2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2003 Hugh Dickins.
10 * Copyright (C) 2002-2003 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * This file is released under the GPL.
17 * This virtual memory filesystem is heavily based on the ramfs. It
18 * extends ramfs by the ability to use swap and honor resource limits
19 * which makes it a completely usable filesystem.
22 #include <linux/config.h>
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/devfs_fs_kernel.h>
28 #include <linux/mman.h>
29 #include <linux/file.h>
30 #include <linux/swap.h>
31 #include <linux/pagemap.h>
32 #include <linux/string.h>
33 #include <linux/slab.h>
34 #include <linux/backing-dev.h>
35 #include <linux/shmem_fs.h>
36 #include <linux/mount.h>
37 #include <linux/writeback.h>
38 #include <linux/vfs.h>
39 #include <linux/blkdev.h>
40 #include <linux/security.h>
41 #include <linux/swapops.h>
42 #include <linux/mempolicy.h>
43 #include <linux/namei.h>
44 #include <asm/uaccess.h>
45 #include <asm/div64.h>
46 #include <asm/pgtable.h>
48 /* This magic number is used in glibc for posix shared memory */
49 #define TMPFS_MAGIC 0x01021994
51 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
52 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
53 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
55 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
56 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
58 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
60 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
61 #define SHMEM_PAGEIN VM_READ
62 #define SHMEM_TRUNCATE VM_WRITE
64 /* Pretend that each entry is of this size in directory's i_size */
65 #define BOGO_DIRENT_SIZE 20
67 /* Keep swapped page count in private field of indirect struct page */
68 #define nr_swapped private
70 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
72 SGP_QUICK, /* don't try more than file page cache lookup */
73 SGP_READ, /* don't exceed i_size, don't allocate page */
74 SGP_CACHE, /* don't exceed i_size, may allocate page */
75 SGP_WRITE, /* may exceed i_size, may allocate page */
78 static int shmem_getpage(struct inode *inode, unsigned long idx,
79 struct page **pagep, enum sgp_type sgp, int *type);
81 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
84 * The above definition of ENTRIES_PER_PAGE, and the use of
85 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
86 * might be reconsidered if it ever diverges from PAGE_SIZE.
88 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
91 static inline void shmem_dir_free(struct page *page)
93 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
96 static struct page **shmem_dir_map(struct page *page)
98 return (struct page **)kmap_atomic(page, KM_USER0);
101 static inline void shmem_dir_unmap(struct page **dir)
103 kunmap_atomic(dir, KM_USER0);
106 static swp_entry_t *shmem_swp_map(struct page *page)
108 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
111 static inline void shmem_swp_balance_unmap(void)
114 * When passing a pointer to an i_direct entry, to code which
115 * also handles indirect entries and so will shmem_swp_unmap,
116 * we must arrange for the preempt count to remain in balance.
117 * What kmap_atomic of a lowmem page does depends on config
118 * and architecture, so pretend to kmap_atomic some lowmem page.
120 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
123 static inline void shmem_swp_unmap(swp_entry_t *entry)
125 kunmap_atomic(entry, KM_USER1);
128 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
130 return sb->s_fs_info;
134 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
135 * for shared memory and for shared anonymous (/dev/zero) mappings
136 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
137 * consistent with the pre-accounting of private mappings ...
139 static inline int shmem_acct_size(unsigned long flags, loff_t size)
141 return (flags & VM_ACCOUNT)?
142 security_vm_enough_memory(VM_ACCT(size)): 0;
145 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
147 if (flags & VM_ACCOUNT)
148 vm_unacct_memory(VM_ACCT(size));
152 * ... whereas tmpfs objects are accounted incrementally as
153 * pages are allocated, in order to allow huge sparse files.
154 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
155 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
157 static inline int shmem_acct_block(unsigned long flags)
159 return (flags & VM_ACCOUNT)?
160 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
163 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
165 if (!(flags & VM_ACCOUNT))
166 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
169 static struct super_operations shmem_ops;
170 static struct address_space_operations shmem_aops;
171 static struct file_operations shmem_file_operations;
172 static struct inode_operations shmem_inode_operations;
173 static struct inode_operations shmem_dir_inode_operations;
174 static struct vm_operations_struct shmem_vm_ops;
176 static struct backing_dev_info shmem_backing_dev_info = {
177 .ra_pages = 0, /* No readahead */
178 .memory_backed = 1, /* Does not contribute to dirty memory */
179 .unplug_io_fn = default_unplug_io_fn,
182 LIST_HEAD(shmem_inodes);
183 static spinlock_t shmem_ilock = SPIN_LOCK_UNLOCKED;
185 static void shmem_free_block(struct inode *inode)
187 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
188 spin_lock(&sbinfo->stat_lock);
189 sbinfo->free_blocks++;
190 inode->i_blocks -= BLOCKS_PER_PAGE;
191 spin_unlock(&sbinfo->stat_lock);
195 * shmem_recalc_inode - recalculate the size of an inode
197 * @inode: inode to recalc
199 * We have to calculate the free blocks since the mm can drop
200 * undirtied hole pages behind our back.
202 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
203 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
205 * It has to be called with the spinlock held.
207 static void shmem_recalc_inode(struct inode *inode)
209 struct shmem_inode_info *info = SHMEM_I(inode);
212 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
214 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
215 info->alloced -= freed;
216 spin_lock(&sbinfo->stat_lock);
217 sbinfo->free_blocks += freed;
218 inode->i_blocks -= freed*BLOCKS_PER_PAGE;
219 spin_unlock(&sbinfo->stat_lock);
220 shmem_unacct_blocks(info->flags, freed);
225 * shmem_swp_entry - find the swap vector position in the info structure
227 * @info: info structure for the inode
228 * @index: index of the page to find
229 * @page: optional page to add to the structure. Has to be preset to
232 * If there is no space allocated yet it will return NULL when
233 * page is NULL, else it will use the page for the needed block,
234 * setting it to NULL on return to indicate that it has been used.
