2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2003 Hugh Dickins.
10 * Copyright (C) 2002-2003 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * This file is released under the GPL.
17 * This virtual memory filesystem is heavily based on the ramfs. It
18 * extends ramfs by the ability to use swap and honor resource limits
19 * which makes it a completely usable filesystem.
22 #include <linux/config.h>
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/devfs_fs_kernel.h>
28 #include <linux/mman.h>
29 #include <linux/file.h>
30 #include <linux/swap.h>
31 #include <linux/pagemap.h>
32 #include <linux/string.h>
33 #include <linux/slab.h>
34 #include <linux/backing-dev.h>
35 #include <linux/shmem_fs.h>
36 #include <linux/mount.h>
37 #include <linux/writeback.h>
38 #include <linux/vfs.h>
39 #include <linux/blkdev.h>
40 #include <linux/security.h>
41 #include <linux/swapops.h>
42 #include <linux/mempolicy.h>
43 #include <linux/namei.h>
44 #include <asm/uaccess.h>
45 #include <asm/div64.h>
46 #include <asm/pgtable.h>
48 /* This magic number is used in glibc for posix shared memory */
49 #define TMPFS_MAGIC 0x01021994
51 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
52 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
53 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
55 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
56 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
58 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
60 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
61 #define SHMEM_PAGEIN VM_READ
62 #define SHMEM_TRUNCATE VM_WRITE
64 /* Pretend that each entry is of this size in directory's i_size */
65 #define BOGO_DIRENT_SIZE 20
67 /* Keep swapped page count in private field of indirect struct page */
68 #define nr_swapped private
70 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
72 SGP_QUICK, /* don't try more than file page cache lookup */
73 SGP_READ, /* don't exceed i_size, don't allocate page */
74 SGP_CACHE, /* don't exceed i_size, may allocate page */
75 SGP_WRITE, /* may exceed i_size, may allocate page */
78 static int shmem_getpage(struct inode *inode, unsigned long idx,
79 struct page **pagep, enum sgp_type sgp, int *type);
81 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
84 * The above definition of ENTRIES_PER_PAGE, and the use of
85 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
86 * might be reconsidered if it ever diverges from PAGE_SIZE.
88 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
91 static inline void shmem_dir_free(struct page *page)
93 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
96 static struct page **shmem_dir_map(struct page *page)
98 return (struct page **)kmap_atomic(page, KM_USER0);
101 static inline void shmem_dir_unmap(struct page **dir)
103 kunmap_atomic(dir, KM_USER0);
106 static swp_entry_t *shmem_swp_map(struct page *page)
108 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
111 static inline void shmem_swp_balance_unmap(void)
114 * When passing a pointer to an i_direct entry, to code which
115 * also handles indirect entries and so will shmem_swp_unmap,
116 * we must arrange for the preempt count to remain in balance.
117 * What kmap_atomic of a lowmem page does depends on config
118 * and architecture, so pretend to kmap_atomic some lowmem page.
120 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
123 static inline void shmem_swp_unmap(swp_entry_t *entry)
125 kunmap_atomic(entry, KM_USER1);
128 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
130 return sb->s_fs_info;
134 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
135 * for shared memory and for shared anonymous (/dev/zero) mappings
136 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
137 * consistent with the pre-accounting of private mappings ...
139 static inline int shmem_acct_size(unsigned long flags, loff_t size)
141 return (flags & VM_ACCOUNT)?
142 security_vm_enough_memory(VM_ACCT(size)): 0;
145 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
147 if (flags & VM_ACCOUNT)
148 vm_unacct_memory(VM_ACCT(size));
152 * ... whereas tmpfs objects are accounted incrementally as
153 * pages are allocated, in order to allow huge sparse files.
154 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
155 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
157 static inline int shmem_acct_block(unsigned long flags)
159 return (flags & VM_ACCOUNT)?
160 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
163 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
165 if (!(flags & VM_ACCOUNT))
166 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
169 static struct super_operations shmem_ops;
170 static struct address_space_operations shmem_aops;
171 static struct file_operations shmem_file_operations;
172 static struct inode_operations shmem_inode_operations;
173 static struct inode_operations shmem_dir_inode_operations;
174 static struct vm_operations_struct shmem_vm_ops;
176 static struct backing_dev_info shmem_backing_dev_info = {
177 .ra_pages = 0, /* No readahead */
178 .memory_backed = 1, /* Does not contribute to dirty memory */
179 .unplug_io_fn = default_unplug_io_fn,
182 LIST_HEAD(shmem_inodes);
183 static spinlock_t shmem_ilock = SPIN_LOCK_UNLOCKED;
185 static void shmem_free_block(struct inode *inode)
187 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
188 spin_lock(&sbinfo->stat_lock);
189 sbinfo->free_blocks++;
190 inode->i_blocks -= BLOCKS_PER_PAGE;
191 spin_unlock(&sbinfo->stat_lock);
195 * shmem_recalc_inode - recalculate the size of an inode
197 * @inode: inode to recalc
199 * We have to calculate the free blocks since the mm can drop
200 * undirtied hole pages behind our back.
202 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
203 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
205 * It has to be called with the spinlock held.
207 static void shmem_recalc_inode(struct inode *inode)
209 struct shmem_inode_info *info = SHMEM_I(inode);
212 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
214 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
215 info->alloced -= freed;
216 spin_lock(&sbinfo->stat_lock);
217 sbinfo->free_blocks += freed;
218 inode->i_blocks -= freed*BLOCKS_PER_PAGE;
219 spin_unlock(&sbinfo->stat_lock);
220 shmem_unacct_blocks(info->flags, freed);
225 * shmem_swp_entry - find the swap vector position in the info structure
227 * @info: info structure for the inode
228 * @index: index of the page to find
229 * @page: optional page to add to the structure. Has to be preset to
232 * If there is no space allocated yet it will return NULL when
233 * page is NULL, else it will use the page for the needed block,
234 * setting it to NULL on return to indicate that it has been used.
