2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/pagemap.h>
29 #include <linux/file.h>
31 #include <linux/export.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/pagevec.h>
55 #include <linux/percpu_counter.h>
56 #include <linux/falloc.h>
57 #include <linux/splice.h>
58 #include <linux/security.h>
59 #include <linux/swapops.h>
60 #include <linux/mempolicy.h>
61 #include <linux/namei.h>
62 #include <linux/ctype.h>
63 #include <linux/migrate.h>
64 #include <linux/highmem.h>
65 #include <linux/seq_file.h>
66 #include <linux/magic.h>
68 #include <asm/uaccess.h>
69 #include <asm/pgtable.h>
71 #define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
72 #define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
74 /* Pretend that each entry is of this size in directory's i_size */
75 #define BOGO_DIRENT_SIZE 20
77 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
78 #define SHORT_SYMLINK_LEN 128
81 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
82 char *name; /* xattr name */
88 * shmem_fallocate and shmem_writepage communicate via inode->i_private
89 * (with i_mutex making sure that it has only one user at a time):
90 * we would prefer not to enlarge the shmem inode just for that.
93 pgoff_t start; /* start of range currently being fallocated */
94 pgoff_t next; /* the next page offset to be fallocated */
95 pgoff_t nr_falloced; /* how many new pages have been fallocated */
96 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
99 /* Flag allocation requirements to shmem_getpage */
101 SGP_READ, /* don't exceed i_size, don't allocate page */
102 SGP_CACHE, /* don't exceed i_size, may allocate page */
103 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
104 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
105 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
109 static unsigned long shmem_default_max_blocks(void)
111 return totalram_pages / 2;
114 static unsigned long shmem_default_max_inodes(void)
116 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
120 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122 struct shmem_inode_info *info, pgoff_t index);
123 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
124 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
126 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
127 struct page **pagep, enum sgp_type sgp, int *fault_type)
129 return shmem_getpage_gfp(inode, index, pagep, sgp,
130 mapping_gfp_mask(inode->i_mapping), fault_type);
133 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
135 return sb->s_fs_info;
139 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
140 * for shared memory and for shared anonymous (/dev/zero) mappings
141 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
142 * consistent with the pre-accounting of private mappings ...
144 static inline int shmem_acct_size(unsigned long flags, loff_t size)
146 return (flags & VM_NORESERVE) ?
147 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
150 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
152 if (!(flags & VM_NORESERVE))
153 vm_unacct_memory(VM_ACCT(size));
157 * ... whereas tmpfs objects are accounted incrementally as
158 * pages are allocated, in order to allow huge sparse files.
159 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
160 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
162 static inline int shmem_acct_block(unsigned long flags)
164 return (flags & VM_NORESERVE) ?
165 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
168 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
170 if (flags & VM_NORESERVE)
171 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
174 static const struct super_operations shmem_ops;
175 static const struct address_space_operations shmem_aops;
176 static const struct file_operations shmem_file_operations;
177 static const struct inode_operations shmem_inode_operations;
178 static const struct inode_operations shmem_dir_inode_operations;
179 static const struct inode_operations shmem_special_inode_operations;
180 static const struct vm_operations_struct shmem_vm_ops;
182 static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
183 .ra_pages = 0, /* No readahead */
184 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
187 static LIST_HEAD(shmem_swaplist);
188 static DEFINE_MUTEX(shmem_swaplist_mutex);
190 static int shmem_reserve_inode(struct super_block *sb)
192 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
193 if (sbinfo->max_inodes) {
194 spin_lock(&sbinfo->stat_lock);
195 if (!sbinfo->free_inodes) {
196 spin_unlock(&sbinfo->stat_lock);
199 sbinfo->free_inodes--;
200 spin_unlock(&sbinfo->stat_lock);
205 static void shmem_free_inode(struct super_block *sb)
207 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
208 if (sbinfo->max_inodes) {
209 spin_lock(&sbinfo->stat_lock);
210 sbinfo->free_inodes++;
211 spin_unlock(&sbinfo->stat_lock);
216 * shmem_recalc_inode - recalculate the block usage of an inode
217 * @inode: inode to recalc
219 * We have to calculate the free blocks since the mm can drop
220 * undirtied hole pages behind our back.
222 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
223 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
225 * It has to be called with the spinlock held.
227 static void shmem_recalc_inode(struct inode *inode)
229 struct shmem_inode_info *info = SHMEM_I(inode);
232 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
235 if (sbinfo->max_blocks)
236 percpu_counter_add(&sbinfo->used_blocks, -freed);
237 info->alloced -= freed;
238 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
239 shmem_unacct_blocks(info->flags, freed);
244 * Replace item expected in radix tree by a new item, while holding tree lock.
246 static int shmem_radix_tree_replace(struct address_space *mapping,
247 pgoff_t index, void *expected, void *replacement)
252 VM_BUG_ON(!expected);
253 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
255 item = radix_tree_deref_slot_protected(pslot,
256 &mapping->tree_lock);
257 if (item != expected)
260 radix_tree_replace_slot(pslot, replacement);
262 radix_tree_delete(&mapping->page_tree, index);
267 * Like add_to_page_cache_locked, but error if expected item has gone.
269 static int shmem_add_to_page_cache(struct page *page,
270 struct address_space *mapping,
271 pgoff_t index, gfp_t gfp, void *expected)
275 VM_BUG_ON(!PageLocked(page));
276 VM_BUG_ON(!PageSwapBacked(page));
279 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
281 page_cache_get(page);
282 page->mapping = mapping;
285 spin_lock_irq(&mapping->tree_lock);
287 error = radix_tree_insert(&mapping->page_tree,
290 error = shmem_radix_tree_replace(mapping, index,
294 __inc_zone_page_state(page, NR_FILE_PAGES);
295 __inc_zone_page_state(page, NR_SHMEM);
296 spin_unlock_irq(&mapping->tree_lock);
298 page->mapping = NULL;
299 spin_unlock_irq(&mapping->tree_lock);
300 page_cache_release(page);
303 radix_tree_preload_end();
306 mem_cgroup_uncharge_cache_page(page);
311 * Like delete_from_page_cache, but substitutes swap for page.
313 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
315 struct address_space *mapping = page->mapping;
318 spin_lock_irq(&mapping->tree_lock);
319 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
320 page->mapping = NULL;
322 __dec_zone_page_state(page, NR_FILE_PAGES);
323 __dec_zone_page_state(page, NR_SHMEM);
324 spin_unlock_irq(&mapping->tree_lock);
325 page_cache_release(page);
330 * Like find_get_pages, but collecting swap entries as well as pages.
332 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
333 pgoff_t start, unsigned int nr_pages,
334 struct page **pages, pgoff_t *indices)
338 unsigned int nr_found;
342 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
343 (void ***)pages, indices, start, nr_pages);
345 for (i = 0; i < nr_found; i++) {
348 page = radix_tree_deref_slot((void **)pages[i]);
351 if (radix_tree_exception(page)) {
352 if (radix_tree_deref_retry(page))
355 * Otherwise, we must be storing a swap entry
356 * here as an exceptional entry: so return it
357 * without attempting to raise page count.
361 if (!page_cache_get_speculative(page))
364 /* Has the page moved? */
365 if (unlikely(page != *((void **)pages[i]))) {
366 page_cache_release(page);
370 indices[ret] = indices[i];
374 if (unlikely(!ret && nr_found))
381 * Remove swap entry from radix tree, free the swap and its page cache.
383 static int shmem_free_swap(struct address_space *mapping,
384 pgoff_t index, void *radswap)
388 spin_lock_irq(&mapping->tree_lock);
389 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
390 spin_unlock_irq(&mapping->tree_lock);
392 free_swap_and_cache(radix_to_swp_entry(radswap));
397 * Pagevec may contain swap entries, so shuffle up pages before releasing.
399 static void shmem_deswap_pagevec(struct pagevec *pvec)
403 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
404 struct page *page = pvec->pages[i];
405 if (!radix_tree_exceptional_entry(page))
406 pvec->pages[j++] = page;
412 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
414 void shmem_unlock_mapping(struct address_space *mapping)
417 pgoff_t indices[PAGEVEC_SIZE];
420 pagevec_init(&pvec, 0);
422 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
424 while (!mapping_unevictable(mapping)) {
426 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
427 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
429 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
430 PAGEVEC_SIZE, pvec.pages, indices);
433 index = indices[pvec.nr - 1] + 1;
434 shmem_deswap_pagevec(&pvec);
435 check_move_unevictable_pages(pvec.pages, pvec.nr);
436 pagevec_release(&pvec);
442 * Remove range of pages and swap entries from radix tree, and free them.
