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 <linux/ctype.h>
49 #include <asm/uaccess.h>
50 #include <asm/div64.h>
51 #include <asm/pgtable.h>
53 /* This magic number is used in glibc for posix shared memory */
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);
514 * If there are no indirect blocks or we are punching a hole
515 * below indirect blocks, nothing to be done.
517 if (!topdir || (punch_hole && (limit <= SHMEM_NR_DIRECT)))
520 BUG_ON(limit <= SHMEM_NR_DIRECT);
521 limit -= SHMEM_NR_DIRECT;
522 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
523 offset = idx % ENTRIES_PER_PAGE;
526 dir = shmem_dir_map(topdir);
527 stage = ENTRIES_PER_PAGEPAGE/2;
528 if (idx < ENTRIES_PER_PAGEPAGE/2) {
530 diroff = idx/ENTRIES_PER_PAGE;
532 dir += ENTRIES_PER_PAGE/2;
533 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
535 stage += ENTRIES_PER_PAGEPAGE;
538 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
539 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
540 if (!diroff && !offset) {
543 list_add(&middir->lru, &pages_to_free);
545 shmem_dir_unmap(dir);
546 dir = shmem_dir_map(middir);
554 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
555 if (unlikely(idx == stage)) {
556 shmem_dir_unmap(dir);
557 dir = shmem_dir_map(topdir) +
558 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
561 idx += ENTRIES_PER_PAGEPAGE;
565 stage = idx + ENTRIES_PER_PAGEPAGE;
569 list_add(&middir->lru, &pages_to_free);
570 shmem_dir_unmap(dir);
572 dir = shmem_dir_map(middir);
575 subdir = dir[diroff];
576 if (subdir && page_private(subdir)) {
578 if (size > ENTRIES_PER_PAGE)
579 size = ENTRIES_PER_PAGE;
580 freed = shmem_map_and_free_swp(subdir,
583 dir = shmem_dir_map(middir);
584 nr_swaps_freed += freed;
586 spin_lock(&info->lock);
587 set_page_private(subdir, page_private(subdir) - freed);
589 spin_unlock(&info->lock);
591 BUG_ON(page_private(subdir) > offset);
595 else if (subdir && !page_private(subdir)) {
598 list_add(&subdir->lru, &pages_to_free);
602 shmem_dir_unmap(dir);
604 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
606 * Call truncate_inode_pages again: racing shmem_unuse_inode
607 * may have swizzled a page in from swap since vmtruncate or
608 * generic_delete_inode did it, before we lowered next_index.
609 * Also, though shmem_getpage checks i_size before adding to
610 * cache, no recheck after: so fix the narrow window there too.
612 truncate_inode_pages_range(inode->i_mapping, start, end);
615 spin_lock(&info->lock);
616 info->flags &= ~SHMEM_TRUNCATE;
617 info->swapped -= nr_swaps_freed;
618 if (nr_pages_to_free)
619 shmem_free_blocks(inode, nr_pages_to_free);
620 shmem_recalc_inode(inode);
621 spin_unlock(&info->lock);
624 * Empty swap vector directory pages to be freed?
626 if (!list_empty(&pages_to_free)) {
627 pages_to_free.prev->next = NULL;
628 shmem_free_pages(pages_to_free.next);
632 static void shmem_truncate(struct inode *inode)
634 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
637 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
639 struct inode *inode = dentry->d_inode;
640 struct page *page = NULL;
643 if (attr->ia_valid & ATTR_SIZE) {
644 if (attr->ia_size < inode->i_size) {
646 * If truncating down to a partial page, then
647 * if that page is already allocated, hold it
648 * in memory until the truncation is over, so
649 * truncate_partial_page cannnot miss it were
650 * it assigned to swap.
652 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
653 (void) shmem_getpage(inode,
654 attr->ia_size>>PAGE_CACHE_SHIFT,
655 &page, SGP_READ, NULL);
658 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
659 * detect if any pages might have been added to cache
660 * after truncate_inode_pages. But we needn't bother
661 * if it's being fully truncated to zero-length: the
662 * nrpages check is efficient enough in that case.
