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/exportfs.h>
31 #include <linux/generic_acl.h>
33 #include <linux/mman.h>
34 #include <linux/file.h>
35 #include <linux/swap.h>
36 #include <linux/pagemap.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/backing-dev.h>
40 #include <linux/shmem_fs.h>
41 #include <linux/mount.h>
42 #include <linux/writeback.h>
43 #include <linux/vfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/security.h>
46 #include <linux/swapops.h>
47 #include <linux/mempolicy.h>
48 #include <linux/namei.h>
49 #include <linux/ctype.h>
50 #include <linux/migrate.h>
51 #include <linux/highmem.h>
52 #include <linux/backing-dev.h>
54 #include <asm/uaccess.h>
55 #include <asm/div64.h>
56 #include <asm/pgtable.h>
58 /* This magic number is used in glibc for posix shared memory */
59 #define TMPFS_MAGIC 0x01021994
61 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
62 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
63 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
65 #define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
66 #define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
68 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
70 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
71 #define SHMEM_PAGEIN VM_READ
72 #define SHMEM_TRUNCATE VM_WRITE
74 /* Definition to limit shmem_truncate's steps between cond_rescheds */
75 #define LATENCY_LIMIT 64
77 /* Pretend that each entry is of this size in directory's i_size */
78 #define BOGO_DIRENT_SIZE 20
80 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
82 SGP_QUICK, /* don't try more than file page cache lookup */
83 SGP_READ, /* don't exceed i_size, don't allocate page */
84 SGP_CACHE, /* don't exceed i_size, may allocate page */
85 SGP_WRITE, /* may exceed i_size, may allocate page */
86 SGP_NOPAGE, /* same as SGP_CACHE, return with page locked */
89 static int shmem_getpage(struct inode *inode, unsigned long idx,
90 struct page **pagep, enum sgp_type sgp, int *type);
92 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
95 * The above definition of ENTRIES_PER_PAGE, and the use of
96 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
97 * might be reconsidered if it ever diverges from PAGE_SIZE.
99 * __GFP_MOVABLE is masked out as swap vectors cannot move
101 return alloc_pages((gfp_mask & ~__GFP_MOVABLE) | __GFP_ZERO,
102 PAGE_CACHE_SHIFT-PAGE_SHIFT);
105 static inline void shmem_dir_free(struct page *page)
107 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
110 static struct page **shmem_dir_map(struct page *page)
112 return (struct page **)kmap_atomic(page, KM_USER0);
115 static inline void shmem_dir_unmap(struct page **dir)
117 kunmap_atomic(dir, KM_USER0);
120 static swp_entry_t *shmem_swp_map(struct page *page)
122 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
125 static inline void shmem_swp_balance_unmap(void)
128 * When passing a pointer to an i_direct entry, to code which
129 * also handles indirect entries and so will shmem_swp_unmap,
130 * we must arrange for the preempt count to remain in balance.
131 * What kmap_atomic of a lowmem page does depends on config
132 * and architecture, so pretend to kmap_atomic some lowmem page.
134 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
137 static inline void shmem_swp_unmap(swp_entry_t *entry)
139 kunmap_atomic(entry, KM_USER1);
142 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
144 return sb->s_fs_info;
148 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
149 * for shared memory and for shared anonymous (/dev/zero) mappings
150 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
151 * consistent with the pre-accounting of private mappings ...
153 static inline int shmem_acct_size(unsigned long flags, loff_t size)
155 return (flags & VM_ACCOUNT)?
156 security_vm_enough_memory(VM_ACCT(size)): 0;
159 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
161 if (flags & VM_ACCOUNT)
162 vm_unacct_memory(VM_ACCT(size));
166 * ... whereas tmpfs objects are accounted incrementally as
167 * pages are allocated, in order to allow huge sparse files.
168 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
169 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
171 static inline int shmem_acct_block(unsigned long flags)
173 return (flags & VM_ACCOUNT)?
174 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
177 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
179 if (!(flags & VM_ACCOUNT))
180 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
183 static const struct super_operations shmem_ops;
184 static const struct address_space_operations shmem_aops;
185 static const struct file_operations shmem_file_operations;
186 static const struct inode_operations shmem_inode_operations;
187 static const struct inode_operations shmem_dir_inode_operations;
188 static const struct inode_operations shmem_special_inode_operations;
189 static struct vm_operations_struct shmem_vm_ops;
191 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
192 .ra_pages = 0, /* No readahead */
193 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
194 .unplug_io_fn = default_unplug_io_fn,
197 static LIST_HEAD(shmem_swaplist);
198 static DEFINE_SPINLOCK(shmem_swaplist_lock);
200 static void shmem_free_blocks(struct inode *inode, long pages)
202 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
203 if (sbinfo->max_blocks) {
204 spin_lock(&sbinfo->stat_lock);
205 sbinfo->free_blocks += pages;
206 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
207 spin_unlock(&sbinfo->stat_lock);
212 * shmem_recalc_inode - recalculate the size of an inode
214 * @inode: inode to recalc
216 * We have to calculate the free blocks since the mm can drop
217 * undirtied hole pages behind our back.
219 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
220 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
222 * It has to be called with the spinlock held.
224 static void shmem_recalc_inode(struct inode *inode)
226 struct shmem_inode_info *info = SHMEM_I(inode);
229 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
231 info->alloced -= freed;
232 shmem_unacct_blocks(info->flags, freed);
233 shmem_free_blocks(inode, freed);
238 * shmem_swp_entry - find the swap vector position in the info structure
240 * @info: info structure for the inode
241 * @index: index of the page to find
242 * @page: optional page to add to the structure. Has to be preset to
245 * If there is no space allocated yet it will return NULL when
246 * page is NULL, else it will use the page for the needed block,
247 * setting it to NULL on return to indicate that it has been used.
249 * The swap vector is organized the following way:
251 * There are SHMEM_NR_DIRECT entries directly stored in the
252 * shmem_inode_info structure. So small files do not need an addional
255 * For pages with index > SHMEM_NR_DIRECT there is the pointer
256 * i_indirect which points to a page which holds in the first half
257 * doubly indirect blocks, in the second half triple indirect blocks:
259 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
260 * following layout (for SHMEM_NR_DIRECT == 16):
262 * i_indirect -> dir --> 16-19
275 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
277 unsigned long offset;
281 if (index < SHMEM_NR_DIRECT) {
282 shmem_swp_balance_unmap();
283 return info->i_direct+index;
285 if (!info->i_indirect) {
287 info->i_indirect = *page;
290 return NULL; /* need another page */
293 index -= SHMEM_NR_DIRECT;
294 offset = index % ENTRIES_PER_PAGE;
295 index /= ENTRIES_PER_PAGE;
296 dir = shmem_dir_map(info->i_indirect);
298 if (index >= ENTRIES_PER_PAGE/2) {
299 index -= ENTRIES_PER_PAGE/2;
300 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
301 index %= ENTRIES_PER_PAGE;
308 shmem_dir_unmap(dir);
309 return NULL; /* need another page */
311 shmem_dir_unmap(dir);
312 dir = shmem_dir_map(subdir);
318 if (!page || !(subdir = *page)) {
319 shmem_dir_unmap(dir);
320 return NULL; /* need a page */
325 shmem_dir_unmap(dir);
326 return shmem_swp_map(subdir) + offset;
329 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
331 long incdec = value? 1: -1;
334 info->swapped += incdec;
335 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
336 struct page *page = kmap_atomic_to_page(entry);
337 set_page_private(page, page_private(page) + incdec);
342 * shmem_swp_alloc - get the position of the swap entry for the page.
