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>
18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 * This file is released under the GPL.
24 #include <linux/init.h>
25 #include <linux/vfs.h>
26 #include <linux/mount.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
30 #include <linux/module.h>
31 #include <linux/percpu_counter.h>
32 #include <linux/swap.h>
34 static struct vfsmount *shm_mnt;
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
43 #include <linux/xattr.h>
44 #include <linux/exportfs.h>
45 #include <linux/posix_acl.h>
46 #include <linux/generic_acl.h>
47 #include <linux/mman.h>
48 #include <linux/string.h>
49 #include <linux/slab.h>
50 #include <linux/backing-dev.h>
51 #include <linux/shmem_fs.h>
52 #include <linux/writeback.h>
53 #include <linux/blkdev.h>
54 #include <linux/splice.h>
55 #include <linux/security.h>
56 #include <linux/swapops.h>
57 #include <linux/mempolicy.h>
58 #include <linux/namei.h>
59 #include <linux/ctype.h>
60 #include <linux/migrate.h>
61 #include <linux/highmem.h>
62 #include <linux/seq_file.h>
63 #include <linux/magic.h>
65 #include <asm/uaccess.h>
66 #include <asm/div64.h>
67 #include <asm/pgtable.h>
70 * The maximum size of a shmem/tmpfs file is limited by the maximum size of
71 * its triple-indirect swap vector - see illustration at shmem_swp_entry().
73 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
74 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum
75 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
76 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
78 * We use / and * instead of shifts in the definitions below, so that the swap
79 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
81 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
82 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
84 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
85 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
87 #define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
88 #define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
90 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
91 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
93 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
94 #define SHMEM_PAGEIN VM_READ
95 #define SHMEM_TRUNCATE VM_WRITE
97 /* Definition to limit shmem_truncate's steps between cond_rescheds */
98 #define LATENCY_LIMIT 64
100 /* Pretend that each entry is of this size in directory's i_size */
101 #define BOGO_DIRENT_SIZE 20
104 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
105 char *name; /* xattr name */
110 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
112 SGP_READ, /* don't exceed i_size, don't allocate page */
113 SGP_CACHE, /* don't exceed i_size, may allocate page */
114 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
115 SGP_WRITE, /* may exceed i_size, may allocate page */
119 static unsigned long shmem_default_max_blocks(void)
121 return totalram_pages / 2;
124 static unsigned long shmem_default_max_inodes(void)
126 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
130 static int shmem_getpage(struct inode *inode, unsigned long idx,
131 struct page **pagep, enum sgp_type sgp, int *type);
133 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
136 * The above definition of ENTRIES_PER_PAGE, and the use of
137 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
138 * might be reconsidered if it ever diverges from PAGE_SIZE.
140 * Mobility flags are masked out as swap vectors cannot move
142 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
143 PAGE_CACHE_SHIFT-PAGE_SHIFT);
146 static inline void shmem_dir_free(struct page *page)
148 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
151 static struct page **shmem_dir_map(struct page *page)
153 return (struct page **)kmap_atomic(page, KM_USER0);
156 static inline void shmem_dir_unmap(struct page **dir)
158 kunmap_atomic(dir, KM_USER0);
161 static swp_entry_t *shmem_swp_map(struct page *page)
163 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
166 static inline void shmem_swp_balance_unmap(void)
169 * When passing a pointer to an i_direct entry, to code which
170 * also handles indirect entries and so will shmem_swp_unmap,
171 * we must arrange for the preempt count to remain in balance.
172 * What kmap_atomic of a lowmem page does depends on config
173 * and architecture, so pretend to kmap_atomic some lowmem page.
175 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
178 static inline void shmem_swp_unmap(swp_entry_t *entry)
180 kunmap_atomic(entry, KM_USER1);
183 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
185 return sb->s_fs_info;
189 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
190 * for shared memory and for shared anonymous (/dev/zero) mappings
191 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
192 * consistent with the pre-accounting of private mappings ...
194 static inline int shmem_acct_size(unsigned long flags, loff_t size)
196 return (flags & VM_NORESERVE) ?
197 0 : security_vm_enough_memory_kern(VM_ACCT(size));
200 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
202 if (!(flags & VM_NORESERVE))
203 vm_unacct_memory(VM_ACCT(size));
207 * ... whereas tmpfs objects are accounted incrementally as
208 * pages are allocated, in order to allow huge sparse files.
209 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
210 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
212 static inline int shmem_acct_block(unsigned long flags)
214 return (flags & VM_NORESERVE) ?
215 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
218 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
220 if (flags & VM_NORESERVE)
221 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
224 static const struct super_operations shmem_ops;
225 static const struct address_space_operations shmem_aops;
226 static const struct file_operations shmem_file_operations;
227 static const struct inode_operations shmem_inode_operations;
228 static const struct inode_operations shmem_dir_inode_operations;
229 static const struct inode_operations shmem_special_inode_operations;
230 static const struct vm_operations_struct shmem_vm_ops;
232 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
233 .ra_pages = 0, /* No readahead */
234 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
237 static LIST_HEAD(shmem_swaplist);
238 static DEFINE_MUTEX(shmem_swaplist_mutex);
240 static void shmem_free_blocks(struct inode *inode, long pages)
242 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
243 if (sbinfo->max_blocks) {
244 percpu_counter_add(&sbinfo->used_blocks, -pages);
245 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
249 static int shmem_reserve_inode(struct super_block *sb)
251 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
252 if (sbinfo->max_inodes) {
253 spin_lock(&sbinfo->stat_lock);
254 if (!sbinfo->free_inodes) {
255 spin_unlock(&sbinfo->stat_lock);
258 sbinfo->free_inodes--;
259 spin_unlock(&sbinfo->stat_lock);
264 static void shmem_free_inode(struct super_block *sb)
266 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
267 if (sbinfo->max_inodes) {
268 spin_lock(&sbinfo->stat_lock);
269 sbinfo->free_inodes++;
270 spin_unlock(&sbinfo->stat_lock);
275 * shmem_recalc_inode - recalculate the size of an inode
276 * @inode: inode to recalc
278 * We have to calculate the free blocks since the mm can drop
279 * undirtied hole pages behind our back.
281 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
282 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
284 * It has to be called with the spinlock held.
286 static void shmem_recalc_inode(struct inode *inode)
288 struct shmem_inode_info *info = SHMEM_I(inode);
291 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
293 info->alloced -= freed;
294 shmem_unacct_blocks(info->flags, freed);
295 shmem_free_blocks(inode, freed);
300 * shmem_swp_entry - find the swap vector position in the info structure
301 * @info: info structure for the inode
302 * @index: index of the page to find
303 * @page: optional page to add to the structure. Has to be preset to
306 * If there is no space allocated yet it will return NULL when
307 * page is NULL, else it will use the page for the needed block,
308 * setting it to NULL on return to indicate that it has been used.
310 * The swap vector is organized the following way:
312 * There are SHMEM_NR_DIRECT entries directly stored in the
313 * shmem_inode_info structure. So small files do not need an addional
316 * For pages with index > SHMEM_NR_DIRECT there is the pointer
317 * i_indirect which points to a page which holds in the first half
318 * doubly indirect blocks, in the second half triple indirect blocks:
320 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
321 * following layout (for SHMEM_NR_DIRECT == 16):
323 * i_indirect -> dir --> 16-19
336 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
338 unsigned long offset;
342 if (index < SHMEM_NR_DIRECT) {
343 shmem_swp_balance_unmap();
344 return info->i_direct+index;
346 if (!info->i_indirect) {
348 info->i_indirect = *page;
351 return NULL; /* need another page */
354 index -= SHMEM_NR_DIRECT;
355 offset = index % ENTRIES_PER_PAGE;
356 index /= ENTRIES_PER_PAGE;
357 dir = shmem_dir_map(info->i_indirect);
359 if (index >= ENTRIES_PER_PAGE/2) {
360 index -= ENTRIES_PER_PAGE/2;
361 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
362 index %= ENTRIES_PER_PAGE;
369 shmem_dir_unmap(dir);
370 return NULL; /* need another page */
372 shmem_dir_unmap(dir);
373 dir = shmem_dir_map(subdir);
379 if (!page || !(subdir = *page)) {
380 shmem_dir_unmap(dir);
381 return NULL; /* need a page */
386 shmem_dir_unmap(dir);
387 return shmem_swp_map(subdir) + offset;
390 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
392 long incdec = value? 1: -1;
395 info->swapped += incdec;
396 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
397 struct page *page = kmap_atomic_to_page(entry);
398 set_page_private(page, page_private(page) + incdec);
403 * shmem_swp_alloc - get the position of the swap entry for the page.
404 * @info: info structure for the inode
405 * @index: index of the page to find
406 * @sgp: check and recheck i_size? skip allocation?
408 * If the entry does not exist, allocate it.
410 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
412 struct inode *inode = &info->vfs_inode;
413 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
414 struct page *page = NULL;
417 if (sgp != SGP_WRITE &&
418 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
419 return ERR_PTR(-EINVAL);
421 while (!(entry = shmem_swp_entry(info, index, &page))) {
423 return shmem_swp_map(ZERO_PAGE(0));
425 * Test used_blocks against 1 less max_blocks, since we have 1 data
426 * page (and perhaps indirect index pages) yet to allocate:
427 * a waste to allocate index if we cannot allocate data.