665 struct shmem_inode_info *info = SHMEM_I(inode);
666 spin_lock(&info->lock);
667 info->flags &= ~SHMEM_PAGEIN;
668 spin_unlock(&info->lock);
673 error = inode_change_ok(inode, attr);
675 error = inode_setattr(inode, attr);
677 page_cache_release(page);
681 static void shmem_delete_inode(struct inode *inode)
683 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
684 struct shmem_inode_info *info = SHMEM_I(inode);
686 if (inode->i_op->truncate == shmem_truncate) {
687 truncate_inode_pages(inode->i_mapping, 0);
688 shmem_unacct_size(info->flags, inode->i_size);
690 shmem_truncate(inode);
691 if (!list_empty(&info->swaplist)) {
692 spin_lock(&shmem_swaplist_lock);
693 list_del_init(&info->swaplist);
694 spin_unlock(&shmem_swaplist_lock);
697 BUG_ON(inode->i_blocks);
698 if (sbinfo->max_inodes) {
699 spin_lock(&sbinfo->stat_lock);
700 sbinfo->free_inodes++;
701 spin_unlock(&sbinfo->stat_lock);
706 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
710 for (ptr = dir; ptr < edir; ptr++) {
711 if (ptr->val == entry.val)
717 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
730 ptr = info->i_direct;
731 spin_lock(&info->lock);
732 limit = info->next_index;
734 if (size > SHMEM_NR_DIRECT)
735 size = SHMEM_NR_DIRECT;
736 offset = shmem_find_swp(entry, ptr, ptr+size);
738 shmem_swp_balance_unmap();
741 if (!info->i_indirect)
744 dir = shmem_dir_map(info->i_indirect);
745 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
747 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
748 if (unlikely(idx == stage)) {
749 shmem_dir_unmap(dir-1);
750 dir = shmem_dir_map(info->i_indirect) +
751 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
754 idx += ENTRIES_PER_PAGEPAGE;
758 stage = idx + ENTRIES_PER_PAGEPAGE;
760 shmem_dir_unmap(dir);
761 dir = shmem_dir_map(subdir);
764 if (subdir && page_private(subdir)) {
765 ptr = shmem_swp_map(subdir);
767 if (size > ENTRIES_PER_PAGE)
768 size = ENTRIES_PER_PAGE;
769 offset = shmem_find_swp(entry, ptr, ptr+size);
771 shmem_dir_unmap(dir);
774 shmem_swp_unmap(ptr);
778 shmem_dir_unmap(dir-1);
780 spin_unlock(&info->lock);
784 inode = &info->vfs_inode;
785 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
786 info->flags |= SHMEM_PAGEIN;
787 shmem_swp_set(info, ptr + offset, 0);
789 shmem_swp_unmap(ptr);
790 spin_unlock(&info->lock);
792 * Decrement swap count even when the entry is left behind:
793 * try_to_unuse will skip over mms, then reincrement count.
800 * shmem_unuse() search for an eventually swapped out shmem page.
802 int shmem_unuse(swp_entry_t entry, struct page *page)
804 struct list_head *p, *next;
805 struct shmem_inode_info *info;
808 spin_lock(&shmem_swaplist_lock);
809 list_for_each_safe(p, next, &shmem_swaplist) {
810 info = list_entry(p, struct shmem_inode_info, swaplist);
812 list_del_init(&info->swaplist);
813 else if (shmem_unuse_inode(info, entry, page)) {
814 /* move head to start search for next from here */
815 list_move_tail(&shmem_swaplist, &info->swaplist);
820 spin_unlock(&shmem_swaplist_lock);
825 * Move the page from the page cache to the swap cache.
827 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
829 struct shmem_inode_info *info;
830 swp_entry_t *entry, swap;
831 struct address_space *mapping;
835 BUG_ON(!PageLocked(page));
836 BUG_ON(page_mapped(page));
838 mapping = page->mapping;
840 inode = mapping->host;
841 info = SHMEM_I(inode);
842 if (info->flags & VM_LOCKED)
844 swap = get_swap_page();
848 spin_lock(&info->lock);
849 shmem_recalc_inode(inode);
850 if (index >= info->next_index) {
851 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
854 entry = shmem_swp_entry(info, index, NULL);
858 if (move_to_swap_cache(page, swap) == 0) {
859 shmem_swp_set(info, entry, swap.val);
860 shmem_swp_unmap(entry);
861 spin_unlock(&info->lock);
862 if (list_empty(&info->swaplist)) {
863 spin_lock(&shmem_swaplist_lock);
864 /* move instead of add in case we're racing */
865 list_move_tail(&info->swaplist, &shmem_swaplist);
866 spin_unlock(&shmem_swaplist_lock);
872 shmem_swp_unmap(entry);
874 spin_unlock(&info->lock);
877 set_page_dirty(page);
878 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
882 static int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
884 char *nodelist = strchr(value, ':');
888 /* NUL-terminate policy string */
890 if (nodelist_parse(nodelist, *policy_nodes))
893 if (!strcmp(value, "default")) {
894 *policy = MPOL_DEFAULT;
895 /* Don't allow a nodelist */
898 } else if (!strcmp(value, "prefer")) {
899 *policy = MPOL_PREFERRED;
900 /* Insist on a nodelist of one node only */
902 char *rest = nodelist;
903 while (isdigit(*rest))
908 } else if (!strcmp(value, "bind")) {
910 /* Insist on a nodelist */
913 } else if (!strcmp(value, "interleave")) {
914 *policy = MPOL_INTERLEAVE;
915 /* Default to nodes online if no nodelist */
917 *policy_nodes = node_online_map;
921 /* Restore string for error message */
927 static struct page *shmem_swapin_async(struct shared_policy *p,
928 swp_entry_t entry, unsigned long idx)
931 struct vm_area_struct pvma;
933 /* Create a pseudo vma that just contains the policy */
934 memset(&pvma, 0, sizeof(struct vm_area_struct));
935 pvma.vm_end = PAGE_SIZE;
937 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
