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
103 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
104 char *name; /* xattr name */
109 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
111 SGP_READ, /* don't exceed i_size, don't allocate page */
112 SGP_CACHE, /* don't exceed i_size, may allocate page */
113 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
114 SGP_WRITE, /* may exceed i_size, may allocate page */
118 static unsigned long shmem_default_max_blocks(void)
120 return totalram_pages / 2;
123 static unsigned long shmem_default_max_inodes(void)
125 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
129 static int shmem_getpage(struct inode *inode, unsigned long idx,
130 struct page **pagep, enum sgp_type sgp, int *type);
132 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
135 * The above definition of ENTRIES_PER_PAGE, and the use of
136 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
137 * might be reconsidered if it ever diverges from PAGE_SIZE.
139 * Mobility flags are masked out as swap vectors cannot move
141 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
142 PAGE_CACHE_SHIFT-PAGE_SHIFT);
145 static inline void shmem_dir_free(struct page *page)
147 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
150 static struct page **shmem_dir_map(struct page *page)
152 return (struct page **)kmap_atomic(page, KM_USER0);
155 static inline void shmem_dir_unmap(struct page **dir)
157 kunmap_atomic(dir, KM_USER0);
160 static swp_entry_t *shmem_swp_map(struct page *page)
162 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
165 static inline void shmem_swp_balance_unmap(void)
168 * When passing a pointer to an i_direct entry, to code which
169 * also handles indirect entries and so will shmem_swp_unmap,
170 * we must arrange for the preempt count to remain in balance.
171 * What kmap_atomic of a lowmem page does depends on config
172 * and architecture, so pretend to kmap_atomic some lowmem page.
174 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
177 static inline void shmem_swp_unmap(swp_entry_t *entry)
179 kunmap_atomic(entry, KM_USER1);
182 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
184 return sb->s_fs_info;
188 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
189 * for shared memory and for shared anonymous (/dev/zero) mappings
190 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
191 * consistent with the pre-accounting of private mappings ...
193 static inline int shmem_acct_size(unsigned long flags, loff_t size)
195 return (flags & VM_NORESERVE) ?
196 0 : security_vm_enough_memory_kern(VM_ACCT(size));
199 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
201 if (!(flags & VM_NORESERVE))
202 vm_unacct_memory(VM_ACCT(size));
206 * ... whereas tmpfs objects are accounted incrementally as
207 * pages are allocated, in order to allow huge sparse files.
208 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
209 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
211 static inline int shmem_acct_block(unsigned long flags)
213 return (flags & VM_NORESERVE) ?
214 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
217 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
219 if (flags & VM_NORESERVE)
220 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
223 static const struct super_operations shmem_ops;
224 static const struct address_space_operations shmem_aops;
225 static const struct file_operations shmem_file_operations;
226 static const struct inode_operations shmem_inode_operations;
227 static const struct inode_operations shmem_dir_inode_operations;
228 static const struct inode_operations shmem_special_inode_operations;
229 static const struct vm_operations_struct shmem_vm_ops;
231 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
232 .ra_pages = 0, /* No readahead */
233 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
236 static LIST_HEAD(shmem_swaplist);
237 static DEFINE_MUTEX(shmem_swaplist_mutex);
239 static void shmem_free_blocks(struct inode *inode, long pages)
241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
242 if (sbinfo->max_blocks) {
243 percpu_counter_add(&sbinfo->used_blocks, -pages);
244 spin_lock(&inode->i_lock);
245 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
246 spin_unlock(&inode->i_lock);
250 static int shmem_reserve_inode(struct super_block *sb)
252 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
253 if (sbinfo->max_inodes) {
254 spin_lock(&sbinfo->stat_lock);
255 if (!sbinfo->free_inodes) {
256 spin_unlock(&sbinfo->stat_lock);
259 sbinfo->free_inodes--;
260 spin_unlock(&sbinfo->stat_lock);
265 static void shmem_free_inode(struct super_block *sb)
267 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
268 if (sbinfo->max_inodes) {
269 spin_lock(&sbinfo->stat_lock);
270 sbinfo->free_inodes++;
271 spin_unlock(&sbinfo->stat_lock);
276 * shmem_recalc_inode - recalculate the size of an inode
277 * @inode: inode to recalc
279 * We have to calculate the free blocks since the mm can drop
280 * undirtied hole pages behind our back.
282 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
283 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
285 * It has to be called with the spinlock held.
287 static void shmem_recalc_inode(struct inode *inode)
289 struct shmem_inode_info *info = SHMEM_I(inode);
292 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
294 info->alloced -= freed;
295 shmem_unacct_blocks(info->flags, freed);
296 shmem_free_blocks(inode, freed);
301 * shmem_swp_entry - find the swap vector position in the info structure
302 * @info: info structure for the inode
303 * @index: index of the page to find
304 * @page: optional page to add to the structure. Has to be preset to
307 * If there is no space allocated yet it will return NULL when
308 * page is NULL, else it will use the page for the needed block,
309 * setting it to NULL on return to indicate that it has been used.
311 * The swap vector is organized the following way:
313 * There are SHMEM_NR_DIRECT entries directly stored in the
314 * shmem_inode_info structure. So small files do not need an addional
317 * For pages with index > SHMEM_NR_DIRECT there is the pointer
318 * i_indirect which points to a page which holds in the first half
319 * doubly indirect blocks, in the second half triple indirect blocks:
321 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
322 * following layout (for SHMEM_NR_DIRECT == 16):
324 * i_indirect -> dir --> 16-19
337 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
339 unsigned long offset;
343 if (index < SHMEM_NR_DIRECT) {
344 shmem_swp_balance_unmap();
345 return info->i_direct+index;
347 if (!info->i_indirect) {
349 info->i_indirect = *page;
352 return NULL; /* need another page */
355 index -= SHMEM_NR_DIRECT;
356 offset = index % ENTRIES_PER_PAGE;
357 index /= ENTRIES_PER_PAGE;
358 dir = shmem_dir_map(info->i_indirect);
360 if (index >= ENTRIES_PER_PAGE/2) {
361 index -= ENTRIES_PER_PAGE/2;
362 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
363 index %= ENTRIES_PER_PAGE;
370 shmem_dir_unmap(dir);
371 return NULL; /* need another page */
373 shmem_dir_unmap(dir);
374 dir = shmem_dir_map(subdir);
380 if (!page || !(subdir = *page)) {
381 shmem_dir_unmap(dir);
382 return NULL; /* need a page */
387 shmem_dir_unmap(dir);
388 return shmem_swp_map(subdir) + offset;
391 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
393 long incdec = value? 1: -1;
396 info->swapped += incdec;
397 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
398 struct page *page = kmap_atomic_to_page(entry);
399 set_page_private(page, page_private(page) + incdec);
404 * shmem_swp_alloc - get the position of the swap entry for the page.
405 * @info: info structure for the inode
406 * @index: index of the page to find
407 * @sgp: check and recheck i_size? skip allocation?
409 * If the entry does not exist, allocate it.
411 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
413 struct inode *inode = &info->vfs_inode;
414 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
415 struct page *page = NULL;
418 if (sgp != SGP_WRITE &&
419 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
420 return ERR_PTR(-EINVAL);
422 while (!(entry = shmem_swp_entry(info, index, &page))) {
424 return shmem_swp_map(ZERO_PAGE(0));
426 * Test used_blocks against 1 less max_blocks, since we have 1 data
427 * page (and perhaps indirect index pages) yet to allocate:
428 * a waste to allocate index if we cannot allocate data.
