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>
53 #include <asm/uaccess.h>
54 #include <asm/div64.h>
55 #include <asm/pgtable.h>
57 /* This magic number is used in glibc for posix shared memory */
58 #define TMPFS_MAGIC 0x01021994
60 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
61 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
62 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
64 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
65 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
67 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
69 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
70 #define SHMEM_PAGEIN VM_READ
71 #define SHMEM_TRUNCATE VM_WRITE
73 /* Definition to limit shmem_truncate's steps between cond_rescheds */
74 #define LATENCY_LIMIT 64
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
79 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
81 SGP_READ, /* don't exceed i_size, don't allocate page */
82 SGP_CACHE, /* don't exceed i_size, may allocate page */
83 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
84 SGP_WRITE, /* may exceed i_size, may allocate page */
87 static int shmem_getpage(struct inode *inode, unsigned long idx,
88 struct page **pagep, enum sgp_type sgp, int *type);
90 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
93 * The above definition of ENTRIES_PER_PAGE, and the use of
94 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
95 * might be reconsidered if it ever diverges from PAGE_SIZE.
97 * Mobility flags are masked out as swap vectors cannot move
99 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
100 PAGE_CACHE_SHIFT-PAGE_SHIFT);
103 static inline void shmem_dir_free(struct page *page)
105 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
108 static struct page **shmem_dir_map(struct page *page)
110 return (struct page **)kmap_atomic(page, KM_USER0);
113 static inline void shmem_dir_unmap(struct page **dir)
115 kunmap_atomic(dir, KM_USER0);
118 static swp_entry_t *shmem_swp_map(struct page *page)
120 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
123 static inline void shmem_swp_balance_unmap(void)
126 * When passing a pointer to an i_direct entry, to code which
127 * also handles indirect entries and so will shmem_swp_unmap,
128 * we must arrange for the preempt count to remain in balance.
129 * What kmap_atomic of a lowmem page does depends on config
130 * and architecture, so pretend to kmap_atomic some lowmem page.
132 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
135 static inline void shmem_swp_unmap(swp_entry_t *entry)
137 kunmap_atomic(entry, KM_USER1);
140 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
142 return sb->s_fs_info;
146 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
147 * for shared memory and for shared anonymous (/dev/zero) mappings
148 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
149 * consistent with the pre-accounting of private mappings ...
151 static inline int shmem_acct_size(unsigned long flags, loff_t size)
153 return (flags & VM_ACCOUNT)?
154 security_vm_enough_memory(VM_ACCT(size)): 0;
157 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
159 if (flags & VM_ACCOUNT)
160 vm_unacct_memory(VM_ACCT(size));
164 * ... whereas tmpfs objects are accounted incrementally as
165 * pages are allocated, in order to allow huge sparse files.
166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169 static inline int shmem_acct_block(unsigned long flags)
171 return (flags & VM_ACCOUNT)?
172 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
175 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
177 if (!(flags & VM_ACCOUNT))
178 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
181 static const struct super_operations shmem_ops;
182 static const struct address_space_operations shmem_aops;
183 static const struct file_operations shmem_file_operations;
184 static const struct inode_operations shmem_inode_operations;
185 static const struct inode_operations shmem_dir_inode_operations;
186 static const struct inode_operations shmem_special_inode_operations;
187 static struct vm_operations_struct shmem_vm_ops;
189 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
190 .ra_pages = 0, /* No readahead */
191 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
192 .unplug_io_fn = default_unplug_io_fn,
195 static LIST_HEAD(shmem_swaplist);
196 static DEFINE_SPINLOCK(shmem_swaplist_lock);
198 static void shmem_free_blocks(struct inode *inode, long pages)
200 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
201 if (sbinfo->max_blocks) {
202 spin_lock(&sbinfo->stat_lock);
203 sbinfo->free_blocks += pages;
204 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
205 spin_unlock(&sbinfo->stat_lock);
209 static int shmem_reserve_inode(struct super_block *sb)
211 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
212 if (sbinfo->max_inodes) {
213 spin_lock(&sbinfo->stat_lock);
214 if (!sbinfo->free_inodes) {
215 spin_unlock(&sbinfo->stat_lock);
218 sbinfo->free_inodes--;
219 spin_unlock(&sbinfo->stat_lock);
224 static void shmem_free_inode(struct super_block *sb)
226 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
227 if (sbinfo->max_inodes) {
228 spin_lock(&sbinfo->stat_lock);
229 sbinfo->free_inodes++;
230 spin_unlock(&sbinfo->stat_lock);
235 * shmem_recalc_inode - recalculate the size of an inode
237 * @inode: inode to recalc
239 * We have to calculate the free blocks since the mm can drop
240 * undirtied hole pages behind our back.
242 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
243 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
245 * It has to be called with the spinlock held.
247 static void shmem_recalc_inode(struct inode *inode)
249 struct shmem_inode_info *info = SHMEM_I(inode);
252 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
254 info->alloced -= freed;
255 shmem_unacct_blocks(info->flags, freed);
256 shmem_free_blocks(inode, freed);
261 * shmem_swp_entry - find the swap vector position in the info structure
263 * @info: info structure for the inode
264 * @index: index of the page to find
265 * @page: optional page to add to the structure. Has to be preset to
268 * If there is no space allocated yet it will return NULL when
269 * page is NULL, else it will use the page for the needed block,
270 * setting it to NULL on return to indicate that it has been used.
272 * The swap vector is organized the following way:
274 * There are SHMEM_NR_DIRECT entries directly stored in the
275 * shmem_inode_info structure. So small files do not need an addional
278 * For pages with index > SHMEM_NR_DIRECT there is the pointer
279 * i_indirect which points to a page which holds in the first half
280 * doubly indirect blocks, in the second half triple indirect blocks:
282 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
283 * following layout (for SHMEM_NR_DIRECT == 16):
285 * i_indirect -> dir --> 16-19
298 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
300 unsigned long offset;
304 if (index < SHMEM_NR_DIRECT) {
305 shmem_swp_balance_unmap();
306 return info->i_direct+index;
308 if (!info->i_indirect) {
310 info->i_indirect = *page;
313 return NULL; /* need another page */
316 index -= SHMEM_NR_DIRECT;
317 offset = index % ENTRIES_PER_PAGE;
318 index /= ENTRIES_PER_PAGE;
319 dir = shmem_dir_map(info->i_indirect);
321 if (index >= ENTRIES_PER_PAGE/2) {
322 index -= ENTRIES_PER_PAGE/2;
323 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
324 index %= ENTRIES_PER_PAGE;
331 shmem_dir_unmap(dir);
332 return NULL; /* need another page */
334 shmem_dir_unmap(dir);
335 dir = shmem_dir_map(subdir);
341 if (!page || !(subdir = *page)) {
342 shmem_dir_unmap(dir);
343 return NULL; /* need a page */
348 shmem_dir_unmap(dir);
349 return shmem_swp_map(subdir) + offset;
352 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
354 long incdec = value? 1: -1;
357 info->swapped += incdec;
358 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
359 struct page *page = kmap_atomic_to_page(entry);
360 set_page_private(page, page_private(page) + incdec);
365 * shmem_swp_alloc - get the position of the swap entry for the page.
