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/module.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_exceptional_entry(page
))
337 /* radix_tree_deref_retry(page) */
340 if (!page_cache_get_speculative(page
))
343 /* Has the page moved? */
344 if (unlikely(page
!= *((void **)pages
[i
]))) {
345 page_cache_release(page
);
349 indices
[ret
] = indices
[i
];
353 if (unlikely(!ret
&& nr_found
))
360 * Remove swap entry from radix tree, free the swap and its page cache.
362 static int shmem_free_swap(struct address_space
*mapping
,
363 pgoff_t index
, void *radswap
)
367 spin_lock_irq(&mapping
->tree_lock
);
368 error
= shmem_radix_tree_replace(mapping
, index
, radswap
, NULL
);
369 spin_unlock_irq(&mapping
->tree_lock
);
371 free_swap_and_cache(radix_to_swp_entry(radswap
));
376 * Pagevec may contain swap entries, so shuffle up pages before releasing.
378 static void shmem_pagevec_release(struct pagevec
*pvec
)
382 for (i
= 0, j
= 0; i
< pagevec_count(pvec
); i
++) {
383 struct page
*page
= pvec
->pages
[i
];
384 if (!radix_tree_exceptional_entry(page
))
385 pvec
->pages
[j
++] = page
;
388 pagevec_release(pvec
);
392 * Remove range of pages and swap entries from radix tree, and free them.
394 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
396 struct address_space
*mapping
= inode
->i_mapping
;
397 struct shmem_inode_info
*info
= SHMEM_I(inode
);
398 pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
399 unsigned partial
= lstart
& (PAGE_CACHE_SIZE
- 1);
400 pgoff_t end
= (lend
>> PAGE_CACHE_SHIFT
);
402 pgoff_t indices
[PAGEVEC_SIZE
];
403 long nr_swaps_freed
= 0;
407 BUG_ON((lend
& (PAGE_CACHE_SIZE
- 1)) != (PAGE_CACHE_SIZE
- 1));
409 pagevec_init(&pvec
, 0);
411 while (index
<= end
) {
412 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
413 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
414 pvec
.pages
, indices
);
417 mem_cgroup_uncharge_start();
418 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
419 struct page
*page
= pvec
.pages
[i
];
425 if (radix_tree_exceptional_entry(page
)) {
426 nr_swaps_freed
+= !shmem_free_swap(mapping
,
431 if (!trylock_page(page
))
433 if (page
->mapping
== mapping
) {
434 VM_BUG_ON(PageWriteback(page
));
435 truncate_inode_page(mapping
, page
);
439 shmem_pagevec_release(&pvec
);
440 mem_cgroup_uncharge_end();
446 struct page
*page
= NULL
;
447 shmem_getpage(inode
, start
- 1, &page
, SGP_READ
, NULL
);
449 zero_user_segment(page
, partial
, PAGE_CACHE_SIZE
);
450 set_page_dirty(page
);
452 page_cache_release(page
);
459 pvec
.nr
= shmem_find_get_pages_and_swap(mapping
, index
,
460 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
461 pvec
.pages
, indices
);
468 if (index
== start
&& indices
[0] > end
) {
469 shmem_pagevec_release(&pvec
);
472 mem_cgroup_uncharge_start();
473 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
474 struct page
*page
= pvec
.pages
[i
];
480 if (radix_tree_exceptional_entry(page
)) {
481 nr_swaps_freed
+= !shmem_free_swap(mapping
,
487 if (page
->mapping
== mapping
) {
488 VM_BUG_ON(PageWriteback(page
));
489 truncate_inode_page(mapping
, page
);
493 shmem_pagevec_release(&pvec
);
494 mem_cgroup_uncharge_end();
498 spin_lock(&info
->lock
);
499 info
->swapped
-= nr_swaps_freed
;
500 shmem_recalc_inode(inode
);
501 spin_unlock(&info
->lock
);
503 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
505 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
507 static int shmem_setattr(struct dentry
*dentry
, struct iattr
*attr
)
509 struct inode
*inode
= dentry
->d_inode
;
512 error
= inode_change_ok(inode
, attr
);
516 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
517 loff_t oldsize
= inode
->i_size
;
518 loff_t newsize
= attr
->ia_size
;
520 if (newsize
!= oldsize
) {
521 i_size_write(inode
, newsize
);
522 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
524 if (newsize
< oldsize
) {
525 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
526 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
527 shmem_truncate_range(inode
, newsize
, (loff_t
)-1);
528 /* unmap again to remove racily COWed private pages */
529 unmap_mapping_range(inode
->i_mapping
, holebegin
, 0, 1);
533 setattr_copy(inode
, attr
);
534 #ifdef CONFIG_TMPFS_POSIX_ACL
535 if (attr
->ia_valid
& ATTR_MODE
)
536 error
= generic_acl_chmod(inode
);
541 static void shmem_evict_inode(struct inode
*inode
)
543 struct shmem_inode_info
*info
= SHMEM_I(inode
);
544 struct shmem_xattr
*xattr
, *nxattr
;
546 if (inode
->i_mapping
->a_ops
== &shmem_aops
) {
547 shmem_unacct_size(info
->flags
, inode
->i_size
);
549 shmem_truncate_range(inode
, 0, (loff_t
)-1);
550 if (!list_empty(&info
->swaplist
)) {
551 mutex_lock(&shmem_swaplist_mutex
);
552 list_del_init(&info
->swaplist
);
553 mutex_unlock(&shmem_swaplist_mutex
);
556 kfree(info
->symlink
);
558 list_for_each_entry_safe(xattr
, nxattr
, &info
->xattr_list
, list
) {
562 BUG_ON(inode
->i_blocks
);
563 shmem_free_inode(inode
->i_sb
);
564 end_writeback(inode
);
568 * If swap found in inode, free it and move page from swapcache to filecache.
570 static int shmem_unuse_inode(struct shmem_inode_info
*info
,
571 swp_entry_t swap
, struct page
*page
)
573 struct address_space
*mapping
= info
->vfs_inode
.i_mapping
;
578 radswap
= swp_to_radix_entry(swap
);
579 index
= radix_tree_locate_item(&mapping
->page_tree
, radswap
);
584 * Move _head_ to start search for next from here.
585 * But be careful: shmem_evict_inode checks list_empty without taking
586 * mutex, and there's an instant in list_move_tail when info->swaplist
587 * would appear empty, if it were the only one on shmem_swaplist.
589 if (shmem_swaplist
.next
!= &info
->swaplist
)
590 list_move_tail(&shmem_swaplist
, &info
->swaplist
);
593 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
594 * but also to hold up shmem_evict_inode(): so inode cannot be freed
595 * beneath us (pagelock doesn't help until the page is in pagecache).
597 error
= shmem_add_to_page_cache(page
, mapping
, index
,
598 GFP_NOWAIT
, radswap
);
599 /* which does mem_cgroup_uncharge_cache_page on error */
601 if (error
!= -ENOMEM
) {
603 * Truncation and eviction use free_swap_and_cache(), which
604 * only does trylock page: if we raced, best clean up here.
606 delete_from_swap_cache(page
);
607 set_page_dirty(page
);
609 spin_lock(&info
->lock
);
611 spin_unlock(&info
->lock
);
614 error
= 1; /* not an error, but entry was found */
620 * Search through swapped inodes to find and replace swap by page.
622 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
624 struct list_head
*this, *next
;
625 struct shmem_inode_info
*info
;
630 * Charge page using GFP_KERNEL while we can wait, before taking
631 * the shmem_swaplist_mutex which might hold up shmem_writepage().
632 * Charged back to the user (not to caller) when swap account is used.
