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/swap.h>
33 static struct vfsmount *shm_mnt;
37 * This virtual memory filesystem is heavily based on the ramfs. It
38 * extends ramfs by the ability to use swap and honor resource limits
39 * which makes it a completely usable filesystem.
42 #include <linux/xattr.h>
43 #include <linux/exportfs.h>
44 #include <linux/posix_acl.h>
45 #include <linux/generic_acl.h>
46 #include <linux/mman.h>
47 #include <linux/string.h>
48 #include <linux/slab.h>
49 #include <linux/backing-dev.h>
50 #include <linux/shmem_fs.h>
51 #include <linux/writeback.h>
52 #include <linux/blkdev.h>
53 #include <linux/security.h>
54 #include <linux/swapops.h>
55 #include <linux/mempolicy.h>
56 #include <linux/namei.h>
57 #include <linux/ctype.h>
58 #include <linux/migrate.h>
59 #include <linux/highmem.h>
60 #include <linux/seq_file.h>
61 #include <linux/magic.h>
63 #include <asm/uaccess.h>
64 #include <asm/div64.h>
65 #include <asm/pgtable.h>
68 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
69 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
71 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
72 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
73 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
74 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
76 * We use / and * instead of shifts in the definitions below, so that the swap
77 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
79 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
80 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
82 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
83 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
85 #define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
86 #define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
88 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
91 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
92 #define SHMEM_PAGEIN VM_READ
93 #define SHMEM_TRUNCATE VM_WRITE
95 /* Definition to limit shmem_truncate's steps between cond_rescheds */
96 #define LATENCY_LIMIT 64
98 /* Pretend that each entry is of this size in directory's i_size */
99 #define BOGO_DIRENT_SIZE 20
101 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
103 SGP_READ, /* don't exceed i_size, don't allocate page */
104 SGP_CACHE, /* don't exceed i_size, may allocate page */
105 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
106 SGP_WRITE, /* may exceed i_size, may allocate page */
110 static unsigned long shmem_default_max_blocks(void)
112 return totalram_pages / 2;
115 static unsigned long shmem_default_max_inodes(void)
117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
121 static int shmem_getpage(struct inode *inode, unsigned long idx,
122 struct page **pagep, enum sgp_type sgp, int *type);
124 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
127 * The above definition of ENTRIES_PER_PAGE, and the use of
128 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
129 * might be reconsidered if it ever diverges from PAGE_SIZE.
131 * Mobility flags are masked out as swap vectors cannot move
133 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
134 PAGE_CACHE_SHIFT-PAGE_SHIFT);
137 static inline void shmem_dir_free(struct page *page)
139 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
142 static struct page **shmem_dir_map(struct page *page)
144 return (struct page **)kmap_atomic(page, KM_USER0);
147 static inline void shmem_dir_unmap(struct page **dir)
149 kunmap_atomic(dir, KM_USER0);
152 static swp_entry_t *shmem_swp_map(struct page *page)
154 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
157 static inline void shmem_swp_balance_unmap(void)
160 * When passing a pointer to an i_direct entry, to code which
161 * also handles indirect entries and so will shmem_swp_unmap,
162 * we must arrange for the preempt count to remain in balance.
163 * What kmap_atomic of a lowmem page does depends on config
164 * and architecture, so pretend to kmap_atomic some lowmem page.
166 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
169 static inline void shmem_swp_unmap(swp_entry_t *entry)
171 kunmap_atomic(entry, KM_USER1);
174 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
176 return sb->s_fs_info;
180 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
181 * for shared memory and for shared anonymous (/dev/zero) mappings
182 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
183 * consistent with the pre-accounting of private mappings ...
185 static inline int shmem_acct_size(unsigned long flags, loff_t size)
187 return (flags & VM_NORESERVE) ?
188 0 : security_vm_enough_memory_kern(VM_ACCT(size));
191 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
193 if (!(flags & VM_NORESERVE))
194 vm_unacct_memory(VM_ACCT(size));
198 * ... whereas tmpfs objects are accounted incrementally as
199 * pages are allocated, in order to allow huge sparse files.
200 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
201 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
203 static inline int shmem_acct_block(unsigned long flags)
205 return (flags & VM_NORESERVE) ?
206 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
209 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
211 if (flags & VM_NORESERVE)
212 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
215 static const struct super_operations shmem_ops;
216 static const struct address_space_operations shmem_aops;
217 static const struct file_operations shmem_file_operations;
218 static const struct inode_operations shmem_inode_operations;
219 static const struct inode_operations shmem_dir_inode_operations;
220 static const struct inode_operations shmem_special_inode_operations;
221 static const struct vm_operations_struct shmem_vm_ops;
223 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
224 .ra_pages = 0, /* No readahead */
225 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
226 .unplug_io_fn = default_unplug_io_fn,
229 static LIST_HEAD(shmem_swaplist);
230 static DEFINE_MUTEX(shmem_swaplist_mutex);
232 static void shmem_free_blocks(struct inode *inode, long pages)
234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
235 if (sbinfo->max_blocks) {
236 spin_lock(&sbinfo->stat_lock);
237 sbinfo->free_blocks += pages;
238 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
239 spin_unlock(&sbinfo->stat_lock);
243 static int shmem_reserve_inode(struct super_block *sb)
245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246 if (sbinfo->max_inodes) {
247 spin_lock(&sbinfo->stat_lock);
248 if (!sbinfo->free_inodes) {
249 spin_unlock(&sbinfo->stat_lock);
252 sbinfo->free_inodes--;
253 spin_unlock(&sbinfo->stat_lock);
258 static void shmem_free_inode(struct super_block *sb)
260 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
261 if (sbinfo->max_inodes) {
262 spin_lock(&sbinfo->stat_lock);
263 sbinfo->free_inodes++;
264 spin_unlock(&sbinfo->stat_lock);
269 * shmem_recalc_inode - recalculate the size of an inode
270 * @inode: inode to recalc
272 * We have to calculate the free blocks since the mm can drop
273 * undirtied hole pages behind our back.
275 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
276 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
278 * It has to be called with the spinlock held.
280 static void shmem_recalc_inode(struct inode *inode)
282 struct shmem_inode_info *info = SHMEM_I(inode);
285 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
287 info->alloced -= freed;
288 shmem_unacct_blocks(info->flags, freed);
289 shmem_free_blocks(inode, freed);
294 * shmem_swp_entry - find the swap vector position in the info structure
295 * @info: info structure for the inode
296 * @index: index of the page to find
297 * @page: optional page to add to the structure. Has to be preset to
300 * If there is no space allocated yet it will return NULL when
301 * page is NULL, else it will use the page for the needed block,
302 * setting it to NULL on return to indicate that it has been used.
304 * The swap vector is organized the following way:
306 * There are SHMEM_NR_DIRECT entries directly stored in the
307 * shmem_inode_info structure. So small files do not need an addional
310 * For pages with index > SHMEM_NR_DIRECT there is the pointer
311 * i_indirect which points to a page which holds in the first half
312 * doubly indirect blocks, in the second half triple indirect blocks:
314 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
315 * following layout (for SHMEM_NR_DIRECT == 16):
317 * i_indirect -> dir --> 16-19
330 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
332 unsigned long offset;
336 if (index < SHMEM_NR_DIRECT) {
337 shmem_swp_balance_unmap();
338 return info->i_direct+index;
340 if (!info->i_indirect) {
342 info->i_indirect = *page;
345 return NULL; /* need another page */
348 index -= SHMEM_NR_DIRECT;
349 offset = index % ENTRIES_PER_PAGE;
350 index /= ENTRIES_PER_PAGE;
351 dir = shmem_dir_map(info->i_indirect);
353 if (index >= ENTRIES_PER_PAGE/2) {
354 index -= ENTRIES_PER_PAGE/2;
355 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
356 index %= ENTRIES_PER_PAGE;
363 shmem_dir_unmap(dir);
364 return NULL; /* need another page */
366 shmem_dir_unmap(dir);
367 dir = shmem_dir_map(subdir);
373 if (!page || !(subdir = *page)) {
374 shmem_dir_unmap(dir);
375 return NULL; /* need a page */
380 shmem_dir_unmap(dir);
381 return shmem_swp_map(subdir) + offset;
384 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
386 long incdec = value? 1: -1;
389 info->swapped += incdec;
390 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
391 struct page *page = kmap_atomic_to_page(entry);
392 set_page_private(page, page_private(page) + incdec);
397 * shmem_swp_alloc - get the position of the swap entry for the page.