366 * If it does not exist allocate the entry.
368 * @info: info structure for the inode
369 * @index: index of the page to find
370 * @sgp: check and recheck i_size? skip allocation?
372 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
374 struct inode *inode = &info->vfs_inode;
375 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
376 struct page *page = NULL;
379 if (sgp != SGP_WRITE &&
380 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
381 return ERR_PTR(-EINVAL);
383 while (!(entry = shmem_swp_entry(info, index, &page))) {
385 return shmem_swp_map(ZERO_PAGE(0));
387 * Test free_blocks against 1 not 0, since we have 1 data
388 * page (and perhaps indirect index pages) yet to allocate:
389 * a waste to allocate index if we cannot allocate data.
391 if (sbinfo->max_blocks) {
392 spin_lock(&sbinfo->stat_lock);
393 if (sbinfo->free_blocks <= 1) {
394 spin_unlock(&sbinfo->stat_lock);
395 return ERR_PTR(-ENOSPC);
397 sbinfo->free_blocks--;
398 inode->i_blocks += BLOCKS_PER_PAGE;
399 spin_unlock(&sbinfo->stat_lock);
402 spin_unlock(&info->lock);
403 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
405 set_page_private(page, 0);
406 spin_lock(&info->lock);
409 shmem_free_blocks(inode, 1);
410 return ERR_PTR(-ENOMEM);
412 if (sgp != SGP_WRITE &&
413 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
414 entry = ERR_PTR(-EINVAL);
417 if (info->next_index <= index)
418 info->next_index = index + 1;
421 /* another task gave its page, or truncated the file */
422 shmem_free_blocks(inode, 1);
423 shmem_dir_free(page);
425 if (info->next_index <= index && !IS_ERR(entry))
426 info->next_index = index + 1;
431 * shmem_free_swp - free some swap entries in a directory
433 * @dir: pointer to the directory
434 * @edir: pointer after last entry of the directory
435 * @punch_lock: pointer to spinlock when needed for the holepunch case
437 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
438 spinlock_t *punch_lock)
440 spinlock_t *punch_unlock = NULL;
444 for (ptr = dir; ptr < edir; ptr++) {
446 if (unlikely(punch_lock)) {
447 punch_unlock = punch_lock;
449 spin_lock(punch_unlock);
453 free_swap_and_cache(*ptr);
454 *ptr = (swp_entry_t){0};
459 spin_unlock(punch_unlock);
463 static int shmem_map_and_free_swp(struct page *subdir, int offset,
464 int limit, struct page ***dir, spinlock_t *punch_lock)
469 ptr = shmem_swp_map(subdir);
470 for (; offset < limit; offset += LATENCY_LIMIT) {
471 int size = limit - offset;
472 if (size > LATENCY_LIMIT)
473 size = LATENCY_LIMIT;
474 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
476 if (need_resched()) {
477 shmem_swp_unmap(ptr);
479 shmem_dir_unmap(*dir);
483 ptr = shmem_swp_map(subdir);
486 shmem_swp_unmap(ptr);
490 static void shmem_free_pages(struct list_head *next)
496 page = container_of(next, struct page, lru);
498 shmem_dir_free(page);
500 if (freed >= LATENCY_LIMIT) {
507 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
509 struct shmem_inode_info *info = SHMEM_I(inode);
514 unsigned long diroff;
520 LIST_HEAD(pages_to_free);
521 long nr_pages_to_free = 0;
522 long nr_swaps_freed = 0;
526 spinlock_t *needs_lock;
527 spinlock_t *punch_lock;
528 unsigned long upper_limit;
530 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
531 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
532 if (idx >= info->next_index)
535 spin_lock(&info->lock);
536 info->flags |= SHMEM_TRUNCATE;
537 if (likely(end == (loff_t) -1)) {
538 limit = info->next_index;
539 upper_limit = SHMEM_MAX_INDEX;
540 info->next_index = idx;
544 if (end + 1 >= inode->i_size) { /* we may free a little more */
545 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
547 upper_limit = SHMEM_MAX_INDEX;
549 limit = (end + 1) >> PAGE_CACHE_SHIFT;
552 needs_lock = &info->lock;
556 topdir = info->i_indirect;
557 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
558 info->i_indirect = NULL;
560 list_add(&topdir->lru, &pages_to_free);
562 spin_unlock(&info->lock);
564 if (info->swapped && idx < SHMEM_NR_DIRECT) {
565 ptr = info->i_direct;
567 if (size > SHMEM_NR_DIRECT)
568 size = SHMEM_NR_DIRECT;
569 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
573 * If there are no indirect blocks or we are punching a hole
574 * below indirect blocks, nothing to be done.
576 if (!topdir || limit <= SHMEM_NR_DIRECT)
580 * The truncation case has already dropped info->lock, and we're safe
581 * because i_size and next_index have already been lowered, preventing
582 * access beyond. But in the punch_hole case, we still need to take
583 * the lock when updating the swap directory, because there might be
584 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
585 * shmem_writepage. However, whenever we find we can remove a whole
586 * directory page (not at the misaligned start or end of the range),
587 * we first NULLify its pointer in the level above, and then have no
588 * need to take the lock when updating its contents: needs_lock and
589 * punch_lock (either pointing to info->lock or NULL) manage this.
592 upper_limit -= SHMEM_NR_DIRECT;
593 limit -= SHMEM_NR_DIRECT;
594 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
595 offset = idx % ENTRIES_PER_PAGE;
598 dir = shmem_dir_map(topdir);
599 stage = ENTRIES_PER_PAGEPAGE/2;
600 if (idx < ENTRIES_PER_PAGEPAGE/2) {
602 diroff = idx/ENTRIES_PER_PAGE;
604 dir += ENTRIES_PER_PAGE/2;
605 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
607 stage += ENTRIES_PER_PAGEPAGE;
610 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
611 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
612 if (!diroff && !offset && upper_limit >= stage) {
614 spin_lock(needs_lock);
616 spin_unlock(needs_lock);
621 list_add(&middir->lru, &pages_to_free);
623 shmem_dir_unmap(dir);
624 dir = shmem_dir_map(middir);
632 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
633 if (unlikely(idx == stage)) {
634 shmem_dir_unmap(dir);
635 dir = shmem_dir_map(topdir) +
636 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
639 idx += ENTRIES_PER_PAGEPAGE;
643 stage = idx + ENTRIES_PER_PAGEPAGE;
646 needs_lock = &info->lock;
647 if (upper_limit >= stage) {
649 spin_lock(needs_lock);
651 spin_unlock(needs_lock);
656 list_add(&middir->lru, &pages_to_free);
658 shmem_dir_unmap(dir);
660 dir = shmem_dir_map(middir);
663 punch_lock = needs_lock;
664 subdir = dir[diroff];
665 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
667 spin_lock(needs_lock);
669 spin_unlock(needs_lock);
674 list_add(&subdir->lru, &pages_to_free);
676 if (subdir && page_private(subdir) /* has swap entries */) {
678 if (size > ENTRIES_PER_PAGE)
679 size = ENTRIES_PER_PAGE;
680 freed = shmem_map_and_free_swp(subdir,
681 offset, size, &dir, punch_lock);
683 dir = shmem_dir_map(middir);
684 nr_swaps_freed += freed;
685 if (offset || punch_lock) {
686 spin_lock(&info->lock);
687 set_page_private(subdir,
688 page_private(subdir) - freed);
689 spin_unlock(&info->lock);
691 BUG_ON(page_private(subdir) != freed);
696 shmem_dir_unmap(dir);
698 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
700 * Call truncate_inode_pages again: racing shmem_unuse_inode
701 * may have swizzled a page in from swap since vmtruncate or
702 * generic_delete_inode did it, before we lowered next_index.
