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/generic_acl.h>
32 #include <linux/mman.h>
33 #include <linux/file.h>
34 #include <linux/swap.h>
35 #include <linux/pagemap.h>
36 #include <linux/string.h>
37 #include <linux/slab.h>
38 #include <linux/backing-dev.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/mount.h>
41 #include <linux/writeback.h>
42 #include <linux/vfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/security.h>
45 #include <linux/swapops.h>
46 #include <linux/mempolicy.h>
47 #include <linux/namei.h>
48 #include <linux/ctype.h>
49 #include <linux/migrate.h>
50 #include <linux/highmem.h>
51 #include <linux/backing-dev.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_QUICK, /* don't try more than file page cache lookup */
82 SGP_READ, /* don't exceed i_size, don't allocate page */
83 SGP_CACHE, /* don't exceed i_size, may allocate page */
84 SGP_WRITE, /* may exceed i_size, may allocate page */
87 static int shmem_getpage(struct inode *inode, unsigned long idx,
88 struct page **pagep, enum sgp_type sgp, int *type);
90 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
93 * The above definition of ENTRIES_PER_PAGE, and the use of
94 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
95 * might be reconsidered if it ever diverges from PAGE_SIZE.
97 * __GFP_MOVABLE is masked out as swap vectors cannot move
99 return alloc_pages((gfp_mask & ~__GFP_MOVABLE) | __GFP_ZERO,
100 PAGE_CACHE_SHIFT-PAGE_SHIFT);
103 static inline void shmem_dir_free(struct page *page)
105 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
108 static struct page **shmem_dir_map(struct page *page)
110 return (struct page **)kmap_atomic(page, KM_USER0);
113 static inline void shmem_dir_unmap(struct page **dir)
115 kunmap_atomic(dir, KM_USER0);
118 static swp_entry_t *shmem_swp_map(struct page *page)
120 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
123 static inline void shmem_swp_balance_unmap(void)
126 * When passing a pointer to an i_direct entry, to code which
127 * also handles indirect entries and so will shmem_swp_unmap,
128 * we must arrange for the preempt count to remain in balance.
129 * What kmap_atomic of a lowmem page does depends on config
130 * and architecture, so pretend to kmap_atomic some lowmem page.
132 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
135 static inline void shmem_swp_unmap(swp_entry_t *entry)
137 kunmap_atomic(entry, KM_USER1);
140 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
142 return sb->s_fs_info;
146 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
147 * for shared memory and for shared anonymous (/dev/zero) mappings
148 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
149 * consistent with the pre-accounting of private mappings ...
151 static inline int shmem_acct_size(unsigned long flags, loff_t size)
153 return (flags & VM_ACCOUNT)?
154 security_vm_enough_memory(VM_ACCT(size)): 0;
157 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
159 if (flags & VM_ACCOUNT)
160 vm_unacct_memory(VM_ACCT(size));
164 * ... whereas tmpfs objects are accounted incrementally as
165 * pages are allocated, in order to allow huge sparse files.
166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169 static inline int shmem_acct_block(unsigned long flags)
171 return (flags & VM_ACCOUNT)?
172 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
175 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
177 if (!(flags & VM_ACCOUNT))
178 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
181 static const struct super_operations shmem_ops;
182 static const struct address_space_operations shmem_aops;
183 static const struct file_operations shmem_file_operations;
184 static const struct inode_operations shmem_inode_operations;
185 static const struct inode_operations shmem_dir_inode_operations;
186 static const struct inode_operations shmem_special_inode_operations;
187 static struct vm_operations_struct shmem_vm_ops;
189 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
190 .ra_pages = 0, /* No readahead */
191 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
192 .unplug_io_fn = default_unplug_io_fn,
195 static LIST_HEAD(shmem_swaplist);
196 static DEFINE_SPINLOCK(shmem_swaplist_lock);
198 static void shmem_free_blocks(struct inode *inode, long pages)
200 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
201 if (sbinfo->max_blocks) {
202 spin_lock(&sbinfo->stat_lock);
203 sbinfo->free_blocks += pages;
204 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
205 spin_unlock(&sbinfo->stat_lock);
210 * shmem_recalc_inode - recalculate the size of an inode
212 * @inode: inode to recalc
214 * We have to calculate the free blocks since the mm can drop
215 * undirtied hole pages behind our back.
217 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
218 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
220 * It has to be called with the spinlock held.
222 static void shmem_recalc_inode(struct inode *inode)
224 struct shmem_inode_info *info = SHMEM_I(inode);
227 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
229 info->alloced -= freed;
230 shmem_unacct_blocks(info->flags, freed);
231 shmem_free_blocks(inode, freed);
236 * shmem_swp_entry - find the swap vector position in the info structure
238 * @info: info structure for the inode
239 * @index: index of the page to find
240 * @page: optional page to add to the structure. Has to be preset to
243 * If there is no space allocated yet it will return NULL when
244 * page is NULL, else it will use the page for the needed block,
245 * setting it to NULL on return to indicate that it has been used.
247 * The swap vector is organized the following way:
249 * There are SHMEM_NR_DIRECT entries directly stored in the
250 * shmem_inode_info structure. So small files do not need an addional
253 * For pages with index > SHMEM_NR_DIRECT there is the pointer
254 * i_indirect which points to a page which holds in the first half
255 * doubly indirect blocks, in the second half triple indirect blocks:
257 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
258 * following layout (for SHMEM_NR_DIRECT == 16):
260 * i_indirect -> dir --> 16-19
273 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
275 unsigned long offset;
279 if (index < SHMEM_NR_DIRECT) {
280 shmem_swp_balance_unmap();
281 return info->i_direct+index;
283 if (!info->i_indirect) {
285 info->i_indirect = *page;
288 return NULL; /* need another page */
291 index -= SHMEM_NR_DIRECT;
292 offset = index % ENTRIES_PER_PAGE;
293 index /= ENTRIES_PER_PAGE;
294 dir = shmem_dir_map(info->i_indirect);
296 if (index >= ENTRIES_PER_PAGE/2) {
297 index -= ENTRIES_PER_PAGE/2;
298 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
299 index %= ENTRIES_PER_PAGE;
306 shmem_dir_unmap(dir);
307 return NULL; /* need another page */
309 shmem_dir_unmap(dir);
310 dir = shmem_dir_map(subdir);
316 if (!page || !(subdir = *page)) {
317 shmem_dir_unmap(dir);
318 return NULL; /* need a page */
323 shmem_dir_unmap(dir);
324 return shmem_swp_map(subdir) + offset;
327 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
329 long incdec = value? 1: -1;
332 info->swapped += incdec;
333 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
334 struct page *page = kmap_atomic_to_page(entry);
335 set_page_private(page, page_private(page) + incdec);
340 * shmem_swp_alloc - get the position of the swap entry for the page.
341 * If it does not exist allocate the entry.
343 * @info: info structure for the inode
344 * @index: index of the page to find
345 * @sgp: check and recheck i_size? skip allocation?
347 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
349 struct inode *inode = &info->vfs_inode;
350 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
351 struct page *page = NULL;
354 if (sgp != SGP_WRITE &&
355 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
356 return ERR_PTR(-EINVAL);
358 while (!(entry = shmem_swp_entry(info, index, &page))) {
360 return shmem_swp_map(ZERO_PAGE(0));
362 * Test free_blocks against 1 not 0, since we have 1 data
363 * page (and perhaps indirect index pages) yet to allocate:
364 * a waste to allocate index if we cannot allocate data.