429 if (sbinfo->max_blocks) {
430 if (percpu_counter_compare(&sbinfo->used_blocks,
431 sbinfo->max_blocks - 1) >= 0)
432 return ERR_PTR(-ENOSPC);
433 percpu_counter_inc(&sbinfo->used_blocks);
434 inode->i_blocks += BLOCKS_PER_PAGE;
437 spin_unlock(&info->lock);
438 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
439 spin_lock(&info->lock);
442 shmem_free_blocks(inode, 1);
443 return ERR_PTR(-ENOMEM);
445 if (sgp != SGP_WRITE &&
446 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
447 entry = ERR_PTR(-EINVAL);
450 if (info->next_index <= index)
451 info->next_index = index + 1;
454 /* another task gave its page, or truncated the file */
455 shmem_free_blocks(inode, 1);
456 shmem_dir_free(page);
458 if (info->next_index <= index && !IS_ERR(entry))
459 info->next_index = index + 1;
464 * shmem_free_swp - free some swap entries in a directory
465 * @dir: pointer to the directory
466 * @edir: pointer after last entry of the directory
467 * @punch_lock: pointer to spinlock when needed for the holepunch case
469 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
470 spinlock_t *punch_lock)
472 spinlock_t *punch_unlock = NULL;
476 for (ptr = dir; ptr < edir; ptr++) {
478 if (unlikely(punch_lock)) {
479 punch_unlock = punch_lock;
481 spin_lock(punch_unlock);
485 free_swap_and_cache(*ptr);
486 *ptr = (swp_entry_t){0};
491 spin_unlock(punch_unlock);
495 static int shmem_map_and_free_swp(struct page *subdir, int offset,
496 int limit, struct page ***dir, spinlock_t *punch_lock)
501 ptr = shmem_swp_map(subdir);
502 for (; offset < limit; offset += LATENCY_LIMIT) {
503 int size = limit - offset;
504 if (size > LATENCY_LIMIT)
505 size = LATENCY_LIMIT;
506 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
508 if (need_resched()) {
509 shmem_swp_unmap(ptr);
511 shmem_dir_unmap(*dir);
515 ptr = shmem_swp_map(subdir);
518 shmem_swp_unmap(ptr);
522 static void shmem_free_pages(struct list_head *next)
528 page = container_of(next, struct page, lru);
530 shmem_dir_free(page);
532 if (freed >= LATENCY_LIMIT) {
539 void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
541 struct shmem_inode_info *info = SHMEM_I(inode);
546 unsigned long diroff;
552 LIST_HEAD(pages_to_free);
553 long nr_pages_to_free = 0;
554 long nr_swaps_freed = 0;
558 spinlock_t *needs_lock;
559 spinlock_t *punch_lock;
560 unsigned long upper_limit;
562 truncate_inode_pages_range(inode->i_mapping, start, end);
564 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
565 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
566 if (idx >= info->next_index)
569 spin_lock(&info->lock);
570 info->flags |= SHMEM_TRUNCATE;
571 if (likely(end == (loff_t) -1)) {
572 limit = info->next_index;
573 upper_limit = SHMEM_MAX_INDEX;
574 info->next_index = idx;
578 if (end + 1 >= inode->i_size) { /* we may free a little more */
579 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
581 upper_limit = SHMEM_MAX_INDEX;
583 limit = (end + 1) >> PAGE_CACHE_SHIFT;
586 needs_lock = &info->lock;
590 topdir = info->i_indirect;
591 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
592 info->i_indirect = NULL;
594 list_add(&topdir->lru, &pages_to_free);
596 spin_unlock(&info->lock);
598 if (info->swapped && idx < SHMEM_NR_DIRECT) {
599 ptr = info->i_direct;
601 if (size > SHMEM_NR_DIRECT)
602 size = SHMEM_NR_DIRECT;
603 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
607 * If there are no indirect blocks or we are punching a hole
608 * below indirect blocks, nothing to be done.
610 if (!topdir || limit <= SHMEM_NR_DIRECT)
614 * The truncation case has already dropped info->lock, and we're safe
615 * because i_size and next_index have already been lowered, preventing
616 * access beyond. But in the punch_hole case, we still need to take
617 * the lock when updating the swap directory, because there might be
618 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
619 * shmem_writepage. However, whenever we find we can remove a whole
620 * directory page (not at the misaligned start or end of the range),
621 * we first NULLify its pointer in the level above, and then have no
622 * need to take the lock when updating its contents: needs_lock and
623 * punch_lock (either pointing to info->lock or NULL) manage this.
626 upper_limit -= SHMEM_NR_DIRECT;
627 limit -= SHMEM_NR_DIRECT;
628 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
629 offset = idx % ENTRIES_PER_PAGE;
632 dir = shmem_dir_map(topdir);
633 stage = ENTRIES_PER_PAGEPAGE/2;
634 if (idx < ENTRIES_PER_PAGEPAGE/2) {
636 diroff = idx/ENTRIES_PER_PAGE;
638 dir += ENTRIES_PER_PAGE/2;
639 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
641 stage += ENTRIES_PER_PAGEPAGE;
644 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
645 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
646 if (!diroff && !offset && upper_limit >= stage) {
648 spin_lock(needs_lock);
650 spin_unlock(needs_lock);
655 list_add(&middir->lru, &pages_to_free);
657 shmem_dir_unmap(dir);
658 dir = shmem_dir_map(middir);
666 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
667 if (unlikely(idx == stage)) {
668 shmem_dir_unmap(dir);
669 dir = shmem_dir_map(topdir) +
670 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
673 idx += ENTRIES_PER_PAGEPAGE;
677 stage = idx + ENTRIES_PER_PAGEPAGE;
680 needs_lock = &info->lock;
681 if (upper_limit >= stage) {
683 spin_lock(needs_lock);
685 spin_unlock(needs_lock);
690 list_add(&middir->lru, &pages_to_free);
692 shmem_dir_unmap(dir);
694 dir = shmem_dir_map(middir);
697 punch_lock = needs_lock;
698 subdir = dir[diroff];
699 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
701 spin_lock(needs_lock);
703 spin_unlock(needs_lock);
708 list_add(&subdir->lru, &pages_to_free);
710 if (subdir && page_private(subdir) /* has swap entries */) {
712 if (size > ENTRIES_PER_PAGE)
713 size = ENTRIES_PER_PAGE;
714 freed = shmem_map_and_free_swp(subdir,
715 offset, size, &dir, punch_lock);
717 dir = shmem_dir_map(middir);
718 nr_swaps_freed += freed;
719 if (offset || punch_lock) {
720 spin_lock(&info->lock);
721 set_page_private(subdir,
722 page_private(subdir) - freed);
723 spin_unlock(&info->lock);
725 BUG_ON(page_private(subdir) != freed);
730 shmem_dir_unmap(dir);
732 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
734 * Call truncate_inode_pages again: racing shmem_unuse_inode
735 * may have swizzled a page in from swap since
736 * truncate_pagecache or generic_delete_inode did it, before we
737 * lowered next_index. Also, though shmem_getpage checks
738 * i_size before adding to cache, no recheck after: so fix the
739 * narrow window there too.
741 truncate_inode_pages_range(inode->i_mapping, start, end);
744 spin_lock(&info->lock);
745 info->flags &= ~SHMEM_TRUNCATE;
746 info->swapped -= nr_swaps_freed;
747 if (nr_pages_to_free)
748 shmem_free_blocks(inode, nr_pages_to_free);
749 shmem_recalc_inode(inode);
750 spin_unlock(&info->lock);
753 * Empty swap vector directory pages to be freed?
755 if (!list_empty(&pages_to_free)) {
756 pages_to_free.prev->next = NULL;
757 shmem_free_pages(pages_to_free.next);
760 EXPORT_SYMBOL_GPL(shmem_truncate_range);
762 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
764 struct inode *inode = dentry->d_inode;
767 error = inode_change_ok(inode, attr);
771 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
772 loff_t oldsize = inode->i_size;
773 loff_t newsize = attr->ia_size;
774 struct page *page = NULL;
776 if (newsize < oldsize) {
778 * If truncating down to a partial page, then
779 * if that page is already allocated, hold it
780 * in memory until the truncation is over, so
781 * truncate_partial_page cannot miss it were
782 * it assigned to swap.
784 if (newsize & (PAGE_CACHE_SIZE-1)) {
785 (void) shmem_getpage(inode,
786 newsize >> PAGE_CACHE_SHIFT,
787 &page, SGP_READ, NULL);
792 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
793 * detect if any pages might have been added to cache
794 * after truncate_inode_pages. But we needn't bother
795 * if it's being fully truncated to zero-length: the
796 * nrpages check is efficient enough in that case.
