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 <linux/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 const struct super_operations shmem_ops;
185 static const struct address_space_operations shmem_aops;
186 static const struct file_operations shmem_file_operations;
187 static const struct inode_operations shmem_inode_operations;
188 static const struct inode_operations shmem_dir_inode_operations;
189 static const struct inode_operations shmem_special_inode_operations;
190 static const struct vm_operations_struct shmem_vm_ops;
191 static struct file_system_type shmem_fs_type;
193 static LIST_HEAD(shmem_swaplist);
194 static DEFINE_MUTEX(shmem_swaplist_mutex);
196 static int shmem_reserve_inode(struct super_block *sb)
198 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
199 if (sbinfo->max_inodes) {
200 spin_lock(&sbinfo->stat_lock);
201 if (!sbinfo->free_inodes) {
202 spin_unlock(&sbinfo->stat_lock);
205 sbinfo->free_inodes--;
206 spin_unlock(&sbinfo->stat_lock);
211 static void shmem_free_inode(struct super_block *sb)
213 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
214 if (sbinfo->max_inodes) {
215 spin_lock(&sbinfo->stat_lock);
216 sbinfo->free_inodes++;
217 spin_unlock(&sbinfo->stat_lock);
222 * shmem_recalc_inode - recalculate the block usage of an inode
223 * @inode: inode to recalc
225 * We have to calculate the free blocks since the mm can drop
226 * undirtied hole pages behind our back.
228 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
229 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
231 * It has to be called with the spinlock held.
233 static void shmem_recalc_inode(struct inode *inode)
235 struct shmem_inode_info *info = SHMEM_I(inode);
238 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
240 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
241 if (sbinfo->max_blocks)
242 percpu_counter_add(&sbinfo->used_blocks, -freed);
243 info->alloced -= freed;
244 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
245 shmem_unacct_blocks(info->flags, freed);
249 bool shmem_charge(struct inode *inode, long pages)
251 struct shmem_inode_info *info = SHMEM_I(inode);
252 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
255 if (shmem_acct_block(info->flags, pages))
257 spin_lock_irqsave(&info->lock, flags);
258 info->alloced += pages;
259 inode->i_blocks += pages * BLOCKS_PER_PAGE;
260 shmem_recalc_inode(inode);
261 spin_unlock_irqrestore(&info->lock, flags);
262 inode->i_mapping->nrpages += pages;
264 if (!sbinfo->max_blocks)
266 if (percpu_counter_compare(&sbinfo->used_blocks,
267 sbinfo->max_blocks - pages) > 0) {
268 inode->i_mapping->nrpages -= pages;
269 spin_lock_irqsave(&info->lock, flags);
270 info->alloced -= pages;
271 shmem_recalc_inode(inode);
272 spin_unlock_irqrestore(&info->lock, flags);
273 shmem_unacct_blocks(info->flags, pages);
276 percpu_counter_add(&sbinfo->used_blocks, pages);
280 void shmem_uncharge(struct inode *inode, long pages)
282 struct shmem_inode_info *info = SHMEM_I(inode);
283 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
286 spin_lock_irqsave(&info->lock, flags);
287 info->alloced -= pages;
288 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
289 shmem_recalc_inode(inode);
290 spin_unlock_irqrestore(&info->lock, flags);
292 if (sbinfo->max_blocks)
293 percpu_counter_sub(&sbinfo->used_blocks, pages);
294 shmem_unacct_blocks(info->flags, pages);
298 * Replace item expected in radix tree by a new item, while holding tree lock.
300 static int shmem_radix_tree_replace(struct address_space *mapping,
301 pgoff_t index, void *expected, void *replacement)
303 struct radix_tree_node *node;
307 VM_BUG_ON(!expected);
308 VM_BUG_ON(!replacement);
309 item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot);
312 if (item != expected)
314 __radix_tree_replace(&mapping->page_tree, node, pslot,
315 replacement, NULL, NULL);
320 * Sometimes, before we decide whether to proceed or to fail, we must check
321 * that an entry was not already brought back from swap by a racing thread.
323 * Checking page is not enough: by the time a SwapCache page is locked, it
324 * might be reused, and again be SwapCache, using the same swap as before.
326 static bool shmem_confirm_swap(struct address_space *mapping,
327 pgoff_t index, swp_entry_t swap)
332 item = radix_tree_lookup(&mapping->page_tree, index);
334 return item == swp_to_radix_entry(swap);
338 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
341 * disables huge pages for the mount;
343 * enables huge pages for the mount;
344 * SHMEM_HUGE_WITHIN_SIZE:
345 * only allocate huge pages if the page will be fully within i_size,
346 * also respect fadvise()/madvise() hints;
348 * only allocate huge pages if requested with fadvise()/madvise();
351 #define SHMEM_HUGE_NEVER 0
352 #define SHMEM_HUGE_ALWAYS 1
353 #define SHMEM_HUGE_WITHIN_SIZE 2
354 #define SHMEM_HUGE_ADVISE 3
358 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
361 * disables huge on shm_mnt and all mounts, for emergency use;
363 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
366 #define SHMEM_HUGE_DENY (-1)
367 #define SHMEM_HUGE_FORCE (-2)
369 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
370 /* ifdef here to avoid bloating shmem.o when not necessary */
372 int shmem_huge __read_mostly;
374 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
375 static int shmem_parse_huge(const char *str)
377 if (!strcmp(str, "never"))
378 return SHMEM_HUGE_NEVER;
379 if (!strcmp(str, "always"))
380 return SHMEM_HUGE_ALWAYS;
381 if (!strcmp(str, "within_size"))
382 return SHMEM_HUGE_WITHIN_SIZE;
383 if (!strcmp(str, "advise"))
384 return SHMEM_HUGE_ADVISE;
385 if (!strcmp(str, "deny"))
386 return SHMEM_HUGE_DENY;
387 if (!strcmp(str, "force"))
388 return SHMEM_HUGE_FORCE;
392 static const char *shmem_format_huge(int huge)
395 case SHMEM_HUGE_NEVER:
397 case SHMEM_HUGE_ALWAYS:
399 case SHMEM_HUGE_WITHIN_SIZE:
400 return "within_size";
401 case SHMEM_HUGE_ADVISE:
403 case SHMEM_HUGE_DENY:
405 case SHMEM_HUGE_FORCE:
414 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
415 struct shrink_control *sc, unsigned long nr_to_split)
417 LIST_HEAD(list), *pos, *next;
418 LIST_HEAD(to_remove);
420 struct shmem_inode_info *info;
422 unsigned long batch = sc ? sc->nr_to_scan : 128;
423 int removed = 0, split = 0;
425 if (list_empty(&sbinfo->shrinklist))
428 spin_lock(&sbinfo->shrinklist_lock);
429 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
430 info = list_entry(pos, struct shmem_inode_info, shrinklist);
433 inode = igrab(&info->vfs_inode);
435 /* inode is about to be evicted */
437 list_del_init(&info->shrinklist);
442 /* Check if there's anything to gain */
443 if (round_up(inode->i_size, PAGE_SIZE) ==
444 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
445 list_move(&info->shrinklist, &to_remove);
450 list_move(&info->shrinklist, &list);
455 spin_unlock(&sbinfo->shrinklist_lock);
457 list_for_each_safe(pos, next, &to_remove) {
458 info = list_entry(pos, struct shmem_inode_info, shrinklist);
459 inode = &info->vfs_inode;
460 list_del_init(&info->shrinklist);
464 list_for_each_safe(pos, next, &list) {
467 info = list_entry(pos, struct shmem_inode_info, shrinklist);
468 inode = &info->vfs_inode;
470 if (nr_to_split && split >= nr_to_split) {
475 page = find_lock_page(inode->i_mapping,
476 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
480 if (!PageTransHuge(page)) {
486 ret = split_huge_page(page);
491 /* split failed: leave it on the list */
498 list_del_init(&info->shrinklist);
503 spin_lock(&sbinfo->shrinklist_lock);
504 list_splice_tail(&list, &sbinfo->shrinklist);
505 sbinfo->shrinklist_len -= removed;
506 spin_unlock(&sbinfo->shrinklist_lock);
511 static long shmem_unused_huge_scan(struct super_block *sb,
512 struct shrink_control *sc)
514 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
516 if (!READ_ONCE(sbinfo->shrinklist_len))
519 return shmem_unused_huge_shrink(sbinfo, sc, 0);
522 static long shmem_unused_huge_count(struct super_block *sb,
523 struct shrink_control *sc)
525 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
526 return READ_ONCE(sbinfo->shrinklist_len);
528 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
530 #define shmem_huge SHMEM_HUGE_DENY
532 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
533 struct shrink_control *sc, unsigned long nr_to_split)
537 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
540 * Like add_to_page_cache_locked, but error if expected item has gone.
542 static int shmem_add_to_page_cache(struct page *page,
543 struct address_space *mapping,
544 pgoff_t index, void *expected)
546 int error, nr = hpage_nr_pages(page);
548 VM_BUG_ON_PAGE(PageTail(page), page);
549 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
550 VM_BUG_ON_PAGE(!PageLocked(page), page);
551 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
552 VM_BUG_ON(expected && PageTransHuge(page));
554 page_ref_add(page, nr);
555 page->mapping = mapping;
558 spin_lock_irq(&mapping->tree_lock);
559 if (PageTransHuge(page)) {
560 void __rcu **results;
565 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
566 &results, &idx, index, 1) &&
567 idx < index + HPAGE_PMD_NR) {
572 for (i = 0; i < HPAGE_PMD_NR; i++) {
573 error = radix_tree_insert(&mapping->page_tree,
574 index + i, page + i);
577 count_vm_event(THP_FILE_ALLOC);
579 } else if (!expected) {
580 error = radix_tree_insert(&mapping->page_tree, index, page);
582 error = shmem_radix_tree_replace(mapping, index, expected,
587 mapping->nrpages += nr;
588 if (PageTransHuge(page))
589 __inc_node_page_state(page, NR_SHMEM_THPS);
590 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
591 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
592 spin_unlock_irq(&mapping->tree_lock);
594 page->mapping = NULL;
595 spin_unlock_irq(&mapping->tree_lock);
596 page_ref_sub(page, nr);
602 * Like delete_from_page_cache, but substitutes swap for page.
604 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
606 struct address_space *mapping = page->mapping;
609 VM_BUG_ON_PAGE(PageCompound(page), page);
611 spin_lock_irq(&mapping->tree_lock);
612 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
613 page->mapping = NULL;
615 __dec_node_page_state(page, NR_FILE_PAGES);
616 __dec_node_page_state(page, NR_SHMEM);
617 spin_unlock_irq(&mapping->tree_lock);
623 * Remove swap entry from radix tree, free the swap and its page cache.
625 static int shmem_free_swap(struct address_space *mapping,
626 pgoff_t index, void *radswap)
630 spin_lock_irq(&mapping->tree_lock);
631 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
632 spin_unlock_irq(&mapping->tree_lock);
635 free_swap_and_cache(radix_to_swp_entry(radswap));
640 * Determine (in bytes) how many of the shmem object's pages mapped by the
641 * given offsets are swapped out.
643 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
644 * as long as the inode doesn't go away and racy results are not a problem.
