2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
35 #include <linux/khugepaged.h>
37 static struct vfsmount *shm_mnt;
41 * This virtual memory filesystem is heavily based on the ramfs. It
42 * extends ramfs by the ability to use swap and honor resource limits
43 * which makes it a completely usable filesystem.
46 #include <linux/xattr.h>
47 #include <linux/exportfs.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/mman.h>
51 #include <linux/string.h>
52 #include <linux/slab.h>
53 #include <linux/backing-dev.h>
54 #include <linux/shmem_fs.h>
55 #include <linux/writeback.h>
56 #include <linux/blkdev.h>
57 #include <linux/pagevec.h>
58 #include <linux/percpu_counter.h>
59 #include <linux/falloc.h>
60 #include <linux/splice.h>
61 #include <linux/security.h>
62 #include <linux/swapops.h>
63 #include <linux/mempolicy.h>
64 #include <linux/namei.h>
65 #include <linux/ctype.h>
66 #include <linux/migrate.h>
67 #include <linux/highmem.h>
68 #include <linux/seq_file.h>
69 #include <linux/magic.h>
70 #include <linux/syscalls.h>
71 #include <linux/fcntl.h>
72 #include <uapi/linux/memfd.h>
74 #include <asm/uaccess.h>
75 #include <asm/pgtable.h>
79 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
80 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
82 /* Pretend that each entry is of this size in directory's i_size */
83 #define BOGO_DIRENT_SIZE 20
85 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
86 #define SHORT_SYMLINK_LEN 128
89 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
90 * inode->i_private (with i_mutex making sure that it has only one user at
91 * a time): we would prefer not to enlarge the shmem inode just for that.
94 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
95 pgoff_t start; /* start of range currently being fallocated */
96 pgoff_t next; /* the next page offset to be fallocated */
97 pgoff_t nr_falloced; /* how many new pages have been fallocated */
98 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
102 static unsigned long shmem_default_max_blocks(void)
104 return totalram_pages / 2;
107 static unsigned long shmem_default_max_inodes(void)
109 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
113 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
114 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
115 struct shmem_inode_info *info, pgoff_t index);
116 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
117 struct page **pagep, enum sgp_type sgp,
118 gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
120 int shmem_getpage(struct inode *inode, pgoff_t index,
121 struct page **pagep, enum sgp_type sgp)
123 return shmem_getpage_gfp(inode, index, pagep, sgp,
124 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
127 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
129 return sb->s_fs_info;
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
138 static inline int shmem_acct_size(unsigned long flags, loff_t size)
140 return (flags & VM_NORESERVE) ?
141 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
144 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
146 if (!(flags & VM_NORESERVE))
147 vm_unacct_memory(VM_ACCT(size));
150 static inline int shmem_reacct_size(unsigned long flags,
151 loff_t oldsize, loff_t newsize)
153 if (!(flags & VM_NORESERVE)) {
154 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
155 return security_vm_enough_memory_mm(current->mm,
156 VM_ACCT(newsize) - VM_ACCT(oldsize));
157 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
158 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
164 * ... whereas tmpfs objects are accounted incrementally as
165 * pages are allocated, in order to allow large sparse files.
166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169 static inline int shmem_acct_block(unsigned long flags, long pages)
171 if (!(flags & VM_NORESERVE))
174 return security_vm_enough_memory_mm(current->mm,
175 pages * VM_ACCT(PAGE_SIZE));
178 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
180 if (flags & VM_NORESERVE)
181 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
184 static 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;
419 struct shmem_inode_info *info;
421 unsigned long batch = sc ? sc->nr_to_scan : 128;
422 int removed = 0, split = 0;
424 if (list_empty(&sbinfo->shrinklist))
427 spin_lock(&sbinfo->shrinklist_lock);
428 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
429 info = list_entry(pos, struct shmem_inode_info, shrinklist);
432 inode = igrab(&info->vfs_inode);
434 /* inode is about to be evicted */
436 list_del_init(&info->shrinklist);
441 /* Check if there's anything to gain */
442 if (round_up(inode->i_size, PAGE_SIZE) ==
443 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
444 list_del_init(&info->shrinklist);
450 list_move(&info->shrinklist, &list);
455 spin_unlock(&sbinfo->shrinklist_lock);
457 list_for_each_safe(pos, next, &list) {
460 info = list_entry(pos, struct shmem_inode_info, shrinklist);
461 inode = &info->vfs_inode;
463 if (nr_to_split && split >= nr_to_split) {
468 page = find_lock_page(inode->i_mapping,
469 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
473 if (!PageTransHuge(page)) {
479 ret = split_huge_page(page);
484 /* split failed: leave it on the list */
491 list_del_init(&info->shrinklist);
496 spin_lock(&sbinfo->shrinklist_lock);
497 list_splice_tail(&list, &sbinfo->shrinklist);
498 sbinfo->shrinklist_len -= removed;
499 spin_unlock(&sbinfo->shrinklist_lock);
504 static long shmem_unused_huge_scan(struct super_block *sb,
505 struct shrink_control *sc)
507 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
509 if (!READ_ONCE(sbinfo->shrinklist_len))
512 return shmem_unused_huge_shrink(sbinfo, sc, 0);
515 static long shmem_unused_huge_count(struct super_block *sb,
516 struct shrink_control *sc)
518 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
519 return READ_ONCE(sbinfo->shrinklist_len);
521 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
523 #define shmem_huge SHMEM_HUGE_DENY
525 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
526 struct shrink_control *sc, unsigned long nr_to_split)
530 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
533 * Like add_to_page_cache_locked, but error if expected item has gone.
535 static int shmem_add_to_page_cache(struct page *page,
536 struct address_space *mapping,
537 pgoff_t index, void *expected)
539 int error, nr = hpage_nr_pages(page);
541 VM_BUG_ON_PAGE(PageTail(page), page);
542 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
543 VM_BUG_ON_PAGE(!PageLocked(page), page);
544 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
545 VM_BUG_ON(expected && PageTransHuge(page));
547 page_ref_add(page, nr);
548 page->mapping = mapping;
551 spin_lock_irq(&mapping->tree_lock);
552 if (PageTransHuge(page)) {
553 void __rcu **results;
558 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
559 &results, &idx, index, 1) &&
560 idx < index + HPAGE_PMD_NR) {
565 for (i = 0; i < HPAGE_PMD_NR; i++) {
566 error = radix_tree_insert(&mapping->page_tree,
567 index + i, page + i);
570 count_vm_event(THP_FILE_ALLOC);
572 } else if (!expected) {
573 error = radix_tree_insert(&mapping->page_tree, index, page);
575 error = shmem_radix_tree_replace(mapping, index, expected,
580 mapping->nrpages += nr;
581 if (PageTransHuge(page))
582 __inc_node_page_state(page, NR_SHMEM_THPS);
583 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
584 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
585 spin_unlock_irq(&mapping->tree_lock);
587 page->mapping = NULL;
588 spin_unlock_irq(&mapping->tree_lock);
589 page_ref_sub(page, nr);
595 * Like delete_from_page_cache, but substitutes swap for page.
597 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
599 struct address_space *mapping = page->mapping;
602 VM_BUG_ON_PAGE(PageCompound(page), page);
604 spin_lock_irq(&mapping->tree_lock);
605 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
606 page->mapping = NULL;
608 __dec_node_page_state(page, NR_FILE_PAGES);
609 __dec_node_page_state(page, NR_SHMEM);
610 spin_unlock_irq(&mapping->tree_lock);
616 * Remove swap entry from radix tree, free the swap and its page cache.
