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 */
50 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
51 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
52 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
54 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
55 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
57 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
59 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
60 #define SHMEM_PAGEIN VM_READ
61 #define SHMEM_TRUNCATE VM_WRITE
63 /* Pretend that each entry is of this size in directory's i_size */
64 #define BOGO_DIRENT_SIZE 20
66 /* Keep swapped page count in private field of indirect struct page */
67 #define nr_swapped private
69 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
71 SGP_QUICK, /* don't try more than file page cache lookup */
72 SGP_READ, /* don't exceed i_size, don't allocate page */
73 SGP_CACHE, /* don't exceed i_size, may allocate page */
74 SGP_WRITE, /* may exceed i_size, may allocate page */
77 static int shmem_getpage(struct inode *inode, unsigned long idx,
78 struct page **pagep, enum sgp_type sgp, int *type);
80 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
83 * The above definition of ENTRIES_PER_PAGE, and the use of
84 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
85 * might be reconsidered if it ever diverges from PAGE_SIZE.
87 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
90 static inline void shmem_dir_free(struct page *page)
92 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
95 static struct page **shmem_dir_map(struct page *page)
97 return (struct page **)kmap_atomic(page, KM_USER0);
100 static inline void shmem_dir_unmap(struct page **dir)
102 kunmap_atomic(dir, KM_USER0);
105 static swp_entry_t *shmem_swp_map(struct page *page)
107 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
110 static inline void shmem_swp_balance_unmap(void)
113 * When passing a pointer to an i_direct entry, to code which
114 * also handles indirect entries and so will shmem_swp_unmap,
115 * we must arrange for the preempt count to remain in balance.
116 * What kmap_atomic of a lowmem page does depends on config
117 * and architecture, so pretend to kmap_atomic some lowmem page.
119 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
122 static inline void shmem_swp_unmap(swp_entry_t *entry)
124 kunmap_atomic(entry, KM_USER1);
127 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
129 return sb->s_fs_info;
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
138 static inline int shmem_acct_size(unsigned long flags, loff_t size)
140 return (flags & VM_ACCOUNT)?
141 security_vm_enough_memory(VM_ACCT(size)): 0;
144 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
146 if (flags & VM_ACCOUNT)
147 vm_unacct_memory(VM_ACCT(size));
151 * ... whereas tmpfs objects are accounted incrementally as
152 * pages are allocated, in order to allow huge sparse files.
153 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
154 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
156 static inline int shmem_acct_block(unsigned long flags)
158 return (flags & VM_ACCOUNT)?
159 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
162 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
164 if (!(flags & VM_ACCOUNT))
165 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
168 static struct super_operations shmem_ops;
169 static struct address_space_operations shmem_aops;
170 static struct file_operations shmem_file_operations;
171 static struct inode_operations shmem_inode_operations;
172 static struct inode_operations shmem_dir_inode_operations;
173 static struct vm_operations_struct shmem_vm_ops;
175 static struct backing_dev_info shmem_backing_dev_info = {
176 .ra_pages = 0, /* No readahead */
177 .memory_backed = 1, /* Does not contribute to dirty memory */
178 .unplug_io_fn = default_unplug_io_fn,
181 LIST_HEAD(shmem_inodes);
182 static spinlock_t shmem_ilock = SPIN_LOCK_UNLOCKED;
184 static void shmem_free_block(struct inode *inode)
186 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
187 spin_lock(&sbinfo->stat_lock);
188 sbinfo->free_blocks++;
189 inode->i_blocks -= BLOCKS_PER_PAGE;
190 spin_unlock(&sbinfo->stat_lock);
194 * shmem_recalc_inode - recalculate the size of an inode
196 * @inode: inode to recalc
198 * We have to calculate the free blocks since the mm can drop
199 * undirtied hole pages behind our back.
201 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
202 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
204 * It has to be called with the spinlock held.
206 static void shmem_recalc_inode(struct inode *inode)
208 struct shmem_inode_info *info = SHMEM_I(inode);
211 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
214 info->alloced -= freed;
215 spin_lock(&sbinfo->stat_lock);
216 sbinfo->free_blocks += freed;
217 inode->i_blocks -= freed*BLOCKS_PER_PAGE;
218 spin_unlock(&sbinfo->stat_lock);
219 shmem_unacct_blocks(info->flags, freed);
224 * shmem_swp_entry - find the swap vector position in the info structure
226 * @info: info structure for the inode
227 * @index: index of the page to find
228 * @page: optional page to add to the structure. Has to be preset to
231 * If there is no space allocated yet it will return NULL when
232 * page is NULL, else it will use the page for the needed block,
233 * setting it to NULL on return to indicate that it has been used.
235 * The swap vector is organized the following way:
237 * There are SHMEM_NR_DIRECT entries directly stored in the
238 * shmem_inode_info structure. So small files do not need an addional
241 * For pages with index > SHMEM_NR_DIRECT there is the pointer
242 * i_indirect which points to a page which holds in the first half
243 * doubly indirect blocks, in the second half triple indirect blocks:
245 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
246 * following layout (for SHMEM_NR_DIRECT == 16):
248 * i_indirect -> dir --> 16-19
261 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
263 unsigned long offset;
267 if (index < SHMEM_NR_DIRECT) {
268 shmem_swp_balance_unmap();
269 return info->i_direct+index;
271 if (!info->i_indirect) {
273 info->i_indirect = *page;
276 return NULL; /* need another page */
279 index -= SHMEM_NR_DIRECT;
280 offset = index % ENTRIES_PER_PAGE;
281 index /= ENTRIES_PER_PAGE;
282 dir = shmem_dir_map(info->i_indirect);
284 if (index >= ENTRIES_PER_PAGE/2) {
285 index -= ENTRIES_PER_PAGE/2;
286 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
287 index %= ENTRIES_PER_PAGE;
294 shmem_dir_unmap(dir);
295 return NULL; /* need another page */
297 shmem_dir_unmap(dir);
298 dir = shmem_dir_map(subdir);
304 if (!page || !(subdir = *page)) {
305 shmem_dir_unmap(dir);
306 return NULL; /* need a page */
311 shmem_dir_unmap(dir);
312 return shmem_swp_map(subdir) + offset;
315 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
317 long incdec = value? 1: -1;
320 info->swapped += incdec;
321 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
322 kmap_atomic_to_page(entry)->nr_swapped += incdec;
326 * shmem_swp_alloc - get the position of the swap entry for the page.
