2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/pagemap.h>
29 #include <linux/file.h>
31 #include <linux/export.h>
32 #include <linux/swap.h>
34 static struct vfsmount
*shm_mnt
;
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
43 #include <linux/xattr.h>
44 #include <linux/exportfs.h>
45 #include <linux/posix_acl.h>
46 #include <linux/generic_acl.h>
47 #include <linux/mman.h>
48 #include <linux/string.h>
49 #include <linux/slab.h>
50 #include <linux/backing-dev.h>
51 #include <linux/shmem_fs.h>
52 #include <linux/writeback.h>
53 #include <linux/blkdev.h>
54 #include <linux/pagevec.h>
55 #include <linux/percpu_counter.h>
56 #include <linux/splice.h>
57 #include <linux/security.h>
58 #include <linux/swapops.h>
59 #include <linux/mempolicy.h>
60 #include <linux/namei.h>
61 #include <linux/ctype.h>
62 #include <linux/migrate.h>
63 #include <linux/highmem.h>
64 #include <linux/seq_file.h>
65 #include <linux/magic.h>
67 #include <asm/uaccess.h>
68 #include <asm/pgtable.h>
70 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
71 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
73 /* Pretend that each entry is of this size in directory's i_size */
74 #define BOGO_DIRENT_SIZE 20
76 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
77 #define SHORT_SYMLINK_LEN 128
80 struct list_head list
; /* anchored by shmem_inode_info->xattr_list */
81 char *name
; /* xattr name */
86 /* Flag allocation requirements to shmem_getpage */
88 SGP_READ
, /* don't exceed i_size, don't allocate page */
89 SGP_CACHE
, /* don't exceed i_size, may allocate page */
90 SGP_DIRTY
, /* like SGP_CACHE, but set new page dirty */
91 SGP_WRITE
, /* may exceed i_size, may allocate page */
95 static unsigned long shmem_default_max_blocks(void)
97 return totalram_pages
/ 2;
100 static unsigned long shmem_default_max_inodes(void)
102 return min(totalram_pages
- totalhigh_pages
, totalram_pages
/ 2);
106 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
107 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
);
109 static inline int shmem_getpage(struct inode
*inode
, pgoff_t index
,
110 struct page
**pagep
, enum sgp_type sgp
, int *fault_type
)
112 return shmem_getpage_gfp(inode
, index
, pagep
, sgp
,
113 mapping_gfp_mask(inode
->i_mapping
), fault_type
);
116 static inline struct shmem_sb_info
*SHMEM_SB(struct super_block
*sb
)
118 return sb
->s_fs_info
;
122 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
123 * for shared memory and for shared anonymous (/dev/zero) mappings
124 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
125 * consistent with the pre-accounting of private mappings ...
127 static inline int shmem_acct_size(unsigned long flags
, loff_t size
)
129 return (flags
& VM_NORESERVE
) ?
130 0 : security_vm_enough_memory_kern(VM_ACCT(size
));
133 static inline void shmem_unacct_size(unsigned long flags
, loff_t size
)
135 if (!(flags
& VM_NORESERVE
))
136 vm_unacct_memory(VM_ACCT(size
));
140 * ... whereas tmpfs objects are accounted incrementally as
141 * pages are allocated, in order to allow huge sparse files.
142 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
143 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
145 static inline int shmem_acct_block(unsigned long flags
)
147 return (flags
& VM_NORESERVE
) ?
148 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE
)) : 0;
151 static inline void shmem_unacct_blocks(unsigned long flags
, long pages
)
153 if (flags
& VM_NORESERVE
)
154 vm_unacct_memory(pages
* VM_ACCT(PAGE_CACHE_SIZE
));
157 static const struct super_operations shmem_ops
;
158 static const struct address_space_operations shmem_aops
;
159 static const struct file_operations shmem_file_operations
;
160 static const struct inode_operations shmem_inode_operations
;
161 static const struct inode_operations shmem_dir_inode_operations
;
162 static const struct inode_operations shmem_special_inode_operations
;
163 static const struct vm_operations_struct shmem_vm_ops
;
165 static struct backing_dev_info shmem_backing_dev_info __read_mostly
= {
166 .ra_pages
= 0, /* No readahead */
167 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
| BDI_CAP_SWAP_BACKED
,
170 static LIST_HEAD(shmem_swaplist
);
171 static DEFINE_MUTEX(shmem_swaplist_mutex
);
173 static int shmem_reserve_inode(struct super_block
*sb
)
175 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
176 if (sbinfo
->max_inodes
) {
177 spin_lock(&sbinfo
->stat_lock
);
178 if (!sbinfo
->free_inodes
) {
179 spin_unlock(&sbinfo
->stat_lock
);
182 sbinfo
->free_inodes
--;
183 spin_unlock(&sbinfo
->stat_lock
);
188 static void shmem_free_inode(struct super_block
*sb
)
190 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
191 if (sbinfo
->max_inodes
) {
192 spin_lock(&sbinfo
->stat_lock
);
193 sbinfo
->free_inodes
++;
194 spin_unlock(&sbinfo
->stat_lock
);
199 * shmem_recalc_inode - recalculate the block usage of an inode
200 * @inode: inode to recalc
202 * We have to calculate the free blocks since the mm can drop
203 * undirtied hole pages behind our back.
205 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
206 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
208 * It has to be called with the spinlock held.
210 static void shmem_recalc_inode(struct inode
*inode
)
212 struct shmem_inode_info
*info
= SHMEM_I(inode
);
215 freed
= info
->alloced
- info
->swapped
- inode
->i_mapping
->nrpages
;
217 struct shmem_sb_info
*sbinfo
= SHMEM_SB(inode
->i_sb
);
218 if (sbinfo
->max_blocks
)
219 percpu_counter_add(&sbinfo
->used_blocks
, -freed
);
220 info
->alloced
-= freed
;
221 inode
->i_blocks
-= freed
* BLOCKS_PER_PAGE
;
222 shmem_unacct_blocks(info
->flags
, freed
);
227 * Replace item expected in radix tree by a new item, while holding tree lock.
229 static int shmem_radix_tree_replace(struct address_space
*mapping
,
230 pgoff_t index
, void *expected
, void *replacement
)
235 VM_BUG_ON(!expected
);
236 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
, index
);
238 item
= radix_tree_deref_slot_protected(pslot
,
239 &mapping
->tree_lock
);
240 if (item
!= expected
)
243 radix_tree_replace_slot(pslot
, replacement
);
245 radix_tree_delete(&mapping
->page_tree
, index
);
250 * Like add_to_page_cache_locked, but error if expected item has gone.
252 static int shmem_add_to_page_cache(struct page
*page
,
253 struct address_space
*mapping
,
254 pgoff_t index
, gfp_t gfp
, void *expected
)
258 VM_BUG_ON(!PageLocked(page
));
259 VM_BUG_ON(!PageSwapBacked(page
));
262 error
= radix_tree_preload(gfp
& GFP_RECLAIM_MASK
);
264 page_cache_get(page
);
265 page
->mapping
= mapping
;
268 spin_lock_irq(&mapping
->tree_lock
);
270 error
= radix_tree_insert(&mapping
->page_tree
,
273 error
= shmem_radix_tree_replace(mapping
, index
,
277 __inc_zone_page_state(page
, NR_FILE_PAGES
);
278 __inc_zone_page_state(page
, NR_SHMEM
);
279 spin_unlock_irq(&mapping
->tree_lock
);
281 page
->mapping
= NULL
;
282 spin_unlock_irq(&mapping
->tree_lock
);
283 page_cache_release(page
);
286 radix_tree_preload_end();
289 mem_cgroup_uncharge_cache_page(page
);
294 * Like delete_from_page_cache, but substitutes swap for page.