938 page = read_swap_cache_async(entry, &pvma, 0);
939 mpol_free(pvma.vm_policy);
943 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
946 struct shared_policy *p = &info->policy;
949 unsigned long offset;
951 num = valid_swaphandles(entry, &offset);
952 for (i = 0; i < num; offset++, i++) {
953 page = shmem_swapin_async(p,
954 swp_entry(swp_type(entry), offset), idx);
957 page_cache_release(page);
959 lru_add_drain(); /* Push any new pages onto the LRU now */
960 return shmem_swapin_async(p, entry, idx);
964 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
967 struct vm_area_struct pvma;
970 memset(&pvma, 0, sizeof(struct vm_area_struct));
971 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
973 pvma.vm_end = PAGE_SIZE;
974 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
975 mpol_free(pvma.vm_policy);
979 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
984 static inline struct page *
985 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
987 swapin_readahead(entry, 0, NULL);
988 return read_swap_cache_async(entry, NULL, 0);
991 static inline struct page *
992 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
994 return alloc_page(gfp | __GFP_ZERO);
999 * shmem_getpage - either get the page from swap or allocate a new one
1001 * If we allocate a new one we do not mark it dirty. That's up to the
1002 * vm. If we swap it in we mark it dirty since we also free the swap
1003 * entry since a page cannot live in both the swap and page cache
1005 static int shmem_getpage(struct inode *inode, unsigned long idx,
1006 struct page **pagep, enum sgp_type sgp, int *type)
1008 struct address_space *mapping = inode->i_mapping;
1009 struct shmem_inode_info *info = SHMEM_I(inode);
1010 struct shmem_sb_info *sbinfo;
1011 struct page *filepage = *pagep;
1012 struct page *swappage;
1017 if (idx >= SHMEM_MAX_INDEX)
1020 * Normally, filepage is NULL on entry, and either found
1021 * uptodate immediately, or allocated and zeroed, or read
1022 * in under swappage, which is then assigned to filepage.
1023 * But shmem_prepare_write passes in a locked filepage,
1024 * which may be found not uptodate by other callers too,
1025 * and may need to be copied from the swappage read in.
1029 filepage = find_lock_page(mapping, idx);
1030 if (filepage && PageUptodate(filepage))
1033 if (sgp == SGP_QUICK)
1036 spin_lock(&info->lock);
1037 shmem_recalc_inode(inode);
1038 entry = shmem_swp_alloc(info, idx, sgp);
1039 if (IS_ERR(entry)) {
1040 spin_unlock(&info->lock);
1041 error = PTR_ERR(entry);
1047 /* Look it up and read it in.. */
1048 swappage = lookup_swap_cache(swap);
1050 shmem_swp_unmap(entry);
1051 spin_unlock(&info->lock);
1052 /* here we actually do the io */
1053 if (type && *type == VM_FAULT_MINOR) {
1054 inc_page_state(pgmajfault);
1055 *type = VM_FAULT_MAJOR;
1057 swappage = shmem_swapin(info, swap, idx);
1059 spin_lock(&info->lock);
1060 entry = shmem_swp_alloc(info, idx, sgp);
1062 error = PTR_ERR(entry);
1064 if (entry->val == swap.val)
1066 shmem_swp_unmap(entry);
1068 spin_unlock(&info->lock);
1073 wait_on_page_locked(swappage);
1074 page_cache_release(swappage);
1078 /* We have to do this with page locked to prevent races */
1079 if (TestSetPageLocked(swappage)) {
1080 shmem_swp_unmap(entry);
1081 spin_unlock(&info->lock);
1082 wait_on_page_locked(swappage);
1083 page_cache_release(swappage);
1086 if (!PageSwapCache(swappage)) {
1087 /* Page migration has occured */
1088 shmem_swp_unmap(entry);
1089 spin_unlock(&info->lock);
1090 unlock_page(swappage);
1091 page_cache_release(swappage);
1094 if (PageWriteback(swappage)) {
1095 shmem_swp_unmap(entry);
1096 spin_unlock(&info->lock);
1097 wait_on_page_writeback(swappage);
1098 unlock_page(swappage);
1099 page_cache_release(swappage);
1102 if (!PageUptodate(swappage)) {
1103 shmem_swp_unmap(entry);
1104 spin_unlock(&info->lock);
1105 unlock_page(swappage);
1106 page_cache_release(swappage);
1112 shmem_swp_set(info, entry, 0);
1113 shmem_swp_unmap(entry);
1114 delete_from_swap_cache(swappage);
1115 spin_unlock(&info->lock);
1116 copy_highpage(filepage, swappage);
1117 unlock_page(swappage);
1118 page_cache_release(swappage);
1119 flush_dcache_page(filepage);
1120 SetPageUptodate(filepage);
1121 set_page_dirty(filepage);
1123 } else if (!(error = move_from_swap_cache(
1124 swappage, idx, mapping))) {
1125 info->flags |= SHMEM_PAGEIN;
1126 shmem_swp_set(info, entry, 0);
1127 shmem_swp_unmap(entry);
1128 spin_unlock(&info->lock);
1129 filepage = swappage;
1132 shmem_swp_unmap(entry);
1133 spin_unlock(&info->lock);
1134 unlock_page(swappage);
1135 page_cache_release(swappage);
1136 if (error == -ENOMEM) {
1137 /* let kswapd refresh zone for GFP_ATOMICs */
1138 blk_congestion_wait(WRITE, HZ/50);
1142 } else if (sgp == SGP_READ && !filepage) {
1143 shmem_swp_unmap(entry);
1144 filepage = find_get_page(mapping, idx);
1146 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1147 spin_unlock(&info->lock);
1148 wait_on_page_locked(filepage);
1149 page_cache_release(filepage);
1153 spin_unlock(&info->lock);
1155 shmem_swp_unmap(entry);
1156 sbinfo = SHMEM_SB(inode->i_sb);
1157 if (sbinfo->max_blocks) {
1158 spin_lock(&sbinfo->stat_lock);
