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 * Sometimes, before we decide whether to proceed or to fail, we must check
268 * that an entry was not already brought back from swap by a racing thread.
270 * Checking page is not enough: by the time a SwapCache page is locked, it
271 * might be reused, and again be SwapCache, using the same swap as before.
273 static bool shmem_confirm_swap(struct address_space *mapping,
274 pgoff_t index, swp_entry_t swap)
279 item = radix_tree_lookup(&mapping->page_tree, index);
281 return item == swp_to_radix_entry(swap);
285 * Like add_to_page_cache_locked, but error if expected item has gone.
287 static int shmem_add_to_page_cache(struct page *page,
288 struct address_space *mapping,
289 pgoff_t index, gfp_t gfp, void *expected)
293 VM_BUG_ON(!PageLocked(page));
294 VM_BUG_ON(!PageSwapBacked(page));
297 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
299 page_cache_get(page);
300 page->mapping = mapping;
303 spin_lock_irq(&mapping->tree_lock);
305 error = radix_tree_insert(&mapping->page_tree,
308 error = shmem_radix_tree_replace(mapping, index,
312 __inc_zone_page_state(page, NR_FILE_PAGES);
313 __inc_zone_page_state(page, NR_SHMEM);
314 spin_unlock_irq(&mapping->tree_lock);
316 page->mapping = NULL;
317 spin_unlock_irq(&mapping->tree_lock);
318 page_cache_release(page);
321 radix_tree_preload_end();
324 mem_cgroup_uncharge_cache_page(page);
329 * Like delete_from_page_cache, but substitutes swap for page.
331 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
333 struct address_space *mapping = page->mapping;
336 spin_lock_irq(&mapping->tree_lock);
337 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
338 page->mapping = NULL;
340 __dec_zone_page_state(page, NR_FILE_PAGES);
341 __dec_zone_page_state(page, NR_SHMEM);
342 spin_unlock_irq(&mapping->tree_lock);
343 page_cache_release(page);
348 * Like find_get_pages, but collecting swap entries as well as pages.
350 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
351 pgoff_t start, unsigned int nr_pages,
352 struct page **pages, pgoff_t *indices)
356 unsigned int nr_found;
360 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
361 (void ***)pages, indices, start, nr_pages);
363 for (i = 0; i < nr_found; i++) {
366 page = radix_tree_deref_slot((void **)pages[i]);
369 if (radix_tree_exception(page)) {
370 if (radix_tree_deref_retry(page))
373 * Otherwise, we must be storing a swap entry
374 * here as an exceptional entry: so return it
375 * without attempting to raise page count.
379 if (!page_cache_get_speculative(page))
382 /* Has the page moved? */
383 if (unlikely(page != *((void **)pages[i]))) {
384 page_cache_release(page);
388 indices[ret] = indices[i];
392 if (unlikely(!ret && nr_found))
399 * Remove swap entry from radix tree, free the swap and its page cache.
401 static int shmem_free_swap(struct address_space *mapping,
402 pgoff_t index, void *radswap)
406 spin_lock_irq(&mapping->tree_lock);
407 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
408 spin_unlock_irq(&mapping->tree_lock);
410 free_swap_and_cache(radix_to_swp_entry(radswap));
415 * Pagevec may contain swap entries, so shuffle up pages before releasing.
417 static void shmem_deswap_pagevec(struct pagevec *pvec)
421 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
422 struct page *page = pvec->pages[i];
423 if (!radix_tree_exceptional_entry(page))
424 pvec->pages[j++] = page;
430 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
432 void shmem_unlock_mapping(struct address_space *mapping)
435 pgoff_t indices[PAGEVEC_SIZE];
438 pagevec_init(&pvec, 0);
440 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
442 while (!mapping_unevictable(mapping)) {
444 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
445 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
447 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
448 PAGEVEC_SIZE, pvec.pages, indices);
451 index = indices[pvec.nr - 1] + 1;
452 shmem_deswap_pagevec(&pvec);
453 check_move_unevictable_pages(pvec.pages, pvec.nr);
454 pagevec_release(&pvec);
460 * Remove range of pages and swap entries from radix tree, and free them.
461 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
463 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
466 struct address_space *mapping = inode->i_mapping;
467 struct shmem_inode_info *info = SHMEM_I(inode);
468 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
469 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
470 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
471 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
473 pgoff_t indices[PAGEVEC_SIZE];
474 long nr_swaps_freed = 0;
479 end = -1; /* unsigned, so actually very big */
481 pagevec_init(&pvec, 0);
483 while (index < end) {
484 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
485 min(end - index, (pgoff_t)PAGEVEC_SIZE),
486 pvec.pages, indices);
489 mem_cgroup_uncharge_start();
490 for (i = 0; i < pagevec_count(&pvec); i++) {
491 struct page *page = pvec.pages[i];
497 if (radix_tree_exceptional_entry(page)) {
500 nr_swaps_freed += !shmem_free_swap(mapping,
505 if (!trylock_page(page))
507 if (!unfalloc || !PageUptodate(page)) {
508 if (page->mapping == mapping) {
509 VM_BUG_ON(PageWriteback(page));
510 truncate_inode_page(mapping, page);
515 shmem_deswap_pagevec(&pvec);
516 pagevec_release(&pvec);
517 mem_cgroup_uncharge_end();
523 struct page *page = NULL;
524 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
526 unsigned int top = PAGE_CACHE_SIZE;
531 zero_user_segment(page, partial_start, top);
532 set_page_dirty(page);
534 page_cache_release(page);
538 struct page *page = NULL;
539 shmem_getpage(inode, end, &page, SGP_READ, NULL);
541 zero_user_segment(page, 0, partial_end);
542 set_page_dirty(page);
544 page_cache_release(page);
553 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
554 min(end - index, (pgoff_t)PAGEVEC_SIZE),
555 pvec.pages, indices);
557 if (index == start || unfalloc)
562 if ((index == start || unfalloc) && indices[0] >= end) {
563 shmem_deswap_pagevec(&pvec);
564 pagevec_release(&pvec);
567 mem_cgroup_uncharge_start();
568 for (i = 0; i < pagevec_count(&pvec); i++) {
569 struct page *page = pvec.pages[i];
575 if (radix_tree_exceptional_entry(page)) {
578 nr_swaps_freed += !shmem_free_swap(mapping,
584 if (!unfalloc || !PageUptodate(page)) {
585 if (page->mapping == mapping) {
586 VM_BUG_ON(PageWriteback(page));
587 truncate_inode_page(mapping, page);
592 shmem_deswap_pagevec(&pvec);
593 pagevec_release(&pvec);
594 mem_cgroup_uncharge_end();
598 spin_lock(&info->lock);
599 info->swapped -= nr_swaps_freed;
600 shmem_recalc_inode(inode);
601 spin_unlock(&info->lock);
604 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
606 shmem_undo_range(inode, lstart, lend, false);
607 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
609 EXPORT_SYMBOL_GPL(shmem_truncate_range);
611 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
613 struct inode *inode = dentry->d_inode;
616 error = inode_change_ok(inode, attr);
620 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
621 loff_t oldsize = inode->i_size;
622 loff_t newsize = attr->ia_size;
624 if (newsize != oldsize) {
625 i_size_write(inode, newsize);
626 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
628 if (newsize < oldsize) {
629 loff_t holebegin = round_up(newsize, PAGE_SIZE);
630 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
631 shmem_truncate_range(inode, newsize, (loff_t)-1);
632 /* unmap again to remove racily COWed private pages */
633 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
637 setattr_copy(inode, attr);
638 #ifdef CONFIG_TMPFS_POSIX_ACL
639 if (attr->ia_valid & ATTR_MODE)
640 error = generic_acl_chmod(inode);
645 static void shmem_evict_inode(struct inode *inode)
647 struct shmem_inode_info *info = SHMEM_I(inode);
648 struct shmem_xattr *xattr, *nxattr;
650 if (inode->i_mapping->a_ops == &shmem_aops) {
651 shmem_unacct_size(info->flags, inode->i_size);
653 shmem_truncate_range(inode, 0, (loff_t)-1);
654 if (!list_empty(&info->swaplist)) {
655 mutex_lock(&shmem_swaplist_mutex);
656 list_del_init(&info->swaplist);
657 mutex_unlock(&shmem_swaplist_mutex);
660 kfree(info->symlink);
662 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
666 BUG_ON(inode->i_blocks);
667 shmem_free_inode(inode->i_sb);
672 * If swap found in inode, free it and move page from swapcache to filecache.