343 * If it does not exist allocate the entry.
345 * @info: info structure for the inode
346 * @index: index of the page to find
347 * @sgp: check and recheck i_size? skip allocation?
349 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
351 struct inode *inode = &info->vfs_inode;
352 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
353 struct page *page = NULL;
356 if (sgp != SGP_WRITE &&
357 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
358 return ERR_PTR(-EINVAL);
360 while (!(entry = shmem_swp_entry(info, index, &page))) {
362 return shmem_swp_map(ZERO_PAGE(0));
364 * Test free_blocks against 1 not 0, since we have 1 data
365 * page (and perhaps indirect index pages) yet to allocate:
366 * a waste to allocate index if we cannot allocate data.
368 if (sbinfo->max_blocks) {
369 spin_lock(&sbinfo->stat_lock);
370 if (sbinfo->free_blocks <= 1) {
371 spin_unlock(&sbinfo->stat_lock);
372 return ERR_PTR(-ENOSPC);
374 sbinfo->free_blocks--;
375 inode->i_blocks += BLOCKS_PER_PAGE;
376 spin_unlock(&sbinfo->stat_lock);
379 spin_unlock(&info->lock);
380 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
382 set_page_private(page, 0);
383 spin_lock(&info->lock);
386 shmem_free_blocks(inode, 1);
387 return ERR_PTR(-ENOMEM);
389 if (sgp != SGP_WRITE &&
390 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
391 entry = ERR_PTR(-EINVAL);
394 if (info->next_index <= index)
395 info->next_index = index + 1;
398 /* another task gave its page, or truncated the file */
399 shmem_free_blocks(inode, 1);
400 shmem_dir_free(page);
402 if (info->next_index <= index && !IS_ERR(entry))
403 info->next_index = index + 1;
408 * shmem_free_swp - free some swap entries in a directory
410 * @dir: pointer to the directory
411 * @edir: pointer after last entry of the directory
412 * @punch_lock: pointer to spinlock when needed for the holepunch case
414 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
415 spinlock_t *punch_lock)
417 spinlock_t *punch_unlock = NULL;
421 for (ptr = dir; ptr < edir; ptr++) {
423 if (unlikely(punch_lock)) {
424 punch_unlock = punch_lock;
426 spin_lock(punch_unlock);
430 free_swap_and_cache(*ptr);
431 *ptr = (swp_entry_t){0};
436 spin_unlock(punch_unlock);
440 static int shmem_map_and_free_swp(struct page *subdir, int offset,
441 int limit, struct page ***dir, spinlock_t *punch_lock)
446 ptr = shmem_swp_map(subdir);
447 for (; offset < limit; offset += LATENCY_LIMIT) {
448 int size = limit - offset;
449 if (size > LATENCY_LIMIT)
450 size = LATENCY_LIMIT;
451 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
453 if (need_resched()) {
454 shmem_swp_unmap(ptr);
456 shmem_dir_unmap(*dir);
460 ptr = shmem_swp_map(subdir);
463 shmem_swp_unmap(ptr);
467 static void shmem_free_pages(struct list_head *next)
473 page = container_of(next, struct page, lru);
475 shmem_dir_free(page);
477 if (freed >= LATENCY_LIMIT) {
484 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
486 struct shmem_inode_info *info = SHMEM_I(inode);
491 unsigned long diroff;
497 LIST_HEAD(pages_to_free);
498 long nr_pages_to_free = 0;
499 long nr_swaps_freed = 0;
503 spinlock_t *needs_lock;
504 spinlock_t *punch_lock;
505 unsigned long upper_limit;
507 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
508 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
509 if (idx >= info->next_index)
512 spin_lock(&info->lock);
513 info->flags |= SHMEM_TRUNCATE;
514 if (likely(end == (loff_t) -1)) {
515 limit = info->next_index;
516 upper_limit = SHMEM_MAX_INDEX;
517 info->next_index = idx;
521 if (end + 1 >= inode->i_size) { /* we may free a little more */
522 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
524 upper_limit = SHMEM_MAX_INDEX;
526 limit = (end + 1) >> PAGE_CACHE_SHIFT;
529 needs_lock = &info->lock;
533 topdir = info->i_indirect;
534 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
535 info->i_indirect = NULL;
537 list_add(&topdir->lru, &pages_to_free);
539 spin_unlock(&info->lock);
541 if (info->swapped && idx < SHMEM_NR_DIRECT) {
542 ptr = info->i_direct;
544 if (size > SHMEM_NR_DIRECT)
545 size = SHMEM_NR_DIRECT;
546 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
550 * If there are no indirect blocks or we are punching a hole
551 * below indirect blocks, nothing to be done.
553 if (!topdir || limit <= SHMEM_NR_DIRECT)
557 * The truncation case has already dropped info->lock, and we're safe
558 * because i_size and next_index have already been lowered, preventing
559 * access beyond. But in the punch_hole case, we still need to take
560 * the lock when updating the swap directory, because there might be
561 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
562 * shmem_writepage. However, whenever we find we can remove a whole
563 * directory page (not at the misaligned start or end of the range),
564 * we first NULLify its pointer in the level above, and then have no
565 * need to take the lock when updating its contents: needs_lock and
566 * punch_lock (either pointing to info->lock or NULL) manage this.
569 upper_limit -= SHMEM_NR_DIRECT;
570 limit -= SHMEM_NR_DIRECT;
571 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
572 offset = idx % ENTRIES_PER_PAGE;
575 dir = shmem_dir_map(topdir);
576 stage = ENTRIES_PER_PAGEPAGE/2;
577 if (idx < ENTRIES_PER_PAGEPAGE/2) {
579 diroff = idx/ENTRIES_PER_PAGE;
581 dir += ENTRIES_PER_PAGE/2;
582 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
584 stage += ENTRIES_PER_PAGEPAGE;
587 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
588 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
589 if (!diroff && !offset && upper_limit >= stage) {
591 spin_lock(needs_lock);
593 spin_unlock(needs_lock);
598 list_add(&middir->lru, &pages_to_free);
600 shmem_dir_unmap(dir);
601 dir = shmem_dir_map(middir);
609 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
610 if (unlikely(idx == stage)) {
611 shmem_dir_unmap(dir);
612 dir = shmem_dir_map(topdir) +
613 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
616 idx += ENTRIES_PER_PAGEPAGE;
620 stage = idx + ENTRIES_PER_PAGEPAGE;
623 needs_lock = &info->lock;
624 if (upper_limit >= stage) {
626 spin_lock(needs_lock);
628 spin_unlock(needs_lock);
633 list_add(&middir->lru, &pages_to_free);
635 shmem_dir_unmap(dir);
637 dir = shmem_dir_map(middir);
640 punch_lock = needs_lock;
641 subdir = dir[diroff];
642 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
644 spin_lock(needs_lock);
646 spin_unlock(needs_lock);
651 list_add(&subdir->lru, &pages_to_free);
653 if (subdir && page_private(subdir) /* has swap entries */) {
655 if (size > ENTRIES_PER_PAGE)
656 size = ENTRIES_PER_PAGE;
657 freed = shmem_map_and_free_swp(subdir,
658 offset, size, &dir, punch_lock);
660 dir = shmem_dir_map(middir);
661 nr_swaps_freed += freed;
662 if (offset || punch_lock) {
663 spin_lock(&info->lock);
664 set_page_private(subdir,
665 page_private(subdir) - freed);
666 spin_unlock(&info->lock);
668 BUG_ON(page_private(subdir) != freed);
673 shmem_dir_unmap(dir);
675 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
677 * Call truncate_inode_pages again: racing shmem_unuse_inode
678 * may have swizzled a page in from swap since vmtruncate or
679 * generic_delete_inode did it, before we lowered next_index.