430 if (sbinfo->max_blocks) {
431 if (percpu_counter_compare(&sbinfo->used_blocks,
432 sbinfo->max_blocks - 1) >= 0)
433 return ERR_PTR(-ENOSPC);
434 percpu_counter_inc(&sbinfo->used_blocks);
435 spin_lock(&inode->i_lock);
436 inode->i_blocks += BLOCKS_PER_PAGE;
437 spin_unlock(&inode->i_lock);
440 spin_unlock(&info->lock);
441 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
442 spin_lock(&info->lock);
445 shmem_free_blocks(inode, 1);
446 return ERR_PTR(-ENOMEM);
448 if (sgp != SGP_WRITE &&
449 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
450 entry = ERR_PTR(-EINVAL);
453 if (info->next_index <= index)
454 info->next_index = index + 1;
457 /* another task gave its page, or truncated the file */
458 shmem_free_blocks(inode, 1);
459 shmem_dir_free(page);
461 if (info->next_index <= index && !IS_ERR(entry))
462 info->next_index = index + 1;
467 * shmem_free_swp - free some swap entries in a directory
468 * @dir: pointer to the directory
469 * @edir: pointer after last entry of the directory
470 * @punch_lock: pointer to spinlock when needed for the holepunch case
472 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
473 spinlock_t *punch_lock)
475 spinlock_t *punch_unlock = NULL;
479 for (ptr = dir; ptr < edir; ptr++) {
481 if (unlikely(punch_lock)) {
482 punch_unlock = punch_lock;
484 spin_lock(punch_unlock);
488 free_swap_and_cache(*ptr);
489 *ptr = (swp_entry_t){0};
494 spin_unlock(punch_unlock);
498 static int shmem_map_and_free_swp(struct page *subdir, int offset,
499 int limit, struct page ***dir, spinlock_t *punch_lock)
504 ptr = shmem_swp_map(subdir);
505 for (; offset < limit; offset += LATENCY_LIMIT) {
506 int size = limit - offset;
507 if (size > LATENCY_LIMIT)
508 size = LATENCY_LIMIT;
509 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
511 if (need_resched()) {
512 shmem_swp_unmap(ptr);
514 shmem_dir_unmap(*dir);
518 ptr = shmem_swp_map(subdir);
521 shmem_swp_unmap(ptr);
525 static void shmem_free_pages(struct list_head *next)
531 page = container_of(next, struct page, lru);
533 shmem_dir_free(page);
535 if (freed >= LATENCY_LIMIT) {
542 void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
544 struct shmem_inode_info *info = SHMEM_I(inode);
549 unsigned long diroff;
555 LIST_HEAD(pages_to_free);
556 long nr_pages_to_free = 0;
557 long nr_swaps_freed = 0;
561 spinlock_t *needs_lock;
562 spinlock_t *punch_lock;
563 unsigned long upper_limit;
565 truncate_inode_pages_range(inode->i_mapping, start, end);
567 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
568 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
569 if (idx >= info->next_index)
572 spin_lock(&info->lock);
573 info->flags |= SHMEM_TRUNCATE;
574 if (likely(end == (loff_t) -1)) {
575 limit = info->next_index;
576 upper_limit = SHMEM_MAX_INDEX;
577 info->next_index = idx;
581 if (end + 1 >= inode->i_size) { /* we may free a little more */
582 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
584 upper_limit = SHMEM_MAX_INDEX;
586 limit = (end + 1) >> PAGE_CACHE_SHIFT;
589 needs_lock = &info->lock;
593 topdir = info->i_indirect;
594 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
595 info->i_indirect = NULL;
597 list_add(&topdir->lru, &pages_to_free);
599 spin_unlock(&info->lock);
601 if (info->swapped && idx < SHMEM_NR_DIRECT) {
602 ptr = info->i_direct;
604 if (size > SHMEM_NR_DIRECT)
605 size = SHMEM_NR_DIRECT;
606 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
610 * If there are no indirect blocks or we are punching a hole
611 * below indirect blocks, nothing to be done.
613 if (!topdir || limit <= SHMEM_NR_DIRECT)
617 * The truncation case has already dropped info->lock, and we're safe
618 * because i_size and next_index have already been lowered, preventing
619 * access beyond. But in the punch_hole case, we still need to take
620 * the lock when updating the swap directory, because there might be
621 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
622 * shmem_writepage. However, whenever we find we can remove a whole
623 * directory page (not at the misaligned start or end of the range),
624 * we first NULLify its pointer in the level above, and then have no
625 * need to take the lock when updating its contents: needs_lock and
626 * punch_lock (either pointing to info->lock or NULL) manage this.
629 upper_limit -= SHMEM_NR_DIRECT;
630 limit -= SHMEM_NR_DIRECT;
631 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
632 offset = idx % ENTRIES_PER_PAGE;
635 dir = shmem_dir_map(topdir);
636 stage = ENTRIES_PER_PAGEPAGE/2;
637 if (idx < ENTRIES_PER_PAGEPAGE/2) {
639 diroff = idx/ENTRIES_PER_PAGE;
641 dir += ENTRIES_PER_PAGE/2;
642 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
644 stage += ENTRIES_PER_PAGEPAGE;
647 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
648 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
649 if (!diroff && !offset && upper_limit >= stage) {
651 spin_lock(needs_lock);
653 spin_unlock(needs_lock);
658 list_add(&middir->lru, &pages_to_free);
660 shmem_dir_unmap(dir);
661 dir = shmem_dir_map(middir);
669 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
670 if (unlikely(idx == stage)) {
671 shmem_dir_unmap(dir);
672 dir = shmem_dir_map(topdir) +
673 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
676 idx += ENTRIES_PER_PAGEPAGE;
680 stage = idx + ENTRIES_PER_PAGEPAGE;
683 needs_lock = &info->lock;
684 if (upper_limit >= stage) {
686 spin_lock(needs_lock);
688 spin_unlock(needs_lock);
693 list_add(&middir->lru, &pages_to_free);
695 shmem_dir_unmap(dir);
697 dir = shmem_dir_map(middir);
700 punch_lock = needs_lock;
701 subdir = dir[diroff];
702 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
704 spin_lock(needs_lock);
706 spin_unlock(needs_lock);
711 list_add(&subdir->lru, &pages_to_free);
713 if (subdir && page_private(subdir) /* has swap entries */) {
715 if (size > ENTRIES_PER_PAGE)
716 size = ENTRIES_PER_PAGE;
717 freed = shmem_map_and_free_swp(subdir,
718 offset, size, &dir, punch_lock);
720 dir = shmem_dir_map(middir);
721 nr_swaps_freed += freed;
722 if (offset || punch_lock) {
723 spin_lock(&info->lock);
724 set_page_private(subdir,
725 page_private(subdir) - freed);
726 spin_unlock(&info->lock);
728 BUG_ON(page_private(subdir) != freed);
733 shmem_dir_unmap(dir);
735 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
737 * Call truncate_inode_pages again: racing shmem_unuse_inode
738 * may have swizzled a page in from swap since
739 * truncate_pagecache or generic_delete_inode did it, before we
740 * lowered next_index. Also, though shmem_getpage checks
741 * i_size before adding to cache, no recheck after: so fix the
742 * narrow window there too.
744 truncate_inode_pages_range(inode->i_mapping, start, end);
747 spin_lock(&info->lock);
748 info->flags &= ~SHMEM_TRUNCATE;
749 info->swapped -= nr_swaps_freed;
750 if (nr_pages_to_free)
751 shmem_free_blocks(inode, nr_pages_to_free);
752 shmem_recalc_inode(inode);
753 spin_unlock(&info->lock);
756 * Empty swap vector directory pages to be freed?
758 if (!list_empty(&pages_to_free)) {
759 pages_to_free.prev->next = NULL;
760 shmem_free_pages(pages_to_free.next);
763 EXPORT_SYMBOL_GPL(shmem_truncate_range);
765 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
767 struct inode *inode = dentry->d_inode;
770 error = inode_change_ok(inode, attr);
774 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
775 loff_t oldsize = inode->i_size;
776 loff_t newsize = attr->ia_size;
777 struct page *page = NULL;
779 if (newsize < oldsize) {
781 * If truncating down to a partial page, then
782 * if that page is already allocated, hold it
783 * in memory until the truncation is over, so
784 * truncate_partial_page cannot miss it were
785 * it assigned to swap.
787 if (newsize & (PAGE_CACHE_SIZE-1)) {
788 (void) shmem_getpage(inode,
789 newsize >> PAGE_CACHE_SHIFT,
790 &page, SGP_READ, NULL);
795 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
796 * detect if any pages might have been added to cache
797 * after truncate_inode_pages. But we needn't bother
798 * if it's being fully truncated to zero-length: the
799 * nrpages check is efficient enough in that case.