398 * @info: info structure for the inode
399 * @index: index of the page to find
400 * @sgp: check and recheck i_size? skip allocation?
402 * If the entry does not exist, allocate it.
404 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
406 struct inode *inode = &info->vfs_inode;
407 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
408 struct page *page = NULL;
411 if (sgp != SGP_WRITE &&
412 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
413 return ERR_PTR(-EINVAL);
415 while (!(entry = shmem_swp_entry(info, index, &page))) {
417 return shmem_swp_map(ZERO_PAGE(0));
419 * Test free_blocks against 1 not 0, since we have 1 data
420 * page (and perhaps indirect index pages) yet to allocate:
421 * a waste to allocate index if we cannot allocate data.
423 if (sbinfo->max_blocks) {
424 spin_lock(&sbinfo->stat_lock);
425 if (sbinfo->free_blocks <= 1) {
426 spin_unlock(&sbinfo->stat_lock);
427 return ERR_PTR(-ENOSPC);
429 sbinfo->free_blocks--;
430 inode->i_blocks += BLOCKS_PER_PAGE;
431 spin_unlock(&sbinfo->stat_lock);
434 spin_unlock(&info->lock);
435 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
436 spin_lock(&info->lock);
439 shmem_free_blocks(inode, 1);
440 return ERR_PTR(-ENOMEM);
442 if (sgp != SGP_WRITE &&
443 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
444 entry = ERR_PTR(-EINVAL);
447 if (info->next_index <= index)
448 info->next_index = index + 1;
451 /* another task gave its page, or truncated the file */
452 shmem_free_blocks(inode, 1);
453 shmem_dir_free(page);
455 if (info->next_index <= index && !IS_ERR(entry))
456 info->next_index = index + 1;
461 * shmem_free_swp - free some swap entries in a directory
462 * @dir: pointer to the directory
463 * @edir: pointer after last entry of the directory
464 * @punch_lock: pointer to spinlock when needed for the holepunch case
466 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
467 spinlock_t *punch_lock)
469 spinlock_t *punch_unlock = NULL;
473 for (ptr = dir; ptr < edir; ptr++) {
475 if (unlikely(punch_lock)) {
476 punch_unlock = punch_lock;
478 spin_lock(punch_unlock);
482 free_swap_and_cache(*ptr);
483 *ptr = (swp_entry_t){0};
488 spin_unlock(punch_unlock);
492 static int shmem_map_and_free_swp(struct page *subdir, int offset,
493 int limit, struct page ***dir, spinlock_t *punch_lock)
498 ptr = shmem_swp_map(subdir);
499 for (; offset < limit; offset += LATENCY_LIMIT) {
500 int size = limit - offset;
501 if (size > LATENCY_LIMIT)
502 size = LATENCY_LIMIT;
503 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
505 if (need_resched()) {
506 shmem_swp_unmap(ptr);
508 shmem_dir_unmap(*dir);
512 ptr = shmem_swp_map(subdir);
515 shmem_swp_unmap(ptr);
519 static void shmem_free_pages(struct list_head *next)
525 page = container_of(next, struct page, lru);
527 shmem_dir_free(page);
529 if (freed >= LATENCY_LIMIT) {
536 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
538 struct shmem_inode_info *info = SHMEM_I(inode);
543 unsigned long diroff;
549 LIST_HEAD(pages_to_free);
550 long nr_pages_to_free = 0;
551 long nr_swaps_freed = 0;
555 spinlock_t *needs_lock;
556 spinlock_t *punch_lock;
557 unsigned long upper_limit;
559 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
560 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
561 if (idx >= info->next_index)
564 spin_lock(&info->lock);
565 info->flags |= SHMEM_TRUNCATE;
566 if (likely(end == (loff_t) -1)) {
567 limit = info->next_index;
568 upper_limit = SHMEM_MAX_INDEX;
569 info->next_index = idx;
573 if (end + 1 >= inode->i_size) { /* we may free a little more */
574 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
576 upper_limit = SHMEM_MAX_INDEX;
578 limit = (end + 1) >> PAGE_CACHE_SHIFT;
581 needs_lock = &info->lock;
585 topdir = info->i_indirect;
586 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
587 info->i_indirect = NULL;
589 list_add(&topdir->lru, &pages_to_free);
591 spin_unlock(&info->lock);
593 if (info->swapped && idx < SHMEM_NR_DIRECT) {
594 ptr = info->i_direct;
596 if (size > SHMEM_NR_DIRECT)
597 size = SHMEM_NR_DIRECT;
598 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
602 * If there are no indirect blocks or we are punching a hole
603 * below indirect blocks, nothing to be done.
605 if (!topdir || limit <= SHMEM_NR_DIRECT)
609 * The truncation case has already dropped info->lock, and we're safe
610 * because i_size and next_index have already been lowered, preventing
611 * access beyond. But in the punch_hole case, we still need to take
612 * the lock when updating the swap directory, because there might be
613 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
614 * shmem_writepage. However, whenever we find we can remove a whole
615 * directory page (not at the misaligned start or end of the range),
616 * we first NULLify its pointer in the level above, and then have no
617 * need to take the lock when updating its contents: needs_lock and
618 * punch_lock (either pointing to info->lock or NULL) manage this.
621 upper_limit -= SHMEM_NR_DIRECT;
622 limit -= SHMEM_NR_DIRECT;
623 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
624 offset = idx % ENTRIES_PER_PAGE;
627 dir = shmem_dir_map(topdir);
628 stage = ENTRIES_PER_PAGEPAGE/2;
629 if (idx < ENTRIES_PER_PAGEPAGE/2) {
631 diroff = idx/ENTRIES_PER_PAGE;
633 dir += ENTRIES_PER_PAGE/2;
634 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
636 stage += ENTRIES_PER_PAGEPAGE;
639 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
640 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
641 if (!diroff && !offset && upper_limit >= stage) {
643 spin_lock(needs_lock);
645 spin_unlock(needs_lock);
650 list_add(&middir->lru, &pages_to_free);
652 shmem_dir_unmap(dir);
653 dir = shmem_dir_map(middir);
661 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
662 if (unlikely(idx == stage)) {
663 shmem_dir_unmap(dir);
664 dir = shmem_dir_map(topdir) +
665 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
668 idx += ENTRIES_PER_PAGEPAGE;
672 stage = idx + ENTRIES_PER_PAGEPAGE;
675 needs_lock = &info->lock;
676 if (upper_limit >= stage) {
678 spin_lock(needs_lock);
680 spin_unlock(needs_lock);
685 list_add(&middir->lru, &pages_to_free);
687 shmem_dir_unmap(dir);
689 dir = shmem_dir_map(middir);
692 punch_lock = needs_lock;
693 subdir = dir[diroff];
694 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
696 spin_lock(needs_lock);
698 spin_unlock(needs_lock);
703 list_add(&subdir->lru, &pages_to_free);
705 if (subdir && page_private(subdir) /* has swap entries */) {
707 if (size > ENTRIES_PER_PAGE)
708 size = ENTRIES_PER_PAGE;
709 freed = shmem_map_and_free_swp(subdir,
710 offset, size, &dir, punch_lock);
712 dir = shmem_dir_map(middir);
713 nr_swaps_freed += freed;
714 if (offset || punch_lock) {
715 spin_lock(&info->lock);
716 set_page_private(subdir,
717 page_private(subdir) - freed);
718 spin_unlock(&info->lock);
720 BUG_ON(page_private(subdir) != freed);
725 shmem_dir_unmap(dir);
727 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
729 * Call truncate_inode_pages again: racing shmem_unuse_inode
730 * may have swizzled a page in from swap since
731 * truncate_pagecache or generic_delete_inode did it, before we
732 * lowered next_index. Also, though shmem_getpage checks
733 * i_size before adding to cache, no recheck after: so fix the
734 * narrow window there too.