674 static int shmem_unuse_inode(struct shmem_inode_info *info,
675 swp_entry_t swap, struct page **pagep)
677 struct address_space *mapping = info->vfs_inode.i_mapping;
683 radswap = swp_to_radix_entry(swap);
684 index = radix_tree_locate_item(&mapping->page_tree, radswap);
689 * Move _head_ to start search for next from here.
690 * But be careful: shmem_evict_inode checks list_empty without taking
691 * mutex, and there's an instant in list_move_tail when info->swaplist
692 * would appear empty, if it were the only one on shmem_swaplist.
694 if (shmem_swaplist.next != &info->swaplist)
695 list_move_tail(&shmem_swaplist, &info->swaplist);
697 gfp = mapping_gfp_mask(mapping);
698 if (shmem_should_replace_page(*pagep, gfp)) {
699 mutex_unlock(&shmem_swaplist_mutex);
700 error = shmem_replace_page(pagep, gfp, info, index);
701 mutex_lock(&shmem_swaplist_mutex);
703 * We needed to drop mutex to make that restrictive page
704 * allocation, but the inode might have been freed while we
705 * dropped it: although a racing shmem_evict_inode() cannot
706 * complete without emptying the radix_tree, our page lock
707 * on this swapcache page is not enough to prevent that -
708 * free_swap_and_cache() of our swap entry will only
709 * trylock_page(), removing swap from radix_tree whatever.
711 * We must not proceed to shmem_add_to_page_cache() if the
712 * inode has been freed, but of course we cannot rely on
713 * inode or mapping or info to check that. However, we can
714 * safely check if our swap entry is still in use (and here
715 * it can't have got reused for another page): if it's still
716 * in use, then the inode cannot have been freed yet, and we
717 * can safely proceed (if it's no longer in use, that tells
718 * nothing about the inode, but we don't need to unuse swap).
720 if (!page_swapcount(*pagep))
725 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
726 * but also to hold up shmem_evict_inode(): so inode cannot be freed
727 * beneath us (pagelock doesn't help until the page is in pagecache).
730 error = shmem_add_to_page_cache(*pagep, mapping, index,
731 GFP_NOWAIT, radswap);
732 if (error != -ENOMEM) {
734 * Truncation and eviction use free_swap_and_cache(), which
735 * only does trylock page: if we raced, best clean up here.
737 delete_from_swap_cache(*pagep);
738 set_page_dirty(*pagep);
740 spin_lock(&info->lock);
742 spin_unlock(&info->lock);
745 error = 1; /* not an error, but entry was found */
751 * Search through swapped inodes to find and replace swap by page.
753 int shmem_unuse(swp_entry_t swap, struct page *page)
755 struct list_head *this, *next;
756 struct shmem_inode_info *info;
761 * There's a faint possibility that swap page was replaced before
762 * caller locked it: caller will come back later with the right page.
764 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
768 * Charge page using GFP_KERNEL while we can wait, before taking
769 * the shmem_swaplist_mutex which might hold up shmem_writepage().
770 * Charged back to the user (not to caller) when swap account is used.
772 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
775 /* No radix_tree_preload: swap entry keeps a place for page in tree */
777 mutex_lock(&shmem_swaplist_mutex);
778 list_for_each_safe(this, next, &shmem_swaplist) {
779 info = list_entry(this, struct shmem_inode_info, swaplist);
781 found = shmem_unuse_inode(info, swap, &page);
783 list_del_init(&info->swaplist);
788 mutex_unlock(&shmem_swaplist_mutex);
794 page_cache_release(page);
799 * Move the page from the page cache to the swap cache.
801 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
803 struct shmem_inode_info *info;
804 struct address_space *mapping;
809 BUG_ON(!PageLocked(page));
810 mapping = page->mapping;
812 inode = mapping->host;
813 info = SHMEM_I(inode);
814 if (info->flags & VM_LOCKED)
816 if (!total_swap_pages)
820 * shmem_backing_dev_info's capabilities prevent regular writeback or
821 * sync from ever calling shmem_writepage; but a stacking filesystem
822 * might use ->writepage of its underlying filesystem, in which case
823 * tmpfs should write out to swap only in response to memory pressure,
824 * and not for the writeback threads or sync.
826 if (!wbc->for_reclaim) {
827 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
832 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
833 * value into swapfile.c, the only way we can correctly account for a
834 * fallocated page arriving here is now to initialize it and write it.
836 * That's okay for a page already fallocated earlier, but if we have
837 * not yet completed the fallocation, then (a) we want to keep track
838 * of this page in case we have to undo it, and (b) it may not be a
839 * good idea to continue anyway, once we're pushing into swap. So
840 * reactivate the page, and let shmem_fallocate() quit when too many.
842 if (!PageUptodate(page)) {
843 if (inode->i_private) {
844 struct shmem_falloc *shmem_falloc;
845 spin_lock(&inode->i_lock);
846 shmem_falloc = inode->i_private;
848 index >= shmem_falloc->start &&
849 index < shmem_falloc->next)
850 shmem_falloc->nr_unswapped++;
853 spin_unlock(&inode->i_lock);
857 clear_highpage(page);
858 flush_dcache_page(page);
859 SetPageUptodate(page);
862 swap = get_swap_page();
867 * Add inode to shmem_unuse()'s list of swapped-out inodes,
868 * if it's not already there. Do it now before the page is
869 * moved to swap cache, when its pagelock no longer protects
870 * the inode from eviction. But don't unlock the mutex until
871 * we've incremented swapped, because shmem_unuse_inode() will
872 * prune a !swapped inode from the swaplist under this mutex.