296 static void shmem_delete_from_page_cache(struct page
*page
, void *radswap
)
298 struct address_space
*mapping
= page
->mapping
;
301 spin_lock_irq(&mapping
->tree_lock
);
302 error
= shmem_radix_tree_replace(mapping
, page
->index
, page
, radswap
);
303 page
->mapping
= NULL
;
305 __dec_zone_page_state(page
, NR_FILE_PAGES
);
306 __dec_zone_page_state(page
, NR_SHMEM
);
307 spin_unlock_irq(&mapping
->tree_lock
);
308 page_cache_release(page
);
313 * Like find_get_pages, but collecting swap entries as well as pages.
315 static unsigned shmem_find_get_pages_and_swap(struct address_space
*mapping
,
316 pgoff_t start
, unsigned int nr_pages
,
317 struct page
**pages
, pgoff_t
*indices
)
321 unsigned int nr_found
;
325 nr_found
= radix_tree_gang_lookup_slot(&mapping
->page_tree
,
326 (void ***)pages
, indices
, start
, nr_pages
);
328 for (i
= 0; i
< nr_found
; i
++) {
331 page
= radix_tree_deref_slot((void **)pages
[i
]);
334 if (radix_tree_exception(page
)) {
335 if (radix_tree_deref_retry(page
))
338 * Otherwise, we must be storing a swap entry
339 * here as an exceptional entry: so return it
340 * without attempting to raise page count.
344 if (!page_cache_get_speculative(page
))
347 /* Has the page moved? */
348 if (unlikely(page
!= *((void **)pages
[i
]))) {
349 page_cache_release(page
);
353 indices
[ret
] = indices
[i
];
357 if (unlikely(!ret
&& nr_found
))
364 * Remove swap entry from radix tree, free the swap and its page cache.
366 static int shmem_free_swap(struct address_space
*mapping
,
367 pgoff_t index
, void *radswap
)
371 spin_lock_irq(&mapping
->tree_lock
);
372 error
= shmem_radix_tree_replace(mapping
, index
, radswap
, NULL
);
373 spin_unlock_irq(&mapping
->tree_lock
);
375 free_swap_and_cache(radix_to_swp_entry(radswap
));
380 * Pagevec may contain swap entries, so shuffle up pages before releasing.
382 static void shmem_deswap_pagevec(struct pagevec
*pvec
)
386 for (i
= 0, j
= 0; i
< pagevec_count(pvec
); i
++) {
387 struct page
*page
= pvec
->pages
[i
];
388 if (!radix_tree_exceptional_entry(page
))
389 pvec
->pages
[j
++] = page
;
395 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
397 void shmem_unlock_mapping(struct address_space
*mapping
)
400 pgoff_t indices
[PAGEVEC_SIZE
];
403 pagevec_init(&pvec
, 0);
405 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
407 while (!mapping_unevictable(mapping
)) {
409 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
410 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
412 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
413 PAGEVEC_SIZE
, pvec
.pages
, indices
);
416 index
= indices
[pvec
.nr
- 1] + 1;
417 shmem_deswap_pagevec(&pvec
);
418 check_move_unevictable_pages(pvec
.pages
, pvec
.nr
);
419 pagevec_release(&pvec
);
425 * Remove range of pages and swap entries from radix tree, and free them.
427 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
429 struct address_space
*mapping
= inode
->i_mapping
;
430 struct shmem_inode_info
*info
= SHMEM_I(inode
);
431 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
432 unsigned partial
= lstart
& (PAGE_CACHE_SIZE
- 1);
433 pgoff_t end
= (lend
>> PAGE_CACHE_SHIFT
);
435 pgoff_t indices
[PAGEVEC_SIZE
];
436 long nr_swaps_freed
= 0;
440 BUG_ON((lend
& (PAGE_CACHE_SIZE
- 1)) != (PAGE_CACHE_SIZE
- 1));
442 pagevec_init(&pvec
, 0);
444 while (index
<= end
) {
445 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
446 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
447 pvec
.pages
, indices
);
450 mem_cgroup_uncharge_start();
451 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
452 struct page
*page
= pvec
.pages
[i
];
458 if (radix_tree_exceptional_entry(page
)) {
459 nr_swaps_freed
+= !shmem_free_swap(mapping
,
464 if (!trylock_page(page
))
466 if (page
->mapping
== mapping
) {
467 VM_BUG_ON(PageWriteback(page
));
468 truncate_inode_page(mapping
, page
);
472 shmem_deswap_pagevec(&pvec
);
473 pagevec_release(&pvec
);
474 mem_cgroup_uncharge_end();
480 struct page
*page
= NULL
;
481 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
483 zero_user_segment(page
, partial
, PAGE_CACHE_SIZE
);
484 set_page_dirty(page
);
486 page_cache_release(page
);
493 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
494 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
495 pvec
.pages
, indices
);
502 if (index
== start
&& indices
[0] > end
) {
503 shmem_deswap_pagevec(&pvec
);
504 pagevec_release(&pvec
);
507 mem_cgroup_uncharge_start();
508 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
509 struct page
*page
= pvec
.pages
[i
];
515 if (radix_tree_exceptional_entry(page
)) {
516 nr_swaps_freed
+= !shmem_free_swap(mapping
,
522 if (page
->mapping
== mapping
) {
523 VM_BUG_ON(PageWriteback(page
));
524 truncate_inode_page(mapping
, page
);
528 shmem_deswap_pagevec(&pvec
);
529 pagevec_release(&pvec
);
530 mem_cgroup_uncharge_end();
534 spin_lock(&info
->lock
);
535 info
->swapped
-= nr_swaps_freed
;
536 shmem_recalc_inode(inode
);
537 spin_unlock(&info
->lock
);
539 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
541 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
543 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
545 struct inode
*inode
= dentry
->d_inode
;
548 error
= inode_change_ok(inode
, attr
);
552 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
553 loff_t oldsize
= inode
->i_size
;
554 loff_t newsize
= attr
->ia_size
;
556 if (newsize
!= oldsize
) {
557 i_size_write(inode
, newsize
);
558 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
560 if (newsize
< oldsize
) {
561 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
562 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
563 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
564 /* unmap again to remove racily COWed private pages */
565 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
569 setattr_copy(inode
, attr
);
570 #ifdef CONFIG_TMPFS_POSIX_ACL
571 if (attr
->ia_valid
& ATTR_MODE
)
572 error
= generic_acl_chmod(inode
);
577 static void shmem_evict_inode(struct inode
*inode
)
579 struct shmem_inode_info
*info
= SHMEM_I(inode
);
580 struct shmem_xattr
*xattr
, *nxattr
;
582 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
583 shmem_unacct_size(info
->flags
, inode
->i_size
);
585 shmem_truncate_range(inode
, 0, (loff_t
)-1);
586 if (!list_empty(&info
->swaplist
)) {
587 mutex_lock(&shmem_swaplist_mutex
);
588 list_del_init(&info
->swaplist
);
589 mutex_unlock(&shmem_swaplist_mutex
);
592 kfree(info
->symlink
);
594 list_for_each_entry_safe(xattr
, nxattr
, &info
->xattr_list
, list
) {
598 BUG_ON(inode
->i_blocks
);
599 shmem_free_inode(inode
->i_sb
);
600 end_writeback(inode
);
604 * If swap found in inode, free it and move page from swapcache to filecache.
606 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
607 swp_entry_t swap
, struct page
*page
)
609 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
614 radswap
= swp_to_radix_entry(swap
);
615 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
620 * Move _head_ to start search for next from here.