236 * The swap vector is organized the following way:
238 * There are SHMEM_NR_DIRECT entries directly stored in the
239 * shmem_inode_info structure. So small files do not need an addional
242 * For pages with index > SHMEM_NR_DIRECT there is the pointer
243 * i_indirect which points to a page which holds in the first half
244 * doubly indirect blocks, in the second half triple indirect blocks:
246 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
247 * following layout (for SHMEM_NR_DIRECT == 16):
249 * i_indirect -> dir --> 16-19
262 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
264 unsigned long offset;
268 if (index < SHMEM_NR_DIRECT) {
269 shmem_swp_balance_unmap();
270 return info->i_direct+index;
272 if (!info->i_indirect) {
274 info->i_indirect = *page;
277 return NULL; /* need another page */
280 index -= SHMEM_NR_DIRECT;
281 offset = index % ENTRIES_PER_PAGE;
282 index /= ENTRIES_PER_PAGE;
283 dir = shmem_dir_map(info->i_indirect);
285 if (index >= ENTRIES_PER_PAGE/2) {
286 index -= ENTRIES_PER_PAGE/2;
287 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
288 index %= ENTRIES_PER_PAGE;
295 shmem_dir_unmap(dir);
296 return NULL; /* need another page */
298 shmem_dir_unmap(dir);
299 dir = shmem_dir_map(subdir);
305 if (!page || !(subdir = *page)) {
306 shmem_dir_unmap(dir);
307 return NULL; /* need a page */
312 shmem_dir_unmap(dir);
313 return shmem_swp_map(subdir) + offset;
316 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
318 long incdec = value? 1: -1;
321 info->swapped += incdec;
322 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
323 kmap_atomic_to_page(entry)->nr_swapped += incdec;
327 * shmem_swp_alloc - get the position of the swap entry for the page.
328 * If it does not exist allocate the entry.
330 * @info: info structure for the inode
331 * @index: index of the page to find
332 * @sgp: check and recheck i_size? skip allocation?
334 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
336 struct inode *inode = &info->vfs_inode;
337 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
338 struct page *page = NULL;
341 if (sgp != SGP_WRITE &&
342 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
343 return ERR_PTR(-EINVAL);
345 while (!(entry = shmem_swp_entry(info, index, &page))) {
347 return shmem_swp_map(ZERO_PAGE(0));
349 * Test free_blocks against 1 not 0, since we have 1 data
350 * page (and perhaps indirect index pages) yet to allocate:
351 * a waste to allocate index if we cannot allocate data.
353 spin_lock(&sbinfo->stat_lock);
354 if (sbinfo->free_blocks <= 1) {
355 spin_unlock(&sbinfo->stat_lock);
356 return ERR_PTR(-ENOSPC);
358 sbinfo->free_blocks--;
359 inode->i_blocks += BLOCKS_PER_PAGE;
360 spin_unlock(&sbinfo->stat_lock);
362 spin_unlock(&info->lock);
363 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
365 clear_highpage(page);
366 page->nr_swapped = 0;
368 spin_lock(&info->lock);
371 shmem_free_block(inode);
372 return ERR_PTR(-ENOMEM);
374 if (sgp != SGP_WRITE &&
375 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
376 entry = ERR_PTR(-EINVAL);
379 if (info->next_index <= index)
380 info->next_index = index + 1;
383 /* another task gave its page, or truncated the file */
384 shmem_free_block(inode);
385 shmem_dir_free(page);
387 if (info->next_index <= index && !IS_ERR(entry))
388 info->next_index = index + 1;
393 * shmem_free_swp - free some swap entries in a directory
395 * @dir: pointer to the directory
396 * @edir: pointer after last entry of the directory
398 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
403 for (ptr = dir; ptr < edir; ptr++) {
405 free_swap_and_cache(*ptr);
406 *ptr = (swp_entry_t){0};
413 static void shmem_truncate(struct inode *inode)
415 struct shmem_inode_info *info = SHMEM_I(inode);
427 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
428 idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
429 if (idx >= info->next_index)
432 spin_lock(&info->lock);
433 info->flags |= SHMEM_TRUNCATE;
434 limit = info->next_index;
435 info->next_index = idx;
436 if (info->swapped && idx < SHMEM_NR_DIRECT) {
437 ptr = info->i_direct;
439 if (size > SHMEM_NR_DIRECT)
440 size = SHMEM_NR_DIRECT;
441 info->swapped -= shmem_free_swp(ptr+idx, ptr+size);
443 if (!info->i_indirect)
446 BUG_ON(limit <= SHMEM_NR_DIRECT);
447 limit -= SHMEM_NR_DIRECT;
448 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
449 offset = idx % ENTRIES_PER_PAGE;
453 dir = shmem_dir_map(info->i_indirect);
454 stage = ENTRIES_PER_PAGEPAGE/2;
455 if (idx < ENTRIES_PER_PAGEPAGE/2)
456 dir += idx/ENTRIES_PER_PAGE;
458 dir += ENTRIES_PER_PAGE/2;
459 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
461 stage += ENTRIES_PER_PAGEPAGE;
464 size = ((idx - ENTRIES_PER_PAGEPAGE/2) %
465 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
466 if (!size && !offset) {
470 shmem_dir_unmap(dir);
471 dir = shmem_dir_map(subdir) + size;
478 for (; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
479 if (unlikely(idx == stage)) {
480 shmem_dir_unmap(dir-1);
481 dir = shmem_dir_map(info->i_indirect) +
482 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
485 idx += ENTRIES_PER_PAGEPAGE;
489 stage = idx + ENTRIES_PER_PAGEPAGE;
492 shmem_dir_unmap(dir);
494 shmem_dir_free(empty);
495 shmem_free_block(inode);
498 cond_resched_lock(&info->lock);
499 dir = shmem_dir_map(subdir);
502 if (subdir && subdir->nr_swapped) {
503 ptr = shmem_swp_map(subdir);
505 if (size > ENTRIES_PER_PAGE)
506 size = ENTRIES_PER_PAGE;
507 freed = shmem_free_swp(ptr+offset, ptr+size);
508 shmem_swp_unmap(ptr);
509 info->swapped -= freed;
510 subdir->nr_swapped -= freed;
511 BUG_ON(subdir->nr_swapped > offset);
517 shmem_dir_free(subdir);
518 shmem_free_block(inode);
522 shmem_dir_unmap(dir-1);
524 shmem_dir_free(empty);
525 shmem_free_block(inode);
527 if (info->next_index <= SHMEM_NR_DIRECT) {
528 shmem_dir_free(info->i_indirect);
529 info->i_indirect = NULL;
530 shmem_free_block(inode);
533 BUG_ON(info->swapped > info->next_index);
534 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
536 * Call truncate_inode_pages again: racing shmem_unuse_inode
537 * may have swizzled a page in from swap since vmtruncate or
538 * generic_delete_inode did it, before we lowered next_index.
