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>
18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 * This file is released under the GPL.
24 #include <linux/init.h>
25 #include <linux/vfs.h>
26 #include <linux/mount.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
30 #include <linux/module.h>
31 #include <linux/percpu_counter.h>
32 #include <linux/swap.h>
34 static struct vfsmount *shm_mnt;
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
43 #include <linux/xattr.h>
44 #include <linux/exportfs.h>
45 #include <linux/posix_acl.h>
46 #include <linux/generic_acl.h>
47 #include <linux/mman.h>
48 #include <linux/string.h>
49 #include <linux/slab.h>
50 #include <linux/backing-dev.h>
51 #include <linux/shmem_fs.h>
52 #include <linux/writeback.h>
53 #include <linux/blkdev.h>
54 #include <linux/security.h>
55 #include <linux/swapops.h>
56 #include <linux/mempolicy.h>
57 #include <linux/namei.h>
58 #include <linux/ctype.h>
59 #include <linux/migrate.h>
60 #include <linux/highmem.h>
61 #include <linux/seq_file.h>
62 #include <linux/magic.h>
64 #include <asm/uaccess.h>
65 #include <asm/div64.h>
66 #include <asm/pgtable.h>
69 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
70 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
72 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
73 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
74 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
75 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
77 * We use / and * instead of shifts in the definitions below, so that the swap
78 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
80 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
81 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
83 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
84 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
86 #define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
87 #define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
89 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
90 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
92 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
93 #define SHMEM_PAGEIN VM_READ
94 #define SHMEM_TRUNCATE VM_WRITE
96 /* Definition to limit shmem_truncate's steps between cond_rescheds */
97 #define LATENCY_LIMIT 64
99 /* Pretend that each entry is of this size in directory's i_size */
100 #define BOGO_DIRENT_SIZE 20
102 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
104 SGP_READ, /* don't exceed i_size, don't allocate page */
105 SGP_CACHE, /* don't exceed i_size, may allocate page */
106 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
107 SGP_WRITE, /* may exceed i_size, may allocate page */
111 static unsigned long shmem_default_max_blocks(void)
113 return totalram_pages / 2;
116 static unsigned long shmem_default_max_inodes(void)
118 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
122 static int shmem_getpage(struct inode *inode, unsigned long idx,
123 struct page **pagep, enum sgp_type sgp, int *type);
125 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
128 * The above definition of ENTRIES_PER_PAGE, and the use of
129 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
130 * might be reconsidered if it ever diverges from PAGE_SIZE.
132 * Mobility flags are masked out as swap vectors cannot move
134 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
135 PAGE_CACHE_SHIFT-PAGE_SHIFT);
138 static inline void shmem_dir_free(struct page *page)
140 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
143 static struct page **shmem_dir_map(struct page *page)
145 return (struct page **)kmap_atomic(page, KM_USER0);
148 static inline void shmem_dir_unmap(struct page **dir)
150 kunmap_atomic(dir, KM_USER0);
153 static swp_entry_t *shmem_swp_map(struct page *page)
155 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
158 static inline void shmem_swp_balance_unmap(void)
161 * When passing a pointer to an i_direct entry, to code which
162 * also handles indirect entries and so will shmem_swp_unmap,
163 * we must arrange for the preempt count to remain in balance.
164 * What kmap_atomic of a lowmem page does depends on config
165 * and architecture, so pretend to kmap_atomic some lowmem page.
167 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
170 static inline void shmem_swp_unmap(swp_entry_t *entry)
172 kunmap_atomic(entry, KM_USER1);
175 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
177 return sb->s_fs_info;
181 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
182 * for shared memory and for shared anonymous (/dev/zero) mappings
183 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
184 * consistent with the pre-accounting of private mappings ...
186 static inline int shmem_acct_size(unsigned long flags, loff_t size)
188 return (flags & VM_NORESERVE) ?
189 0 : security_vm_enough_memory_kern(VM_ACCT(size));
192 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
194 if (!(flags & VM_NORESERVE))
195 vm_unacct_memory(VM_ACCT(size));
199 * ... whereas tmpfs objects are accounted incrementally as
200 * pages are allocated, in order to allow huge sparse files.
201 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
202 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
204 static inline int shmem_acct_block(unsigned long flags)
206 return (flags & VM_NORESERVE) ?
207 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
210 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
212 if (flags & VM_NORESERVE)
213 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
216 static const struct super_operations shmem_ops;
217 static const struct address_space_operations shmem_aops;
218 static const struct file_operations shmem_file_operations;
219 static const struct inode_operations shmem_inode_operations;
220 static const struct inode_operations shmem_dir_inode_operations;
221 static const struct inode_operations shmem_special_inode_operations;
222 static const struct vm_operations_struct shmem_vm_ops;
224 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
225 .ra_pages = 0, /* No readahead */
226 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
227 .unplug_io_fn = default_unplug_io_fn,
230 static LIST_HEAD(shmem_swaplist);
231 static DEFINE_MUTEX(shmem_swaplist_mutex);
233 static void shmem_free_blocks(struct inode *inode, long pages)
235 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
236 if (sbinfo->max_blocks) {
237 percpu_counter_add(&sbinfo->used_blocks, -pages);
238 spin_lock(&inode->i_lock);
239 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
240 spin_unlock(&inode->i_lock);
244 static int shmem_reserve_inode(struct super_block *sb)
246 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
247 if (sbinfo->max_inodes) {
248 spin_lock(&sbinfo->stat_lock);
249 if (!sbinfo->free_inodes) {
250 spin_unlock(&sbinfo->stat_lock);
253 sbinfo->free_inodes--;
254 spin_unlock(&sbinfo->stat_lock);
259 static void shmem_free_inode(struct super_block *sb)
261 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
262 if (sbinfo->max_inodes) {
263 spin_lock(&sbinfo->stat_lock);
264 sbinfo->free_inodes++;
265 spin_unlock(&sbinfo->stat_lock);
270 * shmem_recalc_inode - recalculate the size of an inode
271 * @inode: inode to recalc
273 * We have to calculate the free blocks since the mm can drop
274 * undirtied hole pages behind our back.
276 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
277 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
279 * It has to be called with the spinlock held.
281 static void shmem_recalc_inode(struct inode *inode)
283 struct shmem_inode_info *info = SHMEM_I(inode);
286 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
288 info->alloced -= freed;
289 shmem_unacct_blocks(info->flags, freed);
290 shmem_free_blocks(inode, freed);
295 * shmem_swp_entry - find the swap vector position in the info structure
296 * @info: info structure for the inode
297 * @index: index of the page to find
298 * @page: optional page to add to the structure. Has to be preset to
301 * If there is no space allocated yet it will return NULL when
302 * page is NULL, else it will use the page for the needed block,
303 * setting it to NULL on return to indicate that it has been used.
305 * The swap vector is organized the following way:
307 * There are SHMEM_NR_DIRECT entries directly stored in the
308 * shmem_inode_info structure. So small files do not need an addional
311 * For pages with index > SHMEM_NR_DIRECT there is the pointer
312 * i_indirect which points to a page which holds in the first half
313 * doubly indirect blocks, in the second half triple indirect blocks:
315 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
316 * following layout (for SHMEM_NR_DIRECT == 16):
318 * i_indirect -> dir --> 16-19
331 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
333 unsigned long offset;
337 if (index < SHMEM_NR_DIRECT) {
338 shmem_swp_balance_unmap();
339 return info->i_direct+index;
341 if (!info->i_indirect) {
343 info->i_indirect = *page;
346 return NULL; /* need another page */
349 index -= SHMEM_NR_DIRECT;
350 offset = index % ENTRIES_PER_PAGE;
351 index /= ENTRIES_PER_PAGE;
352 dir = shmem_dir_map(info->i_indirect);
354 if (index >= ENTRIES_PER_PAGE/2) {
355 index -= ENTRIES_PER_PAGE/2;
356 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
357 index %= ENTRIES_PER_PAGE;
364 shmem_dir_unmap(dir);
365 return NULL; /* need another page */
367 shmem_dir_unmap(dir);
368 dir = shmem_dir_map(subdir);
374 if (!page || !(subdir = *page)) {
375 shmem_dir_unmap(dir);
376 return NULL; /* need a page */
381 shmem_dir_unmap(dir);
382 return shmem_swp_map(subdir) + offset;
385 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
387 long incdec = value? 1: -1;
390 info->swapped += incdec;
391 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
392 struct page *page = kmap_atomic_to_page(entry);
393 set_page_private(page, page_private(page) + incdec);
398 * shmem_swp_alloc - get the position of the swap entry for the page.