327 * If it does not exist allocate the entry.
329 * @info: info structure for the inode
330 * @index: index of the page to find
331 * @sgp: check and recheck i_size? skip allocation?
333 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
335 struct inode *inode = &info->vfs_inode;
336 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
337 struct page *page = NULL;
340 if (sgp != SGP_WRITE &&
341 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
342 return ERR_PTR(-EINVAL);
344 while (!(entry = shmem_swp_entry(info, index, &page))) {
346 return shmem_swp_map(ZERO_PAGE(0));
348 * Test free_blocks against 1 not 0, since we have 1 data
349 * page (and perhaps indirect index pages) yet to allocate:
350 * a waste to allocate index if we cannot allocate data.
352 spin_lock(&sbinfo->stat_lock);
353 if (sbinfo->free_blocks <= 1) {
354 spin_unlock(&sbinfo->stat_lock);
355 return ERR_PTR(-ENOSPC);
357 sbinfo->free_blocks--;
358 inode->i_blocks += BLOCKS_PER_PAGE;
359 spin_unlock(&sbinfo->stat_lock);
361 spin_unlock(&info->lock);
362 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
364 clear_highpage(page);
365 page->nr_swapped = 0;
367 spin_lock(&info->lock);
370 shmem_free_block(inode);
371 return ERR_PTR(-ENOMEM);
373 if (sgp != SGP_WRITE &&
374 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
375 entry = ERR_PTR(-EINVAL);
378 if (info->next_index <= index)
379 info->next_index = index + 1;
382 /* another task gave its page, or truncated the file */
383 shmem_free_block(inode);
384 shmem_dir_free(page);
386 if (info->next_index <= index && !IS_ERR(entry))
387 info->next_index = index + 1;
392 * shmem_free_swp - free some swap entries in a directory
394 * @dir: pointer to the directory
395 * @edir: pointer after last entry of the directory
397 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
402 for (ptr = dir; ptr < edir; ptr++) {
404 free_swap_and_cache(*ptr);
405 *ptr = (swp_entry_t){0};
412 static void shmem_truncate(struct inode *inode)
414 struct shmem_inode_info *info = SHMEM_I(inode);
426 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
427 idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
428 if (idx >= info->next_index)
431 spin_lock(&info->lock);
432 info->flags |= SHMEM_TRUNCATE;
433 limit = info->next_index;
434 info->next_index = idx;
435 if (info->swapped && idx < SHMEM_NR_DIRECT) {
436 ptr = info->i_direct;
438 if (size > SHMEM_NR_DIRECT)
439 size = SHMEM_NR_DIRECT;
440 info->swapped -= shmem_free_swp(ptr+idx, ptr+size);
442 if (!info->i_indirect)
445 BUG_ON(limit <= SHMEM_NR_DIRECT);
446 limit -= SHMEM_NR_DIRECT;
447 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
448 offset = idx % ENTRIES_PER_PAGE;
452 dir = shmem_dir_map(info->i_indirect);
453 stage = ENTRIES_PER_PAGEPAGE/2;
454 if (idx < ENTRIES_PER_PAGEPAGE/2)
455 dir += idx/ENTRIES_PER_PAGE;
457 dir += ENTRIES_PER_PAGE/2;
458 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
460 stage += ENTRIES_PER_PAGEPAGE;
463 size = ((idx - ENTRIES_PER_PAGEPAGE/2) %
464 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
465 if (!size && !offset) {
469 shmem_dir_unmap(dir);
470 dir = shmem_dir_map(subdir) + size;
477 for (; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
478 if (unlikely(idx == stage)) {
479 shmem_dir_unmap(dir-1);
480 dir = shmem_dir_map(info->i_indirect) +
481 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
484 idx += ENTRIES_PER_PAGEPAGE;
488 stage = idx + ENTRIES_PER_PAGEPAGE;
491 shmem_dir_unmap(dir);
493 shmem_dir_free(empty);
494 shmem_free_block(inode);
497 cond_resched_lock(&info->lock);
498 dir = shmem_dir_map(subdir);
501 if (subdir && subdir->nr_swapped) {
502 ptr = shmem_swp_map(subdir);
504 if (size > ENTRIES_PER_PAGE)
505 size = ENTRIES_PER_PAGE;
506 freed = shmem_free_swp(ptr+offset, ptr+size);
507 shmem_swp_unmap(ptr);
508 info->swapped -= freed;
509 subdir->nr_swapped -= freed;
510 BUG_ON(subdir->nr_swapped > offset);
516 shmem_dir_free(subdir);
517 shmem_free_block(inode);
521 shmem_dir_unmap(dir-1);
523 shmem_dir_free(empty);
524 shmem_free_block(inode);
526 if (info->next_index <= SHMEM_NR_DIRECT) {
527 shmem_dir_free(info->i_indirect);
528 info->i_indirect = NULL;
529 shmem_free_block(inode);
532 BUG_ON(info->swapped > info->next_index);
533 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
535 * Call truncate_inode_pages again: racing shmem_unuse_inode
536 * may have swizzled a page in from swap since vmtruncate or
537 * generic_delete_inode did it, before we lowered next_index.
