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/config.h>
27 #include <linux/module.h>
28 #include <linux/init.h>
29 #include <linux/devfs_fs_kernel.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 <asm/uaccess.h>
49 #include <asm/div64.h>
50 #include <asm/pgtable.h>
52 /* This magic number is used in glibc for posix shared memory */
53 #define TMPFS_MAGIC 0x01021994
55 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
56 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
57 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
59 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
60 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
62 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
64 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
65 #define SHMEM_PAGEIN VM_READ
66 #define SHMEM_TRUNCATE VM_WRITE
68 /* Definition to limit shmem_truncate's steps between cond_rescheds */
69 #define LATENCY_LIMIT 64
71 /* Pretend that each entry is of this size in directory's i_size */
72 #define BOGO_DIRENT_SIZE 20
74 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
76 SGP_QUICK, /* don't try more than file page cache lookup */
77 SGP_READ, /* don't exceed i_size, don't allocate page */
78 SGP_CACHE, /* don't exceed i_size, may allocate page */
79 SGP_WRITE, /* may exceed i_size, may allocate page */
82 static int shmem_getpage(struct inode *inode, unsigned long idx,
83 struct page **pagep, enum sgp_type sgp, int *type);
85 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
88 * The above definition of ENTRIES_PER_PAGE, and the use of
89 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
90 * might be reconsidered if it ever diverges from PAGE_SIZE.
92 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
95 static inline void shmem_dir_free(struct page *page)
97 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
100 static struct page **shmem_dir_map(struct page *page)
102 return (struct page **)kmap_atomic(page, KM_USER0);
105 static inline void shmem_dir_unmap(struct page **dir)
107 kunmap_atomic(dir, KM_USER0);
110 static swp_entry_t *shmem_swp_map(struct page *page)
112 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
115 static inline void shmem_swp_balance_unmap(void)
118 * When passing a pointer to an i_direct entry, to code which
119 * also handles indirect entries and so will shmem_swp_unmap,
120 * we must arrange for the preempt count to remain in balance.
121 * What kmap_atomic of a lowmem page does depends on config
122 * and architecture, so pretend to kmap_atomic some lowmem page.
124 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
127 static inline void shmem_swp_unmap(swp_entry_t *entry)
129 kunmap_atomic(entry, KM_USER1);
132 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
134 return sb->s_fs_info;
138 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
139 * for shared memory and for shared anonymous (/dev/zero) mappings
140 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
141 * consistent with the pre-accounting of private mappings ...
143 static inline int shmem_acct_size(unsigned long flags, loff_t size)
145 return (flags & VM_ACCOUNT)?
146 security_vm_enough_memory(VM_ACCT(size)): 0;
149 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
151 if (flags & VM_ACCOUNT)
152 vm_unacct_memory(VM_ACCT(size));
156 * ... whereas tmpfs objects are accounted incrementally as
157 * pages are allocated, in order to allow huge sparse files.
158 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
159 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
161 static inline int shmem_acct_block(unsigned long flags)
163 return (flags & VM_ACCOUNT)?
164 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
167 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
169 if (!(flags & VM_ACCOUNT))
170 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
173 static struct super_operations shmem_ops;
174 static struct address_space_operations shmem_aops;
175 static struct file_operations shmem_file_operations;
176 static struct inode_operations shmem_inode_operations;
177 static struct inode_operations shmem_dir_inode_operations;
178 static struct vm_operations_struct shmem_vm_ops;
180 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
181 .ra_pages = 0, /* No readahead */
182 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
183 .unplug_io_fn = default_unplug_io_fn,
186 static LIST_HEAD(shmem_swaplist);
187 static DEFINE_SPINLOCK(shmem_swaplist_lock);
189 static void shmem_free_blocks(struct inode *inode, long pages)
191 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
192 if (sbinfo->max_blocks) {
193 spin_lock(&sbinfo->stat_lock);
194 sbinfo->free_blocks += pages;
195 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
196 spin_unlock(&sbinfo->stat_lock);
201 * shmem_recalc_inode - recalculate the size of an inode
203 * @inode: inode to recalc
205 * We have to calculate the free blocks since the mm can drop
206 * undirtied hole pages behind our back.
208 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
209 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
211 * It has to be called with the spinlock held.
213 static void shmem_recalc_inode(struct inode *inode)
215 struct shmem_inode_info *info = SHMEM_I(inode);
218 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
220 info->alloced -= freed;
221 shmem_unacct_blocks(info->flags, freed);
222 shmem_free_blocks(inode, freed);
227 * shmem_swp_entry - find the swap vector position in the info structure
229 * @info: info structure for the inode
230 * @index: index of the page to find
231 * @page: optional page to add to the structure. Has to be preset to
234 * If there is no space allocated yet it will return NULL when
235 * page is NULL, else it will use the page for the needed block,
236 * setting it to NULL on return to indicate that it has been used.
238 * The swap vector is organized the following way:
240 * There are SHMEM_NR_DIRECT entries directly stored in the
241 * shmem_inode_info structure. So small files do not need an addional
244 * For pages with index > SHMEM_NR_DIRECT there is the pointer
245 * i_indirect which points to a page which holds in the first half
246 * doubly indirect blocks, in the second half triple indirect blocks:
248 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
249 * following layout (for SHMEM_NR_DIRECT == 16):
251 * i_indirect -> dir --> 16-19
264 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
266 unsigned long offset;
270 if (index < SHMEM_NR_DIRECT) {
271 shmem_swp_balance_unmap();
272 return info->i_direct+index;
274 if (!info->i_indirect) {
276 info->i_indirect = *page;
279 return NULL; /* need another page */
282 index -= SHMEM_NR_DIRECT;
283 offset = index % ENTRIES_PER_PAGE;
284 index /= ENTRIES_PER_PAGE;
285 dir = shmem_dir_map(info->i_indirect);
287 if (index >= ENTRIES_PER_PAGE/2) {
288 index -= ENTRIES_PER_PAGE/2;
289 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
290 index %= ENTRIES_PER_PAGE;
297 shmem_dir_unmap(dir);
298 return NULL; /* need another page */
300 shmem_dir_unmap(dir);
301 dir = shmem_dir_map(subdir);
307 if (!page || !(subdir = *page)) {
308 shmem_dir_unmap(dir);
309 return NULL; /* need a page */
314 shmem_dir_unmap(dir);
315 return shmem_swp_map(subdir) + offset;
318 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
320 long incdec = value? 1: -1;
323 info->swapped += incdec;
324 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
325 struct page *page = kmap_atomic_to_page(entry);
326 set_page_private(page, page_private(page) + incdec);
331 * shmem_swp_alloc - get the position of the swap entry for the page.
