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/swap.h>
33 static struct vfsmount *shm_mnt;
37 * This virtual memory filesystem is heavily based on the ramfs. It
38 * extends ramfs by the ability to use swap and honor resource limits
39 * which makes it a completely usable filesystem.
42 #include <linux/xattr.h>
43 #include <linux/exportfs.h>
44 #include <linux/posix_acl.h>
45 #include <linux/generic_acl.h>
46 #include <linux/mman.h>
47 #include <linux/string.h>
48 #include <linux/slab.h>
49 #include <linux/backing-dev.h>
50 #include <linux/shmem_fs.h>
51 #include <linux/writeback.h>
52 #include <linux/blkdev.h>
53 #include <linux/pagevec.h>
54 #include <linux/percpu_counter.h>
55 #include <linux/splice.h>
56 #include <linux/security.h>
57 #include <linux/swapops.h>
58 #include <linux/mempolicy.h>
59 #include <linux/namei.h>
60 #include <linux/ctype.h>
61 #include <linux/migrate.h>
62 #include <linux/highmem.h>
63 #include <linux/seq_file.h>
64 #include <linux/magic.h>
66 #include <asm/uaccess.h>
67 #include <asm/pgtable.h>
69 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
70 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
72 /* Pretend that each entry is of this size in directory's i_size */
73 #define BOGO_DIRENT_SIZE 20
76 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
77 char *name; /* xattr name */
82 /* Flag allocation requirements to shmem_getpage */
84 SGP_READ, /* don't exceed i_size, don't allocate page */
85 SGP_CACHE, /* don't exceed i_size, may allocate page */
86 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
87 SGP_WRITE, /* may exceed i_size, may allocate page */
91 static unsigned long shmem_default_max_blocks(void)
93 return totalram_pages / 2;
96 static unsigned long shmem_default_max_inodes(void)
98 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
102 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
103 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
105 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
106 struct page **pagep, enum sgp_type sgp, int *fault_type)
108 return shmem_getpage_gfp(inode, index, pagep, sgp,
109 mapping_gfp_mask(inode->i_mapping), fault_type);
112 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
114 return sb->s_fs_info;
118 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
119 * for shared memory and for shared anonymous (/dev/zero) mappings
120 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
121 * consistent with the pre-accounting of private mappings ...
123 static inline int shmem_acct_size(unsigned long flags, loff_t size)
125 return (flags & VM_NORESERVE) ?
126 0 : security_vm_enough_memory_kern(VM_ACCT(size));
129 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
131 if (!(flags & VM_NORESERVE))
132 vm_unacct_memory(VM_ACCT(size));
136 * ... whereas tmpfs objects are accounted incrementally as
137 * pages are allocated, in order to allow huge sparse files.
138 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
139 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
141 static inline int shmem_acct_block(unsigned long flags)
143 return (flags & VM_NORESERVE) ?
144 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
147 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
149 if (flags & VM_NORESERVE)
150 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
153 static const struct super_operations shmem_ops;
154 static const struct address_space_operations shmem_aops;
155 static const struct file_operations shmem_file_operations;
156 static const struct inode_operations shmem_inode_operations;
157 static const struct inode_operations shmem_dir_inode_operations;
158 static const struct inode_operations shmem_special_inode_operations;
159 static const struct vm_operations_struct shmem_vm_ops;
161 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
162 .ra_pages = 0, /* No readahead */
163 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
166 static LIST_HEAD(shmem_swaplist);
167 static DEFINE_MUTEX(shmem_swaplist_mutex);
169 static int shmem_reserve_inode(struct super_block *sb)
171 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
172 if (sbinfo->max_inodes) {
173 spin_lock(&sbinfo->stat_lock);
174 if (!sbinfo->free_inodes) {
175 spin_unlock(&sbinfo->stat_lock);
178 sbinfo->free_inodes--;
179 spin_unlock(&sbinfo->stat_lock);
184 static void shmem_free_inode(struct super_block *sb)
186 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
187 if (sbinfo->max_inodes) {
188 spin_lock(&sbinfo->stat_lock);
189 sbinfo->free_inodes++;
190 spin_unlock(&sbinfo->stat_lock);
195 * shmem_recalc_inode - recalculate the block usage of an inode
196 * @inode: inode to recalc
198 * We have to calculate the free blocks since the mm can drop
199 * undirtied hole pages behind our back.
201 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
202 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
204 * It has to be called with the spinlock held.
206 static void shmem_recalc_inode(struct inode *inode)
208 struct shmem_inode_info *info = SHMEM_I(inode);
211 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
214 if (sbinfo->max_blocks)
215 percpu_counter_add(&sbinfo->used_blocks, -freed);
216 info->alloced -= freed;
217 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
218 shmem_unacct_blocks(info->flags, freed);
222 static void shmem_put_swap(struct shmem_inode_info *info, pgoff_t index,
225 if (index < SHMEM_NR_DIRECT)
226 info->i_direct[index] = swap;
229 static swp_entry_t shmem_get_swap(struct shmem_inode_info *info, pgoff_t index)
231 return (index < SHMEM_NR_DIRECT) ?
232 info->i_direct[index] : (swp_entry_t){0};
236 * Replace item expected in radix tree by a new item, while holding tree lock.
238 static int shmem_radix_tree_replace(struct address_space *mapping,
239 pgoff_t index, void *expected, void *replacement)
244 VM_BUG_ON(!expected);
245 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
247 item = radix_tree_deref_slot_protected(pslot,
248 &mapping->tree_lock);
249 if (item != expected)
252 radix_tree_replace_slot(pslot, replacement);
254 radix_tree_delete(&mapping->page_tree, index);
259 * Like add_to_page_cache_locked, but error if expected item has gone.
261 static int shmem_add_to_page_cache(struct page *page,
262 struct address_space *mapping,
263 pgoff_t index, gfp_t gfp, void *expected)
267 VM_BUG_ON(!PageLocked(page));
268 VM_BUG_ON(!PageSwapBacked(page));
270 error = mem_cgroup_cache_charge(page, current->mm,
271 gfp & GFP_RECLAIM_MASK);
275 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
277 page_cache_get(page);
278 page->mapping = mapping;
281 spin_lock_irq(&mapping->tree_lock);
283 error = radix_tree_insert(&mapping->page_tree,
286 error = shmem_radix_tree_replace(mapping, index,
290 __inc_zone_page_state(page, NR_FILE_PAGES);
291 __inc_zone_page_state(page, NR_SHMEM);
292 spin_unlock_irq(&mapping->tree_lock);
294 page->mapping = NULL;
295 spin_unlock_irq(&mapping->tree_lock);
296 page_cache_release(page);
299 radix_tree_preload_end();
302 mem_cgroup_uncharge_cache_page(page);
308 * Like find_get_pages, but collecting swap entries as well as pages.
310 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
311 pgoff_t start, unsigned int nr_pages,
312 struct page **pages, pgoff_t *indices)
316 unsigned int nr_found;
320 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
321 (void ***)pages, indices, start, nr_pages);
323 for (i = 0; i < nr_found; i++) {
326 page = radix_tree_deref_slot((void **)pages[i]);
329 if (radix_tree_exception(page)) {
330 if (radix_tree_exceptional_entry(page))
332 /* radix_tree_deref_retry(page) */
335 if (!page_cache_get_speculative(page))
338 /* Has the page moved? */
339 if (unlikely(page != *((void **)pages[i]))) {
340 page_cache_release(page);
344 indices[ret] = indices[i];
348 if (unlikely(!ret && nr_found))
355 * Lockless lookup of swap entry in radix tree, avoiding refcount on pages.