680 * Also, though shmem_getpage checks i_size before adding to
681 * cache, no recheck after: so fix the narrow window there too.
683 * Recalling truncate_inode_pages_range and unmap_mapping_range
684 * every time for punch_hole (which never got a chance to clear
685 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
686 * yet hardly ever necessary: try to optimize them out later.
688 truncate_inode_pages_range(inode->i_mapping, start, end);
690 unmap_mapping_range(inode->i_mapping, start,
694 spin_lock(&info->lock);
695 info->flags &= ~SHMEM_TRUNCATE;
696 info->swapped -= nr_swaps_freed;
697 if (nr_pages_to_free)
698 shmem_free_blocks(inode, nr_pages_to_free);
699 shmem_recalc_inode(inode);
700 spin_unlock(&info->lock);
703 * Empty swap vector directory pages to be freed?
705 if (!list_empty(&pages_to_free)) {
706 pages_to_free.prev->next = NULL;
707 shmem_free_pages(pages_to_free.next);
711 static void shmem_truncate(struct inode *inode)
713 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
716 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
718 struct inode *inode = dentry->d_inode;
719 struct page *page = NULL;
722 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
723 if (attr->ia_size < inode->i_size) {
725 * If truncating down to a partial page, then
726 * if that page is already allocated, hold it
727 * in memory until the truncation is over, so
728 * truncate_partial_page cannnot miss it were
729 * it assigned to swap.
731 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
732 (void) shmem_getpage(inode,
733 attr->ia_size>>PAGE_CACHE_SHIFT,
734 &page, SGP_READ, NULL);
737 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
738 * detect if any pages might have been added to cache
739 * after truncate_inode_pages. But we needn't bother
740 * if it's being fully truncated to zero-length: the
741 * nrpages check is efficient enough in that case.
744 struct shmem_inode_info *info = SHMEM_I(inode);
745 spin_lock(&info->lock);
746 info->flags &= ~SHMEM_PAGEIN;
747 spin_unlock(&info->lock);
752 error = inode_change_ok(inode, attr);
754 error = inode_setattr(inode, attr);
755 #ifdef CONFIG_TMPFS_POSIX_ACL
756 if (!error && (attr->ia_valid & ATTR_MODE))
757 error = generic_acl_chmod(inode, &shmem_acl_ops);
760 page_cache_release(page);
764 static void shmem_delete_inode(struct inode *inode)
766 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
767 struct shmem_inode_info *info = SHMEM_I(inode);
769 if (inode->i_op->truncate == shmem_truncate) {
770 truncate_inode_pages(inode->i_mapping, 0);
771 shmem_unacct_size(info->flags, inode->i_size);
773 shmem_truncate(inode);
774 if (!list_empty(&info->swaplist)) {
775 spin_lock(&shmem_swaplist_lock);
776 list_del_init(&info->swaplist);
777 spin_unlock(&shmem_swaplist_lock);
780 BUG_ON(inode->i_blocks);
781 if (sbinfo->max_inodes) {
782 spin_lock(&sbinfo->stat_lock);
783 sbinfo->free_inodes++;
784 spin_unlock(&sbinfo->stat_lock);
789 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
793 for (ptr = dir; ptr < edir; ptr++) {
794 if (ptr->val == entry.val)
800 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
813 ptr = info->i_direct;
814 spin_lock(&info->lock);
815 limit = info->next_index;
817 if (size > SHMEM_NR_DIRECT)
818 size = SHMEM_NR_DIRECT;
819 offset = shmem_find_swp(entry, ptr, ptr+size);
821 shmem_swp_balance_unmap();
824 if (!info->i_indirect)
827 dir = shmem_dir_map(info->i_indirect);
828 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
830 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
831 if (unlikely(idx == stage)) {
832 shmem_dir_unmap(dir-1);
833 dir = shmem_dir_map(info->i_indirect) +
834 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
837 idx += ENTRIES_PER_PAGEPAGE;
841 stage = idx + ENTRIES_PER_PAGEPAGE;
843 shmem_dir_unmap(dir);
844 dir = shmem_dir_map(subdir);
847 if (subdir && page_private(subdir)) {
848 ptr = shmem_swp_map(subdir);
850 if (size > ENTRIES_PER_PAGE)
851 size = ENTRIES_PER_PAGE;
852 offset = shmem_find_swp(entry, ptr, ptr+size);
854 shmem_dir_unmap(dir);
857 shmem_swp_unmap(ptr);
861 shmem_dir_unmap(dir-1);
863 spin_unlock(&info->lock);
867 inode = &info->vfs_inode;
868 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
869 info->flags |= SHMEM_PAGEIN;
870 shmem_swp_set(info, ptr + offset, 0);
872 shmem_swp_unmap(ptr);
873 spin_unlock(&info->lock);
875 * Decrement swap count even when the entry is left behind:
876 * try_to_unuse will skip over mms, then reincrement count.
883 * shmem_unuse() search for an eventually swapped out shmem page.
885 int shmem_unuse(swp_entry_t entry, struct page *page)
887 struct list_head *p, *next;
888 struct shmem_inode_info *info;
891 spin_lock(&shmem_swaplist_lock);
892 list_for_each_safe(p, next, &shmem_swaplist) {
893 info = list_entry(p, struct shmem_inode_info, swaplist);
895 list_del_init(&info->swaplist);
896 else if (shmem_unuse_inode(info, entry, page)) {
897 /* move head to start search for next from here */
898 list_move_tail(&shmem_swaplist, &info->swaplist);
903 spin_unlock(&shmem_swaplist_lock);
908 * Move the page from the page cache to the swap cache.
