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_MUTEX(shmem_swaplist_mutex);
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 mutex_lock(&shmem_swaplist_mutex);
800 list_del_init(&info->swaplist);
801 mutex_unlock(&shmem_swaplist_mutex);
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);
843 if (!info->i_indirect)
846 dir = shmem_dir_map(info->i_indirect);
847 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
849 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
850 if (unlikely(idx == stage)) {
851 shmem_dir_unmap(dir-1);
852 if (cond_resched_lock(&info->lock)) {
853 /* check it has not been truncated */
854 if (limit > info->next_index) {
855 limit = info->next_index;
860 dir = shmem_dir_map(info->i_indirect) +
861 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
864 idx += ENTRIES_PER_PAGEPAGE;
868 stage = idx + ENTRIES_PER_PAGEPAGE;
870 shmem_dir_unmap(dir);
871 dir = shmem_dir_map(subdir);
874 if (subdir && page_private(subdir)) {
875 ptr = shmem_swp_map(subdir);
877 if (size > ENTRIES_PER_PAGE)
878 size = ENTRIES_PER_PAGE;
879 offset = shmem_find_swp(entry, ptr, ptr+size);
880 shmem_swp_unmap(ptr);
882 shmem_dir_unmap(dir);
888 shmem_dir_unmap(dir-1);
890 spin_unlock(&info->lock);
894 inode = igrab(&info->vfs_inode);
895 spin_unlock(&info->lock);
897 /* move head to start search for next from here */
898 list_move_tail(&shmem_swaplist, &info->swaplist);
899 mutex_unlock(&shmem_swaplist_mutex);
904 error = radix_tree_preload(GFP_KERNEL);
909 spin_lock(&info->lock);
910 ptr = shmem_swp_entry(info, idx, NULL);
911 if (ptr && ptr->val == entry.val)
912 error = add_to_page_cache(page, inode->i_mapping,
914 if (error == -EEXIST) {
915 struct page *filepage = find_get_page(inode->i_mapping, idx);
919 * There might be a more uptodate page coming down
920 * from a stacked writepage: forget our swappage if so.
922 if (PageUptodate(filepage))
924 page_cache_release(filepage);
928 delete_from_swap_cache(page);
929 set_page_dirty(page);
930 info->flags |= SHMEM_PAGEIN;
931 shmem_swp_set(info, ptr, 0);
933 error = 1; /* not an error, but entry was found */
936 shmem_swp_unmap(ptr);
937 spin_unlock(&info->lock);
938 radix_tree_preload_end();
941 page_cache_release(page);
942 iput(inode); /* allows for NULL */
947 * shmem_unuse() search for an eventually swapped out shmem page.
949 int shmem_unuse(swp_entry_t entry, struct page *page)
951 struct list_head *p, *next;
952 struct shmem_inode_info *info;
955 mutex_lock(&shmem_swaplist_mutex);
956 list_for_each_safe(p, next, &shmem_swaplist) {
957 info = list_entry(p, struct shmem_inode_info, swaplist);
959 found = shmem_unuse_inode(info, entry, page);
961 list_del_init(&info->swaplist);
966 mutex_unlock(&shmem_swaplist_mutex);
967 out: return found; /* 0 or 1 or -ENOMEM */
971 * Move the page from the page cache to the swap cache.
973 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
975 struct shmem_inode_info *info;
976 swp_entry_t *entry, swap;
977 struct address_space *mapping;
981 BUG_ON(!PageLocked(page));
982 mapping = page->mapping;
984 inode = mapping->host;
985 info = SHMEM_I(inode);
986 if (info->flags & VM_LOCKED)
988 if (!total_swap_pages)
992 * shmem_backing_dev_info's capabilities prevent regular writeback or
993 * sync from ever calling shmem_writepage; but a stacking filesystem
994 * may use the ->writepage of its underlying filesystem, in which case
995 * tmpfs should write out to swap only in response to memory pressure,
996 * and not for pdflush or sync. However, in those cases, we do still
997 * want to check if there's a redundant swappage to be discarded.
999 if (wbc->for_reclaim)
1000 swap = get_swap_page();
1004 spin_lock(&info->lock);
1005 if (index >= info->next_index) {
1006 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1009 entry = shmem_swp_entry(info, index, NULL);
1012 * The more uptodate page coming down from a stacked
1013 * writepage should replace our old swappage.