399 * @info: info structure for the inode
400 * @index: index of the page to find
401 * @sgp: check and recheck i_size? skip allocation?
403 * If the entry does not exist, allocate it.
405 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
407 struct inode *inode = &info->vfs_inode;
408 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
409 struct page *page = NULL;
412 if (sgp != SGP_WRITE &&
413 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
414 return ERR_PTR(-EINVAL);
416 while (!(entry = shmem_swp_entry(info, index, &page))) {
418 return shmem_swp_map(ZERO_PAGE(0));
420 * Test used_blocks against 1 less max_blocks, since we have 1 data
421 * page (and perhaps indirect index pages) yet to allocate:
422 * a waste to allocate index if we cannot allocate data.
424 if (sbinfo->max_blocks) {
425 if (percpu_counter_compare(&sbinfo->used_blocks, (sbinfo->max_blocks - 1)) > 0)
426 return ERR_PTR(-ENOSPC);
427 percpu_counter_inc(&sbinfo->used_blocks);
428 spin_lock(&inode->i_lock);
429 inode->i_blocks += BLOCKS_PER_PAGE;
430 spin_unlock(&inode->i_lock);
433 spin_unlock(&info->lock);
434 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
435 spin_lock(&info->lock);
438 shmem_free_blocks(inode, 1);
439 return ERR_PTR(-ENOMEM);
441 if (sgp != SGP_WRITE &&
442 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
443 entry = ERR_PTR(-EINVAL);
446 if (info->next_index <= index)
447 info->next_index = index + 1;
450 /* another task gave its page, or truncated the file */
451 shmem_free_blocks(inode, 1);
452 shmem_dir_free(page);
454 if (info->next_index <= index && !IS_ERR(entry))
455 info->next_index = index + 1;
460 * shmem_free_swp - free some swap entries in a directory
461 * @dir: pointer to the directory
462 * @edir: pointer after last entry of the directory
463 * @punch_lock: pointer to spinlock when needed for the holepunch case
465 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
466 spinlock_t *punch_lock)
468 spinlock_t *punch_unlock = NULL;
472 for (ptr = dir; ptr < edir; ptr++) {
474 if (unlikely(punch_lock)) {
475 punch_unlock = punch_lock;
477 spin_lock(punch_unlock);
481 free_swap_and_cache(*ptr);
482 *ptr = (swp_entry_t){0};
487 spin_unlock(punch_unlock);
491 static int shmem_map_and_free_swp(struct page *subdir, int offset,
492 int limit, struct page ***dir, spinlock_t *punch_lock)
497 ptr = shmem_swp_map(subdir);
498 for (; offset < limit; offset += LATENCY_LIMIT) {
499 int size = limit - offset;
500 if (size > LATENCY_LIMIT)
501 size = LATENCY_LIMIT;
502 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
504 if (need_resched()) {
505 shmem_swp_unmap(ptr);
507 shmem_dir_unmap(*dir);
511 ptr = shmem_swp_map(subdir);
514 shmem_swp_unmap(ptr);
518 static void shmem_free_pages(struct list_head *next)
524 page = container_of(next, struct page, lru);
526 shmem_dir_free(page);
528 if (freed >= LATENCY_LIMIT) {
535 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
537 struct shmem_inode_info *info = SHMEM_I(inode);
542 unsigned long diroff;
548 LIST_HEAD(pages_to_free);
549 long nr_pages_to_free = 0;
550 long nr_swaps_freed = 0;
554 spinlock_t *needs_lock;
555 spinlock_t *punch_lock;
556 unsigned long upper_limit;
558 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
559 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
560 if (idx >= info->next_index)
563 spin_lock(&info->lock);
564 info->flags |= SHMEM_TRUNCATE;
565 if (likely(end == (loff_t) -1)) {
566 limit = info->next_index;
567 upper_limit = SHMEM_MAX_INDEX;
568 info->next_index = idx;
572 if (end + 1 >= inode->i_size) { /* we may free a little more */
573 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
575 upper_limit = SHMEM_MAX_INDEX;
577 limit = (end + 1) >> PAGE_CACHE_SHIFT;
580 needs_lock = &info->lock;
584 topdir = info->i_indirect;
585 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
586 info->i_indirect = NULL;
588 list_add(&topdir->lru, &pages_to_free);
590 spin_unlock(&info->lock);
592 if (info->swapped && idx < SHMEM_NR_DIRECT) {
593 ptr = info->i_direct;
595 if (size > SHMEM_NR_DIRECT)
596 size = SHMEM_NR_DIRECT;
597 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
601 * If there are no indirect blocks or we are punching a hole
602 * below indirect blocks, nothing to be done.
604 if (!topdir || limit <= SHMEM_NR_DIRECT)
608 * The truncation case has already dropped info->lock, and we're safe
609 * because i_size and next_index have already been lowered, preventing
610 * access beyond. But in the punch_hole case, we still need to take
611 * the lock when updating the swap directory, because there might be
612 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
613 * shmem_writepage. However, whenever we find we can remove a whole
614 * directory page (not at the misaligned start or end of the range),
615 * we first NULLify its pointer in the level above, and then have no
616 * need to take the lock when updating its contents: needs_lock and
617 * punch_lock (either pointing to info->lock or NULL) manage this.
620 upper_limit -= SHMEM_NR_DIRECT;
621 limit -= SHMEM_NR_DIRECT;
622 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
623 offset = idx % ENTRIES_PER_PAGE;
626 dir = shmem_dir_map(topdir);
627 stage = ENTRIES_PER_PAGEPAGE/2;
628 if (idx < ENTRIES_PER_PAGEPAGE/2) {
630 diroff = idx/ENTRIES_PER_PAGE;
632 dir += ENTRIES_PER_PAGE/2;
633 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
635 stage += ENTRIES_PER_PAGEPAGE;
638 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
639 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
640 if (!diroff && !offset && upper_limit >= stage) {
642 spin_lock(needs_lock);
644 spin_unlock(needs_lock);
649 list_add(&middir->lru, &pages_to_free);
651 shmem_dir_unmap(dir);
652 dir = shmem_dir_map(middir);
660 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
661 if (unlikely(idx == stage)) {
662 shmem_dir_unmap(dir);
663 dir = shmem_dir_map(topdir) +
664 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
667 idx += ENTRIES_PER_PAGEPAGE;
671 stage = idx + ENTRIES_PER_PAGEPAGE;
674 needs_lock = &info->lock;
675 if (upper_limit >= stage) {
677 spin_lock(needs_lock);
679 spin_unlock(needs_lock);
684 list_add(&middir->lru, &pages_to_free);
686 shmem_dir_unmap(dir);
688 dir = shmem_dir_map(middir);
691 punch_lock = needs_lock;
692 subdir = dir[diroff];
693 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
695 spin_lock(needs_lock);
697 spin_unlock(needs_lock);
702 list_add(&subdir->lru, &pages_to_free);
704 if (subdir && page_private(subdir) /* has swap entries */) {
706 if (size > ENTRIES_PER_PAGE)
707 size = ENTRIES_PER_PAGE;
708 freed = shmem_map_and_free_swp(subdir,
709 offset, size, &dir, punch_lock);
711 dir = shmem_dir_map(middir);
712 nr_swaps_freed += freed;
713 if (offset || punch_lock) {
714 spin_lock(&info->lock);
715 set_page_private(subdir,
716 page_private(subdir) - freed);
717 spin_unlock(&info->lock);
719 BUG_ON(page_private(subdir) != freed);
724 shmem_dir_unmap(dir);
726 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
728 * Call truncate_inode_pages again: racing shmem_unuse_inode
729 * may have swizzled a page in from swap since
730 * truncate_pagecache or generic_delete_inode did it, before we
731 * lowered next_index. Also, though shmem_getpage checks
732 * i_size before adding to cache, no recheck after: so fix the
733 * narrow window there too.
