2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 * This file is released under the GPL.
21 * This virtual memory filesystem is heavily based on the ramfs. It
22 * extends ramfs by the ability to use swap and honor resource limits
23 * which makes it a completely usable filesystem.
26 #include <linux/module.h>
27 #include <linux/init.h>
29 #include <linux/xattr.h>
30 #include <linux/exportfs.h>
31 #include <linux/generic_acl.h>
33 #include <linux/mman.h>
34 #include <linux/file.h>
35 #include <linux/swap.h>
36 #include <linux/pagemap.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/backing-dev.h>
40 #include <linux/shmem_fs.h>
41 #include <linux/mount.h>
42 #include <linux/writeback.h>
43 #include <linux/vfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/security.h>
46 #include <linux/swapops.h>
47 #include <linux/mempolicy.h>
48 #include <linux/namei.h>
49 #include <linux/ctype.h>
50 #include <linux/migrate.h>
51 #include <linux/highmem.h>
53 #include <asm/uaccess.h>
54 #include <asm/div64.h>
55 #include <asm/pgtable.h>
57 /* This magic number is used in glibc for posix shared memory */
58 #define TMPFS_MAGIC 0x01021994
60 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
61 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
62 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
64 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
65 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
67 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
69 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
70 #define SHMEM_PAGEIN VM_READ
71 #define SHMEM_TRUNCATE VM_WRITE
73 /* Definition to limit shmem_truncate's steps between cond_rescheds */
74 #define LATENCY_LIMIT 64
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
79 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
81 SGP_READ, /* don't exceed i_size, don't allocate page */
82 SGP_CACHE, /* don't exceed i_size, may allocate page */
83 SGP_WRITE, /* may exceed i_size, may allocate page */
86 static int shmem_getpage(struct inode *inode, unsigned long idx,
87 struct page **pagep, enum sgp_type sgp, int *type);
89 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
92 * The above definition of ENTRIES_PER_PAGE, and the use of
93 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
94 * might be reconsidered if it ever diverges from PAGE_SIZE.
96 * Mobility flags are masked out as swap vectors cannot move
98 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
99 PAGE_CACHE_SHIFT-PAGE_SHIFT);
102 static inline void shmem_dir_free(struct page *page)
104 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
107 static struct page **shmem_dir_map(struct page *page)
109 return (struct page **)kmap_atomic(page, KM_USER0);
112 static inline void shmem_dir_unmap(struct page **dir)
114 kunmap_atomic(dir, KM_USER0);
117 static swp_entry_t *shmem_swp_map(struct page *page)
119 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
122 static inline void shmem_swp_balance_unmap(void)
125 * When passing a pointer to an i_direct entry, to code which
126 * also handles indirect entries and so will shmem_swp_unmap,
127 * we must arrange for the preempt count to remain in balance.
128 * What kmap_atomic of a lowmem page does depends on config
129 * and architecture, so pretend to kmap_atomic some lowmem page.
131 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
134 static inline void shmem_swp_unmap(swp_entry_t *entry)
136 kunmap_atomic(entry, KM_USER1);
139 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
141 return sb->s_fs_info;
145 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
146 * for shared memory and for shared anonymous (/dev/zero) mappings
147 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
148 * consistent with the pre-accounting of private mappings ...
150 static inline int shmem_acct_size(unsigned long flags, loff_t size)
152 return (flags & VM_ACCOUNT)?
153 security_vm_enough_memory(VM_ACCT(size)): 0;
156 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
158 if (flags & VM_ACCOUNT)
159 vm_unacct_memory(VM_ACCT(size));
163 * ... whereas tmpfs objects are accounted incrementally as
164 * pages are allocated, in order to allow huge sparse files.
165 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
166 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
168 static inline int shmem_acct_block(unsigned long flags)
170 return (flags & VM_ACCOUNT)?
171 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
174 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
176 if (!(flags & VM_ACCOUNT))
177 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
180 static const struct super_operations shmem_ops;
181 static const struct address_space_operations shmem_aops;
182 static const struct file_operations shmem_file_operations;
183 static const struct inode_operations shmem_inode_operations;
184 static const struct inode_operations shmem_dir_inode_operations;
185 static const struct inode_operations shmem_special_inode_operations;
186 static struct vm_operations_struct shmem_vm_ops;
188 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
189 .ra_pages = 0, /* No readahead */
190 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
191 .unplug_io_fn = default_unplug_io_fn,
194 static LIST_HEAD(shmem_swaplist);
195 static DEFINE_SPINLOCK(shmem_swaplist_lock);
197 static void shmem_free_blocks(struct inode *inode, long pages)
199 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
200 if (sbinfo->max_blocks) {
201 spin_lock(&sbinfo->stat_lock);
202 sbinfo->free_blocks += pages;
203 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
204 spin_unlock(&sbinfo->stat_lock);
209 * shmem_recalc_inode - recalculate the size of an inode
211 * @inode: inode to recalc
213 * We have to calculate the free blocks since the mm can drop
214 * undirtied hole pages behind our back.
216 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
217 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
219 * It has to be called with the spinlock held.
221 static void shmem_recalc_inode(struct inode *inode)
223 struct shmem_inode_info *info = SHMEM_I(inode);
226 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
228 info->alloced -= freed;
229 shmem_unacct_blocks(info->flags, freed);
230 shmem_free_blocks(inode, freed);
235 * shmem_swp_entry - find the swap vector position in the info structure
237 * @info: info structure for the inode
238 * @index: index of the page to find
239 * @page: optional page to add to the structure. Has to be preset to
242 * If there is no space allocated yet it will return NULL when
243 * page is NULL, else it will use the page for the needed block,
244 * setting it to NULL on return to indicate that it has been used.
246 * The swap vector is organized the following way:
248 * There are SHMEM_NR_DIRECT entries directly stored in the
249 * shmem_inode_info structure. So small files do not need an addional
252 * For pages with index > SHMEM_NR_DIRECT there is the pointer
253 * i_indirect which points to a page which holds in the first half
254 * doubly indirect blocks, in the second half triple indirect blocks:
256 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
257 * following layout (for SHMEM_NR_DIRECT == 16):
259 * i_indirect -> dir --> 16-19
272 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
274 unsigned long offset;
278 if (index < SHMEM_NR_DIRECT) {
279 shmem_swp_balance_unmap();
280 return info->i_direct+index;
282 if (!info->i_indirect) {
284 info->i_indirect = *page;
287 return NULL; /* need another page */
290 index -= SHMEM_NR_DIRECT;
291 offset = index % ENTRIES_PER_PAGE;
292 index /= ENTRIES_PER_PAGE;
293 dir = shmem_dir_map(info->i_indirect);
295 if (index >= ENTRIES_PER_PAGE/2) {
296 index -= ENTRIES_PER_PAGE/2;
297 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
298 index %= ENTRIES_PER_PAGE;
305 shmem_dir_unmap(dir);
306 return NULL; /* need another page */
308 shmem_dir_unmap(dir);
309 dir = shmem_dir_map(subdir);
315 if (!page || !(subdir = *page)) {
316 shmem_dir_unmap(dir);
317 return NULL; /* need a page */
322 shmem_dir_unmap(dir);
323 return shmem_swp_map(subdir) + offset;
326 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
328 long incdec = value? 1: -1;
331 info->swapped += incdec;
332 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
333 struct page *page = kmap_atomic_to_page(entry);
334 set_page_private(page, page_private(page) + incdec);
339 * shmem_swp_alloc - get the position of the swap entry for the page.