646 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
647 pgoff_t start, pgoff_t end)
649 struct radix_tree_iter iter;
652 unsigned long swapped = 0;
656 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
657 if (iter.index >= end)
660 page = radix_tree_deref_slot(slot);
662 if (radix_tree_deref_retry(page)) {
663 slot = radix_tree_iter_retry(&iter);
667 if (radix_tree_exceptional_entry(page))
670 if (need_resched()) {
671 slot = radix_tree_iter_resume(slot, &iter);
678 return swapped << PAGE_SHIFT;
682 * Determine (in bytes) how many of the shmem object's pages mapped by the
683 * given vma is swapped out.
685 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
686 * as long as the inode doesn't go away and racy results are not a problem.
688 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
690 struct inode *inode = file_inode(vma->vm_file);
691 struct shmem_inode_info *info = SHMEM_I(inode);
692 struct address_space *mapping = inode->i_mapping;
693 unsigned long swapped;
695 /* Be careful as we don't hold info->lock */
696 swapped = READ_ONCE(info->swapped);
699 * The easier cases are when the shmem object has nothing in swap, or
700 * the vma maps it whole. Then we can simply use the stats that we
706 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
707 return swapped << PAGE_SHIFT;
709 /* Here comes the more involved part */
710 return shmem_partial_swap_usage(mapping,
711 linear_page_index(vma, vma->vm_start),
712 linear_page_index(vma, vma->vm_end));
716 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
718 void shmem_unlock_mapping(struct address_space *mapping)
721 pgoff_t indices[PAGEVEC_SIZE];
724 pagevec_init(&pvec, 0);
726 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
728 while (!mapping_unevictable(mapping)) {
730 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
731 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
733 pvec.nr = find_get_entries(mapping, index,
734 PAGEVEC_SIZE, pvec.pages, indices);
737 index = indices[pvec.nr - 1] + 1;
738 pagevec_remove_exceptionals(&pvec);
739 check_move_unevictable_pages(pvec.pages, pvec.nr);
740 pagevec_release(&pvec);
746 * Remove range of pages and swap entries from radix tree, and free them.
747 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
749 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
752 struct address_space *mapping = inode->i_mapping;
753 struct shmem_inode_info *info = SHMEM_I(inode);
754 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
755 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
756 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
757 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
759 pgoff_t indices[PAGEVEC_SIZE];
760 long nr_swaps_freed = 0;
765 end = -1; /* unsigned, so actually very big */
767 pagevec_init(&pvec, 0);
769 while (index < end) {
770 pvec.nr = find_get_entries(mapping, index,
771 min(end - index, (pgoff_t)PAGEVEC_SIZE),
772 pvec.pages, indices);
775 for (i = 0; i < pagevec_count(&pvec); i++) {
776 struct page *page = pvec.pages[i];
782 if (radix_tree_exceptional_entry(page)) {
785 nr_swaps_freed += !shmem_free_swap(mapping,
790 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
792 if (!trylock_page(page))
795 if (PageTransTail(page)) {
796 /* Middle of THP: zero out the page */
797 clear_highpage(page);
800 } else if (PageTransHuge(page)) {
801 if (index == round_down(end, HPAGE_PMD_NR)) {
803 * Range ends in the middle of THP:
806 clear_highpage(page);
810 index += HPAGE_PMD_NR - 1;
811 i += HPAGE_PMD_NR - 1;
814 if (!unfalloc || !PageUptodate(page)) {
815 VM_BUG_ON_PAGE(PageTail(page), page);
816 if (page_mapping(page) == mapping) {
817 VM_BUG_ON_PAGE(PageWriteback(page), page);
818 truncate_inode_page(mapping, page);
823 pagevec_remove_exceptionals(&pvec);
824 pagevec_release(&pvec);
830 struct page *page = NULL;
831 shmem_getpage(inode, start - 1, &page, SGP_READ);
833 unsigned int top = PAGE_SIZE;
838 zero_user_segment(page, partial_start, top);
839 set_page_dirty(page);
845 struct page *page = NULL;
846 shmem_getpage(inode, end, &page, SGP_READ);
848 zero_user_segment(page, 0, partial_end);
849 set_page_dirty(page);
858 while (index < end) {
861 pvec.nr = find_get_entries(mapping, index,
862 min(end - index, (pgoff_t)PAGEVEC_SIZE),
863 pvec.pages, indices);
865 /* If all gone or hole-punch or unfalloc, we're done */
866 if (index == start || end != -1)
868 /* But if truncating, restart to make sure all gone */
872 for (i = 0; i < pagevec_count(&pvec); i++) {
873 struct page *page = pvec.pages[i];
879 if (radix_tree_exceptional_entry(page)) {
882 if (shmem_free_swap(mapping, index, page)) {
883 /* Swap was replaced by page: retry */
893 if (PageTransTail(page)) {
894 /* Middle of THP: zero out the page */
895 clear_highpage(page);
898 * Partial thp truncate due 'start' in middle
899 * of THP: don't need to look on these pages
900 * again on !pvec.nr restart.
902 if (index != round_down(end, HPAGE_PMD_NR))
905 } else if (PageTransHuge(page)) {
906 if (index == round_down(end, HPAGE_PMD_NR)) {
908 * Range ends in the middle of THP:
911 clear_highpage(page);
915 index += HPAGE_PMD_NR - 1;
916 i += HPAGE_PMD_NR - 1;
919 if (!unfalloc || !PageUptodate(page)) {
920 VM_BUG_ON_PAGE(PageTail(page), page);
921 if (page_mapping(page) == mapping) {
922 VM_BUG_ON_PAGE(PageWriteback(page), page);
923 truncate_inode_page(mapping, page);
925 /* Page was replaced by swap: retry */
933 pagevec_remove_exceptionals(&pvec);
934 pagevec_release(&pvec);
938 spin_lock_irq(&info->lock);
939 info->swapped -= nr_swaps_freed;
940 shmem_recalc_inode(inode);
941 spin_unlock_irq(&info->lock);
944 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
946 shmem_undo_range(inode, lstart, lend, false);
947 inode->i_ctime = inode->i_mtime = current_time(inode);
949 EXPORT_SYMBOL_GPL(shmem_truncate_range);
951 static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
954 struct inode *inode = dentry->d_inode;
955 struct shmem_inode_info *info = SHMEM_I(inode);
957 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
958 spin_lock_irq(&info->lock);
959 shmem_recalc_inode(inode);
960 spin_unlock_irq(&info->lock);
962 generic_fillattr(inode, stat);
966 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
968 struct inode *inode = d_inode(dentry);
969 struct shmem_inode_info *info = SHMEM_I(inode);
970 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
973 error = setattr_prepare(dentry, attr);
977 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
978 loff_t oldsize = inode->i_size;
979 loff_t newsize = attr->ia_size;
981 /* protected by i_mutex */
982 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
983 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
986 if (newsize != oldsize) {
987 error = shmem_reacct_size(SHMEM_I(inode)->flags,
991 i_size_write(inode, newsize);
992 inode->i_ctime = inode->i_mtime = current_time(inode);
994 if (newsize <= oldsize) {
995 loff_t holebegin = round_up(newsize, PAGE_SIZE);
996 if (oldsize > holebegin)
997 unmap_mapping_range(inode->i_mapping,
1000 shmem_truncate_range(inode,
1001 newsize, (loff_t)-1);
1002 /* unmap again to remove racily COWed private pages */
1003 if (oldsize > holebegin)
1004 unmap_mapping_range(inode->i_mapping,
1008 * Part of the huge page can be beyond i_size: subject
1009 * to shrink under memory pressure.
1011 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1012 spin_lock(&sbinfo->shrinklist_lock);
1013 if (list_empty(&info->shrinklist)) {
1014 list_add_tail(&info->shrinklist,
1015 &sbinfo->shrinklist);
1016 sbinfo->shrinklist_len++;
1018 spin_unlock(&sbinfo->shrinklist_lock);
1023 setattr_copy(inode, attr);
1024 if (attr->ia_valid & ATTR_MODE)
1025 error = posix_acl_chmod(inode, inode->i_mode);
1029 static void shmem_evict_inode(struct inode *inode)
1031 struct shmem_inode_info *info = SHMEM_I(inode);
1032 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1034 if (inode->i_mapping->a_ops == &shmem_aops) {
1035 shmem_unacct_size(info->flags, inode->i_size);
1037 shmem_truncate_range(inode, 0, (loff_t)-1);
1038 if (!list_empty(&info->shrinklist)) {
1039 spin_lock(&sbinfo->shrinklist_lock);
1040 if (!list_empty(&info->shrinklist)) {
1041 list_del_init(&info->shrinklist);
1042 sbinfo->shrinklist_len--;
1044 spin_unlock(&sbinfo->shrinklist_lock);
1046 if (!list_empty(&info->swaplist)) {
1047 mutex_lock(&shmem_swaplist_mutex);
1048 list_del_init(&info->swaplist);
1049 mutex_unlock(&shmem_swaplist_mutex);
1053 simple_xattrs_free(&info->xattrs);
1054 WARN_ON(inode->i_blocks);
1055 shmem_free_inode(inode->i_sb);
1059 static unsigned long find_swap_entry(struct radix_tree_root *root, void *item)
1061 struct radix_tree_iter iter;
1063 unsigned long found = -1;
1064 unsigned int checked = 0;
1067 radix_tree_for_each_slot(slot, root, &iter, 0) {
1068 if (*slot == item) {
1073 if ((checked % 4096) != 0)
1075 slot = radix_tree_iter_resume(slot, &iter);
1084 * If swap found in inode, free it and move page from swapcache to filecache.
1086 static int shmem_unuse_inode(struct shmem_inode_info *info,
1087 swp_entry_t swap, struct page **pagep)
1089 struct address_space *mapping = info->vfs_inode.i_mapping;
1095 radswap = swp_to_radix_entry(swap);
1096 index = find_swap_entry(&mapping->page_tree, radswap);
1098 return -EAGAIN; /* tell shmem_unuse we found nothing */
1101 * Move _head_ to start search for next from here.
1102 * But be careful: shmem_evict_inode checks list_empty without taking
1103 * mutex, and there's an instant in list_move_tail when info->swaplist
1104 * would appear empty, if it were the only one on shmem_swaplist.
1106 if (shmem_swaplist.next != &info->swaplist)
1107 list_move_tail(&shmem_swaplist, &info->swaplist);
1109 gfp = mapping_gfp_mask(mapping);
1110 if (shmem_should_replace_page(*pagep, gfp)) {
1111 mutex_unlock(&shmem_swaplist_mutex);
1112 error = shmem_replace_page(pagep, gfp, info, index);
1113 mutex_lock(&shmem_swaplist_mutex);
1115 * We needed to drop mutex to make that restrictive page
1116 * allocation, but the inode might have been freed while we
1117 * dropped it: although a racing shmem_evict_inode() cannot
1118 * complete without emptying the radix_tree, our page lock
1119 * on this swapcache page is not enough to prevent that -
1120 * free_swap_and_cache() of our swap entry will only
1121 * trylock_page(), removing swap from radix_tree whatever.
1123 * We must not proceed to shmem_add_to_page_cache() if the
1124 * inode has been freed, but of course we cannot rely on
1125 * inode or mapping or info to check that. However, we can
1126 * safely check if our swap entry is still in use (and here
1127 * it can't have got reused for another page): if it's still
1128 * in use, then the inode cannot have been freed yet, and we
1129 * can safely proceed (if it's no longer in use, that tells
1130 * nothing about the inode, but we don't need to unuse swap).