366 if (sbinfo->max_blocks) {
367 spin_lock(&sbinfo->stat_lock);
368 if (sbinfo->free_blocks <= 1) {
369 spin_unlock(&sbinfo->stat_lock);
370 return ERR_PTR(-ENOSPC);
372 sbinfo->free_blocks--;
373 inode->i_blocks += BLOCKS_PER_PAGE;
374 spin_unlock(&sbinfo->stat_lock);
377 spin_unlock(&info->lock);
378 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
380 set_page_private(page, 0);
381 spin_lock(&info->lock);
384 shmem_free_blocks(inode, 1);
385 return ERR_PTR(-ENOMEM);
387 if (sgp != SGP_WRITE &&
388 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
389 entry = ERR_PTR(-EINVAL);
392 if (info->next_index <= index)
393 info->next_index = index + 1;
396 /* another task gave its page, or truncated the file */
397 shmem_free_blocks(inode, 1);
398 shmem_dir_free(page);
400 if (info->next_index <= index && !IS_ERR(entry))
401 info->next_index = index + 1;
406 * shmem_free_swp - free some swap entries in a directory
408 * @dir: pointer to the directory
409 * @edir: pointer after last entry of the directory
410 * @punch_lock: pointer to spinlock when needed for the holepunch case
412 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
413 spinlock_t *punch_lock)
415 spinlock_t *punch_unlock = NULL;
419 for (ptr = dir; ptr < edir; ptr++) {
421 if (unlikely(punch_lock)) {
422 punch_unlock = punch_lock;
424 spin_lock(punch_unlock);
428 free_swap_and_cache(*ptr);
429 *ptr = (swp_entry_t){0};
434 spin_unlock(punch_unlock);
438 static int shmem_map_and_free_swp(struct page *subdir, int offset,
439 int limit, struct page ***dir, spinlock_t *punch_lock)
444 ptr = shmem_swp_map(subdir);
445 for (; offset < limit; offset += LATENCY_LIMIT) {
446 int size = limit - offset;
447 if (size > LATENCY_LIMIT)
448 size = LATENCY_LIMIT;
449 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
451 if (need_resched()) {
452 shmem_swp_unmap(ptr);
454 shmem_dir_unmap(*dir);
458 ptr = shmem_swp_map(subdir);
461 shmem_swp_unmap(ptr);
465 static void shmem_free_pages(struct list_head *next)
471 page = container_of(next, struct page, lru);
473 shmem_dir_free(page);
475 if (freed >= LATENCY_LIMIT) {
482 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
484 struct shmem_inode_info *info = SHMEM_I(inode);
489 unsigned long diroff;
495 LIST_HEAD(pages_to_free);
496 long nr_pages_to_free = 0;
497 long nr_swaps_freed = 0;
501 spinlock_t *needs_lock;
502 spinlock_t *punch_lock;
503 unsigned long upper_limit;
505 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
506 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
507 if (idx >= info->next_index)
510 spin_lock(&info->lock);
511 info->flags |= SHMEM_TRUNCATE;
512 if (likely(end == (loff_t) -1)) {
513 limit = info->next_index;
514 upper_limit = SHMEM_MAX_INDEX;
515 info->next_index = idx;
519 if (end + 1 >= inode->i_size) { /* we may free a little more */
520 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
522 upper_limit = SHMEM_MAX_INDEX;
524 limit = (end + 1) >> PAGE_CACHE_SHIFT;
527 needs_lock = &info->lock;
531 topdir = info->i_indirect;
532 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
533 info->i_indirect = NULL;
535 list_add(&topdir->lru, &pages_to_free);
537 spin_unlock(&info->lock);
539 if (info->swapped && idx < SHMEM_NR_DIRECT) {
540 ptr = info->i_direct;
542 if (size > SHMEM_NR_DIRECT)
543 size = SHMEM_NR_DIRECT;
544 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
548 * If there are no indirect blocks or we are punching a hole
549 * below indirect blocks, nothing to be done.
551 if (!topdir || limit <= SHMEM_NR_DIRECT)
555 * The truncation case has already dropped info->lock, and we're safe
556 * because i_size and next_index have already been lowered, preventing
557 * access beyond. But in the punch_hole case, we still need to take
558 * the lock when updating the swap directory, because there might be
559 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
560 * shmem_writepage. However, whenever we find we can remove a whole
561 * directory page (not at the misaligned start or end of the range),
562 * we first NULLify its pointer in the level above, and then have no
563 * need to take the lock when updating its contents: needs_lock and
564 * punch_lock (either pointing to info->lock or NULL) manage this.
567 upper_limit -= SHMEM_NR_DIRECT;
568 limit -= SHMEM_NR_DIRECT;
569 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
570 offset = idx % ENTRIES_PER_PAGE;
573 dir = shmem_dir_map(topdir);
574 stage = ENTRIES_PER_PAGEPAGE/2;
575 if (idx < ENTRIES_PER_PAGEPAGE/2) {
577 diroff = idx/ENTRIES_PER_PAGE;
579 dir += ENTRIES_PER_PAGE/2;
580 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
582 stage += ENTRIES_PER_PAGEPAGE;
585 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
586 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
587 if (!diroff && !offset && upper_limit >= stage) {
589 spin_lock(needs_lock);
591 spin_unlock(needs_lock);
596 list_add(&middir->lru, &pages_to_free);
598 shmem_dir_unmap(dir);
599 dir = shmem_dir_map(middir);
607 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
608 if (unlikely(idx == stage)) {
609 shmem_dir_unmap(dir);
610 dir = shmem_dir_map(topdir) +
611 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
614 idx += ENTRIES_PER_PAGEPAGE;
618 stage = idx + ENTRIES_PER_PAGEPAGE;
621 needs_lock = &info->lock;
622 if (upper_limit >= stage) {
624 spin_lock(needs_lock);
626 spin_unlock(needs_lock);
631 list_add(&middir->lru, &pages_to_free);
633 shmem_dir_unmap(dir);
635 dir = shmem_dir_map(middir);
638 punch_lock = needs_lock;
639 subdir = dir[diroff];
640 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
642 spin_lock(needs_lock);
644 spin_unlock(needs_lock);
649 list_add(&subdir->lru, &pages_to_free);
651 if (subdir && page_private(subdir) /* has swap entries */) {
653 if (size > ENTRIES_PER_PAGE)
654 size = ENTRIES_PER_PAGE;
655 freed = shmem_map_and_free_swp(subdir,
656 offset, size, &dir, punch_lock);
658 dir = shmem_dir_map(middir);
659 nr_swaps_freed += freed;
660 if (offset || punch_lock) {
661 spin_lock(&info->lock);
662 set_page_private(subdir,
663 page_private(subdir) - freed);
664 spin_unlock(&info->lock);
666 BUG_ON(page_private(subdir) != freed);
671 shmem_dir_unmap(dir);
673 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
675 * Call truncate_inode_pages again: racing shmem_unuse_inode
676 * may have swizzled a page in from swap since vmtruncate or
677 * generic_delete_inode did it, before we lowered next_index.