538 * Also, though shmem_getpage checks i_size before adding to
539 * cache, no recheck after: so fix the narrow window there too.
541 spin_unlock(&info->lock);
542 truncate_inode_pages(inode->i_mapping, inode->i_size);
543 spin_lock(&info->lock);
545 info->flags &= ~SHMEM_TRUNCATE;
546 shmem_recalc_inode(inode);
547 spin_unlock(&info->lock);
550 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
552 struct inode *inode = dentry->d_inode;
553 struct page *page = NULL;
556 if (attr->ia_valid & ATTR_SIZE) {
557 if (attr->ia_size < inode->i_size) {
559 * If truncating down to a partial page, then
560 * if that page is already allocated, hold it
561 * in memory until the truncation is over, so
562 * truncate_partial_page cannnot miss it were
563 * it assigned to swap.
565 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
566 (void) shmem_getpage(inode,
567 attr->ia_size>>PAGE_CACHE_SHIFT,
568 &page, SGP_READ, NULL);
571 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
572 * detect if any pages might have been added to cache
573 * after truncate_inode_pages. But we needn't bother
574 * if it's being fully truncated to zero-length: the
575 * nrpages check is efficient enough in that case.
578 struct shmem_inode_info *info = SHMEM_I(inode);
579 spin_lock(&info->lock);
580 info->flags &= ~SHMEM_PAGEIN;
581 spin_unlock(&info->lock);
586 error = inode_change_ok(inode, attr);
588 error = inode_setattr(inode, attr);
590 page_cache_release(page);
594 static void shmem_delete_inode(struct inode *inode)
596 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
597 struct shmem_inode_info *info = SHMEM_I(inode);
599 if (inode->i_op->truncate == shmem_truncate) {
600 spin_lock(&shmem_ilock);
601 list_del(&info->list);
602 spin_unlock(&shmem_ilock);
603 shmem_unacct_size(info->flags, inode->i_size);
605 shmem_truncate(inode);
607 BUG_ON(inode->i_blocks);
608 spin_lock(&sbinfo->stat_lock);
609 sbinfo->free_inodes++;
610 spin_unlock(&sbinfo->stat_lock);
614 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
618 for (ptr = dir; ptr < edir; ptr++) {
619 if (ptr->val == entry.val)
625 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
638 ptr = info->i_direct;
639 spin_lock(&info->lock);
640 limit = info->next_index;
642 if (size > SHMEM_NR_DIRECT)
643 size = SHMEM_NR_DIRECT;
644 offset = shmem_find_swp(entry, ptr, ptr+size);
646 shmem_swp_balance_unmap();
649 if (!info->i_indirect)
651 /* we might be racing with shmem_truncate */
652 if (limit <= SHMEM_NR_DIRECT)
655 dir = shmem_dir_map(info->i_indirect);
656 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
658 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
659 if (unlikely(idx == stage)) {
660 shmem_dir_unmap(dir-1);
661 dir = shmem_dir_map(info->i_indirect) +
662 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
665 idx += ENTRIES_PER_PAGEPAGE;
669 stage = idx + ENTRIES_PER_PAGEPAGE;
671 shmem_dir_unmap(dir);
672 dir = shmem_dir_map(subdir);
675 if (subdir && subdir->nr_swapped) {
676 ptr = shmem_swp_map(subdir);
678 if (size > ENTRIES_PER_PAGE)
679 size = ENTRIES_PER_PAGE;
680 offset = shmem_find_swp(entry, ptr, ptr+size);
682 shmem_dir_unmap(dir);
685 shmem_swp_unmap(ptr);
689 shmem_dir_unmap(dir-1);
691 spin_unlock(&info->lock);
695 inode = &info->vfs_inode;
696 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
697 info->flags |= SHMEM_PAGEIN;
698 shmem_swp_set(info, ptr + offset, 0);
700 shmem_swp_unmap(ptr);
701 spin_unlock(&info->lock);
703 * Decrement swap count even when the entry is left behind:
704 * try_to_unuse will skip over mms, then reincrement count.
711 * shmem_unuse() search for an eventually swapped out shmem page.
713 int shmem_unuse(swp_entry_t entry, struct page *page)
716 struct shmem_inode_info *info;
719 spin_lock(&shmem_ilock);
720 list_for_each(p, &shmem_inodes) {
721 info = list_entry(p, struct shmem_inode_info, list);
723 if (info->swapped && shmem_unuse_inode(info, entry, page)) {
724 /* move head to start search for next from here */
725 list_move_tail(&shmem_inodes, &info->list);
730 spin_unlock(&shmem_ilock);
735 * Move the page from the page cache to the swap cache.
