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>
74 #include <asm/uaccess.h>
75 #include <asm/pgtable.h>
79 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
80 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
82 /* Pretend that each entry is of this size in directory's i_size */
83 #define BOGO_DIRENT_SIZE 20
85 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
86 #define SHORT_SYMLINK_LEN 128
89 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
90 * inode->i_private (with i_mutex making sure that it has only one user at
91 * a time): we would prefer not to enlarge the shmem inode just for that.
94 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
95 pgoff_t start; /* start of range currently being fallocated */
96 pgoff_t next; /* the next page offset to be fallocated */
97 pgoff_t nr_falloced; /* how many new pages have been fallocated */
98 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
102 static unsigned long shmem_default_max_blocks(void)
104 return totalram_pages / 2;
107 static unsigned long shmem_default_max_inodes(void)
109 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
113 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
114 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
115 struct shmem_inode_info *info, pgoff_t index);
116 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
117 struct page **pagep, enum sgp_type sgp,
118 gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
120 int shmem_getpage(struct inode *inode, pgoff_t index,
121 struct page **pagep, enum sgp_type sgp)
123 return shmem_getpage_gfp(inode, index, pagep, sgp,
124 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
127 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
129 return sb->s_fs_info;
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
138 static inline int shmem_acct_size(unsigned long flags, loff_t size)
140 return (flags & VM_NORESERVE) ?
141 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
144 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
146 if (!(flags & VM_NORESERVE))
147 vm_unacct_memory(VM_ACCT(size));
150 static inline int shmem_reacct_size(unsigned long flags,
151 loff_t oldsize, loff_t newsize)
153 if (!(flags & VM_NORESERVE)) {
154 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
155 return security_vm_enough_memory_mm(current->mm,
156 VM_ACCT(newsize) - VM_ACCT(oldsize));
157 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
158 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
164 * ... whereas tmpfs objects are accounted incrementally as
165 * pages are allocated, in order to allow large sparse files.
166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169 static inline int shmem_acct_block(unsigned long flags, long pages)
171 if (!(flags & VM_NORESERVE))
174 return security_vm_enough_memory_mm(current->mm,
175 pages * VM_ACCT(PAGE_SIZE));
178 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
180 if (flags & VM_NORESERVE)
181 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
184 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
186 struct shmem_inode_info *info = SHMEM_I(inode);
187 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
189 if (shmem_acct_block(info->flags, pages))
192 if (sbinfo->max_blocks) {
193 if (percpu_counter_compare(&sbinfo->used_blocks,
194 sbinfo->max_blocks - pages) > 0)
196 percpu_counter_add(&sbinfo->used_blocks, pages);
202 shmem_unacct_blocks(info->flags, pages);
206 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
208 struct shmem_inode_info *info = SHMEM_I(inode);
209 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
211 if (sbinfo->max_blocks)
212 percpu_counter_sub(&sbinfo->used_blocks, pages);
213 shmem_unacct_blocks(info->flags, pages);
216 static const struct super_operations shmem_ops;
217 static const struct address_space_operations shmem_aops;
218 static const struct file_operations shmem_file_operations;
219 static const struct inode_operations shmem_inode_operations;
220 static const struct inode_operations shmem_dir_inode_operations;
221 static const struct inode_operations shmem_special_inode_operations;
222 static const struct vm_operations_struct shmem_vm_ops;
223 static struct file_system_type shmem_fs_type;
225 static LIST_HEAD(shmem_swaplist);
226 static DEFINE_MUTEX(shmem_swaplist_mutex);
228 static int shmem_reserve_inode(struct super_block *sb)
230 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
231 if (sbinfo->max_inodes) {
232 spin_lock(&sbinfo->stat_lock);
233 if (!sbinfo->free_inodes) {
234 spin_unlock(&sbinfo->stat_lock);
237 sbinfo->free_inodes--;
238 spin_unlock(&sbinfo->stat_lock);
243 static void shmem_free_inode(struct super_block *sb)
245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246 if (sbinfo->max_inodes) {
247 spin_lock(&sbinfo->stat_lock);
248 sbinfo->free_inodes++;
249 spin_unlock(&sbinfo->stat_lock);
254 * shmem_recalc_inode - recalculate the block usage of an inode
255 * @inode: inode to recalc
257 * We have to calculate the free blocks since the mm can drop
258 * undirtied hole pages behind our back.
260 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
261 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
263 * It has to be called with the spinlock held.
265 static void shmem_recalc_inode(struct inode *inode)
267 struct shmem_inode_info *info = SHMEM_I(inode);
270 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
272 info->alloced -= freed;
273 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
274 shmem_inode_unacct_blocks(inode, freed);
278 bool shmem_charge(struct inode *inode, long pages)
280 struct shmem_inode_info *info = SHMEM_I(inode);
283 if (!shmem_inode_acct_block(inode, pages))
286 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
287 inode->i_mapping->nrpages += pages;
289 spin_lock_irqsave(&info->lock, flags);
290 info->alloced += pages;
291 inode->i_blocks += pages * BLOCKS_PER_PAGE;
292 shmem_recalc_inode(inode);
293 spin_unlock_irqrestore(&info->lock, flags);
298 void shmem_uncharge(struct inode *inode, long pages)
300 struct shmem_inode_info *info = SHMEM_I(inode);
303 /* nrpages adjustment done by __delete_from_page_cache() or caller */
305 spin_lock_irqsave(&info->lock, flags);
306 info->alloced -= pages;
307 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
308 shmem_recalc_inode(inode);
309 spin_unlock_irqrestore(&info->lock, flags);
311 shmem_inode_unacct_blocks(inode, pages);
315 * Replace item expected in radix tree by a new item, while holding tree lock.
317 static int shmem_radix_tree_replace(struct address_space *mapping,
318 pgoff_t index, void *expected, void *replacement)
323 VM_BUG_ON(!expected);
324 VM_BUG_ON(!replacement);
325 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
328 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
329 if (item != expected)
331 radix_tree_replace_slot(pslot, replacement);
336 * Sometimes, before we decide whether to proceed or to fail, we must check
337 * that an entry was not already brought back from swap by a racing thread.
339 * Checking page is not enough: by the time a SwapCache page is locked, it
340 * might be reused, and again be SwapCache, using the same swap as before.
342 static bool shmem_confirm_swap(struct address_space *mapping,
343 pgoff_t index, swp_entry_t swap)
348 item = radix_tree_lookup(&mapping->page_tree, index);
350 return item == swp_to_radix_entry(swap);
354 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
357 * disables huge pages for the mount;
359 * enables huge pages for the mount;
360 * SHMEM_HUGE_WITHIN_SIZE:
361 * only allocate huge pages if the page will be fully within i_size,
362 * also respect fadvise()/madvise() hints;
364 * only allocate huge pages if requested with fadvise()/madvise();
367 #define SHMEM_HUGE_NEVER 0
368 #define SHMEM_HUGE_ALWAYS 1
369 #define SHMEM_HUGE_WITHIN_SIZE 2
370 #define SHMEM_HUGE_ADVISE 3
374 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
377 * disables huge on shm_mnt and all mounts, for emergency use;
379 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
382 #define SHMEM_HUGE_DENY (-1)
383 #define SHMEM_HUGE_FORCE (-2)
385 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
386 /* ifdef here to avoid bloating shmem.o when not necessary */
388 int shmem_huge __read_mostly;
390 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
391 static int shmem_parse_huge(const char *str)
393 if (!strcmp(str, "never"))
394 return SHMEM_HUGE_NEVER;
395 if (!strcmp(str, "always"))
396 return SHMEM_HUGE_ALWAYS;
397 if (!strcmp(str, "within_size"))
398 return SHMEM_HUGE_WITHIN_SIZE;
399 if (!strcmp(str, "advise"))
400 return SHMEM_HUGE_ADVISE;
401 if (!strcmp(str, "deny"))
402 return SHMEM_HUGE_DENY;
403 if (!strcmp(str, "force"))
404 return SHMEM_HUGE_FORCE;
408 static const char *shmem_format_huge(int huge)
411 case SHMEM_HUGE_NEVER:
413 case SHMEM_HUGE_ALWAYS:
415 case SHMEM_HUGE_WITHIN_SIZE:
416 return "within_size";
417 case SHMEM_HUGE_ADVISE:
419 case SHMEM_HUGE_DENY:
421 case SHMEM_HUGE_FORCE:
430 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
431 struct shrink_control *sc, unsigned long nr_to_split)
433 LIST_HEAD(list), *pos, *next;
434 LIST_HEAD(to_remove);
436 struct shmem_inode_info *info;
438 unsigned long batch = sc ? sc->nr_to_scan : 128;
439 int removed = 0, split = 0;
441 if (list_empty(&sbinfo->shrinklist))
444 spin_lock(&sbinfo->shrinklist_lock);
445 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
446 info = list_entry(pos, struct shmem_inode_info, shrinklist);
449 inode = igrab(&info->vfs_inode);
451 /* inode is about to be evicted */
453 list_del_init(&info->shrinklist);
458 /* Check if there's anything to gain */
459 if (round_up(inode->i_size, PAGE_SIZE) ==
460 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
461 list_move(&info->shrinklist, &to_remove);
466 list_move(&info->shrinklist, &list);
471 spin_unlock(&sbinfo->shrinklist_lock);
473 list_for_each_safe(pos, next, &to_remove) {
474 info = list_entry(pos, struct shmem_inode_info, shrinklist);
475 inode = &info->vfs_inode;
476 list_del_init(&info->shrinklist);
480 list_for_each_safe(pos, next, &list) {
483 info = list_entry(pos, struct shmem_inode_info, shrinklist);
484 inode = &info->vfs_inode;
486 if (nr_to_split && split >= nr_to_split)
489 page = find_get_page(inode->i_mapping,
490 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
494 /* No huge page at the end of the file: nothing to split */
495 if (!PageTransHuge(page)) {
501 * Leave the inode on the list if we failed to lock
502 * the page at this time.
504 * Waiting for the lock may lead to deadlock in the
507 if (!trylock_page(page)) {
512 ret = split_huge_page(page);
516 /* If split failed leave the inode on the list */
522 list_del_init(&info->shrinklist);
528 spin_lock(&sbinfo->shrinklist_lock);
529 list_splice_tail(&list, &sbinfo->shrinklist);
530 sbinfo->shrinklist_len -= removed;
531 spin_unlock(&sbinfo->shrinklist_lock);
536 static long shmem_unused_huge_scan(struct super_block *sb,
537 struct shrink_control *sc)
539 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
541 if (!READ_ONCE(sbinfo->shrinklist_len))
544 return shmem_unused_huge_shrink(sbinfo, sc, 0);
547 static long shmem_unused_huge_count(struct super_block *sb,
548 struct shrink_control *sc)
550 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
551 return READ_ONCE(sbinfo->shrinklist_len);
553 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
555 #define shmem_huge SHMEM_HUGE_DENY
557 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
558 struct shrink_control *sc, unsigned long nr_to_split)
562 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
565 * Like add_to_page_cache_locked, but error if expected item has gone.