678 * Also, though shmem_getpage checks i_size before adding to
679 * cache, no recheck after: so fix the narrow window there too.
681 * Recalling truncate_inode_pages_range and unmap_mapping_range
682 * every time for punch_hole (which never got a chance to clear
683 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
684 * yet hardly ever necessary: try to optimize them out later.
686 truncate_inode_pages_range(inode->i_mapping, start, end);
688 unmap_mapping_range(inode->i_mapping, start,
692 spin_lock(&info->lock);
693 info->flags &= ~SHMEM_TRUNCATE;
694 info->swapped -= nr_swaps_freed;
695 if (nr_pages_to_free)
696 shmem_free_blocks(inode, nr_pages_to_free);
697 shmem_recalc_inode(inode);
698 spin_unlock(&info->lock);
701 * Empty swap vector directory pages to be freed?
703 if (!list_empty(&pages_to_free)) {
704 pages_to_free.prev->next = NULL;
705 shmem_free_pages(pages_to_free.next);
709 static void shmem_truncate(struct inode *inode)
711 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
714 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
716 struct inode *inode = dentry->d_inode;
717 struct page *page = NULL;
720 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
721 if (attr->ia_size < inode->i_size) {
723 * If truncating down to a partial page, then
724 * if that page is already allocated, hold it
725 * in memory until the truncation is over, so
726 * truncate_partial_page cannnot miss it were
727 * it assigned to swap.
729 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
730 (void) shmem_getpage(inode,
731 attr->ia_size>>PAGE_CACHE_SHIFT,
732 &page, SGP_READ, NULL);
735 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
736 * detect if any pages might have been added to cache
737 * after truncate_inode_pages. But we needn't bother
738 * if it's being fully truncated to zero-length: the
739 * nrpages check is efficient enough in that case.
742 struct shmem_inode_info *info = SHMEM_I(inode);
743 spin_lock(&info->lock);
744 info->flags &= ~SHMEM_PAGEIN;
745 spin_unlock(&info->lock);
750 error = inode_change_ok(inode, attr);
752 error = inode_setattr(inode, attr);
753 #ifdef CONFIG_TMPFS_POSIX_ACL
754 if (!error && (attr->ia_valid & ATTR_MODE))
755 error = generic_acl_chmod(inode, &shmem_acl_ops);
758 page_cache_release(page);
762 static void shmem_delete_inode(struct inode *inode)
764 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
765 struct shmem_inode_info *info = SHMEM_I(inode);
767 if (inode->i_op->truncate == shmem_truncate) {
768 truncate_inode_pages(inode->i_mapping, 0);
769 shmem_unacct_size(info->flags, inode->i_size);
771 shmem_truncate(inode);
772 if (!list_empty(&info->swaplist)) {
773 spin_lock(&shmem_swaplist_lock);
774 list_del_init(&info->swaplist);
775 spin_unlock(&shmem_swaplist_lock);
778 BUG_ON(inode->i_blocks);
779 if (sbinfo->max_inodes) {
780 spin_lock(&sbinfo->stat_lock);
781 sbinfo->free_inodes++;
782 spin_unlock(&sbinfo->stat_lock);
787 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
791 for (ptr = dir; ptr < edir; ptr++) {
792 if (ptr->val == entry.val)
798 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
811 ptr = info->i_direct;
812 spin_lock(&info->lock);
813 limit = info->next_index;
815 if (size > SHMEM_NR_DIRECT)
816 size = SHMEM_NR_DIRECT;
817 offset = shmem_find_swp(entry, ptr, ptr+size);
819 shmem_swp_balance_unmap();
822 if (!info->i_indirect)
825 dir = shmem_dir_map(info->i_indirect);
826 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
828 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
829 if (unlikely(idx == stage)) {
830 shmem_dir_unmap(dir-1);
831 dir = shmem_dir_map(info->i_indirect) +
832 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
835 idx += ENTRIES_PER_PAGEPAGE;
839 stage = idx + ENTRIES_PER_PAGEPAGE;
841 shmem_dir_unmap(dir);
842 dir = shmem_dir_map(subdir);
845 if (subdir && page_private(subdir)) {
846 ptr = shmem_swp_map(subdir);
848 if (size > ENTRIES_PER_PAGE)
849 size = ENTRIES_PER_PAGE;
850 offset = shmem_find_swp(entry, ptr, ptr+size);
852 shmem_dir_unmap(dir);
855 shmem_swp_unmap(ptr);
859 shmem_dir_unmap(dir-1);
861 spin_unlock(&info->lock);
865 inode = &info->vfs_inode;
866 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
867 info->flags |= SHMEM_PAGEIN;
868 shmem_swp_set(info, ptr + offset, 0);
870 shmem_swp_unmap(ptr);
871 spin_unlock(&info->lock);
873 * Decrement swap count even when the entry is left behind:
874 * try_to_unuse will skip over mms, then reincrement count.
881 * shmem_unuse() search for an eventually swapped out shmem page.
883 int shmem_unuse(swp_entry_t entry, struct page *page)
885 struct list_head *p, *next;
886 struct shmem_inode_info *info;
889 spin_lock(&shmem_swaplist_lock);
890 list_for_each_safe(p, next, &shmem_swaplist) {
891 info = list_entry(p, struct shmem_inode_info, swaplist);
893 list_del_init(&info->swaplist);
894 else if (shmem_unuse_inode(info, entry, page)) {
895 /* move head to start search for next from here */
896 list_move_tail(&shmem_swaplist, &info->swaplist);
901 spin_unlock(&shmem_swaplist_lock);
906 * Move the page from the page cache to the swap cache.