735 * Recalling truncate_inode_pages_range and unmap_mapping_range
736 * every time for punch_hole (which never got a chance to clear
737 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
738 * yet hardly ever necessary: try to optimize them out later.
740 truncate_inode_pages_range(inode->i_mapping, start, end);
742 unmap_mapping_range(inode->i_mapping, start,
746 spin_lock(&info->lock);
747 info->flags &= ~SHMEM_TRUNCATE;
748 info->swapped -= nr_swaps_freed;
749 if (nr_pages_to_free)
750 shmem_free_blocks(inode, nr_pages_to_free);
751 shmem_recalc_inode(inode);
752 spin_unlock(&info->lock);
755 * Empty swap vector directory pages to be freed?
757 if (!list_empty(&pages_to_free)) {
758 pages_to_free.prev->next = NULL;
759 shmem_free_pages(pages_to_free.next);
763 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
765 struct inode *inode = dentry->d_inode;
766 loff_t newsize = attr->ia_size;
769 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)
770 && newsize != inode->i_size) {
771 struct page *page = NULL;
773 if (newsize < inode->i_size) {
775 * If truncating down to a partial page, then
776 * if that page is already allocated, hold it
777 * in memory until the truncation is over, so
778 * truncate_partial_page cannnot miss it were
779 * it assigned to swap.
781 if (newsize & (PAGE_CACHE_SIZE-1)) {
782 (void) shmem_getpage(inode,
783 newsize >> PAGE_CACHE_SHIFT,
784 &page, SGP_READ, NULL);
789 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
790 * detect if any pages might have been added to cache
791 * after truncate_inode_pages. But we needn't bother
792 * if it's being fully truncated to zero-length: the
793 * nrpages check is efficient enough in that case.
796 struct shmem_inode_info *info = SHMEM_I(inode);
797 spin_lock(&info->lock);
798 info->flags &= ~SHMEM_PAGEIN;
799 spin_unlock(&info->lock);
803 error = simple_setsize(inode, newsize);
805 page_cache_release(page);
808 shmem_truncate_range(inode, newsize, (loff_t)-1);
811 error = inode_change_ok(inode, attr);
813 generic_setattr(inode, attr);
814 #ifdef CONFIG_TMPFS_POSIX_ACL
815 if (!error && (attr->ia_valid & ATTR_MODE))
816 error = generic_acl_chmod(inode);
821 static void shmem_delete_inode(struct inode *inode)
823 struct shmem_inode_info *info = SHMEM_I(inode);
825 if (inode->i_mapping->a_ops == &shmem_aops) {
826 truncate_inode_pages(inode->i_mapping, 0);
827 shmem_unacct_size(info->flags, inode->i_size);
829 shmem_truncate_range(inode, 0, (loff_t)-1);
830 if (!list_empty(&info->swaplist)) {
831 mutex_lock(&shmem_swaplist_mutex);
832 list_del_init(&info->swaplist);
833 mutex_unlock(&shmem_swaplist_mutex);
836 BUG_ON(inode->i_blocks);
837 shmem_free_inode(inode->i_sb);
841 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
845 for (ptr = dir; ptr < edir; ptr++) {
846 if (ptr->val == entry.val)
852 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
866 ptr = info->i_direct;
867 spin_lock(&info->lock);
868 if (!info->swapped) {
869 list_del_init(&info->swaplist);
872 limit = info->next_index;
874 if (size > SHMEM_NR_DIRECT)
875 size = SHMEM_NR_DIRECT;
876 offset = shmem_find_swp(entry, ptr, ptr+size);
879 if (!info->i_indirect)
882 dir = shmem_dir_map(info->i_indirect);
883 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
885 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
886 if (unlikely(idx == stage)) {
887 shmem_dir_unmap(dir-1);
888 if (cond_resched_lock(&info->lock)) {
889 /* check it has not been truncated */
890 if (limit > info->next_index) {
891 limit = info->next_index;
896 dir = shmem_dir_map(info->i_indirect) +
897 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
900 idx += ENTRIES_PER_PAGEPAGE;
904 stage = idx + ENTRIES_PER_PAGEPAGE;
906 shmem_dir_unmap(dir);
907 dir = shmem_dir_map(subdir);
910 if (subdir && page_private(subdir)) {
911 ptr = shmem_swp_map(subdir);
913 if (size > ENTRIES_PER_PAGE)
914 size = ENTRIES_PER_PAGE;
915 offset = shmem_find_swp(entry, ptr, ptr+size);
916 shmem_swp_unmap(ptr);
918 shmem_dir_unmap(dir);
924 shmem_dir_unmap(dir-1);
926 spin_unlock(&info->lock);
930 inode = igrab(&info->vfs_inode);
931 spin_unlock(&info->lock);
934 * Move _head_ to start search for next from here.
935 * But be careful: shmem_delete_inode checks list_empty without taking
936 * mutex, and there's an instant in list_move_tail when info->swaplist
937 * would appear empty, if it were the only one on shmem_swaplist. We
938 * could avoid doing it if inode NULL; or use this minor optimization.
940 if (shmem_swaplist.next != &info->swaplist)
941 list_move_tail(&shmem_swaplist, &info->swaplist);
942 mutex_unlock(&shmem_swaplist_mutex);
948 * Charge page using GFP_KERNEL while we can wait.
949 * Charged back to the user(not to caller) when swap account is used.
950 * add_to_page_cache() will be called with GFP_NOWAIT.
952 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
955 error = radix_tree_preload(GFP_KERNEL);
957 mem_cgroup_uncharge_cache_page(page);
962 spin_lock(&info->lock);
963 ptr = shmem_swp_entry(info, idx, NULL);
964 if (ptr && ptr->val == entry.val) {
965 error = add_to_page_cache_locked(page, inode->i_mapping,
967 /* does mem_cgroup_uncharge_cache_page on error */
968 } else /* we must compensate for our precharge above */
969 mem_cgroup_uncharge_cache_page(page);
971 if (error == -EEXIST) {
972 struct page *filepage = find_get_page(inode->i_mapping, idx);
976 * There might be a more uptodate page coming down
977 * from a stacked writepage: forget our swappage if so.
979 if (PageUptodate(filepage))
981 page_cache_release(filepage);
985 delete_from_swap_cache(page);
986 set_page_dirty(page);
987 info->flags |= SHMEM_PAGEIN;
988 shmem_swp_set(info, ptr, 0);
990 error = 1; /* not an error, but entry was found */
993 shmem_swp_unmap(ptr);
994 spin_unlock(&info->lock);
995 radix_tree_preload_end();
998 page_cache_release(page);
999 iput(inode); /* allows for NULL */
1004 * shmem_unuse() search for an eventually swapped out shmem page.
1006 int shmem_unuse(swp_entry_t entry, struct page *page)
1008 struct list_head *p, *next;
1009 struct shmem_inode_info *info;
1012 mutex_lock(&shmem_swaplist_mutex);
1013 list_for_each_safe(p, next, &shmem_swaplist) {
1014 info = list_entry(p, struct shmem_inode_info, swaplist);
1015 found = shmem_unuse_inode(info, entry, page);
1020 mutex_unlock(&shmem_swaplist_mutex);
1022 * Can some race bring us here? We've been holding page lock,
1023 * so I think not; but would rather try again later than BUG()
1026 page_cache_release(page);
1028 return (found < 0) ? found : 0;
1032 * Move the page from the page cache to the swap cache.