332 * If it does not exist allocate the entry.
334 * @info: info structure for the inode
335 * @index: index of the page to find
336 * @sgp: check and recheck i_size? skip allocation?
338 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
340 struct inode *inode = &info->vfs_inode;
341 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
342 struct page *page = NULL;
345 if (sgp != SGP_WRITE &&
346 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
347 return ERR_PTR(-EINVAL);
349 while (!(entry = shmem_swp_entry(info, index, &page))) {
351 return shmem_swp_map(ZERO_PAGE(0));
353 * Test free_blocks against 1 not 0, since we have 1 data
354 * page (and perhaps indirect index pages) yet to allocate:
355 * a waste to allocate index if we cannot allocate data.
357 if (sbinfo->max_blocks) {
358 spin_lock(&sbinfo->stat_lock);
359 if (sbinfo->free_blocks <= 1) {
360 spin_unlock(&sbinfo->stat_lock);
361 return ERR_PTR(-ENOSPC);
363 sbinfo->free_blocks--;
364 inode->i_blocks += BLOCKS_PER_PAGE;
365 spin_unlock(&sbinfo->stat_lock);
368 spin_unlock(&info->lock);
369 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
371 set_page_private(page, 0);
372 spin_lock(&info->lock);
375 shmem_free_blocks(inode, 1);
376 return ERR_PTR(-ENOMEM);
378 if (sgp != SGP_WRITE &&
379 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
380 entry = ERR_PTR(-EINVAL);
383 if (info->next_index <= index)
384 info->next_index = index + 1;
387 /* another task gave its page, or truncated the file */
388 shmem_free_blocks(inode, 1);
389 shmem_dir_free(page);
391 if (info->next_index <= index && !IS_ERR(entry))
392 info->next_index = index + 1;
397 * shmem_free_swp - free some swap entries in a directory
399 * @dir: pointer to the directory
400 * @edir: pointer after last entry of the directory
402 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
407 for (ptr = dir; ptr < edir; ptr++) {
409 free_swap_and_cache(*ptr);
410 *ptr = (swp_entry_t){0};
417 static int shmem_map_and_free_swp(struct page *subdir,
418 int offset, int limit, struct page ***dir)
423 ptr = shmem_swp_map(subdir);
424 for (; offset < limit; offset += LATENCY_LIMIT) {
425 int size = limit - offset;
426 if (size > LATENCY_LIMIT)
427 size = LATENCY_LIMIT;
428 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
429 if (need_resched()) {
430 shmem_swp_unmap(ptr);
432 shmem_dir_unmap(*dir);
436 ptr = shmem_swp_map(subdir);
439 shmem_swp_unmap(ptr);
443 static void shmem_free_pages(struct list_head *next)
449 page = container_of(next, struct page, lru);
451 shmem_dir_free(page);
453 if (freed >= LATENCY_LIMIT) {
460 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
462 struct shmem_inode_info *info = SHMEM_I(inode);
467 unsigned long diroff;
473 LIST_HEAD(pages_to_free);
474 long nr_pages_to_free = 0;
475 long nr_swaps_freed = 0;
480 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
481 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
482 if (idx >= info->next_index)
485 spin_lock(&info->lock);
486 info->flags |= SHMEM_TRUNCATE;
487 if (likely(end == (loff_t) -1)) {
488 limit = info->next_index;
489 info->next_index = idx;
491 limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
492 if (limit > info->next_index)
493 limit = info->next_index;
497 topdir = info->i_indirect;
498 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
499 info->i_indirect = NULL;
501 list_add(&topdir->lru, &pages_to_free);
503 spin_unlock(&info->lock);
505 if (info->swapped && idx < SHMEM_NR_DIRECT) {
506 ptr = info->i_direct;
508 if (size > SHMEM_NR_DIRECT)
509 size = SHMEM_NR_DIRECT;
510 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
515 BUG_ON(limit <= SHMEM_NR_DIRECT);
516 limit -= SHMEM_NR_DIRECT;
517 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
518 offset = idx % ENTRIES_PER_PAGE;
521 dir = shmem_dir_map(topdir);
522 stage = ENTRIES_PER_PAGEPAGE/2;
523 if (idx < ENTRIES_PER_PAGEPAGE/2) {
525 diroff = idx/ENTRIES_PER_PAGE;
527 dir += ENTRIES_PER_PAGE/2;
528 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
530 stage += ENTRIES_PER_PAGEPAGE;
533 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
534 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
535 if (!diroff && !offset) {
538 list_add(&middir->lru, &pages_to_free);
540 shmem_dir_unmap(dir);
541 dir = shmem_dir_map(middir);
549 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
550 if (unlikely(idx == stage)) {
551 shmem_dir_unmap(dir);
552 dir = shmem_dir_map(topdir) +
553 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
556 idx += ENTRIES_PER_PAGEPAGE;
560 stage = idx + ENTRIES_PER_PAGEPAGE;
564 list_add(&middir->lru, &pages_to_free);
565 shmem_dir_unmap(dir);
567 dir = shmem_dir_map(middir);
570 subdir = dir[diroff];
571 if (subdir && page_private(subdir)) {
573 if (size > ENTRIES_PER_PAGE)
574 size = ENTRIES_PER_PAGE;
575 freed = shmem_map_and_free_swp(subdir,
578 dir = shmem_dir_map(middir);
579 nr_swaps_freed += freed;
581 spin_lock(&info->lock);
582 set_page_private(subdir, page_private(subdir) - freed);
584 spin_unlock(&info->lock);
586 BUG_ON(page_private(subdir) > offset);
590 else if (subdir && !page_private(subdir)) {
593 list_add(&subdir->lru, &pages_to_free);
597 shmem_dir_unmap(dir);
599 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
601 * Call truncate_inode_pages again: racing shmem_unuse_inode
602 * may have swizzled a page in from swap since vmtruncate or
603 * generic_delete_inode did it, before we lowered next_index.