621 * But be careful: shmem_evict_inode checks list_empty without taking
622 * mutex, and there's an instant in list_move_tail when info->swaplist
623 * would appear empty, if it were the only one on shmem_swaplist.
625 if (shmem_swaplist
.next
!= &info
->swaplist
)
626 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
629 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
630 * but also to hold up shmem_evict_inode(): so inode cannot be freed
631 * beneath us (pagelock doesn't help until the page is in pagecache).
633 error
= shmem_add_to_page_cache(page
, mapping
, index
,
634 GFP_NOWAIT
, radswap
);
635 /* which does mem_cgroup_uncharge_cache_page on error */
637 if (error
!= -ENOMEM
) {
639 * Truncation and eviction use free_swap_and_cache(), which
640 * only does trylock page: if we raced, best clean up here.
642 delete_from_swap_cache(page
);
643 set_page_dirty(page
);
645 spin_lock(&info
->lock
);
647 spin_unlock(&info
->lock
);
650 error
= 1; /* not an error, but entry was found */
656 * Search through swapped inodes to find and replace swap by page.
658 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
660 struct list_head
*this, *next
;
661 struct shmem_inode_info
*info
;
666 * Charge page using GFP_KERNEL while we can wait, before taking
667 * the shmem_swaplist_mutex which might hold up shmem_writepage().
668 * Charged back to the user (not to caller) when swap account is used.
670 error
= mem_cgroup_cache_charge(page
, current
->mm
, GFP_KERNEL
);
673 /* No radix_tree_preload: swap entry keeps a place for page in tree */
675 mutex_lock(&shmem_swaplist_mutex
);
676 list_for_each_safe(this, next
, &shmem_swaplist
) {
677 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
679 found
= shmem_unuse_inode(info
, swap
, page
);
681 list_del_init(&info
->swaplist
);
686 mutex_unlock(&shmem_swaplist_mutex
);
689 mem_cgroup_uncharge_cache_page(page
);
694 page_cache_release(page
);
699 * Move the page from the page cache to the swap cache.
701 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
703 struct shmem_inode_info
*info
;
704 struct address_space
*mapping
;
709 BUG_ON(!PageLocked(page
));
710 mapping
= page
->mapping
;
712 inode
= mapping
->host
;
713 info
= SHMEM_I(inode
);
714 if (info
->flags
& VM_LOCKED
)
716 if (!total_swap_pages
)
720 * shmem_backing_dev_info's capabilities prevent regular writeback or
721 * sync from ever calling shmem_writepage; but a stacking filesystem
722 * might use ->writepage of its underlying filesystem, in which case
723 * tmpfs should write out to swap only in response to memory pressure,
724 * and not for the writeback threads or sync.
726 if (!wbc
->for_reclaim
) {
727 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
730 swap
= get_swap_page();
735 * Add inode to shmem_unuse()'s list of swapped-out inodes,
736 * if it's not already there. Do it now before the page is
737 * moved to swap cache, when its pagelock no longer protects
738 * the inode from eviction. But don't unlock the mutex until
739 * we've incremented swapped, because shmem_unuse_inode() will
740 * prune a !swapped inode from the swaplist under this mutex.
742 mutex_lock(&shmem_swaplist_mutex
);
743 if (list_empty(&info
->swaplist
))
744 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
746 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
747 swap_shmem_alloc(swap
);
748 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
750 spin_lock(&info
->lock
);
752 shmem_recalc_inode(inode
);
753 spin_unlock(&info
->lock
);
755 mutex_unlock(&shmem_swaplist_mutex
);
756 BUG_ON(page_mapped(page
));
757 swap_writepage(page
, wbc
);
761 mutex_unlock(&shmem_swaplist_mutex
);
762 swapcache_free(swap
, NULL
);
764 set_page_dirty(page
);
765 if (wbc
->for_reclaim
)
766 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
773 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
777 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
778 return; /* show nothing */
780 mpol_to_str(buffer
, sizeof(buffer
), mpol
, 1);
782 seq_printf(seq
, ",mpol=%s", buffer
);
785 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
787 struct mempolicy
*mpol
= NULL
;
789 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
792 spin_unlock(&sbinfo
->stat_lock
);
796 #endif /* CONFIG_TMPFS */
798 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
799 struct shmem_inode_info
*info
, pgoff_t index
)
801 struct mempolicy mpol
, *spol
;
802 struct vm_area_struct pvma
;
804 spol
= mpol_cond_copy(&mpol
,
805 mpol_shared_policy_lookup(&info
->policy
, index
));
807 /* Create a pseudo vma that just contains the policy */
809 pvma
.vm_pgoff
= index
;
811 pvma
.vm_policy
= spol
;
812 return swapin_readahead(swap
, gfp
, &pvma
, 0);
815 static struct page
*shmem_alloc_page(gfp_t gfp
,
816 struct shmem_inode_info
*info
, pgoff_t index
)
818 struct vm_area_struct pvma
;
820 /* Create a pseudo vma that just contains the policy */
822 pvma
.vm_pgoff
= index
;
824 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
827 * alloc_page_vma() will drop the shared policy reference
829 return alloc_page_vma(gfp
, &pvma
, 0);
831 #else /* !CONFIG_NUMA */
833 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
836 #endif /* CONFIG_TMPFS */
838 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
839 struct shmem_inode_info
*info
, pgoff_t index
)
841 return swapin_readahead(swap
, gfp
, NULL
, 0);
844 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
845 struct shmem_inode_info
*info
, pgoff_t index
)
847 return alloc_page(gfp
);
849 #endif /* CONFIG_NUMA */
851 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
852 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
859 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
861 * If we allocate a new one we do not mark it dirty. That's up to the
862 * vm. If we swap it in we mark it dirty since we also free the swap
863 * entry since a page cannot live in both the swap and page cache
865 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
866 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
868 struct address_space
*mapping
= inode
->i_mapping
;
869 struct shmem_inode_info
*info
;
870 struct shmem_sb_info
*sbinfo
;
876 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
880 page
= find_lock_page(mapping
, index
);
881 if (radix_tree_exceptional_entry(page
)) {
882 swap
= radix_to_swp_entry(page
);
886 if (sgp
!= SGP_WRITE
&&
887 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
892 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
894 * Once we can get the page lock, it must be uptodate:
895 * if there were an error in reading back from swap,
896 * the page would not be inserted into the filecache.
898 BUG_ON(page
&& !PageUptodate(page
));
904 * Fast cache lookup did not find it:
905 * bring it back from swap or allocate.