874 mutex_lock(&shmem_swaplist_mutex);
875 if (list_empty(&info->swaplist))
876 list_add_tail(&info->swaplist, &shmem_swaplist);
878 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
879 swap_shmem_alloc(swap);
880 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
882 spin_lock(&info->lock);
884 shmem_recalc_inode(inode);
885 spin_unlock(&info->lock);
887 mutex_unlock(&shmem_swaplist_mutex);
888 BUG_ON(page_mapped(page));
889 swap_writepage(page, wbc);
893 mutex_unlock(&shmem_swaplist_mutex);
894 swapcache_free(swap, NULL);
896 set_page_dirty(page);
897 if (wbc->for_reclaim)
898 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
905 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
909 if (!mpol || mpol->mode == MPOL_DEFAULT)
910 return; /* show nothing */
912 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
914 seq_printf(seq, ",mpol=%s", buffer);
917 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
919 struct mempolicy *mpol = NULL;
921 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
924 spin_unlock(&sbinfo->stat_lock);
928 #endif /* CONFIG_TMPFS */
930 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
931 struct shmem_inode_info *info, pgoff_t index)
933 struct mempolicy mpol, *spol;
934 struct vm_area_struct pvma;
936 spol = mpol_cond_copy(&mpol,
937 mpol_shared_policy_lookup(&info->policy, index));
939 /* Create a pseudo vma that just contains the policy */
941 pvma.vm_pgoff = index;
943 pvma.vm_policy = spol;
944 return swapin_readahead(swap, gfp, &pvma, 0);
947 static struct page *shmem_alloc_page(gfp_t gfp,
948 struct shmem_inode_info *info, pgoff_t index)
950 struct vm_area_struct pvma;
952 /* Create a pseudo vma that just contains the policy */
954 pvma.vm_pgoff = index;
956 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
959 * alloc_page_vma() will drop the shared policy reference
961 return alloc_page_vma(gfp, &pvma, 0);
963 #else /* !CONFIG_NUMA */
965 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
968 #endif /* CONFIG_TMPFS */
970 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
971 struct shmem_inode_info *info, pgoff_t index)
973 return swapin_readahead(swap, gfp, NULL, 0);
976 static inline struct page *shmem_alloc_page(gfp_t gfp,
977 struct shmem_inode_info *info, pgoff_t index)
979 return alloc_page(gfp);
981 #endif /* CONFIG_NUMA */
983 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
984 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
991 * When a page is moved from swapcache to shmem filecache (either by the
992 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
993 * shmem_unuse_inode()), it may have been read in earlier from swap, in
994 * ignorance of the mapping it belongs to. If that mapping has special
995 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
996 * we may need to copy to a suitable page before moving to filecache.
998 * In a future release, this may well be extended to respect cpuset and
999 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1000 * but for now it is a simple matter of zone.
1002 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1004 return page_zonenum(page) > gfp_zone(gfp);
1007 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1008 struct shmem_inode_info *info, pgoff_t index)
1010 struct page *oldpage, *newpage;
1011 struct address_space *swap_mapping;
1016 swap_index = page_private(oldpage);
1017 swap_mapping = page_mapping(oldpage);
1020 * We have arrived here because our zones are constrained, so don't
1021 * limit chance of success by further cpuset and node constraints.
1023 gfp &= ~GFP_CONSTRAINT_MASK;
1024 newpage = shmem_alloc_page(gfp, info, index);
1028 page_cache_get(newpage);
1029 copy_highpage(newpage, oldpage);
1030 flush_dcache_page(newpage);
1032 __set_page_locked(newpage);
1033 SetPageUptodate(newpage);
1034 SetPageSwapBacked(newpage);
1035 set_page_private(newpage, swap_index);
1036 SetPageSwapCache(newpage);
1039 * Our caller will very soon move newpage out of swapcache, but it's
1040 * a nice clean interface for us to replace oldpage by newpage there.
1042 spin_lock_irq(&swap_mapping->tree_lock);
1043 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1046 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1047 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1049 spin_unlock_irq(&swap_mapping->tree_lock);
1051 if (unlikely(error)) {
1053 * Is this possible? I think not, now that our callers check
1054 * both PageSwapCache and page_private after getting page lock;
1055 * but be defensive. Reverse old to newpage for clear and free.
1059 mem_cgroup_replace_page_cache(oldpage, newpage);
1060 lru_cache_add_anon(newpage);
1064 ClearPageSwapCache(oldpage);
1065 set_page_private(oldpage, 0);
1067 unlock_page(oldpage);
1068 page_cache_release(oldpage);
1069 page_cache_release(oldpage);
1074 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1076 * If we allocate a new one we do not mark it dirty. That's up to the
1077 * vm. If we swap it in we mark it dirty since we also free the swap
1078 * entry since a page cannot live in both the swap and page cache
1080 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1081 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1083 struct address_space *mapping = inode->i_mapping;
1084 struct shmem_inode_info *info;
1085 struct shmem_sb_info *sbinfo;
1092 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1096 page = find_lock_page(mapping, index);
1097 if (radix_tree_exceptional_entry(page)) {
1098 swap = radix_to_swp_entry(page);
1102 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1103 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1108 /* fallocated page? */
1109 if (page && !PageUptodate(page)) {
1110 if (sgp != SGP_READ)
1113 page_cache_release(page);
1116 if (page || (sgp == SGP_READ && !swap.val)) {
1122 * Fast cache lookup did not find it:
1123 * bring it back from swap or allocate.
1125 info = SHMEM_I(inode);
1126 sbinfo = SHMEM_SB(inode->i_sb);
1129 /* Look it up and read it in.. */
1130 page = lookup_swap_cache(swap);
1132 /* here we actually do the io */
1134 *fault_type |= VM_FAULT_MAJOR;
1135 page = shmem_swapin(swap, gfp, info, index);
1142 /* We have to do this with page locked to prevent races */
1144 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1145 !shmem_confirm_swap(mapping, index, swap)) {
1146 error = -EEXIST; /* try again */
1149 if (!PageUptodate(page)) {
1153 wait_on_page_writeback(page);
1155 if (shmem_should_replace_page(page, gfp)) {
1156 error = shmem_replace_page(&page, gfp, info, index);
1161 error = mem_cgroup_cache_charge(page, current->mm,
1162 gfp & GFP_RECLAIM_MASK);
1164 error = shmem_add_to_page_cache(page, mapping, index,
1165 gfp, swp_to_radix_entry(swap));
1166 /* We already confirmed swap, and make no allocation */
1172 spin_lock(&info->lock);
1174 shmem_recalc_inode(inode);
1175 spin_unlock(&info->lock);
1177 delete_from_swap_cache(page);
1178 set_page_dirty(page);
1182 if (shmem_acct_block(info->flags)) {
1186 if (sbinfo->max_blocks) {
1187 if (percpu_counter_compare(&sbinfo->used_blocks,
1188 sbinfo->max_blocks) >= 0) {
1192 percpu_counter_inc(&sbinfo->used_blocks);
1195 page = shmem_alloc_page(gfp, info, index);
1201 SetPageSwapBacked(page);
1202 __set_page_locked(page);
1203 error = mem_cgroup_cache_charge(page, current->mm,
1204 gfp & GFP_RECLAIM_MASK);
1206 error = shmem_add_to_page_cache(page, mapping, index,
1210 lru_cache_add_anon(page);
1212 spin_lock(&info->lock);
1214 inode->i_blocks += BLOCKS_PER_PAGE;
1215 shmem_recalc_inode(inode);
1216 spin_unlock(&info->lock);
1220 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1222 if (sgp == SGP_FALLOC)
1226 * Let SGP_WRITE caller clear ends if write does not fill page;
1227 * but SGP_FALLOC on a page fallocated earlier must initialize
1228 * it now, lest undo on failure cancel our earlier guarantee.
