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/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
35 #include <linux/khugepaged.h>
37 static struct vfsmount *shm_mnt;
41 * This virtual memory filesystem is heavily based on the ramfs. It
42 * extends ramfs by the ability to use swap and honor resource limits
43 * which makes it a completely usable filesystem.
46 #include <linux/xattr.h>
47 #include <linux/exportfs.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/mman.h>
51 #include <linux/string.h>
52 #include <linux/slab.h>
53 #include <linux/backing-dev.h>
54 #include <linux/shmem_fs.h>
55 #include <linux/writeback.h>
56 #include <linux/blkdev.h>
57 #include <linux/pagevec.h>
58 #include <linux/percpu_counter.h>
59 #include <linux/falloc.h>
60 #include <linux/splice.h>
61 #include <linux/security.h>
62 #include <linux/swapops.h>
63 #include <linux/mempolicy.h>
64 #include <linux/namei.h>
65 #include <linux/ctype.h>
66 #include <linux/migrate.h>
67 #include <linux/highmem.h>
68 #include <linux/seq_file.h>
69 #include <linux/magic.h>
70 #include <linux/syscalls.h>
71 #include <linux/fcntl.h>
72 #include <uapi/linux/memfd.h>
73 #include <linux/rmap.h>
75 #include <linux/uaccess.h>
76 #include <asm/pgtable.h>
80 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
81 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
83 /* Pretend that each entry is of this size in directory's i_size */
84 #define BOGO_DIRENT_SIZE 20
86 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
87 #define SHORT_SYMLINK_LEN 128
90 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
91 * inode->i_private (with i_mutex making sure that it has only one user at
92 * a time): we would prefer not to enlarge the shmem inode just for that.
95 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
96 pgoff_t start; /* start of range currently being fallocated */
97 pgoff_t next; /* the next page offset to be fallocated */
98 pgoff_t nr_falloced; /* how many new pages have been fallocated */
99 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
103 static unsigned long shmem_default_max_blocks(void)
105 return totalram_pages / 2;
108 static unsigned long shmem_default_max_inodes(void)
110 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
114 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
115 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
116 struct shmem_inode_info *info, pgoff_t index);
117 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
118 struct page **pagep, enum sgp_type sgp,
119 gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
121 int shmem_getpage(struct inode *inode, pgoff_t index,
122 struct page **pagep, enum sgp_type sgp)
124 return shmem_getpage_gfp(inode, index, pagep, sgp,
125 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
128 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
130 return sb->s_fs_info;
134 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
135 * for shared memory and for shared anonymous (/dev/zero) mappings
136 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
137 * consistent with the pre-accounting of private mappings ...
139 static inline int shmem_acct_size(unsigned long flags, loff_t size)
141 return (flags & VM_NORESERVE) ?
142 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
145 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
147 if (!(flags & VM_NORESERVE))
148 vm_unacct_memory(VM_ACCT(size));
151 static inline int shmem_reacct_size(unsigned long flags,
152 loff_t oldsize, loff_t newsize)
154 if (!(flags & VM_NORESERVE)) {
155 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
156 return security_vm_enough_memory_mm(current->mm,
157 VM_ACCT(newsize) - VM_ACCT(oldsize));
158 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
159 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
165 * ... whereas tmpfs objects are accounted incrementally as
166 * pages are allocated, in order to allow large sparse files.
167 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
168 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
170 static inline int shmem_acct_block(unsigned long flags, long pages)
172 if (!(flags & VM_NORESERVE))
175 return security_vm_enough_memory_mm(current->mm,
176 pages * VM_ACCT(PAGE_SIZE));
179 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
181 if (flags & VM_NORESERVE)
182 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
185 static const struct super_operations shmem_ops;
186 static const struct address_space_operations shmem_aops;
187 static const struct file_operations shmem_file_operations;
188 static const struct inode_operations shmem_inode_operations;
189 static const struct inode_operations shmem_dir_inode_operations;
190 static const struct inode_operations shmem_special_inode_operations;
191 static const struct vm_operations_struct shmem_vm_ops;
192 static struct file_system_type shmem_fs_type;
194 bool vma_is_shmem(struct vm_area_struct *vma)
196 return vma->vm_ops == &shmem_vm_ops;
199 static LIST_HEAD(shmem_swaplist);
200 static DEFINE_MUTEX(shmem_swaplist_mutex);
202 static int shmem_reserve_inode(struct super_block *sb)
204 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
205 if (sbinfo->max_inodes) {
206 spin_lock(&sbinfo->stat_lock);
207 if (!sbinfo->free_inodes) {
208 spin_unlock(&sbinfo->stat_lock);
211 sbinfo->free_inodes--;
212 spin_unlock(&sbinfo->stat_lock);
217 static void shmem_free_inode(struct super_block *sb)
219 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
220 if (sbinfo->max_inodes) {
221 spin_lock(&sbinfo->stat_lock);
222 sbinfo->free_inodes++;
223 spin_unlock(&sbinfo->stat_lock);
228 * shmem_recalc_inode - recalculate the block usage of an inode
229 * @inode: inode to recalc
231 * We have to calculate the free blocks since the mm can drop
232 * undirtied hole pages behind our back.
234 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
235 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
237 * It has to be called with the spinlock held.
239 static void shmem_recalc_inode(struct inode *inode)
241 struct shmem_inode_info *info = SHMEM_I(inode);
244 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
246 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
247 if (sbinfo->max_blocks)
248 percpu_counter_add(&sbinfo->used_blocks, -freed);
249 info->alloced -= freed;
250 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
251 shmem_unacct_blocks(info->flags, freed);
255 bool shmem_charge(struct inode *inode, long pages)
257 struct shmem_inode_info *info = SHMEM_I(inode);
258 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
261 if (shmem_acct_block(info->flags, pages))
263 spin_lock_irqsave(&info->lock, flags);
264 info->alloced += pages;
265 inode->i_blocks += pages * BLOCKS_PER_PAGE;
266 shmem_recalc_inode(inode);
267 spin_unlock_irqrestore(&info->lock, flags);
268 inode->i_mapping->nrpages += pages;
270 if (!sbinfo->max_blocks)
272 if (percpu_counter_compare(&sbinfo->used_blocks,
273 sbinfo->max_blocks - pages) > 0) {
274 inode->i_mapping->nrpages -= pages;
275 spin_lock_irqsave(&info->lock, flags);
276 info->alloced -= pages;
277 shmem_recalc_inode(inode);
278 spin_unlock_irqrestore(&info->lock, flags);
279 shmem_unacct_blocks(info->flags, pages);
282 percpu_counter_add(&sbinfo->used_blocks, pages);
286 void shmem_uncharge(struct inode *inode, long pages)
288 struct shmem_inode_info *info = SHMEM_I(inode);
289 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
292 spin_lock_irqsave(&info->lock, flags);
293 info->alloced -= pages;
294 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
295 shmem_recalc_inode(inode);
296 spin_unlock_irqrestore(&info->lock, flags);
298 if (sbinfo->max_blocks)
299 percpu_counter_sub(&sbinfo->used_blocks, pages);
300 shmem_unacct_blocks(info->flags, pages);
304 * Replace item expected in radix tree by a new item, while holding tree lock.
306 static int shmem_radix_tree_replace(struct address_space *mapping,
307 pgoff_t index, void *expected, void *replacement)
309 struct radix_tree_node *node;
313 VM_BUG_ON(!expected);
314 VM_BUG_ON(!replacement);
315 item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot);
318 if (item != expected)
320 __radix_tree_replace(&mapping->page_tree, node, pslot,
321 replacement, NULL, NULL);
326 * Sometimes, before we decide whether to proceed or to fail, we must check
327 * that an entry was not already brought back from swap by a racing thread.
329 * Checking page is not enough: by the time a SwapCache page is locked, it
330 * might be reused, and again be SwapCache, using the same swap as before.
332 static bool shmem_confirm_swap(struct address_space *mapping,
333 pgoff_t index, swp_entry_t swap)
338 item = radix_tree_lookup(&mapping->page_tree, index);
340 return item == swp_to_radix_entry(swap);
344 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
347 * disables huge pages for the mount;
349 * enables huge pages for the mount;
350 * SHMEM_HUGE_WITHIN_SIZE:
351 * only allocate huge pages if the page will be fully within i_size,
352 * also respect fadvise()/madvise() hints;
354 * only allocate huge pages if requested with fadvise()/madvise();
357 #define SHMEM_HUGE_NEVER 0
358 #define SHMEM_HUGE_ALWAYS 1
359 #define SHMEM_HUGE_WITHIN_SIZE 2
360 #define SHMEM_HUGE_ADVISE 3
364 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
367 * disables huge on shm_mnt and all mounts, for emergency use;
369 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
372 #define SHMEM_HUGE_DENY (-1)
373 #define SHMEM_HUGE_FORCE (-2)
375 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
376 /* ifdef here to avoid bloating shmem.o when not necessary */
378 int shmem_huge __read_mostly;
380 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
381 static int shmem_parse_huge(const char *str)
383 if (!strcmp(str, "never"))
384 return SHMEM_HUGE_NEVER;
385 if (!strcmp(str, "always"))
386 return SHMEM_HUGE_ALWAYS;
387 if (!strcmp(str, "within_size"))
388 return SHMEM_HUGE_WITHIN_SIZE;
389 if (!strcmp(str, "advise"))
390 return SHMEM_HUGE_ADVISE;
391 if (!strcmp(str, "deny"))
392 return SHMEM_HUGE_DENY;
393 if (!strcmp(str, "force"))
394 return SHMEM_HUGE_FORCE;
398 static const char *shmem_format_huge(int huge)
401 case SHMEM_HUGE_NEVER:
403 case SHMEM_HUGE_ALWAYS:
405 case SHMEM_HUGE_WITHIN_SIZE:
406 return "within_size";
407 case SHMEM_HUGE_ADVISE:
409 case SHMEM_HUGE_DENY:
411 case SHMEM_HUGE_FORCE:
420 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
421 struct shrink_control *sc, unsigned long nr_to_split)
423 LIST_HEAD(list), *pos, *next;
424 LIST_HEAD(to_remove);
426 struct shmem_inode_info *info;
428 unsigned long batch = sc ? sc->nr_to_scan : 128;
429 int removed = 0, split = 0;
431 if (list_empty(&sbinfo->shrinklist))
434 spin_lock(&sbinfo->shrinklist_lock);
435 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
436 info = list_entry(pos, struct shmem_inode_info, shrinklist);
439 inode = igrab(&info->vfs_inode);
441 /* inode is about to be evicted */
443 list_del_init(&info->shrinklist);
448 /* Check if there's anything to gain */
449 if (round_up(inode->i_size, PAGE_SIZE) ==
450 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
451 list_move(&info->shrinklist, &to_remove);
456 list_move(&info->shrinklist, &list);
461 spin_unlock(&sbinfo->shrinklist_lock);
463 list_for_each_safe(pos, next, &to_remove) {
464 info = list_entry(pos, struct shmem_inode_info, shrinklist);
465 inode = &info->vfs_inode;
466 list_del_init(&info->shrinklist);
470 list_for_each_safe(pos, next, &list) {
473 info = list_entry(pos, struct shmem_inode_info, shrinklist);
474 inode = &info->vfs_inode;
476 if (nr_to_split && split >= nr_to_split) {
481 page = find_lock_page(inode->i_mapping,
482 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
486 if (!PageTransHuge(page)) {
492 ret = split_huge_page(page);
497 /* split failed: leave it on the list */
504 list_del_init(&info->shrinklist);
509 spin_lock(&sbinfo->shrinklist_lock);
510 list_splice_tail(&list, &sbinfo->shrinklist);
511 sbinfo->shrinklist_len -= removed;
512 spin_unlock(&sbinfo->shrinklist_lock);
517 static long shmem_unused_huge_scan(struct super_block *sb,
518 struct shrink_control *sc)
520 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
522 if (!READ_ONCE(sbinfo->shrinklist_len))
525 return shmem_unused_huge_shrink(sbinfo, sc, 0);
528 static long shmem_unused_huge_count(struct super_block *sb,
529 struct shrink_control *sc)
531 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
532 return READ_ONCE(sbinfo->shrinklist_len);
534 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
536 #define shmem_huge SHMEM_HUGE_DENY
538 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
539 struct shrink_control *sc, unsigned long nr_to_split)
543 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
546 * Like add_to_page_cache_locked, but error if expected item has gone.