908 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
910 struct shmem_inode_info *info;
911 swp_entry_t *entry, swap;
912 struct address_space *mapping;
916 BUG_ON(!PageLocked(page));
917 BUG_ON(page_mapped(page));
919 mapping = page->mapping;
921 inode = mapping->host;
922 info = SHMEM_I(inode);
923 if (info->flags & VM_LOCKED)
925 swap = get_swap_page();
929 spin_lock(&info->lock);
930 shmem_recalc_inode(inode);
931 if (index >= info->next_index) {
932 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
935 entry = shmem_swp_entry(info, index, NULL);
939 if (move_to_swap_cache(page, swap) == 0) {
940 shmem_swp_set(info, entry, swap.val);
941 shmem_swp_unmap(entry);
942 spin_unlock(&info->lock);
943 if (list_empty(&info->swaplist)) {
944 spin_lock(&shmem_swaplist_lock);
945 /* move instead of add in case we're racing */
946 list_move_tail(&info->swaplist, &shmem_swaplist);
947 spin_unlock(&shmem_swaplist_lock);
953 shmem_swp_unmap(entry);
955 spin_unlock(&info->lock);
958 set_page_dirty(page);
959 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
963 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
965 char *nodelist = strchr(value, ':');
969 /* NUL-terminate policy string */
971 if (nodelist_parse(nodelist, *policy_nodes))
973 if (!nodes_subset(*policy_nodes, node_online_map))
976 if (!strcmp(value, "default")) {
977 *policy = MPOL_DEFAULT;
978 /* Don't allow a nodelist */
981 } else if (!strcmp(value, "prefer")) {
982 *policy = MPOL_PREFERRED;
983 /* Insist on a nodelist of one node only */
985 char *rest = nodelist;
986 while (isdigit(*rest))
991 } else if (!strcmp(value, "bind")) {
993 /* Insist on a nodelist */
996 } else if (!strcmp(value, "interleave")) {
997 *policy = MPOL_INTERLEAVE;
998 /* Default to nodes online if no nodelist */
1000 *policy_nodes = node_online_map;
1004 /* Restore string for error message */
1010 static struct page *shmem_swapin_async(struct shared_policy *p,
1011 swp_entry_t entry, unsigned long idx)
1014 struct vm_area_struct pvma;
1016 /* Create a pseudo vma that just contains the policy */
1017 memset(&pvma, 0, sizeof(struct vm_area_struct));
1018 pvma.vm_end = PAGE_SIZE;
1019 pvma.vm_pgoff = idx;
1020 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
1021 page = read_swap_cache_async(entry, &pvma, 0);
1022 mpol_free(pvma.vm_policy);
1026 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
1029 struct shared_policy *p = &info->policy;
1032 unsigned long offset;
1034 num = valid_swaphandles(entry, &offset);
1035 for (i = 0; i < num; offset++, i++) {
1036 page = shmem_swapin_async(p,
1037 swp_entry(swp_type(entry), offset), idx);
1040 page_cache_release(page);
1042 lru_add_drain(); /* Push any new pages onto the LRU now */
1043 return shmem_swapin_async(p, entry, idx);
1046 static struct page *
1047 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
1050 struct vm_area_struct pvma;
1053 memset(&pvma, 0, sizeof(struct vm_area_struct));
1054 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1055 pvma.vm_pgoff = idx;
1056 pvma.vm_end = PAGE_SIZE;
1057 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
1058 mpol_free(pvma.vm_policy);
1062 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
1067 static inline struct page *
1068 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
1070 swapin_readahead(entry, 0, NULL);
1071 return read_swap_cache_async(entry, NULL, 0);
1074 static inline struct page *
1075 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
1077 return alloc_page(gfp | __GFP_ZERO);
1082 * shmem_getpage - either get the page from swap or allocate a new one
1084 * If we allocate a new one we do not mark it dirty. That's up to the
1085 * vm. If we swap it in we mark it dirty since we also free the swap
1086 * entry since a page cannot live in both the swap and page cache
1088 static int shmem_getpage(struct inode *inode, unsigned long idx,
1089 struct page **pagep, enum sgp_type sgp, int *type)
1091 struct address_space *mapping = inode->i_mapping;
1092 struct shmem_inode_info *info = SHMEM_I(inode);
1093 struct shmem_sb_info *sbinfo;
1094 struct page *filepage = *pagep;
1095 struct page *swappage;
1100 if (idx >= SHMEM_MAX_INDEX)
1103 * Normally, filepage is NULL on entry, and either found
1104 * uptodate immediately, or allocated and zeroed, or read
1105 * in under swappage, which is then assigned to filepage.
1106 * But shmem_readpage and shmem_prepare_write pass in a locked
1107 * filepage, which may be found not uptodate by other callers
1108 * too, and may need to be copied from the swappage read in.
1112 filepage = find_lock_page(mapping, idx);
1113 if (filepage && PageUptodate(filepage))
1116 if (sgp == SGP_QUICK)
1119 spin_lock(&info->lock);
1120 shmem_recalc_inode(inode);
1121 entry = shmem_swp_alloc(info, idx, sgp);
1122 if (IS_ERR(entry)) {
1123 spin_unlock(&info->lock);
1124 error = PTR_ERR(entry);
1130 /* Look it up and read it in.. */
1131 swappage = lookup_swap_cache(swap);
1133 shmem_swp_unmap(entry);
1134 /* here we actually do the io */
1135 if (type && *type == VM_FAULT_MINOR) {
1136 __count_vm_event(PGMAJFAULT);
1137 *type = VM_FAULT_MAJOR;
1139 spin_unlock(&info->lock);
1140 swappage = shmem_swapin(info, swap, idx);
1142 spin_lock(&info->lock);
1143 entry = shmem_swp_alloc(info, idx, sgp);
1145 error = PTR_ERR(entry);
1147 if (entry->val == swap.val)
1149 shmem_swp_unmap(entry);
1151 spin_unlock(&info->lock);
1156 wait_on_page_locked(swappage);
1157 page_cache_release(swappage);
1161 /* We have to do this with page locked to prevent races */
1162 if (TestSetPageLocked(swappage)) {
1163 shmem_swp_unmap(entry);
1164 spin_unlock(&info->lock);
1165 wait_on_page_locked(swappage);
1166 page_cache_release(swappage);
1169 if (PageWriteback(swappage)) {
1170 shmem_swp_unmap(entry);
1171 spin_unlock(&info->lock);
1172 wait_on_page_writeback(swappage);
1173 unlock_page(swappage);
1174 page_cache_release(swappage);
1177 if (!PageUptodate(swappage)) {
1178 shmem_swp_unmap(entry);
1179 spin_unlock(&info->lock);
1180 unlock_page(swappage);
1181 page_cache_release(swappage);
1187 shmem_swp_set(info, entry, 0);
1188 shmem_swp_unmap(entry);
1189 delete_from_swap_cache(swappage);
1190 spin_unlock(&info->lock);
1191 copy_highpage(filepage, swappage);
1192 unlock_page(swappage);
1193 page_cache_release(swappage);
1194 flush_dcache_page(filepage);
1195 SetPageUptodate(filepage);
1196 set_page_dirty(filepage);
1198 } else if (!(error = move_from_swap_cache(
1199 swappage, idx, mapping))) {
1200 info->flags |= SHMEM_PAGEIN;
1201 shmem_swp_set(info, entry, 0);
1202 shmem_swp_unmap(entry);
1203 spin_unlock(&info->lock);
1204 filepage = swappage;
1207 shmem_swp_unmap(entry);
1208 spin_unlock(&info->lock);
1209 unlock_page(swappage);
1210 page_cache_release(swappage);
1211 if (error == -ENOMEM) {
1212 /* let kswapd refresh zone for GFP_ATOMICs */
1213 congestion_wait(WRITE, HZ/50);
1217 } else if (sgp == SGP_READ && !filepage) {
1218 shmem_swp_unmap(entry);
1219 filepage = find_get_page(mapping, idx);
1221 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1222 spin_unlock(&info->lock);
1223 wait_on_page_locked(filepage);
1224 page_cache_release(filepage);
1228 spin_unlock(&info->lock);
1230 shmem_swp_unmap(entry);
1231 sbinfo = SHMEM_SB(inode->i_sb);
1232 if (sbinfo->max_blocks) {
1233 spin_lock(&sbinfo->stat_lock);
1234 if (sbinfo->free_blocks == 0 ||
1235 shmem_acct_block(info->flags)) {
1236 spin_unlock(&sbinfo->stat_lock);
1237 spin_unlock(&info->lock);
1241 sbinfo->free_blocks--;
1242 inode->i_blocks += BLOCKS_PER_PAGE;
1243 spin_unlock(&sbinfo->stat_lock);
1244 } else if (shmem_acct_block(info->flags)) {
1245 spin_unlock(&info->lock);
1251 spin_unlock(&info->lock);
1252 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
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 flush_dcache_page(filepage);
1287 SetPageUptodate(filepage);