1034 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1036 struct shmem_inode_info *info;
1037 swp_entry_t *entry, swap;
1038 struct address_space *mapping;
1039 unsigned long index;
1040 struct inode *inode;
1042 BUG_ON(!PageLocked(page));
1043 mapping = page->mapping;
1044 index = page->index;
1045 inode = mapping->host;
1046 info = SHMEM_I(inode);
1047 if (info->flags & VM_LOCKED)
1049 if (!total_swap_pages)
1053 * shmem_backing_dev_info's capabilities prevent regular writeback or
1054 * sync from ever calling shmem_writepage; but a stacking filesystem
1055 * may use the ->writepage of its underlying filesystem, in which case
1056 * tmpfs should write out to swap only in response to memory pressure,
1057 * and not for the writeback threads or sync. However, in those cases,
1058 * we do still want to check if there's a redundant swappage to be
1061 if (wbc->for_reclaim)
1062 swap = get_swap_page();
1066 spin_lock(&info->lock);
1067 if (index >= info->next_index) {
1068 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1071 entry = shmem_swp_entry(info, index, NULL);
1074 * The more uptodate page coming down from a stacked
1075 * writepage should replace our old swappage.
1077 free_swap_and_cache(*entry);
1078 shmem_swp_set(info, entry, 0);
1080 shmem_recalc_inode(inode);
1082 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1083 remove_from_page_cache(page);
1084 shmem_swp_set(info, entry, swap.val);
1085 shmem_swp_unmap(entry);
1086 if (list_empty(&info->swaplist))
1087 inode = igrab(inode);
1090 spin_unlock(&info->lock);
1091 swap_shmem_alloc(swap);
1092 BUG_ON(page_mapped(page));
1093 page_cache_release(page); /* pagecache ref */
1094 swap_writepage(page, wbc);
1096 mutex_lock(&shmem_swaplist_mutex);
1097 /* move instead of add in case we're racing */
1098 list_move_tail(&info->swaplist, &shmem_swaplist);
1099 mutex_unlock(&shmem_swaplist_mutex);
1105 shmem_swp_unmap(entry);
1107 spin_unlock(&info->lock);
1109 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1110 * clear SWAP_HAS_CACHE flag.
1112 swapcache_free(swap, NULL);
1114 set_page_dirty(page);
1115 if (wbc->for_reclaim)
1116 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1123 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1127 if (!mpol || mpol->mode == MPOL_DEFAULT)
1128 return; /* show nothing */
1130 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1132 seq_printf(seq, ",mpol=%s", buffer);
1135 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1137 struct mempolicy *mpol = NULL;
1139 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1140 mpol = sbinfo->mpol;
1142 spin_unlock(&sbinfo->stat_lock);
1146 #endif /* CONFIG_TMPFS */
1148 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1149 struct shmem_inode_info *info, unsigned long idx)
1151 struct mempolicy mpol, *spol;
1152 struct vm_area_struct pvma;
1155 spol = mpol_cond_copy(&mpol,
1156 mpol_shared_policy_lookup(&info->policy, idx));
1158 /* Create a pseudo vma that just contains the policy */
1160 pvma.vm_pgoff = idx;
1162 pvma.vm_policy = spol;
1163 page = swapin_readahead(entry, gfp, &pvma, 0);
1167 static struct page *shmem_alloc_page(gfp_t gfp,
1168 struct shmem_inode_info *info, unsigned long idx)
1170 struct vm_area_struct pvma;
1172 /* Create a pseudo vma that just contains the policy */
1174 pvma.vm_pgoff = idx;
1176 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1179 * alloc_page_vma() will drop the shared policy reference
1181 return alloc_page_vma(gfp, &pvma, 0);
1183 #else /* !CONFIG_NUMA */
1185 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1188 #endif /* CONFIG_TMPFS */
1190 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1191 struct shmem_inode_info *info, unsigned long idx)
1193 return swapin_readahead(entry, gfp, NULL, 0);
1196 static inline struct page *shmem_alloc_page(gfp_t gfp,
1197 struct shmem_inode_info *info, unsigned long idx)
1199 return alloc_page(gfp);
1201 #endif /* CONFIG_NUMA */
1203 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1204 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1211 * shmem_getpage - either get the page from swap or allocate a new one
1213 * If we allocate a new one we do not mark it dirty. That's up to the
1214 * vm. If we swap it in we mark it dirty since we also free the swap
1215 * entry since a page cannot live in both the swap and page cache
1217 static int shmem_getpage(struct inode *inode, unsigned long idx,
1218 struct page **pagep, enum sgp_type sgp, int *type)
1220 struct address_space *mapping = inode->i_mapping;
1221 struct shmem_inode_info *info = SHMEM_I(inode);
1222 struct shmem_sb_info *sbinfo;
1223 struct page *filepage = *pagep;
1224 struct page *swappage;
1230 if (idx >= SHMEM_MAX_INDEX)
1237 * Normally, filepage is NULL on entry, and either found
1238 * uptodate immediately, or allocated and zeroed, or read
1239 * in under swappage, which is then assigned to filepage.
1240 * But shmem_readpage (required for splice) passes in a locked
1241 * filepage, which may be found not uptodate by other callers
1242 * too, and may need to be copied from the swappage read in.
1246 filepage = find_lock_page(mapping, idx);
1247 if (filepage && PageUptodate(filepage))
1250 gfp = mapping_gfp_mask(mapping);
1253 * Try to preload while we can wait, to not make a habit of
1254 * draining atomic reserves; but don't latch on to this cpu.
1256 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1259 radix_tree_preload_end();
1262 spin_lock(&info->lock);
1263 shmem_recalc_inode(inode);
1264 entry = shmem_swp_alloc(info, idx, sgp);
1265 if (IS_ERR(entry)) {
1266 spin_unlock(&info->lock);
1267 error = PTR_ERR(entry);
1273 /* Look it up and read it in.. */
1274 swappage = lookup_swap_cache(swap);
1276 shmem_swp_unmap(entry);
1277 /* here we actually do the io */
1278 if (type && !(*type & VM_FAULT_MAJOR)) {
1279 __count_vm_event(PGMAJFAULT);
1280 *type |= VM_FAULT_MAJOR;
1282 spin_unlock(&info->lock);
1283 swappage = shmem_swapin(swap, gfp, info, idx);
1285 spin_lock(&info->lock);
1286 entry = shmem_swp_alloc(info, idx, sgp);
1288 error = PTR_ERR(entry);
1290 if (entry->val == swap.val)
1292 shmem_swp_unmap(entry);
1294 spin_unlock(&info->lock);
1299 wait_on_page_locked(swappage);
1300 page_cache_release(swappage);
1304 /* We have to do this with page locked to prevent races */
1305 if (!trylock_page(swappage)) {
1306 shmem_swp_unmap(entry);
1307 spin_unlock(&info->lock);
1308 wait_on_page_locked(swappage);
1309 page_cache_release(swappage);
1312 if (PageWriteback(swappage)) {
1313 shmem_swp_unmap(entry);
1314 spin_unlock(&info->lock);
1315 wait_on_page_writeback(swappage);
1316 unlock_page(swappage);
1317 page_cache_release(swappage);
1320 if (!PageUptodate(swappage)) {
1321 shmem_swp_unmap(entry);
1322 spin_unlock(&info->lock);
1323 unlock_page(swappage);
1324 page_cache_release(swappage);
1330 shmem_swp_set(info, entry, 0);
1331 shmem_swp_unmap(entry);
1332 delete_from_swap_cache(swappage);
1333 spin_unlock(&info->lock);
1334 copy_highpage(filepage, swappage);
1335 unlock_page(swappage);
1336 page_cache_release(swappage);
1337 flush_dcache_page(filepage);
1338 SetPageUptodate(filepage);
1339 set_page_dirty(filepage);
1341 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1342 idx, GFP_NOWAIT))) {
1343 info->flags |= SHMEM_PAGEIN;
1344 shmem_swp_set(info, entry, 0);
1345 shmem_swp_unmap(entry);
1346 delete_from_swap_cache(swappage);
1347 spin_unlock(&info->lock);
1348 filepage = swappage;
1349 set_page_dirty(filepage);
1352 shmem_swp_unmap(entry);
1353 spin_unlock(&info->lock);
1354 if (error == -ENOMEM) {
1356 * reclaim from proper memory cgroup and
1357 * call memcg's OOM if needed.