567 static int shmem_add_to_page_cache(struct page *page,
568 struct address_space *mapping,
569 pgoff_t index, void *expected)
571 int error, nr = hpage_nr_pages(page);
573 VM_BUG_ON_PAGE(PageTail(page), page);
574 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
575 VM_BUG_ON_PAGE(!PageLocked(page), page);
576 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
577 VM_BUG_ON(expected && PageTransHuge(page));
579 page_ref_add(page, nr);
580 page->mapping = mapping;
583 spin_lock_irq(&mapping->tree_lock);
584 if (PageTransHuge(page)) {
585 void __rcu **results;
590 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
591 &results, &idx, index, 1) &&
592 idx < index + HPAGE_PMD_NR) {
597 for (i = 0; i < HPAGE_PMD_NR; i++) {
598 error = radix_tree_insert(&mapping->page_tree,
599 index + i, page + i);
602 count_vm_event(THP_FILE_ALLOC);
604 } else if (!expected) {
605 error = radix_tree_insert(&mapping->page_tree, index, page);
607 error = shmem_radix_tree_replace(mapping, index, expected,
612 mapping->nrpages += nr;
613 if (PageTransHuge(page))
614 __inc_node_page_state(page, NR_SHMEM_THPS);
615 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
616 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
617 spin_unlock_irq(&mapping->tree_lock);
619 page->mapping = NULL;
620 spin_unlock_irq(&mapping->tree_lock);
621 page_ref_sub(page, nr);
627 * Like delete_from_page_cache, but substitutes swap for page.
629 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
631 struct address_space *mapping = page->mapping;
634 VM_BUG_ON_PAGE(PageCompound(page), page);
636 spin_lock_irq(&mapping->tree_lock);
637 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
638 page->mapping = NULL;
640 __dec_node_page_state(page, NR_FILE_PAGES);
641 __dec_node_page_state(page, NR_SHMEM);
642 spin_unlock_irq(&mapping->tree_lock);
648 * Remove swap entry from radix tree, free the swap and its page cache.
650 static int shmem_free_swap(struct address_space *mapping,
651 pgoff_t index, void *radswap)
655 spin_lock_irq(&mapping->tree_lock);
656 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
657 spin_unlock_irq(&mapping->tree_lock);
660 free_swap_and_cache(radix_to_swp_entry(radswap));
665 * Determine (in bytes) how many of the shmem object's pages mapped by the
666 * given offsets are swapped out.
668 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
669 * as long as the inode doesn't go away and racy results are not a problem.
671 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
672 pgoff_t start, pgoff_t end)
674 struct radix_tree_iter iter;
677 unsigned long swapped = 0;
681 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
682 if (iter.index >= end)
685 page = radix_tree_deref_slot(slot);
687 if (radix_tree_deref_retry(page)) {
688 slot = radix_tree_iter_retry(&iter);
692 if (radix_tree_exceptional_entry(page))
695 if (need_resched()) {
697 slot = radix_tree_iter_next(&iter);
703 return swapped << PAGE_SHIFT;
707 * Determine (in bytes) how many of the shmem object's pages mapped by the
708 * given vma is swapped out.
710 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
711 * as long as the inode doesn't go away and racy results are not a problem.
713 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
715 struct inode *inode = file_inode(vma->vm_file);
716 struct shmem_inode_info *info = SHMEM_I(inode);
717 struct address_space *mapping = inode->i_mapping;
718 unsigned long swapped;
720 /* Be careful as we don't hold info->lock */
721 swapped = READ_ONCE(info->swapped);
724 * The easier cases are when the shmem object has nothing in swap, or
725 * the vma maps it whole. Then we can simply use the stats that we
731 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
732 return swapped << PAGE_SHIFT;
734 /* Here comes the more involved part */
735 return shmem_partial_swap_usage(mapping,
736 linear_page_index(vma, vma->vm_start),
737 linear_page_index(vma, vma->vm_end));
741 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
743 void shmem_unlock_mapping(struct address_space *mapping)
746 pgoff_t indices[PAGEVEC_SIZE];
749 pagevec_init(&pvec, 0);
751 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
753 while (!mapping_unevictable(mapping)) {
755 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
756 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
758 pvec.nr = find_get_entries(mapping, index,
759 PAGEVEC_SIZE, pvec.pages, indices);
762 index = indices[pvec.nr - 1] + 1;
763 pagevec_remove_exceptionals(&pvec);
764 check_move_unevictable_pages(pvec.pages, pvec.nr);
765 pagevec_release(&pvec);
771 * Remove range of pages and swap entries from radix tree, and free them.
772 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
774 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
777 struct address_space *mapping = inode->i_mapping;
778 struct shmem_inode_info *info = SHMEM_I(inode);
779 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
780 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
781 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
782 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
784 pgoff_t indices[PAGEVEC_SIZE];
785 long nr_swaps_freed = 0;
790 end = -1; /* unsigned, so actually very big */
792 pagevec_init(&pvec, 0);
794 while (index < end) {
795 pvec.nr = find_get_entries(mapping, index,
796 min(end - index, (pgoff_t)PAGEVEC_SIZE),
797 pvec.pages, indices);
800 for (i = 0; i < pagevec_count(&pvec); i++) {
801 struct page *page = pvec.pages[i];
807 if (radix_tree_exceptional_entry(page)) {
810 nr_swaps_freed += !shmem_free_swap(mapping,
815 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
817 if (!trylock_page(page))
820 if (PageTransTail(page)) {
821 /* Middle of THP: zero out the page */
822 clear_highpage(page);
825 } else if (PageTransHuge(page)) {
826 if (index == round_down(end, HPAGE_PMD_NR)) {
828 * Range ends in the middle of THP:
831 clear_highpage(page);
835 index += HPAGE_PMD_NR - 1;
836 i += HPAGE_PMD_NR - 1;
839 if (!unfalloc || !PageUptodate(page)) {
840 VM_BUG_ON_PAGE(PageTail(page), page);
841 if (page_mapping(page) == mapping) {
842 VM_BUG_ON_PAGE(PageWriteback(page), page);
843 truncate_inode_page(mapping, page);
848 pagevec_remove_exceptionals(&pvec);
849 pagevec_release(&pvec);
855 struct page *page = NULL;
856 shmem_getpage(inode, start - 1, &page, SGP_READ);
858 unsigned int top = PAGE_SIZE;
863 zero_user_segment(page, partial_start, top);
864 set_page_dirty(page);
870 struct page *page = NULL;
871 shmem_getpage(inode, end, &page, SGP_READ);
873 zero_user_segment(page, 0, partial_end);
874 set_page_dirty(page);
883 while (index < end) {
886 pvec.nr = find_get_entries(mapping, index,
887 min(end - index, (pgoff_t)PAGEVEC_SIZE),
888 pvec.pages, indices);
890 /* If all gone or hole-punch or unfalloc, we're done */
891 if (index == start || end != -1)
893 /* But if truncating, restart to make sure all gone */
897 for (i = 0; i < pagevec_count(&pvec); i++) {
898 struct page *page = pvec.pages[i];
904 if (radix_tree_exceptional_entry(page)) {
907 if (shmem_free_swap(mapping, index, page)) {
908 /* Swap was replaced by page: retry */
918 if (PageTransTail(page)) {
919 /* Middle of THP: zero out the page */
920 clear_highpage(page);
923 * Partial thp truncate due 'start' in middle
924 * of THP: don't need to look on these pages
925 * again on !pvec.nr restart.
927 if (index != round_down(end, HPAGE_PMD_NR))
930 } else if (PageTransHuge(page)) {
931 if (index == round_down(end, HPAGE_PMD_NR)) {
933 * Range ends in the middle of THP:
936 clear_highpage(page);
940 index += HPAGE_PMD_NR - 1;
941 i += HPAGE_PMD_NR - 1;
944 if (!unfalloc || !PageUptodate(page)) {
945 VM_BUG_ON_PAGE(PageTail(page), page);
946 if (page_mapping(page) == mapping) {
947 VM_BUG_ON_PAGE(PageWriteback(page), page);
948 truncate_inode_page(mapping, page);
950 /* Page was replaced by swap: retry */
958 pagevec_remove_exceptionals(&pvec);
959 pagevec_release(&pvec);
963 spin_lock_irq(&info->lock);
964 info->swapped -= nr_swaps_freed;
965 shmem_recalc_inode(inode);
966 spin_unlock_irq(&info->lock);
969 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
971 shmem_undo_range(inode, lstart, lend, false);
972 inode->i_ctime = inode->i_mtime = current_time(inode);
974 EXPORT_SYMBOL_GPL(shmem_truncate_range);
976 static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
979 struct inode *inode = dentry->d_inode;
980 struct shmem_inode_info *info = SHMEM_I(inode);
982 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
983 spin_lock_irq(&info->lock);
984 shmem_recalc_inode(inode);
985 spin_unlock_irq(&info->lock);
987 generic_fillattr(inode, stat);
991 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
993 struct inode *inode = d_inode(dentry);
994 struct shmem_inode_info *info = SHMEM_I(inode);
995 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
998 error = setattr_prepare(dentry, attr);
1002 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1003 loff_t oldsize = inode->i_size;
1004 loff_t newsize = attr->ia_size;
1006 /* protected by i_mutex */
1007 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1008 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1011 if (newsize != oldsize) {
1012 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1016 i_size_write(inode, newsize);
1017 inode->i_ctime = inode->i_mtime = current_time(inode);
1019 if (newsize <= oldsize) {
1020 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1021 if (oldsize > holebegin)
1022 unmap_mapping_range(inode->i_mapping,
1025 shmem_truncate_range(inode,
1026 newsize, (loff_t)-1);
1027 /* unmap again to remove racily COWed private pages */
1028 if (oldsize > holebegin)
1029 unmap_mapping_range(inode->i_mapping,
1033 * Part of the huge page can be beyond i_size: subject
1034 * to shrink under memory pressure.
1036 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1037 spin_lock(&sbinfo->shrinklist_lock);
1039 * _careful to defend against unlocked access to
1040 * ->shrink_list in shmem_unused_huge_shrink()
1042 if (list_empty_careful(&info->shrinklist)) {
1043 list_add_tail(&info->shrinklist,
1044 &sbinfo->shrinklist);
1045 sbinfo->shrinklist_len++;
1047 spin_unlock(&sbinfo->shrinklist_lock);
1052 setattr_copy(inode, attr);
1053 if (attr->ia_valid & ATTR_MODE)
1054 error = posix_acl_chmod(inode, inode->i_mode);
1058 static void shmem_evict_inode(struct inode *inode)
1060 struct shmem_inode_info *info = SHMEM_I(inode);
1061 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1063 if (inode->i_mapping->a_ops == &shmem_aops) {
1064 shmem_unacct_size(info->flags, inode->i_size);
1066 shmem_truncate_range(inode, 0, (loff_t)-1);
1067 if (!list_empty(&info->shrinklist)) {
1068 spin_lock(&sbinfo->shrinklist_lock);
1069 if (!list_empty(&info->shrinklist)) {
1070 list_del_init(&info->shrinklist);
1071 sbinfo->shrinklist_len--;
1073 spin_unlock(&sbinfo->shrinklist_lock);
1075 if (!list_empty(&info->swaplist)) {
1076 mutex_lock(&shmem_swaplist_mutex);
1077 list_del_init(&info->swaplist);
1078 mutex_unlock(&shmem_swaplist_mutex);
1082 simple_xattrs_free(&info->xattrs);
1083 WARN_ON(inode->i_blocks);
1084 shmem_free_inode(inode->i_sb);
1089 * If swap found in inode, free it and move page from swapcache to filecache.
1091 static int shmem_unuse_inode(struct shmem_inode_info *info,
1092 swp_entry_t swap, struct page **pagep)
1094 struct address_space *mapping = info->vfs_inode.i_mapping;
1100 radswap = swp_to_radix_entry(swap);
1101 index = radix_tree_locate_item(&mapping->page_tree, radswap);
1103 return -EAGAIN; /* tell shmem_unuse we found nothing */
1106 * Move _head_ to start search for next from here.