236 * The swap vector is organized the following way:
238 * There are SHMEM_NR_DIRECT entries directly stored in the
239 * shmem_inode_info structure. So small files do not need an addional
242 * For pages with index > SHMEM_NR_DIRECT there is the pointer
243 * i_indirect which points to a page which holds in the first half
244 * doubly indirect blocks, in the second half triple indirect blocks:
246 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
247 * following layout (for SHMEM_NR_DIRECT == 16):
249 * i_indirect -> dir --> 16-19
262 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
264 unsigned long offset;
268 if (index < SHMEM_NR_DIRECT) {
269 shmem_swp_balance_unmap();
270 return info->i_direct+index;
272 if (!info->i_indirect) {
274 info->i_indirect = *page;
277 return NULL; /* need another page */
280 index -= SHMEM_NR_DIRECT;
281 offset = index % ENTRIES_PER_PAGE;
282 index /= ENTRIES_PER_PAGE;
283 dir = shmem_dir_map(info->i_indirect);
285 if (index >= ENTRIES_PER_PAGE/2) {
286 index -= ENTRIES_PER_PAGE/2;
287 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
288 index %= ENTRIES_PER_PAGE;
295 shmem_dir_unmap(dir);
296 return NULL; /* need another page */
298 shmem_dir_unmap(dir);
299 dir = shmem_dir_map(subdir);
305 if (!page || !(subdir = *page)) {
306 shmem_dir_unmap(dir);
307 return NULL; /* need a page */
312 shmem_dir_unmap(dir);
313 return shmem_swp_map(subdir) + offset;
316 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
318 long incdec = value? 1: -1;
321 info->swapped += incdec;
322 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
323 kmap_atomic_to_page(entry)->nr_swapped += incdec;
327 * shmem_swp_alloc - get the position of the swap entry for the page.
328 * If it does not exist allocate the entry.
330 * @info: info structure for the inode
331 * @index: index of the page to find
332 * @sgp: check and recheck i_size? skip allocation?
334 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
336 struct inode *inode = &info->vfs_inode;
337 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
338 struct page *page = NULL;
340 static const swp_entry_t unswapped = { 0 };
342 if (sgp != SGP_WRITE &&
343 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
344 return ERR_PTR(-EINVAL);
346 while (!(entry = shmem_swp_entry(info, index, &page))) {
348 return (swp_entry_t *) &unswapped;
350 * Test free_blocks against 1 not 0, since we have 1 data
351 * page (and perhaps indirect index pages) yet to allocate:
352 * a waste to allocate index if we cannot allocate data.
354 spin_lock(&sbinfo->stat_lock);
355 if (sbinfo->free_blocks <= 1) {
356 spin_unlock(&sbinfo->stat_lock);
357 return ERR_PTR(-ENOSPC);
359 sbinfo->free_blocks--;
360 inode->i_blocks += BLOCKS_PER_PAGE;
361 spin_unlock(&sbinfo->stat_lock);
363 spin_unlock(&info->lock);
364 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
366 clear_highpage(page);
367 page->nr_swapped = 0;
369 spin_lock(&info->lock);
372 shmem_free_block(inode);
373 return ERR_PTR(-ENOMEM);
375 if (sgp != SGP_WRITE &&
376 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
377 entry = ERR_PTR(-EINVAL);
380 if (info->next_index <= index)
381 info->next_index = index + 1;
384 /* another task gave its page, or truncated the file */
385 shmem_free_block(inode);
386 shmem_dir_free(page);
388 if (info->next_index <= index && !IS_ERR(entry))
389 info->next_index = index + 1;
394 * shmem_free_swp - free some swap entries in a directory
396 * @dir: pointer to the directory
397 * @edir: pointer after last entry of the directory
399 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
404 for (ptr = dir; ptr < edir; ptr++) {
406 free_swap_and_cache(*ptr);
407 *ptr = (swp_entry_t){0};
414 static void shmem_truncate(struct inode *inode)
416 struct shmem_inode_info *info = SHMEM_I(inode);
428 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
429 idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
430 if (idx >= info->next_index)
433 spin_lock(&info->lock);
434 info->flags |= SHMEM_TRUNCATE;
435 limit = info->next_index;
436 info->next_index = idx;
437 if (info->swapped && idx < SHMEM_NR_DIRECT) {
438 ptr = info->i_direct;
440 if (size > SHMEM_NR_DIRECT)
441 size = SHMEM_NR_DIRECT;
442 info->swapped -= shmem_free_swp(ptr+idx, ptr+size);
444 if (!info->i_indirect)
447 BUG_ON(limit <= SHMEM_NR_DIRECT);
448 limit -= SHMEM_NR_DIRECT;
449 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
450 offset = idx % ENTRIES_PER_PAGE;
454 dir = shmem_dir_map(info->i_indirect);
455 stage = ENTRIES_PER_PAGEPAGE/2;
456 if (idx < ENTRIES_PER_PAGEPAGE/2)
457 dir += idx/ENTRIES_PER_PAGE;
459 dir += ENTRIES_PER_PAGE/2;
460 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
462 stage += ENTRIES_PER_PAGEPAGE;
465 size = ((idx - ENTRIES_PER_PAGEPAGE/2) %
466 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
467 if (!size && !offset) {
471 shmem_dir_unmap(dir);
472 dir = shmem_dir_map(subdir) + size;
479 for (; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
480 if (unlikely(idx == stage)) {
481 shmem_dir_unmap(dir-1);
482 dir = shmem_dir_map(info->i_indirect) +
483 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
486 idx += ENTRIES_PER_PAGEPAGE;
490 stage = idx + ENTRIES_PER_PAGEPAGE;
493 shmem_dir_unmap(dir);
495 shmem_dir_free(empty);
496 shmem_free_block(inode);
499 cond_resched_lock(&info->lock);
500 dir = shmem_dir_map(subdir);
503 if (subdir && subdir->nr_swapped) {
504 ptr = shmem_swp_map(subdir);
506 if (size > ENTRIES_PER_PAGE)
507 size = ENTRIES_PER_PAGE;
508 freed = shmem_free_swp(ptr+offset, ptr+size);
509 shmem_swp_unmap(ptr);
510 info->swapped -= freed;
511 subdir->nr_swapped -= freed;
512 BUG_ON(subdir->nr_swapped > offset);
518 shmem_dir_free(subdir);
519 shmem_free_block(inode);
523 shmem_dir_unmap(dir-1);
525 shmem_dir_free(empty);
526 shmem_free_block(inode);
528 if (info->next_index <= SHMEM_NR_DIRECT) {
529 shmem_dir_free(info->i_indirect);
530 info->i_indirect = NULL;
531 shmem_free_block(inode);
534 BUG_ON(info->swapped > info->next_index);
535 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
537 * Call truncate_inode_pages again: racing shmem_unuse_inode
538 * may have swizzled a page in from swap since vmtruncate or
539 * generic_delete_inode did it, before we lowered next_index.
