2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
11 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 * This file is released under the GPL.
21 * This virtual memory filesystem is heavily based on the ramfs. It
22 * extends ramfs by the ability to use swap and honor resource limits
23 * which makes it a completely usable filesystem.
26 #include <linux/module.h>
27 #include <linux/init.h>
29 #include <linux/xattr.h>
30 #include <linux/generic_acl.h>
32 #include <linux/mman.h>
33 #include <linux/file.h>
34 #include <linux/swap.h>
35 #include <linux/pagemap.h>
36 #include <linux/string.h>
37 #include <linux/slab.h>
38 #include <linux/backing-dev.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/mount.h>
41 #include <linux/writeback.h>
42 #include <linux/vfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/security.h>
45 #include <linux/swapops.h>
46 #include <linux/mempolicy.h>
47 #include <linux/namei.h>
48 #include <linux/ctype.h>
49 #include <linux/migrate.h>
50 #include <linux/highmem.h>
51 #include <linux/backing-dev.h>
53 #include <asm/uaccess.h>
54 #include <asm/div64.h>
55 #include <asm/pgtable.h>
57 /* This magic number is used in glibc for posix shared memory */
59 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
60 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
61 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
63 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
64 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
66 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
68 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
69 #define SHMEM_PAGEIN VM_READ
70 #define SHMEM_TRUNCATE VM_WRITE
72 /* Definition to limit shmem_truncate's steps between cond_rescheds */
73 #define LATENCY_LIMIT 64
75 /* Pretend that each entry is of this size in directory's i_size */
76 #define BOGO_DIRENT_SIZE 20
78 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
80 SGP_QUICK, /* don't try more than file page cache lookup */
81 SGP_READ, /* don't exceed i_size, don't allocate page */
82 SGP_CACHE, /* don't exceed i_size, may allocate page */
83 SGP_WRITE, /* may exceed i_size, may allocate page */
86 static int shmem_getpage(struct inode *inode, unsigned long idx,
87 struct page **pagep, enum sgp_type sgp, int *type);
89 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
92 * The above definition of ENTRIES_PER_PAGE, and the use of
93 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
94 * might be reconsidered if it ever diverges from PAGE_SIZE.
96 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
99 static inline void shmem_dir_free(struct page *page)
101 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
104 static struct page **shmem_dir_map(struct page *page)
106 return (struct page **)kmap_atomic(page, KM_USER0);
109 static inline void shmem_dir_unmap(struct page **dir)
111 kunmap_atomic(dir, KM_USER0);
114 static swp_entry_t *shmem_swp_map(struct page *page)
116 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
119 static inline void shmem_swp_balance_unmap(void)
122 * When passing a pointer to an i_direct entry, to code which
123 * also handles indirect entries and so will shmem_swp_unmap,
124 * we must arrange for the preempt count to remain in balance.
125 * What kmap_atomic of a lowmem page does depends on config
126 * and architecture, so pretend to kmap_atomic some lowmem page.
128 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
131 static inline void shmem_swp_unmap(swp_entry_t *entry)
133 kunmap_atomic(entry, KM_USER1);
136 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
138 return sb->s_fs_info;
142 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
143 * for shared memory and for shared anonymous (/dev/zero) mappings
144 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
145 * consistent with the pre-accounting of private mappings ...
147 static inline int shmem_acct_size(unsigned long flags, loff_t size)
149 return (flags & VM_ACCOUNT)?
150 security_vm_enough_memory(VM_ACCT(size)): 0;
153 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
155 if (flags & VM_ACCOUNT)
156 vm_unacct_memory(VM_ACCT(size));
160 * ... whereas tmpfs objects are accounted incrementally as
161 * pages are allocated, in order to allow huge sparse files.
162 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
163 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
165 static inline int shmem_acct_block(unsigned long flags)
167 return (flags & VM_ACCOUNT)?
168 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
171 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
173 if (!(flags & VM_ACCOUNT))
174 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
177 static struct super_operations shmem_ops;
178 static const struct address_space_operations shmem_aops;
179 static const struct file_operations shmem_file_operations;
180 static struct inode_operations shmem_inode_operations;
181 static struct inode_operations shmem_dir_inode_operations;
182 static struct inode_operations shmem_special_inode_operations;
183 static struct vm_operations_struct shmem_vm_ops;
185 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
186 .ra_pages = 0, /* No readahead */
187 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
188 .unplug_io_fn = default_unplug_io_fn,
191 static LIST_HEAD(shmem_swaplist);
192 static DEFINE_SPINLOCK(shmem_swaplist_lock);
194 static void shmem_free_blocks(struct inode *inode, long pages)
196 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
197 if (sbinfo->max_blocks) {
198 spin_lock(&sbinfo->stat_lock);
199 sbinfo->free_blocks += pages;
200 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
201 spin_unlock(&sbinfo->stat_lock);
206 * shmem_recalc_inode - recalculate the size of an inode
208 * @inode: inode to recalc
210 * We have to calculate the free blocks since the mm can drop
211 * undirtied hole pages behind our back.
213 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
214 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
216 * It has to be called with the spinlock held.
218 static void shmem_recalc_inode(struct inode *inode)
220 struct shmem_inode_info *info = SHMEM_I(inode);
223 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
225 info->alloced -= freed;
226 shmem_unacct_blocks(info->flags, freed);
227 shmem_free_blocks(inode, freed);
232 * shmem_swp_entry - find the swap vector position in the info structure
234 * @info: info structure for the inode
235 * @index: index of the page to find
236 * @page: optional page to add to the structure. Has to be preset to
239 * If there is no space allocated yet it will return NULL when
240 * page is NULL, else it will use the page for the needed block,
241 * setting it to NULL on return to indicate that it has been used.
243 * The swap vector is organized the following way:
245 * There are SHMEM_NR_DIRECT entries directly stored in the
246 * shmem_inode_info structure. So small files do not need an addional
249 * For pages with index > SHMEM_NR_DIRECT there is the pointer
250 * i_indirect which points to a page which holds in the first half
251 * doubly indirect blocks, in the second half triple indirect blocks:
253 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
254 * following layout (for SHMEM_NR_DIRECT == 16):
256 * i_indirect -> dir --> 16-19
269 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
271 unsigned long offset;
275 if (index < SHMEM_NR_DIRECT) {
276 shmem_swp_balance_unmap();
277 return info->i_direct+index;
279 if (!info->i_indirect) {
281 info->i_indirect = *page;
284 return NULL; /* need another page */
287 index -= SHMEM_NR_DIRECT;
288 offset = index % ENTRIES_PER_PAGE;
289 index /= ENTRIES_PER_PAGE;
290 dir = shmem_dir_map(info->i_indirect);
292 if (index >= ENTRIES_PER_PAGE/2) {
293 index -= ENTRIES_PER_PAGE/2;
294 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
295 index %= ENTRIES_PER_PAGE;
302 shmem_dir_unmap(dir);
303 return NULL; /* need another page */
305 shmem_dir_unmap(dir);
306 dir = shmem_dir_map(subdir);
312 if (!page || !(subdir = *page)) {
313 shmem_dir_unmap(dir);
314 return NULL; /* need a page */
319 shmem_dir_unmap(dir);
320 return shmem_swp_map(subdir) + offset;
323 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
325 long incdec = value? 1: -1;
328 info->swapped += incdec;
329 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
330 struct page *page = kmap_atomic_to_page(entry);
331 set_page_private(page, page_private(page) + incdec);
336 * shmem_swp_alloc - get the position of the swap entry for the page.