604 * Also, though shmem_getpage checks i_size before adding to
605 * cache, no recheck after: so fix the narrow window there too.
607 truncate_inode_pages_range(inode->i_mapping, start, end);
610 spin_lock(&info->lock);
611 info->flags &= ~SHMEM_TRUNCATE;
612 info->swapped -= nr_swaps_freed;
613 if (nr_pages_to_free)
614 shmem_free_blocks(inode, nr_pages_to_free);
615 shmem_recalc_inode(inode);
616 spin_unlock(&info->lock);
619 * Empty swap vector directory pages to be freed?
621 if (!list_empty(&pages_to_free)) {
622 pages_to_free.prev->next = NULL;
623 shmem_free_pages(pages_to_free.next);
627 static void shmem_truncate(struct inode *inode)
629 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
632 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
634 struct inode *inode = dentry->d_inode;
635 struct page *page = NULL;
638 if (attr->ia_valid & ATTR_SIZE) {
639 if (attr->ia_size < inode->i_size) {
641 * If truncating down to a partial page, then
642 * if that page is already allocated, hold it
643 * in memory until the truncation is over, so
644 * truncate_partial_page cannnot miss it were
645 * it assigned to swap.
647 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
648 (void) shmem_getpage(inode,
649 attr->ia_size>>PAGE_CACHE_SHIFT,
650 &page, SGP_READ, NULL);
653 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
654 * detect if any pages might have been added to cache
655 * after truncate_inode_pages. But we needn't bother
656 * if it's being fully truncated to zero-length: the
657 * nrpages check is efficient enough in that case.
660 struct shmem_inode_info *info = SHMEM_I(inode);
661 spin_lock(&info->lock);
662 info->flags &= ~SHMEM_PAGEIN;
663 spin_unlock(&info->lock);
668 error = inode_change_ok(inode, attr);
670 error = inode_setattr(inode, attr);
672 page_cache_release(page);
676 static void shmem_delete_inode(struct inode *inode)
678 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
679 struct shmem_inode_info *info = SHMEM_I(inode);
681 if (inode->i_op->truncate == shmem_truncate) {
682 truncate_inode_pages(inode->i_mapping, 0);
683 shmem_unacct_size(info->flags, inode->i_size);
685 shmem_truncate(inode);
686 if (!list_empty(&info->swaplist)) {
687 spin_lock(&shmem_swaplist_lock);
688 list_del_init(&info->swaplist);
689 spin_unlock(&shmem_swaplist_lock);
692 BUG_ON(inode->i_blocks);
693 if (sbinfo->max_inodes) {
694 spin_lock(&sbinfo->stat_lock);
695 sbinfo->free_inodes++;
696 spin_unlock(&sbinfo->stat_lock);
701 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
705 for (ptr = dir; ptr < edir; ptr++) {
706 if (ptr->val == entry.val)
712 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
725 ptr = info->i_direct;
726 spin_lock(&info->lock);
727 limit = info->next_index;
729 if (size > SHMEM_NR_DIRECT)
730 size = SHMEM_NR_DIRECT;
731 offset = shmem_find_swp(entry, ptr, ptr+size);
733 shmem_swp_balance_unmap();
736 if (!info->i_indirect)
739 dir = shmem_dir_map(info->i_indirect);
740 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
742 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
743 if (unlikely(idx == stage)) {
744 shmem_dir_unmap(dir-1);
745 dir = shmem_dir_map(info->i_indirect) +
746 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
749 idx += ENTRIES_PER_PAGEPAGE;
753 stage = idx + ENTRIES_PER_PAGEPAGE;
755 shmem_dir_unmap(dir);
756 dir = shmem_dir_map(subdir);
759 if (subdir && page_private(subdir)) {
760 ptr = shmem_swp_map(subdir);
762 if (size > ENTRIES_PER_PAGE)
763 size = ENTRIES_PER_PAGE;
764 offset = shmem_find_swp(entry, ptr, ptr+size);
766 shmem_dir_unmap(dir);
769 shmem_swp_unmap(ptr);
773 shmem_dir_unmap(dir-1);
775 spin_unlock(&info->lock);
779 inode = &info->vfs_inode;
780 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
781 info->flags |= SHMEM_PAGEIN;
782 shmem_swp_set(info, ptr + offset, 0);
784 shmem_swp_unmap(ptr);
785 spin_unlock(&info->lock);
787 * Decrement swap count even when the entry is left behind:
788 * try_to_unuse will skip over mms, then reincrement count.
795 * shmem_unuse() search for an eventually swapped out shmem page.
797 int shmem_unuse(swp_entry_t entry, struct page *page)
799 struct list_head *p, *next;
800 struct shmem_inode_info *info;
803 spin_lock(&shmem_swaplist_lock);
804 list_for_each_safe(p, next, &shmem_swaplist) {
805 info = list_entry(p, struct shmem_inode_info, swaplist);
807 list_del_init(&info->swaplist);
808 else if (shmem_unuse_inode(info, entry, page)) {
809 /* move head to start search for next from here */
810 list_move_tail(&shmem_swaplist, &info->swaplist);
815 spin_unlock(&shmem_swaplist_lock);
820 * Move the page from the page cache to the swap cache.