618 static int shmem_free_swap(struct address_space *mapping,
619 pgoff_t index, void *radswap)
623 spin_lock_irq(&mapping->tree_lock);
624 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
625 spin_unlock_irq(&mapping->tree_lock);
628 free_swap_and_cache(radix_to_swp_entry(radswap));
633 * Determine (in bytes) how many of the shmem object's pages mapped by the
634 * given offsets are swapped out.
636 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
637 * as long as the inode doesn't go away and racy results are not a problem.
639 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
640 pgoff_t start, pgoff_t end)
642 struct radix_tree_iter iter;
645 unsigned long swapped = 0;
649 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
650 if (iter.index >= end)
653 page = radix_tree_deref_slot(slot);
655 if (radix_tree_deref_retry(page)) {
656 slot = radix_tree_iter_retry(&iter);
660 if (radix_tree_exceptional_entry(page))
663 if (need_resched()) {
665 slot = radix_tree_iter_next(&iter);
671 return swapped << PAGE_SHIFT;
675 * Determine (in bytes) how many of the shmem object's pages mapped by the
676 * given vma is swapped out.
678 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
679 * as long as the inode doesn't go away and racy results are not a problem.
681 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
683 struct inode *inode = file_inode(vma->vm_file);
684 struct shmem_inode_info *info = SHMEM_I(inode);
685 struct address_space *mapping = inode->i_mapping;
686 unsigned long swapped;
688 /* Be careful as we don't hold info->lock */
689 swapped = READ_ONCE(info->swapped);
692 * The easier cases are when the shmem object has nothing in swap, or
693 * the vma maps it whole. Then we can simply use the stats that we
699 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
700 return swapped << PAGE_SHIFT;
702 /* Here comes the more involved part */
703 return shmem_partial_swap_usage(mapping,
704 linear_page_index(vma, vma->vm_start),
705 linear_page_index(vma, vma->vm_end));
709 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
711 void shmem_unlock_mapping(struct address_space *mapping)
714 pgoff_t indices[PAGEVEC_SIZE];
717 pagevec_init(&pvec, 0);
719 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
721 while (!mapping_unevictable(mapping)) {
723 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
724 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
726 pvec.nr = find_get_entries(mapping, index,
727 PAGEVEC_SIZE, pvec.pages, indices);
730 index = indices[pvec.nr - 1] + 1;
731 pagevec_remove_exceptionals(&pvec);
732 check_move_unevictable_pages(pvec.pages, pvec.nr);
733 pagevec_release(&pvec);
739 * Remove range of pages and swap entries from radix tree, and free them.
740 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
742 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
745 struct address_space *mapping = inode->i_mapping;
746 struct shmem_inode_info *info = SHMEM_I(inode);
747 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
748 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
749 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
750 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
752 pgoff_t indices[PAGEVEC_SIZE];
753 long nr_swaps_freed = 0;
758 end = -1; /* unsigned, so actually very big */
760 pagevec_init(&pvec, 0);
762 while (index < end) {
763 pvec.nr = find_get_entries(mapping, index,
764 min(end - index, (pgoff_t)PAGEVEC_SIZE),
765 pvec.pages, indices);
768 for (i = 0; i < pagevec_count(&pvec); i++) {
769 struct page *page = pvec.pages[i];
775 if (radix_tree_exceptional_entry(page)) {
778 nr_swaps_freed += !shmem_free_swap(mapping,
783 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
785 if (!trylock_page(page))
788 if (PageTransTail(page)) {
789 /* Middle of THP: zero out the page */
790 clear_highpage(page);
793 } else if (PageTransHuge(page)) {
794 if (index == round_down(end, HPAGE_PMD_NR)) {
796 * Range ends in the middle of THP:
799 clear_highpage(page);
803 index += HPAGE_PMD_NR - 1;
804 i += HPAGE_PMD_NR - 1;
807 if (!unfalloc || !PageUptodate(page)) {
808 VM_BUG_ON_PAGE(PageTail(page), page);
809 if (page_mapping(page) == mapping) {
810 VM_BUG_ON_PAGE(PageWriteback(page), page);
811 truncate_inode_page(mapping, page);
816 pagevec_remove_exceptionals(&pvec);
817 pagevec_release(&pvec);
823 struct page *page = NULL;
824 shmem_getpage(inode, start - 1, &page, SGP_READ);
826 unsigned int top = PAGE_SIZE;
831 zero_user_segment(page, partial_start, top);
832 set_page_dirty(page);
838 struct page *page = NULL;
839 shmem_getpage(inode, end, &page, SGP_READ);
841 zero_user_segment(page, 0, partial_end);
842 set_page_dirty(page);
851 while (index < end) {
854 pvec.nr = find_get_entries(mapping, index,
855 min(end - index, (pgoff_t)PAGEVEC_SIZE),
856 pvec.pages, indices);
858 /* If all gone or hole-punch or unfalloc, we're done */
859 if (index == start || end != -1)
861 /* But if truncating, restart to make sure all gone */
865 for (i = 0; i < pagevec_count(&pvec); i++) {
866 struct page *page = pvec.pages[i];
872 if (radix_tree_exceptional_entry(page)) {
875 if (shmem_free_swap(mapping, index, page)) {
876 /* Swap was replaced by page: retry */
886 if (PageTransTail(page)) {
887 /* Middle of THP: zero out the page */
888 clear_highpage(page);
891 * Partial thp truncate due 'start' in middle
892 * of THP: don't need to look on these pages
893 * again on !pvec.nr restart.
895 if (index != round_down(end, HPAGE_PMD_NR))
898 } else if (PageTransHuge(page)) {
899 if (index == round_down(end, HPAGE_PMD_NR)) {
901 * Range ends in the middle of THP:
904 clear_highpage(page);
908 index += HPAGE_PMD_NR - 1;
909 i += HPAGE_PMD_NR - 1;
912 if (!unfalloc || !PageUptodate(page)) {
913 VM_BUG_ON_PAGE(PageTail(page), page);
914 if (page_mapping(page) == mapping) {
915 VM_BUG_ON_PAGE(PageWriteback(page), page);
916 truncate_inode_page(mapping, page);
918 /* Page was replaced by swap: retry */
926 pagevec_remove_exceptionals(&pvec);
927 pagevec_release(&pvec);
931 spin_lock_irq(&info->lock);
932 info->swapped -= nr_swaps_freed;
933 shmem_recalc_inode(inode);
934 spin_unlock_irq(&info->lock);
937 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
939 shmem_undo_range(inode, lstart, lend, false);
940 inode->i_ctime = inode->i_mtime = current_time(inode);
942 EXPORT_SYMBOL_GPL(shmem_truncate_range);
944 static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
947 struct inode *inode = dentry->d_inode;
948 struct shmem_inode_info *info = SHMEM_I(inode);
950 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
951 spin_lock_irq(&info->lock);
952 shmem_recalc_inode(inode);
953 spin_unlock_irq(&info->lock);
955 generic_fillattr(inode, stat);
959 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
961 struct inode *inode = d_inode(dentry);
962 struct shmem_inode_info *info = SHMEM_I(inode);
963 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
966 error = setattr_prepare(dentry, attr);
970 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
971 loff_t oldsize = inode->i_size;
972 loff_t newsize = attr->ia_size;
974 /* protected by i_mutex */
975 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
976 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
979 if (newsize != oldsize) {
980 error = shmem_reacct_size(SHMEM_I(inode)->flags,
984 i_size_write(inode, newsize);
985 inode->i_ctime = inode->i_mtime = current_time(inode);
987 if (newsize <= oldsize) {
988 loff_t holebegin = round_up(newsize, PAGE_SIZE);
989 if (oldsize > holebegin)
990 unmap_mapping_range(inode->i_mapping,
993 shmem_truncate_range(inode,
994 newsize, (loff_t)-1);
995 /* unmap again to remove racily COWed private pages */
996 if (oldsize > holebegin)
997 unmap_mapping_range(inode->i_mapping,
1001 * Part of the huge page can be beyond i_size: subject
1002 * to shrink under memory pressure.