337 * If it does not exist allocate the entry.
339 * @info: info structure for the inode
340 * @index: index of the page to find
341 * @sgp: check and recheck i_size? skip allocation?
343 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
345 struct inode *inode = &info->vfs_inode;
346 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
347 struct page *page = NULL;
350 if (sgp != SGP_WRITE &&
351 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
352 return ERR_PTR(-EINVAL);
354 while (!(entry = shmem_swp_entry(info, index, &page))) {
356 return shmem_swp_map(ZERO_PAGE(0));
358 * Test free_blocks against 1 not 0, since we have 1 data
359 * page (and perhaps indirect index pages) yet to allocate:
360 * a waste to allocate index if we cannot allocate data.
362 if (sbinfo->max_blocks) {
363 spin_lock(&sbinfo->stat_lock);
364 if (sbinfo->free_blocks <= 1) {
365 spin_unlock(&sbinfo->stat_lock);
366 return ERR_PTR(-ENOSPC);
368 sbinfo->free_blocks--;
369 inode->i_blocks += BLOCKS_PER_PAGE;
370 spin_unlock(&sbinfo->stat_lock);
373 spin_unlock(&info->lock);
374 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
376 set_page_private(page, 0);
377 spin_lock(&info->lock);
380 shmem_free_blocks(inode, 1);
381 return ERR_PTR(-ENOMEM);
383 if (sgp != SGP_WRITE &&
384 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
385 entry = ERR_PTR(-EINVAL);
388 if (info->next_index <= index)
389 info->next_index = index + 1;
392 /* another task gave its page, or truncated the file */
393 shmem_free_blocks(inode, 1);
394 shmem_dir_free(page);
396 if (info->next_index <= index && !IS_ERR(entry))
397 info->next_index = index + 1;
402 * shmem_free_swp - free some swap entries in a directory
404 * @dir: pointer to the directory
405 * @edir: pointer after last entry of the directory
406 * @punch_lock: pointer to spinlock when needed for the holepunch case
408 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
409 spinlock_t *punch_lock)
411 spinlock_t *punch_unlock = NULL;
415 for (ptr = dir; ptr < edir; ptr++) {
417 if (unlikely(punch_lock)) {
418 punch_unlock = punch_lock;
420 spin_lock(punch_unlock);
424 free_swap_and_cache(*ptr);
425 *ptr = (swp_entry_t){0};
430 spin_unlock(punch_unlock);
434 static int shmem_map_and_free_swp(struct page *subdir, int offset,
435 int limit, struct page ***dir, spinlock_t *punch_lock)
440 ptr = shmem_swp_map(subdir);
441 for (; offset < limit; offset += LATENCY_LIMIT) {
442 int size = limit - offset;
443 if (size > LATENCY_LIMIT)
444 size = LATENCY_LIMIT;
445 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
447 if (need_resched()) {
448 shmem_swp_unmap(ptr);
450 shmem_dir_unmap(*dir);
454 ptr = shmem_swp_map(subdir);
457 shmem_swp_unmap(ptr);
461 static void shmem_free_pages(struct list_head *next)
467 page = container_of(next, struct page, lru);
469 shmem_dir_free(page);
471 if (freed >= LATENCY_LIMIT) {
478 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
480 struct shmem_inode_info *info = SHMEM_I(inode);
485 unsigned long diroff;
491 LIST_HEAD(pages_to_free);
492 long nr_pages_to_free = 0;
493 long nr_swaps_freed = 0;
497 spinlock_t *needs_lock;
498 spinlock_t *punch_lock;
499 unsigned long upper_limit;
501 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
502 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
503 if (idx >= info->next_index)
506 spin_lock(&info->lock);
507 info->flags |= SHMEM_TRUNCATE;
508 if (likely(end == (loff_t) -1)) {
509 limit = info->next_index;
510 upper_limit = SHMEM_MAX_INDEX;
511 info->next_index = idx;
515 if (end + 1 >= inode->i_size) { /* we may free a little more */
516 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
518 upper_limit = SHMEM_MAX_INDEX;
520 limit = (end + 1) >> PAGE_CACHE_SHIFT;
523 needs_lock = &info->lock;
527 topdir = info->i_indirect;
528 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
529 info->i_indirect = NULL;
531 list_add(&topdir->lru, &pages_to_free);
533 spin_unlock(&info->lock);
535 if (info->swapped && idx < SHMEM_NR_DIRECT) {
536 ptr = info->i_direct;
538 if (size > SHMEM_NR_DIRECT)
539 size = SHMEM_NR_DIRECT;
540 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
544 * If there are no indirect blocks or we are punching a hole
545 * below indirect blocks, nothing to be done.
547 if (!topdir || limit <= SHMEM_NR_DIRECT)
551 * The truncation case has already dropped info->lock, and we're safe
552 * because i_size and next_index have already been lowered, preventing
553 * access beyond. But in the punch_hole case, we still need to take
554 * the lock when updating the swap directory, because there might be
555 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
556 * shmem_writepage. However, whenever we find we can remove a whole
557 * directory page (not at the misaligned start or end of the range),
558 * we first NULLify its pointer in the level above, and then have no
559 * need to take the lock when updating its contents: needs_lock and
560 * punch_lock (either pointing to info->lock or NULL) manage this.
563 upper_limit -= SHMEM_NR_DIRECT;
564 limit -= SHMEM_NR_DIRECT;
565 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
566 offset = idx % ENTRIES_PER_PAGE;
569 dir = shmem_dir_map(topdir);
570 stage = ENTRIES_PER_PAGEPAGE/2;
571 if (idx < ENTRIES_PER_PAGEPAGE/2) {
573 diroff = idx/ENTRIES_PER_PAGE;
575 dir += ENTRIES_PER_PAGE/2;
576 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
578 stage += ENTRIES_PER_PAGEPAGE;
581 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
582 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
583 if (!diroff && !offset && upper_limit >= stage) {
585 spin_lock(needs_lock);
587 spin_unlock(needs_lock);
592 list_add(&middir->lru, &pages_to_free);
594 shmem_dir_unmap(dir);
595 dir = shmem_dir_map(middir);
603 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
604 if (unlikely(idx == stage)) {
605 shmem_dir_unmap(dir);
606 dir = shmem_dir_map(topdir) +
607 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
610 idx += ENTRIES_PER_PAGEPAGE;
614 stage = idx + ENTRIES_PER_PAGEPAGE;
617 needs_lock = &info->lock;
618 if (upper_limit >= stage) {
620 spin_lock(needs_lock);
622 spin_unlock(needs_lock);
627 list_add(&middir->lru, &pages_to_free);
629 shmem_dir_unmap(dir);
631 dir = shmem_dir_map(middir);
634 punch_lock = needs_lock;
635 subdir = dir[diroff];
636 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
638 spin_lock(needs_lock);
640 spin_unlock(needs_lock);
645 list_add(&subdir->lru, &pages_to_free);
647 if (subdir && page_private(subdir) /* has swap entries */) {
649 if (size > ENTRIES_PER_PAGE)
650 size = ENTRIES_PER_PAGE;
651 freed = shmem_map_and_free_swp(subdir,
652 offset, size, &dir, punch_lock);
654 dir = shmem_dir_map(middir);
655 nr_swaps_freed += freed;
656 if (offset || punch_lock) {
657 spin_lock(&info->lock);
658 set_page_private(subdir,
659 page_private(subdir) - freed);
660 spin_unlock(&info->lock);
662 BUG_ON(page_private(subdir) != freed);
667 shmem_dir_unmap(dir);
669 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
671 * Call truncate_inode_pages again: racing shmem_unuse_inode
672 * may have swizzled a page in from swap since vmtruncate or
673 * generic_delete_inode did it, before we lowered next_index.