443 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
445 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
448 struct address_space *mapping = inode->i_mapping;
449 struct shmem_inode_info *info = SHMEM_I(inode);
450 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
451 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
452 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
453 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
455 pgoff_t indices[PAGEVEC_SIZE];
456 long nr_swaps_freed = 0;
461 end = -1; /* unsigned, so actually very big */
463 pagevec_init(&pvec, 0);
465 while (index < end) {
466 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
467 min(end - index, (pgoff_t)PAGEVEC_SIZE),
468 pvec.pages, indices);
471 mem_cgroup_uncharge_start();
472 for (i = 0; i < pagevec_count(&pvec); i++) {
473 struct page *page = pvec.pages[i];
479 if (radix_tree_exceptional_entry(page)) {
482 nr_swaps_freed += !shmem_free_swap(mapping,
487 if (!trylock_page(page))
489 if (!unfalloc || !PageUptodate(page)) {
490 if (page->mapping == mapping) {
491 VM_BUG_ON(PageWriteback(page));
492 truncate_inode_page(mapping, page);
497 shmem_deswap_pagevec(&pvec);
498 pagevec_release(&pvec);
499 mem_cgroup_uncharge_end();
505 struct page *page = NULL;
506 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
508 unsigned int top = PAGE_CACHE_SIZE;
513 zero_user_segment(page, partial_start, top);
514 set_page_dirty(page);
516 page_cache_release(page);
520 struct page *page = NULL;
521 shmem_getpage(inode, end, &page, SGP_READ, NULL);
523 zero_user_segment(page, 0, partial_end);
524 set_page_dirty(page);
526 page_cache_release(page);
535 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
536 min(end - index, (pgoff_t)PAGEVEC_SIZE),
537 pvec.pages, indices);
539 if (index == start || unfalloc)
544 if ((index == start || unfalloc) && indices[0] >= end) {
545 shmem_deswap_pagevec(&pvec);
546 pagevec_release(&pvec);
549 mem_cgroup_uncharge_start();
550 for (i = 0; i < pagevec_count(&pvec); i++) {
551 struct page *page = pvec.pages[i];
557 if (radix_tree_exceptional_entry(page)) {
560 nr_swaps_freed += !shmem_free_swap(mapping,
566 if (!unfalloc || !PageUptodate(page)) {
567 if (page->mapping == mapping) {
568 VM_BUG_ON(PageWriteback(page));
569 truncate_inode_page(mapping, page);
574 shmem_deswap_pagevec(&pvec);
575 pagevec_release(&pvec);
576 mem_cgroup_uncharge_end();
580 spin_lock(&info->lock);
581 info->swapped -= nr_swaps_freed;
582 shmem_recalc_inode(inode);
583 spin_unlock(&info->lock);
586 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
588 shmem_undo_range(inode, lstart, lend, false);
589 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
591 EXPORT_SYMBOL_GPL(shmem_truncate_range);
593 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
595 struct inode *inode = dentry->d_inode;
598 error = inode_change_ok(inode, attr);
602 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
603 loff_t oldsize = inode->i_size;
604 loff_t newsize = attr->ia_size;
606 if (newsize != oldsize) {
607 i_size_write(inode, newsize);
608 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
610 if (newsize < oldsize) {
611 loff_t holebegin = round_up(newsize, PAGE_SIZE);
612 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
613 shmem_truncate_range(inode, newsize, (loff_t)-1);
614 /* unmap again to remove racily COWed private pages */
615 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
619 setattr_copy(inode, attr);
620 #ifdef CONFIG_TMPFS_POSIX_ACL
621 if (attr->ia_valid & ATTR_MODE)
622 error = generic_acl_chmod(inode);
627 static void shmem_evict_inode(struct inode *inode)
629 struct shmem_inode_info *info = SHMEM_I(inode);
630 struct shmem_xattr *xattr, *nxattr;
632 if (inode->i_mapping->a_ops == &shmem_aops) {
633 shmem_unacct_size(info->flags, inode->i_size);
635 shmem_truncate_range(inode, 0, (loff_t)-1);
636 if (!list_empty(&info->swaplist)) {
637 mutex_lock(&shmem_swaplist_mutex);
638 list_del_init(&info->swaplist);
639 mutex_unlock(&shmem_swaplist_mutex);
642 kfree(info->symlink);
644 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
648 BUG_ON(inode->i_blocks);
649 shmem_free_inode(inode->i_sb);
654 * If swap found in inode, free it and move page from swapcache to filecache.
656 static int shmem_unuse_inode(struct shmem_inode_info *info,
657 swp_entry_t swap, struct page **pagep)
659 struct address_space *mapping = info->vfs_inode.i_mapping;
665 radswap = swp_to_radix_entry(swap);
666 index = radix_tree_locate_item(&mapping->page_tree, radswap);
671 * Move _head_ to start search for next from here.
672 * But be careful: shmem_evict_inode checks list_empty without taking
673 * mutex, and there's an instant in list_move_tail when info->swaplist
674 * would appear empty, if it were the only one on shmem_swaplist.
676 if (shmem_swaplist.next != &info->swaplist)
677 list_move_tail(&shmem_swaplist, &info->swaplist);
679 gfp = mapping_gfp_mask(mapping);
680 if (shmem_should_replace_page(*pagep, gfp)) {
681 mutex_unlock(&shmem_swaplist_mutex);
682 error = shmem_replace_page(pagep, gfp, info, index);
683 mutex_lock(&shmem_swaplist_mutex);
685 * We needed to drop mutex to make that restrictive page
686 * allocation; but the inode might already be freed by now,
687 * and we cannot refer to inode or mapping or info to check.
688 * However, we do hold page lock on the PageSwapCache page,
689 * so can check if that still has our reference remaining.
691 if (!page_swapcount(*pagep))
696 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
697 * but also to hold up shmem_evict_inode(): so inode cannot be freed
698 * beneath us (pagelock doesn't help until the page is in pagecache).
701 error = shmem_add_to_page_cache(*pagep, mapping, index,
702 GFP_NOWAIT, radswap);
703 if (error != -ENOMEM) {
705 * Truncation and eviction use free_swap_and_cache(), which
706 * only does trylock page: if we raced, best clean up here.
708 delete_from_swap_cache(*pagep);
709 set_page_dirty(*pagep);
711 spin_lock(&info->lock);
713 spin_unlock(&info->lock);
716 error = 1; /* not an error, but entry was found */
722 * Search through swapped inodes to find and replace swap by page.
724 int shmem_unuse(swp_entry_t swap, struct page *page)
726 struct list_head *this, *next;
727 struct shmem_inode_info *info;
732 * There's a faint possibility that swap page was replaced before
733 * caller locked it: it will come back later with the right page.
735 if (unlikely(!PageSwapCache(page)))
739 * Charge page using GFP_KERNEL while we can wait, before taking
740 * the shmem_swaplist_mutex which might hold up shmem_writepage().
741 * Charged back to the user (not to caller) when swap account is used.
743 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
746 /* No radix_tree_preload: swap entry keeps a place for page in tree */
748 mutex_lock(&shmem_swaplist_mutex);
749 list_for_each_safe(this, next, &shmem_swaplist) {
750 info = list_entry(this, struct shmem_inode_info, swaplist);
752 found = shmem_unuse_inode(info, swap, &page);
754 list_del_init(&info->swaplist);
759 mutex_unlock(&shmem_swaplist_mutex);
765 page_cache_release(page);
770 * Move the page from the page cache to the swap cache.
772 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
774 struct shmem_inode_info *info;
775 struct address_space *mapping;
780 BUG_ON(!PageLocked(page));
781 mapping = page->mapping;
783 inode = mapping->host;
784 info = SHMEM_I(inode);
785 if (info->flags & VM_LOCKED)
787 if (!total_swap_pages)
791 * shmem_backing_dev_info's capabilities prevent regular writeback or
792 * sync from ever calling shmem_writepage; but a stacking filesystem
793 * might use ->writepage of its underlying filesystem, in which case
794 * tmpfs should write out to swap only in response to memory pressure,
795 * and not for the writeback threads or sync.