802 struct shmem_inode_info *info = SHMEM_I(inode);
803 spin_lock(&info->lock);
804 info->flags &= ~SHMEM_PAGEIN;
805 spin_unlock(&info->lock);
808 if (newsize != oldsize) {
809 i_size_write(inode, newsize);
810 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
812 if (newsize < oldsize) {
813 loff_t holebegin = round_up(newsize, PAGE_SIZE);
814 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
815 shmem_truncate_range(inode, newsize, (loff_t)-1);
816 /* unmap again to remove racily COWed private pages */
817 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
820 page_cache_release(page);
823 setattr_copy(inode, attr);
824 #ifdef CONFIG_TMPFS_POSIX_ACL
825 if (attr->ia_valid & ATTR_MODE)
826 error = generic_acl_chmod(inode);
831 static void shmem_evict_inode(struct inode *inode)
833 struct shmem_inode_info *info = SHMEM_I(inode);
834 struct shmem_xattr *xattr, *nxattr;
836 if (inode->i_mapping->a_ops == &shmem_aops) {
837 shmem_unacct_size(info->flags, inode->i_size);
839 shmem_truncate_range(inode, 0, (loff_t)-1);
840 if (!list_empty(&info->swaplist)) {
841 mutex_lock(&shmem_swaplist_mutex);
842 list_del_init(&info->swaplist);
843 mutex_unlock(&shmem_swaplist_mutex);
847 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
851 BUG_ON(inode->i_blocks);
852 shmem_free_inode(inode->i_sb);
853 end_writeback(inode);
856 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
860 for (ptr = dir; ptr < edir; ptr++) {
861 if (ptr->val == entry.val)
867 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
869 struct address_space *mapping;
881 ptr = info->i_direct;
882 spin_lock(&info->lock);
883 if (!info->swapped) {
884 list_del_init(&info->swaplist);
887 limit = info->next_index;
889 if (size > SHMEM_NR_DIRECT)
890 size = SHMEM_NR_DIRECT;
891 offset = shmem_find_swp(entry, ptr, ptr+size);
893 shmem_swp_balance_unmap();
896 if (!info->i_indirect)
899 dir = shmem_dir_map(info->i_indirect);
900 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
902 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
903 if (unlikely(idx == stage)) {
904 shmem_dir_unmap(dir-1);
905 if (cond_resched_lock(&info->lock)) {
906 /* check it has not been truncated */
907 if (limit > info->next_index) {
908 limit = info->next_index;
913 dir = shmem_dir_map(info->i_indirect) +
914 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
917 idx += ENTRIES_PER_PAGEPAGE;
921 stage = idx + ENTRIES_PER_PAGEPAGE;
923 shmem_dir_unmap(dir);
924 dir = shmem_dir_map(subdir);
927 if (subdir && page_private(subdir)) {
928 ptr = shmem_swp_map(subdir);
930 if (size > ENTRIES_PER_PAGE)
931 size = ENTRIES_PER_PAGE;
932 offset = shmem_find_swp(entry, ptr, ptr+size);
933 shmem_swp_unmap(ptr);
935 shmem_dir_unmap(dir);
936 ptr = shmem_swp_map(subdir);
942 shmem_dir_unmap(dir-1);
944 spin_unlock(&info->lock);
951 * Move _head_ to start search for next from here.
952 * But be careful: shmem_evict_inode checks list_empty without taking
953 * mutex, and there's an instant in list_move_tail when info->swaplist
954 * would appear empty, if it were the only one on shmem_swaplist. We
955 * could avoid doing it if inode NULL; or use this minor optimization.
957 if (shmem_swaplist.next != &info->swaplist)
958 list_move_tail(&shmem_swaplist, &info->swaplist);
961 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
962 * but also to hold up shmem_evict_inode(): so inode cannot be freed
963 * beneath us (pagelock doesn't help until the page is in pagecache).
965 mapping = info->vfs_inode.i_mapping;
966 error = add_to_page_cache_locked(page, mapping, idx, GFP_NOWAIT);
967 /* which does mem_cgroup_uncharge_cache_page on error */
969 if (error == -EEXIST) {
970 struct page *filepage = find_get_page(mapping, idx);
974 * There might be a more uptodate page coming down
975 * from a stacked writepage: forget our swappage if so.
977 if (PageUptodate(filepage))
979 page_cache_release(filepage);
983 delete_from_swap_cache(page);
984 set_page_dirty(page);
985 info->flags |= SHMEM_PAGEIN;
986 shmem_swp_set(info, ptr, 0);
988 error = 1; /* not an error, but entry was found */
990 shmem_swp_unmap(ptr);
991 spin_unlock(&info->lock);
996 * shmem_unuse() search for an eventually swapped out shmem page.
998 int shmem_unuse(swp_entry_t entry, struct page *page)
1000 struct list_head *p, *next;
1001 struct shmem_inode_info *info;
1006 * Charge page using GFP_KERNEL while we can wait, before taking
1007 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1008 * Charged back to the user (not to caller) when swap account is used.
1009 * add_to_page_cache() will be called with GFP_NOWAIT.
1011 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
1015 * Try to preload while we can wait, to not make a habit of
1016 * draining atomic reserves; but don't latch on to this cpu,
1017 * it's okay if sometimes we get rescheduled after this.
1019 error = radix_tree_preload(GFP_KERNEL);
1022 radix_tree_preload_end();
1024 mutex_lock(&shmem_swaplist_mutex);
1025 list_for_each_safe(p, next, &shmem_swaplist) {
1026 info = list_entry(p, struct shmem_inode_info, swaplist);
1027 found = shmem_unuse_inode(info, entry, page);
1032 mutex_unlock(&shmem_swaplist_mutex);
1036 mem_cgroup_uncharge_cache_page(page);
1041 page_cache_release(page);
1046 * Move the page from the page cache to the swap cache.
1048 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1050 struct shmem_inode_info *info;
1051 swp_entry_t *entry, swap;
1052 struct address_space *mapping;
1053 unsigned long index;
1054 struct inode *inode;
1056 BUG_ON(!PageLocked(page));
1057 mapping = page->mapping;
1058 index = page->index;
1059 inode = mapping->host;
1060 info = SHMEM_I(inode);
1061 if (info->flags & VM_LOCKED)
1063 #ifdef CONFIG_ZRAM_FOR_ANDROID
1065 * Modification for compcache
1066 * shmem_writepage can be reason of kernel panic when using swap.
1067 * This modification prevent using swap by shmem.
1071 if (!total_swap_pages)
1076 * shmem_backing_dev_info's capabilities prevent regular writeback or
1077 * sync from ever calling shmem_writepage; but a stacking filesystem
1078 * may use the ->writepage of its underlying filesystem, in which case
1079 * tmpfs should write out to swap only in response to memory pressure,
1080 * and not for the writeback threads or sync. However, in those cases,
1081 * we do still want to check if there's a redundant swappage to be
1084 if (wbc->for_reclaim)
1085 swap = get_swap_page();
1090 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1091 * if it's not already there. Do it now because we cannot take
1092 * mutex while holding spinlock, and must do so before the page
1093 * is moved to swap cache, when its pagelock no longer protects
1094 * the inode from eviction. But don't unlock the mutex until
1095 * we've taken the spinlock, because shmem_unuse_inode() will
1096 * prune a !swapped inode from the swaplist under both locks.
1099 mutex_lock(&shmem_swaplist_mutex);
1100 if (list_empty(&info->swaplist))
1101 list_add_tail(&info->swaplist, &shmem_swaplist);
1104 spin_lock(&info->lock);
1106 mutex_unlock(&shmem_swaplist_mutex);
1108 if (index >= info->next_index) {
1109 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1112 entry = shmem_swp_entry(info, index, NULL);
1115 * The more uptodate page coming down from a stacked
1116 * writepage should replace our old swappage.
1118 free_swap_and_cache(*entry);
1119 shmem_swp_set(info, entry, 0);
1121 shmem_recalc_inode(inode);
1123 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1124 delete_from_page_cache(page);
1125 shmem_swp_set(info, entry, swap.val);
1126 shmem_swp_unmap(entry);
1127 swap_shmem_alloc(swap);
1128 spin_unlock(&info->lock);
1129 BUG_ON(page_mapped(page));
1130 swap_writepage(page, wbc);
1134 shmem_swp_unmap(entry);
1136 spin_unlock(&info->lock);
1138 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1139 * clear SWAP_HAS_CACHE flag.