634 error
= mem_cgroup_cache_charge(page
, current
->mm
, GFP_KERNEL
);
637 /* No radix_tree_preload: swap entry keeps a place for page in tree */
639 mutex_lock(&shmem_swaplist_mutex
);
640 list_for_each_safe(this, next
, &shmem_swaplist
) {
641 info
= list_entry(this, struct shmem_inode_info
, swaplist
);
643 found
= shmem_unuse_inode(info
, swap
, page
);
645 list_del_init(&info
->swaplist
);
650 mutex_unlock(&shmem_swaplist_mutex
);
653 mem_cgroup_uncharge_cache_page(page
);
658 page_cache_release(page
);
663 * Move the page from the page cache to the swap cache.
665 static int shmem_writepage(struct page
*page
, struct writeback_control
*wbc
)
667 struct shmem_inode_info
*info
;
668 struct address_space
*mapping
;
673 BUG_ON(!PageLocked(page
));
674 mapping
= page
->mapping
;
676 inode
= mapping
->host
;
677 info
= SHMEM_I(inode
);
678 if (info
->flags
& VM_LOCKED
)
680 if (!total_swap_pages
)
684 * shmem_backing_dev_info's capabilities prevent regular writeback or
685 * sync from ever calling shmem_writepage; but a stacking filesystem
686 * might use ->writepage of its underlying filesystem, in which case
687 * tmpfs should write out to swap only in response to memory pressure,
688 * and not for the writeback threads or sync.
690 if (!wbc
->for_reclaim
) {
691 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
694 swap
= get_swap_page();
699 * Add inode to shmem_unuse()'s list of swapped-out inodes,
700 * if it's not already there. Do it now before the page is
701 * moved to swap cache, when its pagelock no longer protects
702 * the inode from eviction. But don't unlock the mutex until
703 * we've incremented swapped, because shmem_unuse_inode() will
704 * prune a !swapped inode from the swaplist under this mutex.
706 mutex_lock(&shmem_swaplist_mutex
);
707 if (list_empty(&info
->swaplist
))
708 list_add_tail(&info
->swaplist
, &shmem_swaplist
);
710 if (add_to_swap_cache(page
, swap
, GFP_ATOMIC
) == 0) {
711 swap_shmem_alloc(swap
);
712 shmem_delete_from_page_cache(page
, swp_to_radix_entry(swap
));
714 spin_lock(&info
->lock
);
716 shmem_recalc_inode(inode
);
717 spin_unlock(&info
->lock
);
719 mutex_unlock(&shmem_swaplist_mutex
);
720 BUG_ON(page_mapped(page
));
721 swap_writepage(page
, wbc
);
725 mutex_unlock(&shmem_swaplist_mutex
);
726 swapcache_free(swap
, NULL
);
728 set_page_dirty(page
);
729 if (wbc
->for_reclaim
)
730 return AOP_WRITEPAGE_ACTIVATE
; /* Return with page locked */
737 static void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
741 if (!mpol
|| mpol
->mode
== MPOL_DEFAULT
)
742 return; /* show nothing */
744 mpol_to_str(buffer
, sizeof(buffer
), mpol
, 1);
746 seq_printf(seq
, ",mpol=%s", buffer
);
749 static struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
751 struct mempolicy
*mpol
= NULL
;
753 spin_lock(&sbinfo
->stat_lock
); /* prevent replace/use races */
756 spin_unlock(&sbinfo
->stat_lock
);
760 #endif /* CONFIG_TMPFS */
762 static struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
763 struct shmem_inode_info
*info
, pgoff_t index
)
765 struct mempolicy mpol
, *spol
;
766 struct vm_area_struct pvma
;
768 spol
= mpol_cond_copy(&mpol
,
769 mpol_shared_policy_lookup(&info
->policy
, index
));
771 /* Create a pseudo vma that just contains the policy */
773 pvma
.vm_pgoff
= index
;
775 pvma
.vm_policy
= spol
;
776 return swapin_readahead(swap
, gfp
, &pvma
, 0);
779 static struct page
*shmem_alloc_page(gfp_t gfp
,
780 struct shmem_inode_info
*info
, pgoff_t index
)
782 struct vm_area_struct pvma
;
784 /* Create a pseudo vma that just contains the policy */
786 pvma
.vm_pgoff
= index
;
788 pvma
.vm_policy
= mpol_shared_policy_lookup(&info
->policy
, index
);
791 * alloc_page_vma() will drop the shared policy reference
793 return alloc_page_vma(gfp
, &pvma
, 0);
795 #else /* !CONFIG_NUMA */
797 static inline void shmem_show_mpol(struct seq_file
*seq
, struct mempolicy
*mpol
)
800 #endif /* CONFIG_TMPFS */
802 static inline struct page
*shmem_swapin(swp_entry_t swap
, gfp_t gfp
,
803 struct shmem_inode_info
*info
, pgoff_t index
)
805 return swapin_readahead(swap
, gfp
, NULL
, 0);
808 static inline struct page
*shmem_alloc_page(gfp_t gfp
,
809 struct shmem_inode_info
*info
, pgoff_t index
)
811 return alloc_page(gfp
);
813 #endif /* CONFIG_NUMA */
815 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
816 static inline struct mempolicy
*shmem_get_sbmpol(struct shmem_sb_info
*sbinfo
)
823 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
825 * If we allocate a new one we do not mark it dirty. That's up to the
826 * vm. If we swap it in we mark it dirty since we also free the swap
827 * entry since a page cannot live in both the swap and page cache
829 static int shmem_getpage_gfp(struct inode
*inode
, pgoff_t index
,
830 struct page
**pagep
, enum sgp_type sgp
, gfp_t gfp
, int *fault_type
)
832 struct address_space
*mapping
= inode
->i_mapping
;
833 struct shmem_inode_info
*info
;
834 struct shmem_sb_info
*sbinfo
;
840 if (index
> (MAX_LFS_FILESIZE
>> PAGE_CACHE_SHIFT
))
844 page
= find_lock_page(mapping
, index
);
845 if (radix_tree_exceptional_entry(page
)) {
846 swap
= radix_to_swp_entry(page
);
850 if (sgp
!= SGP_WRITE
&&
851 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
856 if (page
|| (sgp
== SGP_READ
&& !swap
.val
)) {
858 * Once we can get the page lock, it must be uptodate:
859 * if there were an error in reading back from swap,
860 * the page would not be inserted into the filecache.
862 BUG_ON(page
&& !PageUptodate(page
));
868 * Fast cache lookup did not find it:
869 * bring it back from swap or allocate.