737 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
739 struct shmem_inode_info *info;
740 swp_entry_t *entry, swap;
741 struct address_space *mapping;
745 BUG_ON(!PageLocked(page));
746 BUG_ON(page_mapped(page));
748 mapping = page->mapping;
750 inode = mapping->host;
751 info = SHMEM_I(inode);
752 if (info->flags & VM_LOCKED)
754 swap = get_swap_page();
758 spin_lock(&info->lock);
759 shmem_recalc_inode(inode);
760 if (index >= info->next_index) {
761 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
764 entry = shmem_swp_entry(info, index, NULL);
768 if (move_to_swap_cache(page, swap) == 0) {
769 shmem_swp_set(info, entry, swap.val);
770 shmem_swp_unmap(entry);
771 spin_unlock(&info->lock);
776 shmem_swp_unmap(entry);
778 spin_unlock(&info->lock);
781 set_page_dirty(page);
782 return WRITEPAGE_ACTIVATE; /* Return with the page locked */
786 static struct page *shmem_swapin_async(struct shared_policy *p,
787 swp_entry_t entry, unsigned long idx)
790 struct vm_area_struct pvma;
792 /* Create a pseudo vma that just contains the policy */
793 memset(&pvma, 0, sizeof(struct vm_area_struct));
794 pvma.vm_end = PAGE_SIZE;
796 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
797 page = read_swap_cache_async(entry, &pvma, 0);
798 mpol_free(pvma.vm_policy);
802 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
805 struct shared_policy *p = &info->policy;
808 unsigned long offset;
810 num = valid_swaphandles(entry, &offset);
811 for (i = 0; i < num; offset++, i++) {
812 page = shmem_swapin_async(p,
813 swp_entry(swp_type(entry), offset), idx);
816 page_cache_release(page);
818 lru_add_drain(); /* Push any new pages onto the LRU now */
819 return shmem_swapin_async(p, entry, idx);
823 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
826 struct vm_area_struct pvma;
829 memset(&pvma, 0, sizeof(struct vm_area_struct));
830 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
832 pvma.vm_end = PAGE_SIZE;
833 page = alloc_page_vma(gfp, &pvma, 0);
834 mpol_free(pvma.vm_policy);
838 static inline struct page *
839 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
841 swapin_readahead(entry, 0, NULL);
842 return read_swap_cache_async(entry, NULL, 0);
845 static inline struct page *
846 shmem_alloc_page(unsigned long gfp,struct shmem_inode_info *info,
849 return alloc_page(gfp);
854 * shmem_getpage - either get the page from swap or allocate a new one
856 * If we allocate a new one we do not mark it dirty. That's up to the
857 * vm. If we swap it in we mark it dirty since we also free the swap
858 * entry since a page cannot live in both the swap and page cache
860 static int shmem_getpage(struct inode *inode, unsigned long idx,
861 struct page **pagep, enum sgp_type sgp, int *type)
863 struct address_space *mapping = inode->i_mapping;
864 struct shmem_inode_info *info = SHMEM_I(inode);
865 struct shmem_sb_info *sbinfo;
866 struct page *filepage = *pagep;
867 struct page *swappage;
870 int error, majmin = VM_FAULT_MINOR;
872 if (idx >= SHMEM_MAX_INDEX)
875 * Normally, filepage is NULL on entry, and either found
876 * uptodate immediately, or allocated and zeroed, or read
877 * in under swappage, which is then assigned to filepage.
878 * But shmem_prepare_write passes in a locked filepage,
879 * which may be found not uptodate by other callers too,
880 * and may need to be copied from the swappage read in.
884 filepage = find_lock_page(mapping, idx);
885 if (filepage && PageUptodate(filepage))
888 if (sgp == SGP_QUICK)
891 spin_lock(&info->lock);
892 shmem_recalc_inode(inode);
893 entry = shmem_swp_alloc(info, idx, sgp);
895 spin_unlock(&info->lock);
896 error = PTR_ERR(entry);
902 /* Look it up and read it in.. */
903 swappage = lookup_swap_cache(swap);
905 shmem_swp_unmap(entry);
906 spin_unlock(&info->lock);
907 /* here we actually do the io */
908 if (majmin == VM_FAULT_MINOR && type)
909 inc_page_state(pgmajfault);
910 majmin = VM_FAULT_MAJOR;
911 swappage = shmem_swapin(info, swap, idx);
913 spin_lock(&info->lock);
914 entry = shmem_swp_alloc(info, idx, sgp);
916 error = PTR_ERR(entry);
918 if (entry->val == swap.val)
920 shmem_swp_unmap(entry);
922 spin_unlock(&info->lock);
927 wait_on_page_locked(swappage);
928 page_cache_release(swappage);
932 /* We have to do this with page locked to prevent races */
933 if (TestSetPageLocked(swappage)) {
934 shmem_swp_unmap(entry);
935 spin_unlock(&info->lock);
936 wait_on_page_locked(swappage);
937 page_cache_release(swappage);
940 if (PageWriteback(swappage)) {
941 shmem_swp_unmap(entry);
942 spin_unlock(&info->lock);
943 wait_on_page_writeback(swappage);
944 unlock_page(swappage);
945 page_cache_release(swappage);
948 if (!PageUptodate(swappage)) {
949 shmem_swp_unmap(entry);
950 spin_unlock(&info->lock);
951 unlock_page(swappage);
952 page_cache_release(swappage);
958 shmem_swp_set(info, entry, 0);
959 shmem_swp_unmap(entry);
960 delete_from_swap_cache(swappage);
961 spin_unlock(&info->lock);
962 copy_highpage(filepage, swappage);
963 unlock_page(swappage);
964 page_cache_release(swappage);
965 flush_dcache_page(filepage);
966 SetPageUptodate(filepage);
967 set_page_dirty(filepage);
969 } else if (!(error = move_from_swap_cache(
970 swappage, idx, mapping))) {
971 info->flags |= SHMEM_PAGEIN;
972 shmem_swp_set(info, entry, 0);
973 shmem_swp_unmap(entry);
974 spin_unlock(&info->lock);
978 shmem_swp_unmap(entry);
979 spin_unlock(&info->lock);
980 unlock_page(swappage);
981 page_cache_release(swappage);
982 if (error == -ENOMEM) {
983 /* let kswapd refresh zone for GFP_ATOMICs */
984 blk_congestion_wait(WRITE, HZ/50);
988 } else if (sgp == SGP_READ && !filepage) {
989 shmem_swp_unmap(entry);
990 filepage = find_get_page(mapping, idx);
992 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