703 * Also, though shmem_getpage checks i_size before adding to
704 * cache, no recheck after: so fix the narrow window there too.
706 * Recalling truncate_inode_pages_range and unmap_mapping_range
707 * every time for punch_hole (which never got a chance to clear
708 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
709 * yet hardly ever necessary: try to optimize them out later.
711 truncate_inode_pages_range(inode->i_mapping, start, end);
713 unmap_mapping_range(inode->i_mapping, start,
717 spin_lock(&info->lock);
718 info->flags &= ~SHMEM_TRUNCATE;
719 info->swapped -= nr_swaps_freed;
720 if (nr_pages_to_free)
721 shmem_free_blocks(inode, nr_pages_to_free);
722 shmem_recalc_inode(inode);
723 spin_unlock(&info->lock);
726 * Empty swap vector directory pages to be freed?
728 if (!list_empty(&pages_to_free)) {
729 pages_to_free.prev->next = NULL;
730 shmem_free_pages(pages_to_free.next);
734 static void shmem_truncate(struct inode *inode)
736 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
739 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
741 struct inode *inode = dentry->d_inode;
742 struct page *page = NULL;
745 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
746 if (attr->ia_size < inode->i_size) {
748 * If truncating down to a partial page, then
749 * if that page is already allocated, hold it
750 * in memory until the truncation is over, so
751 * truncate_partial_page cannnot miss it were
752 * it assigned to swap.
754 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
755 (void) shmem_getpage(inode,
756 attr->ia_size>>PAGE_CACHE_SHIFT,
757 &page, SGP_READ, NULL);
762 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
763 * detect if any pages might have been added to cache
764 * after truncate_inode_pages. But we needn't bother
765 * if it's being fully truncated to zero-length: the
766 * nrpages check is efficient enough in that case.
769 struct shmem_inode_info *info = SHMEM_I(inode);
770 spin_lock(&info->lock);
771 info->flags &= ~SHMEM_PAGEIN;
772 spin_unlock(&info->lock);
777 error = inode_change_ok(inode, attr);
779 error = inode_setattr(inode, attr);
780 #ifdef CONFIG_TMPFS_POSIX_ACL
781 if (!error && (attr->ia_valid & ATTR_MODE))
782 error = generic_acl_chmod(inode, &shmem_acl_ops);
785 page_cache_release(page);
789 static void shmem_delete_inode(struct inode *inode)
791 struct shmem_inode_info *info = SHMEM_I(inode);
793 if (inode->i_op->truncate == shmem_truncate) {
794 truncate_inode_pages(inode->i_mapping, 0);
795 shmem_unacct_size(info->flags, inode->i_size);
797 shmem_truncate(inode);
798 if (!list_empty(&info->swaplist)) {
799 spin_lock(&shmem_swaplist_lock);
800 list_del_init(&info->swaplist);
801 spin_unlock(&shmem_swaplist_lock);
804 BUG_ON(inode->i_blocks);
805 shmem_free_inode(inode->i_sb);
809 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
813 for (ptr = dir; ptr < edir; ptr++) {
814 if (ptr->val == entry.val)
820 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
834 ptr = info->i_direct;
835 spin_lock(&info->lock);
836 limit = info->next_index;
838 if (size > SHMEM_NR_DIRECT)
839 size = SHMEM_NR_DIRECT;
840 offset = shmem_find_swp(entry, ptr, ptr+size);
842 shmem_swp_balance_unmap();
845 if (!info->i_indirect)
848 dir = shmem_dir_map(info->i_indirect);
849 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
851 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
852 if (unlikely(idx == stage)) {
853 shmem_dir_unmap(dir-1);
854 dir = shmem_dir_map(info->i_indirect) +
855 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
858 idx += ENTRIES_PER_PAGEPAGE;
862 stage = idx + ENTRIES_PER_PAGEPAGE;
864 shmem_dir_unmap(dir);
865 dir = shmem_dir_map(subdir);
868 if (subdir && page_private(subdir)) {
869 ptr = shmem_swp_map(subdir);
871 if (size > ENTRIES_PER_PAGE)
872 size = ENTRIES_PER_PAGE;
873 offset = shmem_find_swp(entry, ptr, ptr+size);
875 shmem_dir_unmap(dir);
878 shmem_swp_unmap(ptr);
882 shmem_dir_unmap(dir-1);
884 spin_unlock(&info->lock);
888 inode = &info->vfs_inode;
889 error = add_to_page_cache(page, inode->i_mapping, idx, GFP_ATOMIC);
890 if (error == -EEXIST) {
891 struct page *filepage = find_get_page(inode->i_mapping, idx);
894 * There might be a more uptodate page coming down
895 * from a stacked writepage: forget our swappage if so.
897 if (PageUptodate(filepage))
899 page_cache_release(filepage);
903 delete_from_swap_cache(page);
904 set_page_dirty(page);
905 info->flags |= SHMEM_PAGEIN;
906 shmem_swp_set(info, ptr + offset, 0);
908 shmem_swp_unmap(ptr);
909 spin_unlock(&info->lock);
911 * Decrement swap count even when the entry is left behind:
912 * try_to_unuse will skip over mms, then reincrement count.
919 * shmem_unuse() search for an eventually swapped out shmem page.
921 int shmem_unuse(swp_entry_t entry, struct page *page)
923 struct list_head *p, *next;
924 struct shmem_inode_info *info;
927 spin_lock(&shmem_swaplist_lock);
928 list_for_each_safe(p, next, &shmem_swaplist) {
929 info = list_entry(p, struct shmem_inode_info, swaplist);
931 list_del_init(&info->swaplist);
932 else if (shmem_unuse_inode(info, entry, page)) {
933 /* move head to start search for next from here */
934 list_move_tail(&shmem_swaplist, &info->swaplist);
939 spin_unlock(&shmem_swaplist_lock);
944 * Move the page from the page cache to the swap cache.
946 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
948 struct shmem_inode_info *info;
949 swp_entry_t *entry, swap;
950 struct address_space *mapping;
954 BUG_ON(!PageLocked(page));
955 mapping = page->mapping;
957 inode = mapping->host;
958 info = SHMEM_I(inode);
959 if (info->flags & VM_LOCKED)
961 if (!total_swap_pages)
965 * shmem_backing_dev_info's capabilities prevent regular writeback or
966 * sync from ever calling shmem_writepage; but a stacking filesystem
967 * may use the ->writepage of its underlying filesystem, in which case
968 * tmpfs should write out to swap only in response to memory pressure,
969 * and not for pdflush or sync. However, in those cases, we do still
970 * want to check if there's a redundant swappage to be discarded.