1159 if (sbinfo->free_blocks == 0 ||
1160 shmem_acct_block(info->flags)) {
1161 spin_unlock(&sbinfo->stat_lock);
1162 spin_unlock(&info->lock);
1166 sbinfo->free_blocks--;
1167 inode->i_blocks += BLOCKS_PER_PAGE;
1168 spin_unlock(&sbinfo->stat_lock);
1169 } else if (shmem_acct_block(info->flags)) {
1170 spin_unlock(&info->lock);
1176 spin_unlock(&info->lock);
1177 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1181 shmem_unacct_blocks(info->flags, 1);
1182 shmem_free_blocks(inode, 1);
1187 spin_lock(&info->lock);
1188 entry = shmem_swp_alloc(info, idx, sgp);
1190 error = PTR_ERR(entry);
1193 shmem_swp_unmap(entry);
1195 if (error || swap.val || 0 != add_to_page_cache_lru(
1196 filepage, mapping, idx, GFP_ATOMIC)) {
1197 spin_unlock(&info->lock);
1198 page_cache_release(filepage);
1199 shmem_unacct_blocks(info->flags, 1);
1200 shmem_free_blocks(inode, 1);
1206 info->flags |= SHMEM_PAGEIN;
1210 spin_unlock(&info->lock);
1211 flush_dcache_page(filepage);
1212 SetPageUptodate(filepage);
1215 if (*pagep != filepage) {
1216 unlock_page(filepage);
1222 if (*pagep != filepage) {
1223 unlock_page(filepage);
1224 page_cache_release(filepage);
1229 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1231 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1232 struct page *page = NULL;
1236 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1237 idx += vma->vm_pgoff;
1238 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1239 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1240 return NOPAGE_SIGBUS;
1242 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1244 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1246 mark_page_accessed(page);
1250 static int shmem_populate(struct vm_area_struct *vma,
1251 unsigned long addr, unsigned long len,
1252 pgprot_t prot, unsigned long pgoff, int nonblock)
1254 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1255 struct mm_struct *mm = vma->vm_mm;
1256 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1259 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1260 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1263 while ((long) len > 0) {
1264 struct page *page = NULL;
1267 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1269 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1272 /* Page may still be null, but only if nonblock was set. */
1274 mark_page_accessed(page);
1275 err = install_page(mm, vma, addr, page, prot);
1277 page_cache_release(page);
1280 } else if (vma->vm_flags & VM_NONLINEAR) {
1281 /* No page was found just because we can't read it in
1282 * now (being here implies nonblock != 0), but the page
1283 * may exist, so set the PTE to fault it in later. */
1284 err = install_file_pte(mm, vma, addr, pgoff, prot);
1297 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1299 struct inode *i = vma->vm_file->f_dentry->d_inode;
1300 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1304 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1306 struct inode *i = vma->vm_file->f_dentry->d_inode;
1309 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1310 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1314 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1316 struct inode *inode = file->f_dentry->d_inode;
1317 struct shmem_inode_info *info = SHMEM_I(inode);
1318 int retval = -ENOMEM;
1320 spin_lock(&info->lock);
1321 if (lock && !(info->flags & VM_LOCKED)) {
1322 if (!user_shm_lock(inode->i_size, user))
1324 info->flags |= VM_LOCKED;
1326 if (!lock && (info->flags & VM_LOCKED) && user) {
1327 user_shm_unlock(inode->i_size, user);
1328 info->flags &= ~VM_LOCKED;
1332 spin_unlock(&info->lock);
1336 int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1338 file_accessed(file);
1339 vma->vm_ops = &shmem_vm_ops;
1343 static struct inode *
1344 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1346 struct inode *inode;
1347 struct shmem_inode_info *info;
1348 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1350 if (sbinfo->max_inodes) {
1351 spin_lock(&sbinfo->stat_lock);
1352 if (!sbinfo->free_inodes) {
1353 spin_unlock(&sbinfo->stat_lock);
1356 sbinfo->free_inodes--;
1357 spin_unlock(&sbinfo->stat_lock);
1360 inode = new_inode(sb);
1362 inode->i_mode = mode;
1363 inode->i_uid = current->fsuid;
1364 inode->i_gid = current->fsgid;
1365 inode->i_blksize = PAGE_CACHE_SIZE;
1366 inode->i_blocks = 0;
1367 inode->i_mapping->a_ops = &shmem_aops;
1368 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1369 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1370 info = SHMEM_I(inode);
1371 memset(info, 0, (char *)inode - (char *)info);
1372 spin_lock_init(&info->lock);
1373 INIT_LIST_HEAD(&info->swaplist);
1375 switch (mode & S_IFMT) {
1377 init_special_inode(inode, mode, dev);
1380 inode->i_op = &shmem_inode_operations;
1381 inode->i_fop = &shmem_file_operations;
1382 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1383 &sbinfo->policy_nodes);
1387 /* Some things misbehave if size == 0 on a directory */
1388 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1389 inode->i_op = &shmem_dir_inode_operations;
1390 inode->i_fop = &simple_dir_operations;
1394 * Must not load anything in the rbtree,
1395 * mpol_free_shared_policy will not be called.