822 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
824 struct shmem_inode_info *info;
825 swp_entry_t *entry, swap;
826 struct address_space *mapping;
830 BUG_ON(!PageLocked(page));
831 BUG_ON(page_mapped(page));
833 mapping = page->mapping;
835 inode = mapping->host;
836 info = SHMEM_I(inode);
837 if (info->flags & VM_LOCKED)
839 swap = get_swap_page();
843 spin_lock(&info->lock);
844 shmem_recalc_inode(inode);
845 if (index >= info->next_index) {
846 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
849 entry = shmem_swp_entry(info, index, NULL);
853 if (move_to_swap_cache(page, swap) == 0) {
854 shmem_swp_set(info, entry, swap.val);
855 shmem_swp_unmap(entry);
856 spin_unlock(&info->lock);
857 if (list_empty(&info->swaplist)) {
858 spin_lock(&shmem_swaplist_lock);
859 /* move instead of add in case we're racing */
860 list_move_tail(&info->swaplist, &shmem_swaplist);
861 spin_unlock(&shmem_swaplist_lock);
867 shmem_swp_unmap(entry);
869 spin_unlock(&info->lock);
872 set_page_dirty(page);
873 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
877 static struct page *shmem_swapin_async(struct shared_policy *p,
878 swp_entry_t entry, unsigned long idx)
881 struct vm_area_struct pvma;
883 /* Create a pseudo vma that just contains the policy */
884 memset(&pvma, 0, sizeof(struct vm_area_struct));
885 pvma.vm_end = PAGE_SIZE;
887 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
888 page = read_swap_cache_async(entry, &pvma, 0);
889 mpol_free(pvma.vm_policy);
893 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
896 struct shared_policy *p = &info->policy;
899 unsigned long offset;
901 num = valid_swaphandles(entry, &offset);
902 for (i = 0; i < num; offset++, i++) {
903 page = shmem_swapin_async(p,
904 swp_entry(swp_type(entry), offset), idx);
907 page_cache_release(page);
909 lru_add_drain(); /* Push any new pages onto the LRU now */
910 return shmem_swapin_async(p, entry, idx);
914 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
917 struct vm_area_struct pvma;
920 memset(&pvma, 0, sizeof(struct vm_area_struct));
921 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
923 pvma.vm_end = PAGE_SIZE;
924 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
925 mpol_free(pvma.vm_policy);
929 static inline struct page *
930 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
932 swapin_readahead(entry, 0, NULL);
933 return read_swap_cache_async(entry, NULL, 0);
936 static inline struct page *
937 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
939 return alloc_page(gfp | __GFP_ZERO);
944 * shmem_getpage - either get the page from swap or allocate a new one
946 * If we allocate a new one we do not mark it dirty. That's up to the
947 * vm. If we swap it in we mark it dirty since we also free the swap
948 * entry since a page cannot live in both the swap and page cache
950 static int shmem_getpage(struct inode *inode, unsigned long idx,
951 struct page **pagep, enum sgp_type sgp, int *type)
953 struct address_space *mapping = inode->i_mapping;
954 struct shmem_inode_info *info = SHMEM_I(inode);
955 struct shmem_sb_info *sbinfo;
956 struct page *filepage = *pagep;
957 struct page *swappage;
962 if (idx >= SHMEM_MAX_INDEX)
965 * Normally, filepage is NULL on entry, and either found
966 * uptodate immediately, or allocated and zeroed, or read
967 * in under swappage, which is then assigned to filepage.
968 * But shmem_prepare_write passes in a locked filepage,
969 * which may be found not uptodate by other callers too,
970 * and may need to be copied from the swappage read in.
974 filepage = find_lock_page(mapping, idx);
975 if (filepage && PageUptodate(filepage))
978 if (sgp == SGP_QUICK)
981 spin_lock(&info->lock);
982 shmem_recalc_inode(inode);
983 entry = shmem_swp_alloc(info, idx, sgp);
985 spin_unlock(&info->lock);
986 error = PTR_ERR(entry);
992 /* Look it up and read it in.. */
993 swappage = lookup_swap_cache(swap);
995 shmem_swp_unmap(entry);
996 spin_unlock(&info->lock);
997 /* here we actually do the io */
998 if (type && *type == VM_FAULT_MINOR) {
999 inc_page_state(pgmajfault);
1000 *type = VM_FAULT_MAJOR;
1002 swappage = shmem_swapin(info, swap, idx);
1004 spin_lock(&info->lock);
1005 entry = shmem_swp_alloc(info, idx, sgp);
1007 error = PTR_ERR(entry);
1009 if (entry->val == swap.val)
1011 shmem_swp_unmap(entry);
1013 spin_unlock(&info->lock);
1018 wait_on_page_locked(swappage);
1019 page_cache_release(swappage);
1023 /* We have to do this with page locked to prevent races */
1024 if (TestSetPageLocked(swappage)) {
1025 shmem_swp_unmap(entry);
1026 spin_unlock(&info->lock);
1027 wait_on_page_locked(swappage);
1028 page_cache_release(swappage);
1031 if (PageWriteback(swappage)) {
1032 shmem_swp_unmap(entry);
1033 spin_unlock(&info->lock);
1034 wait_on_page_writeback(swappage);
1035 unlock_page(swappage);
1036 page_cache_release(swappage);
1039 if (!PageUptodate(swappage)) {
1040 shmem_swp_unmap(entry);
1041 spin_unlock(&info->lock);
1042 unlock_page(swappage);
1043 page_cache_release(swappage);
1049 shmem_swp_set(info, entry, 0);
1050 shmem_swp_unmap(entry);
1051 delete_from_swap_cache(swappage);
1052 spin_unlock(&info->lock);
1053 copy_highpage(filepage, swappage);
1054 unlock_page(swappage);
1055 page_cache_release(swappage);
1056 flush_dcache_page(filepage);
1057 SetPageUptodate(filepage);
1058 set_page_dirty(filepage);
1060 } else if (!(error = move_from_swap_cache(
1061 swappage, idx, mapping))) {
1062 info->flags |= SHMEM_PAGEIN;
1063 shmem_swp_set(info, entry, 0);
1064 shmem_swp_unmap(entry);
1065 spin_unlock(&info->lock);
1066 filepage = swappage;
1069 shmem_swp_unmap(entry);
1070 spin_unlock(&info->lock);
1071 unlock_page(swappage);
1072 page_cache_release(swappage);
1073 if (error == -ENOMEM) {
1074 /* let kswapd refresh zone for GFP_ATOMICs */
1075 blk_congestion_wait(WRITE, HZ/50);
1079 } else if (sgp == SGP_READ && !filepage) {
1080 shmem_swp_unmap(entry);
1081 filepage = find_get_page(mapping, idx);