357 static pgoff_t shmem_find_swap(struct address_space *mapping, void *radswap)
359 void **slots[PAGEVEC_SIZE];
360 pgoff_t indices[PAGEVEC_SIZE];
361 unsigned int nr_found;
367 pgoff_t index = indices[nr_found - 1] + 1;
371 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
372 slots, indices, index, PAGEVEC_SIZE);
373 for (i = 0; i < nr_found; i++) {
374 void *item = radix_tree_deref_slot(slots[i]);
375 if (radix_tree_deref_retry(item)) {
379 if (item == radswap) {
391 * Remove swap entry from radix tree, free the swap and its page cache.
393 static int shmem_free_swap(struct address_space *mapping,
394 pgoff_t index, void *radswap)
398 spin_lock_irq(&mapping->tree_lock);
399 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
400 spin_unlock_irq(&mapping->tree_lock);
402 free_swap_and_cache(radix_to_swp_entry(radswap));
407 * Pagevec may contain swap entries, so shuffle up pages before releasing.
409 static void shmem_pagevec_release(struct pagevec *pvec)
413 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
414 struct page *page = pvec->pages[i];
415 if (!radix_tree_exceptional_entry(page))
416 pvec->pages[j++] = page;
419 pagevec_release(pvec);
423 * Remove range of pages and swap entries from radix tree, and free them.
425 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
427 struct address_space *mapping = inode->i_mapping;
428 struct shmem_inode_info *info = SHMEM_I(inode);
429 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
430 unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
431 pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
433 pgoff_t indices[PAGEVEC_SIZE];
434 long nr_swaps_freed = 0;
438 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
440 pagevec_init(&pvec, 0);
442 while (index <= end) {
443 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
444 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
445 pvec.pages, indices);
448 mem_cgroup_uncharge_start();
449 for (i = 0; i < pagevec_count(&pvec); i++) {
450 struct page *page = pvec.pages[i];
456 if (radix_tree_exceptional_entry(page)) {
457 nr_swaps_freed += !shmem_free_swap(mapping,
462 if (!trylock_page(page))
464 if (page->mapping == mapping) {
465 VM_BUG_ON(PageWriteback(page));
466 truncate_inode_page(mapping, page);
470 shmem_pagevec_release(&pvec);
471 mem_cgroup_uncharge_end();
477 struct page *page = NULL;
478 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
480 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
481 set_page_dirty(page);
483 page_cache_release(page);
490 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
491 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
492 pvec.pages, indices);
499 if (index == start && indices[0] > end) {
500 shmem_pagevec_release(&pvec);
503 mem_cgroup_uncharge_start();
504 for (i = 0; i < pagevec_count(&pvec); i++) {
505 struct page *page = pvec.pages[i];
511 if (radix_tree_exceptional_entry(page)) {
512 nr_swaps_freed += !shmem_free_swap(mapping,
518 if (page->mapping == mapping) {
519 VM_BUG_ON(PageWriteback(page));
520 truncate_inode_page(mapping, page);
524 shmem_pagevec_release(&pvec);
525 mem_cgroup_uncharge_end();
529 spin_lock(&info->lock);
530 info->swapped -= nr_swaps_freed;
531 shmem_recalc_inode(inode);
532 spin_unlock(&info->lock);
534 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
536 EXPORT_SYMBOL_GPL(shmem_truncate_range);
538 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
540 struct inode *inode = dentry->d_inode;
543 error = inode_change_ok(inode, attr);
547 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
548 loff_t oldsize = inode->i_size;
549 loff_t newsize = attr->ia_size;
551 if (newsize != oldsize) {
552 i_size_write(inode, newsize);
553 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
555 if (newsize < oldsize) {
556 loff_t holebegin = round_up(newsize, PAGE_SIZE);
557 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
558 shmem_truncate_range(inode, newsize, (loff_t)-1);
559 /* unmap again to remove racily COWed private pages */
560 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
564 setattr_copy(inode, attr);
565 #ifdef CONFIG_TMPFS_POSIX_ACL
566 if (attr->ia_valid & ATTR_MODE)
567 error = generic_acl_chmod(inode);
572 static void shmem_evict_inode(struct inode *inode)
574 struct shmem_inode_info *info = SHMEM_I(inode);
575 struct shmem_xattr *xattr, *nxattr;
577 if (inode->i_mapping->a_ops == &shmem_aops) {
578 shmem_unacct_size(info->flags, inode->i_size);
580 shmem_truncate_range(inode, 0, (loff_t)-1);
581 if (!list_empty(&info->swaplist)) {
582 mutex_lock(&shmem_swaplist_mutex);
583 list_del_init(&info->swaplist);
584 mutex_unlock(&shmem_swaplist_mutex);
588 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
592 BUG_ON(inode->i_blocks);
593 shmem_free_inode(inode->i_sb);
594 end_writeback(inode);
598 * If swap found in inode, free it and move page from swapcache to filecache.
600 static int shmem_unuse_inode(struct shmem_inode_info *info,
601 swp_entry_t swap, struct page *page)
603 struct address_space *mapping = info->vfs_inode.i_mapping;
608 radswap = swp_to_radix_entry(swap);
609 index = shmem_find_swap(mapping, radswap);
614 * Move _head_ to start search for next from here.
615 * But be careful: shmem_evict_inode checks list_empty without taking
616 * mutex, and there's an instant in list_move_tail when info->swaplist
617 * would appear empty, if it were the only one on shmem_swaplist.
619 if (shmem_swaplist.next != &info->swaplist)
620 list_move_tail(&shmem_swaplist, &info->swaplist);
623 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
624 * but also to hold up shmem_evict_inode(): so inode cannot be freed
625 * beneath us (pagelock doesn't help until the page is in pagecache).
627 error = shmem_add_to_page_cache(page, mapping, index,
628 GFP_NOWAIT, radswap);
629 /* which does mem_cgroup_uncharge_cache_page on error */
631 if (error != -ENOMEM) {
633 * Truncation and eviction use free_swap_and_cache(), which
634 * only does trylock page: if we raced, best clean up here.
636 delete_from_swap_cache(page);
637 set_page_dirty(page);
639 spin_lock(&info->lock);
641 spin_unlock(&info->lock);
644 error = 1; /* not an error, but entry was found */
650 * Search through swapped inodes to find and replace swap by page.
652 int shmem_unuse(swp_entry_t swap, struct page *page)
654 struct list_head *this, *next;
655 struct shmem_inode_info *info;
660 * Charge page using GFP_KERNEL while we can wait, before taking
661 * the shmem_swaplist_mutex which might hold up shmem_writepage().
662 * Charged back to the user (not to caller) when swap account is used.
663 * shmem_add_to_page_cache() will be called with GFP_NOWAIT.