736 * Recalling truncate_inode_pages_range and unmap_mapping_range
737 * every time for punch_hole (which never got a chance to clear
738 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
739 * yet hardly ever necessary: try to optimize them out later.
741 truncate_inode_pages_range(inode->i_mapping, start, end);
743 unmap_mapping_range(inode->i_mapping, start,
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);
764 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
766 struct inode *inode = dentry->d_inode;
767 loff_t newsize = attr->ia_size;
770 error = inode_change_ok(inode, attr);
774 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)
775 && newsize != inode->i_size) {
776 struct page *page = NULL;
778 if (newsize < inode->i_size) {
780 * If truncating down to a partial page, then
781 * if that page is already allocated, hold it
782 * in memory until the truncation is over, so
783 * truncate_partial_page cannnot miss it were
784 * it assigned to swap.
786 if (newsize & (PAGE_CACHE_SIZE-1)) {
787 (void) shmem_getpage(inode,
788 newsize >> PAGE_CACHE_SHIFT,
789 &page, SGP_READ, NULL);
794 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
795 * detect if any pages might have been added to cache
796 * after truncate_inode_pages. But we needn't bother
797 * if it's being fully truncated to zero-length: the
798 * nrpages check is efficient enough in that case.
801 struct shmem_inode_info *info = SHMEM_I(inode);
802 spin_lock(&info->lock);
803 info->flags &= ~SHMEM_PAGEIN;
804 spin_unlock(&info->lock);
808 error = simple_setsize(inode, newsize);
810 page_cache_release(page);
813 shmem_truncate_range(inode, newsize, (loff_t)-1);
816 setattr_copy(inode, attr);
817 #ifdef CONFIG_TMPFS_POSIX_ACL
818 if (attr->ia_valid & ATTR_MODE)
819 error = generic_acl_chmod(inode);
824 static void shmem_delete_inode(struct inode *inode)
826 struct shmem_inode_info *info = SHMEM_I(inode);
828 if (inode->i_mapping->a_ops == &shmem_aops) {
829 truncate_inode_pages(inode->i_mapping, 0);
830 shmem_unacct_size(info->flags, inode->i_size);
832 shmem_truncate_range(inode, 0, (loff_t)-1);
833 if (!list_empty(&info->swaplist)) {
834 mutex_lock(&shmem_swaplist_mutex);
835 list_del_init(&info->swaplist);
836 mutex_unlock(&shmem_swaplist_mutex);
839 BUG_ON(inode->i_blocks);
840 shmem_free_inode(inode->i_sb);
844 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
848 for (ptr = dir; ptr < edir; ptr++) {
849 if (ptr->val == entry.val)
855 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
869 ptr = info->i_direct;
870 spin_lock(&info->lock);
871 if (!info->swapped) {
872 list_del_init(&info->swaplist);
875 limit = info->next_index;
877 if (size > SHMEM_NR_DIRECT)
878 size = SHMEM_NR_DIRECT;
879 offset = shmem_find_swp(entry, ptr, ptr+size);
882 if (!info->i_indirect)
885 dir = shmem_dir_map(info->i_indirect);
886 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
888 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
889 if (unlikely(idx == stage)) {
890 shmem_dir_unmap(dir-1);
891 if (cond_resched_lock(&info->lock)) {
892 /* check it has not been truncated */
893 if (limit > info->next_index) {
894 limit = info->next_index;
899 dir = shmem_dir_map(info->i_indirect) +
900 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
903 idx += ENTRIES_PER_PAGEPAGE;
907 stage = idx + ENTRIES_PER_PAGEPAGE;
909 shmem_dir_unmap(dir);
910 dir = shmem_dir_map(subdir);
913 if (subdir && page_private(subdir)) {
914 ptr = shmem_swp_map(subdir);
916 if (size > ENTRIES_PER_PAGE)
917 size = ENTRIES_PER_PAGE;
918 offset = shmem_find_swp(entry, ptr, ptr+size);
919 shmem_swp_unmap(ptr);
921 shmem_dir_unmap(dir);
927 shmem_dir_unmap(dir-1);
929 spin_unlock(&info->lock);
933 inode = igrab(&info->vfs_inode);
934 spin_unlock(&info->lock);
937 * Move _head_ to start search for next from here.
938 * But be careful: shmem_delete_inode checks list_empty without taking
939 * mutex, and there's an instant in list_move_tail when info->swaplist
940 * would appear empty, if it were the only one on shmem_swaplist. We
941 * could avoid doing it if inode NULL; or use this minor optimization.
943 if (shmem_swaplist.next != &info->swaplist)
944 list_move_tail(&shmem_swaplist, &info->swaplist);
945 mutex_unlock(&shmem_swaplist_mutex);
951 * Charge page using GFP_KERNEL while we can wait.
952 * Charged back to the user(not to caller) when swap account is used.
953 * add_to_page_cache() will be called with GFP_NOWAIT.
955 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
958 error = radix_tree_preload(GFP_KERNEL);
960 mem_cgroup_uncharge_cache_page(page);
965 spin_lock(&info->lock);
966 ptr = shmem_swp_entry(info, idx, NULL);
967 if (ptr && ptr->val == entry.val) {
968 error = add_to_page_cache_locked(page, inode->i_mapping,
970 /* does mem_cgroup_uncharge_cache_page on error */
971 } else /* we must compensate for our precharge above */
972 mem_cgroup_uncharge_cache_page(page);
974 if (error == -EEXIST) {
975 struct page *filepage = find_get_page(inode->i_mapping, idx);
979 * There might be a more uptodate page coming down
980 * from a stacked writepage: forget our swappage if so.
982 if (PageUptodate(filepage))
984 page_cache_release(filepage);
988 delete_from_swap_cache(page);
989 set_page_dirty(page);
990 info->flags |= SHMEM_PAGEIN;
991 shmem_swp_set(info, ptr, 0);
993 error = 1; /* not an error, but entry was found */
996 shmem_swp_unmap(ptr);
997 spin_unlock(&info->lock);
998 radix_tree_preload_end();
1001 page_cache_release(page);
1002 iput(inode); /* allows for NULL */
1007 * shmem_unuse() search for an eventually swapped out shmem page.
1009 int shmem_unuse(swp_entry_t entry, struct page *page)
1011 struct list_head *p, *next;
1012 struct shmem_inode_info *info;
1015 mutex_lock(&shmem_swaplist_mutex);
1016 list_for_each_safe(p, next, &shmem_swaplist) {
1017 info = list_entry(p, struct shmem_inode_info, swaplist);
1018 found = shmem_unuse_inode(info, entry, page);
1023 mutex_unlock(&shmem_swaplist_mutex);
1025 * Can some race bring us here? We've been holding page lock,
1026 * so I think not; but would rather try again later than BUG()
1029 page_cache_release(page);
1031 return (found < 0) ? found : 0;
1035 * Move the page from the page cache to the swap cache.
