2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2004 Hugh Dickins.
10 * Copyright (C) 2002-2004 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * This file is released under the GPL.
17 * This virtual memory filesystem is heavily based on the ramfs. It
18 * extends ramfs by the ability to use swap and honor resource limits
19 * which makes it a completely usable filesystem.
22 #include <linux/config.h>
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/devfs_fs_kernel.h>
28 #include <linux/mman.h>
29 #include <linux/file.h>
30 #include <linux/swap.h>
31 #include <linux/pagemap.h>
32 #include <linux/string.h>
33 #include <linux/slab.h>
34 #include <linux/backing-dev.h>
35 #include <linux/shmem_fs.h>
36 #include <linux/mount.h>
37 #include <linux/writeback.h>
38 #include <linux/vfs.h>
39 #include <linux/blkdev.h>
40 #include <linux/security.h>
41 #include <linux/swapops.h>
42 #include <linux/mempolicy.h>
43 #include <linux/namei.h>
44 #include <asm/uaccess.h>
45 #include <asm/div64.h>
46 #include <asm/pgtable.h>
48 /* This magic number is used in glibc for posix shared memory */
49 #define TMPFS_MAGIC 0x01021994
51 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
52 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
53 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
55 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
56 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
58 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
60 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
61 #define SHMEM_PAGEIN VM_READ
62 #define SHMEM_TRUNCATE VM_WRITE
64 /* Pretend that each entry is of this size in directory's i_size */
65 #define BOGO_DIRENT_SIZE 20
67 /* Keep swapped page count in private field of indirect struct page */
68 #define nr_swapped private
70 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
72 SGP_QUICK, /* don't try more than file page cache lookup */
73 SGP_READ, /* don't exceed i_size, don't allocate page */
74 SGP_CACHE, /* don't exceed i_size, may allocate page */
75 SGP_WRITE, /* may exceed i_size, may allocate page */
78 static int shmem_getpage(struct inode *inode, unsigned long idx,
79 struct page **pagep, enum sgp_type sgp, int *type);
81 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
84 * The above definition of ENTRIES_PER_PAGE, and the use of
85 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
86 * might be reconsidered if it ever diverges from PAGE_SIZE.
88 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
91 static inline void shmem_dir_free(struct page *page)
93 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
96 static struct page **shmem_dir_map(struct page *page)
98 return (struct page **)kmap_atomic(page, KM_USER0);
101 static inline void shmem_dir_unmap(struct page **dir)
103 kunmap_atomic(dir, KM_USER0);
106 static swp_entry_t *shmem_swp_map(struct page *page)
108 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
111 static inline void shmem_swp_balance_unmap(void)
114 * When passing a pointer to an i_direct entry, to code which
115 * also handles indirect entries and so will shmem_swp_unmap,
116 * we must arrange for the preempt count to remain in balance.
117 * What kmap_atomic of a lowmem page does depends on config
118 * and architecture, so pretend to kmap_atomic some lowmem page.
120 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
123 static inline void shmem_swp_unmap(swp_entry_t *entry)
125 kunmap_atomic(entry, KM_USER1);
128 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
130 return sb->s_fs_info;
134 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
135 * for shared memory and for shared anonymous (/dev/zero) mappings
136 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
137 * consistent with the pre-accounting of private mappings ...
139 static inline int shmem_acct_size(unsigned long flags, loff_t size)
141 return (flags & VM_ACCOUNT)?
142 security_vm_enough_memory(VM_ACCT(size)): 0;
145 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
147 if (flags & VM_ACCOUNT)
148 vm_unacct_memory(VM_ACCT(size));
152 * ... whereas tmpfs objects are accounted incrementally as
153 * pages are allocated, in order to allow huge sparse files.
154 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
155 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
157 static inline int shmem_acct_block(unsigned long flags)
159 return (flags & VM_ACCOUNT)?
160 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
163 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
165 if (!(flags & VM_ACCOUNT))
166 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
169 static struct super_operations shmem_ops;
170 static struct address_space_operations shmem_aops;
171 static struct file_operations shmem_file_operations;
172 static struct inode_operations shmem_inode_operations;
173 static struct inode_operations shmem_dir_inode_operations;
174 static struct vm_operations_struct shmem_vm_ops;
176 static struct backing_dev_info shmem_backing_dev_info = {
177 .ra_pages = 0, /* No readahead */
178 .memory_backed = 1, /* Does not contribute to dirty memory */
179 .unplug_io_fn = default_unplug_io_fn,
182 static LIST_HEAD(shmem_swaplist);
183 static spinlock_t shmem_swaplist_lock = SPIN_LOCK_UNLOCKED;
185 static void shmem_free_block(struct inode *inode)
187 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
189 spin_lock(&sbinfo->stat_lock);
190 sbinfo->free_blocks++;
191 inode->i_blocks -= BLOCKS_PER_PAGE;
192 spin_unlock(&sbinfo->stat_lock);
197 * shmem_recalc_inode - recalculate the size of an inode
199 * @inode: inode to recalc
201 * We have to calculate the free blocks since the mm can drop
202 * undirtied hole pages behind our back.
204 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
205 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
207 * It has to be called with the spinlock held.