907 info
= SHMEM_I(inode
);
908 sbinfo
= SHMEM_SB(inode
->i_sb
);
911 /* Look it up and read it in.. */
912 page
= lookup_swap_cache(swap
);
914 /* here we actually do the io */
916 *fault_type
|= VM_FAULT_MAJOR
;
917 page
= shmem_swapin(swap
, gfp
, info
, index
);
924 /* We have to do this with page locked to prevent races */
926 if (!PageUptodate(page
)) {
930 wait_on_page_writeback(page
);
932 /* Someone may have already done it for us */
934 if (page
->mapping
== mapping
&&
935 page
->index
== index
)
941 error
= mem_cgroup_cache_charge(page
, current
->mm
,
942 gfp
& GFP_RECLAIM_MASK
);
944 error
= shmem_add_to_page_cache(page
, mapping
, index
,
945 gfp
, swp_to_radix_entry(swap
));
949 spin_lock(&info
->lock
);
951 shmem_recalc_inode(inode
);
952 spin_unlock(&info
->lock
);
954 delete_from_swap_cache(page
);
955 set_page_dirty(page
);
959 if (shmem_acct_block(info
->flags
)) {
963 if (sbinfo
->max_blocks
) {
964 if (percpu_counter_compare(&sbinfo
->used_blocks
,
965 sbinfo
->max_blocks
) >= 0) {
969 percpu_counter_inc(&sbinfo
->used_blocks
);
972 page
= shmem_alloc_page(gfp
, info
, index
);
978 SetPageSwapBacked(page
);
979 __set_page_locked(page
);
980 error
= mem_cgroup_cache_charge(page
, current
->mm
,
981 gfp
& GFP_RECLAIM_MASK
);
983 error
= shmem_add_to_page_cache(page
, mapping
, index
,
987 lru_cache_add_anon(page
);
989 spin_lock(&info
->lock
);
991 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
992 shmem_recalc_inode(inode
);
993 spin_unlock(&info
->lock
);
995 clear_highpage(page
);
996 flush_dcache_page(page
);
997 SetPageUptodate(page
);
998 if (sgp
== SGP_DIRTY
)
999 set_page_dirty(page
);
1002 /* Perhaps the file has been truncated since we checked */
1003 if (sgp
!= SGP_WRITE
&&
1004 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
1015 ClearPageDirty(page
);
1016 delete_from_page_cache(page
);
1017 spin_lock(&info
->lock
);
1019 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
1020 spin_unlock(&info
->lock
);
1022 if (sbinfo
->max_blocks
)
1023 percpu_counter_add(&sbinfo
->used_blocks
, -1);
1025 shmem_unacct_blocks(info
->flags
, 1);
1027 if (swap
.val
&& error
!= -EINVAL
) {
1028 struct page
*test
= find_get_page(mapping
, index
);
1029 if (test
&& !radix_tree_exceptional_entry(test
))
1030 page_cache_release(test
);
1031 /* Have another try if the entry has changed */
1032 if (test
!= swp_to_radix_entry(swap
))
1037 page_cache_release(page
);
1039 if (error
== -ENOSPC
&& !once
++) {
1040 info
= SHMEM_I(inode
);
1041 spin_lock(&info
->lock
);
1042 shmem_recalc_inode(inode
);
1043 spin_unlock(&info
->lock
);
1046 if (error
== -EEXIST
)
1051 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1053 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1055 int ret
= VM_FAULT_LOCKED
;
1057 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1059 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1061 if (ret
& VM_FAULT_MAJOR
) {
1062 count_vm_event(PGMAJFAULT
);
1063 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1069 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1071 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1072 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1075 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1078 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1081 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1082 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1086 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1088 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1089 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1090 int retval
= -ENOMEM
;
1092 spin_lock(&info
->lock
);
1093 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1094 if (!user_shm_lock(inode
->i_size
, user
))
1096 info
->flags
|= VM_LOCKED
;
1097 mapping_set_unevictable(file
->f_mapping
);
1099 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1100 user_shm_unlock(inode
->i_size
, user
);
1101 info
->flags
&= ~VM_LOCKED
;
1102 mapping_clear_unevictable(file
->f_mapping
);
1107 spin_unlock(&info
->lock
);
1111 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1113 file_accessed(file
);
1114 vma
->vm_ops
= &shmem_vm_ops
;
1115 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1119 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1120 umode_t mode
, dev_t dev
, unsigned long flags
)
1122 struct inode
*inode
;
1123 struct shmem_inode_info
*info
;
1124 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1126 if (shmem_reserve_inode(sb
))
1129 inode
= new_inode(sb
);
1131 inode
->i_ino
= get_next_ino();
1132 inode_init_owner(inode
, dir
, mode
);
1133 inode
->i_blocks
= 0;
1134 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1135 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1136 inode
->i_generation
= get_seconds();
1137 info
= SHMEM_I(inode
);
1138 memset(info
, 0, (char *)inode
- (char *)info
);
1139 spin_lock_init(&info
->lock
);
1140 info
->flags
= flags
& VM_NORESERVE
;
1141 INIT_LIST_HEAD(&info
->swaplist
);
1142 INIT_LIST_HEAD(&info
->xattr_list
);
1143 cache_no_acl(inode
);
1145 switch (mode
& S_IFMT
) {
1147 inode
->i_op
= &shmem_special_inode_operations
;
1148 init_special_inode(inode
, mode
, dev
);
1151 inode
->i_mapping
->a_ops
= &shmem_aops
;
1152 inode
->i_op
= &shmem_inode_operations
;
1153 inode
->i_fop
= &shmem_file_operations
;
1154 mpol_shared_policy_init(&info
->policy
,
1155 shmem_get_sbmpol(sbinfo
));
1159 /* Some things misbehave if size == 0 on a directory */
1160 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1161 inode
->i_op
= &shmem_dir_inode_operations
;
1162 inode
->i_fop
= &simple_dir_operations
;
1166 * Must not load anything in the rbtree,
1167 * mpol_free_shared_policy will not be called.
1169 mpol_shared_policy_init(&info
->policy
, NULL
);
1173 shmem_free_inode(sb
);
1178 static const struct inode_operations shmem_symlink_inode_operations
;
1179 static const struct inode_operations shmem_short_symlink_operations
;
1181 #ifdef CONFIG_TMPFS_XATTR
1182 static int shmem_initxattrs(struct inode
*, const struct xattr
*, void *);
1184 #define shmem_initxattrs NULL
1188 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1189 loff_t pos
, unsigned len
, unsigned flags
,
1190 struct page
**pagep
, void **fsdata
)
1192 struct inode
*inode
= mapping
->host
;
1193 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1194 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1198 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1199 loff_t pos
, unsigned len
, unsigned copied
,
1200 struct page
*page
, void *fsdata
)
1202 struct inode
*inode
= mapping
->host
;
1204 if (pos
+ copied
> inode
->i_size
)
1205 i_size_write(inode
, pos
+ copied
);
1207 set_page_dirty(page
);
1209 page_cache_release(page
);
1214 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1216 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
1217 struct address_space
*mapping
= inode
->i_mapping
;
1219 unsigned long offset
;
1220 enum sgp_type sgp
= SGP_READ
;
1223 * Might this read be for a stacking filesystem? Then when reading
1224 * holes of a sparse file, we actually need to allocate those pages,
1225 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1227 if (segment_eq(get_fs(), KERNEL_DS
))
1230 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1231 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1234 struct page
*page
= NULL
;
1236 unsigned long nr
, ret
;
1237 loff_t i_size
= i_size_read(inode
);
1239 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1240 if (index
> end_index
)
1242 if (index
== end_index
) {
1243 nr
= i_size
& ~PAGE_CACHE_MASK
;
1248 desc
->error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1250 if (desc
->error
== -EINVAL
)
1258 * We must evaluate after, since reads (unlike writes)
1259 * are called without i_mutex protection against truncate
1261 nr
= PAGE_CACHE_SIZE
;
1262 i_size
= i_size_read(inode
);
1263 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1264 if (index
== end_index
) {
1265 nr
= i_size
& ~PAGE_CACHE_MASK
;
1268 page_cache_release(page
);
1276 * If users can be writing to this page using arbitrary
1277 * virtual addresses, take care about potential aliasing
1278 * before reading the page on the kernel side.
1280 if (mapping_writably_mapped(mapping
))
1281 flush_dcache_page(page
);
1283 * Mark the page accessed if we read the beginning.
1286 mark_page_accessed(page
);
1288 page
= ZERO_PAGE(0);
1289 page_cache_get(page
);
1293 * Ok, we have the page, and it's up-to-date, so
1294 * now we can copy it to user space...
1296 * The actor routine returns how many bytes were actually used..
