_page_gfp(mapping
, index
, gfp
);
3074 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);