209 static void shmem_recalc_inode(struct inode *inode)
211 struct shmem_inode_info *info = SHMEM_I(inode);
214 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
216 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
217 info->alloced -= freed;
218 shmem_unacct_blocks(info->flags, freed);
220 spin_lock(&sbinfo->stat_lock);
221 sbinfo->free_blocks += freed;
222 inode->i_blocks -= freed*BLOCKS_PER_PAGE;
223 spin_unlock(&sbinfo->stat_lock);
229 * shmem_swp_entry - find the swap vector position in the info structure
231 * @info: info structure for the inode
232 * @index: index of the page to find
233 * @page: optional page to add to the structure. Has to be preset to
236 * If there is no space allocated yet it will return NULL when
237 * page is NULL, else it will use the page for the needed block,
238 * setting it to NULL on return to indicate that it has been used.
240 * The swap vector is organized the following way:
242 * There are SHMEM_NR_DIRECT entries directly stored in the
243 * shmem_inode_info structure. So small files do not need an addional
246 * For pages with index > SHMEM_NR_DIRECT there is the pointer
247 * i_indirect which points to a page which holds in the first half
248 * doubly indirect blocks, in the second half triple indirect blocks:
250 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
251 * following layout (for SHMEM_NR_DIRECT == 16):
253 * i_indirect -> dir --> 16-19
266 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
268 unsigned long offset;
272 if (index < SHMEM_NR_DIRECT) {
273 shmem_swp_balance_unmap();
274 return info->i_direct+index;
276 if (!info->i_indirect) {
278 info->i_indirect = *page;
281 return NULL; /* need another page */
284 index -= SHMEM_NR_DIRECT;
285 offset = index % ENTRIES_PER_PAGE;
286 index /= ENTRIES_PER_PAGE;
287 dir = shmem_dir_map(info->i_indirect);
289 if (index >= ENTRIES_PER_PAGE/2) {
290 index -= ENTRIES_PER_PAGE/2;
291 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
292 index %= ENTRIES_PER_PAGE;
299 shmem_dir_unmap(dir);
300 return NULL; /* need another page */
302 shmem_dir_unmap(dir);
303 dir = shmem_dir_map(subdir);
309 if (!page || !(subdir = *page)) {
310 shmem_dir_unmap(dir);
311 return NULL; /* need a page */
316 shmem_dir_unmap(dir);
317 return shmem_swp_map(subdir) + offset;
320 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
322 long incdec = value? 1: -1;
325 info->swapped += incdec;
326 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
327 kmap_atomic_to_page(entry)->nr_swapped += incdec;
331 * shmem_swp_alloc - get the position of the swap entry for the page.
332 * If it does not exist allocate the entry.
334 * @info: info structure for the inode
335 * @index: index of the page to find
336 * @sgp: check and recheck i_size? skip allocation?
338 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
340 struct inode *inode = &info->vfs_inode;
341 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
342 struct page *page = NULL;
345 if (sgp != SGP_WRITE &&
346 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
347 return ERR_PTR(-EINVAL);
349 while (!(entry = shmem_swp_entry(info, index, &page))) {
351 return shmem_swp_map(ZERO_PAGE(0));
353 * Test free_blocks against 1 not 0, since we have 1 data
354 * page (and perhaps indirect index pages) yet to allocate:
355 * a waste to allocate index if we cannot allocate data.
358 spin_lock(&sbinfo->stat_lock);
359 if (sbinfo->free_blocks <= 1) {
360 spin_unlock(&sbinfo->stat_lock);
361 return ERR_PTR(-ENOSPC);
363 sbinfo->free_blocks--;
364 inode->i_blocks += BLOCKS_PER_PAGE;
365 spin_unlock(&sbinfo->stat_lock);
368 spin_unlock(&info->lock);
369 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
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 shmem_unacct_size(info->flags, inode->i_size);
609 shmem_truncate(inode);
610 if (!list_empty(&info->swaplist)) {
611 spin_lock(&shmem_swaplist_lock);
612 list_del_init(&info->swaplist);
613 spin_unlock(&shmem_swaplist_lock);
617 BUG_ON(inode->i_blocks);
618 spin_lock(&sbinfo->stat_lock);
619 sbinfo->free_inodes++;
620 spin_unlock(&sbinfo->stat_lock);
625 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
629 for (ptr = dir; ptr < edir; ptr++) {
630 if (ptr->val == entry.val)
636 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
649 ptr = info->i_direct;
650 spin_lock(&info->lock);
651 limit = info->next_index;
653 if (size > SHMEM_NR_DIRECT)
654 size = SHMEM_NR_DIRECT;
655 offset = shmem_find_swp(entry, ptr, ptr+size);
657 shmem_swp_balance_unmap();
660 if (!info->i_indirect)
662 /* we might be racing with shmem_truncate */
663 if (limit <= SHMEM_NR_DIRECT)
666 dir = shmem_dir_map(info->i_indirect);
667 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
669 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
670 if (unlikely(idx == stage)) {
671 shmem_dir_unmap(dir-1);
672 dir = shmem_dir_map(info->i_indirect) +
673 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
676 idx += ENTRIES_PER_PAGEPAGE;
680 stage = idx + ENTRIES_PER_PAGEPAGE;
682 shmem_dir_unmap(dir);
683 dir = shmem_dir_map(subdir);
686 if (subdir && subdir->nr_swapped) {
687 ptr = shmem_swp_map(subdir);
689 if (size > ENTRIES_PER_PAGE)
690 size = ENTRIES_PER_PAGE;
691 offset = shmem_find_swp(entry, ptr, ptr+size);
693 shmem_dir_unmap(dir);
696 shmem_swp_unmap(ptr);
700 shmem_dir_unmap(dir-1);
702 spin_unlock(&info->lock);
706 inode = &info->vfs_inode;
707 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
708 info->flags |= SHMEM_PAGEIN;
709 shmem_swp_set(info, ptr + offset, 0);
711 shmem_swp_unmap(ptr);
712 spin_unlock(&info->lock);
714 * Decrement swap count even when the entry is left behind:
715 * try_to_unuse will skip over mms, then reincrement count.