1083 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1084 spin_unlock(&info->lock);
1085 wait_on_page_locked(filepage);
1086 page_cache_release(filepage);
1090 spin_unlock(&info->lock);
1092 shmem_swp_unmap(entry);
1093 sbinfo = SHMEM_SB(inode->i_sb);
1094 if (sbinfo->max_blocks) {
1095 spin_lock(&sbinfo->stat_lock);
1096 if (sbinfo->free_blocks == 0 ||
1097 shmem_acct_block(info->flags)) {
1098 spin_unlock(&sbinfo->stat_lock);
1099 spin_unlock(&info->lock);
1103 sbinfo->free_blocks--;
1104 inode->i_blocks += BLOCKS_PER_PAGE;
1105 spin_unlock(&sbinfo->stat_lock);
1106 } else if (shmem_acct_block(info->flags)) {
1107 spin_unlock(&info->lock);
1113 spin_unlock(&info->lock);
1114 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1118 shmem_unacct_blocks(info->flags, 1);
1119 shmem_free_blocks(inode, 1);
1124 spin_lock(&info->lock);
1125 entry = shmem_swp_alloc(info, idx, sgp);
1127 error = PTR_ERR(entry);
1130 shmem_swp_unmap(entry);
1132 if (error || swap.val || 0 != add_to_page_cache_lru(
1133 filepage, mapping, idx, GFP_ATOMIC)) {
1134 spin_unlock(&info->lock);
1135 page_cache_release(filepage);
1136 shmem_unacct_blocks(info->flags, 1);
1137 shmem_free_blocks(inode, 1);
1143 info->flags |= SHMEM_PAGEIN;
1147 spin_unlock(&info->lock);
1148 flush_dcache_page(filepage);
1149 SetPageUptodate(filepage);
1152 if (*pagep != filepage) {
1153 unlock_page(filepage);
1159 if (*pagep != filepage) {
1160 unlock_page(filepage);
1161 page_cache_release(filepage);
1166 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1168 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1169 struct page *page = NULL;
1173 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1174 idx += vma->vm_pgoff;
1175 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1176 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1177 return NOPAGE_SIGBUS;
1179 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1181 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1183 mark_page_accessed(page);
1187 static int shmem_populate(struct vm_area_struct *vma,
1188 unsigned long addr, unsigned long len,
1189 pgprot_t prot, unsigned long pgoff, int nonblock)
1191 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1192 struct mm_struct *mm = vma->vm_mm;
1193 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1196 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1197 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1200 while ((long) len > 0) {
1201 struct page *page = NULL;
1204 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1206 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1209 /* Page may still be null, but only if nonblock was set. */
1211 mark_page_accessed(page);
1212 err = install_page(mm, vma, addr, page, prot);
1214 page_cache_release(page);
1217 } else if (vma->vm_flags & VM_NONLINEAR) {
1218 /* No page was found just because we can't read it in
1219 * now (being here implies nonblock != 0), but the page
1220 * may exist, so set the PTE to fault it in later. */
1221 err = install_file_pte(mm, vma, addr, pgoff, prot);
1234 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1236 struct inode *i = vma->vm_file->f_dentry->d_inode;
1237 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1241 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1243 struct inode *i = vma->vm_file->f_dentry->d_inode;
1246 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1247 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1251 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1253 struct inode *inode = file->f_dentry->d_inode;
1254 struct shmem_inode_info *info = SHMEM_I(inode);
1255 int retval = -ENOMEM;
1257 spin_lock(&info->lock);
1258 if (lock && !(info->flags & VM_LOCKED)) {
1259 if (!user_shm_lock(inode->i_size, user))
1261 info->flags |= VM_LOCKED;
1263 if (!lock && (info->flags & VM_LOCKED) && user) {
1264 user_shm_unlock(inode->i_size, user);
1265 info->flags &= ~VM_LOCKED;
1269 spin_unlock(&info->lock);
1273 int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1275 file_accessed(file);
1276 vma->vm_ops = &shmem_vm_ops;
1280 static struct inode *
1281 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1283 struct inode *inode;
1284 struct shmem_inode_info *info;
1285 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1287 if (sbinfo->max_inodes) {
1288 spin_lock(&sbinfo->stat_lock);
1289 if (!sbinfo->free_inodes) {
1290 spin_unlock(&sbinfo->stat_lock);
1293 sbinfo->free_inodes--;
1294 spin_unlock(&sbinfo->stat_lock);
1297 inode = new_inode(sb);
1299 inode->i_mode = mode;
1300 inode->i_uid = current->fsuid;
1301 inode->i_gid = current->fsgid;
1302 inode->i_blksize = PAGE_CACHE_SIZE;
1303 inode->i_blocks = 0;
1304 inode->i_mapping->a_ops = &shmem_aops;
1305 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1306 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1307 info = SHMEM_I(inode);
1308 memset(info, 0, (char *)inode - (char *)info);
1309 spin_lock_init(&info->lock);
1310 INIT_LIST_HEAD(&info->swaplist);
1312 switch (mode & S_IFMT) {
1314 init_special_inode(inode, mode, dev);
1317 inode->i_op = &shmem_inode_operations;
1318 inode->i_fop = &shmem_file_operations;
1319 mpol_shared_policy_init(&info->policy);
1323 /* Some things misbehave if size == 0 on a directory */
1324 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1325 inode->i_op = &shmem_dir_inode_operations;
1326 inode->i_fop = &simple_dir_operations;
1330 * Must not load anything in the rbtree,
1331 * mpol_free_shared_policy will not be called.
1333 mpol_shared_policy_init(&info->policy);
1336 } else if (sbinfo->max_inodes) {
1337 spin_lock(&sbinfo->stat_lock);
1338 sbinfo->free_inodes++;
1339 spin_unlock(&sbinfo->stat_lock);
1345 static struct inode_operations shmem_symlink_inode_operations;
1346 static struct inode_operations shmem_symlink_inline_operations;
1349 * Normally tmpfs makes no use of shmem_prepare_write, but it
1350 * lets a tmpfs file be used read-write below the loop driver.
