2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 * This file is released under the GPL.
21 * This virtual memory filesystem is heavily based on the ramfs. It
22 * extends ramfs by the ability to use swap and honor resource limits
23 * which makes it a completely usable filesystem.
26 #include <linux/module.h>
27 #include <linux/init.h>
29 #include <linux/xattr.h>
30 #include <linux/exportfs.h>
31 #include <linux/generic_acl.h>
33 #include <linux/mman.h>
34 #include <linux/file.h>
35 #include <linux/swap.h>
36 #include <linux/pagemap.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/backing-dev.h>
40 #include <linux/shmem_fs.h>
41 #include <linux/mount.h>
42 #include <linux/writeback.h>
43 #include <linux/vfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/security.h>
46 #include <linux/swapops.h>
47 #include <linux/mempolicy.h>
48 #include <linux/namei.h>
49 #include <linux/ctype.h>
50 #include <linux/migrate.h>
51 #include <linux/highmem.h>
52 #include <linux/seq_file.h>
54 #include <asm/uaccess.h>
55 #include <asm/div64.h>
56 #include <asm/pgtable.h>
58 /* This magic number is used in glibc for posix shared memory */
59 #define TMPFS_MAGIC 0x01021994
61 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
62 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
63 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
65 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
66 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
68 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
70 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
71 #define SHMEM_PAGEIN VM_READ
72 #define SHMEM_TRUNCATE VM_WRITE
74 /* Definition to limit shmem_truncate's steps between cond_rescheds */
75 #define LATENCY_LIMIT 64
77 /* Pretend that each entry is of this size in directory's i_size */
78 #define BOGO_DIRENT_SIZE 20
80 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
82 SGP_READ, /* don't exceed i_size, don't allocate page */
83 SGP_CACHE, /* don't exceed i_size, may allocate page */
84 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
85 SGP_WRITE, /* may exceed i_size, may allocate page */
89 static unsigned long shmem_default_max_blocks(void)
91 return totalram_pages / 2;
94 static unsigned long shmem_default_max_inodes(void)
96 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
100 static int shmem_getpage(struct inode *inode, unsigned long idx,
101 struct page **pagep, enum sgp_type sgp, int *type);
103 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
106 * The above definition of ENTRIES_PER_PAGE, and the use of
107 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
108 * might be reconsidered if it ever diverges from PAGE_SIZE.
110 * Mobility flags are masked out as swap vectors cannot move
112 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
113 PAGE_CACHE_SHIFT-PAGE_SHIFT);
116 static inline void shmem_dir_free(struct page *page)
118 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
121 static struct page **shmem_dir_map(struct page *page)
123 return (struct page **)kmap_atomic(page, KM_USER0);
126 static inline void shmem_dir_unmap(struct page **dir)
128 kunmap_atomic(dir, KM_USER0);
131 static swp_entry_t *shmem_swp_map(struct page *page)
133 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
136 static inline void shmem_swp_balance_unmap(void)
139 * When passing a pointer to an i_direct entry, to code which
140 * also handles indirect entries and so will shmem_swp_unmap,
141 * we must arrange for the preempt count to remain in balance.
142 * What kmap_atomic of a lowmem page does depends on config
143 * and architecture, so pretend to kmap_atomic some lowmem page.
145 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
148 static inline void shmem_swp_unmap(swp_entry_t *entry)
150 kunmap_atomic(entry, KM_USER1);
153 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
155 return sb->s_fs_info;
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
164 static inline int shmem_acct_size(unsigned long flags, loff_t size)
166 return (flags & VM_ACCOUNT)?
167 security_vm_enough_memory(VM_ACCT(size)): 0;
170 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
172 if (flags & VM_ACCOUNT)
173 vm_unacct_memory(VM_ACCT(size));
177 * ... whereas tmpfs objects are accounted incrementally as
178 * pages are allocated, in order to allow huge sparse files.
179 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
180 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
182 static inline int shmem_acct_block(unsigned long flags)
184 return (flags & VM_ACCOUNT)?
185 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
188 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
190 if (!(flags & VM_ACCOUNT))
191 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
194 static const struct super_operations shmem_ops;
195 static const struct address_space_operations shmem_aops;
196 static const struct file_operations shmem_file_operations;
197 static const struct inode_operations shmem_inode_operations;
198 static const struct inode_operations shmem_dir_inode_operations;
199 static const struct inode_operations shmem_special_inode_operations;
200 static struct vm_operations_struct shmem_vm_ops;
202 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
203 .ra_pages = 0, /* No readahead */
204 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
205 .unplug_io_fn = default_unplug_io_fn,
208 static LIST_HEAD(shmem_swaplist);
209 static DEFINE_MUTEX(shmem_swaplist_mutex);
211 static void shmem_free_blocks(struct inode *inode, long pages)
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
214 if (sbinfo->max_blocks) {
215 spin_lock(&sbinfo->stat_lock);
216 sbinfo->free_blocks += pages;
217 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
218 spin_unlock(&sbinfo->stat_lock);
222 static int shmem_reserve_inode(struct super_block *sb)
224 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
225 if (sbinfo->max_inodes) {
226 spin_lock(&sbinfo->stat_lock);
227 if (!sbinfo->free_inodes) {
228 spin_unlock(&sbinfo->stat_lock);
231 sbinfo->free_inodes--;
232 spin_unlock(&sbinfo->stat_lock);
237 static void shmem_free_inode(struct super_block *sb)
239 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
240 if (sbinfo->max_inodes) {
241 spin_lock(&sbinfo->stat_lock);
242 sbinfo->free_inodes++;
243 spin_unlock(&sbinfo->stat_lock);
248 * shmem_recalc_inode - recalculate the size of an inode
249 * @inode: inode to recalc
251 * We have to calculate the free blocks since the mm can drop
252 * undirtied hole pages behind our back.
254 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
255 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
257 * It has to be called with the spinlock held.
259 static void shmem_recalc_inode(struct inode *inode)
261 struct shmem_inode_info *info = SHMEM_I(inode);
264 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
266 info->alloced -= freed;
267 shmem_unacct_blocks(info->flags, freed);
268 shmem_free_blocks(inode, freed);
273 * shmem_swp_entry - find the swap vector position in the info structure
274 * @info: info structure for the inode
275 * @index: index of the page to find
276 * @page: optional page to add to the structure. Has to be preset to
279 * If there is no space allocated yet it will return NULL when
280 * page is NULL, else it will use the page for the needed block,
281 * setting it to NULL on return to indicate that it has been used.
283 * The swap vector is organized the following way:
285 * There are SHMEM_NR_DIRECT entries directly stored in the
286 * shmem_inode_info structure. So small files do not need an addional
289 * For pages with index > SHMEM_NR_DIRECT there is the pointer
290 * i_indirect which points to a page which holds in the first half
291 * doubly indirect blocks, in the second half triple indirect blocks:
293 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
294 * following layout (for SHMEM_NR_DIRECT == 16):
296 * i_indirect -> dir --> 16-19
309 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
311 unsigned long offset;
315 if (index < SHMEM_NR_DIRECT) {
316 shmem_swp_balance_unmap();
317 return info->i_direct+index;
319 if (!info->i_indirect) {
321 info->i_indirect = *page;
324 return NULL; /* need another page */
327 index -= SHMEM_NR_DIRECT;
328 offset = index % ENTRIES_PER_PAGE;
329 index /= ENTRIES_PER_PAGE;
330 dir = shmem_dir_map(info->i_indirect);
332 if (index >= ENTRIES_PER_PAGE/2) {
333 index -= ENTRIES_PER_PAGE/2;
334 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
335 index %= ENTRIES_PER_PAGE;
342 shmem_dir_unmap(dir);
343 return NULL; /* need another page */
345 shmem_dir_unmap(dir);
346 dir = shmem_dir_map(subdir);
352 if (!page || !(subdir = *page)) {
353 shmem_dir_unmap(dir);
354 return NULL; /* need a page */
359 shmem_dir_unmap(dir);
360 return shmem_swp_map(subdir) + offset;
363 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
365 long incdec = value? 1: -1;
368 info->swapped += incdec;
369 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
370 struct page *page = kmap_atomic_to_page(entry);
371 set_page_private(page, page_private(page) + incdec);
376 * shmem_swp_alloc - get the position of the swap entry for the page.