722 * shmem_unuse() search for an eventually swapped out shmem page.
724 int shmem_unuse(swp_entry_t entry, struct page *page)
726 struct list_head *p, *next;
727 struct shmem_inode_info *info;
730 spin_lock(&shmem_swaplist_lock);
731 list_for_each_safe(p, next, &shmem_swaplist) {
732 info = list_entry(p, struct shmem_inode_info, swaplist);
734 list_del_init(&info->swaplist);
735 else if (shmem_unuse_inode(info, entry, page)) {
736 /* move head to start search for next from here */
737 list_move_tail(&shmem_swaplist, &info->swaplist);
742 spin_unlock(&shmem_swaplist_lock);
747 * Move the page from the page cache to the swap cache.
749 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
751 struct shmem_inode_info *info;
752 swp_entry_t *entry, swap;
753 struct address_space *mapping;
757 BUG_ON(!PageLocked(page));
758 BUG_ON(page_mapped(page));
760 mapping = page->mapping;
762 inode = mapping->host;
763 info = SHMEM_I(inode);
764 if (info->flags & VM_LOCKED)
766 swap = get_swap_page();
770 spin_lock(&info->lock);
771 shmem_recalc_inode(inode);
772 if (index >= info->next_index) {
773 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
776 entry = shmem_swp_entry(info, index, NULL);
780 if (move_to_swap_cache(page, swap) == 0) {
781 shmem_swp_set(info, entry, swap.val);
782 shmem_swp_unmap(entry);
783 spin_unlock(&info->lock);
784 if (list_empty(&info->swaplist)) {
785 spin_lock(&shmem_swaplist_lock);
786 /* move instead of add in case we're racing */
787 list_move_tail(&info->swaplist, &shmem_swaplist);
788 spin_unlock(&shmem_swaplist_lock);
794 shmem_swp_unmap(entry);
796 spin_unlock(&info->lock);
799 set_page_dirty(page);
800 return WRITEPAGE_ACTIVATE; /* Return with the page locked */
804 static struct page *shmem_swapin_async(struct shared_policy *p,
805 swp_entry_t entry, unsigned long idx)
808 struct vm_area_struct pvma;
810 /* Create a pseudo vma that just contains the policy */
811 memset(&pvma, 0, sizeof(struct vm_area_struct));
812 pvma.vm_end = PAGE_SIZE;
814 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
815 page = read_swap_cache_async(entry, &pvma, 0);
816 mpol_free(pvma.vm_policy);
820 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
823 struct shared_policy *p = &info->policy;
826 unsigned long offset;
828 num = valid_swaphandles(entry, &offset);
829 for (i = 0; i < num; offset++, i++) {
830 page = shmem_swapin_async(p,
831 swp_entry(swp_type(entry), offset), idx);
834 page_cache_release(page);
836 lru_add_drain(); /* Push any new pages onto the LRU now */
837 return shmem_swapin_async(p, entry, idx);
841 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
844 struct vm_area_struct pvma;
847 memset(&pvma, 0, sizeof(struct vm_area_struct));
848 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
850 pvma.vm_end = PAGE_SIZE;
851 page = alloc_page_vma(gfp, &pvma, 0);
852 mpol_free(pvma.vm_policy);
856 static inline struct page *
857 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
859 swapin_readahead(entry, 0, NULL);
860 return read_swap_cache_async(entry, NULL, 0);
863 static inline struct page *
864 shmem_alloc_page(unsigned long gfp,struct shmem_inode_info *info,
867 return alloc_page(gfp);
872 * shmem_getpage - either get the page from swap or allocate a new one
874 * If we allocate a new one we do not mark it dirty. That's up to the
875 * vm. If we swap it in we mark it dirty since we also free the swap
876 * entry since a page cannot live in both the swap and page cache
878 static int shmem_getpage(struct inode *inode, unsigned long idx,
879 struct page **pagep, enum sgp_type sgp, int *type)
881 struct address_space *mapping = inode->i_mapping;
882 struct shmem_inode_info *info = SHMEM_I(inode);
883 struct shmem_sb_info *sbinfo;
884 struct page *filepage = *pagep;
885 struct page *swappage;
890 if (idx >= SHMEM_MAX_INDEX)
893 * Normally, filepage is NULL on entry, and either found
894 * uptodate immediately, or allocated and zeroed, or read
895 * in under swappage, which is then assigned to filepage.
896 * But shmem_prepare_write passes in a locked filepage,
897 * which may be found not uptodate by other callers too,
898 * and may need to be copied from the swappage read in.