797 if (!wbc->for_reclaim) {
798 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
803 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
804 * value into swapfile.c, the only way we can correctly account for a
805 * fallocated page arriving here is now to initialize it and write it.
807 * That's okay for a page already fallocated earlier, but if we have
808 * not yet completed the fallocation, then (a) we want to keep track
809 * of this page in case we have to undo it, and (b) it may not be a
810 * good idea to continue anyway, once we're pushing into swap. So
811 * reactivate the page, and let shmem_fallocate() quit when too many.
813 if (!PageUptodate(page)) {
814 if (inode->i_private) {
815 struct shmem_falloc *shmem_falloc;
816 spin_lock(&inode->i_lock);
817 shmem_falloc = inode->i_private;
819 index >= shmem_falloc->start &&
820 index < shmem_falloc->next)
821 shmem_falloc->nr_unswapped++;
824 spin_unlock(&inode->i_lock);
828 clear_highpage(page);
829 flush_dcache_page(page);
830 SetPageUptodate(page);
833 swap = get_swap_page();
838 * Add inode to shmem_unuse()'s list of swapped-out inodes,
839 * if it's not already there. Do it now before the page is
840 * moved to swap cache, when its pagelock no longer protects
841 * the inode from eviction. But don't unlock the mutex until
842 * we've incremented swapped, because shmem_unuse_inode() will
843 * prune a !swapped inode from the swaplist under this mutex.
845 mutex_lock(&shmem_swaplist_mutex);
846 if (list_empty(&info->swaplist))
847 list_add_tail(&info->swaplist, &shmem_swaplist);
849 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
850 swap_shmem_alloc(swap);
851 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
853 spin_lock(&info->lock);
855 shmem_recalc_inode(inode);
856 spin_unlock(&info->lock);
858 mutex_unlock(&shmem_swaplist_mutex);
859 BUG_ON(page_mapped(page));
860 swap_writepage(page, wbc);
864 mutex_unlock(&shmem_swaplist_mutex);
865 swapcache_free(swap, NULL);
867 set_page_dirty(page);
868 if (wbc->for_reclaim)
869 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
876 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
880 if (!mpol || mpol->mode == MPOL_DEFAULT)
881 return; /* show nothing */
883 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
885 seq_printf(seq, ",mpol=%s", buffer);
888 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
890 struct mempolicy *mpol = NULL;
892 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
895 spin_unlock(&sbinfo->stat_lock);
899 #endif /* CONFIG_TMPFS */
901 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
902 struct shmem_inode_info *info, pgoff_t index)
904 struct mempolicy mpol, *spol;
905 struct vm_area_struct pvma;
907 spol = mpol_cond_copy(&mpol,
908 mpol_shared_policy_lookup(&info->policy, index));
910 /* Create a pseudo vma that just contains the policy */
912 pvma.vm_pgoff = index;
914 pvma.vm_policy = spol;
915 return swapin_readahead(swap, gfp, &pvma, 0);
918 static struct page *shmem_alloc_page(gfp_t gfp,
919 struct shmem_inode_info *info, pgoff_t index)
921 struct vm_area_struct pvma;
923 /* Create a pseudo vma that just contains the policy */
925 pvma.vm_pgoff = index;
927 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
930 * alloc_page_vma() will drop the shared policy reference
932 return alloc_page_vma(gfp, &pvma, 0);
934 #else /* !CONFIG_NUMA */
936 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
939 #endif /* CONFIG_TMPFS */
941 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
942 struct shmem_inode_info *info, pgoff_t index)
944 return swapin_readahead(swap, gfp, NULL, 0);
947 static inline struct page *shmem_alloc_page(gfp_t gfp,
948 struct shmem_inode_info *info, pgoff_t index)
950 return alloc_page(gfp);
952 #endif /* CONFIG_NUMA */
954 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
955 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
962 * When a page is moved from swapcache to shmem filecache (either by the
963 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
964 * shmem_unuse_inode()), it may have been read in earlier from swap, in
965 * ignorance of the mapping it belongs to. If that mapping has special
966 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
967 * we may need to copy to a suitable page before moving to filecache.
969 * In a future release, this may well be extended to respect cpuset and
970 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
971 * but for now it is a simple matter of zone.
973 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
975 return page_zonenum(page) > gfp_zone(gfp);
978 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
979 struct shmem_inode_info *info, pgoff_t index)
981 struct page *oldpage, *newpage;
982 struct address_space *swap_mapping;
987 swap_index = page_private(oldpage);
988 swap_mapping = page_mapping(oldpage);
991 * We have arrived here because our zones are constrained, so don't
992 * limit chance of success by further cpuset and node constraints.
994 gfp &= ~GFP_CONSTRAINT_MASK;
995 newpage = shmem_alloc_page(gfp, info, index);
998 VM_BUG_ON(shmem_should_replace_page(newpage, gfp));
1001 page_cache_get(newpage);
1002 copy_highpage(newpage, oldpage);
1004 VM_BUG_ON(!PageLocked(oldpage));
1005 __set_page_locked(newpage);
1006 VM_BUG_ON(!PageUptodate(oldpage));
1007 SetPageUptodate(newpage);
1008 VM_BUG_ON(!PageSwapBacked(oldpage));
1009 SetPageSwapBacked(newpage);
1010 VM_BUG_ON(!swap_index);
1011 set_page_private(newpage, swap_index);
1012 VM_BUG_ON(!PageSwapCache(oldpage));
1013 SetPageSwapCache(newpage);
1016 * Our caller will very soon move newpage out of swapcache, but it's
1017 * a nice clean interface for us to replace oldpage by newpage there.
1019 spin_lock_irq(&swap_mapping->tree_lock);
1020 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1022 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1023 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1024 spin_unlock_irq(&swap_mapping->tree_lock);
1027 mem_cgroup_replace_page_cache(oldpage, newpage);
1028 lru_cache_add_anon(newpage);
1030 ClearPageSwapCache(oldpage);
1031 set_page_private(oldpage, 0);
1033 unlock_page(oldpage);
1034 page_cache_release(oldpage);
1035 page_cache_release(oldpage);
1040 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1042 * If we allocate a new one we do not mark it dirty. That's up to the
1043 * vm. If we swap it in we mark it dirty since we also free the swap
1044 * entry since a page cannot live in both the swap and page cache
1046 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1047 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1049 struct address_space *mapping = inode->i_mapping;
1050 struct shmem_inode_info *info;
1051 struct shmem_sb_info *sbinfo;
1058 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1062 page = find_lock_page(mapping, index);
1063 if (radix_tree_exceptional_entry(page)) {
1064 swap = radix_to_swp_entry(page);
1068 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1069 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1074 /* fallocated page? */
1075 if (page && !PageUptodate(page)) {
1076 if (sgp != SGP_READ)
1079 page_cache_release(page);
1082 if (page || (sgp == SGP_READ && !swap.val)) {
1088 * Fast cache lookup did not find it:
1089 * bring it back from swap or allocate.
1091 info = SHMEM_I(inode);
1092 sbinfo = SHMEM_SB(inode->i_sb);
1095 /* Look it up and read it in.. */
1096 page = lookup_swap_cache(swap);
1098 /* here we actually do the io */
1100 *fault_type |= VM_FAULT_MAJOR;
1101 page = shmem_swapin(swap, gfp, info, index);
1108 /* We have to do this with page locked to prevent races */
1110 if (!PageSwapCache(page) || page->mapping) {
1111 error = -EEXIST; /* try again */
1114 if (!PageUptodate(page)) {
1118 wait_on_page_writeback(page);
1120 if (shmem_should_replace_page(page, gfp)) {
1121 error = shmem_replace_page(&page, gfp, info, index);
1126 error = mem_cgroup_cache_charge(page, current->mm,
1127 gfp & GFP_RECLAIM_MASK);
1129 error = shmem_add_to_page_cache(page, mapping, index,
1130 gfp, swp_to_radix_entry(swap));
1134 spin_lock(&info->lock);
1136 shmem_recalc_inode(inode);
1137 spin_unlock(&info->lock);
1139 delete_from_swap_cache(page);
1140 set_page_dirty(page);
1144 if (shmem_acct_block(info->flags)) {
1148 if (sbinfo->max_blocks) {
1149 if (percpu_counter_compare(&sbinfo->used_blocks,
1150 sbinfo->max_blocks) >= 0) {
1154 percpu_counter_inc(&sbinfo->used_blocks);
1157 page = shmem_alloc_page(gfp, info, index);
1163 SetPageSwapBacked(page);
1164 __set_page_locked(page);
1165 error = mem_cgroup_cache_charge(page, current->mm,
1166 gfp & GFP_RECLAIM_MASK);
1168 error = shmem_add_to_page_cache(page, mapping, index,
1172 lru_cache_add_anon(page);
1174 spin_lock(&info->lock);
1176 inode->i_blocks += BLOCKS_PER_PAGE;
1177 shmem_recalc_inode(inode);
1178 spin_unlock(&info->lock);
1182 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1184 if (sgp == SGP_FALLOC)
1188 * Let SGP_WRITE caller clear ends if write does not fill page;
1189 * but SGP_FALLOC on a page fallocated earlier must initialize
1190 * it now, lest undo on failure cancel our earlier guarantee.