972 if (wbc->for_reclaim)
973 swap = get_swap_page();
977 spin_lock(&info->lock);
978 if (index >= info->next_index) {
979 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
982 entry = shmem_swp_entry(info, index, NULL);
985 * The more uptodate page coming down from a stacked
986 * writepage should replace our old swappage.
988 free_swap_and_cache(*entry);
989 shmem_swp_set(info, entry, 0);
991 shmem_recalc_inode(inode);
993 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
994 remove_from_page_cache(page);
995 shmem_swp_set(info, entry, swap.val);
996 shmem_swp_unmap(entry);
997 spin_unlock(&info->lock);
998 if (list_empty(&info->swaplist)) {
999 spin_lock(&shmem_swaplist_lock);
1000 /* move instead of add in case we're racing */
1001 list_move_tail(&info->swaplist, &shmem_swaplist);
1002 spin_unlock(&shmem_swaplist_lock);
1004 swap_duplicate(swap);
1005 BUG_ON(page_mapped(page));
1006 page_cache_release(page); /* pagecache ref */
1007 set_page_dirty(page);
1012 shmem_swp_unmap(entry);
1014 spin_unlock(&info->lock);
1017 set_page_dirty(page);
1018 if (wbc->for_reclaim)
1019 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1025 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
1027 char *nodelist = strchr(value, ':');
1031 /* NUL-terminate policy string */
1033 if (nodelist_parse(nodelist, *policy_nodes))
1035 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
1038 if (!strcmp(value, "default")) {
1039 *policy = MPOL_DEFAULT;
1040 /* Don't allow a nodelist */
1043 } else if (!strcmp(value, "prefer")) {
1044 *policy = MPOL_PREFERRED;
1045 /* Insist on a nodelist of one node only */
1047 char *rest = nodelist;
1048 while (isdigit(*rest))
1053 } else if (!strcmp(value, "bind")) {
1054 *policy = MPOL_BIND;
1055 /* Insist on a nodelist */
1058 } else if (!strcmp(value, "interleave")) {
1059 *policy = MPOL_INTERLEAVE;
1061 * Default to online nodes with memory if no nodelist
1064 *policy_nodes = node_states[N_HIGH_MEMORY];
1068 /* Restore string for error message */
1074 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1075 struct shmem_inode_info *info, unsigned long idx)
1077 struct vm_area_struct pvma;
1080 /* Create a pseudo vma that just contains the policy */
1082 pvma.vm_pgoff = idx;
1084 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1085 page = swapin_readahead(entry, gfp, &pvma, 0);
1086 mpol_free(pvma.vm_policy);
1090 static struct page *shmem_alloc_page(gfp_t gfp,
1091 struct shmem_inode_info *info, unsigned long idx)
1093 struct vm_area_struct pvma;
1096 /* Create a pseudo vma that just contains the policy */
1098 pvma.vm_pgoff = idx;
1100 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1101 page = alloc_page_vma(gfp, &pvma, 0);
1102 mpol_free(pvma.vm_policy);
1106 static inline int shmem_parse_mpol(char *value, int *policy,
1107 nodemask_t *policy_nodes)
1112 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1113 struct shmem_inode_info *info, unsigned long idx)
1115 return swapin_readahead(entry, gfp, NULL, 0);
1118 static inline struct page *shmem_alloc_page(gfp_t gfp,
1119 struct shmem_inode_info *info, unsigned long idx)
1121 return alloc_page(gfp);
1126 * shmem_getpage - either get the page from swap or allocate a new one
1128 * If we allocate a new one we do not mark it dirty. That's up to the
1129 * vm. If we swap it in we mark it dirty since we also free the swap
1130 * entry since a page cannot live in both the swap and page cache
1132 static int shmem_getpage(struct inode *inode, unsigned long idx,
1133 struct page **pagep, enum sgp_type sgp, int *type)
1135 struct address_space *mapping = inode->i_mapping;
1136 struct shmem_inode_info *info = SHMEM_I(inode);
1137 struct shmem_sb_info *sbinfo;
1138 struct page *filepage = *pagep;
1139 struct page *swappage;
1145 if (idx >= SHMEM_MAX_INDEX)
1152 * Normally, filepage is NULL on entry, and either found
1153 * uptodate immediately, or allocated and zeroed, or read
1154 * in under swappage, which is then assigned to filepage.
1155 * But shmem_readpage (required for splice) passes in a locked
1156 * filepage, which may be found not uptodate by other callers
1157 * too, and may need to be copied from the swappage read in.
1161 filepage = find_lock_page(mapping, idx);
1162 if (filepage && PageUptodate(filepage))
1165 gfp = mapping_gfp_mask(mapping);
1167 spin_lock(&info->lock);
1168 shmem_recalc_inode(inode);
1169 entry = shmem_swp_alloc(info, idx, sgp);
1170 if (IS_ERR(entry)) {
1171 spin_unlock(&info->lock);
1172 error = PTR_ERR(entry);
1178 /* Look it up and read it in.. */
1179 swappage = lookup_swap_cache(swap);
1181 shmem_swp_unmap(entry);
1182 /* here we actually do the io */
1183 if (type && !(*type & VM_FAULT_MAJOR)) {
1184 __count_vm_event(PGMAJFAULT);
1185 *type |= VM_FAULT_MAJOR;
1187 spin_unlock(&info->lock);
1188 swappage = shmem_swapin(swap, gfp, info, idx);
1190 spin_lock(&info->lock);
1191 entry = shmem_swp_alloc(info, idx, sgp);
1193 error = PTR_ERR(entry);
1195 if (entry->val == swap.val)
1197 shmem_swp_unmap(entry);
1199 spin_unlock(&info->lock);
1204 wait_on_page_locked(swappage);
1205 page_cache_release(swappage);
1209 /* We have to do this with page locked to prevent races */
1210 if (TestSetPageLocked(swappage)) {
1211 shmem_swp_unmap(entry);
1212 spin_unlock(&info->lock);
1213 wait_on_page_locked(swappage);
1214 page_cache_release(swappage);
1217 if (PageWriteback(swappage)) {
1218 shmem_swp_unmap(entry);
1219 spin_unlock(&info->lock);
1220 wait_on_page_writeback(swappage);
1221 unlock_page(swappage);
1222 page_cache_release(swappage);
1225 if (!PageUptodate(swappage)) {
1226 shmem_swp_unmap(entry);
1227 spin_unlock(&info->lock);
1228 unlock_page(swappage);
1229 page_cache_release(swappage);
1235 shmem_swp_set(info, entry, 0);
1236 shmem_swp_unmap(entry);
1237 delete_from_swap_cache(swappage);
1238 spin_unlock(&info->lock);
1239 copy_highpage(filepage, swappage);
1240 unlock_page(swappage);
1241 page_cache_release(swappage);
1242 flush_dcache_page(filepage);
1243 SetPageUptodate(filepage);
1244 set_page_dirty(filepage);
1246 } else if (!(error = add_to_page_cache(
1247 swappage, mapping, idx, GFP_ATOMIC))) {
1248 info->flags |= SHMEM_PAGEIN;
1249 shmem_swp_set(info, entry, 0);
1250 shmem_swp_unmap(entry);
1251 delete_from_swap_cache(swappage);
1252 spin_unlock(&info->lock);
1253 filepage = swappage;
1254 set_page_dirty(filepage);
1257 shmem_swp_unmap(entry);
1258 spin_unlock(&info->lock);
1259 unlock_page(swappage);
1260 page_cache_release(swappage);
1261 if (error == -ENOMEM) {
1262 /* let kswapd refresh zone for GFP_ATOMICs */
1263 congestion_wait(WRITE, HZ/50);
1267 } else if (sgp == SGP_READ && !filepage) {
1268 shmem_swp_unmap(entry);
1269 filepage = find_get_page(mapping, idx);
1271 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1272 spin_unlock(&info->lock);
1273 wait_on_page_locked(filepage);