340 * If it does not exist allocate the entry.
342 * @info: info structure for the inode
343 * @index: index of the page to find
344 * @sgp: check and recheck i_size? skip allocation?
346 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
348 struct inode *inode = &info->vfs_inode;
349 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
350 struct page *page = NULL;
353 if (sgp != SGP_WRITE &&
354 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
355 return ERR_PTR(-EINVAL);
357 while (!(entry = shmem_swp_entry(info, index, &page))) {
359 return shmem_swp_map(ZERO_PAGE(0));
361 * Test free_blocks against 1 not 0, since we have 1 data
362 * page (and perhaps indirect index pages) yet to allocate:
363 * a waste to allocate index if we cannot allocate data.
365 if (sbinfo->max_blocks) {
366 spin_lock(&sbinfo->stat_lock);
367 if (sbinfo->free_blocks <= 1) {
368 spin_unlock(&sbinfo->stat_lock);
369 return ERR_PTR(-ENOSPC);
371 sbinfo->free_blocks--;
372 inode->i_blocks += BLOCKS_PER_PAGE;
373 spin_unlock(&sbinfo->stat_lock);
376 spin_unlock(&info->lock);
377 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
379 set_page_private(page, 0);
380 spin_lock(&info->lock);
383 shmem_free_blocks(inode, 1);
384 return ERR_PTR(-ENOMEM);
386 if (sgp != SGP_WRITE &&
387 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
388 entry = ERR_PTR(-EINVAL);
391 if (info->next_index <= index)
392 info->next_index = index + 1;
395 /* another task gave its page, or truncated the file */
396 shmem_free_blocks(inode, 1);
397 shmem_dir_free(page);
399 if (info->next_index <= index && !IS_ERR(entry))
400 info->next_index = index + 1;
405 * shmem_free_swp - free some swap entries in a directory
407 * @dir: pointer to the directory
408 * @edir: pointer after last entry of the directory
409 * @punch_lock: pointer to spinlock when needed for the holepunch case
411 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
412 spinlock_t *punch_lock)
414 spinlock_t *punch_unlock = NULL;
418 for (ptr = dir; ptr < edir; ptr++) {
420 if (unlikely(punch_lock)) {
421 punch_unlock = punch_lock;
423 spin_lock(punch_unlock);
427 free_swap_and_cache(*ptr);
428 *ptr = (swp_entry_t){0};
433 spin_unlock(punch_unlock);
437 static int shmem_map_and_free_swp(struct page *subdir, int offset,
438 int limit, struct page ***dir, spinlock_t *punch_lock)
443 ptr = shmem_swp_map(subdir);
444 for (; offset < limit; offset += LATENCY_LIMIT) {
445 int size = limit - offset;
446 if (size > LATENCY_LIMIT)
447 size = LATENCY_LIMIT;
448 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
450 if (need_resched()) {
451 shmem_swp_unmap(ptr);
453 shmem_dir_unmap(*dir);
457 ptr = shmem_swp_map(subdir);
460 shmem_swp_unmap(ptr);
464 static void shmem_free_pages(struct list_head *next)
470 page = container_of(next, struct page, lru);
472 shmem_dir_free(page);
474 if (freed >= LATENCY_LIMIT) {
481 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
483 struct shmem_inode_info *info = SHMEM_I(inode);
488 unsigned long diroff;
494 LIST_HEAD(pages_to_free);
495 long nr_pages_to_free = 0;
496 long nr_swaps_freed = 0;
500 spinlock_t *needs_lock;
501 spinlock_t *punch_lock;
502 unsigned long upper_limit;
504 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
505 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
506 if (idx >= info->next_index)
509 spin_lock(&info->lock);
510 info->flags |= SHMEM_TRUNCATE;
511 if (likely(end == (loff_t) -1)) {
512 limit = info->next_index;
513 upper_limit = SHMEM_MAX_INDEX;
514 info->next_index = idx;
518 if (end + 1 >= inode->i_size) { /* we may free a little more */
519 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
521 upper_limit = SHMEM_MAX_INDEX;
523 limit = (end + 1) >> PAGE_CACHE_SHIFT;
526 needs_lock = &info->lock;
530 topdir = info->i_indirect;
531 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
532 info->i_indirect = NULL;
534 list_add(&topdir->lru, &pages_to_free);
536 spin_unlock(&info->lock);
538 if (info->swapped && idx < SHMEM_NR_DIRECT) {
539 ptr = info->i_direct;
541 if (size > SHMEM_NR_DIRECT)
542 size = SHMEM_NR_DIRECT;
543 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
547 * If there are no indirect blocks or we are punching a hole
548 * below indirect blocks, nothing to be done.
550 if (!topdir || limit <= SHMEM_NR_DIRECT)
554 * The truncation case has already dropped info->lock, and we're safe
555 * because i_size and next_index have already been lowered, preventing
556 * access beyond. But in the punch_hole case, we still need to take
557 * the lock when updating the swap directory, because there might be
558 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
559 * shmem_writepage. However, whenever we find we can remove a whole
560 * directory page (not at the misaligned start or end of the range),
561 * we first NULLify its pointer in the level above, and then have no
562 * need to take the lock when updating its contents: needs_lock and
563 * punch_lock (either pointing to info->lock or NULL) manage this.
566 upper_limit -= SHMEM_NR_DIRECT;
567 limit -= SHMEM_NR_DIRECT;
568 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
569 offset = idx % ENTRIES_PER_PAGE;
572 dir = shmem_dir_map(topdir);
573 stage = ENTRIES_PER_PAGEPAGE/2;
574 if (idx < ENTRIES_PER_PAGEPAGE/2) {
576 diroff = idx/ENTRIES_PER_PAGE;
578 dir += ENTRIES_PER_PAGE/2;
579 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
581 stage += ENTRIES_PER_PAGEPAGE;
584 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
585 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
586 if (!diroff && !offset && upper_limit >= stage) {
588 spin_lock(needs_lock);
590 spin_unlock(needs_lock);
595 list_add(&middir->lru, &pages_to_free);
597 shmem_dir_unmap(dir);
598 dir = shmem_dir_map(middir);
606 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
607 if (unlikely(idx == stage)) {
608 shmem_dir_unmap(dir);
609 dir = shmem_dir_map(topdir) +
610 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
613 idx += ENTRIES_PER_PAGEPAGE;
617 stage = idx + ENTRIES_PER_PAGEPAGE;
620 needs_lock = &info->lock;
621 if (upper_limit >= stage) {
623 spin_lock(needs_lock);
625 spin_unlock(needs_lock);
630 list_add(&middir->lru, &pages_to_free);
632 shmem_dir_unmap(dir);
634 dir = shmem_dir_map(middir);
637 punch_lock = needs_lock;
638 subdir = dir[diroff];
639 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
641 spin_lock(needs_lock);
643 spin_unlock(needs_lock);
648 list_add(&subdir->lru, &pages_to_free);
650 if (subdir && page_private(subdir) /* has swap entries */) {
652 if (size > ENTRIES_PER_PAGE)
653 size = ENTRIES_PER_PAGE;
654 freed = shmem_map_and_free_swp(subdir,
655 offset, size, &dir, punch_lock);
657 dir = shmem_dir_map(middir);
658 nr_swaps_freed += freed;
659 if (offset || punch_lock) {
660 spin_lock(&info->lock);
661 set_page_private(subdir,
662 page_private(subdir) - freed);
663 spin_unlock(&info->lock);
665 BUG_ON(page_private(subdir) != freed);
670 shmem_dir_unmap(dir);
672 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
674 * Call truncate_inode_pages again: racing shmem_unuse_inode
675 * may have swizzled a page in from swap since vmtruncate or
676 * generic_delete_inode did it, before we lowered next_index.
