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 <asm/uaccess.h>
44 #include <asm/div64.h>
45 #include <asm/pgtable.h>
47 /* This magic number is used in glibc for posix shared memory */
48 #define TMPFS_MAGIC 0x01021994
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)
108 * We have to avoid the unconditional inc_preempt_count()
109 * in kmap_atomic(), since shmem_swp_unmap() will also be
110 * applied to the low memory addresses within i_direct[].
111 * PageHighMem and high_memory tests are good for all arches
112 * and configs: highmem_start_page and FIXADDR_START are not.
114 return PageHighMem(page)?
115 (swp_entry_t *)kmap_atomic(page, KM_USER1):
116 (swp_entry_t *)page_address(page);
119 static inline void shmem_swp_unmap(swp_entry_t *entry)
121 if (entry >= (swp_entry_t *)high_memory)
122 kunmap_atomic(entry, KM_USER1);
125 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
127 return sb->s_fs_info;
131 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
132 * for shared memory and for shared anonymous (/dev/zero) mappings
133 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
134 * consistent with the pre-accounting of private mappings ...
136 static inline int shmem_acct_size(unsigned long flags, loff_t size)
138 return (flags & VM_ACCOUNT)?
139 security_vm_enough_memory(VM_ACCT(size)): 0;
142 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
144 if (flags & VM_ACCOUNT)
145 vm_unacct_memory(VM_ACCT(size));
149 * ... whereas tmpfs objects are accounted incrementally as
150 * pages are allocated, in order to allow huge sparse files.
151 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
152 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
154 static inline int shmem_acct_block(unsigned long flags)
156 return (flags & VM_ACCOUNT)?
157 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
160 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
162 if (!(flags & VM_ACCOUNT))
163 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
166 static struct super_operations shmem_ops;
167 static struct address_space_operations shmem_aops;
168 static struct file_operations shmem_file_operations;
169 static struct inode_operations shmem_inode_operations;
170 static struct inode_operations shmem_dir_inode_operations;
171 static struct vm_operations_struct shmem_vm_ops;
173 static struct backing_dev_info shmem_backing_dev_info = {
174 .ra_pages = 0, /* No readahead */
175 .memory_backed = 1, /* Does not contribute to dirty memory */
176 .unplug_io_fn = default_unplug_io_fn,
179 LIST_HEAD(shmem_inodes);
180 static spinlock_t shmem_ilock = SPIN_LOCK_UNLOCKED;
182 static void shmem_free_block(struct inode *inode)
184 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
185 spin_lock(&sbinfo->stat_lock);
186 sbinfo->free_blocks++;
187 inode->i_blocks -= BLOCKS_PER_PAGE;
188 spin_unlock(&sbinfo->stat_lock);
192 * shmem_recalc_inode - recalculate the size of an inode
194 * @inode: inode to recalc
196 * We have to calculate the free blocks since the mm can drop
197 * undirtied hole pages behind our back.
199 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
200 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
202 * It has to be called with the spinlock held.
204 static void shmem_recalc_inode(struct inode *inode)
206 struct shmem_inode_info *info = SHMEM_I(inode);
209 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
211 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
212 info->alloced -= freed;
213 spin_lock(&sbinfo->stat_lock);
214 sbinfo->free_blocks += freed;
215 inode->i_blocks -= freed*BLOCKS_PER_PAGE;
216 spin_unlock(&sbinfo->stat_lock);
217 shmem_unacct_blocks(info->flags, freed);
222 * shmem_swp_entry - find the swap vector position in the info structure
224 * @info: info structure for the inode
225 * @index: index of the page to find
226 * @page: optional page to add to the structure. Has to be preset to
229 * If there is no space allocated yet it will return NULL when
230 * page is NULL, else it will use the page for the needed block,
231 * setting it to NULL on return to indicate that it has been used.
233 * The swap vector is organized the following way:
235 * There are SHMEM_NR_DIRECT entries directly stored in the
236 * shmem_inode_info structure. So small files do not need an addional
239 * For pages with index > SHMEM_NR_DIRECT there is the pointer
240 * i_indirect which points to a page which holds in the first half
241 * doubly indirect blocks, in the second half triple indirect blocks:
243 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
244 * following layout (for SHMEM_NR_DIRECT == 16):
246 * i_indirect -> dir --> 16-19
259 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
261 unsigned long offset;
265 if (index < SHMEM_NR_DIRECT)
266 return info->i_direct+index;
267 if (!info->i_indirect) {
269 info->i_indirect = *page;
272 return NULL; /* need another page */
275 index -= SHMEM_NR_DIRECT;
276 offset = index % ENTRIES_PER_PAGE;
277 index /= ENTRIES_PER_PAGE;
278 dir = shmem_dir_map(info->i_indirect);
280 if (index >= ENTRIES_PER_PAGE/2) {
281 index -= ENTRIES_PER_PAGE/2;
282 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
283 index %= ENTRIES_PER_PAGE;
290 shmem_dir_unmap(dir);
291 return NULL; /* need another page */
293 shmem_dir_unmap(dir);
294 dir = shmem_dir_map(subdir);
300 if (!page || !(subdir = *page)) {
301 shmem_dir_unmap(dir);
302 return NULL; /* need a page */
307 shmem_dir_unmap(dir);
310 * With apologies... caller shmem_swp_alloc passes non-NULL
311 * page (though perhaps NULL *page); and now we know that this
312 * indirect page has been allocated, we can shortcut the final
313 * kmap if we know it contains no swap entries, as is commonly
314 * the case: return pointer to a 0 which doesn't need kmapping.
316 return (page && !subdir->nr_swapped)?