1004 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1005 spin_lock(&sbinfo->shrinklist_lock);
1006 if (list_empty(&info->shrinklist)) {
1007 list_add_tail(&info->shrinklist,
1008 &sbinfo->shrinklist);
1009 sbinfo->shrinklist_len++;
1011 spin_unlock(&sbinfo->shrinklist_lock);
1016 setattr_copy(inode, attr);
1017 if (attr->ia_valid & ATTR_MODE)
1018 error = posix_acl_chmod(inode, inode->i_mode);
1022 static void shmem_evict_inode(struct inode *inode)
1024 struct shmem_inode_info *info = SHMEM_I(inode);
1025 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1027 if (inode->i_mapping->a_ops == &shmem_aops) {
1028 shmem_unacct_size(info->flags, inode->i_size);
1030 shmem_truncate_range(inode, 0, (loff_t)-1);
1031 if (!list_empty(&info->shrinklist)) {
1032 spin_lock(&sbinfo->shrinklist_lock);
1033 if (!list_empty(&info->shrinklist)) {
1034 list_del_init(&info->shrinklist);
1035 sbinfo->shrinklist_len--;
1037 spin_unlock(&sbinfo->shrinklist_lock);
1039 if (!list_empty(&info->swaplist)) {
1040 mutex_lock(&shmem_swaplist_mutex);
1041 list_del_init(&info->swaplist);
1042 mutex_unlock(&shmem_swaplist_mutex);
1046 simple_xattrs_free(&info->xattrs);
1047 WARN_ON(inode->i_blocks);
1048 shmem_free_inode(inode->i_sb);
1053 * If swap found in inode, free it and move page from swapcache to filecache.
1055 static int shmem_unuse_inode(struct shmem_inode_info *info,
1056 swp_entry_t swap, struct page **pagep)
1058 struct address_space *mapping = info->vfs_inode.i_mapping;
1064 radswap = swp_to_radix_entry(swap);
1065 index = radix_tree_locate_item(&mapping->page_tree, radswap);
1067 return -EAGAIN; /* tell shmem_unuse we found nothing */
1070 * Move _head_ to start search for next from here.
1071 * But be careful: shmem_evict_inode checks list_empty without taking
1072 * mutex, and there's an instant in list_move_tail when info->swaplist
1073 * would appear empty, if it were the only one on shmem_swaplist.
1075 if (shmem_swaplist.next != &info->swaplist)
1076 list_move_tail(&shmem_swaplist, &info->swaplist);
1078 gfp = mapping_gfp_mask(mapping);
1079 if (shmem_should_replace_page(*pagep, gfp)) {
1080 mutex_unlock(&shmem_swaplist_mutex);
1081 error = shmem_replace_page(pagep, gfp, info, index);
1082 mutex_lock(&shmem_swaplist_mutex);
1084 * We needed to drop mutex to make that restrictive page
1085 * allocation, but the inode might have been freed while we
1086 * dropped it: although a racing shmem_evict_inode() cannot
1087 * complete without emptying the radix_tree, our page lock
1088 * on this swapcache page is not enough to prevent that -
1089 * free_swap_and_cache() of our swap entry will only
1090 * trylock_page(), removing swap from radix_tree whatever.
1092 * We must not proceed to shmem_add_to_page_cache() if the
1093 * inode has been freed, but of course we cannot rely on
1094 * inode or mapping or info to check that. However, we can
1095 * safely check if our swap entry is still in use (and here
1096 * it can't have got reused for another page): if it's still
1097 * in use, then the inode cannot have been freed yet, and we
1098 * can safely proceed (if it's no longer in use, that tells
1099 * nothing about the inode, but we don't need to unuse swap).
1101 if (!page_swapcount(*pagep))
1106 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1107 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1108 * beneath us (pagelock doesn't help until the page is in pagecache).
1111 error = shmem_add_to_page_cache(*pagep, mapping, index,
1113 if (error != -ENOMEM) {
1115 * Truncation and eviction use free_swap_and_cache(), which
1116 * only does trylock page: if we raced, best clean up here.
1118 delete_from_swap_cache(*pagep);
1119 set_page_dirty(*pagep);
1121 spin_lock_irq(&info->lock);
1123 spin_unlock_irq(&info->lock);
1131 * Search through swapped inodes to find and replace swap by page.
1133 int shmem_unuse(swp_entry_t swap, struct page *page)
1135 struct list_head *this, *next;
1136 struct shmem_inode_info *info;
1137 struct mem_cgroup *memcg;
1141 * There's a faint possibility that swap page was replaced before
1142 * caller locked it: caller will come back later with the right page.
1144 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1148 * Charge page using GFP_KERNEL while we can wait, before taking
1149 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1150 * Charged back to the user (not to caller) when swap account is used.
1152 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1156 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1159 mutex_lock(&shmem_swaplist_mutex);
1160 list_for_each_safe(this, next, &shmem_swaplist) {
1161 info = list_entry(this, struct shmem_inode_info, swaplist);
1163 error = shmem_unuse_inode(info, swap, &page);
1165 list_del_init(&info->swaplist);
1167 if (error != -EAGAIN)
1169 /* found nothing in this: move on to search the next */
1171 mutex_unlock(&shmem_swaplist_mutex);
1174 if (error != -ENOMEM)
1176 mem_cgroup_cancel_charge(page, memcg, false);
1178 mem_cgroup_commit_charge(page, memcg, true, false);
1186 * Move the page from the page cache to the swap cache.
1188 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1190 struct shmem_inode_info *info;
1191 struct address_space *mapping;
1192 struct inode *inode;
1196 VM_BUG_ON_PAGE(PageCompound(page), page);
1197 BUG_ON(!PageLocked(page));
1198 mapping = page->mapping;
1199 index = page->index;
1200 inode = mapping->host;
1201 info = SHMEM_I(inode);
1202 if (info->flags & VM_LOCKED)
1204 if (!total_swap_pages)
1208 * Our capabilities prevent regular writeback or sync from ever calling
1209 * shmem_writepage; but a stacking filesystem might use ->writepage of
1210 * its underlying filesystem, in which case tmpfs should write out to
1211 * swap only in response to memory pressure, and not for the writeback
1214 if (!wbc->for_reclaim) {
1215 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1220 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1221 * value into swapfile.c, the only way we can correctly account for a
1222 * fallocated page arriving here is now to initialize it and write it.
1224 * That's okay for a page already fallocated earlier, but if we have
1225 * not yet completed the fallocation, then (a) we want to keep track
1226 * of this page in case we have to undo it, and (b) it may not be a
1227 * good idea to continue anyway, once we're pushing into swap. So
1228 * reactivate the page, and let shmem_fallocate() quit when too many.