548 static int shmem_add_to_page_cache(struct page *page,
549 struct address_space *mapping,
550 pgoff_t index, void *expected)
552 int error, nr = hpage_nr_pages(page);
554 VM_BUG_ON_PAGE(PageTail(page), page);
555 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
556 VM_BUG_ON_PAGE(!PageLocked(page), page);
557 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
558 VM_BUG_ON(expected && PageTransHuge(page));
560 page_ref_add(page, nr);
561 page->mapping = mapping;
564 spin_lock_irq(&mapping->tree_lock);
565 if (PageTransHuge(page)) {
566 void __rcu **results;
571 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
572 &results, &idx, index, 1) &&
573 idx < index + HPAGE_PMD_NR) {
578 for (i = 0; i < HPAGE_PMD_NR; i++) {
579 error = radix_tree_insert(&mapping->page_tree,
580 index + i, page + i);
583 count_vm_event(THP_FILE_ALLOC);
585 } else if (!expected) {
586 error = radix_tree_insert(&mapping->page_tree, index, page);
588 error = shmem_radix_tree_replace(mapping, index, expected,
593 mapping->nrpages += nr;
594 if (PageTransHuge(page))
595 __inc_node_page_state(page, NR_SHMEM_THPS);
596 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
597 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
598 spin_unlock_irq(&mapping->tree_lock);
600 page->mapping = NULL;
601 spin_unlock_irq(&mapping->tree_lock);
602 page_ref_sub(page, nr);
608 * Like delete_from_page_cache, but substitutes swap for page.
610 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
612 struct address_space *mapping = page->mapping;
615 VM_BUG_ON_PAGE(PageCompound(page), page);
617 spin_lock_irq(&mapping->tree_lock);
618 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
619 page->mapping = NULL;
621 __dec_node_page_state(page, NR_FILE_PAGES);
622 __dec_node_page_state(page, NR_SHMEM);
623 spin_unlock_irq(&mapping->tree_lock);
629 * Remove swap entry from radix tree, free the swap and its page cache.
631 static int shmem_free_swap(struct address_space *mapping,
632 pgoff_t index, void *radswap)
636 spin_lock_irq(&mapping->tree_lock);
637 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
638 spin_unlock_irq(&mapping->tree_lock);
641 free_swap_and_cache(radix_to_swp_entry(radswap));
646 * Determine (in bytes) how many of the shmem object's pages mapped by the
647 * given offsets are swapped out.
649 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
650 * as long as the inode doesn't go away and racy results are not a problem.
652 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
653 pgoff_t start, pgoff_t end)
655 struct radix_tree_iter iter;
658 unsigned long swapped = 0;
662 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
663 if (iter.index >= end)
666 page = radix_tree_deref_slot(slot);
668 if (radix_tree_deref_retry(page)) {
669 slot = radix_tree_iter_retry(&iter);
673 if (radix_tree_exceptional_entry(page))
676 if (need_resched()) {
677 slot = radix_tree_iter_resume(slot, &iter);
684 return swapped << PAGE_SHIFT;
688 * Determine (in bytes) how many of the shmem object's pages mapped by the
689 * given vma is swapped out.
691 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
692 * as long as the inode doesn't go away and racy results are not a problem.
694 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
696 struct inode *inode = file_inode(vma->vm_file);
697 struct shmem_inode_info *info = SHMEM_I(inode);
698 struct address_space *mapping = inode->i_mapping;
699 unsigned long swapped;
701 /* Be careful as we don't hold info->lock */
702 swapped = READ_ONCE(info->swapped);
705 * The easier cases are when the shmem object has nothing in swap, or
706 * the vma maps it whole. Then we can simply use the stats that we
712 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
713 return swapped << PAGE_SHIFT;
715 /* Here comes the more involved part */
716 return shmem_partial_swap_usage(mapping,
717 linear_page_index(vma, vma->vm_start),
718 linear_page_index(vma, vma->vm_end));
722 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
724 void shmem_unlock_mapping(struct address_space *mapping)
727 pgoff_t indices[PAGEVEC_SIZE];
730 pagevec_init(&pvec, 0);
732 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
734 while (!mapping_unevictable(mapping)) {
736 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
737 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
739 pvec.nr = find_get_entries(mapping, index,
740 PAGEVEC_SIZE, pvec.pages, indices);
743 index = indices[pvec.nr - 1] + 1;
744 pagevec_remove_exceptionals(&pvec);
745 check_move_unevictable_pages(pvec.pages, pvec.nr);
746 pagevec_release(&pvec);
752 * Remove range of pages and swap entries from radix tree, and free them.
753 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
755 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
758 struct address_space *mapping = inode->i_mapping;
759 struct shmem_inode_info *info = SHMEM_I(inode);
760 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
761 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
762 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
763 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
765 pgoff_t indices[PAGEVEC_SIZE];
766 long nr_swaps_freed = 0;
771 end = -1; /* unsigned, so actually very big */
773 pagevec_init(&pvec, 0);
775 while (index < end) {
776 pvec.nr = find_get_entries(mapping, index,
777 min(end - index, (pgoff_t)PAGEVEC_SIZE),
778 pvec.pages, indices);
781 for (i = 0; i < pagevec_count(&pvec); i++) {
782 struct page *page = pvec.pages[i];
788 if (radix_tree_exceptional_entry(page)) {
791 nr_swaps_freed += !shmem_free_swap(mapping,
796 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
798 if (!trylock_page(page))
801 if (PageTransTail(page)) {
802 /* Middle of THP: zero out the page */
803 clear_highpage(page);
806 } else if (PageTransHuge(page)) {
807 if (index == round_down(end, HPAGE_PMD_NR)) {
809 * Range ends in the middle of THP:
812 clear_highpage(page);
816 index += HPAGE_PMD_NR - 1;
817 i += HPAGE_PMD_NR - 1;
820 if (!unfalloc || !PageUptodate(page)) {
821 VM_BUG_ON_PAGE(PageTail(page), page);
822 if (page_mapping(page) == mapping) {
823 VM_BUG_ON_PAGE(PageWriteback(page), page);
824 truncate_inode_page(mapping, page);
829 pagevec_remove_exceptionals(&pvec);
830 pagevec_release(&pvec);
836 struct page *page = NULL;
837 shmem_getpage(inode, start - 1, &page, SGP_READ);
839 unsigned int top = PAGE_SIZE;
844 zero_user_segment(page, partial_start, top);
845 set_page_dirty(page);
851 struct page *page = NULL;
852 shmem_getpage(inode, end, &page, SGP_READ);
854 zero_user_segment(page, 0, partial_end);
855 set_page_dirty(page);
864 while (index < end) {
867 pvec.nr = find_get_entries(mapping, index,
868 min(end - index, (pgoff_t)PAGEVEC_SIZE),
869 pvec.pages, indices);
871 /* If all gone or hole-punch or unfalloc, we're done */
872 if (index == start || end != -1)
874 /* But if truncating, restart to make sure all gone */
878 for (i = 0; i < pagevec_count(&pvec); i++) {
879 struct page *page = pvec.pages[i];
885 if (radix_tree_exceptional_entry(page)) {
888 if (shmem_free_swap(mapping, index, page)) {
889 /* Swap was replaced by page: retry */
899 if (PageTransTail(page)) {
900 /* Middle of THP: zero out the page */
901 clear_highpage(page);
904 * Partial thp truncate due 'start' in middle
905 * of THP: don't need to look on these pages
906 * again on !pvec.nr restart.
908 if (index != round_down(end, HPAGE_PMD_NR))
911 } else if (PageTransHuge(page)) {
912 if (index == round_down(end, HPAGE_PMD_NR)) {
914 * Range ends in the middle of THP:
917 clear_highpage(page);
921 index += HPAGE_PMD_NR - 1;
922 i += HPAGE_PMD_NR - 1;
925 if (!unfalloc || !PageUptodate(page)) {
926 VM_BUG_ON_PAGE(PageTail(page), page);
927 if (page_mapping(page) == mapping) {
928 VM_BUG_ON_PAGE(PageWriteback(page), page);
929 truncate_inode_page(mapping, page);
931 /* Page was replaced by swap: retry */
939 pagevec_remove_exceptionals(&pvec);
940 pagevec_release(&pvec);
944 spin_lock_irq(&info->lock);
945 info->swapped -= nr_swaps_freed;
946 shmem_recalc_inode(inode);
947 spin_unlock_irq(&info->lock);
950 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
952 shmem_undo_range(inode, lstart, lend, false);
953 inode->i_ctime = inode->i_mtime = current_time(inode);
955 EXPORT_SYMBOL_GPL(shmem_truncate_range);
957 static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
960 struct inode *inode = dentry->d_inode;
961 struct shmem_inode_info *info = SHMEM_I(inode);
963 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
964 spin_lock_irq(&info->lock);
965 shmem_recalc_inode(inode);
966 spin_unlock_irq(&info->lock);
968 generic_fillattr(inode, stat);
972 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
974 struct inode *inode = d_inode(dentry);
975 struct shmem_inode_info *info = SHMEM_I(inode);
976 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
979 error = setattr_prepare(dentry, attr);
983 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
984 loff_t oldsize = inode->i_size;
985 loff_t newsize = attr->ia_size;
987 /* protected by i_mutex */
988 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
989 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
992 if (newsize != oldsize) {
993 error = shmem_reacct_size(SHMEM_I(inode)->flags,
997 i_size_write(inode, newsize);
998 inode->i_ctime = inode->i_mtime = current_time(inode);
1000 if (newsize <= oldsize) {
1001 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1002 if (oldsize > holebegin)
1003 unmap_mapping_range(inode->i_mapping,
1006 shmem_truncate_range(inode,
1007 newsize, (loff_t)-1);
1008 /* unmap again to remove racily COWed private pages */
1009 if (oldsize > holebegin)
1010 unmap_mapping_range(inode->i_mapping,
1014 * Part of the huge page can be beyond i_size: subject
1015 * to shrink under memory pressure.
1017 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1018 spin_lock(&sbinfo->shrinklist_lock);
1019 if (list_empty(&info->shrinklist)) {
1020 list_add_tail(&info->shrinklist,
1021 &sbinfo->shrinklist);
1022 sbinfo->shrinklist_len++;
1024 spin_unlock(&sbinfo->shrinklist_lock);
1029 setattr_copy(inode, attr);
1030 if (attr->ia_valid & ATTR_MODE)
1031 error = posix_acl_chmod(inode, inode->i_mode);
1035 static void shmem_evict_inode(struct inode *inode)
1037 struct shmem_inode_info *info = SHMEM_I(inode);
1038 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1040 if (inode->i_mapping->a_ops == &shmem_aops) {
1041 shmem_unacct_size(info->flags, inode->i_size);
1043 shmem_truncate_range(inode, 0, (loff_t)-1);
1044 if (!list_empty(&info->shrinklist)) {
1045 spin_lock(&sbinfo->shrinklist_lock);
1046 if (!list_empty(&info->shrinklist)) {
1047 list_del_init(&info->shrinklist);
1048 sbinfo->shrinklist_len--;
1050 spin_unlock(&sbinfo->shrinklist_lock);
1052 if (!list_empty(&info->swaplist)) {
1053 mutex_lock(&shmem_swaplist_mutex);
1054 list_del_init(&info->swaplist);
1055 mutex_unlock(&shmem_swaplist_mutex);
1059 simple_xattrs_free(&info->xattrs);
1060 WARN_ON(inode->i_blocks);
1061 shmem_free_inode(inode->i_sb);
1065 static unsigned long find_swap_entry(struct radix_tree_root *root, void *item)
1067 struct radix_tree_iter iter;
1069 unsigned long found = -1;
1070 unsigned int checked = 0;
1073 radix_tree_for_each_slot(slot, root, &iter, 0) {
1074 if (*slot == item) {
1079 if ((checked % 4096) != 0)
1081 slot = radix_tree_iter_resume(slot, &iter);
1090 * If swap found in inode, free it and move page from swapcache to filecache.