539 * Also, though shmem_getpage checks i_size before adding to
540 * cache, no recheck after: so fix the narrow window there too.
542 spin_unlock(&info->lock);
543 truncate_inode_pages(inode->i_mapping, inode->i_size);
544 spin_lock(&info->lock);
546 info->flags &= ~SHMEM_TRUNCATE;
547 shmem_recalc_inode(inode);
548 spin_unlock(&info->lock);
551 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
553 struct inode *inode = dentry->d_inode;
554 struct page *page = NULL;
557 if (attr->ia_valid & ATTR_SIZE) {
558 if (attr->ia_size < inode->i_size) {
560 * If truncating down to a partial page, then
561 * if that page is already allocated, hold it
562 * in memory until the truncation is over, so
563 * truncate_partial_page cannnot miss it were
564 * it assigned to swap.
566 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
567 (void) shmem_getpage(inode,
568 attr->ia_size>>PAGE_CACHE_SHIFT,
569 &page, SGP_READ, NULL);
572 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
573 * detect if any pages might have been added to cache
574 * after truncate_inode_pages. But we needn't bother
575 * if it's being fully truncated to zero-length: the
576 * nrpages check is efficient enough in that case.
579 struct shmem_inode_info *info = SHMEM_I(inode);
580 spin_lock(&info->lock);
581 info->flags &= ~SHMEM_PAGEIN;
582 spin_unlock(&info->lock);
587 error = inode_change_ok(inode, attr);
589 error = inode_setattr(inode, attr);
591 page_cache_release(page);
595 static void shmem_delete_inode(struct inode *inode)
597 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
598 struct shmem_inode_info *info = SHMEM_I(inode);
600 if (inode->i_op->truncate == shmem_truncate) {
601 spin_lock(&shmem_ilock);
602 list_del(&info->list);
603 spin_unlock(&shmem_ilock);
604 shmem_unacct_size(info->flags, inode->i_size);
606 shmem_truncate(inode);
608 BUG_ON(inode->i_blocks);
609 spin_lock(&sbinfo->stat_lock);
610 sbinfo->free_inodes++;
611 spin_unlock(&sbinfo->stat_lock);
615 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
619 for (ptr = dir; ptr < edir; ptr++) {
620 if (ptr->val == entry.val)
626 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
639 ptr = info->i_direct;
640 spin_lock(&info->lock);
641 limit = info->next_index;
643 if (size > SHMEM_NR_DIRECT)
644 size = SHMEM_NR_DIRECT;
645 offset = shmem_find_swp(entry, ptr, ptr+size);
647 shmem_swp_balance_unmap();
650 if (!info->i_indirect)
652 /* we might be racing with shmem_truncate */
653 if (limit <= SHMEM_NR_DIRECT)
656 dir = shmem_dir_map(info->i_indirect);
657 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
659 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
660 if (unlikely(idx == stage)) {
661 shmem_dir_unmap(dir-1);
662 dir = shmem_dir_map(info->i_indirect) +
663 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
666 idx += ENTRIES_PER_PAGEPAGE;
670 stage = idx + ENTRIES_PER_PAGEPAGE;
672 shmem_dir_unmap(dir);
673 dir = shmem_dir_map(subdir);
676 if (subdir && subdir->nr_swapped) {
677 ptr = shmem_swp_map(subdir);
679 if (size > ENTRIES_PER_PAGE)
680 size = ENTRIES_PER_PAGE;
681 offset = shmem_find_swp(entry, ptr, ptr+size);
683 shmem_dir_unmap(dir);
686 shmem_swp_unmap(ptr);
690 shmem_dir_unmap(dir-1);
692 spin_unlock(&info->lock);
696 inode = &info->vfs_inode;
697 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
698 info->flags |= SHMEM_PAGEIN;
699 shmem_swp_set(info, ptr + offset, 0);
701 shmem_swp_unmap(ptr);
702 spin_unlock(&info->lock);
704 * Decrement swap count even when the entry is left behind:
705 * try_to_unuse will skip over mms, then reincrement count.
712 * shmem_unuse() search for an eventually swapped out shmem page.
714 int shmem_unuse(swp_entry_t entry, struct page *page)
717 struct shmem_inode_info *info;
720 spin_lock(&shmem_ilock);
721 list_for_each(p, &shmem_inodes) {
722 info = list_entry(p, struct shmem_inode_info, list);
724 if (info->swapped && shmem_unuse_inode(info, entry, page)) {
725 /* move head to start search for next from here */
726 list_move_tail(&shmem_inodes, &info->list);
731 spin_unlock(&shmem_ilock);
736 * Move the page from the page cache to the swap cache.