674 * Also, though shmem_getpage checks i_size before adding to
675 * cache, no recheck after: so fix the narrow window there too.
677 * Recalling truncate_inode_pages_range and unmap_mapping_range
678 * every time for punch_hole (which never got a chance to clear
679 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
680 * yet hardly ever necessary: try to optimize them out later.
682 truncate_inode_pages_range(inode->i_mapping, start, end);
684 unmap_mapping_range(inode->i_mapping, start,
688 spin_lock(&info->lock);
689 info->flags &= ~SHMEM_TRUNCATE;
690 info->swapped -= nr_swaps_freed;
691 if (nr_pages_to_free)
692 shmem_free_blocks(inode, nr_pages_to_free);
693 shmem_recalc_inode(inode);
694 spin_unlock(&info->lock);
697 * Empty swap vector directory pages to be freed?
699 if (!list_empty(&pages_to_free)) {
700 pages_to_free.prev->next = NULL;
701 shmem_free_pages(pages_to_free.next);
705 static void shmem_truncate(struct inode *inode)
707 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
710 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
712 struct inode *inode = dentry->d_inode;
713 struct page *page = NULL;
716 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
717 if (attr->ia_size < inode->i_size) {
719 * If truncating down to a partial page, then
720 * if that page is already allocated, hold it
721 * in memory until the truncation is over, so
722 * truncate_partial_page cannnot miss it were
723 * it assigned to swap.
725 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
726 (void) shmem_getpage(inode,
727 attr->ia_size>>PAGE_CACHE_SHIFT,
728 &page, SGP_READ, NULL);
731 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
732 * detect if any pages might have been added to cache
733 * after truncate_inode_pages. But we needn't bother
734 * if it's being fully truncated to zero-length: the
735 * nrpages check is efficient enough in that case.
738 struct shmem_inode_info *info = SHMEM_I(inode);
739 spin_lock(&info->lock);
740 info->flags &= ~SHMEM_PAGEIN;
741 spin_unlock(&info->lock);
746 error = inode_change_ok(inode, attr);
748 error = inode_setattr(inode, attr);
749 #ifdef CONFIG_TMPFS_POSIX_ACL
750 if (!error && (attr->ia_valid & ATTR_MODE))
751 error = generic_acl_chmod(inode, &shmem_acl_ops);
754 page_cache_release(page);
758 static void shmem_delete_inode(struct inode *inode)
760 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
761 struct shmem_inode_info *info = SHMEM_I(inode);
763 if (inode->i_op->truncate == shmem_truncate) {
764 truncate_inode_pages(inode->i_mapping, 0);
765 shmem_unacct_size(info->flags, inode->i_size);
767 shmem_truncate(inode);
768 if (!list_empty(&info->swaplist)) {
769 spin_lock(&shmem_swaplist_lock);
770 list_del_init(&info->swaplist);
771 spin_unlock(&shmem_swaplist_lock);
774 BUG_ON(inode->i_blocks);
775 if (sbinfo->max_inodes) {
776 spin_lock(&sbinfo->stat_lock);
777 sbinfo->free_inodes++;
778 spin_unlock(&sbinfo->stat_lock);
783 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
787 for (ptr = dir; ptr < edir; ptr++) {
788 if (ptr->val == entry.val)
794 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
807 ptr = info->i_direct;
808 spin_lock(&info->lock);
809 limit = info->next_index;
811 if (size > SHMEM_NR_DIRECT)
812 size = SHMEM_NR_DIRECT;
813 offset = shmem_find_swp(entry, ptr, ptr+size);
815 shmem_swp_balance_unmap();
818 if (!info->i_indirect)
821 dir = shmem_dir_map(info->i_indirect);
822 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
824 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
825 if (unlikely(idx == stage)) {
826 shmem_dir_unmap(dir-1);
827 dir = shmem_dir_map(info->i_indirect) +
828 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
831 idx += ENTRIES_PER_PAGEPAGE;
835 stage = idx + ENTRIES_PER_PAGEPAGE;
837 shmem_dir_unmap(dir);
838 dir = shmem_dir_map(subdir);
841 if (subdir && page_private(subdir)) {
842 ptr = shmem_swp_map(subdir);
844 if (size > ENTRIES_PER_PAGE)
845 size = ENTRIES_PER_PAGE;
846 offset = shmem_find_swp(entry, ptr, ptr+size);
848 shmem_dir_unmap(dir);
851 shmem_swp_unmap(ptr);
855 shmem_dir_unmap(dir-1);
857 spin_unlock(&info->lock);
861 inode = &info->vfs_inode;
862 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
863 info->flags |= SHMEM_PAGEIN;
864 shmem_swp_set(info, ptr + offset, 0);
866 shmem_swp_unmap(ptr);
867 spin_unlock(&info->lock);
869 * Decrement swap count even when the entry is left behind:
870 * try_to_unuse will skip over mms, then reincrement count.
877 * shmem_unuse() search for an eventually swapped out shmem page.
879 int shmem_unuse(swp_entry_t entry, struct page *page)
881 struct list_head *p, *next;
882 struct shmem_inode_info *info;
885 spin_lock(&shmem_swaplist_lock);
886 list_for_each_safe(p, next, &shmem_swaplist) {
887 info = list_entry(p, struct shmem_inode_info, swaplist);
889 list_del_init(&info->swaplist);
890 else if (shmem_unuse_inode(info, entry, page)) {
891 /* move head to start search for next from here */
892 list_move_tail(&shmem_swaplist, &info->swaplist);
897 spin_unlock(&shmem_swaplist_lock);
902 * Move the page from the page cache to the swap cache.