540 * Also, though shmem_getpage checks i_size before adding to
541 * cache, no recheck after: so fix the narrow window there too.
543 spin_unlock(&info->lock);
544 truncate_inode_pages(inode->i_mapping, inode->i_size);
545 spin_lock(&info->lock);
547 info->flags &= ~SHMEM_TRUNCATE;
548 shmem_recalc_inode(inode);
549 spin_unlock(&info->lock);
552 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
554 struct inode *inode = dentry->d_inode;
555 struct page *page = NULL;
558 if (attr->ia_valid & ATTR_SIZE) {
559 if (attr->ia_size < inode->i_size) {
561 * If truncating down to a partial page, then
562 * if that page is already allocated, hold it
563 * in memory until the truncation is over, so
564 * truncate_partial_page cannnot miss it were
565 * it assigned to swap.
567 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
568 (void) shmem_getpage(inode,
569 attr->ia_size>>PAGE_CACHE_SHIFT,
570 &page, SGP_READ, NULL);
573 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
574 * detect if any pages might have been added to cache
575 * after truncate_inode_pages. But we needn't bother
576 * if it's being fully truncated to zero-length: the
577 * nrpages check is efficient enough in that case.
580 struct shmem_inode_info *info = SHMEM_I(inode);
581 spin_lock(&info->lock);
582 info->flags &= ~SHMEM_PAGEIN;
583 spin_unlock(&info->lock);
588 error = inode_change_ok(inode, attr);
590 error = inode_setattr(inode, attr);
592 page_cache_release(page);
596 static void shmem_delete_inode(struct inode *inode)
598 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
599 struct shmem_inode_info *info = SHMEM_I(inode);
601 if (inode->i_op->truncate == shmem_truncate) {
602 spin_lock(&shmem_ilock);
603 list_del(&info->list);
604 spin_unlock(&shmem_ilock);
605 shmem_unacct_size(info->flags, inode->i_size);
607 shmem_truncate(inode);
609 BUG_ON(inode->i_blocks);
610 spin_lock(&sbinfo->stat_lock);
611 sbinfo->free_inodes++;
612 spin_unlock(&sbinfo->stat_lock);
616 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
620 for (ptr = dir; ptr < edir; ptr++) {
621 if (ptr->val == entry.val)
627 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
640 ptr = info->i_direct;
641 spin_lock(&info->lock);
642 limit = info->next_index;
644 if (size > SHMEM_NR_DIRECT)
645 size = SHMEM_NR_DIRECT;
646 offset = shmem_find_swp(entry, ptr, ptr+size);
648 shmem_swp_balance_unmap();
651 if (!info->i_indirect)
653 /* we might be racing with shmem_truncate */
654 if (limit <= SHMEM_NR_DIRECT)
657 dir = shmem_dir_map(info->i_indirect);
658 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
660 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
661 if (unlikely(idx == stage)) {
662 shmem_dir_unmap(dir-1);
663 dir = shmem_dir_map(info->i_indirect) +
664 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
667 idx += ENTRIES_PER_PAGEPAGE;
671 stage = idx + ENTRIES_PER_PAGEPAGE;
673 shmem_dir_unmap(dir);
674 dir = shmem_dir_map(subdir);
677 if (subdir && subdir->nr_swapped) {
678 ptr = shmem_swp_map(subdir);
680 if (size > ENTRIES_PER_PAGE)
681 size = ENTRIES_PER_PAGE;
682 offset = shmem_find_swp(entry, ptr, ptr+size);
684 shmem_dir_unmap(dir);
687 shmem_swp_unmap(ptr);
691 shmem_dir_unmap(dir-1);
693 spin_unlock(&info->lock);
697 inode = &info->vfs_inode;
698 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
699 info->flags |= SHMEM_PAGEIN;
700 shmem_swp_set(info, ptr + offset, 0);
702 shmem_swp_unmap(ptr);
703 spin_unlock(&info->lock);
705 * Decrement swap count even when the entry is left behind:
706 * try_to_unuse will skip over mms, then reincrement count.
713 * shmem_unuse() search for an eventually swapped out shmem page.
715 int shmem_unuse(swp_entry_t entry, struct page *page)
718 struct shmem_inode_info *info;
721 spin_lock(&shmem_ilock);
722 list_for_each(p, &shmem_inodes) {
723 info = list_entry(p, struct shmem_inode_info, list);
725 if (info->swapped && shmem_unuse_inode(info, entry, page)) {
726 /* move head to start search for next from here */
727 list_move_tail(&shmem_inodes, &info->list);
732 spin_unlock(&shmem_ilock);
737 * Move the page from the page cache to the swap cache.