1141 swapcache_free(swap, NULL);
1143 set_page_dirty(page);
1144 if (wbc->for_reclaim)
1145 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1152 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1156 if (!mpol || mpol->mode == MPOL_DEFAULT)
1157 return; /* show nothing */
1159 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1161 seq_printf(seq, ",mpol=%s", buffer);
1164 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1166 struct mempolicy *mpol = NULL;
1168 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1169 mpol = sbinfo->mpol;
1171 spin_unlock(&sbinfo->stat_lock);
1175 #endif /* CONFIG_TMPFS */
1177 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1178 struct shmem_inode_info *info, unsigned long idx)
1180 struct vm_area_struct pvma;
1183 /* Create a pseudo vma that just contains the policy */
1185 pvma.vm_pgoff = idx;
1187 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1189 page = swapin_readahead(entry, gfp, &pvma, 0);
1191 /* Drop reference taken by mpol_shared_policy_lookup() */
1192 mpol_cond_put(pvma.vm_policy);
1197 static struct page *shmem_alloc_page(gfp_t gfp,
1198 struct shmem_inode_info *info, unsigned long idx)
1200 struct vm_area_struct pvma;
1203 /* Create a pseudo vma that just contains the policy */
1205 pvma.vm_pgoff = idx;
1207 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1209 page = alloc_page_vma(gfp, &pvma, 0);
1211 /* Drop reference taken by mpol_shared_policy_lookup() */
1212 mpol_cond_put(pvma.vm_policy);
1216 #else /* !CONFIG_NUMA */
1218 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1221 #endif /* CONFIG_TMPFS */
1223 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1224 struct shmem_inode_info *info, unsigned long idx)
1226 return swapin_readahead(entry, gfp, NULL, 0);
1229 static inline struct page *shmem_alloc_page(gfp_t gfp,
1230 struct shmem_inode_info *info, unsigned long idx)
1232 return alloc_page(gfp);
1234 #endif /* CONFIG_NUMA */
1236 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1237 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1244 * shmem_getpage - either get the page from swap or allocate a new one
1246 * If we allocate a new one we do not mark it dirty. That's up to the
1247 * vm. If we swap it in we mark it dirty since we also free the swap
1248 * entry since a page cannot live in both the swap and page cache
1250 static int shmem_getpage(struct inode *inode, unsigned long idx,
1251 struct page **pagep, enum sgp_type sgp, int *type)
1253 struct address_space *mapping = inode->i_mapping;
1254 struct shmem_inode_info *info = SHMEM_I(inode);
1255 struct shmem_sb_info *sbinfo;
1256 struct page *filepage = *pagep;
1257 struct page *swappage;
1258 struct page *prealloc_page = NULL;
1264 if (idx >= SHMEM_MAX_INDEX)
1271 * Normally, filepage is NULL on entry, and either found
1272 * uptodate immediately, or allocated and zeroed, or read
1273 * in under swappage, which is then assigned to filepage.
1274 * But shmem_readpage (required for splice) passes in a locked
1275 * filepage, which may be found not uptodate by other callers
1276 * too, and may need to be copied from the swappage read in.
1280 filepage = find_lock_page(mapping, idx);
1281 if (filepage && PageUptodate(filepage))
1283 gfp = mapping_gfp_mask(mapping);
1286 * Try to preload while we can wait, to not make a habit of
1287 * draining atomic reserves; but don't latch on to this cpu.
1289 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1292 radix_tree_preload_end();
1293 if (sgp != SGP_READ && !prealloc_page) {
1294 /* We don't care if this fails */
1295 prealloc_page = shmem_alloc_page(gfp, info, idx);
1296 if (prealloc_page) {
1297 if (mem_cgroup_cache_charge(prealloc_page,
1298 current->mm, GFP_KERNEL)) {
1299 page_cache_release(prealloc_page);
1300 prealloc_page = NULL;
1307 spin_lock(&info->lock);
1308 shmem_recalc_inode(inode);
1309 entry = shmem_swp_alloc(info, idx, sgp);
1310 if (IS_ERR(entry)) {
1311 spin_unlock(&info->lock);
1312 error = PTR_ERR(entry);
1318 /* Look it up and read it in.. */
1319 swappage = lookup_swap_cache(swap);
1321 shmem_swp_unmap(entry);
1322 spin_unlock(&info->lock);
1323 /* here we actually do the io */
1325 *type |= VM_FAULT_MAJOR;
1326 swappage = shmem_swapin(swap, gfp, info, idx);
1328 spin_lock(&info->lock);
1329 entry = shmem_swp_alloc(info, idx, sgp);
1331 error = PTR_ERR(entry);
1333 if (entry->val == swap.val)
1335 shmem_swp_unmap(entry);
1337 spin_unlock(&info->lock);
1342 wait_on_page_locked(swappage);
1343 page_cache_release(swappage);
1347 /* We have to do this with page locked to prevent races */
1348 if (!trylock_page(swappage)) {
1349 shmem_swp_unmap(entry);
1350 spin_unlock(&info->lock);
1351 wait_on_page_locked(swappage);
1352 page_cache_release(swappage);
1355 if (PageWriteback(swappage)) {
1356 shmem_swp_unmap(entry);
1357 spin_unlock(&info->lock);
1358 wait_on_page_writeback(swappage);
1359 unlock_page(swappage);
1360 page_cache_release(swappage);
1363 if (!PageUptodate(swappage)) {
1364 shmem_swp_unmap(entry);
1365 spin_unlock(&info->lock);
1366 unlock_page(swappage);
1367 page_cache_release(swappage);
1373 shmem_swp_set(info, entry, 0);
1374 shmem_swp_unmap(entry);
1375 delete_from_swap_cache(swappage);
1376 spin_unlock(&info->lock);
1377 copy_highpage(filepage, swappage);
1378 unlock_page(swappage);
1379 page_cache_release(swappage);
1380 flush_dcache_page(filepage);
1381 SetPageUptodate(filepage);
1382 set_page_dirty(filepage);
1384 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1385 idx, GFP_NOWAIT))) {
1386 info->flags |= SHMEM_PAGEIN;
1387 shmem_swp_set(info, entry, 0);
1388 shmem_swp_unmap(entry);
1389 delete_from_swap_cache(swappage);
1390 spin_unlock(&info->lock);
1391 filepage = swappage;
1392 set_page_dirty(filepage);
1395 shmem_swp_unmap(entry);
1396 spin_unlock(&info->lock);
1397 if (error == -ENOMEM) {
1399 * reclaim from proper memory cgroup and
1400 * call memcg's OOM if needed.
1402 error = mem_cgroup_shmem_charge_fallback(
1407 unlock_page(swappage);
1408 page_cache_release(swappage);
1412 unlock_page(swappage);
1413 page_cache_release(swappage);
1416 } else if (sgp == SGP_READ && !filepage) {
1417 shmem_swp_unmap(entry);
1418 filepage = find_get_page(mapping, idx);
1420 (!PageUptodate(filepage) || !trylock_page(filepage))) {
1421 spin_unlock(&info->lock);
1422 wait_on_page_locked(filepage);
1423 page_cache_release(filepage);
1427 spin_unlock(&info->lock);
1429 shmem_swp_unmap(entry);
1430 sbinfo = SHMEM_SB(inode->i_sb);
1431 if (sbinfo->max_blocks) {
1432 if (percpu_counter_compare(&sbinfo->used_blocks,
1433 sbinfo->max_blocks) >= 0 ||
1434 shmem_acct_block(info->flags))
1436 percpu_counter_inc(&sbinfo->used_blocks);
1437 spin_lock(&inode->i_lock);
1438 inode->i_blocks += BLOCKS_PER_PAGE;
1439 spin_unlock(&inode->i_lock);
1440 } else if (shmem_acct_block(info->flags))
1446 if (!prealloc_page) {
1447 spin_unlock(&info->lock);
1448 filepage = shmem_alloc_page(gfp, info, idx);
1450 shmem_unacct_blocks(info->flags, 1);
1451 shmem_free_blocks(inode, 1);
1455 SetPageSwapBacked(filepage);
1458 * Precharge page while we can wait, compensate
1461 error = mem_cgroup_cache_charge(filepage,
1462 current->mm, GFP_KERNEL);
1464 page_cache_release(filepage);
1465 shmem_unacct_blocks(info->flags, 1);
1466 shmem_free_blocks(inode, 1);
1471 spin_lock(&info->lock);
1473 filepage = prealloc_page;
1474 prealloc_page = NULL;
1475 SetPageSwapBacked(filepage);
1478 entry = shmem_swp_alloc(info, idx, sgp);
1480 error = PTR_ERR(entry);
1483 shmem_swp_unmap(entry);
1485 ret = error || swap.val;
1487 mem_cgroup_uncharge_cache_page(filepage);
1489 ret = add_to_page_cache_lru(filepage, mapping,
1492 * At add_to_page_cache_lru() failure, uncharge will
1493 * be done automatically.
1496 spin_unlock(&info->lock);
1497 page_cache_release(filepage);
1498 shmem_unacct_blocks(info->flags, 1);
1499 shmem_free_blocks(inode, 1);
1505 info->flags |= SHMEM_PAGEIN;
1509 spin_unlock(&info->lock);
1510 clear_highpage(filepage);
1511 flush_dcache_page(filepage);
1512 SetPageUptodate(filepage);
1513 if (sgp == SGP_DIRTY)
1514 set_page_dirty(filepage);
1523 * Perhaps the page was brought in from swap between find_lock_page
1524 * and taking info->lock? We allow for that at add_to_page_cache_lru,
1525 * but must also avoid reporting a spurious ENOSPC while working on a
1526 * full tmpfs. (When filepage has been passed in to shmem_getpage, it
1527 * is already in page cache, which prevents this race from occurring.)