677 * Also, though shmem_getpage checks i_size before adding to
678 * cache, no recheck after: so fix the narrow window there too.
680 * Recalling truncate_inode_pages_range and unmap_mapping_range
681 * every time for punch_hole (which never got a chance to clear
682 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
683 * yet hardly ever necessary: try to optimize them out later.
685 truncate_inode_pages_range(inode->i_mapping, start, end);
687 unmap_mapping_range(inode->i_mapping, start,
691 spin_lock(&info->lock);
692 info->flags &= ~SHMEM_TRUNCATE;
693 info->swapped -= nr_swaps_freed;
694 if (nr_pages_to_free)
695 shmem_free_blocks(inode, nr_pages_to_free);
696 shmem_recalc_inode(inode);
697 spin_unlock(&info->lock);
700 * Empty swap vector directory pages to be freed?
702 if (!list_empty(&pages_to_free)) {
703 pages_to_free.prev->next = NULL;
704 shmem_free_pages(pages_to_free.next);
708 static void shmem_truncate(struct inode *inode)
710 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
713 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
715 struct inode *inode = dentry->d_inode;
716 struct page *page = NULL;
719 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
720 if (attr->ia_size < inode->i_size) {
722 * If truncating down to a partial page, then
723 * if that page is already allocated, hold it
724 * in memory until the truncation is over, so
725 * truncate_partial_page cannnot miss it were
726 * it assigned to swap.
728 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
729 (void) shmem_getpage(inode,
730 attr->ia_size>>PAGE_CACHE_SHIFT,
731 &page, SGP_READ, NULL);
736 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
737 * detect if any pages might have been added to cache
738 * after truncate_inode_pages. But we needn't bother
739 * if it's being fully truncated to zero-length: the
740 * nrpages check is efficient enough in that case.
743 struct shmem_inode_info *info = SHMEM_I(inode);
744 spin_lock(&info->lock);
745 info->flags &= ~SHMEM_PAGEIN;
746 spin_unlock(&info->lock);
751 error = inode_change_ok(inode, attr);
753 error = inode_setattr(inode, attr);
754 #ifdef CONFIG_TMPFS_POSIX_ACL
755 if (!error && (attr->ia_valid & ATTR_MODE))
756 error = generic_acl_chmod(inode, &shmem_acl_ops);
759 page_cache_release(page);
763 static void shmem_delete_inode(struct inode *inode)
765 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
766 struct shmem_inode_info *info = SHMEM_I(inode);
768 if (inode->i_op->truncate == shmem_truncate) {
769 truncate_inode_pages(inode->i_mapping, 0);
770 shmem_unacct_size(info->flags, inode->i_size);
772 shmem_truncate(inode);
773 if (!list_empty(&info->swaplist)) {
774 spin_lock(&shmem_swaplist_lock);
775 list_del_init(&info->swaplist);
776 spin_unlock(&shmem_swaplist_lock);
779 BUG_ON(inode->i_blocks);
780 if (sbinfo->max_inodes) {
781 spin_lock(&sbinfo->stat_lock);
782 sbinfo->free_inodes++;
783 spin_unlock(&sbinfo->stat_lock);
788 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
792 for (ptr = dir; ptr < edir; ptr++) {
793 if (ptr->val == entry.val)
799 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
812 ptr = info->i_direct;
813 spin_lock(&info->lock);
814 limit = info->next_index;
816 if (size > SHMEM_NR_DIRECT)
817 size = SHMEM_NR_DIRECT;
818 offset = shmem_find_swp(entry, ptr, ptr+size);
820 shmem_swp_balance_unmap();
823 if (!info->i_indirect)
826 dir = shmem_dir_map(info->i_indirect);
827 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
829 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
830 if (unlikely(idx == stage)) {
831 shmem_dir_unmap(dir-1);
832 dir = shmem_dir_map(info->i_indirect) +
833 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
836 idx += ENTRIES_PER_PAGEPAGE;
840 stage = idx + ENTRIES_PER_PAGEPAGE;
842 shmem_dir_unmap(dir);
843 dir = shmem_dir_map(subdir);
846 if (subdir && page_private(subdir)) {
847 ptr = shmem_swp_map(subdir);
849 if (size > ENTRIES_PER_PAGE)
850 size = ENTRIES_PER_PAGE;
851 offset = shmem_find_swp(entry, ptr, ptr+size);
853 shmem_dir_unmap(dir);
856 shmem_swp_unmap(ptr);
860 shmem_dir_unmap(dir-1);
862 spin_unlock(&info->lock);
866 inode = &info->vfs_inode;
867 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
868 info->flags |= SHMEM_PAGEIN;
869 shmem_swp_set(info, ptr + offset, 0);
871 shmem_swp_unmap(ptr);
872 spin_unlock(&info->lock);
874 * Decrement swap count even when the entry is left behind:
875 * try_to_unuse will skip over mms, then reincrement count.
882 * shmem_unuse() search for an eventually swapped out shmem page.
884 int shmem_unuse(swp_entry_t entry, struct page *page)
886 struct list_head *p, *next;
887 struct shmem_inode_info *info;
890 spin_lock(&shmem_swaplist_lock);
891 list_for_each_safe(p, next, &shmem_swaplist) {
892 info = list_entry(p, struct shmem_inode_info, swaplist);
894 list_del_init(&info->swaplist);
895 else if (shmem_unuse_inode(info, entry, page)) {
896 /* move head to start search for next from here */
897 list_move_tail(&shmem_swaplist, &info->swaplist);
902 spin_unlock(&shmem_swaplist_lock);
907 * Move the page from the page cache to the swap cache.
909 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
911 struct shmem_inode_info *info;
912 swp_entry_t *entry, swap;
913 struct address_space *mapping;
917 BUG_ON(!PageLocked(page));
919 * shmem_backing_dev_info's capabilities prevent regular writeback or
920 * sync from ever calling shmem_writepage; but a stacking filesystem
921 * may use the ->writepage of its underlying filesystem, in which case
922 * we want to do nothing when that underlying filesystem is tmpfs
923 * (writing out to swap is useful as a response to memory pressure, but
924 * of no use to stabilize the data) - just redirty the page, unlock it
925 * and claim success in this case. AOP_WRITEPAGE_ACTIVATE, and the
926 * page_mapped check below, must be avoided unless we're in reclaim.