317 (swp_entry_t *)&subdir->nr_swapped:
318 shmem_swp_map(subdir) + offset;
321 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
323 long incdec = value? 1: -1;
326 info->swapped += incdec;
327 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
328 kmap_atomic_to_page(entry)->nr_swapped += incdec;
332 * shmem_swp_alloc - get the position of the swap entry for the page.
333 * If it does not exist allocate the entry.
335 * @info: info structure for the inode
336 * @index: index of the page to find
337 * @sgp: check and recheck i_size? skip allocation?
339 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
341 struct inode *inode = &info->vfs_inode;
342 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
343 struct page *page = NULL;
345 static const swp_entry_t unswapped = { 0 };
347 if (sgp != SGP_WRITE &&
348 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
349 return ERR_PTR(-EINVAL);
351 while (!(entry = shmem_swp_entry(info, index, &page))) {
353 return (swp_entry_t *) &unswapped;
355 * Test free_blocks against 1 not 0, since we have 1 data
356 * page (and perhaps indirect index pages) yet to allocate:
357 * a waste to allocate index if we cannot allocate data.
359 spin_lock(&sbinfo->stat_lock);
360 if (sbinfo->free_blocks <= 1) {
361 spin_unlock(&sbinfo->stat_lock);
362 return ERR_PTR(-ENOSPC);
364 sbinfo->free_blocks--;
365 inode->i_blocks += BLOCKS_PER_PAGE;
366 spin_unlock(&sbinfo->stat_lock);
368 spin_unlock(&info->lock);
369 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
371 clear_highpage(page);
372 page->nr_swapped = 0;
374 spin_lock(&info->lock);
377 shmem_free_block(inode);
378 return ERR_PTR(-ENOMEM);
380 if (sgp != SGP_WRITE &&
381 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
382 entry = ERR_PTR(-EINVAL);
385 if (info->next_index <= index)
386 info->next_index = index + 1;
389 /* another task gave its page, or truncated the file */
390 shmem_free_block(inode);
391 shmem_dir_free(page);
393 if (info->next_index <= index && !IS_ERR(entry))
394 info->next_index = index + 1;
399 * shmem_free_swp - free some swap entries in a directory
401 * @dir: pointer to the directory
402 * @edir: pointer after last entry of the directory
404 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
409 for (ptr = dir; ptr < edir; ptr++) {
411 free_swap_and_cache(*ptr);
412 *ptr = (swp_entry_t){0};
419 static void shmem_truncate(struct inode *inode)
421 struct shmem_inode_info *info = SHMEM_I(inode);
433 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
434 idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
435 if (idx >= info->next_index)
438 spin_lock(&info->lock);
439 info->flags |= SHMEM_TRUNCATE;
440 limit = info->next_index;
441 info->next_index = idx;
442 if (info->swapped && idx < SHMEM_NR_DIRECT) {
443 ptr = info->i_direct;
445 if (size > SHMEM_NR_DIRECT)
446 size = SHMEM_NR_DIRECT;
447 info->swapped -= shmem_free_swp(ptr+idx, ptr+size);
449 if (!info->i_indirect)
452 BUG_ON(limit <= SHMEM_NR_DIRECT);
453 limit -= SHMEM_NR_DIRECT;
454 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
455 offset = idx % ENTRIES_PER_PAGE;
459 dir = shmem_dir_map(info->i_indirect);
460 stage = ENTRIES_PER_PAGEPAGE/2;
461 if (idx < ENTRIES_PER_PAGEPAGE/2)
462 dir += idx/ENTRIES_PER_PAGE;
464 dir += ENTRIES_PER_PAGE/2;
465 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
467 stage += ENTRIES_PER_PAGEPAGE;
470 size = ((idx - ENTRIES_PER_PAGEPAGE/2) %
471 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
472 if (!size && !offset) {
476 shmem_dir_unmap(dir);
477 dir = shmem_dir_map(subdir) + size;
484 for (; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
485 if (unlikely(idx == stage)) {
486 shmem_dir_unmap(dir-1);
487 dir = shmem_dir_map(info->i_indirect) +
488 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
491 idx += ENTRIES_PER_PAGEPAGE;
495 stage = idx + ENTRIES_PER_PAGEPAGE;
498 shmem_dir_unmap(dir);
500 shmem_dir_free(empty);
501 shmem_free_block(inode);
504 cond_resched_lock(&info->lock);
505 dir = shmem_dir_map(subdir);
508 if (subdir && subdir->nr_swapped) {
509 ptr = shmem_swp_map(subdir);
511 if (size > ENTRIES_PER_PAGE)
512 size = ENTRIES_PER_PAGE;
513 freed = shmem_free_swp(ptr+offset, ptr+size);
514 shmem_swp_unmap(ptr);
515 info->swapped -= freed;
516 subdir->nr_swapped -= freed;
517 BUG_ON(subdir->nr_swapped > offset);
523 shmem_dir_free(subdir);
524 shmem_free_block(inode);
528 shmem_dir_unmap(dir-1);
530 shmem_dir_free(empty);
531 shmem_free_block(inode);
533 if (info->next_index <= SHMEM_NR_DIRECT) {
534 shmem_dir_free(info->i_indirect);
535 info->i_indirect = NULL;
536 shmem_free_block(inode);
539 BUG_ON(info->swapped > info->next_index);
540 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
542 * Call truncate_inode_pages again: racing shmem_unuse_inode
543 * may have swizzled a page in from swap since vmtruncate or
544 * generic_delete_inode did it, before we lowered next_index.
545 * Also, though shmem_getpage checks i_size before adding to
546 * cache, no recheck after: so fix the narrow window there too.