1132 if (!page_swapcount(*pagep))
1137 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1138 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1139 * beneath us (pagelock doesn't help until the page is in pagecache).
1142 error = shmem_add_to_page_cache(*pagep, mapping, index,
1144 if (error != -ENOMEM) {
1146 * Truncation and eviction use free_swap_and_cache(), which
1147 * only does trylock page: if we raced, best clean up here.
1149 delete_from_swap_cache(*pagep);
1150 set_page_dirty(*pagep);
1152 spin_lock_irq(&info->lock);
1154 spin_unlock_irq(&info->lock);
1162 * Search through swapped inodes to find and replace swap by page.
1164 int shmem_unuse(swp_entry_t swap, struct page *page)
1166 struct list_head *this, *next;
1167 struct shmem_inode_info *info;
1168 struct mem_cgroup *memcg;
1172 * There's a faint possibility that swap page was replaced before
1173 * caller locked it: caller will come back later with the right page.
1175 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1179 * Charge page using GFP_KERNEL while we can wait, before taking
1180 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1181 * Charged back to the user (not to caller) when swap account is used.
1183 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1187 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1190 mutex_lock(&shmem_swaplist_mutex);
1191 list_for_each_safe(this, next, &shmem_swaplist) {
1192 info = list_entry(this, struct shmem_inode_info, swaplist);
1194 error = shmem_unuse_inode(info, swap, &page);
1196 list_del_init(&info->swaplist);
1198 if (error != -EAGAIN)
1200 /* found nothing in this: move on to search the next */
1202 mutex_unlock(&shmem_swaplist_mutex);
1205 if (error != -ENOMEM)
1207 mem_cgroup_cancel_charge(page, memcg, false);
1209 mem_cgroup_commit_charge(page, memcg, true, false);
1217 * Move the page from the page cache to the swap cache.
1219 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1221 struct shmem_inode_info *info;
1222 struct address_space *mapping;
1223 struct inode *inode;
1227 VM_BUG_ON_PAGE(PageCompound(page), page);
1228 BUG_ON(!PageLocked(page));
1229 mapping = page->mapping;
1230 index = page->index;
1231 inode = mapping->host;
1232 info = SHMEM_I(inode);
1233 if (info->flags & VM_LOCKED)
1235 if (!total_swap_pages)
1239 * Our capabilities prevent regular writeback or sync from ever calling
1240 * shmem_writepage; but a stacking filesystem might use ->writepage of
1241 * its underlying filesystem, in which case tmpfs should write out to
1242 * swap only in response to memory pressure, and not for the writeback
1245 if (!wbc->for_reclaim) {
1246 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1251 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1252 * value into swapfile.c, the only way we can correctly account for a
1253 * fallocated page arriving here is now to initialize it and write it.
1255 * That's okay for a page already fallocated earlier, but if we have
1256 * not yet completed the fallocation, then (a) we want to keep track
1257 * of this page in case we have to undo it, and (b) it may not be a
1258 * good idea to continue anyway, once we're pushing into swap. So
1259 * reactivate the page, and let shmem_fallocate() quit when too many.
1261 if (!PageUptodate(page)) {
1262 if (inode->i_private) {
1263 struct shmem_falloc *shmem_falloc;
1264 spin_lock(&inode->i_lock);
1265 shmem_falloc = inode->i_private;
1267 !shmem_falloc->waitq &&
1268 index >= shmem_falloc->start &&
1269 index < shmem_falloc->next)
1270 shmem_falloc->nr_unswapped++;
1272 shmem_falloc = NULL;
1273 spin_unlock(&inode->i_lock);
1277 clear_highpage(page);
1278 flush_dcache_page(page);
1279 SetPageUptodate(page);
1282 swap = get_swap_page();
1286 if (mem_cgroup_try_charge_swap(page, swap))
1290 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1291 * if it's not already there. Do it now before the page is
1292 * moved to swap cache, when its pagelock no longer protects
1293 * the inode from eviction. But don't unlock the mutex until
1294 * we've incremented swapped, because shmem_unuse_inode() will
1295 * prune a !swapped inode from the swaplist under this mutex.
1297 mutex_lock(&shmem_swaplist_mutex);
1298 if (list_empty(&info->swaplist))
1299 list_add_tail(&info->swaplist, &shmem_swaplist);
1301 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1302 spin_lock_irq(&info->lock);
1303 shmem_recalc_inode(inode);
1305 spin_unlock_irq(&info->lock);
1307 swap_shmem_alloc(swap);
1308 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1310 mutex_unlock(&shmem_swaplist_mutex);
1311 BUG_ON(page_mapped(page));
1312 swap_writepage(page, wbc);
1316 mutex_unlock(&shmem_swaplist_mutex);
1318 swapcache_free(swap);
1320 set_page_dirty(page);
1321 if (wbc->for_reclaim)
1322 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1327 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1328 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1332 if (!mpol || mpol->mode == MPOL_DEFAULT)
1333 return; /* show nothing */
1335 mpol_to_str(buffer, sizeof(buffer), mpol);
1337 seq_printf(seq, ",mpol=%s", buffer);
1340 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1342 struct mempolicy *mpol = NULL;
1344 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1345 mpol = sbinfo->mpol;
1347 spin_unlock(&sbinfo->stat_lock);
1351 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1352 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1355 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1359 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1361 #define vm_policy vm_private_data
1364 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1365 struct shmem_inode_info *info, pgoff_t index)
1367 /* Create a pseudo vma that just contains the policy */
1369 /* Bias interleave by inode number to distribute better across nodes */
1370 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1372 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1375 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1377 /* Drop reference taken by mpol_shared_policy_lookup() */
1378 mpol_cond_put(vma->vm_policy);
1381 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1382 struct shmem_inode_info *info, pgoff_t index)
1384 struct vm_area_struct pvma;
1387 shmem_pseudo_vma_init(&pvma, info, index);
1388 page = swapin_readahead(swap, gfp, &pvma, 0);
1389 shmem_pseudo_vma_destroy(&pvma);
1394 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1395 struct shmem_inode_info *info, pgoff_t index)
1397 struct vm_area_struct pvma;
1398 struct inode *inode = &info->vfs_inode;
1399 struct address_space *mapping = inode->i_mapping;
1400 pgoff_t idx, hindex;
1401 void __rcu **results;
1404 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1407 hindex = round_down(index, HPAGE_PMD_NR);
1409 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1410 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1416 shmem_pseudo_vma_init(&pvma, info, hindex);
1417 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1418 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1419 shmem_pseudo_vma_destroy(&pvma);
1421 prep_transhuge_page(page);
1425 static struct page *shmem_alloc_page(gfp_t gfp,
1426 struct shmem_inode_info *info, pgoff_t index)
1428 struct vm_area_struct pvma;
1431 shmem_pseudo_vma_init(&pvma, info, index);
1432 page = alloc_page_vma(gfp, &pvma, 0);
1433 shmem_pseudo_vma_destroy(&pvma);
1438 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1439 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1440 pgoff_t index, bool huge)
1446 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1448 nr = huge ? HPAGE_PMD_NR : 1;
1450 if (shmem_acct_block(info->flags, nr))
1452 if (sbinfo->max_blocks) {
1453 if (percpu_counter_compare(&sbinfo->used_blocks,
1454 sbinfo->max_blocks - nr) > 0)
1456 percpu_counter_add(&sbinfo->used_blocks, nr);
1460 page = shmem_alloc_hugepage(gfp, info, index);
1462 page = shmem_alloc_page(gfp, info, index);
1464 __SetPageLocked(page);
1465 __SetPageSwapBacked(page);
1470 if (sbinfo->max_blocks)
1471 percpu_counter_add(&sbinfo->used_blocks, -nr);
1473 shmem_unacct_blocks(info->flags, nr);
1475 return ERR_PTR(err);
1479 * When a page is moved from swapcache to shmem filecache (either by the
1480 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1481 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1482 * ignorance of the mapping it belongs to. If that mapping has special
1483 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1484 * we may need to copy to a suitable page before moving to filecache.
1486 * In a future release, this may well be extended to respect cpuset and
1487 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1488 * but for now it is a simple matter of zone.
1490 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1492 return page_zonenum(page) > gfp_zone(gfp);
1495 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1496 struct shmem_inode_info *info, pgoff_t index)
1498 struct page *oldpage, *newpage;
1499 struct address_space *swap_mapping;
1504 swap_index = page_private(oldpage);
1505 swap_mapping = page_mapping(oldpage);
1508 * We have arrived here because our zones are constrained, so don't
1509 * limit chance of success by further cpuset and node constraints.
1511 gfp &= ~GFP_CONSTRAINT_MASK;
1512 newpage = shmem_alloc_page(gfp, info, index);
1517 copy_highpage(newpage, oldpage);
1518 flush_dcache_page(newpage);
1520 __SetPageLocked(newpage);
1521 __SetPageSwapBacked(newpage);
1522 SetPageUptodate(newpage);
1523 set_page_private(newpage, swap_index);
1524 SetPageSwapCache(newpage);
1527 * Our caller will very soon move newpage out of swapcache, but it's
1528 * a nice clean interface for us to replace oldpage by newpage there.
1530 spin_lock_irq(&swap_mapping->tree_lock);
1531 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1534 __inc_node_page_state(newpage, NR_FILE_PAGES);
1535 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1537 spin_unlock_irq(&swap_mapping->tree_lock);
1539 if (unlikely(error)) {
1541 * Is this possible? I think not, now that our callers check
1542 * both PageSwapCache and page_private after getting page lock;
1543 * but be defensive. Reverse old to newpage for clear and free.
1547 mem_cgroup_migrate(oldpage, newpage);
1548 lru_cache_add_anon(newpage);
1552 ClearPageSwapCache(oldpage);
1553 set_page_private(oldpage, 0);
1555 unlock_page(oldpage);
1562 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1564 * If we allocate a new one we do not mark it dirty. That's up to the
1565 * vm. If we swap it in we mark it dirty since we also free the swap
1566 * entry since a page cannot live in both the swap and page cache.
1568 * fault_mm and fault_type are only supplied by shmem_fault:
1569 * otherwise they are NULL.
1571 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1572 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1573 struct mm_struct *fault_mm, int *fault_type)
1575 struct address_space *mapping = inode->i_mapping;
1576 struct shmem_inode_info *info = SHMEM_I(inode);
1577 struct shmem_sb_info *sbinfo;
1578 struct mm_struct *charge_mm;
1579 struct mem_cgroup *memcg;
1582 enum sgp_type sgp_huge = sgp;
1583 pgoff_t hindex = index;
1588 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1590 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1594 page = find_lock_entry(mapping, index);
1595 if (radix_tree_exceptional_entry(page)) {
1596 swap = radix_to_swp_entry(page);
1600 if (sgp <= SGP_CACHE &&
1601 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1606 if (page && sgp == SGP_WRITE)
1607 mark_page_accessed(page);
1609 /* fallocated page? */
1610 if (page && !PageUptodate(page)) {
1611 if (sgp != SGP_READ)
1617 if (page || (sgp == SGP_READ && !swap.val)) {
1623 * Fast cache lookup did not find it:
1624 * bring it back from swap or allocate.