1230 if (!PageUptodate(page)) {
1231 if (inode->i_private) {
1232 struct shmem_falloc *shmem_falloc;
1233 spin_lock(&inode->i_lock);
1234 shmem_falloc = inode->i_private;
1236 !shmem_falloc->waitq &&
1237 index >= shmem_falloc->start &&
1238 index < shmem_falloc->next)
1239 shmem_falloc->nr_unswapped++;
1241 shmem_falloc = NULL;
1242 spin_unlock(&inode->i_lock);
1246 clear_highpage(page);
1247 flush_dcache_page(page);
1248 SetPageUptodate(page);
1251 swap = get_swap_page();
1255 if (mem_cgroup_try_charge_swap(page, swap))
1259 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1260 * if it's not already there. Do it now before the page is
1261 * moved to swap cache, when its pagelock no longer protects
1262 * the inode from eviction. But don't unlock the mutex until
1263 * we've incremented swapped, because shmem_unuse_inode() will
1264 * prune a !swapped inode from the swaplist under this mutex.
1266 mutex_lock(&shmem_swaplist_mutex);
1267 if (list_empty(&info->swaplist))
1268 list_add_tail(&info->swaplist, &shmem_swaplist);
1270 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1271 spin_lock_irq(&info->lock);
1272 shmem_recalc_inode(inode);
1274 spin_unlock_irq(&info->lock);
1276 swap_shmem_alloc(swap);
1277 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1279 mutex_unlock(&shmem_swaplist_mutex);
1280 BUG_ON(page_mapped(page));
1281 swap_writepage(page, wbc);
1285 mutex_unlock(&shmem_swaplist_mutex);
1287 swapcache_free(swap);
1289 set_page_dirty(page);
1290 if (wbc->for_reclaim)
1291 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1296 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1297 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1301 if (!mpol || mpol->mode == MPOL_DEFAULT)
1302 return; /* show nothing */
1304 mpol_to_str(buffer, sizeof(buffer), mpol);
1306 seq_printf(seq, ",mpol=%s", buffer);
1309 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1311 struct mempolicy *mpol = NULL;
1313 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1314 mpol = sbinfo->mpol;
1316 spin_unlock(&sbinfo->stat_lock);
1320 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1321 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1324 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1328 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1330 #define vm_policy vm_private_data
1333 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1334 struct shmem_inode_info *info, pgoff_t index)
1336 /* Create a pseudo vma that just contains the policy */
1338 /* Bias interleave by inode number to distribute better across nodes */
1339 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1341 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1344 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1346 /* Drop reference taken by mpol_shared_policy_lookup() */
1347 mpol_cond_put(vma->vm_policy);
1350 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1351 struct shmem_inode_info *info, pgoff_t index)
1353 struct vm_area_struct pvma;
1356 shmem_pseudo_vma_init(&pvma, info, index);
1357 page = swapin_readahead(swap, gfp, &pvma, 0);
1358 shmem_pseudo_vma_destroy(&pvma);
1363 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1364 struct shmem_inode_info *info, pgoff_t index)
1366 struct vm_area_struct pvma;
1367 struct inode *inode = &info->vfs_inode;
1368 struct address_space *mapping = inode->i_mapping;
1369 pgoff_t idx, hindex;
1370 void __rcu **results;
1373 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1376 hindex = round_down(index, HPAGE_PMD_NR);
1378 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1379 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1385 shmem_pseudo_vma_init(&pvma, info, hindex);
1386 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1387 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1388 shmem_pseudo_vma_destroy(&pvma);
1390 prep_transhuge_page(page);
1394 static struct page *shmem_alloc_page(gfp_t gfp,
1395 struct shmem_inode_info *info, pgoff_t index)
1397 struct vm_area_struct pvma;
1400 shmem_pseudo_vma_init(&pvma, info, index);
1401 page = alloc_page_vma(gfp, &pvma, 0);
1402 shmem_pseudo_vma_destroy(&pvma);
1407 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1408 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1409 pgoff_t index, bool huge)
1415 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1417 nr = huge ? HPAGE_PMD_NR : 1;
1419 if (shmem_acct_block(info->flags, nr))
1421 if (sbinfo->max_blocks) {
1422 if (percpu_counter_compare(&sbinfo->used_blocks,
1423 sbinfo->max_blocks - nr) > 0)
1425 percpu_counter_add(&sbinfo->used_blocks, nr);
1429 page = shmem_alloc_hugepage(gfp, info, index);
1431 page = shmem_alloc_page(gfp, info, index);
1433 __SetPageLocked(page);
1434 __SetPageSwapBacked(page);
1439 if (sbinfo->max_blocks)
1440 percpu_counter_add(&sbinfo->used_blocks, -nr);
1442 shmem_unacct_blocks(info->flags, nr);
1444 return ERR_PTR(err);
1448 * When a page is moved from swapcache to shmem filecache (either by the
1449 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1450 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1451 * ignorance of the mapping it belongs to. If that mapping has special
1452 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1453 * we may need to copy to a suitable page before moving to filecache.
1455 * In a future release, this may well be extended to respect cpuset and
1456 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1457 * but for now it is a simple matter of zone.
1459 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1461 return page_zonenum(page) > gfp_zone(gfp);
1464 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1465 struct shmem_inode_info *info, pgoff_t index)
1467 struct page *oldpage, *newpage;
1468 struct address_space *swap_mapping;
1473 swap_index = page_private(oldpage);
1474 swap_mapping = page_mapping(oldpage);
1477 * We have arrived here because our zones are constrained, so don't
1478 * limit chance of success by further cpuset and node constraints.
1480 gfp &= ~GFP_CONSTRAINT_MASK;
1481 newpage = shmem_alloc_page(gfp, info, index);
1486 copy_highpage(newpage, oldpage);
1487 flush_dcache_page(newpage);
1489 __SetPageLocked(newpage);
1490 __SetPageSwapBacked(newpage);
1491 SetPageUptodate(newpage);
1492 set_page_private(newpage, swap_index);
1493 SetPageSwapCache(newpage);
1496 * Our caller will very soon move newpage out of swapcache, but it's
1497 * a nice clean interface for us to replace oldpage by newpage there.
1499 spin_lock_irq(&swap_mapping->tree_lock);
1500 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1503 __inc_node_page_state(newpage, NR_FILE_PAGES);
1504 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1506 spin_unlock_irq(&swap_mapping->tree_lock);
1508 if (unlikely(error)) {
1510 * Is this possible? I think not, now that our callers check
1511 * both PageSwapCache and page_private after getting page lock;
1512 * but be defensive. Reverse old to newpage for clear and free.
1516 mem_cgroup_migrate(oldpage, newpage);
1517 lru_cache_add_anon(newpage);
1521 ClearPageSwapCache(oldpage);
1522 set_page_private(oldpage, 0);
1524 unlock_page(oldpage);
1531 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1533 * If we allocate a new one we do not mark it dirty. That's up to the
1534 * vm. If we swap it in we mark it dirty since we also free the swap
1535 * entry since a page cannot live in both the swap and page cache.
1537 * fault_mm and fault_type are only supplied by shmem_fault:
1538 * otherwise they are NULL.
1540 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1541 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1542 struct mm_struct *fault_mm, int *fault_type)
1544 struct address_space *mapping = inode->i_mapping;
1545 struct shmem_inode_info *info = SHMEM_I(inode);
1546 struct shmem_sb_info *sbinfo;
1547 struct mm_struct *charge_mm;
1548 struct mem_cgroup *memcg;
1551 enum sgp_type sgp_huge = sgp;
1552 pgoff_t hindex = index;
1557 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1559 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1563 page = find_lock_entry(mapping, index);
1564 if (radix_tree_exceptional_entry(page)) {
1565 swap = radix_to_swp_entry(page);
1569 if (sgp <= SGP_CACHE &&
1570 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1575 if (page && sgp == SGP_WRITE)
1576 mark_page_accessed(page);
1578 /* fallocated page? */
1579 if (page && !PageUptodate(page)) {
1580 if (sgp != SGP_READ)
1586 if (page || (sgp == SGP_READ && !swap.val)) {
1592 * Fast cache lookup did not find it:
1593 * bring it back from swap or allocate.