1037 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1039 struct shmem_inode_info *info;
1040 swp_entry_t *entry, swap;
1041 struct address_space *mapping;
1042 unsigned long index;
1043 struct inode *inode;
1045 BUG_ON(!PageLocked(page));
1046 mapping = page->mapping;
1047 index = page->index;
1048 inode = mapping->host;
1049 info = SHMEM_I(inode);
1050 if (info->flags & VM_LOCKED)
1052 if (!total_swap_pages)
1056 * shmem_backing_dev_info's capabilities prevent regular writeback or
1057 * sync from ever calling shmem_writepage; but a stacking filesystem
1058 * may use the ->writepage of its underlying filesystem, in which case
1059 * tmpfs should write out to swap only in response to memory pressure,
1060 * and not for the writeback threads or sync. However, in those cases,
1061 * we do still want to check if there's a redundant swappage to be
1064 if (wbc->for_reclaim)
1065 swap = get_swap_page();
1069 spin_lock(&info->lock);
1070 if (index >= info->next_index) {
1071 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1074 entry = shmem_swp_entry(info, index, NULL);
1077 * The more uptodate page coming down from a stacked
1078 * writepage should replace our old swappage.
1080 free_swap_and_cache(*entry);
1081 shmem_swp_set(info, entry, 0);
1083 shmem_recalc_inode(inode);
1085 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1086 remove_from_page_cache(page);
1087 shmem_swp_set(info, entry, swap.val);
1088 shmem_swp_unmap(entry);
1089 if (list_empty(&info->swaplist))
1090 inode = igrab(inode);
1093 spin_unlock(&info->lock);
1094 swap_shmem_alloc(swap);
1095 BUG_ON(page_mapped(page));
1096 page_cache_release(page); /* pagecache ref */
1097 swap_writepage(page, wbc);
1099 mutex_lock(&shmem_swaplist_mutex);
1100 /* move instead of add in case we're racing */
1101 list_move_tail(&info->swaplist, &shmem_swaplist);
1102 mutex_unlock(&shmem_swaplist_mutex);
1108 shmem_swp_unmap(entry);
1110 spin_unlock(&info->lock);
1112 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1113 * clear SWAP_HAS_CACHE flag.
1115 swapcache_free(swap, NULL);
1117 set_page_dirty(page);
1118 if (wbc->for_reclaim)
1119 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1126 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1130 if (!mpol || mpol->mode == MPOL_DEFAULT)
1131 return; /* show nothing */
1133 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1135 seq_printf(seq, ",mpol=%s", buffer);
1138 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1140 struct mempolicy *mpol = NULL;
1142 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1143 mpol = sbinfo->mpol;
1145 spin_unlock(&sbinfo->stat_lock);
1149 #endif /* CONFIG_TMPFS */
1151 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1152 struct shmem_inode_info *info, unsigned long idx)
1154 struct mempolicy mpol, *spol;
1155 struct vm_area_struct pvma;
1158 spol = mpol_cond_copy(&mpol,
1159 mpol_shared_policy_lookup(&info->policy, idx));
1161 /* Create a pseudo vma that just contains the policy */
1163 pvma.vm_pgoff = idx;
1165 pvma.vm_policy = spol;
1166 page = swapin_readahead(entry, gfp, &pvma, 0);
1170 static struct page *shmem_alloc_page(gfp_t gfp,
1171 struct shmem_inode_info *info, unsigned long idx)
1173 struct vm_area_struct pvma;
1175 /* Create a pseudo vma that just contains the policy */
1177 pvma.vm_pgoff = idx;
1179 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1182 * alloc_page_vma() will drop the shared policy reference
1184 return alloc_page_vma(gfp, &pvma, 0);
1186 #else /* !CONFIG_NUMA */
1188 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1191 #endif /* CONFIG_TMPFS */
1193 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1194 struct shmem_inode_info *info, unsigned long idx)
1196 return swapin_readahead(entry, gfp, NULL, 0);
1199 static inline struct page *shmem_alloc_page(gfp_t gfp,
1200 struct shmem_inode_info *info, unsigned long idx)
1202 return alloc_page(gfp);
1204 #endif /* CONFIG_NUMA */
1206 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1207 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1214 * shmem_getpage - either get the page from swap or allocate a new one
1216 * If we allocate a new one we do not mark it dirty. That's up to the
1217 * vm. If we swap it in we mark it dirty since we also free the swap
1218 * entry since a page cannot live in both the swap and page cache
1220 static int shmem_getpage(struct inode *inode, unsigned long idx,
1221 struct page **pagep, enum sgp_type sgp, int *type)
1223 struct address_space *mapping = inode->i_mapping;
1224 struct shmem_inode_info *info = SHMEM_I(inode);
1225 struct shmem_sb_info *sbinfo;
1226 struct page *filepage = *pagep;
1227 struct page *swappage;
1233 if (idx >= SHMEM_MAX_INDEX)
1240 * Normally, filepage is NULL on entry, and either found
1241 * uptodate immediately, or allocated and zeroed, or read
1242 * in under swappage, which is then assigned to filepage.
1243 * But shmem_readpage (required for splice) passes in a locked
1244 * filepage, which may be found not uptodate by other callers
1245 * too, and may need to be copied from the swappage read in.
1249 filepage = find_lock_page(mapping, idx);
1250 if (filepage && PageUptodate(filepage))
1253 gfp = mapping_gfp_mask(mapping);
1256 * Try to preload while we can wait, to not make a habit of
1257 * draining atomic reserves; but don't latch on to this cpu.
1259 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1262 radix_tree_preload_end();
1265 spin_lock(&info->lock);
1266 shmem_recalc_inode(inode);
1267 entry = shmem_swp_alloc(info, idx, sgp);
1268 if (IS_ERR(entry)) {
1269 spin_unlock(&info->lock);
1270 error = PTR_ERR(entry);
1276 /* Look it up and read it in.. */
1277 swappage = lookup_swap_cache(swap);
1279 shmem_swp_unmap(entry);
1280 /* here we actually do the io */
1281 if (type && !(*type & VM_FAULT_MAJOR)) {
1282 __count_vm_event(PGMAJFAULT);
1283 *type |= VM_FAULT_MAJOR;
1285 spin_unlock(&info->lock);
1286 swappage = shmem_swapin(swap, gfp, info, idx);
1288 spin_lock(&info->lock);
1289 entry = shmem_swp_alloc(info, idx, sgp);
1291 error = PTR_ERR(entry);
1293 if (entry->val == swap.val)
1295 shmem_swp_unmap(entry);
1297 spin_unlock(&info->lock);
1302 wait_on_page_locked(swappage);
1303 page_cache_release(swappage);
1307 /* We have to do this with page locked to prevent races */
1308 if (!trylock_page(swappage)) {
1309 shmem_swp_unmap(entry);
1310 spin_unlock(&info->lock);
1311 wait_on_page_locked(swappage);
1312 page_cache_release(swappage);
1315 if (PageWriteback(swappage)) {
1316 shmem_swp_unmap(entry);
1317 spin_unlock(&info->lock);
1318 wait_on_page_writeback(swappage);
1319 unlock_page(swappage);
1320 page_cache_release(swappage);
1323 if (!PageUptodate(swappage)) {
1324 shmem_swp_unmap(entry);
1325 spin_unlock(&info->lock);
1326 unlock_page(swappage);
1327 page_cache_release(swappage);
1333 shmem_swp_set(info, entry, 0);
1334 shmem_swp_unmap(entry);
1335 delete_from_swap_cache(swappage);
1336 spin_unlock(&info->lock);
1337 copy_highpage(filepage, swappage);
1338 unlock_page(swappage);
1339 page_cache_release(swappage);
1340 flush_dcache_page(filepage);
1341 SetPageUptodate(filepage);
1342 set_page_dirty(filepage);
1344 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1345 idx, GFP_NOWAIT))) {
1346 info->flags |= SHMEM_PAGEIN;
1347 shmem_swp_set(info, entry, 0);
1348 shmem_swp_unmap(entry);
1349 delete_from_swap_cache(swappage);
1350 spin_unlock(&info->lock);
1351 filepage = swappage;
1352 set_page_dirty(filepage);
1355 shmem_swp_unmap(entry);
1356 spin_unlock(&info->lock);
1357 if (error == -ENOMEM) {
1359 * reclaim from proper memory cgroup and
1360 * call memcg's OOM if needed.