910 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
912 struct shmem_inode_info *info;
913 swp_entry_t *entry, swap;
914 struct address_space *mapping;
918 BUG_ON(!PageLocked(page));
919 BUG_ON(page_mapped(page));
921 mapping = page->mapping;
923 inode = mapping->host;
924 info = SHMEM_I(inode);
925 if (info->flags & VM_LOCKED)
927 swap = get_swap_page();
931 spin_lock(&info->lock);
932 shmem_recalc_inode(inode);
933 if (index >= info->next_index) {
934 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
937 entry = shmem_swp_entry(info, index, NULL);
941 if (move_to_swap_cache(page, swap) == 0) {
942 shmem_swp_set(info, entry, swap.val);
943 shmem_swp_unmap(entry);
944 spin_unlock(&info->lock);
945 if (list_empty(&info->swaplist)) {
946 spin_lock(&shmem_swaplist_lock);
947 /* move instead of add in case we're racing */
948 list_move_tail(&info->swaplist, &shmem_swaplist);
949 spin_unlock(&shmem_swaplist_lock);
955 shmem_swp_unmap(entry);
957 spin_unlock(&info->lock);
960 set_page_dirty(page);
961 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
965 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
967 char *nodelist = strchr(value, ':');
971 /* NUL-terminate policy string */
973 if (nodelist_parse(nodelist, *policy_nodes))
975 if (!nodes_subset(*policy_nodes, node_online_map))
978 if (!strcmp(value, "default")) {
979 *policy = MPOL_DEFAULT;
980 /* Don't allow a nodelist */
983 } else if (!strcmp(value, "prefer")) {
984 *policy = MPOL_PREFERRED;
985 /* Insist on a nodelist of one node only */
987 char *rest = nodelist;
988 while (isdigit(*rest))
993 } else if (!strcmp(value, "bind")) {
995 /* Insist on a nodelist */
998 } else if (!strcmp(value, "interleave")) {
999 *policy = MPOL_INTERLEAVE;
1000 /* Default to nodes online if no nodelist */
1002 *policy_nodes = node_online_map;
1006 /* Restore string for error message */
1012 static struct page *shmem_swapin_async(struct shared_policy *p,
1013 swp_entry_t entry, unsigned long idx)
1016 struct vm_area_struct pvma;
1018 /* Create a pseudo vma that just contains the policy */
1019 memset(&pvma, 0, sizeof(struct vm_area_struct));
1020 pvma.vm_end = PAGE_SIZE;
1021 pvma.vm_pgoff = idx;
1022 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
1023 page = read_swap_cache_async(entry, &pvma, 0);
1024 mpol_free(pvma.vm_policy);
1028 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
1031 struct shared_policy *p = &info->policy;
1034 unsigned long offset;
1036 num = valid_swaphandles(entry, &offset);
1037 for (i = 0; i < num; offset++, i++) {
1038 page = shmem_swapin_async(p,
1039 swp_entry(swp_type(entry), offset), idx);
1042 page_cache_release(page);
1044 lru_add_drain(); /* Push any new pages onto the LRU now */
1045 return shmem_swapin_async(p, entry, idx);
1048 static struct page *
1049 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
1052 struct vm_area_struct pvma;
1055 memset(&pvma, 0, sizeof(struct vm_area_struct));
1056 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1057 pvma.vm_pgoff = idx;
1058 pvma.vm_end = PAGE_SIZE;
1059 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
1060 mpol_free(pvma.vm_policy);
1064 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
1069 static inline struct page *
1070 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
1072 swapin_readahead(entry, 0, NULL);
1073 return read_swap_cache_async(entry, NULL, 0);
1076 static inline struct page *
1077 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
1079 return alloc_page(gfp | __GFP_ZERO);
1084 * shmem_getpage - either get the page from swap or allocate a new one
1086 * If we allocate a new one we do not mark it dirty. That's up to the
1087 * vm. If we swap it in we mark it dirty since we also free the swap
1088 * entry since a page cannot live in both the swap and page cache
1090 static int shmem_getpage(struct inode *inode, unsigned long idx,
1091 struct page **pagep, enum sgp_type sgp, int *type)
1093 struct address_space *mapping = inode->i_mapping;
1094 struct shmem_inode_info *info = SHMEM_I(inode);
1095 struct shmem_sb_info *sbinfo;
1096 struct page *filepage = *pagep;
1097 struct page *swappage;
1102 if (idx >= SHMEM_MAX_INDEX)
1105 * Normally, filepage is NULL on entry, and either found
1106 * uptodate immediately, or allocated and zeroed, or read
1107 * in under swappage, which is then assigned to filepage.
1108 * But shmem_readpage and shmem_prepare_write pass in a locked
1109 * filepage, which may be found not uptodate by other callers
1110 * too, and may need to be copied from the swappage read in.
1114 filepage = find_lock_page(mapping, idx);
1115 if (filepage && PageUptodate(filepage))
1118 if (sgp == SGP_QUICK)
1121 spin_lock(&info->lock);
1122 shmem_recalc_inode(inode);
1123 entry = shmem_swp_alloc(info, idx, sgp);
1124 if (IS_ERR(entry)) {
1125 spin_unlock(&info->lock);
1126 error = PTR_ERR(entry);
1132 /* Look it up and read it in.. */
1133 swappage = lookup_swap_cache(swap);
1135 shmem_swp_unmap(entry);
1136 /* here we actually do the io */
1137 if (type && *type == VM_FAULT_MINOR) {
1138 __count_vm_event(PGMAJFAULT);
1139 *type = VM_FAULT_MAJOR;
1141 spin_unlock(&info->lock);
1142 swappage = shmem_swapin(info, swap, idx);
1144 spin_lock(&info->lock);
1145 entry = shmem_swp_alloc(info, idx, sgp);
1147 error = PTR_ERR(entry);
1149 if (entry->val == swap.val)
1151 shmem_swp_unmap(entry);
1153 spin_unlock(&info->lock);
1158 wait_on_page_locked(swappage);
1159 page_cache_release(swappage);
1163 /* We have to do this with page locked to prevent races */
1164 if (TestSetPageLocked(swappage)) {
1165 shmem_swp_unmap(entry);
1166 spin_unlock(&info->lock);
1167 wait_on_page_locked(swappage);
1168 page_cache_release(swappage);
1171 if (PageWriteback(swappage)) {
1172 shmem_swp_unmap(entry);
1173 spin_unlock(&info->lock);
1174 wait_on_page_writeback(swappage);
1175 unlock_page(swappage);
1176 page_cache_release(swappage);
1179 if (!PageUptodate(swappage)) {
1180 shmem_swp_unmap(entry);
1181 spin_unlock(&info->lock);
1182 unlock_page(swappage);
1183 page_cache_release(swappage);
1189 shmem_swp_set(info, entry, 0);
1190 shmem_swp_unmap(entry);
1191 delete_from_swap_cache(swappage);
1192 spin_unlock(&info->lock);
1193 copy_highpage(filepage, swappage);
1194 unlock_page(swappage);
1195 page_cache_release(swappage);
1196 flush_dcache_page(filepage);
1197 SetPageUptodate(filepage);
1198 set_page_dirty(filepage);
1200 } else if (!(error = move_from_swap_cache(
1201 swappage, idx, mapping))) {
1202 info->flags |= SHMEM_PAGEIN;
1203 shmem_swp_set(info, entry, 0);
1204 shmem_swp_unmap(entry);
1205 spin_unlock(&info->lock);
1206 filepage = swappage;
1209 shmem_swp_unmap(entry);
1210 spin_unlock(&info->lock);
1211 unlock_page(swappage);
1212 page_cache_release(swappage);
1213 if (error == -ENOMEM) {
1214 /* let kswapd refresh zone for GFP_ATOMICs */
1215 congestion_wait(WRITE, HZ/50);
1219 } else if (sgp == SGP_READ && !filepage) {
1220 shmem_swp_unmap(entry);
1221 filepage = find_get_page(mapping, idx);
1223 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1224 spin_unlock(&info->lock);
1225 wait_on_page_locked(filepage);
1226 page_cache_release(filepage);
1230 spin_unlock(&info->lock);
1232 shmem_swp_unmap(entry);
1233 sbinfo = SHMEM_SB(inode->i_sb);
1234 if (sbinfo->max_blocks) {
1235 spin_lock(&sbinfo->stat_lock);
1236 if (sbinfo->free_blocks == 0 ||
1237 shmem_acct_block(info->flags)) {
1238 spin_unlock(&sbinfo->stat_lock);
1239 spin_unlock(&info->lock);
1243 sbinfo->free_blocks--;
1244 inode->i_blocks += BLOCKS_PER_PAGE;
1245 spin_unlock(&sbinfo->stat_lock);