1192 if (sgp != SGP_WRITE) {
1193 clear_highpage(page);
1194 flush_dcache_page(page);
1195 SetPageUptodate(page);
1197 if (sgp == SGP_DIRTY)
1198 set_page_dirty(page);
1201 /* Perhaps the file has been truncated since we checked */
1202 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1203 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1217 info = SHMEM_I(inode);
1218 ClearPageDirty(page);
1219 delete_from_page_cache(page);
1220 spin_lock(&info->lock);
1222 inode->i_blocks -= BLOCKS_PER_PAGE;
1223 spin_unlock(&info->lock);
1225 sbinfo = SHMEM_SB(inode->i_sb);
1226 if (sbinfo->max_blocks)
1227 percpu_counter_add(&sbinfo->used_blocks, -1);
1229 shmem_unacct_blocks(info->flags, 1);
1231 if (swap.val && error != -EINVAL) {
1232 struct page *test = find_get_page(mapping, index);
1233 if (test && !radix_tree_exceptional_entry(test))
1234 page_cache_release(test);
1235 /* Have another try if the entry has changed */
1236 if (test != swp_to_radix_entry(swap))
1241 page_cache_release(page);
1243 if (error == -ENOSPC && !once++) {
1244 info = SHMEM_I(inode);
1245 spin_lock(&info->lock);
1246 shmem_recalc_inode(inode);
1247 spin_unlock(&info->lock);
1250 if (error == -EEXIST)
1255 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1257 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1259 int ret = VM_FAULT_LOCKED;
1261 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1263 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1265 if (ret & VM_FAULT_MAJOR) {
1266 count_vm_event(PGMAJFAULT);
1267 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1273 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1275 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1276 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1279 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1282 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1285 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1286 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1290 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1292 struct inode *inode = file->f_path.dentry->d_inode;
1293 struct shmem_inode_info *info = SHMEM_I(inode);
1294 int retval = -ENOMEM;
1296 spin_lock(&info->lock);
1297 if (lock && !(info->flags & VM_LOCKED)) {
1298 if (!user_shm_lock(inode->i_size, user))
1300 info->flags |= VM_LOCKED;
1301 mapping_set_unevictable(file->f_mapping);
1303 if (!lock && (info->flags & VM_LOCKED) && user) {
1304 user_shm_unlock(inode->i_size, user);
1305 info->flags &= ~VM_LOCKED;
1306 mapping_clear_unevictable(file->f_mapping);
1311 spin_unlock(&info->lock);
1315 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1317 file_accessed(file);
1318 vma->vm_ops = &shmem_vm_ops;
1319 vma->vm_flags |= VM_CAN_NONLINEAR;
1323 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1324 umode_t mode, dev_t dev, unsigned long flags)
1326 struct inode *inode;
1327 struct shmem_inode_info *info;
1328 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1330 if (shmem_reserve_inode(sb))
1333 inode = new_inode(sb);
1335 inode->i_ino = get_next_ino();
1336 inode_init_owner(inode, dir, mode);
1337 inode->i_blocks = 0;
1338 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1339 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1340 inode->i_generation = get_seconds();
1341 info = SHMEM_I(inode);
1342 memset(info, 0, (char *)inode - (char *)info);
1343 spin_lock_init(&info->lock);
1344 info->flags = flags & VM_NORESERVE;
1345 INIT_LIST_HEAD(&info->swaplist);
1346 INIT_LIST_HEAD(&info->xattr_list);
1347 cache_no_acl(inode);
1349 switch (mode & S_IFMT) {
1351 inode->i_op = &shmem_special_inode_operations;
1352 init_special_inode(inode, mode, dev);
1355 inode->i_mapping->a_ops = &shmem_aops;
1356 inode->i_op = &shmem_inode_operations;
1357 inode->i_fop = &shmem_file_operations;
1358 mpol_shared_policy_init(&info->policy,
1359 shmem_get_sbmpol(sbinfo));
1363 /* Some things misbehave if size == 0 on a directory */
1364 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1365 inode->i_op = &shmem_dir_inode_operations;
1366 inode->i_fop = &simple_dir_operations;
1370 * Must not load anything in the rbtree,
1371 * mpol_free_shared_policy will not be called.
1373 mpol_shared_policy_init(&info->policy, NULL);
1377 shmem_free_inode(sb);
1382 static const struct inode_operations shmem_symlink_inode_operations;
1383 static const struct inode_operations shmem_short_symlink_operations;
1385 #ifdef CONFIG_TMPFS_XATTR
1386 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1388 #define shmem_initxattrs NULL
1392 shmem_write_begin(struct file *file, struct address_space *mapping,
1393 loff_t pos, unsigned len, unsigned flags,
1394 struct page **pagep, void **fsdata)
1396 struct inode *inode = mapping->host;
1397 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1398 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1402 shmem_write_end(struct file *file, struct address_space *mapping,
1403 loff_t pos, unsigned len, unsigned copied,
1404 struct page *page, void *fsdata)
1406 struct inode *inode = mapping->host;
1408 if (pos + copied > inode->i_size)
1409 i_size_write(inode, pos + copied);
1411 if (!PageUptodate(page)) {
1412 if (copied < PAGE_CACHE_SIZE) {
1413 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1414 zero_user_segments(page, 0, from,
1415 from + copied, PAGE_CACHE_SIZE);
1417 SetPageUptodate(page);
1419 set_page_dirty(page);
1421 page_cache_release(page);
1426 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1428 struct inode *inode = filp->f_path.dentry->d_inode;
1429 struct address_space *mapping = inode->i_mapping;
1431 unsigned long offset;
1432 enum sgp_type sgp = SGP_READ;
1435 * Might this read be for a stacking filesystem? Then when reading
1436 * holes of a sparse file, we actually need to allocate those pages,
1437 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1439 if (segment_eq(get_fs(), KERNEL_DS))
1442 index = *ppos >> PAGE_CACHE_SHIFT;
1443 offset = *ppos & ~PAGE_CACHE_MASK;
1446 struct page *page = NULL;
1448 unsigned long nr, ret;
1449 loff_t i_size = i_size_read(inode);
1451 end_index = i_size >> PAGE_CACHE_SHIFT;
1452 if (index > end_index)
1454 if (index == end_index) {
1455 nr = i_size & ~PAGE_CACHE_MASK;
1460 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1462 if (desc->error == -EINVAL)
1470 * We must evaluate after, since reads (unlike writes)
1471 * are called without i_mutex protection against truncate
1473 nr = PAGE_CACHE_SIZE;
1474 i_size = i_size_read(inode);
1475 end_index = i_size >> PAGE_CACHE_SHIFT;
1476 if (index == end_index) {
1477 nr = i_size & ~PAGE_CACHE_MASK;
1480 page_cache_release(page);
1488 * If users can be writing to this page using arbitrary
1489 * virtual addresses, take care about potential aliasing
1490 * before reading the page on the kernel side.
1492 if (mapping_writably_mapped(mapping))
1493 flush_dcache_page(page);
1495 * Mark the page accessed if we read the beginning.
1498 mark_page_accessed(page);
1500 page = ZERO_PAGE(0);
1501 page_cache_get(page);
1505 * Ok, we have the page, and it's up-to-date, so
1506 * now we can copy it to user space...
1508 * The actor routine returns how many bytes were actually used..
1509 * NOTE! This may not be the same as how much of a user buffer
1510 * we filled up (we may be padding etc), so we can only update
1511 * "pos" here (the actor routine has to update the user buffer
1512 * pointers and the remaining count).