1230 if (sgp != SGP_WRITE) {
1231 clear_highpage(page);
1232 flush_dcache_page(page);
1233 SetPageUptodate(page);
1235 if (sgp == SGP_DIRTY)
1236 set_page_dirty(page);
1239 /* Perhaps the file has been truncated since we checked */
1240 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1241 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1255 info = SHMEM_I(inode);
1256 ClearPageDirty(page);
1257 delete_from_page_cache(page);
1258 spin_lock(&info->lock);
1260 inode->i_blocks -= BLOCKS_PER_PAGE;
1261 spin_unlock(&info->lock);
1263 sbinfo = SHMEM_SB(inode->i_sb);
1264 if (sbinfo->max_blocks)
1265 percpu_counter_add(&sbinfo->used_blocks, -1);
1267 shmem_unacct_blocks(info->flags, 1);
1269 if (swap.val && error != -EINVAL &&
1270 !shmem_confirm_swap(mapping, index, swap))
1275 page_cache_release(page);
1277 if (error == -ENOSPC && !once++) {
1278 info = SHMEM_I(inode);
1279 spin_lock(&info->lock);
1280 shmem_recalc_inode(inode);
1281 spin_unlock(&info->lock);
1284 if (error == -EEXIST) /* from above or from radix_tree_insert */
1289 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1291 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1293 int ret = VM_FAULT_LOCKED;
1295 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1297 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1299 if (ret & VM_FAULT_MAJOR) {
1300 count_vm_event(PGMAJFAULT);
1301 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1307 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1309 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1310 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1313 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1316 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1319 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1320 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1324 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1326 struct inode *inode = file->f_path.dentry->d_inode;
1327 struct shmem_inode_info *info = SHMEM_I(inode);
1328 int retval = -ENOMEM;
1330 spin_lock(&info->lock);
1331 if (lock && !(info->flags & VM_LOCKED)) {
1332 if (!user_shm_lock(inode->i_size, user))
1334 info->flags |= VM_LOCKED;
1335 mapping_set_unevictable(file->f_mapping);
1337 if (!lock && (info->flags & VM_LOCKED) && user) {
1338 user_shm_unlock(inode->i_size, user);
1339 info->flags &= ~VM_LOCKED;
1340 mapping_clear_unevictable(file->f_mapping);
1345 spin_unlock(&info->lock);
1349 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1351 file_accessed(file);
1352 vma->vm_ops = &shmem_vm_ops;
1353 vma->vm_flags |= VM_CAN_NONLINEAR;
1357 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1358 umode_t mode, dev_t dev, unsigned long flags)
1360 struct inode *inode;
1361 struct shmem_inode_info *info;
1362 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1364 if (shmem_reserve_inode(sb))
1367 inode = new_inode(sb);
1369 inode->i_ino = get_next_ino();
1370 inode_init_owner(inode, dir, mode);
1371 inode->i_blocks = 0;
1372 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1373 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1374 inode->i_generation = get_seconds();
1375 info = SHMEM_I(inode);
1376 memset(info, 0, (char *)inode - (char *)info);
1377 spin_lock_init(&info->lock);
1378 info->flags = flags & VM_NORESERVE;
1379 INIT_LIST_HEAD(&info->swaplist);
1380 INIT_LIST_HEAD(&info->xattr_list);
1381 cache_no_acl(inode);
1383 switch (mode & S_IFMT) {
1385 inode->i_op = &shmem_special_inode_operations;
1386 init_special_inode(inode, mode, dev);
1389 inode->i_mapping->a_ops = &shmem_aops;
1390 inode->i_op = &shmem_inode_operations;
1391 inode->i_fop = &shmem_file_operations;
1392 mpol_shared_policy_init(&info->policy,
1393 shmem_get_sbmpol(sbinfo));
1397 /* Some things misbehave if size == 0 on a directory */
1398 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1399 inode->i_op = &shmem_dir_inode_operations;
1400 inode->i_fop = &simple_dir_operations;
1404 * Must not load anything in the rbtree,
1405 * mpol_free_shared_policy will not be called.
1407 mpol_shared_policy_init(&info->policy, NULL);
1411 shmem_free_inode(sb);
1416 static const struct inode_operations shmem_symlink_inode_operations;
1417 static const struct inode_operations shmem_short_symlink_operations;
1419 #ifdef CONFIG_TMPFS_XATTR
1420 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1422 #define shmem_initxattrs NULL
1426 shmem_write_begin(struct file *file, struct address_space *mapping,
1427 loff_t pos, unsigned len, unsigned flags,
1428 struct page **pagep, void **fsdata)
1430 struct inode *inode = mapping->host;
1431 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1432 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1436 shmem_write_end(struct file *file, struct address_space *mapping,
1437 loff_t pos, unsigned len, unsigned copied,
1438 struct page *page, void *fsdata)
1440 struct inode *inode = mapping->host;
1442 if (pos + copied > inode->i_size)
1443 i_size_write(inode, pos + copied);
1445 if (!PageUptodate(page)) {
1446 if (copied < PAGE_CACHE_SIZE) {
1447 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1448 zero_user_segments(page, 0, from,
1449 from + copied, PAGE_CACHE_SIZE);
1451 SetPageUptodate(page);
1453 set_page_dirty(page);
1455 page_cache_release(page);
1460 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1462 struct inode *inode = filp->f_path.dentry->d_inode;
1463 struct address_space *mapping = inode->i_mapping;
1465 unsigned long offset;
1466 enum sgp_type sgp = SGP_READ;
1469 * Might this read be for a stacking filesystem? Then when reading
1470 * holes of a sparse file, we actually need to allocate those pages,
1471 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1473 if (segment_eq(get_fs(), KERNEL_DS))
1476 index = *ppos >> PAGE_CACHE_SHIFT;
1477 offset = *ppos & ~PAGE_CACHE_MASK;
1480 struct page *page = NULL;
1482 unsigned long nr, ret;
1483 loff_t i_size = i_size_read(inode);
1485 end_index = i_size >> PAGE_CACHE_SHIFT;
1486 if (index > end_index)
1488 if (index == end_index) {
1489 nr = i_size & ~PAGE_CACHE_MASK;
1494 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1496 if (desc->error == -EINVAL)
1504 * We must evaluate after, since reads (unlike writes)
1505 * are called without i_mutex protection against truncate
1507 nr = PAGE_CACHE_SIZE;
1508 i_size = i_size_read(inode);
1509 end_index = i_size >> PAGE_CACHE_SHIFT;
1510 if (index == end_index) {
1511 nr = i_size & ~PAGE_CACHE_MASK;
1514 page_cache_release(page);
1522 * If users can be writing to this page using arbitrary
1523 * virtual addresses, take care about potential aliasing
1524 * before reading the page on the kernel side.
1526 if (mapping_writably_mapped(mapping))
1527 flush_dcache_page(page);
1529 * Mark the page accessed if we read the beginning.
1532 mark_page_accessed(page);
1534 page = ZERO_PAGE(0);
1535 page_cache_get(page);
1539 * Ok, we have the page, and it's up-to-date, so
1540 * now we can copy it to user space...
1542 * The actor routine returns how many bytes were actually used..
1543 * NOTE! This may not be the same as how much of a user buffer
1544 * we filled up (we may be padding etc), so we can only update
1545 * "pos" here (the actor routine has to update the user buffer
1546 * pointers and the remaining count).