1015 free_swap_and_cache(*entry);
1016 shmem_swp_set(info, entry, 0);
1018 shmem_recalc_inode(inode);
1020 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1021 remove_from_page_cache(page);
1022 shmem_swp_set(info, entry, swap.val);
1023 shmem_swp_unmap(entry);
1024 spin_unlock(&info->lock);
1025 if (list_empty(&info->swaplist)) {
1026 mutex_lock(&shmem_swaplist_mutex);
1027 /* move instead of add in case we're racing */
1028 list_move_tail(&info->swaplist, &shmem_swaplist);
1029 mutex_unlock(&shmem_swaplist_mutex);
1031 swap_duplicate(swap);
1032 BUG_ON(page_mapped(page));
1033 page_cache_release(page); /* pagecache ref */
1034 set_page_dirty(page);
1039 shmem_swp_unmap(entry);
1041 spin_unlock(&info->lock);
1044 set_page_dirty(page);
1045 if (wbc->for_reclaim)
1046 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1052 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
1054 char *nodelist = strchr(value, ':');
1058 /* NUL-terminate policy string */
1060 if (nodelist_parse(nodelist, *policy_nodes))
1062 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
1065 if (!strcmp(value, "default")) {
1066 *policy = MPOL_DEFAULT;
1067 /* Don't allow a nodelist */
1070 } else if (!strcmp(value, "prefer")) {
1071 *policy = MPOL_PREFERRED;
1072 /* Insist on a nodelist of one node only */
1074 char *rest = nodelist;
1075 while (isdigit(*rest))
1080 } else if (!strcmp(value, "bind")) {
1081 *policy = MPOL_BIND;
1082 /* Insist on a nodelist */
1085 } else if (!strcmp(value, "interleave")) {
1086 *policy = MPOL_INTERLEAVE;
1088 * Default to online nodes with memory if no nodelist
1091 *policy_nodes = node_states[N_HIGH_MEMORY];
1095 /* Restore string for error message */
1101 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1102 struct shmem_inode_info *info, unsigned long idx)
1104 struct vm_area_struct pvma;
1107 /* Create a pseudo vma that just contains the policy */
1109 pvma.vm_pgoff = idx;
1111 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1112 page = swapin_readahead(entry, gfp, &pvma, 0);
1113 mpol_free(pvma.vm_policy);
1117 static struct page *shmem_alloc_page(gfp_t gfp,
1118 struct shmem_inode_info *info, unsigned long idx)
1120 struct vm_area_struct pvma;
1123 /* Create a pseudo vma that just contains the policy */
1125 pvma.vm_pgoff = idx;
1127 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1128 page = alloc_page_vma(gfp, &pvma, 0);
1129 mpol_free(pvma.vm_policy);
1133 static inline int shmem_parse_mpol(char *value, int *policy,
1134 nodemask_t *policy_nodes)
1139 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1140 struct shmem_inode_info *info, unsigned long idx)
1142 return swapin_readahead(entry, gfp, NULL, 0);
1145 static inline struct page *shmem_alloc_page(gfp_t gfp,
1146 struct shmem_inode_info *info, unsigned long idx)
1148 return alloc_page(gfp);
1153 * shmem_getpage - either get the page from swap or allocate a new one
1155 * If we allocate a new one we do not mark it dirty. That's up to the
1156 * vm. If we swap it in we mark it dirty since we also free the swap
1157 * entry since a page cannot live in both the swap and page cache
1159 static int shmem_getpage(struct inode *inode, unsigned long idx,
1160 struct page **pagep, enum sgp_type sgp, int *type)
1162 struct address_space *mapping = inode->i_mapping;
1163 struct shmem_inode_info *info = SHMEM_I(inode);
1164 struct shmem_sb_info *sbinfo;
1165 struct page *filepage = *pagep;
1166 struct page *swappage;
1172 if (idx >= SHMEM_MAX_INDEX)
1179 * Normally, filepage is NULL on entry, and either found
1180 * uptodate immediately, or allocated and zeroed, or read
1181 * in under swappage, which is then assigned to filepage.
1182 * But shmem_readpage (required for splice) passes in a locked
1183 * filepage, which may be found not uptodate by other callers
1184 * too, and may need to be copied from the swappage read in.
1188 filepage = find_lock_page(mapping, idx);
1189 if (filepage && PageUptodate(filepage))
1192 gfp = mapping_gfp_mask(mapping);
1195 * Try to preload while we can wait, to not make a habit of
1196 * draining atomic reserves; but don't latch on to this cpu.