377 * @info: info structure for the inode
378 * @index: index of the page to find
379 * @sgp: check and recheck i_size? skip allocation?
381 * If the entry does not exist, allocate it.
383 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
385 struct inode *inode = &info->vfs_inode;
386 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
387 struct page *page = NULL;
390 if (sgp != SGP_WRITE &&
391 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
392 return ERR_PTR(-EINVAL);
394 while (!(entry = shmem_swp_entry(info, index, &page))) {
396 return shmem_swp_map(ZERO_PAGE(0));
398 * Test free_blocks against 1 not 0, since we have 1 data
399 * page (and perhaps indirect index pages) yet to allocate:
400 * a waste to allocate index if we cannot allocate data.
402 if (sbinfo->max_blocks) {
403 spin_lock(&sbinfo->stat_lock);
404 if (sbinfo->free_blocks <= 1) {
405 spin_unlock(&sbinfo->stat_lock);
406 return ERR_PTR(-ENOSPC);
408 sbinfo->free_blocks--;
409 inode->i_blocks += BLOCKS_PER_PAGE;
410 spin_unlock(&sbinfo->stat_lock);
413 spin_unlock(&info->lock);
414 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
416 set_page_private(page, 0);
417 spin_lock(&info->lock);
420 shmem_free_blocks(inode, 1);
421 return ERR_PTR(-ENOMEM);
423 if (sgp != SGP_WRITE &&
424 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
425 entry = ERR_PTR(-EINVAL);
428 if (info->next_index <= index)
429 info->next_index = index + 1;
432 /* another task gave its page, or truncated the file */
433 shmem_free_blocks(inode, 1);
434 shmem_dir_free(page);
436 if (info->next_index <= index && !IS_ERR(entry))
437 info->next_index = index + 1;
442 * shmem_free_swp - free some swap entries in a directory
443 * @dir: pointer to the directory
444 * @edir: pointer after last entry of the directory
445 * @punch_lock: pointer to spinlock when needed for the holepunch case
447 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
448 spinlock_t *punch_lock)
450 spinlock_t *punch_unlock = NULL;
454 for (ptr = dir; ptr < edir; ptr++) {
456 if (unlikely(punch_lock)) {
457 punch_unlock = punch_lock;
459 spin_lock(punch_unlock);
463 free_swap_and_cache(*ptr);
464 *ptr = (swp_entry_t){0};
469 spin_unlock(punch_unlock);
473 static int shmem_map_and_free_swp(struct page *subdir, int offset,
474 int limit, struct page ***dir, spinlock_t *punch_lock)
479 ptr = shmem_swp_map(subdir);
480 for (; offset < limit; offset += LATENCY_LIMIT) {
481 int size = limit - offset;
482 if (size > LATENCY_LIMIT)
483 size = LATENCY_LIMIT;
484 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
486 if (need_resched()) {
487 shmem_swp_unmap(ptr);
489 shmem_dir_unmap(*dir);
493 ptr = shmem_swp_map(subdir);
496 shmem_swp_unmap(ptr);
500 static void shmem_free_pages(struct list_head *next)
506 page = container_of(next, struct page, lru);
508 shmem_dir_free(page);
510 if (freed >= LATENCY_LIMIT) {
517 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
519 struct shmem_inode_info *info = SHMEM_I(inode);
524 unsigned long diroff;
530 LIST_HEAD(pages_to_free);
531 long nr_pages_to_free = 0;
532 long nr_swaps_freed = 0;
536 spinlock_t *needs_lock;
537 spinlock_t *punch_lock;
538 unsigned long upper_limit;
540 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
541 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
542 if (idx >= info->next_index)
545 spin_lock(&info->lock);
546 info->flags |= SHMEM_TRUNCATE;
547 if (likely(end == (loff_t) -1)) {
548 limit = info->next_index;
549 upper_limit = SHMEM_MAX_INDEX;
550 info->next_index = idx;
554 if (end + 1 >= inode->i_size) { /* we may free a little more */
555 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
557 upper_limit = SHMEM_MAX_INDEX;
559 limit = (end + 1) >> PAGE_CACHE_SHIFT;
562 needs_lock = &info->lock;
566 topdir = info->i_indirect;
567 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
568 info->i_indirect = NULL;
570 list_add(&topdir->lru, &pages_to_free);
572 spin_unlock(&info->lock);
574 if (info->swapped && idx < SHMEM_NR_DIRECT) {
575 ptr = info->i_direct;
577 if (size > SHMEM_NR_DIRECT)
578 size = SHMEM_NR_DIRECT;
579 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
583 * If there are no indirect blocks or we are punching a hole
584 * below indirect blocks, nothing to be done.
586 if (!topdir || limit <= SHMEM_NR_DIRECT)
590 * The truncation case has already dropped info->lock, and we're safe
591 * because i_size and next_index have already been lowered, preventing
592 * access beyond. But in the punch_hole case, we still need to take
593 * the lock when updating the swap directory, because there might be
594 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
595 * shmem_writepage. However, whenever we find we can remove a whole
596 * directory page (not at the misaligned start or end of the range),
597 * we first NULLify its pointer in the level above, and then have no
598 * need to take the lock when updating its contents: needs_lock and
599 * punch_lock (either pointing to info->lock or NULL) manage this.
602 upper_limit -= SHMEM_NR_DIRECT;
603 limit -= SHMEM_NR_DIRECT;
604 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
605 offset = idx % ENTRIES_PER_PAGE;
608 dir = shmem_dir_map(topdir);
609 stage = ENTRIES_PER_PAGEPAGE/2;
610 if (idx < ENTRIES_PER_PAGEPAGE/2) {
612 diroff = idx/ENTRIES_PER_PAGE;
614 dir += ENTRIES_PER_PAGE/2;
615 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
617 stage += ENTRIES_PER_PAGEPAGE;
620 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
621 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
622 if (!diroff && !offset && upper_limit >= stage) {
624 spin_lock(needs_lock);
626 spin_unlock(needs_lock);
631 list_add(&middir->lru, &pages_to_free);
633 shmem_dir_unmap(dir);
634 dir = shmem_dir_map(middir);
642 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
643 if (unlikely(idx == stage)) {
644 shmem_dir_unmap(dir);
645 dir = shmem_dir_map(topdir) +
646 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
649 idx += ENTRIES_PER_PAGEPAGE;
653 stage = idx + ENTRIES_PER_PAGEPAGE;
656 needs_lock = &info->lock;
657 if (upper_limit >= stage) {
659 spin_lock(needs_lock);
661 spin_unlock(needs_lock);
666 list_add(&middir->lru, &pages_to_free);
668 shmem_dir_unmap(dir);
670 dir = shmem_dir_map(middir);
673 punch_lock = needs_lock;
674 subdir = dir[diroff];
675 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
677 spin_lock(needs_lock);
679 spin_unlock(needs_lock);
684 list_add(&subdir->lru, &pages_to_free);
686 if (subdir && page_private(subdir) /* has swap entries */) {
688 if (size > ENTRIES_PER_PAGE)
689 size = ENTRIES_PER_PAGE;
690 freed = shmem_map_and_free_swp(subdir,
691 offset, size, &dir, punch_lock);
693 dir = shmem_dir_map(middir);
694 nr_swaps_freed += freed;
695 if (offset || punch_lock) {
696 spin_lock(&info->lock);
697 set_page_private(subdir,
698 page_private(subdir) - freed);
699 spin_unlock(&info->lock);
701 BUG_ON(page_private(subdir) != freed);
706 shmem_dir_unmap(dir);
708 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
710 * Call truncate_inode_pages again: racing shmem_unuse_inode
711 * may have swizzled a page in from swap since vmtruncate or
712 * generic_delete_inode did it, before we lowered next_index.