928 if (!wbc->for_reclaim) {
929 set_page_dirty(page);
933 BUG_ON(page_mapped(page));
935 mapping = page->mapping;
937 inode = mapping->host;
938 info = SHMEM_I(inode);
939 if (info->flags & VM_LOCKED)
941 swap = get_swap_page();
945 spin_lock(&info->lock);
946 shmem_recalc_inode(inode);
947 if (index >= info->next_index) {
948 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
951 entry = shmem_swp_entry(info, index, NULL);
955 if (move_to_swap_cache(page, swap) == 0) {
956 shmem_swp_set(info, entry, swap.val);
957 shmem_swp_unmap(entry);
958 spin_unlock(&info->lock);
959 if (list_empty(&info->swaplist)) {
960 spin_lock(&shmem_swaplist_lock);
961 /* move instead of add in case we're racing */
962 list_move_tail(&info->swaplist, &shmem_swaplist);
963 spin_unlock(&shmem_swaplist_lock);
969 shmem_swp_unmap(entry);
971 spin_unlock(&info->lock);
974 set_page_dirty(page);
975 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
979 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
981 char *nodelist = strchr(value, ':');
985 /* NUL-terminate policy string */
987 if (nodelist_parse(nodelist, *policy_nodes))
989 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
992 if (!strcmp(value, "default")) {
993 *policy = MPOL_DEFAULT;
994 /* Don't allow a nodelist */
997 } else if (!strcmp(value, "prefer")) {
998 *policy = MPOL_PREFERRED;
999 /* Insist on a nodelist of one node only */
1001 char *rest = nodelist;
1002 while (isdigit(*rest))
1007 } else if (!strcmp(value, "bind")) {
1008 *policy = MPOL_BIND;
1009 /* Insist on a nodelist */
1012 } else if (!strcmp(value, "interleave")) {
1013 *policy = MPOL_INTERLEAVE;
1015 * Default to online nodes with memory if no nodelist
1018 *policy_nodes = node_states[N_HIGH_MEMORY];
1022 /* Restore string for error message */
1028 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1029 struct shmem_inode_info *info, unsigned long idx)
1031 struct vm_area_struct pvma;
1034 /* Create a pseudo vma that just contains the policy */
1036 pvma.vm_pgoff = idx;
1038 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1039 page = swapin_readahead(entry, gfp, &pvma, 0);
1040 mpol_free(pvma.vm_policy);
1044 static struct page *shmem_alloc_page(gfp_t gfp,
1045 struct shmem_inode_info *info, unsigned long idx)
1047 struct vm_area_struct pvma;
1050 /* Create a pseudo vma that just contains the policy */
1052 pvma.vm_pgoff = idx;
1054 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1055 page = alloc_page_vma(gfp, &pvma, 0);
1056 mpol_free(pvma.vm_policy);
1060 static inline int shmem_parse_mpol(char *value, int *policy,
1061 nodemask_t *policy_nodes)
1066 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1067 struct shmem_inode_info *info, unsigned long idx)
1069 return swapin_readahead(entry, gfp, NULL, 0);
1072 static inline struct page *shmem_alloc_page(gfp_t gfp,
1073 struct shmem_inode_info *info, unsigned long idx)
1075 return alloc_page(gfp);
1080 * shmem_getpage - either get the page from swap or allocate a new one
1082 * If we allocate a new one we do not mark it dirty. That's up to the
1083 * vm. If we swap it in we mark it dirty since we also free the swap
1084 * entry since a page cannot live in both the swap and page cache
1086 static int shmem_getpage(struct inode *inode, unsigned long idx,
1087 struct page **pagep, enum sgp_type sgp, int *type)
1089 struct address_space *mapping = inode->i_mapping;
1090 struct shmem_inode_info *info = SHMEM_I(inode);
1091 struct shmem_sb_info *sbinfo;
1092 struct page *filepage = *pagep;
1093 struct page *swappage;
1099 if (idx >= SHMEM_MAX_INDEX)
1106 * Normally, filepage is NULL on entry, and either found
1107 * uptodate immediately, or allocated and zeroed, or read
1108 * in under swappage, which is then assigned to filepage.
1109 * But shmem_readpage (required for splice) passes in a locked
1110 * filepage, which may be found not uptodate by other callers
1111 * too, and may need to be copied from the swappage read in.
1115 filepage = find_lock_page(mapping, idx);
1116 if (filepage && PageUptodate(filepage))
1119 gfp = mapping_gfp_mask(mapping);
1121 spin_lock(&info->lock);
1122 shmem_recalc_inode(inode);
1123 entry = shmem_swp_alloc(info, idx, sgp);
1124 if (IS_ERR(entry)) {
1125 spin_unlock(&info->lock);
1126 error = PTR_ERR(entry);
1132 /* Look it up and read it in.. */
1133 swappage = lookup_swap_cache(swap);
1135 shmem_swp_unmap(entry);
1136 /* here we actually do the io */
1137 if (type && !(*type & VM_FAULT_MAJOR)) {
1138 __count_vm_event(PGMAJFAULT);
1139 *type |= VM_FAULT_MAJOR;
1141 spin_unlock(&info->lock);
1142 swappage = shmem_swapin(swap, gfp, info, idx);
1144 spin_lock(&info->lock);
1145 entry = shmem_swp_alloc(info, idx, sgp);
1147 error = PTR_ERR(entry);
1149 if (entry->val == swap.val)
1151 shmem_swp_unmap(entry);
1153 spin_unlock(&info->lock);
1158 wait_on_page_locked(swappage);
1159 page_cache_release(swappage);
1163 /* We have to do this with page locked to prevent races */
1164 if (TestSetPageLocked(swappage)) {
1165 shmem_swp_unmap(entry);
1166 spin_unlock(&info->lock);
1167 wait_on_page_locked(swappage);
1168 page_cache_release(swappage);
1171 if (PageWriteback(swappage)) {
1172 shmem_swp_unmap(entry);
1173 spin_unlock(&info->lock);
1174 wait_on_page_writeback(swappage);
1175 unlock_page(swappage);
1176 page_cache_release(swappage);
1179 if (!PageUptodate(swappage)) {
1180 shmem_swp_unmap(entry);
1181 spin_unlock(&info->lock);
1182 unlock_page(swappage);
1183 page_cache_release(swappage);
1189 shmem_swp_set(info, entry, 0);
1190 shmem_swp_unmap(entry);
1191 delete_from_swap_cache(swappage);
1192 spin_unlock(&info->lock);
1193 copy_highpage(filepage, swappage);
1194 unlock_page(swappage);
1195 page_cache_release(swappage);
1196 flush_dcache_page(filepage);
1197 SetPageUptodate(filepage);
1198 set_page_dirty(filepage);
1200 } else if (!(error = move_from_swap_cache(
1201 swappage, idx, mapping))) {
1202 info->flags |= SHMEM_PAGEIN;
1203 shmem_swp_set(info, entry, 0);
1204 shmem_swp_unmap(entry);
1205 spin_unlock(&info->lock);
1206 filepage = swappage;
1209 shmem_swp_unmap(entry);
1210 spin_unlock(&info->lock);
1211 unlock_page(swappage);
1212 page_cache_release(swappage);
1213 if (error == -ENOMEM) {
1214 /* let kswapd refresh zone for GFP_ATOMICs */
1215 congestion_wait(WRITE, HZ/50);
1219 } else if (sgp == SGP_READ && !filepage) {
1220 shmem_swp_unmap(entry);
1221 filepage = find_get_page(mapping, idx);
1223 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1224 spin_unlock(&info->lock);
1225 wait_on_page_locked(filepage);
1226 page_cache_release(filepage);
1230 spin_unlock(&info->lock);
1232 shmem_swp_unmap(entry);
1233 sbinfo = SHMEM_SB(inode->i_sb);