904 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
906 struct shmem_inode_info *info;
907 swp_entry_t *entry, swap;
908 struct address_space *mapping;
912 BUG_ON(!PageLocked(page));
913 BUG_ON(page_mapped(page));
915 mapping = page->mapping;
917 inode = mapping->host;
918 info = SHMEM_I(inode);
919 if (info->flags & VM_LOCKED)
921 swap = get_swap_page();
925 spin_lock(&info->lock);
926 shmem_recalc_inode(inode);
927 if (index >= info->next_index) {
928 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
931 entry = shmem_swp_entry(info, index, NULL);
935 if (move_to_swap_cache(page, swap) == 0) {
936 shmem_swp_set(info, entry, swap.val);
937 shmem_swp_unmap(entry);
938 spin_unlock(&info->lock);
939 if (list_empty(&info->swaplist)) {
940 spin_lock(&shmem_swaplist_lock);
941 /* move instead of add in case we're racing */
942 list_move_tail(&info->swaplist, &shmem_swaplist);
943 spin_unlock(&shmem_swaplist_lock);
949 shmem_swp_unmap(entry);
951 spin_unlock(&info->lock);
954 set_page_dirty(page);
955 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
959 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
961 char *nodelist = strchr(value, ':');
965 /* NUL-terminate policy string */
967 if (nodelist_parse(nodelist, *policy_nodes))
970 if (!strcmp(value, "default")) {
971 *policy = MPOL_DEFAULT;
972 /* Don't allow a nodelist */
975 } else if (!strcmp(value, "prefer")) {
976 *policy = MPOL_PREFERRED;
977 /* Insist on a nodelist of one node only */
979 char *rest = nodelist;
980 while (isdigit(*rest))
985 } else if (!strcmp(value, "bind")) {
987 /* Insist on a nodelist */
990 } else if (!strcmp(value, "interleave")) {
991 *policy = MPOL_INTERLEAVE;
992 /* Default to nodes online if no nodelist */
994 *policy_nodes = node_online_map;
998 /* Restore string for error message */
1004 static struct page *shmem_swapin_async(struct shared_policy *p,
1005 swp_entry_t entry, unsigned long idx)
1008 struct vm_area_struct pvma;
1010 /* Create a pseudo vma that just contains the policy */
1011 memset(&pvma, 0, sizeof(struct vm_area_struct));
1012 pvma.vm_end = PAGE_SIZE;
1013 pvma.vm_pgoff = idx;
1014 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
1015 page = read_swap_cache_async(entry, &pvma, 0);
1016 mpol_free(pvma.vm_policy);
1020 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
1023 struct shared_policy *p = &info->policy;
1026 unsigned long offset;
1028 num = valid_swaphandles(entry, &offset);
1029 for (i = 0; i < num; offset++, i++) {
1030 page = shmem_swapin_async(p,
1031 swp_entry(swp_type(entry), offset), idx);
1034 page_cache_release(page);
1036 lru_add_drain(); /* Push any new pages onto the LRU now */
1037 return shmem_swapin_async(p, entry, idx);
1040 static struct page *
1041 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
1044 struct vm_area_struct pvma;
1047 memset(&pvma, 0, sizeof(struct vm_area_struct));
1048 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1049 pvma.vm_pgoff = idx;
1050 pvma.vm_end = PAGE_SIZE;
1051 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
1052 mpol_free(pvma.vm_policy);
1056 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
1061 static inline struct page *
1062 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
1064 swapin_readahead(entry, 0, NULL);
1065 return read_swap_cache_async(entry, NULL, 0);
1068 static inline struct page *
1069 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
1071 return alloc_page(gfp | __GFP_ZERO);
1076 * shmem_getpage - either get the page from swap or allocate a new one
1078 * If we allocate a new one we do not mark it dirty. That's up to the
1079 * vm. If we swap it in we mark it dirty since we also free the swap
1080 * entry since a page cannot live in both the swap and page cache
1082 static int shmem_getpage(struct inode *inode, unsigned long idx,
1083 struct page **pagep, enum sgp_type sgp, int *type)
1085 struct address_space *mapping = inode->i_mapping;
1086 struct shmem_inode_info *info = SHMEM_I(inode);
1087 struct shmem_sb_info *sbinfo;
1088 struct page *filepage = *pagep;
1089 struct page *swappage;
1094 if (idx >= SHMEM_MAX_INDEX)
1097 * Normally, filepage is NULL on entry, and either found
1098 * uptodate immediately, or allocated and zeroed, or read
1099 * in under swappage, which is then assigned to filepage.
1100 * But shmem_prepare_write passes in a locked filepage,
1101 * which may be found not uptodate by other callers too,
1102 * and may need to be copied from the swappage read in.
1106 filepage = find_lock_page(mapping, idx);
1107 if (filepage && PageUptodate(filepage))
1110 if (sgp == SGP_QUICK)
1113 spin_lock(&info->lock);
1114 shmem_recalc_inode(inode);
1115 entry = shmem_swp_alloc(info, idx, sgp);
1116 if (IS_ERR(entry)) {
1117 spin_unlock(&info->lock);
1118 error = PTR_ERR(entry);
1124 /* Look it up and read it in.. */
1125 swappage = lookup_swap_cache(swap);
1127 shmem_swp_unmap(entry);
1128 /* here we actually do the io */
1129 if (type && *type == VM_FAULT_MINOR) {
1130 __count_vm_event(PGMAJFAULT);
1131 *type = VM_FAULT_MAJOR;
1133 spin_unlock(&info->lock);
1134 swappage = shmem_swapin(info, swap, idx);
1136 spin_lock(&info->lock);
1137 entry = shmem_swp_alloc(info, idx, sgp);
1139 error = PTR_ERR(entry);
1141 if (entry->val == swap.val)
1143 shmem_swp_unmap(entry);
1145 spin_unlock(&info->lock);
1150 wait_on_page_locked(swappage);
1151 page_cache_release(swappage);
1155 /* We have to do this with page locked to prevent races */
1156 if (TestSetPageLocked(swappage)) {
1157 shmem_swp_unmap(entry);
1158 spin_unlock(&info->lock);
1159 wait_on_page_locked(swappage);
1160 page_cache_release(swappage);
1163 if (PageWriteback(swappage)) {
1164 shmem_swp_unmap(entry);
1165 spin_unlock(&info->lock);
1166 wait_on_page_writeback(swappage);
1167 unlock_page(swappage);
1168 page_cache_release(swappage);
1171 if (!PageUptodate(swappage)) {
1172 shmem_swp_unmap(entry);
1173 spin_unlock(&info->lock);
1174 unlock_page(swappage);
1175 page_cache_release(swappage);
1181 shmem_swp_set(info, entry, 0);
1182 shmem_swp_unmap(entry);
1183 delete_from_swap_cache(swappage);
1184 spin_unlock(&info->lock);
1185 copy_highpage(filepage, swappage);
1186 unlock_page(swappage);
1187 page_cache_release(swappage);
1188 flush_dcache_page(filepage);
1189 SetPageUptodate(filepage);
1190 set_page_dirty(filepage);
1192 } else if (!(error = move_from_swap_cache(
1193 swappage, idx, mapping))) {
1194 info->flags |= SHMEM_PAGEIN;
1195 shmem_swp_set(info, entry, 0);
1196 shmem_swp_unmap(entry);
1197 spin_unlock(&info->lock);
1198 filepage = swappage;
1201 shmem_swp_unmap(entry);
1202 spin_unlock(&info->lock);
1203 unlock_page(swappage);
1204 page_cache_release(swappage);
1205 if (error == -ENOMEM) {
1206 /* let kswapd refresh zone for GFP_ATOMICs */
1207 congestion_wait(WRITE, HZ/50);
1211 } else if (sgp == SGP_READ && !filepage) {
1212 shmem_swp_unmap(entry);
1213 filepage = find_get_page(mapping, idx);
1215 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1216 spin_unlock(&info->lock);
1217 wait_on_page_locked(filepage);
1218 page_cache_release(filepage);
1222 spin_unlock(&info->lock);
1224 shmem_swp_unmap(entry);
1225 sbinfo = SHMEM_SB(inode->i_sb);
1226 if (sbinfo->max_blocks) {
1227 spin_lock(&sbinfo->stat_lock);
1228 if (sbinfo->free_blocks == 0 ||
1229 shmem_acct_block(info->flags)) {
1230 spin_unlock(&sbinfo->stat_lock);
1231 spin_unlock(&info->lock);
1235 sbinfo->free_blocks--;
1236 inode->i_blocks += BLOCKS_PER_PAGE;
1237 spin_unlock(&sbinfo->stat_lock);
1238 } else if (shmem_acct_block(info->flags)) {
1239 spin_unlock(&info->lock);
1245 spin_unlock(&info->lock);
1246 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1250 shmem_unacct_blocks(info->flags, 1);
1251 shmem_free_blocks(inode, 1);
1256 spin_lock(&info->lock);
1257 entry = shmem_swp_alloc(info, idx, sgp);
1259 error = PTR_ERR(entry);
1262 shmem_swp_unmap(entry);