1297 * NOTE! This may not be the same as how much of a user buffer
1298 * we filled up (we may be padding etc), so we can only update
1299 * "pos" here (the actor routine has to update the user buffer
1300 * pointers and the remaining count).
1302 ret
= actor(desc
, page
, offset
, nr
);
1304 index
+= offset
>> PAGE_CACHE_SHIFT
;
1305 offset
&= ~PAGE_CACHE_MASK
;
1307 page_cache_release(page
);
1308 if (ret
!= nr
|| !desc
->count
)
1314 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1315 file_accessed(filp
);
1318 static ssize_t
shmem_file_aio_read(struct kiocb
*iocb
,
1319 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
1321 struct file
*filp
= iocb
->ki_filp
;
1325 loff_t
*ppos
= &iocb
->ki_pos
;
1327 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1331 for (seg
= 0; seg
< nr_segs
; seg
++) {
1332 read_descriptor_t desc
;
1335 desc
.arg
.buf
= iov
[seg
].iov_base
;
1336 desc
.count
= iov
[seg
].iov_len
;
1337 if (desc
.count
== 0)
1340 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1341 retval
+= desc
.written
;
1343 retval
= retval
?: desc
.error
;
1352 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1353 struct pipe_inode_info
*pipe
, size_t len
,
1356 struct address_space
*mapping
= in
->f_mapping
;
1357 struct inode
*inode
= mapping
->host
;
1358 unsigned int loff
, nr_pages
, req_pages
;
1359 struct page
*pages
[PIPE_DEF_BUFFERS
];
1360 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1362 pgoff_t index
, end_index
;
1365 struct splice_pipe_desc spd
= {
1369 .ops
= &page_cache_pipe_buf_ops
,
1370 .spd_release
= spd_release_page
,
1373 isize
= i_size_read(inode
);
1374 if (unlikely(*ppos
>= isize
))
1377 left
= isize
- *ppos
;
1378 if (unlikely(left
< len
))
1381 if (splice_grow_spd(pipe
, &spd
))
1384 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1385 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1386 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1387 nr_pages
= min(req_pages
, pipe
->buffers
);
1389 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1390 nr_pages
, spd
.pages
);
1391 index
+= spd
.nr_pages
;
1394 while (spd
.nr_pages
< nr_pages
) {
1395 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1399 spd
.pages
[spd
.nr_pages
++] = page
;
1403 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1404 nr_pages
= spd
.nr_pages
;
1407 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1408 unsigned int this_len
;
1413 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1414 page
= spd
.pages
[page_nr
];
1416 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1417 error
= shmem_getpage(inode
, index
, &page
,
1422 page_cache_release(spd
.pages
[page_nr
]);
1423 spd
.pages
[page_nr
] = page
;
1426 isize
= i_size_read(inode
);
1427 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1428 if (unlikely(!isize
|| index
> end_index
))
1431 if (end_index
== index
) {
1434 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1438 this_len
= min(this_len
, plen
- loff
);
1442 spd
.partial
[page_nr
].offset
= loff
;
1443 spd
.partial
[page_nr
].len
= this_len
;
1450 while (page_nr
< nr_pages
)
1451 page_cache_release(spd
.pages
[page_nr
++]);
1454 error
= splice_to_pipe(pipe
, &spd
);
1456 splice_shrink_spd(pipe
, &spd
);
1465 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1467 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
1469 buf
->f_type
= TMPFS_MAGIC
;
1470 buf
->f_bsize
= PAGE_CACHE_SIZE
;
1471 buf
->f_namelen
= NAME_MAX
;
1472 if (sbinfo
->max_blocks
) {
1473 buf
->f_blocks
= sbinfo
->max_blocks
;
1475 buf
->f_bfree
= sbinfo
->max_blocks
-
1476 percpu_counter_sum(&sbinfo
->used_blocks
);
1478 if (sbinfo
->max_inodes
) {
1479 buf
->f_files
= sbinfo
->max_inodes
;
1480 buf
->f_ffree
= sbinfo
->free_inodes
;
1482 /* else leave those fields 0 like simple_statfs */
1487 * File creation. Allocate an inode, and we're done..
1490 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
1492 struct inode
*inode
;
1493 int error
= -ENOSPC
;
1495 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
1497 error
= security_inode_init_security(inode
, dir
,
1499 shmem_initxattrs
, NULL
);
1501 if (error
!= -EOPNOTSUPP
) {
1506 #ifdef CONFIG_TMPFS_POSIX_ACL
1507 error
= generic_acl_init(inode
, dir
);
1515 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1516 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1517 d_instantiate(dentry
, inode
);
1518 dget(dentry
); /* Extra count - pin the dentry in core */
1523 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
1527 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
1533 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
1534 struct nameidata
*nd
)
1536 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
1542 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1544 struct inode
*inode
= old_dentry
->d_inode
;
1548 * No ordinary (disk based) filesystem counts links as inodes;
1549 * but each new link needs a new dentry, pinning lowmem, and
1550 * tmpfs dentries cannot be pruned until they are unlinked.
1552 ret
= shmem_reserve_inode(inode
->i_sb
);
1556 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1557 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1559 ihold(inode
); /* New dentry reference */
1560 dget(dentry
); /* Extra pinning count for the created dentry */
1561 d_instantiate(dentry
, inode
);
1566 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
1568 struct inode
*inode
= dentry
->d_inode
;
1570 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
1571 shmem_free_inode(inode
->i_sb
);
1573 dir
->i_size
-= BOGO_DIRENT_SIZE
;
1574 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1576 dput(dentry
); /* Undo the count from "create" - this does all the work */
1580 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1582 if (!simple_empty(dentry
))
1585 drop_nlink(dentry
->d_inode
);
1587 return shmem_unlink(dir
, dentry
);
1591 * The VFS layer already does all the dentry stuff for rename,
1592 * we just have to decrement the usage count for the target if
1593 * it exists so that the VFS layer correctly free's it when it
1596 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
1598 struct inode
*inode
= old_dentry
->d_inode
;
1599 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
1601 if (!simple_empty(new_dentry
))
1604 if (new_dentry
->d_inode
) {
1605 (void) shmem_unlink(new_dir
, new_dentry
);
1607 drop_nlink(old_dir
);
1608 } else if (they_are_dirs
) {
1609 drop_nlink(old_dir
);
1613 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
1614 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
1615 old_dir
->i_ctime
= old_dir
->i_mtime
=
1616 new_dir
->i_ctime
= new_dir
->i_mtime
=
1617 inode
->i_ctime
= CURRENT_TIME
;
1621 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1625 struct inode
*inode
;
1628 struct shmem_inode_info
*info
;
1630 len
= strlen(symname
) + 1;
1631 if (len
> PAGE_CACHE_SIZE
)
1632 return -ENAMETOOLONG
;
1634 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
1638 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
,
1639 shmem_initxattrs
, NULL
);
1641 if (error
!= -EOPNOTSUPP
) {
1648 info
= SHMEM_I(inode
);
1649 inode
->i_size
= len
-1;
1650 if (len
<= SHORT_SYMLINK_LEN
) {
1651 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
1652 if (!info
->symlink
) {
1656 inode
->i_op
= &shmem_short_symlink_operations
;
1658 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
1663 inode
->i_mapping
->a_ops
= &shmem_aops
;
1664 inode
->i_op
= &shmem_symlink_inode_operations
;
1665 kaddr
= kmap_atomic(page
);
1666 memcpy(kaddr
, symname
, len
);
1667 kunmap_atomic(kaddr
);
1668 set_page_dirty(page
);
1670 page_cache_release(page
);
1672 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1673 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1674 d_instantiate(dentry
, inode
);
1679 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
1681 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
1685 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1687 struct page
*page
= NULL
;
1688 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
1689 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
1695 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
1697 if (!IS_ERR(nd_get_link(nd
))) {
1698 struct page
*page
= cookie
;
1700 mark_page_accessed(page
);
1701 page_cache_release(page
);
1705 #ifdef CONFIG_TMPFS_XATTR
1707 * Superblocks without xattr inode operations may get some security.* xattr
1708 * support from the LSM "for free". As soon as we have any other xattrs
1709 * like ACLs, we also need to implement the security.* handlers at
1710 * filesystem level, though.