1234 if (sbinfo->max_blocks) {
1235 spin_lock(&sbinfo->stat_lock);
1236 if (sbinfo->free_blocks == 0 ||
1237 shmem_acct_block(info->flags)) {
1238 spin_unlock(&sbinfo->stat_lock);
1239 spin_unlock(&info->lock);
1243 sbinfo->free_blocks--;
1244 inode->i_blocks += BLOCKS_PER_PAGE;
1245 spin_unlock(&sbinfo->stat_lock);
1246 } else if (shmem_acct_block(info->flags)) {
1247 spin_unlock(&info->lock);
1253 spin_unlock(&info->lock);
1254 filepage = shmem_alloc_page(gfp, info, idx);
1256 shmem_unacct_blocks(info->flags, 1);
1257 shmem_free_blocks(inode, 1);
1262 spin_lock(&info->lock);
1263 entry = shmem_swp_alloc(info, idx, sgp);
1265 error = PTR_ERR(entry);
1268 shmem_swp_unmap(entry);
1270 if (error || swap.val || 0 != add_to_page_cache_lru(
1271 filepage, mapping, idx, GFP_ATOMIC)) {
1272 spin_unlock(&info->lock);
1273 page_cache_release(filepage);
1274 shmem_unacct_blocks(info->flags, 1);
1275 shmem_free_blocks(inode, 1);
1281 info->flags |= SHMEM_PAGEIN;
1285 spin_unlock(&info->lock);
1286 clear_highpage(filepage);
1287 flush_dcache_page(filepage);
1288 SetPageUptodate(filepage);
1295 if (*pagep != filepage) {
1296 unlock_page(filepage);
1297 page_cache_release(filepage);
1302 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1304 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1308 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1309 return VM_FAULT_SIGBUS;
1311 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1313 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1315 mark_page_accessed(vmf->page);
1316 return ret | VM_FAULT_LOCKED;
1320 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1322 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1323 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1326 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1329 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1332 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1333 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1337 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1339 struct inode *inode = file->f_path.dentry->d_inode;
1340 struct shmem_inode_info *info = SHMEM_I(inode);
1341 int retval = -ENOMEM;
1343 spin_lock(&info->lock);
1344 if (lock && !(info->flags & VM_LOCKED)) {
1345 if (!user_shm_lock(inode->i_size, user))
1347 info->flags |= VM_LOCKED;
1349 if (!lock && (info->flags & VM_LOCKED) && user) {
1350 user_shm_unlock(inode->i_size, user);
1351 info->flags &= ~VM_LOCKED;
1355 spin_unlock(&info->lock);
1359 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1361 file_accessed(file);
1362 vma->vm_ops = &shmem_vm_ops;
1363 vma->vm_flags |= VM_CAN_NONLINEAR;
1367 static struct inode *
1368 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1370 struct inode *inode;
1371 struct shmem_inode_info *info;
1372 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1374 if (sbinfo->max_inodes) {
1375 spin_lock(&sbinfo->stat_lock);
1376 if (!sbinfo->free_inodes) {
1377 spin_unlock(&sbinfo->stat_lock);
1380 sbinfo->free_inodes--;
1381 spin_unlock(&sbinfo->stat_lock);
1384 inode = new_inode(sb);
1386 inode->i_mode = mode;
1387 inode->i_uid = current->fsuid;
1388 inode->i_gid = current->fsgid;
1389 inode->i_blocks = 0;
1390 inode->i_mapping->a_ops = &shmem_aops;
1391 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1392 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1393 inode->i_generation = get_seconds();
1394 info = SHMEM_I(inode);
1395 memset(info, 0, (char *)inode - (char *)info);
1396 spin_lock_init(&info->lock);
1397 INIT_LIST_HEAD(&info->swaplist);
1399 switch (mode & S_IFMT) {
1401 inode->i_op = &shmem_special_inode_operations;
1402 init_special_inode(inode, mode, dev);
1405 inode->i_op = &shmem_inode_operations;
1406 inode->i_fop = &shmem_file_operations;
1407 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1408 &sbinfo->policy_nodes);
1412 /* Some things misbehave if size == 0 on a directory */
1413 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1414 inode->i_op = &shmem_dir_inode_operations;
1415 inode->i_fop = &simple_dir_operations;
1419 * Must not load anything in the rbtree,
1420 * mpol_free_shared_policy will not be called.
1422 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1426 } else if (sbinfo->max_inodes) {
1427 spin_lock(&sbinfo->stat_lock);
1428 sbinfo->free_inodes++;
1429 spin_unlock(&sbinfo->stat_lock);
1435 static const struct inode_operations shmem_symlink_inode_operations;
1436 static const struct inode_operations shmem_symlink_inline_operations;
1439 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1440 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1441 * below the loop driver, in the generic fashion that many filesystems support.
1443 static int shmem_readpage(struct file *file, struct page *page)
1445 struct inode *inode = page->mapping->host;
1446 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1452 shmem_write_begin(struct file *file, struct address_space *mapping,
1453 loff_t pos, unsigned len, unsigned flags,
1454 struct page **pagep, void **fsdata)
1456 struct inode *inode = mapping->host;
1457 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1459 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1463 shmem_write_end(struct file *file, struct address_space *mapping,
1464 loff_t pos, unsigned len, unsigned copied,
1465 struct page *page, void *fsdata)
1467 struct inode *inode = mapping->host;
1469 if (pos + copied > inode->i_size)
1470 i_size_write(inode, pos + copied);
1473 set_page_dirty(page);
1474 page_cache_release(page);
1479 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1481 struct inode *inode = filp->f_path.dentry->d_inode;
1482 struct address_space *mapping = inode->i_mapping;
1483 unsigned long index, offset;
1485 index = *ppos >> PAGE_CACHE_SHIFT;
1486 offset = *ppos & ~PAGE_CACHE_MASK;
1489 struct page *page = NULL;
1490 unsigned long end_index, nr, ret;
1491 loff_t i_size = i_size_read(inode);
1493 end_index = i_size >> PAGE_CACHE_SHIFT;
1494 if (index > end_index)
1496 if (index == end_index) {
1497 nr = i_size & ~PAGE_CACHE_MASK;
1502 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1504 if (desc->error == -EINVAL)
1512 * We must evaluate after, since reads (unlike writes)
1513 * are called without i_mutex protection against truncate
1515 nr = PAGE_CACHE_SIZE;
1516 i_size = i_size_read(inode);
1517 end_index = i_size >> PAGE_CACHE_SHIFT;
1518 if (index == end_index) {
1519 nr = i_size & ~PAGE_CACHE_MASK;
1522 page_cache_release(page);
1530 * If users can be writing to this page using arbitrary
1531 * virtual addresses, take care about potential aliasing
1532 * before reading the page on the kernel side.