548 spin_unlock(&info->lock);
549 truncate_inode_pages(inode->i_mapping, inode->i_size);
550 spin_lock(&info->lock);
552 info->flags &= ~SHMEM_TRUNCATE;
553 shmem_recalc_inode(inode);
554 spin_unlock(&info->lock);
557 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
559 struct inode *inode = dentry->d_inode;
560 struct page *page = NULL;
563 if (attr->ia_valid & ATTR_SIZE) {
564 if (attr->ia_size < inode->i_size) {
566 * If truncating down to a partial page, then
567 * if that page is already allocated, hold it
568 * in memory until the truncation is over, so
569 * truncate_partial_page cannnot miss it were
570 * it assigned to swap.
572 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
573 (void) shmem_getpage(inode,
574 attr->ia_size>>PAGE_CACHE_SHIFT,
575 &page, SGP_READ, NULL);
578 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
579 * detect if any pages might have been added to cache
580 * after truncate_inode_pages. But we needn't bother
581 * if it's being fully truncated to zero-length: the
582 * nrpages check is efficient enough in that case.
585 struct shmem_inode_info *info = SHMEM_I(inode);
586 spin_lock(&info->lock);
587 info->flags &= ~SHMEM_PAGEIN;
588 spin_unlock(&info->lock);
593 error = inode_change_ok(inode, attr);
595 error = inode_setattr(inode, attr);
597 page_cache_release(page);
601 static void shmem_delete_inode(struct inode *inode)
603 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
604 struct shmem_inode_info *info = SHMEM_I(inode);
606 if (inode->i_op->truncate == shmem_truncate) {
607 spin_lock(&shmem_ilock);
608 list_del(&info->list);
609 spin_unlock(&shmem_ilock);
610 shmem_unacct_size(info->flags, inode->i_size);
612 shmem_truncate(inode);
614 BUG_ON(inode->i_blocks);
615 spin_lock(&sbinfo->stat_lock);
616 sbinfo->free_inodes++;
617 spin_unlock(&sbinfo->stat_lock);
621 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
625 for (ptr = dir; ptr < edir; ptr++) {
626 if (ptr->val == entry.val)
632 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
645 ptr = info->i_direct;
646 spin_lock(&info->lock);
647 limit = info->next_index;
649 if (size > SHMEM_NR_DIRECT)
650 size = SHMEM_NR_DIRECT;
651 offset = shmem_find_swp(entry, ptr, ptr+size);
654 if (!info->i_indirect)
656 /* we might be racing with shmem_truncate */
657 if (limit <= SHMEM_NR_DIRECT)
660 dir = shmem_dir_map(info->i_indirect);
661 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
663 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
664 if (unlikely(idx == stage)) {
665 shmem_dir_unmap(dir-1);
666 dir = shmem_dir_map(info->i_indirect) +
667 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
670 idx += ENTRIES_PER_PAGEPAGE;
674 stage = idx + ENTRIES_PER_PAGEPAGE;
676 shmem_dir_unmap(dir);
677 dir = shmem_dir_map(subdir);
680 if (subdir && subdir->nr_swapped) {
681 ptr = shmem_swp_map(subdir);
683 if (size > ENTRIES_PER_PAGE)
684 size = ENTRIES_PER_PAGE;
685 offset = shmem_find_swp(entry, ptr, ptr+size);
687 shmem_dir_unmap(dir);
690 shmem_swp_unmap(ptr);
694 shmem_dir_unmap(dir-1);
696 spin_unlock(&info->lock);
700 inode = &info->vfs_inode;
701 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
702 info->flags |= SHMEM_PAGEIN;
703 shmem_swp_set(info, ptr + offset, 0);
705 shmem_swp_unmap(ptr);
706 spin_unlock(&info->lock);
708 * Decrement swap count even when the entry is left behind:
709 * try_to_unuse will skip over mms, then reincrement count.
716 * shmem_unuse() search for an eventually swapped out shmem page.
718 int shmem_unuse(swp_entry_t entry, struct page *page)
721 struct shmem_inode_info *info;
724 spin_lock(&shmem_ilock);
725 list_for_each(p, &shmem_inodes) {
726 info = list_entry(p, struct shmem_inode_info, list);
728 if (info->swapped && shmem_unuse_inode(info, entry, page)) {
729 /* move head to start search for next from here */
730 list_move_tail(&shmem_inodes, &info->list);
735 spin_unlock(&shmem_ilock);
740 * Move the page from the page cache to the swap cache.