1626 sbinfo = SHMEM_SB(inode->i_sb);
1627 charge_mm = fault_mm ? : current->mm;
1630 /* Look it up and read it in.. */
1631 page = lookup_swap_cache(swap);
1633 /* Or update major stats only when swapin succeeds?? */
1635 *fault_type |= VM_FAULT_MAJOR;
1636 count_vm_event(PGMAJFAULT);
1637 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1639 /* Here we actually start the io */
1640 page = shmem_swapin(swap, gfp, info, index);
1647 /* We have to do this with page locked to prevent races */
1649 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1650 !shmem_confirm_swap(mapping, index, swap)) {
1651 error = -EEXIST; /* try again */
1654 if (!PageUptodate(page)) {
1658 wait_on_page_writeback(page);
1660 if (shmem_should_replace_page(page, gfp)) {
1661 error = shmem_replace_page(&page, gfp, info, index);
1666 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1669 error = shmem_add_to_page_cache(page, mapping, index,
1670 swp_to_radix_entry(swap));
1672 * We already confirmed swap under page lock, and make
1673 * no memory allocation here, so usually no possibility
1674 * of error; but free_swap_and_cache() only trylocks a
1675 * page, so it is just possible that the entry has been
1676 * truncated or holepunched since swap was confirmed.
1677 * shmem_undo_range() will have done some of the
1678 * unaccounting, now delete_from_swap_cache() will do
1680 * Reset swap.val? No, leave it so "failed" goes back to
1681 * "repeat": reading a hole and writing should succeed.
1684 mem_cgroup_cancel_charge(page, memcg, false);
1685 delete_from_swap_cache(page);
1691 mem_cgroup_commit_charge(page, memcg, true, false);
1693 spin_lock_irq(&info->lock);
1695 shmem_recalc_inode(inode);
1696 spin_unlock_irq(&info->lock);
1698 if (sgp == SGP_WRITE)
1699 mark_page_accessed(page);
1701 delete_from_swap_cache(page);
1702 set_page_dirty(page);
1706 /* shmem_symlink() */
1707 if (mapping->a_ops != &shmem_aops)
1709 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1711 if (shmem_huge == SHMEM_HUGE_FORCE)
1713 switch (sbinfo->huge) {
1716 case SHMEM_HUGE_NEVER:
1718 case SHMEM_HUGE_WITHIN_SIZE:
1719 off = round_up(index, HPAGE_PMD_NR);
1720 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1721 if (i_size >= HPAGE_PMD_SIZE &&
1722 i_size >> PAGE_SHIFT >= off)
1725 case SHMEM_HUGE_ADVISE:
1726 if (sgp_huge == SGP_HUGE)
1728 /* TODO: implement fadvise() hints */
1733 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1736 alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1741 error = PTR_ERR(page);
1743 if (error != -ENOSPC)
1746 * Try to reclaim some spece by splitting a huge page
1747 * beyond i_size on the filesystem.
1751 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1752 if (ret == SHRINK_STOP)
1760 if (PageTransHuge(page))
1761 hindex = round_down(index, HPAGE_PMD_NR);
1765 if (sgp == SGP_WRITE)
1766 __SetPageReferenced(page);
1768 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1769 PageTransHuge(page));
1772 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1773 compound_order(page));
1775 error = shmem_add_to_page_cache(page, mapping, hindex,
1777 radix_tree_preload_end();
1780 mem_cgroup_cancel_charge(page, memcg,
1781 PageTransHuge(page));
1784 mem_cgroup_commit_charge(page, memcg, false,
1785 PageTransHuge(page));
1786 lru_cache_add_anon(page);
1788 spin_lock_irq(&info->lock);
1789 info->alloced += 1 << compound_order(page);
1790 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1791 shmem_recalc_inode(inode);
1792 spin_unlock_irq(&info->lock);
1795 if (PageTransHuge(page) &&
1796 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1797 hindex + HPAGE_PMD_NR - 1) {
1799 * Part of the huge page is beyond i_size: subject
1800 * to shrink under memory pressure.
1802 spin_lock(&sbinfo->shrinklist_lock);
1803 if (list_empty(&info->shrinklist)) {
1804 list_add_tail(&info->shrinklist,
1805 &sbinfo->shrinklist);
1806 sbinfo->shrinklist_len++;
1808 spin_unlock(&sbinfo->shrinklist_lock);
1812 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1814 if (sgp == SGP_FALLOC)
1818 * Let SGP_WRITE caller clear ends if write does not fill page;
1819 * but SGP_FALLOC on a page fallocated earlier must initialize
1820 * it now, lest undo on failure cancel our earlier guarantee.
1822 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1823 struct page *head = compound_head(page);
1826 for (i = 0; i < (1 << compound_order(head)); i++) {
1827 clear_highpage(head + i);
1828 flush_dcache_page(head + i);
1830 SetPageUptodate(head);
1834 /* Perhaps the file has been truncated since we checked */
1835 if (sgp <= SGP_CACHE &&
1836 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1838 ClearPageDirty(page);
1839 delete_from_page_cache(page);
1840 spin_lock_irq(&info->lock);
1841 shmem_recalc_inode(inode);
1842 spin_unlock_irq(&info->lock);
1847 *pagep = page + index - hindex;
1854 if (sbinfo->max_blocks)
1855 percpu_counter_sub(&sbinfo->used_blocks,
1856 1 << compound_order(page));
1857 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1859 if (PageTransHuge(page)) {
1865 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1872 if (error == -ENOSPC && !once++) {
1873 spin_lock_irq(&info->lock);
1874 shmem_recalc_inode(inode);
1875 spin_unlock_irq(&info->lock);
1878 if (error == -EEXIST) /* from above or from radix_tree_insert */
1884 * This is like autoremove_wake_function, but it removes the wait queue
1885 * entry unconditionally - even if something else had already woken the
1888 static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1890 int ret = default_wake_function(wait, mode, sync, key);
1891 list_del_init(&wait->task_list);
1895 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1897 struct inode *inode = file_inode(vma->vm_file);
1898 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1901 int ret = VM_FAULT_LOCKED;
1904 * Trinity finds that probing a hole which tmpfs is punching can
1905 * prevent the hole-punch from ever completing: which in turn
1906 * locks writers out with its hold on i_mutex. So refrain from
1907 * faulting pages into the hole while it's being punched. Although
1908 * shmem_undo_range() does remove the additions, it may be unable to
1909 * keep up, as each new page needs its own unmap_mapping_range() call,
1910 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1912 * It does not matter if we sometimes reach this check just before the
1913 * hole-punch begins, so that one fault then races with the punch:
1914 * we just need to make racing faults a rare case.
1916 * The implementation below would be much simpler if we just used a
1917 * standard mutex or completion: but we cannot take i_mutex in fault,
1918 * and bloating every shmem inode for this unlikely case would be sad.
1920 if (unlikely(inode->i_private)) {
1921 struct shmem_falloc *shmem_falloc;
1923 spin_lock(&inode->i_lock);
1924 shmem_falloc = inode->i_private;
1926 shmem_falloc->waitq &&
1927 vmf->pgoff >= shmem_falloc->start &&
1928 vmf->pgoff < shmem_falloc->next) {
1929 wait_queue_head_t *shmem_falloc_waitq;
1930 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1932 ret = VM_FAULT_NOPAGE;
1933 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1934 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1935 /* It's polite to up mmap_sem if we can */
1936 up_read(&vma->vm_mm->mmap_sem);
1937 ret = VM_FAULT_RETRY;
1940 shmem_falloc_waitq = shmem_falloc->waitq;
1941 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1942 TASK_UNINTERRUPTIBLE);
1943 spin_unlock(&inode->i_lock);
1947 * shmem_falloc_waitq points into the shmem_fallocate()
1948 * stack of the hole-punching task: shmem_falloc_waitq
1949 * is usually invalid by the time we reach here, but
1950 * finish_wait() does not dereference it in that case;
1951 * though i_lock needed lest racing with wake_up_all().
1953 spin_lock(&inode->i_lock);
1954 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1955 spin_unlock(&inode->i_lock);
1958 spin_unlock(&inode->i_lock);
1962 if (vma->vm_flags & VM_HUGEPAGE)
1964 else if (vma->vm_flags & VM_NOHUGEPAGE)
1967 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1968 gfp, vma->vm_mm, &ret);
1970 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1974 unsigned long shmem_get_unmapped_area(struct file *file,
1975 unsigned long uaddr, unsigned long len,
1976 unsigned long pgoff, unsigned long flags)
1978 unsigned long (*get_area)(struct file *,
1979 unsigned long, unsigned long, unsigned long, unsigned long);
1981 unsigned long offset;
1982 unsigned long inflated_len;
1983 unsigned long inflated_addr;
1984 unsigned long inflated_offset;
1986 if (len > TASK_SIZE)
1989 get_area = current->mm->get_unmapped_area;
1990 addr = get_area(file, uaddr, len, pgoff, flags);
1992 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1994 if (IS_ERR_VALUE(addr))
1996 if (addr & ~PAGE_MASK)
1998 if (addr > TASK_SIZE - len)
2001 if (shmem_huge == SHMEM_HUGE_DENY)
2003 if (len < HPAGE_PMD_SIZE)
2005 if (flags & MAP_FIXED)
2008 * Our priority is to support MAP_SHARED mapped hugely;
2009 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2010 * But if caller specified an address hint, respect that as before.
2015 if (shmem_huge != SHMEM_HUGE_FORCE) {
2016 struct super_block *sb;
2019 VM_BUG_ON(file->f_op != &shmem_file_operations);
2020 sb = file_inode(file)->i_sb;
2023 * Called directly from mm/mmap.c, or drivers/char/mem.c
2024 * for "/dev/zero", to create a shared anonymous object.