1530 struct page *page = find_get_page(mapping, idx);
1532 spin_unlock(&info->lock);
1533 page_cache_release(page);
1537 spin_unlock(&info->lock);
1540 if (*pagep != filepage) {
1541 unlock_page(filepage);
1542 page_cache_release(filepage);
1545 if (prealloc_page) {
1546 mem_cgroup_uncharge_cache_page(prealloc_page);
1547 page_cache_release(prealloc_page);
1552 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1554 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1558 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1559 return VM_FAULT_SIGBUS;
1561 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1563 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1564 if (ret & VM_FAULT_MAJOR) {
1565 count_vm_event(PGMAJFAULT);
1566 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1568 return ret | VM_FAULT_LOCKED;
1572 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1574 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1575 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1578 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1581 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1584 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1585 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1589 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1591 struct inode *inode = file->f_path.dentry->d_inode;
1592 struct shmem_inode_info *info = SHMEM_I(inode);
1593 int retval = -ENOMEM;
1595 spin_lock(&info->lock);
1596 if (lock && !(info->flags & VM_LOCKED)) {
1597 if (!user_shm_lock(inode->i_size, user))
1599 info->flags |= VM_LOCKED;
1600 mapping_set_unevictable(file->f_mapping);
1602 if (!lock && (info->flags & VM_LOCKED) && user) {
1603 user_shm_unlock(inode->i_size, user);
1604 info->flags &= ~VM_LOCKED;
1605 mapping_clear_unevictable(file->f_mapping);
1606 scan_mapping_unevictable_pages(file->f_mapping);
1611 spin_unlock(&info->lock);
1615 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1617 file_accessed(file);
1618 vma->vm_ops = &shmem_vm_ops;
1619 vma->vm_flags |= VM_CAN_NONLINEAR;
1623 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1624 int mode, dev_t dev, unsigned long flags)
1626 struct inode *inode;
1627 struct shmem_inode_info *info;
1628 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1630 if (shmem_reserve_inode(sb))
1633 inode = new_inode(sb);
1635 inode->i_ino = get_next_ino();
1636 inode_init_owner(inode, dir, mode);
1637 inode->i_blocks = 0;
1638 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1639 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1640 inode->i_generation = get_seconds();
1641 info = SHMEM_I(inode);
1642 memset(info, 0, (char *)inode - (char *)info);
1643 spin_lock_init(&info->lock);
1644 info->flags = flags & VM_NORESERVE;
1645 INIT_LIST_HEAD(&info->swaplist);
1646 INIT_LIST_HEAD(&info->xattr_list);
1647 cache_no_acl(inode);
1649 switch (mode & S_IFMT) {
1651 inode->i_op = &shmem_special_inode_operations;
1652 init_special_inode(inode, mode, dev);
1655 inode->i_mapping->a_ops = &shmem_aops;
1656 inode->i_op = &shmem_inode_operations;
1657 inode->i_fop = &shmem_file_operations;
1658 mpol_shared_policy_init(&info->policy,
1659 shmem_get_sbmpol(sbinfo));
1663 /* Some things misbehave if size == 0 on a directory */
1664 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1665 inode->i_op = &shmem_dir_inode_operations;
1666 inode->i_fop = &simple_dir_operations;
1670 * Must not load anything in the rbtree,
1671 * mpol_free_shared_policy will not be called.
1673 mpol_shared_policy_init(&info->policy, NULL);
1677 shmem_free_inode(sb);
1682 static const struct inode_operations shmem_symlink_inode_operations;
1683 static const struct inode_operations shmem_symlink_inline_operations;
1686 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1687 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1688 * below the loop driver, in the generic fashion that many filesystems support.
1690 static int shmem_readpage(struct file *file, struct page *page)
1692 struct inode *inode = page->mapping->host;
1693 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1698 #ifdef CONFIG_TMPFS_XATTR
1699 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1701 #define shmem_initxattrs NULL
1705 shmem_write_begin(struct file *file, struct address_space *mapping,
1706 loff_t pos, unsigned len, unsigned flags,
1707 struct page **pagep, void **fsdata)
1709 struct inode *inode = mapping->host;
1710 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1712 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1716 shmem_write_end(struct file *file, struct address_space *mapping,
1717 loff_t pos, unsigned len, unsigned copied,
1718 struct page *page, void *fsdata)
1720 struct inode *inode = mapping->host;
1722 if (pos + copied > inode->i_size)
1723 i_size_write(inode, pos + copied);
1725 set_page_dirty(page);
1727 page_cache_release(page);
1732 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1734 struct inode *inode = filp->f_path.dentry->d_inode;
1735 struct address_space *mapping = inode->i_mapping;
1736 unsigned long index, offset;
1737 enum sgp_type sgp = SGP_READ;
1740 * Might this read be for a stacking filesystem? Then when reading
1741 * holes of a sparse file, we actually need to allocate those pages,
1742 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1744 if (segment_eq(get_fs(), KERNEL_DS))
1747 index = *ppos >> PAGE_CACHE_SHIFT;
1748 offset = *ppos & ~PAGE_CACHE_MASK;
1751 struct page *page = NULL;
1752 unsigned long end_index, nr, ret;
1753 loff_t i_size = i_size_read(inode);
1755 end_index = i_size >> PAGE_CACHE_SHIFT;
1756 if (index > end_index)
1758 if (index == end_index) {
1759 nr = i_size & ~PAGE_CACHE_MASK;
1764 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1766 if (desc->error == -EINVAL)
1774 * We must evaluate after, since reads (unlike writes)
1775 * are called without i_mutex protection against truncate
1777 nr = PAGE_CACHE_SIZE;
1778 i_size = i_size_read(inode);
1779 end_index = i_size >> PAGE_CACHE_SHIFT;
1780 if (index == end_index) {
1781 nr = i_size & ~PAGE_CACHE_MASK;
1784 page_cache_release(page);
1792 * If users can be writing to this page using arbitrary
1793 * virtual addresses, take care about potential aliasing
1794 * before reading the page on the kernel side.
1796 if (mapping_writably_mapped(mapping))
1797 flush_dcache_page(page);
1799 * Mark the page accessed if we read the beginning.
1802 mark_page_accessed(page);
1804 page = ZERO_PAGE(0);
1805 page_cache_get(page);
1809 * Ok, we have the page, and it's up-to-date, so
1810 * now we can copy it to user space...
1812 * The actor routine returns how many bytes were actually used..
1813 * NOTE! This may not be the same as how much of a user buffer
1814 * we filled up (we may be padding etc), so we can only update
1815 * "pos" here (the actor routine has to update the user buffer
1816 * pointers and the remaining count).
1818 ret = actor(desc, page, offset, nr);
1820 index += offset >> PAGE_CACHE_SHIFT;
1821 offset &= ~PAGE_CACHE_MASK;
1823 page_cache_release(page);
1824 if (ret != nr || !desc->count)
1830 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1831 file_accessed(filp);
1834 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1835 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1837 struct file *filp = iocb->ki_filp;
1841 loff_t *ppos = &iocb->ki_pos;
1843 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1847 for (seg = 0; seg < nr_segs; seg++) {
1848 read_descriptor_t desc;
1851 desc.arg.buf = iov[seg].iov_base;
1852 desc.count = iov[seg].iov_len;
1853 if (desc.count == 0)
1856 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1857 retval += desc.written;
1859 retval = retval ?: desc.error;
1868 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1870 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1872 buf->f_type = TMPFS_MAGIC;
1873 buf->f_bsize = PAGE_CACHE_SIZE;
1874 buf->f_namelen = NAME_MAX;
1875 if (sbinfo->max_blocks) {
1876 buf->f_blocks = sbinfo->max_blocks;
1877 buf->f_bavail = buf->f_bfree =
1878 sbinfo->max_blocks - percpu_counter_sum(&sbinfo->used_blocks);
1880 if (sbinfo->max_inodes) {
1881 buf->f_files = sbinfo->max_inodes;
1882 buf->f_ffree = sbinfo->free_inodes;
1884 /* else leave those fields 0 like simple_statfs */
1889 * File creation. Allocate an inode, and we're done..
1892 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1894 struct inode *inode;
1895 int error = -ENOSPC;
1897 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1899 error = security_inode_init_security(inode, dir,
1901 shmem_initxattrs, NULL);
1903 if (error != -EOPNOTSUPP) {
1908 #ifdef CONFIG_TMPFS_POSIX_ACL
1909 error = generic_acl_init(inode, dir);
1917 dir->i_size += BOGO_DIRENT_SIZE;
1918 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1919 d_instantiate(dentry, inode);
1920 dget(dentry); /* Extra count - pin the dentry in core */
1925 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1929 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1935 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1936 struct nameidata *nd)
1938 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1944 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1946 struct inode *inode = old_dentry->d_inode;
1950 * No ordinary (disk based) filesystem counts links as inodes;
1951 * but each new link needs a new dentry, pinning lowmem, and
1952 * tmpfs dentries cannot be pruned until they are unlinked.