1290 if (*pagep != filepage) {
1291 unlock_page(filepage);
1297 if (*pagep != filepage) {
1298 unlock_page(filepage);
1299 page_cache_release(filepage);
1304 static struct page *shmem_nopage(struct vm_area_struct *vma,
1305 unsigned long address, int *type)
1307 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1308 struct page *page = NULL;
1312 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1313 idx += vma->vm_pgoff;
1314 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1315 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1316 return NOPAGE_SIGBUS;
1318 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1320 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1322 mark_page_accessed(page);
1326 static int shmem_populate(struct vm_area_struct *vma,
1327 unsigned long addr, unsigned long len,
1328 pgprot_t prot, unsigned long pgoff, int nonblock)
1330 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1331 struct mm_struct *mm = vma->vm_mm;
1332 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1335 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1336 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1339 while ((long) len > 0) {
1340 struct page *page = NULL;
1343 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1345 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1348 /* Page may still be null, but only if nonblock was set. */
1350 mark_page_accessed(page);
1351 err = install_page(mm, vma, addr, page, prot);
1353 page_cache_release(page);
1356 } else if (vma->vm_flags & VM_NONLINEAR) {
1357 /* No page was found just because we can't read it in
1358 * now (being here implies nonblock != 0), but the page
1359 * may exist, so set the PTE to fault it in later. */
1360 err = install_file_pte(mm, vma, addr, pgoff, prot);
1373 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1375 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1376 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1380 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1382 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1385 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1386 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1390 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1392 struct inode *inode = file->f_path.dentry->d_inode;
1393 struct shmem_inode_info *info = SHMEM_I(inode);
1394 int retval = -ENOMEM;
1396 spin_lock(&info->lock);
1397 if (lock && !(info->flags & VM_LOCKED)) {
1398 if (!user_shm_lock(inode->i_size, user))
1400 info->flags |= VM_LOCKED;
1402 if (!lock && (info->flags & VM_LOCKED) && user) {
1403 user_shm_unlock(inode->i_size, user);
1404 info->flags &= ~VM_LOCKED;
1408 spin_unlock(&info->lock);
1412 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1414 file_accessed(file);
1415 vma->vm_ops = &shmem_vm_ops;
1419 static struct inode *
1420 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1422 struct inode *inode;
1423 struct shmem_inode_info *info;
1424 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1426 if (sbinfo->max_inodes) {
1427 spin_lock(&sbinfo->stat_lock);
1428 if (!sbinfo->free_inodes) {
1429 spin_unlock(&sbinfo->stat_lock);
1432 sbinfo->free_inodes--;
1433 spin_unlock(&sbinfo->stat_lock);
1436 inode = new_inode(sb);
1438 inode->i_mode = mode;
1439 inode->i_uid = current->fsuid;
1440 inode->i_gid = current->fsgid;
1441 inode->i_blocks = 0;
1442 inode->i_mapping->a_ops = &shmem_aops;
1443 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1444 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1445 inode->i_generation = get_seconds();
1446 info = SHMEM_I(inode);
1447 memset(info, 0, (char *)inode - (char *)info);
1448 spin_lock_init(&info->lock);
1449 INIT_LIST_HEAD(&info->swaplist);
1451 switch (mode & S_IFMT) {
1453 inode->i_op = &shmem_special_inode_operations;
1454 init_special_inode(inode, mode, dev);
1457 inode->i_op = &shmem_inode_operations;
1458 inode->i_fop = &shmem_file_operations;
1459 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1460 &sbinfo->policy_nodes);
1464 /* Some things misbehave if size == 0 on a directory */
1465 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1466 inode->i_op = &shmem_dir_inode_operations;
1467 inode->i_fop = &simple_dir_operations;
1471 * Must not load anything in the rbtree,
1472 * mpol_free_shared_policy will not be called.
1474 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1478 } else if (sbinfo->max_inodes) {
1479 spin_lock(&sbinfo->stat_lock);
1480 sbinfo->free_inodes++;
1481 spin_unlock(&sbinfo->stat_lock);
1487 static const struct inode_operations shmem_symlink_inode_operations;
1488 static const struct inode_operations shmem_symlink_inline_operations;
1491 * Normally tmpfs avoids the use of shmem_readpage and shmem_prepare_write;
1492 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1493 * below the loop driver, in the generic fashion that many filesystems support.
1495 static int shmem_readpage(struct file *file, struct page *page)
1497 struct inode *inode = page->mapping->host;
1498 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1504 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1506 struct inode *inode = page->mapping->host;
1507 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1511 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1513 struct inode *inode = file->f_path.dentry->d_inode;
1515 unsigned long written;
1518 if ((ssize_t) count < 0)
1521 if (!access_ok(VERIFY_READ, buf, count))
1524 mutex_lock(&inode->i_mutex);
1529 err = generic_write_checks(file, &pos, &count, 0);
1533 err = remove_suid(file->f_path.dentry);
1537 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1540 struct page *page = NULL;
1541 unsigned long bytes, index, offset;
1545 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1546 index = pos >> PAGE_CACHE_SHIFT;
1547 bytes = PAGE_CACHE_SIZE - offset;
1552 * We don't hold page lock across copy from user -
1553 * what would it guard against? - so no deadlock here.
1554 * But it still may be a good idea to prefault below.
1557 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1562 if (PageHighMem(page)) {
1563 volatile unsigned char dummy;
1564 __get_user(dummy, buf);
1565 __get_user(dummy, buf + bytes - 1);
1567 kaddr = kmap_atomic(page, KM_USER0);
1568 left = __copy_from_user_inatomic(kaddr + offset,
1570 kunmap_atomic(kaddr, KM_USER0);
1574 left = __copy_from_user(kaddr + offset, buf, bytes);
1582 if (pos > inode->i_size)
1583 i_size_write(inode, pos);
1585 flush_dcache_page(page);
1586 set_page_dirty(page);
1587 mark_page_accessed(page);
1588 page_cache_release(page);
1598 * Our dirty pages are not counted in nr_dirty,
1599 * and we do not attempt to balance dirty pages.
1609 mutex_unlock(&inode->i_mutex);
1613 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1615 struct inode *inode = filp->f_path.dentry->d_inode;
1616 struct address_space *mapping = inode->i_mapping;
1617 unsigned long index, offset;
1619 index = *ppos >> PAGE_CACHE_SHIFT;
1620 offset = *ppos & ~PAGE_CACHE_MASK;
1623 struct page *page = NULL;
1624 unsigned long end_index, nr, ret;
1625 loff_t i_size = i_size_read(inode);
1627 end_index = i_size >> PAGE_CACHE_SHIFT;
1628 if (index > end_index)
1630 if (index == end_index) {
1631 nr = i_size & ~PAGE_CACHE_MASK;
1636 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1638 if (desc->error == -EINVAL)
1644 * We must evaluate after, since reads (unlike writes)
1645 * are called without i_mutex protection against truncate
1647 nr = PAGE_CACHE_SIZE;
1648 i_size = i_size_read(inode);
1649 end_index = i_size >> PAGE_CACHE_SHIFT;
1650 if (index == end_index) {
1651 nr = i_size & ~PAGE_CACHE_MASK;
1654 page_cache_release(page);
1662 * If users can be writing to this page using arbitrary
1663 * virtual addresses, take care about potential aliasing
1664 * before reading the page on the kernel side.