2026 if (IS_ERR(shm_mnt))
2028 sb = shm_mnt->mnt_sb;
2030 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2034 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2035 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2037 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2040 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2041 if (inflated_len > TASK_SIZE)
2043 if (inflated_len < len)
2046 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2047 if (IS_ERR_VALUE(inflated_addr))
2049 if (inflated_addr & ~PAGE_MASK)
2052 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2053 inflated_addr += offset - inflated_offset;
2054 if (inflated_offset > offset)
2055 inflated_addr += HPAGE_PMD_SIZE;
2057 if (inflated_addr > TASK_SIZE - len)
2059 return inflated_addr;
2063 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2065 struct inode *inode = file_inode(vma->vm_file);
2066 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2069 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2072 struct inode *inode = file_inode(vma->vm_file);
2075 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2076 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2080 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2082 struct inode *inode = file_inode(file);
2083 struct shmem_inode_info *info = SHMEM_I(inode);
2084 int retval = -ENOMEM;
2086 spin_lock_irq(&info->lock);
2087 if (lock && !(info->flags & VM_LOCKED)) {
2088 if (!user_shm_lock(inode->i_size, user))
2090 info->flags |= VM_LOCKED;
2091 mapping_set_unevictable(file->f_mapping);
2093 if (!lock && (info->flags & VM_LOCKED) && user) {
2094 user_shm_unlock(inode->i_size, user);
2095 info->flags &= ~VM_LOCKED;
2096 mapping_clear_unevictable(file->f_mapping);
2101 spin_unlock_irq(&info->lock);
2105 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2107 file_accessed(file);
2108 vma->vm_ops = &shmem_vm_ops;
2109 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2110 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2111 (vma->vm_end & HPAGE_PMD_MASK)) {
2112 khugepaged_enter(vma, vma->vm_flags);
2117 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2118 umode_t mode, dev_t dev, unsigned long flags)
2120 struct inode *inode;
2121 struct shmem_inode_info *info;
2122 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2124 if (shmem_reserve_inode(sb))
2127 inode = new_inode(sb);
2129 inode->i_ino = get_next_ino();
2130 inode_init_owner(inode, dir, mode);
2131 inode->i_blocks = 0;
2132 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2133 inode->i_generation = get_seconds();
2134 info = SHMEM_I(inode);
2135 memset(info, 0, (char *)inode - (char *)info);
2136 spin_lock_init(&info->lock);
2137 info->seals = F_SEAL_SEAL;
2138 info->flags = flags & VM_NORESERVE;
2139 INIT_LIST_HEAD(&info->shrinklist);
2140 INIT_LIST_HEAD(&info->swaplist);
2141 simple_xattrs_init(&info->xattrs);
2142 cache_no_acl(inode);
2144 switch (mode & S_IFMT) {
2146 inode->i_op = &shmem_special_inode_operations;
2147 init_special_inode(inode, mode, dev);
2150 inode->i_mapping->a_ops = &shmem_aops;
2151 inode->i_op = &shmem_inode_operations;
2152 inode->i_fop = &shmem_file_operations;
2153 mpol_shared_policy_init(&info->policy,
2154 shmem_get_sbmpol(sbinfo));
2158 /* Some things misbehave if size == 0 on a directory */
2159 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2160 inode->i_op = &shmem_dir_inode_operations;
2161 inode->i_fop = &simple_dir_operations;
2165 * Must not load anything in the rbtree,
2166 * mpol_free_shared_policy will not be called.
2168 mpol_shared_policy_init(&info->policy, NULL);
2172 shmem_free_inode(sb);
2176 bool shmem_mapping(struct address_space *mapping)
2178 return mapping->a_ops == &shmem_aops;
2182 static const struct inode_operations shmem_symlink_inode_operations;
2183 static const struct inode_operations shmem_short_symlink_operations;
2185 #ifdef CONFIG_TMPFS_XATTR
2186 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2188 #define shmem_initxattrs NULL
2192 shmem_write_begin(struct file *file, struct address_space *mapping,
2193 loff_t pos, unsigned len, unsigned flags,
2194 struct page **pagep, void **fsdata)
2196 struct inode *inode = mapping->host;
2197 struct shmem_inode_info *info = SHMEM_I(inode);
2198 pgoff_t index = pos >> PAGE_SHIFT;
2200 /* i_mutex is held by caller */
2201 if (unlikely(info->seals)) {
2202 if (info->seals & F_SEAL_WRITE)
2204 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2208 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2212 shmem_write_end(struct file *file, struct address_space *mapping,
2213 loff_t pos, unsigned len, unsigned copied,
2214 struct page *page, void *fsdata)
2216 struct inode *inode = mapping->host;
2218 if (pos + copied > inode->i_size)
2219 i_size_write(inode, pos + copied);
2221 if (!PageUptodate(page)) {
2222 struct page *head = compound_head(page);
2223 if (PageTransCompound(page)) {
2226 for (i = 0; i < HPAGE_PMD_NR; i++) {
2227 if (head + i == page)
2229 clear_highpage(head + i);
2230 flush_dcache_page(head + i);
2233 if (copied < PAGE_SIZE) {
2234 unsigned from = pos & (PAGE_SIZE - 1);
2235 zero_user_segments(page, 0, from,
2236 from + copied, PAGE_SIZE);
2238 SetPageUptodate(head);
2240 set_page_dirty(page);
2247 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2249 struct file *file = iocb->ki_filp;
2250 struct inode *inode = file_inode(file);
2251 struct address_space *mapping = inode->i_mapping;
2253 unsigned long offset;
2254 enum sgp_type sgp = SGP_READ;
2257 loff_t *ppos = &iocb->ki_pos;
2260 * Might this read be for a stacking filesystem? Then when reading
2261 * holes of a sparse file, we actually need to allocate those pages,
2262 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2264 if (!iter_is_iovec(to))
2267 index = *ppos >> PAGE_SHIFT;
2268 offset = *ppos & ~PAGE_MASK;
2271 struct page *page = NULL;
2273 unsigned long nr, ret;
2274 loff_t i_size = i_size_read(inode);
2276 end_index = i_size >> PAGE_SHIFT;
2277 if (index > end_index)
2279 if (index == end_index) {
2280 nr = i_size & ~PAGE_MASK;
2285 error = shmem_getpage(inode, index, &page, sgp);
2287 if (error == -EINVAL)
2292 if (sgp == SGP_CACHE)
2293 set_page_dirty(page);
2298 * We must evaluate after, since reads (unlike writes)
2299 * are called without i_mutex protection against truncate
2302 i_size = i_size_read(inode);
2303 end_index = i_size >> PAGE_SHIFT;
2304 if (index == end_index) {
2305 nr = i_size & ~PAGE_MASK;
2316 * If users can be writing to this page using arbitrary
2317 * virtual addresses, take care about potential aliasing
2318 * before reading the page on the kernel side.
2320 if (mapping_writably_mapped(mapping))
2321 flush_dcache_page(page);
2323 * Mark the page accessed if we read the beginning.
2326 mark_page_accessed(page);
2328 page = ZERO_PAGE(0);
2333 * Ok, we have the page, and it's up-to-date, so
2334 * now we can copy it to user space...
2336 ret = copy_page_to_iter(page, offset, nr, to);
2339 index += offset >> PAGE_SHIFT;
2340 offset &= ~PAGE_MASK;
2343 if (!iov_iter_count(to))
2352 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2353 file_accessed(file);
2354 return retval ? retval : error;
2358 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2360 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2361 pgoff_t index, pgoff_t end, int whence)
2364 struct pagevec pvec;
2365 pgoff_t indices[PAGEVEC_SIZE];
2369 pagevec_init(&pvec, 0);
2370 pvec.nr = 1; /* start small: we may be there already */
2372 pvec.nr = find_get_entries(mapping, index,
2373 pvec.nr, pvec.pages, indices);
2375 if (whence == SEEK_DATA)
2379 for (i = 0; i < pvec.nr; i++, index++) {
2380 if (index < indices[i]) {
2381 if (whence == SEEK_HOLE) {
2387 page = pvec.pages[i];
2388 if (page && !radix_tree_exceptional_entry(page)) {
2389 if (!PageUptodate(page))
2393 (page && whence == SEEK_DATA) ||
2394 (!page && whence == SEEK_HOLE)) {
2399 pagevec_remove_exceptionals(&pvec);
2400 pagevec_release(&pvec);
2401 pvec.nr = PAGEVEC_SIZE;
2407 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2409 struct address_space *mapping = file->f_mapping;
2410 struct inode *inode = mapping->host;
2414 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2415 return generic_file_llseek_size(file, offset, whence,
2416 MAX_LFS_FILESIZE, i_size_read(inode));
2418 /* We're holding i_mutex so we can access i_size directly */
2422 else if (offset >= inode->i_size)
2425 start = offset >> PAGE_SHIFT;
2426 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2427 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2428 new_offset <<= PAGE_SHIFT;
2429 if (new_offset > offset) {
2430 if (new_offset < inode->i_size)
2431 offset = new_offset;
2432 else if (whence == SEEK_DATA)
2435 offset = inode->i_size;
2440 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2441 inode_unlock(inode);
2446 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2447 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2449 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2450 #define LAST_SCAN 4 /* about 150ms max */
2452 static void shmem_tag_pins(struct address_space *mapping)
2454 struct radix_tree_iter iter;
2463 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2464 page = radix_tree_deref_slot(slot);
2465 if (!page || radix_tree_exception(page)) {
2466 if (radix_tree_deref_retry(page)) {
2467 slot = radix_tree_iter_retry(&iter);
2470 } else if (page_count(page) - page_mapcount(page) > 1) {
2471 spin_lock_irq(&mapping->tree_lock);
2472 radix_tree_tag_set(&mapping->page_tree, iter.index,
2474 spin_unlock_irq(&mapping->tree_lock);
2477 if (need_resched()) {
2478 slot = radix_tree_iter_resume(slot, &iter);
2486 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2487 * via get_user_pages(), drivers might have some pending I/O without any active
2488 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2489 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2490 * them to be dropped.
2491 * The caller must guarantee that no new user will acquire writable references
2492 * to those pages to avoid races.
2494 static int shmem_wait_for_pins(struct address_space *mapping)
2496 struct radix_tree_iter iter;
2502 shmem_tag_pins(mapping);
2505 for (scan = 0; scan <= LAST_SCAN; scan++) {
2506 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2510 lru_add_drain_all();
2511 else if (schedule_timeout_killable((HZ << scan) / 200))
2516 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2517 start, SHMEM_TAG_PINNED) {
2519 page = radix_tree_deref_slot(slot);
2520 if (radix_tree_exception(page)) {
2521 if (radix_tree_deref_retry(page)) {
2522 slot = radix_tree_iter_retry(&iter);
2530 page_count(page) - page_mapcount(page) != 1) {
2531 if (scan < LAST_SCAN)
2532 goto continue_resched;
2535 * On the last scan, we clean up all those tags
2536 * we inserted; but make a note that we still
2537 * found pages pinned.
2542 spin_lock_irq(&mapping->tree_lock);
2543 radix_tree_tag_clear(&mapping->page_tree,
2544 iter.index, SHMEM_TAG_PINNED);
2545 spin_unlock_irq(&mapping->tree_lock);
2547 if (need_resched()) {
2548 slot = radix_tree_iter_resume(slot, &iter);
2558 #define F_ALL_SEALS (F_SEAL_SEAL | \
2563 int shmem_add_seals(struct file *file, unsigned int seals)
2565 struct inode *inode = file_inode(file);
2566 struct shmem_inode_info *info = SHMEM_I(inode);
2571 * Sealing allows multiple parties to share a shmem-file but restrict
2572 * access to a specific subset of file operations. Seals can only be
2573 * added, but never removed. This way, mutually untrusted parties can
2574 * share common memory regions with a well-defined policy. A malicious
2575 * peer can thus never perform unwanted operations on a shared object.