1954 ret = shmem_reserve_inode(inode->i_sb);
1958 dir->i_size += BOGO_DIRENT_SIZE;
1959 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1961 ihold(inode); /* New dentry reference */
1962 dget(dentry); /* Extra pinning count for the created dentry */
1963 d_instantiate(dentry, inode);
1968 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1970 struct inode *inode = dentry->d_inode;
1972 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1973 shmem_free_inode(inode->i_sb);
1975 dir->i_size -= BOGO_DIRENT_SIZE;
1976 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1978 dput(dentry); /* Undo the count from "create" - this does all the work */
1982 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1984 if (!simple_empty(dentry))
1987 drop_nlink(dentry->d_inode);
1989 return shmem_unlink(dir, dentry);
1993 * The VFS layer already does all the dentry stuff for rename,
1994 * we just have to decrement the usage count for the target if
1995 * it exists so that the VFS layer correctly free's it when it
1998 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2000 struct inode *inode = old_dentry->d_inode;
2001 int they_are_dirs = S_ISDIR(inode->i_mode);
2003 if (!simple_empty(new_dentry))
2006 if (new_dentry->d_inode) {
2007 (void) shmem_unlink(new_dir, new_dentry);
2009 drop_nlink(old_dir);
2010 } else if (they_are_dirs) {
2011 drop_nlink(old_dir);
2015 old_dir->i_size -= BOGO_DIRENT_SIZE;
2016 new_dir->i_size += BOGO_DIRENT_SIZE;
2017 old_dir->i_ctime = old_dir->i_mtime =
2018 new_dir->i_ctime = new_dir->i_mtime =
2019 inode->i_ctime = CURRENT_TIME;
2023 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2027 struct inode *inode;
2028 struct page *page = NULL;
2030 struct shmem_inode_info *info;
2032 len = strlen(symname) + 1;
2033 if (len > PAGE_CACHE_SIZE)
2034 return -ENAMETOOLONG;
2036 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2040 error = security_inode_init_security(inode, dir, &dentry->d_name,
2041 shmem_initxattrs, NULL);
2043 if (error != -EOPNOTSUPP) {
2050 info = SHMEM_I(inode);
2051 inode->i_size = len-1;
2052 if (len <= SHMEM_SYMLINK_INLINE_LEN) {
2054 memcpy(info->inline_symlink, symname, len);
2055 inode->i_op = &shmem_symlink_inline_operations;
2057 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2062 inode->i_mapping->a_ops = &shmem_aops;
2063 inode->i_op = &shmem_symlink_inode_operations;
2064 kaddr = kmap_atomic(page, KM_USER0);
2065 memcpy(kaddr, symname, len);
2066 kunmap_atomic(kaddr, KM_USER0);
2067 set_page_dirty(page);
2069 page_cache_release(page);
2071 dir->i_size += BOGO_DIRENT_SIZE;
2072 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2073 d_instantiate(dentry, inode);
2078 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
2080 nd_set_link(nd, SHMEM_I(dentry->d_inode)->inline_symlink);
2084 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2086 struct page *page = NULL;
2087 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2088 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2094 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2096 if (!IS_ERR(nd_get_link(nd))) {
2097 struct page *page = cookie;
2099 mark_page_accessed(page);
2100 page_cache_release(page);
2104 #ifdef CONFIG_TMPFS_XATTR
2106 * Superblocks without xattr inode operations may get some security.* xattr
2107 * support from the LSM "for free". As soon as we have any other xattrs
2108 * like ACLs, we also need to implement the security.* handlers at
2109 * filesystem level, though.
2113 * Allocate new xattr and copy in the value; but leave the name to callers.
2115 static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2117 struct shmem_xattr *new_xattr;
2121 len = sizeof(*new_xattr) + size;
2122 if (len <= sizeof(*new_xattr))
2125 new_xattr = kmalloc(len, GFP_KERNEL);
2129 new_xattr->size = size;
2130 memcpy(new_xattr->value, value, size);
2135 * Callback for security_inode_init_security() for acquiring xattrs.
2137 static int shmem_initxattrs(struct inode *inode,
2138 const struct xattr *xattr_array,
2141 struct shmem_inode_info *info = SHMEM_I(inode);
2142 const struct xattr *xattr;
2143 struct shmem_xattr *new_xattr;
2146 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2147 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2151 len = strlen(xattr->name) + 1;
2152 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2154 if (!new_xattr->name) {
2159 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2160 XATTR_SECURITY_PREFIX_LEN);
2161 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2164 spin_lock(&info->lock);
2165 list_add(&new_xattr->list, &info->xattr_list);
2166 spin_unlock(&info->lock);
2172 static int shmem_xattr_get(struct dentry *dentry, const char *name,
2173 void *buffer, size_t size)
2175 struct shmem_inode_info *info;
2176 struct shmem_xattr *xattr;
2179 info = SHMEM_I(dentry->d_inode);
2181 spin_lock(&info->lock);
2182 list_for_each_entry(xattr, &info->xattr_list, list) {
2183 if (strcmp(name, xattr->name))
2188 if (size < xattr->size)
2191 memcpy(buffer, xattr->value, xattr->size);
2195 spin_unlock(&info->lock);
2199 static int shmem_xattr_set(struct inode *inode, const char *name,
2200 const void *value, size_t size, int flags)
2202 struct shmem_inode_info *info = SHMEM_I(inode);
2203 struct shmem_xattr *xattr;
2204 struct shmem_xattr *new_xattr = NULL;
2207 /* value == NULL means remove */
2209 new_xattr = shmem_xattr_alloc(value, size);
2213 new_xattr->name = kstrdup(name, GFP_KERNEL);
2214 if (!new_xattr->name) {
2220 spin_lock(&info->lock);
2221 list_for_each_entry(xattr, &info->xattr_list, list) {
2222 if (!strcmp(name, xattr->name)) {
2223 if (flags & XATTR_CREATE) {
2226 } else if (new_xattr) {
2227 list_replace(&xattr->list, &new_xattr->list);
2229 list_del(&xattr->list);
2234 if (flags & XATTR_REPLACE) {
2238 list_add(&new_xattr->list, &info->xattr_list);
2242 spin_unlock(&info->lock);
2250 static const struct xattr_handler *shmem_xattr_handlers[] = {
2251 #ifdef CONFIG_TMPFS_POSIX_ACL
2252 &generic_acl_access_handler,
2253 &generic_acl_default_handler,
2258 static int shmem_xattr_validate(const char *name)
2260 struct { const char *prefix; size_t len; } arr[] = {
2261 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2262 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2266 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2267 size_t preflen = arr[i].len;
2268 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2277 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2278 void *buffer, size_t size)
2283 * If this is a request for a synthetic attribute in the system.*
2284 * namespace use the generic infrastructure to resolve a handler
2285 * for it via sb->s_xattr.
2287 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2288 return generic_getxattr(dentry, name, buffer, size);
2290 err = shmem_xattr_validate(name);
2294 return shmem_xattr_get(dentry, name, buffer, size);
2297 static int shmem_setxattr(struct dentry *dentry, const char *name,
2298 const void *value, size_t size, int flags)
2303 * If this is a request for a synthetic attribute in the system.*
2304 * namespace use the generic infrastructure to resolve a handler
2305 * for it via sb->s_xattr.
2307 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2308 return generic_setxattr(dentry, name, value, size, flags);
2310 err = shmem_xattr_validate(name);
2315 value = ""; /* empty EA, do not remove */
2317 return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
2321 static int shmem_removexattr(struct dentry *dentry, const char *name)
2326 * If this is a request for a synthetic attribute in the system.*
2327 * namespace use the generic infrastructure to resolve a handler
2328 * for it via sb->s_xattr.