1264 if (error || swap.val || 0 != add_to_page_cache_lru(
1265 filepage, mapping, idx, GFP_ATOMIC)) {
1266 spin_unlock(&info->lock);
1267 page_cache_release(filepage);
1268 shmem_unacct_blocks(info->flags, 1);
1269 shmem_free_blocks(inode, 1);
1275 info->flags |= SHMEM_PAGEIN;
1279 spin_unlock(&info->lock);
1280 flush_dcache_page(filepage);
1281 SetPageUptodate(filepage);
1284 if (*pagep != filepage) {
1285 unlock_page(filepage);
1291 if (*pagep != filepage) {
1292 unlock_page(filepage);
1293 page_cache_release(filepage);
1298 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1300 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1301 struct page *page = NULL;
1305 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1306 idx += vma->vm_pgoff;
1307 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1308 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1309 return NOPAGE_SIGBUS;
1311 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1313 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1315 mark_page_accessed(page);
1319 static int shmem_populate(struct vm_area_struct *vma,
1320 unsigned long addr, unsigned long len,
1321 pgprot_t prot, unsigned long pgoff, int nonblock)
1323 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1324 struct mm_struct *mm = vma->vm_mm;
1325 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1328 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1329 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1332 while ((long) len > 0) {
1333 struct page *page = NULL;
1336 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1338 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1341 /* Page may still be null, but only if nonblock was set. */
1343 mark_page_accessed(page);
1344 err = install_page(mm, vma, addr, page, prot);
1346 page_cache_release(page);
1349 } else if (vma->vm_flags & VM_NONLINEAR) {
1350 /* No page was found just because we can't read it in
1351 * now (being here implies nonblock != 0), but the page
1352 * may exist, so set the PTE to fault it in later. */
1353 err = install_file_pte(mm, vma, addr, pgoff, prot);
1366 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1368 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1369 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1373 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1375 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1378 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1379 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1383 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1385 struct inode *inode = file->f_path.dentry->d_inode;
1386 struct shmem_inode_info *info = SHMEM_I(inode);
1387 int retval = -ENOMEM;
1389 spin_lock(&info->lock);
1390 if (lock && !(info->flags & VM_LOCKED)) {
1391 if (!user_shm_lock(inode->i_size, user))
1393 info->flags |= VM_LOCKED;
1395 if (!lock && (info->flags & VM_LOCKED) && user) {
1396 user_shm_unlock(inode->i_size, user);
1397 info->flags &= ~VM_LOCKED;
1401 spin_unlock(&info->lock);
1405 int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1407 file_accessed(file);
1408 vma->vm_ops = &shmem_vm_ops;
1412 static struct inode *
1413 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1415 struct inode *inode;
1416 struct shmem_inode_info *info;
1417 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1419 if (sbinfo->max_inodes) {
1420 spin_lock(&sbinfo->stat_lock);
1421 if (!sbinfo->free_inodes) {
1422 spin_unlock(&sbinfo->stat_lock);
1425 sbinfo->free_inodes--;
1426 spin_unlock(&sbinfo->stat_lock);
1429 inode = new_inode(sb);
1431 inode->i_mode = mode;
1432 inode->i_uid = current->fsuid;
1433 inode->i_gid = current->fsgid;
1434 inode->i_blocks = 0;
1435 inode->i_mapping->a_ops = &shmem_aops;
1436 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1437 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1438 inode->i_generation = get_seconds();
1439 info = SHMEM_I(inode);
1440 memset(info, 0, (char *)inode - (char *)info);
1441 spin_lock_init(&info->lock);
1442 INIT_LIST_HEAD(&info->swaplist);
1444 switch (mode & S_IFMT) {
1446 inode->i_op = &shmem_special_inode_operations;
1447 init_special_inode(inode, mode, dev);
1450 inode->i_op = &shmem_inode_operations;
1451 inode->i_fop = &shmem_file_operations;
1452 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1453 &sbinfo->policy_nodes);
1457 /* Some things misbehave if size == 0 on a directory */
1458 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1459 inode->i_op = &shmem_dir_inode_operations;
1460 inode->i_fop = &simple_dir_operations;
1464 * Must not load anything in the rbtree,
1465 * mpol_free_shared_policy will not be called.
1467 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1471 } else if (sbinfo->max_inodes) {
1472 spin_lock(&sbinfo->stat_lock);
1473 sbinfo->free_inodes++;
1474 spin_unlock(&sbinfo->stat_lock);
1480 static struct inode_operations shmem_symlink_inode_operations;
1481 static struct inode_operations shmem_symlink_inline_operations;
1484 * Normally tmpfs makes no use of shmem_prepare_write, but it
1485 * lets a tmpfs file be used read-write below the loop driver.
1488 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1490 struct inode *inode = page->mapping->host;
1491 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1495 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1497 struct inode *inode = file->f_path.dentry->d_inode;
1499 unsigned long written;
1502 if ((ssize_t) count < 0)
1505 if (!access_ok(VERIFY_READ, buf, count))
1508 mutex_lock(&inode->i_mutex);
1513 err = generic_write_checks(file, &pos, &count, 0);
1517 err = remove_suid(file->f_path.dentry);
1521 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1524 struct page *page = NULL;
1525 unsigned long bytes, index, offset;
1529 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1530 index = pos >> PAGE_CACHE_SHIFT;
1531 bytes = PAGE_CACHE_SIZE - offset;
1536 * We don't hold page lock across copy from user -
1537 * what would it guard against? - so no deadlock here.
1538 * But it still may be a good idea to prefault below.
1541 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1546 if (PageHighMem(page)) {
1547 volatile unsigned char dummy;
1548 __get_user(dummy, buf);
1549 __get_user(dummy, buf + bytes - 1);
1551 kaddr = kmap_atomic(page, KM_USER0);
1552 left = __copy_from_user_inatomic(kaddr + offset,
1554 kunmap_atomic(kaddr, KM_USER0);
1558 left = __copy_from_user(kaddr + offset, buf, bytes);
1566 if (pos > inode->i_size)
1567 i_size_write(inode, pos);
1569 flush_dcache_page(page);
1570 set_page_dirty(page);
1571 mark_page_accessed(page);
1572 page_cache_release(page);
1582 * Our dirty pages are not counted in nr_dirty,
1583 * and we do not attempt to balance dirty pages.
1593 mutex_unlock(&inode->i_mutex);
1597 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1599 struct inode *inode = filp->f_path.dentry->d_inode;
1600 struct address_space *mapping = inode->i_mapping;
1601 unsigned long index, offset;
1603 index = *ppos >> PAGE_CACHE_SHIFT;
1604 offset = *ppos & ~PAGE_CACHE_MASK;
1607 struct page *page = NULL;
1608 unsigned long end_index, nr, ret;
1609 loff_t i_size = i_size_read(inode);
1611 end_index = i_size >> PAGE_CACHE_SHIFT;
1612 if (index > end_index)
1614 if (index == end_index) {
1615 nr = i_size & ~PAGE_CACHE_MASK;
1620 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1622 if (desc->error == -EINVAL)
1628 * We must evaluate after, since reads (unlike writes)
1629 * are called without i_mutex protection against truncate
1631 nr = PAGE_CACHE_SIZE;
1632 i_size = i_size_read(inode);
1633 end_index = i_size >> PAGE_CACHE_SHIFT;
1634 if (index == end_index) {
1635 nr = i_size & ~PAGE_CACHE_MASK;
1638 page_cache_release(page);
1646 * If users can be writing to this page using arbitrary
1647 * virtual addresses, take care about potential aliasing
1648 * before reading the page on the kernel side.