902 filepage = find_lock_page(mapping, idx);
903 if (filepage && PageUptodate(filepage))
906 if (sgp == SGP_QUICK)
909 spin_lock(&info->lock);
910 shmem_recalc_inode(inode);
911 entry = shmem_swp_alloc(info, idx, sgp);
913 spin_unlock(&info->lock);
914 error = PTR_ERR(entry);
920 /* Look it up and read it in.. */
921 swappage = lookup_swap_cache(swap);
923 shmem_swp_unmap(entry);
924 spin_unlock(&info->lock);
925 /* here we actually do the io */
926 if (type && *type == VM_FAULT_MINOR) {
927 inc_page_state(pgmajfault);
928 *type = VM_FAULT_MAJOR;
930 swappage = shmem_swapin(info, swap, idx);
932 spin_lock(&info->lock);
933 entry = shmem_swp_alloc(info, idx, sgp);
935 error = PTR_ERR(entry);
937 if (entry->val == swap.val)
939 shmem_swp_unmap(entry);
941 spin_unlock(&info->lock);
946 wait_on_page_locked(swappage);
947 page_cache_release(swappage);
951 /* We have to do this with page locked to prevent races */
952 if (TestSetPageLocked(swappage)) {
953 shmem_swp_unmap(entry);
954 spin_unlock(&info->lock);
955 wait_on_page_locked(swappage);
956 page_cache_release(swappage);
959 if (PageWriteback(swappage)) {
960 shmem_swp_unmap(entry);
961 spin_unlock(&info->lock);
962 wait_on_page_writeback(swappage);
963 unlock_page(swappage);
964 page_cache_release(swappage);
967 if (!PageUptodate(swappage)) {
968 shmem_swp_unmap(entry);
969 spin_unlock(&info->lock);
970 unlock_page(swappage);
971 page_cache_release(swappage);
977 shmem_swp_set(info, entry, 0);
978 shmem_swp_unmap(entry);
979 delete_from_swap_cache(swappage);
980 spin_unlock(&info->lock);
981 copy_highpage(filepage, swappage);
982 unlock_page(swappage);
983 page_cache_release(swappage);
984 flush_dcache_page(filepage);
985 SetPageUptodate(filepage);
986 set_page_dirty(filepage);
988 } else if (!(error = move_from_swap_cache(
989 swappage, idx, mapping))) {
990 info->flags |= SHMEM_PAGEIN;
991 shmem_swp_set(info, entry, 0);
992 shmem_swp_unmap(entry);
993 spin_unlock(&info->lock);
997 shmem_swp_unmap(entry);
998 spin_unlock(&info->lock);
999 unlock_page(swappage);
1000 page_cache_release(swappage);
1001 if (error == -ENOMEM) {
1002 /* let kswapd refresh zone for GFP_ATOMICs */
1003 blk_congestion_wait(WRITE, HZ/50);
1007 } else if (sgp == SGP_READ && !filepage) {
1008 shmem_swp_unmap(entry);
1009 filepage = find_get_page(mapping, idx);
1011 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1012 spin_unlock(&info->lock);
1013 wait_on_page_locked(filepage);
1014 page_cache_release(filepage);
1018 spin_unlock(&info->lock);
1020 shmem_swp_unmap(entry);
1021 sbinfo = SHMEM_SB(inode->i_sb);
1023 spin_lock(&sbinfo->stat_lock);
1024 if (sbinfo->free_blocks == 0 ||
1025 shmem_acct_block(info->flags)) {
1026 spin_unlock(&sbinfo->stat_lock);
1027 spin_unlock(&info->lock);
1031 sbinfo->free_blocks--;
1032 inode->i_blocks += BLOCKS_PER_PAGE;
1033 spin_unlock(&sbinfo->stat_lock);
1034 } else if (shmem_acct_block(info->flags)) {
1035 spin_unlock(&info->lock);
1041 spin_unlock(&info->lock);
1042 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1046 shmem_unacct_blocks(info->flags, 1);
1047 shmem_free_block(inode);
1052 spin_lock(&info->lock);
1053 entry = shmem_swp_alloc(info, idx, sgp);
1055 error = PTR_ERR(entry);
1058 shmem_swp_unmap(entry);
1060 if (error || swap.val || 0 != add_to_page_cache_lru(
1061 filepage, mapping, idx, GFP_ATOMIC)) {
1062 spin_unlock(&info->lock);
1063 page_cache_release(filepage);
1064 shmem_unacct_blocks(info->flags, 1);
1065 shmem_free_block(inode);
1071 info->flags |= SHMEM_PAGEIN;
1075 spin_unlock(&info->lock);
1076 clear_highpage(filepage);
1077 flush_dcache_page(filepage);
1078 SetPageUptodate(filepage);
1081 if (*pagep != filepage) {
1082 unlock_page(filepage);
1088 if (*pagep != filepage) {
1089 unlock_page(filepage);
1090 page_cache_release(filepage);
1095 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1097 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1098 struct page *page = NULL;
1102 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1103 idx += vma->vm_pgoff;
1104 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1106 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1108 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1110 mark_page_accessed(page);
1114 static int shmem_populate(struct vm_area_struct *vma,
1115 unsigned long addr, unsigned long len,
1116 pgprot_t prot, unsigned long pgoff, int nonblock)
1118 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1119 struct mm_struct *mm = vma->vm_mm;
1120 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1123 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1124 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1127 while ((long) len > 0) {
1128 struct page *page = NULL;
1131 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1133 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1137 mark_page_accessed(page);
1138 err = install_page(mm, vma, addr, page, prot);