1107 * But be careful: shmem_evict_inode checks list_empty without taking
1108 * mutex, and there's an instant in list_move_tail when info->swaplist
1109 * would appear empty, if it were the only one on shmem_swaplist.
1111 if (shmem_swaplist.next != &info->swaplist)
1112 list_move_tail(&shmem_swaplist, &info->swaplist);
1114 gfp = mapping_gfp_mask(mapping);
1115 if (shmem_should_replace_page(*pagep, gfp)) {
1116 mutex_unlock(&shmem_swaplist_mutex);
1117 error = shmem_replace_page(pagep, gfp, info, index);
1118 mutex_lock(&shmem_swaplist_mutex);
1120 * We needed to drop mutex to make that restrictive page
1121 * allocation, but the inode might have been freed while we
1122 * dropped it: although a racing shmem_evict_inode() cannot
1123 * complete without emptying the radix_tree, our page lock
1124 * on this swapcache page is not enough to prevent that -
1125 * free_swap_and_cache() of our swap entry will only
1126 * trylock_page(), removing swap from radix_tree whatever.
1128 * We must not proceed to shmem_add_to_page_cache() if the
1129 * inode has been freed, but of course we cannot rely on
1130 * inode or mapping or info to check that. However, we can
1131 * safely check if our swap entry is still in use (and here
1132 * it can't have got reused for another page): if it's still
1133 * in use, then the inode cannot have been freed yet, and we
1134 * can safely proceed (if it's no longer in use, that tells
1135 * nothing about the inode, but we don't need to unuse swap).
1137 if (!page_swapcount(*pagep))
1142 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1143 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1144 * beneath us (pagelock doesn't help until the page is in pagecache).
1147 error = shmem_add_to_page_cache(*pagep, mapping, index,
1149 if (error != -ENOMEM) {
1151 * Truncation and eviction use free_swap_and_cache(), which
1152 * only does trylock page: if we raced, best clean up here.
1154 delete_from_swap_cache(*pagep);
1155 set_page_dirty(*pagep);
1157 spin_lock_irq(&info->lock);
1159 spin_unlock_irq(&info->lock);
1167 * Search through swapped inodes to find and replace swap by page.
1169 int shmem_unuse(swp_entry_t swap, struct page *page)
1171 struct list_head *this, *next;
1172 struct shmem_inode_info *info;
1173 struct mem_cgroup *memcg;
1177 * There's a faint possibility that swap page was replaced before
1178 * caller locked it: caller will come back later with the right page.
1180 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1184 * Charge page using GFP_KERNEL while we can wait, before taking
1185 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1186 * Charged back to the user (not to caller) when swap account is used.
1188 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1192 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1195 mutex_lock(&shmem_swaplist_mutex);
1196 list_for_each_safe(this, next, &shmem_swaplist) {
1197 info = list_entry(this, struct shmem_inode_info, swaplist);
1199 error = shmem_unuse_inode(info, swap, &page);
1201 list_del_init(&info->swaplist);
1203 if (error != -EAGAIN)
1205 /* found nothing in this: move on to search the next */
1207 mutex_unlock(&shmem_swaplist_mutex);
1210 if (error != -ENOMEM)
1212 mem_cgroup_cancel_charge(page, memcg, false);
1214 mem_cgroup_commit_charge(page, memcg, true, false);
1222 * Move the page from the page cache to the swap cache.
1224 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1226 struct shmem_inode_info *info;
1227 struct address_space *mapping;
1228 struct inode *inode;
1232 VM_BUG_ON_PAGE(PageCompound(page), page);
1233 BUG_ON(!PageLocked(page));
1234 mapping = page->mapping;
1235 index = page->index;
1236 inode = mapping->host;
1237 info = SHMEM_I(inode);
1238 if (info->flags & VM_LOCKED)
1240 if (!total_swap_pages)
1244 * Our capabilities prevent regular writeback or sync from ever calling
1245 * shmem_writepage; but a stacking filesystem might use ->writepage of
1246 * its underlying filesystem, in which case tmpfs should write out to
1247 * swap only in response to memory pressure, and not for the writeback
1250 if (!wbc->for_reclaim) {
1251 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1256 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1257 * value into swapfile.c, the only way we can correctly account for a
1258 * fallocated page arriving here is now to initialize it and write it.
1260 * That's okay for a page already fallocated earlier, but if we have
1261 * not yet completed the fallocation, then (a) we want to keep track
1262 * of this page in case we have to undo it, and (b) it may not be a
1263 * good idea to continue anyway, once we're pushing into swap. So
1264 * reactivate the page, and let shmem_fallocate() quit when too many.
1266 if (!PageUptodate(page)) {
1267 if (inode->i_private) {
1268 struct shmem_falloc *shmem_falloc;
1269 spin_lock(&inode->i_lock);
1270 shmem_falloc = inode->i_private;
1272 !shmem_falloc->waitq &&
1273 index >= shmem_falloc->start &&
1274 index < shmem_falloc->next)
1275 shmem_falloc->nr_unswapped++;
1277 shmem_falloc = NULL;
1278 spin_unlock(&inode->i_lock);
1282 clear_highpage(page);
1283 flush_dcache_page(page);
1284 SetPageUptodate(page);
1287 swap = get_swap_page();
1291 if (mem_cgroup_try_charge_swap(page, swap))
1295 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1296 * if it's not already there. Do it now before the page is
1297 * moved to swap cache, when its pagelock no longer protects
1298 * the inode from eviction. But don't unlock the mutex until
1299 * we've incremented swapped, because shmem_unuse_inode() will
1300 * prune a !swapped inode from the swaplist under this mutex.
1302 mutex_lock(&shmem_swaplist_mutex);
1303 if (list_empty(&info->swaplist))
1304 list_add_tail(&info->swaplist, &shmem_swaplist);
1306 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1307 spin_lock_irq(&info->lock);
1308 shmem_recalc_inode(inode);
1310 spin_unlock_irq(&info->lock);
1312 swap_shmem_alloc(swap);
1313 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1315 mutex_unlock(&shmem_swaplist_mutex);
1316 BUG_ON(page_mapped(page));
1317 swap_writepage(page, wbc);
1321 mutex_unlock(&shmem_swaplist_mutex);
1323 swapcache_free(swap);
1325 set_page_dirty(page);
1326 if (wbc->for_reclaim)
1327 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1332 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1333 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1337 if (!mpol || mpol->mode == MPOL_DEFAULT)
1338 return; /* show nothing */
1340 mpol_to_str(buffer, sizeof(buffer), mpol);
1342 seq_printf(seq, ",mpol=%s", buffer);
1345 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1347 struct mempolicy *mpol = NULL;
1349 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1350 mpol = sbinfo->mpol;
1352 spin_unlock(&sbinfo->stat_lock);
1356 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1357 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1360 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1364 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1366 #define vm_policy vm_private_data
1369 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1370 struct shmem_inode_info *info, pgoff_t index)
1372 /* Create a pseudo vma that just contains the policy */
1374 /* Bias interleave by inode number to distribute better across nodes */
1375 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1377 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1380 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1382 /* Drop reference taken by mpol_shared_policy_lookup() */
1383 mpol_cond_put(vma->vm_policy);
1386 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1387 struct shmem_inode_info *info, pgoff_t index)
1389 struct vm_area_struct pvma;
1392 shmem_pseudo_vma_init(&pvma, info, index);
1393 page = swapin_readahead(swap, gfp, &pvma, 0);
1394 shmem_pseudo_vma_destroy(&pvma);
1399 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1400 struct shmem_inode_info *info, pgoff_t index)
1402 struct vm_area_struct pvma;
1403 struct inode *inode = &info->vfs_inode;
1404 struct address_space *mapping = inode->i_mapping;
1405 pgoff_t idx, hindex;
1406 void __rcu **results;
1409 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1412 hindex = round_down(index, HPAGE_PMD_NR);
1414 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1415 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1421 shmem_pseudo_vma_init(&pvma, info, hindex);
1422 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1423 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1424 shmem_pseudo_vma_destroy(&pvma);
1426 prep_transhuge_page(page);
1430 static struct page *shmem_alloc_page(gfp_t gfp,
1431 struct shmem_inode_info *info, pgoff_t index)
1433 struct vm_area_struct pvma;
1436 shmem_pseudo_vma_init(&pvma, info, index);
1437 page = alloc_page_vma(gfp, &pvma, 0);
1438 shmem_pseudo_vma_destroy(&pvma);
1443 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1444 struct inode *inode,
1445 pgoff_t index, bool huge)
1447 struct shmem_inode_info *info = SHMEM_I(inode);
1452 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1454 nr = huge ? HPAGE_PMD_NR : 1;
1456 if (!shmem_inode_acct_block(inode, nr))
1460 page = shmem_alloc_hugepage(gfp, info, index);
1462 page = shmem_alloc_page(gfp, info, index);
1464 __SetPageLocked(page);
1465 __SetPageSwapBacked(page);
1470 shmem_inode_unacct_blocks(inode, nr);
1472 return ERR_PTR(err);
1476 * When a page is moved from swapcache to shmem filecache (either by the
1477 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1478 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1479 * ignorance of the mapping it belongs to. If that mapping has special
1480 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1481 * we may need to copy to a suitable page before moving to filecache.
1483 * In a future release, this may well be extended to respect cpuset and
1484 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1485 * but for now it is a simple matter of zone.
1487 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1489 return page_zonenum(page) > gfp_zone(gfp);
1492 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1493 struct shmem_inode_info *info, pgoff_t index)
1495 struct page *oldpage, *newpage;
1496 struct address_space *swap_mapping;
1502 entry.val = page_private(oldpage);
1503 swap_index = swp_offset(entry);
1504 swap_mapping = page_mapping(oldpage);
1507 * We have arrived here because our zones are constrained, so don't
1508 * limit chance of success by further cpuset and node constraints.
1510 gfp &= ~GFP_CONSTRAINT_MASK;
1511 newpage = shmem_alloc_page(gfp, info, index);
1516 copy_highpage(newpage, oldpage);
1517 flush_dcache_page(newpage);
1519 __SetPageLocked(newpage);
1520 __SetPageSwapBacked(newpage);
1521 SetPageUptodate(newpage);
1522 set_page_private(newpage, entry.val);
1523 SetPageSwapCache(newpage);
1526 * Our caller will very soon move newpage out of swapcache, but it's
1527 * a nice clean interface for us to replace oldpage by newpage there.
1529 spin_lock_irq(&swap_mapping->tree_lock);
1530 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1533 __inc_node_page_state(newpage, NR_FILE_PAGES);
1534 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1536 spin_unlock_irq(&swap_mapping->tree_lock);
1538 if (unlikely(error)) {
1540 * Is this possible? I think not, now that our callers check
1541 * both PageSwapCache and page_private after getting page lock;
1542 * but be defensive. Reverse old to newpage for clear and free.
1546 mem_cgroup_migrate(oldpage, newpage);
1547 lru_cache_add_anon(newpage);
1551 ClearPageSwapCache(oldpage);
1552 set_page_private(oldpage, 0);
1554 unlock_page(oldpage);
1561 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1563 * If we allocate a new one we do not mark it dirty. That's up to the
1564 * vm. If we swap it in we mark it dirty since we also free the swap
1565 * entry since a page cannot live in both the swap and page cache.
1567 * fault_mm and fault_type are only supplied by shmem_fault:
1568 * otherwise they are NULL.