1548 ret = actor(desc, page, offset, nr);
1550 index += offset >> PAGE_CACHE_SHIFT;
1551 offset &= ~PAGE_CACHE_MASK;
1553 page_cache_release(page);
1554 if (ret != nr || !desc->count)
1560 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1561 file_accessed(filp);
1564 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1565 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1567 struct file *filp = iocb->ki_filp;
1571 loff_t *ppos = &iocb->ki_pos;
1573 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1577 for (seg = 0; seg < nr_segs; seg++) {
1578 read_descriptor_t desc;
1581 desc.arg.buf = iov[seg].iov_base;
1582 desc.count = iov[seg].iov_len;
1583 if (desc.count == 0)
1586 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1587 retval += desc.written;
1589 retval = retval ?: desc.error;
1598 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1599 struct pipe_inode_info *pipe, size_t len,
1602 struct address_space *mapping = in->f_mapping;
1603 struct inode *inode = mapping->host;
1604 unsigned int loff, nr_pages, req_pages;
1605 struct page *pages[PIPE_DEF_BUFFERS];
1606 struct partial_page partial[PIPE_DEF_BUFFERS];
1608 pgoff_t index, end_index;
1611 struct splice_pipe_desc spd = {
1614 .nr_pages_max = PIPE_DEF_BUFFERS,
1616 .ops = &page_cache_pipe_buf_ops,
1617 .spd_release = spd_release_page,
1620 isize = i_size_read(inode);
1621 if (unlikely(*ppos >= isize))
1624 left = isize - *ppos;
1625 if (unlikely(left < len))
1628 if (splice_grow_spd(pipe, &spd))
1631 index = *ppos >> PAGE_CACHE_SHIFT;
1632 loff = *ppos & ~PAGE_CACHE_MASK;
1633 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1634 nr_pages = min(req_pages, pipe->buffers);
1636 spd.nr_pages = find_get_pages_contig(mapping, index,
1637 nr_pages, spd.pages);
1638 index += spd.nr_pages;
1641 while (spd.nr_pages < nr_pages) {
1642 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1646 spd.pages[spd.nr_pages++] = page;
1650 index = *ppos >> PAGE_CACHE_SHIFT;
1651 nr_pages = spd.nr_pages;
1654 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1655 unsigned int this_len;
1660 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1661 page = spd.pages[page_nr];
1663 if (!PageUptodate(page) || page->mapping != mapping) {
1664 error = shmem_getpage(inode, index, &page,
1669 page_cache_release(spd.pages[page_nr]);
1670 spd.pages[page_nr] = page;
1673 isize = i_size_read(inode);
1674 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1675 if (unlikely(!isize || index > end_index))
1678 if (end_index == index) {
1681 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1685 this_len = min(this_len, plen - loff);
1689 spd.partial[page_nr].offset = loff;
1690 spd.partial[page_nr].len = this_len;
1697 while (page_nr < nr_pages)
1698 page_cache_release(spd.pages[page_nr++]);
1701 error = splice_to_pipe(pipe, &spd);
1703 splice_shrink_spd(&spd);
1712 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1715 struct inode *inode = file->f_path.dentry->d_inode;
1716 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1717 struct shmem_falloc shmem_falloc;
1718 pgoff_t start, index, end;
1721 mutex_lock(&inode->i_mutex);
1723 if (mode & FALLOC_FL_PUNCH_HOLE) {
1724 struct address_space *mapping = file->f_mapping;
1725 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1726 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1728 if ((u64)unmap_end > (u64)unmap_start)
1729 unmap_mapping_range(mapping, unmap_start,
1730 1 + unmap_end - unmap_start, 0);
1731 shmem_truncate_range(inode, offset, offset + len - 1);
1732 /* No need to unmap again: hole-punching leaves COWed pages */
1737 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1738 error = inode_newsize_ok(inode, offset + len);
1742 start = offset >> PAGE_CACHE_SHIFT;
1743 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1744 /* Try to avoid a swapstorm if len is impossible to satisfy */
1745 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1750 shmem_falloc.start = start;
1751 shmem_falloc.next = start;
1752 shmem_falloc.nr_falloced = 0;
1753 shmem_falloc.nr_unswapped = 0;
1754 spin_lock(&inode->i_lock);
1755 inode->i_private = &shmem_falloc;
1756 spin_unlock(&inode->i_lock);
1758 for (index = start; index < end; index++) {
1762 * Good, the fallocate(2) manpage permits EINTR: we may have
1763 * been interrupted because we are using up too much memory.
1765 if (signal_pending(current))
1767 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1770 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1773 /* Remove the !PageUptodate pages we added */
1774 shmem_undo_range(inode,
1775 (loff_t)start << PAGE_CACHE_SHIFT,
1776 (loff_t)index << PAGE_CACHE_SHIFT, true);
1781 * Inform shmem_writepage() how far we have reached.
1782 * No need for lock or barrier: we have the page lock.
1784 shmem_falloc.next++;
1785 if (!PageUptodate(page))
1786 shmem_falloc.nr_falloced++;
1789 * If !PageUptodate, leave it that way so that freeable pages
1790 * can be recognized if we need to rollback on error later.
1791 * But set_page_dirty so that memory pressure will swap rather
1792 * than free the pages we are allocating (and SGP_CACHE pages
1793 * might still be clean: we now need to mark those dirty too).
1795 set_page_dirty(page);
1797 page_cache_release(page);
1801 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1802 i_size_write(inode, offset + len);
1803 inode->i_ctime = CURRENT_TIME;
1805 spin_lock(&inode->i_lock);
1806 inode->i_private = NULL;
1807 spin_unlock(&inode->i_lock);
1809 mutex_unlock(&inode->i_mutex);
1813 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1815 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1817 buf->f_type = TMPFS_MAGIC;
1818 buf->f_bsize = PAGE_CACHE_SIZE;
1819 buf->f_namelen = NAME_MAX;
1820 if (sbinfo->max_blocks) {
1821 buf->f_blocks = sbinfo->max_blocks;
1823 buf->f_bfree = sbinfo->max_blocks -
1824 percpu_counter_sum(&sbinfo->used_blocks);
1826 if (sbinfo->max_inodes) {
1827 buf->f_files = sbinfo->max_inodes;
1828 buf->f_ffree = sbinfo->free_inodes;
1830 /* else leave those fields 0 like simple_statfs */
1835 * File creation. Allocate an inode, and we're done..
1838 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1840 struct inode *inode;
1841 int error = -ENOSPC;
1843 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1845 error = security_inode_init_security(inode, dir,
1847 shmem_initxattrs, NULL);
1849 if (error != -EOPNOTSUPP) {
1854 #ifdef CONFIG_TMPFS_POSIX_ACL
1855 error = generic_acl_init(inode, dir);
1863 dir->i_size += BOGO_DIRENT_SIZE;
1864 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1865 d_instantiate(dentry, inode);
1866 dget(dentry); /* Extra count - pin the dentry in core */
1871 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1875 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1881 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1882 struct nameidata *nd)
1884 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1890 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1892 struct inode *inode = old_dentry->d_inode;
1896 * No ordinary (disk based) filesystem counts links as inodes;
1897 * but each new link needs a new dentry, pinning lowmem, and
1898 * tmpfs dentries cannot be pruned until they are unlinked.
1900 ret = shmem_reserve_inode(inode->i_sb);
1904 dir->i_size += BOGO_DIRENT_SIZE;
1905 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1907 ihold(inode); /* New dentry reference */
1908 dget(dentry); /* Extra pinning count for the created dentry */
1909 d_instantiate(dentry, inode);
1914 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1916 struct inode *inode = dentry->d_inode;
1918 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1919 shmem_free_inode(inode->i_sb);
1921 dir->i_size -= BOGO_DIRENT_SIZE;
1922 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1924 dput(dentry); /* Undo the count from "create" - this does all the work */
1928 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1930 if (!simple_empty(dentry))
1933 drop_nlink(dentry->d_inode);
1935 return shmem_unlink(dir, dentry);
1939 * The VFS layer already does all the dentry stuff for rename,
1940 * we just have to decrement the usage count for the target if
1941 * it exists so that the VFS layer correctly free's it when it
1944 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1946 struct inode *inode = old_dentry->d_inode;
1947 int they_are_dirs = S_ISDIR(inode->i_mode);
1949 if (!simple_empty(new_dentry))
1952 if (new_dentry->d_inode) {
1953 (void) shmem_unlink(new_dir, new_dentry);
1955 drop_nlink(old_dir);
1956 } else if (they_are_dirs) {
1957 drop_nlink(old_dir);
1961 old_dir->i_size -= BOGO_DIRENT_SIZE;
1962 new_dir->i_size += BOGO_DIRENT_SIZE;
1963 old_dir->i_ctime = old_dir->i_mtime =
1964 new_dir->i_ctime = new_dir->i_mtime =
1965 inode->i_ctime = CURRENT_TIME;
1969 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1973 struct inode *inode;
1976 struct shmem_inode_info *info;
1978 len = strlen(symname) + 1;
1979 if (len > PAGE_CACHE_SIZE)
1980 return -ENAMETOOLONG;
1982 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1986 error = security_inode_init_security(inode, dir, &dentry->d_name,
1987 shmem_initxattrs, NULL);
1989 if (error != -EOPNOTSUPP) {
1996 info = SHMEM_I(inode);
1997 inode->i_size = len-1;
1998 if (len <= SHORT_SYMLINK_LEN) {
1999 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2000 if (!info->symlink) {
2004 inode->i_op = &shmem_short_symlink_operations;
2006 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2011 inode->i_mapping->a_ops = &shmem_aops;
2012 inode->i_op = &shmem_symlink_inode_operations;
2013 kaddr = kmap_atomic(page);
2014 memcpy(kaddr, symname, len);
2015 kunmap_atomic(kaddr);
2016 SetPageUptodate(page);
2017 set_page_dirty(page);
2019 page_cache_release(page);
2021 dir->i_size += BOGO_DIRENT_SIZE;
2022 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2023 d_instantiate(dentry, inode);
2028 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2030 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2034 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2036 struct page *page = NULL;
2037 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2038 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2044 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2046 if (!IS_ERR(nd_get_link(nd))) {
2047 struct page *page = cookie;
2049 mark_page_accessed(page);
2050 page_cache_release(page);
2054 #ifdef CONFIG_TMPFS_XATTR
2056 * Superblocks without xattr inode operations may get some security.* xattr
2057 * support from the LSM "for free". As soon as we have any other xattrs
2058 * like ACLs, we also need to implement the security.* handlers at
2059 * filesystem level, though.