871 info
= SHMEM_I(inode
);
872 sbinfo
= SHMEM_SB(inode
->i_sb
);
875 /* Look it up and read it in.. */
876 page
= lookup_swap_cache(swap
);
878 /* here we actually do the io */
880 *fault_type
|= VM_FAULT_MAJOR
;
881 page
= shmem_swapin(swap
, gfp
, info
, index
);
888 /* We have to do this with page locked to prevent races */
890 if (!PageUptodate(page
)) {
894 wait_on_page_writeback(page
);
896 /* Someone may have already done it for us */
898 if (page
->mapping
== mapping
&&
899 page
->index
== index
)
905 error
= mem_cgroup_cache_charge(page
, current
->mm
,
906 gfp
& GFP_RECLAIM_MASK
);
908 error
= shmem_add_to_page_cache(page
, mapping
, index
,
909 gfp
, swp_to_radix_entry(swap
));
913 spin_lock(&info
->lock
);
915 shmem_recalc_inode(inode
);
916 spin_unlock(&info
->lock
);
918 delete_from_swap_cache(page
);
919 set_page_dirty(page
);
923 if (shmem_acct_block(info
->flags
)) {
927 if (sbinfo
->max_blocks
) {
928 if (percpu_counter_compare(&sbinfo
->used_blocks
,
929 sbinfo
->max_blocks
) >= 0) {
933 percpu_counter_inc(&sbinfo
->used_blocks
);
936 page
= shmem_alloc_page(gfp
, info
, index
);
942 SetPageSwapBacked(page
);
943 __set_page_locked(page
);
944 error
= mem_cgroup_cache_charge(page
, current
->mm
,
945 gfp
& GFP_RECLAIM_MASK
);
947 error
= shmem_add_to_page_cache(page
, mapping
, index
,
951 lru_cache_add_anon(page
);
953 spin_lock(&info
->lock
);
955 inode
->i_blocks
+= BLOCKS_PER_PAGE
;
956 shmem_recalc_inode(inode
);
957 spin_unlock(&info
->lock
);
959 clear_highpage(page
);
960 flush_dcache_page(page
);
961 SetPageUptodate(page
);
962 if (sgp
== SGP_DIRTY
)
963 set_page_dirty(page
);
966 /* Perhaps the file has been truncated since we checked */
967 if (sgp
!= SGP_WRITE
&&
968 ((loff_t
)index
<< PAGE_CACHE_SHIFT
) >= i_size_read(inode
)) {
979 ClearPageDirty(page
);
980 delete_from_page_cache(page
);
981 spin_lock(&info
->lock
);
983 inode
->i_blocks
-= BLOCKS_PER_PAGE
;
984 spin_unlock(&info
->lock
);
986 if (sbinfo
->max_blocks
)
987 percpu_counter_add(&sbinfo
->used_blocks
, -1);
989 shmem_unacct_blocks(info
->flags
, 1);
991 if (swap
.val
&& error
!= -EINVAL
) {
992 struct page
*test
= find_get_page(mapping
, index
);
993 if (test
&& !radix_tree_exceptional_entry(test
))
994 page_cache_release(test
);
995 /* Have another try if the entry has changed */
996 if (test
!= swp_to_radix_entry(swap
))
1001 page_cache_release(page
);
1003 if (error
== -ENOSPC
&& !once
++) {
1004 info
= SHMEM_I(inode
);
1005 spin_lock(&info
->lock
);
1006 shmem_recalc_inode(inode
);
1007 spin_unlock(&info
->lock
);
1010 if (error
== -EEXIST
)
1015 static int shmem_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1017 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1019 int ret
= VM_FAULT_LOCKED
;
1021 error
= shmem_getpage(inode
, vmf
->pgoff
, &vmf
->page
, SGP_CACHE
, &ret
);
1023 return ((error
== -ENOMEM
) ? VM_FAULT_OOM
: VM_FAULT_SIGBUS
);
1025 if (ret
& VM_FAULT_MAJOR
) {
1026 count_vm_event(PGMAJFAULT
);
1027 mem_cgroup_count_vm_event(vma
->vm_mm
, PGMAJFAULT
);
1033 static int shmem_set_policy(struct vm_area_struct
*vma
, struct mempolicy
*mpol
)
1035 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1036 return mpol_set_shared_policy(&SHMEM_I(inode
)->policy
, vma
, mpol
);
1039 static struct mempolicy
*shmem_get_policy(struct vm_area_struct
*vma
,
1042 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
1045 index
= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
1046 return mpol_shared_policy_lookup(&SHMEM_I(inode
)->policy
, index
);
1050 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
1052 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1053 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1054 int retval
= -ENOMEM
;
1056 spin_lock(&info
->lock
);
1057 if (lock
&& !(info
->flags
& VM_LOCKED
)) {
1058 if (!user_shm_lock(inode
->i_size
, user
))
1060 info
->flags
|= VM_LOCKED
;
1061 mapping_set_unevictable(file
->f_mapping
);
1063 if (!lock
&& (info
->flags
& VM_LOCKED
) && user
) {
1064 user_shm_unlock(inode
->i_size
, user
);
1065 info
->flags
&= ~VM_LOCKED
;
1066 mapping_clear_unevictable(file
->f_mapping
);
1067 scan_mapping_unevictable_pages(file
->f_mapping
);
1072 spin_unlock(&info
->lock
);
1076 static int shmem_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1078 file_accessed(file
);
1079 vma
->vm_ops
= &shmem_vm_ops
;
1080 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1084 static struct inode
*shmem_get_inode(struct super_block
*sb
, const struct inode
*dir
,
1085 int mode
, dev_t dev
, unsigned long flags
)
1087 struct inode
*inode
;
1088 struct shmem_inode_info
*info
;
1089 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
1091 if (shmem_reserve_inode(sb
))
1094 inode
= new_inode(sb
);
1096 inode
->i_ino
= get_next_ino();
1097 inode_init_owner(inode
, dir
, mode
);
1098 inode
->i_blocks
= 0;
1099 inode
->i_mapping
->backing_dev_info
= &shmem_backing_dev_info
;
1100 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
1101 inode
->i_generation
= get_seconds();
1102 info
= SHMEM_I(inode
);
1103 memset(info
, 0, (char *)inode
- (char *)info
);
1104 spin_lock_init(&info
->lock
);
1105 info
->flags
= flags
& VM_NORESERVE
;
1106 INIT_LIST_HEAD(&info
->swaplist
);
1107 INIT_LIST_HEAD(&info
->xattr_list
);
1108 cache_no_acl(inode
);
1110 switch (mode
& S_IFMT
) {
1112 inode
->i_op
= &shmem_special_inode_operations
;
1113 init_special_inode(inode
, mode
, dev
);
1116 inode
->i_mapping
->a_ops
= &shmem_aops
;
1117 inode
->i_op
= &shmem_inode_operations
;
1118 inode
->i_fop
= &shmem_file_operations
;
1119 mpol_shared_policy_init(&info
->policy
,
1120 shmem_get_sbmpol(sbinfo
));
1124 /* Some things misbehave if size == 0 on a directory */
1125 inode
->i_size
= 2 * BOGO_DIRENT_SIZE
;
1126 inode
->i_op
= &shmem_dir_inode_operations
;
1127 inode
->i_fop
= &simple_dir_operations
;
1131 * Must not load anything in the rbtree,
1132 * mpol_free_shared_policy will not be called.
1134 mpol_shared_policy_init(&info
->policy
, NULL
);
1138 shmem_free_inode(sb
);
1143 static const struct inode_operations shmem_symlink_inode_operations
;
1144 static const struct inode_operations shmem_short_symlink_operations
;
1147 shmem_write_begin(struct file
*file
, struct address_space
*mapping
,
1148 loff_t pos
, unsigned len
, unsigned flags
,
1149 struct page
**pagep
, void **fsdata
)
1151 struct inode
*inode
= mapping
->host
;
1152 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1153 return shmem_getpage(inode
, index
, pagep
, SGP_WRITE
, NULL
);
1157 shmem_write_end(struct file
*file
, struct address_space
*mapping
,
1158 loff_t pos
, unsigned len
, unsigned copied
,
1159 struct page
*page
, void *fsdata
)
1161 struct inode
*inode
= mapping
->host
;
1163 if (pos
+ copied
> inode
->i_size
)
1164 i_size_write(inode
, pos
+ copied
);
1166 set_page_dirty(page
);
1168 page_cache_release(page
);
1173 static void do_shmem_file_read(struct file
*filp
, loff_t
*ppos
, read_descriptor_t
*desc
, read_actor_t actor
)
1175 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
1176 struct address_space
*mapping
= inode
->i_mapping
;
1178 unsigned long offset
;
1179 enum sgp_type sgp
= SGP_READ
;
1182 * Might this read be for a stacking filesystem? Then when reading
1183 * holes of a sparse file, we actually need to allocate those pages,
1184 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1186 if (segment_eq(get_fs(), KERNEL_DS
))
1189 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1190 offset
= *ppos
& ~PAGE_CACHE_MASK
;
1193 struct page
*page
= NULL
;
1195 unsigned long nr
, ret
;
1196 loff_t i_size
= i_size_read(inode
);
1198 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1199 if (index
> end_index
)
1201 if (index
== end_index
) {
1202 nr
= i_size
& ~PAGE_CACHE_MASK
;
1207 desc
->error
= shmem_getpage(inode
, index
, &page
, sgp
, NULL
);
1209 if (desc
->error
== -EINVAL
)
1217 * We must evaluate after, since reads (unlike writes)
1218 * are called without i_mutex protection against truncate
1220 nr
= PAGE_CACHE_SIZE
;
1221 i_size
= i_size_read(inode
);
1222 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
1223 if (index
== end_index
) {
1224 nr
= i_size
& ~PAGE_CACHE_MASK
;
1227 page_cache_release(page
);
1235 * If users can be writing to this page using arbitrary
1236 * virtual addresses, take care about potential aliasing
1237 * before reading the page on the kernel side.