993 spin_unlock(&info->lock);
994 wait_on_page_locked(filepage);
995 page_cache_release(filepage);
999 spin_unlock(&info->lock);
1001 shmem_swp_unmap(entry);
1002 sbinfo = SHMEM_SB(inode->i_sb);
1003 spin_lock(&sbinfo->stat_lock);
1004 if (sbinfo->free_blocks == 0 || shmem_acct_block(info->flags)) {
1005 spin_unlock(&sbinfo->stat_lock);
1006 spin_unlock(&info->lock);
1010 sbinfo->free_blocks--;
1011 inode->i_blocks += BLOCKS_PER_PAGE;
1012 spin_unlock(&sbinfo->stat_lock);
1015 spin_unlock(&info->lock);
1016 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1020 shmem_unacct_blocks(info->flags, 1);
1021 shmem_free_block(inode);
1026 spin_lock(&info->lock);
1027 entry = shmem_swp_alloc(info, idx, sgp);
1029 error = PTR_ERR(entry);
1032 shmem_swp_unmap(entry);
1034 if (error || swap.val || 0 != add_to_page_cache_lru(
1035 filepage, mapping, idx, GFP_ATOMIC)) {
1036 spin_unlock(&info->lock);
1037 page_cache_release(filepage);
1038 shmem_unacct_blocks(info->flags, 1);
1039 shmem_free_block(inode);
1045 info->flags |= SHMEM_PAGEIN;
1049 spin_unlock(&info->lock);
1050 clear_highpage(filepage);
1051 flush_dcache_page(filepage);
1052 SetPageUptodate(filepage);
1057 unlock_page(filepage);
1060 *pagep = ZERO_PAGE(0);
1067 if (*pagep != filepage) {
1068 unlock_page(filepage);
1069 page_cache_release(filepage);
1074 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1076 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1077 struct page *page = NULL;
1081 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1082 idx += vma->vm_pgoff;
1083 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1085 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1087 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1089 mark_page_accessed(page);
1093 static int shmem_populate(struct vm_area_struct *vma,
1094 unsigned long addr, unsigned long len,
1095 pgprot_t prot, unsigned long pgoff, int nonblock)
1097 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1098 struct mm_struct *mm = vma->vm_mm;
1099 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1102 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1103 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1106 while ((long) len > 0) {
1107 struct page *page = NULL;
1110 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1112 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1116 mark_page_accessed(page);
1117 err = install_page(mm, vma, addr, page, prot);
1119 page_cache_release(page);
1122 } else if (nonblock) {
1123 err = install_file_pte(mm, vma, addr, pgoff, prot);
1136 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1138 struct inode *i = vma->vm_file->f_dentry->d_inode;
1139 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1143 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1145 struct inode *i = vma->vm_file->f_dentry->d_inode;
1148 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1149 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1153 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1155 struct inode *inode = file->f_dentry->d_inode;
1156 struct shmem_inode_info *info = SHMEM_I(inode);
1157 int retval = -ENOMEM;
1159 spin_lock(&info->lock);
1160 if (lock && !(info->flags & VM_LOCKED)) {
1161 if (!user_shm_lock(inode->i_size, user))
1163 info->flags |= VM_LOCKED;
1165 if (!lock && (info->flags & VM_LOCKED) && user) {
1166 user_shm_unlock(inode->i_size, user);
1167 info->flags &= ~VM_LOCKED;
1171 spin_unlock(&info->lock);
1175 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1177 file_accessed(file);
1178 vma->vm_ops = &shmem_vm_ops;
1182 static struct inode *
1183 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1185 struct inode *inode;
1186 struct shmem_inode_info *info;
1187 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1189 spin_lock(&sbinfo->stat_lock);
1190 if (!sbinfo->free_inodes) {
1191 spin_unlock(&sbinfo->stat_lock);
1194 sbinfo->free_inodes--;
1195 spin_unlock(&sbinfo->stat_lock);
1197 inode = new_inode(sb);
1199 inode->i_mode = mode;
1200 inode->i_uid = current->fsuid;
1201 inode->i_gid = current->fsgid;
1202 inode->i_blksize = PAGE_CACHE_SIZE;
1203 inode->i_blocks = 0;
1204 inode->i_mapping->a_ops = &shmem_aops;
1205 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1206 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1207 info = SHMEM_I(inode);
1208 memset(info, 0, (char *)inode - (char *)info);
1209 spin_lock_init(&info->lock);
1210 mpol_shared_policy_init(&info->policy);
1211 switch (mode & S_IFMT) {
1213 init_special_inode(inode, mode, dev);
1216 inode->i_op = &shmem_inode_operations;
1217 inode->i_fop = &shmem_file_operations;
1218 spin_lock(&shmem_ilock);
1219 list_add_tail(&info->list, &shmem_inodes);
1220 spin_unlock(&shmem_ilock);
1224 /* Some things misbehave if size == 0 on a directory */
1225 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1226 inode->i_op = &shmem_dir_inode_operations;
1227 inode->i_fop = &simple_dir_operations;
1236 static int shmem_set_size(struct shmem_sb_info *info,
1237 unsigned long max_blocks, unsigned long max_inodes)
1240 unsigned long blocks, inodes;
1242 spin_lock(&info->stat_lock);
1243 blocks = info->max_blocks - info->free_blocks;
1244 inodes = info->max_inodes - info->free_inodes;
1246 if (max_blocks < blocks)
1248 if (max_inodes < inodes)
1251 info->max_blocks = max_blocks;
1252 info->free_blocks = max_blocks - blocks;
1253 info->max_inodes = max_inodes;
1254 info->free_inodes = max_inodes - inodes;
1256 spin_unlock(&info->stat_lock);
1262 static struct inode_operations shmem_symlink_inode_operations;
1263 static struct inode_operations shmem_symlink_inline_operations;
1266 * Normally tmpfs makes no use of shmem_prepare_write, but it
1267 * lets a tmpfs file be used read-write below the loop driver.