1092 static int shmem_unuse_inode(struct shmem_inode_info *info,
1093 swp_entry_t swap, struct page **pagep)
1095 struct address_space *mapping = info->vfs_inode.i_mapping;
1101 radswap = swp_to_radix_entry(swap);
1102 index = find_swap_entry(&mapping->page_tree, radswap);
1104 return -EAGAIN; /* tell shmem_unuse we found nothing */
1107 * Move _head_ to start search for next from here.
1108 * But be careful: shmem_evict_inode checks list_empty without taking
1109 * mutex, and there's an instant in list_move_tail when info->swaplist
1110 * would appear empty, if it were the only one on shmem_swaplist.
1112 if (shmem_swaplist.next != &info->swaplist)
1113 list_move_tail(&shmem_swaplist, &info->swaplist);
1115 gfp = mapping_gfp_mask(mapping);
1116 if (shmem_should_replace_page(*pagep, gfp)) {
1117 mutex_unlock(&shmem_swaplist_mutex);
1118 error = shmem_replace_page(pagep, gfp, info, index);
1119 mutex_lock(&shmem_swaplist_mutex);
1121 * We needed to drop mutex to make that restrictive page
1122 * allocation, but the inode might have been freed while we
1123 * dropped it: although a racing shmem_evict_inode() cannot
1124 * complete without emptying the radix_tree, our page lock
1125 * on this swapcache page is not enough to prevent that -
1126 * free_swap_and_cache() of our swap entry will only
1127 * trylock_page(), removing swap from radix_tree whatever.
1129 * We must not proceed to shmem_add_to_page_cache() if the
1130 * inode has been freed, but of course we cannot rely on
1131 * inode or mapping or info to check that. However, we can
1132 * safely check if our swap entry is still in use (and here
1133 * it can't have got reused for another page): if it's still
1134 * in use, then the inode cannot have been freed yet, and we
1135 * can safely proceed (if it's no longer in use, that tells
1136 * nothing about the inode, but we don't need to unuse swap).
1138 if (!page_swapcount(*pagep))
1143 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1144 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1145 * beneath us (pagelock doesn't help until the page is in pagecache).
1148 error = shmem_add_to_page_cache(*pagep, mapping, index,
1150 if (error != -ENOMEM) {
1152 * Truncation and eviction use free_swap_and_cache(), which
1153 * only does trylock page: if we raced, best clean up here.
1155 delete_from_swap_cache(*pagep);
1156 set_page_dirty(*pagep);
1158 spin_lock_irq(&info->lock);
1160 spin_unlock_irq(&info->lock);
1168 * Search through swapped inodes to find and replace swap by page.
1170 int shmem_unuse(swp_entry_t swap, struct page *page)
1172 struct list_head *this, *next;
1173 struct shmem_inode_info *info;
1174 struct mem_cgroup *memcg;
1178 * There's a faint possibility that swap page was replaced before
1179 * caller locked it: caller will come back later with the right page.
1181 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1185 * Charge page using GFP_KERNEL while we can wait, before taking
1186 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1187 * Charged back to the user (not to caller) when swap account is used.
1189 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1193 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1196 mutex_lock(&shmem_swaplist_mutex);
1197 list_for_each_safe(this, next, &shmem_swaplist) {
1198 info = list_entry(this, struct shmem_inode_info, swaplist);
1200 error = shmem_unuse_inode(info, swap, &page);
1202 list_del_init(&info->swaplist);
1204 if (error != -EAGAIN)
1206 /* found nothing in this: move on to search the next */
1208 mutex_unlock(&shmem_swaplist_mutex);
1211 if (error != -ENOMEM)
1213 mem_cgroup_cancel_charge(page, memcg, false);
1215 mem_cgroup_commit_charge(page, memcg, true, false);
1223 * Move the page from the page cache to the swap cache.
1225 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1227 struct shmem_inode_info *info;
1228 struct address_space *mapping;
1229 struct inode *inode;
1233 VM_BUG_ON_PAGE(PageCompound(page), page);
1234 BUG_ON(!PageLocked(page));
1235 mapping = page->mapping;
1236 index = page->index;
1237 inode = mapping->host;
1238 info = SHMEM_I(inode);
1239 if (info->flags & VM_LOCKED)
1241 if (!total_swap_pages)
1245 * Our capabilities prevent regular writeback or sync from ever calling
1246 * shmem_writepage; but a stacking filesystem might use ->writepage of
1247 * its underlying filesystem, in which case tmpfs should write out to
1248 * swap only in response to memory pressure, and not for the writeback
1251 if (!wbc->for_reclaim) {
1252 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1257 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1258 * value into swapfile.c, the only way we can correctly account for a
1259 * fallocated page arriving here is now to initialize it and write it.
1261 * That's okay for a page already fallocated earlier, but if we have
1262 * not yet completed the fallocation, then (a) we want to keep track
1263 * of this page in case we have to undo it, and (b) it may not be a
1264 * good idea to continue anyway, once we're pushing into swap. So
1265 * reactivate the page, and let shmem_fallocate() quit when too many.
1267 if (!PageUptodate(page)) {
1268 if (inode->i_private) {
1269 struct shmem_falloc *shmem_falloc;
1270 spin_lock(&inode->i_lock);
1271 shmem_falloc = inode->i_private;
1273 !shmem_falloc->waitq &&
1274 index >= shmem_falloc->start &&
1275 index < shmem_falloc->next)
1276 shmem_falloc->nr_unswapped++;
1278 shmem_falloc = NULL;
1279 spin_unlock(&inode->i_lock);
1283 clear_highpage(page);
1284 flush_dcache_page(page);
1285 SetPageUptodate(page);
1288 swap = get_swap_page();
1292 if (mem_cgroup_try_charge_swap(page, swap))
1296 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1297 * if it's not already there. Do it now before the page is
1298 * moved to swap cache, when its pagelock no longer protects
1299 * the inode from eviction. But don't unlock the mutex until
1300 * we've incremented swapped, because shmem_unuse_inode() will
1301 * prune a !swapped inode from the swaplist under this mutex.
1303 mutex_lock(&shmem_swaplist_mutex);
1304 if (list_empty(&info->swaplist))
1305 list_add_tail(&info->swaplist, &shmem_swaplist);
1307 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1308 spin_lock_irq(&info->lock);
1309 shmem_recalc_inode(inode);
1311 spin_unlock_irq(&info->lock);
1313 swap_shmem_alloc(swap);
1314 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1316 mutex_unlock(&shmem_swaplist_mutex);
1317 BUG_ON(page_mapped(page));
1318 swap_writepage(page, wbc);
1322 mutex_unlock(&shmem_swaplist_mutex);
1324 swapcache_free(swap);
1326 set_page_dirty(page);
1327 if (wbc->for_reclaim)
1328 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1333 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1334 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1338 if (!mpol || mpol->mode == MPOL_DEFAULT)
1339 return; /* show nothing */
1341 mpol_to_str(buffer, sizeof(buffer), mpol);
1343 seq_printf(seq, ",mpol=%s", buffer);
1346 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1348 struct mempolicy *mpol = NULL;
1350 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1351 mpol = sbinfo->mpol;
1353 spin_unlock(&sbinfo->stat_lock);
1357 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1358 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1361 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1365 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1367 #define vm_policy vm_private_data
1370 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1371 struct shmem_inode_info *info, pgoff_t index)
1373 /* Create a pseudo vma that just contains the policy */
1375 /* Bias interleave by inode number to distribute better across nodes */
1376 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1378 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1381 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1383 /* Drop reference taken by mpol_shared_policy_lookup() */
1384 mpol_cond_put(vma->vm_policy);
1387 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1388 struct shmem_inode_info *info, pgoff_t index)
1390 struct vm_area_struct pvma;
1393 shmem_pseudo_vma_init(&pvma, info, index);
1394 page = swapin_readahead(swap, gfp, &pvma, 0);
1395 shmem_pseudo_vma_destroy(&pvma);
1400 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1401 struct shmem_inode_info *info, pgoff_t index)
1403 struct vm_area_struct pvma;
1404 struct inode *inode = &info->vfs_inode;
1405 struct address_space *mapping = inode->i_mapping;
1406 pgoff_t idx, hindex;
1407 void __rcu **results;
1410 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1413 hindex = round_down(index, HPAGE_PMD_NR);
1415 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1416 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1422 shmem_pseudo_vma_init(&pvma, info, hindex);
1423 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1424 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1425 shmem_pseudo_vma_destroy(&pvma);
1427 prep_transhuge_page(page);
1431 static struct page *shmem_alloc_page(gfp_t gfp,
1432 struct shmem_inode_info *info, pgoff_t index)
1434 struct vm_area_struct pvma;
1437 shmem_pseudo_vma_init(&pvma, info, index);
1438 page = alloc_page_vma(gfp, &pvma, 0);
1439 shmem_pseudo_vma_destroy(&pvma);
1444 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1445 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1446 pgoff_t index, bool huge)
1452 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1454 nr = huge ? HPAGE_PMD_NR : 1;
1456 if (shmem_acct_block(info->flags, nr))
1458 if (sbinfo->max_blocks) {
1459 if (percpu_counter_compare(&sbinfo->used_blocks,
1460 sbinfo->max_blocks - nr) > 0)
1462 percpu_counter_add(&sbinfo->used_blocks, nr);
1466 page = shmem_alloc_hugepage(gfp, info, index);
1468 page = shmem_alloc_page(gfp, info, index);
1470 __SetPageLocked(page);
1471 __SetPageSwapBacked(page);
1476 if (sbinfo->max_blocks)
1477 percpu_counter_add(&sbinfo->used_blocks, -nr);
1479 shmem_unacct_blocks(info->flags, nr);
1481 return ERR_PTR(err);
1485 * When a page is moved from swapcache to shmem filecache (either by the
1486 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1487 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1488 * ignorance of the mapping it belongs to. If that mapping has special
1489 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1490 * we may need to copy to a suitable page before moving to filecache.
1492 * In a future release, this may well be extended to respect cpuset and
1493 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1494 * but for now it is a simple matter of zone.
1496 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1498 return page_zonenum(page) > gfp_zone(gfp);
1501 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1502 struct shmem_inode_info *info, pgoff_t index)
1504 struct page *oldpage, *newpage;
1505 struct address_space *swap_mapping;
1510 swap_index = page_private(oldpage);
1511 swap_mapping = page_mapping(oldpage);
1514 * We have arrived here because our zones are constrained, so don't
1515 * limit chance of success by further cpuset and node constraints.
1517 gfp &= ~GFP_CONSTRAINT_MASK;
1518 newpage = shmem_alloc_page(gfp, info, index);
1523 copy_highpage(newpage, oldpage);
1524 flush_dcache_page(newpage);
1526 __SetPageLocked(newpage);
1527 __SetPageSwapBacked(newpage);
1528 SetPageUptodate(newpage);
1529 set_page_private(newpage, swap_index);
1530 SetPageSwapCache(newpage);
1533 * Our caller will very soon move newpage out of swapcache, but it's
1534 * a nice clean interface for us to replace oldpage by newpage there.
1536 spin_lock_irq(&swap_mapping->tree_lock);
1537 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1540 __inc_node_page_state(newpage, NR_FILE_PAGES);
1541 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1543 spin_unlock_irq(&swap_mapping->tree_lock);
1545 if (unlikely(error)) {
1547 * Is this possible? I think not, now that our callers check
1548 * both PageSwapCache and page_private after getting page lock;
1549 * but be defensive. Reverse old to newpage for clear and free.