799 struct shmem_inode_info *info = SHMEM_I(inode);
800 spin_lock(&info->lock);
801 info->flags &= ~SHMEM_PAGEIN;
802 spin_unlock(&info->lock);
805 if (newsize != oldsize) {
806 i_size_write(inode, newsize);
807 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
809 if (newsize < oldsize) {
810 loff_t holebegin = round_up(newsize, PAGE_SIZE);
811 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
812 shmem_truncate_range(inode, newsize, (loff_t)-1);
813 /* unmap again to remove racily COWed private pages */
814 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
817 page_cache_release(page);
820 setattr_copy(inode, attr);
821 #ifdef CONFIG_TMPFS_POSIX_ACL
822 if (attr->ia_valid & ATTR_MODE)
823 error = generic_acl_chmod(inode);
828 static void shmem_evict_inode(struct inode *inode)
830 struct shmem_inode_info *info = SHMEM_I(inode);
831 struct shmem_xattr *xattr, *nxattr;
833 if (inode->i_mapping->a_ops == &shmem_aops) {
834 shmem_unacct_size(info->flags, inode->i_size);
836 shmem_truncate_range(inode, 0, (loff_t)-1);
837 if (!list_empty(&info->swaplist)) {
838 mutex_lock(&shmem_swaplist_mutex);
839 list_del_init(&info->swaplist);
840 mutex_unlock(&shmem_swaplist_mutex);
844 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
848 BUG_ON(inode->i_blocks);
849 shmem_free_inode(inode->i_sb);
850 end_writeback(inode);
853 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
857 for (ptr = dir; ptr < edir; ptr++) {
858 if (ptr->val == entry.val)
864 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
866 struct address_space *mapping;
878 ptr = info->i_direct;
879 spin_lock(&info->lock);
880 if (!info->swapped) {
881 list_del_init(&info->swaplist);
884 limit = info->next_index;
886 if (size > SHMEM_NR_DIRECT)
887 size = SHMEM_NR_DIRECT;
888 offset = shmem_find_swp(entry, ptr, ptr+size);
890 shmem_swp_balance_unmap();
893 if (!info->i_indirect)
896 dir = shmem_dir_map(info->i_indirect);
897 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
899 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
900 if (unlikely(idx == stage)) {
901 shmem_dir_unmap(dir-1);
902 if (cond_resched_lock(&info->lock)) {
903 /* check it has not been truncated */
904 if (limit > info->next_index) {
905 limit = info->next_index;
910 dir = shmem_dir_map(info->i_indirect) +
911 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
914 idx += ENTRIES_PER_PAGEPAGE;
918 stage = idx + ENTRIES_PER_PAGEPAGE;
920 shmem_dir_unmap(dir);
921 dir = shmem_dir_map(subdir);
924 if (subdir && page_private(subdir)) {
925 ptr = shmem_swp_map(subdir);
927 if (size > ENTRIES_PER_PAGE)
928 size = ENTRIES_PER_PAGE;
929 offset = shmem_find_swp(entry, ptr, ptr+size);
930 shmem_swp_unmap(ptr);
932 shmem_dir_unmap(dir);
933 ptr = shmem_swp_map(subdir);
939 shmem_dir_unmap(dir-1);
941 spin_unlock(&info->lock);
948 * Move _head_ to start search for next from here.
949 * But be careful: shmem_evict_inode checks list_empty without taking
950 * mutex, and there's an instant in list_move_tail when info->swaplist
951 * would appear empty, if it were the only one on shmem_swaplist. We
952 * could avoid doing it if inode NULL; or use this minor optimization.
954 if (shmem_swaplist.next != &info->swaplist)
955 list_move_tail(&shmem_swaplist, &info->swaplist);
958 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
959 * but also to hold up shmem_evict_inode(): so inode cannot be freed
960 * beneath us (pagelock doesn't help until the page is in pagecache).
962 mapping = info->vfs_inode.i_mapping;
963 error = add_to_page_cache_locked(page, mapping, idx, GFP_NOWAIT);
964 /* which does mem_cgroup_uncharge_cache_page on error */
966 if (error == -EEXIST) {
967 struct page *filepage = find_get_page(mapping, idx);
971 * There might be a more uptodate page coming down
972 * from a stacked writepage: forget our swappage if so.
974 if (PageUptodate(filepage))
976 page_cache_release(filepage);
980 delete_from_swap_cache(page);
981 set_page_dirty(page);
982 info->flags |= SHMEM_PAGEIN;
983 shmem_swp_set(info, ptr, 0);
985 error = 1; /* not an error, but entry was found */
987 shmem_swp_unmap(ptr);
988 spin_unlock(&info->lock);
993 * shmem_unuse() search for an eventually swapped out shmem page.
995 int shmem_unuse(swp_entry_t entry, struct page *page)
997 struct list_head *p, *next;
998 struct shmem_inode_info *info;
1003 * Charge page using GFP_KERNEL while we can wait, before taking
1004 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1005 * Charged back to the user (not to caller) when swap account is used.
1006 * add_to_page_cache() will be called with GFP_NOWAIT.
1008 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
1012 * Try to preload while we can wait, to not make a habit of
1013 * draining atomic reserves; but don't latch on to this cpu,
1014 * it's okay if sometimes we get rescheduled after this.
1016 error = radix_tree_preload(GFP_KERNEL);
1019 radix_tree_preload_end();
1021 mutex_lock(&shmem_swaplist_mutex);
1022 list_for_each_safe(p, next, &shmem_swaplist) {
1023 info = list_entry(p, struct shmem_inode_info, swaplist);
1024 found = shmem_unuse_inode(info, entry, page);
1029 mutex_unlock(&shmem_swaplist_mutex);
1033 mem_cgroup_uncharge_cache_page(page);
1038 page_cache_release(page);
1043 * Move the page from the page cache to the swap cache.
1045 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1047 struct shmem_inode_info *info;
1048 swp_entry_t *entry, swap;
1049 struct address_space *mapping;
1050 unsigned long index;
1051 struct inode *inode;
1053 BUG_ON(!PageLocked(page));
1054 mapping = page->mapping;
1055 index = page->index;
1056 inode = mapping->host;
1057 info = SHMEM_I(inode);
1058 if (info->flags & VM_LOCKED)
1060 if (!total_swap_pages)
1064 * shmem_backing_dev_info's capabilities prevent regular writeback or
1065 * sync from ever calling shmem_writepage; but a stacking filesystem
1066 * may use the ->writepage of its underlying filesystem, in which case
1067 * tmpfs should write out to swap only in response to memory pressure,
1068 * and not for the writeback threads or sync. However, in those cases,
1069 * we do still want to check if there's a redundant swappage to be
1072 if (wbc->for_reclaim)
1073 swap = get_swap_page();
1078 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1079 * if it's not already there. Do it now because we cannot take
1080 * mutex while holding spinlock, and must do so before the page
1081 * is moved to swap cache, when its pagelock no longer protects
1082 * the inode from eviction. But don't unlock the mutex until
1083 * we've taken the spinlock, because shmem_unuse_inode() will
1084 * prune a !swapped inode from the swaplist under both locks.
1087 mutex_lock(&shmem_swaplist_mutex);
1088 if (list_empty(&info->swaplist))
1089 list_add_tail(&info->swaplist, &shmem_swaplist);
1092 spin_lock(&info->lock);
1094 mutex_unlock(&shmem_swaplist_mutex);
1096 if (index >= info->next_index) {
1097 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1100 entry = shmem_swp_entry(info, index, NULL);
1103 * The more uptodate page coming down from a stacked
1104 * writepage should replace our old swappage.
1106 free_swap_and_cache(*entry);
1107 shmem_swp_set(info, entry, 0);
1109 shmem_recalc_inode(inode);
1111 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1112 delete_from_page_cache(page);
1113 shmem_swp_set(info, entry, swap.val);
1114 shmem_swp_unmap(entry);
1115 swap_shmem_alloc(swap);
1116 spin_unlock(&info->lock);
1117 BUG_ON(page_mapped(page));
1118 swap_writepage(page, wbc);
1122 shmem_swp_unmap(entry);
1124 spin_unlock(&info->lock);
1126 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1127 * clear SWAP_HAS_CACHE flag.
1129 swapcache_free(swap, NULL);
1131 set_page_dirty(page);
1132 if (wbc->for_reclaim)
1133 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1140 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1144 if (!mpol || mpol->mode == MPOL_DEFAULT)
1145 return; /* show nothing */
1147 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1149 seq_printf(seq, ",mpol=%s", buffer);
1152 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1154 struct mempolicy *mpol = NULL;
1156 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1157 mpol = sbinfo->mpol;
1159 spin_unlock(&sbinfo->stat_lock);
1163 #endif /* CONFIG_TMPFS */
1165 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1166 struct shmem_inode_info *info, unsigned long idx)
1168 struct mempolicy mpol, *spol;
1169 struct vm_area_struct pvma;
1172 spol = mpol_cond_copy(&mpol,
1173 mpol_shared_policy_lookup(&info->policy, idx));
1175 /* Create a pseudo vma that just contains the policy */
1177 pvma.vm_pgoff = idx;
1179 pvma.vm_policy = spol;
1180 page = swapin_readahead(entry, gfp, &pvma, 0);
1184 static struct page *shmem_alloc_page(gfp_t gfp,
1185 struct shmem_inode_info *info, unsigned long idx)
1187 struct vm_area_struct pvma;
1189 /* Create a pseudo vma that just contains the policy */
1191 pvma.vm_pgoff = idx;
1193 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1196 * alloc_page_vma() will drop the shared policy reference
1198 return alloc_page_vma(gfp, &pvma, 0);
1200 #else /* !CONFIG_NUMA */
1202 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1205 #endif /* CONFIG_TMPFS */
1207 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1208 struct shmem_inode_info *info, unsigned long idx)
1210 return swapin_readahead(entry, gfp, NULL, 0);
1213 static inline struct page *shmem_alloc_page(gfp_t gfp,
1214 struct shmem_inode_info *info, unsigned long idx)
1216 return alloc_page(gfp);
1218 #endif /* CONFIG_NUMA */
1220 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1221 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1228 * shmem_getpage - either get the page from swap or allocate a new one
1230 * If we allocate a new one we do not mark it dirty. That's up to the
1231 * vm. If we swap it in we mark it dirty since we also free the swap
1232 * entry since a page cannot live in both the swap and page cache
1234 static int shmem_getpage(struct inode *inode, unsigned long idx,
1235 struct page **pagep, enum sgp_type sgp, int *type)
1237 struct address_space *mapping = inode->i_mapping;
1238 struct shmem_inode_info *info = SHMEM_I(inode);
1239 struct shmem_sb_info *sbinfo;
1240 struct page *filepage = *pagep;
1241 struct page *swappage;
1242 struct page *prealloc_page = NULL;
1248 if (idx >= SHMEM_MAX_INDEX)
1255 * Normally, filepage is NULL on entry, and either found
1256 * uptodate immediately, or allocated and zeroed, or read
1257 * in under swappage, which is then assigned to filepage.