742 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
744 struct shmem_inode_info *info;
745 swp_entry_t *entry, swap;
746 struct address_space *mapping;
750 BUG_ON(!PageLocked(page));
751 BUG_ON(page_mapped(page));
753 mapping = page->mapping;
755 inode = mapping->host;
756 info = SHMEM_I(inode);
757 if (info->flags & VM_LOCKED)
759 swap = get_swap_page();
763 spin_lock(&info->lock);
764 shmem_recalc_inode(inode);
765 if (index >= info->next_index) {
766 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
769 entry = shmem_swp_entry(info, index, NULL);
773 if (move_to_swap_cache(page, swap) == 0) {
774 shmem_swp_set(info, entry, swap.val);
775 shmem_swp_unmap(entry);
776 spin_unlock(&info->lock);
781 shmem_swp_unmap(entry);
783 spin_unlock(&info->lock);
786 set_page_dirty(page);
787 return WRITEPAGE_ACTIVATE; /* Return with the page locked */
791 static struct page *shmem_swapin_async(struct shared_policy *p,
792 swp_entry_t entry, unsigned long idx)
795 struct vm_area_struct pvma;
797 /* Create a pseudo vma that just contains the policy */
798 memset(&pvma, 0, sizeof(struct vm_area_struct));
799 pvma.vm_end = PAGE_SIZE;
801 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
802 page = read_swap_cache_async(entry, &pvma, 0);
803 mpol_free(pvma.vm_policy);
807 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
810 struct shared_policy *p = &info->policy;
813 unsigned long offset;
815 num = valid_swaphandles(entry, &offset);
816 for (i = 0; i < num; offset++, i++) {
817 page = shmem_swapin_async(p,
818 swp_entry(swp_type(entry), offset), idx);
821 page_cache_release(page);
823 lru_add_drain(); /* Push any new pages onto the LRU now */
824 return shmem_swapin_async(p, entry, idx);
828 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
831 struct vm_area_struct pvma;
834 memset(&pvma, 0, sizeof(struct vm_area_struct));
835 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
837 pvma.vm_end = PAGE_SIZE;
838 page = alloc_page_vma(gfp, &pvma, 0);
839 mpol_free(pvma.vm_policy);
843 static inline struct page *
844 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
846 swapin_readahead(entry, 0, NULL);
847 return read_swap_cache_async(entry, NULL, 0);
850 static inline struct page *
851 shmem_alloc_page(unsigned long gfp,struct shmem_inode_info *info,
854 return alloc_page(gfp);
859 * shmem_getpage - either get the page from swap or allocate a new one
861 * If we allocate a new one we do not mark it dirty. That's up to the
862 * vm. If we swap it in we mark it dirty since we also free the swap
863 * entry since a page cannot live in both the swap and page cache
865 static int shmem_getpage(struct inode *inode, unsigned long idx,
866 struct page **pagep, enum sgp_type sgp, int *type)
868 struct address_space *mapping = inode->i_mapping;
869 struct shmem_inode_info *info = SHMEM_I(inode);
870 struct shmem_sb_info *sbinfo;
871 struct page *filepage = *pagep;
872 struct page *swappage;
875 int error, majmin = VM_FAULT_MINOR;
877 if (idx >= SHMEM_MAX_INDEX)
880 * Normally, filepage is NULL on entry, and either found
881 * uptodate immediately, or allocated and zeroed, or read
882 * in under swappage, which is then assigned to filepage.
883 * But shmem_prepare_write passes in a locked filepage,
884 * which may be found not uptodate by other callers too,
885 * and may need to be copied from the swappage read in.
889 filepage = find_lock_page(mapping, idx);
890 if (filepage && PageUptodate(filepage))
893 if (sgp == SGP_QUICK)
896 spin_lock(&info->lock);
897 shmem_recalc_inode(inode);
898 entry = shmem_swp_alloc(info, idx, sgp);
900 spin_unlock(&info->lock);
901 error = PTR_ERR(entry);
907 /* Look it up and read it in.. */
908 swappage = lookup_swap_cache(swap);
910 shmem_swp_unmap(entry);
911 spin_unlock(&info->lock);
912 /* here we actually do the io */
913 if (majmin == VM_FAULT_MINOR && type)
914 inc_page_state(pgmajfault);
915 majmin = VM_FAULT_MAJOR;
916 swappage = shmem_swapin(info, swap, idx);
918 spin_lock(&info->lock);
919 entry = shmem_swp_alloc(info, idx, sgp);
921 error = PTR_ERR(entry);
923 if (entry->val == swap.val)
925 shmem_swp_unmap(entry);
927 spin_unlock(&info->lock);
932 wait_on_page_locked(swappage);
933 page_cache_release(swappage);
937 /* We have to do this with page locked to prevent races */
938 if (TestSetPageLocked(swappage)) {
939 shmem_swp_unmap(entry);
940 spin_unlock(&info->lock);
941 wait_on_page_locked(swappage);
942 page_cache_release(swappage);
945 if (PageWriteback(swappage)) {
946 shmem_swp_unmap(entry);
947 spin_unlock(&info->lock);
948 wait_on_page_writeback(swappage);
949 unlock_page(swappage);
950 page_cache_release(swappage);
953 if (!PageUptodate(swappage)) {
954 shmem_swp_unmap(entry);
955 spin_unlock(&info->lock);
956 unlock_page(swappage);
957 page_cache_release(swappage);
963 shmem_swp_set(info, entry, 0);
964 shmem_swp_unmap(entry);
965 delete_from_swap_cache(swappage);
966 spin_unlock(&info->lock);
967 copy_highpage(filepage, swappage);
968 unlock_page(swappage);
969 page_cache_release(swappage);
970 flush_dcache_page(filepage);
971 SetPageUptodate(filepage);
972 set_page_dirty(filepage);
974 } else if (!(error = move_from_swap_cache(
975 swappage, idx, mapping))) {
976 info->flags |= SHMEM_PAGEIN;
977 shmem_swp_set(info, entry, 0);
978 shmem_swp_unmap(entry);
979 spin_unlock(&info->lock);
983 shmem_swp_unmap(entry);
984 spin_unlock(&info->lock);
985 unlock_page(swappage);
986 page_cache_release(swappage);
987 if (error == -ENOMEM) {
988 /* let kswapd refresh zone for GFP_ATOMICs */
989 blk_congestion_wait(WRITE, HZ/50);
993 } else if (sgp == SGP_READ && !filepage) {
994 shmem_swp_unmap(entry);
995 filepage = find_get_page(mapping, idx);