1198 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1201 radix_tree_preload_end();
1204 spin_lock(&info->lock);
1205 shmem_recalc_inode(inode);
1206 entry = shmem_swp_alloc(info, idx, sgp);
1207 if (IS_ERR(entry)) {
1208 spin_unlock(&info->lock);
1209 error = PTR_ERR(entry);
1215 /* Look it up and read it in.. */
1216 swappage = lookup_swap_cache(swap);
1218 shmem_swp_unmap(entry);
1219 /* here we actually do the io */
1220 if (type && !(*type & VM_FAULT_MAJOR)) {
1221 __count_vm_event(PGMAJFAULT);
1222 *type |= VM_FAULT_MAJOR;
1224 spin_unlock(&info->lock);
1225 swappage = shmem_swapin(swap, gfp, info, idx);
1227 spin_lock(&info->lock);
1228 entry = shmem_swp_alloc(info, idx, sgp);
1230 error = PTR_ERR(entry);
1232 if (entry->val == swap.val)
1234 shmem_swp_unmap(entry);
1236 spin_unlock(&info->lock);
1241 wait_on_page_locked(swappage);
1242 page_cache_release(swappage);
1246 /* We have to do this with page locked to prevent races */
1247 if (TestSetPageLocked(swappage)) {
1248 shmem_swp_unmap(entry);
1249 spin_unlock(&info->lock);
1250 wait_on_page_locked(swappage);
1251 page_cache_release(swappage);
1254 if (PageWriteback(swappage)) {
1255 shmem_swp_unmap(entry);
1256 spin_unlock(&info->lock);
1257 wait_on_page_writeback(swappage);
1258 unlock_page(swappage);
1259 page_cache_release(swappage);
1262 if (!PageUptodate(swappage)) {
1263 shmem_swp_unmap(entry);
1264 spin_unlock(&info->lock);
1265 unlock_page(swappage);
1266 page_cache_release(swappage);
1272 shmem_swp_set(info, entry, 0);
1273 shmem_swp_unmap(entry);
1274 delete_from_swap_cache(swappage);
1275 spin_unlock(&info->lock);
1276 copy_highpage(filepage, swappage);
1277 unlock_page(swappage);
1278 page_cache_release(swappage);
1279 flush_dcache_page(filepage);
1280 SetPageUptodate(filepage);
1281 set_page_dirty(filepage);
1283 } else if (!(error = add_to_page_cache(
1284 swappage, mapping, idx, GFP_NOWAIT))) {
1285 info->flags |= SHMEM_PAGEIN;
1286 shmem_swp_set(info, entry, 0);
1287 shmem_swp_unmap(entry);
1288 delete_from_swap_cache(swappage);
1289 spin_unlock(&info->lock);
1290 filepage = swappage;
1291 set_page_dirty(filepage);
1294 shmem_swp_unmap(entry);
1295 spin_unlock(&info->lock);
1296 unlock_page(swappage);
1297 page_cache_release(swappage);
1300 } else if (sgp == SGP_READ && !filepage) {
1301 shmem_swp_unmap(entry);
1302 filepage = find_get_page(mapping, idx);
1304 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1305 spin_unlock(&info->lock);
1306 wait_on_page_locked(filepage);
1307 page_cache_release(filepage);
1311 spin_unlock(&info->lock);
1313 shmem_swp_unmap(entry);
1314 sbinfo = SHMEM_SB(inode->i_sb);
1315 if (sbinfo->max_blocks) {
1316 spin_lock(&sbinfo->stat_lock);
1317 if (sbinfo->free_blocks == 0 ||
1318 shmem_acct_block(info->flags)) {
1319 spin_unlock(&sbinfo->stat_lock);
1320 spin_unlock(&info->lock);
1324 sbinfo->free_blocks--;
1325 inode->i_blocks += BLOCKS_PER_PAGE;
1326 spin_unlock(&sbinfo->stat_lock);
1327 } else if (shmem_acct_block(info->flags)) {
1328 spin_unlock(&info->lock);
1334 spin_unlock(&info->lock);
1335 filepage = shmem_alloc_page(gfp, info, idx);
1337 shmem_unacct_blocks(info->flags, 1);
1338 shmem_free_blocks(inode, 1);
1343 spin_lock(&info->lock);
1344 entry = shmem_swp_alloc(info, idx, sgp);
1346 error = PTR_ERR(entry);
1349 shmem_swp_unmap(entry);
1351 if (error || swap.val || 0 != add_to_page_cache_lru(
1352 filepage, mapping, idx, GFP_NOWAIT)) {
1353 spin_unlock(&info->lock);
1354 page_cache_release(filepage);
1355 shmem_unacct_blocks(info->flags, 1);
1356 shmem_free_blocks(inode, 1);
1362 info->flags |= SHMEM_PAGEIN;
1366 spin_unlock(&info->lock);
1367 clear_highpage(filepage);
1368 flush_dcache_page(filepage);
1369 SetPageUptodate(filepage);
1370 if (sgp == SGP_DIRTY)
1371 set_page_dirty(filepage);
1378 if (*pagep != filepage) {
1379 unlock_page(filepage);
1380 page_cache_release(filepage);
1385 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1387 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1391 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1392 return VM_FAULT_SIGBUS;
1394 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1396 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1398 mark_page_accessed(vmf->page);
1399 return ret | VM_FAULT_LOCKED;
1403 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1405 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1406 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1409 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1412 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1415 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1416 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1420 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1422 struct inode *inode = file->f_path.dentry->d_inode;
1423 struct shmem_inode_info *info = SHMEM_I(inode);
1424 int retval = -ENOMEM;
1426 spin_lock(&info->lock);
1427 if (lock && !(info->flags & VM_LOCKED)) {
1428 if (!user_shm_lock(inode->i_size, user))
1430 info->flags |= VM_LOCKED;
1432 if (!lock && (info->flags & VM_LOCKED) && user) {
1433 user_shm_unlock(inode->i_size, user);
1434 info->flags &= ~VM_LOCKED;
1438 spin_unlock(&info->lock);
1442 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1444 file_accessed(file);
1445 vma->vm_ops = &shmem_vm_ops;
1446 vma->vm_flags |= VM_CAN_NONLINEAR;
1450 static struct inode *
1451 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1453 struct inode *inode;
1454 struct shmem_inode_info *info;
1455 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1457 if (shmem_reserve_inode(sb))
1460 inode = new_inode(sb);
1462 inode->i_mode = mode;
1463 inode->i_uid = current->fsuid;
1464 inode->i_gid = current->fsgid;
1465 inode->i_blocks = 0;
1466 inode->i_mapping->a_ops = &shmem_aops;
1467 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1468 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1469 inode->i_generation = get_seconds();
1470 info = SHMEM_I(inode);
1471 memset(info, 0, (char *)inode - (char *)info);
1472 spin_lock_init(&info->lock);
1473 INIT_LIST_HEAD(&info->swaplist);
1475 switch (mode & S_IFMT) {
1477 inode->i_op = &shmem_special_inode_operations;
1478 init_special_inode(inode, mode, dev);
1481 inode->i_op = &shmem_inode_operations;
1482 inode->i_fop = &shmem_file_operations;
1483 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1484 &sbinfo->policy_nodes);
1488 /* Some things misbehave if size == 0 on a directory */
1489 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1490 inode->i_op = &shmem_dir_inode_operations;
1491 inode->i_fop = &simple_dir_operations;
1495 * Must not load anything in the rbtree,
1496 * mpol_free_shared_policy will not be called.