1514 ret = actor(desc, page, offset, nr);
1516 index += offset >> PAGE_CACHE_SHIFT;
1517 offset &= ~PAGE_CACHE_MASK;
1519 page_cache_release(page);
1520 if (ret != nr || !desc->count)
1526 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1527 file_accessed(filp);
1530 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1531 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1533 struct file *filp = iocb->ki_filp;
1537 loff_t *ppos = &iocb->ki_pos;
1539 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1543 for (seg = 0; seg < nr_segs; seg++) {
1544 read_descriptor_t desc;
1547 desc.arg.buf = iov[seg].iov_base;
1548 desc.count = iov[seg].iov_len;
1549 if (desc.count == 0)
1552 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1553 retval += desc.written;
1555 retval = retval ?: desc.error;
1564 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1565 struct pipe_inode_info *pipe, size_t len,
1568 struct address_space *mapping = in->f_mapping;
1569 struct inode *inode = mapping->host;
1570 unsigned int loff, nr_pages, req_pages;
1571 struct page *pages[PIPE_DEF_BUFFERS];
1572 struct partial_page partial[PIPE_DEF_BUFFERS];
1574 pgoff_t index, end_index;
1577 struct splice_pipe_desc spd = {
1580 .nr_pages_max = PIPE_DEF_BUFFERS,
1582 .ops = &page_cache_pipe_buf_ops,
1583 .spd_release = spd_release_page,
1586 isize = i_size_read(inode);
1587 if (unlikely(*ppos >= isize))
1590 left = isize - *ppos;
1591 if (unlikely(left < len))
1594 if (splice_grow_spd(pipe, &spd))
1597 index = *ppos >> PAGE_CACHE_SHIFT;
1598 loff = *ppos & ~PAGE_CACHE_MASK;
1599 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1600 nr_pages = min(req_pages, pipe->buffers);
1602 spd.nr_pages = find_get_pages_contig(mapping, index,
1603 nr_pages, spd.pages);
1604 index += spd.nr_pages;
1607 while (spd.nr_pages < nr_pages) {
1608 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1612 spd.pages[spd.nr_pages++] = page;
1616 index = *ppos >> PAGE_CACHE_SHIFT;
1617 nr_pages = spd.nr_pages;
1620 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1621 unsigned int this_len;
1626 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1627 page = spd.pages[page_nr];
1629 if (!PageUptodate(page) || page->mapping != mapping) {
1630 error = shmem_getpage(inode, index, &page,
1635 page_cache_release(spd.pages[page_nr]);
1636 spd.pages[page_nr] = page;
1639 isize = i_size_read(inode);
1640 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1641 if (unlikely(!isize || index > end_index))
1644 if (end_index == index) {
1647 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1651 this_len = min(this_len, plen - loff);
1655 spd.partial[page_nr].offset = loff;
1656 spd.partial[page_nr].len = this_len;
1663 while (page_nr < nr_pages)
1664 page_cache_release(spd.pages[page_nr++]);
1667 error = splice_to_pipe(pipe, &spd);
1669 splice_shrink_spd(&spd);
1679 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1681 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1682 pgoff_t index, pgoff_t end, int origin)
1685 struct pagevec pvec;
1686 pgoff_t indices[PAGEVEC_SIZE];
1690 pagevec_init(&pvec, 0);
1691 pvec.nr = 1; /* start small: we may be there already */
1693 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1694 pvec.nr, pvec.pages, indices);
1696 if (origin == SEEK_DATA)
1700 for (i = 0; i < pvec.nr; i++, index++) {
1701 if (index < indices[i]) {
1702 if (origin == SEEK_HOLE) {
1708 page = pvec.pages[i];
1709 if (page && !radix_tree_exceptional_entry(page)) {
1710 if (!PageUptodate(page))
1714 (page && origin == SEEK_DATA) ||
1715 (!page && origin == SEEK_HOLE)) {
1720 shmem_deswap_pagevec(&pvec);
1721 pagevec_release(&pvec);
1722 pvec.nr = PAGEVEC_SIZE;
1728 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int origin)
1730 struct address_space *mapping;
1731 struct inode *inode;
1735 if (origin != SEEK_DATA && origin != SEEK_HOLE)
1736 return generic_file_llseek_size(file, offset, origin,
1738 mapping = file->f_mapping;
1739 inode = mapping->host;
1740 mutex_lock(&inode->i_mutex);
1741 /* We're holding i_mutex so we can access i_size directly */
1745 else if (offset >= inode->i_size)
1748 start = offset >> PAGE_CACHE_SHIFT;
1749 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1750 new_offset = shmem_seek_hole_data(mapping, start, end, origin);
1751 new_offset <<= PAGE_CACHE_SHIFT;
1752 if (new_offset > offset) {
1753 if (new_offset < inode->i_size)
1754 offset = new_offset;
1755 else if (origin == SEEK_DATA)
1758 offset = inode->i_size;
1762 if (offset >= 0 && offset != file->f_pos) {
1763 file->f_pos = offset;
1764 file->f_version = 0;
1766 mutex_unlock(&inode->i_mutex);
1770 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1773 struct inode *inode = file->f_path.dentry->d_inode;
1774 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1775 struct shmem_falloc shmem_falloc;
1776 pgoff_t start, index, end;
1779 mutex_lock(&inode->i_mutex);
1781 if (mode & FALLOC_FL_PUNCH_HOLE) {
1782 struct address_space *mapping = file->f_mapping;
1783 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1784 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1786 if ((u64)unmap_end > (u64)unmap_start)
1787 unmap_mapping_range(mapping, unmap_start,
1788 1 + unmap_end - unmap_start, 0);
1789 shmem_truncate_range(inode, offset, offset + len - 1);
1790 /* No need to unmap again: hole-punching leaves COWed pages */
1795 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1796 error = inode_newsize_ok(inode, offset + len);
1800 start = offset >> PAGE_CACHE_SHIFT;
1801 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1802 /* Try to avoid a swapstorm if len is impossible to satisfy */
1803 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1808 shmem_falloc.start = start;
1809 shmem_falloc.next = start;
1810 shmem_falloc.nr_falloced = 0;
1811 shmem_falloc.nr_unswapped = 0;
1812 spin_lock(&inode->i_lock);
1813 inode->i_private = &shmem_falloc;
1814 spin_unlock(&inode->i_lock);
1816 for (index = start; index < end; index++) {
1820 * Good, the fallocate(2) manpage permits EINTR: we may have
1821 * been interrupted because we are using up too much memory.
1823 if (signal_pending(current))
1825 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1828 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1831 /* Remove the !PageUptodate pages we added */
1832 shmem_undo_range(inode,
1833 (loff_t)start << PAGE_CACHE_SHIFT,
1834 (loff_t)index << PAGE_CACHE_SHIFT, true);
1839 * Inform shmem_writepage() how far we have reached.
1840 * No need for lock or barrier: we have the page lock.
1842 shmem_falloc.next++;
1843 if (!PageUptodate(page))
1844 shmem_falloc.nr_falloced++;
1847 * If !PageUptodate, leave it that way so that freeable pages
1848 * can be recognized if we need to rollback on error later.
1849 * But set_page_dirty so that memory pressure will swap rather
1850 * than free the pages we are allocating (and SGP_CACHE pages
1851 * might still be clean: we now need to mark those dirty too).
1853 set_page_dirty(page);
1855 page_cache_release(page);
1859 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1860 i_size_write(inode, offset + len);
1861 inode->i_ctime = CURRENT_TIME;
1863 spin_lock(&inode->i_lock);
1864 inode->i_private = NULL;
1865 spin_unlock(&inode->i_lock);
1867 mutex_unlock(&inode->i_mutex);
1871 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1873 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1875 buf->f_type = TMPFS_MAGIC;
1876 buf->f_bsize = PAGE_CACHE_SIZE;
1877 buf->f_namelen = NAME_MAX;
1878 if (sbinfo->max_blocks) {
1879 buf->f_blocks = sbinfo->max_blocks;
1881 buf->f_bfree = sbinfo->max_blocks -
1882 percpu_counter_sum(&sbinfo->used_blocks);
1884 if (sbinfo->max_inodes) {
1885 buf->f_files = sbinfo->max_inodes;
1886 buf->f_ffree = sbinfo->free_inodes;
1888 /* else leave those fields 0 like simple_statfs */
1893 * File creation. Allocate an inode, and we're done..