665 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
668 /* No radix_tree_preload: swap entry keeps a place for page in tree */
670 mutex_lock(&shmem_swaplist_mutex);
671 list_for_each_safe(this, next, &shmem_swaplist) {
672 info = list_entry(this, struct shmem_inode_info, swaplist);
673 if (!info->swapped) {
674 spin_lock(&info->lock);
676 list_del_init(&info->swaplist);
677 spin_unlock(&info->lock);
680 found = shmem_unuse_inode(info, swap, page);
685 mutex_unlock(&shmem_swaplist_mutex);
688 mem_cgroup_uncharge_cache_page(page);
693 page_cache_release(page);
698 * Move the page from the page cache to the swap cache.
700 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
702 struct shmem_inode_info *info;
703 swp_entry_t swap, oswap;
704 struct address_space *mapping;
708 BUG_ON(!PageLocked(page));
709 mapping = page->mapping;
711 inode = mapping->host;
712 info = SHMEM_I(inode);
713 if (info->flags & VM_LOCKED)
715 if (!total_swap_pages)
719 * shmem_backing_dev_info's capabilities prevent regular writeback or
720 * sync from ever calling shmem_writepage; but a stacking filesystem
721 * might use ->writepage of its underlying filesystem, in which case
722 * tmpfs should write out to swap only in response to memory pressure,
723 * and not for the writeback threads or sync.
725 if (!wbc->for_reclaim) {
726 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
731 * Disable even the toy swapping implementation, while we convert
732 * functions one by one to having swap entries in the radix tree.
734 if (index < ULONG_MAX)
737 swap = get_swap_page();
742 * Add inode to shmem_unuse()'s list of swapped-out inodes,
743 * if it's not already there. Do it now because we cannot take
744 * mutex while holding spinlock, and must do so before the page
745 * is moved to swap cache, when its pagelock no longer protects
746 * the inode from eviction. But don't unlock the mutex until
747 * we've taken the spinlock, because shmem_unuse_inode() will
748 * prune a !swapped inode from the swaplist under both locks.
750 mutex_lock(&shmem_swaplist_mutex);
751 if (list_empty(&info->swaplist))
752 list_add_tail(&info->swaplist, &shmem_swaplist);
754 spin_lock(&info->lock);
755 mutex_unlock(&shmem_swaplist_mutex);
757 oswap = shmem_get_swap(info, index);
759 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
760 free_swap_and_cache(oswap);
761 shmem_put_swap(info, index, (swp_entry_t){0});
764 shmem_recalc_inode(inode);
766 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
767 delete_from_page_cache(page);
768 shmem_put_swap(info, index, swap);
770 swap_shmem_alloc(swap);
771 spin_unlock(&info->lock);
772 BUG_ON(page_mapped(page));
773 swap_writepage(page, wbc);
777 spin_unlock(&info->lock);
778 swapcache_free(swap, NULL);
780 set_page_dirty(page);
781 if (wbc->for_reclaim)
782 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
789 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
793 if (!mpol || mpol->mode == MPOL_DEFAULT)
794 return; /* show nothing */
796 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
798 seq_printf(seq, ",mpol=%s", buffer);
801 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
803 struct mempolicy *mpol = NULL;
805 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
808 spin_unlock(&sbinfo->stat_lock);
812 #endif /* CONFIG_TMPFS */
814 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
815 struct shmem_inode_info *info, pgoff_t index)
817 struct mempolicy mpol, *spol;
818 struct vm_area_struct pvma;
820 spol = mpol_cond_copy(&mpol,
821 mpol_shared_policy_lookup(&info->policy, index));
823 /* Create a pseudo vma that just contains the policy */
825 pvma.vm_pgoff = index;
827 pvma.vm_policy = spol;
828 return swapin_readahead(swap, gfp, &pvma, 0);
831 static struct page *shmem_alloc_page(gfp_t gfp,
832 struct shmem_inode_info *info, pgoff_t index)
834 struct vm_area_struct pvma;
836 /* Create a pseudo vma that just contains the policy */
838 pvma.vm_pgoff = index;
840 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
843 * alloc_page_vma() will drop the shared policy reference
845 return alloc_page_vma(gfp, &pvma, 0);
847 #else /* !CONFIG_NUMA */
849 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
852 #endif /* CONFIG_TMPFS */
854 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
855 struct shmem_inode_info *info, pgoff_t index)
857 return swapin_readahead(swap, gfp, NULL, 0);
860 static inline struct page *shmem_alloc_page(gfp_t gfp,
861 struct shmem_inode_info *info, pgoff_t index)
863 return alloc_page(gfp);
865 #endif /* CONFIG_NUMA */
867 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
868 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
875 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
877 * If we allocate a new one we do not mark it dirty. That's up to the
878 * vm. If we swap it in we mark it dirty since we also free the swap
879 * entry since a page cannot live in both the swap and page cache
881 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
882 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
884 struct address_space *mapping = inode->i_mapping;
885 struct shmem_inode_info *info;
886 struct shmem_sb_info *sbinfo;
892 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
896 page = find_lock_page(mapping, index);
897 if (radix_tree_exceptional_entry(page)) {
898 swap = radix_to_swp_entry(page);
902 if (sgp != SGP_WRITE &&
903 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
908 if (page || (sgp == SGP_READ && !swap.val)) {
910 * Once we can get the page lock, it must be uptodate:
911 * if there were an error in reading back from swap,
912 * the page would not be inserted into the filecache.
914 BUG_ON(page && !PageUptodate(page));
920 * Fast cache lookup did not find it:
921 * bring it back from swap or allocate.