739 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
741 struct shmem_inode_info *info;
742 swp_entry_t *entry, swap;
743 struct address_space *mapping;
747 BUG_ON(!PageLocked(page));
748 BUG_ON(page_mapped(page));
750 mapping = page->mapping;
752 inode = mapping->host;
753 info = SHMEM_I(inode);
754 if (info->flags & VM_LOCKED)
756 swap = get_swap_page();
760 spin_lock(&info->lock);
761 shmem_recalc_inode(inode);
762 if (index >= info->next_index) {
763 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
766 entry = shmem_swp_entry(info, index, NULL);
770 if (move_to_swap_cache(page, swap) == 0) {
771 shmem_swp_set(info, entry, swap.val);
772 shmem_swp_unmap(entry);
773 spin_unlock(&info->lock);
778 shmem_swp_unmap(entry);
780 spin_unlock(&info->lock);
783 set_page_dirty(page);
784 return WRITEPAGE_ACTIVATE; /* Return with the page locked */
788 static struct page *shmem_swapin_async(struct shared_policy *p,
789 swp_entry_t entry, unsigned long idx)
792 struct vm_area_struct pvma;
794 /* Create a pseudo vma that just contains the policy */
795 memset(&pvma, 0, sizeof(struct vm_area_struct));
796 pvma.vm_end = PAGE_SIZE;
798 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
799 page = read_swap_cache_async(entry, &pvma, 0);
800 mpol_free(pvma.vm_policy);
804 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
807 struct shared_policy *p = &info->policy;
810 unsigned long offset;
812 num = valid_swaphandles(entry, &offset);
813 for (i = 0; i < num; offset++, i++) {
814 page = shmem_swapin_async(p,
815 swp_entry(swp_type(entry), offset), idx);
818 page_cache_release(page);
820 lru_add_drain(); /* Push any new pages onto the LRU now */
821 return shmem_swapin_async(p, entry, idx);
825 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
828 struct vm_area_struct pvma;
831 memset(&pvma, 0, sizeof(struct vm_area_struct));
832 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
834 pvma.vm_end = PAGE_SIZE;
835 page = alloc_page_vma(gfp, &pvma, 0);
836 mpol_free(pvma.vm_policy);
840 static inline struct page *
841 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
843 swapin_readahead(entry, 0, NULL);
844 return read_swap_cache_async(entry, NULL, 0);
847 static inline struct page *
848 shmem_alloc_page(unsigned long gfp,struct shmem_inode_info *info,
851 return alloc_page(gfp);
856 * shmem_getpage - either get the page from swap or allocate a new one
858 * If we allocate a new one we do not mark it dirty. That's up to the
859 * vm. If we swap it in we mark it dirty since we also free the swap
860 * entry since a page cannot live in both the swap and page cache
862 static int shmem_getpage(struct inode *inode, unsigned long idx,
863 struct page **pagep, enum sgp_type sgp, int *type)
865 struct address_space *mapping = inode->i_mapping;
866 struct shmem_inode_info *info = SHMEM_I(inode);
867 struct shmem_sb_info *sbinfo;
868 struct page *filepage = *pagep;
869 struct page *swappage;
872 int error, majmin = VM_FAULT_MINOR;
874 if (idx >= SHMEM_MAX_INDEX)
877 * Normally, filepage is NULL on entry, and either found
878 * uptodate immediately, or allocated and zeroed, or read
879 * in under swappage, which is then assigned to filepage.
880 * But shmem_prepare_write passes in a locked filepage,
881 * which may be found not uptodate by other callers too,
882 * and may need to be copied from the swappage read in.
886 filepage = find_lock_page(mapping, idx);
887 if (filepage && PageUptodate(filepage))
890 if (sgp == SGP_QUICK)
893 spin_lock(&info->lock);
894 shmem_recalc_inode(inode);
895 entry = shmem_swp_alloc(info, idx, sgp);
897 spin_unlock(&info->lock);
898 error = PTR_ERR(entry);
904 /* Look it up and read it in.. */
905 swappage = lookup_swap_cache(swap);
907 shmem_swp_unmap(entry);
908 spin_unlock(&info->lock);
909 /* here we actually do the io */
910 if (majmin == VM_FAULT_MINOR && type)
911 inc_page_state(pgmajfault);
912 majmin = VM_FAULT_MAJOR;
913 swappage = shmem_swapin(info, swap, idx);
915 spin_lock(&info->lock);
916 entry = shmem_swp_alloc(info, idx, sgp);
918 error = PTR_ERR(entry);
920 if (entry->val == swap.val)
922 shmem_swp_unmap(entry);
924 spin_unlock(&info->lock);
929 wait_on_page_locked(swappage);
930 page_cache_release(swappage);
934 /* We have to do this with page locked to prevent races */
935 if (TestSetPageLocked(swappage)) {
936 shmem_swp_unmap(entry);
937 spin_unlock(&info->lock);
938 wait_on_page_locked(swappage);
939 page_cache_release(swappage);
942 if (PageWriteback(swappage)) {
943 shmem_swp_unmap(entry);
944 spin_unlock(&info->lock);
945 wait_on_page_writeback(swappage);
946 unlock_page(swappage);
947 page_cache_release(swappage);
950 if (!PageUptodate(swappage)) {
951 shmem_swp_unmap(entry);
952 spin_unlock(&info->lock);
953 unlock_page(swappage);
954 page_cache_release(swappage);
960 shmem_swp_set(info, entry, 0);
961 shmem_swp_unmap(entry);
962 delete_from_swap_cache(swappage);
963 spin_unlock(&info->lock);
964 copy_highpage(filepage, swappage);
965 unlock_page(swappage);
966 page_cache_release(swappage);
967 flush_dcache_page(filepage);
968 SetPageUptodate(filepage);
969 set_page_dirty(filepage);
971 } else if (!(error = move_from_swap_cache(