1534 if (mapping_writably_mapped(mapping))
1535 flush_dcache_page(page);
1537 * Mark the page accessed if we read the beginning.
1540 mark_page_accessed(page);
1542 page = ZERO_PAGE(0);
1543 page_cache_get(page);
1547 * Ok, we have the page, and it's up-to-date, so
1548 * now we can copy it to user space...
1550 * The actor routine returns how many bytes were actually used..
1551 * NOTE! This may not be the same as how much of a user buffer
1552 * we filled up (we may be padding etc), so we can only update
1553 * "pos" here (the actor routine has to update the user buffer
1554 * pointers and the remaining count).
1556 ret = actor(desc, page, offset, nr);
1558 index += offset >> PAGE_CACHE_SHIFT;
1559 offset &= ~PAGE_CACHE_MASK;
1561 page_cache_release(page);
1562 if (ret != nr || !desc->count)
1568 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1569 file_accessed(filp);
1572 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1574 read_descriptor_t desc;
1576 if ((ssize_t) count < 0)
1578 if (!access_ok(VERIFY_WRITE, buf, count))
1588 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1590 return desc.written;
1594 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1596 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1598 buf->f_type = TMPFS_MAGIC;
1599 buf->f_bsize = PAGE_CACHE_SIZE;
1600 buf->f_namelen = NAME_MAX;
1601 spin_lock(&sbinfo->stat_lock);
1602 if (sbinfo->max_blocks) {
1603 buf->f_blocks = sbinfo->max_blocks;
1604 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1606 if (sbinfo->max_inodes) {
1607 buf->f_files = sbinfo->max_inodes;
1608 buf->f_ffree = sbinfo->free_inodes;
1610 /* else leave those fields 0 like simple_statfs */
1611 spin_unlock(&sbinfo->stat_lock);
1616 * File creation. Allocate an inode, and we're done..
1619 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1621 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1622 int error = -ENOSPC;
1625 error = security_inode_init_security(inode, dir, NULL, NULL,
1628 if (error != -EOPNOTSUPP) {
1633 error = shmem_acl_init(inode, dir);
1638 if (dir->i_mode & S_ISGID) {
1639 inode->i_gid = dir->i_gid;
1641 inode->i_mode |= S_ISGID;
1643 dir->i_size += BOGO_DIRENT_SIZE;
1644 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1645 d_instantiate(dentry, inode);
1646 dget(dentry); /* Extra count - pin the dentry in core */
1651 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1655 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1661 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1662 struct nameidata *nd)
1664 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1670 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1672 struct inode *inode = old_dentry->d_inode;
1673 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1676 * No ordinary (disk based) filesystem counts links as inodes;
1677 * but each new link needs a new dentry, pinning lowmem, and
1678 * tmpfs dentries cannot be pruned until they are unlinked.
1680 if (sbinfo->max_inodes) {
1681 spin_lock(&sbinfo->stat_lock);
1682 if (!sbinfo->free_inodes) {
1683 spin_unlock(&sbinfo->stat_lock);
1686 sbinfo->free_inodes--;
1687 spin_unlock(&sbinfo->stat_lock);
1690 dir->i_size += BOGO_DIRENT_SIZE;
1691 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1693 atomic_inc(&inode->i_count); /* New dentry reference */
1694 dget(dentry); /* Extra pinning count for the created dentry */
1695 d_instantiate(dentry, inode);
1699 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1701 struct inode *inode = dentry->d_inode;
1703 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1704 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1705 if (sbinfo->max_inodes) {
1706 spin_lock(&sbinfo->stat_lock);
1707 sbinfo->free_inodes++;
1708 spin_unlock(&sbinfo->stat_lock);
1712 dir->i_size -= BOGO_DIRENT_SIZE;
1713 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1715 dput(dentry); /* Undo the count from "create" - this does all the work */
1719 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1721 if (!simple_empty(dentry))
1724 drop_nlink(dentry->d_inode);
1726 return shmem_unlink(dir, dentry);
1730 * The VFS layer already does all the dentry stuff for rename,
1731 * we just have to decrement the usage count for the target if
1732 * it exists so that the VFS layer correctly free's it when it
1735 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1737 struct inode *inode = old_dentry->d_inode;
1738 int they_are_dirs = S_ISDIR(inode->i_mode);
1740 if (!simple_empty(new_dentry))
1743 if (new_dentry->d_inode) {
1744 (void) shmem_unlink(new_dir, new_dentry);
1746 drop_nlink(old_dir);
1747 } else if (they_are_dirs) {
1748 drop_nlink(old_dir);
1752 old_dir->i_size -= BOGO_DIRENT_SIZE;
1753 new_dir->i_size += BOGO_DIRENT_SIZE;
1754 old_dir->i_ctime = old_dir->i_mtime =
1755 new_dir->i_ctime = new_dir->i_mtime =
1756 inode->i_ctime = CURRENT_TIME;
1760 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1764 struct inode *inode;
1765 struct page *page = NULL;
1767 struct shmem_inode_info *info;
1769 len = strlen(symname) + 1;
1770 if (len > PAGE_CACHE_SIZE)
1771 return -ENAMETOOLONG;
1773 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1777 error = security_inode_init_security(inode, dir, NULL, NULL,
1780 if (error != -EOPNOTSUPP) {
1787 info = SHMEM_I(inode);
1788 inode->i_size = len-1;
1789 if (len <= (char *)inode - (char *)info) {
1791 memcpy(info, symname, len);
1792 inode->i_op = &shmem_symlink_inline_operations;
1794 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1800 inode->i_op = &shmem_symlink_inode_operations;
1801 kaddr = kmap_atomic(page, KM_USER0);
1802 memcpy(kaddr, symname, len);
1803 kunmap_atomic(kaddr, KM_USER0);
1804 set_page_dirty(page);
1805 page_cache_release(page);
1807 if (dir->i_mode & S_ISGID)
1808 inode->i_gid = dir->i_gid;
1809 dir->i_size += BOGO_DIRENT_SIZE;
1810 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1811 d_instantiate(dentry, inode);
1816 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1818 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1822 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1824 struct page *page = NULL;
1825 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1826 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1832 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1834 if (!IS_ERR(nd_get_link(nd))) {
1835 struct page *page = cookie;
1837 mark_page_accessed(page);
1838 page_cache_release(page);
1842 static const struct inode_operations shmem_symlink_inline_operations = {
1843 .readlink = generic_readlink,
1844 .follow_link = shmem_follow_link_inline,
1847 static const struct inode_operations shmem_symlink_inode_operations = {
1848 .truncate = shmem_truncate,
1849 .readlink = generic_readlink,
1850 .follow_link = shmem_follow_link,
1851 .put_link = shmem_put_link,
1854 #ifdef CONFIG_TMPFS_POSIX_ACL
1856 * Superblocks without xattr inode operations will get security.* xattr
1857 * support from the VFS "for free". As soon as we have any other xattrs
1858 * like ACLs, we also need to implement the security.* handlers at
1859 * filesystem level, though.