923 info = SHMEM_I(inode);
924 sbinfo = SHMEM_SB(inode->i_sb);
927 /* Look it up and read it in.. */
928 page = lookup_swap_cache(swap);
930 /* here we actually do the io */
932 *fault_type |= VM_FAULT_MAJOR;
933 page = shmem_swapin(swap, gfp, info, index);
940 /* We have to do this with page locked to prevent races */
942 if (!PageUptodate(page)) {
946 wait_on_page_writeback(page);
948 /* Someone may have already done it for us */
950 if (page->mapping == mapping &&
951 page->index == index)
957 error = shmem_add_to_page_cache(page, mapping, index,
958 gfp, swp_to_radix_entry(swap));
962 spin_lock(&info->lock);
964 shmem_recalc_inode(inode);
965 spin_unlock(&info->lock);
967 delete_from_swap_cache(page);
968 set_page_dirty(page);
972 if (shmem_acct_block(info->flags)) {
976 if (sbinfo->max_blocks) {
977 if (percpu_counter_compare(&sbinfo->used_blocks,
978 sbinfo->max_blocks) >= 0) {
982 percpu_counter_inc(&sbinfo->used_blocks);
985 page = shmem_alloc_page(gfp, info, index);
991 SetPageSwapBacked(page);
992 __set_page_locked(page);
993 error = shmem_add_to_page_cache(page, mapping, index,
997 lru_cache_add_anon(page);
999 spin_lock(&info->lock);
1001 inode->i_blocks += BLOCKS_PER_PAGE;
1002 shmem_recalc_inode(inode);
1003 spin_unlock(&info->lock);
1005 clear_highpage(page);
1006 flush_dcache_page(page);
1007 SetPageUptodate(page);
1008 if (sgp == SGP_DIRTY)
1009 set_page_dirty(page);
1012 /* Perhaps the file has been truncated since we checked */
1013 if (sgp != SGP_WRITE &&
1014 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1025 ClearPageDirty(page);
1026 delete_from_page_cache(page);
1027 spin_lock(&info->lock);
1029 inode->i_blocks -= BLOCKS_PER_PAGE;
1030 spin_unlock(&info->lock);
1032 if (sbinfo->max_blocks)
1033 percpu_counter_add(&sbinfo->used_blocks, -1);
1035 shmem_unacct_blocks(info->flags, 1);
1037 if (swap.val && error != -EINVAL) {
1038 struct page *test = find_get_page(mapping, index);
1039 if (test && !radix_tree_exceptional_entry(test))
1040 page_cache_release(test);
1041 /* Have another try if the entry has changed */
1042 if (test != swp_to_radix_entry(swap))
1047 page_cache_release(page);
1049 if (error == -ENOSPC && !once++) {
1050 info = SHMEM_I(inode);
1051 spin_lock(&info->lock);
1052 shmem_recalc_inode(inode);
1053 spin_unlock(&info->lock);
1056 if (error == -EEXIST)
1061 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1063 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1065 int ret = VM_FAULT_LOCKED;
1067 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1069 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1071 if (ret & VM_FAULT_MAJOR) {
1072 count_vm_event(PGMAJFAULT);
1073 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1079 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1081 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1082 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1085 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1088 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1091 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1092 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1096 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1098 struct inode *inode = file->f_path.dentry->d_inode;
1099 struct shmem_inode_info *info = SHMEM_I(inode);
1100 int retval = -ENOMEM;
1102 spin_lock(&info->lock);
1103 if (lock && !(info->flags & VM_LOCKED)) {
1104 if (!user_shm_lock(inode->i_size, user))
1106 info->flags |= VM_LOCKED;
1107 mapping_set_unevictable(file->f_mapping);
1109 if (!lock && (info->flags & VM_LOCKED) && user) {
1110 user_shm_unlock(inode->i_size, user);
1111 info->flags &= ~VM_LOCKED;
1112 mapping_clear_unevictable(file->f_mapping);
1113 scan_mapping_unevictable_pages(file->f_mapping);
1118 spin_unlock(&info->lock);
1122 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1124 file_accessed(file);
1125 vma->vm_ops = &shmem_vm_ops;
1126 vma->vm_flags |= VM_CAN_NONLINEAR;
1130 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1131 int mode, dev_t dev, unsigned long flags)
1133 struct inode *inode;
1134 struct shmem_inode_info *info;
1135 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1137 if (shmem_reserve_inode(sb))
1140 inode = new_inode(sb);
1142 inode->i_ino = get_next_ino();
1143 inode_init_owner(inode, dir, mode);
1144 inode->i_blocks = 0;
1145 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1146 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1147 inode->i_generation = get_seconds();
1148 info = SHMEM_I(inode);
1149 memset(info, 0, (char *)inode - (char *)info);
1150 spin_lock_init(&info->lock);
1151 info->flags = flags & VM_NORESERVE;
1152 INIT_LIST_HEAD(&info->swaplist);
1153 INIT_LIST_HEAD(&info->xattr_list);
1154 cache_no_acl(inode);
1156 switch (mode & S_IFMT) {
1158 inode->i_op = &shmem_special_inode_operations;
1159 init_special_inode(inode, mode, dev);
1162 inode->i_mapping->a_ops = &shmem_aops;
1163 inode->i_op = &shmem_inode_operations;
1164 inode->i_fop = &shmem_file_operations;
1165 mpol_shared_policy_init(&info->policy,
1166 shmem_get_sbmpol(sbinfo));
1170 /* Some things misbehave if size == 0 on a directory */
1171 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1172 inode->i_op = &shmem_dir_inode_operations;
1173 inode->i_fop = &simple_dir_operations;
1177 * Must not load anything in the rbtree,
1178 * mpol_free_shared_policy will not be called.
1180 mpol_shared_policy_init(&info->policy, NULL);
1184 shmem_free_inode(sb);
1189 static const struct inode_operations shmem_symlink_inode_operations;
1190 static const struct inode_operations shmem_symlink_inline_operations;
1193 shmem_write_begin(struct file *file, struct address_space *mapping,
1194 loff_t pos, unsigned len, unsigned flags,
1195 struct page **pagep, void **fsdata)
1197 struct inode *inode = mapping->host;
1198 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1199 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1203 shmem_write_end(struct file *file, struct address_space *mapping,
1204 loff_t pos, unsigned len, unsigned copied,
1205 struct page *page, void *fsdata)
1207 struct inode *inode = mapping->host;
1209 if (pos + copied > inode->i_size)
1210 i_size_write(inode, pos + copied);
1212 set_page_dirty(page);
1214 page_cache_release(page);
1219 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1221 struct inode *inode = filp->f_path.dentry->d_inode;
1222 struct address_space *mapping = inode->i_mapping;
1224 unsigned long offset;
1225 enum sgp_type sgp = SGP_READ;
1228 * Might this read be for a stacking filesystem? Then when reading
1229 * holes of a sparse file, we actually need to allocate those pages,
1230 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1232 if (segment_eq(get_fs(), KERNEL_DS))
1235 index = *ppos >> PAGE_CACHE_SHIFT;
1236 offset = *ppos & ~PAGE_CACHE_MASK;
1239 struct page *page = NULL;
1241 unsigned long nr, ret;
1242 loff_t i_size = i_size_read(inode);
1244 end_index = i_size >> PAGE_CACHE_SHIFT;
1245 if (index > end_index)
1247 if (index == end_index) {
1248 nr = i_size & ~PAGE_CACHE_MASK;
1253 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1255 if (desc->error == -EINVAL)
1263 * We must evaluate after, since reads (unlike writes)
1264 * are called without i_mutex protection against truncate
1266 nr = PAGE_CACHE_SIZE;
1267 i_size = i_size_read(inode);
1268 end_index = i_size >> PAGE_CACHE_SHIFT;
1269 if (index == end_index) {
1270 nr = i_size & ~PAGE_CACHE_MASK;
1273 page_cache_release(page);
1281 * If users can be writing to this page using arbitrary
1282 * virtual addresses, take care about potential aliasing
1283 * before reading the page on the kernel side.
1285 if (mapping_writably_mapped(mapping))
1286 flush_dcache_page(page);
1288 * Mark the page accessed if we read the beginning.
1291 mark_page_accessed(page);
1293 page = ZERO_PAGE(0);
1294 page_cache_get(page);
1298 * Ok, we have the page, and it's up-to-date, so
1299 * now we can copy it to user space...
1301 * The actor routine returns how many bytes were actually used..
1302 * NOTE! This may not be the same as how much of a user buffer
1303 * we filled up (we may be padding etc), so we can only update
1304 * "pos" here (the actor routine has to update the user buffer
1305 * pointers and the remaining count).