1246 } else if (shmem_acct_block(info->flags)) {
1247 spin_unlock(&info->lock);
1253 spin_unlock(&info->lock);
1254 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1258 shmem_unacct_blocks(info->flags, 1);
1259 shmem_free_blocks(inode, 1);
1264 spin_lock(&info->lock);
1265 entry = shmem_swp_alloc(info, idx, sgp);
1267 error = PTR_ERR(entry);
1270 shmem_swp_unmap(entry);
1272 if (error || swap.val || 0 != add_to_page_cache_lru(
1273 filepage, mapping, idx, GFP_ATOMIC)) {
1274 spin_unlock(&info->lock);
1275 page_cache_release(filepage);
1276 shmem_unacct_blocks(info->flags, 1);
1277 shmem_free_blocks(inode, 1);
1283 info->flags |= SHMEM_PAGEIN;
1287 spin_unlock(&info->lock);
1288 flush_dcache_page(filepage);
1289 SetPageUptodate(filepage);
1292 if (*pagep != filepage) {
1294 if (sgp != SGP_NOPAGE)
1295 unlock_page(filepage);
1301 if (*pagep != filepage) {
1302 unlock_page(filepage);
1303 page_cache_release(filepage);
1308 static struct page *shmem_nopage(struct vm_area_struct *vma,
1309 unsigned long address, int *type)
1311 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1312 struct page *page = NULL;
1316 BUG_ON(!(vma->vm_flags & VM_CAN_INVALIDATE));
1318 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1319 idx += vma->vm_pgoff;
1320 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1321 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1322 return NOPAGE_SIGBUS;
1324 error = shmem_getpage(inode, idx, &page, SGP_NOPAGE, type);
1326 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1328 mark_page_accessed(page);
1332 static int shmem_populate(struct vm_area_struct *vma,
1333 unsigned long addr, unsigned long len,
1334 pgprot_t prot, unsigned long pgoff, int nonblock)
1336 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1337 struct mm_struct *mm = vma->vm_mm;
1338 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1341 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1342 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1345 while ((long) len > 0) {
1346 struct page *page = NULL;
1349 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1351 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1354 /* Page may still be null, but only if nonblock was set. */
1356 mark_page_accessed(page);
1357 err = install_page(mm, vma, addr, page, prot);
1359 page_cache_release(page);
1362 } else if (vma->vm_flags & VM_NONLINEAR) {
1363 /* No page was found just because we can't read it in
1364 * now (being here implies nonblock != 0), but the page
1365 * may exist, so set the PTE to fault it in later. */
1366 err = install_file_pte(mm, vma, addr, pgoff, prot);
1379 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1381 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1382 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1386 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1388 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1391 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1392 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1396 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1398 struct inode *inode = file->f_path.dentry->d_inode;
1399 struct shmem_inode_info *info = SHMEM_I(inode);
1400 int retval = -ENOMEM;
1402 spin_lock(&info->lock);
1403 if (lock && !(info->flags & VM_LOCKED)) {
1404 if (!user_shm_lock(inode->i_size, user))
1406 info->flags |= VM_LOCKED;
1408 if (!lock && (info->flags & VM_LOCKED) && user) {
1409 user_shm_unlock(inode->i_size, user);
1410 info->flags &= ~VM_LOCKED;
1414 spin_unlock(&info->lock);
1418 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1420 file_accessed(file);
1421 vma->vm_ops = &shmem_vm_ops;
1422 vma->vm_flags |= VM_CAN_INVALIDATE;
1426 static struct inode *
1427 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1429 struct inode *inode;
1430 struct shmem_inode_info *info;
1431 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1433 if (sbinfo->max_inodes) {
1434 spin_lock(&sbinfo->stat_lock);
1435 if (!sbinfo->free_inodes) {
1436 spin_unlock(&sbinfo->stat_lock);
1439 sbinfo->free_inodes--;
1440 spin_unlock(&sbinfo->stat_lock);
1443 inode = new_inode(sb);
1445 inode->i_mode = mode;
1446 inode->i_uid = current->fsuid;
1447 inode->i_gid = current->fsgid;
1448 inode->i_blocks = 0;
1449 inode->i_mapping->a_ops = &shmem_aops;
1450 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1451 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1452 inode->i_generation = get_seconds();
1453 info = SHMEM_I(inode);
1454 memset(info, 0, (char *)inode - (char *)info);
1455 spin_lock_init(&info->lock);
1456 INIT_LIST_HEAD(&info->swaplist);
1458 switch (mode & S_IFMT) {
1460 inode->i_op = &shmem_special_inode_operations;
1461 init_special_inode(inode, mode, dev);
1464 inode->i_op = &shmem_inode_operations;
1465 inode->i_fop = &shmem_file_operations;
1466 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1467 &sbinfo->policy_nodes);
1471 /* Some things misbehave if size == 0 on a directory */
1472 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1473 inode->i_op = &shmem_dir_inode_operations;
1474 inode->i_fop = &simple_dir_operations;
1478 * Must not load anything in the rbtree,
1479 * mpol_free_shared_policy will not be called.
1481 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1485 } else if (sbinfo->max_inodes) {
1486 spin_lock(&sbinfo->stat_lock);
1487 sbinfo->free_inodes++;
1488 spin_unlock(&sbinfo->stat_lock);
1494 static const struct inode_operations shmem_symlink_inode_operations;
1495 static const struct inode_operations shmem_symlink_inline_operations;
1498 * Normally tmpfs avoids the use of shmem_readpage and shmem_prepare_write;
1499 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1500 * below the loop driver, in the generic fashion that many filesystems support.
1502 static int shmem_readpage(struct file *file, struct page *page)
1504 struct inode *inode = page->mapping->host;
1505 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1511 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1513 struct inode *inode = page->mapping->host;
1514 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1518 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1520 struct inode *inode = file->f_path.dentry->d_inode;
1522 unsigned long written;
1525 if ((ssize_t) count < 0)
1528 if (!access_ok(VERIFY_READ, buf, count))
1531 mutex_lock(&inode->i_mutex);
1536 err = generic_write_checks(file, &pos, &count, 0);
1540 err = remove_suid(file->f_path.dentry);
1544 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1547 struct page *page = NULL;
1548 unsigned long bytes, index, offset;
1552 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1553 index = pos >> PAGE_CACHE_SHIFT;
1554 bytes = PAGE_CACHE_SIZE - offset;
1559 * We don't hold page lock across copy from user -
1560 * what would it guard against? - so no deadlock here.
1561 * But it still may be a good idea to prefault below.
1564 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1569 if (PageHighMem(page)) {
1570 volatile unsigned char dummy;
1571 __get_user(dummy, buf);
1572 __get_user(dummy, buf + bytes - 1);
1574 kaddr = kmap_atomic(page, KM_USER0);
1575 left = __copy_from_user_inatomic(kaddr + offset,
1577 kunmap_atomic(kaddr, KM_USER0);
1581 left = __copy_from_user(kaddr + offset, buf, bytes);
1589 if (pos > inode->i_size)
1590 i_size_write(inode, pos);
1592 flush_dcache_page(page);
1593 set_page_dirty(page);
1594 mark_page_accessed(page);
1595 page_cache_release(page);
1605 * Our dirty pages are not counted in nr_dirty,
1606 * and we do not attempt to balance dirty pages.