2063 * Allocate new xattr and copy in the value; but leave the name to callers.
2065 static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2067 struct shmem_xattr *new_xattr;
2071 len = sizeof(*new_xattr) + size;
2072 if (len <= sizeof(*new_xattr))
2075 new_xattr = kmalloc(len, GFP_KERNEL);
2079 new_xattr->size = size;
2080 memcpy(new_xattr->value, value, size);
2085 * Callback for security_inode_init_security() for acquiring xattrs.
2087 static int shmem_initxattrs(struct inode *inode,
2088 const struct xattr *xattr_array,
2091 struct shmem_inode_info *info = SHMEM_I(inode);
2092 const struct xattr *xattr;
2093 struct shmem_xattr *new_xattr;
2096 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2097 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2101 len = strlen(xattr->name) + 1;
2102 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2104 if (!new_xattr->name) {
2109 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2110 XATTR_SECURITY_PREFIX_LEN);
2111 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2114 spin_lock(&info->lock);
2115 list_add(&new_xattr->list, &info->xattr_list);
2116 spin_unlock(&info->lock);
2122 static int shmem_xattr_get(struct dentry *dentry, const char *name,
2123 void *buffer, size_t size)
2125 struct shmem_inode_info *info;
2126 struct shmem_xattr *xattr;
2129 info = SHMEM_I(dentry->d_inode);
2131 spin_lock(&info->lock);
2132 list_for_each_entry(xattr, &info->xattr_list, list) {
2133 if (strcmp(name, xattr->name))
2138 if (size < xattr->size)
2141 memcpy(buffer, xattr->value, xattr->size);
2145 spin_unlock(&info->lock);
2149 static int shmem_xattr_set(struct inode *inode, const char *name,
2150 const void *value, size_t size, int flags)
2152 struct shmem_inode_info *info = SHMEM_I(inode);
2153 struct shmem_xattr *xattr;
2154 struct shmem_xattr *new_xattr = NULL;
2157 /* value == NULL means remove */
2159 new_xattr = shmem_xattr_alloc(value, size);
2163 new_xattr->name = kstrdup(name, GFP_KERNEL);
2164 if (!new_xattr->name) {
2170 spin_lock(&info->lock);
2171 list_for_each_entry(xattr, &info->xattr_list, list) {
2172 if (!strcmp(name, xattr->name)) {
2173 if (flags & XATTR_CREATE) {
2176 } else if (new_xattr) {
2177 list_replace(&xattr->list, &new_xattr->list);
2179 list_del(&xattr->list);
2184 if (flags & XATTR_REPLACE) {
2188 list_add(&new_xattr->list, &info->xattr_list);
2192 spin_unlock(&info->lock);
2199 static const struct xattr_handler *shmem_xattr_handlers[] = {
2200 #ifdef CONFIG_TMPFS_POSIX_ACL
2201 &generic_acl_access_handler,
2202 &generic_acl_default_handler,
2207 static int shmem_xattr_validate(const char *name)
2209 struct { const char *prefix; size_t len; } arr[] = {
2210 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2211 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2215 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2216 size_t preflen = arr[i].len;
2217 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2226 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2227 void *buffer, size_t size)
2232 * If this is a request for a synthetic attribute in the system.*
2233 * namespace use the generic infrastructure to resolve a handler
2234 * for it via sb->s_xattr.
2236 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2237 return generic_getxattr(dentry, name, buffer, size);
2239 err = shmem_xattr_validate(name);
2243 return shmem_xattr_get(dentry, name, buffer, size);
2246 static int shmem_setxattr(struct dentry *dentry, const char *name,
2247 const void *value, size_t size, int flags)
2252 * If this is a request for a synthetic attribute in the system.*
2253 * namespace use the generic infrastructure to resolve a handler
2254 * for it via sb->s_xattr.
2256 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2257 return generic_setxattr(dentry, name, value, size, flags);
2259 err = shmem_xattr_validate(name);
2264 value = ""; /* empty EA, do not remove */
2266 return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
2270 static int shmem_removexattr(struct dentry *dentry, const char *name)
2275 * If this is a request for a synthetic attribute in the system.*
2276 * namespace use the generic infrastructure to resolve a handler
2277 * for it via sb->s_xattr.
2279 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2280 return generic_removexattr(dentry, name);
2282 err = shmem_xattr_validate(name);
2286 return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
2289 static bool xattr_is_trusted(const char *name)
2291 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2294 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2296 bool trusted = capable(CAP_SYS_ADMIN);
2297 struct shmem_xattr *xattr;
2298 struct shmem_inode_info *info;
2301 info = SHMEM_I(dentry->d_inode);
2303 spin_lock(&info->lock);
2304 list_for_each_entry(xattr, &info->xattr_list, list) {
2307 /* skip "trusted." attributes for unprivileged callers */
2308 if (!trusted && xattr_is_trusted(xattr->name))
2311 len = strlen(xattr->name) + 1;
2318 memcpy(buffer, xattr->name, len);
2322 spin_unlock(&info->lock);
2326 #endif /* CONFIG_TMPFS_XATTR */
2328 static const struct inode_operations shmem_short_symlink_operations = {
2329 .readlink = generic_readlink,
2330 .follow_link = shmem_follow_short_symlink,
2331 #ifdef CONFIG_TMPFS_XATTR
2332 .setxattr = shmem_setxattr,
2333 .getxattr = shmem_getxattr,
2334 .listxattr = shmem_listxattr,
2335 .removexattr = shmem_removexattr,
2339 static const struct inode_operations shmem_symlink_inode_operations = {
2340 .readlink = generic_readlink,
2341 .follow_link = shmem_follow_link,
2342 .put_link = shmem_put_link,
2343 #ifdef CONFIG_TMPFS_XATTR
2344 .setxattr = shmem_setxattr,
2345 .getxattr = shmem_getxattr,
2346 .listxattr = shmem_listxattr,
2347 .removexattr = shmem_removexattr,
2351 static struct dentry *shmem_get_parent(struct dentry *child)
2353 return ERR_PTR(-ESTALE);
2356 static int shmem_match(struct inode *ino, void *vfh)
2360 inum = (inum << 32) | fh[1];
2361 return ino->i_ino == inum && fh[0] == ino->i_generation;
2364 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2365 struct fid *fid, int fh_len, int fh_type)
2367 struct inode *inode;
2368 struct dentry *dentry = NULL;
2369 u64 inum = fid->raw[2];
2370 inum = (inum << 32) | fid->raw[1];
2375 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2376 shmem_match, fid->raw);
2378 dentry = d_find_alias(inode);
2385 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2386 struct inode *parent)
2393 if (inode_unhashed(inode)) {
2394 /* Unfortunately insert_inode_hash is not idempotent,
2395 * so as we hash inodes here rather than at creation
2396 * time, we need a lock to ensure we only try
2399 static DEFINE_SPINLOCK(lock);
2401 if (inode_unhashed(inode))
2402 __insert_inode_hash(inode,
2403 inode->i_ino + inode->i_generation);
2407 fh[0] = inode->i_generation;
2408 fh[1] = inode->i_ino;
2409 fh[2] = ((__u64)inode->i_ino) >> 32;
2415 static const struct export_operations shmem_export_ops = {
2416 .get_parent = shmem_get_parent,
2417 .encode_fh = shmem_encode_fh,
2418 .fh_to_dentry = shmem_fh_to_dentry,
2421 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2424 char *this_char, *value, *rest;
2428 while (options != NULL) {
2429 this_char = options;
2432 * NUL-terminate this option: unfortunately,
2433 * mount options form a comma-separated list,
2434 * but mpol's nodelist may also contain commas.