1595 sbinfo = SHMEM_SB(inode->i_sb);
1596 charge_mm = fault_mm ? : current->mm;
1599 /* Look it up and read it in.. */
1600 page = lookup_swap_cache(swap);
1602 /* Or update major stats only when swapin succeeds?? */
1604 *fault_type |= VM_FAULT_MAJOR;
1605 count_vm_event(PGMAJFAULT);
1606 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1608 /* Here we actually start the io */
1609 page = shmem_swapin(swap, gfp, info, index);
1616 /* We have to do this with page locked to prevent races */
1618 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1619 !shmem_confirm_swap(mapping, index, swap)) {
1620 error = -EEXIST; /* try again */
1623 if (!PageUptodate(page)) {
1627 wait_on_page_writeback(page);
1629 if (shmem_should_replace_page(page, gfp)) {
1630 error = shmem_replace_page(&page, gfp, info, index);
1635 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1638 error = shmem_add_to_page_cache(page, mapping, index,
1639 swp_to_radix_entry(swap));
1641 * We already confirmed swap under page lock, and make
1642 * no memory allocation here, so usually no possibility
1643 * of error; but free_swap_and_cache() only trylocks a
1644 * page, so it is just possible that the entry has been
1645 * truncated or holepunched since swap was confirmed.
1646 * shmem_undo_range() will have done some of the
1647 * unaccounting, now delete_from_swap_cache() will do
1649 * Reset swap.val? No, leave it so "failed" goes back to
1650 * "repeat": reading a hole and writing should succeed.
1653 mem_cgroup_cancel_charge(page, memcg, false);
1654 delete_from_swap_cache(page);
1660 mem_cgroup_commit_charge(page, memcg, true, false);
1662 spin_lock_irq(&info->lock);
1664 shmem_recalc_inode(inode);
1665 spin_unlock_irq(&info->lock);
1667 if (sgp == SGP_WRITE)
1668 mark_page_accessed(page);
1670 delete_from_swap_cache(page);
1671 set_page_dirty(page);
1675 /* shmem_symlink() */
1676 if (mapping->a_ops != &shmem_aops)
1678 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1680 if (shmem_huge == SHMEM_HUGE_FORCE)
1682 switch (sbinfo->huge) {
1685 case SHMEM_HUGE_NEVER:
1687 case SHMEM_HUGE_WITHIN_SIZE:
1688 off = round_up(index, HPAGE_PMD_NR);
1689 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1690 if (i_size >= HPAGE_PMD_SIZE &&
1691 i_size >> PAGE_SHIFT >= off)
1694 case SHMEM_HUGE_ADVISE:
1695 if (sgp_huge == SGP_HUGE)
1697 /* TODO: implement fadvise() hints */
1702 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1705 alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1710 error = PTR_ERR(page);
1712 if (error != -ENOSPC)
1715 * Try to reclaim some spece by splitting a huge page
1716 * beyond i_size on the filesystem.
1720 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1721 if (ret == SHRINK_STOP)
1729 if (PageTransHuge(page))
1730 hindex = round_down(index, HPAGE_PMD_NR);
1734 if (sgp == SGP_WRITE)
1735 __SetPageReferenced(page);
1737 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1738 PageTransHuge(page));
1741 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1742 compound_order(page));
1744 error = shmem_add_to_page_cache(page, mapping, hindex,
1746 radix_tree_preload_end();
1749 mem_cgroup_cancel_charge(page, memcg,
1750 PageTransHuge(page));
1753 mem_cgroup_commit_charge(page, memcg, false,
1754 PageTransHuge(page));
1755 lru_cache_add_anon(page);
1757 spin_lock_irq(&info->lock);
1758 info->alloced += 1 << compound_order(page);
1759 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1760 shmem_recalc_inode(inode);
1761 spin_unlock_irq(&info->lock);
1764 if (PageTransHuge(page) &&
1765 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1766 hindex + HPAGE_PMD_NR - 1) {
1768 * Part of the huge page is beyond i_size: subject
1769 * to shrink under memory pressure.
1771 spin_lock(&sbinfo->shrinklist_lock);
1772 if (list_empty(&info->shrinklist)) {
1773 list_add_tail(&info->shrinklist,
1774 &sbinfo->shrinklist);
1775 sbinfo->shrinklist_len++;
1777 spin_unlock(&sbinfo->shrinklist_lock);
1781 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1783 if (sgp == SGP_FALLOC)
1787 * Let SGP_WRITE caller clear ends if write does not fill page;
1788 * but SGP_FALLOC on a page fallocated earlier must initialize
1789 * it now, lest undo on failure cancel our earlier guarantee.
1791 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1792 struct page *head = compound_head(page);
1795 for (i = 0; i < (1 << compound_order(head)); i++) {
1796 clear_highpage(head + i);
1797 flush_dcache_page(head + i);
1799 SetPageUptodate(head);
1803 /* Perhaps the file has been truncated since we checked */
1804 if (sgp <= SGP_CACHE &&
1805 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1807 ClearPageDirty(page);
1808 delete_from_page_cache(page);
1809 spin_lock_irq(&info->lock);
1810 shmem_recalc_inode(inode);
1811 spin_unlock_irq(&info->lock);
1816 *pagep = page + index - hindex;
1823 if (sbinfo->max_blocks)
1824 percpu_counter_sub(&sbinfo->used_blocks,
1825 1 << compound_order(page));
1826 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1828 if (PageTransHuge(page)) {
1834 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1841 if (error == -ENOSPC && !once++) {
1842 spin_lock_irq(&info->lock);
1843 shmem_recalc_inode(inode);
1844 spin_unlock_irq(&info->lock);
1847 if (error == -EEXIST) /* from above or from radix_tree_insert */
1853 * This is like autoremove_wake_function, but it removes the wait queue
1854 * entry unconditionally - even if something else had already woken the
1857 static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1859 int ret = default_wake_function(wait, mode, sync, key);
1860 list_del_init(&wait->task_list);
1864 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1866 struct inode *inode = file_inode(vma->vm_file);
1867 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1870 int ret = VM_FAULT_LOCKED;
1873 * Trinity finds that probing a hole which tmpfs is punching can
1874 * prevent the hole-punch from ever completing: which in turn
1875 * locks writers out with its hold on i_mutex. So refrain from
1876 * faulting pages into the hole while it's being punched. Although
1877 * shmem_undo_range() does remove the additions, it may be unable to
1878 * keep up, as each new page needs its own unmap_mapping_range() call,
1879 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1881 * It does not matter if we sometimes reach this check just before the
1882 * hole-punch begins, so that one fault then races with the punch:
1883 * we just need to make racing faults a rare case.
1885 * The implementation below would be much simpler if we just used a
1886 * standard mutex or completion: but we cannot take i_mutex in fault,
1887 * and bloating every shmem inode for this unlikely case would be sad.