713 * Also, though shmem_getpage checks i_size before adding to
714 * cache, no recheck after: so fix the narrow window there too.
716 * Recalling truncate_inode_pages_range and unmap_mapping_range
717 * every time for punch_hole (which never got a chance to clear
718 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
719 * yet hardly ever necessary: try to optimize them out later.
721 truncate_inode_pages_range(inode->i_mapping, start, end);
723 unmap_mapping_range(inode->i_mapping, start,
727 spin_lock(&info->lock);
728 info->flags &= ~SHMEM_TRUNCATE;
729 info->swapped -= nr_swaps_freed;
730 if (nr_pages_to_free)
731 shmem_free_blocks(inode, nr_pages_to_free);
732 shmem_recalc_inode(inode);
733 spin_unlock(&info->lock);
736 * Empty swap vector directory pages to be freed?
738 if (!list_empty(&pages_to_free)) {
739 pages_to_free.prev->next = NULL;
740 shmem_free_pages(pages_to_free.next);
744 static void shmem_truncate(struct inode *inode)
746 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
749 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
751 struct inode *inode = dentry->d_inode;
752 struct page *page = NULL;
755 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
756 if (attr->ia_size < inode->i_size) {
758 * If truncating down to a partial page, then
759 * if that page is already allocated, hold it
760 * in memory until the truncation is over, so
761 * truncate_partial_page cannnot miss it were
762 * it assigned to swap.
764 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
765 (void) shmem_getpage(inode,
766 attr->ia_size>>PAGE_CACHE_SHIFT,
767 &page, SGP_READ, NULL);
772 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
773 * detect if any pages might have been added to cache
774 * after truncate_inode_pages. But we needn't bother
775 * if it's being fully truncated to zero-length: the
776 * nrpages check is efficient enough in that case.
779 struct shmem_inode_info *info = SHMEM_I(inode);
780 spin_lock(&info->lock);
781 info->flags &= ~SHMEM_PAGEIN;
782 spin_unlock(&info->lock);
787 error = inode_change_ok(inode, attr);
789 error = inode_setattr(inode, attr);
790 #ifdef CONFIG_TMPFS_POSIX_ACL
791 if (!error && (attr->ia_valid & ATTR_MODE))
792 error = generic_acl_chmod(inode, &shmem_acl_ops);
795 page_cache_release(page);
799 static void shmem_delete_inode(struct inode *inode)
801 struct shmem_inode_info *info = SHMEM_I(inode);
803 if (inode->i_op->truncate == shmem_truncate) {
804 truncate_inode_pages(inode->i_mapping, 0);
805 shmem_unacct_size(info->flags, inode->i_size);
807 shmem_truncate(inode);
808 if (!list_empty(&info->swaplist)) {
809 mutex_lock(&shmem_swaplist_mutex);
810 list_del_init(&info->swaplist);
811 mutex_unlock(&shmem_swaplist_mutex);
814 BUG_ON(inode->i_blocks);
815 shmem_free_inode(inode->i_sb);
819 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
823 for (ptr = dir; ptr < edir; ptr++) {
824 if (ptr->val == entry.val)
830 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
844 ptr = info->i_direct;
845 spin_lock(&info->lock);
846 if (!info->swapped) {
847 list_del_init(&info->swaplist);
850 limit = info->next_index;
852 if (size > SHMEM_NR_DIRECT)
853 size = SHMEM_NR_DIRECT;
854 offset = shmem_find_swp(entry, ptr, ptr+size);
857 if (!info->i_indirect)
860 dir = shmem_dir_map(info->i_indirect);
861 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
863 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
864 if (unlikely(idx == stage)) {
865 shmem_dir_unmap(dir-1);
866 if (cond_resched_lock(&info->lock)) {
867 /* check it has not been truncated */
868 if (limit > info->next_index) {
869 limit = info->next_index;
874 dir = shmem_dir_map(info->i_indirect) +
875 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
878 idx += ENTRIES_PER_PAGEPAGE;
882 stage = idx + ENTRIES_PER_PAGEPAGE;
884 shmem_dir_unmap(dir);
885 dir = shmem_dir_map(subdir);
888 if (subdir && page_private(subdir)) {
889 ptr = shmem_swp_map(subdir);
891 if (size > ENTRIES_PER_PAGE)
892 size = ENTRIES_PER_PAGE;
893 offset = shmem_find_swp(entry, ptr, ptr+size);
894 shmem_swp_unmap(ptr);
896 shmem_dir_unmap(dir);
902 shmem_dir_unmap(dir-1);
904 spin_unlock(&info->lock);
908 inode = igrab(&info->vfs_inode);
909 spin_unlock(&info->lock);
912 * Move _head_ to start search for next from here.
913 * But be careful: shmem_delete_inode checks list_empty without taking
914 * mutex, and there's an instant in list_move_tail when info->swaplist
915 * would appear empty, if it were the only one on shmem_swaplist. We
916 * could avoid doing it if inode NULL; or use this minor optimization.
918 if (shmem_swaplist.next != &info->swaplist)
919 list_move_tail(&shmem_swaplist, &info->swaplist);
920 mutex_unlock(&shmem_swaplist_mutex);
925 /* Precharge page while we can wait, compensate afterwards */
926 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
929 error = radix_tree_preload(GFP_KERNEL);
934 spin_lock(&info->lock);
935 ptr = shmem_swp_entry(info, idx, NULL);
936 if (ptr && ptr->val == entry.val)
937 error = add_to_page_cache(page, inode->i_mapping,
939 if (error == -EEXIST) {
940 struct page *filepage = find_get_page(inode->i_mapping, idx);
944 * There might be a more uptodate page coming down
945 * from a stacked writepage: forget our swappage if so.
947 if (PageUptodate(filepage))
949 page_cache_release(filepage);
953 delete_from_swap_cache(page);
954 set_page_dirty(page);
955 info->flags |= SHMEM_PAGEIN;
956 shmem_swp_set(info, ptr, 0);
958 error = 1; /* not an error, but entry was found */
961 shmem_swp_unmap(ptr);
962 spin_unlock(&info->lock);
963 radix_tree_preload_end();
965 mem_cgroup_uncharge_page(page);
968 page_cache_release(page);
969 iput(inode); /* allows for NULL */
974 * shmem_unuse() search for an eventually swapped out shmem page.
976 int shmem_unuse(swp_entry_t entry, struct page *page)
978 struct list_head *p, *next;
979 struct shmem_inode_info *info;
982 mutex_lock(&shmem_swaplist_mutex);
983 list_for_each_safe(p, next, &shmem_swaplist) {
984 info = list_entry(p, struct shmem_inode_info, swaplist);
985 found = shmem_unuse_inode(info, entry, page);
990 mutex_unlock(&shmem_swaplist_mutex);
991 out: return found; /* 0 or 1 or -ENOMEM */
995 * Move the page from the page cache to the swap cache.