1270 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1272 struct inode *inode = page->mapping->host;
1273 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1277 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1279 struct inode *inode = file->f_dentry->d_inode;
1281 unsigned long written;
1284 if ((ssize_t) count < 0)
1287 if (!access_ok(VERIFY_READ, buf, count))
1290 down(&inode->i_sem);
1295 err = generic_write_checks(file, &pos, &count, 0);
1299 err = remove_suid(file->f_dentry);
1303 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1306 struct page *page = NULL;
1307 unsigned long bytes, index, offset;
1311 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1312 index = pos >> PAGE_CACHE_SHIFT;
1313 bytes = PAGE_CACHE_SIZE - offset;
1318 * We don't hold page lock across copy from user -
1319 * what would it guard against? - so no deadlock here.
1320 * But it still may be a good idea to prefault below.
1323 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1328 if (PageHighMem(page)) {
1329 volatile unsigned char dummy;
1330 __get_user(dummy, buf);
1331 __get_user(dummy, buf + bytes - 1);
1333 kaddr = kmap_atomic(page, KM_USER0);
1334 left = __copy_from_user(kaddr + offset, buf, bytes);
1335 kunmap_atomic(kaddr, KM_USER0);
1339 left = __copy_from_user(kaddr + offset, buf, bytes);
1347 if (pos > inode->i_size)
1348 i_size_write(inode, pos);
1350 flush_dcache_page(page);
1351 set_page_dirty(page);
1352 mark_page_accessed(page);
1353 page_cache_release(page);
1363 * Our dirty pages are not counted in nr_dirty,
1364 * and we do not attempt to balance dirty pages.
1378 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1380 struct inode *inode = filp->f_dentry->d_inode;
1381 struct address_space *mapping = inode->i_mapping;
1382 unsigned long index, offset;
1384 index = *ppos >> PAGE_CACHE_SHIFT;
1385 offset = *ppos & ~PAGE_CACHE_MASK;
1388 struct page *page = NULL;
1389 unsigned long end_index, nr, ret;
1390 loff_t i_size = i_size_read(inode);
1392 end_index = i_size >> PAGE_CACHE_SHIFT;
1393 if (index > end_index)
1395 if (index == end_index) {
1396 nr = i_size & ~PAGE_CACHE_MASK;
1401 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1403 if (desc->error == -EINVAL)
1409 * We must evaluate after, since reads (unlike writes)
1410 * are called without i_sem protection against truncate
1412 nr = PAGE_CACHE_SIZE;
1413 i_size = i_size_read(inode);
1414 end_index = i_size >> PAGE_CACHE_SHIFT;
1415 if (index == end_index) {
1416 nr = i_size & ~PAGE_CACHE_MASK;
1418 page_cache_release(page);
1424 if (page != ZERO_PAGE(0)) {
1426 * If users can be writing to this page using arbitrary
1427 * virtual addresses, take care about potential aliasing
1428 * before reading the page on the kernel side.
1430 if (mapping_writably_mapped(mapping))
1431 flush_dcache_page(page);
1433 * Mark the page accessed if we read the beginning.
1436 mark_page_accessed(page);
1440 * Ok, we have the page, and it's up-to-date, so
1441 * now we can copy it to user space...
1443 * The actor routine returns how many bytes were actually used..
1444 * NOTE! This may not be the same as how much of a user buffer
1445 * we filled up (we may be padding etc), so we can only update
1446 * "pos" here (the actor routine has to update the user buffer
1447 * pointers and the remaining count).
1449 ret = actor(desc, page, offset, nr);
1451 index += offset >> PAGE_CACHE_SHIFT;
1452 offset &= ~PAGE_CACHE_MASK;
1454 page_cache_release(page);
1455 if (ret != nr || !desc->count)
1461 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1462 file_accessed(filp);
1465 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1467 read_descriptor_t desc;
1469 if ((ssize_t) count < 0)
1471 if (!access_ok(VERIFY_WRITE, buf, count))
1481 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1483 return desc.written;
1487 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1488 size_t count, read_actor_t actor, void *target)
1490 read_descriptor_t desc;
1497 desc.arg.data = target;
1500 do_shmem_file_read(in_file, ppos, &desc, actor);
1502 return desc.written;
1506 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1508 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1510 buf->f_type = TMPFS_SUPER_MAGIC;
1511 buf->f_bsize = PAGE_CACHE_SIZE;
1512 spin_lock(&sbinfo->stat_lock);
1513 buf->f_blocks = sbinfo->max_blocks;
1514 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1515 buf->f_files = sbinfo->max_inodes;
1516 buf->f_ffree = sbinfo->free_inodes;
1517 spin_unlock(&sbinfo->stat_lock);
1518 buf->f_namelen = NAME_MAX;
1523 * File creation. Allocate an inode, and we're done..