1666 if (mapping_writably_mapped(mapping))
1667 flush_dcache_page(page);
1669 * Mark the page accessed if we read the beginning.
1672 mark_page_accessed(page);
1674 page = ZERO_PAGE(0);
1675 page_cache_get(page);
1679 * Ok, we have the page, and it's up-to-date, so
1680 * now we can copy it to user space...
1682 * The actor routine returns how many bytes were actually used..
1683 * NOTE! This may not be the same as how much of a user buffer
1684 * we filled up (we may be padding etc), so we can only update
1685 * "pos" here (the actor routine has to update the user buffer
1686 * pointers and the remaining count).
1688 ret = actor(desc, page, offset, nr);
1690 index += offset >> PAGE_CACHE_SHIFT;
1691 offset &= ~PAGE_CACHE_MASK;
1693 page_cache_release(page);
1694 if (ret != nr || !desc->count)
1700 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1701 file_accessed(filp);
1704 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1706 read_descriptor_t desc;
1708 if ((ssize_t) count < 0)
1710 if (!access_ok(VERIFY_WRITE, buf, count))
1720 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1722 return desc.written;
1726 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1728 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1730 buf->f_type = TMPFS_MAGIC;
1731 buf->f_bsize = PAGE_CACHE_SIZE;
1732 buf->f_namelen = NAME_MAX;
1733 spin_lock(&sbinfo->stat_lock);
1734 if (sbinfo->max_blocks) {
1735 buf->f_blocks = sbinfo->max_blocks;
1736 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1738 if (sbinfo->max_inodes) {
1739 buf->f_files = sbinfo->max_inodes;
1740 buf->f_ffree = sbinfo->free_inodes;
1742 /* else leave those fields 0 like simple_statfs */
1743 spin_unlock(&sbinfo->stat_lock);
1748 * File creation. Allocate an inode, and we're done..
1751 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1753 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1754 int error = -ENOSPC;
1757 error = security_inode_init_security(inode, dir, NULL, NULL,
1760 if (error != -EOPNOTSUPP) {
1765 error = shmem_acl_init(inode, dir);
1770 if (dir->i_mode & S_ISGID) {
1771 inode->i_gid = dir->i_gid;
1773 inode->i_mode |= S_ISGID;
1775 dir->i_size += BOGO_DIRENT_SIZE;
1776 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1777 d_instantiate(dentry, inode);
1778 dget(dentry); /* Extra count - pin the dentry in core */
1783 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1787 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1793 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1794 struct nameidata *nd)
1796 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1802 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1804 struct inode *inode = old_dentry->d_inode;
1805 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1808 * No ordinary (disk based) filesystem counts links as inodes;
1809 * but each new link needs a new dentry, pinning lowmem, and
1810 * tmpfs dentries cannot be pruned until they are unlinked.
1812 if (sbinfo->max_inodes) {
1813 spin_lock(&sbinfo->stat_lock);
1814 if (!sbinfo->free_inodes) {
1815 spin_unlock(&sbinfo->stat_lock);
1818 sbinfo->free_inodes--;
1819 spin_unlock(&sbinfo->stat_lock);
1822 dir->i_size += BOGO_DIRENT_SIZE;
1823 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1825 atomic_inc(&inode->i_count); /* New dentry reference */
1826 dget(dentry); /* Extra pinning count for the created dentry */
1827 d_instantiate(dentry, inode);
1831 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1833 struct inode *inode = dentry->d_inode;
1835 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1836 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1837 if (sbinfo->max_inodes) {
1838 spin_lock(&sbinfo->stat_lock);
1839 sbinfo->free_inodes++;
1840 spin_unlock(&sbinfo->stat_lock);
1844 dir->i_size -= BOGO_DIRENT_SIZE;
1845 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1847 dput(dentry); /* Undo the count from "create" - this does all the work */
1851 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1853 if (!simple_empty(dentry))
1856 drop_nlink(dentry->d_inode);
1858 return shmem_unlink(dir, dentry);
1862 * The VFS layer already does all the dentry stuff for rename,
1863 * we just have to decrement the usage count for the target if
1864 * it exists so that the VFS layer correctly free's it when it
1867 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1869 struct inode *inode = old_dentry->d_inode;
1870 int they_are_dirs = S_ISDIR(inode->i_mode);
1872 if (!simple_empty(new_dentry))
1875 if (new_dentry->d_inode) {
1876 (void) shmem_unlink(new_dir, new_dentry);
1878 drop_nlink(old_dir);
1879 } else if (they_are_dirs) {
1880 drop_nlink(old_dir);
1884 old_dir->i_size -= BOGO_DIRENT_SIZE;
1885 new_dir->i_size += BOGO_DIRENT_SIZE;
1886 old_dir->i_ctime = old_dir->i_mtime =
1887 new_dir->i_ctime = new_dir->i_mtime =
1888 inode->i_ctime = CURRENT_TIME;
1892 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1896 struct inode *inode;
1897 struct page *page = NULL;
1899 struct shmem_inode_info *info;
1901 len = strlen(symname) + 1;
1902 if (len > PAGE_CACHE_SIZE)
1903 return -ENAMETOOLONG;
1905 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1909 error = security_inode_init_security(inode, dir, NULL, NULL,
1912 if (error != -EOPNOTSUPP) {
1919 info = SHMEM_I(inode);
1920 inode->i_size = len-1;
1921 if (len <= (char *)inode - (char *)info) {
1923 memcpy(info, symname, len);
1924 inode->i_op = &shmem_symlink_inline_operations;
1926 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1931 inode->i_op = &shmem_symlink_inode_operations;
1932 kaddr = kmap_atomic(page, KM_USER0);
1933 memcpy(kaddr, symname, len);
1934 kunmap_atomic(kaddr, KM_USER0);
1935 set_page_dirty(page);
1936 page_cache_release(page);
1938 if (dir->i_mode & S_ISGID)
1939 inode->i_gid = dir->i_gid;
1940 dir->i_size += BOGO_DIRENT_SIZE;
1941 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1942 d_instantiate(dentry, inode);
1947 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1949 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1953 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1955 struct page *page = NULL;
1956 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1957 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1961 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1963 if (!IS_ERR(nd_get_link(nd))) {
1964 struct page *page = cookie;
1966 mark_page_accessed(page);
1967 page_cache_release(page);
1971 static const struct inode_operations shmem_symlink_inline_operations = {
1972 .readlink = generic_readlink,
1973 .follow_link = shmem_follow_link_inline,
1976 static const struct inode_operations shmem_symlink_inode_operations = {
1977 .truncate = shmem_truncate,
1978 .readlink = generic_readlink,
1979 .follow_link = shmem_follow_link,
1980 .put_link = shmem_put_link,
1983 #ifdef CONFIG_TMPFS_POSIX_ACL
1985 * Superblocks without xattr inode operations will get security.* xattr
1986 * support from the VFS "for free". As soon as we have any other xattrs
1987 * like ACLs, we also need to implement the security.* handlers at
1988 * filesystem level, though.