1362 error = mem_cgroup_shmem_charge_fallback(
1367 unlock_page(swappage);
1368 page_cache_release(swappage);
1372 unlock_page(swappage);
1373 page_cache_release(swappage);
1376 } else if (sgp == SGP_READ && !filepage) {
1377 shmem_swp_unmap(entry);
1378 filepage = find_get_page(mapping, idx);
1380 (!PageUptodate(filepage) || !trylock_page(filepage))) {
1381 spin_unlock(&info->lock);
1382 wait_on_page_locked(filepage);
1383 page_cache_release(filepage);
1387 spin_unlock(&info->lock);
1389 shmem_swp_unmap(entry);
1390 sbinfo = SHMEM_SB(inode->i_sb);
1391 if (sbinfo->max_blocks) {
1392 spin_lock(&sbinfo->stat_lock);
1393 if (sbinfo->free_blocks == 0 ||
1394 shmem_acct_block(info->flags)) {
1395 spin_unlock(&sbinfo->stat_lock);
1396 spin_unlock(&info->lock);
1400 sbinfo->free_blocks--;
1401 inode->i_blocks += BLOCKS_PER_PAGE;
1402 spin_unlock(&sbinfo->stat_lock);
1403 } else if (shmem_acct_block(info->flags)) {
1404 spin_unlock(&info->lock);
1412 spin_unlock(&info->lock);
1413 filepage = shmem_alloc_page(gfp, info, idx);
1415 shmem_unacct_blocks(info->flags, 1);
1416 shmem_free_blocks(inode, 1);
1420 SetPageSwapBacked(filepage);
1422 /* Precharge page while we can wait, compensate after */
1423 error = mem_cgroup_cache_charge(filepage, current->mm,
1426 page_cache_release(filepage);
1427 shmem_unacct_blocks(info->flags, 1);
1428 shmem_free_blocks(inode, 1);
1433 spin_lock(&info->lock);
1434 entry = shmem_swp_alloc(info, idx, sgp);
1436 error = PTR_ERR(entry);
1439 shmem_swp_unmap(entry);
1441 ret = error || swap.val;
1443 mem_cgroup_uncharge_cache_page(filepage);
1445 ret = add_to_page_cache_lru(filepage, mapping,
1448 * At add_to_page_cache_lru() failure, uncharge will
1449 * be done automatically.
1452 spin_unlock(&info->lock);
1453 page_cache_release(filepage);
1454 shmem_unacct_blocks(info->flags, 1);
1455 shmem_free_blocks(inode, 1);
1461 info->flags |= SHMEM_PAGEIN;
1465 spin_unlock(&info->lock);
1466 clear_highpage(filepage);
1467 flush_dcache_page(filepage);
1468 SetPageUptodate(filepage);
1469 if (sgp == SGP_DIRTY)
1470 set_page_dirty(filepage);
1477 if (*pagep != filepage) {
1478 unlock_page(filepage);
1479 page_cache_release(filepage);
1484 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1486 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1490 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1491 return VM_FAULT_SIGBUS;
1493 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1495 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1497 return ret | VM_FAULT_LOCKED;
1501 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1503 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1504 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1507 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1510 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1513 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1514 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1518 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1520 struct inode *inode = file->f_path.dentry->d_inode;
1521 struct shmem_inode_info *info = SHMEM_I(inode);
1522 int retval = -ENOMEM;
1524 spin_lock(&info->lock);
1525 if (lock && !(info->flags & VM_LOCKED)) {
1526 if (!user_shm_lock(inode->i_size, user))
1528 info->flags |= VM_LOCKED;
1529 mapping_set_unevictable(file->f_mapping);
1531 if (!lock && (info->flags & VM_LOCKED) && user) {
1532 user_shm_unlock(inode->i_size, user);
1533 info->flags &= ~VM_LOCKED;
1534 mapping_clear_unevictable(file->f_mapping);
1535 scan_mapping_unevictable_pages(file->f_mapping);
1540 spin_unlock(&info->lock);
1544 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1546 file_accessed(file);
1547 vma->vm_ops = &shmem_vm_ops;
1548 vma->vm_flags |= VM_CAN_NONLINEAR;
1552 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1553 int mode, dev_t dev, unsigned long flags)
1555 struct inode *inode;
1556 struct shmem_inode_info *info;
1557 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1559 if (shmem_reserve_inode(sb))
1562 inode = new_inode(sb);
1564 inode_init_owner(inode, dir, mode);
1565 inode->i_blocks = 0;
1566 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1567 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1568 inode->i_generation = get_seconds();
1569 info = SHMEM_I(inode);
1570 memset(info, 0, (char *)inode - (char *)info);
1571 spin_lock_init(&info->lock);
1572 info->flags = flags & VM_NORESERVE;
1573 INIT_LIST_HEAD(&info->swaplist);
1574 cache_no_acl(inode);
1576 switch (mode & S_IFMT) {
1578 inode->i_op = &shmem_special_inode_operations;
1579 init_special_inode(inode, mode, dev);
1582 inode->i_mapping->a_ops = &shmem_aops;
1583 inode->i_op = &shmem_inode_operations;
1584 inode->i_fop = &shmem_file_operations;
1585 mpol_shared_policy_init(&info->policy,
1586 shmem_get_sbmpol(sbinfo));
1590 /* Some things misbehave if size == 0 on a directory */
1591 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1592 inode->i_op = &shmem_dir_inode_operations;
1593 inode->i_fop = &simple_dir_operations;
1597 * Must not load anything in the rbtree,
1598 * mpol_free_shared_policy will not be called.
1600 mpol_shared_policy_init(&info->policy, NULL);
1604 shmem_free_inode(sb);
1609 static const struct inode_operations shmem_symlink_inode_operations;
1610 static const struct inode_operations shmem_symlink_inline_operations;
1613 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1614 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1615 * below the loop driver, in the generic fashion that many filesystems support.
1617 static int shmem_readpage(struct file *file, struct page *page)
1619 struct inode *inode = page->mapping->host;
1620 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1626 shmem_write_begin(struct file *file, struct address_space *mapping,
1627 loff_t pos, unsigned len, unsigned flags,
1628 struct page **pagep, void **fsdata)
1630 struct inode *inode = mapping->host;
1631 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1633 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1637 shmem_write_end(struct file *file, struct address_space *mapping,
1638 loff_t pos, unsigned len, unsigned copied,
1639 struct page *page, void *fsdata)
1641 struct inode *inode = mapping->host;
1643 if (pos + copied > inode->i_size)
1644 i_size_write(inode, pos + copied);
1646 set_page_dirty(page);
1648 page_cache_release(page);
1653 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1655 struct inode *inode = filp->f_path.dentry->d_inode;
1656 struct address_space *mapping = inode->i_mapping;
1657 unsigned long index, offset;
1658 enum sgp_type sgp = SGP_READ;
1661 * Might this read be for a stacking filesystem? Then when reading
1662 * holes of a sparse file, we actually need to allocate those pages,
1663 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1665 if (segment_eq(get_fs(), KERNEL_DS))
1668 index = *ppos >> PAGE_CACHE_SHIFT;
1669 offset = *ppos & ~PAGE_CACHE_MASK;
1672 struct page *page = NULL;
1673 unsigned long end_index, nr, ret;
1674 loff_t i_size = i_size_read(inode);
1676 end_index = i_size >> PAGE_CACHE_SHIFT;
1677 if (index > end_index)
1679 if (index == end_index) {
1680 nr = i_size & ~PAGE_CACHE_MASK;
1685 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1687 if (desc->error == -EINVAL)
1695 * We must evaluate after, since reads (unlike writes)
1696 * are called without i_mutex protection against truncate
1698 nr = PAGE_CACHE_SIZE;
1699 i_size = i_size_read(inode);
1700 end_index = i_size >> PAGE_CACHE_SHIFT;
1701 if (index == end_index) {
1702 nr = i_size & ~PAGE_CACHE_MASK;
1705 page_cache_release(page);
1713 * If users can be writing to this page using arbitrary
1714 * virtual addresses, take care about potential aliasing
1715 * before reading the page on the kernel side.