972 swappage, idx, mapping))) {
973 info->flags |= SHMEM_PAGEIN;
974 shmem_swp_set(info, entry, 0);
975 shmem_swp_unmap(entry);
976 spin_unlock(&info->lock);
980 shmem_swp_unmap(entry);
981 spin_unlock(&info->lock);
982 unlock_page(swappage);
983 page_cache_release(swappage);
984 if (error == -ENOMEM) {
985 /* let kswapd refresh zone for GFP_ATOMICs */
986 blk_congestion_wait(WRITE, HZ/50);
990 } else if (sgp == SGP_READ && !filepage) {
991 shmem_swp_unmap(entry);
992 filepage = find_get_page(mapping, idx);
994 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
995 spin_unlock(&info->lock);
996 wait_on_page_locked(filepage);
997 page_cache_release(filepage);
1001 spin_unlock(&info->lock);
1003 shmem_swp_unmap(entry);
1004 sbinfo = SHMEM_SB(inode->i_sb);
1005 spin_lock(&sbinfo->stat_lock);
1006 if (sbinfo->free_blocks == 0 || shmem_acct_block(info->flags)) {
1007 spin_unlock(&sbinfo->stat_lock);
1008 spin_unlock(&info->lock);
1012 sbinfo->free_blocks--;
1013 inode->i_blocks += BLOCKS_PER_PAGE;
1014 spin_unlock(&sbinfo->stat_lock);
1017 spin_unlock(&info->lock);
1018 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1022 shmem_unacct_blocks(info->flags, 1);
1023 shmem_free_block(inode);
1028 spin_lock(&info->lock);
1029 entry = shmem_swp_alloc(info, idx, sgp);
1031 error = PTR_ERR(entry);
1034 shmem_swp_unmap(entry);
1036 if (error || swap.val || 0 != add_to_page_cache_lru(
1037 filepage, mapping, idx, GFP_ATOMIC)) {
1038 spin_unlock(&info->lock);
1039 page_cache_release(filepage);
1040 shmem_unacct_blocks(info->flags, 1);
1041 shmem_free_block(inode);
1047 info->flags |= SHMEM_PAGEIN;
1051 spin_unlock(&info->lock);
1052 clear_highpage(filepage);
1053 flush_dcache_page(filepage);
1054 SetPageUptodate(filepage);
1059 unlock_page(filepage);
1062 *pagep = ZERO_PAGE(0);
1069 if (*pagep != filepage) {
1070 unlock_page(filepage);
1071 page_cache_release(filepage);
1076 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1078 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1079 struct page *page = NULL;
1083 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1084 idx += vma->vm_pgoff;
1085 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1087 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1089 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1091 mark_page_accessed(page);
1095 static int shmem_populate(struct vm_area_struct *vma,
1096 unsigned long addr, unsigned long len,
1097 pgprot_t prot, unsigned long pgoff, int nonblock)
1099 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1100 struct mm_struct *mm = vma->vm_mm;
1101 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1104 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1105 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1108 while ((long) len > 0) {
1109 struct page *page = NULL;
1112 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1114 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1118 mark_page_accessed(page);
1119 err = install_page(mm, vma, addr, page, prot);
1121 page_cache_release(page);
1124 } else if (nonblock) {
1126 * If a nonlinear mapping then store the file page
1127 * offset in the pte.
1129 if (pgoff != linear_page_index(vma, addr)) {
1130 err = install_file_pte(mm, vma, addr, pgoff, prot);
1144 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1146 struct inode *i = vma->vm_file->f_dentry->d_inode;
1147 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1151 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1153 struct inode *i = vma->vm_file->f_dentry->d_inode;
1156 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1157 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1161 /* Protects current->user->locked_shm from concurrent access */
1162 static spinlock_t shmem_lock_user = SPIN_LOCK_UNLOCKED;
1164 int shmem_lock(struct file *file, int lock, struct user_struct * user)
1166 struct inode *inode = file->f_dentry->d_inode;
1167 struct shmem_inode_info *info = SHMEM_I(inode);
1168 unsigned long lock_limit, locked;
1169 int retval = -ENOMEM;
1171 spin_lock(&info->lock);
1172 spin_lock(&shmem_lock_user);
1173 if (lock && !(info->flags & VM_LOCKED)) {
1174 locked = inode->i_size >> PAGE_SHIFT;
1175 locked += user->locked_shm;
1176 lock_limit = current->rlim[RLIMIT_MEMLOCK].rlim_cur;
1177 lock_limit >>= PAGE_SHIFT;
1178 if ((locked > lock_limit) && !capable(CAP_IPC_LOCK))
1180 /* for this branch user == current->user so it won't go away under us */
1181 atomic_inc(&user->__count);
1182 user->locked_shm = locked;
1184 if (!lock && (info->flags & VM_LOCKED) && user) {
1185 locked = inode->i_size >> PAGE_SHIFT;
1186 user->locked_shm -= locked;
1190 info->flags |= VM_LOCKED;
1192 info->flags &= ~VM_LOCKED;
1195 spin_unlock(&shmem_lock_user);
1196 spin_unlock(&info->lock);
1200 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1202 file_accessed(file);
1203 vma->vm_ops = &shmem_vm_ops;
1207 static struct inode *
1208 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1210 struct inode *inode;
1211 struct shmem_inode_info *info;
1212 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1214 spin_lock(&sbinfo->stat_lock);