1991 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1992 size_t list_len, const char *name,
1995 return security_inode_listsecurity(inode, list, list_len);
1998 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1999 void *buffer, size_t size)
2001 if (strcmp(name, "") == 0)
2003 return security_inode_getsecurity(inode, name, buffer, size,
2007 static int shmem_xattr_security_set(struct inode *inode, const char *name,
2008 const void *value, size_t size, int flags)
2010 if (strcmp(name, "") == 0)
2012 return security_inode_setsecurity(inode, name, value, size, flags);
2015 static struct xattr_handler shmem_xattr_security_handler = {
2016 .prefix = XATTR_SECURITY_PREFIX,
2017 .list = shmem_xattr_security_list,
2018 .get = shmem_xattr_security_get,
2019 .set = shmem_xattr_security_set,
2022 static struct xattr_handler *shmem_xattr_handlers[] = {
2023 &shmem_xattr_acl_access_handler,
2024 &shmem_xattr_acl_default_handler,
2025 &shmem_xattr_security_handler,
2030 static struct dentry *shmem_get_parent(struct dentry *child)
2032 return ERR_PTR(-ESTALE);
2035 static int shmem_match(struct inode *ino, void *vfh)
2039 inum = (inum << 32) | fh[1];
2040 return ino->i_ino == inum && fh[0] == ino->i_generation;
2043 static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
2045 struct dentry *de = NULL;
2046 struct inode *inode;
2049 inum = (inum << 32) | fh[1];
2051 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
2053 de = d_find_alias(inode);
2057 return de? de: ERR_PTR(-ESTALE);
2060 static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
2062 int (*acceptable)(void *context, struct dentry *de),
2066 return ERR_PTR(-ESTALE);
2068 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
2072 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2075 struct inode *inode = dentry->d_inode;
2080 if (hlist_unhashed(&inode->i_hash)) {
2081 /* Unfortunately insert_inode_hash is not idempotent,
2082 * so as we hash inodes here rather than at creation
2083 * time, we need a lock to ensure we only try
2086 static DEFINE_SPINLOCK(lock);
2088 if (hlist_unhashed(&inode->i_hash))
2089 __insert_inode_hash(inode,
2090 inode->i_ino + inode->i_generation);
2094 fh[0] = inode->i_generation;
2095 fh[1] = inode->i_ino;
2096 fh[2] = ((__u64)inode->i_ino) >> 32;
2102 static struct export_operations shmem_export_ops = {
2103 .get_parent = shmem_get_parent,
2104 .get_dentry = shmem_get_dentry,
2105 .encode_fh = shmem_encode_fh,
2106 .decode_fh = shmem_decode_fh,
2109 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2110 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2111 int *policy, nodemask_t *policy_nodes)
2113 char *this_char, *value, *rest;
2115 while (options != NULL) {
2116 this_char = options;
2119 * NUL-terminate this option: unfortunately,
2120 * mount options form a comma-separated list,
2121 * but mpol's nodelist may also contain commas.
2123 options = strchr(options, ',');
2124 if (options == NULL)
2127 if (!isdigit(*options)) {
2134 if ((value = strchr(this_char,'=')) != NULL) {
2138 "tmpfs: No value for mount option '%s'\n",
2143 if (!strcmp(this_char,"size")) {
2144 unsigned long long size;
2145 size = memparse(value,&rest);
2147 size <<= PAGE_SHIFT;
2148 size *= totalram_pages;
2154 *blocks = size >> PAGE_CACHE_SHIFT;
2155 } else if (!strcmp(this_char,"nr_blocks")) {
2156 *blocks = memparse(value,&rest);
2159 } else if (!strcmp(this_char,"nr_inodes")) {
2160 *inodes = memparse(value,&rest);
2163 } else if (!strcmp(this_char,"mode")) {
2166 *mode = simple_strtoul(value,&rest,8);
2169 } else if (!strcmp(this_char,"uid")) {
2172 *uid = simple_strtoul(value,&rest,0);
2175 } else if (!strcmp(this_char,"gid")) {
2178 *gid = simple_strtoul(value,&rest,0);
2181 } else if (!strcmp(this_char,"mpol")) {
2182 if (shmem_parse_mpol(value,policy,policy_nodes))
2185 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2193 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2199 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2201 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2202 unsigned long max_blocks = sbinfo->max_blocks;
2203 unsigned long max_inodes = sbinfo->max_inodes;
2204 int policy = sbinfo->policy;
2205 nodemask_t policy_nodes = sbinfo->policy_nodes;
2206 unsigned long blocks;
2207 unsigned long inodes;
2208 int error = -EINVAL;
2210 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2211 &max_inodes, &policy, &policy_nodes))
2214 spin_lock(&sbinfo->stat_lock);
2215 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2216 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2217 if (max_blocks < blocks)
2219 if (max_inodes < inodes)
2222 * Those tests also disallow limited->unlimited while any are in
2223 * use, so i_blocks will always be zero when max_blocks is zero;
2224 * but we must separately disallow unlimited->limited, because
2225 * in that case we have no record of how much is already in use.
2227 if (max_blocks && !sbinfo->max_blocks)
2229 if (max_inodes && !sbinfo->max_inodes)
2233 sbinfo->max_blocks = max_blocks;
2234 sbinfo->free_blocks = max_blocks - blocks;
2235 sbinfo->max_inodes = max_inodes;
2236 sbinfo->free_inodes = max_inodes - inodes;
2237 sbinfo->policy = policy;
2238 sbinfo->policy_nodes = policy_nodes;
2240 spin_unlock(&sbinfo->stat_lock);
2245 static void shmem_put_super(struct super_block *sb)
2247 kfree(sb->s_fs_info);
2248 sb->s_fs_info = NULL;
2251 static int shmem_fill_super(struct super_block *sb,
2252 void *data, int silent)
2254 struct inode *inode;
2255 struct dentry *root;
2256 int mode = S_IRWXUGO | S_ISVTX;
2257 uid_t uid = current->fsuid;
2258 gid_t gid = current->fsgid;
2260 struct shmem_sb_info *sbinfo;
2261 unsigned long blocks = 0;
2262 unsigned long inodes = 0;
2263 int policy = MPOL_DEFAULT;
2264 nodemask_t policy_nodes = node_online_map;
2268 * Per default we only allow half of the physical ram per
2269 * tmpfs instance, limiting inodes to one per page of lowmem;
2270 * but the internal instance is left unlimited.