997 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
998 spin_unlock(&info->lock);
999 wait_on_page_locked(filepage);
1000 page_cache_release(filepage);
1004 spin_unlock(&info->lock);
1006 shmem_swp_unmap(entry);
1007 sbinfo = SHMEM_SB(inode->i_sb);
1008 spin_lock(&sbinfo->stat_lock);
1009 if (sbinfo->free_blocks == 0 || shmem_acct_block(info->flags)) {
1010 spin_unlock(&sbinfo->stat_lock);
1011 spin_unlock(&info->lock);
1015 sbinfo->free_blocks--;
1016 inode->i_blocks += BLOCKS_PER_PAGE;
1017 spin_unlock(&sbinfo->stat_lock);
1020 spin_unlock(&info->lock);
1021 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1025 shmem_unacct_blocks(info->flags, 1);
1026 shmem_free_block(inode);
1031 spin_lock(&info->lock);
1032 entry = shmem_swp_alloc(info, idx, sgp);
1034 error = PTR_ERR(entry);
1037 shmem_swp_unmap(entry);
1039 if (error || swap.val || 0 != add_to_page_cache_lru(
1040 filepage, mapping, idx, GFP_ATOMIC)) {
1041 spin_unlock(&info->lock);
1042 page_cache_release(filepage);
1043 shmem_unacct_blocks(info->flags, 1);
1044 shmem_free_block(inode);
1050 info->flags |= SHMEM_PAGEIN;
1054 spin_unlock(&info->lock);
1055 clear_highpage(filepage);
1056 flush_dcache_page(filepage);
1057 SetPageUptodate(filepage);
1062 unlock_page(filepage);
1065 *pagep = ZERO_PAGE(0);
1072 if (*pagep != filepage) {
1073 unlock_page(filepage);
1074 page_cache_release(filepage);
1079 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1081 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1082 struct page *page = NULL;
1086 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1087 idx += vma->vm_pgoff;
1088 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1090 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1092 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1094 mark_page_accessed(page);
1098 static int shmem_populate(struct vm_area_struct *vma,
1099 unsigned long addr, unsigned long len,
1100 pgprot_t prot, unsigned long pgoff, int nonblock)
1102 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1103 struct mm_struct *mm = vma->vm_mm;
1104 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1107 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1108 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1111 while ((long) len > 0) {
1112 struct page *page = NULL;
1115 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1117 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1121 mark_page_accessed(page);
1122 err = install_page(mm, vma, addr, page, prot);
1124 page_cache_release(page);
1127 } else if (nonblock) {
1129 * If a nonlinear mapping then store the file page
1130 * offset in the pte.
1132 if (pgoff != linear_page_index(vma, addr)) {
1133 err = install_file_pte(mm, vma, addr, pgoff, prot);
1147 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1149 struct inode *i = vma->vm_file->f_dentry->d_inode;
1150 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1154 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1156 struct inode *i = vma->vm_file->f_dentry->d_inode;
1159 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1160 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1164 void shmem_lock(struct file *file, int lock)
1166 struct inode *inode = file->f_dentry->d_inode;
1167 struct shmem_inode_info *info = SHMEM_I(inode);
1169 spin_lock(&info->lock);
1171 info->flags |= VM_LOCKED;
1173 info->flags &= ~VM_LOCKED;
1174 spin_unlock(&info->lock);
1177 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1179 file_accessed(file);
1180 vma->vm_ops = &shmem_vm_ops;
1184 static struct inode *
1185 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1187 struct inode *inode;
1188 struct shmem_inode_info *info;
1189 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1191 spin_lock(&sbinfo->stat_lock);
1192 if (!sbinfo->free_inodes) {
1193 spin_unlock(&sbinfo->stat_lock);
1196 sbinfo->free_inodes--;
1197 spin_unlock(&sbinfo->stat_lock);
1199 inode = new_inode(sb);
1201 inode->i_mode = mode;
1202 inode->i_uid = current->fsuid;
1203 inode->i_gid = current->fsgid;
1204 inode->i_blksize = PAGE_CACHE_SIZE;
1205 inode->i_blocks = 0;
1206 inode->i_mapping->a_ops = &shmem_aops;
1207 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1208 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1209 info = SHMEM_I(inode);
1210 memset(info, 0, (char *)inode - (char *)info);
1211 spin_lock_init(&info->lock);
1212 mpol_shared_policy_init(&info->policy);
1213 switch (mode & S_IFMT) {
1215 init_special_inode(inode, mode, dev);
1218 inode->i_op = &shmem_inode_operations;
1219 inode->i_fop = &shmem_file_operations;
1220 spin_lock(&shmem_ilock);
1221 list_add_tail(&info->list, &shmem_inodes);
1222 spin_unlock(&shmem_ilock);
1226 /* Some things misbehave if size == 0 on a directory */
1227 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1228 inode->i_op = &shmem_dir_inode_operations;
1229 inode->i_fop = &simple_dir_operations;
1238 static int shmem_set_size(struct shmem_sb_info *info,
1239 unsigned long max_blocks, unsigned long max_inodes)
1242 unsigned long blocks, inodes;
1244 spin_lock(&info->stat_lock);
1245 blocks = info->max_blocks - info->free_blocks;
1246 inodes = info->max_inodes - info->free_inodes;
1248 if (max_blocks < blocks)
1250 if (max_inodes < inodes)
1253 info->max_blocks = max_blocks;
1254 info->free_blocks = max_blocks - blocks;
1255 info->max_inodes = max_inodes;
1256 info->free_inodes = max_inodes - inodes;
1258 spin_unlock(&info->stat_lock);
1264 static struct inode_operations shmem_symlink_inode_operations;
1265 static struct inode_operations shmem_symlink_inline_operations;
1268 * Normally tmpfs makes no use of shmem_prepare_write, but it
1269 * lets a tmpfs file be used read-write below the loop driver.