1353 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1355 struct inode *inode = page->mapping->host;
1356 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1360 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1362 struct inode *inode = file->f_dentry->d_inode;
1364 unsigned long written;
1367 if ((ssize_t) count < 0)
1370 if (!access_ok(VERIFY_READ, buf, count))
1373 down(&inode->i_sem);
1378 err = generic_write_checks(file, &pos, &count, 0);
1382 err = remove_suid(file->f_dentry);
1386 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1389 struct page *page = NULL;
1390 unsigned long bytes, index, offset;
1394 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1395 index = pos >> PAGE_CACHE_SHIFT;
1396 bytes = PAGE_CACHE_SIZE - offset;
1401 * We don't hold page lock across copy from user -
1402 * what would it guard against? - so no deadlock here.
1403 * But it still may be a good idea to prefault below.
1406 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1411 if (PageHighMem(page)) {
1412 volatile unsigned char dummy;
1413 __get_user(dummy, buf);
1414 __get_user(dummy, buf + bytes - 1);
1416 kaddr = kmap_atomic(page, KM_USER0);
1417 left = __copy_from_user_inatomic(kaddr + offset,
1419 kunmap_atomic(kaddr, KM_USER0);
1423 left = __copy_from_user(kaddr + offset, buf, bytes);
1431 if (pos > inode->i_size)
1432 i_size_write(inode, pos);
1434 flush_dcache_page(page);
1435 set_page_dirty(page);
1436 mark_page_accessed(page);
1437 page_cache_release(page);
1447 * Our dirty pages are not counted in nr_dirty,
1448 * and we do not attempt to balance dirty pages.
1462 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1464 struct inode *inode = filp->f_dentry->d_inode;
1465 struct address_space *mapping = inode->i_mapping;
1466 unsigned long index, offset;
1468 index = *ppos >> PAGE_CACHE_SHIFT;
1469 offset = *ppos & ~PAGE_CACHE_MASK;
1472 struct page *page = NULL;
1473 unsigned long end_index, nr, ret;
1474 loff_t i_size = i_size_read(inode);
1476 end_index = i_size >> PAGE_CACHE_SHIFT;
1477 if (index > end_index)
1479 if (index == end_index) {
1480 nr = i_size & ~PAGE_CACHE_MASK;
1485 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1487 if (desc->error == -EINVAL)
1493 * We must evaluate after, since reads (unlike writes)
1494 * are called without i_sem protection against truncate
1496 nr = PAGE_CACHE_SIZE;
1497 i_size = i_size_read(inode);
1498 end_index = i_size >> PAGE_CACHE_SHIFT;
1499 if (index == end_index) {
1500 nr = i_size & ~PAGE_CACHE_MASK;
1503 page_cache_release(page);
1511 * If users can be writing to this page using arbitrary
1512 * virtual addresses, take care about potential aliasing
1513 * before reading the page on the kernel side.
1515 if (mapping_writably_mapped(mapping))
1516 flush_dcache_page(page);
1518 * Mark the page accessed if we read the beginning.
1521 mark_page_accessed(page);
1523 page = ZERO_PAGE(0);
1524 page_cache_get(page);
1528 * Ok, we have the page, and it's up-to-date, so
1529 * now we can copy it to user space...
1531 * The actor routine returns how many bytes were actually used..
1532 * NOTE! This may not be the same as how much of a user buffer
1533 * we filled up (we may be padding etc), so we can only update
1534 * "pos" here (the actor routine has to update the user buffer
1535 * pointers and the remaining count).
1537 ret = actor(desc, page, offset, nr);
1539 index += offset >> PAGE_CACHE_SHIFT;
1540 offset &= ~PAGE_CACHE_MASK;
1542 page_cache_release(page);
1543 if (ret != nr || !desc->count)
1549 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1550 file_accessed(filp);
1553 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1555 read_descriptor_t desc;
1557 if ((ssize_t) count < 0)
1559 if (!access_ok(VERIFY_WRITE, buf, count))
1569 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1571 return desc.written;
1575 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1576 size_t count, read_actor_t actor, void *target)
1578 read_descriptor_t desc;
1585 desc.arg.data = target;
1588 do_shmem_file_read(in_file, ppos, &desc, actor);
1590 return desc.written;
1594 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1596 struct shmem_sb_info *sbinfo = SHMEM_SB(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) {
1634 if (dir->i_mode & S_ISGID) {
1635 inode->i_gid = dir->i_gid;
1637 inode->i_mode |= S_ISGID;
1639 dir->i_size += BOGO_DIRENT_SIZE;
1640 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1641 d_instantiate(dentry, inode);
1642 dget(dentry); /* Extra count - pin the dentry in core */
1647 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1651 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1657 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1658 struct nameidata *nd)
1660 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1666 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1668 struct inode *inode = old_dentry->d_inode;
1669 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1672 * No ordinary (disk based) filesystem counts links as inodes;
1673 * but each new link needs a new dentry, pinning lowmem, and
1674 * tmpfs dentries cannot be pruned until they are unlinked.