1140 page_cache_release(page);
1143 } else if (nonblock) {
1144 err = install_file_pte(mm, vma, addr, pgoff, prot);
1157 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1159 struct inode *i = vma->vm_file->f_dentry->d_inode;
1160 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1164 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1166 struct inode *i = vma->vm_file->f_dentry->d_inode;
1169 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1170 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1174 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1176 struct inode *inode = file->f_dentry->d_inode;
1177 struct shmem_inode_info *info = SHMEM_I(inode);
1178 int retval = -ENOMEM;
1180 spin_lock(&info->lock);
1181 if (lock && !(info->flags & VM_LOCKED)) {
1182 if (!user_shm_lock(inode->i_size, user))
1184 info->flags |= VM_LOCKED;
1186 if (!lock && (info->flags & VM_LOCKED) && user) {
1187 user_shm_unlock(inode->i_size, user);
1188 info->flags &= ~VM_LOCKED;
1192 spin_unlock(&info->lock);
1196 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1198 file_accessed(file);
1199 vma->vm_ops = &shmem_vm_ops;
1203 static struct inode *
1204 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1206 struct inode *inode;
1207 struct shmem_inode_info *info;
1208 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1211 spin_lock(&sbinfo->stat_lock);
1212 if (!sbinfo->free_inodes) {
1213 spin_unlock(&sbinfo->stat_lock);
1216 sbinfo->free_inodes--;
1217 spin_unlock(&sbinfo->stat_lock);
1220 inode = new_inode(sb);
1222 inode->i_mode = mode;
1223 inode->i_uid = current->fsuid;
1224 inode->i_gid = current->fsgid;
1225 inode->i_blksize = PAGE_CACHE_SIZE;
1226 inode->i_blocks = 0;
1227 inode->i_mapping->a_ops = &shmem_aops;
1228 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1229 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1230 info = SHMEM_I(inode);
1231 memset(info, 0, (char *)inode - (char *)info);
1232 spin_lock_init(&info->lock);
1233 mpol_shared_policy_init(&info->policy);
1234 INIT_LIST_HEAD(&info->swaplist);
1236 switch (mode & S_IFMT) {
1238 init_special_inode(inode, mode, dev);
1241 inode->i_op = &shmem_inode_operations;
1242 inode->i_fop = &shmem_file_operations;
1246 /* Some things misbehave if size == 0 on a directory */
1247 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1248 inode->i_op = &shmem_dir_inode_operations;
1249 inode->i_fop = &simple_dir_operations;
1260 static int shmem_set_size(struct shmem_sb_info *sbinfo,
1261 unsigned long max_blocks, unsigned long max_inodes)
1264 unsigned long blocks, inodes;
1266 spin_lock(&sbinfo->stat_lock);
1267 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
1268 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1270 if (max_blocks < blocks)
1272 if (max_inodes < inodes)
1275 sbinfo->max_blocks = max_blocks;
1276 sbinfo->free_blocks = max_blocks - blocks;
1277 sbinfo->max_inodes = max_inodes;
1278 sbinfo->free_inodes = max_inodes - inodes;
1280 spin_unlock(&sbinfo->stat_lock);
1284 static struct inode_operations shmem_symlink_inode_operations;
1285 static struct inode_operations shmem_symlink_inline_operations;
1288 * Normally tmpfs makes no use of shmem_prepare_write, but it
1289 * lets a tmpfs file be used read-write below the loop driver.
1292 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1294 struct inode *inode = page->mapping->host;
1295 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1299 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1301 struct inode *inode = file->f_dentry->d_inode;
1303 unsigned long written;
1306 if ((ssize_t) count < 0)
1309 if (!access_ok(VERIFY_READ, buf, count))
1312 down(&inode->i_sem);
1317 err = generic_write_checks(file, &pos, &count, 0);
1321 err = remove_suid(file->f_dentry);
1325 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1328 struct page *page = NULL;
1329 unsigned long bytes, index, offset;
1333 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1334 index = pos >> PAGE_CACHE_SHIFT;
1335 bytes = PAGE_CACHE_SIZE - offset;
1340 * We don't hold page lock across copy from user -
1341 * what would it guard against? - so no deadlock here.
1342 * But it still may be a good idea to prefault below.
1345 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1350 if (PageHighMem(page)) {
1351 volatile unsigned char dummy;
1352 __get_user(dummy, buf);
1353 __get_user(dummy, buf + bytes - 1);
1355 kaddr = kmap_atomic(page, KM_USER0);
1356 left = __copy_from_user_inatomic(kaddr + offset,
1358 kunmap_atomic(kaddr, KM_USER0);
1362 left = __copy_from_user(kaddr + offset, buf, bytes);
1370 if (pos > inode->i_size)
1371 i_size_write(inode, pos);
1373 flush_dcache_page(page);
1374 set_page_dirty(page);
1375 mark_page_accessed(page);
1376 page_cache_release(page);
1386 * Our dirty pages are not counted in nr_dirty,
1387 * and we do not attempt to balance dirty pages.