1397 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1401 } else if (sbinfo->max_inodes) {
1402 spin_lock(&sbinfo->stat_lock);
1403 sbinfo->free_inodes++;
1404 spin_unlock(&sbinfo->stat_lock);
1410 static struct inode_operations shmem_symlink_inode_operations;
1411 static struct inode_operations shmem_symlink_inline_operations;
1414 * Normally tmpfs makes no use of shmem_prepare_write, but it
1415 * lets a tmpfs file be used read-write below the loop driver.
1418 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1420 struct inode *inode = page->mapping->host;
1421 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1425 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1427 struct inode *inode = file->f_dentry->d_inode;
1429 unsigned long written;
1432 if ((ssize_t) count < 0)
1435 if (!access_ok(VERIFY_READ, buf, count))
1438 mutex_lock(&inode->i_mutex);
1443 err = generic_write_checks(file, &pos, &count, 0);
1447 err = remove_suid(file->f_dentry);
1451 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1454 struct page *page = NULL;
1455 unsigned long bytes, index, offset;
1459 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1460 index = pos >> PAGE_CACHE_SHIFT;
1461 bytes = PAGE_CACHE_SIZE - offset;
1466 * We don't hold page lock across copy from user -
1467 * what would it guard against? - so no deadlock here.
1468 * But it still may be a good idea to prefault below.
1471 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1476 if (PageHighMem(page)) {
1477 volatile unsigned char dummy;
1478 __get_user(dummy, buf);
1479 __get_user(dummy, buf + bytes - 1);
1481 kaddr = kmap_atomic(page, KM_USER0);
1482 left = __copy_from_user_inatomic(kaddr + offset,
1484 kunmap_atomic(kaddr, KM_USER0);
1488 left = __copy_from_user(kaddr + offset, buf, bytes);
1496 if (pos > inode->i_size)
1497 i_size_write(inode, pos);
1499 flush_dcache_page(page);
1500 set_page_dirty(page);
1501 mark_page_accessed(page);
1502 page_cache_release(page);
1512 * Our dirty pages are not counted in nr_dirty,
1513 * and we do not attempt to balance dirty pages.
1523 mutex_unlock(&inode->i_mutex);
1527 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1529 struct inode *inode = filp->f_dentry->d_inode;
1530 struct address_space *mapping = inode->i_mapping;
1531 unsigned long index, offset;
1533 index = *ppos >> PAGE_CACHE_SHIFT;
1534 offset = *ppos & ~PAGE_CACHE_MASK;
1537 struct page *page = NULL;
1538 unsigned long end_index, nr, ret;
1539 loff_t i_size = i_size_read(inode);
1541 end_index = i_size >> PAGE_CACHE_SHIFT;
1542 if (index > end_index)
1544 if (index == end_index) {
1545 nr = i_size & ~PAGE_CACHE_MASK;
1550 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1552 if (desc->error == -EINVAL)
1558 * We must evaluate after, since reads (unlike writes)
1559 * are called without i_mutex protection against truncate
1561 nr = PAGE_CACHE_SIZE;
1562 i_size = i_size_read(inode);
1563 end_index = i_size >> PAGE_CACHE_SHIFT;
1564 if (index == end_index) {
1565 nr = i_size & ~PAGE_CACHE_MASK;
1568 page_cache_release(page);
1576 * If users can be writing to this page using arbitrary
1577 * virtual addresses, take care about potential aliasing
1578 * before reading the page on the kernel side.
1580 if (mapping_writably_mapped(mapping))
1581 flush_dcache_page(page);
1583 * Mark the page accessed if we read the beginning.
1586 mark_page_accessed(page);
1588 page = ZERO_PAGE(0);
1589 page_cache_get(page);
1593 * Ok, we have the page, and it's up-to-date, so
1594 * now we can copy it to user space...
1596 * The actor routine returns how many bytes were actually used..
1597 * NOTE! This may not be the same as how much of a user buffer
1598 * we filled up (we may be padding etc), so we can only update
1599 * "pos" here (the actor routine has to update the user buffer
1600 * pointers and the remaining count).
1602 ret = actor(desc, page, offset, nr);
1604 index += offset >> PAGE_CACHE_SHIFT;
1605 offset &= ~PAGE_CACHE_MASK;
1607 page_cache_release(page);
1608 if (ret != nr || !desc->count)
1614 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1615 file_accessed(filp);
1618 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1620 read_descriptor_t desc;
1622 if ((ssize_t) count < 0)
1624 if (!access_ok(VERIFY_WRITE, buf, count))
1634 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1636 return desc.written;
1640 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1641 size_t count, read_actor_t actor, void *target)
1643 read_descriptor_t desc;
1650 desc.arg.data = target;
1653 do_shmem_file_read(in_file, ppos, &desc, actor);
1655 return desc.written;
1659 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1661 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1663 buf->f_type = TMPFS_SUPER_MAGIC;
1664 buf->f_bsize = PAGE_CACHE_SIZE;
1665 buf->f_namelen = NAME_MAX;
1666 spin_lock(&sbinfo->stat_lock);
1667 if (sbinfo->max_blocks) {
1668 buf->f_blocks = sbinfo->max_blocks;
1669 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1671 if (sbinfo->max_inodes) {
1672 buf->f_files = sbinfo->max_inodes;
1673 buf->f_ffree = sbinfo->free_inodes;
1675 /* else leave those fields 0 like simple_statfs */
1676 spin_unlock(&sbinfo->stat_lock);
1681 * File creation. Allocate an inode, and we're done..