1526 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1528 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1529 int error = -ENOSPC;
1532 if (dir->i_mode & S_ISGID) {
1533 inode->i_gid = dir->i_gid;
1535 inode->i_mode |= S_ISGID;
1537 dir->i_size += BOGO_DIRENT_SIZE;
1538 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1539 d_instantiate(dentry, inode);
1540 dget(dentry); /* Extra count - pin the dentry in core */
1546 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1550 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1556 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1557 struct nameidata *nd)
1559 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1565 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1567 struct inode *inode = old_dentry->d_inode;
1569 dir->i_size += BOGO_DIRENT_SIZE;
1570 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1572 atomic_inc(&inode->i_count); /* New dentry reference */
1573 dget(dentry); /* Extra pinning count for the created dentry */
1574 d_instantiate(dentry, inode);
1578 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1580 struct inode *inode = dentry->d_inode;
1582 dir->i_size -= BOGO_DIRENT_SIZE;
1583 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1585 dput(dentry); /* Undo the count from "create" - this does all the work */
1589 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1591 if (!simple_empty(dentry))
1595 return shmem_unlink(dir, dentry);
1599 * The VFS layer already does all the dentry stuff for rename,
1600 * we just have to decrement the usage count for the target if
1601 * it exists so that the VFS layer correctly free's it when it
1604 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1606 struct inode *inode = old_dentry->d_inode;
1607 int they_are_dirs = S_ISDIR(inode->i_mode);
1609 if (!simple_empty(new_dentry))
1612 if (new_dentry->d_inode) {
1613 (void) shmem_unlink(new_dir, new_dentry);
1616 } else if (they_are_dirs) {
1621 old_dir->i_size -= BOGO_DIRENT_SIZE;
1622 new_dir->i_size += BOGO_DIRENT_SIZE;
1623 old_dir->i_ctime = old_dir->i_mtime =
1624 new_dir->i_ctime = new_dir->i_mtime =
1625 inode->i_ctime = CURRENT_TIME;
1629 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1633 struct inode *inode;
1634 struct page *page = NULL;
1636 struct shmem_inode_info *info;
1638 len = strlen(symname) + 1;
1639 if (len > PAGE_CACHE_SIZE)
1640 return -ENAMETOOLONG;
1642 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1646 info = SHMEM_I(inode);
1647 inode->i_size = len-1;
1648 if (len <= (char *)inode - (char *)info) {
1650 memcpy(info, symname, len);
1651 inode->i_op = &shmem_symlink_inline_operations;
1653 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1658 inode->i_op = &shmem_symlink_inode_operations;
1659 spin_lock(&shmem_ilock);
1660 list_add_tail(&info->list, &shmem_inodes);
1661 spin_unlock(&shmem_ilock);
1662 kaddr = kmap_atomic(page, KM_USER0);
1663 memcpy(kaddr, symname, len);
1664 kunmap_atomic(kaddr, KM_USER0);
1665 set_page_dirty(page);
1666 page_cache_release(page);
1668 if (dir->i_mode & S_ISGID)
1669 inode->i_gid = dir->i_gid;
1670 dir->i_size += BOGO_DIRENT_SIZE;
1671 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1672 d_instantiate(dentry, inode);
1677 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1679 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1683 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1685 struct page *page = NULL;
1686 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1687 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1691 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1693 if (!IS_ERR(nd_get_link(nd))) {
1696 page = find_get_page(dentry->d_inode->i_mapping, 0);
1700 mark_page_accessed(page);
1701 page_cache_release(page);
1702 page_cache_release(page);
1706 static struct inode_operations shmem_symlink_inline_operations = {
1707 .readlink = generic_readlink,
1708 .follow_link = shmem_follow_link_inline,
1711 static struct inode_operations shmem_symlink_inode_operations = {
1712 .truncate = shmem_truncate,
1713 .readlink = generic_readlink,
1714 .follow_link = shmem_follow_link,
1715 .put_link = shmem_put_link,
1718 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1720 char *this_char, *value, *rest;
1722 while ((this_char = strsep(&options, ",")) != NULL) {
1725 if ((value = strchr(this_char,'=')) != NULL) {
1729 "tmpfs: No value for mount option '%s'\n",
1734 if (!strcmp(this_char,"size")) {
1735 unsigned long long size;
1736 size = memparse(value,&rest);
1738 size <<= PAGE_SHIFT;
1739 size *= totalram_pages;
1745 *blocks = size >> PAGE_CACHE_SHIFT;
1746 } else if (!strcmp(this_char,"nr_blocks")) {
1747 *blocks = memparse(value,&rest);
1750 } else if (!strcmp(this_char,"nr_inodes")) {
1751 *inodes = memparse(value,&rest);
1754 } else if (!strcmp(this_char,"mode")) {
1757 *mode = simple_strtoul(value,&rest,8);
1760 } else if (!strcmp(this_char,"uid")) {
1763 *uid = simple_strtoul(value,&rest,0);
1766 } else if (!strcmp(this_char,"gid")) {
1769 *gid = simple_strtoul(value,&rest,0);
1773 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1781 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1787 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1789 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1790 unsigned long max_blocks = sbinfo->max_blocks;
1791 unsigned long max_inodes = sbinfo->max_inodes;
1793 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
1795 return shmem_set_size(sbinfo, max_blocks, max_inodes);
1799 static int shmem_fill_super(struct super_block *sb,
1800 void *data, int silent)
1802 struct inode *inode;
1803 struct dentry *root;
1804 unsigned long blocks, inodes;
1805 int mode = S_IRWXUGO | S_ISVTX;
1806 uid_t uid = current->fsuid;
1807 gid_t gid = current->fsgid;
1808 struct shmem_sb_info *sbinfo;