1274 page_cache_release(filepage);
1278 spin_unlock(&info->lock);
1280 shmem_swp_unmap(entry);
1281 sbinfo = SHMEM_SB(inode->i_sb);
1282 if (sbinfo->max_blocks) {
1283 spin_lock(&sbinfo->stat_lock);
1284 if (sbinfo->free_blocks == 0 ||
1285 shmem_acct_block(info->flags)) {
1286 spin_unlock(&sbinfo->stat_lock);
1287 spin_unlock(&info->lock);
1291 sbinfo->free_blocks--;
1292 inode->i_blocks += BLOCKS_PER_PAGE;
1293 spin_unlock(&sbinfo->stat_lock);
1294 } else if (shmem_acct_block(info->flags)) {
1295 spin_unlock(&info->lock);
1301 spin_unlock(&info->lock);
1302 filepage = shmem_alloc_page(gfp, info, idx);
1304 shmem_unacct_blocks(info->flags, 1);
1305 shmem_free_blocks(inode, 1);
1310 spin_lock(&info->lock);
1311 entry = shmem_swp_alloc(info, idx, sgp);
1313 error = PTR_ERR(entry);
1316 shmem_swp_unmap(entry);
1318 if (error || swap.val || 0 != add_to_page_cache_lru(
1319 filepage, mapping, idx, GFP_ATOMIC)) {
1320 spin_unlock(&info->lock);
1321 page_cache_release(filepage);
1322 shmem_unacct_blocks(info->flags, 1);
1323 shmem_free_blocks(inode, 1);
1329 info->flags |= SHMEM_PAGEIN;
1333 spin_unlock(&info->lock);
1334 clear_highpage(filepage);
1335 flush_dcache_page(filepage);
1336 SetPageUptodate(filepage);
1337 if (sgp == SGP_DIRTY)
1338 set_page_dirty(filepage);
1345 if (*pagep != filepage) {
1346 unlock_page(filepage);
1347 page_cache_release(filepage);
1352 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1354 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1358 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1359 return VM_FAULT_SIGBUS;
1361 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1363 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1365 mark_page_accessed(vmf->page);
1366 return ret | VM_FAULT_LOCKED;
1370 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1372 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1373 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1376 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1379 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1382 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1383 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1387 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1389 struct inode *inode = file->f_path.dentry->d_inode;
1390 struct shmem_inode_info *info = SHMEM_I(inode);
1391 int retval = -ENOMEM;
1393 spin_lock(&info->lock);
1394 if (lock && !(info->flags & VM_LOCKED)) {
1395 if (!user_shm_lock(inode->i_size, user))
1397 info->flags |= VM_LOCKED;
1399 if (!lock && (info->flags & VM_LOCKED) && user) {
1400 user_shm_unlock(inode->i_size, user);
1401 info->flags &= ~VM_LOCKED;
1405 spin_unlock(&info->lock);
1409 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1411 file_accessed(file);
1412 vma->vm_ops = &shmem_vm_ops;
1413 vma->vm_flags |= VM_CAN_NONLINEAR;
1417 static struct inode *
1418 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1420 struct inode *inode;
1421 struct shmem_inode_info *info;
1422 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1424 if (shmem_reserve_inode(sb))
1427 inode = new_inode(sb);
1429 inode->i_mode = mode;
1430 inode->i_uid = current->fsuid;
1431 inode->i_gid = current->fsgid;
1432 inode->i_blocks = 0;
1433 inode->i_mapping->a_ops = &shmem_aops;
1434 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1435 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1436 inode->i_generation = get_seconds();
1437 info = SHMEM_I(inode);
1438 memset(info, 0, (char *)inode - (char *)info);
1439 spin_lock_init(&info->lock);
1440 INIT_LIST_HEAD(&info->swaplist);
1442 switch (mode & S_IFMT) {
1444 inode->i_op = &shmem_special_inode_operations;
1445 init_special_inode(inode, mode, dev);
1448 inode->i_op = &shmem_inode_operations;
1449 inode->i_fop = &shmem_file_operations;
1450 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1451 &sbinfo->policy_nodes);
1455 /* Some things misbehave if size == 0 on a directory */
1456 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1457 inode->i_op = &shmem_dir_inode_operations;
1458 inode->i_fop = &simple_dir_operations;
1462 * Must not load anything in the rbtree,
1463 * mpol_free_shared_policy will not be called.
1465 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1470 shmem_free_inode(sb);
1475 static const struct inode_operations shmem_symlink_inode_operations;
1476 static const struct inode_operations shmem_symlink_inline_operations;
1479 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1480 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1481 * below the loop driver, in the generic fashion that many filesystems support.
1483 static int shmem_readpage(struct file *file, struct page *page)
1485 struct inode *inode = page->mapping->host;
1486 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1492 shmem_write_begin(struct file *file, struct address_space *mapping,
1493 loff_t pos, unsigned len, unsigned flags,
1494 struct page **pagep, void **fsdata)
1496 struct inode *inode = mapping->host;
1497 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1499 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1503 shmem_write_end(struct file *file, struct address_space *mapping,
1504 loff_t pos, unsigned len, unsigned copied,
1505 struct page *page, void *fsdata)
1507 struct inode *inode = mapping->host;
1509 if (pos + copied > inode->i_size)
1510 i_size_write(inode, pos + copied);
1513 set_page_dirty(page);
1514 page_cache_release(page);
1519 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1521 struct inode *inode = filp->f_path.dentry->d_inode;
1522 struct address_space *mapping = inode->i_mapping;
1523 unsigned long index, offset;
1524 enum sgp_type sgp = SGP_READ;
1527 * Might this read be for a stacking filesystem? Then when reading
1528 * holes of a sparse file, we actually need to allocate those pages,
1529 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1531 if (segment_eq(get_fs(), KERNEL_DS))
1534 index = *ppos >> PAGE_CACHE_SHIFT;
1535 offset = *ppos & ~PAGE_CACHE_MASK;
1538 struct page *page = NULL;
1539 unsigned long end_index, nr, ret;
1540 loff_t i_size = i_size_read(inode);
1542 end_index = i_size >> PAGE_CACHE_SHIFT;
1543 if (index > end_index)
1545 if (index == end_index) {
1546 nr = i_size & ~PAGE_CACHE_MASK;
1551 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1553 if (desc->error == -EINVAL)
1561 * We must evaluate after, since reads (unlike writes)
1562 * are called without i_mutex protection against truncate
1564 nr = PAGE_CACHE_SIZE;
1565 i_size = i_size_read(inode);
1566 end_index = i_size >> PAGE_CACHE_SHIFT;
1567 if (index == end_index) {
1568 nr = i_size & ~PAGE_CACHE_MASK;
1571 page_cache_release(page);
1579 * If users can be writing to this page using arbitrary
1580 * virtual addresses, take care about potential aliasing
1581 * before reading the page on the kernel side.