1401 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1403 struct inode *inode = filp->f_dentry->d_inode;
1404 struct address_space *mapping = inode->i_mapping;
1405 unsigned long index, offset;
1407 index = *ppos >> PAGE_CACHE_SHIFT;
1408 offset = *ppos & ~PAGE_CACHE_MASK;
1411 struct page *page = NULL;
1412 unsigned long end_index, nr, ret;
1413 loff_t i_size = i_size_read(inode);
1415 end_index = i_size >> PAGE_CACHE_SHIFT;
1416 if (index > end_index)
1418 if (index == end_index) {
1419 nr = i_size & ~PAGE_CACHE_MASK;
1424 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1426 if (desc->error == -EINVAL)
1432 * We must evaluate after, since reads (unlike writes)
1433 * are called without i_sem protection against truncate
1435 nr = PAGE_CACHE_SIZE;
1436 i_size = i_size_read(inode);
1437 end_index = i_size >> PAGE_CACHE_SHIFT;
1438 if (index == end_index) {
1439 nr = i_size & ~PAGE_CACHE_MASK;
1442 page_cache_release(page);
1450 * If users can be writing to this page using arbitrary
1451 * virtual addresses, take care about potential aliasing
1452 * before reading the page on the kernel side.
1454 if (mapping_writably_mapped(mapping))
1455 flush_dcache_page(page);
1457 * Mark the page accessed if we read the beginning.
1460 mark_page_accessed(page);
1462 page = ZERO_PAGE(0);
1465 * Ok, we have the page, and it's up-to-date, so
1466 * now we can copy it to user space...
1468 * The actor routine returns how many bytes were actually used..
1469 * NOTE! This may not be the same as how much of a user buffer
1470 * we filled up (we may be padding etc), so we can only update
1471 * "pos" here (the actor routine has to update the user buffer
1472 * pointers and the remaining count).
1474 ret = actor(desc, page, offset, nr);
1476 index += offset >> PAGE_CACHE_SHIFT;
1477 offset &= ~PAGE_CACHE_MASK;
1479 page_cache_release(page);
1480 if (ret != nr || !desc->count)
1486 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1487 file_accessed(filp);
1490 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1492 read_descriptor_t desc;
1494 if ((ssize_t) count < 0)
1496 if (!access_ok(VERIFY_WRITE, buf, count))
1506 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1508 return desc.written;
1512 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1513 size_t count, read_actor_t actor, void *target)
1515 read_descriptor_t desc;
1522 desc.arg.data = target;
1525 do_shmem_file_read(in_file, ppos, &desc, actor);
1527 return desc.written;
1531 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1533 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1535 buf->f_type = TMPFS_MAGIC;
1536 buf->f_bsize = PAGE_CACHE_SIZE;
1537 buf->f_namelen = NAME_MAX;
1539 spin_lock(&sbinfo->stat_lock);
1540 buf->f_blocks = sbinfo->max_blocks;
1541 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1542 buf->f_files = sbinfo->max_inodes;
1543 buf->f_ffree = sbinfo->free_inodes;
1544 spin_unlock(&sbinfo->stat_lock);
1546 /* else leave those fields 0 like simple_statfs */
1551 * File creation. Allocate an inode, and we're done..
1554 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1556 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1557 int error = -ENOSPC;
1560 if (dir->i_mode & S_ISGID) {
1561 inode->i_gid = dir->i_gid;
1563 inode->i_mode |= S_ISGID;
1565 dir->i_size += BOGO_DIRENT_SIZE;
1566 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1567 d_instantiate(dentry, inode);
1568 dget(dentry); /* Extra count - pin the dentry in core */
1574 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1578 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1584 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1585 struct nameidata *nd)
1587 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1593 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1595 struct inode *inode = old_dentry->d_inode;
1596 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1599 * No ordinary (disk based) filesystem counts links as inodes;
1600 * but each new link needs a new dentry, pinning lowmem, and
1601 * tmpfs dentries cannot be pruned until they are unlinked.
1604 spin_lock(&sbinfo->stat_lock);
1605 if (!sbinfo->free_inodes) {
1606 spin_unlock(&sbinfo->stat_lock);
1609 sbinfo->free_inodes--;
1610 spin_unlock(&sbinfo->stat_lock);
1613 dir->i_size += BOGO_DIRENT_SIZE;
1614 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1616 atomic_inc(&inode->i_count); /* New dentry reference */
1617 dget(dentry); /* Extra pinning count for the created dentry */
1618 d_instantiate(dentry, inode);
1622 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1624 struct inode *inode = dentry->d_inode;
1626 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1627 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1629 spin_lock(&sbinfo->stat_lock);
1630 sbinfo->free_inodes++;
1631 spin_unlock(&sbinfo->stat_lock);
1635 dir->i_size -= BOGO_DIRENT_SIZE;