1359 error = mem_cgroup_shmem_charge_fallback(
1364 unlock_page(swappage);
1365 page_cache_release(swappage);
1369 unlock_page(swappage);
1370 page_cache_release(swappage);
1373 } else if (sgp == SGP_READ && !filepage) {
1374 shmem_swp_unmap(entry);
1375 filepage = find_get_page(mapping, idx);
1377 (!PageUptodate(filepage) || !trylock_page(filepage))) {
1378 spin_unlock(&info->lock);
1379 wait_on_page_locked(filepage);
1380 page_cache_release(filepage);
1384 spin_unlock(&info->lock);
1386 shmem_swp_unmap(entry);
1387 sbinfo = SHMEM_SB(inode->i_sb);
1388 if (sbinfo->max_blocks) {
1389 if ((percpu_counter_compare(&sbinfo->used_blocks, sbinfo->max_blocks) > 0) ||
1390 shmem_acct_block(info->flags)) {
1391 spin_unlock(&info->lock);
1395 percpu_counter_inc(&sbinfo->used_blocks);
1396 spin_lock(&inode->i_lock);
1397 inode->i_blocks += BLOCKS_PER_PAGE;
1398 spin_unlock(&inode->i_lock);
1399 } else if (shmem_acct_block(info->flags)) {
1400 spin_unlock(&info->lock);
1408 spin_unlock(&info->lock);
1409 filepage = shmem_alloc_page(gfp, info, idx);
1411 shmem_unacct_blocks(info->flags, 1);
1412 shmem_free_blocks(inode, 1);
1416 SetPageSwapBacked(filepage);
1418 /* Precharge page while we can wait, compensate after */
1419 error = mem_cgroup_cache_charge(filepage, current->mm,
1422 page_cache_release(filepage);
1423 shmem_unacct_blocks(info->flags, 1);
1424 shmem_free_blocks(inode, 1);
1429 spin_lock(&info->lock);
1430 entry = shmem_swp_alloc(info, idx, sgp);
1432 error = PTR_ERR(entry);
1435 shmem_swp_unmap(entry);
1437 ret = error || swap.val;
1439 mem_cgroup_uncharge_cache_page(filepage);
1441 ret = add_to_page_cache_lru(filepage, mapping,
1444 * At add_to_page_cache_lru() failure, uncharge will
1445 * be done automatically.
1448 spin_unlock(&info->lock);
1449 page_cache_release(filepage);
1450 shmem_unacct_blocks(info->flags, 1);
1451 shmem_free_blocks(inode, 1);
1457 info->flags |= SHMEM_PAGEIN;
1461 spin_unlock(&info->lock);
1462 clear_highpage(filepage);
1463 flush_dcache_page(filepage);
1464 SetPageUptodate(filepage);
1465 if (sgp == SGP_DIRTY)
1466 set_page_dirty(filepage);
1473 if (*pagep != filepage) {
1474 unlock_page(filepage);
1475 page_cache_release(filepage);
1480 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1482 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1486 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1487 return VM_FAULT_SIGBUS;
1489 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1491 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1493 return ret | VM_FAULT_LOCKED;
1497 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1499 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1500 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1503 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1506 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1509 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1510 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1514 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1516 struct inode *inode = file->f_path.dentry->d_inode;
1517 struct shmem_inode_info *info = SHMEM_I(inode);
1518 int retval = -ENOMEM;
1520 spin_lock(&info->lock);
1521 if (lock && !(info->flags & VM_LOCKED)) {
1522 if (!user_shm_lock(inode->i_size, user))
1524 info->flags |= VM_LOCKED;
1525 mapping_set_unevictable(file->f_mapping);
1527 if (!lock && (info->flags & VM_LOCKED) && user) {
1528 user_shm_unlock(inode->i_size, user);
1529 info->flags &= ~VM_LOCKED;
1530 mapping_clear_unevictable(file->f_mapping);
1531 scan_mapping_unevictable_pages(file->f_mapping);
1536 spin_unlock(&info->lock);
1540 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1542 file_accessed(file);
1543 vma->vm_ops = &shmem_vm_ops;
1544 vma->vm_flags |= VM_CAN_NONLINEAR;
1548 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1549 int mode, dev_t dev, unsigned long flags)
1551 struct inode *inode;
1552 struct shmem_inode_info *info;
1553 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1555 if (shmem_reserve_inode(sb))
1558 inode = new_inode(sb);
1560 inode_init_owner(inode, dir, mode);
1561 inode->i_blocks = 0;
1562 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1563 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1564 inode->i_generation = get_seconds();
1565 info = SHMEM_I(inode);
1566 memset(info, 0, (char *)inode - (char *)info);
1567 spin_lock_init(&info->lock);
1568 info->flags = flags & VM_NORESERVE;
1569 INIT_LIST_HEAD(&info->swaplist);
1570 cache_no_acl(inode);
1572 switch (mode & S_IFMT) {
1574 inode->i_op = &shmem_special_inode_operations;
1575 init_special_inode(inode, mode, dev);
1578 inode->i_mapping->a_ops = &shmem_aops;
1579 inode->i_op = &shmem_inode_operations;
1580 inode->i_fop = &shmem_file_operations;
1581 mpol_shared_policy_init(&info->policy,
1582 shmem_get_sbmpol(sbinfo));
1586 /* Some things misbehave if size == 0 on a directory */
1587 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1588 inode->i_op = &shmem_dir_inode_operations;
1589 inode->i_fop = &simple_dir_operations;
1593 * Must not load anything in the rbtree,
1594 * mpol_free_shared_policy will not be called.
1596 mpol_shared_policy_init(&info->policy, NULL);
1600 shmem_free_inode(sb);
1605 static const struct inode_operations shmem_symlink_inode_operations;
1606 static const struct inode_operations shmem_symlink_inline_operations;
1609 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1610 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1611 * below the loop driver, in the generic fashion that many filesystems support.
1613 static int shmem_readpage(struct file *file, struct page *page)
1615 struct inode *inode = page->mapping->host;
1616 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1622 shmem_write_begin(struct file *file, struct address_space *mapping,
1623 loff_t pos, unsigned len, unsigned flags,
1624 struct page **pagep, void **fsdata)
1626 struct inode *inode = mapping->host;
1627 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1629 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1633 shmem_write_end(struct file *file, struct address_space *mapping,
1634 loff_t pos, unsigned len, unsigned copied,
1635 struct page *page, void *fsdata)
1637 struct inode *inode = mapping->host;
1639 if (pos + copied > inode->i_size)
1640 i_size_write(inode, pos + copied);
1642 set_page_dirty(page);
1644 page_cache_release(page);
1649 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1651 struct inode *inode = filp->f_path.dentry->d_inode;
1652 struct address_space *mapping = inode->i_mapping;
1653 unsigned long index, offset;
1654 enum sgp_type sgp = SGP_READ;
1657 * Might this read be for a stacking filesystem? Then when reading
1658 * holes of a sparse file, we actually need to allocate those pages,
1659 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1661 if (segment_eq(get_fs(), KERNEL_DS))
1664 index = *ppos >> PAGE_CACHE_SHIFT;
1665 offset = *ppos & ~PAGE_CACHE_MASK;
1668 struct page *page = NULL;
1669 unsigned long end_index, nr, ret;
1670 loff_t i_size = i_size_read(inode);
1672 end_index = i_size >> PAGE_CACHE_SHIFT;
1673 if (index > end_index)
1675 if (index == end_index) {
1676 nr = i_size & ~PAGE_CACHE_MASK;
1681 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1683 if (desc->error == -EINVAL)
1691 * We must evaluate after, since reads (unlike writes)
1692 * are called without i_mutex protection against truncate
1694 nr = PAGE_CACHE_SIZE;
1695 i_size = i_size_read(inode);
1696 end_index = i_size >> PAGE_CACHE_SHIFT;
1697 if (index == end_index) {
1698 nr = i_size & ~PAGE_CACHE_MASK;
1701 page_cache_release(page);
1709 * If users can be writing to this page using arbitrary
1710 * virtual addresses, take care about potential aliasing
1711 * before reading the page on the kernel side.