1215 if (!sbinfo->free_inodes) {
1216 spin_unlock(&sbinfo->stat_lock);
1219 sbinfo->free_inodes--;
1220 spin_unlock(&sbinfo->stat_lock);
1222 inode = new_inode(sb);
1224 inode->i_mode = mode;
1225 inode->i_uid = current->fsuid;
1226 inode->i_gid = current->fsgid;
1227 inode->i_blksize = PAGE_CACHE_SIZE;
1228 inode->i_blocks = 0;
1229 inode->i_mapping->a_ops = &shmem_aops;
1230 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1231 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1232 info = SHMEM_I(inode);
1233 memset(info, 0, (char *)inode - (char *)info);
1234 spin_lock_init(&info->lock);
1235 mpol_shared_policy_init(&info->policy);
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;
1243 spin_lock(&shmem_ilock);
1244 list_add_tail(&info->list, &shmem_inodes);
1245 spin_unlock(&shmem_ilock);
1249 /* Some things misbehave if size == 0 on a directory */
1250 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1251 inode->i_op = &shmem_dir_inode_operations;
1252 inode->i_fop = &simple_dir_operations;
1261 static int shmem_set_size(struct shmem_sb_info *info,
1262 unsigned long max_blocks, unsigned long max_inodes)
1265 unsigned long blocks, inodes;
1267 spin_lock(&info->stat_lock);
1268 blocks = info->max_blocks - info->free_blocks;
1269 inodes = info->max_inodes - info->free_inodes;
1271 if (max_blocks < blocks)
1273 if (max_inodes < inodes)
1276 info->max_blocks = max_blocks;
1277 info->free_blocks = max_blocks - blocks;
1278 info->max_inodes = max_inodes;
1279 info->free_inodes = max_inodes - inodes;
1281 spin_unlock(&info->stat_lock);
1287 static struct inode_operations shmem_symlink_inode_operations;
1288 static struct inode_operations shmem_symlink_inline_operations;
1291 * Normally tmpfs makes no use of shmem_prepare_write, but it
1292 * lets a tmpfs file be used read-write below the loop driver.
1295 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1297 struct inode *inode = page->mapping->host;
1298 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1302 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1304 struct inode *inode = file->f_dentry->d_inode;
1306 unsigned long written;
1309 if ((ssize_t) count < 0)
1312 if (!access_ok(VERIFY_READ, buf, count))
1315 down(&inode->i_sem);
1320 err = generic_write_checks(file, &pos, &count, 0);
1324 err = remove_suid(file->f_dentry);
1328 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1331 struct page *page = NULL;
1332 unsigned long bytes, index, offset;
1336 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1337 index = pos >> PAGE_CACHE_SHIFT;
1338 bytes = PAGE_CACHE_SIZE - offset;
1343 * We don't hold page lock across copy from user -
1344 * what would it guard against? - so no deadlock here.
1345 * But it still may be a good idea to prefault below.
1348 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1353 if (PageHighMem(page)) {
1354 volatile unsigned char dummy;
1355 __get_user(dummy, buf);
1356 __get_user(dummy, buf + bytes - 1);
1358 kaddr = kmap_atomic(page, KM_USER0);
1359 left = __copy_from_user(kaddr + offset, buf, bytes);
1360 kunmap_atomic(kaddr, KM_USER0);
1364 left = __copy_from_user(kaddr + offset, buf, bytes);
1372 if (pos > inode->i_size)
1373 i_size_write(inode, pos);
1375 flush_dcache_page(page);
1376 set_page_dirty(page);
1377 mark_page_accessed(page);
1378 page_cache_release(page);
1388 * Our dirty pages are not counted in nr_dirty,
1389 * and we do not attempt to balance dirty pages.
1403 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1405 struct inode *inode = filp->f_dentry->d_inode;
1406 struct address_space *mapping = inode->i_mapping;
1407 unsigned long index, offset;
1409 index = *ppos >> PAGE_CACHE_SHIFT;
1410 offset = *ppos & ~PAGE_CACHE_MASK;
1413 struct page *page = NULL;
1414 unsigned long end_index, nr, ret;
1415 loff_t i_size = i_size_read(inode);
1417 end_index = i_size >> PAGE_CACHE_SHIFT;
1418 if (index > end_index)
1420 if (index == end_index) {
1421 nr = i_size & ~PAGE_CACHE_MASK;
1426 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1428 if (desc->error == -EINVAL)
1434 * We must evaluate after, since reads (unlike writes)
1435 * are called without i_sem protection against truncate
1437 nr = PAGE_CACHE_SIZE;
1438 i_size = i_size_read(inode);
1439 end_index = i_size >> PAGE_CACHE_SHIFT;
1440 if (index == end_index) {
1441 nr = i_size & ~PAGE_CACHE_MASK;
1443 page_cache_release(page);
1449 if (page != ZERO_PAGE(0)) {
1451 * If users can be writing to this page using arbitrary
1452 * virtual addresses, take care about potential aliasing
1453 * before reading the page on the kernel side.
1455 if (mapping_writably_mapped(mapping))
1456 flush_dcache_page(page);
1458 * Mark the page accessed if we read the beginning.
1461 mark_page_accessed(page);
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 spin_lock(&sbinfo->stat_lock);
1538 buf->f_blocks = sbinfo->max_blocks;
1539 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1540 buf->f_files = sbinfo->max_inodes;
1541 buf->f_ffree = sbinfo->free_inodes;
1542 spin_unlock(&sbinfo->stat_lock);
1543 buf->f_namelen = NAME_MAX;
1548 * File creation. Allocate an inode, and we're done..