1553 mem_cgroup_migrate(oldpage, newpage);
1554 lru_cache_add_anon(newpage);
1558 ClearPageSwapCache(oldpage);
1559 set_page_private(oldpage, 0);
1561 unlock_page(oldpage);
1568 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1570 * If we allocate a new one we do not mark it dirty. That's up to the
1571 * vm. If we swap it in we mark it dirty since we also free the swap
1572 * entry since a page cannot live in both the swap and page cache.
1574 * fault_mm and fault_type are only supplied by shmem_fault:
1575 * otherwise they are NULL.
1577 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1578 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1579 struct mm_struct *fault_mm, int *fault_type)
1581 struct address_space *mapping = inode->i_mapping;
1582 struct shmem_inode_info *info = SHMEM_I(inode);
1583 struct shmem_sb_info *sbinfo;
1584 struct mm_struct *charge_mm;
1585 struct mem_cgroup *memcg;
1588 enum sgp_type sgp_huge = sgp;
1589 pgoff_t hindex = index;
1594 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1596 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1600 page = find_lock_entry(mapping, index);
1601 if (radix_tree_exceptional_entry(page)) {
1602 swap = radix_to_swp_entry(page);
1606 if (sgp <= SGP_CACHE &&
1607 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1612 if (page && sgp == SGP_WRITE)
1613 mark_page_accessed(page);
1615 /* fallocated page? */
1616 if (page && !PageUptodate(page)) {
1617 if (sgp != SGP_READ)
1623 if (page || (sgp == SGP_READ && !swap.val)) {
1629 * Fast cache lookup did not find it:
1630 * bring it back from swap or allocate.
1632 sbinfo = SHMEM_SB(inode->i_sb);
1633 charge_mm = fault_mm ? : current->mm;
1636 /* Look it up and read it in.. */
1637 page = lookup_swap_cache(swap);
1639 /* Or update major stats only when swapin succeeds?? */
1641 *fault_type |= VM_FAULT_MAJOR;
1642 count_vm_event(PGMAJFAULT);
1643 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1645 /* Here we actually start the io */
1646 page = shmem_swapin(swap, gfp, info, index);
1653 /* We have to do this with page locked to prevent races */
1655 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1656 !shmem_confirm_swap(mapping, index, swap)) {
1657 error = -EEXIST; /* try again */
1660 if (!PageUptodate(page)) {
1664 wait_on_page_writeback(page);
1666 if (shmem_should_replace_page(page, gfp)) {
1667 error = shmem_replace_page(&page, gfp, info, index);
1672 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1675 error = shmem_add_to_page_cache(page, mapping, index,
1676 swp_to_radix_entry(swap));
1678 * We already confirmed swap under page lock, and make
1679 * no memory allocation here, so usually no possibility
1680 * of error; but free_swap_and_cache() only trylocks a
1681 * page, so it is just possible that the entry has been
1682 * truncated or holepunched since swap was confirmed.
1683 * shmem_undo_range() will have done some of the
1684 * unaccounting, now delete_from_swap_cache() will do
1686 * Reset swap.val? No, leave it so "failed" goes back to
1687 * "repeat": reading a hole and writing should succeed.
1690 mem_cgroup_cancel_charge(page, memcg, false);
1691 delete_from_swap_cache(page);
1697 mem_cgroup_commit_charge(page, memcg, true, false);
1699 spin_lock_irq(&info->lock);
1701 shmem_recalc_inode(inode);
1702 spin_unlock_irq(&info->lock);
1704 if (sgp == SGP_WRITE)
1705 mark_page_accessed(page);
1707 delete_from_swap_cache(page);
1708 set_page_dirty(page);
1712 /* shmem_symlink() */
1713 if (mapping->a_ops != &shmem_aops)
1715 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1717 if (shmem_huge == SHMEM_HUGE_FORCE)
1719 switch (sbinfo->huge) {
1722 case SHMEM_HUGE_NEVER:
1724 case SHMEM_HUGE_WITHIN_SIZE:
1725 off = round_up(index, HPAGE_PMD_NR);
1726 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1727 if (i_size >= HPAGE_PMD_SIZE &&
1728 i_size >> PAGE_SHIFT >= off)
1731 case SHMEM_HUGE_ADVISE:
1732 if (sgp_huge == SGP_HUGE)
1734 /* TODO: implement fadvise() hints */
1739 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1742 alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1747 error = PTR_ERR(page);
1749 if (error != -ENOSPC)
1752 * Try to reclaim some spece by splitting a huge page
1753 * beyond i_size on the filesystem.
1757 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1758 if (ret == SHRINK_STOP)
1766 if (PageTransHuge(page))
1767 hindex = round_down(index, HPAGE_PMD_NR);
1771 if (sgp == SGP_WRITE)
1772 __SetPageReferenced(page);
1774 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1775 PageTransHuge(page));
1778 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1779 compound_order(page));
1781 error = shmem_add_to_page_cache(page, mapping, hindex,
1783 radix_tree_preload_end();
1786 mem_cgroup_cancel_charge(page, memcg,
1787 PageTransHuge(page));
1790 mem_cgroup_commit_charge(page, memcg, false,
1791 PageTransHuge(page));
1792 lru_cache_add_anon(page);
1794 spin_lock_irq(&info->lock);
1795 info->alloced += 1 << compound_order(page);
1796 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1797 shmem_recalc_inode(inode);
1798 spin_unlock_irq(&info->lock);
1801 if (PageTransHuge(page) &&
1802 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1803 hindex + HPAGE_PMD_NR - 1) {
1805 * Part of the huge page is beyond i_size: subject
1806 * to shrink under memory pressure.
1808 spin_lock(&sbinfo->shrinklist_lock);
1809 if (list_empty(&info->shrinklist)) {
1810 list_add_tail(&info->shrinklist,
1811 &sbinfo->shrinklist);
1812 sbinfo->shrinklist_len++;
1814 spin_unlock(&sbinfo->shrinklist_lock);
1818 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1820 if (sgp == SGP_FALLOC)
1824 * Let SGP_WRITE caller clear ends if write does not fill page;
1825 * but SGP_FALLOC on a page fallocated earlier must initialize
1826 * it now, lest undo on failure cancel our earlier guarantee.
1828 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1829 struct page *head = compound_head(page);
1832 for (i = 0; i < (1 << compound_order(head)); i++) {
1833 clear_highpage(head + i);
1834 flush_dcache_page(head + i);
1836 SetPageUptodate(head);
1840 /* Perhaps the file has been truncated since we checked */
1841 if (sgp <= SGP_CACHE &&
1842 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1844 ClearPageDirty(page);
1845 delete_from_page_cache(page);
1846 spin_lock_irq(&info->lock);
1847 shmem_recalc_inode(inode);
1848 spin_unlock_irq(&info->lock);
1853 *pagep = page + index - hindex;
1860 if (sbinfo->max_blocks)
1861 percpu_counter_sub(&sbinfo->used_blocks,
1862 1 << compound_order(page));
1863 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1865 if (PageTransHuge(page)) {
1871 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1878 if (error == -ENOSPC && !once++) {
1879 spin_lock_irq(&info->lock);
1880 shmem_recalc_inode(inode);
1881 spin_unlock_irq(&info->lock);
1884 if (error == -EEXIST) /* from above or from radix_tree_insert */
1890 * This is like autoremove_wake_function, but it removes the wait queue
1891 * entry unconditionally - even if something else had already woken the
1894 static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1896 int ret = default_wake_function(wait, mode, sync, key);
1897 list_del_init(&wait->task_list);
1901 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1903 struct inode *inode = file_inode(vma->vm_file);
1904 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1907 int ret = VM_FAULT_LOCKED;
1910 * Trinity finds that probing a hole which tmpfs is punching can
1911 * prevent the hole-punch from ever completing: which in turn
1912 * locks writers out with its hold on i_mutex. So refrain from
1913 * faulting pages into the hole while it's being punched. Although
1914 * shmem_undo_range() does remove the additions, it may be unable to
1915 * keep up, as each new page needs its own unmap_mapping_range() call,
1916 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1918 * It does not matter if we sometimes reach this check just before the
1919 * hole-punch begins, so that one fault then races with the punch:
1920 * we just need to make racing faults a rare case.
1922 * The implementation below would be much simpler if we just used a
1923 * standard mutex or completion: but we cannot take i_mutex in fault,
1924 * and bloating every shmem inode for this unlikely case would be sad.
1926 if (unlikely(inode->i_private)) {
1927 struct shmem_falloc *shmem_falloc;
1929 spin_lock(&inode->i_lock);
1930 shmem_falloc = inode->i_private;
1932 shmem_falloc->waitq &&
1933 vmf->pgoff >= shmem_falloc->start &&
1934 vmf->pgoff < shmem_falloc->next) {
1935 wait_queue_head_t *shmem_falloc_waitq;
1936 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1938 ret = VM_FAULT_NOPAGE;
1939 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1940 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1941 /* It's polite to up mmap_sem if we can */
1942 up_read(&vma->vm_mm->mmap_sem);
1943 ret = VM_FAULT_RETRY;
1946 shmem_falloc_waitq = shmem_falloc->waitq;
1947 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1948 TASK_UNINTERRUPTIBLE);
1949 spin_unlock(&inode->i_lock);
1953 * shmem_falloc_waitq points into the shmem_fallocate()
1954 * stack of the hole-punching task: shmem_falloc_waitq
1955 * is usually invalid by the time we reach here, but
1956 * finish_wait() does not dereference it in that case;
1957 * though i_lock needed lest racing with wake_up_all().
1959 spin_lock(&inode->i_lock);
1960 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1961 spin_unlock(&inode->i_lock);
1964 spin_unlock(&inode->i_lock);
1968 if (vma->vm_flags & VM_HUGEPAGE)
1970 else if (vma->vm_flags & VM_NOHUGEPAGE)
1973 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1974 gfp, vma->vm_mm, &ret);
1976 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1980 unsigned long shmem_get_unmapped_area(struct file *file,
1981 unsigned long uaddr, unsigned long len,
1982 unsigned long pgoff, unsigned long flags)
1984 unsigned long (*get_area)(struct file *,
1985 unsigned long, unsigned long, unsigned long, unsigned long);
1987 unsigned long offset;
1988 unsigned long inflated_len;
1989 unsigned long inflated_addr;
1990 unsigned long inflated_offset;
1992 if (len > TASK_SIZE)
1995 get_area = current->mm->get_unmapped_area;
1996 addr = get_area(file, uaddr, len, pgoff, flags);
1998 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2000 if (IS_ERR_VALUE(addr))
2002 if (addr & ~PAGE_MASK)
2004 if (addr > TASK_SIZE - len)
2007 if (shmem_huge == SHMEM_HUGE_DENY)
2009 if (len < HPAGE_PMD_SIZE)
2011 if (flags & MAP_FIXED)
2014 * Our priority is to support MAP_SHARED mapped hugely;
2015 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2016 * But if caller specified an address hint, respect that as before.
2021 if (shmem_huge != SHMEM_HUGE_FORCE) {
2022 struct super_block *sb;
2025 VM_BUG_ON(file->f_op != &shmem_file_operations);
2026 sb = file_inode(file)->i_sb;
2029 * Called directly from mm/mmap.c, or drivers/char/mem.c
2030 * for "/dev/zero", to create a shared anonymous object.