738 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
740 struct shmem_inode_info *info;
741 swp_entry_t *entry, swap;
742 struct address_space *mapping;
746 BUG_ON(!PageLocked(page));
747 BUG_ON(page_mapped(page));
749 mapping = page->mapping;
751 inode = mapping->host;
752 info = SHMEM_I(inode);
753 if (info->flags & VM_LOCKED)
755 swap = get_swap_page();
759 spin_lock(&info->lock);
760 shmem_recalc_inode(inode);
761 if (index >= info->next_index) {
762 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
765 entry = shmem_swp_entry(info, index, NULL);
769 if (move_to_swap_cache(page, swap) == 0) {
770 shmem_swp_set(info, entry, swap.val);
771 shmem_swp_unmap(entry);
772 spin_unlock(&info->lock);
777 shmem_swp_unmap(entry);
779 spin_unlock(&info->lock);
782 set_page_dirty(page);
783 return WRITEPAGE_ACTIVATE; /* Return with the page locked */
787 static struct page *shmem_swapin_async(struct shared_policy *p,
788 swp_entry_t entry, unsigned long idx)
791 struct vm_area_struct pvma;
793 /* Create a pseudo vma that just contains the policy */
794 memset(&pvma, 0, sizeof(struct vm_area_struct));
795 pvma.vm_end = PAGE_SIZE;
797 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
798 page = read_swap_cache_async(entry, &pvma, 0);
799 mpol_free(pvma.vm_policy);
803 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
806 struct shared_policy *p = &info->policy;
809 unsigned long offset;
811 num = valid_swaphandles(entry, &offset);
812 for (i = 0; i < num; offset++, i++) {
813 page = shmem_swapin_async(p,
814 swp_entry(swp_type(entry), offset), idx);
817 page_cache_release(page);
819 lru_add_drain(); /* Push any new pages onto the LRU now */
820 return shmem_swapin_async(p, entry, idx);
824 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
827 struct vm_area_struct pvma;
830 memset(&pvma, 0, sizeof(struct vm_area_struct));
831 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
833 pvma.vm_end = PAGE_SIZE;
834 page = alloc_page_vma(gfp, &pvma, 0);
835 mpol_free(pvma.vm_policy);
839 static inline struct page *
840 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
842 swapin_readahead(entry, 0, NULL);
843 return read_swap_cache_async(entry, NULL, 0);
846 static inline struct page *
847 shmem_alloc_page(unsigned long gfp,struct shmem_inode_info *info,
850 return alloc_page(gfp);
855 * shmem_getpage - either get the page from swap or allocate a new one
857 * If we allocate a new one we do not mark it dirty. That's up to the
858 * vm. If we swap it in we mark it dirty since we also free the swap
859 * entry since a page cannot live in both the swap and page cache
861 static int shmem_getpage(struct inode *inode, unsigned long idx,
862 struct page **pagep, enum sgp_type sgp, int *type)
864 struct address_space *mapping = inode->i_mapping;
865 struct shmem_inode_info *info = SHMEM_I(inode);
866 struct shmem_sb_info *sbinfo;
867 struct page *filepage = *pagep;
868 struct page *swappage;
871 int error, majmin = VM_FAULT_MINOR;
873 if (idx >= SHMEM_MAX_INDEX)
876 * Normally, filepage is NULL on entry, and either found
877 * uptodate immediately, or allocated and zeroed, or read
878 * in under swappage, which is then assigned to filepage.
879 * But shmem_prepare_write passes in a locked filepage,
880 * which may be found not uptodate by other callers too,
881 * and may need to be copied from the swappage read in.
885 filepage = find_lock_page(mapping, idx);
886 if (filepage && PageUptodate(filepage))
889 if (sgp == SGP_QUICK)
892 spin_lock(&info->lock);
893 shmem_recalc_inode(inode);
894 entry = shmem_swp_alloc(info, idx, sgp);
896 spin_unlock(&info->lock);
897 error = PTR_ERR(entry);
903 /* Look it up and read it in.. */
904 swappage = lookup_swap_cache(swap);
906 shmem_swp_unmap(entry);
907 spin_unlock(&info->lock);
908 /* here we actually do the io */
909 if (majmin == VM_FAULT_MINOR && type)
910 inc_page_state(pgmajfault);
911 majmin = VM_FAULT_MAJOR;
912 swappage = shmem_swapin(info, swap, idx);
914 spin_lock(&info->lock);
915 entry = shmem_swp_alloc(info, idx, sgp);
917 error = PTR_ERR(entry);
919 if (entry->val == swap.val)
921 shmem_swp_unmap(entry);
923 spin_unlock(&info->lock);
928 wait_on_page_locked(swappage);
929 page_cache_release(swappage);
933 /* We have to do this with page locked to prevent races */
934 if (TestSetPageLocked(swappage)) {
935 shmem_swp_unmap(entry);
936 spin_unlock(&info->lock);
937 wait_on_page_locked(swappage);
938 page_cache_release(swappage);
941 if (PageWriteback(swappage)) {
942 shmem_swp_unmap(entry);
943 spin_unlock(&info->lock);
944 wait_on_page_writeback(swappage);
945 unlock_page(swappage);
946 page_cache_release(swappage);
949 if (!PageUptodate(swappage)) {
950 shmem_swp_unmap(entry);
951 spin_unlock(&info->lock);
952 unlock_page(swappage);
953 page_cache_release(swappage);
959 shmem_swp_set(info, entry, 0);
960 shmem_swp_unmap(entry);
961 delete_from_swap_cache(swappage);
962 spin_unlock(&info->lock);
963 copy_highpage(filepage, swappage);
964 unlock_page(swappage);
965 page_cache_release(swappage);
966 flush_dcache_page(filepage);
967 SetPageUptodate(filepage);
968 set_page_dirty(filepage);
970 } else if (!(error = move_from_swap_cache(
971 swappage, idx, mapping))) {
972 info->flags |= SHMEM_PAGEIN;
973 shmem_swp_set(info, entry, 0);
974 shmem_swp_unmap(entry);
975 spin_unlock(&info->lock);
979 shmem_swp_unmap(entry);
980 spin_unlock(&info->lock);
981 unlock_page(swappage);
982 page_cache_release(swappage);
983 if (error == -ENOMEM) {
984 /* let kswapd refresh zone for GFP_ATOMICs */
985 blk_congestion_wait(WRITE, HZ/50);
989 } else if (sgp == SGP_READ && !filepage) {
990 shmem_swp_unmap(entry);
991 filepage = find_get_page(mapping, idx);
993 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
994 spin_unlock(&info->lock);
995 wait_on_page_locked(filepage);
996 page_cache_release(filepage);
1000 spin_unlock(&info->lock);
1002 shmem_swp_unmap(entry);
1003 sbinfo = SHMEM_SB(inode->i_sb);
1004 spin_lock(&sbinfo->stat_lock);
1005 if (sbinfo->free_blocks == 0 || shmem_acct_block(info->flags)) {
1006 spin_unlock(&sbinfo->stat_lock);
1007 spin_unlock(&info->lock);