1714 * Allocate new xattr and copy in the value; but leave the name to callers.
1716 static struct shmem_xattr
*shmem_xattr_alloc(const void *value
, size_t size
)
1718 struct shmem_xattr
*new_xattr
;
1722 len
= sizeof(*new_xattr
) + size
;
1723 if (len
<= sizeof(*new_xattr
))
1726 new_xattr
= kmalloc(len
, GFP_KERNEL
);
1730 new_xattr
->size
= size
;
1731 memcpy(new_xattr
->value
, value
, size
);
1736 * Callback for security_inode_init_security() for acquiring xattrs.
1738 static int shmem_initxattrs(struct inode
*inode
,
1739 const struct xattr
*xattr_array
,
1742 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1743 const struct xattr
*xattr
;
1744 struct shmem_xattr
*new_xattr
;
1747 for (xattr
= xattr_array
; xattr
->name
!= NULL
; xattr
++) {
1748 new_xattr
= shmem_xattr_alloc(xattr
->value
, xattr
->value_len
);
1752 len
= strlen(xattr
->name
) + 1;
1753 new_xattr
->name
= kmalloc(XATTR_SECURITY_PREFIX_LEN
+ len
,
1755 if (!new_xattr
->name
) {
1760 memcpy(new_xattr
->name
, XATTR_SECURITY_PREFIX
,
1761 XATTR_SECURITY_PREFIX_LEN
);
1762 memcpy(new_xattr
->name
+ XATTR_SECURITY_PREFIX_LEN
,
1765 spin_lock(&info
->lock
);
1766 list_add(&new_xattr
->list
, &info
->xattr_list
);
1767 spin_unlock(&info
->lock
);
1773 static int shmem_xattr_get(struct dentry
*dentry
, const char *name
,
1774 void *buffer
, size_t size
)
1776 struct shmem_inode_info
*info
;
1777 struct shmem_xattr
*xattr
;
1780 info
= SHMEM_I(dentry
->d_inode
);
1782 spin_lock(&info
->lock
);
1783 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
1784 if (strcmp(name
, xattr
->name
))
1789 if (size
< xattr
->size
)
1792 memcpy(buffer
, xattr
->value
, xattr
->size
);
1796 spin_unlock(&info
->lock
);
1800 static int shmem_xattr_set(struct inode
*inode
, const char *name
,
1801 const void *value
, size_t size
, int flags
)
1803 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1804 struct shmem_xattr
*xattr
;
1805 struct shmem_xattr
*new_xattr
= NULL
;
1808 /* value == NULL means remove */
1810 new_xattr
= shmem_xattr_alloc(value
, size
);
1814 new_xattr
->name
= kstrdup(name
, GFP_KERNEL
);
1815 if (!new_xattr
->name
) {
1821 spin_lock(&info
->lock
);
1822 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
1823 if (!strcmp(name
, xattr
->name
)) {
1824 if (flags
& XATTR_CREATE
) {
1827 } else if (new_xattr
) {
1828 list_replace(&xattr
->list
, &new_xattr
->list
);
1830 list_del(&xattr
->list
);
1835 if (flags
& XATTR_REPLACE
) {
1839 list_add(&new_xattr
->list
, &info
->xattr_list
);
1843 spin_unlock(&info
->lock
);
1850 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
1851 #ifdef CONFIG_TMPFS_POSIX_ACL
1852 &generic_acl_access_handler
,
1853 &generic_acl_default_handler
,
1858 static int shmem_xattr_validate(const char *name
)
1860 struct { const char *prefix
; size_t len
; } arr
[] = {
1861 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
1862 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
1866 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
1867 size_t preflen
= arr
[i
].len
;
1868 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
1877 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
1878 void *buffer
, size_t size
)
1883 * If this is a request for a synthetic attribute in the system.*
1884 * namespace use the generic infrastructure to resolve a handler
1885 * for it via sb->s_xattr.
1887 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
1888 return generic_getxattr(dentry
, name
, buffer
, size
);
1890 err
= shmem_xattr_validate(name
);
1894 return shmem_xattr_get(dentry
, name
, buffer
, size
);
1897 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
1898 const void *value
, size_t size
, int flags
)
1903 * If this is a request for a synthetic attribute in the system.*
1904 * namespace use the generic infrastructure to resolve a handler
1905 * for it via sb->s_xattr.
1907 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
1908 return generic_setxattr(dentry
, name
, value
, size
, flags
);
1910 err
= shmem_xattr_validate(name
);
1915 value
= ""; /* empty EA, do not remove */
1917 return shmem_xattr_set(dentry
->d_inode
, name
, value
, size
, flags
);
1921 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
1926 * If this is a request for a synthetic attribute in the system.*
1927 * namespace use the generic infrastructure to resolve a handler
1928 * for it via sb->s_xattr.
1930 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
1931 return generic_removexattr(dentry
, name
);
1933 err
= shmem_xattr_validate(name
);
1937 return shmem_xattr_set(dentry
->d_inode
, name
, NULL
, 0, XATTR_REPLACE
);
1940 static bool xattr_is_trusted(const char *name
)
1942 return !strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
);
1945 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
1947 bool trusted
= capable(CAP_SYS_ADMIN
);
1948 struct shmem_xattr
*xattr
;
1949 struct shmem_inode_info
*info
;
1952 info
= SHMEM_I(dentry
->d_inode
);
1954 spin_lock(&info
->lock
);
1955 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
1958 /* skip "trusted." attributes for unprivileged callers */
1959 if (!trusted
&& xattr_is_trusted(xattr
->name
))
1962 len
= strlen(xattr
->name
) + 1;
1969 memcpy(buffer
, xattr
->name
, len
);
1973 spin_unlock(&info
->lock
);
1977 #endif /* CONFIG_TMPFS_XATTR */
1979 static const struct inode_operations shmem_short_symlink_operations
= {
1980 .readlink
= generic_readlink
,
1981 .follow_link
= shmem_follow_short_symlink
,
1982 #ifdef CONFIG_TMPFS_XATTR
1983 .setxattr
= shmem_setxattr
,
1984 .getxattr
= shmem_getxattr
,
1985 .listxattr
= shmem_listxattr
,
1986 .removexattr
= shmem_removexattr
,
1990 static const struct inode_operations shmem_symlink_inode_operations
= {
1991 .readlink
= generic_readlink
,
1992 .follow_link
= shmem_follow_link
,
1993 .put_link
= shmem_put_link
,
1994 #ifdef CONFIG_TMPFS_XATTR
1995 .setxattr
= shmem_setxattr
,
1996 .getxattr
= shmem_getxattr
,
1997 .listxattr
= shmem_listxattr
,
1998 .removexattr
= shmem_removexattr
,
2002 static struct dentry
*shmem_get_parent(struct dentry
*child
)
2004 return ERR_PTR(-ESTALE
);
2007 static int shmem_match(struct inode
*ino
, void *vfh
)
2011 inum
= (inum
<< 32) | fh
[1];
2012 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
2015 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
2016 struct fid
*fid
, int fh_len
, int fh_type
)
2018 struct inode
*inode
;
2019 struct dentry
*dentry
= NULL
;
2020 u64 inum
= fid
->raw
[2];
2021 inum
= (inum
<< 32) | fid
->raw
[1];
2026 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
2027 shmem_match
, fid
->raw
);
2029 dentry
= d_find_alias(inode
);
2036 static int shmem_encode_fh(struct dentry
*dentry
, __u32
*fh
, int *len
,
2039 struct inode
*inode
= dentry
->d_inode
;
2046 if (inode_unhashed(inode
)) {
2047 /* Unfortunately insert_inode_hash is not idempotent,
2048 * so as we hash inodes here rather than at creation
2049 * time, we need a lock to ensure we only try
2052 static DEFINE_SPINLOCK(lock
);
2054 if (inode_unhashed(inode
))
2055 __insert_inode_hash(inode
,
2056 inode
->i_ino
+ inode
->i_generation
);
2060 fh
[0] = inode
->i_generation
;
2061 fh
[1] = inode
->i_ino
;
2062 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
2068 static const struct export_operations shmem_export_ops
= {
2069 .get_parent
= shmem_get_parent
,
2070 .encode_fh
= shmem_encode_fh
,
2071 .fh_to_dentry
= shmem_fh_to_dentry
,
2074 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
2077 char *this_char
, *value
, *rest
;
2079 while (options
!= NULL
) {
2080 this_char
= options
;
2083 * NUL-terminate this option: unfortunately,
2084 * mount options form a comma-separated list,
2085 * but mpol's nodelist may also contain commas.