2577 * Seals are only supported on special shmem-files and always affect
2578 * the whole underlying inode. Once a seal is set, it may prevent some
2579 * kinds of access to the file. Currently, the following seals are
2581 * SEAL_SEAL: Prevent further seals from being set on this file
2582 * SEAL_SHRINK: Prevent the file from shrinking
2583 * SEAL_GROW: Prevent the file from growing
2584 * SEAL_WRITE: Prevent write access to the file
2586 * As we don't require any trust relationship between two parties, we
2587 * must prevent seals from being removed. Therefore, sealing a file
2588 * only adds a given set of seals to the file, it never touches
2589 * existing seals. Furthermore, the "setting seals"-operation can be
2590 * sealed itself, which basically prevents any further seal from being
2593 * Semantics of sealing are only defined on volatile files. Only
2594 * anonymous shmem files support sealing. More importantly, seals are
2595 * never written to disk. Therefore, there's no plan to support it on
2599 if (file->f_op != &shmem_file_operations)
2601 if (!(file->f_mode & FMODE_WRITE))
2603 if (seals & ~(unsigned int)F_ALL_SEALS)
2608 if (info->seals & F_SEAL_SEAL) {
2613 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2614 error = mapping_deny_writable(file->f_mapping);
2618 error = shmem_wait_for_pins(file->f_mapping);
2620 mapping_allow_writable(file->f_mapping);
2625 info->seals |= seals;
2629 inode_unlock(inode);
2632 EXPORT_SYMBOL_GPL(shmem_add_seals);
2634 int shmem_get_seals(struct file *file)
2636 if (file->f_op != &shmem_file_operations)
2639 return SHMEM_I(file_inode(file))->seals;
2641 EXPORT_SYMBOL_GPL(shmem_get_seals);
2643 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2649 /* disallow upper 32bit */
2653 error = shmem_add_seals(file, arg);
2656 error = shmem_get_seals(file);
2666 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2669 struct inode *inode = file_inode(file);
2670 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2671 struct shmem_inode_info *info = SHMEM_I(inode);
2672 struct shmem_falloc shmem_falloc;
2673 pgoff_t start, index, end;
2676 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2681 if (mode & FALLOC_FL_PUNCH_HOLE) {
2682 struct address_space *mapping = file->f_mapping;
2683 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2684 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2685 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2687 /* protected by i_mutex */
2688 if (info->seals & F_SEAL_WRITE) {
2693 shmem_falloc.waitq = &shmem_falloc_waitq;
2694 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2695 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2696 spin_lock(&inode->i_lock);
2697 inode->i_private = &shmem_falloc;
2698 spin_unlock(&inode->i_lock);
2700 if ((u64)unmap_end > (u64)unmap_start)
2701 unmap_mapping_range(mapping, unmap_start,
2702 1 + unmap_end - unmap_start, 0);
2703 shmem_truncate_range(inode, offset, offset + len - 1);
2704 /* No need to unmap again: hole-punching leaves COWed pages */
2706 spin_lock(&inode->i_lock);
2707 inode->i_private = NULL;
2708 wake_up_all(&shmem_falloc_waitq);
2709 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
2710 spin_unlock(&inode->i_lock);
2715 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2716 error = inode_newsize_ok(inode, offset + len);
2720 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2725 start = offset >> PAGE_SHIFT;
2726 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2727 /* Try to avoid a swapstorm if len is impossible to satisfy */
2728 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2733 shmem_falloc.waitq = NULL;
2734 shmem_falloc.start = start;
2735 shmem_falloc.next = start;
2736 shmem_falloc.nr_falloced = 0;
2737 shmem_falloc.nr_unswapped = 0;
2738 spin_lock(&inode->i_lock);
2739 inode->i_private = &shmem_falloc;
2740 spin_unlock(&inode->i_lock);
2742 for (index = start; index < end; index++) {
2746 * Good, the fallocate(2) manpage permits EINTR: we may have
2747 * been interrupted because we are using up too much memory.
2749 if (signal_pending(current))
2751 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2754 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2756 /* Remove the !PageUptodate pages we added */
2757 if (index > start) {
2758 shmem_undo_range(inode,
2759 (loff_t)start << PAGE_SHIFT,
2760 ((loff_t)index << PAGE_SHIFT) - 1, true);
2766 * Inform shmem_writepage() how far we have reached.
2767 * No need for lock or barrier: we have the page lock.
2769 shmem_falloc.next++;
2770 if (!PageUptodate(page))
2771 shmem_falloc.nr_falloced++;
2774 * If !PageUptodate, leave it that way so that freeable pages
2775 * can be recognized if we need to rollback on error later.
2776 * But set_page_dirty so that memory pressure will swap rather
2777 * than free the pages we are allocating (and SGP_CACHE pages
2778 * might still be clean: we now need to mark those dirty too).
2780 set_page_dirty(page);
2786 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2787 i_size_write(inode, offset + len);
2788 inode->i_ctime = current_time(inode);
2790 spin_lock(&inode->i_lock);
2791 inode->i_private = NULL;
2792 spin_unlock(&inode->i_lock);
2794 inode_unlock(inode);
2798 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2800 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2802 buf->f_type = TMPFS_MAGIC;
2803 buf->f_bsize = PAGE_SIZE;
2804 buf->f_namelen = NAME_MAX;
2805 if (sbinfo->max_blocks) {
2806 buf->f_blocks = sbinfo->max_blocks;
2808 buf->f_bfree = sbinfo->max_blocks -
2809 percpu_counter_sum(&sbinfo->used_blocks);
2811 if (sbinfo->max_inodes) {
2812 buf->f_files = sbinfo->max_inodes;
2813 buf->f_ffree = sbinfo->free_inodes;
2815 /* else leave those fields 0 like simple_statfs */
2820 * File creation. Allocate an inode, and we're done..
2823 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2825 struct inode *inode;
2826 int error = -ENOSPC;
2828 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2830 error = simple_acl_create(dir, inode);
2833 error = security_inode_init_security(inode, dir,
2835 shmem_initxattrs, NULL);
2836 if (error && error != -EOPNOTSUPP)
2840 dir->i_size += BOGO_DIRENT_SIZE;
2841 dir->i_ctime = dir->i_mtime = current_time(dir);
2842 d_instantiate(dentry, inode);
2843 dget(dentry); /* Extra count - pin the dentry in core */
2852 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2854 struct inode *inode;
2855 int error = -ENOSPC;
2857 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2859 error = security_inode_init_security(inode, dir,
2861 shmem_initxattrs, NULL);
2862 if (error && error != -EOPNOTSUPP)
2864 error = simple_acl_create(dir, inode);
2867 d_tmpfile(dentry, inode);
2875 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2879 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2885 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2888 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2894 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2896 struct inode *inode = d_inode(old_dentry);
2900 * No ordinary (disk based) filesystem counts links as inodes;
2901 * but each new link needs a new dentry, pinning lowmem, and
2902 * tmpfs dentries cannot be pruned until they are unlinked.
2904 ret = shmem_reserve_inode(inode->i_sb);
2908 dir->i_size += BOGO_DIRENT_SIZE;
2909 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2911 ihold(inode); /* New dentry reference */
2912 dget(dentry); /* Extra pinning count for the created dentry */
2913 d_instantiate(dentry, inode);
2918 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2920 struct inode *inode = d_inode(dentry);
2922 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2923 shmem_free_inode(inode->i_sb);
2925 dir->i_size -= BOGO_DIRENT_SIZE;
2926 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2928 dput(dentry); /* Undo the count from "create" - this does all the work */
2932 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2934 if (!simple_empty(dentry))
2937 drop_nlink(d_inode(dentry));
2939 return shmem_unlink(dir, dentry);
2942 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2944 bool old_is_dir = d_is_dir(old_dentry);
2945 bool new_is_dir = d_is_dir(new_dentry);
2947 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2949 drop_nlink(old_dir);
2952 drop_nlink(new_dir);
2956 old_dir->i_ctime = old_dir->i_mtime =
2957 new_dir->i_ctime = new_dir->i_mtime =
2958 d_inode(old_dentry)->i_ctime =
2959 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2964 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2966 struct dentry *whiteout;
2969 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2973 error = shmem_mknod(old_dir, whiteout,
2974 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2980 * Cheat and hash the whiteout while the old dentry is still in
2981 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2983 * d_lookup() will consistently find one of them at this point,
2984 * not sure which one, but that isn't even important.
2991 * The VFS layer already does all the dentry stuff for rename,
2992 * we just have to decrement the usage count for the target if
2993 * it exists so that the VFS layer correctly free's it when it
2996 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
2998 struct inode *inode = d_inode(old_dentry);
2999 int they_are_dirs = S_ISDIR(inode->i_mode);
3001 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3004 if (flags & RENAME_EXCHANGE)
3005 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3007 if (!simple_empty(new_dentry))
3010 if (flags & RENAME_WHITEOUT) {
3013 error = shmem_whiteout(old_dir, old_dentry);
3018 if (d_really_is_positive(new_dentry)) {
3019 (void) shmem_unlink(new_dir, new_dentry);
3020 if (they_are_dirs) {
3021 drop_nlink(d_inode(new_dentry));
3022 drop_nlink(old_dir);
3024 } else if (they_are_dirs) {
3025 drop_nlink(old_dir);
3029 old_dir->i_size -= BOGO_DIRENT_SIZE;
3030 new_dir->i_size += BOGO_DIRENT_SIZE;
3031 old_dir->i_ctime = old_dir->i_mtime =
3032 new_dir->i_ctime = new_dir->i_mtime =
3033 inode->i_ctime = current_time(old_dir);
3037 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3041 struct inode *inode;
3043 struct shmem_inode_info *info;
3045 len = strlen(symname) + 1;
3046 if (len > PAGE_SIZE)
3047 return -ENAMETOOLONG;
3049 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3053 error = security_inode_init_security(inode, dir, &dentry->d_name,
3054 shmem_initxattrs, NULL);
3056 if (error != -EOPNOTSUPP) {
3063 info = SHMEM_I(inode);
3064 inode->i_size = len-1;
3065 if (len <= SHORT_SYMLINK_LEN) {
3066 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3067 if (!inode->i_link) {
3071 inode->i_op = &shmem_short_symlink_operations;
3073 inode_nohighmem(inode);
3074 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3079 inode->i_mapping->a_ops = &shmem_aops;
3080 inode->i_op = &shmem_symlink_inode_operations;
3081 memcpy(page_address(page), symname, len);
3082 SetPageUptodate(page);
3083 set_page_dirty(page);
3087 dir->i_size += BOGO_DIRENT_SIZE;
3088 dir->i_ctime = dir->i_mtime = current_time(dir);
3089 d_instantiate(dentry, inode);
3094 static void shmem_put_link(void *arg)
3096 mark_page_accessed(arg);
3100 static const char *shmem_get_link(struct dentry *dentry,
3101 struct inode *inode,
3102 struct delayed_call *done)
3104 struct page *page = NULL;
3107 page = find_get_page(inode->i_mapping, 0);
3109 return ERR_PTR(-ECHILD);
3110 if (!PageUptodate(page)) {
3112 return ERR_PTR(-ECHILD);
3115 error = shmem_getpage(inode, 0, &page, SGP_READ);
3117 return ERR_PTR(error);
3120 set_delayed_call(done, shmem_put_link, page);
3121 return page_address(page);
3124 #ifdef CONFIG_TMPFS_XATTR
3126 * Superblocks without xattr inode operations may get some security.* xattr
3127 * support from the LSM "for free". As soon as we have any other xattrs
3128 * like ACLs, we also need to implement the security.* handlers at
3129 * filesystem level, though.
3133 * Callback for security_inode_init_security() for acquiring xattrs.