1676 if (sbinfo->max_inodes) {
1677 spin_lock(&sbinfo->stat_lock);
1678 if (!sbinfo->free_inodes) {
1679 spin_unlock(&sbinfo->stat_lock);
1682 sbinfo->free_inodes--;
1683 spin_unlock(&sbinfo->stat_lock);
1686 dir->i_size += BOGO_DIRENT_SIZE;
1687 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1689 atomic_inc(&inode->i_count); /* New dentry reference */
1690 dget(dentry); /* Extra pinning count for the created dentry */
1691 d_instantiate(dentry, inode);
1695 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1697 struct inode *inode = dentry->d_inode;
1699 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1700 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1701 if (sbinfo->max_inodes) {
1702 spin_lock(&sbinfo->stat_lock);
1703 sbinfo->free_inodes++;
1704 spin_unlock(&sbinfo->stat_lock);
1708 dir->i_size -= BOGO_DIRENT_SIZE;
1709 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1711 dput(dentry); /* Undo the count from "create" - this does all the work */
1715 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1717 if (!simple_empty(dentry))
1721 return shmem_unlink(dir, dentry);
1725 * The VFS layer already does all the dentry stuff for rename,
1726 * we just have to decrement the usage count for the target if
1727 * it exists so that the VFS layer correctly free's it when it
1730 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1732 struct inode *inode = old_dentry->d_inode;
1733 int they_are_dirs = S_ISDIR(inode->i_mode);
1735 if (!simple_empty(new_dentry))
1738 if (new_dentry->d_inode) {
1739 (void) shmem_unlink(new_dir, new_dentry);
1742 } else if (they_are_dirs) {
1747 old_dir->i_size -= BOGO_DIRENT_SIZE;
1748 new_dir->i_size += BOGO_DIRENT_SIZE;
1749 old_dir->i_ctime = old_dir->i_mtime =
1750 new_dir->i_ctime = new_dir->i_mtime =
1751 inode->i_ctime = CURRENT_TIME;
1755 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1759 struct inode *inode;
1760 struct page *page = NULL;
1762 struct shmem_inode_info *info;
1764 len = strlen(symname) + 1;
1765 if (len > PAGE_CACHE_SIZE)
1766 return -ENAMETOOLONG;
1768 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1772 error = security_inode_init_security(inode, dir, NULL, NULL,
1775 if (error != -EOPNOTSUPP) {
1782 info = SHMEM_I(inode);
1783 inode->i_size = len-1;
1784 if (len <= (char *)inode - (char *)info) {
1786 memcpy(info, symname, len);
1787 inode->i_op = &shmem_symlink_inline_operations;
1789 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1794 inode->i_op = &shmem_symlink_inode_operations;
1795 kaddr = kmap_atomic(page, KM_USER0);
1796 memcpy(kaddr, symname, len);
1797 kunmap_atomic(kaddr, KM_USER0);
1798 set_page_dirty(page);
1799 page_cache_release(page);
1801 if (dir->i_mode & S_ISGID)
1802 inode->i_gid = dir->i_gid;
1803 dir->i_size += BOGO_DIRENT_SIZE;
1804 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1805 d_instantiate(dentry, inode);
1810 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1812 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1816 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1818 struct page *page = NULL;
1819 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1820 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1824 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1826 if (!IS_ERR(nd_get_link(nd))) {
1827 struct page *page = cookie;
1829 mark_page_accessed(page);
1830 page_cache_release(page);
1834 static struct inode_operations shmem_symlink_inline_operations = {
1835 .readlink = generic_readlink,
1836 .follow_link = shmem_follow_link_inline,
1839 static struct inode_operations shmem_symlink_inode_operations = {
1840 .truncate = shmem_truncate,
1841 .readlink = generic_readlink,
1842 .follow_link = shmem_follow_link,
1843 .put_link = shmem_put_link,
1846 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1848 char *this_char, *value, *rest;
1850 while ((this_char = strsep(&options, ",")) != NULL) {
1853 if ((value = strchr(this_char,'=')) != NULL) {
1857 "tmpfs: No value for mount option '%s'\n",
1862 if (!strcmp(this_char,"size")) {
1863 unsigned long long size;
1864 size = memparse(value,&rest);
1866 size <<= PAGE_SHIFT;
1867 size *= totalram_pages;
1873 *blocks = size >> PAGE_CACHE_SHIFT;
1874 } else if (!strcmp(this_char,"nr_blocks")) {
1875 *blocks = memparse(value,&rest);
1878 } else if (!strcmp(this_char,"nr_inodes")) {
1879 *inodes = memparse(value,&rest);
1882 } else if (!strcmp(this_char,"mode")) {
1885 *mode = simple_strtoul(value,&rest,8);
1888 } else if (!strcmp(this_char,"uid")) {
1891 *uid = simple_strtoul(value,&rest,0);
1894 } else if (!strcmp(this_char,"gid")) {
1897 *gid = simple_strtoul(value,&rest,0);
1901 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1909 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1915 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1917 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1918 unsigned long max_blocks = sbinfo->max_blocks;
1919 unsigned long max_inodes = sbinfo->max_inodes;
1920 unsigned long blocks;
1921 unsigned long inodes;
1922 int error = -EINVAL;
1924 if (shmem_parse_options(data, NULL, NULL, NULL,
1925 &max_blocks, &max_inodes))
1928 spin_lock(&sbinfo->stat_lock);
1929 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
1930 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1931 if (max_blocks < blocks)
1933 if (max_inodes < inodes)
1936 * Those tests also disallow limited->unlimited while any are in
1937 * use, so i_blocks will always be zero when max_blocks is zero;
1938 * but we must separately disallow unlimited->limited, because
1939 * in that case we have no record of how much is already in use.
1941 if (max_blocks && !sbinfo->max_blocks)
1943 if (max_inodes && !sbinfo->max_inodes)
1947 sbinfo->max_blocks = max_blocks;
1948 sbinfo->free_blocks = max_blocks - blocks;
1949 sbinfo->max_inodes = max_inodes;
1950 sbinfo->free_inodes = max_inodes - inodes;
1952 spin_unlock(&sbinfo->stat_lock);
1957 static void shmem_put_super(struct super_block *sb)
1959 kfree(sb->s_fs_info);
1960 sb->s_fs_info = NULL;
1963 static int shmem_fill_super(struct super_block *sb,
1964 void *data, int silent)
1966 struct inode *inode;
1967 struct dentry *root;
1968 int mode = S_IRWXUGO | S_ISVTX;
1969 uid_t uid = current->fsuid;
1970 gid_t gid = current->fsgid;
1972 struct shmem_sb_info *sbinfo;
1973 unsigned long blocks = 0;
1974 unsigned long inodes = 0;
1978 * Per default we only allow half of the physical ram per
1979 * tmpfs instance, limiting inodes to one per page of lowmem;
1980 * but the internal instance is left unlimited.