1713 if (mapping_writably_mapped(mapping))
1714 flush_dcache_page(page);
1716 * Mark the page accessed if we read the beginning.
1719 mark_page_accessed(page);
1721 page = ZERO_PAGE(0);
1722 page_cache_get(page);
1726 * Ok, we have the page, and it's up-to-date, so
1727 * now we can copy it to user space...
1729 * The actor routine returns how many bytes were actually used..
1730 * NOTE! This may not be the same as how much of a user buffer
1731 * we filled up (we may be padding etc), so we can only update
1732 * "pos" here (the actor routine has to update the user buffer
1733 * pointers and the remaining count).
1735 ret = actor(desc, page, offset, nr);
1737 index += offset >> PAGE_CACHE_SHIFT;
1738 offset &= ~PAGE_CACHE_MASK;
1740 page_cache_release(page);
1741 if (ret != nr || !desc->count)
1747 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1748 file_accessed(filp);
1751 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1752 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1754 struct file *filp = iocb->ki_filp;
1758 loff_t *ppos = &iocb->ki_pos;
1760 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1764 for (seg = 0; seg < nr_segs; seg++) {
1765 read_descriptor_t desc;
1768 desc.arg.buf = iov[seg].iov_base;
1769 desc.count = iov[seg].iov_len;
1770 if (desc.count == 0)
1773 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1774 retval += desc.written;
1776 retval = retval ?: desc.error;
1785 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1787 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1789 buf->f_type = TMPFS_MAGIC;
1790 buf->f_bsize = PAGE_CACHE_SIZE;
1791 buf->f_namelen = NAME_MAX;
1792 if (sbinfo->max_blocks) {
1793 buf->f_blocks = sbinfo->max_blocks;
1794 buf->f_bavail = buf->f_bfree =
1795 sbinfo->max_blocks - percpu_counter_sum(&sbinfo->used_blocks);
1797 if (sbinfo->max_inodes) {
1798 buf->f_files = sbinfo->max_inodes;
1799 buf->f_ffree = sbinfo->free_inodes;
1801 /* else leave those fields 0 like simple_statfs */
1806 * File creation. Allocate an inode, and we're done..
1809 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1811 struct inode *inode;
1812 int error = -ENOSPC;
1814 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1816 error = security_inode_init_security(inode, dir, NULL, NULL,
1819 if (error != -EOPNOTSUPP) {
1824 #ifdef CONFIG_TMPFS_POSIX_ACL
1825 error = generic_acl_init(inode, dir);
1833 dir->i_size += BOGO_DIRENT_SIZE;
1834 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1835 d_instantiate(dentry, inode);
1836 dget(dentry); /* Extra count - pin the dentry in core */
1841 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1845 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1851 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1852 struct nameidata *nd)
1854 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1860 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1862 struct inode *inode = old_dentry->d_inode;
1866 * No ordinary (disk based) filesystem counts links as inodes;
1867 * but each new link needs a new dentry, pinning lowmem, and
1868 * tmpfs dentries cannot be pruned until they are unlinked.
1870 ret = shmem_reserve_inode(inode->i_sb);
1874 dir->i_size += BOGO_DIRENT_SIZE;
1875 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1877 atomic_inc(&inode->i_count); /* New dentry reference */
1878 dget(dentry); /* Extra pinning count for the created dentry */
1879 d_instantiate(dentry, inode);
1884 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1886 struct inode *inode = dentry->d_inode;
1888 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1889 shmem_free_inode(inode->i_sb);
1891 dir->i_size -= BOGO_DIRENT_SIZE;
1892 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1894 dput(dentry); /* Undo the count from "create" - this does all the work */
1898 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1900 if (!simple_empty(dentry))
1903 drop_nlink(dentry->d_inode);
1905 return shmem_unlink(dir, dentry);
1909 * The VFS layer already does all the dentry stuff for rename,
1910 * we just have to decrement the usage count for the target if
1911 * it exists so that the VFS layer correctly free's it when it
1914 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1916 struct inode *inode = old_dentry->d_inode;
1917 int they_are_dirs = S_ISDIR(inode->i_mode);
1919 if (!simple_empty(new_dentry))
1922 if (new_dentry->d_inode) {
1923 (void) shmem_unlink(new_dir, new_dentry);
1925 drop_nlink(old_dir);
1926 } else if (they_are_dirs) {
1927 drop_nlink(old_dir);
1931 old_dir->i_size -= BOGO_DIRENT_SIZE;
1932 new_dir->i_size += BOGO_DIRENT_SIZE;
1933 old_dir->i_ctime = old_dir->i_mtime =
1934 new_dir->i_ctime = new_dir->i_mtime =
1935 inode->i_ctime = CURRENT_TIME;
1939 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1943 struct inode *inode;
1944 struct page *page = NULL;
1946 struct shmem_inode_info *info;
1948 len = strlen(symname) + 1;
1949 if (len > PAGE_CACHE_SIZE)
1950 return -ENAMETOOLONG;
1952 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1956 error = security_inode_init_security(inode, dir, NULL, NULL,
1959 if (error != -EOPNOTSUPP) {
1966 info = SHMEM_I(inode);
1967 inode->i_size = len-1;
1968 if (len <= (char *)inode - (char *)info) {
1970 memcpy(info, symname, len);
1971 inode->i_op = &shmem_symlink_inline_operations;
1973 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1978 inode->i_mapping->a_ops = &shmem_aops;
1979 inode->i_op = &shmem_symlink_inode_operations;
1980 kaddr = kmap_atomic(page, KM_USER0);
1981 memcpy(kaddr, symname, len);
1982 kunmap_atomic(kaddr, KM_USER0);
1983 set_page_dirty(page);
1985 page_cache_release(page);
1987 dir->i_size += BOGO_DIRENT_SIZE;
1988 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1989 d_instantiate(dentry, inode);
1994 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1996 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
2000 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2002 struct page *page = NULL;
2003 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2004 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2010 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2012 if (!IS_ERR(nd_get_link(nd))) {
2013 struct page *page = cookie;
2015 mark_page_accessed(page);
2016 page_cache_release(page);
2020 static const struct inode_operations shmem_symlink_inline_operations = {
2021 .readlink = generic_readlink,
2022 .follow_link = shmem_follow_link_inline,
2025 static const struct inode_operations shmem_symlink_inode_operations = {
2026 .readlink = generic_readlink,
2027 .follow_link = shmem_follow_link,
2028 .put_link = shmem_put_link,
2031 #ifdef CONFIG_TMPFS_POSIX_ACL
2033 * Superblocks without xattr inode operations will get security.* xattr
2034 * support from the VFS "for free". As soon as we have any other xattrs
2035 * like ACLs, we also need to implement the security.* handlers at
2036 * filesystem level, though.