1862 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1863 size_t list_len, const char *name,
1866 return security_inode_listsecurity(inode, list, list_len);
1869 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1870 void *buffer, size_t size)
1872 if (strcmp(name, "") == 0)
1874 return security_inode_getsecurity(inode, name, buffer, size,
1878 static int shmem_xattr_security_set(struct inode *inode, const char *name,
1879 const void *value, size_t size, int flags)
1881 if (strcmp(name, "") == 0)
1883 return security_inode_setsecurity(inode, name, value, size, flags);
1886 static struct xattr_handler shmem_xattr_security_handler = {
1887 .prefix = XATTR_SECURITY_PREFIX,
1888 .list = shmem_xattr_security_list,
1889 .get = shmem_xattr_security_get,
1890 .set = shmem_xattr_security_set,
1893 static struct xattr_handler *shmem_xattr_handlers[] = {
1894 &shmem_xattr_acl_access_handler,
1895 &shmem_xattr_acl_default_handler,
1896 &shmem_xattr_security_handler,
1901 static struct dentry *shmem_get_parent(struct dentry *child)
1903 return ERR_PTR(-ESTALE);
1906 static int shmem_match(struct inode *ino, void *vfh)
1910 inum = (inum << 32) | fh[1];
1911 return ino->i_ino == inum && fh[0] == ino->i_generation;
1914 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1915 struct fid *fid, int fh_len, int fh_type)
1917 struct inode *inode;
1918 struct dentry *dentry = NULL;
1919 u64 inum = fid->raw[2];
1920 inum = (inum << 32) | fid->raw[1];
1925 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1926 shmem_match, fid->raw);
1928 dentry = d_find_alias(inode);
1935 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1938 struct inode *inode = dentry->d_inode;
1943 if (hlist_unhashed(&inode->i_hash)) {
1944 /* Unfortunately insert_inode_hash is not idempotent,
1945 * so as we hash inodes here rather than at creation
1946 * time, we need a lock to ensure we only try
1949 static DEFINE_SPINLOCK(lock);
1951 if (hlist_unhashed(&inode->i_hash))
1952 __insert_inode_hash(inode,
1953 inode->i_ino + inode->i_generation);
1957 fh[0] = inode->i_generation;
1958 fh[1] = inode->i_ino;
1959 fh[2] = ((__u64)inode->i_ino) >> 32;
1965 static const struct export_operations shmem_export_ops = {
1966 .get_parent = shmem_get_parent,
1967 .encode_fh = shmem_encode_fh,
1968 .fh_to_dentry = shmem_fh_to_dentry,
1971 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
1972 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
1973 int *policy, nodemask_t *policy_nodes)
1975 char *this_char, *value, *rest;
1977 while (options != NULL) {
1978 this_char = options;
1981 * NUL-terminate this option: unfortunately,
1982 * mount options form a comma-separated list,
1983 * but mpol's nodelist may also contain commas.
1985 options = strchr(options, ',');
1986 if (options == NULL)
1989 if (!isdigit(*options)) {
1996 if ((value = strchr(this_char,'=')) != NULL) {
2000 "tmpfs: No value for mount option '%s'\n",
2005 if (!strcmp(this_char,"size")) {
2006 unsigned long long size;
2007 size = memparse(value,&rest);
2009 size <<= PAGE_SHIFT;
2010 size *= totalram_pages;
2016 *blocks = size >> PAGE_CACHE_SHIFT;
2017 } else if (!strcmp(this_char,"nr_blocks")) {
2018 *blocks = memparse(value,&rest);
2021 } else if (!strcmp(this_char,"nr_inodes")) {
2022 *inodes = memparse(value,&rest);
2025 } else if (!strcmp(this_char,"mode")) {
2028 *mode = simple_strtoul(value,&rest,8);
2031 } else if (!strcmp(this_char,"uid")) {
2034 *uid = simple_strtoul(value,&rest,0);
2037 } else if (!strcmp(this_char,"gid")) {
2040 *gid = simple_strtoul(value,&rest,0);
2043 } else if (!strcmp(this_char,"mpol")) {
2044 if (shmem_parse_mpol(value,policy,policy_nodes))
2047 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2055 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2061 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2063 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2064 unsigned long max_blocks = sbinfo->max_blocks;
2065 unsigned long max_inodes = sbinfo->max_inodes;
2066 int policy = sbinfo->policy;
2067 nodemask_t policy_nodes = sbinfo->policy_nodes;
2068 unsigned long blocks;
2069 unsigned long inodes;
2070 int error = -EINVAL;
2072 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2073 &max_inodes, &policy, &policy_nodes))
2076 spin_lock(&sbinfo->stat_lock);
2077 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2078 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2079 if (max_blocks < blocks)
2081 if (max_inodes < inodes)
2084 * Those tests also disallow limited->unlimited while any are in
2085 * use, so i_blocks will always be zero when max_blocks is zero;
2086 * but we must separately disallow unlimited->limited, because
2087 * in that case we have no record of how much is already in use.
2089 if (max_blocks && !sbinfo->max_blocks)
2091 if (max_inodes && !sbinfo->max_inodes)
2095 sbinfo->max_blocks = max_blocks;
2096 sbinfo->free_blocks = max_blocks - blocks;
2097 sbinfo->max_inodes = max_inodes;
2098 sbinfo->free_inodes = max_inodes - inodes;
2099 sbinfo->policy = policy;
2100 sbinfo->policy_nodes = policy_nodes;
2102 spin_unlock(&sbinfo->stat_lock);
2107 static void shmem_put_super(struct super_block *sb)
2109 kfree(sb->s_fs_info);
2110 sb->s_fs_info = NULL;
2113 static int shmem_fill_super(struct super_block *sb,
2114 void *data, int silent)
2116 struct inode *inode;
2117 struct dentry *root;
2118 int mode = S_IRWXUGO | S_ISVTX;
2119 uid_t uid = current->fsuid;
2120 gid_t gid = current->fsgid;
2122 struct shmem_sb_info *sbinfo;
2123 unsigned long blocks = 0;
2124 unsigned long inodes = 0;
2125 int policy = MPOL_DEFAULT;
2126 nodemask_t policy_nodes = node_states[N_HIGH_MEMORY];
2130 * Per default we only allow half of the physical ram per
2131 * tmpfs instance, limiting inodes to one per page of lowmem;
2132 * but the internal instance is left unlimited.