1811 sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
1814 sb->s_fs_info = sbinfo;
1815 memset(sbinfo, 0, sizeof(struct shmem_sb_info));
1818 * Per default we only allow half of the physical ram per
1821 blocks = inodes = totalram_pages / 2;
1824 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, &inodes)) {
1829 sb->s_flags |= MS_NOUSER;
1832 spin_lock_init(&sbinfo->stat_lock);
1833 sbinfo->max_blocks = blocks;
1834 sbinfo->free_blocks = blocks;
1835 sbinfo->max_inodes = inodes;
1836 sbinfo->free_inodes = inodes;
1837 sb->s_maxbytes = SHMEM_MAX_BYTES;
1838 sb->s_blocksize = PAGE_CACHE_SIZE;
1839 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1840 sb->s_magic = TMPFS_SUPER_MAGIC;
1841 sb->s_op = &shmem_ops;
1842 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1847 root = d_alloc_root(inode);
1857 sb->s_fs_info = NULL;
1861 static void shmem_put_super(struct super_block *sb)
1863 kfree(sb->s_fs_info);
1864 sb->s_fs_info = NULL;
1867 static kmem_cache_t *shmem_inode_cachep;
1869 static struct inode *shmem_alloc_inode(struct super_block *sb)
1871 struct shmem_inode_info *p;
1872 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
1875 return &p->vfs_inode;
1878 static void shmem_destroy_inode(struct inode *inode)
1880 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
1881 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
1884 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
1886 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
1888 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1889 SLAB_CTOR_CONSTRUCTOR) {
1890 inode_init_once(&p->vfs_inode);
1894 static int init_inodecache(void)
1896 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
1897 sizeof(struct shmem_inode_info),
1898 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
1900 if (shmem_inode_cachep == NULL)
1905 static void destroy_inodecache(void)
1907 if (kmem_cache_destroy(shmem_inode_cachep))
1908 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
1911 static struct address_space_operations shmem_aops = {
1912 .writepage = shmem_writepage,
1913 .set_page_dirty = __set_page_dirty_nobuffers,
1915 .prepare_write = shmem_prepare_write,
1916 .commit_write = simple_commit_write,
1920 static struct file_operations shmem_file_operations = {
1923 .llseek = generic_file_llseek,
1924 .read = shmem_file_read,
1925 .write = shmem_file_write,
1926 .fsync = simple_sync_file,
1927 .sendfile = shmem_file_sendfile,
1931 static struct inode_operations shmem_inode_operations = {
1932 .truncate = shmem_truncate,
1933 .setattr = shmem_notify_change,
1936 static struct inode_operations shmem_dir_inode_operations = {
1938 .create = shmem_create,
1939 .lookup = simple_lookup,
1941 .unlink = shmem_unlink,
1942 .symlink = shmem_symlink,
1943 .mkdir = shmem_mkdir,
1944 .rmdir = shmem_rmdir,
1945 .mknod = shmem_mknod,
1946 .rename = shmem_rename,
1950 static struct super_operations shmem_ops = {
1951 .alloc_inode = shmem_alloc_inode,
1952 .destroy_inode = shmem_destroy_inode,
1954 .statfs = shmem_statfs,
1955 .remount_fs = shmem_remount_fs,
1957 .delete_inode = shmem_delete_inode,
1958 .drop_inode = generic_delete_inode,
1959 .put_super = shmem_put_super,
1962 static struct vm_operations_struct shmem_vm_ops = {
1963 .nopage = shmem_nopage,
1964 .populate = shmem_populate,
1966 .set_policy = shmem_set_policy,
1967 .get_policy = shmem_get_policy,
1971 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
1972 int flags, const char *dev_name, void *data)
1974 return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
1977 static struct file_system_type tmpfs_fs_type = {
1978 .owner = THIS_MODULE,
1980 .get_sb = shmem_get_sb,
1981 .kill_sb = kill_litter_super,
1983 static struct vfsmount *shm_mnt;
1985 static int __init init_tmpfs(void)
1989 error = init_inodecache();
1993 error = register_filesystem(&tmpfs_fs_type);
1995 printk(KERN_ERR "Could not register tmpfs\n");
1999 devfs_mk_dir("shm");
2001 shm_mnt = kern_mount(&tmpfs_fs_type);
2002 if (IS_ERR(shm_mnt)) {
2003 error = PTR_ERR(shm_mnt);
2004 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2008 /* The internal instance should not do size checking */
2009 shmem_set_size(SHMEM_SB(shm_mnt->mnt_sb), ULONG_MAX, ULONG_MAX);
2013 unregister_filesystem(&tmpfs_fs_type);
2015 destroy_inodecache();
2017 shm_mnt = ERR_PTR(error);
2020 module_init(init_tmpfs)
2023 * shmem_file_setup - get an unlinked file living in tmpfs
2025 * @name: name for dentry (to be seen in /proc/<pid>/maps
2026 * @size: size to be set for the file
2029 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2033 struct inode *inode;
2034 struct dentry *dentry, *root;
2037 if (IS_ERR(shm_mnt))
2038 return (void *)shm_mnt;
2040 if (size > SHMEM_MAX_BYTES)
2041 return ERR_PTR(-EINVAL);
2043 if (shmem_acct_size(flags, size))
2044 return ERR_PTR(-ENOMEM);
2048 this.len = strlen(name);
2049 this.hash = 0; /* will go */
2050 root = shm_mnt->mnt_root;
2051 dentry = d_alloc(root, &this);
2056 file = get_empty_filp();
2061 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2065 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2066 d_instantiate(dentry, inode);
2067 inode->i_size = size;
2068 inode->i_nlink = 0; /* It is unlinked */
2069 file->f_vfsmnt = mntget(shm_mnt);
2070 file->f_dentry = dentry;
2071 file->f_mapping = inode->i_mapping;
2072 file->f_op = &shmem_file_operations;
2073 file->f_mode = FMODE_WRITE | FMODE_READ;
2081 shmem_unacct_size(flags, size);
2082 return ERR_PTR(error);
2086 * shmem_zero_setup - setup a shared anonymous mapping
2088 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2090 int shmem_zero_setup(struct vm_area_struct *vma)
2093 loff_t size = vma->vm_end - vma->vm_start;
2095 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2097 return PTR_ERR(file);
2101 vma->vm_file = file;
2102 vma->vm_ops = &shmem_vm_ops;
2106 EXPORT_SYMBOL(shmem_file_setup);