1896 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1898 struct inode *inode;
1899 int error = -ENOSPC;
1901 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1903 error = security_inode_init_security(inode, dir,
1905 shmem_initxattrs, NULL);
1907 if (error != -EOPNOTSUPP) {
1912 #ifdef CONFIG_TMPFS_POSIX_ACL
1913 error = generic_acl_init(inode, dir);
1921 dir->i_size += BOGO_DIRENT_SIZE;
1922 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1923 d_instantiate(dentry, inode);
1924 dget(dentry); /* Extra count - pin the dentry in core */
1929 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1933 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1939 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1940 struct nameidata *nd)
1942 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1948 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1950 struct inode *inode = old_dentry->d_inode;
1954 * No ordinary (disk based) filesystem counts links as inodes;
1955 * but each new link needs a new dentry, pinning lowmem, and
1956 * tmpfs dentries cannot be pruned until they are unlinked.
1958 ret = shmem_reserve_inode(inode->i_sb);
1962 dir->i_size += BOGO_DIRENT_SIZE;
1963 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1965 ihold(inode); /* New dentry reference */
1966 dget(dentry); /* Extra pinning count for the created dentry */
1967 d_instantiate(dentry, inode);
1972 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1974 struct inode *inode = dentry->d_inode;
1976 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1977 shmem_free_inode(inode->i_sb);
1979 dir->i_size -= BOGO_DIRENT_SIZE;
1980 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1982 dput(dentry); /* Undo the count from "create" - this does all the work */
1986 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1988 if (!simple_empty(dentry))
1991 drop_nlink(dentry->d_inode);
1993 return shmem_unlink(dir, dentry);
1997 * The VFS layer already does all the dentry stuff for rename,
1998 * we just have to decrement the usage count for the target if
1999 * it exists so that the VFS layer correctly free's it when it
2002 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2004 struct inode *inode = old_dentry->d_inode;
2005 int they_are_dirs = S_ISDIR(inode->i_mode);
2007 if (!simple_empty(new_dentry))
2010 if (new_dentry->d_inode) {
2011 (void) shmem_unlink(new_dir, new_dentry);
2013 drop_nlink(old_dir);
2014 } else if (they_are_dirs) {
2015 drop_nlink(old_dir);
2019 old_dir->i_size -= BOGO_DIRENT_SIZE;
2020 new_dir->i_size += BOGO_DIRENT_SIZE;
2021 old_dir->i_ctime = old_dir->i_mtime =
2022 new_dir->i_ctime = new_dir->i_mtime =
2023 inode->i_ctime = CURRENT_TIME;
2027 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2031 struct inode *inode;
2034 struct shmem_inode_info *info;
2036 len = strlen(symname) + 1;
2037 if (len > PAGE_CACHE_SIZE)
2038 return -ENAMETOOLONG;
2040 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2044 error = security_inode_init_security(inode, dir, &dentry->d_name,
2045 shmem_initxattrs, NULL);
2047 if (error != -EOPNOTSUPP) {
2054 info = SHMEM_I(inode);
2055 inode->i_size = len-1;
2056 if (len <= SHORT_SYMLINK_LEN) {
2057 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2058 if (!info->symlink) {
2062 inode->i_op = &shmem_short_symlink_operations;
2064 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2069 inode->i_mapping->a_ops = &shmem_aops;
2070 inode->i_op = &shmem_symlink_inode_operations;
2071 kaddr = kmap_atomic(page);
2072 memcpy(kaddr, symname, len);
2073 kunmap_atomic(kaddr);
2074 SetPageUptodate(page);
2075 set_page_dirty(page);
2077 page_cache_release(page);
2079 dir->i_size += BOGO_DIRENT_SIZE;
2080 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2081 d_instantiate(dentry, inode);
2086 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2088 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2092 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2094 struct page *page = NULL;
2095 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2096 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2102 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2104 if (!IS_ERR(nd_get_link(nd))) {
2105 struct page *page = cookie;
2107 mark_page_accessed(page);
2108 page_cache_release(page);
2112 #ifdef CONFIG_TMPFS_XATTR
2114 * Superblocks without xattr inode operations may get some security.* xattr
2115 * support from the LSM "for free". As soon as we have any other xattrs
2116 * like ACLs, we also need to implement the security.* handlers at
2117 * filesystem level, though.
2121 * Allocate new xattr and copy in the value; but leave the name to callers.
2123 static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2125 struct shmem_xattr *new_xattr;
2129 len = sizeof(*new_xattr) + size;
2130 if (len <= sizeof(*new_xattr))
2133 new_xattr = kmalloc(len, GFP_KERNEL);
2137 new_xattr->size = size;
2138 memcpy(new_xattr->value, value, size);
2143 * Callback for security_inode_init_security() for acquiring xattrs.
2145 static int shmem_initxattrs(struct inode *inode,
2146 const struct xattr *xattr_array,
2149 struct shmem_inode_info *info = SHMEM_I(inode);
2150 const struct xattr *xattr;
2151 struct shmem_xattr *new_xattr;
2154 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2155 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2159 len = strlen(xattr->name) + 1;
2160 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2162 if (!new_xattr->name) {
2167 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2168 XATTR_SECURITY_PREFIX_LEN);
2169 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2172 spin_lock(&info->lock);
2173 list_add(&new_xattr->list, &info->xattr_list);
2174 spin_unlock(&info->lock);
2180 static int shmem_xattr_get(struct dentry *dentry, const char *name,
2181 void *buffer, size_t size)
2183 struct shmem_inode_info *info;
2184 struct shmem_xattr *xattr;
2187 info = SHMEM_I(dentry->d_inode);
2189 spin_lock(&info->lock);
2190 list_for_each_entry(xattr, &info->xattr_list, list) {
2191 if (strcmp(name, xattr->name))
2196 if (size < xattr->size)
2199 memcpy(buffer, xattr->value, xattr->size);
2203 spin_unlock(&info->lock);
2207 static int shmem_xattr_set(struct inode *inode, const char *name,
2208 const void *value, size_t size, int flags)
2210 struct shmem_inode_info *info = SHMEM_I(inode);
2211 struct shmem_xattr *xattr;
2212 struct shmem_xattr *new_xattr = NULL;
2215 /* value == NULL means remove */
2217 new_xattr = shmem_xattr_alloc(value, size);
2221 new_xattr->name = kstrdup(name, GFP_KERNEL);
2222 if (!new_xattr->name) {
2228 spin_lock(&info->lock);
2229 list_for_each_entry(xattr, &info->xattr_list, list) {
2230 if (!strcmp(name, xattr->name)) {
2231 if (flags & XATTR_CREATE) {
2234 } else if (new_xattr) {
2235 list_replace(&xattr->list, &new_xattr->list);
2237 list_del(&xattr->list);
2242 if (flags & XATTR_REPLACE) {
2246 list_add(&new_xattr->list, &info->xattr_list);
2250 spin_unlock(&info->lock);
2257 static const struct xattr_handler *shmem_xattr_handlers[] = {
2258 #ifdef CONFIG_TMPFS_POSIX_ACL
2259 &generic_acl_access_handler,
2260 &generic_acl_default_handler,
2265 static int shmem_xattr_validate(const char *name)
2267 struct { const char *prefix; size_t len; } arr[] = {
2268 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2269 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2273 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2274 size_t preflen = arr[i].len;
2275 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2284 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2285 void *buffer, size_t size)
2290 * If this is a request for a synthetic attribute in the system.*
2291 * namespace use the generic infrastructure to resolve a handler
2292 * for it via sb->s_xattr.
2294 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2295 return generic_getxattr(dentry, name, buffer, size);
2297 err = shmem_xattr_validate(name);
2301 return shmem_xattr_get(dentry, name, buffer, size);
2304 static int shmem_setxattr(struct dentry *dentry, const char *name,
2305 const void *value, size_t size, int flags)
2310 * If this is a request for a synthetic attribute in the system.*
2311 * namespace use the generic infrastructure to resolve a handler
2312 * for it via sb->s_xattr.
2314 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2315 return generic_setxattr(dentry, name, value, size, flags);
2317 err = shmem_xattr_validate(name);
2322 value = ""; /* empty EA, do not remove */
2324 return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
2328 static int shmem_removexattr(struct dentry *dentry, const char *name)
2333 * If this is a request for a synthetic attribute in the system.*
2334 * namespace use the generic infrastructure to resolve a handler
2335 * for it via sb->s_xattr.