1498 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1503 shmem_free_inode(sb);
1508 static const struct inode_operations shmem_symlink_inode_operations;
1509 static const struct inode_operations shmem_symlink_inline_operations;
1512 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1513 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1514 * below the loop driver, in the generic fashion that many filesystems support.
1516 static int shmem_readpage(struct file *file, struct page *page)
1518 struct inode *inode = page->mapping->host;
1519 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1525 shmem_write_begin(struct file *file, struct address_space *mapping,
1526 loff_t pos, unsigned len, unsigned flags,
1527 struct page **pagep, void **fsdata)
1529 struct inode *inode = mapping->host;
1530 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1532 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1536 shmem_write_end(struct file *file, struct address_space *mapping,
1537 loff_t pos, unsigned len, unsigned copied,
1538 struct page *page, void *fsdata)
1540 struct inode *inode = mapping->host;
1542 if (pos + copied > inode->i_size)
1543 i_size_write(inode, pos + copied);
1546 set_page_dirty(page);
1547 page_cache_release(page);
1552 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1554 struct inode *inode = filp->f_path.dentry->d_inode;
1555 struct address_space *mapping = inode->i_mapping;
1556 unsigned long index, offset;
1557 enum sgp_type sgp = SGP_READ;
1560 * Might this read be for a stacking filesystem? Then when reading
1561 * holes of a sparse file, we actually need to allocate those pages,
1562 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1564 if (segment_eq(get_fs(), KERNEL_DS))
1567 index = *ppos >> PAGE_CACHE_SHIFT;
1568 offset = *ppos & ~PAGE_CACHE_MASK;
1571 struct page *page = NULL;
1572 unsigned long end_index, nr, ret;
1573 loff_t i_size = i_size_read(inode);
1575 end_index = i_size >> PAGE_CACHE_SHIFT;
1576 if (index > end_index)
1578 if (index == end_index) {
1579 nr = i_size & ~PAGE_CACHE_MASK;
1584 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1586 if (desc->error == -EINVAL)
1594 * We must evaluate after, since reads (unlike writes)
1595 * are called without i_mutex protection against truncate
1597 nr = PAGE_CACHE_SIZE;
1598 i_size = i_size_read(inode);
1599 end_index = i_size >> PAGE_CACHE_SHIFT;
1600 if (index == end_index) {
1601 nr = i_size & ~PAGE_CACHE_MASK;
1604 page_cache_release(page);
1612 * If users can be writing to this page using arbitrary
1613 * virtual addresses, take care about potential aliasing
1614 * before reading the page on the kernel side.
1616 if (mapping_writably_mapped(mapping))
1617 flush_dcache_page(page);
1619 * Mark the page accessed if we read the beginning.
1622 mark_page_accessed(page);
1624 page = ZERO_PAGE(0);
1625 page_cache_get(page);
1629 * Ok, we have the page, and it's up-to-date, so
1630 * now we can copy it to user space...
1632 * The actor routine returns how many bytes were actually used..
1633 * NOTE! This may not be the same as how much of a user buffer
1634 * we filled up (we may be padding etc), so we can only update
1635 * "pos" here (the actor routine has to update the user buffer
1636 * pointers and the remaining count).