1636 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1638 dput(dentry); /* Undo the count from "create" - this does all the work */
1642 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1644 if (!simple_empty(dentry))
1648 return shmem_unlink(dir, dentry);
1652 * The VFS layer already does all the dentry stuff for rename,
1653 * we just have to decrement the usage count for the target if
1654 * it exists so that the VFS layer correctly free's it when it
1657 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1659 struct inode *inode = old_dentry->d_inode;
1660 int they_are_dirs = S_ISDIR(inode->i_mode);
1662 if (!simple_empty(new_dentry))
1665 if (new_dentry->d_inode) {
1666 (void) shmem_unlink(new_dir, new_dentry);
1669 } else if (they_are_dirs) {
1674 old_dir->i_size -= BOGO_DIRENT_SIZE;
1675 new_dir->i_size += BOGO_DIRENT_SIZE;
1676 old_dir->i_ctime = old_dir->i_mtime =
1677 new_dir->i_ctime = new_dir->i_mtime =
1678 inode->i_ctime = CURRENT_TIME;
1682 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1686 struct inode *inode;
1687 struct page *page = NULL;
1689 struct shmem_inode_info *info;
1691 len = strlen(symname) + 1;
1692 if (len > PAGE_CACHE_SIZE)
1693 return -ENAMETOOLONG;
1695 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1699 info = SHMEM_I(inode);
1700 inode->i_size = len-1;
1701 if (len <= (char *)inode - (char *)info) {
1703 memcpy(info, symname, len);
1704 inode->i_op = &shmem_symlink_inline_operations;
1706 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1711 inode->i_op = &shmem_symlink_inode_operations;
1712 kaddr = kmap_atomic(page, KM_USER0);
1713 memcpy(kaddr, symname, len);
1714 kunmap_atomic(kaddr, KM_USER0);
1715 set_page_dirty(page);
1716 page_cache_release(page);
1718 if (dir->i_mode & S_ISGID)
1719 inode->i_gid = dir->i_gid;
1720 dir->i_size += BOGO_DIRENT_SIZE;
1721 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1722 d_instantiate(dentry, inode);
1727 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1729 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1733 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1735 struct page *page = NULL;
1736 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1737 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1741 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1743 if (!IS_ERR(nd_get_link(nd))) {
1746 page = find_get_page(dentry->d_inode->i_mapping, 0);
1750 mark_page_accessed(page);
1751 page_cache_release(page);
1752 page_cache_release(page);
1756 static struct inode_operations shmem_symlink_inline_operations = {
1757 .readlink = generic_readlink,
1758 .follow_link = shmem_follow_link_inline,
1761 static struct inode_operations shmem_symlink_inode_operations = {
1762 .truncate = shmem_truncate,
1763 .readlink = generic_readlink,
1764 .follow_link = shmem_follow_link,
1765 .put_link = shmem_put_link,
1768 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1770 char *this_char, *value, *rest;
1772 while ((this_char = strsep(&options, ",")) != NULL) {
1775 if ((value = strchr(this_char,'=')) != NULL) {
1779 "tmpfs: No value for mount option '%s'\n",
1784 if (!strcmp(this_char,"size")) {
1785 unsigned long long size;
1786 size = memparse(value,&rest);
1788 size <<= PAGE_SHIFT;
1789 size *= totalram_pages;
1795 *blocks = size >> PAGE_CACHE_SHIFT;
1796 } else if (!strcmp(this_char,"nr_blocks")) {
1797 *blocks = memparse(value,&rest);
1800 } else if (!strcmp(this_char,"nr_inodes")) {
1801 *inodes = memparse(value,&rest);
1804 } else if (!strcmp(this_char,"mode")) {
1807 *mode = simple_strtoul(value,&rest,8);
1810 } else if (!strcmp(this_char,"uid")) {
1813 *uid = simple_strtoul(value,&rest,0);
1816 } else if (!strcmp(this_char,"gid")) {
1819 *gid = simple_strtoul(value,&rest,0);
1823 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1831 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1837 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1839 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1840 unsigned long max_blocks = 0;
1841 unsigned long max_inodes = 0;
1844 max_blocks = sbinfo->max_blocks;
1845 max_inodes = sbinfo->max_inodes;
1847 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
1849 /* Keep it simple: disallow limited <-> unlimited remount */
1850 if ((max_blocks || max_inodes) == !sbinfo)
1852 /* But allow the pointless unlimited -> unlimited remount */
1855 return shmem_set_size(sbinfo, max_blocks, max_inodes);
1859 static void shmem_put_super(struct super_block *sb)
1861 kfree(sb->s_fs_info);
1862 sb->s_fs_info = NULL;
1865 static int shmem_fill_super(struct super_block *sb,
1866 void *data, int silent)
1868 struct inode *inode;
1869 struct dentry *root;
1870 int mode = S_IRWXUGO | S_ISVTX;
1871 uid_t uid = current->fsuid;
1872 gid_t gid = current->fsgid;
1876 unsigned long blocks = 0;
1877 unsigned long inodes = 0;
1880 * Per default we only allow half of the physical ram per
1881 * tmpfs instance, limiting inodes to one per page of lowmem;
1882 * but the internal instance is left unlimited.