2330 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2331 return generic_removexattr(dentry, name);
2333 err = shmem_xattr_validate(name);
2337 return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
2340 static bool xattr_is_trusted(const char *name)
2342 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2345 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2347 bool trusted = capable(CAP_SYS_ADMIN);
2348 struct shmem_xattr *xattr;
2349 struct shmem_inode_info *info;
2352 info = SHMEM_I(dentry->d_inode);
2354 spin_lock(&info->lock);
2355 list_for_each_entry(xattr, &info->xattr_list, list) {
2358 /* skip "trusted." attributes for unprivileged callers */
2359 if (!trusted && xattr_is_trusted(xattr->name))
2362 len = strlen(xattr->name) + 1;
2369 memcpy(buffer, xattr->name, len);
2373 spin_unlock(&info->lock);
2377 #endif /* CONFIG_TMPFS_XATTR */
2379 static const struct inode_operations shmem_symlink_inline_operations = {
2380 .readlink = generic_readlink,
2381 .follow_link = shmem_follow_link_inline,
2382 #ifdef CONFIG_TMPFS_XATTR
2383 .setxattr = shmem_setxattr,
2384 .getxattr = shmem_getxattr,
2385 .listxattr = shmem_listxattr,
2386 .removexattr = shmem_removexattr,
2390 static const struct inode_operations shmem_symlink_inode_operations = {
2391 .readlink = generic_readlink,
2392 .follow_link = shmem_follow_link,
2393 .put_link = shmem_put_link,
2394 #ifdef CONFIG_TMPFS_XATTR
2395 .setxattr = shmem_setxattr,
2396 .getxattr = shmem_getxattr,
2397 .listxattr = shmem_listxattr,
2398 .removexattr = shmem_removexattr,
2402 static struct dentry *shmem_get_parent(struct dentry *child)
2404 return ERR_PTR(-ESTALE);
2407 static int shmem_match(struct inode *ino, void *vfh)
2411 inum = (inum << 32) | fh[1];
2412 return ino->i_ino == inum && fh[0] == ino->i_generation;
2415 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2416 struct fid *fid, int fh_len, int fh_type)
2418 struct inode *inode;
2419 struct dentry *dentry = NULL;
2426 inum = (inum << 32) | fid->raw[1];
2428 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2429 shmem_match, fid->raw);
2431 dentry = d_find_alias(inode);
2438 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2441 struct inode *inode = dentry->d_inode;
2448 if (inode_unhashed(inode)) {
2449 /* Unfortunately insert_inode_hash is not idempotent,
2450 * so as we hash inodes here rather than at creation
2451 * time, we need a lock to ensure we only try
2454 static DEFINE_SPINLOCK(lock);
2456 if (inode_unhashed(inode))
2457 __insert_inode_hash(inode,
2458 inode->i_ino + inode->i_generation);
2462 fh[0] = inode->i_generation;
2463 fh[1] = inode->i_ino;
2464 fh[2] = ((__u64)inode->i_ino) >> 32;
2470 static const struct export_operations shmem_export_ops = {
2471 .get_parent = shmem_get_parent,
2472 .encode_fh = shmem_encode_fh,
2473 .fh_to_dentry = shmem_fh_to_dentry,
2476 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2479 char *this_char, *value, *rest;
2481 while (options != NULL) {
2482 this_char = options;
2485 * NUL-terminate this option: unfortunately,
2486 * mount options form a comma-separated list,
2487 * but mpol's nodelist may also contain commas.
2489 options = strchr(options, ',');
2490 if (options == NULL)
2493 if (!isdigit(*options)) {
2500 if ((value = strchr(this_char,'=')) != NULL) {
2504 "tmpfs: No value for mount option '%s'\n",
2509 if (!strcmp(this_char,"size")) {
2510 unsigned long long size;
2511 size = memparse(value,&rest);
2513 size <<= PAGE_SHIFT;
2514 size *= totalram_pages;
2520 sbinfo->max_blocks =
2521 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2522 } else if (!strcmp(this_char,"nr_blocks")) {
2523 sbinfo->max_blocks = memparse(value, &rest);
2526 } else if (!strcmp(this_char,"nr_inodes")) {
2527 sbinfo->max_inodes = memparse(value, &rest);
2530 } else if (!strcmp(this_char,"mode")) {
2533 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2536 } else if (!strcmp(this_char,"uid")) {
2539 sbinfo->uid = simple_strtoul(value, &rest, 0);
2542 } else if (!strcmp(this_char,"gid")) {
2545 sbinfo->gid = simple_strtoul(value, &rest, 0);
2548 } else if (!strcmp(this_char,"mpol")) {
2549 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2552 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2560 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2566 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2568 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2569 struct shmem_sb_info config = *sbinfo;
2570 unsigned long inodes;
2571 int error = -EINVAL;
2574 if (shmem_parse_options(data, &config, true))
2577 spin_lock(&sbinfo->stat_lock);
2578 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2579 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2581 if (config.max_inodes < inodes)
2584 * Those tests also disallow limited->unlimited while any are in
2585 * use, so i_blocks will always be zero when max_blocks is zero;
2586 * but we must separately disallow unlimited->limited, because
2587 * in that case we have no record of how much is already in use.
2589 if (config.max_blocks && !sbinfo->max_blocks)
2591 if (config.max_inodes && !sbinfo->max_inodes)
2595 sbinfo->max_blocks = config.max_blocks;
2596 sbinfo->max_inodes = config.max_inodes;
2597 sbinfo->free_inodes = config.max_inodes - inodes;
2600 * Preserve previous mempolicy unless mpol remount option was specified.
2603 mpol_put(sbinfo->mpol);
2604 sbinfo->mpol = config.mpol; /* transfers initial ref */
2607 spin_unlock(&sbinfo->stat_lock);
2611 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2613 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2615 if (sbinfo->max_blocks != shmem_default_max_blocks())
2616 seq_printf(seq, ",size=%luk",
2617 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2618 if (sbinfo->max_inodes != shmem_default_max_inodes())
2619 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2620 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2621 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2622 if (sbinfo->uid != 0)
2623 seq_printf(seq, ",uid=%u", sbinfo->uid);
2624 if (sbinfo->gid != 0)
2625 seq_printf(seq, ",gid=%u", sbinfo->gid);
2626 shmem_show_mpol(seq, sbinfo->mpol);
2629 #endif /* CONFIG_TMPFS */
2631 static void shmem_put_super(struct super_block *sb)
2633 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2635 percpu_counter_destroy(&sbinfo->used_blocks);
2637 sb->s_fs_info = NULL;
2640 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2642 struct inode *inode;
2643 struct dentry *root;
2644 struct shmem_sb_info *sbinfo;
2647 /* Round up to L1_CACHE_BYTES to resist false sharing */
2648 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2649 L1_CACHE_BYTES), GFP_KERNEL);
2653 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2654 sbinfo->uid = current_fsuid();
2655 sbinfo->gid = current_fsgid();
2656 sb->s_fs_info = sbinfo;
2660 * Per default we only allow half of the physical ram per
2661 * tmpfs instance, limiting inodes to one per page of lowmem;
2662 * but the internal instance is left unlimited.
2664 if (!(sb->s_flags & MS_NOUSER)) {
2665 sbinfo->max_blocks = shmem_default_max_blocks();
2666 sbinfo->max_inodes = shmem_default_max_inodes();
2667 if (shmem_parse_options(data, sbinfo, false)) {
2672 sb->s_export_op = &shmem_export_ops;
2674 sb->s_flags |= MS_NOUSER;
2677 spin_lock_init(&sbinfo->stat_lock);
2678 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2680 sbinfo->free_inodes = sbinfo->max_inodes;
2682 sb->s_maxbytes = SHMEM_MAX_BYTES;
2683 sb->s_blocksize = PAGE_CACHE_SIZE;
2684 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2685 sb->s_magic = TMPFS_MAGIC;
2686 sb->s_op = &shmem_ops;
2687 sb->s_time_gran = 1;
2688 #ifdef CONFIG_TMPFS_XATTR
2689 sb->s_xattr = shmem_xattr_handlers;
2691 #ifdef CONFIG_TMPFS_POSIX_ACL
2692 sb->s_flags |= MS_POSIXACL;
2695 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2698 inode->i_uid = sbinfo->uid;
2699 inode->i_gid = sbinfo->gid;
2700 root = d_alloc_root(inode);
2709 shmem_put_super(sb);
2713 static struct kmem_cache *shmem_inode_cachep;
2715 static struct inode *shmem_alloc_inode(struct super_block *sb)
2717 struct shmem_inode_info *p;
2718 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2721 return &p->vfs_inode;
2724 static void shmem_i_callback(struct rcu_head *head)
2726 struct inode *inode = container_of(head, struct inode, i_rcu);
2727 INIT_LIST_HEAD(&inode->i_dentry);
2728 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2731 static void shmem_destroy_inode(struct inode *inode)
2733 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2734 /* only struct inode is valid if it's an inline symlink */
2735 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2737 call_rcu(&inode->i_rcu, shmem_i_callback);
2740 static void init_once(void *foo)
2742 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2744 inode_init_once(&p->vfs_inode);
2747 static int init_inodecache(void)
2749 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2750 sizeof(struct shmem_inode_info),
2751 0, SLAB_PANIC, init_once);