1889 if (unlikely(inode->i_private)) {
1890 struct shmem_falloc *shmem_falloc;
1892 spin_lock(&inode->i_lock);
1893 shmem_falloc = inode->i_private;
1895 shmem_falloc->waitq &&
1896 vmf->pgoff >= shmem_falloc->start &&
1897 vmf->pgoff < shmem_falloc->next) {
1898 wait_queue_head_t *shmem_falloc_waitq;
1899 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1901 ret = VM_FAULT_NOPAGE;
1902 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1903 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1904 /* It's polite to up mmap_sem if we can */
1905 up_read(&vma->vm_mm->mmap_sem);
1906 ret = VM_FAULT_RETRY;
1909 shmem_falloc_waitq = shmem_falloc->waitq;
1910 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1911 TASK_UNINTERRUPTIBLE);
1912 spin_unlock(&inode->i_lock);
1916 * shmem_falloc_waitq points into the shmem_fallocate()
1917 * stack of the hole-punching task: shmem_falloc_waitq
1918 * is usually invalid by the time we reach here, but
1919 * finish_wait() does not dereference it in that case;
1920 * though i_lock needed lest racing with wake_up_all().
1922 spin_lock(&inode->i_lock);
1923 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1924 spin_unlock(&inode->i_lock);
1927 spin_unlock(&inode->i_lock);
1931 if (vma->vm_flags & VM_HUGEPAGE)
1933 else if (vma->vm_flags & VM_NOHUGEPAGE)
1936 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1937 gfp, vma->vm_mm, &ret);
1939 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1943 unsigned long shmem_get_unmapped_area(struct file *file,
1944 unsigned long uaddr, unsigned long len,
1945 unsigned long pgoff, unsigned long flags)
1947 unsigned long (*get_area)(struct file *,
1948 unsigned long, unsigned long, unsigned long, unsigned long);
1950 unsigned long offset;
1951 unsigned long inflated_len;
1952 unsigned long inflated_addr;
1953 unsigned long inflated_offset;
1955 if (len > TASK_SIZE)
1958 get_area = current->mm->get_unmapped_area;
1959 addr = get_area(file, uaddr, len, pgoff, flags);
1961 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1963 if (IS_ERR_VALUE(addr))
1965 if (addr & ~PAGE_MASK)
1967 if (addr > TASK_SIZE - len)
1970 if (shmem_huge == SHMEM_HUGE_DENY)
1972 if (len < HPAGE_PMD_SIZE)
1974 if (flags & MAP_FIXED)
1977 * Our priority is to support MAP_SHARED mapped hugely;
1978 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1979 * But if caller specified an address hint, respect that as before.
1984 if (shmem_huge != SHMEM_HUGE_FORCE) {
1985 struct super_block *sb;
1988 VM_BUG_ON(file->f_op != &shmem_file_operations);
1989 sb = file_inode(file)->i_sb;
1992 * Called directly from mm/mmap.c, or drivers/char/mem.c
1993 * for "/dev/zero", to create a shared anonymous object.
1995 if (IS_ERR(shm_mnt))
1997 sb = shm_mnt->mnt_sb;
1999 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2003 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2004 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2006 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2009 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2010 if (inflated_len > TASK_SIZE)
2012 if (inflated_len < len)
2015 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2016 if (IS_ERR_VALUE(inflated_addr))
2018 if (inflated_addr & ~PAGE_MASK)
2021 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2022 inflated_addr += offset - inflated_offset;
2023 if (inflated_offset > offset)
2024 inflated_addr += HPAGE_PMD_SIZE;
2026 if (inflated_addr > TASK_SIZE - len)
2028 return inflated_addr;
2032 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2034 struct inode *inode = file_inode(vma->vm_file);
2035 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2038 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2041 struct inode *inode = file_inode(vma->vm_file);
2044 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2045 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2049 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2051 struct inode *inode = file_inode(file);
2052 struct shmem_inode_info *info = SHMEM_I(inode);
2053 int retval = -ENOMEM;
2055 spin_lock_irq(&info->lock);
2056 if (lock && !(info->flags & VM_LOCKED)) {
2057 if (!user_shm_lock(inode->i_size, user))
2059 info->flags |= VM_LOCKED;
2060 mapping_set_unevictable(file->f_mapping);
2062 if (!lock && (info->flags & VM_LOCKED) && user) {
2063 user_shm_unlock(inode->i_size, user);
2064 info->flags &= ~VM_LOCKED;
2065 mapping_clear_unevictable(file->f_mapping);
2070 spin_unlock_irq(&info->lock);
2074 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2076 file_accessed(file);
2077 vma->vm_ops = &shmem_vm_ops;
2078 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2079 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2080 (vma->vm_end & HPAGE_PMD_MASK)) {
2081 khugepaged_enter(vma, vma->vm_flags);
2086 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2087 umode_t mode, dev_t dev, unsigned long flags)
2089 struct inode *inode;
2090 struct shmem_inode_info *info;
2091 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2093 if (shmem_reserve_inode(sb))
2096 inode = new_inode(sb);
2098 inode->i_ino = get_next_ino();
2099 inode_init_owner(inode, dir, mode);
2100 inode->i_blocks = 0;
2101 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2102 inode->i_generation = get_seconds();
2103 info = SHMEM_I(inode);
2104 memset(info, 0, (char *)inode - (char *)info);
2105 spin_lock_init(&info->lock);
2106 info->seals = F_SEAL_SEAL;
2107 info->flags = flags & VM_NORESERVE;
2108 INIT_LIST_HEAD(&info->shrinklist);
2109 INIT_LIST_HEAD(&info->swaplist);
2110 simple_xattrs_init(&info->xattrs);
2111 cache_no_acl(inode);
2113 switch (mode & S_IFMT) {
2115 inode->i_op = &shmem_special_inode_operations;
2116 init_special_inode(inode, mode, dev);
2119 inode->i_mapping->a_ops = &shmem_aops;
2120 inode->i_op = &shmem_inode_operations;
2121 inode->i_fop = &shmem_file_operations;
2122 mpol_shared_policy_init(&info->policy,
2123 shmem_get_sbmpol(sbinfo));
2127 /* Some things misbehave if size == 0 on a directory */
2128 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2129 inode->i_op = &shmem_dir_inode_operations;
2130 inode->i_fop = &simple_dir_operations;
2134 * Must not load anything in the rbtree,
2135 * mpol_free_shared_policy will not be called.