997 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
999 struct shmem_inode_info *info;
1000 swp_entry_t *entry, swap;
1001 struct address_space *mapping;
1002 unsigned long index;
1003 struct inode *inode;
1005 BUG_ON(!PageLocked(page));
1006 mapping = page->mapping;
1007 index = page->index;
1008 inode = mapping->host;
1009 info = SHMEM_I(inode);
1010 if (info->flags & VM_LOCKED)
1012 if (!total_swap_pages)
1016 * shmem_backing_dev_info's capabilities prevent regular writeback or
1017 * sync from ever calling shmem_writepage; but a stacking filesystem
1018 * may use the ->writepage of its underlying filesystem, in which case
1019 * tmpfs should write out to swap only in response to memory pressure,
1020 * and not for pdflush or sync. However, in those cases, we do still
1021 * want to check if there's a redundant swappage to be discarded.
1023 if (wbc->for_reclaim)
1024 swap = get_swap_page();
1028 spin_lock(&info->lock);
1029 if (index >= info->next_index) {
1030 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1033 entry = shmem_swp_entry(info, index, NULL);
1036 * The more uptodate page coming down from a stacked
1037 * writepage should replace our old swappage.
1039 free_swap_and_cache(*entry);
1040 shmem_swp_set(info, entry, 0);
1042 shmem_recalc_inode(inode);
1044 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1045 remove_from_page_cache(page);
1046 shmem_swp_set(info, entry, swap.val);
1047 shmem_swp_unmap(entry);
1048 if (list_empty(&info->swaplist))
1049 inode = igrab(inode);
1052 spin_unlock(&info->lock);
1053 swap_duplicate(swap);
1054 BUG_ON(page_mapped(page));
1055 page_cache_release(page); /* pagecache ref */
1056 set_page_dirty(page);
1059 mutex_lock(&shmem_swaplist_mutex);
1060 /* move instead of add in case we're racing */
1061 list_move_tail(&info->swaplist, &shmem_swaplist);
1062 mutex_unlock(&shmem_swaplist_mutex);
1068 shmem_swp_unmap(entry);
1070 spin_unlock(&info->lock);
1073 set_page_dirty(page);
1074 if (wbc->for_reclaim)
1075 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1082 static int shmem_parse_mpol(char *value, unsigned short *policy,
1083 unsigned short *mode_flags, nodemask_t *policy_nodes)
1085 char *nodelist = strchr(value, ':');
1086 char *flags = strchr(value, '=');
1090 /* NUL-terminate policy string */
1092 if (nodelist_parse(nodelist, *policy_nodes))
1094 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
1099 if (!strcmp(value, "default")) {
1100 *policy = MPOL_DEFAULT;
1101 /* Don't allow a nodelist */
1104 } else if (!strcmp(value, "prefer")) {
1105 *policy = MPOL_PREFERRED;
1106 /* Insist on a nodelist of one node only */
1108 char *rest = nodelist;
1109 while (isdigit(*rest))
1114 } else if (!strcmp(value, "bind")) {
1115 *policy = MPOL_BIND;
1116 /* Insist on a nodelist */
1119 } else if (!strcmp(value, "interleave")) {
1120 *policy = MPOL_INTERLEAVE;
1122 * Default to online nodes with memory if no nodelist
1125 *policy_nodes = node_states[N_HIGH_MEMORY];
1131 /* Restore string for error message */
1137 static void shmem_show_mpol(struct seq_file *seq, unsigned short policy,
1138 unsigned short flags, const nodemask_t policy_nodes)
1140 char *policy_string;
1143 case MPOL_PREFERRED:
1144 policy_string = "prefer";
1147 policy_string = "bind";
1149 case MPOL_INTERLEAVE:
1150 policy_string = "interleave";
1157 seq_printf(seq, ",mpol=%s", policy_string);
1159 if (policy != MPOL_INTERLEAVE ||
1160 !nodes_equal(policy_nodes, node_states[N_HIGH_MEMORY])) {
1164 len = nodelist_scnprintf(buffer, sizeof(buffer), policy_nodes);
1165 if (len < sizeof(buffer))
1166 seq_printf(seq, ":%s", buffer);
1168 seq_printf(seq, ":?");
1171 #endif /* CONFIG_TMPFS */
1173 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1174 struct shmem_inode_info *info, unsigned long idx)
1176 struct vm_area_struct pvma;
1179 /* Create a pseudo vma that just contains the policy */
1181 pvma.vm_pgoff = idx;
1183 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1184 page = swapin_readahead(entry, gfp, &pvma, 0);
1185 mpol_free(pvma.vm_policy);
1189 static struct page *shmem_alloc_page(gfp_t gfp,
1190 struct shmem_inode_info *info, unsigned long idx)
1192 struct vm_area_struct pvma;
1195 /* Create a pseudo vma that just contains the policy */
1197 pvma.vm_pgoff = idx;
1199 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1200 page = alloc_page_vma(gfp, &pvma, 0);
1201 mpol_free(pvma.vm_policy);
1204 #else /* !CONFIG_NUMA */
1206 static inline int shmem_parse_mpol(char *value, unsigned short *policy,
1207 unsigned short *mode_flags, nodemask_t *policy_nodes)
1212 static inline void shmem_show_mpol(struct seq_file *seq, unsigned short policy,
1213 unsigned short flags, const nodemask_t policy_nodes)
1216 #endif /* CONFIG_TMPFS */
1218 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1219 struct shmem_inode_info *info, unsigned long idx)
1221 return swapin_readahead(entry, gfp, NULL, 0);
1224 static inline struct page *shmem_alloc_page(gfp_t gfp,
1225 struct shmem_inode_info *info, unsigned long idx)
1227 return alloc_page(gfp);
1229 #endif /* CONFIG_NUMA */
1232 * shmem_getpage - either get the page from swap or allocate a new one
1234 * If we allocate a new one we do not mark it dirty. That's up to the
1235 * vm. If we swap it in we mark it dirty since we also free the swap
1236 * entry since a page cannot live in both the swap and page cache
1238 static int shmem_getpage(struct inode *inode, unsigned long idx,
1239 struct page **pagep, enum sgp_type sgp, int *type)
1241 struct address_space *mapping = inode->i_mapping;
1242 struct shmem_inode_info *info = SHMEM_I(inode);
1243 struct shmem_sb_info *sbinfo;
1244 struct page *filepage = *pagep;
1245 struct page *swappage;
1251 if (idx >= SHMEM_MAX_INDEX)
1258 * Normally, filepage is NULL on entry, and either found
1259 * uptodate immediately, or allocated and zeroed, or read
1260 * in under swappage, which is then assigned to filepage.
1261 * But shmem_readpage (required for splice) passes in a locked
1262 * filepage, which may be found not uptodate by other callers
1263 * too, and may need to be copied from the swappage read in.
1267 filepage = find_lock_page(mapping, idx);
1268 if (filepage && PageUptodate(filepage))
1271 gfp = mapping_gfp_mask(mapping);
1274 * Try to preload while we can wait, to not make a habit of
1275 * draining atomic reserves; but don't latch on to this cpu.