3135 static int shmem_initxattrs(struct inode *inode,
3136 const struct xattr *xattr_array,
3139 struct shmem_inode_info *info = SHMEM_I(inode);
3140 const struct xattr *xattr;
3141 struct simple_xattr *new_xattr;
3144 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3145 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3149 len = strlen(xattr->name) + 1;
3150 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3152 if (!new_xattr->name) {
3157 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3158 XATTR_SECURITY_PREFIX_LEN);
3159 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3162 simple_xattr_list_add(&info->xattrs, new_xattr);
3168 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3169 struct dentry *unused, struct inode *inode,
3170 const char *name, void *buffer, size_t size)
3172 struct shmem_inode_info *info = SHMEM_I(inode);
3174 name = xattr_full_name(handler, name);
3175 return simple_xattr_get(&info->xattrs, name, buffer, size);
3178 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3179 struct dentry *unused, struct inode *inode,
3180 const char *name, const void *value,
3181 size_t size, int flags)
3183 struct shmem_inode_info *info = SHMEM_I(inode);
3185 name = xattr_full_name(handler, name);
3186 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3189 static const struct xattr_handler shmem_security_xattr_handler = {
3190 .prefix = XATTR_SECURITY_PREFIX,
3191 .get = shmem_xattr_handler_get,
3192 .set = shmem_xattr_handler_set,
3195 static const struct xattr_handler shmem_trusted_xattr_handler = {
3196 .prefix = XATTR_TRUSTED_PREFIX,
3197 .get = shmem_xattr_handler_get,
3198 .set = shmem_xattr_handler_set,
3201 static const struct xattr_handler *shmem_xattr_handlers[] = {
3202 #ifdef CONFIG_TMPFS_POSIX_ACL
3203 &posix_acl_access_xattr_handler,
3204 &posix_acl_default_xattr_handler,
3206 &shmem_security_xattr_handler,
3207 &shmem_trusted_xattr_handler,
3211 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3213 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3214 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3216 #endif /* CONFIG_TMPFS_XATTR */
3218 static const struct inode_operations shmem_short_symlink_operations = {
3219 .get_link = simple_get_link,
3220 #ifdef CONFIG_TMPFS_XATTR
3221 .listxattr = shmem_listxattr,
3225 static const struct inode_operations shmem_symlink_inode_operations = {
3226 .get_link = shmem_get_link,
3227 #ifdef CONFIG_TMPFS_XATTR
3228 .listxattr = shmem_listxattr,
3232 static struct dentry *shmem_get_parent(struct dentry *child)
3234 return ERR_PTR(-ESTALE);
3237 static int shmem_match(struct inode *ino, void *vfh)
3241 inum = (inum << 32) | fh[1];
3242 return ino->i_ino == inum && fh[0] == ino->i_generation;
3245 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3246 struct fid *fid, int fh_len, int fh_type)
3248 struct inode *inode;
3249 struct dentry *dentry = NULL;
3256 inum = (inum << 32) | fid->raw[1];
3258 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3259 shmem_match, fid->raw);
3261 dentry = d_find_alias(inode);
3268 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3269 struct inode *parent)
3273 return FILEID_INVALID;
3276 if (inode_unhashed(inode)) {
3277 /* Unfortunately insert_inode_hash is not idempotent,
3278 * so as we hash inodes here rather than at creation
3279 * time, we need a lock to ensure we only try
3282 static DEFINE_SPINLOCK(lock);
3284 if (inode_unhashed(inode))
3285 __insert_inode_hash(inode,
3286 inode->i_ino + inode->i_generation);
3290 fh[0] = inode->i_generation;
3291 fh[1] = inode->i_ino;
3292 fh[2] = ((__u64)inode->i_ino) >> 32;
3298 static const struct export_operations shmem_export_ops = {
3299 .get_parent = shmem_get_parent,
3300 .encode_fh = shmem_encode_fh,
3301 .fh_to_dentry = shmem_fh_to_dentry,
3304 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3307 char *this_char, *value, *rest;
3308 struct mempolicy *mpol = NULL;
3312 while (options != NULL) {
3313 this_char = options;
3316 * NUL-terminate this option: unfortunately,
3317 * mount options form a comma-separated list,
3318 * but mpol's nodelist may also contain commas.
3320 options = strchr(options, ',');
3321 if (options == NULL)
3324 if (!isdigit(*options)) {
3331 if ((value = strchr(this_char,'=')) != NULL) {
3334 pr_err("tmpfs: No value for mount option '%s'\n",
3339 if (!strcmp(this_char,"size")) {
3340 unsigned long long size;
3341 size = memparse(value,&rest);
3343 size <<= PAGE_SHIFT;
3344 size *= totalram_pages;
3350 sbinfo->max_blocks =
3351 DIV_ROUND_UP(size, PAGE_SIZE);
3352 } else if (!strcmp(this_char,"nr_blocks")) {
3353 sbinfo->max_blocks = memparse(value, &rest);
3356 } else if (!strcmp(this_char,"nr_inodes")) {
3357 sbinfo->max_inodes = memparse(value, &rest);
3360 } else if (!strcmp(this_char,"mode")) {
3363 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3366 } else if (!strcmp(this_char,"uid")) {
3369 uid = simple_strtoul(value, &rest, 0);
3372 sbinfo->uid = make_kuid(current_user_ns(), uid);
3373 if (!uid_valid(sbinfo->uid))
3375 } else if (!strcmp(this_char,"gid")) {
3378 gid = simple_strtoul(value, &rest, 0);
3381 sbinfo->gid = make_kgid(current_user_ns(), gid);
3382 if (!gid_valid(sbinfo->gid))
3384 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3385 } else if (!strcmp(this_char, "huge")) {
3387 huge = shmem_parse_huge(value);
3390 if (!has_transparent_hugepage() &&
3391 huge != SHMEM_HUGE_NEVER)
3393 sbinfo->huge = huge;
3396 } else if (!strcmp(this_char,"mpol")) {
3399 if (mpol_parse_str(value, &mpol))
3403 pr_err("tmpfs: Bad mount option %s\n", this_char);
3407 sbinfo->mpol = mpol;
3411 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3419 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3421 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3422 struct shmem_sb_info config = *sbinfo;
3423 unsigned long inodes;
3424 int error = -EINVAL;
3427 if (shmem_parse_options(data, &config, true))
3430 spin_lock(&sbinfo->stat_lock);
3431 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3432 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3434 if (config.max_inodes < inodes)
3437 * Those tests disallow limited->unlimited while any are in use;
3438 * but we must separately disallow unlimited->limited, because
3439 * in that case we have no record of how much is already in use.
3441 if (config.max_blocks && !sbinfo->max_blocks)
3443 if (config.max_inodes && !sbinfo->max_inodes)
3447 sbinfo->huge = config.huge;
3448 sbinfo->max_blocks = config.max_blocks;
3449 sbinfo->max_inodes = config.max_inodes;
3450 sbinfo->free_inodes = config.max_inodes - inodes;
3453 * Preserve previous mempolicy unless mpol remount option was specified.
3456 mpol_put(sbinfo->mpol);
3457 sbinfo->mpol = config.mpol; /* transfers initial ref */
3460 spin_unlock(&sbinfo->stat_lock);
3464 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3466 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3468 if (sbinfo->max_blocks != shmem_default_max_blocks())
3469 seq_printf(seq, ",size=%luk",
3470 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3471 if (sbinfo->max_inodes != shmem_default_max_inodes())
3472 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3473 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3474 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3475 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3476 seq_printf(seq, ",uid=%u",
3477 from_kuid_munged(&init_user_ns, sbinfo->uid));
3478 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3479 seq_printf(seq, ",gid=%u",
3480 from_kgid_munged(&init_user_ns, sbinfo->gid));
3481 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3482 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3484 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3486 shmem_show_mpol(seq, sbinfo->mpol);
3490 #define MFD_NAME_PREFIX "memfd:"
3491 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3492 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3494 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3496 SYSCALL_DEFINE2(memfd_create,
3497 const char __user *, uname,
3498 unsigned int, flags)
3500 struct shmem_inode_info *info;
3506 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3509 /* length includes terminating zero */
3510 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3513 if (len > MFD_NAME_MAX_LEN + 1)
3516 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3520 strcpy(name, MFD_NAME_PREFIX);
3521 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3526 /* terminating-zero may have changed after strnlen_user() returned */
3527 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3532 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3538 file = shmem_file_setup(name, 0, VM_NORESERVE);
3540 error = PTR_ERR(file);
3543 info = SHMEM_I(file_inode(file));
3544 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3545 file->f_flags |= O_RDWR | O_LARGEFILE;
3546 if (flags & MFD_ALLOW_SEALING)
3547 info->seals &= ~F_SEAL_SEAL;
3549 fd_install(fd, file);
3560 #endif /* CONFIG_TMPFS */
3562 static void shmem_put_super(struct super_block *sb)
3564 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3566 percpu_counter_destroy(&sbinfo->used_blocks);
3567 mpol_put(sbinfo->mpol);
3569 sb->s_fs_info = NULL;
3572 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3574 struct inode *inode;
3575 struct shmem_sb_info *sbinfo;
3578 /* Round up to L1_CACHE_BYTES to resist false sharing */
3579 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3580 L1_CACHE_BYTES), GFP_KERNEL);
3584 sbinfo->mode = S_IRWXUGO | S_ISVTX;
3585 sbinfo->uid = current_fsuid();
3586 sbinfo->gid = current_fsgid();
3587 sb->s_fs_info = sbinfo;
3591 * Per default we only allow half of the physical ram per
3592 * tmpfs instance, limiting inodes to one per page of lowmem;
3593 * but the internal instance is left unlimited.