1616 mutex_unlock(&inode->i_mutex);
1620 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1622 struct inode *inode = filp->f_path.dentry->d_inode;
1623 struct address_space *mapping = inode->i_mapping;
1624 unsigned long index, offset;
1626 index = *ppos >> PAGE_CACHE_SHIFT;
1627 offset = *ppos & ~PAGE_CACHE_MASK;
1630 struct page *page = NULL;
1631 unsigned long end_index, nr, ret;
1632 loff_t i_size = i_size_read(inode);
1634 end_index = i_size >> PAGE_CACHE_SHIFT;
1635 if (index > end_index)
1637 if (index == end_index) {
1638 nr = i_size & ~PAGE_CACHE_MASK;
1643 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1645 if (desc->error == -EINVAL)
1651 * We must evaluate after, since reads (unlike writes)
1652 * are called without i_mutex protection against truncate
1654 nr = PAGE_CACHE_SIZE;
1655 i_size = i_size_read(inode);
1656 end_index = i_size >> PAGE_CACHE_SHIFT;
1657 if (index == end_index) {
1658 nr = i_size & ~PAGE_CACHE_MASK;
1661 page_cache_release(page);
1669 * If users can be writing to this page using arbitrary
1670 * virtual addresses, take care about potential aliasing
1671 * before reading the page on the kernel side.
1673 if (mapping_writably_mapped(mapping))
1674 flush_dcache_page(page);
1676 * Mark the page accessed if we read the beginning.
1679 mark_page_accessed(page);
1681 page = ZERO_PAGE(0);
1682 page_cache_get(page);
1686 * Ok, we have the page, and it's up-to-date, so
1687 * now we can copy it to user space...
1689 * The actor routine returns how many bytes were actually used..
1690 * NOTE! This may not be the same as how much of a user buffer
1691 * we filled up (we may be padding etc), so we can only update
1692 * "pos" here (the actor routine has to update the user buffer
1693 * pointers and the remaining count).
1695 ret = actor(desc, page, offset, nr);
1697 index += offset >> PAGE_CACHE_SHIFT;
1698 offset &= ~PAGE_CACHE_MASK;
1700 page_cache_release(page);
1701 if (ret != nr || !desc->count)
1707 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1708 file_accessed(filp);
1711 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1713 read_descriptor_t desc;
1715 if ((ssize_t) count < 0)
1717 if (!access_ok(VERIFY_WRITE, buf, count))
1727 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1729 return desc.written;
1733 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1735 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1737 buf->f_type = TMPFS_MAGIC;
1738 buf->f_bsize = PAGE_CACHE_SIZE;
1739 buf->f_namelen = NAME_MAX;
1740 spin_lock(&sbinfo->stat_lock);
1741 if (sbinfo->max_blocks) {
1742 buf->f_blocks = sbinfo->max_blocks;
1743 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1745 if (sbinfo->max_inodes) {
1746 buf->f_files = sbinfo->max_inodes;
1747 buf->f_ffree = sbinfo->free_inodes;
1749 /* else leave those fields 0 like simple_statfs */
1750 spin_unlock(&sbinfo->stat_lock);
1755 * File creation. Allocate an inode, and we're done..
1758 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1760 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1761 int error = -ENOSPC;
1764 error = security_inode_init_security(inode, dir, NULL, NULL,
1767 if (error != -EOPNOTSUPP) {
1772 error = shmem_acl_init(inode, dir);
1777 if (dir->i_mode & S_ISGID) {
1778 inode->i_gid = dir->i_gid;
1780 inode->i_mode |= S_ISGID;
1782 dir->i_size += BOGO_DIRENT_SIZE;
1783 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1784 d_instantiate(dentry, inode);
1785 dget(dentry); /* Extra count - pin the dentry in core */
1790 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1794 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1800 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1801 struct nameidata *nd)
1803 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1809 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1811 struct inode *inode = old_dentry->d_inode;
1812 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1815 * No ordinary (disk based) filesystem counts links as inodes;
1816 * but each new link needs a new dentry, pinning lowmem, and
1817 * tmpfs dentries cannot be pruned until they are unlinked.
1819 if (sbinfo->max_inodes) {
1820 spin_lock(&sbinfo->stat_lock);
1821 if (!sbinfo->free_inodes) {
1822 spin_unlock(&sbinfo->stat_lock);
1825 sbinfo->free_inodes--;
1826 spin_unlock(&sbinfo->stat_lock);
1829 dir->i_size += BOGO_DIRENT_SIZE;
1830 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1832 atomic_inc(&inode->i_count); /* New dentry reference */
1833 dget(dentry); /* Extra pinning count for the created dentry */
1834 d_instantiate(dentry, inode);
1838 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1840 struct inode *inode = dentry->d_inode;
1842 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1843 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1844 if (sbinfo->max_inodes) {
1845 spin_lock(&sbinfo->stat_lock);
1846 sbinfo->free_inodes++;
1847 spin_unlock(&sbinfo->stat_lock);
1851 dir->i_size -= BOGO_DIRENT_SIZE;
1852 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1854 dput(dentry); /* Undo the count from "create" - this does all the work */
1858 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1860 if (!simple_empty(dentry))
1863 drop_nlink(dentry->d_inode);
1865 return shmem_unlink(dir, dentry);
1869 * The VFS layer already does all the dentry stuff for rename,
1870 * we just have to decrement the usage count for the target if
1871 * it exists so that the VFS layer correctly free's it when it
1874 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1876 struct inode *inode = old_dentry->d_inode;
1877 int they_are_dirs = S_ISDIR(inode->i_mode);
1879 if (!simple_empty(new_dentry))
1882 if (new_dentry->d_inode) {
1883 (void) shmem_unlink(new_dir, new_dentry);
1885 drop_nlink(old_dir);
1886 } else if (they_are_dirs) {
1887 drop_nlink(old_dir);
1891 old_dir->i_size -= BOGO_DIRENT_SIZE;
1892 new_dir->i_size += BOGO_DIRENT_SIZE;
1893 old_dir->i_ctime = old_dir->i_mtime =
1894 new_dir->i_ctime = new_dir->i_mtime =
1895 inode->i_ctime = CURRENT_TIME;
1899 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1903 struct inode *inode;
1904 struct page *page = NULL;
1906 struct shmem_inode_info *info;
1908 len = strlen(symname) + 1;
1909 if (len > PAGE_CACHE_SIZE)
1910 return -ENAMETOOLONG;
1912 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1916 error = security_inode_init_security(inode, dir, NULL, NULL,
1919 if (error != -EOPNOTSUPP) {
1926 info = SHMEM_I(inode);
1927 inode->i_size = len-1;
1928 if (len <= (char *)inode - (char *)info) {
1930 memcpy(info, symname, len);
1931 inode->i_op = &shmem_symlink_inline_operations;
1933 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1938 inode->i_op = &shmem_symlink_inode_operations;
1939 kaddr = kmap_atomic(page, KM_USER0);
1940 memcpy(kaddr, symname, len);
1941 kunmap_atomic(kaddr, KM_USER0);
1942 set_page_dirty(page);
1943 page_cache_release(page);
1945 if (dir->i_mode & S_ISGID)
1946 inode->i_gid = dir->i_gid;
1947 dir->i_size += BOGO_DIRENT_SIZE;
1948 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1949 d_instantiate(dentry, inode);
1954 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1956 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1960 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1962 struct page *page = NULL;
1963 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1964 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1968 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1970 if (!IS_ERR(nd_get_link(nd))) {
1971 struct page *page = cookie;
1973 mark_page_accessed(page);
1974 page_cache_release(page);
1978 static const struct inode_operations shmem_symlink_inline_operations = {
1979 .readlink = generic_readlink,
1980 .follow_link = shmem_follow_link_inline,
1983 static const struct inode_operations shmem_symlink_inode_operations = {
1984 .truncate = shmem_truncate,
1985 .readlink = generic_readlink,
1986 .follow_link = shmem_follow_link,
1987 .put_link = shmem_put_link,
1990 #ifdef CONFIG_TMPFS_POSIX_ACL
1992 * Superblocks without xattr inode operations will get security.* xattr
1993 * support from the VFS "for free". As soon as we have any other xattrs
1994 * like ACLs, we also need to implement the security.* handlers at
1995 * filesystem level, though.