1239 if (mapping_writably_mapped(mapping
))
1240 flush_dcache_page(page
);
1242 * Mark the page accessed if we read the beginning.
1245 mark_page_accessed(page
);
1247 page
= ZERO_PAGE(0);
1248 page_cache_get(page
);
1252 * Ok, we have the page, and it's up-to-date, so
1253 * now we can copy it to user space...
1255 * The actor routine returns how many bytes were actually used..
1256 * NOTE! This may not be the same as how much of a user buffer
1257 * we filled up (we may be padding etc), so we can only update
1258 * "pos" here (the actor routine has to update the user buffer
1259 * pointers and the remaining count).
1261 ret
= actor(desc
, page
, offset
, nr
);
1263 index
+= offset
>> PAGE_CACHE_SHIFT
;
1264 offset
&= ~PAGE_CACHE_MASK
;
1266 page_cache_release(page
);
1267 if (ret
!= nr
|| !desc
->count
)
1273 *ppos
= ((loff_t
) index
<< PAGE_CACHE_SHIFT
) + offset
;
1274 file_accessed(filp
);
1277 static ssize_t
shmem_file_aio_read(struct kiocb
*iocb
,
1278 const struct iovec
*iov
, unsigned long nr_segs
, loff_t pos
)
1280 struct file
*filp
= iocb
->ki_filp
;
1284 loff_t
*ppos
= &iocb
->ki_pos
;
1286 retval
= generic_segment_checks(iov
, &nr_segs
, &count
, VERIFY_WRITE
);
1290 for (seg
= 0; seg
< nr_segs
; seg
++) {
1291 read_descriptor_t desc
;
1294 desc
.arg
.buf
= iov
[seg
].iov_base
;
1295 desc
.count
= iov
[seg
].iov_len
;
1296 if (desc
.count
== 0)
1299 do_shmem_file_read(filp
, ppos
, &desc
, file_read_actor
);
1300 retval
+= desc
.written
;
1302 retval
= retval
?: desc
.error
;
1311 static ssize_t
shmem_file_splice_read(struct file
*in
, loff_t
*ppos
,
1312 struct pipe_inode_info
*pipe
, size_t len
,
1315 struct address_space
*mapping
= in
->f_mapping
;
1316 struct inode
*inode
= mapping
->host
;
1317 unsigned int loff
, nr_pages
, req_pages
;
1318 struct page
*pages
[PIPE_DEF_BUFFERS
];
1319 struct partial_page partial
[PIPE_DEF_BUFFERS
];
1321 pgoff_t index
, end_index
;
1324 struct splice_pipe_desc spd
= {
1328 .ops
= &page_cache_pipe_buf_ops
,
1329 .spd_release
= spd_release_page
,
1332 isize
= i_size_read(inode
);
1333 if (unlikely(*ppos
>= isize
))
1336 left
= isize
- *ppos
;
1337 if (unlikely(left
< len
))
1340 if (splice_grow_spd(pipe
, &spd
))
1343 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1344 loff
= *ppos
& ~PAGE_CACHE_MASK
;
1345 req_pages
= (len
+ loff
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1346 nr_pages
= min(req_pages
, pipe
->buffers
);
1348 spd
.nr_pages
= find_get_pages_contig(mapping
, index
,
1349 nr_pages
, spd
.pages
);
1350 index
+= spd
.nr_pages
;
1353 while (spd
.nr_pages
< nr_pages
) {
1354 error
= shmem_getpage(inode
, index
, &page
, SGP_CACHE
, NULL
);
1358 spd
.pages
[spd
.nr_pages
++] = page
;
1362 index
= *ppos
>> PAGE_CACHE_SHIFT
;
1363 nr_pages
= spd
.nr_pages
;
1366 for (page_nr
= 0; page_nr
< nr_pages
; page_nr
++) {
1367 unsigned int this_len
;
1372 this_len
= min_t(unsigned long, len
, PAGE_CACHE_SIZE
- loff
);
1373 page
= spd
.pages
[page_nr
];
1375 if (!PageUptodate(page
) || page
->mapping
!= mapping
) {
1376 error
= shmem_getpage(inode
, index
, &page
,
1381 page_cache_release(spd
.pages
[page_nr
]);
1382 spd
.pages
[page_nr
] = page
;
1385 isize
= i_size_read(inode
);
1386 end_index
= (isize
- 1) >> PAGE_CACHE_SHIFT
;
1387 if (unlikely(!isize
|| index
> end_index
))
1390 if (end_index
== index
) {
1393 plen
= ((isize
- 1) & ~PAGE_CACHE_MASK
) + 1;
1397 this_len
= min(this_len
, plen
- loff
);
1401 spd
.partial
[page_nr
].offset
= loff
;
1402 spd
.partial
[page_nr
].len
= this_len
;
1409 while (page_nr
< nr_pages
)
1410 page_cache_release(spd
.pages
[page_nr
++]);
1413 error
= splice_to_pipe(pipe
, &spd
);
1415 splice_shrink_spd(pipe
, &spd
);
1424 static int shmem_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
1426 struct shmem_sb_info
*sbinfo
= SHMEM_SB(dentry
->d_sb
);
1428 buf
->f_type
= TMPFS_MAGIC
;
1429 buf
->f_bsize
= PAGE_CACHE_SIZE
;
1430 buf
->f_namelen
= NAME_MAX
;
1431 if (sbinfo
->max_blocks
) {
1432 buf
->f_blocks
= sbinfo
->max_blocks
;
1434 buf
->f_bfree
= sbinfo
->max_blocks
-
1435 percpu_counter_sum(&sbinfo
->used_blocks
);
1437 if (sbinfo
->max_inodes
) {
1438 buf
->f_files
= sbinfo
->max_inodes
;
1439 buf
->f_ffree
= sbinfo
->free_inodes
;
1441 /* else leave those fields 0 like simple_statfs */
1446 * File creation. Allocate an inode, and we're done..
1449 shmem_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
1451 struct inode
*inode
;
1452 int error
= -ENOSPC
;
1454 inode
= shmem_get_inode(dir
->i_sb
, dir
, mode
, dev
, VM_NORESERVE
);
1456 error
= security_inode_init_security(inode
, dir
,
1457 &dentry
->d_name
, NULL
,
1460 if (error
!= -EOPNOTSUPP
) {
1465 #ifdef CONFIG_TMPFS_POSIX_ACL
1466 error
= generic_acl_init(inode
, dir
);
1474 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1475 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1476 d_instantiate(dentry
, inode
);
1477 dget(dentry
); /* Extra count - pin the dentry in core */
1482 static int shmem_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1486 if ((error
= shmem_mknod(dir
, dentry
, mode
| S_IFDIR
, 0)))
1492 static int shmem_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1493 struct nameidata
*nd
)
1495 return shmem_mknod(dir
, dentry
, mode
| S_IFREG
, 0);
1501 static int shmem_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*dentry
)
1503 struct inode
*inode
= old_dentry
->d_inode
;
1507 * No ordinary (disk based) filesystem counts links as inodes;
1508 * but each new link needs a new dentry, pinning lowmem, and
1509 * tmpfs dentries cannot be pruned until they are unlinked.