2134 if (!(sb->s_flags & MS_NOUSER)) {
2135 blocks = totalram_pages / 2;
2136 inodes = totalram_pages - totalhigh_pages;
2137 if (inodes > blocks)
2139 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2140 &inodes, &policy, &policy_nodes))
2143 sb->s_export_op = &shmem_export_ops;
2145 sb->s_flags |= MS_NOUSER;
2148 /* Round up to L1_CACHE_BYTES to resist false sharing */
2149 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2150 L1_CACHE_BYTES), GFP_KERNEL);
2154 spin_lock_init(&sbinfo->stat_lock);
2155 sbinfo->max_blocks = blocks;
2156 sbinfo->free_blocks = blocks;
2157 sbinfo->max_inodes = inodes;
2158 sbinfo->free_inodes = inodes;
2159 sbinfo->policy = policy;
2160 sbinfo->policy_nodes = policy_nodes;
2162 sb->s_fs_info = sbinfo;
2163 sb->s_maxbytes = SHMEM_MAX_BYTES;
2164 sb->s_blocksize = PAGE_CACHE_SIZE;
2165 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2166 sb->s_magic = TMPFS_MAGIC;
2167 sb->s_op = &shmem_ops;
2168 sb->s_time_gran = 1;
2169 #ifdef CONFIG_TMPFS_POSIX_ACL
2170 sb->s_xattr = shmem_xattr_handlers;
2171 sb->s_flags |= MS_POSIXACL;
2174 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2179 root = d_alloc_root(inode);
2188 shmem_put_super(sb);
2192 static struct kmem_cache *shmem_inode_cachep;
2194 static struct inode *shmem_alloc_inode(struct super_block *sb)
2196 struct shmem_inode_info *p;
2197 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2200 return &p->vfs_inode;
2203 static void shmem_destroy_inode(struct inode *inode)
2205 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2206 /* only struct inode is valid if it's an inline symlink */
2207 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2209 shmem_acl_destroy_inode(inode);
2210 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2213 static void init_once(struct kmem_cache *cachep, void *foo)
2215 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2217 inode_init_once(&p->vfs_inode);
2218 #ifdef CONFIG_TMPFS_POSIX_ACL
2220 p->i_default_acl = NULL;
2224 static int init_inodecache(void)
2226 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2227 sizeof(struct shmem_inode_info),
2228 0, SLAB_PANIC, init_once);
2232 static void destroy_inodecache(void)
2234 kmem_cache_destroy(shmem_inode_cachep);
2237 static const struct address_space_operations shmem_aops = {
2238 .writepage = shmem_writepage,
2239 .set_page_dirty = __set_page_dirty_no_writeback,
2241 .readpage = shmem_readpage,
2242 .write_begin = shmem_write_begin,
2243 .write_end = shmem_write_end,
2245 .migratepage = migrate_page,
2248 static const struct file_operations shmem_file_operations = {
2251 .llseek = generic_file_llseek,
2252 .read = shmem_file_read,
2253 .write = do_sync_write,
2254 .aio_write = generic_file_aio_write,
2255 .fsync = simple_sync_file,
2256 .splice_read = generic_file_splice_read,
2257 .splice_write = generic_file_splice_write,
2261 static const struct inode_operations shmem_inode_operations = {
2262 .truncate = shmem_truncate,
2263 .setattr = shmem_notify_change,
2264 .truncate_range = shmem_truncate_range,
2265 #ifdef CONFIG_TMPFS_POSIX_ACL
2266 .setxattr = generic_setxattr,
2267 .getxattr = generic_getxattr,
2268 .listxattr = generic_listxattr,
2269 .removexattr = generic_removexattr,
2270 .permission = shmem_permission,
2275 static const struct inode_operations shmem_dir_inode_operations = {
2277 .create = shmem_create,
2278 .lookup = simple_lookup,
2280 .unlink = shmem_unlink,
2281 .symlink = shmem_symlink,
2282 .mkdir = shmem_mkdir,
2283 .rmdir = shmem_rmdir,
2284 .mknod = shmem_mknod,
2285 .rename = shmem_rename,
2287 #ifdef CONFIG_TMPFS_POSIX_ACL
2288 .setattr = shmem_notify_change,
2289 .setxattr = generic_setxattr,
2290 .getxattr = generic_getxattr,
2291 .listxattr = generic_listxattr,
2292 .removexattr = generic_removexattr,
2293 .permission = shmem_permission,
2297 static const struct inode_operations shmem_special_inode_operations = {
2298 #ifdef CONFIG_TMPFS_POSIX_ACL
2299 .setattr = shmem_notify_change,
2300 .setxattr = generic_setxattr,
2301 .getxattr = generic_getxattr,
2302 .listxattr = generic_listxattr,
2303 .removexattr = generic_removexattr,
2304 .permission = shmem_permission,
2308 static const struct super_operations shmem_ops = {
2309 .alloc_inode = shmem_alloc_inode,
2310 .destroy_inode = shmem_destroy_inode,
2312 .statfs = shmem_statfs,
2313 .remount_fs = shmem_remount_fs,
2315 .delete_inode = shmem_delete_inode,
2316 .drop_inode = generic_delete_inode,
2317 .put_super = shmem_put_super,
2320 static struct vm_operations_struct shmem_vm_ops = {
2321 .fault = shmem_fault,
2323 .set_policy = shmem_set_policy,
2324 .get_policy = shmem_get_policy,
2329 static int shmem_get_sb(struct file_system_type *fs_type,
2330 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2332 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2335 static struct file_system_type tmpfs_fs_type = {
2336 .owner = THIS_MODULE,
2338 .get_sb = shmem_get_sb,
2339 .kill_sb = kill_litter_super,
2341 static struct vfsmount *shm_mnt;
2343 static int __init init_tmpfs(void)
2347 error = bdi_init(&shmem_backing_dev_info);
2351 error = init_inodecache();
2355 error = register_filesystem(&tmpfs_fs_type);
2357 printk(KERN_ERR "Could not register tmpfs\n");
2361 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2362 tmpfs_fs_type.name, NULL);
2363 if (IS_ERR(shm_mnt)) {
2364 error = PTR_ERR(shm_mnt);
2365 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2371 unregister_filesystem(&tmpfs_fs_type);
2373 destroy_inodecache();
2375 bdi_destroy(&shmem_backing_dev_info);
2377 shm_mnt = ERR_PTR(error);
2380 module_init(init_tmpfs)
2383 * shmem_file_setup - get an unlinked file living in tmpfs
2385 * @name: name for dentry (to be seen in /proc/<pid>/maps
2386 * @size: size to be set for the file
2389 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2393 struct inode *inode;
2394 struct dentry *dentry, *root;
2397 if (IS_ERR(shm_mnt))
2398 return (void *)shm_mnt;
2400 if (size < 0 || size > SHMEM_MAX_BYTES)
2401 return ERR_PTR(-EINVAL);
2403 if (shmem_acct_size(flags, size))
2404 return ERR_PTR(-ENOMEM);
2408 this.len = strlen(name);
2409 this.hash = 0; /* will go */
2410 root = shm_mnt->mnt_root;
2411 dentry = d_alloc(root, &this);
2416 file = get_empty_filp();
2421 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2425 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2426 d_instantiate(dentry, inode);
2427 inode->i_size = size;
2428 inode->i_nlink = 0; /* It is unlinked */
2429 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2430 &shmem_file_operations);
2438 shmem_unacct_size(flags, size);
2439 return ERR_PTR(error);
2443 * shmem_zero_setup - setup a shared anonymous mapping
2445 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2447 int shmem_zero_setup(struct vm_area_struct *vma)
2450 loff_t size = vma->vm_end - vma->vm_start;
2452 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2454 return PTR_ERR(file);
2458 vma->vm_file = file;
2459 vma->vm_ops = &shmem_vm_ops;