1011 sbinfo->free_blocks--;
1012 inode->i_blocks += BLOCKS_PER_PAGE;
1013 spin_unlock(&sbinfo->stat_lock);
1016 spin_unlock(&info->lock);
1017 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1021 shmem_unacct_blocks(info->flags, 1);
1022 shmem_free_block(inode);
1027 spin_lock(&info->lock);
1028 entry = shmem_swp_alloc(info, idx, sgp);
1030 error = PTR_ERR(entry);
1033 shmem_swp_unmap(entry);
1035 if (error || swap.val || 0 != add_to_page_cache_lru(
1036 filepage, mapping, idx, GFP_ATOMIC)) {
1037 spin_unlock(&info->lock);
1038 page_cache_release(filepage);
1039 shmem_unacct_blocks(info->flags, 1);
1040 shmem_free_block(inode);
1046 info->flags |= SHMEM_PAGEIN;
1050 spin_unlock(&info->lock);
1051 clear_highpage(filepage);
1052 flush_dcache_page(filepage);
1053 SetPageUptodate(filepage);
1058 unlock_page(filepage);
1061 *pagep = ZERO_PAGE(0);
1068 if (*pagep != filepage) {
1069 unlock_page(filepage);
1070 page_cache_release(filepage);
1075 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1077 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1078 struct page *page = NULL;
1082 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1083 idx += vma->vm_pgoff;
1084 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1086 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1088 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1090 mark_page_accessed(page);
1094 static int shmem_populate(struct vm_area_struct *vma,
1095 unsigned long addr, unsigned long len,
1096 pgprot_t prot, unsigned long pgoff, int nonblock)
1098 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1099 struct mm_struct *mm = vma->vm_mm;
1100 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1103 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1104 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1107 while ((long) len > 0) {
1108 struct page *page = NULL;
1111 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1113 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1117 mark_page_accessed(page);
1118 err = install_page(mm, vma, addr, page, prot);
1120 page_cache_release(page);
1123 } else if (nonblock) {
1125 * If a nonlinear mapping then store the file page
1126 * offset in the pte.
1128 if (pgoff != linear_page_index(vma, addr)) {
1129 err = install_file_pte(mm, vma, addr, pgoff, prot);
1143 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1145 struct inode *i = vma->vm_file->f_dentry->d_inode;
1146 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1150 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1152 struct inode *i = vma->vm_file->f_dentry->d_inode;
1155 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1156 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1160 /* Protects current->user->locked_shm from concurrent access */
1161 static spinlock_t shmem_lock_user = SPIN_LOCK_UNLOCKED;
1163 int shmem_lock(struct file *file, int lock, struct user_struct * user)
1165 struct inode *inode = file->f_dentry->d_inode;
1166 struct shmem_inode_info *info = SHMEM_I(inode);
1167 unsigned long lock_limit, locked;
1168 int retval = -ENOMEM;
1170 spin_lock(&info->lock);
1171 spin_lock(&shmem_lock_user);
1172 if (lock && !(info->flags & VM_LOCKED)) {
1173 locked = inode->i_size >> PAGE_SHIFT;
1174 locked += user->locked_shm;
1175 lock_limit = current->rlim[RLIMIT_MEMLOCK].rlim_cur;
1176 lock_limit >>= PAGE_SHIFT;
1177 if ((locked > lock_limit) && !capable(CAP_IPC_LOCK))
1179 /* for this branch user == current->user so it won't go away under us */
1180 atomic_inc(&user->__count);
1181 user->locked_shm = locked;
1183 if (!lock && (info->flags & VM_LOCKED) && user) {
1184 locked = inode->i_size >> PAGE_SHIFT;
1185 user->locked_shm -= locked;
1189 info->flags |= VM_LOCKED;
1191 info->flags &= ~VM_LOCKED;
1194 spin_unlock(&shmem_lock_user);
1195 spin_unlock(&info->lock);
1199 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1201 file_accessed(file);
1202 vma->vm_ops = &shmem_vm_ops;
1206 static struct inode *
1207 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1209 struct inode *inode;
1210 struct shmem_inode_info *info;
1211 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1213 spin_lock(&sbinfo->stat_lock);
1214 if (!sbinfo->free_inodes) {
1215 spin_unlock(&sbinfo->stat_lock);
1218 sbinfo->free_inodes--;
1219 spin_unlock(&sbinfo->stat_lock);
1221 inode = new_inode(sb);
1223 inode->i_mode = mode;
1224 inode->i_uid = current->fsuid;
1225 inode->i_gid = current->fsgid;
1226 inode->i_blksize = PAGE_CACHE_SIZE;
1227 inode->i_blocks = 0;
1228 inode->i_mapping->a_ops = &shmem_aops;
1229 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1230 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1231 info = SHMEM_I(inode);
1232 memset(info, 0, (char *)inode - (char *)info);
1233 spin_lock_init(&info->lock);
1234 mpol_shared_policy_init(&info->policy);
1235 switch (mode & S_IFMT) {
1237 init_special_inode(inode, mode, dev);
1240 inode->i_op = &shmem_inode_operations;
1241 inode->i_fop = &shmem_file_operations;
1242 spin_lock(&shmem_ilock);
1243 list_add_tail(&info->list, &shmem_inodes);
1244 spin_unlock(&shmem_ilock);
1248 /* Some things misbehave if size == 0 on a directory */
1249 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1250 inode->i_op = &shmem_dir_inode_operations;
1251 inode->i_fop = &simple_dir_operations;
1260 static int shmem_set_size(struct shmem_sb_info *info,
1261 unsigned long max_blocks, unsigned long max_inodes)
1264 unsigned long blocks, inodes;
1266 spin_lock(&info->stat_lock);
1267 blocks = info->max_blocks - info->free_blocks;
1268 inodes = info->max_inodes - info->free_inodes;
1270 if (max_blocks < blocks)
1272 if (max_inodes < inodes)
1275 info->max_blocks = max_blocks;
1276 info->free_blocks = max_blocks - blocks;
1277 info->max_inodes = max_inodes;
1278 info->free_inodes = max_inodes - inodes;
1280 spin_unlock(&info->stat_lock);
1286 static struct inode_operations shmem_symlink_inode_operations;
1287 static struct inode_operations shmem_symlink_inline_operations;
1290 * Normally tmpfs makes no use of shmem_prepare_write, but it
1291 * lets a tmpfs file be used read-write below the loop driver.