1717 if (mapping_writably_mapped(mapping))
1718 flush_dcache_page(page);
1720 * Mark the page accessed if we read the beginning.
1723 mark_page_accessed(page);
1725 page = ZERO_PAGE(0);
1726 page_cache_get(page);
1730 * Ok, we have the page, and it's up-to-date, so
1731 * now we can copy it to user space...
1733 * The actor routine returns how many bytes were actually used..
1734 * NOTE! This may not be the same as how much of a user buffer
1735 * we filled up (we may be padding etc), so we can only update
1736 * "pos" here (the actor routine has to update the user buffer
1737 * pointers and the remaining count).
1739 ret = actor(desc, page, offset, nr);
1741 index += offset >> PAGE_CACHE_SHIFT;
1742 offset &= ~PAGE_CACHE_MASK;
1744 page_cache_release(page);
1745 if (ret != nr || !desc->count)
1751 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1752 file_accessed(filp);
1755 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1756 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1758 struct file *filp = iocb->ki_filp;
1762 loff_t *ppos = &iocb->ki_pos;
1764 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1768 for (seg = 0; seg < nr_segs; seg++) {
1769 read_descriptor_t desc;
1772 desc.arg.buf = iov[seg].iov_base;
1773 desc.count = iov[seg].iov_len;
1774 if (desc.count == 0)
1777 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1778 retval += desc.written;
1780 retval = retval ?: desc.error;
1789 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1791 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1793 buf->f_type = TMPFS_MAGIC;
1794 buf->f_bsize = PAGE_CACHE_SIZE;
1795 buf->f_namelen = NAME_MAX;
1796 spin_lock(&sbinfo->stat_lock);
1797 if (sbinfo->max_blocks) {
1798 buf->f_blocks = sbinfo->max_blocks;
1799 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1801 if (sbinfo->max_inodes) {
1802 buf->f_files = sbinfo->max_inodes;
1803 buf->f_ffree = sbinfo->free_inodes;
1805 /* else leave those fields 0 like simple_statfs */
1806 spin_unlock(&sbinfo->stat_lock);
1811 * File creation. Allocate an inode, and we're done..
1814 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1816 struct inode *inode;
1817 int error = -ENOSPC;
1819 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1821 error = security_inode_init_security(inode, dir, NULL, NULL,
1824 if (error != -EOPNOTSUPP) {
1829 #ifdef CONFIG_TMPFS_POSIX_ACL
1830 error = generic_acl_init(inode, dir);
1838 dir->i_size += BOGO_DIRENT_SIZE;
1839 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1840 d_instantiate(dentry, inode);
1841 dget(dentry); /* Extra count - pin the dentry in core */
1846 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1850 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1856 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1857 struct nameidata *nd)
1859 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1865 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1867 struct inode *inode = old_dentry->d_inode;
1871 * No ordinary (disk based) filesystem counts links as inodes;
1872 * but each new link needs a new dentry, pinning lowmem, and
1873 * tmpfs dentries cannot be pruned until they are unlinked.
1875 ret = shmem_reserve_inode(inode->i_sb);
1879 dir->i_size += BOGO_DIRENT_SIZE;
1880 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1882 atomic_inc(&inode->i_count); /* New dentry reference */
1883 dget(dentry); /* Extra pinning count for the created dentry */
1884 d_instantiate(dentry, inode);
1889 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1891 struct inode *inode = dentry->d_inode;
1893 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1894 shmem_free_inode(inode->i_sb);
1896 dir->i_size -= BOGO_DIRENT_SIZE;
1897 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1899 dput(dentry); /* Undo the count from "create" - this does all the work */
1903 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1905 if (!simple_empty(dentry))
1908 drop_nlink(dentry->d_inode);
1910 return shmem_unlink(dir, dentry);
1914 * The VFS layer already does all the dentry stuff for rename,
1915 * we just have to decrement the usage count for the target if
1916 * it exists so that the VFS layer correctly free's it when it
1919 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1921 struct inode *inode = old_dentry->d_inode;
1922 int they_are_dirs = S_ISDIR(inode->i_mode);
1924 if (!simple_empty(new_dentry))
1927 if (new_dentry->d_inode) {
1928 (void) shmem_unlink(new_dir, new_dentry);
1930 drop_nlink(old_dir);
1931 } else if (they_are_dirs) {
1932 drop_nlink(old_dir);
1936 old_dir->i_size -= BOGO_DIRENT_SIZE;
1937 new_dir->i_size += BOGO_DIRENT_SIZE;
1938 old_dir->i_ctime = old_dir->i_mtime =
1939 new_dir->i_ctime = new_dir->i_mtime =
1940 inode->i_ctime = CURRENT_TIME;
1944 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1948 struct inode *inode;
1949 struct page *page = NULL;
1951 struct shmem_inode_info *info;
1953 len = strlen(symname) + 1;
1954 if (len > PAGE_CACHE_SIZE)
1955 return -ENAMETOOLONG;
1957 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1961 error = security_inode_init_security(inode, dir, NULL, NULL,
1964 if (error != -EOPNOTSUPP) {
1971 info = SHMEM_I(inode);
1972 inode->i_size = len-1;
1973 if (len <= (char *)inode - (char *)info) {
1975 memcpy(info, symname, len);
1976 inode->i_op = &shmem_symlink_inline_operations;
1978 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1983 inode->i_mapping->a_ops = &shmem_aops;
1984 inode->i_op = &shmem_symlink_inode_operations;
1985 kaddr = kmap_atomic(page, KM_USER0);
1986 memcpy(kaddr, symname, len);
1987 kunmap_atomic(kaddr, KM_USER0);
1988 set_page_dirty(page);
1990 page_cache_release(page);
1992 dir->i_size += BOGO_DIRENT_SIZE;
1993 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1994 d_instantiate(dentry, inode);
1999 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
2001 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
2005 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2007 struct page *page = NULL;
2008 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2009 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2015 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2017 if (!IS_ERR(nd_get_link(nd))) {
2018 struct page *page = cookie;
2020 mark_page_accessed(page);
2021 page_cache_release(page);
2025 static const struct inode_operations shmem_symlink_inline_operations = {
2026 .readlink = generic_readlink,
2027 .follow_link = shmem_follow_link_inline,
2030 static const struct inode_operations shmem_symlink_inode_operations = {
2031 .readlink = generic_readlink,
2032 .follow_link = shmem_follow_link,
2033 .put_link = shmem_put_link,
2036 #ifdef CONFIG_TMPFS_POSIX_ACL
2038 * Superblocks without xattr inode operations will get security.* xattr
2039 * support from the VFS "for free". As soon as we have any other xattrs
2040 * like ACLs, we also need to implement the security.* handlers at
2041 * filesystem level, though.