1258 * But shmem_readpage (required for splice) passes in a locked
1259 * filepage, which may be found not uptodate by other callers
1260 * too, and may need to be copied from the swappage read in.
1264 filepage = find_lock_page(mapping, idx);
1265 if (filepage && PageUptodate(filepage))
1267 gfp = mapping_gfp_mask(mapping);
1270 * Try to preload while we can wait, to not make a habit of
1271 * draining atomic reserves; but don't latch on to this cpu.
1273 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1276 radix_tree_preload_end();
1277 if (sgp != SGP_READ && !prealloc_page) {
1278 /* We don't care if this fails */
1279 prealloc_page = shmem_alloc_page(gfp, info, idx);
1280 if (prealloc_page) {
1281 if (mem_cgroup_cache_charge(prealloc_page,
1282 current->mm, GFP_KERNEL)) {
1283 page_cache_release(prealloc_page);
1284 prealloc_page = NULL;
1291 spin_lock(&info->lock);
1292 shmem_recalc_inode(inode);
1293 entry = shmem_swp_alloc(info, idx, sgp);
1294 if (IS_ERR(entry)) {
1295 spin_unlock(&info->lock);
1296 error = PTR_ERR(entry);
1302 /* Look it up and read it in.. */
1303 swappage = lookup_swap_cache(swap);
1305 shmem_swp_unmap(entry);
1306 spin_unlock(&info->lock);
1307 /* here we actually do the io */
1309 *type |= VM_FAULT_MAJOR;
1310 swappage = shmem_swapin(swap, gfp, info, idx);
1312 spin_lock(&info->lock);
1313 entry = shmem_swp_alloc(info, idx, sgp);
1315 error = PTR_ERR(entry);
1317 if (entry->val == swap.val)
1319 shmem_swp_unmap(entry);
1321 spin_unlock(&info->lock);
1326 wait_on_page_locked(swappage);
1327 page_cache_release(swappage);
1331 /* We have to do this with page locked to prevent races */
1332 if (!trylock_page(swappage)) {
1333 shmem_swp_unmap(entry);
1334 spin_unlock(&info->lock);
1335 wait_on_page_locked(swappage);
1336 page_cache_release(swappage);
1339 if (PageWriteback(swappage)) {
1340 shmem_swp_unmap(entry);
1341 spin_unlock(&info->lock);
1342 wait_on_page_writeback(swappage);
1343 unlock_page(swappage);
1344 page_cache_release(swappage);
1347 if (!PageUptodate(swappage)) {
1348 shmem_swp_unmap(entry);
1349 spin_unlock(&info->lock);
1350 unlock_page(swappage);
1351 page_cache_release(swappage);
1357 shmem_swp_set(info, entry, 0);
1358 shmem_swp_unmap(entry);
1359 delete_from_swap_cache(swappage);
1360 spin_unlock(&info->lock);
1361 copy_highpage(filepage, swappage);
1362 unlock_page(swappage);
1363 page_cache_release(swappage);
1364 flush_dcache_page(filepage);
1365 SetPageUptodate(filepage);
1366 set_page_dirty(filepage);
1368 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1369 idx, GFP_NOWAIT))) {
1370 info->flags |= SHMEM_PAGEIN;
1371 shmem_swp_set(info, entry, 0);
1372 shmem_swp_unmap(entry);
1373 delete_from_swap_cache(swappage);
1374 spin_unlock(&info->lock);
1375 filepage = swappage;
1376 set_page_dirty(filepage);
1379 shmem_swp_unmap(entry);
1380 spin_unlock(&info->lock);
1381 if (error == -ENOMEM) {
1383 * reclaim from proper memory cgroup and
1384 * call memcg's OOM if needed.
1386 error = mem_cgroup_shmem_charge_fallback(
1391 unlock_page(swappage);
1392 page_cache_release(swappage);
1396 unlock_page(swappage);
1397 page_cache_release(swappage);
1400 } else if (sgp == SGP_READ && !filepage) {
1401 shmem_swp_unmap(entry);
1402 filepage = find_get_page(mapping, idx);
1404 (!PageUptodate(filepage) || !trylock_page(filepage))) {
1405 spin_unlock(&info->lock);
1406 wait_on_page_locked(filepage);
1407 page_cache_release(filepage);
1411 spin_unlock(&info->lock);
1413 shmem_swp_unmap(entry);
1414 sbinfo = SHMEM_SB(inode->i_sb);
1415 if (sbinfo->max_blocks) {
1416 if (percpu_counter_compare(&sbinfo->used_blocks,
1417 sbinfo->max_blocks) >= 0 ||
1418 shmem_acct_block(info->flags))
1420 percpu_counter_inc(&sbinfo->used_blocks);
1421 inode->i_blocks += BLOCKS_PER_PAGE;
1422 } else if (shmem_acct_block(info->flags))
1428 if (!prealloc_page) {
1429 spin_unlock(&info->lock);
1430 filepage = shmem_alloc_page(gfp, info, idx);
1432 spin_lock(&info->lock);
1433 shmem_unacct_blocks(info->flags, 1);
1434 shmem_free_blocks(inode, 1);
1435 spin_unlock(&info->lock);
1439 SetPageSwapBacked(filepage);
1442 * Precharge page while we can wait, compensate
1445 error = mem_cgroup_cache_charge(filepage,
1446 current->mm, GFP_KERNEL);
1448 page_cache_release(filepage);
1449 spin_lock(&info->lock);
1450 shmem_unacct_blocks(info->flags, 1);
1451 shmem_free_blocks(inode, 1);
1452 spin_unlock(&info->lock);
1457 spin_lock(&info->lock);
1459 filepage = prealloc_page;
1460 prealloc_page = NULL;
1461 SetPageSwapBacked(filepage);
1464 entry = shmem_swp_alloc(info, idx, sgp);
1466 error = PTR_ERR(entry);
1469 shmem_swp_unmap(entry);
1471 ret = error || swap.val;
1473 mem_cgroup_uncharge_cache_page(filepage);
1475 ret = add_to_page_cache_lru(filepage, mapping,
1478 * At add_to_page_cache_lru() failure, uncharge will
1479 * be done automatically.
1482 shmem_unacct_blocks(info->flags, 1);
1483 shmem_free_blocks(inode, 1);
1484 spin_unlock(&info->lock);
1485 page_cache_release(filepage);
1491 info->flags |= SHMEM_PAGEIN;
1495 spin_unlock(&info->lock);
1496 clear_highpage(filepage);
1497 flush_dcache_page(filepage);
1498 SetPageUptodate(filepage);
1499 if (sgp == SGP_DIRTY)
1500 set_page_dirty(filepage);
1509 * Perhaps the page was brought in from swap between find_lock_page
1510 * and taking info->lock? We allow for that at add_to_page_cache_lru,
1511 * but must also avoid reporting a spurious ENOSPC while working on a
1512 * full tmpfs. (When filepage has been passed in to shmem_getpage, it
1513 * is already in page cache, which prevents this race from occurring.)