1294 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1296 struct inode *inode = page->mapping->host;
1297 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1301 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1303 struct inode *inode = file->f_dentry->d_inode;
1305 unsigned long written;
1308 if ((ssize_t) count < 0)
1311 if (!access_ok(VERIFY_READ, buf, count))
1314 down(&inode->i_sem);
1319 err = generic_write_checks(file, &pos, &count, 0);
1323 err = remove_suid(file->f_dentry);
1327 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1330 struct page *page = NULL;
1331 unsigned long bytes, index, offset;
1335 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1336 index = pos >> PAGE_CACHE_SHIFT;
1337 bytes = PAGE_CACHE_SIZE - offset;
1342 * We don't hold page lock across copy from user -
1343 * what would it guard against? - so no deadlock here.
1344 * But it still may be a good idea to prefault below.
1347 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1352 if (PageHighMem(page)) {
1353 volatile unsigned char dummy;
1354 __get_user(dummy, buf);
1355 __get_user(dummy, buf + bytes - 1);
1357 kaddr = kmap_atomic(page, KM_USER0);
1358 left = __copy_from_user(kaddr + offset, buf, bytes);
1359 kunmap_atomic(kaddr, KM_USER0);
1363 left = __copy_from_user(kaddr + offset, buf, bytes);
1371 if (pos > inode->i_size)
1372 i_size_write(inode, pos);
1374 flush_dcache_page(page);
1375 set_page_dirty(page);
1376 mark_page_accessed(page);
1377 page_cache_release(page);
1387 * Our dirty pages are not counted in nr_dirty,
1388 * and we do not attempt to balance dirty pages.
1402 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1404 struct inode *inode = filp->f_dentry->d_inode;
1405 struct address_space *mapping = inode->i_mapping;
1406 unsigned long index, offset;
1408 index = *ppos >> PAGE_CACHE_SHIFT;
1409 offset = *ppos & ~PAGE_CACHE_MASK;
1412 struct page *page = NULL;
1413 unsigned long end_index, nr, ret;
1414 loff_t i_size = i_size_read(inode);
1416 end_index = i_size >> PAGE_CACHE_SHIFT;
1417 if (index > end_index)
1419 if (index == end_index) {
1420 nr = i_size & ~PAGE_CACHE_MASK;
1425 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1427 if (desc->error == -EINVAL)
1433 * We must evaluate after, since reads (unlike writes)
1434 * are called without i_sem protection against truncate
1436 nr = PAGE_CACHE_SIZE;
1437 i_size = i_size_read(inode);
1438 end_index = i_size >> PAGE_CACHE_SHIFT;
1439 if (index == end_index) {
1440 nr = i_size & ~PAGE_CACHE_MASK;
1442 page_cache_release(page);
1448 if (page != ZERO_PAGE(0)) {
1450 * If users can be writing to this page using arbitrary
1451 * virtual addresses, take care about potential aliasing
1452 * before reading the page on the kernel side.
1454 if (mapping_writably_mapped(mapping))
1455 flush_dcache_page(page);
1457 * Mark the page accessed if we read the beginning.
1460 mark_page_accessed(page);
1464 * Ok, we have the page, and it's up-to-date, so
1465 * now we can copy it to user space...
1467 * The actor routine returns how many bytes were actually used..
1468 * NOTE! This may not be the same as how much of a user buffer
1469 * we filled up (we may be padding etc), so we can only update
1470 * "pos" here (the actor routine has to update the user buffer
1471 * pointers and the remaining count).
1473 ret = actor(desc, page, offset, nr);
1475 index += offset >> PAGE_CACHE_SHIFT;
1476 offset &= ~PAGE_CACHE_MASK;
1478 page_cache_release(page);
1479 if (ret != nr || !desc->count)
1485 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1486 file_accessed(filp);
1489 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1491 read_descriptor_t desc;
1493 if ((ssize_t) count < 0)
1495 if (!access_ok(VERIFY_WRITE, buf, count))
1505 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1507 return desc.written;
1511 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1512 size_t count, read_actor_t actor, void *target)
1514 read_descriptor_t desc;
1521 desc.arg.data = target;
1524 do_shmem_file_read(in_file, ppos, &desc, actor);
1526 return desc.written;
1530 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1532 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1534 buf->f_type = TMPFS_MAGIC;
1535 buf->f_bsize = PAGE_CACHE_SIZE;
1536 spin_lock(&sbinfo->stat_lock);
1537 buf->f_blocks = sbinfo->max_blocks;
1538 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1539 buf->f_files = sbinfo->max_inodes;
1540 buf->f_ffree = sbinfo->free_inodes;
1541 spin_unlock(&sbinfo->stat_lock);
1542 buf->f_namelen = NAME_MAX;
1547 * File creation. Allocate an inode, and we're done..