1272 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1274 struct inode *inode = page->mapping->host;
1275 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1279 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1281 struct inode *inode = file->f_dentry->d_inode;
1283 unsigned long written;
1286 if ((ssize_t) count < 0)
1289 if (!access_ok(VERIFY_READ, buf, count))
1292 down(&inode->i_sem);
1297 err = generic_write_checks(file, &pos, &count, 0);
1301 err = remove_suid(file->f_dentry);
1305 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1308 struct page *page = NULL;
1309 unsigned long bytes, index, offset;
1313 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1314 index = pos >> PAGE_CACHE_SHIFT;
1315 bytes = PAGE_CACHE_SIZE - offset;
1320 * We don't hold page lock across copy from user -
1321 * what would it guard against? - so no deadlock here.
1322 * But it still may be a good idea to prefault below.
1325 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1330 if (PageHighMem(page)) {
1331 volatile unsigned char dummy;
1332 __get_user(dummy, buf);
1333 __get_user(dummy, buf + bytes - 1);
1335 kaddr = kmap_atomic(page, KM_USER0);
1336 left = __copy_from_user(kaddr + offset, buf, bytes);
1337 kunmap_atomic(kaddr, KM_USER0);
1341 left = __copy_from_user(kaddr + offset, buf, bytes);
1349 if (pos > inode->i_size)
1350 i_size_write(inode, pos);
1352 flush_dcache_page(page);
1353 set_page_dirty(page);
1354 mark_page_accessed(page);
1355 page_cache_release(page);
1365 * Our dirty pages are not counted in nr_dirty,
1366 * and we do not attempt to balance dirty pages.
1380 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1382 struct inode *inode = filp->f_dentry->d_inode;
1383 struct address_space *mapping = inode->i_mapping;
1384 unsigned long index, offset;
1386 index = *ppos >> PAGE_CACHE_SHIFT;
1387 offset = *ppos & ~PAGE_CACHE_MASK;
1390 struct page *page = NULL;
1391 unsigned long end_index, nr, ret;
1392 loff_t i_size = i_size_read(inode);
1394 end_index = i_size >> PAGE_CACHE_SHIFT;
1395 if (index > end_index)
1397 if (index == end_index) {
1398 nr = i_size & ~PAGE_CACHE_MASK;
1403 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1405 if (desc->error == -EINVAL)
1411 * We must evaluate after, since reads (unlike writes)
1412 * are called without i_sem protection against truncate
1414 nr = PAGE_CACHE_SIZE;
1415 i_size = i_size_read(inode);
1416 end_index = i_size >> PAGE_CACHE_SHIFT;
1417 if (index == end_index) {
1418 nr = i_size & ~PAGE_CACHE_MASK;
1420 page_cache_release(page);
1426 if (page != ZERO_PAGE(0)) {
1428 * If users can be writing to this page using arbitrary
1429 * virtual addresses, take care about potential aliasing
1430 * before reading the page on the kernel side.
1432 if (mapping_writably_mapped(mapping))
1433 flush_dcache_page(page);
1435 * Mark the page accessed if we read the beginning.
1438 mark_page_accessed(page);
1442 * Ok, we have the page, and it's up-to-date, so
1443 * now we can copy it to user space...
1445 * The actor routine returns how many bytes were actually used..
1446 * NOTE! This may not be the same as how much of a user buffer
1447 * we filled up (we may be padding etc), so we can only update
1448 * "pos" here (the actor routine has to update the user buffer
1449 * pointers and the remaining count).
1451 ret = actor(desc, page, offset, nr);
1453 index += offset >> PAGE_CACHE_SHIFT;
1454 offset &= ~PAGE_CACHE_MASK;
1456 page_cache_release(page);
1457 if (ret != nr || !desc->count)
1463 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1464 file_accessed(filp);
1467 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1469 read_descriptor_t desc;
1471 if ((ssize_t) count < 0)
1473 if (!access_ok(VERIFY_WRITE, buf, count))
1483 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1485 return desc.written;
1489 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1490 size_t count, read_actor_t actor, void __user *target)
1492 read_descriptor_t desc;
1502 do_shmem_file_read(in_file, ppos, &desc, actor);
1504 return desc.written;
1508 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1510 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1512 buf->f_type = TMPFS_MAGIC;
1513 buf->f_bsize = PAGE_CACHE_SIZE;
1514 spin_lock(&sbinfo->stat_lock);
1515 buf->f_blocks = sbinfo->max_blocks;
1516 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1517 buf->f_files = sbinfo->max_inodes;
1518 buf->f_ffree = sbinfo->free_inodes;
1519 spin_unlock(&sbinfo->stat_lock);
1520 buf->f_namelen = NAME_MAX;
1525 * File creation. Allocate an inode, and we're done..