1516 struct page *page = find_get_page(mapping, idx);
1518 spin_unlock(&info->lock);
1519 page_cache_release(page);
1523 spin_unlock(&info->lock);
1526 if (*pagep != filepage) {
1527 unlock_page(filepage);
1528 page_cache_release(filepage);
1531 if (prealloc_page) {
1532 mem_cgroup_uncharge_cache_page(prealloc_page);
1533 page_cache_release(prealloc_page);
1538 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1540 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1544 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1545 return VM_FAULT_SIGBUS;
1547 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1549 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1550 if (ret & VM_FAULT_MAJOR) {
1551 count_vm_event(PGMAJFAULT);
1552 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1554 return ret | VM_FAULT_LOCKED;
1558 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1560 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1561 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1564 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1567 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1570 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1571 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1575 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1577 struct inode *inode = file->f_path.dentry->d_inode;
1578 struct shmem_inode_info *info = SHMEM_I(inode);
1579 int retval = -ENOMEM;
1581 spin_lock(&info->lock);
1582 if (lock && !(info->flags & VM_LOCKED)) {
1583 if (!user_shm_lock(inode->i_size, user))
1585 info->flags |= VM_LOCKED;
1586 mapping_set_unevictable(file->f_mapping);
1588 if (!lock && (info->flags & VM_LOCKED) && user) {
1589 user_shm_unlock(inode->i_size, user);
1590 info->flags &= ~VM_LOCKED;
1591 mapping_clear_unevictable(file->f_mapping);
1592 scan_mapping_unevictable_pages(file->f_mapping);
1597 spin_unlock(&info->lock);
1601 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1603 file_accessed(file);
1604 vma->vm_ops = &shmem_vm_ops;
1605 vma->vm_flags |= VM_CAN_NONLINEAR;
1609 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1610 int mode, dev_t dev, unsigned long flags)
1612 struct inode *inode;
1613 struct shmem_inode_info *info;
1614 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1616 if (shmem_reserve_inode(sb))
1619 inode = new_inode(sb);
1621 inode->i_ino = get_next_ino();
1622 inode_init_owner(inode, dir, mode);
1623 inode->i_blocks = 0;
1624 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1625 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1626 inode->i_generation = get_seconds();
1627 info = SHMEM_I(inode);
1628 memset(info, 0, (char *)inode - (char *)info);
1629 spin_lock_init(&info->lock);
1630 info->flags = flags & VM_NORESERVE;
1631 INIT_LIST_HEAD(&info->swaplist);
1632 INIT_LIST_HEAD(&info->xattr_list);
1633 cache_no_acl(inode);
1635 switch (mode & S_IFMT) {
1637 inode->i_op = &shmem_special_inode_operations;
1638 init_special_inode(inode, mode, dev);
1641 inode->i_mapping->a_ops = &shmem_aops;
1642 inode->i_op = &shmem_inode_operations;
1643 inode->i_fop = &shmem_file_operations;
1644 mpol_shared_policy_init(&info->policy,
1645 shmem_get_sbmpol(sbinfo));
1649 /* Some things misbehave if size == 0 on a directory */
1650 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1651 inode->i_op = &shmem_dir_inode_operations;
1652 inode->i_fop = &simple_dir_operations;
1656 * Must not load anything in the rbtree,
1657 * mpol_free_shared_policy will not be called.
1659 mpol_shared_policy_init(&info->policy, NULL);
1663 shmem_free_inode(sb);
1668 static const struct inode_operations shmem_symlink_inode_operations;
1669 static const struct inode_operations shmem_symlink_inline_operations;
1672 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1673 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1674 * below the loop driver, in the generic fashion that many filesystems support.
1676 static int shmem_readpage(struct file *file, struct page *page)
1678 struct inode *inode = page->mapping->host;
1679 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1685 shmem_write_begin(struct file *file, struct address_space *mapping,
1686 loff_t pos, unsigned len, unsigned flags,
1687 struct page **pagep, void **fsdata)
1689 struct inode *inode = mapping->host;
1690 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1692 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1696 shmem_write_end(struct file *file, struct address_space *mapping,
1697 loff_t pos, unsigned len, unsigned copied,
1698 struct page *page, void *fsdata)
1700 struct inode *inode = mapping->host;
1702 if (pos + copied > inode->i_size)
1703 i_size_write(inode, pos + copied);
1705 set_page_dirty(page);
1707 page_cache_release(page);
1712 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1714 struct inode *inode = filp->f_path.dentry->d_inode;
1715 struct address_space *mapping = inode->i_mapping;
1716 unsigned long index, offset;
1717 enum sgp_type sgp = SGP_READ;
1720 * Might this read be for a stacking filesystem? Then when reading
1721 * holes of a sparse file, we actually need to allocate those pages,
1722 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1724 if (segment_eq(get_fs(), KERNEL_DS))
1727 index = *ppos >> PAGE_CACHE_SHIFT;
1728 offset = *ppos & ~PAGE_CACHE_MASK;
1731 struct page *page = NULL;
1732 unsigned long end_index, nr, ret;
1733 loff_t i_size = i_size_read(inode);
1735 end_index = i_size >> PAGE_CACHE_SHIFT;
1736 if (index > end_index)
1738 if (index == end_index) {
1739 nr = i_size & ~PAGE_CACHE_MASK;
1744 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1746 if (desc->error == -EINVAL)
1754 * We must evaluate after, since reads (unlike writes)
1755 * are called without i_mutex protection against truncate
1757 nr = PAGE_CACHE_SIZE;
1758 i_size = i_size_read(inode);
1759 end_index = i_size >> PAGE_CACHE_SHIFT;
1760 if (index == end_index) {
1761 nr = i_size & ~PAGE_CACHE_MASK;
1764 page_cache_release(page);
1772 * If users can be writing to this page using arbitrary
1773 * virtual addresses, take care about potential aliasing
1774 * before reading the page on the kernel side.
1776 if (mapping_writably_mapped(mapping))
1777 flush_dcache_page(page);
1779 * Mark the page accessed if we read the beginning.
1782 mark_page_accessed(page);
1784 page = ZERO_PAGE(0);
1785 page_cache_get(page);
1789 * Ok, we have the page, and it's up-to-date, so
1790 * now we can copy it to user space...
1792 * The actor routine returns how many bytes were actually used..
1793 * NOTE! This may not be the same as how much of a user buffer
1794 * we filled up (we may be padding etc), so we can only update
1795 * "pos" here (the actor routine has to update the user buffer
1796 * pointers and the remaining count).
1798 ret = actor(desc, page, offset, nr);
1800 index += offset >> PAGE_CACHE_SHIFT;
1801 offset &= ~PAGE_CACHE_MASK;
1803 page_cache_release(page);
1804 if (ret != nr || !desc->count)
1810 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1811 file_accessed(filp);
1814 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1815 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1817 struct file *filp = iocb->ki_filp;
1821 loff_t *ppos = &iocb->ki_pos;
1823 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1827 for (seg = 0; seg < nr_segs; seg++) {
1828 read_descriptor_t desc;
1831 desc.arg.buf = iov[seg].iov_base;
1832 desc.count = iov[seg].iov_len;
1833 if (desc.count == 0)
1836 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1837 retval += desc.written;
1839 retval = retval ?: desc.error;
1848 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1849 struct pipe_inode_info *pipe, size_t len,
1852 struct address_space *mapping = in->f_mapping;
1853 unsigned int loff, nr_pages, req_pages;
1854 struct page *pages[PIPE_DEF_BUFFERS];
1855 struct partial_page partial[PIPE_DEF_BUFFERS];
1857 pgoff_t index, end_index;
1860 struct splice_pipe_desc spd = {
1864 .ops = &page_cache_pipe_buf_ops,
1865 .spd_release = spd_release_page,
1868 isize = i_size_read(in->f_mapping->host);
1869 if (unlikely(*ppos >= isize))
1872 left = isize - *ppos;
1873 if (unlikely(left < len))
1876 if (splice_grow_spd(pipe, &spd))
1879 index = *ppos >> PAGE_CACHE_SHIFT;
1880 loff = *ppos & ~PAGE_CACHE_MASK;
1881 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1882 nr_pages = min(req_pages, pipe->buffers);
1885 * Lookup the (hopefully) full range of pages we need.
1887 spd.nr_pages = find_get_pages_contig(mapping, index,
1888 nr_pages, spd.pages);
1889 index += spd.nr_pages;
1892 * If find_get_pages_contig() returned fewer pages than we needed,
1893 * readahead/allocate the rest and fill in the holes.
1895 if (spd.nr_pages < nr_pages)
1896 page_cache_sync_readahead(mapping, &in->f_ra, in,
1897 index, req_pages - spd.nr_pages);
1900 while (spd.nr_pages < nr_pages) {
1902 * Page could be there, find_get_pages_contig() breaks on
1905 page = find_get_page(mapping, index);
1908 * page didn't exist, allocate one.
1910 page = page_cache_alloc_cold(mapping);
1914 error = add_to_page_cache_lru(page, mapping, index,
1916 if (unlikely(error)) {
1917 page_cache_release(page);
1918 if (error == -EEXIST)
1923 * add_to_page_cache() locks the page, unlock it
1924 * to avoid convoluting the logic below even more.
1929 spd.pages[spd.nr_pages++] = page;
1934 * Now loop over the map and see if we need to start IO on any
1935 * pages, fill in the partial map, etc.
1937 index = *ppos >> PAGE_CACHE_SHIFT;
1938 nr_pages = spd.nr_pages;
1940 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1941 unsigned int this_len;
1947 * this_len is the max we'll use from this page
1949 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1950 page = spd.pages[page_nr];
1952 if (PageReadahead(page))
1953 page_cache_async_readahead(mapping, &in->f_ra, in,
1954 page, index, req_pages - page_nr);
1957 * If the page isn't uptodate, we may need to start io on it
1959 if (!PageUptodate(page)) {
1963 * Page was truncated, or invalidated by the
1964 * filesystem. Redo the find/create, but this time the
1965 * page is kept locked, so there's no chance of another
1966 * race with truncate/invalidate.
1968 if (!page->mapping) {
1970 page = find_or_create_page(mapping, index,
1971 mapping_gfp_mask(mapping));
1977 page_cache_release(spd.pages[page_nr]);
1978 spd.pages[page_nr] = page;
1981 * page was already under io and is now done, great
1983 if (PageUptodate(page)) {
1989 * need to read in the page
1991 error = mapping->a_ops->readpage(in, page);
1992 if (unlikely(error)) {
1994 * We really should re-lookup the page here,
1995 * but it complicates things a lot. Instead
1996 * lets just do what we already stored, and
1997 * we'll get it the next time we are called.