1570 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1571 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1572 struct mm_struct *fault_mm, int *fault_type)
1574 struct address_space *mapping = inode->i_mapping;
1575 struct shmem_inode_info *info = SHMEM_I(inode);
1576 struct shmem_sb_info *sbinfo;
1577 struct mm_struct *charge_mm;
1578 struct mem_cgroup *memcg;
1581 enum sgp_type sgp_huge = sgp;
1582 pgoff_t hindex = index;
1587 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1589 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1593 page = find_lock_entry(mapping, index);
1594 if (radix_tree_exceptional_entry(page)) {
1595 swap = radix_to_swp_entry(page);
1599 if (sgp <= SGP_CACHE &&
1600 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1605 if (page && sgp == SGP_WRITE)
1606 mark_page_accessed(page);
1608 /* fallocated page? */
1609 if (page && !PageUptodate(page)) {
1610 if (sgp != SGP_READ)
1616 if (page || (sgp == SGP_READ && !swap.val)) {
1622 * Fast cache lookup did not find it:
1623 * bring it back from swap or allocate.
1625 sbinfo = SHMEM_SB(inode->i_sb);
1626 charge_mm = fault_mm ? : current->mm;
1629 /* Look it up and read it in.. */
1630 page = lookup_swap_cache(swap);
1632 /* Or update major stats only when swapin succeeds?? */
1634 *fault_type |= VM_FAULT_MAJOR;
1635 count_vm_event(PGMAJFAULT);
1636 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1638 /* Here we actually start the io */
1639 page = shmem_swapin(swap, gfp, info, index);
1646 /* We have to do this with page locked to prevent races */
1648 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1649 !shmem_confirm_swap(mapping, index, swap)) {
1650 error = -EEXIST; /* try again */
1653 if (!PageUptodate(page)) {
1657 wait_on_page_writeback(page);
1659 if (shmem_should_replace_page(page, gfp)) {
1660 error = shmem_replace_page(&page, gfp, info, index);
1665 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1668 error = shmem_add_to_page_cache(page, mapping, index,
1669 swp_to_radix_entry(swap));
1671 * We already confirmed swap under page lock, and make
1672 * no memory allocation here, so usually no possibility
1673 * of error; but free_swap_and_cache() only trylocks a
1674 * page, so it is just possible that the entry has been
1675 * truncated or holepunched since swap was confirmed.
1676 * shmem_undo_range() will have done some of the
1677 * unaccounting, now delete_from_swap_cache() will do
1679 * Reset swap.val? No, leave it so "failed" goes back to
1680 * "repeat": reading a hole and writing should succeed.
1683 mem_cgroup_cancel_charge(page, memcg, false);
1684 delete_from_swap_cache(page);
1690 mem_cgroup_commit_charge(page, memcg, true, false);
1692 spin_lock_irq(&info->lock);
1694 shmem_recalc_inode(inode);
1695 spin_unlock_irq(&info->lock);
1697 if (sgp == SGP_WRITE)
1698 mark_page_accessed(page);
1700 delete_from_swap_cache(page);
1701 set_page_dirty(page);
1705 /* shmem_symlink() */
1706 if (mapping->a_ops != &shmem_aops)
1708 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1710 if (shmem_huge == SHMEM_HUGE_FORCE)
1712 switch (sbinfo->huge) {
1715 case SHMEM_HUGE_NEVER:
1717 case SHMEM_HUGE_WITHIN_SIZE:
1718 off = round_up(index, HPAGE_PMD_NR);
1719 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1720 if (i_size >= HPAGE_PMD_SIZE &&
1721 i_size >> PAGE_SHIFT >= off)
1724 case SHMEM_HUGE_ADVISE:
1725 if (sgp_huge == SGP_HUGE)
1727 /* TODO: implement fadvise() hints */
1732 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1734 alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode,
1739 error = PTR_ERR(page);
1741 if (error != -ENOSPC)
1744 * Try to reclaim some spece by splitting a huge page
1745 * beyond i_size on the filesystem.
1749 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1750 if (ret == SHRINK_STOP)
1758 if (PageTransHuge(page))
1759 hindex = round_down(index, HPAGE_PMD_NR);
1763 if (sgp == SGP_WRITE)
1764 __SetPageReferenced(page);
1766 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1767 PageTransHuge(page));
1770 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1771 compound_order(page));
1773 error = shmem_add_to_page_cache(page, mapping, hindex,
1775 radix_tree_preload_end();
1778 mem_cgroup_cancel_charge(page, memcg,
1779 PageTransHuge(page));
1782 mem_cgroup_commit_charge(page, memcg, false,
1783 PageTransHuge(page));
1784 lru_cache_add_anon(page);
1786 spin_lock_irq(&info->lock);
1787 info->alloced += 1 << compound_order(page);
1788 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1789 shmem_recalc_inode(inode);
1790 spin_unlock_irq(&info->lock);
1793 if (PageTransHuge(page) &&
1794 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1795 hindex + HPAGE_PMD_NR - 1) {
1797 * Part of the huge page is beyond i_size: subject
1798 * to shrink under memory pressure.
1800 spin_lock(&sbinfo->shrinklist_lock);
1802 * _careful to defend against unlocked access to
1803 * ->shrink_list in shmem_unused_huge_shrink()
1805 if (list_empty_careful(&info->shrinklist)) {
1806 list_add_tail(&info->shrinklist,
1807 &sbinfo->shrinklist);
1808 sbinfo->shrinklist_len++;
1810 spin_unlock(&sbinfo->shrinklist_lock);
1814 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1816 if (sgp == SGP_FALLOC)
1820 * Let SGP_WRITE caller clear ends if write does not fill page;
1821 * but SGP_FALLOC on a page fallocated earlier must initialize
1822 * it now, lest undo on failure cancel our earlier guarantee.
1824 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1825 struct page *head = compound_head(page);
1828 for (i = 0; i < (1 << compound_order(head)); i++) {
1829 clear_highpage(head + i);
1830 flush_dcache_page(head + i);
1832 SetPageUptodate(head);
1836 /* Perhaps the file has been truncated since we checked */
1837 if (sgp <= SGP_CACHE &&
1838 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1840 ClearPageDirty(page);
1841 delete_from_page_cache(page);
1842 spin_lock_irq(&info->lock);
1843 shmem_recalc_inode(inode);
1844 spin_unlock_irq(&info->lock);
1849 *pagep = page + index - hindex;
1856 shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
1858 if (PageTransHuge(page)) {
1864 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1871 if (error == -ENOSPC && !once++) {
1872 spin_lock_irq(&info->lock);
1873 shmem_recalc_inode(inode);
1874 spin_unlock_irq(&info->lock);
1877 if (error == -EEXIST) /* from above or from radix_tree_insert */
1883 * This is like autoremove_wake_function, but it removes the wait queue
1884 * entry unconditionally - even if something else had already woken the
1887 static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1889 int ret = default_wake_function(wait, mode, sync, key);
1890 list_del_init(&wait->task_list);
1894 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1896 struct inode *inode = file_inode(vma->vm_file);
1897 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1900 int ret = VM_FAULT_LOCKED;
1903 * Trinity finds that probing a hole which tmpfs is punching can
1904 * prevent the hole-punch from ever completing: which in turn
1905 * locks writers out with its hold on i_mutex. So refrain from
1906 * faulting pages into the hole while it's being punched. Although
1907 * shmem_undo_range() does remove the additions, it may be unable to
1908 * keep up, as each new page needs its own unmap_mapping_range() call,
1909 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1911 * It does not matter if we sometimes reach this check just before the
1912 * hole-punch begins, so that one fault then races with the punch:
1913 * we just need to make racing faults a rare case.
1915 * The implementation below would be much simpler if we just used a
1916 * standard mutex or completion: but we cannot take i_mutex in fault,
1917 * and bloating every shmem inode for this unlikely case would be sad.
1919 if (unlikely(inode->i_private)) {
1920 struct shmem_falloc *shmem_falloc;
1922 spin_lock(&inode->i_lock);
1923 shmem_falloc = inode->i_private;
1925 shmem_falloc->waitq &&
1926 vmf->pgoff >= shmem_falloc->start &&
1927 vmf->pgoff < shmem_falloc->next) {
1928 wait_queue_head_t *shmem_falloc_waitq;
1929 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1931 ret = VM_FAULT_NOPAGE;
1932 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1933 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1934 /* It's polite to up mmap_sem if we can */
1935 up_read(&vma->vm_mm->mmap_sem);
1936 ret = VM_FAULT_RETRY;
1939 shmem_falloc_waitq = shmem_falloc->waitq;
1940 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1941 TASK_UNINTERRUPTIBLE);
1942 spin_unlock(&inode->i_lock);
1946 * shmem_falloc_waitq points into the shmem_fallocate()
1947 * stack of the hole-punching task: shmem_falloc_waitq
1948 * is usually invalid by the time we reach here, but
1949 * finish_wait() does not dereference it in that case;
1950 * though i_lock needed lest racing with wake_up_all().
1952 spin_lock(&inode->i_lock);
1953 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1954 spin_unlock(&inode->i_lock);
1957 spin_unlock(&inode->i_lock);
1961 if (vma->vm_flags & VM_HUGEPAGE)
1963 else if (vma->vm_flags & VM_NOHUGEPAGE)
1966 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1967 gfp, vma->vm_mm, &ret);
1969 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1973 unsigned long shmem_get_unmapped_area(struct file *file,
1974 unsigned long uaddr, unsigned long len,
1975 unsigned long pgoff, unsigned long flags)
1977 unsigned long (*get_area)(struct file *,
1978 unsigned long, unsigned long, unsigned long, unsigned long);
1980 unsigned long offset;
1981 unsigned long inflated_len;
1982 unsigned long inflated_addr;
1983 unsigned long inflated_offset;
1985 if (len > TASK_SIZE)
1988 get_area = current->mm->get_unmapped_area;
1989 addr = get_area(file, uaddr, len, pgoff, flags);
1991 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1993 if (IS_ERR_VALUE(addr))
1995 if (addr & ~PAGE_MASK)
1997 if (addr > TASK_SIZE - len)
2000 if (shmem_huge == SHMEM_HUGE_DENY)
2002 if (len < HPAGE_PMD_SIZE)
2004 if (flags & MAP_FIXED)
2007 * Our priority is to support MAP_SHARED mapped hugely;
2008 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2009 * But if caller specified an address hint, respect that as before.
2014 if (shmem_huge != SHMEM_HUGE_FORCE) {
2015 struct super_block *sb;
2018 VM_BUG_ON(file->f_op != &shmem_file_operations);
2019 sb = file_inode(file)->i_sb;
2022 * Called directly from mm/mmap.c, or drivers/char/mem.c
2023 * for "/dev/zero", to create a shared anonymous object.
2025 if (IS_ERR(shm_mnt))
2027 sb = shm_mnt->mnt_sb;
2029 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2033 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2034 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2036 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2039 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2040 if (inflated_len > TASK_SIZE)
2042 if (inflated_len < len)
2045 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2046 if (IS_ERR_VALUE(inflated_addr))
2048 if (inflated_addr & ~PAGE_MASK)
2051 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2052 inflated_addr += offset - inflated_offset;
2053 if (inflated_offset > offset)
2054 inflated_addr += HPAGE_PMD_SIZE;
2056 if (inflated_addr > TASK_SIZE - len)
2058 return inflated_addr;
2062 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2064 struct inode *inode = file_inode(vma->vm_file);
2065 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2068 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2071 struct inode *inode = file_inode(vma->vm_file);
2074 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2075 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2079 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2081 struct inode *inode = file_inode(file);
2082 struct shmem_inode_info *info = SHMEM_I(inode);
2083 int retval = -ENOMEM;
2086 * What serializes the accesses to info->flags?