1684 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1686 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1687 int error = -ENOSPC;
1690 error = security_inode_init_security(inode, dir, NULL, NULL,
1693 if (error != -EOPNOTSUPP) {
1699 if (dir->i_mode & S_ISGID) {
1700 inode->i_gid = dir->i_gid;
1702 inode->i_mode |= S_ISGID;
1704 dir->i_size += BOGO_DIRENT_SIZE;
1705 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1706 d_instantiate(dentry, inode);
1707 dget(dentry); /* Extra count - pin the dentry in core */
1712 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1716 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1722 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1723 struct nameidata *nd)
1725 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1731 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1733 struct inode *inode = old_dentry->d_inode;
1734 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1737 * No ordinary (disk based) filesystem counts links as inodes;
1738 * but each new link needs a new dentry, pinning lowmem, and
1739 * tmpfs dentries cannot be pruned until they are unlinked.
1741 if (sbinfo->max_inodes) {
1742 spin_lock(&sbinfo->stat_lock);
1743 if (!sbinfo->free_inodes) {
1744 spin_unlock(&sbinfo->stat_lock);
1747 sbinfo->free_inodes--;
1748 spin_unlock(&sbinfo->stat_lock);
1751 dir->i_size += BOGO_DIRENT_SIZE;
1752 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1754 atomic_inc(&inode->i_count); /* New dentry reference */
1755 dget(dentry); /* Extra pinning count for the created dentry */
1756 d_instantiate(dentry, inode);
1760 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1762 struct inode *inode = dentry->d_inode;
1764 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1765 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1766 if (sbinfo->max_inodes) {
1767 spin_lock(&sbinfo->stat_lock);
1768 sbinfo->free_inodes++;
1769 spin_unlock(&sbinfo->stat_lock);
1773 dir->i_size -= BOGO_DIRENT_SIZE;
1774 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1776 dput(dentry); /* Undo the count from "create" - this does all the work */
1780 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1782 if (!simple_empty(dentry))
1786 return shmem_unlink(dir, dentry);
1790 * The VFS layer already does all the dentry stuff for rename,
1791 * we just have to decrement the usage count for the target if
1792 * it exists so that the VFS layer correctly free's it when it
1795 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1797 struct inode *inode = old_dentry->d_inode;
1798 int they_are_dirs = S_ISDIR(inode->i_mode);
1800 if (!simple_empty(new_dentry))
1803 if (new_dentry->d_inode) {
1804 (void) shmem_unlink(new_dir, new_dentry);
1807 } else if (they_are_dirs) {
1812 old_dir->i_size -= BOGO_DIRENT_SIZE;
1813 new_dir->i_size += BOGO_DIRENT_SIZE;
1814 old_dir->i_ctime = old_dir->i_mtime =
1815 new_dir->i_ctime = new_dir->i_mtime =
1816 inode->i_ctime = CURRENT_TIME;
1820 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1824 struct inode *inode;
1825 struct page *page = NULL;
1827 struct shmem_inode_info *info;
1829 len = strlen(symname) + 1;
1830 if (len > PAGE_CACHE_SIZE)
1831 return -ENAMETOOLONG;
1833 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1837 error = security_inode_init_security(inode, dir, NULL, NULL,
1840 if (error != -EOPNOTSUPP) {
1847 info = SHMEM_I(inode);
1848 inode->i_size = len-1;
1849 if (len <= (char *)inode - (char *)info) {
1851 memcpy(info, symname, len);
1852 inode->i_op = &shmem_symlink_inline_operations;
1854 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1859 inode->i_op = &shmem_symlink_inode_operations;
1860 kaddr = kmap_atomic(page, KM_USER0);
1861 memcpy(kaddr, symname, len);
1862 kunmap_atomic(kaddr, KM_USER0);
1863 set_page_dirty(page);
1864 page_cache_release(page);
1866 if (dir->i_mode & S_ISGID)
1867 inode->i_gid = dir->i_gid;
1868 dir->i_size += BOGO_DIRENT_SIZE;
1869 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1870 d_instantiate(dentry, inode);
1875 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1877 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1881 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1883 struct page *page = NULL;
1884 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1885 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1889 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1891 if (!IS_ERR(nd_get_link(nd))) {
1892 struct page *page = cookie;
1894 mark_page_accessed(page);
1895 page_cache_release(page);
1899 static struct inode_operations shmem_symlink_inline_operations = {
1900 .readlink = generic_readlink,
1901 .follow_link = shmem_follow_link_inline,
1904 static struct inode_operations shmem_symlink_inode_operations = {
1905 .truncate = shmem_truncate,
1906 .readlink = generic_readlink,
1907 .follow_link = shmem_follow_link,
1908 .put_link = shmem_put_link,
1911 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
1912 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
1913 int *policy, nodemask_t *policy_nodes)
1915 char *this_char, *value, *rest;
1917 while (options != NULL) {
1918 this_char = options;
1921 * NUL-terminate this option: unfortunately,
1922 * mount options form a comma-separated list,
1923 * but mpol's nodelist may also contain commas.