1583 if (mapping_writably_mapped(mapping))
1584 flush_dcache_page(page);
1586 * Mark the page accessed if we read the beginning.
1589 mark_page_accessed(page);
1591 page = ZERO_PAGE(0);
1592 page_cache_get(page);
1596 * Ok, we have the page, and it's up-to-date, so
1597 * now we can copy it to user space...
1599 * The actor routine returns how many bytes were actually used..
1600 * NOTE! This may not be the same as how much of a user buffer
1601 * we filled up (we may be padding etc), so we can only update
1602 * "pos" here (the actor routine has to update the user buffer
1603 * pointers and the remaining count).
1605 ret = actor(desc, page, offset, nr);
1607 index += offset >> PAGE_CACHE_SHIFT;
1608 offset &= ~PAGE_CACHE_MASK;
1610 page_cache_release(page);
1611 if (ret != nr || !desc->count)
1617 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1618 file_accessed(filp);
1621 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1623 read_descriptor_t desc;
1625 if ((ssize_t) count < 0)
1627 if (!access_ok(VERIFY_WRITE, buf, count))
1637 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1639 return desc.written;
1643 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1645 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1647 buf->f_type = TMPFS_MAGIC;
1648 buf->f_bsize = PAGE_CACHE_SIZE;
1649 buf->f_namelen = NAME_MAX;
1650 spin_lock(&sbinfo->stat_lock);
1651 if (sbinfo->max_blocks) {
1652 buf->f_blocks = sbinfo->max_blocks;
1653 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1655 if (sbinfo->max_inodes) {
1656 buf->f_files = sbinfo->max_inodes;
1657 buf->f_ffree = sbinfo->free_inodes;
1659 /* else leave those fields 0 like simple_statfs */
1660 spin_unlock(&sbinfo->stat_lock);
1665 * File creation. Allocate an inode, and we're done..
1668 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1670 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1671 int error = -ENOSPC;
1674 error = security_inode_init_security(inode, dir, NULL, NULL,
1677 if (error != -EOPNOTSUPP) {
1682 error = shmem_acl_init(inode, dir);
1687 if (dir->i_mode & S_ISGID) {
1688 inode->i_gid = dir->i_gid;
1690 inode->i_mode |= S_ISGID;
1692 dir->i_size += BOGO_DIRENT_SIZE;
1693 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1694 d_instantiate(dentry, inode);
1695 dget(dentry); /* Extra count - pin the dentry in core */
1700 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1704 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1710 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1711 struct nameidata *nd)
1713 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1719 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1721 struct inode *inode = old_dentry->d_inode;
1725 * No ordinary (disk based) filesystem counts links as inodes;
1726 * but each new link needs a new dentry, pinning lowmem, and
1727 * tmpfs dentries cannot be pruned until they are unlinked.
1729 ret = shmem_reserve_inode(inode->i_sb);
1733 dir->i_size += BOGO_DIRENT_SIZE;
1734 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1736 atomic_inc(&inode->i_count); /* New dentry reference */
1737 dget(dentry); /* Extra pinning count for the created dentry */
1738 d_instantiate(dentry, inode);
1743 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1745 struct inode *inode = dentry->d_inode;
1747 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1748 shmem_free_inode(inode->i_sb);
1750 dir->i_size -= BOGO_DIRENT_SIZE;
1751 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1753 dput(dentry); /* Undo the count from "create" - this does all the work */
1757 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1759 if (!simple_empty(dentry))
1762 drop_nlink(dentry->d_inode);
1764 return shmem_unlink(dir, dentry);
1768 * The VFS layer already does all the dentry stuff for rename,
1769 * we just have to decrement the usage count for the target if
1770 * it exists so that the VFS layer correctly free's it when it
1773 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1775 struct inode *inode = old_dentry->d_inode;
1776 int they_are_dirs = S_ISDIR(inode->i_mode);
1778 if (!simple_empty(new_dentry))
1781 if (new_dentry->d_inode) {
1782 (void) shmem_unlink(new_dir, new_dentry);
1784 drop_nlink(old_dir);
1785 } else if (they_are_dirs) {
1786 drop_nlink(old_dir);
1790 old_dir->i_size -= BOGO_DIRENT_SIZE;
1791 new_dir->i_size += BOGO_DIRENT_SIZE;
1792 old_dir->i_ctime = old_dir->i_mtime =
1793 new_dir->i_ctime = new_dir->i_mtime =
1794 inode->i_ctime = CURRENT_TIME;
1798 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1802 struct inode *inode;
1803 struct page *page = NULL;
1805 struct shmem_inode_info *info;
1807 len = strlen(symname) + 1;
1808 if (len > PAGE_CACHE_SIZE)
1809 return -ENAMETOOLONG;
1811 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1815 error = security_inode_init_security(inode, dir, NULL, NULL,
1818 if (error != -EOPNOTSUPP) {
1825 info = SHMEM_I(inode);
1826 inode->i_size = len-1;
1827 if (len <= (char *)inode - (char *)info) {
1829 memcpy(info, symname, len);
1830 inode->i_op = &shmem_symlink_inline_operations;
1832 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1838 inode->i_op = &shmem_symlink_inode_operations;
1839 kaddr = kmap_atomic(page, KM_USER0);
1840 memcpy(kaddr, symname, len);
1841 kunmap_atomic(kaddr, KM_USER0);
1842 set_page_dirty(page);
1843 page_cache_release(page);
1845 if (dir->i_mode & S_ISGID)
1846 inode->i_gid = dir->i_gid;
1847 dir->i_size += BOGO_DIRENT_SIZE;
1848 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1849 d_instantiate(dentry, inode);
1854 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1856 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1860 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1862 struct page *page = NULL;
1863 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1864 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1870 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1872 if (!IS_ERR(nd_get_link(nd))) {
1873 struct page *page = cookie;
1875 mark_page_accessed(page);
1876 page_cache_release(page);
1880 static const struct inode_operations shmem_symlink_inline_operations = {
1881 .readlink = generic_readlink,
1882 .follow_link = shmem_follow_link_inline,
1885 static const struct inode_operations shmem_symlink_inode_operations = {
1886 .truncate = shmem_truncate,
1887 .readlink = generic_readlink,
1888 .follow_link = shmem_follow_link,
1889 .put_link = shmem_put_link,
1892 #ifdef CONFIG_TMPFS_POSIX_ACL
1894 * Superblocks without xattr inode operations will get security.* xattr
1895 * support from the VFS "for free". As soon as we have any other xattrs
1896 * like ACLs, we also need to implement the security.* handlers at
1897 * filesystem level, though.