1982 if (!(sb->s_flags & MS_NOUSER)) {
1983 blocks = totalram_pages / 2;
1984 inodes = totalram_pages - totalhigh_pages;
1985 if (inodes > blocks)
1987 if (shmem_parse_options(data, &mode, &uid, &gid,
1992 sb->s_flags |= MS_NOUSER;
1995 /* Round up to L1_CACHE_BYTES to resist false sharing */
1996 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
1997 L1_CACHE_BYTES), GFP_KERNEL);
2001 spin_lock_init(&sbinfo->stat_lock);
2002 sbinfo->max_blocks = blocks;
2003 sbinfo->free_blocks = blocks;
2004 sbinfo->max_inodes = inodes;
2005 sbinfo->free_inodes = inodes;
2007 sb->s_fs_info = sbinfo;
2008 sb->s_maxbytes = SHMEM_MAX_BYTES;
2009 sb->s_blocksize = PAGE_CACHE_SIZE;
2010 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2011 sb->s_magic = TMPFS_MAGIC;
2012 sb->s_op = &shmem_ops;
2014 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2019 root = d_alloc_root(inode);
2028 shmem_put_super(sb);
2032 static kmem_cache_t *shmem_inode_cachep;
2034 static struct inode *shmem_alloc_inode(struct super_block *sb)
2036 struct shmem_inode_info *p;
2037 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2040 return &p->vfs_inode;
2043 static void shmem_destroy_inode(struct inode *inode)
2045 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2046 /* only struct inode is valid if it's an inline symlink */
2047 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2049 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2052 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
2054 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2056 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2057 SLAB_CTOR_CONSTRUCTOR) {
2058 inode_init_once(&p->vfs_inode);
2062 static int init_inodecache(void)
2064 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2065 sizeof(struct shmem_inode_info),
2066 0, 0, init_once, NULL);
2067 if (shmem_inode_cachep == NULL)
2072 static void destroy_inodecache(void)
2074 if (kmem_cache_destroy(shmem_inode_cachep))
2075 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2078 static struct address_space_operations shmem_aops = {
2079 .writepage = shmem_writepage,
2080 .set_page_dirty = __set_page_dirty_nobuffers,
2082 .prepare_write = shmem_prepare_write,
2083 .commit_write = simple_commit_write,
2087 static struct file_operations shmem_file_operations = {
2090 .llseek = generic_file_llseek,
2091 .read = shmem_file_read,
2092 .write = shmem_file_write,
2093 .fsync = simple_sync_file,
2094 .sendfile = shmem_file_sendfile,
2098 static struct inode_operations shmem_inode_operations = {
2099 .truncate = shmem_truncate,
2100 .setattr = shmem_notify_change,
2101 .truncate_range = shmem_truncate_range,
2104 static struct inode_operations shmem_dir_inode_operations = {
2106 .create = shmem_create,
2107 .lookup = simple_lookup,
2109 .unlink = shmem_unlink,
2110 .symlink = shmem_symlink,
2111 .mkdir = shmem_mkdir,
2112 .rmdir = shmem_rmdir,
2113 .mknod = shmem_mknod,
2114 .rename = shmem_rename,
2118 static struct super_operations shmem_ops = {
2119 .alloc_inode = shmem_alloc_inode,
2120 .destroy_inode = shmem_destroy_inode,
2122 .statfs = shmem_statfs,
2123 .remount_fs = shmem_remount_fs,
2125 .delete_inode = shmem_delete_inode,
2126 .drop_inode = generic_delete_inode,
2127 .put_super = shmem_put_super,
2130 static struct vm_operations_struct shmem_vm_ops = {
2131 .nopage = shmem_nopage,
2132 .populate = shmem_populate,
2134 .set_policy = shmem_set_policy,
2135 .get_policy = shmem_get_policy,
2140 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2141 int flags, const char *dev_name, void *data)
2143 return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2146 static struct file_system_type tmpfs_fs_type = {
2147 .owner = THIS_MODULE,
2149 .get_sb = shmem_get_sb,
2150 .kill_sb = kill_litter_super,
2152 static struct vfsmount *shm_mnt;
2154 static int __init init_tmpfs(void)
2158 error = init_inodecache();
2162 error = register_filesystem(&tmpfs_fs_type);
2164 printk(KERN_ERR "Could not register tmpfs\n");
2168 devfs_mk_dir("shm");
2170 shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2171 tmpfs_fs_type.name, NULL);
2172 if (IS_ERR(shm_mnt)) {
2173 error = PTR_ERR(shm_mnt);
2174 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2180 unregister_filesystem(&tmpfs_fs_type);
2182 destroy_inodecache();
2184 shm_mnt = ERR_PTR(error);
2187 module_init(init_tmpfs)
2190 * shmem_file_setup - get an unlinked file living in tmpfs
2192 * @name: name for dentry (to be seen in /proc/<pid>/maps
2193 * @size: size to be set for the file
2196 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2200 struct inode *inode;
2201 struct dentry *dentry, *root;
2204 if (IS_ERR(shm_mnt))
2205 return (void *)shm_mnt;
2207 if (size < 0 || size > SHMEM_MAX_BYTES)
2208 return ERR_PTR(-EINVAL);
2210 if (shmem_acct_size(flags, size))
2211 return ERR_PTR(-ENOMEM);
2215 this.len = strlen(name);
2216 this.hash = 0; /* will go */
2217 root = shm_mnt->mnt_root;
2218 dentry = d_alloc(root, &this);
2223 file = get_empty_filp();
2228 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2232 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2233 d_instantiate(dentry, inode);
2234 inode->i_size = size;
2235 inode->i_nlink = 0; /* It is unlinked */
2236 file->f_vfsmnt = mntget(shm_mnt);
2237 file->f_dentry = dentry;
2238 file->f_mapping = inode->i_mapping;
2239 file->f_op = &shmem_file_operations;
2240 file->f_mode = FMODE_WRITE | FMODE_READ;
2248 shmem_unacct_size(flags, size);
2249 return ERR_PTR(error);
2253 * shmem_zero_setup - setup a shared anonymous mapping
2255 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2257 int shmem_zero_setup(struct vm_area_struct *vma)
2260 loff_t size = vma->vm_end - vma->vm_start;
2262 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2264 return PTR_ERR(file);
2268 vma->vm_file = file;
2269 vma->vm_ops = &shmem_vm_ops;