1998 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1999 size_t list_len, const char *name,
2002 return security_inode_listsecurity(inode, list, list_len);
2005 static int shmem_xattr_security_get(struct inode *inode, const char *name,
2006 void *buffer, size_t size)
2008 if (strcmp(name, "") == 0)
2010 return security_inode_getsecurity(inode, name, buffer, size,
2014 static int shmem_xattr_security_set(struct inode *inode, const char *name,
2015 const void *value, size_t size, int flags)
2017 if (strcmp(name, "") == 0)
2019 return security_inode_setsecurity(inode, name, value, size, flags);
2022 static struct xattr_handler shmem_xattr_security_handler = {
2023 .prefix = XATTR_SECURITY_PREFIX,
2024 .list = shmem_xattr_security_list,
2025 .get = shmem_xattr_security_get,
2026 .set = shmem_xattr_security_set,
2029 static struct xattr_handler *shmem_xattr_handlers[] = {
2030 &shmem_xattr_acl_access_handler,
2031 &shmem_xattr_acl_default_handler,
2032 &shmem_xattr_security_handler,
2037 static struct dentry *shmem_get_parent(struct dentry *child)
2039 return ERR_PTR(-ESTALE);
2042 static int shmem_match(struct inode *ino, void *vfh)
2046 inum = (inum << 32) | fh[1];
2047 return ino->i_ino == inum && fh[0] == ino->i_generation;
2050 static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
2052 struct dentry *de = NULL;
2053 struct inode *inode;
2056 inum = (inum << 32) | fh[1];
2058 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
2060 de = d_find_alias(inode);
2064 return de? de: ERR_PTR(-ESTALE);
2067 static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
2069 int (*acceptable)(void *context, struct dentry *de),
2073 return ERR_PTR(-ESTALE);
2075 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
2079 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2082 struct inode *inode = dentry->d_inode;
2087 if (hlist_unhashed(&inode->i_hash)) {
2088 /* Unfortunately insert_inode_hash is not idempotent,
2089 * so as we hash inodes here rather than at creation
2090 * time, we need a lock to ensure we only try
2093 static DEFINE_SPINLOCK(lock);
2095 if (hlist_unhashed(&inode->i_hash))
2096 __insert_inode_hash(inode,
2097 inode->i_ino + inode->i_generation);
2101 fh[0] = inode->i_generation;
2102 fh[1] = inode->i_ino;
2103 fh[2] = ((__u64)inode->i_ino) >> 32;
2109 static struct export_operations shmem_export_ops = {
2110 .get_parent = shmem_get_parent,
2111 .get_dentry = shmem_get_dentry,
2112 .encode_fh = shmem_encode_fh,
2113 .decode_fh = shmem_decode_fh,
2116 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2117 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2118 int *policy, nodemask_t *policy_nodes)
2120 char *this_char, *value, *rest;
2122 while (options != NULL) {
2123 this_char = options;
2126 * NUL-terminate this option: unfortunately,
2127 * mount options form a comma-separated list,
2128 * but mpol's nodelist may also contain commas.
2130 options = strchr(options, ',');
2131 if (options == NULL)
2134 if (!isdigit(*options)) {
2141 if ((value = strchr(this_char,'=')) != NULL) {
2145 "tmpfs: No value for mount option '%s'\n",
2150 if (!strcmp(this_char,"size")) {
2151 unsigned long long size;
2152 size = memparse(value,&rest);
2154 size <<= PAGE_SHIFT;
2155 size *= totalram_pages;
2161 *blocks = size >> PAGE_CACHE_SHIFT;
2162 } else if (!strcmp(this_char,"nr_blocks")) {
2163 *blocks = memparse(value,&rest);
2166 } else if (!strcmp(this_char,"nr_inodes")) {
2167 *inodes = memparse(value,&rest);
2170 } else if (!strcmp(this_char,"mode")) {
2173 *mode = simple_strtoul(value,&rest,8);
2176 } else if (!strcmp(this_char,"uid")) {
2179 *uid = simple_strtoul(value,&rest,0);
2182 } else if (!strcmp(this_char,"gid")) {
2185 *gid = simple_strtoul(value,&rest,0);
2188 } else if (!strcmp(this_char,"mpol")) {
2189 if (shmem_parse_mpol(value,policy,policy_nodes))
2192 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2200 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2206 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2208 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2209 unsigned long max_blocks = sbinfo->max_blocks;
2210 unsigned long max_inodes = sbinfo->max_inodes;
2211 int policy = sbinfo->policy;
2212 nodemask_t policy_nodes = sbinfo->policy_nodes;
2213 unsigned long blocks;
2214 unsigned long inodes;
2215 int error = -EINVAL;
2217 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2218 &max_inodes, &policy, &policy_nodes))
2221 spin_lock(&sbinfo->stat_lock);
2222 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2223 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2224 if (max_blocks < blocks)
2226 if (max_inodes < inodes)
2229 * Those tests also disallow limited->unlimited while any are in
2230 * use, so i_blocks will always be zero when max_blocks is zero;
2231 * but we must separately disallow unlimited->limited, because
2232 * in that case we have no record of how much is already in use.
2234 if (max_blocks && !sbinfo->max_blocks)
2236 if (max_inodes && !sbinfo->max_inodes)
2240 sbinfo->max_blocks = max_blocks;
2241 sbinfo->free_blocks = max_blocks - blocks;
2242 sbinfo->max_inodes = max_inodes;
2243 sbinfo->free_inodes = max_inodes - inodes;
2244 sbinfo->policy = policy;
2245 sbinfo->policy_nodes = policy_nodes;
2247 spin_unlock(&sbinfo->stat_lock);
2252 static void shmem_put_super(struct super_block *sb)
2254 kfree(sb->s_fs_info);
2255 sb->s_fs_info = NULL;
2258 static int shmem_fill_super(struct super_block *sb,
2259 void *data, int silent)
2261 struct inode *inode;
2262 struct dentry *root;
2263 int mode = S_IRWXUGO | S_ISVTX;
2264 uid_t uid = current->fsuid;
2265 gid_t gid = current->fsgid;
2267 struct shmem_sb_info *sbinfo;
2268 unsigned long blocks = 0;
2269 unsigned long inodes = 0;
2270 int policy = MPOL_DEFAULT;
2271 nodemask_t policy_nodes = node_online_map;
2275 * Per default we only allow half of the physical ram per
2276 * tmpfs instance, limiting inodes to one per page of lowmem;
2277 * but the internal instance is left unlimited.