1277 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1280 radix_tree_preload_end();
1283 spin_lock(&info->lock);
1284 shmem_recalc_inode(inode);
1285 entry = shmem_swp_alloc(info, idx, sgp);
1286 if (IS_ERR(entry)) {
1287 spin_unlock(&info->lock);
1288 error = PTR_ERR(entry);
1294 /* Look it up and read it in.. */
1295 swappage = lookup_swap_cache(swap);
1297 shmem_swp_unmap(entry);
1298 /* here we actually do the io */
1299 if (type && !(*type & VM_FAULT_MAJOR)) {
1300 __count_vm_event(PGMAJFAULT);
1301 *type |= VM_FAULT_MAJOR;
1303 spin_unlock(&info->lock);
1304 swappage = shmem_swapin(swap, gfp, info, idx);
1306 spin_lock(&info->lock);
1307 entry = shmem_swp_alloc(info, idx, sgp);
1309 error = PTR_ERR(entry);
1311 if (entry->val == swap.val)
1313 shmem_swp_unmap(entry);
1315 spin_unlock(&info->lock);
1320 wait_on_page_locked(swappage);
1321 page_cache_release(swappage);
1325 /* We have to do this with page locked to prevent races */
1326 if (TestSetPageLocked(swappage)) {
1327 shmem_swp_unmap(entry);
1328 spin_unlock(&info->lock);
1329 wait_on_page_locked(swappage);
1330 page_cache_release(swappage);
1333 if (PageWriteback(swappage)) {
1334 shmem_swp_unmap(entry);
1335 spin_unlock(&info->lock);
1336 wait_on_page_writeback(swappage);
1337 unlock_page(swappage);
1338 page_cache_release(swappage);
1341 if (!PageUptodate(swappage)) {
1342 shmem_swp_unmap(entry);
1343 spin_unlock(&info->lock);
1344 unlock_page(swappage);
1345 page_cache_release(swappage);
1351 shmem_swp_set(info, entry, 0);
1352 shmem_swp_unmap(entry);
1353 delete_from_swap_cache(swappage);
1354 spin_unlock(&info->lock);
1355 copy_highpage(filepage, swappage);
1356 unlock_page(swappage);
1357 page_cache_release(swappage);
1358 flush_dcache_page(filepage);
1359 SetPageUptodate(filepage);
1360 set_page_dirty(filepage);
1362 } else if (!(error = add_to_page_cache(
1363 swappage, mapping, idx, GFP_NOWAIT))) {
1364 info->flags |= SHMEM_PAGEIN;
1365 shmem_swp_set(info, entry, 0);
1366 shmem_swp_unmap(entry);
1367 delete_from_swap_cache(swappage);
1368 spin_unlock(&info->lock);
1369 filepage = swappage;
1370 set_page_dirty(filepage);
1373 shmem_swp_unmap(entry);
1374 spin_unlock(&info->lock);
1375 unlock_page(swappage);
1376 if (error == -ENOMEM) {
1377 /* allow reclaim from this memory cgroup */
1378 error = mem_cgroup_cache_charge(swappage,
1379 current->mm, gfp & ~__GFP_HIGHMEM);
1381 page_cache_release(swappage);
1384 mem_cgroup_uncharge_page(swappage);
1386 page_cache_release(swappage);
1389 } else if (sgp == SGP_READ && !filepage) {
1390 shmem_swp_unmap(entry);
1391 filepage = find_get_page(mapping, idx);
1393 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1394 spin_unlock(&info->lock);
1395 wait_on_page_locked(filepage);
1396 page_cache_release(filepage);
1400 spin_unlock(&info->lock);
1402 shmem_swp_unmap(entry);
1403 sbinfo = SHMEM_SB(inode->i_sb);
1404 if (sbinfo->max_blocks) {
1405 spin_lock(&sbinfo->stat_lock);
1406 if (sbinfo->free_blocks == 0 ||
1407 shmem_acct_block(info->flags)) {
1408 spin_unlock(&sbinfo->stat_lock);
1409 spin_unlock(&info->lock);
1413 sbinfo->free_blocks--;
1414 inode->i_blocks += BLOCKS_PER_PAGE;
1415 spin_unlock(&sbinfo->stat_lock);
1416 } else if (shmem_acct_block(info->flags)) {
1417 spin_unlock(&info->lock);
1423 spin_unlock(&info->lock);
1424 filepage = shmem_alloc_page(gfp, info, idx);
1426 shmem_unacct_blocks(info->flags, 1);
1427 shmem_free_blocks(inode, 1);
1432 /* Precharge page while we can wait, compensate after */
1433 error = mem_cgroup_cache_charge(filepage, current->mm,
1434 gfp & ~__GFP_HIGHMEM);
1436 page_cache_release(filepage);
1437 shmem_unacct_blocks(info->flags, 1);
1438 shmem_free_blocks(inode, 1);
1443 spin_lock(&info->lock);
1444 entry = shmem_swp_alloc(info, idx, sgp);
1446 error = PTR_ERR(entry);
1449 shmem_swp_unmap(entry);
1451 if (error || swap.val || 0 != add_to_page_cache_lru(
1452 filepage, mapping, idx, GFP_NOWAIT)) {
1453 spin_unlock(&info->lock);
1454 mem_cgroup_uncharge_page(filepage);
1455 page_cache_release(filepage);
1456 shmem_unacct_blocks(info->flags, 1);
1457 shmem_free_blocks(inode, 1);
1463 mem_cgroup_uncharge_page(filepage);
1464 info->flags |= SHMEM_PAGEIN;
1468 spin_unlock(&info->lock);
1469 clear_highpage(filepage);
1470 flush_dcache_page(filepage);
1471 SetPageUptodate(filepage);
1472 if (sgp == SGP_DIRTY)
1473 set_page_dirty(filepage);
1480 if (*pagep != filepage) {
1481 unlock_page(filepage);
1482 page_cache_release(filepage);
1487 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1489 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1493 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1494 return VM_FAULT_SIGBUS;
1496 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1498 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1500 mark_page_accessed(vmf->page);
1501 return ret | VM_FAULT_LOCKED;
1505 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1507 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1508 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1511 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1514 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1517 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1518 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1522 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1524 struct inode *inode = file->f_path.dentry->d_inode;
1525 struct shmem_inode_info *info = SHMEM_I(inode);
1526 int retval = -ENOMEM;
1528 spin_lock(&info->lock);
1529 if (lock && !(info->flags & VM_LOCKED)) {
1530 if (!user_shm_lock(inode->i_size, user))
1532 info->flags |= VM_LOCKED;
1534 if (!lock && (info->flags & VM_LOCKED) && user) {
1535 user_shm_unlock(inode->i_size, user);
1536 info->flags &= ~VM_LOCKED;
1540 spin_unlock(&info->lock);
1544 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1546 file_accessed(file);
1547 vma->vm_ops = &shmem_vm_ops;
1548 vma->vm_flags |= VM_CAN_NONLINEAR;
1552 static struct inode *
1553 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1555 struct inode *inode;
1556 struct shmem_inode_info *info;
1557 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1559 if (shmem_reserve_inode(sb))
1562 inode = new_inode(sb);
1564 inode->i_mode = mode;
1565 inode->i_uid = current->fsuid;
1566 inode->i_gid = current->fsgid;
1567 inode->i_blocks = 0;
1568 inode->i_mapping->a_ops = &shmem_aops;
1569 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1570 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1571 inode->i_generation = get_seconds();
1572 info = SHMEM_I(inode);
1573 memset(info, 0, (char *)inode - (char *)info);
1574 spin_lock_init(&info->lock);
1575 INIT_LIST_HEAD(&info->swaplist);
1577 switch (mode & S_IFMT) {
1579 inode->i_op = &shmem_special_inode_operations;
1580 init_special_inode(inode, mode, dev);
1583 inode->i_op = &shmem_inode_operations;
1584 inode->i_fop = &shmem_file_operations;
1585 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1586 sbinfo->flags, &sbinfo->policy_nodes);
1590 /* Some things misbehave if size == 0 on a directory */
1591 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1592 inode->i_op = &shmem_dir_inode_operations;
1593 inode->i_fop = &simple_dir_operations;
1597 * Must not load anything in the rbtree,
1598 * mpol_free_shared_policy will not be called.
1600 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0,
1605 shmem_free_inode(sb);
1610 static const struct inode_operations shmem_symlink_inode_operations;
1611 static const struct inode_operations shmem_symlink_inline_operations;
1614 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1615 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1616 * below the loop driver, in the generic fashion that many filesystems support.