2032 if (IS_ERR(shm_mnt))
2034 sb = shm_mnt->mnt_sb;
2036 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2040 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2041 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2043 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2046 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2047 if (inflated_len > TASK_SIZE)
2049 if (inflated_len < len)
2052 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2053 if (IS_ERR_VALUE(inflated_addr))
2055 if (inflated_addr & ~PAGE_MASK)
2058 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2059 inflated_addr += offset - inflated_offset;
2060 if (inflated_offset > offset)
2061 inflated_addr += HPAGE_PMD_SIZE;
2063 if (inflated_addr > TASK_SIZE - len)
2065 return inflated_addr;
2069 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2071 struct inode *inode = file_inode(vma->vm_file);
2072 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2075 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2078 struct inode *inode = file_inode(vma->vm_file);
2081 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2082 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2086 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2088 struct inode *inode = file_inode(file);
2089 struct shmem_inode_info *info = SHMEM_I(inode);
2090 int retval = -ENOMEM;
2092 spin_lock_irq(&info->lock);
2093 if (lock && !(info->flags & VM_LOCKED)) {
2094 if (!user_shm_lock(inode->i_size, user))
2096 info->flags |= VM_LOCKED;
2097 mapping_set_unevictable(file->f_mapping);
2099 if (!lock && (info->flags & VM_LOCKED) && user) {
2100 user_shm_unlock(inode->i_size, user);
2101 info->flags &= ~VM_LOCKED;
2102 mapping_clear_unevictable(file->f_mapping);
2107 spin_unlock_irq(&info->lock);
2111 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2113 file_accessed(file);
2114 vma->vm_ops = &shmem_vm_ops;
2115 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2116 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2117 (vma->vm_end & HPAGE_PMD_MASK)) {
2118 khugepaged_enter(vma, vma->vm_flags);
2123 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2124 umode_t mode, dev_t dev, unsigned long flags)
2126 struct inode *inode;
2127 struct shmem_inode_info *info;
2128 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2130 if (shmem_reserve_inode(sb))
2133 inode = new_inode(sb);
2135 inode->i_ino = get_next_ino();
2136 inode_init_owner(inode, dir, mode);
2137 inode->i_blocks = 0;
2138 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2139 inode->i_generation = get_seconds();
2140 info = SHMEM_I(inode);
2141 memset(info, 0, (char *)inode - (char *)info);
2142 spin_lock_init(&info->lock);
2143 info->seals = F_SEAL_SEAL;
2144 info->flags = flags & VM_NORESERVE;
2145 INIT_LIST_HEAD(&info->shrinklist);
2146 INIT_LIST_HEAD(&info->swaplist);
2147 simple_xattrs_init(&info->xattrs);
2148 cache_no_acl(inode);
2150 switch (mode & S_IFMT) {
2152 inode->i_op = &shmem_special_inode_operations;
2153 init_special_inode(inode, mode, dev);
2156 inode->i_mapping->a_ops = &shmem_aops;
2157 inode->i_op = &shmem_inode_operations;
2158 inode->i_fop = &shmem_file_operations;
2159 mpol_shared_policy_init(&info->policy,
2160 shmem_get_sbmpol(sbinfo));
2164 /* Some things misbehave if size == 0 on a directory */
2165 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2166 inode->i_op = &shmem_dir_inode_operations;
2167 inode->i_fop = &simple_dir_operations;
2171 * Must not load anything in the rbtree,
2172 * mpol_free_shared_policy will not be called.
2174 mpol_shared_policy_init(&info->policy, NULL);
2178 shmem_free_inode(sb);
2182 bool shmem_mapping(struct address_space *mapping)
2184 return mapping->a_ops == &shmem_aops;
2187 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2189 struct vm_area_struct *dst_vma,
2190 unsigned long dst_addr,
2191 unsigned long src_addr,
2192 struct page **pagep)
2194 struct inode *inode = file_inode(dst_vma->vm_file);
2195 struct shmem_inode_info *info = SHMEM_I(inode);
2196 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2197 struct address_space *mapping = inode->i_mapping;
2198 gfp_t gfp = mapping_gfp_mask(mapping);
2199 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2200 struct mem_cgroup *memcg;
2204 pte_t _dst_pte, *dst_pte;
2209 if (shmem_acct_block(info->flags, 1))
2211 if (sbinfo->max_blocks) {
2212 if (percpu_counter_compare(&sbinfo->used_blocks,
2213 sbinfo->max_blocks) >= 0)
2214 goto out_unacct_blocks;
2215 percpu_counter_inc(&sbinfo->used_blocks);
2218 page = shmem_alloc_page(gfp, info, pgoff);
2220 goto out_dec_used_blocks;
2222 page_kaddr = kmap_atomic(page);
2223 ret = copy_from_user(page_kaddr, (const void __user *)src_addr,
2225 kunmap_atomic(page_kaddr);
2227 /* fallback to copy_from_user outside mmap_sem */
2228 if (unlikely(ret)) {
2230 /* don't free the page */
2238 ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false);
2242 ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
2244 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL);
2245 radix_tree_preload_end();
2248 goto out_release_uncharge;
2250 mem_cgroup_commit_charge(page, memcg, false, false);
2252 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2253 if (dst_vma->vm_flags & VM_WRITE)
2254 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2257 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2258 if (!pte_none(*dst_pte))
2259 goto out_release_uncharge_unlock;
2261 __SetPageUptodate(page);
2263 lru_cache_add_anon(page);
2265 spin_lock(&info->lock);
2267 inode->i_blocks += BLOCKS_PER_PAGE;
2268 shmem_recalc_inode(inode);
2269 spin_unlock(&info->lock);
2271 inc_mm_counter(dst_mm, mm_counter_file(page));
2272 page_add_file_rmap(page, false);
2273 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2275 /* No need to invalidate - it was non-present before */
2276 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2278 pte_unmap_unlock(dst_pte, ptl);
2282 out_release_uncharge_unlock:
2283 pte_unmap_unlock(dst_pte, ptl);
2284 out_release_uncharge:
2285 mem_cgroup_cancel_charge(page, memcg, false);
2288 out_dec_used_blocks:
2289 if (sbinfo->max_blocks)
2290 percpu_counter_add(&sbinfo->used_blocks, -1);
2292 shmem_unacct_blocks(info->flags, 1);
2297 static const struct inode_operations shmem_symlink_inode_operations;
2298 static const struct inode_operations shmem_short_symlink_operations;
2300 #ifdef CONFIG_TMPFS_XATTR
2301 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2303 #define shmem_initxattrs NULL
2307 shmem_write_begin(struct file *file, struct address_space *mapping,
2308 loff_t pos, unsigned len, unsigned flags,
2309 struct page **pagep, void **fsdata)
2311 struct inode *inode = mapping->host;
2312 struct shmem_inode_info *info = SHMEM_I(inode);
2313 pgoff_t index = pos >> PAGE_SHIFT;
2315 /* i_mutex is held by caller */
2316 if (unlikely(info->seals)) {
2317 if (info->seals & F_SEAL_WRITE)
2319 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2323 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2327 shmem_write_end(struct file *file, struct address_space *mapping,
2328 loff_t pos, unsigned len, unsigned copied,
2329 struct page *page, void *fsdata)
2331 struct inode *inode = mapping->host;
2333 if (pos + copied > inode->i_size)
2334 i_size_write(inode, pos + copied);
2336 if (!PageUptodate(page)) {
2337 struct page *head = compound_head(page);
2338 if (PageTransCompound(page)) {
2341 for (i = 0; i < HPAGE_PMD_NR; i++) {
2342 if (head + i == page)
2344 clear_highpage(head + i);
2345 flush_dcache_page(head + i);
2348 if (copied < PAGE_SIZE) {
2349 unsigned from = pos & (PAGE_SIZE - 1);
2350 zero_user_segments(page, 0, from,
2351 from + copied, PAGE_SIZE);
2353 SetPageUptodate(head);
2355 set_page_dirty(page);
2362 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2364 struct file *file = iocb->ki_filp;
2365 struct inode *inode = file_inode(file);
2366 struct address_space *mapping = inode->i_mapping;
2368 unsigned long offset;
2369 enum sgp_type sgp = SGP_READ;
2372 loff_t *ppos = &iocb->ki_pos;
2375 * Might this read be for a stacking filesystem? Then when reading
2376 * holes of a sparse file, we actually need to allocate those pages,
2377 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2379 if (!iter_is_iovec(to))
2382 index = *ppos >> PAGE_SHIFT;
2383 offset = *ppos & ~PAGE_MASK;
2386 struct page *page = NULL;
2388 unsigned long nr, ret;
2389 loff_t i_size = i_size_read(inode);
2391 end_index = i_size >> PAGE_SHIFT;
2392 if (index > end_index)
2394 if (index == end_index) {
2395 nr = i_size & ~PAGE_MASK;
2400 error = shmem_getpage(inode, index, &page, sgp);
2402 if (error == -EINVAL)
2407 if (sgp == SGP_CACHE)
2408 set_page_dirty(page);
2413 * We must evaluate after, since reads (unlike writes)
2414 * are called without i_mutex protection against truncate
2417 i_size = i_size_read(inode);
2418 end_index = i_size >> PAGE_SHIFT;
2419 if (index == end_index) {
2420 nr = i_size & ~PAGE_MASK;
2431 * If users can be writing to this page using arbitrary
2432 * virtual addresses, take care about potential aliasing
2433 * before reading the page on the kernel side.
2435 if (mapping_writably_mapped(mapping))
2436 flush_dcache_page(page);
2438 * Mark the page accessed if we read the beginning.
2441 mark_page_accessed(page);
2443 page = ZERO_PAGE(0);
2448 * Ok, we have the page, and it's up-to-date, so
2449 * now we can copy it to user space...
2451 ret = copy_page_to_iter(page, offset, nr, to);
2454 index += offset >> PAGE_SHIFT;
2455 offset &= ~PAGE_MASK;
2458 if (!iov_iter_count(to))
2467 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2468 file_accessed(file);
2469 return retval ? retval : error;
2473 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2475 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2476 pgoff_t index, pgoff_t end, int whence)
2479 struct pagevec pvec;
2480 pgoff_t indices[PAGEVEC_SIZE];
2484 pagevec_init(&pvec, 0);
2485 pvec.nr = 1; /* start small: we may be there already */
2487 pvec.nr = find_get_entries(mapping, index,
2488 pvec.nr, pvec.pages, indices);
2490 if (whence == SEEK_DATA)
2494 for (i = 0; i < pvec.nr; i++, index++) {
2495 if (index < indices[i]) {
2496 if (whence == SEEK_HOLE) {
2502 page = pvec.pages[i];
2503 if (page && !radix_tree_exceptional_entry(page)) {
2504 if (!PageUptodate(page))
2508 (page && whence == SEEK_DATA) ||
2509 (!page && whence == SEEK_HOLE)) {
2514 pagevec_remove_exceptionals(&pvec);
2515 pagevec_release(&pvec);
2516 pvec.nr = PAGEVEC_SIZE;
2522 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2524 struct address_space *mapping = file->f_mapping;
2525 struct inode *inode = mapping->host;
2529 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2530 return generic_file_llseek_size(file, offset, whence,
2531 MAX_LFS_FILESIZE, i_size_read(inode));
2533 /* We're holding i_mutex so we can access i_size directly */
2537 else if (offset >= inode->i_size)
2540 start = offset >> PAGE_SHIFT;
2541 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2542 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2543 new_offset <<= PAGE_SHIFT;
2544 if (new_offset > offset) {
2545 if (new_offset < inode->i_size)
2546 offset = new_offset;
2547 else if (whence == SEEK_DATA)
2550 offset = inode->i_size;
2555 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2556 inode_unlock(inode);
2561 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2562 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2564 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2565 #define LAST_SCAN 4 /* about 150ms max */
2567 static void shmem_tag_pins(struct address_space *mapping)
2569 struct radix_tree_iter iter;
2578 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2579 page = radix_tree_deref_slot(slot);
2580 if (!page || radix_tree_exception(page)) {
2581 if (radix_tree_deref_retry(page)) {
2582 slot = radix_tree_iter_retry(&iter);
2585 } else if (page_count(page) - page_mapcount(page) > 1) {
2586 spin_lock_irq(&mapping->tree_lock);
2587 radix_tree_tag_set(&mapping->page_tree, iter.index,
2589 spin_unlock_irq(&mapping->tree_lock);
2592 if (need_resched()) {
2593 slot = radix_tree_iter_resume(slot, &iter);
2601 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2602 * via get_user_pages(), drivers might have some pending I/O without any active
2603 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2604 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2605 * them to be dropped.
2606 * The caller must guarantee that no new user will acquire writable references
2607 * to those pages to avoid races.