2039 static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
2040 size_t list_len, const char *name,
2041 size_t name_len, int handler_flags)
2043 return security_inode_listsecurity(dentry->d_inode, list, list_len);
2046 static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2047 void *buffer, size_t size, int handler_flags)
2049 if (strcmp(name, "") == 0)
2051 return xattr_getsecurity(dentry->d_inode, name, buffer, size);
2054 static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2055 const void *value, size_t size, int flags, int handler_flags)
2057 if (strcmp(name, "") == 0)
2059 return security_inode_setsecurity(dentry->d_inode, name, value,
2063 static const struct xattr_handler shmem_xattr_security_handler = {
2064 .prefix = XATTR_SECURITY_PREFIX,
2065 .list = shmem_xattr_security_list,
2066 .get = shmem_xattr_security_get,
2067 .set = shmem_xattr_security_set,
2070 static const struct xattr_handler *shmem_xattr_handlers[] = {
2071 &generic_acl_access_handler,
2072 &generic_acl_default_handler,
2073 &shmem_xattr_security_handler,
2078 static struct dentry *shmem_get_parent(struct dentry *child)
2080 return ERR_PTR(-ESTALE);
2083 static int shmem_match(struct inode *ino, void *vfh)
2087 inum = (inum << 32) | fh[1];
2088 return ino->i_ino == inum && fh[0] == ino->i_generation;
2091 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2092 struct fid *fid, int fh_len, int fh_type)
2094 struct inode *inode;
2095 struct dentry *dentry = NULL;
2096 u64 inum = fid->raw[2];
2097 inum = (inum << 32) | fid->raw[1];
2102 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2103 shmem_match, fid->raw);
2105 dentry = d_find_alias(inode);
2112 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2115 struct inode *inode = dentry->d_inode;
2120 if (hlist_unhashed(&inode->i_hash)) {
2121 /* Unfortunately insert_inode_hash is not idempotent,
2122 * so as we hash inodes here rather than at creation
2123 * time, we need a lock to ensure we only try
2126 static DEFINE_SPINLOCK(lock);
2128 if (hlist_unhashed(&inode->i_hash))
2129 __insert_inode_hash(inode,
2130 inode->i_ino + inode->i_generation);
2134 fh[0] = inode->i_generation;
2135 fh[1] = inode->i_ino;
2136 fh[2] = ((__u64)inode->i_ino) >> 32;
2142 static const struct export_operations shmem_export_ops = {
2143 .get_parent = shmem_get_parent,
2144 .encode_fh = shmem_encode_fh,
2145 .fh_to_dentry = shmem_fh_to_dentry,
2148 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2151 char *this_char, *value, *rest;
2153 while (options != NULL) {
2154 this_char = options;
2157 * NUL-terminate this option: unfortunately,
2158 * mount options form a comma-separated list,
2159 * but mpol's nodelist may also contain commas.
2161 options = strchr(options, ',');
2162 if (options == NULL)
2165 if (!isdigit(*options)) {
2172 if ((value = strchr(this_char,'=')) != NULL) {
2176 "tmpfs: No value for mount option '%s'\n",
2181 if (!strcmp(this_char,"size")) {
2182 unsigned long long size;
2183 size = memparse(value,&rest);
2185 size <<= PAGE_SHIFT;
2186 size *= totalram_pages;
2192 sbinfo->max_blocks =
2193 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2194 } else if (!strcmp(this_char,"nr_blocks")) {
2195 sbinfo->max_blocks = memparse(value, &rest);
2198 } else if (!strcmp(this_char,"nr_inodes")) {
2199 sbinfo->max_inodes = memparse(value, &rest);
2202 } else if (!strcmp(this_char,"mode")) {
2205 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2208 } else if (!strcmp(this_char,"uid")) {
2211 sbinfo->uid = simple_strtoul(value, &rest, 0);
2214 } else if (!strcmp(this_char,"gid")) {
2217 sbinfo->gid = simple_strtoul(value, &rest, 0);
2220 } else if (!strcmp(this_char,"mpol")) {
2221 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2224 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2232 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2238 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2240 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2241 struct shmem_sb_info config = *sbinfo;
2242 unsigned long inodes;
2243 int error = -EINVAL;
2245 if (shmem_parse_options(data, &config, true))
2248 spin_lock(&sbinfo->stat_lock);
2249 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2250 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2252 if (config.max_inodes < inodes)
2255 * Those tests also disallow limited->unlimited while any are in
2256 * use, so i_blocks will always be zero when max_blocks is zero;
2257 * but we must separately disallow unlimited->limited, because
2258 * in that case we have no record of how much is already in use.
2260 if (config.max_blocks && !sbinfo->max_blocks)
2262 if (config.max_inodes && !sbinfo->max_inodes)
2266 sbinfo->max_blocks = config.max_blocks;
2267 sbinfo->max_inodes = config.max_inodes;
2268 sbinfo->free_inodes = config.max_inodes - inodes;
2270 mpol_put(sbinfo->mpol);
2271 sbinfo->mpol = config.mpol; /* transfers initial ref */
2273 spin_unlock(&sbinfo->stat_lock);
2277 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2279 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2281 if (sbinfo->max_blocks != shmem_default_max_blocks())
2282 seq_printf(seq, ",size=%luk",
2283 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2284 if (sbinfo->max_inodes != shmem_default_max_inodes())
2285 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2286 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2287 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2288 if (sbinfo->uid != 0)
2289 seq_printf(seq, ",uid=%u", sbinfo->uid);
2290 if (sbinfo->gid != 0)
2291 seq_printf(seq, ",gid=%u", sbinfo->gid);
2292 shmem_show_mpol(seq, sbinfo->mpol);
2295 #endif /* CONFIG_TMPFS */
2297 static void shmem_put_super(struct super_block *sb)
2299 kfree(sb->s_fs_info);
2300 sb->s_fs_info = NULL;
2303 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2305 struct inode *inode;
2306 struct dentry *root;
2307 struct shmem_sb_info *sbinfo;
2310 /* Round up to L1_CACHE_BYTES to resist false sharing */
2311 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2312 L1_CACHE_BYTES), GFP_KERNEL);
2316 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2317 sbinfo->uid = current_fsuid();
2318 sbinfo->gid = current_fsgid();
2319 sb->s_fs_info = sbinfo;
2323 * Per default we only allow half of the physical ram per
2324 * tmpfs instance, limiting inodes to one per page of lowmem;
2325 * but the internal instance is left unlimited.