1307 ret = actor(desc, page, offset, nr);
1309 index += offset >> PAGE_CACHE_SHIFT;
1310 offset &= ~PAGE_CACHE_MASK;
1312 page_cache_release(page);
1313 if (ret != nr || !desc->count)
1319 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1320 file_accessed(filp);
1323 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1324 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1326 struct file *filp = iocb->ki_filp;
1330 loff_t *ppos = &iocb->ki_pos;
1332 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1336 for (seg = 0; seg < nr_segs; seg++) {
1337 read_descriptor_t desc;
1340 desc.arg.buf = iov[seg].iov_base;
1341 desc.count = iov[seg].iov_len;
1342 if (desc.count == 0)
1345 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1346 retval += desc.written;
1348 retval = retval ?: desc.error;
1357 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1358 struct pipe_inode_info *pipe, size_t len,
1361 struct address_space *mapping = in->f_mapping;
1362 struct inode *inode = mapping->host;
1363 unsigned int loff, nr_pages, req_pages;
1364 struct page *pages[PIPE_DEF_BUFFERS];
1365 struct partial_page partial[PIPE_DEF_BUFFERS];
1367 pgoff_t index, end_index;
1370 struct splice_pipe_desc spd = {
1374 .ops = &page_cache_pipe_buf_ops,
1375 .spd_release = spd_release_page,
1378 isize = i_size_read(inode);
1379 if (unlikely(*ppos >= isize))
1382 left = isize - *ppos;
1383 if (unlikely(left < len))
1386 if (splice_grow_spd(pipe, &spd))
1389 index = *ppos >> PAGE_CACHE_SHIFT;
1390 loff = *ppos & ~PAGE_CACHE_MASK;
1391 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1392 nr_pages = min(req_pages, pipe->buffers);
1394 spd.nr_pages = find_get_pages_contig(mapping, index,
1395 nr_pages, spd.pages);
1396 index += spd.nr_pages;
1399 while (spd.nr_pages < nr_pages) {
1400 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1404 spd.pages[spd.nr_pages++] = page;
1408 index = *ppos >> PAGE_CACHE_SHIFT;
1409 nr_pages = spd.nr_pages;
1412 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1413 unsigned int this_len;
1418 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1419 page = spd.pages[page_nr];
1421 if (!PageUptodate(page) || page->mapping != mapping) {
1422 error = shmem_getpage(inode, index, &page,
1427 page_cache_release(spd.pages[page_nr]);
1428 spd.pages[page_nr] = page;
1431 isize = i_size_read(inode);
1432 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1433 if (unlikely(!isize || index > end_index))
1436 if (end_index == index) {
1439 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1443 this_len = min(this_len, plen - loff);
1447 spd.partial[page_nr].offset = loff;
1448 spd.partial[page_nr].len = this_len;
1455 while (page_nr < nr_pages)
1456 page_cache_release(spd.pages[page_nr++]);
1459 error = splice_to_pipe(pipe, &spd);
1461 splice_shrink_spd(pipe, &spd);
1470 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1472 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1474 buf->f_type = TMPFS_MAGIC;
1475 buf->f_bsize = PAGE_CACHE_SIZE;
1476 buf->f_namelen = NAME_MAX;
1477 if (sbinfo->max_blocks) {
1478 buf->f_blocks = sbinfo->max_blocks;
1480 buf->f_bfree = sbinfo->max_blocks -
1481 percpu_counter_sum(&sbinfo->used_blocks);
1483 if (sbinfo->max_inodes) {
1484 buf->f_files = sbinfo->max_inodes;
1485 buf->f_ffree = sbinfo->free_inodes;
1487 /* else leave those fields 0 like simple_statfs */
1492 * File creation. Allocate an inode, and we're done..
1495 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1497 struct inode *inode;
1498 int error = -ENOSPC;
1500 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1502 error = security_inode_init_security(inode, dir,
1503 &dentry->d_name, NULL,
1506 if (error != -EOPNOTSUPP) {
1511 #ifdef CONFIG_TMPFS_POSIX_ACL
1512 error = generic_acl_init(inode, dir);
1520 dir->i_size += BOGO_DIRENT_SIZE;
1521 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1522 d_instantiate(dentry, inode);
1523 dget(dentry); /* Extra count - pin the dentry in core */
1528 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1532 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1538 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1539 struct nameidata *nd)
1541 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1547 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1549 struct inode *inode = old_dentry->d_inode;
1553 * No ordinary (disk based) filesystem counts links as inodes;
1554 * but each new link needs a new dentry, pinning lowmem, and
1555 * tmpfs dentries cannot be pruned until they are unlinked.
1557 ret = shmem_reserve_inode(inode->i_sb);
1561 dir->i_size += BOGO_DIRENT_SIZE;
1562 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1564 ihold(inode); /* New dentry reference */
1565 dget(dentry); /* Extra pinning count for the created dentry */
1566 d_instantiate(dentry, inode);
1571 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1573 struct inode *inode = dentry->d_inode;
1575 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1576 shmem_free_inode(inode->i_sb);
1578 dir->i_size -= BOGO_DIRENT_SIZE;
1579 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1581 dput(dentry); /* Undo the count from "create" - this does all the work */
1585 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1587 if (!simple_empty(dentry))
1590 drop_nlink(dentry->d_inode);
1592 return shmem_unlink(dir, dentry);
1596 * The VFS layer already does all the dentry stuff for rename,
1597 * we just have to decrement the usage count for the target if
1598 * it exists so that the VFS layer correctly free's it when it
1601 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1603 struct inode *inode = old_dentry->d_inode;
1604 int they_are_dirs = S_ISDIR(inode->i_mode);
1606 if (!simple_empty(new_dentry))
1609 if (new_dentry->d_inode) {
1610 (void) shmem_unlink(new_dir, new_dentry);
1612 drop_nlink(old_dir);
1613 } else if (they_are_dirs) {
1614 drop_nlink(old_dir);
1618 old_dir->i_size -= BOGO_DIRENT_SIZE;
1619 new_dir->i_size += BOGO_DIRENT_SIZE;
1620 old_dir->i_ctime = old_dir->i_mtime =
1621 new_dir->i_ctime = new_dir->i_mtime =
1622 inode->i_ctime = CURRENT_TIME;
1626 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1630 struct inode *inode;
1633 struct shmem_inode_info *info;
1635 len = strlen(symname) + 1;
1636 if (len > PAGE_CACHE_SIZE)
1637 return -ENAMETOOLONG;
1639 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1643 error = security_inode_init_security(inode, dir, &dentry->d_name, NULL,
1646 if (error != -EOPNOTSUPP) {
1653 info = SHMEM_I(inode);
1654 inode->i_size = len-1;
1655 if (len <= SHMEM_SYMLINK_INLINE_LEN) {
1657 memcpy(info->inline_symlink, symname, len);
1658 inode->i_op = &shmem_symlink_inline_operations;
1660 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1665 inode->i_mapping->a_ops = &shmem_aops;
1666 inode->i_op = &shmem_symlink_inode_operations;
1667 kaddr = kmap_atomic(page, KM_USER0);
1668 memcpy(kaddr, symname, len);
1669 kunmap_atomic(kaddr, KM_USER0);
1670 set_page_dirty(page);
1672 page_cache_release(page);
1674 dir->i_size += BOGO_DIRENT_SIZE;
1675 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1676 d_instantiate(dentry, inode);
1681 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1683 nd_set_link(nd, SHMEM_I(dentry->d_inode)->inline_symlink);
1687 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1689 struct page *page = NULL;
1690 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1691 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
1697 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1699 if (!IS_ERR(nd_get_link(nd))) {
1700 struct page *page = cookie;
1702 mark_page_accessed(page);
1703 page_cache_release(page);
1707 #ifdef CONFIG_TMPFS_XATTR
1709 * Superblocks without xattr inode operations may get some security.* xattr
1710 * support from the LSM "for free". As soon as we have any other xattrs
1711 * like ACLs, we also need to implement the security.* handlers at
1712 * filesystem level, though.