2609 static int shmem_wait_for_pins(struct address_space *mapping)
2611 struct radix_tree_iter iter;
2617 shmem_tag_pins(mapping);
2620 for (scan = 0; scan <= LAST_SCAN; scan++) {
2621 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2625 lru_add_drain_all();
2626 else if (schedule_timeout_killable((HZ << scan) / 200))
2631 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2632 start, SHMEM_TAG_PINNED) {
2634 page = radix_tree_deref_slot(slot);
2635 if (radix_tree_exception(page)) {
2636 if (radix_tree_deref_retry(page)) {
2637 slot = radix_tree_iter_retry(&iter);
2645 page_count(page) - page_mapcount(page) != 1) {
2646 if (scan < LAST_SCAN)
2647 goto continue_resched;
2650 * On the last scan, we clean up all those tags
2651 * we inserted; but make a note that we still
2652 * found pages pinned.
2657 spin_lock_irq(&mapping->tree_lock);
2658 radix_tree_tag_clear(&mapping->page_tree,
2659 iter.index, SHMEM_TAG_PINNED);
2660 spin_unlock_irq(&mapping->tree_lock);
2662 if (need_resched()) {
2663 slot = radix_tree_iter_resume(slot, &iter);
2673 #define F_ALL_SEALS (F_SEAL_SEAL | \
2678 int shmem_add_seals(struct file *file, unsigned int seals)
2680 struct inode *inode = file_inode(file);
2681 struct shmem_inode_info *info = SHMEM_I(inode);
2686 * Sealing allows multiple parties to share a shmem-file but restrict
2687 * access to a specific subset of file operations. Seals can only be
2688 * added, but never removed. This way, mutually untrusted parties can
2689 * share common memory regions with a well-defined policy. A malicious
2690 * peer can thus never perform unwanted operations on a shared object.
2692 * Seals are only supported on special shmem-files and always affect
2693 * the whole underlying inode. Once a seal is set, it may prevent some
2694 * kinds of access to the file. Currently, the following seals are
2696 * SEAL_SEAL: Prevent further seals from being set on this file
2697 * SEAL_SHRINK: Prevent the file from shrinking
2698 * SEAL_GROW: Prevent the file from growing
2699 * SEAL_WRITE: Prevent write access to the file
2701 * As we don't require any trust relationship between two parties, we
2702 * must prevent seals from being removed. Therefore, sealing a file
2703 * only adds a given set of seals to the file, it never touches
2704 * existing seals. Furthermore, the "setting seals"-operation can be
2705 * sealed itself, which basically prevents any further seal from being
2708 * Semantics of sealing are only defined on volatile files. Only
2709 * anonymous shmem files support sealing. More importantly, seals are
2710 * never written to disk. Therefore, there's no plan to support it on
2714 if (file->f_op != &shmem_file_operations)
2716 if (!(file->f_mode & FMODE_WRITE))
2718 if (seals & ~(unsigned int)F_ALL_SEALS)
2723 if (info->seals & F_SEAL_SEAL) {
2728 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2729 error = mapping_deny_writable(file->f_mapping);
2733 error = shmem_wait_for_pins(file->f_mapping);
2735 mapping_allow_writable(file->f_mapping);
2740 info->seals |= seals;
2744 inode_unlock(inode);
2747 EXPORT_SYMBOL_GPL(shmem_add_seals);
2749 int shmem_get_seals(struct file *file)
2751 if (file->f_op != &shmem_file_operations)
2754 return SHMEM_I(file_inode(file))->seals;
2756 EXPORT_SYMBOL_GPL(shmem_get_seals);
2758 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2764 /* disallow upper 32bit */
2768 error = shmem_add_seals(file, arg);
2771 error = shmem_get_seals(file);
2781 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2784 struct inode *inode = file_inode(file);
2785 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2786 struct shmem_inode_info *info = SHMEM_I(inode);
2787 struct shmem_falloc shmem_falloc;
2788 pgoff_t start, index, end;
2791 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2796 if (mode & FALLOC_FL_PUNCH_HOLE) {
2797 struct address_space *mapping = file->f_mapping;
2798 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2799 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2800 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2802 /* protected by i_mutex */
2803 if (info->seals & F_SEAL_WRITE) {
2808 shmem_falloc.waitq = &shmem_falloc_waitq;
2809 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2810 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2811 spin_lock(&inode->i_lock);
2812 inode->i_private = &shmem_falloc;
2813 spin_unlock(&inode->i_lock);
2815 if ((u64)unmap_end > (u64)unmap_start)
2816 unmap_mapping_range(mapping, unmap_start,
2817 1 + unmap_end - unmap_start, 0);
2818 shmem_truncate_range(inode, offset, offset + len - 1);
2819 /* No need to unmap again: hole-punching leaves COWed pages */
2821 spin_lock(&inode->i_lock);
2822 inode->i_private = NULL;
2823 wake_up_all(&shmem_falloc_waitq);
2824 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
2825 spin_unlock(&inode->i_lock);
2830 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2831 error = inode_newsize_ok(inode, offset + len);
2835 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2840 start = offset >> PAGE_SHIFT;
2841 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2842 /* Try to avoid a swapstorm if len is impossible to satisfy */
2843 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2848 shmem_falloc.waitq = NULL;
2849 shmem_falloc.start = start;
2850 shmem_falloc.next = start;
2851 shmem_falloc.nr_falloced = 0;
2852 shmem_falloc.nr_unswapped = 0;
2853 spin_lock(&inode->i_lock);
2854 inode->i_private = &shmem_falloc;
2855 spin_unlock(&inode->i_lock);
2857 for (index = start; index < end; index++) {
2861 * Good, the fallocate(2) manpage permits EINTR: we may have
2862 * been interrupted because we are using up too much memory.
2864 if (signal_pending(current))
2866 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2869 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2871 /* Remove the !PageUptodate pages we added */
2872 if (index > start) {
2873 shmem_undo_range(inode,
2874 (loff_t)start << PAGE_SHIFT,
2875 ((loff_t)index << PAGE_SHIFT) - 1, true);
2881 * Inform shmem_writepage() how far we have reached.
2882 * No need for lock or barrier: we have the page lock.
2884 shmem_falloc.next++;
2885 if (!PageUptodate(page))
2886 shmem_falloc.nr_falloced++;
2889 * If !PageUptodate, leave it that way so that freeable pages
2890 * can be recognized if we need to rollback on error later.
2891 * But set_page_dirty so that memory pressure will swap rather
2892 * than free the pages we are allocating (and SGP_CACHE pages
2893 * might still be clean: we now need to mark those dirty too).
2895 set_page_dirty(page);
2901 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2902 i_size_write(inode, offset + len);
2903 inode->i_ctime = current_time(inode);
2905 spin_lock(&inode->i_lock);
2906 inode->i_private = NULL;
2907 spin_unlock(&inode->i_lock);
2909 inode_unlock(inode);
2913 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2915 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2917 buf->f_type = TMPFS_MAGIC;
2918 buf->f_bsize = PAGE_SIZE;
2919 buf->f_namelen = NAME_MAX;
2920 if (sbinfo->max_blocks) {
2921 buf->f_blocks = sbinfo->max_blocks;
2923 buf->f_bfree = sbinfo->max_blocks -
2924 percpu_counter_sum(&sbinfo->used_blocks);
2926 if (sbinfo->max_inodes) {
2927 buf->f_files = sbinfo->max_inodes;
2928 buf->f_ffree = sbinfo->free_inodes;
2930 /* else leave those fields 0 like simple_statfs */
2935 * File creation. Allocate an inode, and we're done..
2938 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2940 struct inode *inode;
2941 int error = -ENOSPC;
2943 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2945 error = simple_acl_create(dir, inode);
2948 error = security_inode_init_security(inode, dir,
2950 shmem_initxattrs, NULL);
2951 if (error && error != -EOPNOTSUPP)
2955 dir->i_size += BOGO_DIRENT_SIZE;
2956 dir->i_ctime = dir->i_mtime = current_time(dir);
2957 d_instantiate(dentry, inode);
2958 dget(dentry); /* Extra count - pin the dentry in core */
2967 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2969 struct inode *inode;
2970 int error = -ENOSPC;
2972 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2974 error = security_inode_init_security(inode, dir,
2976 shmem_initxattrs, NULL);
2977 if (error && error != -EOPNOTSUPP)
2979 error = simple_acl_create(dir, inode);
2982 d_tmpfile(dentry, inode);
2990 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2994 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
3000 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
3003 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
3009 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3011 struct inode *inode = d_inode(old_dentry);
3015 * No ordinary (disk based) filesystem counts links as inodes;
3016 * but each new link needs a new dentry, pinning lowmem, and
3017 * tmpfs dentries cannot be pruned until they are unlinked.
3019 ret = shmem_reserve_inode(inode->i_sb);
3023 dir->i_size += BOGO_DIRENT_SIZE;
3024 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3026 ihold(inode); /* New dentry reference */
3027 dget(dentry); /* Extra pinning count for the created dentry */
3028 d_instantiate(dentry, inode);
3033 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3035 struct inode *inode = d_inode(dentry);
3037 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3038 shmem_free_inode(inode->i_sb);
3040 dir->i_size -= BOGO_DIRENT_SIZE;
3041 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3043 dput(dentry); /* Undo the count from "create" - this does all the work */
3047 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3049 if (!simple_empty(dentry))
3052 drop_nlink(d_inode(dentry));
3054 return shmem_unlink(dir, dentry);
3057 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3059 bool old_is_dir = d_is_dir(old_dentry);
3060 bool new_is_dir = d_is_dir(new_dentry);
3062 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3064 drop_nlink(old_dir);
3067 drop_nlink(new_dir);
3071 old_dir->i_ctime = old_dir->i_mtime =
3072 new_dir->i_ctime = new_dir->i_mtime =
3073 d_inode(old_dentry)->i_ctime =
3074 d_inode(new_dentry)->i_ctime = current_time(old_dir);
3079 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3081 struct dentry *whiteout;
3084 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3088 error = shmem_mknod(old_dir, whiteout,
3089 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3095 * Cheat and hash the whiteout while the old dentry is still in
3096 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3098 * d_lookup() will consistently find one of them at this point,
3099 * not sure which one, but that isn't even important.
3106 * The VFS layer already does all the dentry stuff for rename,
3107 * we just have to decrement the usage count for the target if
3108 * it exists so that the VFS layer correctly free's it when it
3111 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3113 struct inode *inode = d_inode(old_dentry);
3114 int they_are_dirs = S_ISDIR(inode->i_mode);
3116 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3119 if (flags & RENAME_EXCHANGE)
3120 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3122 if (!simple_empty(new_dentry))
3125 if (flags & RENAME_WHITEOUT) {
3128 error = shmem_whiteout(old_dir, old_dentry);
3133 if (d_really_is_positive(new_dentry)) {
3134 (void) shmem_unlink(new_dir, new_dentry);
3135 if (they_are_dirs) {
3136 drop_nlink(d_inode(new_dentry));
3137 drop_nlink(old_dir);
3139 } else if (they_are_dirs) {
3140 drop_nlink(old_dir);
3144 old_dir->i_size -= BOGO_DIRENT_SIZE;
3145 new_dir->i_size += BOGO_DIRENT_SIZE;
3146 old_dir->i_ctime = old_dir->i_mtime =
3147 new_dir->i_ctime = new_dir->i_mtime =
3148 inode->i_ctime = current_time(old_dir);
3152 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3156 struct inode *inode;
3158 struct shmem_inode_info *info;
3160 len = strlen(symname) + 1;
3161 if (len > PAGE_SIZE)
3162 return -ENAMETOOLONG;
3164 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3168 error = security_inode_init_security(inode, dir, &dentry->d_name,
3169 shmem_initxattrs, NULL);
3171 if (error != -EOPNOTSUPP) {
3178 info = SHMEM_I(inode);
3179 inode->i_size = len-1;
3180 if (len <= SHORT_SYMLINK_LEN) {
3181 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3182 if (!inode->i_link) {
3186 inode->i_op = &shmem_short_symlink_operations;
3188 inode_nohighmem(inode);
3189 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3194 inode->i_mapping->a_ops = &shmem_aops;
3195 inode->i_op = &shmem_symlink_inode_operations;
3196 memcpy(page_address(page), symname, len);
3197 SetPageUptodate(page);
3198 set_page_dirty(page);
3202 dir->i_size += BOGO_DIRENT_SIZE;
3203 dir->i_ctime = dir->i_mtime = current_time(dir);
3204 d_instantiate(dentry, inode);
3209 static void shmem_put_link(void *arg)
3211 mark_page_accessed(arg);
3215 static const char *shmem_get_link(struct dentry *dentry,
3216 struct inode *inode,
3217 struct delayed_call *done)
3219 struct page *page = NULL;
3222 page = find_get_page(inode->i_mapping, 0);
3224 return ERR_PTR(-ECHILD);
3225 if (!PageUptodate(page)) {
3227 return ERR_PTR(-ECHILD);
3230 error = shmem_getpage(inode, 0, &page, SGP_READ);
3232 return ERR_PTR(error);
3235 set_delayed_call(done, shmem_put_link, page);
3236 return page_address(page);
3239 #ifdef CONFIG_TMPFS_XATTR
3241 * Superblocks without xattr inode operations may get some security.* xattr
3242 * support from the LSM "for free". As soon as we have any other xattrs
3243 * like ACLs, we also need to implement the security.* handlers at
3244 * filesystem level, though.