2087 * ipc_lock_object() when called from shmctl_do_lock(),
2088 * no serialization needed when called from shm_destroy().
2090 if (lock && !(info->flags & VM_LOCKED)) {
2091 if (!user_shm_lock(inode->i_size, user))
2093 info->flags |= VM_LOCKED;
2094 mapping_set_unevictable(file->f_mapping);
2096 if (!lock && (info->flags & VM_LOCKED) && user) {
2097 user_shm_unlock(inode->i_size, user);
2098 info->flags &= ~VM_LOCKED;
2099 mapping_clear_unevictable(file->f_mapping);
2107 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2109 file_accessed(file);
2110 vma->vm_ops = &shmem_vm_ops;
2111 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2112 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2113 (vma->vm_end & HPAGE_PMD_MASK)) {
2114 khugepaged_enter(vma, vma->vm_flags);
2119 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2120 umode_t mode, dev_t dev, unsigned long flags)
2122 struct inode *inode;
2123 struct shmem_inode_info *info;
2124 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2126 if (shmem_reserve_inode(sb))
2129 inode = new_inode(sb);
2131 inode->i_ino = get_next_ino();
2132 inode_init_owner(inode, dir, mode);
2133 inode->i_blocks = 0;
2134 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2135 inode->i_generation = get_seconds();
2136 info = SHMEM_I(inode);
2137 memset(info, 0, (char *)inode - (char *)info);
2138 spin_lock_init(&info->lock);
2139 info->seals = F_SEAL_SEAL;
2140 info->flags = flags & VM_NORESERVE;
2141 INIT_LIST_HEAD(&info->shrinklist);
2142 INIT_LIST_HEAD(&info->swaplist);
2143 simple_xattrs_init(&info->xattrs);
2144 cache_no_acl(inode);
2146 switch (mode & S_IFMT) {
2148 inode->i_op = &shmem_special_inode_operations;
2149 init_special_inode(inode, mode, dev);
2152 inode->i_mapping->a_ops = &shmem_aops;
2153 inode->i_op = &shmem_inode_operations;
2154 inode->i_fop = &shmem_file_operations;
2155 mpol_shared_policy_init(&info->policy,
2156 shmem_get_sbmpol(sbinfo));
2160 /* Some things misbehave if size == 0 on a directory */
2161 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2162 inode->i_op = &shmem_dir_inode_operations;
2163 inode->i_fop = &simple_dir_operations;
2167 * Must not load anything in the rbtree,
2168 * mpol_free_shared_policy will not be called.
2170 mpol_shared_policy_init(&info->policy, NULL);
2174 lockdep_annotate_inode_mutex_key(inode);
2176 shmem_free_inode(sb);
2180 bool shmem_mapping(struct address_space *mapping)
2185 return mapping->host->i_sb->s_op == &shmem_ops;
2189 static const struct inode_operations shmem_symlink_inode_operations;
2190 static const struct inode_operations shmem_short_symlink_operations;
2192 #ifdef CONFIG_TMPFS_XATTR
2193 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2195 #define shmem_initxattrs NULL
2199 shmem_write_begin(struct file *file, struct address_space *mapping,
2200 loff_t pos, unsigned len, unsigned flags,
2201 struct page **pagep, void **fsdata)
2203 struct inode *inode = mapping->host;
2204 struct shmem_inode_info *info = SHMEM_I(inode);
2205 pgoff_t index = pos >> PAGE_SHIFT;
2207 /* i_mutex is held by caller */
2208 if (unlikely(info->seals)) {
2209 if (info->seals & F_SEAL_WRITE)
2211 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2215 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2219 shmem_write_end(struct file *file, struct address_space *mapping,
2220 loff_t pos, unsigned len, unsigned copied,
2221 struct page *page, void *fsdata)
2223 struct inode *inode = mapping->host;
2225 if (pos + copied > inode->i_size)
2226 i_size_write(inode, pos + copied);
2228 if (!PageUptodate(page)) {
2229 struct page *head = compound_head(page);
2230 if (PageTransCompound(page)) {
2233 for (i = 0; i < HPAGE_PMD_NR; i++) {
2234 if (head + i == page)
2236 clear_highpage(head + i);
2237 flush_dcache_page(head + i);
2240 if (copied < PAGE_SIZE) {
2241 unsigned from = pos & (PAGE_SIZE - 1);
2242 zero_user_segments(page, 0, from,
2243 from + copied, PAGE_SIZE);
2245 SetPageUptodate(head);
2247 set_page_dirty(page);
2254 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2256 struct file *file = iocb->ki_filp;
2257 struct inode *inode = file_inode(file);
2258 struct address_space *mapping = inode->i_mapping;
2260 unsigned long offset;
2261 enum sgp_type sgp = SGP_READ;
2264 loff_t *ppos = &iocb->ki_pos;
2267 * Might this read be for a stacking filesystem? Then when reading
2268 * holes of a sparse file, we actually need to allocate those pages,
2269 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2271 if (!iter_is_iovec(to))
2274 index = *ppos >> PAGE_SHIFT;
2275 offset = *ppos & ~PAGE_MASK;
2278 struct page *page = NULL;
2280 unsigned long nr, ret;
2281 loff_t i_size = i_size_read(inode);
2283 end_index = i_size >> PAGE_SHIFT;
2284 if (index > end_index)
2286 if (index == end_index) {
2287 nr = i_size & ~PAGE_MASK;
2292 error = shmem_getpage(inode, index, &page, sgp);
2294 if (error == -EINVAL)
2299 if (sgp == SGP_CACHE)
2300 set_page_dirty(page);
2305 * We must evaluate after, since reads (unlike writes)
2306 * are called without i_mutex protection against truncate
2309 i_size = i_size_read(inode);
2310 end_index = i_size >> PAGE_SHIFT;
2311 if (index == end_index) {
2312 nr = i_size & ~PAGE_MASK;
2323 * If users can be writing to this page using arbitrary
2324 * virtual addresses, take care about potential aliasing
2325 * before reading the page on the kernel side.
2327 if (mapping_writably_mapped(mapping))
2328 flush_dcache_page(page);
2330 * Mark the page accessed if we read the beginning.
2333 mark_page_accessed(page);
2335 page = ZERO_PAGE(0);
2340 * Ok, we have the page, and it's up-to-date, so
2341 * now we can copy it to user space...
2343 ret = copy_page_to_iter(page, offset, nr, to);
2346 index += offset >> PAGE_SHIFT;
2347 offset &= ~PAGE_MASK;
2350 if (!iov_iter_count(to))
2359 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2360 file_accessed(file);
2361 return retval ? retval : error;
2365 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2367 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2368 pgoff_t index, pgoff_t end, int whence)
2371 struct pagevec pvec;
2372 pgoff_t indices[PAGEVEC_SIZE];
2376 pagevec_init(&pvec, 0);
2377 pvec.nr = 1; /* start small: we may be there already */
2379 pvec.nr = find_get_entries(mapping, index,
2380 pvec.nr, pvec.pages, indices);
2382 if (whence == SEEK_DATA)
2386 for (i = 0; i < pvec.nr; i++, index++) {
2387 if (index < indices[i]) {
2388 if (whence == SEEK_HOLE) {
2394 page = pvec.pages[i];
2395 if (page && !radix_tree_exceptional_entry(page)) {
2396 if (!PageUptodate(page))
2400 (page && whence == SEEK_DATA) ||
2401 (!page && whence == SEEK_HOLE)) {
2406 pagevec_remove_exceptionals(&pvec);
2407 pagevec_release(&pvec);
2408 pvec.nr = PAGEVEC_SIZE;
2414 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2416 struct address_space *mapping = file->f_mapping;
2417 struct inode *inode = mapping->host;
2421 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2422 return generic_file_llseek_size(file, offset, whence,
2423 MAX_LFS_FILESIZE, i_size_read(inode));
2425 /* We're holding i_mutex so we can access i_size directly */
2427 if (offset < 0 || offset >= inode->i_size)
2430 start = offset >> PAGE_SHIFT;
2431 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2432 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2433 new_offset <<= PAGE_SHIFT;
2434 if (new_offset > offset) {
2435 if (new_offset < inode->i_size)
2436 offset = new_offset;
2437 else if (whence == SEEK_DATA)
2440 offset = inode->i_size;
2445 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2446 inode_unlock(inode);
2451 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2452 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2454 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2455 #define LAST_SCAN 4 /* about 150ms max */
2457 static void shmem_tag_pins(struct address_space *mapping)
2459 struct radix_tree_iter iter;
2463 unsigned int tagged = 0;
2468 spin_lock_irq(&mapping->tree_lock);
2469 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2470 page = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
2471 if (!page || radix_tree_exception(page)) {
2472 if (radix_tree_deref_retry(page)) {
2473 slot = radix_tree_iter_retry(&iter);
2476 } else if (page_count(page) - page_mapcount(page) > 1) {
2477 radix_tree_tag_set(&mapping->page_tree, iter.index,
2481 if (++tagged % 1024)
2484 slot = radix_tree_iter_next(&iter);
2485 spin_unlock_irq(&mapping->tree_lock);
2487 spin_lock_irq(&mapping->tree_lock);
2489 spin_unlock_irq(&mapping->tree_lock);
2493 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2494 * via get_user_pages(), drivers might have some pending I/O without any active
2495 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2496 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2497 * them to be dropped.
2498 * The caller must guarantee that no new user will acquire writable references
2499 * to those pages to avoid races.
2501 static int shmem_wait_for_pins(struct address_space *mapping)
2503 struct radix_tree_iter iter;
2509 shmem_tag_pins(mapping);
2512 for (scan = 0; scan <= LAST_SCAN; scan++) {
2513 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2517 lru_add_drain_all();
2518 else if (schedule_timeout_killable((HZ << scan) / 200))
2523 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2524 start, SHMEM_TAG_PINNED) {
2526 page = radix_tree_deref_slot(slot);
2527 if (radix_tree_exception(page)) {
2528 if (radix_tree_deref_retry(page)) {
2529 slot = radix_tree_iter_retry(&iter);
2537 page_count(page) - page_mapcount(page) != 1) {
2538 if (scan < LAST_SCAN)
2539 goto continue_resched;
2542 * On the last scan, we clean up all those tags
2543 * we inserted; but make a note that we still
2544 * found pages pinned.
2549 spin_lock_irq(&mapping->tree_lock);
2550 radix_tree_tag_clear(&mapping->page_tree,
2551 iter.index, SHMEM_TAG_PINNED);
2552 spin_unlock_irq(&mapping->tree_lock);
2554 if (need_resched()) {
2556 slot = radix_tree_iter_next(&iter);
2565 #define F_ALL_SEALS (F_SEAL_SEAL | \
2570 int shmem_add_seals(struct file *file, unsigned int seals)
2572 struct inode *inode = file_inode(file);
2573 struct shmem_inode_info *info = SHMEM_I(inode);
2578 * Sealing allows multiple parties to share a shmem-file but restrict
2579 * access to a specific subset of file operations. Seals can only be
2580 * added, but never removed. This way, mutually untrusted parties can
2581 * share common memory regions with a well-defined policy. A malicious
2582 * peer can thus never perform unwanted operations on a shared object.