2337 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2338 return generic_removexattr(dentry, name);
2340 err = shmem_xattr_validate(name);
2344 return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
2347 static bool xattr_is_trusted(const char *name)
2349 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2352 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2354 bool trusted = capable(CAP_SYS_ADMIN);
2355 struct shmem_xattr *xattr;
2356 struct shmem_inode_info *info;
2359 info = SHMEM_I(dentry->d_inode);
2361 spin_lock(&info->lock);
2362 list_for_each_entry(xattr, &info->xattr_list, list) {
2365 /* skip "trusted." attributes for unprivileged callers */
2366 if (!trusted && xattr_is_trusted(xattr->name))
2369 len = strlen(xattr->name) + 1;
2376 memcpy(buffer, xattr->name, len);
2380 spin_unlock(&info->lock);
2384 #endif /* CONFIG_TMPFS_XATTR */
2386 static const struct inode_operations shmem_short_symlink_operations = {
2387 .readlink = generic_readlink,
2388 .follow_link = shmem_follow_short_symlink,
2389 #ifdef CONFIG_TMPFS_XATTR
2390 .setxattr = shmem_setxattr,
2391 .getxattr = shmem_getxattr,
2392 .listxattr = shmem_listxattr,
2393 .removexattr = shmem_removexattr,
2397 static const struct inode_operations shmem_symlink_inode_operations = {
2398 .readlink = generic_readlink,
2399 .follow_link = shmem_follow_link,
2400 .put_link = shmem_put_link,
2401 #ifdef CONFIG_TMPFS_XATTR
2402 .setxattr = shmem_setxattr,
2403 .getxattr = shmem_getxattr,
2404 .listxattr = shmem_listxattr,
2405 .removexattr = shmem_removexattr,
2409 static struct dentry *shmem_get_parent(struct dentry *child)
2411 return ERR_PTR(-ESTALE);
2414 static int shmem_match(struct inode *ino, void *vfh)
2418 inum = (inum << 32) | fh[1];
2419 return ino->i_ino == inum && fh[0] == ino->i_generation;
2422 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2423 struct fid *fid, int fh_len, int fh_type)
2425 struct inode *inode;
2426 struct dentry *dentry = NULL;
2427 u64 inum = fid->raw[2];
2428 inum = (inum << 32) | fid->raw[1];
2433 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2434 shmem_match, fid->raw);
2436 dentry = d_find_alias(inode);
2443 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2444 struct inode *parent)
2451 if (inode_unhashed(inode)) {
2452 /* Unfortunately insert_inode_hash is not idempotent,
2453 * so as we hash inodes here rather than at creation
2454 * time, we need a lock to ensure we only try
2457 static DEFINE_SPINLOCK(lock);
2459 if (inode_unhashed(inode))
2460 __insert_inode_hash(inode,
2461 inode->i_ino + inode->i_generation);
2465 fh[0] = inode->i_generation;
2466 fh[1] = inode->i_ino;
2467 fh[2] = ((__u64)inode->i_ino) >> 32;
2473 static const struct export_operations shmem_export_ops = {
2474 .get_parent = shmem_get_parent,
2475 .encode_fh = shmem_encode_fh,
2476 .fh_to_dentry = shmem_fh_to_dentry,
2479 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2482 char *this_char, *value, *rest;
2486 while (options != NULL) {
2487 this_char = options;
2490 * NUL-terminate this option: unfortunately,
2491 * mount options form a comma-separated list,
2492 * but mpol's nodelist may also contain commas.
2494 options = strchr(options, ',');
2495 if (options == NULL)
2498 if (!isdigit(*options)) {
2505 if ((value = strchr(this_char,'=')) != NULL) {
2509 "tmpfs: No value for mount option '%s'\n",
2514 if (!strcmp(this_char,"size")) {
2515 unsigned long long size;
2516 size = memparse(value,&rest);
2518 size <<= PAGE_SHIFT;
2519 size *= totalram_pages;
2525 sbinfo->max_blocks =
2526 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2527 } else if (!strcmp(this_char,"nr_blocks")) {
2528 sbinfo->max_blocks = memparse(value, &rest);
2531 } else if (!strcmp(this_char,"nr_inodes")) {
2532 sbinfo->max_inodes = memparse(value, &rest);
2535 } else if (!strcmp(this_char,"mode")) {
2538 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2541 } else if (!strcmp(this_char,"uid")) {
2544 uid = simple_strtoul(value, &rest, 0);
2547 sbinfo->uid = make_kuid(current_user_ns(), uid);
2548 if (!uid_valid(sbinfo->uid))
2550 } else if (!strcmp(this_char,"gid")) {
2553 gid = simple_strtoul(value, &rest, 0);
2556 sbinfo->gid = make_kgid(current_user_ns(), gid);
2557 if (!gid_valid(sbinfo->gid))
2559 } else if (!strcmp(this_char,"mpol")) {
2560 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2563 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2571 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2577 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2579 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2580 struct shmem_sb_info config = *sbinfo;
2581 unsigned long inodes;
2582 int error = -EINVAL;
2584 if (shmem_parse_options(data, &config, true))
2587 spin_lock(&sbinfo->stat_lock);
2588 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2589 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2591 if (config.max_inodes < inodes)
2594 * Those tests disallow limited->unlimited while any are in use;
2595 * but we must separately disallow unlimited->limited, because
2596 * in that case we have no record of how much is already in use.
2598 if (config.max_blocks && !sbinfo->max_blocks)
2600 if (config.max_inodes && !sbinfo->max_inodes)
2604 sbinfo->max_blocks = config.max_blocks;
2605 sbinfo->max_inodes = config.max_inodes;
2606 sbinfo->free_inodes = config.max_inodes - inodes;
2608 mpol_put(sbinfo->mpol);
2609 sbinfo->mpol = config.mpol; /* transfers initial ref */
2611 spin_unlock(&sbinfo->stat_lock);
2615 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2617 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2619 if (sbinfo->max_blocks != shmem_default_max_blocks())
2620 seq_printf(seq, ",size=%luk",
2621 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2622 if (sbinfo->max_inodes != shmem_default_max_inodes())
2623 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2624 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2625 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2626 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2627 seq_printf(seq, ",uid=%u",
2628 from_kuid_munged(&init_user_ns, sbinfo->uid));
2629 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2630 seq_printf(seq, ",gid=%u",
2631 from_kgid_munged(&init_user_ns, sbinfo->gid));
2632 shmem_show_mpol(seq, sbinfo->mpol);
2635 #endif /* CONFIG_TMPFS */
2637 static void shmem_put_super(struct super_block *sb)
2639 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2641 percpu_counter_destroy(&sbinfo->used_blocks);
2643 sb->s_fs_info = NULL;
2646 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2648 struct inode *inode;
2649 struct shmem_sb_info *sbinfo;
2652 /* Round up to L1_CACHE_BYTES to resist false sharing */
2653 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2654 L1_CACHE_BYTES), GFP_KERNEL);
2658 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2659 sbinfo->uid = current_fsuid();
2660 sbinfo->gid = current_fsgid();
2661 sb->s_fs_info = sbinfo;
2665 * Per default we only allow half of the physical ram per
2666 * tmpfs instance, limiting inodes to one per page of lowmem;
2667 * but the internal instance is left unlimited.