1638 ret = actor(desc, page, offset, nr);
1640 index += offset >> PAGE_CACHE_SHIFT;
1641 offset &= ~PAGE_CACHE_MASK;
1643 page_cache_release(page);
1644 if (ret != nr || !desc->count)
1650 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1651 file_accessed(filp);
1654 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1656 read_descriptor_t desc;
1658 if ((ssize_t) count < 0)
1660 if (!access_ok(VERIFY_WRITE, buf, count))
1670 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1672 return desc.written;
1676 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1678 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1680 buf->f_type = TMPFS_MAGIC;
1681 buf->f_bsize = PAGE_CACHE_SIZE;
1682 buf->f_namelen = NAME_MAX;
1683 spin_lock(&sbinfo->stat_lock);
1684 if (sbinfo->max_blocks) {
1685 buf->f_blocks = sbinfo->max_blocks;
1686 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1688 if (sbinfo->max_inodes) {
1689 buf->f_files = sbinfo->max_inodes;
1690 buf->f_ffree = sbinfo->free_inodes;
1692 /* else leave those fields 0 like simple_statfs */
1693 spin_unlock(&sbinfo->stat_lock);
1698 * File creation. Allocate an inode, and we're done..
1701 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1703 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1704 int error = -ENOSPC;
1707 error = security_inode_init_security(inode, dir, NULL, NULL,
1710 if (error != -EOPNOTSUPP) {
1715 error = shmem_acl_init(inode, dir);
1720 if (dir->i_mode & S_ISGID) {
1721 inode->i_gid = dir->i_gid;
1723 inode->i_mode |= S_ISGID;
1725 dir->i_size += BOGO_DIRENT_SIZE;
1726 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1727 d_instantiate(dentry, inode);
1728 dget(dentry); /* Extra count - pin the dentry in core */
1733 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1737 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1743 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1744 struct nameidata *nd)
1746 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1752 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1754 struct inode *inode = old_dentry->d_inode;
1758 * No ordinary (disk based) filesystem counts links as inodes;
1759 * but each new link needs a new dentry, pinning lowmem, and
1760 * tmpfs dentries cannot be pruned until they are unlinked.
1762 ret = shmem_reserve_inode(inode->i_sb);
1766 dir->i_size += BOGO_DIRENT_SIZE;
1767 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1769 atomic_inc(&inode->i_count); /* New dentry reference */
1770 dget(dentry); /* Extra pinning count for the created dentry */
1771 d_instantiate(dentry, inode);
1776 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1778 struct inode *inode = dentry->d_inode;
1780 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1781 shmem_free_inode(inode->i_sb);
1783 dir->i_size -= BOGO_DIRENT_SIZE;
1784 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1786 dput(dentry); /* Undo the count from "create" - this does all the work */
1790 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1792 if (!simple_empty(dentry))
1795 drop_nlink(dentry->d_inode);
1797 return shmem_unlink(dir, dentry);
1801 * The VFS layer already does all the dentry stuff for rename,
1802 * we just have to decrement the usage count for the target if
1803 * it exists so that the VFS layer correctly free's it when it
1806 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1808 struct inode *inode = old_dentry->d_inode;
1809 int they_are_dirs = S_ISDIR(inode->i_mode);
1811 if (!simple_empty(new_dentry))
1814 if (new_dentry->d_inode) {
1815 (void) shmem_unlink(new_dir, new_dentry);
1817 drop_nlink(old_dir);
1818 } else if (they_are_dirs) {
1819 drop_nlink(old_dir);
1823 old_dir->i_size -= BOGO_DIRENT_SIZE;
1824 new_dir->i_size += BOGO_DIRENT_SIZE;
1825 old_dir->i_ctime = old_dir->i_mtime =
1826 new_dir->i_ctime = new_dir->i_mtime =
1827 inode->i_ctime = CURRENT_TIME;
1831 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1835 struct inode *inode;
1836 struct page *page = NULL;
1838 struct shmem_inode_info *info;
1840 len = strlen(symname) + 1;
1841 if (len > PAGE_CACHE_SIZE)
1842 return -ENAMETOOLONG;
1844 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1848 error = security_inode_init_security(inode, dir, NULL, NULL,
1851 if (error != -EOPNOTSUPP) {
1858 info = SHMEM_I(inode);
1859 inode->i_size = len-1;
1860 if (len <= (char *)inode - (char *)info) {
1862 memcpy(info, symname, len);
1863 inode->i_op = &shmem_symlink_inline_operations;
1865 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1871 inode->i_op = &shmem_symlink_inode_operations;
1872 kaddr = kmap_atomic(page, KM_USER0);
1873 memcpy(kaddr, symname, len);
1874 kunmap_atomic(kaddr, KM_USER0);
1875 set_page_dirty(page);
1876 page_cache_release(page);
1878 if (dir->i_mode & S_ISGID)
1879 inode->i_gid = dir->i_gid;
1880 dir->i_size += BOGO_DIRENT_SIZE;
1881 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1882 d_instantiate(dentry, inode);
1887 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1889 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1893 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1895 struct page *page = NULL;
1896 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1897 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1903 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1905 if (!IS_ERR(nd_get_link(nd))) {
1906 struct page *page = cookie;
1908 mark_page_accessed(page);
1909 page_cache_release(page);
1913 static const struct inode_operations shmem_symlink_inline_operations = {
1914 .readlink = generic_readlink,
1915 .follow_link = shmem_follow_link_inline,
1918 static const struct inode_operations shmem_symlink_inode_operations = {
1919 .truncate = shmem_truncate,
1920 .readlink = generic_readlink,
1921 .follow_link = shmem_follow_link,
1922 .put_link = shmem_put_link,
1925 #ifdef CONFIG_TMPFS_POSIX_ACL
1927 * Superblocks without xattr inode operations will get security.* xattr
1928 * support from the VFS "for free". As soon as we have any other xattrs
1929 * like ACLs, we also need to implement the security.* handlers at
1930 * filesystem level, though.