2436 options = strchr(options, ',');
2437 if (options == NULL)
2440 if (!isdigit(*options)) {
2447 if ((value = strchr(this_char,'=')) != NULL) {
2451 "tmpfs: No value for mount option '%s'\n",
2456 if (!strcmp(this_char,"size")) {
2457 unsigned long long size;
2458 size = memparse(value,&rest);
2460 size <<= PAGE_SHIFT;
2461 size *= totalram_pages;
2467 sbinfo->max_blocks =
2468 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2469 } else if (!strcmp(this_char,"nr_blocks")) {
2470 sbinfo->max_blocks = memparse(value, &rest);
2473 } else if (!strcmp(this_char,"nr_inodes")) {
2474 sbinfo->max_inodes = memparse(value, &rest);
2477 } else if (!strcmp(this_char,"mode")) {
2480 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2483 } else if (!strcmp(this_char,"uid")) {
2486 uid = simple_strtoul(value, &rest, 0);
2489 sbinfo->uid = make_kuid(current_user_ns(), uid);
2490 if (!uid_valid(sbinfo->uid))
2492 } else if (!strcmp(this_char,"gid")) {
2495 gid = simple_strtoul(value, &rest, 0);
2498 sbinfo->gid = make_kgid(current_user_ns(), gid);
2499 if (!gid_valid(sbinfo->gid))
2501 } else if (!strcmp(this_char,"mpol")) {
2502 if (mpol_parse_str(value, &sbinfo->mpol, 1))
2505 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2513 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2519 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2521 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2522 struct shmem_sb_info config = *sbinfo;
2523 unsigned long inodes;
2524 int error = -EINVAL;
2526 if (shmem_parse_options(data, &config, true))
2529 spin_lock(&sbinfo->stat_lock);
2530 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2531 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2533 if (config.max_inodes < inodes)
2536 * Those tests disallow limited->unlimited while any are in use;
2537 * but we must separately disallow unlimited->limited, because
2538 * in that case we have no record of how much is already in use.
2540 if (config.max_blocks && !sbinfo->max_blocks)
2542 if (config.max_inodes && !sbinfo->max_inodes)
2546 sbinfo->max_blocks = config.max_blocks;
2547 sbinfo->max_inodes = config.max_inodes;
2548 sbinfo->free_inodes = config.max_inodes - inodes;
2550 mpol_put(sbinfo->mpol);
2551 sbinfo->mpol = config.mpol; /* transfers initial ref */
2553 spin_unlock(&sbinfo->stat_lock);
2557 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2559 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2561 if (sbinfo->max_blocks != shmem_default_max_blocks())
2562 seq_printf(seq, ",size=%luk",
2563 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2564 if (sbinfo->max_inodes != shmem_default_max_inodes())
2565 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2566 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2567 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2568 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2569 seq_printf(seq, ",uid=%u",
2570 from_kuid_munged(&init_user_ns, sbinfo->uid));
2571 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2572 seq_printf(seq, ",gid=%u",
2573 from_kgid_munged(&init_user_ns, sbinfo->gid));
2574 shmem_show_mpol(seq, sbinfo->mpol);
2577 #endif /* CONFIG_TMPFS */
2579 static void shmem_put_super(struct super_block *sb)
2581 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2583 percpu_counter_destroy(&sbinfo->used_blocks);
2585 sb->s_fs_info = NULL;
2588 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2590 struct inode *inode;
2591 struct shmem_sb_info *sbinfo;
2594 /* Round up to L1_CACHE_BYTES to resist false sharing */
2595 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2596 L1_CACHE_BYTES), GFP_KERNEL);
2600 sbinfo->mode = S_IRWXUGO | S_ISVTX;
2601 sbinfo->uid = current_fsuid();
2602 sbinfo->gid = current_fsgid();
2603 sb->s_fs_info = sbinfo;
2607 * Per default we only allow half of the physical ram per
2608 * tmpfs instance, limiting inodes to one per page of lowmem;
2609 * but the internal instance is left unlimited.