1551 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1553 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1554 int error = -ENOSPC;
1557 if (dir->i_mode & S_ISGID) {
1558 inode->i_gid = dir->i_gid;
1560 inode->i_mode |= S_ISGID;
1562 dir->i_size += BOGO_DIRENT_SIZE;
1563 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1564 d_instantiate(dentry, inode);
1565 dget(dentry); /* Extra count - pin the dentry in core */
1571 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1575 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1581 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1582 struct nameidata *nd)
1584 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1590 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1592 struct inode *inode = old_dentry->d_inode;
1594 dir->i_size += BOGO_DIRENT_SIZE;
1595 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1597 atomic_inc(&inode->i_count); /* New dentry reference */
1598 dget(dentry); /* Extra pinning count for the created dentry */
1599 d_instantiate(dentry, inode);
1603 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1605 struct inode *inode = dentry->d_inode;
1607 dir->i_size -= BOGO_DIRENT_SIZE;
1608 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1610 dput(dentry); /* Undo the count from "create" - this does all the work */
1614 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1616 if (!simple_empty(dentry))
1620 return shmem_unlink(dir, dentry);
1624 * The VFS layer already does all the dentry stuff for rename,
1625 * we just have to decrement the usage count for the target if
1626 * it exists so that the VFS layer correctly free's it when it
1629 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1631 struct inode *inode = old_dentry->d_inode;
1632 int they_are_dirs = S_ISDIR(inode->i_mode);
1634 if (!simple_empty(new_dentry))
1637 if (new_dentry->d_inode) {
1638 (void) shmem_unlink(new_dir, new_dentry);
1641 } else if (they_are_dirs) {
1646 old_dir->i_size -= BOGO_DIRENT_SIZE;
1647 new_dir->i_size += BOGO_DIRENT_SIZE;
1648 old_dir->i_ctime = old_dir->i_mtime =
1649 new_dir->i_ctime = new_dir->i_mtime =
1650 inode->i_ctime = CURRENT_TIME;
1654 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1658 struct inode *inode;
1659 struct page *page = NULL;
1661 struct shmem_inode_info *info;
1663 len = strlen(symname) + 1;
1664 if (len > PAGE_CACHE_SIZE)
1665 return -ENAMETOOLONG;
1667 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1671 info = SHMEM_I(inode);
1672 inode->i_size = len-1;
1673 if (len <= (char *)inode - (char *)info) {
1675 memcpy(info, symname, len);
1676 inode->i_op = &shmem_symlink_inline_operations;
1678 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1683 inode->i_op = &shmem_symlink_inode_operations;
1684 spin_lock(&shmem_ilock);
1685 list_add_tail(&info->list, &shmem_inodes);
1686 spin_unlock(&shmem_ilock);
1687 kaddr = kmap_atomic(page, KM_USER0);
1688 memcpy(kaddr, symname, len);
1689 kunmap_atomic(kaddr, KM_USER0);
1690 set_page_dirty(page);
1691 page_cache_release(page);
1693 if (dir->i_mode & S_ISGID)
1694 inode->i_gid = dir->i_gid;
1695 dir->i_size += BOGO_DIRENT_SIZE;
1696 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1697 d_instantiate(dentry, inode);
1702 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1704 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1708 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1710 struct page *page = NULL;
1711 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1712 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1716 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1718 if (!IS_ERR(nd_get_link(nd))) {
1721 page = find_get_page(dentry->d_inode->i_mapping, 0);
1725 mark_page_accessed(page);
1726 page_cache_release(page);
1727 page_cache_release(page);
1731 static struct inode_operations shmem_symlink_inline_operations = {
1732 .readlink = generic_readlink,
1733 .follow_link = shmem_follow_link_inline,
1736 static struct inode_operations shmem_symlink_inode_operations = {
1737 .truncate = shmem_truncate,
1738 .readlink = generic_readlink,
1739 .follow_link = shmem_follow_link,
1740 .put_link = shmem_put_link,
1743 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1745 char *this_char, *value, *rest;
1747 while ((this_char = strsep(&options, ",")) != NULL) {
1750 if ((value = strchr(this_char,'=')) != NULL) {
1754 "tmpfs: No value for mount option '%s'\n",
1759 if (!strcmp(this_char,"size")) {
1760 unsigned long long size;
1761 size = memparse(value,&rest);
1763 size <<= PAGE_SHIFT;
1764 size *= totalram_pages;
1770 *blocks = size >> PAGE_CACHE_SHIFT;
1771 } else if (!strcmp(this_char,"nr_blocks")) {
1772 *blocks = memparse(value,&rest);
1775 } else if (!strcmp(this_char,"nr_inodes")) {
1776 *inodes = memparse(value,&rest);
1779 } else if (!strcmp(this_char,"mode")) {
1782 *mode = simple_strtoul(value,&rest,8);
1785 } else if (!strcmp(this_char,"uid")) {
1788 *uid = simple_strtoul(value,&rest,0);
1791 } else if (!strcmp(this_char,"gid")) {
1794 *gid = simple_strtoul(value,&rest,0);
1798 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1806 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1812 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1814 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1815 unsigned long max_blocks = sbinfo->max_blocks;
1816 unsigned long max_inodes = sbinfo->max_inodes;
1818 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
1820 return shmem_set_size(sbinfo, max_blocks, max_inodes);
1824 static int shmem_fill_super(struct super_block *sb,
1825 void *data, int silent)
1827 struct inode *inode;
1828 struct dentry *root;
1829 unsigned long blocks, inodes;
1830 int mode = S_IRWXUGO | S_ISVTX;
1831 uid_t uid = current->fsuid;
1832 gid_t gid = current->fsgid;
1833 struct shmem_sb_info *sbinfo;