1511 ret
= shmem_reserve_inode(inode
->i_sb
);
1515 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1516 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1518 ihold(inode
); /* New dentry reference */
1519 dget(dentry
); /* Extra pinning count for the created dentry */
1520 d_instantiate(dentry
, inode
);
1525 static int shmem_unlink(struct inode
*dir
, struct dentry
*dentry
)
1527 struct inode
*inode
= dentry
->d_inode
;
1529 if (inode
->i_nlink
> 1 && !S_ISDIR(inode
->i_mode
))
1530 shmem_free_inode(inode
->i_sb
);
1532 dir
->i_size
-= BOGO_DIRENT_SIZE
;
1533 inode
->i_ctime
= dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1535 dput(dentry
); /* Undo the count from "create" - this does all the work */
1539 static int shmem_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1541 if (!simple_empty(dentry
))
1544 drop_nlink(dentry
->d_inode
);
1546 return shmem_unlink(dir
, dentry
);
1550 * The VFS layer already does all the dentry stuff for rename,
1551 * we just have to decrement the usage count for the target if
1552 * it exists so that the VFS layer correctly free's it when it
1555 static int shmem_rename(struct inode
*old_dir
, struct dentry
*old_dentry
, struct inode
*new_dir
, struct dentry
*new_dentry
)
1557 struct inode
*inode
= old_dentry
->d_inode
;
1558 int they_are_dirs
= S_ISDIR(inode
->i_mode
);
1560 if (!simple_empty(new_dentry
))
1563 if (new_dentry
->d_inode
) {
1564 (void) shmem_unlink(new_dir
, new_dentry
);
1566 drop_nlink(old_dir
);
1567 } else if (they_are_dirs
) {
1568 drop_nlink(old_dir
);
1572 old_dir
->i_size
-= BOGO_DIRENT_SIZE
;
1573 new_dir
->i_size
+= BOGO_DIRENT_SIZE
;
1574 old_dir
->i_ctime
= old_dir
->i_mtime
=
1575 new_dir
->i_ctime
= new_dir
->i_mtime
=
1576 inode
->i_ctime
= CURRENT_TIME
;
1580 static int shmem_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *symname
)
1584 struct inode
*inode
;
1587 struct shmem_inode_info
*info
;
1589 len
= strlen(symname
) + 1;
1590 if (len
> PAGE_CACHE_SIZE
)
1591 return -ENAMETOOLONG
;
1593 inode
= shmem_get_inode(dir
->i_sb
, dir
, S_IFLNK
|S_IRWXUGO
, 0, VM_NORESERVE
);
1597 error
= security_inode_init_security(inode
, dir
, &dentry
->d_name
, NULL
,
1600 if (error
!= -EOPNOTSUPP
) {
1607 info
= SHMEM_I(inode
);
1608 inode
->i_size
= len
-1;
1609 if (len
<= SHORT_SYMLINK_LEN
) {
1610 info
->symlink
= kmemdup(symname
, len
, GFP_KERNEL
);
1611 if (!info
->symlink
) {
1615 inode
->i_op
= &shmem_short_symlink_operations
;
1617 error
= shmem_getpage(inode
, 0, &page
, SGP_WRITE
, NULL
);
1622 inode
->i_mapping
->a_ops
= &shmem_aops
;
1623 inode
->i_op
= &shmem_symlink_inode_operations
;
1624 kaddr
= kmap_atomic(page
, KM_USER0
);
1625 memcpy(kaddr
, symname
, len
);
1626 kunmap_atomic(kaddr
, KM_USER0
);
1627 set_page_dirty(page
);
1629 page_cache_release(page
);
1631 dir
->i_size
+= BOGO_DIRENT_SIZE
;
1632 dir
->i_ctime
= dir
->i_mtime
= CURRENT_TIME
;
1633 d_instantiate(dentry
, inode
);
1638 static void *shmem_follow_short_symlink(struct dentry
*dentry
, struct nameidata
*nd
)
1640 nd_set_link(nd
, SHMEM_I(dentry
->d_inode
)->symlink
);
1644 static void *shmem_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1646 struct page
*page
= NULL
;
1647 int error
= shmem_getpage(dentry
->d_inode
, 0, &page
, SGP_READ
, NULL
);
1648 nd_set_link(nd
, error
? ERR_PTR(error
) : kmap(page
));
1654 static void shmem_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
1656 if (!IS_ERR(nd_get_link(nd
))) {
1657 struct page
*page
= cookie
;
1659 mark_page_accessed(page
);
1660 page_cache_release(page
);
1664 #ifdef CONFIG_TMPFS_XATTR
1666 * Superblocks without xattr inode operations may get some security.* xattr
1667 * support from the LSM "for free". As soon as we have any other xattrs
1668 * like ACLs, we also need to implement the security.* handlers at
1669 * filesystem level, though.
1672 static int shmem_xattr_get(struct dentry
*dentry
, const char *name
,
1673 void *buffer
, size_t size
)
1675 struct shmem_inode_info
*info
;
1676 struct shmem_xattr
*xattr
;
1679 info
= SHMEM_I(dentry
->d_inode
);
1681 spin_lock(&info
->lock
);
1682 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
1683 if (strcmp(name
, xattr
->name
))
1688 if (size
< xattr
->size
)
1691 memcpy(buffer
, xattr
->value
, xattr
->size
);
1695 spin_unlock(&info
->lock
);
1699 static int shmem_xattr_set(struct dentry
*dentry
, const char *name
,
1700 const void *value
, size_t size
, int flags
)
1702 struct inode
*inode
= dentry
->d_inode
;
1703 struct shmem_inode_info
*info
= SHMEM_I(inode
);
1704 struct shmem_xattr
*xattr
;
1705 struct shmem_xattr
*new_xattr
= NULL
;
1709 /* value == NULL means remove */
1712 len
= sizeof(*new_xattr
) + size
;
1713 if (len
<= sizeof(*new_xattr
))
1716 new_xattr
= kmalloc(len
, GFP_KERNEL
);
1720 new_xattr
->name
= kstrdup(name
, GFP_KERNEL
);
1721 if (!new_xattr
->name
) {
1726 new_xattr
->size
= size
;
1727 memcpy(new_xattr
->value
, value
, size
);
1730 spin_lock(&info
->lock
);
1731 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
1732 if (!strcmp(name
, xattr
->name
)) {
1733 if (flags
& XATTR_CREATE
) {
1736 } else if (new_xattr
) {
1737 list_replace(&xattr
->list
, &new_xattr
->list
);
1739 list_del(&xattr
->list
);
1744 if (flags
& XATTR_REPLACE
) {
1748 list_add(&new_xattr
->list
, &info
->xattr_list
);
1752 spin_unlock(&info
->lock
);
1759 static const struct xattr_handler
*shmem_xattr_handlers
[] = {
1760 #ifdef CONFIG_TMPFS_POSIX_ACL
1761 &generic_acl_access_handler
,
1762 &generic_acl_default_handler
,
1767 static int shmem_xattr_validate(const char *name
)
1769 struct { const char *prefix
; size_t len
; } arr
[] = {
1770 { XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
},
1771 { XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
}
1775 for (i
= 0; i
< ARRAY_SIZE(arr
); i
++) {
1776 size_t preflen
= arr
[i
].len
;
1777 if (strncmp(name
, arr
[i
].prefix
, preflen
) == 0) {
1786 static ssize_t
shmem_getxattr(struct dentry
*dentry
, const char *name
,
1787 void *buffer
, size_t size
)
1792 * If this is a request for a synthetic attribute in the system.*
1793 * namespace use the generic infrastructure to resolve a handler
1794 * for it via sb->s_xattr.