1715 static int shmem_xattr_get(struct dentry *dentry, const char *name,
1716 void *buffer, size_t size)
1718 struct shmem_inode_info *info;
1719 struct shmem_xattr *xattr;
1722 info = SHMEM_I(dentry->d_inode);
1724 spin_lock(&info->lock);
1725 list_for_each_entry(xattr, &info->xattr_list, list) {
1726 if (strcmp(name, xattr->name))
1731 if (size < xattr->size)
1734 memcpy(buffer, xattr->value, xattr->size);
1738 spin_unlock(&info->lock);
1742 static int shmem_xattr_set(struct dentry *dentry, const char *name,
1743 const void *value, size_t size, int flags)
1745 struct inode *inode = dentry->d_inode;
1746 struct shmem_inode_info *info = SHMEM_I(inode);
1747 struct shmem_xattr *xattr;
1748 struct shmem_xattr *new_xattr = NULL;
1752 /* value == NULL means remove */
1755 len = sizeof(*new_xattr) + size;
1756 if (len <= sizeof(*new_xattr))
1759 new_xattr = kmalloc(len, GFP_KERNEL);
1763 new_xattr->name = kstrdup(name, GFP_KERNEL);
1764 if (!new_xattr->name) {
1769 new_xattr->size = size;
1770 memcpy(new_xattr->value, value, size);
1773 spin_lock(&info->lock);
1774 list_for_each_entry(xattr, &info->xattr_list, list) {
1775 if (!strcmp(name, xattr->name)) {
1776 if (flags & XATTR_CREATE) {
1779 } else if (new_xattr) {
1780 list_replace(&xattr->list, &new_xattr->list);
1782 list_del(&xattr->list);
1787 if (flags & XATTR_REPLACE) {
1791 list_add(&new_xattr->list, &info->xattr_list);
1795 spin_unlock(&info->lock);
1802 static const struct xattr_handler *shmem_xattr_handlers[] = {
1803 #ifdef CONFIG_TMPFS_POSIX_ACL
1804 &generic_acl_access_handler,
1805 &generic_acl_default_handler,
1810 static int shmem_xattr_validate(const char *name)
1812 struct { const char *prefix; size_t len; } arr[] = {
1813 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1814 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1818 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1819 size_t preflen = arr[i].len;
1820 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1829 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1830 void *buffer, size_t size)
1835 * If this is a request for a synthetic attribute in the system.*
1836 * namespace use the generic infrastructure to resolve a handler
1837 * for it via sb->s_xattr.
1839 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1840 return generic_getxattr(dentry, name, buffer, size);
1842 err = shmem_xattr_validate(name);
1846 return shmem_xattr_get(dentry, name, buffer, size);
1849 static int shmem_setxattr(struct dentry *dentry, const char *name,
1850 const void *value, size_t size, int flags)
1855 * If this is a request for a synthetic attribute in the system.*
1856 * namespace use the generic infrastructure to resolve a handler
1857 * for it via sb->s_xattr.
1859 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1860 return generic_setxattr(dentry, name, value, size, flags);
1862 err = shmem_xattr_validate(name);
1867 value = ""; /* empty EA, do not remove */
1869 return shmem_xattr_set(dentry, name, value, size, flags);
1873 static int shmem_removexattr(struct dentry *dentry, const char *name)
1878 * If this is a request for a synthetic attribute in the system.*
1879 * namespace use the generic infrastructure to resolve a handler
1880 * for it via sb->s_xattr.
1882 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1883 return generic_removexattr(dentry, name);
1885 err = shmem_xattr_validate(name);
1889 return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
1892 static bool xattr_is_trusted(const char *name)
1894 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1897 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1899 bool trusted = capable(CAP_SYS_ADMIN);
1900 struct shmem_xattr *xattr;
1901 struct shmem_inode_info *info;
1904 info = SHMEM_I(dentry->d_inode);
1906 spin_lock(&info->lock);
1907 list_for_each_entry(xattr, &info->xattr_list, list) {
1910 /* skip "trusted." attributes for unprivileged callers */
1911 if (!trusted && xattr_is_trusted(xattr->name))
1914 len = strlen(xattr->name) + 1;
1921 memcpy(buffer, xattr->name, len);
1925 spin_unlock(&info->lock);
1929 #endif /* CONFIG_TMPFS_XATTR */
1931 static const struct inode_operations shmem_symlink_inline_operations = {
1932 .readlink = generic_readlink,
1933 .follow_link = shmem_follow_link_inline,
1934 #ifdef CONFIG_TMPFS_XATTR
1935 .setxattr = shmem_setxattr,
1936 .getxattr = shmem_getxattr,
1937 .listxattr = shmem_listxattr,
1938 .removexattr = shmem_removexattr,
1942 static const struct inode_operations shmem_symlink_inode_operations = {
1943 .readlink = generic_readlink,
1944 .follow_link = shmem_follow_link,
1945 .put_link = shmem_put_link,
1946 #ifdef CONFIG_TMPFS_XATTR
1947 .setxattr = shmem_setxattr,
1948 .getxattr = shmem_getxattr,
1949 .listxattr = shmem_listxattr,
1950 .removexattr = shmem_removexattr,
1954 static struct dentry *shmem_get_parent(struct dentry *child)
1956 return ERR_PTR(-ESTALE);
1959 static int shmem_match(struct inode *ino, void *vfh)
1963 inum = (inum << 32) | fh[1];
1964 return ino->i_ino == inum && fh[0] == ino->i_generation;
1967 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1968 struct fid *fid, int fh_len, int fh_type)
1970 struct inode *inode;
1971 struct dentry *dentry = NULL;
1972 u64 inum = fid->raw[2];
1973 inum = (inum << 32) | fid->raw[1];
1978 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1979 shmem_match, fid->raw);
1981 dentry = d_find_alias(inode);
1988 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1991 struct inode *inode = dentry->d_inode;
1998 if (inode_unhashed(inode)) {
1999 /* Unfortunately insert_inode_hash is not idempotent,
2000 * so as we hash inodes here rather than at creation
2001 * time, we need a lock to ensure we only try
2004 static DEFINE_SPINLOCK(lock);
2006 if (inode_unhashed(inode))
2007 __insert_inode_hash(inode,
2008 inode->i_ino + inode->i_generation);
2012 fh[0] = inode->i_generation;
2013 fh[1] = inode->i_ino;
2014 fh[2] = ((__u64)inode->i_ino) >> 32;
2020 static const struct export_operations shmem_export_ops = {
2021 .get_parent = shmem_get_parent,
2022 .encode_fh = shmem_encode_fh,
2023 .fh_to_dentry = shmem_fh_to_dentry,
2026 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2029 char *this_char, *value, *rest;
2031 while (options != NULL) {
2032 this_char = options;
2035 * NUL-terminate this option: unfortunately,
2036 * mount options form a comma-separated list,
2037 * but mpol's nodelist may also contain commas.