1884 if (!(sb->s_flags & MS_NOUSER)) {
1885 blocks = totalram_pages / 2;
1886 inodes = totalram_pages - totalhigh_pages;
1887 if (inodes > blocks)
1890 if (shmem_parse_options(data, &mode,
1891 &uid, &gid, &blocks, &inodes))
1895 if (blocks || inodes) {
1896 struct shmem_sb_info *sbinfo;
1897 sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
1900 sb->s_fs_info = sbinfo;
1901 spin_lock_init(&sbinfo->stat_lock);
1902 sbinfo->max_blocks = blocks;
1903 sbinfo->free_blocks = blocks;
1904 sbinfo->max_inodes = inodes;
1905 sbinfo->free_inodes = inodes;
1909 sb->s_maxbytes = SHMEM_MAX_BYTES;
1910 sb->s_blocksize = PAGE_CACHE_SIZE;
1911 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1912 sb->s_magic = TMPFS_MAGIC;
1913 sb->s_op = &shmem_ops;
1914 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1919 root = d_alloc_root(inode);
1928 shmem_put_super(sb);
1932 static kmem_cache_t *shmem_inode_cachep;
1934 static struct inode *shmem_alloc_inode(struct super_block *sb)
1936 struct shmem_inode_info *p;
1937 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
1940 return &p->vfs_inode;
1943 static void shmem_destroy_inode(struct inode *inode)
1945 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
1946 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
1949 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
1951 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
1953 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1954 SLAB_CTOR_CONSTRUCTOR) {
1955 inode_init_once(&p->vfs_inode);
1959 static int init_inodecache(void)
1961 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
1962 sizeof(struct shmem_inode_info),
1963 0, 0, init_once, NULL);
1964 if (shmem_inode_cachep == NULL)
1969 static void destroy_inodecache(void)
1971 if (kmem_cache_destroy(shmem_inode_cachep))
1972 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
1975 static struct address_space_operations shmem_aops = {
1976 .writepage = shmem_writepage,
1977 .set_page_dirty = __set_page_dirty_nobuffers,
1979 .prepare_write = shmem_prepare_write,
1980 .commit_write = simple_commit_write,
1984 static struct file_operations shmem_file_operations = {
1987 .llseek = generic_file_llseek,
1988 .read = shmem_file_read,
1989 .write = shmem_file_write,
1990 .fsync = simple_sync_file,
1991 .sendfile = shmem_file_sendfile,
1995 static struct inode_operations shmem_inode_operations = {
1996 .truncate = shmem_truncate,
1997 .setattr = shmem_notify_change,
2000 static struct inode_operations shmem_dir_inode_operations = {
2002 .create = shmem_create,
2003 .lookup = simple_lookup,
2005 .unlink = shmem_unlink,
2006 .symlink = shmem_symlink,
2007 .mkdir = shmem_mkdir,
2008 .rmdir = shmem_rmdir,
2009 .mknod = shmem_mknod,
2010 .rename = shmem_rename,
2014 static struct super_operations shmem_ops = {
2015 .alloc_inode = shmem_alloc_inode,
2016 .destroy_inode = shmem_destroy_inode,
2018 .statfs = shmem_statfs,
2019 .remount_fs = shmem_remount_fs,
2021 .delete_inode = shmem_delete_inode,
2022 .drop_inode = generic_delete_inode,
2023 .put_super = shmem_put_super,
2026 static struct vm_operations_struct shmem_vm_ops = {
2027 .nopage = shmem_nopage,
2028 .populate = shmem_populate,
2030 .set_policy = shmem_set_policy,
2031 .get_policy = shmem_get_policy,
2035 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2036 int flags, const char *dev_name, void *data)
2038 return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2041 static struct file_system_type tmpfs_fs_type = {
2042 .owner = THIS_MODULE,
2044 .get_sb = shmem_get_sb,
2045 .kill_sb = kill_litter_super,
2047 static struct vfsmount *shm_mnt;
2049 static int __init init_tmpfs(void)
2053 error = init_inodecache();
2057 error = register_filesystem(&tmpfs_fs_type);
2059 printk(KERN_ERR "Could not register tmpfs\n");
2063 devfs_mk_dir("shm");
2065 shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2066 tmpfs_fs_type.name, NULL);
2067 if (IS_ERR(shm_mnt)) {
2068 error = PTR_ERR(shm_mnt);
2069 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2075 unregister_filesystem(&tmpfs_fs_type);
2077 destroy_inodecache();
2079 shm_mnt = ERR_PTR(error);
2082 module_init(init_tmpfs)
2085 * shmem_file_setup - get an unlinked file living in tmpfs
2087 * @name: name for dentry (to be seen in /proc/<pid>/maps
2088 * @size: size to be set for the file
2091 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2095 struct inode *inode;
2096 struct dentry *dentry, *root;
2099 if (IS_ERR(shm_mnt))
2100 return (void *)shm_mnt;
2102 if (size < 0 || size > SHMEM_MAX_BYTES)
2103 return ERR_PTR(-EINVAL);
2105 if (shmem_acct_size(flags, size))
2106 return ERR_PTR(-ENOMEM);
2110 this.len = strlen(name);
2111 this.hash = 0; /* will go */
2112 root = shm_mnt->mnt_root;
2113 dentry = d_alloc(root, &this);
2118 file = get_empty_filp();
2123 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2127 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2128 d_instantiate(dentry, inode);
2129 inode->i_size = size;
2130 inode->i_nlink = 0; /* It is unlinked */
2131 file->f_vfsmnt = mntget(shm_mnt);
2132 file->f_dentry = dentry;
2133 file->f_mapping = inode->i_mapping;
2134 file->f_op = &shmem_file_operations;
2135 file->f_mode = FMODE_WRITE | FMODE_READ;
2143 shmem_unacct_size(flags, size);
2144 return ERR_PTR(error);
2148 * shmem_zero_setup - setup a shared anonymous mapping
2150 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2152 int shmem_zero_setup(struct vm_area_struct *vma)
2155 loff_t size = vma->vm_end - vma->vm_start;
2157 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2159 return PTR_ERR(file);
2163 vma->vm_file = file;
2164 vma->vm_ops = &shmem_vm_ops;
2168 EXPORT_SYMBOL(shmem_file_setup);