2611 if (!(sb->s_flags & MS_NOUSER)) {
2612 sbinfo->max_blocks = shmem_default_max_blocks();
2613 sbinfo->max_inodes = shmem_default_max_inodes();
2614 if (shmem_parse_options(data, sbinfo, false)) {
2619 sb->s_export_op = &shmem_export_ops;
2620 sb->s_flags |= MS_NOSEC;
2622 sb->s_flags |= MS_NOUSER;
2625 spin_lock_init(&sbinfo->stat_lock);
2626 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2628 sbinfo->free_inodes = sbinfo->max_inodes;
2630 sb->s_maxbytes = MAX_LFS_FILESIZE;
2631 sb->s_blocksize = PAGE_CACHE_SIZE;
2632 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2633 sb->s_magic = TMPFS_MAGIC;
2634 sb->s_op = &shmem_ops;
2635 sb->s_time_gran = 1;
2636 #ifdef CONFIG_TMPFS_XATTR
2637 sb->s_xattr = shmem_xattr_handlers;
2639 #ifdef CONFIG_TMPFS_POSIX_ACL
2640 sb->s_flags |= MS_POSIXACL;
2643 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2646 inode->i_uid = sbinfo->uid;
2647 inode->i_gid = sbinfo->gid;
2648 sb->s_root = d_make_root(inode);
2654 shmem_put_super(sb);
2658 static struct kmem_cache *shmem_inode_cachep;
2660 static struct inode *shmem_alloc_inode(struct super_block *sb)
2662 struct shmem_inode_info *info;
2663 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2666 return &info->vfs_inode;
2669 static void shmem_destroy_callback(struct rcu_head *head)
2671 struct inode *inode = container_of(head, struct inode, i_rcu);
2672 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2675 static void shmem_destroy_inode(struct inode *inode)
2677 if (S_ISREG(inode->i_mode))
2678 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2679 call_rcu(&inode->i_rcu, shmem_destroy_callback);
2682 static void shmem_init_inode(void *foo)
2684 struct shmem_inode_info *info = foo;
2685 inode_init_once(&info->vfs_inode);
2688 static int shmem_init_inodecache(void)
2690 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2691 sizeof(struct shmem_inode_info),
2692 0, SLAB_PANIC, shmem_init_inode);
2696 static void shmem_destroy_inodecache(void)
2698 kmem_cache_destroy(shmem_inode_cachep);
2701 static const struct address_space_operations shmem_aops = {
2702 .writepage = shmem_writepage,
2703 .set_page_dirty = __set_page_dirty_no_writeback,
2705 .write_begin = shmem_write_begin,
2706 .write_end = shmem_write_end,
2708 .migratepage = migrate_page,
2709 .error_remove_page = generic_error_remove_page,
2712 static const struct file_operations shmem_file_operations = {
2715 .llseek = generic_file_llseek,
2716 .read = do_sync_read,
2717 .write = do_sync_write,
2718 .aio_read = shmem_file_aio_read,
2719 .aio_write = generic_file_aio_write,
2720 .fsync = noop_fsync,
2721 .splice_read = shmem_file_splice_read,
2722 .splice_write = generic_file_splice_write,
2723 .fallocate = shmem_fallocate,
2727 static const struct inode_operations shmem_inode_operations = {
2728 .setattr = shmem_setattr,
2729 #ifdef CONFIG_TMPFS_XATTR
2730 .setxattr = shmem_setxattr,
2731 .getxattr = shmem_getxattr,
2732 .listxattr = shmem_listxattr,
2733 .removexattr = shmem_removexattr,
2737 static const struct inode_operations shmem_dir_inode_operations = {
2739 .create = shmem_create,
2740 .lookup = simple_lookup,
2742 .unlink = shmem_unlink,
2743 .symlink = shmem_symlink,
2744 .mkdir = shmem_mkdir,
2745 .rmdir = shmem_rmdir,
2746 .mknod = shmem_mknod,
2747 .rename = shmem_rename,
2749 #ifdef CONFIG_TMPFS_XATTR
2750 .setxattr = shmem_setxattr,
2751 .getxattr = shmem_getxattr,
2752 .listxattr = shmem_listxattr,
2753 .removexattr = shmem_removexattr,
2755 #ifdef CONFIG_TMPFS_POSIX_ACL
2756 .setattr = shmem_setattr,
2760 static const struct inode_operations shmem_special_inode_operations = {
2761 #ifdef CONFIG_TMPFS_XATTR
2762 .setxattr = shmem_setxattr,
2763 .getxattr = shmem_getxattr,
2764 .listxattr = shmem_listxattr,
2765 .removexattr = shmem_removexattr,
2767 #ifdef CONFIG_TMPFS_POSIX_ACL
2768 .setattr = shmem_setattr,
2772 static const struct super_operations shmem_ops = {
2773 .alloc_inode = shmem_alloc_inode,
2774 .destroy_inode = shmem_destroy_inode,
2776 .statfs = shmem_statfs,
2777 .remount_fs = shmem_remount_fs,
2778 .show_options = shmem_show_options,
2780 .evict_inode = shmem_evict_inode,
2781 .drop_inode = generic_delete_inode,
2782 .put_super = shmem_put_super,
2785 static const struct vm_operations_struct shmem_vm_ops = {
2786 .fault = shmem_fault,
2788 .set_policy = shmem_set_policy,
2789 .get_policy = shmem_get_policy,
2793 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2794 int flags, const char *dev_name, void *data)
2796 return mount_nodev(fs_type, flags, data, shmem_fill_super);
2799 static struct file_system_type shmem_fs_type = {
2800 .owner = THIS_MODULE,
2802 .mount = shmem_mount,
2803 .kill_sb = kill_litter_super,
2806 int __init shmem_init(void)
2810 error = bdi_init(&shmem_backing_dev_info);
2814 error = shmem_init_inodecache();
2818 error = register_filesystem(&shmem_fs_type);
2820 printk(KERN_ERR "Could not register tmpfs\n");
2824 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2825 shmem_fs_type.name, NULL);
2826 if (IS_ERR(shm_mnt)) {
2827 error = PTR_ERR(shm_mnt);
2828 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2834 unregister_filesystem(&shmem_fs_type);
2836 shmem_destroy_inodecache();
2838 bdi_destroy(&shmem_backing_dev_info);
2840 shm_mnt = ERR_PTR(error);
2844 #else /* !CONFIG_SHMEM */
2847 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2849 * This is intended for small system where the benefits of the full
2850 * shmem code (swap-backed and resource-limited) are outweighed by
2851 * their complexity. On systems without swap this code should be
2852 * effectively equivalent, but much lighter weight.
2855 #include <linux/ramfs.h>
2857 static struct file_system_type shmem_fs_type = {
2859 .mount = ramfs_mount,
2860 .kill_sb = kill_litter_super,
2863 int __init shmem_init(void)
2865 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2867 shm_mnt = kern_mount(&shmem_fs_type);
2868 BUG_ON(IS_ERR(shm_mnt));
2873 int shmem_unuse(swp_entry_t swap, struct page *page)
2878 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2883 void shmem_unlock_mapping(struct address_space *mapping)
2887 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2889 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2891 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2893 #define shmem_vm_ops generic_file_vm_ops
2894 #define shmem_file_operations ramfs_file_operations
2895 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
2896 #define shmem_acct_size(flags, size) 0
2897 #define shmem_unacct_size(flags, size) do {} while (0)
2899 #endif /* CONFIG_SHMEM */
2904 * shmem_file_setup - get an unlinked file living in tmpfs
2905 * @name: name for dentry (to be seen in /proc/<pid>/maps
2906 * @size: size to be set for the file
2907 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2909 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2913 struct inode *inode;
2915 struct dentry *root;
2918 if (IS_ERR(shm_mnt))
2919 return (void *)shm_mnt;
2921 if (size < 0 || size > MAX_LFS_FILESIZE)
2922 return ERR_PTR(-EINVAL);
2924 if (shmem_acct_size(flags, size))
2925 return ERR_PTR(-ENOMEM);
2929 this.len = strlen(name);
2930 this.hash = 0; /* will go */
2931 root = shm_mnt->mnt_root;
2932 path.dentry = d_alloc(root, &this);
2935 path.mnt = mntget(shm_mnt);
2938 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2942 d_instantiate(path.dentry, inode);
2943 inode->i_size = size;
2944 clear_nlink(inode); /* It is unlinked */
2946 error = ramfs_nommu_expand_for_mapping(inode, size);
2952 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2953 &shmem_file_operations);
2962 shmem_unacct_size(flags, size);
2963 return ERR_PTR(error);
2965 EXPORT_SYMBOL_GPL(shmem_file_setup);
2968 * shmem_zero_setup - setup a shared anonymous mapping
2969 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2971 int shmem_zero_setup(struct vm_area_struct *vma)
2974 loff_t size = vma->vm_end - vma->vm_start;
2976 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2978 return PTR_ERR(file);
2982 vma->vm_file = file;
2983 vma->vm_ops = &shmem_vm_ops;
2984 vma->vm_flags |= VM_CAN_NONLINEAR;
2989 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2990 * @mapping: the page's address_space
2991 * @index: the page index
2992 * @gfp: the page allocator flags to use if allocating
2994 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2995 * with any new page allocations done using the specified allocation flags.
2996 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2997 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2998 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3000 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3001 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3003 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3004 pgoff_t index, gfp_t gfp)
3007 struct inode *inode = mapping->host;
3011 BUG_ON(mapping->a_ops != &shmem_aops);
3012 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3014 page = ERR_PTR(error);
3020 * The tiny !SHMEM case uses ramfs without swap
3022 return read_cache_page_gfp(mapping, index, gfp);
3025 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);