2039 options = strchr(options, ',');
2040 if (options == NULL)
2043 if (!isdigit(*options)) {
2050 if ((value = strchr(this_char,'=')) != NULL) {
2054 "tmpfs: No value for mount option '%s'\n",
2059 if (!strcmp(this_char,"size")) {
2060 unsigned long long size;
2061 size = memparse(value,&rest);
2063 size <<= PAGE_SHIFT;
2064 size *= totalram_pages;
2070 sbinfo->max_blocks =
2071 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2072 } else if (!strcmp(this_char,"nr_blocks")) {
2073 sbinfo->max_blocks = memparse(value, &rest);
2076 } else if (!strcmp(this_char,"nr_inodes")) {
2077 sbinfo->max_inodes = memparse(value, &rest);
2080 } else if (!strcmp(this_char,"mode")) {
2083 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2086 } else if (!strcmp(this_char,"uid")) {
2089 sbinfo->uid = simple_strtoul(value, &rest, 0);
2092 } else if (!strcmp(this_char,"gid")) {
2095 sbinfo->gid = simple_strtoul(value, &rest, 0);
2098 } else if (!strcmp(this_char,"mpol")) {
2099 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2102 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2110 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2116 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2118 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2119 struct shmem_sb_info config = *sbinfo;
2120 unsigned long inodes;
2121 int error = -EINVAL;
2123 if (shmem_parse_options(data, &config, true))
2126 spin_lock(&sbinfo->stat_lock);
2127 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2128 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2130 if (config.max_inodes < inodes)
2133 * Those tests disallow limited->unlimited while any are in use;
2134 * but we must separately disallow unlimited->limited, because
2135 * in that case we have no record of how much is already in use.
2137 if (config.max_blocks && !sbinfo->max_blocks)
2139 if (config.max_inodes && !sbinfo->max_inodes)
2143 sbinfo->max_blocks = config.max_blocks;
2144 sbinfo->max_inodes = config.max_inodes;
2145 sbinfo->free_inodes = config.max_inodes - inodes;
2147 mpol_put(sbinfo->mpol);
2148 sbinfo->mpol = config.mpol; /* transfers initial ref */
2150 spin_unlock(&sbinfo->stat_lock);
2154 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2156 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2158 if (sbinfo->max_blocks != shmem_default_max_blocks())
2159 seq_printf(seq, ",size=%luk",
2160 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2161 if (sbinfo->max_inodes != shmem_default_max_inodes())
2162 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2163 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2164 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2165 if (sbinfo->uid != 0)
2166 seq_printf(seq, ",uid=%u", sbinfo->uid);
2167 if (sbinfo->gid != 0)
2168 seq_printf(seq, ",gid=%u", sbinfo->gid);
2169 shmem_show_mpol(seq, sbinfo->mpol);
2172 #endif /* CONFIG_TMPFS */
2174 static void shmem_put_super(struct super_block *sb)
2176 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2178 percpu_counter_destroy(&sbinfo->used_blocks);
2180 sb->s_fs_info = NULL;
2183 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2185 struct inode *inode;
2186 struct dentry *root;
2187 struct shmem_sb_info *sbinfo;
2190 /* Round up to L1_CACHE_BYTES to resist false sharing */
2191 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2192 L1_CACHE_BYTES), GFP_KERNEL);
2196 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2197 sbinfo->uid = current_fsuid();
2198 sbinfo->gid = current_fsgid();
2199 sb->s_fs_info = sbinfo;
2203 * Per default we only allow half of the physical ram per
2204 * tmpfs instance, limiting inodes to one per page of lowmem;
2205 * but the internal instance is left unlimited.
2207 if (!(sb->s_flags & MS_NOUSER)) {
2208 sbinfo->max_blocks = shmem_default_max_blocks();
2209 sbinfo->max_inodes = shmem_default_max_inodes();
2210 if (shmem_parse_options(data, sbinfo, false)) {
2215 sb->s_export_op = &shmem_export_ops;
2217 sb->s_flags |= MS_NOUSER;
2220 spin_lock_init(&sbinfo->stat_lock);
2221 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2223 sbinfo->free_inodes = sbinfo->max_inodes;
2225 sb->s_maxbytes = MAX_LFS_FILESIZE;
2226 sb->s_blocksize = PAGE_CACHE_SIZE;
2227 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2228 sb->s_magic = TMPFS_MAGIC;
2229 sb->s_op = &shmem_ops;
2230 sb->s_time_gran = 1;
2231 #ifdef CONFIG_TMPFS_XATTR
2232 sb->s_xattr = shmem_xattr_handlers;
2234 #ifdef CONFIG_TMPFS_POSIX_ACL
2235 sb->s_flags |= MS_POSIXACL;
2238 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2241 inode->i_uid = sbinfo->uid;
2242 inode->i_gid = sbinfo->gid;
2243 root = d_alloc_root(inode);
2252 shmem_put_super(sb);
2256 static struct kmem_cache *shmem_inode_cachep;
2258 static struct inode *shmem_alloc_inode(struct super_block *sb)
2260 struct shmem_inode_info *info;
2261 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2264 return &info->vfs_inode;
2267 static void shmem_destroy_callback(struct rcu_head *head)
2269 struct inode *inode = container_of(head, struct inode, i_rcu);
2270 INIT_LIST_HEAD(&inode->i_dentry);
2271 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2274 static void shmem_destroy_inode(struct inode *inode)
2276 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2277 /* only struct inode is valid if it's an inline symlink */
2278 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2280 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2283 static void shmem_init_inode(void *foo)
2285 struct shmem_inode_info *info = foo;
2286 inode_init_once(&info->vfs_inode);
2289 static int shmem_init_inodecache(void)
2291 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2292 sizeof(struct shmem_inode_info),
2293 0, SLAB_PANIC, shmem_init_inode);
2297 static void shmem_destroy_inodecache(void)
2299 kmem_cache_destroy(shmem_inode_cachep);
2302 static const struct address_space_operations shmem_aops = {
2303 .writepage = shmem_writepage,
2304 .set_page_dirty = __set_page_dirty_no_writeback,
2306 .write_begin = shmem_write_begin,
2307 .write_end = shmem_write_end,
2309 .migratepage = migrate_page,
2310 .error_remove_page = generic_error_remove_page,
2313 static const struct file_operations shmem_file_operations = {
2316 .llseek = generic_file_llseek,
2317 .read = do_sync_read,
2318 .write = do_sync_write,
2319 .aio_read = shmem_file_aio_read,
2320 .aio_write = generic_file_aio_write,
2321 .fsync = noop_fsync,
2322 .splice_read = shmem_file_splice_read,
2323 .splice_write = generic_file_splice_write,
2327 static const struct inode_operations shmem_inode_operations = {
2328 .setattr = shmem_setattr,
2329 .truncate_range = shmem_truncate_range,
2330 #ifdef CONFIG_TMPFS_XATTR
2331 .setxattr = shmem_setxattr,
2332 .getxattr = shmem_getxattr,
2333 .listxattr = shmem_listxattr,
2334 .removexattr = shmem_removexattr,
2338 static const struct inode_operations shmem_dir_inode_operations = {
2340 .create = shmem_create,
2341 .lookup = simple_lookup,
2343 .unlink = shmem_unlink,
2344 .symlink = shmem_symlink,
2345 .mkdir = shmem_mkdir,
2346 .rmdir = shmem_rmdir,
2347 .mknod = shmem_mknod,
2348 .rename = shmem_rename,
2350 #ifdef CONFIG_TMPFS_XATTR
2351 .setxattr = shmem_setxattr,
2352 .getxattr = shmem_getxattr,
2353 .listxattr = shmem_listxattr,