1933 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1934 size_t list_len, const char *name,
1937 return security_inode_listsecurity(inode, list, list_len);
1940 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1941 void *buffer, size_t size)
1943 if (strcmp(name, "") == 0)
1945 return security_inode_getsecurity(inode, name, buffer, size,
1949 static int shmem_xattr_security_set(struct inode *inode, const char *name,
1950 const void *value, size_t size, int flags)
1952 if (strcmp(name, "") == 0)
1954 return security_inode_setsecurity(inode, name, value, size, flags);
1957 static struct xattr_handler shmem_xattr_security_handler = {
1958 .prefix = XATTR_SECURITY_PREFIX,
1959 .list = shmem_xattr_security_list,
1960 .get = shmem_xattr_security_get,
1961 .set = shmem_xattr_security_set,
1964 static struct xattr_handler *shmem_xattr_handlers[] = {
1965 &shmem_xattr_acl_access_handler,
1966 &shmem_xattr_acl_default_handler,
1967 &shmem_xattr_security_handler,
1972 static struct dentry *shmem_get_parent(struct dentry *child)
1974 return ERR_PTR(-ESTALE);
1977 static int shmem_match(struct inode *ino, void *vfh)
1981 inum = (inum << 32) | fh[1];
1982 return ino->i_ino == inum && fh[0] == ino->i_generation;
1985 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1986 struct fid *fid, int fh_len, int fh_type)
1988 struct inode *inode;
1989 struct dentry *dentry = NULL;
1990 u64 inum = fid->raw[2];
1991 inum = (inum << 32) | fid->raw[1];
1996 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1997 shmem_match, fid->raw);
1999 dentry = d_find_alias(inode);
2006 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2009 struct inode *inode = dentry->d_inode;
2014 if (hlist_unhashed(&inode->i_hash)) {
2015 /* Unfortunately insert_inode_hash is not idempotent,
2016 * so as we hash inodes here rather than at creation
2017 * time, we need a lock to ensure we only try
2020 static DEFINE_SPINLOCK(lock);
2022 if (hlist_unhashed(&inode->i_hash))
2023 __insert_inode_hash(inode,
2024 inode->i_ino + inode->i_generation);
2028 fh[0] = inode->i_generation;
2029 fh[1] = inode->i_ino;
2030 fh[2] = ((__u64)inode->i_ino) >> 32;
2036 static const struct export_operations shmem_export_ops = {
2037 .get_parent = shmem_get_parent,
2038 .encode_fh = shmem_encode_fh,
2039 .fh_to_dentry = shmem_fh_to_dentry,
2042 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2043 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2044 int *policy, nodemask_t *policy_nodes)
2046 char *this_char, *value, *rest;
2048 while (options != NULL) {
2049 this_char = options;
2052 * NUL-terminate this option: unfortunately,
2053 * mount options form a comma-separated list,
2054 * but mpol's nodelist may also contain commas.
2056 options = strchr(options, ',');
2057 if (options == NULL)
2060 if (!isdigit(*options)) {
2067 if ((value = strchr(this_char,'=')) != NULL) {
2071 "tmpfs: No value for mount option '%s'\n",
2076 if (!strcmp(this_char,"size")) {
2077 unsigned long long size;
2078 size = memparse(value,&rest);
2080 size <<= PAGE_SHIFT;
2081 size *= totalram_pages;
2087 *blocks = DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2088 } else if (!strcmp(this_char,"nr_blocks")) {
2089 *blocks = memparse(value,&rest);
2092 } else if (!strcmp(this_char,"nr_inodes")) {
2093 *inodes = memparse(value,&rest);
2096 } else if (!strcmp(this_char,"mode")) {
2099 *mode = simple_strtoul(value,&rest,8);
2102 } else if (!strcmp(this_char,"uid")) {
2105 *uid = simple_strtoul(value,&rest,0);
2108 } else if (!strcmp(this_char,"gid")) {
2111 *gid = simple_strtoul(value,&rest,0);
2114 } else if (!strcmp(this_char,"mpol")) {
2115 if (shmem_parse_mpol(value,policy,policy_nodes))
2118 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2126 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2132 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2134 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2135 unsigned long max_blocks = sbinfo->max_blocks;
2136 unsigned long max_inodes = sbinfo->max_inodes;
2137 int policy = sbinfo->policy;
2138 nodemask_t policy_nodes = sbinfo->policy_nodes;
2139 unsigned long blocks;
2140 unsigned long inodes;
2141 int error = -EINVAL;
2143 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2144 &max_inodes, &policy, &policy_nodes))
2147 spin_lock(&sbinfo->stat_lock);
2148 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2149 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2150 if (max_blocks < blocks)
2152 if (max_inodes < inodes)
2155 * Those tests also disallow limited->unlimited while any are in
2156 * use, so i_blocks will always be zero when max_blocks is zero;
2157 * but we must separately disallow unlimited->limited, because
2158 * in that case we have no record of how much is already in use.