1999 if (error == AOP_TRUNCATED_PAGE)
2007 * i_size must be checked after PageUptodate.
2009 isize = i_size_read(mapping->host);
2010 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
2011 if (unlikely(!isize || index > end_index))
2015 * if this is the last page, see if we need to shrink
2016 * the length and stop
2018 if (end_index == index) {
2022 * max good bytes in this page
2024 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
2029 * force quit after adding this page
2031 this_len = min(this_len, plen - loff);
2035 spd.partial[page_nr].offset = loff;
2036 spd.partial[page_nr].len = this_len;
2044 * Release any pages at the end, if we quit early. 'page_nr' is how far
2045 * we got, 'nr_pages' is how many pages are in the map.
2047 while (page_nr < nr_pages)
2048 page_cache_release(spd.pages[page_nr++]);
2049 in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
2052 error = splice_to_pipe(pipe, &spd);
2054 splice_shrink_spd(pipe, &spd);
2063 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2065 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2067 buf->f_type = TMPFS_MAGIC;
2068 buf->f_bsize = PAGE_CACHE_SIZE;
2069 buf->f_namelen = NAME_MAX;
2070 if (sbinfo->max_blocks) {
2071 buf->f_blocks = sbinfo->max_blocks;
2072 buf->f_bavail = buf->f_bfree =
2073 sbinfo->max_blocks - percpu_counter_sum(&sbinfo->used_blocks);
2075 if (sbinfo->max_inodes) {
2076 buf->f_files = sbinfo->max_inodes;
2077 buf->f_ffree = sbinfo->free_inodes;
2079 /* else leave those fields 0 like simple_statfs */
2084 * File creation. Allocate an inode, and we're done..
2087 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2089 struct inode *inode;
2090 int error = -ENOSPC;
2092 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2094 error = security_inode_init_security(inode, dir,
2095 &dentry->d_name, NULL,
2098 if (error != -EOPNOTSUPP) {
2103 #ifdef CONFIG_TMPFS_POSIX_ACL
2104 error = generic_acl_init(inode, dir);
2112 dir->i_size += BOGO_DIRENT_SIZE;
2113 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2114 d_instantiate(dentry, inode);
2115 dget(dentry); /* Extra count - pin the dentry in core */
2120 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2124 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2130 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
2131 struct nameidata *nd)
2133 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2139 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2141 struct inode *inode = old_dentry->d_inode;
2145 * No ordinary (disk based) filesystem counts links as inodes;
2146 * but each new link needs a new dentry, pinning lowmem, and
2147 * tmpfs dentries cannot be pruned until they are unlinked.
2149 ret = shmem_reserve_inode(inode->i_sb);
2153 dir->i_size += BOGO_DIRENT_SIZE;
2154 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2156 ihold(inode); /* New dentry reference */
2157 dget(dentry); /* Extra pinning count for the created dentry */
2158 d_instantiate(dentry, inode);
2163 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2165 struct inode *inode = dentry->d_inode;
2167 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2168 shmem_free_inode(inode->i_sb);
2170 dir->i_size -= BOGO_DIRENT_SIZE;
2171 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2173 dput(dentry); /* Undo the count from "create" - this does all the work */
2177 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2179 if (!simple_empty(dentry))
2182 drop_nlink(dentry->d_inode);
2184 return shmem_unlink(dir, dentry);
2188 * The VFS layer already does all the dentry stuff for rename,
2189 * we just have to decrement the usage count for the target if
2190 * it exists so that the VFS layer correctly free's it when it
2193 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2195 struct inode *inode = old_dentry->d_inode;
2196 int they_are_dirs = S_ISDIR(inode->i_mode);
2198 if (!simple_empty(new_dentry))
2201 if (new_dentry->d_inode) {
2202 (void) shmem_unlink(new_dir, new_dentry);
2204 drop_nlink(old_dir);
2205 } else if (they_are_dirs) {
2206 drop_nlink(old_dir);
2210 old_dir->i_size -= BOGO_DIRENT_SIZE;
2211 new_dir->i_size += BOGO_DIRENT_SIZE;
2212 old_dir->i_ctime = old_dir->i_mtime =
2213 new_dir->i_ctime = new_dir->i_mtime =
2214 inode->i_ctime = CURRENT_TIME;
2218 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2222 struct inode *inode;
2223 struct page *page = NULL;
2225 struct shmem_inode_info *info;
2227 len = strlen(symname) + 1;
2228 if (len > PAGE_CACHE_SIZE)
2229 return -ENAMETOOLONG;
2231 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2235 error = security_inode_init_security(inode, dir, &dentry->d_name, NULL,
2238 if (error != -EOPNOTSUPP) {
2245 info = SHMEM_I(inode);
2246 inode->i_size = len-1;
2247 if (len <= SHMEM_SYMLINK_INLINE_LEN) {
2249 memcpy(info->inline_symlink, symname, len);
2250 inode->i_op = &shmem_symlink_inline_operations;
2252 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2257 inode->i_mapping->a_ops = &shmem_aops;
2258 inode->i_op = &shmem_symlink_inode_operations;
2259 kaddr = kmap_atomic(page, KM_USER0);
2260 memcpy(kaddr, symname, len);
2261 kunmap_atomic(kaddr, KM_USER0);
2262 set_page_dirty(page);
2264 page_cache_release(page);
2266 dir->i_size += BOGO_DIRENT_SIZE;
2267 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2268 d_instantiate(dentry, inode);
2273 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
2275 nd_set_link(nd, SHMEM_I(dentry->d_inode)->inline_symlink);
2279 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2281 struct page *page = NULL;
2282 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2283 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2289 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2291 if (!IS_ERR(nd_get_link(nd))) {
2292 struct page *page = cookie;
2294 mark_page_accessed(page);
2295 page_cache_release(page);
2299 #ifdef CONFIG_TMPFS_XATTR
2301 * Superblocks without xattr inode operations may get some security.* xattr
2302 * support from the LSM "for free". As soon as we have any other xattrs
2303 * like ACLs, we also need to implement the security.* handlers at
2304 * filesystem level, though.
2307 static int shmem_xattr_get(struct dentry *dentry, const char *name,
2308 void *buffer, size_t size)
2310 struct shmem_inode_info *info;
2311 struct shmem_xattr *xattr;
2314 info = SHMEM_I(dentry->d_inode);
2316 spin_lock(&info->lock);
2317 list_for_each_entry(xattr, &info->xattr_list, list) {
2318 if (strcmp(name, xattr->name))
2323 if (size < xattr->size)
2326 memcpy(buffer, xattr->value, xattr->size);
2330 spin_unlock(&info->lock);
2334 static int shmem_xattr_set(struct dentry *dentry, const char *name,
2335 const void *value, size_t size, int flags)
2337 struct inode *inode = dentry->d_inode;
2338 struct shmem_inode_info *info = SHMEM_I(inode);
2339 struct shmem_xattr *xattr;
2340 struct shmem_xattr *new_xattr = NULL;
2344 /* value == NULL means remove */
2347 len = sizeof(*new_xattr) + size;
2348 if (len <= sizeof(*new_xattr))
2351 new_xattr = kmalloc(len, GFP_KERNEL);
2355 new_xattr->name = kstrdup(name, GFP_KERNEL);
2356 if (!new_xattr->name) {
2361 new_xattr->size = size;
2362 memcpy(new_xattr->value, value, size);
2365 spin_lock(&info->lock);
2366 list_for_each_entry(xattr, &info->xattr_list, list) {
2367 if (!strcmp(name, xattr->name)) {
2368 if (flags & XATTR_CREATE) {
2371 } else if (new_xattr) {
2372 list_replace(&xattr->list, &new_xattr->list);
2374 list_del(&xattr->list);
2379 if (flags & XATTR_REPLACE) {
2383 list_add(&new_xattr->list, &info->xattr_list);
2387 spin_unlock(&info->lock);
2395 static const struct xattr_handler *shmem_xattr_handlers[] = {
2396 #ifdef CONFIG_TMPFS_POSIX_ACL
2397 &generic_acl_access_handler,
2398 &generic_acl_default_handler,
2403 static int shmem_xattr_validate(const char *name)
2405 struct { const char *prefix; size_t len; } arr[] = {
2406 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2407 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2411 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2412 size_t preflen = arr[i].len;
2413 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2422 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2423 void *buffer, size_t size)
2428 * If this is a request for a synthetic attribute in the system.*
2429 * namespace use the generic infrastructure to resolve a handler
2430 * for it via sb->s_xattr.
2432 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2433 return generic_getxattr(dentry, name, buffer, size);
2435 err = shmem_xattr_validate(name);
2439 return shmem_xattr_get(dentry, name, buffer, size);
2442 static int shmem_setxattr(struct dentry *dentry, const char *name,
2443 const void *value, size_t size, int flags)
2448 * If this is a request for a synthetic attribute in the system.*
2449 * namespace use the generic infrastructure to resolve a handler
2450 * for it via sb->s_xattr.
2452 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2453 return generic_setxattr(dentry, name, value, size, flags);
2455 err = shmem_xattr_validate(name);
2460 value = ""; /* empty EA, do not remove */
2462 return shmem_xattr_set(dentry, name, value, size, flags);
2466 static int shmem_removexattr(struct dentry *dentry, const char *name)
2471 * If this is a request for a synthetic attribute in the system.*
2472 * namespace use the generic infrastructure to resolve a handler
2473 * for it via sb->s_xattr.