2272 if (!(sb->s_flags & MS_NOUSER)) {
2273 blocks = totalram_pages / 2;
2274 inodes = totalram_pages - totalhigh_pages;
2275 if (inodes > blocks)
2277 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2278 &inodes, &policy, &policy_nodes))
2281 sb->s_export_op = &shmem_export_ops;
2283 sb->s_flags |= MS_NOUSER;
2286 /* Round up to L1_CACHE_BYTES to resist false sharing */
2287 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2288 L1_CACHE_BYTES), GFP_KERNEL);
2292 spin_lock_init(&sbinfo->stat_lock);
2293 sbinfo->max_blocks = blocks;
2294 sbinfo->free_blocks = blocks;
2295 sbinfo->max_inodes = inodes;
2296 sbinfo->free_inodes = inodes;
2297 sbinfo->policy = policy;
2298 sbinfo->policy_nodes = policy_nodes;
2300 sb->s_fs_info = sbinfo;
2301 sb->s_maxbytes = SHMEM_MAX_BYTES;
2302 sb->s_blocksize = PAGE_CACHE_SIZE;
2303 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2304 sb->s_magic = TMPFS_MAGIC;
2305 sb->s_op = &shmem_ops;
2306 sb->s_time_gran = 1;
2307 #ifdef CONFIG_TMPFS_POSIX_ACL
2308 sb->s_xattr = shmem_xattr_handlers;
2309 sb->s_flags |= MS_POSIXACL;
2312 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2317 root = d_alloc_root(inode);
2326 shmem_put_super(sb);
2330 static struct kmem_cache *shmem_inode_cachep;
2332 static struct inode *shmem_alloc_inode(struct super_block *sb)
2334 struct shmem_inode_info *p;
2335 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2338 return &p->vfs_inode;
2341 static void shmem_destroy_inode(struct inode *inode)
2343 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2344 /* only struct inode is valid if it's an inline symlink */
2345 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2347 shmem_acl_destroy_inode(inode);
2348 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2351 static void init_once(void *foo, struct kmem_cache *cachep,
2352 unsigned long flags)
2354 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2356 inode_init_once(&p->vfs_inode);
2357 #ifdef CONFIG_TMPFS_POSIX_ACL
2359 p->i_default_acl = NULL;
2363 static int init_inodecache(void)
2365 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2366 sizeof(struct shmem_inode_info),
2367 0, 0, init_once, NULL);
2368 if (shmem_inode_cachep == NULL)
2373 static void destroy_inodecache(void)
2375 kmem_cache_destroy(shmem_inode_cachep);
2378 static const struct address_space_operations shmem_aops = {
2379 .writepage = shmem_writepage,
2380 .set_page_dirty = __set_page_dirty_no_writeback,
2382 .readpage = shmem_readpage,
2383 .prepare_write = shmem_prepare_write,
2384 .commit_write = simple_commit_write,
2386 .migratepage = migrate_page,
2389 static const struct file_operations shmem_file_operations = {
2392 .llseek = generic_file_llseek,
2393 .read = shmem_file_read,
2394 .write = shmem_file_write,
2395 .fsync = simple_sync_file,
2396 .splice_read = generic_file_splice_read,
2397 .splice_write = generic_file_splice_write,
2401 static const struct inode_operations shmem_inode_operations = {
2402 .truncate = shmem_truncate,
2403 .setattr = shmem_notify_change,
2404 .truncate_range = shmem_truncate_range,
2405 #ifdef CONFIG_TMPFS_POSIX_ACL
2406 .setxattr = generic_setxattr,
2407 .getxattr = generic_getxattr,
2408 .listxattr = generic_listxattr,
2409 .removexattr = generic_removexattr,
2410 .permission = shmem_permission,
2415 static const struct inode_operations shmem_dir_inode_operations = {
2417 .create = shmem_create,
2418 .lookup = simple_lookup,
2420 .unlink = shmem_unlink,
2421 .symlink = shmem_symlink,
2422 .mkdir = shmem_mkdir,
2423 .rmdir = shmem_rmdir,
2424 .mknod = shmem_mknod,
2425 .rename = shmem_rename,
2427 #ifdef CONFIG_TMPFS_POSIX_ACL
2428 .setattr = shmem_notify_change,
2429 .setxattr = generic_setxattr,
2430 .getxattr = generic_getxattr,
2431 .listxattr = generic_listxattr,
2432 .removexattr = generic_removexattr,
2433 .permission = shmem_permission,
2437 static const struct inode_operations shmem_special_inode_operations = {
2438 #ifdef CONFIG_TMPFS_POSIX_ACL
2439 .setattr = shmem_notify_change,
2440 .setxattr = generic_setxattr,
2441 .getxattr = generic_getxattr,
2442 .listxattr = generic_listxattr,
2443 .removexattr = generic_removexattr,
2444 .permission = shmem_permission,
2448 static const struct super_operations shmem_ops = {
2449 .alloc_inode = shmem_alloc_inode,
2450 .destroy_inode = shmem_destroy_inode,
2452 .statfs = shmem_statfs,
2453 .remount_fs = shmem_remount_fs,
2455 .delete_inode = shmem_delete_inode,
2456 .drop_inode = generic_delete_inode,
2457 .put_super = shmem_put_super,
2460 static struct vm_operations_struct shmem_vm_ops = {
2461 .nopage = shmem_nopage,
2462 .populate = shmem_populate,
2464 .set_policy = shmem_set_policy,
2465 .get_policy = shmem_get_policy,
2470 static int shmem_get_sb(struct file_system_type *fs_type,
2471 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2473 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2476 static struct file_system_type tmpfs_fs_type = {
2477 .owner = THIS_MODULE,
2479 .get_sb = shmem_get_sb,
2480 .kill_sb = kill_litter_super,
2482 static struct vfsmount *shm_mnt;
2484 static int __init init_tmpfs(void)
2488 error = init_inodecache();
2492 error = register_filesystem(&tmpfs_fs_type);
2494 printk(KERN_ERR "Could not register tmpfs\n");
2498 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2499 tmpfs_fs_type.name, NULL);
2500 if (IS_ERR(shm_mnt)) {
2501 error = PTR_ERR(shm_mnt);
2502 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2508 unregister_filesystem(&tmpfs_fs_type);
2510 destroy_inodecache();
2512 shm_mnt = ERR_PTR(error);
2515 module_init(init_tmpfs)
2518 * shmem_file_setup - get an unlinked file living in tmpfs
2520 * @name: name for dentry (to be seen in /proc/<pid>/maps
2521 * @size: size to be set for the file
2524 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2528 struct inode *inode;
2529 struct dentry *dentry, *root;
2532 if (IS_ERR(shm_mnt))
2533 return (void *)shm_mnt;
2535 if (size < 0 || size > SHMEM_MAX_BYTES)
2536 return ERR_PTR(-EINVAL);
2538 if (shmem_acct_size(flags, size))
2539 return ERR_PTR(-ENOMEM);
2543 this.len = strlen(name);
2544 this.hash = 0; /* will go */
2545 root = shm_mnt->mnt_root;
2546 dentry = d_alloc(root, &this);
2551 file = get_empty_filp();
2556 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2560 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2561 d_instantiate(dentry, inode);
2562 inode->i_size = size;
2563 inode->i_nlink = 0; /* It is unlinked */
2564 file->f_path.mnt = mntget(shm_mnt);
2565 file->f_path.dentry = dentry;
2566 file->f_mapping = inode->i_mapping;
2567 file->f_op = &shmem_file_operations;
2568 file->f_mode = FMODE_WRITE | FMODE_READ;
2576 shmem_unacct_size(flags, size);
2577 return ERR_PTR(error);
2581 * shmem_zero_setup - setup a shared anonymous mapping
2583 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2585 int shmem_zero_setup(struct vm_area_struct *vma)
2588 loff_t size = vma->vm_end - vma->vm_start;
2590 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2592 return PTR_ERR(file);
2596 vma->vm_file = file;
2597 vma->vm_ops = &shmem_vm_ops;