1528 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1530 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1531 int error = -ENOSPC;
1534 if (dir->i_mode & S_ISGID) {
1535 inode->i_gid = dir->i_gid;
1537 inode->i_mode |= S_ISGID;
1539 dir->i_size += BOGO_DIRENT_SIZE;
1540 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1541 d_instantiate(dentry, inode);
1542 dget(dentry); /* Extra count - pin the dentry in core */
1548 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1552 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1558 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1559 struct nameidata *nd)
1561 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1567 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1569 struct inode *inode = old_dentry->d_inode;
1571 dir->i_size += BOGO_DIRENT_SIZE;
1572 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1574 atomic_inc(&inode->i_count); /* New dentry reference */
1575 dget(dentry); /* Extra pinning count for the created dentry */
1576 d_instantiate(dentry, inode);
1580 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1582 struct inode *inode = dentry->d_inode;
1584 dir->i_size -= BOGO_DIRENT_SIZE;
1585 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1587 dput(dentry); /* Undo the count from "create" - this does all the work */
1591 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1593 if (!simple_empty(dentry))
1597 return shmem_unlink(dir, dentry);
1601 * The VFS layer already does all the dentry stuff for rename,
1602 * we just have to decrement the usage count for the target if
1603 * it exists so that the VFS layer correctly free's it when it
1606 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1608 struct inode *inode = old_dentry->d_inode;
1609 int they_are_dirs = S_ISDIR(inode->i_mode);
1611 if (!simple_empty(new_dentry))
1614 if (new_dentry->d_inode) {
1615 (void) shmem_unlink(new_dir, new_dentry);
1618 } else if (they_are_dirs) {
1623 old_dir->i_size -= BOGO_DIRENT_SIZE;
1624 new_dir->i_size += BOGO_DIRENT_SIZE;
1625 old_dir->i_ctime = old_dir->i_mtime =
1626 new_dir->i_ctime = new_dir->i_mtime =
1627 inode->i_ctime = CURRENT_TIME;
1631 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1635 struct inode *inode;
1636 struct page *page = NULL;
1638 struct shmem_inode_info *info;
1640 len = strlen(symname) + 1;
1641 if (len > PAGE_CACHE_SIZE)
1642 return -ENAMETOOLONG;
1644 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1648 info = SHMEM_I(inode);
1649 inode->i_size = len-1;
1650 if (len <= (char *)inode - (char *)info) {
1652 memcpy(info, symname, len);
1653 inode->i_op = &shmem_symlink_inline_operations;
1655 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1660 inode->i_op = &shmem_symlink_inode_operations;
1661 spin_lock(&shmem_ilock);
1662 list_add_tail(&info->list, &shmem_inodes);
1663 spin_unlock(&shmem_ilock);
1664 kaddr = kmap_atomic(page, KM_USER0);
1665 memcpy(kaddr, symname, len);
1666 kunmap_atomic(kaddr, KM_USER0);
1667 set_page_dirty(page);
1668 page_cache_release(page);
1670 if (dir->i_mode & S_ISGID)
1671 inode->i_gid = dir->i_gid;
1672 dir->i_size += BOGO_DIRENT_SIZE;
1673 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1674 d_instantiate(dentry, inode);
1679 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1681 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1685 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1687 struct page *page = NULL;
1688 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1689 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1693 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1695 if (!IS_ERR(nd_get_link(nd))) {
1698 page = find_get_page(dentry->d_inode->i_mapping, 0);
1702 mark_page_accessed(page);
1703 page_cache_release(page);
1704 page_cache_release(page);
1708 static struct inode_operations shmem_symlink_inline_operations = {
1709 .readlink = generic_readlink,
1710 .follow_link = shmem_follow_link_inline,
1713 static struct inode_operations shmem_symlink_inode_operations = {
1714 .truncate = shmem_truncate,
1715 .readlink = generic_readlink,
1716 .follow_link = shmem_follow_link,
1717 .put_link = shmem_put_link,
1720 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1722 char *this_char, *value, *rest;
1724 while ((this_char = strsep(&options, ",")) != NULL) {
1727 if ((value = strchr(this_char,'=')) != NULL) {
1731 "tmpfs: No value for mount option '%s'\n",
1736 if (!strcmp(this_char,"size")) {
1737 unsigned long long size;
1738 size = memparse(value,&rest);
1740 size <<= PAGE_SHIFT;
1741 size *= totalram_pages;
1747 *blocks = size >> PAGE_CACHE_SHIFT;
1748 } else if (!strcmp(this_char,"nr_blocks")) {
1749 *blocks = memparse(value,&rest);
1752 } else if (!strcmp(this_char,"nr_inodes")) {
1753 *inodes = memparse(value,&rest);
1756 } else if (!strcmp(this_char,"mode")) {
1759 *mode = simple_strtoul(value,&rest,8);
1762 } else if (!strcmp(this_char,"uid")) {
1765 *uid = simple_strtoul(value,&rest,0);
1768 } else if (!strcmp(this_char,"gid")) {
1771 *gid = simple_strtoul(value,&rest,0);
1775 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1783 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1789 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1791 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1792 unsigned long max_blocks = sbinfo->max_blocks;
1793 unsigned long max_inodes = sbinfo->max_inodes;
1795 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
1797 return shmem_set_size(sbinfo, max_blocks, max_inodes);
1801 static int shmem_fill_super(struct super_block *sb,
1802 void *data, int silent)
1804 struct inode *inode;
1805 struct dentry *root;
1806 unsigned long blocks, inodes;
1807 int mode = S_IRWXUGO | S_ISVTX;
1808 uid_t uid = current->fsuid;
1809 gid_t gid = current->fsgid;
1810 struct shmem_sb_info *sbinfo;