2160 if (max_blocks && !sbinfo->max_blocks)
2162 if (max_inodes && !sbinfo->max_inodes)
2166 sbinfo->max_blocks = max_blocks;
2167 sbinfo->free_blocks = max_blocks - blocks;
2168 sbinfo->max_inodes = max_inodes;
2169 sbinfo->free_inodes = max_inodes - inodes;
2170 sbinfo->policy = policy;
2171 sbinfo->policy_nodes = policy_nodes;
2173 spin_unlock(&sbinfo->stat_lock);
2178 static void shmem_put_super(struct super_block *sb)
2180 kfree(sb->s_fs_info);
2181 sb->s_fs_info = NULL;
2184 static int shmem_fill_super(struct super_block *sb,
2185 void *data, int silent)
2187 struct inode *inode;
2188 struct dentry *root;
2189 int mode = S_IRWXUGO | S_ISVTX;
2190 uid_t uid = current->fsuid;
2191 gid_t gid = current->fsgid;
2193 struct shmem_sb_info *sbinfo;
2194 unsigned long blocks = 0;
2195 unsigned long inodes = 0;
2196 int policy = MPOL_DEFAULT;
2197 nodemask_t policy_nodes = node_states[N_HIGH_MEMORY];
2201 * Per default we only allow half of the physical ram per
2202 * tmpfs instance, limiting inodes to one per page of lowmem;
2203 * but the internal instance is left unlimited.
2205 if (!(sb->s_flags & MS_NOUSER)) {
2206 blocks = totalram_pages / 2;
2207 inodes = totalram_pages - totalhigh_pages;
2208 if (inodes > blocks)
2210 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2211 &inodes, &policy, &policy_nodes))
2214 sb->s_export_op = &shmem_export_ops;
2216 sb->s_flags |= MS_NOUSER;
2219 /* Round up to L1_CACHE_BYTES to resist false sharing */
2220 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2221 L1_CACHE_BYTES), GFP_KERNEL);
2225 spin_lock_init(&sbinfo->stat_lock);
2226 sbinfo->max_blocks = blocks;
2227 sbinfo->free_blocks = blocks;
2228 sbinfo->max_inodes = inodes;
2229 sbinfo->free_inodes = inodes;
2230 sbinfo->policy = policy;
2231 sbinfo->policy_nodes = policy_nodes;
2233 sb->s_fs_info = sbinfo;
2234 sb->s_maxbytes = SHMEM_MAX_BYTES;
2235 sb->s_blocksize = PAGE_CACHE_SIZE;
2236 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2237 sb->s_magic = TMPFS_MAGIC;
2238 sb->s_op = &shmem_ops;
2239 sb->s_time_gran = 1;
2240 #ifdef CONFIG_TMPFS_POSIX_ACL
2241 sb->s_xattr = shmem_xattr_handlers;
2242 sb->s_flags |= MS_POSIXACL;
2245 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2250 root = d_alloc_root(inode);
2259 shmem_put_super(sb);
2263 static struct kmem_cache *shmem_inode_cachep;
2265 static struct inode *shmem_alloc_inode(struct super_block *sb)
2267 struct shmem_inode_info *p;
2268 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2271 return &p->vfs_inode;
2274 static void shmem_destroy_inode(struct inode *inode)
2276 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2277 /* only struct inode is valid if it's an inline symlink */
2278 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2280 shmem_acl_destroy_inode(inode);
2281 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2284 static void init_once(struct kmem_cache *cachep, void *foo)
2286 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2288 inode_init_once(&p->vfs_inode);
2289 #ifdef CONFIG_TMPFS_POSIX_ACL
2291 p->i_default_acl = NULL;
2295 static int init_inodecache(void)
2297 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2298 sizeof(struct shmem_inode_info),
2299 0, SLAB_PANIC, init_once);
2303 static void destroy_inodecache(void)
2305 kmem_cache_destroy(shmem_inode_cachep);
2308 static const struct address_space_operations shmem_aops = {
2309 .writepage = shmem_writepage,
2310 .set_page_dirty = __set_page_dirty_no_writeback,
2312 .readpage = shmem_readpage,
2313 .write_begin = shmem_write_begin,
2314 .write_end = shmem_write_end,
2316 .migratepage = migrate_page,
2319 static const struct file_operations shmem_file_operations = {
2322 .llseek = generic_file_llseek,
2323 .read = shmem_file_read,
2324 .write = do_sync_write,
2325 .aio_write = generic_file_aio_write,
2326 .fsync = simple_sync_file,
2327 .splice_read = generic_file_splice_read,
2328 .splice_write = generic_file_splice_write,
2332 static const struct inode_operations shmem_inode_operations = {