2584 * Seals are only supported on special shmem-files and always affect
2585 * the whole underlying inode. Once a seal is set, it may prevent some
2586 * kinds of access to the file. Currently, the following seals are
2588 * SEAL_SEAL: Prevent further seals from being set on this file
2589 * SEAL_SHRINK: Prevent the file from shrinking
2590 * SEAL_GROW: Prevent the file from growing
2591 * SEAL_WRITE: Prevent write access to the file
2593 * As we don't require any trust relationship between two parties, we
2594 * must prevent seals from being removed. Therefore, sealing a file
2595 * only adds a given set of seals to the file, it never touches
2596 * existing seals. Furthermore, the "setting seals"-operation can be
2597 * sealed itself, which basically prevents any further seal from being
2600 * Semantics of sealing are only defined on volatile files. Only
2601 * anonymous shmem files support sealing. More importantly, seals are
2602 * never written to disk. Therefore, there's no plan to support it on
2606 if (file->f_op != &shmem_file_operations)
2608 if (!(file->f_mode & FMODE_WRITE))
2610 if (seals & ~(unsigned int)F_ALL_SEALS)
2615 if (info->seals & F_SEAL_SEAL) {
2620 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2621 error = mapping_deny_writable(file->f_mapping);
2625 error = shmem_wait_for_pins(file->f_mapping);
2627 mapping_allow_writable(file->f_mapping);
2632 info->seals |= seals;
2636 inode_unlock(inode);
2639 EXPORT_SYMBOL_GPL(shmem_add_seals);
2641 int shmem_get_seals(struct file *file)
2643 if (file->f_op != &shmem_file_operations)
2646 return SHMEM_I(file_inode(file))->seals;
2648 EXPORT_SYMBOL_GPL(shmem_get_seals);
2650 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2656 /* disallow upper 32bit */
2660 error = shmem_add_seals(file, arg);
2663 error = shmem_get_seals(file);
2673 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2676 struct inode *inode = file_inode(file);
2677 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2678 struct shmem_inode_info *info = SHMEM_I(inode);
2679 struct shmem_falloc shmem_falloc;
2680 pgoff_t start, index, end;
2683 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2688 if (mode & FALLOC_FL_PUNCH_HOLE) {
2689 struct address_space *mapping = file->f_mapping;
2690 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2691 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2692 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2694 /* protected by i_mutex */
2695 if (info->seals & F_SEAL_WRITE) {
2700 shmem_falloc.waitq = &shmem_falloc_waitq;
2701 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2702 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2703 spin_lock(&inode->i_lock);
2704 inode->i_private = &shmem_falloc;
2705 spin_unlock(&inode->i_lock);
2707 if ((u64)unmap_end > (u64)unmap_start)
2708 unmap_mapping_range(mapping, unmap_start,
2709 1 + unmap_end - unmap_start, 0);
2710 shmem_truncate_range(inode, offset, offset + len - 1);
2711 /* No need to unmap again: hole-punching leaves COWed pages */
2713 spin_lock(&inode->i_lock);
2714 inode->i_private = NULL;
2715 wake_up_all(&shmem_falloc_waitq);
2716 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
2717 spin_unlock(&inode->i_lock);
2722 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2723 error = inode_newsize_ok(inode, offset + len);
2727 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2732 start = offset >> PAGE_SHIFT;
2733 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2734 /* Try to avoid a swapstorm if len is impossible to satisfy */
2735 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2740 shmem_falloc.waitq = NULL;
2741 shmem_falloc.start = start;
2742 shmem_falloc.next = start;
2743 shmem_falloc.nr_falloced = 0;
2744 shmem_falloc.nr_unswapped = 0;
2745 spin_lock(&inode->i_lock);
2746 inode->i_private = &shmem_falloc;
2747 spin_unlock(&inode->i_lock);
2749 for (index = start; index < end; index++) {
2753 * Good, the fallocate(2) manpage permits EINTR: we may have
2754 * been interrupted because we are using up too much memory.
2756 if (signal_pending(current))
2758 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2761 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2763 /* Remove the !PageUptodate pages we added */
2764 if (index > start) {
2765 shmem_undo_range(inode,
2766 (loff_t)start << PAGE_SHIFT,
2767 ((loff_t)index << PAGE_SHIFT) - 1, true);
2773 * Inform shmem_writepage() how far we have reached.
2774 * No need for lock or barrier: we have the page lock.
2776 shmem_falloc.next++;
2777 if (!PageUptodate(page))
2778 shmem_falloc.nr_falloced++;
2781 * If !PageUptodate, leave it that way so that freeable pages
2782 * can be recognized if we need to rollback on error later.
2783 * But set_page_dirty so that memory pressure will swap rather
2784 * than free the pages we are allocating (and SGP_CACHE pages
2785 * might still be clean: we now need to mark those dirty too).
2787 set_page_dirty(page);
2793 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2794 i_size_write(inode, offset + len);
2795 inode->i_ctime = current_time(inode);
2797 spin_lock(&inode->i_lock);
2798 inode->i_private = NULL;
2799 spin_unlock(&inode->i_lock);
2801 inode_unlock(inode);
2805 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2807 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2809 buf->f_type = TMPFS_MAGIC;
2810 buf->f_bsize = PAGE_SIZE;
2811 buf->f_namelen = NAME_MAX;
2812 if (sbinfo->max_blocks) {
2813 buf->f_blocks = sbinfo->max_blocks;
2815 buf->f_bfree = sbinfo->max_blocks -
2816 percpu_counter_sum(&sbinfo->used_blocks);
2818 if (sbinfo->max_inodes) {
2819 buf->f_files = sbinfo->max_inodes;
2820 buf->f_ffree = sbinfo->free_inodes;
2822 /* else leave those fields 0 like simple_statfs */
2827 * File creation. Allocate an inode, and we're done..
2830 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2832 struct inode *inode;
2833 int error = -ENOSPC;
2835 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2837 error = simple_acl_create(dir, inode);
2840 error = security_inode_init_security(inode, dir,
2842 shmem_initxattrs, NULL);
2843 if (error && error != -EOPNOTSUPP)
2847 dir->i_size += BOGO_DIRENT_SIZE;
2848 dir->i_ctime = dir->i_mtime = current_time(dir);
2849 d_instantiate(dentry, inode);
2850 dget(dentry); /* Extra count - pin the dentry in core */
2859 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2861 struct inode *inode;
2862 int error = -ENOSPC;
2864 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2866 error = security_inode_init_security(inode, dir,
2868 shmem_initxattrs, NULL);
2869 if (error && error != -EOPNOTSUPP)
2871 error = simple_acl_create(dir, inode);
2874 d_tmpfile(dentry, inode);
2882 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2886 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2892 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2895 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2901 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2903 struct inode *inode = d_inode(old_dentry);
2907 * No ordinary (disk based) filesystem counts links as inodes;
2908 * but each new link needs a new dentry, pinning lowmem, and
2909 * tmpfs dentries cannot be pruned until they are unlinked.
2910 * But if an O_TMPFILE file is linked into the tmpfs, the
2911 * first link must skip that, to get the accounting right.
2913 if (inode->i_nlink) {
2914 ret = shmem_reserve_inode(inode->i_sb);
2919 dir->i_size += BOGO_DIRENT_SIZE;
2920 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2922 ihold(inode); /* New dentry reference */
2923 dget(dentry); /* Extra pinning count for the created dentry */
2924 d_instantiate(dentry, inode);
2929 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2931 struct inode *inode = d_inode(dentry);
2933 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2934 shmem_free_inode(inode->i_sb);
2936 dir->i_size -= BOGO_DIRENT_SIZE;
2937 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2939 dput(dentry); /* Undo the count from "create" - this does all the work */
2943 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2945 if (!simple_empty(dentry))
2948 drop_nlink(d_inode(dentry));
2950 return shmem_unlink(dir, dentry);
2953 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2955 bool old_is_dir = d_is_dir(old_dentry);
2956 bool new_is_dir = d_is_dir(new_dentry);
2958 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2960 drop_nlink(old_dir);
2963 drop_nlink(new_dir);
2967 old_dir->i_ctime = old_dir->i_mtime =
2968 new_dir->i_ctime = new_dir->i_mtime =
2969 d_inode(old_dentry)->i_ctime =
2970 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2975 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2977 struct dentry *whiteout;
2980 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2984 error = shmem_mknod(old_dir, whiteout,
2985 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2991 * Cheat and hash the whiteout while the old dentry is still in
2992 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2994 * d_lookup() will consistently find one of them at this point,
2995 * not sure which one, but that isn't even important.
3002 * The VFS layer already does all the dentry stuff for rename,
3003 * we just have to decrement the usage count for the target if
3004 * it exists so that the VFS layer correctly free's it when it
3007 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3009 struct inode *inode = d_inode(old_dentry);
3010 int they_are_dirs = S_ISDIR(inode->i_mode);
3012 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3015 if (flags & RENAME_EXCHANGE)
3016 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3018 if (!simple_empty(new_dentry))
3021 if (flags & RENAME_WHITEOUT) {
3024 error = shmem_whiteout(old_dir, old_dentry);
3029 if (d_really_is_positive(new_dentry)) {
3030 (void) shmem_unlink(new_dir, new_dentry);
3031 if (they_are_dirs) {
3032 drop_nlink(d_inode(new_dentry));
3033 drop_nlink(old_dir);
3035 } else if (they_are_dirs) {
3036 drop_nlink(old_dir);
3040 old_dir->i_size -= BOGO_DIRENT_SIZE;
3041 new_dir->i_size += BOGO_DIRENT_SIZE;
3042 old_dir->i_ctime = old_dir->i_mtime =
3043 new_dir->i_ctime = new_dir->i_mtime =
3044 inode->i_ctime = current_time(old_dir);
3048 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3052 struct inode *inode;
3054 struct shmem_inode_info *info;
3056 len = strlen(symname) + 1;
3057 if (len > PAGE_SIZE)
3058 return -ENAMETOOLONG;
3060 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3064 error = security_inode_init_security(inode, dir, &dentry->d_name,
3065 shmem_initxattrs, NULL);
3067 if (error != -EOPNOTSUPP) {
3074 info = SHMEM_I(inode);
3075 inode->i_size = len-1;
3076 if (len <= SHORT_SYMLINK_LEN) {
3077 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3078 if (!inode->i_link) {
3082 inode->i_op = &shmem_short_symlink_operations;
3084 inode_nohighmem(inode);
3085 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3090 inode->i_mapping->a_ops = &shmem_aops;
3091 inode->i_op = &shmem_symlink_inode_operations;
3092 memcpy(page_address(page), symname, len);
3093 SetPageUptodate(page);
3094 set_page_dirty(page);
3098 dir->i_size += BOGO_DIRENT_SIZE;
3099 dir->i_ctime = dir->i_mtime = current_time(dir);
3100 d_instantiate(dentry, inode);
3105 static void shmem_put_link(void *arg)
3107 mark_page_accessed(arg);
3111 static const char *shmem_get_link(struct dentry *dentry,
3112 struct inode *inode,
3113 struct delayed_call *done)
3115 struct page *page = NULL;
3118 page = find_get_page(inode->i_mapping, 0);
3120 return ERR_PTR(-ECHILD);
3121 if (!PageUptodate(page)) {
3123 return ERR_PTR(-ECHILD);
3126 error = shmem_getpage(inode, 0, &page, SGP_READ);
3128 return ERR_PTR(error);
3131 set_delayed_call(done, shmem_put_link, page);
3132 return page_address(page);
3135 #ifdef CONFIG_TMPFS_XATTR
3137 * Superblocks without xattr inode operations may get some security.* xattr
3138 * support from the LSM "for free". As soon as we have any other xattrs
3139 * like ACLs, we also need to implement the security.* handlers at
3140 * filesystem level, though.