2354 .removexattr = shmem_removexattr,
2356 #ifdef CONFIG_TMPFS_POSIX_ACL
2357 .setattr = shmem_setattr,
2361 static const struct inode_operations shmem_special_inode_operations = {
2362 #ifdef CONFIG_TMPFS_XATTR
2363 .setxattr = shmem_setxattr,
2364 .getxattr = shmem_getxattr,
2365 .listxattr = shmem_listxattr,
2366 .removexattr = shmem_removexattr,
2368 #ifdef CONFIG_TMPFS_POSIX_ACL
2369 .setattr = shmem_setattr,
2373 static const struct super_operations shmem_ops = {
2374 .alloc_inode = shmem_alloc_inode,
2375 .destroy_inode = shmem_destroy_inode,
2377 .statfs = shmem_statfs,
2378 .remount_fs = shmem_remount_fs,
2379 .show_options = shmem_show_options,
2381 .evict_inode = shmem_evict_inode,
2382 .drop_inode = generic_delete_inode,
2383 .put_super = shmem_put_super,
2386 static const struct vm_operations_struct shmem_vm_ops = {
2387 .fault = shmem_fault,
2389 .set_policy = shmem_set_policy,
2390 .get_policy = shmem_get_policy,
2394 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2395 int flags, const char *dev_name, void *data)
2397 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2400 static struct file_system_type shmem_fs_type = {
2401 .owner = THIS_MODULE,
2403 .mount = shmem_mount,
2404 .kill_sb = kill_litter_super,
2407 int __init shmem_init(void)
2411 error = bdi_init(&shmem_backing_dev_info);
2415 error = shmem_init_inodecache();
2419 error = register_filesystem(&shmem_fs_type);
2421 printk(KERN_ERR "Could not register tmpfs\n");
2425 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2426 shmem_fs_type.name, NULL);
2427 if (IS_ERR(shm_mnt)) {
2428 error = PTR_ERR(shm_mnt);
2429 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2435 unregister_filesystem(&shmem_fs_type);
2437 shmem_destroy_inodecache();
2439 bdi_destroy(&shmem_backing_dev_info);
2441 shm_mnt = ERR_PTR(error);
2445 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2447 * mem_cgroup_get_shmem_target - find page or swap assigned to the shmem file
2448 * @inode: the inode to be searched
2449 * @index: the page offset to be searched
2450 * @pagep: the pointer for the found page to be stored
2451 * @swapp: the pointer for the found swap entry to be stored
2453 * If a page is found, refcount of it is incremented. Callers should handle
2456 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t index,
2457 struct page **pagep, swp_entry_t *swapp)
2459 struct shmem_inode_info *info = SHMEM_I(inode);
2460 struct page *page = NULL;
2461 swp_entry_t swap = {0};
2463 if ((index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2466 spin_lock(&info->lock);
2468 swap = shmem_get_swap(info, index);
2470 page = find_get_page(&swapper_space, swap.val);
2473 page = find_get_page(inode->i_mapping, index);
2474 spin_unlock(&info->lock);
2481 #else /* !CONFIG_SHMEM */
2484 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2486 * This is intended for small system where the benefits of the full
2487 * shmem code (swap-backed and resource-limited) are outweighed by
2488 * their complexity. On systems without swap this code should be
2489 * effectively equivalent, but much lighter weight.
2492 #include <linux/ramfs.h>
2494 static struct file_system_type shmem_fs_type = {
2496 .mount = ramfs_mount,
2497 .kill_sb = kill_litter_super,
2500 int __init shmem_init(void)
2502 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2504 shm_mnt = kern_mount(&shmem_fs_type);
2505 BUG_ON(IS_ERR(shm_mnt));
2510 int shmem_unuse(swp_entry_t swap, struct page *page)
2515 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2520 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2522 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2524 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2526 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2528 * mem_cgroup_get_shmem_target - find page or swap assigned to the shmem file
2529 * @inode: the inode to be searched
2530 * @index: the page offset to be searched
2531 * @pagep: the pointer for the found page to be stored
2532 * @swapp: the pointer for the found swap entry to be stored
2534 * If a page is found, refcount of it is incremented. Callers should handle
2537 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t index,
2538 struct page **pagep, swp_entry_t *swapp)
2540 struct page *page = NULL;
2542 if ((index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2544 page = find_get_page(inode->i_mapping, index);
2547 *swapp = (swp_entry_t){0};
2551 #define shmem_vm_ops generic_file_vm_ops
2552 #define shmem_file_operations ramfs_file_operations
2553 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2554 #define shmem_acct_size(flags, size) 0
2555 #define shmem_unacct_size(flags, size) do {} while (0)
2557 #endif /* CONFIG_SHMEM */
2562 * shmem_file_setup - get an unlinked file living in tmpfs
2563 * @name: name for dentry (to be seen in /proc/<pid>/maps
2564 * @size: size to be set for the file
2565 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2567 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2571 struct inode *inode;
2573 struct dentry *root;
2576 if (IS_ERR(shm_mnt))
2577 return (void *)shm_mnt;
2579 if (size < 0 || size > MAX_LFS_FILESIZE)
2580 return ERR_PTR(-EINVAL);
2582 if (shmem_acct_size(flags, size))
2583 return ERR_PTR(-ENOMEM);
2587 this.len = strlen(name);
2588 this.hash = 0; /* will go */
2589 root = shm_mnt->mnt_root;
2590 path.dentry = d_alloc(root, &this);
2593 path.mnt = mntget(shm_mnt);
2596 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2600 d_instantiate(path.dentry, inode);
2601 inode->i_size = size;
2602 inode->i_nlink = 0; /* It is unlinked */
2604 error = ramfs_nommu_expand_for_mapping(inode, size);
2610 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2611 &shmem_file_operations);
2620 shmem_unacct_size(flags, size);
2621 return ERR_PTR(error);
2623 EXPORT_SYMBOL_GPL(shmem_file_setup);
2626 * shmem_zero_setup - setup a shared anonymous mapping
2627 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2629 int shmem_zero_setup(struct vm_area_struct *vma)
2632 loff_t size = vma->vm_end - vma->vm_start;
2634 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2636 return PTR_ERR(file);
2640 vma->vm_file = file;
2641 vma->vm_ops = &shmem_vm_ops;
2642 vma->vm_flags |= VM_CAN_NONLINEAR;
2647 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2648 * @mapping: the page's address_space
2649 * @index: the page index
2650 * @gfp: the page allocator flags to use if allocating
2652 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2653 * with any new page allocations done using the specified allocation flags.
2654 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2655 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2656 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2658 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2659 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2661 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2662 pgoff_t index, gfp_t gfp)
2665 struct inode *inode = mapping->host;
2669 BUG_ON(mapping->a_ops != &shmem_aops);
2670 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2672 page = ERR_PTR(error);
2678 * The tiny !SHMEM case uses ramfs without swap
2680 return read_cache_page_gfp(mapping, index, gfp);
2683 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);