2669 if (!(sb->s_flags & MS_NOUSER)) {
2670 sbinfo->max_blocks = shmem_default_max_blocks();
2671 sbinfo->max_inodes = shmem_default_max_inodes();
2672 if (shmem_parse_options(data, sbinfo, false)) {
2677 sb->s_export_op = &shmem_export_ops;
2678 sb->s_flags |= MS_NOSEC;
2680 sb->s_flags |= MS_NOUSER;
2683 spin_lock_init(&sbinfo->stat_lock);
2684 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2686 sbinfo->free_inodes = sbinfo->max_inodes;
2688 sb->s_maxbytes = MAX_LFS_FILESIZE;
2689 sb->s_blocksize = PAGE_CACHE_SIZE;
2690 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2691 sb->s_magic = TMPFS_MAGIC;
2692 sb->s_op = &shmem_ops;
2693 sb->s_time_gran = 1;
2694 #ifdef CONFIG_TMPFS_XATTR
2695 sb->s_xattr = shmem_xattr_handlers;
2697 #ifdef CONFIG_TMPFS_POSIX_ACL
2698 sb->s_flags |= MS_POSIXACL;
2701 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2704 inode->i_uid = sbinfo->uid;
2705 inode->i_gid = sbinfo->gid;
2706 sb->s_root = d_make_root(inode);
2712 shmem_put_super(sb);
2716 static struct kmem_cache *shmem_inode_cachep;
2718 static struct inode *shmem_alloc_inode(struct super_block *sb)
2720 struct shmem_inode_info *info;
2721 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2724 return &info->vfs_inode;
2727 static void shmem_destroy_callback(struct rcu_head *head)
2729 struct inode *inode = container_of(head, struct inode, i_rcu);
2730 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2733 static void shmem_destroy_inode(struct inode *inode)
2735 if (S_ISREG(inode->i_mode))
2736 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2737 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2740 static void shmem_init_inode(void *foo)
2742 struct shmem_inode_info *info = foo;
2743 inode_init_once(&info->vfs_inode);
2746 static int shmem_init_inodecache(void)
2748 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2749 sizeof(struct shmem_inode_info),
2750 0, SLAB_PANIC, shmem_init_inode);
2754 static void shmem_destroy_inodecache(void)
2756 kmem_cache_destroy(shmem_inode_cachep);
2759 static const struct address_space_operations shmem_aops = {
2760 .writepage = shmem_writepage,
2761 .set_page_dirty = __set_page_dirty_no_writeback,
2763 .write_begin = shmem_write_begin,
2764 .write_end = shmem_write_end,
2766 .migratepage = migrate_page,
2767 .error_remove_page = generic_error_remove_page,
2770 static const struct file_operations shmem_file_operations = {
2773 .llseek = shmem_file_llseek,
2774 .read = do_sync_read,
2775 .write = do_sync_write,
2776 .aio_read = shmem_file_aio_read,
2777 .aio_write = generic_file_aio_write,
2778 .fsync = noop_fsync,
2779 .splice_read = shmem_file_splice_read,
2780 .splice_write = generic_file_splice_write,
2781 .fallocate = shmem_fallocate,
2785 static const struct inode_operations shmem_inode_operations = {
2786 .setattr = shmem_setattr,
2787 #ifdef CONFIG_TMPFS_XATTR
2788 .setxattr = shmem_setxattr,
2789 .getxattr = shmem_getxattr,
2790 .listxattr = shmem_listxattr,
2791 .removexattr = shmem_removexattr,
2795 static const struct inode_operations shmem_dir_inode_operations = {
2797 .create = shmem_create,
2798 .lookup = simple_lookup,
2800 .unlink = shmem_unlink,
2801 .symlink = shmem_symlink,
2802 .mkdir = shmem_mkdir,
2803 .rmdir = shmem_rmdir,
2804 .mknod = shmem_mknod,
2805 .rename = shmem_rename,
2807 #ifdef CONFIG_TMPFS_XATTR
2808 .setxattr = shmem_setxattr,
2809 .getxattr = shmem_getxattr,
2810 .listxattr = shmem_listxattr,
2811 .removexattr = shmem_removexattr,
2813 #ifdef CONFIG_TMPFS_POSIX_ACL
2814 .setattr = shmem_setattr,
2818 static const struct inode_operations shmem_special_inode_operations = {
2819 #ifdef CONFIG_TMPFS_XATTR
2820 .setxattr = shmem_setxattr,
2821 .getxattr = shmem_getxattr,
2822 .listxattr = shmem_listxattr,
2823 .removexattr = shmem_removexattr,
2825 #ifdef CONFIG_TMPFS_POSIX_ACL
2826 .setattr = shmem_setattr,
2830 static const struct super_operations shmem_ops = {
2831 .alloc_inode = shmem_alloc_inode,
2832 .destroy_inode = shmem_destroy_inode,
2834 .statfs = shmem_statfs,
2835 .remount_fs = shmem_remount_fs,
2836 .show_options = shmem_show_options,
2838 .evict_inode = shmem_evict_inode,
2839 .drop_inode = generic_delete_inode,
2840 .put_super = shmem_put_super,
2843 static const struct vm_operations_struct shmem_vm_ops = {
2844 .fault = shmem_fault,
2846 .set_policy = shmem_set_policy,
2847 .get_policy = shmem_get_policy,
2851 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2852 int flags, const char *dev_name, void *data)
2854 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2857 static struct file_system_type shmem_fs_type = {
2858 .owner = THIS_MODULE,
2860 .mount = shmem_mount,
2861 .kill_sb = kill_litter_super,
2864 int __init shmem_init(void)
2868 error = bdi_init(&shmem_backing_dev_info);
2872 error = shmem_init_inodecache();
2876 error = register_filesystem(&shmem_fs_type);
2878 printk(KERN_ERR "Could not register tmpfs\n");
2882 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2883 shmem_fs_type.name, NULL);
2884 if (IS_ERR(shm_mnt)) {
2885 error = PTR_ERR(shm_mnt);
2886 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2892 unregister_filesystem(&shmem_fs_type);
2894 shmem_destroy_inodecache();
2896 bdi_destroy(&shmem_backing_dev_info);
2898 shm_mnt = ERR_PTR(error);
2902 #else /* !CONFIG_SHMEM */
2905 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2907 * This is intended for small system where the benefits of the full
2908 * shmem code (swap-backed and resource-limited) are outweighed by
2909 * their complexity. On systems without swap this code should be
2910 * effectively equivalent, but much lighter weight.
2913 #include <linux/ramfs.h>
2915 static struct file_system_type shmem_fs_type = {
2917 .mount = ramfs_mount,
2918 .kill_sb = kill_litter_super,
2921 int __init shmem_init(void)
2923 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2925 shm_mnt = kern_mount(&shmem_fs_type);
2926 BUG_ON(IS_ERR(shm_mnt));
2931 int shmem_unuse(swp_entry_t swap, struct page *page)
2936 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2941 void shmem_unlock_mapping(struct address_space *mapping)
2945 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2947 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2949 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2951 #define shmem_vm_ops generic_file_vm_ops
2952 #define shmem_file_operations ramfs_file_operations
2953 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2954 #define shmem_acct_size(flags, size) 0
2955 #define shmem_unacct_size(flags, size) do {} while (0)
2957 #endif /* CONFIG_SHMEM */
2962 * shmem_file_setup - get an unlinked file living in tmpfs
2963 * @name: name for dentry (to be seen in /proc/<pid>/maps
2964 * @size: size to be set for the file
2965 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2967 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2971 struct inode *inode;
2973 struct dentry *root;
2976 if (IS_ERR(shm_mnt))
2977 return (void *)shm_mnt;
2979 if (size < 0 || size > MAX_LFS_FILESIZE)
2980 return ERR_PTR(-EINVAL);
2982 if (shmem_acct_size(flags, size))
2983 return ERR_PTR(-ENOMEM);
2987 this.len = strlen(name);
2988 this.hash = 0; /* will go */
2989 root = shm_mnt->mnt_root;
2990 path.dentry = d_alloc(root, &this);
2993 path.mnt = mntget(shm_mnt);
2996 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
3000 d_instantiate(path.dentry, inode);
3001 inode->i_size = size;
3002 clear_nlink(inode); /* It is unlinked */
3004 error = ramfs_nommu_expand_for_mapping(inode, size);
3010 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
3011 &shmem_file_operations);
3020 shmem_unacct_size(flags, size);
3021 return ERR_PTR(error);
3023 EXPORT_SYMBOL_GPL(shmem_file_setup);
3026 * shmem_zero_setup - setup a shared anonymous mapping
3027 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3029 int shmem_zero_setup(struct vm_area_struct *vma)
3032 loff_t size = vma->vm_end - vma->vm_start;
3034 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3036 return PTR_ERR(file);
3040 vma->vm_file = file;
3041 vma->vm_ops = &shmem_vm_ops;
3042 vma->vm_flags |= VM_CAN_NONLINEAR;
3047 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3048 * @mapping: the page's address_space
3049 * @index: the page index
3050 * @gfp: the page allocator flags to use if allocating
3052 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3053 * with any new page allocations done using the specified allocation flags.
3054 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3055 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3056 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3058 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3059 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3061 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3062 pgoff_t index, gfp_t gfp)
3065 struct inode *inode = mapping->host;
3069 BUG_ON(mapping->a_ops != &shmem_aops);
3070 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3072 page = ERR_PTR(error);
3078 * The tiny !SHMEM case uses ramfs without swap
3080 return read_cache_page_gfp(mapping, index, gfp);
3083 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);