1650 if (mapping_writably_mapped(mapping))
1651 flush_dcache_page(page);
1653 * Mark the page accessed if we read the beginning.
1656 mark_page_accessed(page);
1658 page = ZERO_PAGE(0);
1659 page_cache_get(page);
1663 * Ok, we have the page, and it's up-to-date, so
1664 * now we can copy it to user space...
1666 * The actor routine returns how many bytes were actually used..
1667 * NOTE! This may not be the same as how much of a user buffer
1668 * we filled up (we may be padding etc), so we can only update
1669 * "pos" here (the actor routine has to update the user buffer
1670 * pointers and the remaining count).
1672 ret = actor(desc, page, offset, nr);
1674 index += offset >> PAGE_CACHE_SHIFT;
1675 offset &= ~PAGE_CACHE_MASK;
1677 page_cache_release(page);
1678 if (ret != nr || !desc->count)
1684 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1685 file_accessed(filp);
1688 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1690 read_descriptor_t desc;
1692 if ((ssize_t) count < 0)
1694 if (!access_ok(VERIFY_WRITE, buf, count))
1704 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1706 return desc.written;
1710 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1711 size_t count, read_actor_t actor, void *target)
1713 read_descriptor_t desc;
1720 desc.arg.data = target;
1723 do_shmem_file_read(in_file, ppos, &desc, actor);
1725 return desc.written;
1729 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1731 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1733 buf->f_type = TMPFS_SUPER_MAGIC;
1734 buf->f_bsize = PAGE_CACHE_SIZE;
1735 buf->f_namelen = NAME_MAX;
1736 spin_lock(&sbinfo->stat_lock);
1737 if (sbinfo->max_blocks) {
1738 buf->f_blocks = sbinfo->max_blocks;
1739 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1741 if (sbinfo->max_inodes) {
1742 buf->f_files = sbinfo->max_inodes;
1743 buf->f_ffree = sbinfo->free_inodes;
1745 /* else leave those fields 0 like simple_statfs */
1746 spin_unlock(&sbinfo->stat_lock);
1751 * File creation. Allocate an inode, and we're done..
1754 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1756 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1757 int error = -ENOSPC;
1760 error = security_inode_init_security(inode, dir, NULL, NULL,
1763 if (error != -EOPNOTSUPP) {
1768 error = shmem_acl_init(inode, dir);
1773 if (dir->i_mode & S_ISGID) {
1774 inode->i_gid = dir->i_gid;
1776 inode->i_mode |= S_ISGID;
1778 dir->i_size += BOGO_DIRENT_SIZE;
1779 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1780 d_instantiate(dentry, inode);
1781 dget(dentry); /* Extra count - pin the dentry in core */
1786 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1790 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1796 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1797 struct nameidata *nd)
1799 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1805 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1807 struct inode *inode = old_dentry->d_inode;
1808 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1811 * No ordinary (disk based) filesystem counts links as inodes;
1812 * but each new link needs a new dentry, pinning lowmem, and
1813 * tmpfs dentries cannot be pruned until they are unlinked.
1815 if (sbinfo->max_inodes) {
1816 spin_lock(&sbinfo->stat_lock);
1817 if (!sbinfo->free_inodes) {
1818 spin_unlock(&sbinfo->stat_lock);
1821 sbinfo->free_inodes--;
1822 spin_unlock(&sbinfo->stat_lock);
1825 dir->i_size += BOGO_DIRENT_SIZE;
1826 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1828 atomic_inc(&inode->i_count); /* New dentry reference */
1829 dget(dentry); /* Extra pinning count for the created dentry */
1830 d_instantiate(dentry, inode);
1834 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1836 struct inode *inode = dentry->d_inode;
1838 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1839 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1840 if (sbinfo->max_inodes) {
1841 spin_lock(&sbinfo->stat_lock);
1842 sbinfo->free_inodes++;
1843 spin_unlock(&sbinfo->stat_lock);
1847 dir->i_size -= BOGO_DIRENT_SIZE;
1848 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1850 dput(dentry); /* Undo the count from "create" - this does all the work */
1854 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1856 if (!simple_empty(dentry))
1859 drop_nlink(dentry->d_inode);
1861 return shmem_unlink(dir, dentry);
1865 * The VFS layer already does all the dentry stuff for rename,
1866 * we just have to decrement the usage count for the target if
1867 * it exists so that the VFS layer correctly free's it when it
1870 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1872 struct inode *inode = old_dentry->d_inode;
1873 int they_are_dirs = S_ISDIR(inode->i_mode);
1875 if (!simple_empty(new_dentry))
1878 if (new_dentry->d_inode) {
1879 (void) shmem_unlink(new_dir, new_dentry);
1881 drop_nlink(old_dir);
1882 } else if (they_are_dirs) {
1883 drop_nlink(old_dir);
1887 old_dir->i_size -= BOGO_DIRENT_SIZE;
1888 new_dir->i_size += BOGO_DIRENT_SIZE;
1889 old_dir->i_ctime = old_dir->i_mtime =
1890 new_dir->i_ctime = new_dir->i_mtime =
1891 inode->i_ctime = CURRENT_TIME;
1895 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1899 struct inode *inode;
1900 struct page *page = NULL;
1902 struct shmem_inode_info *info;
1904 len = strlen(symname) + 1;
1905 if (len > PAGE_CACHE_SIZE)
1906 return -ENAMETOOLONG;
1908 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1912 error = security_inode_init_security(inode, dir, NULL, NULL,
1915 if (error != -EOPNOTSUPP) {
1922 info = SHMEM_I(inode);
1923 inode->i_size = len-1;
1924 if (len <= (char *)inode - (char *)info) {
1926 memcpy(info, symname, len);
1927 inode->i_op = &shmem_symlink_inline_operations;
1929 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1934 inode->i_op = &shmem_symlink_inode_operations;
1935 kaddr = kmap_atomic(page, KM_USER0);
1936 memcpy(kaddr, symname, len);
1937 kunmap_atomic(kaddr, KM_USER0);
1938 set_page_dirty(page);
1939 page_cache_release(page);
1941 if (dir->i_mode & S_ISGID)
1942 inode->i_gid = dir->i_gid;
1943 dir->i_size += BOGO_DIRENT_SIZE;
1944 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1945 d_instantiate(dentry, inode);
1950 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1952 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1956 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1958 struct page *page = NULL;
1959 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1960 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1964 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1966 if (!IS_ERR(nd_get_link(nd))) {
1967 struct page *page = cookie;
1969 mark_page_accessed(page);
1970 page_cache_release(page);
1974 static struct inode_operations shmem_symlink_inline_operations = {
1975 .readlink = generic_readlink,
1976 .follow_link = shmem_follow_link_inline,
1979 static struct inode_operations shmem_symlink_inode_operations = {
1980 .truncate = shmem_truncate,
1981 .readlink = generic_readlink,
1982 .follow_link = shmem_follow_link,
1983 .put_link = shmem_put_link,
1986 #ifdef CONFIG_TMPFS_POSIX_ACL
1988 * Superblocks without xattr inode operations will get security.* xattr
1989 * support from the VFS "for free". As soon as we have any other xattrs
1990 * like ACLs, we also need to implement the security.* handlers at
1991 * filesystem level, though.