3248 * Callback for security_inode_init_security() for acquiring xattrs.
3250 static int shmem_initxattrs(struct inode *inode,
3251 const struct xattr *xattr_array,
3254 struct shmem_inode_info *info = SHMEM_I(inode);
3255 const struct xattr *xattr;
3256 struct simple_xattr *new_xattr;
3259 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3260 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3264 len = strlen(xattr->name) + 1;
3265 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3267 if (!new_xattr->name) {
3272 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3273 XATTR_SECURITY_PREFIX_LEN);
3274 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3277 simple_xattr_list_add(&info->xattrs, new_xattr);
3283 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3284 struct dentry *unused, struct inode *inode,
3285 const char *name, void *buffer, size_t size)
3287 struct shmem_inode_info *info = SHMEM_I(inode);
3289 name = xattr_full_name(handler, name);
3290 return simple_xattr_get(&info->xattrs, name, buffer, size);
3293 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3294 struct dentry *unused, struct inode *inode,
3295 const char *name, const void *value,
3296 size_t size, int flags)
3298 struct shmem_inode_info *info = SHMEM_I(inode);
3300 name = xattr_full_name(handler, name);
3301 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3304 static const struct xattr_handler shmem_security_xattr_handler = {
3305 .prefix = XATTR_SECURITY_PREFIX,
3306 .get = shmem_xattr_handler_get,
3307 .set = shmem_xattr_handler_set,
3310 static const struct xattr_handler shmem_trusted_xattr_handler = {
3311 .prefix = XATTR_TRUSTED_PREFIX,
3312 .get = shmem_xattr_handler_get,
3313 .set = shmem_xattr_handler_set,
3316 static const struct xattr_handler *shmem_xattr_handlers[] = {
3317 #ifdef CONFIG_TMPFS_POSIX_ACL
3318 &posix_acl_access_xattr_handler,
3319 &posix_acl_default_xattr_handler,
3321 &shmem_security_xattr_handler,
3322 &shmem_trusted_xattr_handler,
3326 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3328 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3329 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3331 #endif /* CONFIG_TMPFS_XATTR */
3333 static const struct inode_operations shmem_short_symlink_operations = {
3334 .get_link = simple_get_link,
3335 #ifdef CONFIG_TMPFS_XATTR
3336 .listxattr = shmem_listxattr,
3340 static const struct inode_operations shmem_symlink_inode_operations = {
3341 .get_link = shmem_get_link,
3342 #ifdef CONFIG_TMPFS_XATTR
3343 .listxattr = shmem_listxattr,
3347 static struct dentry *shmem_get_parent(struct dentry *child)
3349 return ERR_PTR(-ESTALE);
3352 static int shmem_match(struct inode *ino, void *vfh)
3356 inum = (inum << 32) | fh[1];
3357 return ino->i_ino == inum && fh[0] == ino->i_generation;
3360 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3361 struct fid *fid, int fh_len, int fh_type)
3363 struct inode *inode;
3364 struct dentry *dentry = NULL;
3371 inum = (inum << 32) | fid->raw[1];
3373 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3374 shmem_match, fid->raw);
3376 dentry = d_find_alias(inode);
3383 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3384 struct inode *parent)
3388 return FILEID_INVALID;
3391 if (inode_unhashed(inode)) {
3392 /* Unfortunately insert_inode_hash is not idempotent,
3393 * so as we hash inodes here rather than at creation
3394 * time, we need a lock to ensure we only try
3397 static DEFINE_SPINLOCK(lock);
3399 if (inode_unhashed(inode))
3400 __insert_inode_hash(inode,
3401 inode->i_ino + inode->i_generation);
3405 fh[0] = inode->i_generation;
3406 fh[1] = inode->i_ino;
3407 fh[2] = ((__u64)inode->i_ino) >> 32;
3413 static const struct export_operations shmem_export_ops = {
3414 .get_parent = shmem_get_parent,
3415 .encode_fh = shmem_encode_fh,
3416 .fh_to_dentry = shmem_fh_to_dentry,
3419 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3422 char *this_char, *value, *rest;
3423 struct mempolicy *mpol = NULL;
3427 while (options != NULL) {
3428 this_char = options;
3431 * NUL-terminate this option: unfortunately,
3432 * mount options form a comma-separated list,
3433 * but mpol's nodelist may also contain commas.
3435 options = strchr(options, ',');
3436 if (options == NULL)
3439 if (!isdigit(*options)) {
3446 if ((value = strchr(this_char,'=')) != NULL) {
3449 pr_err("tmpfs: No value for mount option '%s'\n",
3454 if (!strcmp(this_char,"size")) {
3455 unsigned long long size;
3456 size = memparse(value,&rest);
3458 size <<= PAGE_SHIFT;
3459 size *= totalram_pages;
3465 sbinfo->max_blocks =
3466 DIV_ROUND_UP(size, PAGE_SIZE);
3467 } else if (!strcmp(this_char,"nr_blocks")) {
3468 sbinfo->max_blocks = memparse(value, &rest);
3471 } else if (!strcmp(this_char,"nr_inodes")) {
3472 sbinfo->max_inodes = memparse(value, &rest);
3475 } else if (!strcmp(this_char,"mode")) {
3478 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3481 } else if (!strcmp(this_char,"uid")) {
3484 uid = simple_strtoul(value, &rest, 0);
3487 sbinfo->uid = make_kuid(current_user_ns(), uid);
3488 if (!uid_valid(sbinfo->uid))
3490 } else if (!strcmp(this_char,"gid")) {
3493 gid = simple_strtoul(value, &rest, 0);
3496 sbinfo->gid = make_kgid(current_user_ns(), gid);
3497 if (!gid_valid(sbinfo->gid))
3499 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3500 } else if (!strcmp(this_char, "huge")) {
3502 huge = shmem_parse_huge(value);
3505 if (!has_transparent_hugepage() &&
3506 huge != SHMEM_HUGE_NEVER)
3508 sbinfo->huge = huge;
3511 } else if (!strcmp(this_char,"mpol")) {
3514 if (mpol_parse_str(value, &mpol))
3518 pr_err("tmpfs: Bad mount option %s\n", this_char);
3522 sbinfo->mpol = mpol;
3526 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3534 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3536 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3537 struct shmem_sb_info config = *sbinfo;
3538 unsigned long inodes;
3539 int error = -EINVAL;
3542 if (shmem_parse_options(data, &config, true))
3545 spin_lock(&sbinfo->stat_lock);
3546 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3547 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3549 if (config.max_inodes < inodes)
3552 * Those tests disallow limited->unlimited while any are in use;
3553 * but we must separately disallow unlimited->limited, because
3554 * in that case we have no record of how much is already in use.
3556 if (config.max_blocks && !sbinfo->max_blocks)
3558 if (config.max_inodes && !sbinfo->max_inodes)
3562 sbinfo->huge = config.huge;
3563 sbinfo->max_blocks = config.max_blocks;
3564 sbinfo->max_inodes = config.max_inodes;
3565 sbinfo->free_inodes = config.max_inodes - inodes;
3568 * Preserve previous mempolicy unless mpol remount option was specified.
3571 mpol_put(sbinfo->mpol);
3572 sbinfo->mpol = config.mpol; /* transfers initial ref */
3575 spin_unlock(&sbinfo->stat_lock);
3579 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3581 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3583 if (sbinfo->max_blocks != shmem_default_max_blocks())
3584 seq_printf(seq, ",size=%luk",
3585 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3586 if (sbinfo->max_inodes != shmem_default_max_inodes())
3587 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3588 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3589 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3590 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3591 seq_printf(seq, ",uid=%u",
3592 from_kuid_munged(&init_user_ns, sbinfo->uid));
3593 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3594 seq_printf(seq, ",gid=%u",
3595 from_kgid_munged(&init_user_ns, sbinfo->gid));
3596 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3597 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3599 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3601 shmem_show_mpol(seq, sbinfo->mpol);
3605 #define MFD_NAME_PREFIX "memfd:"
3606 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3607 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3609 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3611 SYSCALL_DEFINE2(memfd_create,
3612 const char __user *, uname,
3613 unsigned int, flags)
3615 struct shmem_inode_info *info;
3621 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3624 /* length includes terminating zero */
3625 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3628 if (len > MFD_NAME_MAX_LEN + 1)
3631 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3635 strcpy(name, MFD_NAME_PREFIX);
3636 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3641 /* terminating-zero may have changed after strnlen_user() returned */
3642 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3647 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3653 file = shmem_file_setup(name, 0, VM_NORESERVE);
3655 error = PTR_ERR(file);
3658 info = SHMEM_I(file_inode(file));
3659 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3660 file->f_flags |= O_RDWR | O_LARGEFILE;
3661 if (flags & MFD_ALLOW_SEALING)
3662 info->seals &= ~F_SEAL_SEAL;
3664 fd_install(fd, file);
3675 #endif /* CONFIG_TMPFS */
3677 static void shmem_put_super(struct super_block *sb)
3679 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3681 percpu_counter_destroy(&sbinfo->used_blocks);
3682 mpol_put(sbinfo->mpol);
3684 sb->s_fs_info = NULL;
3687 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3689 struct inode *inode;
3690 struct shmem_sb_info *sbinfo;
3693 /* Round up to L1_CACHE_BYTES to resist false sharing */
3694 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3695 L1_CACHE_BYTES), GFP_KERNEL);
3699 sbinfo->mode = S_IRWXUGO | S_ISVTX;
3700 sbinfo->uid = current_fsuid();
3701 sbinfo->gid = current_fsgid();
3702 sb->s_fs_info = sbinfo;
3706 * Per default we only allow half of the physical ram per
3707 * tmpfs instance, limiting inodes to one per page of lowmem;
3708 * but the internal instance is left unlimited.