1618 static int shmem_readpage(struct file *file, struct page *page)
1620 struct inode *inode = page->mapping->host;
1621 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1627 shmem_write_begin(struct file *file, struct address_space *mapping,
1628 loff_t pos, unsigned len, unsigned flags,
1629 struct page **pagep, void **fsdata)
1631 struct inode *inode = mapping->host;
1632 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1634 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1638 shmem_write_end(struct file *file, struct address_space *mapping,
1639 loff_t pos, unsigned len, unsigned copied,
1640 struct page *page, void *fsdata)
1642 struct inode *inode = mapping->host;
1644 if (pos + copied > inode->i_size)
1645 i_size_write(inode, pos + copied);
1648 set_page_dirty(page);
1649 page_cache_release(page);
1654 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1656 struct inode *inode = filp->f_path.dentry->d_inode;
1657 struct address_space *mapping = inode->i_mapping;
1658 unsigned long index, offset;
1659 enum sgp_type sgp = SGP_READ;
1662 * Might this read be for a stacking filesystem? Then when reading
1663 * holes of a sparse file, we actually need to allocate those pages,
1664 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1666 if (segment_eq(get_fs(), KERNEL_DS))
1669 index = *ppos >> PAGE_CACHE_SHIFT;
1670 offset = *ppos & ~PAGE_CACHE_MASK;
1673 struct page *page = NULL;
1674 unsigned long end_index, nr, ret;
1675 loff_t i_size = i_size_read(inode);
1677 end_index = i_size >> PAGE_CACHE_SHIFT;
1678 if (index > end_index)
1680 if (index == end_index) {
1681 nr = i_size & ~PAGE_CACHE_MASK;
1686 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1688 if (desc->error == -EINVAL)
1696 * We must evaluate after, since reads (unlike writes)
1697 * are called without i_mutex protection against truncate
1699 nr = PAGE_CACHE_SIZE;
1700 i_size = i_size_read(inode);
1701 end_index = i_size >> PAGE_CACHE_SHIFT;
1702 if (index == end_index) {
1703 nr = i_size & ~PAGE_CACHE_MASK;
1706 page_cache_release(page);
1714 * If users can be writing to this page using arbitrary
1715 * virtual addresses, take care about potential aliasing
1716 * before reading the page on the kernel side.
1718 if (mapping_writably_mapped(mapping))
1719 flush_dcache_page(page);
1721 * Mark the page accessed if we read the beginning.
1724 mark_page_accessed(page);
1726 page = ZERO_PAGE(0);
1727 page_cache_get(page);
1731 * Ok, we have the page, and it's up-to-date, so
1732 * now we can copy it to user space...
1734 * The actor routine returns how many bytes were actually used..
1735 * NOTE! This may not be the same as how much of a user buffer
1736 * we filled up (we may be padding etc), so we can only update
1737 * "pos" here (the actor routine has to update the user buffer
1738 * pointers and the remaining count).
1740 ret = actor(desc, page, offset, nr);
1742 index += offset >> PAGE_CACHE_SHIFT;
1743 offset &= ~PAGE_CACHE_MASK;
1745 page_cache_release(page);
1746 if (ret != nr || !desc->count)
1752 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1753 file_accessed(filp);
1756 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1758 read_descriptor_t desc;
1760 if ((ssize_t) count < 0)
1762 if (!access_ok(VERIFY_WRITE, buf, count))
1772 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1774 return desc.written;
1778 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1780 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1782 buf->f_type = TMPFS_MAGIC;
1783 buf->f_bsize = PAGE_CACHE_SIZE;
1784 buf->f_namelen = NAME_MAX;
1785 spin_lock(&sbinfo->stat_lock);
1786 if (sbinfo->max_blocks) {
1787 buf->f_blocks = sbinfo->max_blocks;
1788 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1790 if (sbinfo->max_inodes) {
1791 buf->f_files = sbinfo->max_inodes;
1792 buf->f_ffree = sbinfo->free_inodes;
1794 /* else leave those fields 0 like simple_statfs */
1795 spin_unlock(&sbinfo->stat_lock);
1800 * File creation. Allocate an inode, and we're done..
1803 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1805 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1806 int error = -ENOSPC;
1809 error = security_inode_init_security(inode, dir, NULL, NULL,
1812 if (error != -EOPNOTSUPP) {
1817 error = shmem_acl_init(inode, dir);
1822 if (dir->i_mode & S_ISGID) {
1823 inode->i_gid = dir->i_gid;
1825 inode->i_mode |= S_ISGID;
1827 dir->i_size += BOGO_DIRENT_SIZE;
1828 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1829 d_instantiate(dentry, inode);
1830 dget(dentry); /* Extra count - pin the dentry in core */
1835 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1839 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1845 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1846 struct nameidata *nd)
1848 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1854 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1856 struct inode *inode = old_dentry->d_inode;
1860 * No ordinary (disk based) filesystem counts links as inodes;
1861 * but each new link needs a new dentry, pinning lowmem, and
1862 * tmpfs dentries cannot be pruned until they are unlinked.
1864 ret = shmem_reserve_inode(inode->i_sb);
1868 dir->i_size += BOGO_DIRENT_SIZE;
1869 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1871 atomic_inc(&inode->i_count); /* New dentry reference */
1872 dget(dentry); /* Extra pinning count for the created dentry */
1873 d_instantiate(dentry, inode);
1878 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1880 struct inode *inode = dentry->d_inode;
1882 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1883 shmem_free_inode(inode->i_sb);
1885 dir->i_size -= BOGO_DIRENT_SIZE;
1886 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1888 dput(dentry); /* Undo the count from "create" - this does all the work */
1892 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1894 if (!simple_empty(dentry))
1897 drop_nlink(dentry->d_inode);
1899 return shmem_unlink(dir, dentry);
1903 * The VFS layer already does all the dentry stuff for rename,
1904 * we just have to decrement the usage count for the target if
1905 * it exists so that the VFS layer correctly free's it when it
1908 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1910 struct inode *inode = old_dentry->d_inode;
1911 int they_are_dirs = S_ISDIR(inode->i_mode);
1913 if (!simple_empty(new_dentry))
1916 if (new_dentry->d_inode) {
1917 (void) shmem_unlink(new_dir, new_dentry);
1919 drop_nlink(old_dir);
1920 } else if (they_are_dirs) {
1921 drop_nlink(old_dir);
1925 old_dir->i_size -= BOGO_DIRENT_SIZE;
1926 new_dir->i_size += BOGO_DIRENT_SIZE;
1927 old_dir->i_ctime = old_dir->i_mtime =
1928 new_dir->i_ctime = new_dir->i_mtime =
1929 inode->i_ctime = CURRENT_TIME;
1933 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1937 struct inode *inode;
1938 struct page *page = NULL;
1940 struct shmem_inode_info *info;
1942 len = strlen(symname) + 1;
1943 if (len > PAGE_CACHE_SIZE)
1944 return -ENAMETOOLONG;
1946 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1950 error = security_inode_init_security(inode, dir, NULL, NULL,
1953 if (error != -EOPNOTSUPP) {
1960 info = SHMEM_I(inode);
1961 inode->i_size = len-1;
1962 if (len <= (char *)inode - (char *)info) {
1964 memcpy(info, symname, len);
1965 inode->i_op = &shmem_symlink_inline_operations;
1967 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1973 inode->i_op = &shmem_symlink_inode_operations;
1974 kaddr = kmap_atomic(page, KM_USER0);
1975 memcpy(kaddr, symname, len);
1976 kunmap_atomic(kaddr, KM_USER0);
1977 set_page_dirty(page);
1978 page_cache_release(page);
1980 if (dir->i_mode & S_ISGID)
1981 inode->i_gid = dir->i_gid;
1982 dir->i_size += BOGO_DIRENT_SIZE;
1983 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1984 d_instantiate(dentry, inode);
1989 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1991 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1995 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1997 struct page *page = NULL;
1998 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1999 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2005 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2007 if (!IS_ERR(nd_get_link(nd))) {
2008 struct page *page = cookie;
2010 mark_page_accessed(page);
2011 page_cache_release(page);
2015 static const struct inode_operations shmem_symlink_inline_operations = {
2016 .readlink = generic_readlink,
2017 .follow_link = shmem_follow_link_inline,
2020 static const struct inode_operations shmem_symlink_inode_operations = {
2021 .truncate = shmem_truncate,
2022 .readlink = generic_readlink,
2023 .follow_link = shmem_follow_link,
2024 .put_link = shmem_put_link,
2027 #ifdef CONFIG_TMPFS_POSIX_ACL
2029 * Superblocks without xattr inode operations will get security.* xattr
2030 * support from the VFS "for free". As soon as we have any other xattrs
2031 * like ACLs, we also need to implement the security.* handlers at
2032 * filesystem level, though.