3595 if (!(sb->s_flags & MS_KERNMOUNT)) {
3596 sbinfo->max_blocks = shmem_default_max_blocks();
3597 sbinfo->max_inodes = shmem_default_max_inodes();
3598 if (shmem_parse_options(data, sbinfo, false)) {
3603 sb->s_flags |= MS_NOUSER;
3605 sb->s_export_op = &shmem_export_ops;
3606 sb->s_flags |= MS_NOSEC;
3608 sb->s_flags |= MS_NOUSER;
3611 spin_lock_init(&sbinfo->stat_lock);
3612 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3614 sbinfo->free_inodes = sbinfo->max_inodes;
3615 spin_lock_init(&sbinfo->shrinklist_lock);
3616 INIT_LIST_HEAD(&sbinfo->shrinklist);
3618 sb->s_maxbytes = MAX_LFS_FILESIZE;
3619 sb->s_blocksize = PAGE_SIZE;
3620 sb->s_blocksize_bits = PAGE_SHIFT;
3621 sb->s_magic = TMPFS_MAGIC;
3622 sb->s_op = &shmem_ops;
3623 sb->s_time_gran = 1;
3624 #ifdef CONFIG_TMPFS_XATTR
3625 sb->s_xattr = shmem_xattr_handlers;
3627 #ifdef CONFIG_TMPFS_POSIX_ACL
3628 sb->s_flags |= MS_POSIXACL;
3631 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3634 inode->i_uid = sbinfo->uid;
3635 inode->i_gid = sbinfo->gid;
3636 sb->s_root = d_make_root(inode);
3642 shmem_put_super(sb);
3646 static struct kmem_cache *shmem_inode_cachep;
3648 static struct inode *shmem_alloc_inode(struct super_block *sb)
3650 struct shmem_inode_info *info;
3651 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3654 return &info->vfs_inode;
3657 static void shmem_destroy_callback(struct rcu_head *head)
3659 struct inode *inode = container_of(head, struct inode, i_rcu);
3660 if (S_ISLNK(inode->i_mode))
3661 kfree(inode->i_link);
3662 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3665 static void shmem_destroy_inode(struct inode *inode)
3667 if (S_ISREG(inode->i_mode))
3668 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3669 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3672 static void shmem_init_inode(void *foo)
3674 struct shmem_inode_info *info = foo;
3675 inode_init_once(&info->vfs_inode);
3678 static int shmem_init_inodecache(void)
3680 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3681 sizeof(struct shmem_inode_info),
3682 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3686 static void shmem_destroy_inodecache(void)
3688 kmem_cache_destroy(shmem_inode_cachep);
3691 static const struct address_space_operations shmem_aops = {
3692 .writepage = shmem_writepage,
3693 .set_page_dirty = __set_page_dirty_no_writeback,
3695 .write_begin = shmem_write_begin,
3696 .write_end = shmem_write_end,
3698 #ifdef CONFIG_MIGRATION
3699 .migratepage = migrate_page,
3701 .error_remove_page = generic_error_remove_page,
3704 static const struct file_operations shmem_file_operations = {
3706 .get_unmapped_area = shmem_get_unmapped_area,
3708 .llseek = shmem_file_llseek,
3709 .read_iter = shmem_file_read_iter,
3710 .write_iter = generic_file_write_iter,
3711 .fsync = noop_fsync,
3712 .splice_read = generic_file_splice_read,
3713 .splice_write = iter_file_splice_write,
3714 .fallocate = shmem_fallocate,
3718 static const struct inode_operations shmem_inode_operations = {
3719 .getattr = shmem_getattr,
3720 .setattr = shmem_setattr,
3721 #ifdef CONFIG_TMPFS_XATTR
3722 .listxattr = shmem_listxattr,
3723 .set_acl = simple_set_acl,
3727 static const struct inode_operations shmem_dir_inode_operations = {
3729 .create = shmem_create,
3730 .lookup = simple_lookup,
3732 .unlink = shmem_unlink,
3733 .symlink = shmem_symlink,
3734 .mkdir = shmem_mkdir,
3735 .rmdir = shmem_rmdir,
3736 .mknod = shmem_mknod,
3737 .rename = shmem_rename2,
3738 .tmpfile = shmem_tmpfile,
3740 #ifdef CONFIG_TMPFS_XATTR
3741 .listxattr = shmem_listxattr,
3743 #ifdef CONFIG_TMPFS_POSIX_ACL
3744 .setattr = shmem_setattr,
3745 .set_acl = simple_set_acl,
3749 static const struct inode_operations shmem_special_inode_operations = {
3750 #ifdef CONFIG_TMPFS_XATTR
3751 .listxattr = shmem_listxattr,
3753 #ifdef CONFIG_TMPFS_POSIX_ACL
3754 .setattr = shmem_setattr,
3755 .set_acl = simple_set_acl,
3759 static const struct super_operations shmem_ops = {
3760 .alloc_inode = shmem_alloc_inode,
3761 .destroy_inode = shmem_destroy_inode,
3763 .statfs = shmem_statfs,
3764 .remount_fs = shmem_remount_fs,
3765 .show_options = shmem_show_options,
3767 .evict_inode = shmem_evict_inode,
3768 .drop_inode = generic_delete_inode,
3769 .put_super = shmem_put_super,
3770 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3771 .nr_cached_objects = shmem_unused_huge_count,
3772 .free_cached_objects = shmem_unused_huge_scan,
3776 static const struct vm_operations_struct shmem_vm_ops = {
3777 .fault = shmem_fault,
3778 .map_pages = filemap_map_pages,
3780 .set_policy = shmem_set_policy,
3781 .get_policy = shmem_get_policy,
3785 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3786 int flags, const char *dev_name, void *data)
3788 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3791 static struct file_system_type shmem_fs_type = {
3792 .owner = THIS_MODULE,
3794 .mount = shmem_mount,
3795 .kill_sb = kill_litter_super,
3796 .fs_flags = FS_USERNS_MOUNT,
3799 int __init shmem_init(void)
3803 /* If rootfs called this, don't re-init */
3804 if (shmem_inode_cachep)
3807 error = shmem_init_inodecache();
3811 error = register_filesystem(&shmem_fs_type);
3813 pr_err("Could not register tmpfs\n");
3817 shm_mnt = kern_mount(&shmem_fs_type);
3818 if (IS_ERR(shm_mnt)) {
3819 error = PTR_ERR(shm_mnt);
3820 pr_err("Could not kern_mount tmpfs\n");
3824 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3825 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3826 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3828 shmem_huge = 0; /* just in case it was patched */
3833 unregister_filesystem(&shmem_fs_type);
3835 shmem_destroy_inodecache();
3837 shm_mnt = ERR_PTR(error);
3841 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3842 static ssize_t shmem_enabled_show(struct kobject *kobj,
3843 struct kobj_attribute *attr, char *buf)
3847 SHMEM_HUGE_WITHIN_SIZE,
3855 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3856 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3858 count += sprintf(buf + count, fmt,
3859 shmem_format_huge(values[i]));
3861 buf[count - 1] = '\n';
3865 static ssize_t shmem_enabled_store(struct kobject *kobj,
3866 struct kobj_attribute *attr, const char *buf, size_t count)
3871 if (count + 1 > sizeof(tmp))
3873 memcpy(tmp, buf, count);
3875 if (count && tmp[count - 1] == '\n')
3876 tmp[count - 1] = '\0';
3878 huge = shmem_parse_huge(tmp);
3879 if (huge == -EINVAL)
3881 if (!has_transparent_hugepage() &&
3882 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3886 if (shmem_huge < SHMEM_HUGE_DENY)
3887 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3891 struct kobj_attribute shmem_enabled_attr =
3892 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3893 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3895 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3896 bool shmem_huge_enabled(struct vm_area_struct *vma)
3898 struct inode *inode = file_inode(vma->vm_file);
3899 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3903 if (shmem_huge == SHMEM_HUGE_FORCE)
3905 if (shmem_huge == SHMEM_HUGE_DENY)
3907 switch (sbinfo->huge) {
3908 case SHMEM_HUGE_NEVER:
3910 case SHMEM_HUGE_ALWAYS:
3912 case SHMEM_HUGE_WITHIN_SIZE:
3913 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3914 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3915 if (i_size >= HPAGE_PMD_SIZE &&
3916 i_size >> PAGE_SHIFT >= off)
3918 case SHMEM_HUGE_ADVISE:
3919 /* TODO: implement fadvise() hints */
3920 return (vma->vm_flags & VM_HUGEPAGE);
3926 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3928 #else /* !CONFIG_SHMEM */
3931 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3933 * This is intended for small system where the benefits of the full
3934 * shmem code (swap-backed and resource-limited) are outweighed by
3935 * their complexity. On systems without swap this code should be
3936 * effectively equivalent, but much lighter weight.
3939 static struct file_system_type shmem_fs_type = {
3941 .mount = ramfs_mount,
3942 .kill_sb = kill_litter_super,
3943 .fs_flags = FS_USERNS_MOUNT,
3946 int __init shmem_init(void)
3948 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3950 shm_mnt = kern_mount(&shmem_fs_type);
3951 BUG_ON(IS_ERR(shm_mnt));
3956 int shmem_unuse(swp_entry_t swap, struct page *page)
3961 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3966 void shmem_unlock_mapping(struct address_space *mapping)
3971 unsigned long shmem_get_unmapped_area(struct file *file,
3972 unsigned long addr, unsigned long len,
3973 unsigned long pgoff, unsigned long flags)
3975 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3979 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3981 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
3983 EXPORT_SYMBOL_GPL(shmem_truncate_range);
3985 #define shmem_vm_ops generic_file_vm_ops
3986 #define shmem_file_operations ramfs_file_operations
3987 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3988 #define shmem_acct_size(flags, size) 0
3989 #define shmem_unacct_size(flags, size) do {} while (0)
3991 #endif /* CONFIG_SHMEM */
3995 static const struct dentry_operations anon_ops = {
3996 .d_dname = simple_dname
3999 static struct file *__shmem_file_setup(const char *name, loff_t size,
4000 unsigned long flags, unsigned int i_flags)
4003 struct inode *inode;
4005 struct super_block *sb;
4008 if (IS_ERR(shm_mnt))
4009 return ERR_CAST(shm_mnt);
4011 if (size < 0 || size > MAX_LFS_FILESIZE)
4012 return ERR_PTR(-EINVAL);
4014 if (shmem_acct_size(flags, size))
4015 return ERR_PTR(-ENOMEM);
4017 res = ERR_PTR(-ENOMEM);
4019 this.len = strlen(name);
4020 this.hash = 0; /* will go */
4021 sb = shm_mnt->mnt_sb;
4022 path.mnt = mntget(shm_mnt);
4023 path.dentry = d_alloc_pseudo(sb, &this);
4026 d_set_d_op(path.dentry, &anon_ops);
4028 res = ERR_PTR(-ENOSPC);
4029 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4033 inode->i_flags |= i_flags;
4034 d_instantiate(path.dentry, inode);
4035 inode->i_size = size;
4036 clear_nlink(inode); /* It is unlinked */
4037 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4041 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4042 &shmem_file_operations);
4049 shmem_unacct_size(flags, size);
4056 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4057 * kernel internal. There will be NO LSM permission checks against the
4058 * underlying inode. So users of this interface must do LSM checks at a
4059 * higher layer. The users are the big_key and shm implementations. LSM
4060 * checks are provided at the key or shm level rather than the inode.
4061 * @name: name for dentry (to be seen in /proc/<pid>/maps
4062 * @size: size to be set for the file
4063 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4065 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4067 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4071 * shmem_file_setup - get an unlinked file living in tmpfs
4072 * @name: name for dentry (to be seen in /proc/<pid>/maps
4073 * @size: size to be set for the file
4074 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4076 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4078 return __shmem_file_setup(name, size, flags, 0);
4080 EXPORT_SYMBOL_GPL(shmem_file_setup);
4083 * shmem_zero_setup - setup a shared anonymous mapping
4084 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4086 int shmem_zero_setup(struct vm_area_struct *vma)
4089 loff_t size = vma->vm_end - vma->vm_start;
4092 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4093 * between XFS directory reading and selinux: since this file is only
4094 * accessible to the user through its mapping, use S_PRIVATE flag to
4095 * bypass file security, in the same way as shmem_kernel_file_setup().
4097 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4099 return PTR_ERR(file);
4103 vma->vm_file = file;
4104 vma->vm_ops = &shmem_vm_ops;
4106 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4107 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4108 (vma->vm_end & HPAGE_PMD_MASK)) {
4109 khugepaged_enter(vma, vma->vm_flags);
4116 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4117 * @mapping: the page's address_space
4118 * @index: the page index
4119 * @gfp: the page allocator flags to use if allocating
4121 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4122 * with any new page allocations done using the specified allocation flags.
4123 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4124 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4125 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4127 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4128 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4130 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4131 pgoff_t index, gfp_t gfp)
4134 struct inode *inode = mapping->host;
4138 BUG_ON(mapping->a_ops != &shmem_aops);
4139 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4142 page = ERR_PTR(error);
4148 * The tiny !SHMEM case uses ramfs without swap
4150 return read_cache_page_gfp(mapping, index, gfp);
4153 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);