3710 if (!(sb->s_flags & MS_KERNMOUNT)) {
3711 sbinfo->max_blocks = shmem_default_max_blocks();
3712 sbinfo->max_inodes = shmem_default_max_inodes();
3713 if (shmem_parse_options(data, sbinfo, false)) {
3718 sb->s_flags |= MS_NOUSER;
3720 sb->s_export_op = &shmem_export_ops;
3721 sb->s_flags |= MS_NOSEC;
3723 sb->s_flags |= MS_NOUSER;
3726 spin_lock_init(&sbinfo->stat_lock);
3727 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3729 sbinfo->free_inodes = sbinfo->max_inodes;
3730 spin_lock_init(&sbinfo->shrinklist_lock);
3731 INIT_LIST_HEAD(&sbinfo->shrinklist);
3733 sb->s_maxbytes = MAX_LFS_FILESIZE;
3734 sb->s_blocksize = PAGE_SIZE;
3735 sb->s_blocksize_bits = PAGE_SHIFT;
3736 sb->s_magic = TMPFS_MAGIC;
3737 sb->s_op = &shmem_ops;
3738 sb->s_time_gran = 1;
3739 #ifdef CONFIG_TMPFS_XATTR
3740 sb->s_xattr = shmem_xattr_handlers;
3742 #ifdef CONFIG_TMPFS_POSIX_ACL
3743 sb->s_flags |= MS_POSIXACL;
3746 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3749 inode->i_uid = sbinfo->uid;
3750 inode->i_gid = sbinfo->gid;
3751 sb->s_root = d_make_root(inode);
3757 shmem_put_super(sb);
3761 static struct kmem_cache *shmem_inode_cachep;
3763 static struct inode *shmem_alloc_inode(struct super_block *sb)
3765 struct shmem_inode_info *info;
3766 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3769 return &info->vfs_inode;
3772 static void shmem_destroy_callback(struct rcu_head *head)
3774 struct inode *inode = container_of(head, struct inode, i_rcu);
3775 if (S_ISLNK(inode->i_mode))
3776 kfree(inode->i_link);
3777 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3780 static void shmem_destroy_inode(struct inode *inode)
3782 if (S_ISREG(inode->i_mode))
3783 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3784 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3787 static void shmem_init_inode(void *foo)
3789 struct shmem_inode_info *info = foo;
3790 inode_init_once(&info->vfs_inode);
3793 static int shmem_init_inodecache(void)
3795 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3796 sizeof(struct shmem_inode_info),
3797 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3801 static void shmem_destroy_inodecache(void)
3803 kmem_cache_destroy(shmem_inode_cachep);
3806 static const struct address_space_operations shmem_aops = {
3807 .writepage = shmem_writepage,
3808 .set_page_dirty = __set_page_dirty_no_writeback,
3810 .write_begin = shmem_write_begin,
3811 .write_end = shmem_write_end,
3813 #ifdef CONFIG_MIGRATION
3814 .migratepage = migrate_page,
3816 .error_remove_page = generic_error_remove_page,
3819 static const struct file_operations shmem_file_operations = {
3821 .get_unmapped_area = shmem_get_unmapped_area,
3823 .llseek = shmem_file_llseek,
3824 .read_iter = shmem_file_read_iter,
3825 .write_iter = generic_file_write_iter,
3826 .fsync = noop_fsync,
3827 .splice_read = generic_file_splice_read,
3828 .splice_write = iter_file_splice_write,
3829 .fallocate = shmem_fallocate,
3833 static const struct inode_operations shmem_inode_operations = {
3834 .getattr = shmem_getattr,
3835 .setattr = shmem_setattr,
3836 #ifdef CONFIG_TMPFS_XATTR
3837 .listxattr = shmem_listxattr,
3838 .set_acl = simple_set_acl,
3842 static const struct inode_operations shmem_dir_inode_operations = {
3844 .create = shmem_create,
3845 .lookup = simple_lookup,
3847 .unlink = shmem_unlink,
3848 .symlink = shmem_symlink,
3849 .mkdir = shmem_mkdir,
3850 .rmdir = shmem_rmdir,
3851 .mknod = shmem_mknod,
3852 .rename = shmem_rename2,
3853 .tmpfile = shmem_tmpfile,
3855 #ifdef CONFIG_TMPFS_XATTR
3856 .listxattr = shmem_listxattr,
3858 #ifdef CONFIG_TMPFS_POSIX_ACL
3859 .setattr = shmem_setattr,
3860 .set_acl = simple_set_acl,
3864 static const struct inode_operations shmem_special_inode_operations = {
3865 #ifdef CONFIG_TMPFS_XATTR
3866 .listxattr = shmem_listxattr,
3868 #ifdef CONFIG_TMPFS_POSIX_ACL
3869 .setattr = shmem_setattr,
3870 .set_acl = simple_set_acl,
3874 static const struct super_operations shmem_ops = {
3875 .alloc_inode = shmem_alloc_inode,
3876 .destroy_inode = shmem_destroy_inode,
3878 .statfs = shmem_statfs,
3879 .remount_fs = shmem_remount_fs,
3880 .show_options = shmem_show_options,
3882 .evict_inode = shmem_evict_inode,
3883 .drop_inode = generic_delete_inode,
3884 .put_super = shmem_put_super,
3885 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3886 .nr_cached_objects = shmem_unused_huge_count,
3887 .free_cached_objects = shmem_unused_huge_scan,
3891 static const struct vm_operations_struct shmem_vm_ops = {
3892 .fault = shmem_fault,
3893 .map_pages = filemap_map_pages,
3895 .set_policy = shmem_set_policy,
3896 .get_policy = shmem_get_policy,
3900 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3901 int flags, const char *dev_name, void *data)
3903 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3906 static struct file_system_type shmem_fs_type = {
3907 .owner = THIS_MODULE,
3909 .mount = shmem_mount,
3910 .kill_sb = kill_litter_super,
3911 .fs_flags = FS_USERNS_MOUNT,
3914 int __init shmem_init(void)
3918 /* If rootfs called this, don't re-init */
3919 if (shmem_inode_cachep)
3922 error = shmem_init_inodecache();
3926 error = register_filesystem(&shmem_fs_type);
3928 pr_err("Could not register tmpfs\n");
3932 shm_mnt = kern_mount(&shmem_fs_type);
3933 if (IS_ERR(shm_mnt)) {
3934 error = PTR_ERR(shm_mnt);
3935 pr_err("Could not kern_mount tmpfs\n");
3939 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3940 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3941 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3943 shmem_huge = 0; /* just in case it was patched */
3948 unregister_filesystem(&shmem_fs_type);
3950 shmem_destroy_inodecache();
3952 shm_mnt = ERR_PTR(error);
3956 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3957 static ssize_t shmem_enabled_show(struct kobject *kobj,
3958 struct kobj_attribute *attr, char *buf)
3962 SHMEM_HUGE_WITHIN_SIZE,
3970 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3971 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3973 count += sprintf(buf + count, fmt,
3974 shmem_format_huge(values[i]));
3976 buf[count - 1] = '\n';
3980 static ssize_t shmem_enabled_store(struct kobject *kobj,
3981 struct kobj_attribute *attr, const char *buf, size_t count)
3986 if (count + 1 > sizeof(tmp))
3988 memcpy(tmp, buf, count);
3990 if (count && tmp[count - 1] == '\n')
3991 tmp[count - 1] = '\0';
3993 huge = shmem_parse_huge(tmp);
3994 if (huge == -EINVAL)
3996 if (!has_transparent_hugepage() &&
3997 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4001 if (shmem_huge < SHMEM_HUGE_DENY)
4002 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4006 struct kobj_attribute shmem_enabled_attr =
4007 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4008 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4010 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4011 bool shmem_huge_enabled(struct vm_area_struct *vma)
4013 struct inode *inode = file_inode(vma->vm_file);
4014 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4018 if (shmem_huge == SHMEM_HUGE_FORCE)
4020 if (shmem_huge == SHMEM_HUGE_DENY)
4022 switch (sbinfo->huge) {
4023 case SHMEM_HUGE_NEVER:
4025 case SHMEM_HUGE_ALWAYS:
4027 case SHMEM_HUGE_WITHIN_SIZE:
4028 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4029 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4030 if (i_size >= HPAGE_PMD_SIZE &&
4031 i_size >> PAGE_SHIFT >= off)
4033 case SHMEM_HUGE_ADVISE:
4034 /* TODO: implement fadvise() hints */
4035 return (vma->vm_flags & VM_HUGEPAGE);
4041 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4043 #else /* !CONFIG_SHMEM */
4046 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4048 * This is intended for small system where the benefits of the full
4049 * shmem code (swap-backed and resource-limited) are outweighed by
4050 * their complexity. On systems without swap this code should be
4051 * effectively equivalent, but much lighter weight.
4054 static struct file_system_type shmem_fs_type = {
4056 .mount = ramfs_mount,
4057 .kill_sb = kill_litter_super,
4058 .fs_flags = FS_USERNS_MOUNT,
4061 int __init shmem_init(void)
4063 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4065 shm_mnt = kern_mount(&shmem_fs_type);
4066 BUG_ON(IS_ERR(shm_mnt));
4071 int shmem_unuse(swp_entry_t swap, struct page *page)
4076 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4081 void shmem_unlock_mapping(struct address_space *mapping)
4086 unsigned long shmem_get_unmapped_area(struct file *file,
4087 unsigned long addr, unsigned long len,
4088 unsigned long pgoff, unsigned long flags)
4090 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4094 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4096 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4098 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4100 #define shmem_vm_ops generic_file_vm_ops
4101 #define shmem_file_operations ramfs_file_operations
4102 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4103 #define shmem_acct_size(flags, size) 0
4104 #define shmem_unacct_size(flags, size) do {} while (0)
4106 #endif /* CONFIG_SHMEM */
4110 static const struct dentry_operations anon_ops = {
4111 .d_dname = simple_dname
4114 static struct file *__shmem_file_setup(const char *name, loff_t size,
4115 unsigned long flags, unsigned int i_flags)
4118 struct inode *inode;
4120 struct super_block *sb;
4123 if (IS_ERR(shm_mnt))
4124 return ERR_CAST(shm_mnt);
4126 if (size < 0 || size > MAX_LFS_FILESIZE)
4127 return ERR_PTR(-EINVAL);
4129 if (shmem_acct_size(flags, size))
4130 return ERR_PTR(-ENOMEM);
4132 res = ERR_PTR(-ENOMEM);
4134 this.len = strlen(name);
4135 this.hash = 0; /* will go */
4136 sb = shm_mnt->mnt_sb;
4137 path.mnt = mntget(shm_mnt);
4138 path.dentry = d_alloc_pseudo(sb, &this);
4141 d_set_d_op(path.dentry, &anon_ops);
4143 res = ERR_PTR(-ENOSPC);
4144 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4148 inode->i_flags |= i_flags;
4149 d_instantiate(path.dentry, inode);
4150 inode->i_size = size;
4151 clear_nlink(inode); /* It is unlinked */
4152 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4156 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4157 &shmem_file_operations);
4164 shmem_unacct_size(flags, size);
4171 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4172 * kernel internal. There will be NO LSM permission checks against the
4173 * underlying inode. So users of this interface must do LSM checks at a
4174 * higher layer. The users are the big_key and shm implementations. LSM
4175 * checks are provided at the key or shm level rather than the inode.
4176 * @name: name for dentry (to be seen in /proc/<pid>/maps
4177 * @size: size to be set for the file
4178 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4180 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4182 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4186 * shmem_file_setup - get an unlinked file living in tmpfs
4187 * @name: name for dentry (to be seen in /proc/<pid>/maps
4188 * @size: size to be set for the file
4189 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4191 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4193 return __shmem_file_setup(name, size, flags, 0);
4195 EXPORT_SYMBOL_GPL(shmem_file_setup);
4198 * shmem_zero_setup - setup a shared anonymous mapping
4199 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4201 int shmem_zero_setup(struct vm_area_struct *vma)
4204 loff_t size = vma->vm_end - vma->vm_start;
4207 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4208 * between XFS directory reading and selinux: since this file is only
4209 * accessible to the user through its mapping, use S_PRIVATE flag to
4210 * bypass file security, in the same way as shmem_kernel_file_setup().
4212 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4214 return PTR_ERR(file);
4218 vma->vm_file = file;
4219 vma->vm_ops = &shmem_vm_ops;
4221 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4222 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4223 (vma->vm_end & HPAGE_PMD_MASK)) {
4224 khugepaged_enter(vma, vma->vm_flags);
4231 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4232 * @mapping: the page's address_space
4233 * @index: the page index
4234 * @gfp: the page allocator flags to use if allocating
4236 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4237 * with any new page allocations done using the specified allocation flags.
4238 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4239 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4240 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4242 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4243 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4245 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4246 pgoff_t index, gfp_t gfp)
4249 struct inode *inode = mapping->host;
4253 BUG_ON(mapping->a_ops != &shmem_aops);
4254 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4257 page = ERR_PTR(error);
4263 * The tiny !SHMEM case uses ramfs without swap
4265 return read_cache_page_gfp(mapping, index, gfp);
4268 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);