1925 options = strchr(options, ',');
1926 if (options == NULL)
1929 if (!isdigit(*options)) {
1936 if ((value = strchr(this_char,'=')) != NULL) {
1940 "tmpfs: No value for mount option '%s'\n",
1945 if (!strcmp(this_char,"size")) {
1946 unsigned long long size;
1947 size = memparse(value,&rest);
1949 size <<= PAGE_SHIFT;
1950 size *= totalram_pages;
1956 *blocks = size >> PAGE_CACHE_SHIFT;
1957 } else if (!strcmp(this_char,"nr_blocks")) {
1958 *blocks = memparse(value,&rest);
1961 } else if (!strcmp(this_char,"nr_inodes")) {
1962 *inodes = memparse(value,&rest);
1965 } else if (!strcmp(this_char,"mode")) {
1968 *mode = simple_strtoul(value,&rest,8);
1971 } else if (!strcmp(this_char,"uid")) {
1974 *uid = simple_strtoul(value,&rest,0);
1977 } else if (!strcmp(this_char,"gid")) {
1980 *gid = simple_strtoul(value,&rest,0);
1983 } else if (!strcmp(this_char,"mpol")) {
1984 if (shmem_parse_mpol(value,policy,policy_nodes))
1987 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1995 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2001 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2003 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2004 unsigned long max_blocks = sbinfo->max_blocks;
2005 unsigned long max_inodes = sbinfo->max_inodes;
2006 int policy = sbinfo->policy;
2007 nodemask_t policy_nodes = sbinfo->policy_nodes;
2008 unsigned long blocks;
2009 unsigned long inodes;
2010 int error = -EINVAL;
2012 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2013 &max_inodes, &policy, &policy_nodes))
2016 spin_lock(&sbinfo->stat_lock);
2017 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2018 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2019 if (max_blocks < blocks)
2021 if (max_inodes < inodes)
2024 * Those tests also disallow limited->unlimited while any are in
2025 * use, so i_blocks will always be zero when max_blocks is zero;
2026 * but we must separately disallow unlimited->limited, because
2027 * in that case we have no record of how much is already in use.
2029 if (max_blocks && !sbinfo->max_blocks)
2031 if (max_inodes && !sbinfo->max_inodes)
2035 sbinfo->max_blocks = max_blocks;
2036 sbinfo->free_blocks = max_blocks - blocks;
2037 sbinfo->max_inodes = max_inodes;
2038 sbinfo->free_inodes = max_inodes - inodes;
2039 sbinfo->policy = policy;
2040 sbinfo->policy_nodes = policy_nodes;
2042 spin_unlock(&sbinfo->stat_lock);
2047 static void shmem_put_super(struct super_block *sb)
2049 kfree(sb->s_fs_info);
2050 sb->s_fs_info = NULL;
2053 static int shmem_fill_super(struct super_block *sb,
2054 void *data, int silent)
2056 struct inode *inode;
2057 struct dentry *root;
2058 int mode = S_IRWXUGO | S_ISVTX;
2059 uid_t uid = current->fsuid;
2060 gid_t gid = current->fsgid;
2062 struct shmem_sb_info *sbinfo;
2063 unsigned long blocks = 0;
2064 unsigned long inodes = 0;
2065 int policy = MPOL_DEFAULT;
2066 nodemask_t policy_nodes = node_online_map;
2070 * Per default we only allow half of the physical ram per
2071 * tmpfs instance, limiting inodes to one per page of lowmem;
2072 * but the internal instance is left unlimited.