2475 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2476 return generic_removexattr(dentry, name);
2478 err = shmem_xattr_validate(name);
2482 return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
2485 static bool xattr_is_trusted(const char *name)
2487 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2490 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2492 bool trusted = capable(CAP_SYS_ADMIN);
2493 struct shmem_xattr *xattr;
2494 struct shmem_inode_info *info;
2497 info = SHMEM_I(dentry->d_inode);
2499 spin_lock(&info->lock);
2500 list_for_each_entry(xattr, &info->xattr_list, list) {
2503 /* skip "trusted." attributes for unprivileged callers */
2504 if (!trusted && xattr_is_trusted(xattr->name))
2507 len = strlen(xattr->name) + 1;
2514 memcpy(buffer, xattr->name, len);
2518 spin_unlock(&info->lock);
2522 #endif /* CONFIG_TMPFS_XATTR */
2524 static const struct inode_operations shmem_symlink_inline_operations = {
2525 .readlink = generic_readlink,
2526 .follow_link = shmem_follow_link_inline,
2527 #ifdef CONFIG_TMPFS_XATTR
2528 .setxattr = shmem_setxattr,
2529 .getxattr = shmem_getxattr,
2530 .listxattr = shmem_listxattr,
2531 .removexattr = shmem_removexattr,
2535 static const struct inode_operations shmem_symlink_inode_operations = {
2536 .readlink = generic_readlink,
2537 .follow_link = shmem_follow_link,
2538 .put_link = shmem_put_link,
2539 #ifdef CONFIG_TMPFS_XATTR
2540 .setxattr = shmem_setxattr,
2541 .getxattr = shmem_getxattr,
2542 .listxattr = shmem_listxattr,
2543 .removexattr = shmem_removexattr,
2547 static struct dentry *shmem_get_parent(struct dentry *child)
2549 return ERR_PTR(-ESTALE);
2552 static int shmem_match(struct inode *ino, void *vfh)
2556 inum = (inum << 32) | fh[1];
2557 return ino->i_ino == inum && fh[0] == ino->i_generation;
2560 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2561 struct fid *fid, int fh_len, int fh_type)
2563 struct inode *inode;
2564 struct dentry *dentry = NULL;
2565 u64 inum = fid->raw[2];
2566 inum = (inum << 32) | fid->raw[1];
2571 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2572 shmem_match, fid->raw);
2574 dentry = d_find_alias(inode);
2581 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2584 struct inode *inode = dentry->d_inode;
2591 if (inode_unhashed(inode)) {
2592 /* Unfortunately insert_inode_hash is not idempotent,
2593 * so as we hash inodes here rather than at creation
2594 * time, we need a lock to ensure we only try
2597 static DEFINE_SPINLOCK(lock);
2599 if (inode_unhashed(inode))
2600 __insert_inode_hash(inode,
2601 inode->i_ino + inode->i_generation);
2605 fh[0] = inode->i_generation;
2606 fh[1] = inode->i_ino;
2607 fh[2] = ((__u64)inode->i_ino) >> 32;
2613 static const struct export_operations shmem_export_ops = {
2614 .get_parent = shmem_get_parent,
2615 .encode_fh = shmem_encode_fh,
2616 .fh_to_dentry = shmem_fh_to_dentry,
2619 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2622 char *this_char, *value, *rest;
2624 while (options != NULL) {
2625 this_char = options;
2628 * NUL-terminate this option: unfortunately,
2629 * mount options form a comma-separated list,
2630 * but mpol's nodelist may also contain commas.
2632 options = strchr(options, ',');
2633 if (options == NULL)
2636 if (!isdigit(*options)) {
2643 if ((value = strchr(this_char,'=')) != NULL) {
2647 "tmpfs: No value for mount option '%s'\n",
2652 if (!strcmp(this_char,"size")) {
2653 unsigned long long size;
2654 size = memparse(value,&rest);
2656 size <<= PAGE_SHIFT;
2657 size *= totalram_pages;
2663 sbinfo->max_blocks =
2664 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2665 } else if (!strcmp(this_char,"nr_blocks")) {
2666 sbinfo->max_blocks = memparse(value, &rest);
2669 } else if (!strcmp(this_char,"nr_inodes")) {
2670 sbinfo->max_inodes = memparse(value, &rest);
2673 } else if (!strcmp(this_char,"mode")) {
2676 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2679 } else if (!strcmp(this_char,"uid")) {
2682 sbinfo->uid = simple_strtoul(value, &rest, 0);
2685 } else if (!strcmp(this_char,"gid")) {
2688 sbinfo->gid = simple_strtoul(value, &rest, 0);
2691 } else if (!strcmp(this_char,"mpol")) {
2692 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2695 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2703 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2709 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2711 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2712 struct shmem_sb_info config = *sbinfo;
2713 unsigned long inodes;
2714 int error = -EINVAL;
2716 if (shmem_parse_options(data, &config, true))
2719 spin_lock(&sbinfo->stat_lock);
2720 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2721 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2723 if (config.max_inodes < inodes)
2726 * Those tests also disallow limited->unlimited while any are in
2727 * use, so i_blocks will always be zero when max_blocks is zero;
2728 * but we must separately disallow unlimited->limited, because
2729 * in that case we have no record of how much is already in use.
2731 if (config.max_blocks && !sbinfo->max_blocks)
2733 if (config.max_inodes && !sbinfo->max_inodes)
2737 sbinfo->max_blocks = config.max_blocks;
2738 sbinfo->max_inodes = config.max_inodes;
2739 sbinfo->free_inodes = config.max_inodes - inodes;
2741 mpol_put(sbinfo->mpol);
2742 sbinfo->mpol = config.mpol; /* transfers initial ref */
2744 spin_unlock(&sbinfo->stat_lock);
2748 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2750 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2752 if (sbinfo->max_blocks != shmem_default_max_blocks())
2753 seq_printf(seq, ",size=%luk",
2754 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2755 if (sbinfo->max_inodes != shmem_default_max_inodes())
2756 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2757 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2758 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2759 if (sbinfo->uid != 0)
2760 seq_printf(seq, ",uid=%u", sbinfo->uid);
2761 if (sbinfo->gid != 0)
2762 seq_printf(seq, ",gid=%u", sbinfo->gid);
2763 shmem_show_mpol(seq, sbinfo->mpol);
2766 #endif /* CONFIG_TMPFS */
2768 static void shmem_put_super(struct super_block *sb)
2770 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2772 percpu_counter_destroy(&sbinfo->used_blocks);
2774 sb->s_fs_info = NULL;
2777 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2779 struct inode *inode;
2780 struct dentry *root;
2781 struct shmem_sb_info *sbinfo;
2784 /* Round up to L1_CACHE_BYTES to resist false sharing */
2785 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2786 L1_CACHE_BYTES), GFP_KERNEL);
2790 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2791 sbinfo->uid = current_fsuid();
2792 sbinfo->gid = current_fsgid();
2793 sb->s_fs_info = sbinfo;
2797 * Per default we only allow half of the physical ram per
2798 * tmpfs instance, limiting inodes to one per page of lowmem;
2799 * but the internal instance is left unlimited.