2333 .truncate = shmem_truncate,
2334 .setattr = shmem_notify_change,
2335 .truncate_range = shmem_truncate_range,
2336 #ifdef CONFIG_TMPFS_POSIX_ACL
2337 .setxattr = generic_setxattr,
2338 .getxattr = generic_getxattr,
2339 .listxattr = generic_listxattr,
2340 .removexattr = generic_removexattr,
2341 .permission = shmem_permission,
2346 static const struct inode_operations shmem_dir_inode_operations = {
2348 .create = shmem_create,
2349 .lookup = simple_lookup,
2351 .unlink = shmem_unlink,
2352 .symlink = shmem_symlink,
2353 .mkdir = shmem_mkdir,
2354 .rmdir = shmem_rmdir,
2355 .mknod = shmem_mknod,
2356 .rename = shmem_rename,
2358 #ifdef CONFIG_TMPFS_POSIX_ACL
2359 .setattr = shmem_notify_change,
2360 .setxattr = generic_setxattr,
2361 .getxattr = generic_getxattr,
2362 .listxattr = generic_listxattr,
2363 .removexattr = generic_removexattr,
2364 .permission = shmem_permission,
2368 static const struct inode_operations shmem_special_inode_operations = {
2369 #ifdef CONFIG_TMPFS_POSIX_ACL
2370 .setattr = shmem_notify_change,
2371 .setxattr = generic_setxattr,
2372 .getxattr = generic_getxattr,
2373 .listxattr = generic_listxattr,
2374 .removexattr = generic_removexattr,
2375 .permission = shmem_permission,
2379 static const struct super_operations shmem_ops = {
2380 .alloc_inode = shmem_alloc_inode,
2381 .destroy_inode = shmem_destroy_inode,
2383 .statfs = shmem_statfs,
2384 .remount_fs = shmem_remount_fs,
2386 .delete_inode = shmem_delete_inode,
2387 .drop_inode = generic_delete_inode,
2388 .put_super = shmem_put_super,
2391 static struct vm_operations_struct shmem_vm_ops = {
2392 .fault = shmem_fault,
2394 .set_policy = shmem_set_policy,
2395 .get_policy = shmem_get_policy,
2400 static int shmem_get_sb(struct file_system_type *fs_type,
2401 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2403 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2406 static struct file_system_type tmpfs_fs_type = {
2407 .owner = THIS_MODULE,
2409 .get_sb = shmem_get_sb,
2410 .kill_sb = kill_litter_super,
2412 static struct vfsmount *shm_mnt;
2414 static int __init init_tmpfs(void)
2418 error = bdi_init(&shmem_backing_dev_info);
2422 error = init_inodecache();
2426 error = register_filesystem(&tmpfs_fs_type);
2428 printk(KERN_ERR "Could not register tmpfs\n");
2432 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2433 tmpfs_fs_type.name, NULL);
2434 if (IS_ERR(shm_mnt)) {
2435 error = PTR_ERR(shm_mnt);
2436 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2442 unregister_filesystem(&tmpfs_fs_type);
2444 destroy_inodecache();
2446 bdi_destroy(&shmem_backing_dev_info);
2448 shm_mnt = ERR_PTR(error);
2451 module_init(init_tmpfs)
2454 * shmem_file_setup - get an unlinked file living in tmpfs
2456 * @name: name for dentry (to be seen in /proc/<pid>/maps
2457 * @size: size to be set for the file
2460 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2464 struct inode *inode;
2465 struct dentry *dentry, *root;
2468 if (IS_ERR(shm_mnt))
2469 return (void *)shm_mnt;
2471 if (size < 0 || size > SHMEM_MAX_BYTES)
2472 return ERR_PTR(-EINVAL);
2474 if (shmem_acct_size(flags, size))
2475 return ERR_PTR(-ENOMEM);
2479 this.len = strlen(name);
2480 this.hash = 0; /* will go */
2481 root = shm_mnt->mnt_root;
2482 dentry = d_alloc(root, &this);
2487 file = get_empty_filp();
2492 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2496 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2497 d_instantiate(dentry, inode);
2498 inode->i_size = size;
2499 inode->i_nlink = 0; /* It is unlinked */
2500 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2501 &shmem_file_operations);
2509 shmem_unacct_size(flags, size);
2510 return ERR_PTR(error);
2514 * shmem_zero_setup - setup a shared anonymous mapping
2516 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2518 int shmem_zero_setup(struct vm_area_struct *vma)
2521 loff_t size = vma->vm_end - vma->vm_start;
2523 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2525 return PTR_ERR(file);
2529 vma->vm_file = file;
2530 vma->vm_ops = &shmem_vm_ops;