1796 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
1797 return generic_getxattr(dentry
, name
, buffer
, size
);
1799 err
= shmem_xattr_validate(name
);
1803 return shmem_xattr_get(dentry
, name
, buffer
, size
);
1806 static int shmem_setxattr(struct dentry
*dentry
, const char *name
,
1807 const void *value
, size_t size
, int flags
)
1812 * If this is a request for a synthetic attribute in the system.*
1813 * namespace use the generic infrastructure to resolve a handler
1814 * for it via sb->s_xattr.
1816 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
1817 return generic_setxattr(dentry
, name
, value
, size
, flags
);
1819 err
= shmem_xattr_validate(name
);
1824 value
= ""; /* empty EA, do not remove */
1826 return shmem_xattr_set(dentry
, name
, value
, size
, flags
);
1830 static int shmem_removexattr(struct dentry
*dentry
, const char *name
)
1835 * If this is a request for a synthetic attribute in the system.*
1836 * namespace use the generic infrastructure to resolve a handler
1837 * for it via sb->s_xattr.
1839 if (!strncmp(name
, XATTR_SYSTEM_PREFIX
, XATTR_SYSTEM_PREFIX_LEN
))
1840 return generic_removexattr(dentry
, name
);
1842 err
= shmem_xattr_validate(name
);
1846 return shmem_xattr_set(dentry
, name
, NULL
, 0, XATTR_REPLACE
);
1849 static bool xattr_is_trusted(const char *name
)
1851 return !strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
);
1854 static ssize_t
shmem_listxattr(struct dentry
*dentry
, char *buffer
, size_t size
)
1856 bool trusted
= capable(CAP_SYS_ADMIN
);
1857 struct shmem_xattr
*xattr
;
1858 struct shmem_inode_info
*info
;
1861 info
= SHMEM_I(dentry
->d_inode
);
1863 spin_lock(&info
->lock
);
1864 list_for_each_entry(xattr
, &info
->xattr_list
, list
) {
1867 /* skip "trusted." attributes for unprivileged callers */
1868 if (!trusted
&& xattr_is_trusted(xattr
->name
))
1871 len
= strlen(xattr
->name
) + 1;
1878 memcpy(buffer
, xattr
->name
, len
);
1882 spin_unlock(&info
->lock
);
1886 #endif /* CONFIG_TMPFS_XATTR */
1888 static const struct inode_operations shmem_short_symlink_operations
= {
1889 .readlink
= generic_readlink
,
1890 .follow_link
= shmem_follow_short_symlink
,
1891 #ifdef CONFIG_TMPFS_XATTR
1892 .setxattr
= shmem_setxattr
,
1893 .getxattr
= shmem_getxattr
,
1894 .listxattr
= shmem_listxattr
,
1895 .removexattr
= shmem_removexattr
,
1899 static const struct inode_operations shmem_symlink_inode_operations
= {
1900 .readlink
= generic_readlink
,
1901 .follow_link
= shmem_follow_link
,
1902 .put_link
= shmem_put_link
,
1903 #ifdef CONFIG_TMPFS_XATTR
1904 .setxattr
= shmem_setxattr
,
1905 .getxattr
= shmem_getxattr
,
1906 .listxattr
= shmem_listxattr
,
1907 .removexattr
= shmem_removexattr
,
1911 static struct dentry
*shmem_get_parent(struct dentry
*child
)
1913 return ERR_PTR(-ESTALE
);
1916 static int shmem_match(struct inode
*ino
, void *vfh
)
1920 inum
= (inum
<< 32) | fh
[1];
1921 return ino
->i_ino
== inum
&& fh
[0] == ino
->i_generation
;
1924 static struct dentry
*shmem_fh_to_dentry(struct super_block
*sb
,
1925 struct fid
*fid
, int fh_len
, int fh_type
)
1927 struct inode
*inode
;
1928 struct dentry
*dentry
= NULL
;
1929 u64 inum
= fid
->raw
[2];
1930 inum
= (inum
<< 32) | fid
->raw
[1];
1935 inode
= ilookup5(sb
, (unsigned long)(inum
+ fid
->raw
[0]),
1936 shmem_match
, fid
->raw
);
1938 dentry
= d_find_alias(inode
);
1945 static int shmem_encode_fh(struct dentry
*dentry
, __u32
*fh
, int *len
,
1948 struct inode
*inode
= dentry
->d_inode
;
1955 if (inode_unhashed(inode
)) {
1956 /* Unfortunately insert_inode_hash is not idempotent,
1957 * so as we hash inodes here rather than at creation
1958 * time, we need a lock to ensure we only try
1961 static DEFINE_SPINLOCK(lock
);
1963 if (inode_unhashed(inode
))
1964 __insert_inode_hash(inode
,
1965 inode
->i_ino
+ inode
->i_generation
);
1969 fh
[0] = inode
->i_generation
;
1970 fh
[1] = inode
->i_ino
;
1971 fh
[2] = ((__u64
)inode
->i_ino
) >> 32;
1977 static const struct export_operations shmem_export_ops
= {
1978 .get_parent
= shmem_get_parent
,
1979 .encode_fh
= shmem_encode_fh
,
1980 .fh_to_dentry
= shmem_fh_to_dentry
,
1983 static int shmem_parse_options(char *options
, struct shmem_sb_info
*sbinfo
,
1986 char *this_char
, *value
, *rest
;
1988 while (options
!= NULL
) {
1989 this_char
= options
;
1992 * NUL-terminate this option: unfortunately,
1993 * mount options form a comma-separated list,
1994 * but mpol's nodelist may also contain commas.
1996 options
= strchr(options
, ',');
1997 if (options
== NULL
)
2000 if (!isdigit(*options
)) {
2007 if ((value
= strchr(this_char
,'=')) != NULL
) {
2011 "tmpfs: No value for mount option '%s'\n",
2016 if (!strcmp(this_char
,"size")) {
2017 unsigned long long size
;
2018 size
= memparse(value
,&rest
);
2020 size
<<= PAGE_SHIFT
;
2021 size
*= totalram_pages
;
2027 sbinfo
->max_blocks
=
2028 DIV_ROUND_UP(size
, PAGE_CACHE_SIZE
);
2029 } else if (!strcmp(this_char
,"nr_blocks")) {
2030 sbinfo
->max_blocks
= memparse(value
, &rest
);
2033 } else if (!strcmp(this_char
,"nr_inodes")) {
2034 sbinfo
->max_inodes
= memparse(value
, &rest
);
2037 } else if (!strcmp(this_char
,"mode")) {
2040 sbinfo
->mode
= simple_strtoul(value
, &rest
, 8) & 07777;
2043 } else if (!strcmp(this_char
,"uid")) {
2046 sbinfo
->uid
= simple_strtoul(value
, &rest
, 0);
2049 } else if (!strcmp(this_char
,"gid")) {
2052 sbinfo
->gid
= simple_strtoul(value
, &rest
, 0);
2055 } else if (!strcmp(this_char
,"mpol")) {
2056 if (mpol_parse_str(value
, &sbinfo
->mpol
, 1))
2059 printk(KERN_ERR
"tmpfs: Bad mount option %s\n",
2067 printk(KERN_ERR
"tmpfs: Bad value '%s' for mount option '%s'\n",
2073 static int shmem_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
2075 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2076 struct shmem_sb_info config
= *sbinfo
;
2077 unsigned long inodes
;
2078 int error
= -EINVAL
;
2080 if (shmem_parse_options(data
, &config
, true))
2083 spin_lock(&sbinfo
->stat_lock
);
2084 inodes
= sbinfo
->max_inodes
- sbinfo
->free_inodes
;
2085 if (percpu_counter_compare(&sbinfo
->used_blocks
, config
.max_blocks
) > 0)
2087 if (config
.max_inodes
< inodes
)
2090 * Those tests disallow limited->unlimited while any are in use;
2091 * but we must separately disallow unlimited->limited, because
2092 * in that case we have no record of how much is already in use.