2044 static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
2045 size_t list_len, const char *name,
2046 size_t name_len, int handler_flags)
2048 return security_inode_listsecurity(dentry->d_inode, list, list_len);
2051 static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2052 void *buffer, size_t size, int handler_flags)
2054 if (strcmp(name, "") == 0)
2056 return xattr_getsecurity(dentry->d_inode, name, buffer, size);
2059 static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2060 const void *value, size_t size, int flags, int handler_flags)
2062 if (strcmp(name, "") == 0)
2064 return security_inode_setsecurity(dentry->d_inode, name, value,
2068 static const struct xattr_handler shmem_xattr_security_handler = {
2069 .prefix = XATTR_SECURITY_PREFIX,
2070 .list = shmem_xattr_security_list,
2071 .get = shmem_xattr_security_get,
2072 .set = shmem_xattr_security_set,
2075 static const struct xattr_handler *shmem_xattr_handlers[] = {
2076 &generic_acl_access_handler,
2077 &generic_acl_default_handler,
2078 &shmem_xattr_security_handler,
2083 static struct dentry *shmem_get_parent(struct dentry *child)
2085 return ERR_PTR(-ESTALE);
2088 static int shmem_match(struct inode *ino, void *vfh)
2092 inum = (inum << 32) | fh[1];
2093 return ino->i_ino == inum && fh[0] == ino->i_generation;
2096 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2097 struct fid *fid, int fh_len, int fh_type)
2099 struct inode *inode;
2100 struct dentry *dentry = NULL;
2101 u64 inum = fid->raw[2];
2102 inum = (inum << 32) | fid->raw[1];
2107 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2108 shmem_match, fid->raw);
2110 dentry = d_find_alias(inode);
2117 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2120 struct inode *inode = dentry->d_inode;
2125 if (hlist_unhashed(&inode->i_hash)) {
2126 /* Unfortunately insert_inode_hash is not idempotent,
2127 * so as we hash inodes here rather than at creation
2128 * time, we need a lock to ensure we only try
2131 static DEFINE_SPINLOCK(lock);
2133 if (hlist_unhashed(&inode->i_hash))
2134 __insert_inode_hash(inode,
2135 inode->i_ino + inode->i_generation);
2139 fh[0] = inode->i_generation;
2140 fh[1] = inode->i_ino;
2141 fh[2] = ((__u64)inode->i_ino) >> 32;
2147 static const struct export_operations shmem_export_ops = {
2148 .get_parent = shmem_get_parent,
2149 .encode_fh = shmem_encode_fh,
2150 .fh_to_dentry = shmem_fh_to_dentry,
2153 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2156 char *this_char, *value, *rest;
2158 while (options != NULL) {
2159 this_char = options;
2162 * NUL-terminate this option: unfortunately,
2163 * mount options form a comma-separated list,
2164 * but mpol's nodelist may also contain commas.
2166 options = strchr(options, ',');
2167 if (options == NULL)
2170 if (!isdigit(*options)) {
2177 if ((value = strchr(this_char,'=')) != NULL) {
2181 "tmpfs: No value for mount option '%s'\n",
2186 if (!strcmp(this_char,"size")) {
2187 unsigned long long size;
2188 size = memparse(value,&rest);
2190 size <<= PAGE_SHIFT;
2191 size *= totalram_pages;
2197 sbinfo->max_blocks =
2198 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2199 } else if (!strcmp(this_char,"nr_blocks")) {
2200 sbinfo->max_blocks = memparse(value, &rest);
2203 } else if (!strcmp(this_char,"nr_inodes")) {
2204 sbinfo->max_inodes = memparse(value, &rest);
2207 } else if (!strcmp(this_char,"mode")) {
2210 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2213 } else if (!strcmp(this_char,"uid")) {
2216 sbinfo->uid = simple_strtoul(value, &rest, 0);
2219 } else if (!strcmp(this_char,"gid")) {
2222 sbinfo->gid = simple_strtoul(value, &rest, 0);
2225 } else if (!strcmp(this_char,"mpol")) {
2226 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2229 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2237 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2243 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2246 struct shmem_sb_info config = *sbinfo;
2247 unsigned long blocks;
2248 unsigned long inodes;
2249 int error = -EINVAL;
2251 if (shmem_parse_options(data, &config, true))
2254 spin_lock(&sbinfo->stat_lock);
2255 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2256 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2257 if (config.max_blocks < blocks)
2259 if (config.max_inodes < inodes)
2262 * Those tests also disallow limited->unlimited while any are in
2263 * use, so i_blocks will always be zero when max_blocks is zero;
2264 * but we must separately disallow unlimited->limited, because
2265 * in that case we have no record of how much is already in use.
2267 if (config.max_blocks && !sbinfo->max_blocks)
2269 if (config.max_inodes && !sbinfo->max_inodes)
2273 sbinfo->max_blocks = config.max_blocks;
2274 sbinfo->free_blocks = config.max_blocks - blocks;
2275 sbinfo->max_inodes = config.max_inodes;
2276 sbinfo->free_inodes = config.max_inodes - inodes;
2278 mpol_put(sbinfo->mpol);
2279 sbinfo->mpol = config.mpol; /* transfers initial ref */
2281 spin_unlock(&sbinfo->stat_lock);
2285 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2287 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2289 if (sbinfo->max_blocks != shmem_default_max_blocks())
2290 seq_printf(seq, ",size=%luk",
2291 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2292 if (sbinfo->max_inodes != shmem_default_max_inodes())
2293 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2294 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2295 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2296 if (sbinfo->uid != 0)
2297 seq_printf(seq, ",uid=%u", sbinfo->uid);
2298 if (sbinfo->gid != 0)
2299 seq_printf(seq, ",gid=%u", sbinfo->gid);
2300 shmem_show_mpol(seq, sbinfo->mpol);
2303 #endif /* CONFIG_TMPFS */
2305 static void shmem_put_super(struct super_block *sb)
2307 kfree(sb->s_fs_info);
2308 sb->s_fs_info = NULL;
2311 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2313 struct inode *inode;
2314 struct dentry *root;
2315 struct shmem_sb_info *sbinfo;
2318 /* Round up to L1_CACHE_BYTES to resist false sharing */
2319 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2320 L1_CACHE_BYTES), GFP_KERNEL);
2324 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2325 sbinfo->uid = current_fsuid();
2326 sbinfo->gid = current_fsgid();
2327 sb->s_fs_info = sbinfo;
2331 * Per default we only allow half of the physical ram per
2332 * tmpfs instance, limiting inodes to one per page of lowmem;
2333 * but the internal instance is left unlimited.