3144 * Callback for security_inode_init_security() for acquiring xattrs.
3146 static int shmem_initxattrs(struct inode *inode,
3147 const struct xattr *xattr_array,
3150 struct shmem_inode_info *info = SHMEM_I(inode);
3151 const struct xattr *xattr;
3152 struct simple_xattr *new_xattr;
3155 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3156 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3160 len = strlen(xattr->name) + 1;
3161 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3163 if (!new_xattr->name) {
3168 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3169 XATTR_SECURITY_PREFIX_LEN);
3170 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3173 simple_xattr_list_add(&info->xattrs, new_xattr);
3179 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3180 struct dentry *unused, struct inode *inode,
3181 const char *name, void *buffer, size_t size)
3183 struct shmem_inode_info *info = SHMEM_I(inode);
3185 name = xattr_full_name(handler, name);
3186 return simple_xattr_get(&info->xattrs, name, buffer, size);
3189 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3190 struct dentry *unused, struct inode *inode,
3191 const char *name, const void *value,
3192 size_t size, int flags)
3194 struct shmem_inode_info *info = SHMEM_I(inode);
3196 name = xattr_full_name(handler, name);
3197 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3200 static const struct xattr_handler shmem_security_xattr_handler = {
3201 .prefix = XATTR_SECURITY_PREFIX,
3202 .get = shmem_xattr_handler_get,
3203 .set = shmem_xattr_handler_set,
3206 static const struct xattr_handler shmem_trusted_xattr_handler = {
3207 .prefix = XATTR_TRUSTED_PREFIX,
3208 .get = shmem_xattr_handler_get,
3209 .set = shmem_xattr_handler_set,
3212 static const struct xattr_handler *shmem_xattr_handlers[] = {
3213 #ifdef CONFIG_TMPFS_POSIX_ACL
3214 &posix_acl_access_xattr_handler,
3215 &posix_acl_default_xattr_handler,
3217 &shmem_security_xattr_handler,
3218 &shmem_trusted_xattr_handler,
3222 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3224 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3225 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3227 #endif /* CONFIG_TMPFS_XATTR */
3229 static const struct inode_operations shmem_short_symlink_operations = {
3230 .readlink = generic_readlink,
3231 .get_link = simple_get_link,
3232 #ifdef CONFIG_TMPFS_XATTR
3233 .listxattr = shmem_listxattr,
3237 static const struct inode_operations shmem_symlink_inode_operations = {
3238 .readlink = generic_readlink,
3239 .get_link = shmem_get_link,
3240 #ifdef CONFIG_TMPFS_XATTR
3241 .listxattr = shmem_listxattr,
3245 static struct dentry *shmem_get_parent(struct dentry *child)
3247 return ERR_PTR(-ESTALE);
3250 static int shmem_match(struct inode *ino, void *vfh)
3254 inum = (inum << 32) | fh[1];
3255 return ino->i_ino == inum && fh[0] == ino->i_generation;
3258 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3259 struct fid *fid, int fh_len, int fh_type)
3261 struct inode *inode;
3262 struct dentry *dentry = NULL;
3269 inum = (inum << 32) | fid->raw[1];
3271 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3272 shmem_match, fid->raw);
3274 dentry = d_find_alias(inode);
3281 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3282 struct inode *parent)
3286 return FILEID_INVALID;
3289 if (inode_unhashed(inode)) {
3290 /* Unfortunately insert_inode_hash is not idempotent,
3291 * so as we hash inodes here rather than at creation
3292 * time, we need a lock to ensure we only try
3295 static DEFINE_SPINLOCK(lock);
3297 if (inode_unhashed(inode))
3298 __insert_inode_hash(inode,
3299 inode->i_ino + inode->i_generation);
3303 fh[0] = inode->i_generation;
3304 fh[1] = inode->i_ino;
3305 fh[2] = ((__u64)inode->i_ino) >> 32;
3311 static const struct export_operations shmem_export_ops = {
3312 .get_parent = shmem_get_parent,
3313 .encode_fh = shmem_encode_fh,
3314 .fh_to_dentry = shmem_fh_to_dentry,
3317 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3320 char *this_char, *value, *rest;
3321 struct mempolicy *mpol = NULL;
3325 while (options != NULL) {
3326 this_char = options;
3329 * NUL-terminate this option: unfortunately,
3330 * mount options form a comma-separated list,
3331 * but mpol's nodelist may also contain commas.
3333 options = strchr(options, ',');
3334 if (options == NULL)
3337 if (!isdigit(*options)) {
3344 if ((value = strchr(this_char,'=')) != NULL) {
3347 pr_err("tmpfs: No value for mount option '%s'\n",
3352 if (!strcmp(this_char,"size")) {
3353 unsigned long long size;
3354 size = memparse(value,&rest);
3356 size <<= PAGE_SHIFT;
3357 size *= totalram_pages;
3363 sbinfo->max_blocks =
3364 DIV_ROUND_UP(size, PAGE_SIZE);
3365 } else if (!strcmp(this_char,"nr_blocks")) {
3366 sbinfo->max_blocks = memparse(value, &rest);
3369 } else if (!strcmp(this_char,"nr_inodes")) {
3370 sbinfo->max_inodes = memparse(value, &rest);
3373 } else if (!strcmp(this_char,"mode")) {
3376 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3379 } else if (!strcmp(this_char,"uid")) {
3382 uid = simple_strtoul(value, &rest, 0);
3385 sbinfo->uid = make_kuid(current_user_ns(), uid);
3386 if (!uid_valid(sbinfo->uid))
3388 } else if (!strcmp(this_char,"gid")) {
3391 gid = simple_strtoul(value, &rest, 0);
3394 sbinfo->gid = make_kgid(current_user_ns(), gid);
3395 if (!gid_valid(sbinfo->gid))
3397 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3398 } else if (!strcmp(this_char, "huge")) {
3400 huge = shmem_parse_huge(value);
3403 if (!has_transparent_hugepage() &&
3404 huge != SHMEM_HUGE_NEVER)
3406 sbinfo->huge = huge;
3409 } else if (!strcmp(this_char,"mpol")) {
3412 if (mpol_parse_str(value, &mpol))
3416 pr_err("tmpfs: Bad mount option %s\n", this_char);
3420 sbinfo->mpol = mpol;
3424 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3432 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3434 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3435 struct shmem_sb_info config = *sbinfo;
3436 unsigned long inodes;
3437 int error = -EINVAL;
3440 if (shmem_parse_options(data, &config, true))
3443 spin_lock(&sbinfo->stat_lock);
3444 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3445 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3447 if (config.max_inodes < inodes)
3450 * Those tests disallow limited->unlimited while any are in use;
3451 * but we must separately disallow unlimited->limited, because
3452 * in that case we have no record of how much is already in use.
3454 if (config.max_blocks && !sbinfo->max_blocks)
3456 if (config.max_inodes && !sbinfo->max_inodes)
3460 sbinfo->huge = config.huge;
3461 sbinfo->max_blocks = config.max_blocks;
3462 sbinfo->max_inodes = config.max_inodes;
3463 sbinfo->free_inodes = config.max_inodes - inodes;
3466 * Preserve previous mempolicy unless mpol remount option was specified.
3469 mpol_put(sbinfo->mpol);
3470 sbinfo->mpol = config.mpol; /* transfers initial ref */
3473 spin_unlock(&sbinfo->stat_lock);
3477 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3479 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3481 if (sbinfo->max_blocks != shmem_default_max_blocks())
3482 seq_printf(seq, ",size=%luk",
3483 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3484 if (sbinfo->max_inodes != shmem_default_max_inodes())
3485 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3486 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3487 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3488 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3489 seq_printf(seq, ",uid=%u",
3490 from_kuid_munged(&init_user_ns, sbinfo->uid));
3491 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3492 seq_printf(seq, ",gid=%u",
3493 from_kgid_munged(&init_user_ns, sbinfo->gid));
3494 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3495 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3497 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3499 shmem_show_mpol(seq, sbinfo->mpol);
3503 #define MFD_NAME_PREFIX "memfd:"
3504 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3505 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3507 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3509 SYSCALL_DEFINE2(memfd_create,
3510 const char __user *, uname,
3511 unsigned int, flags)
3513 struct shmem_inode_info *info;
3519 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3522 /* length includes terminating zero */
3523 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3526 if (len > MFD_NAME_MAX_LEN + 1)
3529 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3533 strcpy(name, MFD_NAME_PREFIX);
3534 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3539 /* terminating-zero may have changed after strnlen_user() returned */
3540 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3545 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3551 file = shmem_file_setup(name, 0, VM_NORESERVE);
3553 error = PTR_ERR(file);
3556 info = SHMEM_I(file_inode(file));
3557 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3558 file->f_flags |= O_RDWR | O_LARGEFILE;
3559 if (flags & MFD_ALLOW_SEALING)
3560 info->seals &= ~F_SEAL_SEAL;
3562 fd_install(fd, file);
3573 #endif /* CONFIG_TMPFS */
3575 static void shmem_put_super(struct super_block *sb)
3577 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3579 percpu_counter_destroy(&sbinfo->used_blocks);
3580 mpol_put(sbinfo->mpol);
3582 sb->s_fs_info = NULL;
3585 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3587 struct inode *inode;
3588 struct shmem_sb_info *sbinfo;
3591 /* Round up to L1_CACHE_BYTES to resist false sharing */
3592 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3593 L1_CACHE_BYTES), GFP_KERNEL);
3597 sbinfo->mode = S_IRWXUGO | S_ISVTX;
3598 sbinfo->uid = current_fsuid();
3599 sbinfo->gid = current_fsgid();
3600 sb->s_fs_info = sbinfo;
3604 * Per default we only allow half of the physical ram per
3605 * tmpfs instance, limiting inodes to one per page of lowmem;
3606 * but the internal instance is left unlimited.