2137 mpol_shared_policy_init(&info->policy, NULL);
2141 shmem_free_inode(sb);
2145 bool shmem_mapping(struct address_space *mapping)
2150 return mapping->host->i_sb->s_op == &shmem_ops;
2154 static const struct inode_operations shmem_symlink_inode_operations;
2155 static const struct inode_operations shmem_short_symlink_operations;
2157 #ifdef CONFIG_TMPFS_XATTR
2158 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2160 #define shmem_initxattrs NULL
2164 shmem_write_begin(struct file *file, struct address_space *mapping,
2165 loff_t pos, unsigned len, unsigned flags,
2166 struct page **pagep, void **fsdata)
2168 struct inode *inode = mapping->host;
2169 struct shmem_inode_info *info = SHMEM_I(inode);
2170 pgoff_t index = pos >> PAGE_SHIFT;
2172 /* i_mutex is held by caller */
2173 if (unlikely(info->seals)) {
2174 if (info->seals & F_SEAL_WRITE)
2176 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2180 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2184 shmem_write_end(struct file *file, struct address_space *mapping,
2185 loff_t pos, unsigned len, unsigned copied,
2186 struct page *page, void *fsdata)
2188 struct inode *inode = mapping->host;
2190 if (pos + copied > inode->i_size)
2191 i_size_write(inode, pos + copied);
2193 if (!PageUptodate(page)) {
2194 struct page *head = compound_head(page);
2195 if (PageTransCompound(page)) {
2198 for (i = 0; i < HPAGE_PMD_NR; i++) {
2199 if (head + i == page)
2201 clear_highpage(head + i);
2202 flush_dcache_page(head + i);
2205 if (copied < PAGE_SIZE) {
2206 unsigned from = pos & (PAGE_SIZE - 1);
2207 zero_user_segments(page, 0, from,
2208 from + copied, PAGE_SIZE);
2210 SetPageUptodate(head);
2212 set_page_dirty(page);
2219 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2221 struct file *file = iocb->ki_filp;
2222 struct inode *inode = file_inode(file);
2223 struct address_space *mapping = inode->i_mapping;
2225 unsigned long offset;
2226 enum sgp_type sgp = SGP_READ;
2229 loff_t *ppos = &iocb->ki_pos;
2232 * Might this read be for a stacking filesystem? Then when reading
2233 * holes of a sparse file, we actually need to allocate those pages,
2234 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2236 if (!iter_is_iovec(to))
2239 index = *ppos >> PAGE_SHIFT;
2240 offset = *ppos & ~PAGE_MASK;
2243 struct page *page = NULL;
2245 unsigned long nr, ret;
2246 loff_t i_size = i_size_read(inode);
2248 end_index = i_size >> PAGE_SHIFT;
2249 if (index > end_index)
2251 if (index == end_index) {
2252 nr = i_size & ~PAGE_MASK;
2257 error = shmem_getpage(inode, index, &page, sgp);
2259 if (error == -EINVAL)
2264 if (sgp == SGP_CACHE)
2265 set_page_dirty(page);
2270 * We must evaluate after, since reads (unlike writes)
2271 * are called without i_mutex protection against truncate
2274 i_size = i_size_read(inode);
2275 end_index = i_size >> PAGE_SHIFT;
2276 if (index == end_index) {
2277 nr = i_size & ~PAGE_MASK;
2288 * If users can be writing to this page using arbitrary
2289 * virtual addresses, take care about potential aliasing
2290 * before reading the page on the kernel side.
2292 if (mapping_writably_mapped(mapping))
2293 flush_dcache_page(page);
2295 * Mark the page accessed if we read the beginning.
2298 mark_page_accessed(page);
2300 page = ZERO_PAGE(0);
2305 * Ok, we have the page, and it's up-to-date, so
2306 * now we can copy it to user space...
2308 ret = copy_page_to_iter(page, offset, nr, to);
2311 index += offset >> PAGE_SHIFT;
2312 offset &= ~PAGE_MASK;
2315 if (!iov_iter_count(to))
2324 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2325 file_accessed(file);
2326 return retval ? retval : error;
2330 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2332 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2333 pgoff_t index, pgoff_t end, int whence)
2336 struct pagevec pvec;
2337 pgoff_t indices[PAGEVEC_SIZE];
2341 pagevec_init(&pvec, 0);
2342 pvec.nr = 1; /* start small: we may be there already */
2344 pvec.nr = find_get_entries(mapping, index,
2345 pvec.nr, pvec.pages, indices);
2347 if (whence == SEEK_DATA)
2351 for (i = 0; i < pvec.nr; i++, index++) {
2352 if (index < indices[i]) {
2353 if (whence == SEEK_HOLE) {
2359 page = pvec.pages[i];
2360 if (page && !radix_tree_exceptional_entry(page)) {
2361 if (!PageUptodate(page))
2365 (page && whence == SEEK_DATA) ||
2366 (!page && whence == SEEK_HOLE)) {
2371 pagevec_remove_exceptionals(&pvec);
2372 pagevec_release(&pvec);
2373 pvec.nr = PAGEVEC_SIZE;
2379 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2381 struct address_space *mapping = file->f_mapping;
2382 struct inode *inode = mapping->host;
2386 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2387 return generic_file_llseek_size(file, offset, whence,
2388 MAX_LFS_FILESIZE, i_size_read(inode));
2390 /* We're holding i_mutex so we can access i_size directly */
2394 else if (offset >= inode->i_size)
2397 start = offset >> PAGE_SHIFT;
2398 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2399 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2400 new_offset <<= PAGE_SHIFT;
2401 if (new_offset > offset) {
2402 if (new_offset < inode->i_size)
2403 offset = new_offset;
2404 else if (whence == SEEK_DATA)
2407 offset = inode->i_size;
2412 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2413 inode_unlock(inode);
2418 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2419 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2421 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2422 #define LAST_SCAN 4 /* about 150ms max */
2424 static void shmem_tag_pins(struct address_space *mapping)
2426 struct radix_tree_iter iter;
2435 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2436 page = radix_tree_deref_slot(slot);
2437 if (!page || radix_tree_exception(page)) {
2438 if (radix_tree_deref_retry(page)) {
2439 slot = radix_tree_iter_retry(&iter);
2442 } else if (page_count(page) - page_mapcount(page) > 1) {
2443 spin_lock_irq(&mapping->tree_lock);
2444 radix_tree_tag_set(&mapping->page_tree, iter.index,
2446 spin_unlock_irq(&mapping->tree_lock);
2449 if (need_resched()) {
2451 slot = radix_tree_iter_next(&iter);
2458 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2459 * via get_user_pages(), drivers might have some pending I/O without any active
2460 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2461 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2462 * them to be dropped.
2463 * The caller must guarantee that no new user will acquire writable references
2464 * to those pages to avoid races.
2466 static int shmem_wait_for_pins(struct address_space *mapping)
2468 struct radix_tree_iter iter;
2474 shmem_tag_pins(mapping);
2477 for (scan = 0; scan <= LAST_SCAN; scan++) {
2478 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2482 lru_add_drain_all();
2483 else if (schedule_timeout_killable((HZ << scan) / 200))
2488 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2489 start, SHMEM_TAG_PINNED) {
2491 page = radix_tree_deref_slot(slot);
2492 if (radix_tree_exception(page)) {
2493 if (radix_tree_deref_retry(page)) {
2494 slot = radix_tree_iter_retry(&iter);
2502 page_count(page) - page_mapcount(page) != 1) {
2503 if (scan < LAST_SCAN)
2504 goto continue_resched;
2507 * On the last scan, we clean up all those tags
2508 * we inserted; but make a note that we still
2509 * found pages pinned.
2514 spin_lock_irq(&mapping->tree_lock);
2515 radix_tree_tag_clear(&mapping->page_tree,
2516 iter.index, SHMEM_TAG_PINNED);
2517 spin_unlock_irq(&mapping->tree_lock);
2519 if (need_resched()) {
2521 slot = radix_tree_iter_next(&iter);
2530 #define F_ALL_SEALS (F_SEAL_SEAL | \
2535 int shmem_add_seals(struct file *file, unsigned int seals)
2537 struct inode *inode = file_inode(file);
2538 struct shmem_inode_info *info = SHMEM_I(inode);
2543 * Sealing allows multiple parties to share a shmem-file but restrict
2544 * access to a specific subset of file operations. Seals can only be
2545 * added, but never removed. This way, mutually untrusted parties can
2546 * share common memory regions with a well-defined policy. A malicious
2547 * peer can thus never perform unwanted operations on a shared object.