2035 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
2036 size_t list_len, const char *name,
2039 return security_inode_listsecurity(inode, list, list_len);
2042 static int shmem_xattr_security_get(struct inode *inode, const char *name,
2043 void *buffer, size_t size)
2045 if (strcmp(name, "") == 0)
2047 return xattr_getsecurity(inode, name, buffer, size);
2050 static int shmem_xattr_security_set(struct inode *inode, const char *name,
2051 const void *value, size_t size, int flags)
2053 if (strcmp(name, "") == 0)
2055 return security_inode_setsecurity(inode, name, value, size, flags);
2058 static struct xattr_handler shmem_xattr_security_handler = {
2059 .prefix = XATTR_SECURITY_PREFIX,
2060 .list = shmem_xattr_security_list,
2061 .get = shmem_xattr_security_get,
2062 .set = shmem_xattr_security_set,
2065 static struct xattr_handler *shmem_xattr_handlers[] = {
2066 &shmem_xattr_acl_access_handler,
2067 &shmem_xattr_acl_default_handler,
2068 &shmem_xattr_security_handler,
2073 static struct dentry *shmem_get_parent(struct dentry *child)
2075 return ERR_PTR(-ESTALE);
2078 static int shmem_match(struct inode *ino, void *vfh)
2082 inum = (inum << 32) | fh[1];
2083 return ino->i_ino == inum && fh[0] == ino->i_generation;
2086 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2087 struct fid *fid, int fh_len, int fh_type)
2089 struct inode *inode;
2090 struct dentry *dentry = NULL;
2091 u64 inum = fid->raw[2];
2092 inum = (inum << 32) | fid->raw[1];
2097 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2098 shmem_match, fid->raw);
2100 dentry = d_find_alias(inode);
2107 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2110 struct inode *inode = dentry->d_inode;
2115 if (hlist_unhashed(&inode->i_hash)) {
2116 /* Unfortunately insert_inode_hash is not idempotent,
2117 * so as we hash inodes here rather than at creation
2118 * time, we need a lock to ensure we only try
2121 static DEFINE_SPINLOCK(lock);
2123 if (hlist_unhashed(&inode->i_hash))
2124 __insert_inode_hash(inode,
2125 inode->i_ino + inode->i_generation);
2129 fh[0] = inode->i_generation;
2130 fh[1] = inode->i_ino;
2131 fh[2] = ((__u64)inode->i_ino) >> 32;
2137 static const struct export_operations shmem_export_ops = {
2138 .get_parent = shmem_get_parent,
2139 .encode_fh = shmem_encode_fh,
2140 .fh_to_dentry = shmem_fh_to_dentry,
2143 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2146 char *this_char, *value, *rest;
2148 while (options != NULL) {
2149 this_char = options;
2152 * NUL-terminate this option: unfortunately,
2153 * mount options form a comma-separated list,
2154 * but mpol's nodelist may also contain commas.
2156 options = strchr(options, ',');
2157 if (options == NULL)
2160 if (!isdigit(*options)) {
2167 if ((value = strchr(this_char,'=')) != NULL) {
2171 "tmpfs: No value for mount option '%s'\n",
2176 if (!strcmp(this_char,"size")) {
2177 unsigned long long size;
2178 size = memparse(value,&rest);
2180 size <<= PAGE_SHIFT;
2181 size *= totalram_pages;
2187 sbinfo->max_blocks =
2188 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2189 } else if (!strcmp(this_char,"nr_blocks")) {
2190 sbinfo->max_blocks = memparse(value, &rest);
2193 } else if (!strcmp(this_char,"nr_inodes")) {
2194 sbinfo->max_inodes = memparse(value, &rest);
2197 } else if (!strcmp(this_char,"mode")) {
2200 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2203 } else if (!strcmp(this_char,"uid")) {
2206 sbinfo->uid = simple_strtoul(value, &rest, 0);
2209 } else if (!strcmp(this_char,"gid")) {
2212 sbinfo->gid = simple_strtoul(value, &rest, 0);
2215 } else if (!strcmp(this_char,"mpol")) {
2216 if (shmem_parse_mpol(value, &sbinfo->policy,
2217 &sbinfo->flags, &sbinfo->policy_nodes))
2220 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2228 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2234 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2236 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2237 struct shmem_sb_info config = *sbinfo;
2238 unsigned long blocks;
2239 unsigned long inodes;
2240 int error = -EINVAL;
2242 if (shmem_parse_options(data, &config, true))
2245 spin_lock(&sbinfo->stat_lock);
2246 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2247 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2248 if (config.max_blocks < blocks)
2250 if (config.max_inodes < inodes)
2253 * Those tests also disallow limited->unlimited while any are in
2254 * use, so i_blocks will always be zero when max_blocks is zero;
2255 * but we must separately disallow unlimited->limited, because
2256 * in that case we have no record of how much is already in use.
2258 if (config.max_blocks && !sbinfo->max_blocks)
2260 if (config.max_inodes && !sbinfo->max_inodes)
2264 sbinfo->max_blocks = config.max_blocks;
2265 sbinfo->free_blocks = config.max_blocks - blocks;
2266 sbinfo->max_inodes = config.max_inodes;
2267 sbinfo->free_inodes = config.max_inodes - inodes;
2268 sbinfo->policy = config.policy;
2269 sbinfo->flags = config.flags;
2270 sbinfo->policy_nodes = config.policy_nodes;
2272 spin_unlock(&sbinfo->stat_lock);
2276 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2278 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2280 if (sbinfo->max_blocks != shmem_default_max_blocks())
2281 seq_printf(seq, ",size=%luk",
2282 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2283 if (sbinfo->max_inodes != shmem_default_max_inodes())
2284 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2285 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2286 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2287 if (sbinfo->uid != 0)
2288 seq_printf(seq, ",uid=%u", sbinfo->uid);
2289 if (sbinfo->gid != 0)
2290 seq_printf(seq, ",gid=%u", sbinfo->gid);
2291 shmem_show_mpol(seq, sbinfo->policy, sbinfo->flags,
2292 sbinfo->policy_nodes);
2295 #endif /* CONFIG_TMPFS */
2297 static void shmem_put_super(struct super_block *sb)
2299 kfree(sb->s_fs_info);
2300 sb->s_fs_info = NULL;
2303 static int shmem_fill_super(struct super_block *sb,
2304 void *data, int silent)
2306 struct inode *inode;
2307 struct dentry *root;
2308 struct shmem_sb_info *sbinfo;
2311 /* Round up to L1_CACHE_BYTES to resist false sharing */
2312 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2313 L1_CACHE_BYTES), GFP_KERNEL);
2317 sbinfo->max_blocks = 0;
2318 sbinfo->max_inodes = 0;
2319 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2320 sbinfo->uid = current->fsuid;
2321 sbinfo->gid = current->fsgid;
2322 sbinfo->policy = MPOL_DEFAULT;
2324 sbinfo->policy_nodes = node_states[N_HIGH_MEMORY];
2325 sb->s_fs_info = sbinfo;
2329 * Per default we only allow half of the physical ram per
2330 * tmpfs instance, limiting inodes to one per page of lowmem;
2331 * but the internal instance is left unlimited.