2094 if (config
.max_blocks
&& !sbinfo
->max_blocks
)
2096 if (config
.max_inodes
&& !sbinfo
->max_inodes
)
2100 sbinfo
->max_blocks
= config
.max_blocks
;
2101 sbinfo
->max_inodes
= config
.max_inodes
;
2102 sbinfo
->free_inodes
= config
.max_inodes
- inodes
;
2104 mpol_put(sbinfo
->mpol
);
2105 sbinfo
->mpol
= config
.mpol
; /* transfers initial ref */
2107 spin_unlock(&sbinfo
->stat_lock
);
2111 static int shmem_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
2113 struct shmem_sb_info
*sbinfo
= SHMEM_SB(vfs
->mnt_sb
);
2115 if (sbinfo
->max_blocks
!= shmem_default_max_blocks())
2116 seq_printf(seq
, ",size=%luk",
2117 sbinfo
->max_blocks
<< (PAGE_CACHE_SHIFT
- 10));
2118 if (sbinfo
->max_inodes
!= shmem_default_max_inodes())
2119 seq_printf(seq
, ",nr_inodes=%lu", sbinfo
->max_inodes
);
2120 if (sbinfo
->mode
!= (S_IRWXUGO
| S_ISVTX
))
2121 seq_printf(seq
, ",mode=%03o", sbinfo
->mode
);
2122 if (sbinfo
->uid
!= 0)
2123 seq_printf(seq
, ",uid=%u", sbinfo
->uid
);
2124 if (sbinfo
->gid
!= 0)
2125 seq_printf(seq
, ",gid=%u", sbinfo
->gid
);
2126 shmem_show_mpol(seq
, sbinfo
->mpol
);
2129 #endif /* CONFIG_TMPFS */
2131 static void shmem_put_super(struct super_block
*sb
)
2133 struct shmem_sb_info
*sbinfo
= SHMEM_SB(sb
);
2135 percpu_counter_destroy(&sbinfo
->used_blocks
);
2137 sb
->s_fs_info
= NULL
;
2140 int shmem_fill_super(struct super_block
*sb
, void *data
, int silent
)
2142 struct inode
*inode
;
2143 struct dentry
*root
;
2144 struct shmem_sb_info
*sbinfo
;
2147 /* Round up to L1_CACHE_BYTES to resist false sharing */
2148 sbinfo
= kzalloc(max((int)sizeof(struct shmem_sb_info
),
2149 L1_CACHE_BYTES
), GFP_KERNEL
);
2153 sbinfo
->mode
= S_IRWXUGO
| S_ISVTX
;
2154 sbinfo
->uid
= current_fsuid();
2155 sbinfo
->gid
= current_fsgid();
2156 sb
->s_fs_info
= sbinfo
;
2160 * Per default we only allow half of the physical ram per
2161 * tmpfs instance, limiting inodes to one per page of lowmem;
2162 * but the internal instance is left unlimited.
2164 if (!(sb
->s_flags
& MS_NOUSER
)) {
2165 sbinfo
->max_blocks
= shmem_default_max_blocks();
2166 sbinfo
->max_inodes
= shmem_default_max_inodes();
2167 if (shmem_parse_options(data
, sbinfo
, false)) {
2172 sb
->s_export_op
= &shmem_export_ops
;
2174 sb
->s_flags
|= MS_NOUSER
;
2177 spin_lock_init(&sbinfo
->stat_lock
);
2178 if (percpu_counter_init(&sbinfo
->used_blocks
, 0))
2180 sbinfo
->free_inodes
= sbinfo
->max_inodes
;
2182 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
2183 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
2184 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
2185 sb
->s_magic
= TMPFS_MAGIC
;
2186 sb
->s_op
= &shmem_ops
;
2187 sb
->s_time_gran
= 1;
2188 #ifdef CONFIG_TMPFS_XATTR
2189 sb
->s_xattr
= shmem_xattr_handlers
;
2191 #ifdef CONFIG_TMPFS_POSIX_ACL
2192 sb
->s_flags
|= MS_POSIXACL
;
2195 inode
= shmem_get_inode(sb
, NULL
, S_IFDIR
| sbinfo
->mode
, 0, VM_NORESERVE
);
2198 inode
->i_uid
= sbinfo
->uid
;
2199 inode
->i_gid
= sbinfo
->gid
;
2200 root
= d_alloc_root(inode
);
2209 shmem_put_super(sb
);
2213 static struct kmem_cache
*shmem_inode_cachep
;
2215 static struct inode
*shmem_alloc_inode(struct super_block
*sb
)
2217 struct shmem_inode_info
*info
;
2218 info
= kmem_cache_alloc(shmem_inode_cachep
, GFP_KERNEL
);
2221 return &info
->vfs_inode
;
2224 static void shmem_destroy_callback(struct rcu_head
*head
)
2226 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
2227 INIT_LIST_HEAD(&inode
->i_dentry
);
2228 kmem_cache_free(shmem_inode_cachep
, SHMEM_I(inode
));
2231 static void shmem_destroy_inode(struct inode
*inode
)
2233 if ((inode
->i_mode
& S_IFMT
) == S_IFREG
)
2234 mpol_free_shared_policy(&SHMEM_I(inode
)->policy
);
2235 call_rcu(&inode
->i_rcu
, shmem_destroy_callback
);
2238 static void shmem_init_inode(void *foo
)
2240 struct shmem_inode_info
*info
= foo
;
2241 inode_init_once(&info
->vfs_inode
);
2244 static int shmem_init_inodecache(void)
2246 shmem_inode_cachep
= kmem_cache_create("shmem_inode_cache",
2247 sizeof(struct shmem_inode_info
),
2248 0, SLAB_PANIC
, shmem_init_inode
);
2252 static void shmem_destroy_inodecache(void)
2254 kmem_cache_destroy(shmem_inode_cachep
);
2257 static const struct address_space_operations shmem_aops
= {
2258 .writepage
= shmem_writepage
,
2259 .set_page_dirty
= __set_page_dirty_no_writeback
,
2261 .write_begin
= shmem_write_begin
,
2262 .write_end
= shmem_write_end
,
2264 .migratepage
= migrate_page
,
2265 .error_remove_page
= generic_error_remove_page
,
2268 static const struct file_operations shmem_file_operations
= {
2271 .llseek
= generic_file_llseek
,
2272 .read
= do_sync_read
,
2273 .write
= do_sync_write
,
2274 .aio_read
= shmem_file_aio_read
,
2275 .aio_write
= generic_file_aio_write
,
2276 .fsync
= noop_fsync
,
2277 .splice_read
= shmem_file_splice_read
,
2278 .splice_write
= generic_file_splice_write
,
2282 static const struct inode_operations shmem_inode_operations
= {
2283 .setattr
= shmem_setattr
,
2284 .truncate_range
= shmem_truncate_range
,
2285 #ifdef CONFIG_TMPFS_XATTR
2286 .setxattr
= shmem_setxattr
,
2287 .getxattr
= shmem_getxattr
,
2288 .listxattr
= shmem_listxattr
,
2289 .removexattr
= shmem_removexattr
,
2293 static const struct inode_operations shmem_dir_inode_operations
= {
2295 .create
= shmem_create
,
2296 .lookup
= simple_lookup
,
2298 .unlink
= shmem_unlink
,
2299 .symlink
= shmem_symlink
,
2300 .mkdir
= shmem_mkdir
,
2301 .rmdir
= shmem_rmdir
,
2302 .mknod
= shmem_mknod
,
2303 .rename
= shmem_rename
,
2305 #ifdef CONFIG_TMPFS_XATTR
2306 .setxattr
= shmem_setxattr
,
2307 .getxattr
= shmem_getxattr
,
2308 .listxattr
= shmem_listxattr
,
2309 .removexattr
= shmem_removexattr
,
2311 #ifdef CONFIG_TMPFS_POSIX_ACL
2312 .setattr
= shmem_setattr
,
2316 static const struct inode_operations shmem_special_inode_operations