3608 if (!(sb->s_flags & MS_KERNMOUNT)) {
3609 sbinfo->max_blocks = shmem_default_max_blocks();
3610 sbinfo->max_inodes = shmem_default_max_inodes();
3611 if (shmem_parse_options(data, sbinfo, false)) {
3616 sb->s_flags |= MS_NOUSER;
3618 sb->s_export_op = &shmem_export_ops;
3619 sb->s_flags |= MS_NOSEC;
3621 sb->s_flags |= MS_NOUSER;
3624 spin_lock_init(&sbinfo->stat_lock);
3625 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3627 sbinfo->free_inodes = sbinfo->max_inodes;
3628 spin_lock_init(&sbinfo->shrinklist_lock);
3629 INIT_LIST_HEAD(&sbinfo->shrinklist);
3631 sb->s_maxbytes = MAX_LFS_FILESIZE;
3632 sb->s_blocksize = PAGE_SIZE;
3633 sb->s_blocksize_bits = PAGE_SHIFT;
3634 sb->s_magic = TMPFS_MAGIC;
3635 sb->s_op = &shmem_ops;
3636 sb->s_time_gran = 1;
3637 #ifdef CONFIG_TMPFS_XATTR
3638 sb->s_xattr = shmem_xattr_handlers;
3640 #ifdef CONFIG_TMPFS_POSIX_ACL
3641 sb->s_flags |= MS_POSIXACL;
3644 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3647 inode->i_uid = sbinfo->uid;
3648 inode->i_gid = sbinfo->gid;
3649 sb->s_root = d_make_root(inode);
3655 shmem_put_super(sb);
3659 static struct kmem_cache *shmem_inode_cachep;
3661 static struct inode *shmem_alloc_inode(struct super_block *sb)
3663 struct shmem_inode_info *info;
3664 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3667 return &info->vfs_inode;
3670 static void shmem_destroy_callback(struct rcu_head *head)
3672 struct inode *inode = container_of(head, struct inode, i_rcu);
3673 if (S_ISLNK(inode->i_mode))
3674 kfree(inode->i_link);
3675 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3678 static void shmem_destroy_inode(struct inode *inode)
3680 if (S_ISREG(inode->i_mode))
3681 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3682 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3685 static void shmem_init_inode(void *foo)
3687 struct shmem_inode_info *info = foo;
3688 inode_init_once(&info->vfs_inode);
3691 static int shmem_init_inodecache(void)
3693 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3694 sizeof(struct shmem_inode_info),
3695 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3699 static void shmem_destroy_inodecache(void)
3701 kmem_cache_destroy(shmem_inode_cachep);
3704 static const struct address_space_operations shmem_aops = {
3705 .writepage = shmem_writepage,
3706 .set_page_dirty = __set_page_dirty_no_writeback,
3708 .write_begin = shmem_write_begin,
3709 .write_end = shmem_write_end,
3711 #ifdef CONFIG_MIGRATION
3712 .migratepage = migrate_page,
3714 .error_remove_page = generic_error_remove_page,
3717 static const struct file_operations shmem_file_operations = {
3719 .get_unmapped_area = shmem_get_unmapped_area,
3721 .llseek = shmem_file_llseek,
3722 .read_iter = shmem_file_read_iter,
3723 .write_iter = generic_file_write_iter,
3724 .fsync = noop_fsync,
3725 .splice_read = generic_file_splice_read,
3726 .splice_write = iter_file_splice_write,
3727 .fallocate = shmem_fallocate,
3731 static const struct inode_operations shmem_inode_operations = {
3732 .getattr = shmem_getattr,
3733 .setattr = shmem_setattr,
3734 #ifdef CONFIG_TMPFS_XATTR
3735 .listxattr = shmem_listxattr,
3736 .set_acl = simple_set_acl,
3740 static const struct inode_operations shmem_dir_inode_operations = {
3742 .create = shmem_create,
3743 .lookup = simple_lookup,
3745 .unlink = shmem_unlink,
3746 .symlink = shmem_symlink,
3747 .mkdir = shmem_mkdir,
3748 .rmdir = shmem_rmdir,
3749 .mknod = shmem_mknod,
3750 .rename = shmem_rename2,
3751 .tmpfile = shmem_tmpfile,
3753 #ifdef CONFIG_TMPFS_XATTR
3754 .listxattr = shmem_listxattr,
3756 #ifdef CONFIG_TMPFS_POSIX_ACL
3757 .setattr = shmem_setattr,
3758 .set_acl = simple_set_acl,
3762 static const struct inode_operations shmem_special_inode_operations = {
3763 #ifdef CONFIG_TMPFS_XATTR
3764 .listxattr = shmem_listxattr,
3766 #ifdef CONFIG_TMPFS_POSIX_ACL
3767 .setattr = shmem_setattr,
3768 .set_acl = simple_set_acl,
3772 static const struct super_operations shmem_ops = {
3773 .alloc_inode = shmem_alloc_inode,
3774 .destroy_inode = shmem_destroy_inode,
3776 .statfs = shmem_statfs,
3777 .remount_fs = shmem_remount_fs,
3778 .show_options = shmem_show_options,
3780 .evict_inode = shmem_evict_inode,
3781 .drop_inode = generic_delete_inode,
3782 .put_super = shmem_put_super,
3783 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3784 .nr_cached_objects = shmem_unused_huge_count,
3785 .free_cached_objects = shmem_unused_huge_scan,
3789 static const struct vm_operations_struct shmem_vm_ops = {
3790 .fault = shmem_fault,
3791 .map_pages = filemap_map_pages,
3793 .set_policy = shmem_set_policy,
3794 .get_policy = shmem_get_policy,
3798 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3799 int flags, const char *dev_name, void *data)
3801 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3804 static struct file_system_type shmem_fs_type = {
3805 .owner = THIS_MODULE,
3807 .mount = shmem_mount,
3808 .kill_sb = kill_litter_super,
3809 .fs_flags = FS_USERNS_MOUNT,
3812 int __init shmem_init(void)
3816 /* If rootfs called this, don't re-init */
3817 if (shmem_inode_cachep)
3820 error = shmem_init_inodecache();
3824 error = register_filesystem(&shmem_fs_type);
3826 pr_err("Could not register tmpfs\n");
3830 shm_mnt = kern_mount(&shmem_fs_type);
3831 if (IS_ERR(shm_mnt)) {
3832 error = PTR_ERR(shm_mnt);
3833 pr_err("Could not kern_mount tmpfs\n");
3837 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3838 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3839 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3841 shmem_huge = 0; /* just in case it was patched */
3846 unregister_filesystem(&shmem_fs_type);
3848 shmem_destroy_inodecache();
3850 shm_mnt = ERR_PTR(error);
3854 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3855 static ssize_t shmem_enabled_show(struct kobject *kobj,
3856 struct kobj_attribute *attr, char *buf)
3860 SHMEM_HUGE_WITHIN_SIZE,
3868 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3869 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3871 count += sprintf(buf + count, fmt,
3872 shmem_format_huge(values[i]));
3874 buf[count - 1] = '\n';
3878 static ssize_t shmem_enabled_store(struct kobject *kobj,
3879 struct kobj_attribute *attr, const char *buf, size_t count)
3884 if (count + 1 > sizeof(tmp))
3886 memcpy(tmp, buf, count);
3888 if (count && tmp[count - 1] == '\n')
3889 tmp[count - 1] = '\0';
3891 huge = shmem_parse_huge(tmp);
3892 if (huge == -EINVAL)
3894 if (!has_transparent_hugepage() &&
3895 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3899 if (shmem_huge > SHMEM_HUGE_DENY)
3900 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3904 struct kobj_attribute shmem_enabled_attr =
3905 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3906 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3908 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3909 bool shmem_huge_enabled(struct vm_area_struct *vma)
3911 struct inode *inode = file_inode(vma->vm_file);
3912 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3916 if (shmem_huge == SHMEM_HUGE_FORCE)
3918 if (shmem_huge == SHMEM_HUGE_DENY)
3920 switch (sbinfo->huge) {
3921 case SHMEM_HUGE_NEVER:
3923 case SHMEM_HUGE_ALWAYS:
3925 case SHMEM_HUGE_WITHIN_SIZE:
3926 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3927 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3928 if (i_size >= HPAGE_PMD_SIZE &&
3929 i_size >> PAGE_SHIFT >= off)
3931 case SHMEM_HUGE_ADVISE:
3932 /* TODO: implement fadvise() hints */
3933 return (vma->vm_flags & VM_HUGEPAGE);
3939 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3941 #else /* !CONFIG_SHMEM */
3944 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3946 * This is intended for small system where the benefits of the full
3947 * shmem code (swap-backed and resource-limited) are outweighed by
3948 * their complexity. On systems without swap this code should be
3949 * effectively equivalent, but much lighter weight.
3952 static struct file_system_type shmem_fs_type = {
3954 .mount = ramfs_mount,
3955 .kill_sb = kill_litter_super,
3956 .fs_flags = FS_USERNS_MOUNT,
3959 int __init shmem_init(void)
3961 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3963 shm_mnt = kern_mount(&shmem_fs_type);
3964 BUG_ON(IS_ERR(shm_mnt));
3969 int shmem_unuse(swp_entry_t swap, struct page *page)
3974 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3979 void shmem_unlock_mapping(struct address_space *mapping)
3984 unsigned long shmem_get_unmapped_area(struct file *file,
3985 unsigned long addr, unsigned long len,
3986 unsigned long pgoff, unsigned long flags)
3988 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3992 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3994 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
3996 EXPORT_SYMBOL_GPL(shmem_truncate_range);
3998 #define shmem_vm_ops generic_file_vm_ops
3999 #define shmem_file_operations ramfs_file_operations
4000 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4001 #define shmem_acct_size(flags, size) 0
4002 #define shmem_unacct_size(flags, size) do {} while (0)
4004 #endif /* CONFIG_SHMEM */
4008 static const struct dentry_operations anon_ops = {
4009 .d_dname = simple_dname
4012 static struct file *__shmem_file_setup(const char *name, loff_t size,
4013 unsigned long flags, unsigned int i_flags)
4016 struct inode *inode;
4018 struct super_block *sb;
4021 if (IS_ERR(shm_mnt))
4022 return ERR_CAST(shm_mnt);
4024 if (size < 0 || size > MAX_LFS_FILESIZE)
4025 return ERR_PTR(-EINVAL);
4027 if (shmem_acct_size(flags, size))
4028 return ERR_PTR(-ENOMEM);
4030 res = ERR_PTR(-ENOMEM);
4032 this.len = strlen(name);
4033 this.hash = 0; /* will go */
4034 sb = shm_mnt->mnt_sb;
4035 path.mnt = mntget(shm_mnt);
4036 path.dentry = d_alloc_pseudo(sb, &this);
4039 d_set_d_op(path.dentry, &anon_ops);
4041 res = ERR_PTR(-ENOSPC);
4042 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4046 inode->i_flags |= i_flags;
4047 d_instantiate(path.dentry, inode);
4048 inode->i_size = size;
4049 clear_nlink(inode); /* It is unlinked */
4050 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4054 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4055 &shmem_file_operations);
4062 shmem_unacct_size(flags, size);
4069 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4070 * kernel internal. There will be NO LSM permission checks against the
4071 * underlying inode. So users of this interface must do LSM checks at a
4072 * higher layer. The users are the big_key and shm implementations. LSM
4073 * checks are provided at the key or shm level rather than the inode.
4074 * @name: name for dentry (to be seen in /proc/<pid>/maps
4075 * @size: size to be set for the file
4076 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4078 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4080 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4084 * shmem_file_setup - get an unlinked file living in tmpfs
4085 * @name: name for dentry (to be seen in /proc/<pid>/maps
4086 * @size: size to be set for the file
4087 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4089 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4091 return __shmem_file_setup(name, size, flags, 0);
4093 EXPORT_SYMBOL_GPL(shmem_file_setup);
4096 * shmem_zero_setup - setup a shared anonymous mapping
4097 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4099 int shmem_zero_setup(struct vm_area_struct *vma)
4102 loff_t size = vma->vm_end - vma->vm_start;
4105 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4106 * between XFS directory reading and selinux: since this file is only
4107 * accessible to the user through its mapping, use S_PRIVATE flag to
4108 * bypass file security, in the same way as shmem_kernel_file_setup().
4110 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4112 return PTR_ERR(file);
4116 vma->vm_file = file;
4117 vma->vm_ops = &shmem_vm_ops;
4119 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4120 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4121 (vma->vm_end & HPAGE_PMD_MASK)) {
4122 khugepaged_enter(vma, vma->vm_flags);
4129 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4130 * @mapping: the page's address_space
4131 * @index: the page index
4132 * @gfp: the page allocator flags to use if allocating
4134 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4135 * with any new page allocations done using the specified allocation flags.
4136 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4137 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4138 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4140 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4141 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4143 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4144 pgoff_t index, gfp_t gfp)
4147 struct inode *inode = mapping->host;
4151 BUG_ON(mapping->a_ops != &shmem_aops);
4152 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4155 page = ERR_PTR(error);
4161 * The tiny !SHMEM case uses ramfs without swap
4163 return read_cache_page_gfp(mapping, index, gfp);
4166 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);