2087 options
= strchr(options
, ',');
2088 if (options
== NULL
)
2091 if (!isdigit(*options
)) {
2098 if ((value
= strchr(this_char
,'=')) != NULL
) {
2102 "tmpfs: No value for mount option '%s'\n",
2107 if (!strcmp(this_char
,"size")) {
2108 unsigned long long size
;
2109 size
= memparse(value
,&rest
);
2111 size
<<= PAGE_SHIFT
;
2112 size
*= totalram_pages
;
2118 sbinfo
->max_blocks
=
2119 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2120 } else if (!strcmp(this_char
,"nr_blocks")) {
2121 sbinfo
->max_blocks
= memparse(value
, &rest
);
2124 } else if (!strcmp(this_char
,"nr_inodes")) {
2125 sbinfo
->max_inodes
= memparse(value
, &rest
);
2128 } else if (!strcmp(this_char
,"mode")) {
2131 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2134 } else if (!strcmp(this_char
,"uid")) {
2137 sbinfo
->uid
= simple_strtoul(value
, &rest
, 0);
2140 } else if (!strcmp(this_char
,"gid")) {
2143 sbinfo
->gid
= simple_strtoul(value
, &rest
, 0);
2146 } else if (!strcmp(this_char
,"mpol")) {
2147 if (mpol_parse_str(value
, &sbinfo
->mpol
, 1))
2150 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2158 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2164 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2166 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2167 struct shmem_sb_info config
= *sbinfo
;
2168 unsigned long inodes
;
2169 int error
= -EINVAL
;
2171 if (shmem_parse_options(data
, &config
, true))
2174 spin_lock(&sbinfo
->stat_lock
);
2175 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2176 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2178 if (config
.max_inodes
< inodes
)
2181 * Those tests disallow limited->unlimited while any are in use;
2182 * but we must separately disallow unlimited->limited, because
2183 * in that case we have no record of how much is already in use.
2185 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2187 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2191 sbinfo
->max_blocks
= config
.max_blocks
;
2192 sbinfo
->max_inodes
= config
.max_inodes
;
2193 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2195 mpol_put(sbinfo
->mpol
);
2196 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2198 spin_unlock(&sbinfo
->stat_lock
);
2202 static int shmem_show_options(struct seq_file
*seq
, struct dentry
*root
)
2204 struct shmem_sb_info
*sbinfo
= SHMEM_SB(root
->d_sb
);
2206 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2207 seq_printf(seq
, ",size=%luk",
2208 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2209 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2210 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2211 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2212 seq_printf(seq
, ",mode=%03ho", sbinfo
->mode
);
2213 if (sbinfo
->uid
!= 0)
2214 seq_printf(seq
, ",uid=%u", sbinfo
->uid
);
2215 if (sbinfo
->gid
!= 0)
2216 seq_printf(seq
, ",gid=%u", sbinfo
->gid
);
2217 shmem_show_mpol(seq
, sbinfo
->mpol
);
2220 #endif /* CONFIG_TMPFS */
2222 static void shmem_put_super(struct super_block
*sb
)
2224 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2226 percpu_counter_destroy(&sbinfo
->used_blocks
);
2228 sb
->s_fs_info
= NULL
;
2231 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2233 struct inode
*inode
;
2234 struct dentry
*root
;
2235 struct shmem_sb_info
*sbinfo
;
2238 /* Round up to L1_CACHE_BYTES to resist false sharing */
2239 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2240 L1_CACHE_BYTES
), GFP_KERNEL
);
2244 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2245 sbinfo
->uid
= current_fsuid();
2246 sbinfo
->gid
= current_fsgid();
2247 sb
->s_fs_info
= sbinfo
;
2251 * Per default we only allow half of the physical ram per
2252 * tmpfs instance, limiting inodes to one per page of lowmem;
2253 * but the internal instance is left unlimited.