2333 if (!(sb->s_flags & MS_NOUSER)) {
2334 sbinfo->max_blocks = shmem_default_max_blocks();
2335 sbinfo->max_inodes = shmem_default_max_inodes();
2336 if (shmem_parse_options(data, sbinfo, false)) {
2341 sb->s_export_op = &shmem_export_ops;
2343 sb->s_flags |= MS_NOUSER;
2346 spin_lock_init(&sbinfo->stat_lock);
2347 sbinfo->free_blocks = sbinfo->max_blocks;
2348 sbinfo->free_inodes = sbinfo->max_inodes;
2350 sb->s_maxbytes = SHMEM_MAX_BYTES;
2351 sb->s_blocksize = PAGE_CACHE_SIZE;
2352 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2353 sb->s_magic = TMPFS_MAGIC;
2354 sb->s_op = &shmem_ops;
2355 sb->s_time_gran = 1;
2356 #ifdef CONFIG_TMPFS_POSIX_ACL
2357 sb->s_xattr = shmem_xattr_handlers;
2358 sb->s_flags |= MS_POSIXACL;
2361 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0);
2364 inode->i_uid = sbinfo->uid;
2365 inode->i_gid = sbinfo->gid;
2366 root = d_alloc_root(inode);
2375 shmem_put_super(sb);
2379 static struct kmem_cache *shmem_inode_cachep;
2381 static struct inode *shmem_alloc_inode(struct super_block *sb)
2383 struct shmem_inode_info *p;
2384 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2387 return &p->vfs_inode;
2390 static void shmem_destroy_inode(struct inode *inode)
2392 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2393 /* only struct inode is valid if it's an inline symlink */
2394 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2396 shmem_acl_destroy_inode(inode);
2397 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2400 static void init_once(struct kmem_cache *cachep, void *foo)
2402 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2404 inode_init_once(&p->vfs_inode);
2405 #ifdef CONFIG_TMPFS_POSIX_ACL
2407 p->i_default_acl = NULL;
2411 static int init_inodecache(void)
2413 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2414 sizeof(struct shmem_inode_info),
2415 0, SLAB_PANIC, init_once);
2419 static void destroy_inodecache(void)
2421 kmem_cache_destroy(shmem_inode_cachep);
2424 static const struct address_space_operations shmem_aops = {
2425 .writepage = shmem_writepage,
2426 .set_page_dirty = __set_page_dirty_no_writeback,
2428 .readpage = shmem_readpage,
2429 .write_begin = shmem_write_begin,
2430 .write_end = shmem_write_end,
2432 .migratepage = migrate_page,
2435 static const struct file_operations shmem_file_operations = {
2438 .llseek = generic_file_llseek,
2439 .read = shmem_file_read,
2440 .write = do_sync_write,
2441 .aio_write = generic_file_aio_write,
2442 .fsync = simple_sync_file,
2443 .splice_read = generic_file_splice_read,
2444 .splice_write = generic_file_splice_write,
2448 static const struct inode_operations shmem_inode_operations = {
2449 .truncate = shmem_truncate,
2450 .setattr = shmem_notify_change,
2451 .truncate_range = shmem_truncate_range,
2452 #ifdef CONFIG_TMPFS_POSIX_ACL
2453 .setxattr = generic_setxattr,
2454 .getxattr = generic_getxattr,
2455 .listxattr = generic_listxattr,
2456 .removexattr = generic_removexattr,
2457 .permission = shmem_permission,
2462 static const struct inode_operations shmem_dir_inode_operations = {
2464 .create = shmem_create,
2465 .lookup = simple_lookup,
2467 .unlink = shmem_unlink,
2468 .symlink = shmem_symlink,
2469 .mkdir = shmem_mkdir,
2470 .rmdir = shmem_rmdir,
2471 .mknod = shmem_mknod,
2472 .rename = shmem_rename,
2474 #ifdef CONFIG_TMPFS_POSIX_ACL
2475 .setattr = shmem_notify_change,
2476 .setxattr = generic_setxattr,
2477 .getxattr = generic_getxattr,
2478 .listxattr = generic_listxattr,
2479 .removexattr = generic_removexattr,
2480 .permission = shmem_permission,
2484 static const struct inode_operations shmem_special_inode_operations = {
2485 #ifdef CONFIG_TMPFS_POSIX_ACL
2486 .setattr = shmem_notify_change,
2487 .setxattr = generic_setxattr,
2488 .getxattr = generic_getxattr,
2489 .listxattr = generic_listxattr,
2490 .removexattr = generic_removexattr,
2491 .permission = shmem_permission,
2495 static const struct super_operations shmem_ops = {
2496 .alloc_inode = shmem_alloc_inode,
2497 .destroy_inode = shmem_destroy_inode,
2499 .statfs = shmem_statfs,
2500 .remount_fs = shmem_remount_fs,
2501 .show_options = shmem_show_options,
2503 .delete_inode = shmem_delete_inode,
2504 .drop_inode = generic_delete_inode,
2505 .put_super = shmem_put_super,
2508 static struct vm_operations_struct shmem_vm_ops = {
2509 .fault = shmem_fault,
2511 .set_policy = shmem_set_policy,
2512 .get_policy = shmem_get_policy,
2517 static int shmem_get_sb(struct file_system_type *fs_type,
2518 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2520 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2523 static struct file_system_type tmpfs_fs_type = {
2524 .owner = THIS_MODULE,
2526 .get_sb = shmem_get_sb,
2527 .kill_sb = kill_litter_super,
2529 static struct vfsmount *shm_mnt;
2531 static int __init init_tmpfs(void)
2535 error = bdi_init(&shmem_backing_dev_info);
2539 error = init_inodecache();
2543 error = register_filesystem(&tmpfs_fs_type);
2545 printk(KERN_ERR "Could not register tmpfs\n");
2549 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2550 tmpfs_fs_type.name, NULL);
2551 if (IS_ERR(shm_mnt)) {
2552 error = PTR_ERR(shm_mnt);
2553 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2559 unregister_filesystem(&tmpfs_fs_type);
2561 destroy_inodecache();
2563 bdi_destroy(&shmem_backing_dev_info);
2565 shm_mnt = ERR_PTR(error);
2568 module_init(init_tmpfs)
2571 * shmem_file_setup - get an unlinked file living in tmpfs
2572 * @name: name for dentry (to be seen in /proc/<pid>/maps
2573 * @size: size to be set for the file
2576 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2580 struct inode *inode;
2581 struct dentry *dentry, *root;
2584 if (IS_ERR(shm_mnt))
2585 return (void *)shm_mnt;
2587 if (size < 0 || size > SHMEM_MAX_BYTES)
2588 return ERR_PTR(-EINVAL);
2590 if (shmem_acct_size(flags, size))
2591 return ERR_PTR(-ENOMEM);
2595 this.len = strlen(name);
2596 this.hash = 0; /* will go */
2597 root = shm_mnt->mnt_root;
2598 dentry = d_alloc(root, &this);
2603 file = get_empty_filp();
2608 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2612 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2613 d_instantiate(dentry, inode);
2614 inode->i_size = size;
2615 inode->i_nlink = 0; /* It is unlinked */
2616 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2617 &shmem_file_operations);
2625 shmem_unacct_size(flags, size);
2626 return ERR_PTR(error);
2630 * shmem_zero_setup - setup a shared anonymous mapping
2631 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2633 int shmem_zero_setup(struct vm_area_struct *vma)
2636 loff_t size = vma->vm_end - vma->vm_start;
2638 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2640 return PTR_ERR(file);
2644 vma->vm_file = file;
2645 vma->vm_ops = &shmem_vm_ops;