2327 if (!(sb->s_flags & MS_NOUSER)) {
2328 sbinfo->max_blocks = shmem_default_max_blocks();
2329 sbinfo->max_inodes = shmem_default_max_inodes();
2330 if (shmem_parse_options(data, sbinfo, false)) {
2335 sb->s_export_op = &shmem_export_ops;
2337 sb->s_flags |= MS_NOUSER;
2340 spin_lock_init(&sbinfo->stat_lock);
2341 percpu_counter_init(&sbinfo->used_blocks, 0);
2342 sbinfo->free_inodes = sbinfo->max_inodes;
2344 sb->s_maxbytes = SHMEM_MAX_BYTES;
2345 sb->s_blocksize = PAGE_CACHE_SIZE;
2346 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2347 sb->s_magic = TMPFS_MAGIC;
2348 sb->s_op = &shmem_ops;
2349 sb->s_time_gran = 1;
2350 #ifdef CONFIG_TMPFS_POSIX_ACL
2351 sb->s_xattr = shmem_xattr_handlers;
2352 sb->s_flags |= MS_POSIXACL;
2355 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2358 inode->i_uid = sbinfo->uid;
2359 inode->i_gid = sbinfo->gid;
2360 root = d_alloc_root(inode);
2369 shmem_put_super(sb);
2373 static struct kmem_cache *shmem_inode_cachep;
2375 static struct inode *shmem_alloc_inode(struct super_block *sb)
2377 struct shmem_inode_info *p;
2378 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2381 return &p->vfs_inode;
2384 static void shmem_destroy_inode(struct inode *inode)
2386 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2387 /* only struct inode is valid if it's an inline symlink */
2388 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2390 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2393 static void init_once(void *foo)
2395 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2397 inode_init_once(&p->vfs_inode);
2400 static int init_inodecache(void)
2402 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2403 sizeof(struct shmem_inode_info),
2404 0, SLAB_PANIC, init_once);
2408 static void destroy_inodecache(void)
2410 kmem_cache_destroy(shmem_inode_cachep);
2413 static const struct address_space_operations shmem_aops = {
2414 .writepage = shmem_writepage,
2415 .set_page_dirty = __set_page_dirty_no_writeback,
2417 .readpage = shmem_readpage,
2418 .write_begin = shmem_write_begin,
2419 .write_end = shmem_write_end,
2421 .migratepage = migrate_page,
2422 .error_remove_page = generic_error_remove_page,
2425 static const struct file_operations shmem_file_operations = {
2428 .llseek = generic_file_llseek,
2429 .read = do_sync_read,
2430 .write = do_sync_write,
2431 .aio_read = shmem_file_aio_read,
2432 .aio_write = generic_file_aio_write,
2433 .fsync = noop_fsync,
2434 .splice_read = generic_file_splice_read,
2435 .splice_write = generic_file_splice_write,
2439 static const struct inode_operations shmem_inode_operations = {
2440 .setattr = shmem_notify_change,
2441 .truncate_range = shmem_truncate_range,
2442 #ifdef CONFIG_TMPFS_POSIX_ACL
2443 .setxattr = generic_setxattr,
2444 .getxattr = generic_getxattr,
2445 .listxattr = generic_listxattr,
2446 .removexattr = generic_removexattr,
2447 .check_acl = generic_check_acl,
2452 static const struct inode_operations shmem_dir_inode_operations = {
2454 .create = shmem_create,
2455 .lookup = simple_lookup,
2457 .unlink = shmem_unlink,
2458 .symlink = shmem_symlink,
2459 .mkdir = shmem_mkdir,
2460 .rmdir = shmem_rmdir,
2461 .mknod = shmem_mknod,
2462 .rename = shmem_rename,
2464 #ifdef CONFIG_TMPFS_POSIX_ACL
2465 .setattr = shmem_notify_change,
2466 .setxattr = generic_setxattr,
2467 .getxattr = generic_getxattr,
2468 .listxattr = generic_listxattr,
2469 .removexattr = generic_removexattr,
2470 .check_acl = generic_check_acl,
2474 static const struct inode_operations shmem_special_inode_operations = {
2475 #ifdef CONFIG_TMPFS_POSIX_ACL
2476 .setattr = shmem_notify_change,
2477 .setxattr = generic_setxattr,
2478 .getxattr = generic_getxattr,
2479 .listxattr = generic_listxattr,
2480 .removexattr = generic_removexattr,
2481 .check_acl = generic_check_acl,
2485 static const struct super_operations shmem_ops = {
2486 .alloc_inode = shmem_alloc_inode,
2487 .destroy_inode = shmem_destroy_inode,
2489 .statfs = shmem_statfs,
2490 .remount_fs = shmem_remount_fs,
2491 .show_options = shmem_show_options,
2493 .delete_inode = shmem_delete_inode,
2494 .drop_inode = generic_delete_inode,
2495 .put_super = shmem_put_super,
2498 static const struct vm_operations_struct shmem_vm_ops = {
2499 .fault = shmem_fault,
2501 .set_policy = shmem_set_policy,
2502 .get_policy = shmem_get_policy,
2507 static int shmem_get_sb(struct file_system_type *fs_type,
2508 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2510 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2513 static struct file_system_type tmpfs_fs_type = {
2514 .owner = THIS_MODULE,
2516 .get_sb = shmem_get_sb,
2517 .kill_sb = kill_litter_super,
2520 int __init init_tmpfs(void)
2524 error = bdi_init(&shmem_backing_dev_info);
2528 error = init_inodecache();
2532 error = register_filesystem(&tmpfs_fs_type);
2534 printk(KERN_ERR "Could not register tmpfs\n");
2538 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2539 tmpfs_fs_type.name, NULL);
2540 if (IS_ERR(shm_mnt)) {
2541 error = PTR_ERR(shm_mnt);
2542 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2548 unregister_filesystem(&tmpfs_fs_type);
2550 destroy_inodecache();
2552 bdi_destroy(&shmem_backing_dev_info);
2554 shm_mnt = ERR_PTR(error);
2558 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2560 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2561 * @inode: the inode to be searched
2562 * @pgoff: the offset to be searched
2563 * @pagep: the pointer for the found page to be stored
2564 * @ent: the pointer for the found swap entry to be stored
2566 * If a page is found, refcount of it is incremented. Callers should handle
2569 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2570 struct page **pagep, swp_entry_t *ent)
2572 swp_entry_t entry = { .val = 0 }, *ptr;
2573 struct page *page = NULL;
2574 struct shmem_inode_info *info = SHMEM_I(inode);
2576 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2579 spin_lock(&info->lock);
2580 ptr = shmem_swp_entry(info, pgoff, NULL);
2582 if (ptr && ptr->val) {
2583 entry.val = ptr->val;
2584 page = find_get_page(&swapper_space, entry.val);
2587 page = find_get_page(inode->i_mapping, pgoff);
2589 shmem_swp_unmap(ptr);
2590 spin_unlock(&info->lock);
2597 #else /* !CONFIG_SHMEM */
2600 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2602 * This is intended for small system where the benefits of the full
2603 * shmem code (swap-backed and resource-limited) are outweighed by
2604 * their complexity. On systems without swap this code should be
2605 * effectively equivalent, but much lighter weight.
2608 #include <linux/ramfs.h>
2610 static struct file_system_type tmpfs_fs_type = {
2612 .get_sb = ramfs_get_sb,
2613 .kill_sb = kill_litter_super,
2616 int __init init_tmpfs(void)
2618 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2620 shm_mnt = kern_mount(&tmpfs_fs_type);
2621 BUG_ON(IS_ERR(shm_mnt));
2626 int shmem_unuse(swp_entry_t entry, struct page *page)
2631 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2636 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2638 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2639 * @inode: the inode to be searched
2640 * @pgoff: the offset to be searched
2641 * @pagep: the pointer for the found page to be stored
2642 * @ent: the pointer for the found swap entry to be stored
2644 * If a page is found, refcount of it is incremented. Callers should handle
2647 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2648 struct page **pagep, swp_entry_t *ent)
2650 struct page *page = NULL;
2652 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2654 page = find_get_page(inode->i_mapping, pgoff);
2657 *ent = (swp_entry_t){ .val = 0 };
2661 #define shmem_vm_ops generic_file_vm_ops
2662 #define shmem_file_operations ramfs_file_operations
2663 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2664 #define shmem_acct_size(flags, size) 0
2665 #define shmem_unacct_size(flags, size) do {} while (0)
2666 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
2668 #endif /* CONFIG_SHMEM */
2673 * shmem_file_setup - get an unlinked file living in tmpfs
2674 * @name: name for dentry (to be seen in /proc/<pid>/maps
2675 * @size: size to be set for the file
2676 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2678 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2682 struct inode *inode;
2684 struct dentry *root;
2687 if (IS_ERR(shm_mnt))
2688 return (void *)shm_mnt;
2690 if (size < 0 || size > SHMEM_MAX_BYTES)
2691 return ERR_PTR(-EINVAL);
2693 if (shmem_acct_size(flags, size))
2694 return ERR_PTR(-ENOMEM);
2698 this.len = strlen(name);
2699 this.hash = 0; /* will go */
2700 root = shm_mnt->mnt_root;
2701 path.dentry = d_alloc(root, &this);
2704 path.mnt = mntget(shm_mnt);
2707 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2711 d_instantiate(path.dentry, inode);
2712 inode->i_size = size;
2713 inode->i_nlink = 0; /* It is unlinked */
2715 error = ramfs_nommu_expand_for_mapping(inode, size);
2721 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2722 &shmem_file_operations);
2731 shmem_unacct_size(flags, size);
2732 return ERR_PTR(error);
2734 EXPORT_SYMBOL_GPL(shmem_file_setup);
2737 * shmem_zero_setup - setup a shared anonymous mapping
2738 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2740 int shmem_zero_setup(struct vm_area_struct *vma)
2743 loff_t size = vma->vm_end - vma->vm_start;
2745 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2747 return PTR_ERR(file);
2751 vma->vm_file = file;
2752 vma->vm_ops = &shmem_vm_ops;