1836 sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
1839 sb->s_fs_info = sbinfo;
1840 memset(sbinfo, 0, sizeof(struct shmem_sb_info));
1843 * Per default we only allow half of the physical ram per
1846 blocks = inodes = totalram_pages / 2;
1849 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, &inodes)) {
1854 sb->s_flags |= MS_NOUSER;
1857 spin_lock_init(&sbinfo->stat_lock);
1858 sbinfo->max_blocks = blocks;
1859 sbinfo->free_blocks = blocks;
1860 sbinfo->max_inodes = inodes;
1861 sbinfo->free_inodes = inodes;
1862 sb->s_maxbytes = SHMEM_MAX_BYTES;
1863 sb->s_blocksize = PAGE_CACHE_SIZE;
1864 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1865 sb->s_magic = TMPFS_MAGIC;
1866 sb->s_op = &shmem_ops;
1867 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1872 root = d_alloc_root(inode);
1882 sb->s_fs_info = NULL;
1886 static void shmem_put_super(struct super_block *sb)
1888 kfree(sb->s_fs_info);
1889 sb->s_fs_info = NULL;
1892 static kmem_cache_t *shmem_inode_cachep;
1894 static struct inode *shmem_alloc_inode(struct super_block *sb)
1896 struct shmem_inode_info *p;
1897 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
1900 return &p->vfs_inode;
1903 static void shmem_destroy_inode(struct inode *inode)
1905 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
1906 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
1909 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
1911 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
1913 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1914 SLAB_CTOR_CONSTRUCTOR) {
1915 inode_init_once(&p->vfs_inode);
1919 static int init_inodecache(void)
1921 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
1922 sizeof(struct shmem_inode_info),
1923 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
1925 if (shmem_inode_cachep == NULL)
1930 static void destroy_inodecache(void)
1932 if (kmem_cache_destroy(shmem_inode_cachep))
1933 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
1936 static struct address_space_operations shmem_aops = {
1937 .writepage = shmem_writepage,
1938 .set_page_dirty = __set_page_dirty_nobuffers,
1940 .prepare_write = shmem_prepare_write,
1941 .commit_write = simple_commit_write,
1945 static struct file_operations shmem_file_operations = {
1948 .llseek = generic_file_llseek,
1949 .read = shmem_file_read,
1950 .write = shmem_file_write,
1951 .fsync = simple_sync_file,
1952 .sendfile = shmem_file_sendfile,
1956 static struct inode_operations shmem_inode_operations = {
1957 .truncate = shmem_truncate,
1958 .setattr = shmem_notify_change,
1961 static struct inode_operations shmem_dir_inode_operations = {
1963 .create = shmem_create,
1964 .lookup = simple_lookup,
1966 .unlink = shmem_unlink,
1967 .symlink = shmem_symlink,
1968 .mkdir = shmem_mkdir,
1969 .rmdir = shmem_rmdir,
1970 .mknod = shmem_mknod,
1971 .rename = shmem_rename,
1975 static struct super_operations shmem_ops = {
1976 .alloc_inode = shmem_alloc_inode,
1977 .destroy_inode = shmem_destroy_inode,
1979 .statfs = shmem_statfs,
1980 .remount_fs = shmem_remount_fs,
1982 .delete_inode = shmem_delete_inode,
1983 .drop_inode = generic_delete_inode,
1984 .put_super = shmem_put_super,
1987 static struct vm_operations_struct shmem_vm_ops = {
1988 .nopage = shmem_nopage,
1989 .populate = shmem_populate,
1991 .set_policy = shmem_set_policy,
1992 .get_policy = shmem_get_policy,
1996 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
1997 int flags, const char *dev_name, void *data)
1999 return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2002 static struct file_system_type tmpfs_fs_type = {
2003 .owner = THIS_MODULE,
2005 .get_sb = shmem_get_sb,
2006 .kill_sb = kill_litter_super,
2008 static struct vfsmount *shm_mnt;
2010 static int __init init_tmpfs(void)
2014 error = init_inodecache();
2018 error = register_filesystem(&tmpfs_fs_type);
2020 printk(KERN_ERR "Could not register tmpfs\n");
2024 devfs_mk_dir("shm");
2026 shm_mnt = kern_mount(&tmpfs_fs_type);
2027 if (IS_ERR(shm_mnt)) {
2028 error = PTR_ERR(shm_mnt);
2029 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2033 /* The internal instance should not do size checking */
2034 shmem_set_size(SHMEM_SB(shm_mnt->mnt_sb), ULONG_MAX, ULONG_MAX);
2038 unregister_filesystem(&tmpfs_fs_type);
2040 destroy_inodecache();
2042 shm_mnt = ERR_PTR(error);
2045 module_init(init_tmpfs)
2048 * shmem_file_setup - get an unlinked file living in tmpfs
2050 * @name: name for dentry (to be seen in /proc/<pid>/maps
2051 * @size: size to be set for the file
2054 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2058 struct inode *inode;
2059 struct dentry *dentry, *root;
2062 if (IS_ERR(shm_mnt))
2063 return (void *)shm_mnt;
2065 if (size > SHMEM_MAX_BYTES)
2066 return ERR_PTR(-EINVAL);
2068 if (shmem_acct_size(flags, size))
2069 return ERR_PTR(-ENOMEM);
2073 this.len = strlen(name);
2074 this.hash = 0; /* will go */
2075 root = shm_mnt->mnt_root;
2076 dentry = d_alloc(root, &this);
2081 file = get_empty_filp();
2086 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2090 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2091 d_instantiate(dentry, inode);
2092 inode->i_size = size;
2093 inode->i_nlink = 0; /* It is unlinked */
2094 file->f_vfsmnt = mntget(shm_mnt);
2095 file->f_dentry = dentry;
2096 file->f_mapping = inode->i_mapping;
2097 file->f_op = &shmem_file_operations;
2098 file->f_mode = FMODE_WRITE | FMODE_READ;
2106 shmem_unacct_size(flags, size);
2107 return ERR_PTR(error);
2111 * shmem_zero_setup - setup a shared anonymous mapping
2113 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2115 int shmem_zero_setup(struct vm_area_struct *vma)
2118 loff_t size = vma->vm_end - vma->vm_start;
2120 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2122 return PTR_ERR(file);
2126 vma->vm_file = file;
2127 vma->vm_ops = &shmem_vm_ops;
2131 EXPORT_SYMBOL(shmem_file_setup);