2074 if (!(sb->s_flags & MS_NOUSER)) {
2075 blocks = totalram_pages / 2;
2076 inodes = totalram_pages - totalhigh_pages;
2077 if (inodes > blocks)
2079 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2080 &inodes, &policy, &policy_nodes))
2084 sb->s_flags |= MS_NOUSER;
2087 /* Round up to L1_CACHE_BYTES to resist false sharing */
2088 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2089 L1_CACHE_BYTES), GFP_KERNEL);
2093 spin_lock_init(&sbinfo->stat_lock);
2094 sbinfo->max_blocks = blocks;
2095 sbinfo->free_blocks = blocks;
2096 sbinfo->max_inodes = inodes;
2097 sbinfo->free_inodes = inodes;
2098 sbinfo->policy = policy;
2099 sbinfo->policy_nodes = policy_nodes;
2101 sb->s_fs_info = sbinfo;
2102 sb->s_maxbytes = SHMEM_MAX_BYTES;
2103 sb->s_blocksize = PAGE_CACHE_SIZE;
2104 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2105 sb->s_magic = TMPFS_SUPER_MAGIC;
2106 sb->s_op = &shmem_ops;
2107 sb->s_time_gran = 1;
2109 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2114 root = d_alloc_root(inode);
2123 shmem_put_super(sb);
2127 static kmem_cache_t *shmem_inode_cachep;
2129 static struct inode *shmem_alloc_inode(struct super_block *sb)
2131 struct shmem_inode_info *p;
2132 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2135 return &p->vfs_inode;
2138 static void shmem_destroy_inode(struct inode *inode)
2140 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2141 /* only struct inode is valid if it's an inline symlink */
2142 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2144 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2147 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
2149 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2151 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2152 SLAB_CTOR_CONSTRUCTOR) {
2153 inode_init_once(&p->vfs_inode);
2157 static int init_inodecache(void)
2159 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2160 sizeof(struct shmem_inode_info),
2161 0, 0, init_once, NULL);
2162 if (shmem_inode_cachep == NULL)
2167 static void destroy_inodecache(void)
2169 if (kmem_cache_destroy(shmem_inode_cachep))
2170 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2173 static struct address_space_operations shmem_aops = {
2174 .writepage = shmem_writepage,
2175 .set_page_dirty = __set_page_dirty_nobuffers,
2177 .prepare_write = shmem_prepare_write,
2178 .commit_write = simple_commit_write,
2180 .migratepage = migrate_page,
2183 static struct file_operations shmem_file_operations = {
2186 .llseek = generic_file_llseek,
2187 .read = shmem_file_read,
2188 .write = shmem_file_write,
2189 .fsync = simple_sync_file,
2190 .sendfile = shmem_file_sendfile,
2194 static struct inode_operations shmem_inode_operations = {
2195 .truncate = shmem_truncate,
2196 .setattr = shmem_notify_change,
2197 .truncate_range = shmem_truncate_range,
2200 static struct inode_operations shmem_dir_inode_operations = {
2202 .create = shmem_create,
2203 .lookup = simple_lookup,
2205 .unlink = shmem_unlink,
2206 .symlink = shmem_symlink,
2207 .mkdir = shmem_mkdir,
2208 .rmdir = shmem_rmdir,
2209 .mknod = shmem_mknod,
2210 .rename = shmem_rename,
2214 static struct super_operations shmem_ops = {
2215 .alloc_inode = shmem_alloc_inode,
2216 .destroy_inode = shmem_destroy_inode,
2218 .statfs = shmem_statfs,
2219 .remount_fs = shmem_remount_fs,
2221 .delete_inode = shmem_delete_inode,
2222 .drop_inode = generic_delete_inode,
2223 .put_super = shmem_put_super,
2226 static struct vm_operations_struct shmem_vm_ops = {
2227 .nopage = shmem_nopage,
2228 .populate = shmem_populate,
2230 .set_policy = shmem_set_policy,
2231 .get_policy = shmem_get_policy,
2236 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2237 int flags, const char *dev_name, void *data)
2239 return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2242 static struct file_system_type tmpfs_fs_type = {
2243 .owner = THIS_MODULE,
2245 .get_sb = shmem_get_sb,
2246 .kill_sb = kill_litter_super,
2248 static struct vfsmount *shm_mnt;
2250 static int __init init_tmpfs(void)
2254 error = init_inodecache();
2258 error = register_filesystem(&tmpfs_fs_type);
2260 printk(KERN_ERR "Could not register tmpfs\n");
2264 devfs_mk_dir("shm");
2266 shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2267 tmpfs_fs_type.name, NULL);
2268 if (IS_ERR(shm_mnt)) {
2269 error = PTR_ERR(shm_mnt);
2270 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2276 unregister_filesystem(&tmpfs_fs_type);
2278 destroy_inodecache();
2280 shm_mnt = ERR_PTR(error);
2283 module_init(init_tmpfs)
2286 * shmem_file_setup - get an unlinked file living in tmpfs
2288 * @name: name for dentry (to be seen in /proc/<pid>/maps
2289 * @size: size to be set for the file
2292 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2296 struct inode *inode;
2297 struct dentry *dentry, *root;
2300 if (IS_ERR(shm_mnt))
2301 return (void *)shm_mnt;
2303 if (size < 0 || size > SHMEM_MAX_BYTES)
2304 return ERR_PTR(-EINVAL);
2306 if (shmem_acct_size(flags, size))
2307 return ERR_PTR(-ENOMEM);
2311 this.len = strlen(name);
2312 this.hash = 0; /* will go */
2313 root = shm_mnt->mnt_root;
2314 dentry = d_alloc(root, &this);
2319 file = get_empty_filp();
2324 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2328 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2329 d_instantiate(dentry, inode);
2330 inode->i_size = size;
2331 inode->i_nlink = 0; /* It is unlinked */
2332 file->f_vfsmnt = mntget(shm_mnt);
2333 file->f_dentry = dentry;
2334 file->f_mapping = inode->i_mapping;
2335 file->f_op = &shmem_file_operations;
2336 file->f_mode = FMODE_WRITE | FMODE_READ;
2344 shmem_unacct_size(flags, size);
2345 return ERR_PTR(error);
2349 * shmem_zero_setup - setup a shared anonymous mapping
2351 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2353 int shmem_zero_setup(struct vm_area_struct *vma)
2356 loff_t size = vma->vm_end - vma->vm_start;
2358 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2360 return PTR_ERR(file);
2364 vma->vm_file = file;
2365 vma->vm_ops = &shmem_vm_ops;