2335 if (!(sb->s_flags & MS_NOUSER)) {
2336 sbinfo->max_blocks = shmem_default_max_blocks();
2337 sbinfo->max_inodes = shmem_default_max_inodes();
2338 if (shmem_parse_options(data, sbinfo, false)) {
2343 sb->s_export_op = &shmem_export_ops;
2345 sb->s_flags |= MS_NOUSER;
2348 spin_lock_init(&sbinfo->stat_lock);
2349 sbinfo->free_blocks = sbinfo->max_blocks;
2350 sbinfo->free_inodes = sbinfo->max_inodes;
2352 sb->s_maxbytes = SHMEM_MAX_BYTES;
2353 sb->s_blocksize = PAGE_CACHE_SIZE;
2354 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2355 sb->s_magic = TMPFS_MAGIC;
2356 sb->s_op = &shmem_ops;
2357 sb->s_time_gran = 1;
2358 #ifdef CONFIG_TMPFS_POSIX_ACL
2359 sb->s_xattr = shmem_xattr_handlers;
2360 sb->s_flags |= MS_POSIXACL;
2363 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2366 inode->i_uid = sbinfo->uid;
2367 inode->i_gid = sbinfo->gid;
2368 root = d_alloc_root(inode);
2377 shmem_put_super(sb);
2381 static struct kmem_cache *shmem_inode_cachep;
2383 static struct inode *shmem_alloc_inode(struct super_block *sb)
2385 struct shmem_inode_info *p;
2386 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2389 return &p->vfs_inode;
2392 static void shmem_destroy_inode(struct inode *inode)
2394 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2395 /* only struct inode is valid if it's an inline symlink */
2396 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2398 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2401 static void init_once(void *foo)
2403 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2405 inode_init_once(&p->vfs_inode);
2408 static int init_inodecache(void)
2410 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2411 sizeof(struct shmem_inode_info),
2412 0, SLAB_PANIC, init_once);
2416 static void destroy_inodecache(void)
2418 kmem_cache_destroy(shmem_inode_cachep);
2421 static const struct address_space_operations shmem_aops = {
2422 .writepage = shmem_writepage,
2423 .set_page_dirty = __set_page_dirty_no_writeback,
2425 .readpage = shmem_readpage,
2426 .write_begin = shmem_write_begin,
2427 .write_end = shmem_write_end,
2429 .migratepage = migrate_page,
2430 .error_remove_page = generic_error_remove_page,
2433 static const struct file_operations shmem_file_operations = {
2436 .llseek = generic_file_llseek,
2437 .read = do_sync_read,
2438 .write = do_sync_write,
2439 .aio_read = shmem_file_aio_read,
2440 .aio_write = generic_file_aio_write,
2441 .fsync = noop_fsync,
2442 .splice_read = generic_file_splice_read,
2443 .splice_write = generic_file_splice_write,
2447 static const struct inode_operations shmem_inode_operations = {
2448 .setattr = shmem_notify_change,
2449 .truncate_range = shmem_truncate_range,
2450 #ifdef CONFIG_TMPFS_POSIX_ACL
2451 .setxattr = generic_setxattr,
2452 .getxattr = generic_getxattr,
2453 .listxattr = generic_listxattr,
2454 .removexattr = generic_removexattr,
2455 .check_acl = generic_check_acl,
2460 static const struct inode_operations shmem_dir_inode_operations = {
2462 .create = shmem_create,
2463 .lookup = simple_lookup,
2465 .unlink = shmem_unlink,
2466 .symlink = shmem_symlink,
2467 .mkdir = shmem_mkdir,
2468 .rmdir = shmem_rmdir,
2469 .mknod = shmem_mknod,
2470 .rename = shmem_rename,
2472 #ifdef CONFIG_TMPFS_POSIX_ACL
2473 .setattr = shmem_notify_change,
2474 .setxattr = generic_setxattr,
2475 .getxattr = generic_getxattr,
2476 .listxattr = generic_listxattr,
2477 .removexattr = generic_removexattr,
2478 .check_acl = generic_check_acl,
2482 static const struct inode_operations shmem_special_inode_operations = {
2483 #ifdef CONFIG_TMPFS_POSIX_ACL
2484 .setattr = shmem_notify_change,
2485 .setxattr = generic_setxattr,
2486 .getxattr = generic_getxattr,
2487 .listxattr = generic_listxattr,
2488 .removexattr = generic_removexattr,
2489 .check_acl = generic_check_acl,
2493 static const struct super_operations shmem_ops = {
2494 .alloc_inode = shmem_alloc_inode,
2495 .destroy_inode = shmem_destroy_inode,
2497 .statfs = shmem_statfs,
2498 .remount_fs = shmem_remount_fs,
2499 .show_options = shmem_show_options,
2501 .delete_inode = shmem_delete_inode,
2502 .drop_inode = generic_delete_inode,
2503 .put_super = shmem_put_super,
2506 static const struct vm_operations_struct shmem_vm_ops = {
2507 .fault = shmem_fault,
2509 .set_policy = shmem_set_policy,
2510 .get_policy = shmem_get_policy,
2515 static int shmem_get_sb(struct file_system_type *fs_type,
2516 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2518 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2521 static struct file_system_type tmpfs_fs_type = {
2522 .owner = THIS_MODULE,
2524 .get_sb = shmem_get_sb,
2525 .kill_sb = kill_litter_super,
2528 int __init init_tmpfs(void)
2532 error = bdi_init(&shmem_backing_dev_info);
2536 error = init_inodecache();
2540 error = register_filesystem(&tmpfs_fs_type);
2542 printk(KERN_ERR "Could not register tmpfs\n");
2546 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2547 tmpfs_fs_type.name, NULL);
2548 if (IS_ERR(shm_mnt)) {
2549 error = PTR_ERR(shm_mnt);
2550 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2556 unregister_filesystem(&tmpfs_fs_type);
2558 destroy_inodecache();
2560 bdi_destroy(&shmem_backing_dev_info);
2562 shm_mnt = ERR_PTR(error);
2566 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2568 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2569 * @inode: the inode to be searched
2570 * @pgoff: the offset to be searched
2571 * @pagep: the pointer for the found page to be stored
2572 * @ent: the pointer for the found swap entry to be stored
2574 * If a page is found, refcount of it is incremented. Callers should handle
2577 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2578 struct page **pagep, swp_entry_t *ent)
2580 swp_entry_t entry = { .val = 0 }, *ptr;
2581 struct page *page = NULL;
2582 struct shmem_inode_info *info = SHMEM_I(inode);
2584 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2587 spin_lock(&info->lock);
2588 ptr = shmem_swp_entry(info, pgoff, NULL);
2590 if (ptr && ptr->val) {
2591 entry.val = ptr->val;
2592 page = find_get_page(&swapper_space, entry.val);
2595 page = find_get_page(inode->i_mapping, pgoff);
2597 shmem_swp_unmap(ptr);
2598 spin_unlock(&info->lock);
2605 #else /* !CONFIG_SHMEM */
2608 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2610 * This is intended for small system where the benefits of the full
2611 * shmem code (swap-backed and resource-limited) are outweighed by
2612 * their complexity. On systems without swap this code should be
2613 * effectively equivalent, but much lighter weight.
2616 #include <linux/ramfs.h>
2618 static struct file_system_type tmpfs_fs_type = {
2620 .get_sb = ramfs_get_sb,
2621 .kill_sb = kill_litter_super,
2624 int __init init_tmpfs(void)
2626 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2628 shm_mnt = kern_mount(&tmpfs_fs_type);
2629 BUG_ON(IS_ERR(shm_mnt));
2634 int shmem_unuse(swp_entry_t entry, struct page *page)
2639 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2644 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2646 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2647 * @inode: the inode to be searched
2648 * @pgoff: the offset to be searched
2649 * @pagep: the pointer for the found page to be stored
2650 * @ent: the pointer for the found swap entry to be stored
2652 * If a page is found, refcount of it is incremented. Callers should handle
2655 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2656 struct page **pagep, swp_entry_t *ent)
2658 struct page *page = NULL;
2660 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2662 page = find_get_page(inode->i_mapping, pgoff);
2665 *ent = (swp_entry_t){ .val = 0 };
2669 #define shmem_vm_ops generic_file_vm_ops
2670 #define shmem_file_operations ramfs_file_operations
2671 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2672 #define shmem_acct_size(flags, size) 0
2673 #define shmem_unacct_size(flags, size) do {} while (0)
2674 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
2676 #endif /* CONFIG_SHMEM */
2681 * shmem_file_setup - get an unlinked file living in tmpfs
2682 * @name: name for dentry (to be seen in /proc/<pid>/maps
2683 * @size: size to be set for the file
2684 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2686 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2690 struct inode *inode;
2692 struct dentry *root;
2695 if (IS_ERR(shm_mnt))
2696 return (void *)shm_mnt;
2698 if (size < 0 || size > SHMEM_MAX_BYTES)
2699 return ERR_PTR(-EINVAL);
2701 if (shmem_acct_size(flags, size))
2702 return ERR_PTR(-ENOMEM);
2706 this.len = strlen(name);
2707 this.hash = 0; /* will go */
2708 root = shm_mnt->mnt_root;
2709 path.dentry = d_alloc(root, &this);
2712 path.mnt = mntget(shm_mnt);
2715 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2719 d_instantiate(path.dentry, inode);
2720 inode->i_size = size;
2721 inode->i_nlink = 0; /* It is unlinked */
2723 error = ramfs_nommu_expand_for_mapping(inode, size);
2729 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2730 &shmem_file_operations);
2739 shmem_unacct_size(flags, size);
2740 return ERR_PTR(error);
2742 EXPORT_SYMBOL_GPL(shmem_file_setup);
2745 * shmem_zero_setup - setup a shared anonymous mapping
2746 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2748 int shmem_zero_setup(struct vm_area_struct *vma)
2751 loff_t size = vma->vm_end - vma->vm_start;
2753 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2755 return PTR_ERR(file);
2759 vma->vm_file = file;
2760 vma->vm_ops = &shmem_vm_ops;