1994 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1995 size_t list_len, const char *name,
1998 return security_inode_listsecurity(inode, list, list_len);
2001 static int shmem_xattr_security_get(struct inode *inode, const char *name,
2002 void *buffer, size_t size)
2004 if (strcmp(name, "") == 0)
2006 return security_inode_getsecurity(inode, name, buffer, size,
2010 static int shmem_xattr_security_set(struct inode *inode, const char *name,
2011 const void *value, size_t size, int flags)
2013 if (strcmp(name, "") == 0)
2015 return security_inode_setsecurity(inode, name, value, size, flags);
2018 static struct xattr_handler shmem_xattr_security_handler = {
2019 .prefix = XATTR_SECURITY_PREFIX,
2020 .list = shmem_xattr_security_list,
2021 .get = shmem_xattr_security_get,
2022 .set = shmem_xattr_security_set,
2025 static struct xattr_handler *shmem_xattr_handlers[] = {
2026 &shmem_xattr_acl_access_handler,
2027 &shmem_xattr_acl_default_handler,
2028 &shmem_xattr_security_handler,
2033 static struct dentry *shmem_get_parent(struct dentry *child)
2035 return ERR_PTR(-ESTALE);
2038 static int shmem_match(struct inode *ino, void *vfh)
2042 inum = (inum << 32) | fh[1];
2043 return ino->i_ino == inum && fh[0] == ino->i_generation;
2046 static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
2048 struct dentry *de = NULL;
2049 struct inode *inode;
2052 inum = (inum << 32) | fh[1];
2054 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
2056 de = d_find_alias(inode);
2060 return de? de: ERR_PTR(-ESTALE);
2063 static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
2065 int (*acceptable)(void *context, struct dentry *de),
2069 return ERR_PTR(-ESTALE);
2071 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
2075 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2078 struct inode *inode = dentry->d_inode;
2083 if (hlist_unhashed(&inode->i_hash)) {
2084 /* Unfortunately insert_inode_hash is not idempotent,
2085 * so as we hash inodes here rather than at creation
2086 * time, we need a lock to ensure we only try
2089 static DEFINE_SPINLOCK(lock);
2091 if (hlist_unhashed(&inode->i_hash))
2092 __insert_inode_hash(inode,
2093 inode->i_ino + inode->i_generation);
2097 fh[0] = inode->i_generation;
2098 fh[1] = inode->i_ino;
2099 fh[2] = ((__u64)inode->i_ino) >> 32;
2105 static struct export_operations shmem_export_ops = {
2106 .get_parent = shmem_get_parent,
2107 .get_dentry = shmem_get_dentry,
2108 .encode_fh = shmem_encode_fh,
2109 .decode_fh = shmem_decode_fh,
2112 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2113 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2114 int *policy, nodemask_t *policy_nodes)
2116 char *this_char, *value, *rest;
2118 while (options != NULL) {
2119 this_char = options;
2122 * NUL-terminate this option: unfortunately,
2123 * mount options form a comma-separated list,
2124 * but mpol's nodelist may also contain commas.
2126 options = strchr(options, ',');
2127 if (options == NULL)
2130 if (!isdigit(*options)) {
2137 if ((value = strchr(this_char,'=')) != NULL) {
2141 "tmpfs: No value for mount option '%s'\n",
2146 if (!strcmp(this_char,"size")) {
2147 unsigned long long size;
2148 size = memparse(value,&rest);
2150 size <<= PAGE_SHIFT;
2151 size *= totalram_pages;
2157 *blocks = size >> PAGE_CACHE_SHIFT;
2158 } else if (!strcmp(this_char,"nr_blocks")) {
2159 *blocks = memparse(value,&rest);
2162 } else if (!strcmp(this_char,"nr_inodes")) {
2163 *inodes = memparse(value,&rest);
2166 } else if (!strcmp(this_char,"mode")) {
2169 *mode = simple_strtoul(value,&rest,8);
2172 } else if (!strcmp(this_char,"uid")) {
2175 *uid = simple_strtoul(value,&rest,0);
2178 } else if (!strcmp(this_char,"gid")) {
2181 *gid = simple_strtoul(value,&rest,0);
2184 } else if (!strcmp(this_char,"mpol")) {
2185 if (shmem_parse_mpol(value,policy,policy_nodes))
2188 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2196 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2202 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2204 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2205 unsigned long max_blocks = sbinfo->max_blocks;
2206 unsigned long max_inodes = sbinfo->max_inodes;
2207 int policy = sbinfo->policy;
2208 nodemask_t policy_nodes = sbinfo->policy_nodes;
2209 unsigned long blocks;
2210 unsigned long inodes;
2211 int error = -EINVAL;
2213 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2214 &max_inodes, &policy, &policy_nodes))
2217 spin_lock(&sbinfo->stat_lock);
2218 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2219 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2220 if (max_blocks < blocks)
2222 if (max_inodes < inodes)
2225 * Those tests also disallow limited->unlimited while any are in
2226 * use, so i_blocks will always be zero when max_blocks is zero;
2227 * but we must separately disallow unlimited->limited, because
2228 * in that case we have no record of how much is already in use.
2230 if (max_blocks && !sbinfo->max_blocks)
2232 if (max_inodes && !sbinfo->max_inodes)
2236 sbinfo->max_blocks = max_blocks;
2237 sbinfo->free_blocks = max_blocks - blocks;
2238 sbinfo->max_inodes = max_inodes;
2239 sbinfo->free_inodes = max_inodes - inodes;
2240 sbinfo->policy = policy;
2241 sbinfo->policy_nodes = policy_nodes;
2243 spin_unlock(&sbinfo->stat_lock);
2248 static void shmem_put_super(struct super_block *sb)
2250 kfree(sb->s_fs_info);
2251 sb->s_fs_info = NULL;
2254 static int shmem_fill_super(struct super_block *sb,
2255 void *data, int silent)
2257 struct inode *inode;
2258 struct dentry *root;
2259 int mode = S_IRWXUGO | S_ISVTX;
2260 uid_t uid = current->fsuid;
2261 gid_t gid = current->fsgid;
2263 struct shmem_sb_info *sbinfo;
2264 unsigned long blocks = 0;
2265 unsigned long inodes = 0;
2266 int policy = MPOL_DEFAULT;
2267 nodemask_t policy_nodes = node_online_map;
2271 * Per default we only allow half of the physical ram per
2272 * tmpfs instance, limiting inodes to one per page of lowmem;
2273 * but the internal instance is left unlimited.