= {
2317 #ifdef CONFIG_TMPFS_XATTR
2318 .setxattr
= shmem_setxattr
,
2319 .getxattr
= shmem_getxattr
,
2320 .listxattr
= shmem_listxattr
,
2321 .removexattr
= shmem_removexattr
,
2323 #ifdef CONFIG_TMPFS_POSIX_ACL
2324 .setattr
= shmem_setattr
,
2328 static const struct super_operations shmem_ops
= {
2329 .alloc_inode
= shmem_alloc_inode
,
2330 .destroy_inode
= shmem_destroy_inode
,
2332 .statfs
= shmem_statfs
,
2333 .remount_fs
= shmem_remount_fs
,
2334 .show_options
= shmem_show_options
,
2336 .evict_inode
= shmem_evict_inode
,
2337 .drop_inode
= generic_delete_inode
,
2338 .put_super
= shmem_put_super
,
2341 static const struct vm_operations_struct shmem_vm_ops
= {
2342 .fault
= shmem_fault
,
2344 .set_policy
= shmem_set_policy
,
2345 .get_policy
= shmem_get_policy
,
2349 static struct dentry
*shmem_mount(struct file_system_type
*fs_type
,
2350 int flags
, const char *dev_name
, void *data
)
2352 return mount_nodev(fs_type
, flags
, data
, shmem_fill_super
);
2355 static struct file_system_type shmem_fs_type
= {
2356 .owner
= THIS_MODULE
,
2358 .mount
= shmem_mount
,
2359 .kill_sb
= kill_litter_super
,
2362 int __init
shmem_init(void)
2366 error
= bdi_init(&shmem_backing_dev_info
);
2370 error
= shmem_init_inodecache();
2374 error
= register_filesystem(&shmem_fs_type
);
2376 printk(KERN_ERR
"Could not register tmpfs\n");
2380 shm_mnt
= vfs_kern_mount(&shmem_fs_type
, MS_NOUSER
,
2381 shmem_fs_type
.name
, NULL
);
2382 if (IS_ERR(shm_mnt
)) {
2383 error
= PTR_ERR(shm_mnt
);
2384 printk(KERN_ERR
"Could not kern_mount tmpfs\n");
2390 unregister_filesystem(&shmem_fs_type
);
2392 shmem_destroy_inodecache();
2394 bdi_destroy(&shmem_backing_dev_info
);
2396 shm_mnt
= ERR_PTR(error
);
2400 #else /* !CONFIG_SHMEM */
2403 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2405 * This is intended for small system where the benefits of the full
2406 * shmem code (swap-backed and resource-limited) are outweighed by
2407 * their complexity. On systems without swap this code should be
2408 * effectively equivalent, but much lighter weight.
2411 #include <linux/ramfs.h>
2413 static struct file_system_type shmem_fs_type
= {
2415 .mount
= ramfs_mount
,
2416 .kill_sb
= kill_litter_super
,
2419 int __init
shmem_init(void)
2421 BUG_ON(register_filesystem(&shmem_fs_type
) != 0);
2423 shm_mnt
= kern_mount(&shmem_fs_type
);
2424 BUG_ON(IS_ERR(shm_mnt
));
2429 int shmem_unuse(swp_entry_t swap
, struct page
*page
)
2434 int shmem_lock(struct file
*file
, int lock
, struct user_struct
*user
)
2439 void shmem_truncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
2441 truncate_inode_pages_range(inode
->i_mapping
, lstart
, lend
);
2443 EXPORT_SYMBOL_GPL(shmem_truncate_range
);
2445 #define shmem_vm_ops generic_file_vm_ops
2446 #define shmem_file_operations ramfs_file_operations
2447 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2448 #define shmem_acct_size(flags, size) 0
2449 #define shmem_unacct_size(flags, size) do {} while (0)
2451 #endif /* CONFIG_SHMEM */
2456 * shmem_file_setup - get an unlinked file living in tmpfs
2457 * @name: name for dentry (to be seen in /proc/<pid>/maps
2458 * @size: size to be set for the file
2459 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2461 struct file
*shmem_file_setup(const char *name
, loff_t size
, unsigned long flags
)
2465 struct inode
*inode
;
2467 struct dentry
*root
;
2470 if (IS_ERR(shm_mnt
))
2471 return (void *)shm_mnt
;
2473 if (size
< 0 || size
> MAX_LFS_FILESIZE
)
2474 return ERR_PTR(-EINVAL
);
2476 if (shmem_acct_size(flags
, size
))
2477 return ERR_PTR(-ENOMEM
);
2481 this.len
= strlen(name
);
2482 this.hash
= 0; /* will go */
2483 root
= shm_mnt
->mnt_root
;
2484 path
.dentry
= d_alloc(root
, &this);
2487 path
.mnt
= mntget(shm_mnt
);
2490 inode
= shmem_get_inode(root
->d_sb
, NULL
, S_IFREG
| S_IRWXUGO
, 0, flags
);
2494 d_instantiate(path
.dentry
, inode
);
2495 inode
->i_size
= size
;
2496 inode
->i_nlink
= 0; /* It is unlinked */
2498 error
= ramfs_nommu_expand_for_mapping(inode
, size
);
2504 file
= alloc_file(&path
, FMODE_WRITE
| FMODE_READ
,
2505 &shmem_file_operations
);
2514 shmem_unacct_size(flags
, size
);
2515 return ERR_PTR(error
);
2517 EXPORT_SYMBOL_GPL(shmem_file_setup
);
2520 * shmem_zero_setup - setup a shared anonymous mapping
2521 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2523 int shmem_zero_setup(struct vm_area_struct
*vma
)
2526 loff_t size
= vma
->vm_end
- vma
->vm_start
;
2528 file
= shmem_file_setup("dev/zero", size
, vma
->vm_flags
);
2530 return PTR_ERR(file
);
2534 vma
->vm_file
= file
;
2535 vma
->vm_ops
= &shmem_vm_ops
;
2536 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
2541 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2542 * @mapping: the page's address_space
2543 * @index: the page index
2544 * @gfp: the page allocator flags to use if allocating
2546 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2547 * with any new page allocations done using the specified allocation flags.
2548 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2549 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2550 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2552 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2553 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2555 struct page
*shmem_read_mapping_page_gfp(struct address_space
*mapping
,
2556 pgoff_t index
, gfp_t gfp
)
2559 struct inode
*inode
= mapping
->host
;
2563 BUG_ON(mapping
->a_ops
!= &shmem_aops
);
2564 error
= shmem_getpage_gfp(inode
, index
, &page
, SGP_CACHE
, gfp
, NULL
);
2566 page
= ERR_PTR(error
);
2572 * The tiny !SHMEM case uses ramfs without swap
2574 return read_cache_page_gfp(mapping
, index
, gfp
);
2577 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp
);