2255 if (!(sb
->s_flags
& MS_NOUSER
)) {
2256 sbinfo
->max_blocks
= shmem_default_max_blocks();
2257 sbinfo
->max_inodes
= shmem_default_max_inodes();
2258 if (shmem_parse_options(data
, sbinfo
, false)) {
2263 sb
->s_export_op
= &shmem_export_ops
;
2265 sb
->s_flags
|= MS_NOUSER
;
2268 spin_lock_init(&sbinfo
->stat_lock
);
2269 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
2271 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2273 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
2274 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2275 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2276 sb
->s_magic
= TMPFS_MAGIC
;
2277 sb
->s_op
= &shmem_ops
;
2278 sb
->s_time_gran
= 1;
2279 #ifdef CONFIG_TMPFS_XATTR
2280 sb
->s_xattr
= shmem_xattr_handlers
;
2282 #ifdef CONFIG_TMPFS_POSIX_ACL
2283 sb
->s_flags
|= MS_POSIXACL
;
2286 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
2289 inode
->i_uid
= sbinfo
->uid
;
2290 inode
->i_gid
= sbinfo
->gid
;
2291 root
= d_alloc_root(inode
);
2300 shmem_put_super(sb
);
2304 static struct kmem_cache
*shmem_inode_cachep
;
2306 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2308 struct shmem_inode_info
*info
;
2309 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2312 return &info
->vfs_inode
;
2315 static void shmem_destroy_callback(struct rcu_head
*head
)
2317 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
2318 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2321 static void shmem_destroy_inode(struct inode
*inode
)
2323 if (S_ISREG(inode
->i_mode
))
2324 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2325 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
2328 static void shmem_init_inode(void *foo
)
2330 struct shmem_inode_info
*info
= foo
;
2331 inode_init_once(&info
->vfs_inode
);
2334 static int shmem_init_inodecache(void)
2336 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2337 sizeof(struct shmem_inode_info
),
2338 0, SLAB_PANIC
, shmem_init_inode
);
2342 static void shmem_destroy_inodecache(void)
2344 kmem_cache_destroy(shmem_inode_cachep
);
2347 static const struct address_space_operations shmem_aops
= {
2348 .writepage
= shmem_writepage
,
2349 .set_page_dirty
= __set_page_dirty_no_writeback
,
2351 .write_begin
= shmem_write_begin
,
2352 .write_end
= shmem_write_end
,
2354 .migratepage
= migrate_page
,
2355 .error_remove_page
= generic_error_remove_page
,
2358 static const struct file_operations shmem_file_operations
= {
2361 .llseek
= generic_file_llseek
,
2362 .read
= do_sync_read
,
2363 .write
= do_sync_write
,
2364 .aio_read
= shmem_file_aio_read
,
2365 .aio_write
= generic_file_aio_write
,
2366 .fsync
= noop_fsync
,
2367 .splice_read
= shmem_file_splice_read
,
2368 .splice_write
= generic_file_splice_write
,
2372 static const struct inode_operations shmem_inode_operations
= {
2373 .setattr
= shmem_setattr
,
2374 .truncate_range
= shmem_truncate_range
,
2375 #ifdef CONFIG_TMPFS_XATTR
2376 .setxattr
= shmem_setxattr
,
2377 .getxattr
= shmem_getxattr
,
2378 .listxattr
= shmem_listxattr
,
2379 .removexattr
= shmem_removexattr
,
2383 static const struct inode_operations shmem_dir_inode_operations
= {
2385 .create
= shmem_create
,
2386 .lookup
= simple_lookup
,
2388 .unlink
= shmem_unlink
,
2389 .symlink
= shmem_symlink
,
2390 .mkdir
= shmem_mkdir
,
2391 .rmdir
= shmem_rmdir
,
2392 .mknod
= shmem_mknod
,
2393 .rename
= shmem_rename
,
2395 #ifdef CONFIG_TMPFS_XATTR
2396 .setxattr
= shmem_setxattr
,
2397 .getxattr
= shmem_getxattr
,
2398 .listxattr
= shmem_listxattr
,
2399 .removexattr
= shmem_removexattr
,
2401 #ifdef CONFIG_TMPFS_POSIX_ACL
2402 .setattr
= shmem_setattr
,
2406 static const struct inode_operations shmem_special_inode_operations
= {
2407 #ifdef CONFIG_TMPFS_XATTR
2408 .setxattr
= shmem_setxattr
,
2409 .getxattr
= shmem_getxattr
,
2410 .listxattr
= shmem_listxattr
,
2411 .removexattr
= shmem_removexattr
,
2413 #ifdef CONFIG_TMPFS_POSIX_ACL
2414 .setattr
= shmem_setattr
,
2418 static const struct super_operations shmem_ops
= {
2419 .alloc_inode
= shmem_alloc_inode
,
2420 .destroy_inode
= shmem_destroy_inode
,
2422 .statfs
= shmem_statfs
,
2423 .remount_fs
= shmem_remount_fs
,
2424 .show_options
= shmem_show_options
,
2426 .evict_inode
= shmem_evict_inode
,
2427 .drop_inode
= generic_delete_inode
,
2428 .put_super
= shmem_put_super
,
2431 static const struct vm_operations_struct shmem_vm_ops
= {
2432 .fault
= shmem_fault
,
2434 .set_policy
= shmem_set_policy
,
2435 .get_policy
= shmem_get_policy
,
2439 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
2440 int flags
, const char *dev_name
, void *data
)
2442 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
2445 static struct file_system_type shmem_fs_type
= {
2446 .owner
= THIS_MODULE
,
2448 .mount
= shmem_mount
,
2449 .kill_sb
= kill_litter_super
,
2452 int __init
shmem_init(void)
2456 error
= bdi_init(&shmem_backing_dev_info
);
2460 error
= shmem_init_inodecache();
2464 error
= register_filesystem(&shmem_fs_type
);
2466 printk(KERN_ERR
"Could not register tmpfs\n");
2470 shm_mnt
= vfs_kern_mount(&shmem_fs_type
, MS_NOUSER
,
2471 shmem_fs_type
.name
, NULL
);
2472 if (IS_ERR(shm_mnt
)) {
2473 error
= PTR_ERR(shm_mnt
);
2474 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2480 unregister_filesystem(&shmem_fs_type
);
2482 shmem_destroy_inodecache();
2484 bdi_destroy(&shmem_backing_dev_info
);
2486 shm_mnt
= ERR_PTR(error
);
2490 #else /* !CONFIG_SHMEM */
2493 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2495 * This is intended for small system where the benefits of the full
2496 * shmem code (swap-backed and resource-limited) are outweighed by
2497 * their complexity. On systems without swap this code should be
2498 * effectively equivalent, but much lighter weight.
2501 #include <linux/ramfs.h>
2503 static struct file_system_type shmem_fs_type
= {
2505 .mount
= ramfs_mount
,
2506 .kill_sb
= kill_litter_super
,
2509 int __init
shmem_init(void)
2511 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
2513 shm_mnt
= kern_mount(&shmem_fs_type
);
2514 BUG_ON(IS_ERR(shm_mnt
));
2519 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
2524 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2529 void shmem_unlock_mapping(struct address_space
*mapping
)
2533 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
2535 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
2537 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
2539 #define shmem_vm_ops generic_file_vm_ops
2540 #define shmem_file_operations ramfs_file_operations
2541 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2542 #define shmem_acct_size(flags, size) 0
2543 #define shmem_unacct_size(flags, size) do {} while (0)
2545 #endif /* CONFIG_SHMEM */
2550 * shmem_file_setup - get an unlinked file living in tmpfs
2551 * @name: name for dentry (to be seen in /proc/<pid>/maps
2552 * @size: size to be set for the file
2553 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2555 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
2559 struct inode
*inode
;
2561 struct dentry
*root
;
2564 if (IS_ERR(shm_mnt
))
2565 return (void *)shm_mnt
;
2567 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
2568 return ERR_PTR(-EINVAL
);
2570 if (shmem_acct_size(flags
, size
))
2571 return ERR_PTR(-ENOMEM
);
2575 this.len
= strlen(name
);
2576 this.hash
= 0; /* will go */
2577 root
= shm_mnt
->mnt_root
;
2578 path
.dentry
= d_alloc(root
, &this);
2581 path
.mnt
= mntget(shm_mnt
);
2584 inode
= shmem_get_inode(root
->d_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
2588 d_instantiate(path
.dentry
, inode
);
2589 inode
->i_size
= size
;
2590 clear_nlink(inode
); /* It is unlinked */
2592 error
= ramfs_nommu_expand_for_mapping(inode
, size
);
2598 file
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
2599 &shmem_file_operations
);
2608 shmem_unacct_size(flags
, size
);
2609 return ERR_PTR(error
);
2611 EXPORT_SYMBOL_GPL(shmem_file_setup
);
2614 * shmem_zero_setup - setup a shared anonymous mapping
2615 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2617 int shmem_zero_setup(struct vm_area_struct
*vma
)
2620 loff_t size
= vma
->vm_end
- vma
->vm_start
;
2622 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
2624 return PTR_ERR(file
);
2628 vma
->vm_file
= file
;
2629 vma
->vm_ops
= &shmem_vm_ops
;
2630 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
2635 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2636 * @mapping: the page's address_space
2637 * @index: the page index
2638 * @gfp: the page allocator flags to use if allocating
2640 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2641 * with any new page allocations done using the specified allocation flags.
2642 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2643 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2644 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2646 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2647 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2649 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
2650 pgoff_t index
, gfp_t gfp
)
2653 struct inode
*inode
= mapping
->host
;
2657 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
2658 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
2660 page
= ERR_PTR(error
);
2666 * The tiny !SHMEM case uses ramfs without swap
2668 return read_cache_page_gfp(mapping
, index
, gfp
);
2671 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);