1900 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1901 size_t list_len, const char *name,
1904 return security_inode_listsecurity(inode, list, list_len);
1907 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1908 void *buffer, size_t size)
1910 if (strcmp(name, "") == 0)
1912 return security_inode_getsecurity(inode, name, buffer, size,
1916 static int shmem_xattr_security_set(struct inode *inode, const char *name,
1917 const void *value, size_t size, int flags)
1919 if (strcmp(name, "") == 0)
1921 return security_inode_setsecurity(inode, name, value, size, flags);
1924 static struct xattr_handler shmem_xattr_security_handler = {
1925 .prefix = XATTR_SECURITY_PREFIX,
1926 .list = shmem_xattr_security_list,
1927 .get = shmem_xattr_security_get,
1928 .set = shmem_xattr_security_set,
1931 static struct xattr_handler *shmem_xattr_handlers[] = {
1932 &shmem_xattr_acl_access_handler,
1933 &shmem_xattr_acl_default_handler,
1934 &shmem_xattr_security_handler,
1939 static struct dentry *shmem_get_parent(struct dentry *child)
1941 return ERR_PTR(-ESTALE);
1944 static int shmem_match(struct inode *ino, void *vfh)
1948 inum = (inum << 32) | fh[1];
1949 return ino->i_ino == inum && fh[0] == ino->i_generation;
1952 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1953 struct fid *fid, int fh_len, int fh_type)
1955 struct inode *inode;
1956 struct dentry *dentry = NULL;
1957 u64 inum = fid->raw[2];
1958 inum = (inum << 32) | fid->raw[1];
1963 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1964 shmem_match, fid->raw);
1966 dentry = d_find_alias(inode);
1973 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1976 struct inode *inode = dentry->d_inode;
1981 if (hlist_unhashed(&inode->i_hash)) {
1982 /* Unfortunately insert_inode_hash is not idempotent,
1983 * so as we hash inodes here rather than at creation
1984 * time, we need a lock to ensure we only try
1987 static DEFINE_SPINLOCK(lock);
1989 if (hlist_unhashed(&inode->i_hash))
1990 __insert_inode_hash(inode,
1991 inode->i_ino + inode->i_generation);
1995 fh[0] = inode->i_generation;
1996 fh[1] = inode->i_ino;
1997 fh[2] = ((__u64)inode->i_ino) >> 32;
2003 static const struct export_operations shmem_export_ops = {
2004 .get_parent = shmem_get_parent,
2005 .encode_fh = shmem_encode_fh,
2006 .fh_to_dentry = shmem_fh_to_dentry,
2009 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2010 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2011 int *policy, nodemask_t *policy_nodes)
2013 char *this_char, *value, *rest;
2015 while (options != NULL) {
2016 this_char = options;
2019 * NUL-terminate this option: unfortunately,
2020 * mount options form a comma-separated list,
2021 * but mpol's nodelist may also contain commas.
2023 options = strchr(options, ',');
2024 if (options == NULL)
2027 if (!isdigit(*options)) {
2034 if ((value = strchr(this_char,'=')) != NULL) {
2038 "tmpfs: No value for mount option '%s'\n",
2043 if (!strcmp(this_char,"size")) {
2044 unsigned long long size;
2045 size = memparse(value,&rest);
2047 size <<= PAGE_SHIFT;
2048 size *= totalram_pages;
2054 *blocks = DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2055 } else if (!strcmp(this_char,"nr_blocks")) {
2056 *blocks = memparse(value,&rest);
2059 } else if (!strcmp(this_char,"nr_inodes")) {
2060 *inodes = memparse(value,&rest);
2063 } else if (!strcmp(this_char,"mode")) {
2066 *mode = simple_strtoul(value,&rest,8);
2069 } else if (!strcmp(this_char,"uid")) {
2072 *uid = simple_strtoul(value,&rest,0);
2075 } else if (!strcmp(this_char,"gid")) {
2078 *gid = simple_strtoul(value,&rest,0);
2081 } else if (!strcmp(this_char,"mpol")) {
2082 if (shmem_parse_mpol(value,policy,policy_nodes))
2085 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2093 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2099 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2101 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2102 unsigned long max_blocks = sbinfo->max_blocks;
2103 unsigned long max_inodes = sbinfo->max_inodes;
2104 int policy = sbinfo->policy;
2105 nodemask_t policy_nodes = sbinfo->policy_nodes;
2106 unsigned long blocks;
2107 unsigned long inodes;
2108 int error = -EINVAL;
2110 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2111 &max_inodes, &policy, &policy_nodes))
2114 spin_lock(&sbinfo->stat_lock);
2115 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2116 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2117 if (max_blocks < blocks)
2119 if (max_inodes < inodes)
2122 * Those tests also disallow limited->unlimited while any are in
2123 * use, so i_blocks will always be zero when max_blocks is zero;
2124 * but we must separately disallow unlimited->limited, because
2125 * in that case we have no record of how much is already in use.
2127 if (max_blocks && !sbinfo->max_blocks)
2129 if (max_inodes && !sbinfo->max_inodes)
2133 sbinfo->max_blocks = max_blocks;
2134 sbinfo->free_blocks = max_blocks - blocks;
2135 sbinfo->max_inodes = max_inodes;
2136 sbinfo->free_inodes = max_inodes - inodes;
2137 sbinfo->policy = policy;
2138 sbinfo->policy_nodes = policy_nodes;
2140 spin_unlock(&sbinfo->stat_lock);
2145 static void shmem_put_super(struct super_block *sb)
2147 kfree(sb->s_fs_info);
2148 sb->s_fs_info = NULL;
2151 static int shmem_fill_super(struct super_block *sb,
2152 void *data, int silent)
2154 struct inode *inode;
2155 struct dentry *root;
2156 int mode = S_IRWXUGO | S_ISVTX;
2157 uid_t uid = current->fsuid;
2158 gid_t gid = current->fsgid;
2160 struct shmem_sb_info *sbinfo;
2161 unsigned long blocks = 0;
2162 unsigned long inodes = 0;
2163 int policy = MPOL_DEFAULT;
2164 nodemask_t policy_nodes = node_states[N_HIGH_MEMORY];
2168 * Per default we only allow half of the physical ram per
2169 * tmpfs instance, limiting inodes to one per page of lowmem;
2170 * but the internal instance is left unlimited.