2801 if (!(sb->s_flags & MS_NOUSER)) {
2802 sbinfo->max_blocks = shmem_default_max_blocks();
2803 sbinfo->max_inodes = shmem_default_max_inodes();
2804 if (shmem_parse_options(data, sbinfo, false)) {
2809 sb->s_export_op = &shmem_export_ops;
2811 sb->s_flags |= MS_NOUSER;
2814 spin_lock_init(&sbinfo->stat_lock);
2815 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2817 sbinfo->free_inodes = sbinfo->max_inodes;
2819 sb->s_maxbytes = SHMEM_MAX_BYTES;
2820 sb->s_blocksize = PAGE_CACHE_SIZE;
2821 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2822 sb->s_magic = TMPFS_MAGIC;
2823 sb->s_op = &shmem_ops;
2824 sb->s_time_gran = 1;
2825 #ifdef CONFIG_TMPFS_XATTR
2826 sb->s_xattr = shmem_xattr_handlers;
2828 #ifdef CONFIG_TMPFS_POSIX_ACL
2829 sb->s_flags |= MS_POSIXACL;
2832 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2835 inode->i_uid = sbinfo->uid;
2836 inode->i_gid = sbinfo->gid;
2837 root = d_alloc_root(inode);
2846 shmem_put_super(sb);
2850 static struct kmem_cache *shmem_inode_cachep;
2852 static struct inode *shmem_alloc_inode(struct super_block *sb)
2854 struct shmem_inode_info *p;
2855 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2858 return &p->vfs_inode;
2861 static void shmem_i_callback(struct rcu_head *head)
2863 struct inode *inode = container_of(head, struct inode, i_rcu);
2864 INIT_LIST_HEAD(&inode->i_dentry);
2865 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2868 static void shmem_destroy_inode(struct inode *inode)
2870 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2871 /* only struct inode is valid if it's an inline symlink */
2872 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2874 call_rcu(&inode->i_rcu, shmem_i_callback);
2877 static void init_once(void *foo)
2879 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2881 inode_init_once(&p->vfs_inode);
2884 static int init_inodecache(void)
2886 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2887 sizeof(struct shmem_inode_info),
2888 0, SLAB_PANIC, init_once);
2892 static void destroy_inodecache(void)
2894 kmem_cache_destroy(shmem_inode_cachep);
2897 static const struct address_space_operations shmem_aops = {
2898 .writepage = shmem_writepage,
2899 .set_page_dirty = __set_page_dirty_no_writeback,
2901 .readpage = shmem_readpage,
2902 .write_begin = shmem_write_begin,
2903 .write_end = shmem_write_end,
2905 .migratepage = migrate_page,
2906 .error_remove_page = generic_error_remove_page,
2909 static const struct file_operations shmem_file_operations = {
2912 .llseek = generic_file_llseek,
2913 .read = do_sync_read,
2914 .write = do_sync_write,
2915 .aio_read = shmem_file_aio_read,
2916 .aio_write = generic_file_aio_write,
2917 .fsync = noop_fsync,
2918 .splice_read = shmem_file_splice_read,
2919 .splice_write = generic_file_splice_write,
2923 static const struct inode_operations shmem_inode_operations = {
2924 .setattr = shmem_setattr,
2925 .truncate_range = shmem_truncate_range,
2926 #ifdef CONFIG_TMPFS_XATTR
2927 .setxattr = shmem_setxattr,
2928 .getxattr = shmem_getxattr,
2929 .listxattr = shmem_listxattr,
2930 .removexattr = shmem_removexattr,
2932 #ifdef CONFIG_TMPFS_POSIX_ACL
2933 .check_acl = generic_check_acl,
2938 static const struct inode_operations shmem_dir_inode_operations = {
2940 .create = shmem_create,
2941 .lookup = simple_lookup,
2943 .unlink = shmem_unlink,
2944 .symlink = shmem_symlink,
2945 .mkdir = shmem_mkdir,
2946 .rmdir = shmem_rmdir,
2947 .mknod = shmem_mknod,
2948 .rename = shmem_rename,
2950 #ifdef CONFIG_TMPFS_XATTR
2951 .setxattr = shmem_setxattr,
2952 .getxattr = shmem_getxattr,
2953 .listxattr = shmem_listxattr,
2954 .removexattr = shmem_removexattr,
2956 #ifdef CONFIG_TMPFS_POSIX_ACL
2957 .setattr = shmem_setattr,
2958 .check_acl = generic_check_acl,
2962 static const struct inode_operations shmem_special_inode_operations = {
2963 #ifdef CONFIG_TMPFS_XATTR
2964 .setxattr = shmem_setxattr,
2965 .getxattr = shmem_getxattr,
2966 .listxattr = shmem_listxattr,
2967 .removexattr = shmem_removexattr,
2969 #ifdef CONFIG_TMPFS_POSIX_ACL
2970 .setattr = shmem_setattr,
2971 .check_acl = generic_check_acl,
2975 static const struct super_operations shmem_ops = {
2976 .alloc_inode = shmem_alloc_inode,
2977 .destroy_inode = shmem_destroy_inode,
2979 .statfs = shmem_statfs,
2980 .remount_fs = shmem_remount_fs,
2981 .show_options = shmem_show_options,
2983 .evict_inode = shmem_evict_inode,
2984 .drop_inode = generic_delete_inode,
2985 .put_super = shmem_put_super,
2988 static const struct vm_operations_struct shmem_vm_ops = {
2989 .fault = shmem_fault,
2991 .set_policy = shmem_set_policy,
2992 .get_policy = shmem_get_policy,
2997 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2998 int flags, const char *dev_name, void *data)
3000 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3003 static struct file_system_type tmpfs_fs_type = {
3004 .owner = THIS_MODULE,
3006 .mount = shmem_mount,
3007 .kill_sb = kill_litter_super,
3010 int __init init_tmpfs(void)
3014 error = bdi_init(&shmem_backing_dev_info);
3018 error = init_inodecache();
3022 error = register_filesystem(&tmpfs_fs_type);
3024 printk(KERN_ERR "Could not register tmpfs\n");
3028 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
3029 tmpfs_fs_type.name, NULL);
3030 if (IS_ERR(shm_mnt)) {
3031 error = PTR_ERR(shm_mnt);
3032 printk(KERN_ERR "Could not kern_mount tmpfs\n");
3038 unregister_filesystem(&tmpfs_fs_type);
3040 destroy_inodecache();
3042 bdi_destroy(&shmem_backing_dev_info);
3044 shm_mnt = ERR_PTR(error);
3048 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
3050 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
3051 * @inode: the inode to be searched
3052 * @pgoff: the offset to be searched
3053 * @pagep: the pointer for the found page to be stored
3054 * @ent: the pointer for the found swap entry to be stored
3056 * If a page is found, refcount of it is incremented. Callers should handle
3059 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
3060 struct page **pagep, swp_entry_t *ent)
3062 swp_entry_t entry = { .val = 0 }, *ptr;
3063 struct page *page = NULL;
3064 struct shmem_inode_info *info = SHMEM_I(inode);
3066 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
3069 spin_lock(&info->lock);
3070 ptr = shmem_swp_entry(info, pgoff, NULL);
3072 if (ptr && ptr->val) {
3073 entry.val = ptr->val;
3074 page = find_get_page(&swapper_space, entry.val);
3077 page = find_get_page(inode->i_mapping, pgoff);
3079 shmem_swp_unmap(ptr);
3080 spin_unlock(&info->lock);
3087 #else /* !CONFIG_SHMEM */
3090 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3092 * This is intended for small system where the benefits of the full
3093 * shmem code (swap-backed and resource-limited) are outweighed by
3094 * their complexity. On systems without swap this code should be
3095 * effectively equivalent, but much lighter weight.
3098 #include <linux/ramfs.h>
3100 static struct file_system_type tmpfs_fs_type = {
3102 .mount = ramfs_mount,
3103 .kill_sb = kill_litter_super,
3106 int __init init_tmpfs(void)
3108 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
3110 shm_mnt = kern_mount(&tmpfs_fs_type);
3111 BUG_ON(IS_ERR(shm_mnt));
3116 int shmem_unuse(swp_entry_t entry, struct page *page)
3121 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3126 void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
3128 truncate_inode_pages_range(inode->i_mapping, start, end);
3130 EXPORT_SYMBOL_GPL(shmem_truncate_range);
3132 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
3134 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
3135 * @inode: the inode to be searched
3136 * @pgoff: the offset to be searched
3137 * @pagep: the pointer for the found page to be stored
3138 * @ent: the pointer for the found swap entry to be stored
3140 * If a page is found, refcount of it is incremented. Callers should handle
3143 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
3144 struct page **pagep, swp_entry_t *ent)
3146 struct page *page = NULL;
3148 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
3150 page = find_get_page(inode->i_mapping, pgoff);
3153 *ent = (swp_entry_t){ .val = 0 };
3157 #define shmem_vm_ops generic_file_vm_ops
3158 #define shmem_file_operations ramfs_file_operations
3159 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3160 #define shmem_acct_size(flags, size) 0
3161 #define shmem_unacct_size(flags, size) do {} while (0)
3162 #define SHMEM_MAX_BYTES MAX_LFS_FILESIZE
3164 #endif /* CONFIG_SHMEM */
3169 * shmem_file_setup - get an unlinked file living in tmpfs
3170 * @name: name for dentry (to be seen in /proc/<pid>/maps
3171 * @size: size to be set for the file
3172 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3174 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3178 struct inode *inode;
3180 struct dentry *root;
3183 if (IS_ERR(shm_mnt))
3184 return (void *)shm_mnt;
3186 if (size < 0 || size > SHMEM_MAX_BYTES)
3187 return ERR_PTR(-EINVAL);
3189 if (shmem_acct_size(flags, size))
3190 return ERR_PTR(-ENOMEM);
3194 this.len = strlen(name);
3195 this.hash = 0; /* will go */
3196 root = shm_mnt->mnt_root;
3197 path.dentry = d_alloc(root, &this);
3200 path.mnt = mntget(shm_mnt);
3203 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
3207 d_instantiate(path.dentry, inode);
3208 inode->i_size = size;
3209 inode->i_nlink = 0; /* It is unlinked */
3211 error = ramfs_nommu_expand_for_mapping(inode, size);
3217 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
3218 &shmem_file_operations);
3227 shmem_unacct_size(flags, size);
3228 return ERR_PTR(error);
3230 EXPORT_SYMBOL_GPL(shmem_file_setup);
3233 * shmem_zero_setup - setup a shared anonymous mapping
3234 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3236 int shmem_zero_setup(struct vm_area_struct *vma)
3239 loff_t size = vma->vm_end - vma->vm_start;
3241 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3243 return PTR_ERR(file);
3247 vma->vm_file = file;
3248 vma->vm_ops = &shmem_vm_ops;
3249 vma->vm_flags |= VM_CAN_NONLINEAR;
3254 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3255 * @mapping: the page's address_space
3256 * @index: the page index
3257 * @gfp: the page allocator flags to use if allocating
3259 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3260 * with any new page allocations done using the specified allocation flags.
3261 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3262 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3263 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3265 * Provide a stub for those callers to start using now, then later
3266 * flesh it out to call shmem_getpage() with additional gfp mask, when
3267 * shmem_file_splice_read() is added and shmem_readpage() is removed.
3269 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3270 pgoff_t index, gfp_t gfp)
3272 return read_cache_page_gfp(mapping, index, gfp);
3274 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);