1813 sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
1816 sb->s_fs_info = sbinfo;
1817 memset(sbinfo, 0, sizeof(struct shmem_sb_info));
1820 * Per default we only allow half of the physical ram per
1823 blocks = inodes = totalram_pages / 2;
1826 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, &inodes)) {
1831 sb->s_flags |= MS_NOUSER;
1834 spin_lock_init(&sbinfo->stat_lock);
1835 sbinfo->max_blocks = blocks;
1836 sbinfo->free_blocks = blocks;
1837 sbinfo->max_inodes = inodes;
1838 sbinfo->free_inodes = inodes;
1839 sb->s_maxbytes = SHMEM_MAX_BYTES;
1840 sb->s_blocksize = PAGE_CACHE_SIZE;
1841 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1842 sb->s_magic = TMPFS_MAGIC;
1843 sb->s_op = &shmem_ops;
1844 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1849 root = d_alloc_root(inode);
1859 sb->s_fs_info = NULL;
1863 static void shmem_put_super(struct super_block *sb)
1865 kfree(sb->s_fs_info);
1866 sb->s_fs_info = NULL;
1869 static kmem_cache_t *shmem_inode_cachep;
1871 static struct inode *shmem_alloc_inode(struct super_block *sb)
1873 struct shmem_inode_info *p;
1874 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
1877 return &p->vfs_inode;
1880 static void shmem_destroy_inode(struct inode *inode)
1882 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
1883 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
1886 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
1888 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
1890 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1891 SLAB_CTOR_CONSTRUCTOR) {
1892 inode_init_once(&p->vfs_inode);
1896 static int init_inodecache(void)
1898 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
1899 sizeof(struct shmem_inode_info),
1900 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
1902 if (shmem_inode_cachep == NULL)
1907 static void destroy_inodecache(void)
1909 if (kmem_cache_destroy(shmem_inode_cachep))
1910 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
1913 static struct address_space_operations shmem_aops = {
1914 .writepage = shmem_writepage,
1915 .set_page_dirty = __set_page_dirty_nobuffers,
1917 .prepare_write = shmem_prepare_write,
1918 .commit_write = simple_commit_write,
1922 static struct file_operations shmem_file_operations = {
1925 .llseek = generic_file_llseek,
1926 .read = shmem_file_read,
1927 .write = shmem_file_write,
1928 .fsync = simple_sync_file,
1929 .sendfile = shmem_file_sendfile,
1933 static struct inode_operations shmem_inode_operations = {
1934 .truncate = shmem_truncate,
1935 .setattr = shmem_notify_change,
1938 static struct inode_operations shmem_dir_inode_operations = {
1940 .create = shmem_create,
1941 .lookup = simple_lookup,
1943 .unlink = shmem_unlink,
1944 .symlink = shmem_symlink,
1945 .mkdir = shmem_mkdir,
1946 .rmdir = shmem_rmdir,
1947 .mknod = shmem_mknod,
1948 .rename = shmem_rename,
1952 static struct super_operations shmem_ops = {
1953 .alloc_inode = shmem_alloc_inode,
1954 .destroy_inode = shmem_destroy_inode,
1956 .statfs = shmem_statfs,
1957 .remount_fs = shmem_remount_fs,
1959 .delete_inode = shmem_delete_inode,
1960 .drop_inode = generic_delete_inode,
1961 .put_super = shmem_put_super,
1964 static struct vm_operations_struct shmem_vm_ops = {
1965 .nopage = shmem_nopage,
1966 .populate = shmem_populate,
1968 .set_policy = shmem_set_policy,
1969 .get_policy = shmem_get_policy,
1973 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
1974 int flags, const char *dev_name, void *data)
1976 return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
1979 static struct file_system_type tmpfs_fs_type = {
1980 .owner = THIS_MODULE,
1982 .get_sb = shmem_get_sb,
1983 .kill_sb = kill_litter_super,
1985 static struct vfsmount *shm_mnt;
1987 static int __init init_tmpfs(void)
1991 error = init_inodecache();
1995 error = register_filesystem(&tmpfs_fs_type);
1997 printk(KERN_ERR "Could not register tmpfs\n");
2001 devfs_mk_dir("shm");
2003 shm_mnt = kern_mount(&tmpfs_fs_type);
2004 if (IS_ERR(shm_mnt)) {
2005 error = PTR_ERR(shm_mnt);
2006 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2010 /* The internal instance should not do size checking */
2011 shmem_set_size(SHMEM_SB(shm_mnt->mnt_sb), ULONG_MAX, ULONG_MAX);
2015 unregister_filesystem(&tmpfs_fs_type);
2017 destroy_inodecache();
2019 shm_mnt = ERR_PTR(error);
2022 module_init(init_tmpfs)
2025 * shmem_file_setup - get an unlinked file living in tmpfs
2027 * @name: name for dentry (to be seen in /proc/<pid>/maps
2028 * @size: size to be set for the file
2031 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2035 struct inode *inode;
2036 struct dentry *dentry, *root;
2039 if (IS_ERR(shm_mnt))
2040 return (void *)shm_mnt;
2042 if (size > SHMEM_MAX_BYTES)
2043 return ERR_PTR(-EINVAL);
2045 if (shmem_acct_size(flags, size))
2046 return ERR_PTR(-ENOMEM);
2050 this.len = strlen(name);
2051 this.hash = 0; /* will go */
2052 root = shm_mnt->mnt_root;
2053 dentry = d_alloc(root, &this);
2058 file = get_empty_filp();
2063 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2067 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2068 d_instantiate(dentry, inode);
2069 inode->i_size = size;
2070 inode->i_nlink = 0; /* It is unlinked */
2071 file->f_vfsmnt = mntget(shm_mnt);
2072 file->f_dentry = dentry;
2073 file->f_mapping = inode->i_mapping;
2074 file->f_op = &shmem_file_operations;
2075 file->f_mode = FMODE_WRITE | FMODE_READ;
2083 shmem_unacct_size(flags, size);
2084 return ERR_PTR(error);
2088 * shmem_zero_setup - setup a shared anonymous mapping
2090 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2092 int shmem_zero_setup(struct vm_area_struct *vma)
2095 loff_t size = vma->vm_end - vma->vm_start;
2097 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2099 return PTR_ERR(file);
2103 vma->vm_file = file;
2104 vma->vm_ops = &shmem_vm_ops;
2108 EXPORT_SYMBOL(shmem_file_setup);