2279 if (!(sb->s_flags & MS_NOUSER)) {
2280 blocks = totalram_pages / 2;
2281 inodes = totalram_pages - totalhigh_pages;
2282 if (inodes > blocks)
2284 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2285 &inodes, &policy, &policy_nodes))
2288 sb->s_export_op = &shmem_export_ops;
2290 sb->s_flags |= MS_NOUSER;
2293 /* Round up to L1_CACHE_BYTES to resist false sharing */
2294 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2295 L1_CACHE_BYTES), GFP_KERNEL);
2299 spin_lock_init(&sbinfo->stat_lock);
2300 sbinfo->max_blocks = blocks;
2301 sbinfo->free_blocks = blocks;
2302 sbinfo->max_inodes = inodes;
2303 sbinfo->free_inodes = inodes;
2304 sbinfo->policy = policy;
2305 sbinfo->policy_nodes = policy_nodes;
2307 sb->s_fs_info = sbinfo;
2308 sb->s_maxbytes = SHMEM_MAX_BYTES;
2309 sb->s_blocksize = PAGE_CACHE_SIZE;
2310 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2311 sb->s_magic = TMPFS_MAGIC;
2312 sb->s_op = &shmem_ops;
2313 sb->s_time_gran = 1;
2314 #ifdef CONFIG_TMPFS_POSIX_ACL
2315 sb->s_xattr = shmem_xattr_handlers;
2316 sb->s_flags |= MS_POSIXACL;
2319 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2324 root = d_alloc_root(inode);
2333 shmem_put_super(sb);
2337 static struct kmem_cache *shmem_inode_cachep;
2339 static struct inode *shmem_alloc_inode(struct super_block *sb)
2341 struct shmem_inode_info *p;
2342 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2345 return &p->vfs_inode;
2348 static void shmem_destroy_inode(struct inode *inode)
2350 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2351 /* only struct inode is valid if it's an inline symlink */
2352 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2354 shmem_acl_destroy_inode(inode);
2355 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2358 static void init_once(void *foo, struct kmem_cache *cachep,
2359 unsigned long flags)
2361 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2363 inode_init_once(&p->vfs_inode);
2364 #ifdef CONFIG_TMPFS_POSIX_ACL
2366 p->i_default_acl = NULL;
2370 static int init_inodecache(void)
2372 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2373 sizeof(struct shmem_inode_info),
2374 0, 0, init_once, NULL);
2375 if (shmem_inode_cachep == NULL)
2380 static void destroy_inodecache(void)
2382 kmem_cache_destroy(shmem_inode_cachep);
2385 static const struct address_space_operations shmem_aops = {
2386 .writepage = shmem_writepage,
2387 .set_page_dirty = __set_page_dirty_no_writeback,
2389 .readpage = shmem_readpage,
2390 .prepare_write = shmem_prepare_write,
2391 .commit_write = simple_commit_write,
2393 .migratepage = migrate_page,
2396 static const struct file_operations shmem_file_operations = {
2399 .llseek = generic_file_llseek,
2400 .read = shmem_file_read,
2401 .write = shmem_file_write,
2402 .fsync = simple_sync_file,
2403 .splice_read = generic_file_splice_read,
2404 .splice_write = generic_file_splice_write,
2408 static const struct inode_operations shmem_inode_operations = {
2409 .truncate = shmem_truncate,
2410 .setattr = shmem_notify_change,
2411 .truncate_range = shmem_truncate_range,
2412 #ifdef CONFIG_TMPFS_POSIX_ACL
2413 .setxattr = generic_setxattr,
2414 .getxattr = generic_getxattr,
2415 .listxattr = generic_listxattr,
2416 .removexattr = generic_removexattr,
2417 .permission = shmem_permission,
2422 static const struct inode_operations shmem_dir_inode_operations = {
2424 .create = shmem_create,
2425 .lookup = simple_lookup,
2427 .unlink = shmem_unlink,
2428 .symlink = shmem_symlink,
2429 .mkdir = shmem_mkdir,
2430 .rmdir = shmem_rmdir,
2431 .mknod = shmem_mknod,
2432 .rename = shmem_rename,
2434 #ifdef CONFIG_TMPFS_POSIX_ACL
2435 .setattr = shmem_notify_change,
2436 .setxattr = generic_setxattr,
2437 .getxattr = generic_getxattr,
2438 .listxattr = generic_listxattr,
2439 .removexattr = generic_removexattr,
2440 .permission = shmem_permission,
2444 static const struct inode_operations shmem_special_inode_operations = {
2445 #ifdef CONFIG_TMPFS_POSIX_ACL
2446 .setattr = shmem_notify_change,
2447 .setxattr = generic_setxattr,
2448 .getxattr = generic_getxattr,
2449 .listxattr = generic_listxattr,
2450 .removexattr = generic_removexattr,
2451 .permission = shmem_permission,
2455 static const struct super_operations shmem_ops = {
2456 .alloc_inode = shmem_alloc_inode,
2457 .destroy_inode = shmem_destroy_inode,
2459 .statfs = shmem_statfs,
2460 .remount_fs = shmem_remount_fs,
2462 .delete_inode = shmem_delete_inode,
2463 .drop_inode = generic_delete_inode,
2464 .put_super = shmem_put_super,
2467 static struct vm_operations_struct shmem_vm_ops = {
2468 .nopage = shmem_nopage,
2469 .populate = shmem_populate,
2471 .set_policy = shmem_set_policy,
2472 .get_policy = shmem_get_policy,
2477 static int shmem_get_sb(struct file_system_type *fs_type,
2478 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2480 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2483 static struct file_system_type tmpfs_fs_type = {
2484 .owner = THIS_MODULE,
2486 .get_sb = shmem_get_sb,
2487 .kill_sb = kill_litter_super,
2489 static struct vfsmount *shm_mnt;
2491 static int __init init_tmpfs(void)
2495 error = init_inodecache();
2499 error = register_filesystem(&tmpfs_fs_type);
2501 printk(KERN_ERR "Could not register tmpfs\n");
2505 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2506 tmpfs_fs_type.name, NULL);
2507 if (IS_ERR(shm_mnt)) {
2508 error = PTR_ERR(shm_mnt);
2509 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2515 unregister_filesystem(&tmpfs_fs_type);
2517 destroy_inodecache();
2519 shm_mnt = ERR_PTR(error);
2522 module_init(init_tmpfs)
2525 * shmem_file_setup - get an unlinked file living in tmpfs
2527 * @name: name for dentry (to be seen in /proc/<pid>/maps
2528 * @size: size to be set for the file
2531 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2535 struct inode *inode;
2536 struct dentry *dentry, *root;
2539 if (IS_ERR(shm_mnt))
2540 return (void *)shm_mnt;
2542 if (size < 0 || size > SHMEM_MAX_BYTES)
2543 return ERR_PTR(-EINVAL);
2545 if (shmem_acct_size(flags, size))
2546 return ERR_PTR(-ENOMEM);
2550 this.len = strlen(name);
2551 this.hash = 0; /* will go */
2552 root = shm_mnt->mnt_root;
2553 dentry = d_alloc(root, &this);
2558 file = get_empty_filp();
2563 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2567 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2568 d_instantiate(dentry, inode);
2569 inode->i_size = size;
2570 inode->i_nlink = 0; /* It is unlinked */
2571 file->f_path.mnt = mntget(shm_mnt);
2572 file->f_path.dentry = dentry;
2573 file->f_mapping = inode->i_mapping;
2574 file->f_op = &shmem_file_operations;
2575 file->f_mode = FMODE_WRITE | FMODE_READ;
2583 shmem_unacct_size(flags, size);
2584 return ERR_PTR(error);
2588 * shmem_zero_setup - setup a shared anonymous mapping
2590 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2592 int shmem_zero_setup(struct vm_area_struct *vma)
2595 loff_t size = vma->vm_end - vma->vm_start;
2597 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2599 return PTR_ERR(file);
2603 vma->vm_file = file;
2604 vma->vm_ops = &shmem_vm_ops;
2605 vma->vm_flags |= VM_CAN_INVALIDATE;