1550 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1552 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1553 int error = -ENOSPC;
1556 if (dir->i_mode & S_ISGID) {
1557 inode->i_gid = dir->i_gid;
1559 inode->i_mode |= S_ISGID;
1561 dir->i_size += BOGO_DIRENT_SIZE;
1562 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1563 d_instantiate(dentry, inode);
1564 dget(dentry); /* Extra count - pin the dentry in core */
1570 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1574 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1580 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1581 struct nameidata *nd)
1583 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1589 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1591 struct inode *inode = old_dentry->d_inode;
1593 dir->i_size += BOGO_DIRENT_SIZE;
1594 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1596 atomic_inc(&inode->i_count); /* New dentry reference */
1597 dget(dentry); /* Extra pinning count for the created dentry */
1598 d_instantiate(dentry, inode);
1602 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1604 struct inode *inode = dentry->d_inode;
1606 dir->i_size -= BOGO_DIRENT_SIZE;
1607 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1609 dput(dentry); /* Undo the count from "create" - this does all the work */
1613 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1615 if (!simple_empty(dentry))
1619 return shmem_unlink(dir, dentry);
1623 * The VFS layer already does all the dentry stuff for rename,
1624 * we just have to decrement the usage count for the target if
1625 * it exists so that the VFS layer correctly free's it when it
1628 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1630 struct inode *inode = old_dentry->d_inode;
1631 int they_are_dirs = S_ISDIR(inode->i_mode);
1633 if (!simple_empty(new_dentry))
1636 if (new_dentry->d_inode) {
1637 (void) shmem_unlink(new_dir, new_dentry);
1640 } else if (they_are_dirs) {
1645 old_dir->i_size -= BOGO_DIRENT_SIZE;
1646 new_dir->i_size += BOGO_DIRENT_SIZE;
1647 old_dir->i_ctime = old_dir->i_mtime =
1648 new_dir->i_ctime = new_dir->i_mtime =
1649 inode->i_ctime = CURRENT_TIME;
1653 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1657 struct inode *inode;
1658 struct page *page = NULL;
1660 struct shmem_inode_info *info;
1662 len = strlen(symname) + 1;
1663 if (len > PAGE_CACHE_SIZE)
1664 return -ENAMETOOLONG;
1666 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1670 info = SHMEM_I(inode);
1671 inode->i_size = len-1;
1672 if (len <= (char *)inode - (char *)info) {
1674 memcpy(info, symname, len);
1675 inode->i_op = &shmem_symlink_inline_operations;
1677 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1682 inode->i_op = &shmem_symlink_inode_operations;
1683 spin_lock(&shmem_ilock);
1684 list_add_tail(&info->list, &shmem_inodes);
1685 spin_unlock(&shmem_ilock);
1686 kaddr = kmap_atomic(page, KM_USER0);
1687 memcpy(kaddr, symname, len);
1688 kunmap_atomic(kaddr, KM_USER0);
1689 set_page_dirty(page);
1690 page_cache_release(page);
1692 if (dir->i_mode & S_ISGID)
1693 inode->i_gid = dir->i_gid;
1694 dir->i_size += BOGO_DIRENT_SIZE;
1695 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1696 d_instantiate(dentry, inode);
1701 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1703 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1707 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1709 struct page *page = NULL;
1710 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1711 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1715 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1717 if (!IS_ERR(nd_get_link(nd))) {
1720 page = find_get_page(dentry->d_inode->i_mapping, 0);
1724 mark_page_accessed(page);
1725 page_cache_release(page);
1726 page_cache_release(page);
1730 static struct inode_operations shmem_symlink_inline_operations = {
1731 .readlink = generic_readlink,
1732 .follow_link = shmem_follow_link_inline,
1735 static struct inode_operations shmem_symlink_inode_operations = {
1736 .truncate = shmem_truncate,
1737 .readlink = generic_readlink,
1738 .follow_link = shmem_follow_link,
1739 .put_link = shmem_put_link,
1742 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1744 char *this_char, *value, *rest;
1746 while ((this_char = strsep(&options, ",")) != NULL) {
1749 if ((value = strchr(this_char,'=')) != NULL) {
1753 "tmpfs: No value for mount option '%s'\n",
1758 if (!strcmp(this_char,"size")) {
1759 unsigned long long size;
1760 size = memparse(value,&rest);
1762 size <<= PAGE_SHIFT;
1763 size *= totalram_pages;
1769 *blocks = size >> PAGE_CACHE_SHIFT;
1770 } else if (!strcmp(this_char,"nr_blocks")) {
1771 *blocks = memparse(value,&rest);
1774 } else if (!strcmp(this_char,"nr_inodes")) {
1775 *inodes = memparse(value,&rest);
1778 } else if (!strcmp(this_char,"mode")) {
1781 *mode = simple_strtoul(value,&rest,8);
1784 } else if (!strcmp(this_char,"uid")) {
1787 *uid = simple_strtoul(value,&rest,0);
1790 } else if (!strcmp(this_char,"gid")) {
1793 *gid = simple_strtoul(value,&rest,0);
1797 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1805 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1811 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1813 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1814 unsigned long max_blocks = sbinfo->max_blocks;
1815 unsigned long max_inodes = sbinfo->max_inodes;
1817 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
1819 return shmem_set_size(sbinfo, max_blocks, max_inodes);
1823 static int shmem_fill_super(struct super_block *sb,
1824 void *data, int silent)
1826 struct inode *inode;
1827 struct dentry *root;
1828 unsigned long blocks, inodes;
1829 int mode = S_IRWXUGO | S_ISVTX;
1830 uid_t uid = current->fsuid;
1831 gid_t gid = current->fsgid;
1832 struct shmem_sb_info *sbinfo;