2755 static void destroy_inodecache(void)
2757 kmem_cache_destroy(shmem_inode_cachep);
2760 static const struct address_space_operations shmem_aops = {
2761 .writepage = shmem_writepage,
2762 .set_page_dirty = __set_page_dirty_no_writeback,
2764 .readpage = shmem_readpage,
2765 .write_begin = shmem_write_begin,
2766 .write_end = shmem_write_end,
2768 .migratepage = migrate_page,
2769 .error_remove_page = generic_error_remove_page,
2772 static const struct file_operations shmem_file_operations = {
2775 .llseek = generic_file_llseek,
2776 .read = do_sync_read,
2777 .write = do_sync_write,
2778 .aio_read = shmem_file_aio_read,
2779 .aio_write = generic_file_aio_write,
2780 .fsync = noop_fsync,
2781 .splice_read = generic_file_splice_read,
2782 .splice_write = generic_file_splice_write,
2786 static const struct inode_operations shmem_inode_operations = {
2787 .setattr = shmem_setattr,
2788 .truncate_range = shmem_truncate_range,
2789 #ifdef CONFIG_TMPFS_XATTR
2790 .setxattr = shmem_setxattr,
2791 .getxattr = shmem_getxattr,
2792 .listxattr = shmem_listxattr,
2793 .removexattr = shmem_removexattr,
2795 #ifdef CONFIG_TMPFS_POSIX_ACL
2796 .check_acl = generic_check_acl,
2801 static const struct inode_operations shmem_dir_inode_operations = {
2803 .create = shmem_create,
2804 .lookup = simple_lookup,
2806 .unlink = shmem_unlink,
2807 .symlink = shmem_symlink,
2808 .mkdir = shmem_mkdir,
2809 .rmdir = shmem_rmdir,
2810 .mknod = shmem_mknod,
2811 .rename = shmem_rename,
2813 #ifdef CONFIG_TMPFS_XATTR
2814 .setxattr = shmem_setxattr,
2815 .getxattr = shmem_getxattr,
2816 .listxattr = shmem_listxattr,
2817 .removexattr = shmem_removexattr,
2819 #ifdef CONFIG_TMPFS_POSIX_ACL
2820 .setattr = shmem_setattr,
2821 .check_acl = generic_check_acl,
2825 static const struct inode_operations shmem_special_inode_operations = {
2826 #ifdef CONFIG_TMPFS_XATTR
2827 .setxattr = shmem_setxattr,
2828 .getxattr = shmem_getxattr,
2829 .listxattr = shmem_listxattr,
2830 .removexattr = shmem_removexattr,
2832 #ifdef CONFIG_TMPFS_POSIX_ACL
2833 .setattr = shmem_setattr,
2834 .check_acl = generic_check_acl,
2838 static const struct super_operations shmem_ops = {
2839 .alloc_inode = shmem_alloc_inode,
2840 .destroy_inode = shmem_destroy_inode,
2842 .statfs = shmem_statfs,
2843 .remount_fs = shmem_remount_fs,
2844 .show_options = shmem_show_options,
2846 .evict_inode = shmem_evict_inode,
2847 .drop_inode = generic_delete_inode,
2848 .put_super = shmem_put_super,
2851 static const struct vm_operations_struct shmem_vm_ops = {
2852 .fault = shmem_fault,
2854 .set_policy = shmem_set_policy,
2855 .get_policy = shmem_get_policy,
2860 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2861 int flags, const char *dev_name, void *data)
2863 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2866 static struct file_system_type tmpfs_fs_type = {
2867 .owner = THIS_MODULE,
2869 .mount = shmem_mount,
2870 .kill_sb = kill_litter_super,
2873 int __init init_tmpfs(void)
2877 error = bdi_init(&shmem_backing_dev_info);
2881 error = init_inodecache();
2885 error = register_filesystem(&tmpfs_fs_type);
2887 printk(KERN_ERR "Could not register tmpfs\n");
2891 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2892 tmpfs_fs_type.name, NULL);
2893 if (IS_ERR(shm_mnt)) {
2894 error = PTR_ERR(shm_mnt);
2895 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2901 unregister_filesystem(&tmpfs_fs_type);
2903 destroy_inodecache();
2905 bdi_destroy(&shmem_backing_dev_info);
2907 shm_mnt = ERR_PTR(error);
2911 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2913 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2914 * @inode: the inode to be searched
2915 * @pgoff: the offset to be searched
2916 * @pagep: the pointer for the found page to be stored
2917 * @ent: the pointer for the found swap entry to be stored
2919 * If a page is found, refcount of it is incremented. Callers should handle
2922 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2923 struct page **pagep, swp_entry_t *ent)
2925 swp_entry_t entry = { .val = 0 }, *ptr;
2926 struct page *page = NULL;
2927 struct shmem_inode_info *info = SHMEM_I(inode);
2929 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2932 spin_lock(&info->lock);
2933 ptr = shmem_swp_entry(info, pgoff, NULL);
2935 if (ptr && ptr->val) {
2936 entry.val = ptr->val;
2937 page = find_get_page(&swapper_space, entry.val);
2940 page = find_get_page(inode->i_mapping, pgoff);
2942 shmem_swp_unmap(ptr);
2943 spin_unlock(&info->lock);
2950 #else /* !CONFIG_SHMEM */
2953 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2955 * This is intended for small system where the benefits of the full
2956 * shmem code (swap-backed and resource-limited) are outweighed by
2957 * their complexity. On systems without swap this code should be
2958 * effectively equivalent, but much lighter weight.
2961 #include <linux/ramfs.h>
2963 static struct file_system_type tmpfs_fs_type = {
2965 .mount = ramfs_mount,
2966 .kill_sb = kill_litter_super,
2969 int __init init_tmpfs(void)
2971 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2973 shm_mnt = kern_mount(&tmpfs_fs_type);
2974 BUG_ON(IS_ERR(shm_mnt));
2979 int shmem_unuse(swp_entry_t entry, struct page *page)
2984 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2989 void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
2991 truncate_inode_pages_range(inode->i_mapping, start, end);
2993 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2995 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2997 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2998 * @inode: the inode to be searched
2999 * @pgoff: the offset to be searched
3000 * @pagep: the pointer for the found page to be stored
3001 * @ent: the pointer for the found swap entry to be stored
3003 * If a page is found, refcount of it is incremented. Callers should handle
3006 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
3007 struct page **pagep, swp_entry_t *ent)
3009 struct page *page = NULL;
3011 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
3013 page = find_get_page(inode->i_mapping, pgoff);
3016 *ent = (swp_entry_t){ .val = 0 };
3020 #define shmem_vm_ops generic_file_vm_ops
3021 #define shmem_file_operations ramfs_file_operations
3022 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3023 #define shmem_acct_size(flags, size) 0
3024 #define shmem_unacct_size(flags, size) do {} while (0)
3025 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
3027 #endif /* CONFIG_SHMEM */
3032 * shmem_file_setup - get an unlinked file living in tmpfs
3033 * @name: name for dentry (to be seen in /proc/<pid>/maps
3034 * @size: size to be set for the file
3035 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3037 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3041 struct inode *inode;
3043 struct dentry *root;
3046 if (IS_ERR(shm_mnt))
3047 return (void *)shm_mnt;
3049 if (size < 0 || size > SHMEM_MAX_BYTES)
3050 return ERR_PTR(-EINVAL);
3052 if (shmem_acct_size(flags, size))
3053 return ERR_PTR(-ENOMEM);
3057 this.len = strlen(name);
3058 this.hash = 0; /* will go */
3059 root = shm_mnt->mnt_root;
3060 path.dentry = d_alloc(root, &this);
3063 path.mnt = mntget(shm_mnt);
3066 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
3070 d_instantiate(path.dentry, inode);
3071 inode->i_size = size;
3072 inode->i_nlink = 0; /* It is unlinked */
3074 error = ramfs_nommu_expand_for_mapping(inode, size);
3080 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
3081 &shmem_file_operations);
3090 shmem_unacct_size(flags, size);
3091 return ERR_PTR(error);
3093 EXPORT_SYMBOL_GPL(shmem_file_setup);
3095 void shmem_set_file(struct vm_area_struct *vma, struct file *file)
3099 vma->vm_file = file;
3100 vma->vm_ops = &shmem_vm_ops;
3101 vma->vm_flags |= VM_CAN_NONLINEAR;
3105 * shmem_zero_setup - setup a shared anonymous mapping
3106 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3108 int shmem_zero_setup(struct vm_area_struct *vma)
3111 loff_t size = vma->vm_end - vma->vm_start;
3113 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3115 return PTR_ERR(file);
3117 shmem_set_file(vma, file);
3122 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3123 * @mapping: the page's address_space
3124 * @index: the page index
3125 * @gfp: the page allocator flags to use if allocating
3127 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3128 * with any new page allocations done using the specified allocation flags.
3129 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3130 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3131 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3133 * Provide a stub for those callers to start using now, then later
3134 * flesh it out to call shmem_getpage() with additional gfp mask, when
3135 * shmem_file_splice_read() is added and shmem_readpage() is removed.
3137 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3138 pgoff_t index, gfp_t gfp)
3140 return read_cache_page_gfp(mapping, index, gfp);
3142 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);