2275 if (!(sb->s_flags & MS_NOUSER)) {
2276 blocks = totalram_pages / 2;
2277 inodes = totalram_pages - totalhigh_pages;
2278 if (inodes > blocks)
2280 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2281 &inodes, &policy, &policy_nodes))
2284 sb->s_export_op = &shmem_export_ops;
2286 sb->s_flags |= MS_NOUSER;
2289 /* Round up to L1_CACHE_BYTES to resist false sharing */
2290 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2291 L1_CACHE_BYTES), GFP_KERNEL);
2295 spin_lock_init(&sbinfo->stat_lock);
2296 sbinfo->max_blocks = blocks;
2297 sbinfo->free_blocks = blocks;
2298 sbinfo->max_inodes = inodes;
2299 sbinfo->free_inodes = inodes;
2300 sbinfo->policy = policy;
2301 sbinfo->policy_nodes = policy_nodes;
2303 sb->s_fs_info = sbinfo;
2304 sb->s_maxbytes = SHMEM_MAX_BYTES;
2305 sb->s_blocksize = PAGE_CACHE_SIZE;
2306 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2307 sb->s_magic = TMPFS_SUPER_MAGIC;
2308 sb->s_op = &shmem_ops;
2309 sb->s_time_gran = 1;
2310 #ifdef CONFIG_TMPFS_POSIX_ACL
2311 sb->s_xattr = shmem_xattr_handlers;
2312 sb->s_flags |= MS_POSIXACL;
2315 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2320 root = d_alloc_root(inode);
2329 shmem_put_super(sb);
2333 static struct kmem_cache *shmem_inode_cachep;
2335 static struct inode *shmem_alloc_inode(struct super_block *sb)
2337 struct shmem_inode_info *p;
2338 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2341 return &p->vfs_inode;
2344 static void shmem_destroy_inode(struct inode *inode)
2346 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2347 /* only struct inode is valid if it's an inline symlink */
2348 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2350 shmem_acl_destroy_inode(inode);
2351 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2354 static void init_once(void *foo, struct kmem_cache *cachep,
2355 unsigned long flags)
2357 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2359 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2360 SLAB_CTOR_CONSTRUCTOR) {
2361 inode_init_once(&p->vfs_inode);
2362 #ifdef CONFIG_TMPFS_POSIX_ACL
2364 p->i_default_acl = NULL;
2369 static int init_inodecache(void)
2371 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2372 sizeof(struct shmem_inode_info),
2373 0, 0, init_once, NULL);
2374 if (shmem_inode_cachep == NULL)
2379 static void destroy_inodecache(void)
2381 kmem_cache_destroy(shmem_inode_cachep);
2384 static const struct address_space_operations shmem_aops = {
2385 .writepage = shmem_writepage,
2386 .set_page_dirty = __set_page_dirty_nobuffers,
2388 .prepare_write = shmem_prepare_write,
2389 .commit_write = simple_commit_write,
2391 .migratepage = migrate_page,
2394 static const struct file_operations shmem_file_operations = {
2397 .llseek = generic_file_llseek,
2398 .read = shmem_file_read,
2399 .write = shmem_file_write,
2400 .fsync = simple_sync_file,
2401 .sendfile = shmem_file_sendfile,
2405 static struct inode_operations shmem_inode_operations = {
2406 .truncate = shmem_truncate,
2407 .setattr = shmem_notify_change,
2408 .truncate_range = shmem_truncate_range,
2409 #ifdef CONFIG_TMPFS_POSIX_ACL
2410 .setxattr = generic_setxattr,
2411 .getxattr = generic_getxattr,
2412 .listxattr = generic_listxattr,
2413 .removexattr = generic_removexattr,
2414 .permission = shmem_permission,
2419 static struct inode_operations shmem_dir_inode_operations = {
2421 .create = shmem_create,
2422 .lookup = simple_lookup,
2424 .unlink = shmem_unlink,
2425 .symlink = shmem_symlink,
2426 .mkdir = shmem_mkdir,
2427 .rmdir = shmem_rmdir,
2428 .mknod = shmem_mknod,
2429 .rename = shmem_rename,
2431 #ifdef CONFIG_TMPFS_POSIX_ACL
2432 .setattr = shmem_notify_change,
2433 .setxattr = generic_setxattr,
2434 .getxattr = generic_getxattr,
2435 .listxattr = generic_listxattr,
2436 .removexattr = generic_removexattr,
2437 .permission = shmem_permission,
2441 static struct inode_operations shmem_special_inode_operations = {
2442 #ifdef CONFIG_TMPFS_POSIX_ACL
2443 .setattr = shmem_notify_change,
2444 .setxattr = generic_setxattr,
2445 .getxattr = generic_getxattr,
2446 .listxattr = generic_listxattr,
2447 .removexattr = generic_removexattr,
2448 .permission = shmem_permission,
2452 static struct super_operations shmem_ops = {
2453 .alloc_inode = shmem_alloc_inode,
2454 .destroy_inode = shmem_destroy_inode,
2456 .statfs = shmem_statfs,
2457 .remount_fs = shmem_remount_fs,
2459 .delete_inode = shmem_delete_inode,
2460 .drop_inode = generic_delete_inode,
2461 .put_super = shmem_put_super,
2464 static struct vm_operations_struct shmem_vm_ops = {
2465 .nopage = shmem_nopage,
2466 .populate = shmem_populate,
2468 .set_policy = shmem_set_policy,
2469 .get_policy = shmem_get_policy,
2474 static int shmem_get_sb(struct file_system_type *fs_type,
2475 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2477 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2480 static struct file_system_type tmpfs_fs_type = {
2481 .owner = THIS_MODULE,
2483 .get_sb = shmem_get_sb,
2484 .kill_sb = kill_litter_super,
2486 static struct vfsmount *shm_mnt;
2488 static int __init init_tmpfs(void)
2492 error = init_inodecache();
2496 error = register_filesystem(&tmpfs_fs_type);
2498 printk(KERN_ERR "Could not register tmpfs\n");
2502 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2503 tmpfs_fs_type.name, NULL);
2504 if (IS_ERR(shm_mnt)) {
2505 error = PTR_ERR(shm_mnt);
2506 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2512 unregister_filesystem(&tmpfs_fs_type);
2514 destroy_inodecache();
2516 shm_mnt = ERR_PTR(error);
2519 module_init(init_tmpfs)
2522 * shmem_file_setup - get an unlinked file living in tmpfs
2524 * @name: name for dentry (to be seen in /proc/<pid>/maps
2525 * @size: size to be set for the file
2528 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2532 struct inode *inode;
2533 struct dentry *dentry, *root;
2536 if (IS_ERR(shm_mnt))
2537 return (void *)shm_mnt;
2539 if (size < 0 || size > SHMEM_MAX_BYTES)
2540 return ERR_PTR(-EINVAL);
2542 if (shmem_acct_size(flags, size))
2543 return ERR_PTR(-ENOMEM);
2547 this.len = strlen(name);
2548 this.hash = 0; /* will go */
2549 root = shm_mnt->mnt_root;
2550 dentry = d_alloc(root, &this);
2555 file = get_empty_filp();
2560 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2564 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2565 d_instantiate(dentry, inode);
2566 inode->i_size = size;
2567 inode->i_nlink = 0; /* It is unlinked */
2568 file->f_path.mnt = mntget(shm_mnt);
2569 file->f_path.dentry = dentry;
2570 file->f_mapping = inode->i_mapping;
2571 file->f_op = &shmem_file_operations;
2572 file->f_mode = FMODE_WRITE | FMODE_READ;
2580 shmem_unacct_size(flags, size);
2581 return ERR_PTR(error);
2585 * shmem_zero_setup - setup a shared anonymous mapping
2587 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2589 int shmem_zero_setup(struct vm_area_struct *vma)
2592 loff_t size = vma->vm_end - vma->vm_start;
2594 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2596 return PTR_ERR(file);
2600 vma->vm_file = file;
2601 vma->vm_ops = &shmem_vm_ops;