1835 sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
1838 sb->s_fs_info = sbinfo;
1839 memset(sbinfo, 0, sizeof(struct shmem_sb_info));
1842 * Per default we only allow half of the physical ram per
1845 blocks = inodes = totalram_pages / 2;
1848 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, &inodes)) {
1853 sb->s_flags |= MS_NOUSER;
1856 spin_lock_init(&sbinfo->stat_lock);
1857 sbinfo->max_blocks = blocks;
1858 sbinfo->free_blocks = blocks;
1859 sbinfo->max_inodes = inodes;
1860 sbinfo->free_inodes = inodes;
1861 sb->s_maxbytes = SHMEM_MAX_BYTES;
1862 sb->s_blocksize = PAGE_CACHE_SIZE;
1863 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1864 sb->s_magic = TMPFS_MAGIC;
1865 sb->s_op = &shmem_ops;
1866 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1871 root = d_alloc_root(inode);
1881 sb->s_fs_info = NULL;
1885 static void shmem_put_super(struct super_block *sb)
1887 kfree(sb->s_fs_info);
1888 sb->s_fs_info = NULL;
1891 static kmem_cache_t *shmem_inode_cachep;
1893 static struct inode *shmem_alloc_inode(struct super_block *sb)
1895 struct shmem_inode_info *p;
1896 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
1899 return &p->vfs_inode;
1902 static void shmem_destroy_inode(struct inode *inode)
1904 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
1905 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
1908 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
1910 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
1912 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1913 SLAB_CTOR_CONSTRUCTOR) {
1914 inode_init_once(&p->vfs_inode);
1918 static int init_inodecache(void)
1920 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
1921 sizeof(struct shmem_inode_info),
1922 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
1924 if (shmem_inode_cachep == NULL)
1929 static void destroy_inodecache(void)
1931 if (kmem_cache_destroy(shmem_inode_cachep))
1932 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
1935 static struct address_space_operations shmem_aops = {
1936 .writepage = shmem_writepage,
1937 .set_page_dirty = __set_page_dirty_nobuffers,
1939 .prepare_write = shmem_prepare_write,
1940 .commit_write = simple_commit_write,
1944 static struct file_operations shmem_file_operations = {
1947 .llseek = generic_file_llseek,
1948 .read = shmem_file_read,
1949 .write = shmem_file_write,
1950 .fsync = simple_sync_file,
1951 .sendfile = shmem_file_sendfile,
1955 static struct inode_operations shmem_inode_operations = {
1956 .truncate = shmem_truncate,
1957 .setattr = shmem_notify_change,
1960 static struct inode_operations shmem_dir_inode_operations = {
1962 .create = shmem_create,
1963 .lookup = simple_lookup,
1965 .unlink = shmem_unlink,
1966 .symlink = shmem_symlink,
1967 .mkdir = shmem_mkdir,
1968 .rmdir = shmem_rmdir,
1969 .mknod = shmem_mknod,
1970 .rename = shmem_rename,
1974 static struct super_operations shmem_ops = {
1975 .alloc_inode = shmem_alloc_inode,
1976 .destroy_inode = shmem_destroy_inode,
1978 .statfs = shmem_statfs,
1979 .remount_fs = shmem_remount_fs,
1981 .delete_inode = shmem_delete_inode,
1982 .drop_inode = generic_delete_inode,
1983 .put_super = shmem_put_super,
1986 static struct vm_operations_struct shmem_vm_ops = {
1987 .nopage = shmem_nopage,
1988 .populate = shmem_populate,
1990 .set_policy = shmem_set_policy,
1991 .get_policy = shmem_get_policy,
1995 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
1996 int flags, const char *dev_name, void *data)
1998 return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2001 static struct file_system_type tmpfs_fs_type = {
2002 .owner = THIS_MODULE,
2004 .get_sb = shmem_get_sb,
2005 .kill_sb = kill_litter_super,
2007 static struct vfsmount *shm_mnt;
2009 static int __init init_tmpfs(void)
2013 error = init_inodecache();
2017 error = register_filesystem(&tmpfs_fs_type);
2019 printk(KERN_ERR "Could not register tmpfs\n");
2023 devfs_mk_dir("shm");
2025 shm_mnt = kern_mount(&tmpfs_fs_type);
2026 if (IS_ERR(shm_mnt)) {
2027 error = PTR_ERR(shm_mnt);
2028 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2032 /* The internal instance should not do size checking */
2033 shmem_set_size(SHMEM_SB(shm_mnt->mnt_sb), ULONG_MAX, ULONG_MAX);
2037 unregister_filesystem(&tmpfs_fs_type);
2039 destroy_inodecache();
2041 shm_mnt = ERR_PTR(error);
2044 module_init(init_tmpfs)
2047 * shmem_file_setup - get an unlinked file living in tmpfs
2049 * @name: name for dentry (to be seen in /proc/<pid>/maps
2050 * @size: size to be set for the file
2053 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2057 struct inode *inode;
2058 struct dentry *dentry, *root;
2061 if (IS_ERR(shm_mnt))
2062 return (void *)shm_mnt;
2064 if (size > SHMEM_MAX_BYTES)
2065 return ERR_PTR(-EINVAL);
2067 if (shmem_acct_size(flags, size))
2068 return ERR_PTR(-ENOMEM);
2072 this.len = strlen(name);
2073 this.hash = 0; /* will go */
2074 root = shm_mnt->mnt_root;
2075 dentry = d_alloc(root, &this);
2080 file = get_empty_filp();
2085 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2089 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2090 d_instantiate(dentry, inode);
2091 inode->i_size = size;
2092 inode->i_nlink = 0; /* It is unlinked */
2093 file->f_vfsmnt = mntget(shm_mnt);
2094 file->f_dentry = dentry;
2095 file->f_mapping = inode->i_mapping;
2096 file->f_op = &shmem_file_operations;
2097 file->f_mode = FMODE_WRITE | FMODE_READ;
2105 shmem_unacct_size(flags, size);
2106 return ERR_PTR(error);
2110 * shmem_zero_setup - setup a shared anonymous mapping
2112 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2114 int shmem_zero_setup(struct vm_area_struct *vma)
2117 loff_t size = vma->vm_end - vma->vm_start;
2119 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2121 return PTR_ERR(file);
2125 vma->vm_file = file;
2126 vma->vm_ops = &shmem_vm_ops;
2130 EXPORT_SYMBOL(shmem_file_setup);