2172 if (!(sb->s_flags & MS_NOUSER)) {
2173 blocks = totalram_pages / 2;
2174 inodes = totalram_pages - totalhigh_pages;
2175 if (inodes > blocks)
2177 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2178 &inodes, &policy, &policy_nodes))
2181 sb->s_export_op = &shmem_export_ops;
2183 sb->s_flags |= MS_NOUSER;
2186 /* Round up to L1_CACHE_BYTES to resist false sharing */
2187 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2188 L1_CACHE_BYTES), GFP_KERNEL);
2192 spin_lock_init(&sbinfo->stat_lock);
2193 sbinfo->max_blocks = blocks;
2194 sbinfo->free_blocks = blocks;
2195 sbinfo->max_inodes = inodes;
2196 sbinfo->free_inodes = inodes;
2197 sbinfo->policy = policy;
2198 sbinfo->policy_nodes = policy_nodes;
2200 sb->s_fs_info = sbinfo;
2201 sb->s_maxbytes = SHMEM_MAX_BYTES;
2202 sb->s_blocksize = PAGE_CACHE_SIZE;
2203 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2204 sb->s_magic = TMPFS_MAGIC;
2205 sb->s_op = &shmem_ops;
2206 sb->s_time_gran = 1;
2207 #ifdef CONFIG_TMPFS_POSIX_ACL
2208 sb->s_xattr = shmem_xattr_handlers;
2209 sb->s_flags |= MS_POSIXACL;
2212 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2217 root = d_alloc_root(inode);
2226 shmem_put_super(sb);
2230 static struct kmem_cache *shmem_inode_cachep;
2232 static struct inode *shmem_alloc_inode(struct super_block *sb)
2234 struct shmem_inode_info *p;
2235 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2238 return &p->vfs_inode;
2241 static void shmem_destroy_inode(struct inode *inode)
2243 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2244 /* only struct inode is valid if it's an inline symlink */
2245 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2247 shmem_acl_destroy_inode(inode);
2248 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2251 static void init_once(struct kmem_cache *cachep, void *foo)
2253 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2255 inode_init_once(&p->vfs_inode);
2256 #ifdef CONFIG_TMPFS_POSIX_ACL
2258 p->i_default_acl = NULL;
2262 static int init_inodecache(void)
2264 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2265 sizeof(struct shmem_inode_info),
2266 0, SLAB_PANIC, init_once);
2270 static void destroy_inodecache(void)
2272 kmem_cache_destroy(shmem_inode_cachep);
2275 static const struct address_space_operations shmem_aops = {
2276 .writepage = shmem_writepage,
2277 .set_page_dirty = __set_page_dirty_no_writeback,
2279 .readpage = shmem_readpage,
2280 .write_begin = shmem_write_begin,
2281 .write_end = shmem_write_end,
2283 .migratepage = migrate_page,
2286 static const struct file_operations shmem_file_operations = {
2289 .llseek = generic_file_llseek,
2290 .read = shmem_file_read,
2291 .write = do_sync_write,
2292 .aio_write = generic_file_aio_write,
2293 .fsync = simple_sync_file,
2294 .splice_read = generic_file_splice_read,
2295 .splice_write = generic_file_splice_write,
2299 static const struct inode_operations shmem_inode_operations = {
2300 .truncate = shmem_truncate,
2301 .setattr = shmem_notify_change,
2302 .truncate_range = shmem_truncate_range,
2303 #ifdef CONFIG_TMPFS_POSIX_ACL
2304 .setxattr = generic_setxattr,
2305 .getxattr = generic_getxattr,
2306 .listxattr = generic_listxattr,
2307 .removexattr = generic_removexattr,
2308 .permission = shmem_permission,
2313 static const struct inode_operations shmem_dir_inode_operations = {
2315 .create = shmem_create,
2316 .lookup = simple_lookup,
2318 .unlink = shmem_unlink,
2319 .symlink = shmem_symlink,
2320 .mkdir = shmem_mkdir,
2321 .rmdir = shmem_rmdir,
2322 .mknod = shmem_mknod,
2323 .rename = shmem_rename,
2325 #ifdef CONFIG_TMPFS_POSIX_ACL
2326 .setattr = shmem_notify_change,
2327 .setxattr = generic_setxattr,
2328 .getxattr = generic_getxattr,
2329 .listxattr = generic_listxattr,
2330 .removexattr = generic_removexattr,
2331 .permission = shmem_permission,
2335 static const struct inode_operations shmem_special_inode_operations = {
2336 #ifdef CONFIG_TMPFS_POSIX_ACL
2337 .setattr = shmem_notify_change,
2338 .setxattr = generic_setxattr,
2339 .getxattr = generic_getxattr,
2340 .listxattr = generic_listxattr,
2341 .removexattr = generic_removexattr,
2342 .permission = shmem_permission,
2346 static const struct super_operations shmem_ops = {
2347 .alloc_inode = shmem_alloc_inode,
2348 .destroy_inode = shmem_destroy_inode,
2350 .statfs = shmem_statfs,
2351 .remount_fs = shmem_remount_fs,
2353 .delete_inode = shmem_delete_inode,
2354 .drop_inode = generic_delete_inode,
2355 .put_super = shmem_put_super,
2358 static struct vm_operations_struct shmem_vm_ops = {
2359 .fault = shmem_fault,
2361 .set_policy = shmem_set_policy,
2362 .get_policy = shmem_get_policy,
2367 static int shmem_get_sb(struct file_system_type *fs_type,
2368 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2370 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2373 static struct file_system_type tmpfs_fs_type = {
2374 .owner = THIS_MODULE,
2376 .get_sb = shmem_get_sb,
2377 .kill_sb = kill_litter_super,
2379 static struct vfsmount *shm_mnt;
2381 static int __init init_tmpfs(void)
2385 error = bdi_init(&shmem_backing_dev_info);
2389 error = init_inodecache();
2393 error = register_filesystem(&tmpfs_fs_type);
2395 printk(KERN_ERR "Could not register tmpfs\n");
2399 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2400 tmpfs_fs_type.name, NULL);
2401 if (IS_ERR(shm_mnt)) {
2402 error = PTR_ERR(shm_mnt);
2403 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2409 unregister_filesystem(&tmpfs_fs_type);
2411 destroy_inodecache();
2413 bdi_destroy(&shmem_backing_dev_info);
2415 shm_mnt = ERR_PTR(error);
2418 module_init(init_tmpfs)
2421 * shmem_file_setup - get an unlinked file living in tmpfs
2423 * @name: name for dentry (to be seen in /proc/<pid>/maps
2424 * @size: size to be set for the file
2427 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2431 struct inode *inode;
2432 struct dentry *dentry, *root;
2435 if (IS_ERR(shm_mnt))
2436 return (void *)shm_mnt;
2438 if (size < 0 || size > SHMEM_MAX_BYTES)
2439 return ERR_PTR(-EINVAL);
2441 if (shmem_acct_size(flags, size))
2442 return ERR_PTR(-ENOMEM);
2446 this.len = strlen(name);
2447 this.hash = 0; /* will go */
2448 root = shm_mnt->mnt_root;
2449 dentry = d_alloc(root, &this);
2454 file = get_empty_filp();
2459 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2463 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2464 d_instantiate(dentry, inode);
2465 inode->i_size = size;
2466 inode->i_nlink = 0; /* It is unlinked */
2467 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2468 &shmem_file_operations);
2476 shmem_unacct_size(flags, size);
2477 return ERR_PTR(error);
2481 * shmem_zero_setup - setup a shared anonymous mapping
2483 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2485 int shmem_zero_setup(struct vm_area_struct *vma)
2488 loff_t size = vma->vm_end - vma->vm_start;
2490 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2492 return PTR_ERR(file);
2496 vma->vm_file = file;
2497 vma->vm_ops = &shmem_vm_ops;