2549 * Seals are only supported on special shmem-files and always affect
2550 * the whole underlying inode. Once a seal is set, it may prevent some
2551 * kinds of access to the file. Currently, the following seals are
2553 * SEAL_SEAL: Prevent further seals from being set on this file
2554 * SEAL_SHRINK: Prevent the file from shrinking
2555 * SEAL_GROW: Prevent the file from growing
2556 * SEAL_WRITE: Prevent write access to the file
2558 * As we don't require any trust relationship between two parties, we
2559 * must prevent seals from being removed. Therefore, sealing a file
2560 * only adds a given set of seals to the file, it never touches
2561 * existing seals. Furthermore, the "setting seals"-operation can be
2562 * sealed itself, which basically prevents any further seal from being
2565 * Semantics of sealing are only defined on volatile files. Only
2566 * anonymous shmem files support sealing. More importantly, seals are
2567 * never written to disk. Therefore, there's no plan to support it on
2571 if (file->f_op != &shmem_file_operations)
2573 if (!(file->f_mode & FMODE_WRITE))
2575 if (seals & ~(unsigned int)F_ALL_SEALS)
2580 if (info->seals & F_SEAL_SEAL) {
2585 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2586 error = mapping_deny_writable(file->f_mapping);
2590 error = shmem_wait_for_pins(file->f_mapping);
2592 mapping_allow_writable(file->f_mapping);
2597 info->seals |= seals;
2601 inode_unlock(inode);
2604 EXPORT_SYMBOL_GPL(shmem_add_seals);
2606 int shmem_get_seals(struct file *file)
2608 if (file->f_op != &shmem_file_operations)
2611 return SHMEM_I(file_inode(file))->seals;
2613 EXPORT_SYMBOL_GPL(shmem_get_seals);
2615 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2621 /* disallow upper 32bit */
2625 error = shmem_add_seals(file, arg);
2628 error = shmem_get_seals(file);
2638 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2641 struct inode *inode = file_inode(file);
2642 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2643 struct shmem_inode_info *info = SHMEM_I(inode);
2644 struct shmem_falloc shmem_falloc;
2645 pgoff_t start, index, end;
2648 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2653 if (mode & FALLOC_FL_PUNCH_HOLE) {
2654 struct address_space *mapping = file->f_mapping;
2655 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2656 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2657 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2659 /* protected by i_mutex */
2660 if (info->seals & F_SEAL_WRITE) {
2665 shmem_falloc.waitq = &shmem_falloc_waitq;
2666 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2667 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2668 spin_lock(&inode->i_lock);
2669 inode->i_private = &shmem_falloc;
2670 spin_unlock(&inode->i_lock);
2672 if ((u64)unmap_end > (u64)unmap_start)
2673 unmap_mapping_range(mapping, unmap_start,
2674 1 + unmap_end - unmap_start, 0);
2675 shmem_truncate_range(inode, offset, offset + len - 1);
2676 /* No need to unmap again: hole-punching leaves COWed pages */
2678 spin_lock(&inode->i_lock);
2679 inode->i_private = NULL;
2680 wake_up_all(&shmem_falloc_waitq);
2681 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
2682 spin_unlock(&inode->i_lock);
2687 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2688 error = inode_newsize_ok(inode, offset + len);
2692 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2697 start = offset >> PAGE_SHIFT;
2698 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2699 /* Try to avoid a swapstorm if len is impossible to satisfy */
2700 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2705 shmem_falloc.waitq = NULL;
2706 shmem_falloc.start = start;
2707 shmem_falloc.next = start;
2708 shmem_falloc.nr_falloced = 0;
2709 shmem_falloc.nr_unswapped = 0;
2710 spin_lock(&inode->i_lock);
2711 inode->i_private = &shmem_falloc;
2712 spin_unlock(&inode->i_lock);
2714 for (index = start; index < end; index++) {
2718 * Good, the fallocate(2) manpage permits EINTR: we may have
2719 * been interrupted because we are using up too much memory.
2721 if (signal_pending(current))
2723 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2726 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2728 /* Remove the !PageUptodate pages we added */
2729 if (index > start) {
2730 shmem_undo_range(inode,
2731 (loff_t)start << PAGE_SHIFT,
2732 ((loff_t)index << PAGE_SHIFT) - 1, true);
2738 * Inform shmem_writepage() how far we have reached.
2739 * No need for lock or barrier: we have the page lock.
2741 shmem_falloc.next++;
2742 if (!PageUptodate(page))
2743 shmem_falloc.nr_falloced++;
2746 * If !PageUptodate, leave it that way so that freeable pages
2747 * can be recognized if we need to rollback on error later.
2748 * But set_page_dirty so that memory pressure will swap rather
2749 * than free the pages we are allocating (and SGP_CACHE pages
2750 * might still be clean: we now need to mark those dirty too).
2752 set_page_dirty(page);
2758 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2759 i_size_write(inode, offset + len);
2760 inode->i_ctime = current_time(inode);
2762 spin_lock(&inode->i_lock);
2763 inode->i_private = NULL;
2764 spin_unlock(&inode->i_lock);
2766 inode_unlock(inode);
2770 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2772 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2774 buf->f_type = TMPFS_MAGIC;
2775 buf->f_bsize = PAGE_SIZE;
2776 buf->f_namelen = NAME_MAX;
2777 if (sbinfo->max_blocks) {
2778 buf->f_blocks = sbinfo->max_blocks;
2780 buf->f_bfree = sbinfo->max_blocks -
2781 percpu_counter_sum(&sbinfo->used_blocks);
2783 if (sbinfo->max_inodes) {
2784 buf->f_files = sbinfo->max_inodes;
2785 buf->f_ffree = sbinfo->free_inodes;
2787 /* else leave those fields 0 like simple_statfs */
2792 * File creation. Allocate an inode, and we're done..
2795 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2797 struct inode *inode;
2798 int error = -ENOSPC;
2800 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2802 error = simple_acl_create(dir, inode);
2805 error = security_inode_init_security(inode, dir,
2807 shmem_initxattrs, NULL);
2808 if (error && error != -EOPNOTSUPP)
2812 dir->i_size += BOGO_DIRENT_SIZE;
2813 dir->i_ctime = dir->i_mtime = current_time(dir);
2814 d_instantiate(dentry, inode);
2815 dget(dentry); /* Extra count - pin the dentry in core */
2824 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2826 struct inode *inode;
2827 int error = -ENOSPC;
2829 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2831 error = security_inode_init_security(inode, dir,
2833 shmem_initxattrs, NULL);
2834 if (error && error != -EOPNOTSUPP)
2836 error = simple_acl_create(dir, inode);
2839 d_tmpfile(dentry, inode);
2847 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2851 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2857 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2860 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2866 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2868 struct inode *inode = d_inode(old_dentry);
2872 * No ordinary (disk based) filesystem counts links as inodes;
2873 * but each new link needs a new dentry, pinning lowmem, and
2874 * tmpfs dentries cannot be pruned until they are unlinked.
2876 ret = shmem_reserve_inode(inode->i_sb);
2880 dir->i_size += BOGO_DIRENT_SIZE;
2881 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2883 ihold(inode); /* New dentry reference */
2884 dget(dentry); /* Extra pinning count for the created dentry */
2885 d_instantiate(dentry, inode);
2890 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2892 struct inode *inode = d_inode(dentry);
2894 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2895 shmem_free_inode(inode->i_sb);
2897 dir->i_size -= BOGO_DIRENT_SIZE;
2898 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2900 dput(dentry); /* Undo the count from "create" - this does all the work */
2904 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2906 if (!simple_empty(dentry))
2909 drop_nlink(d_inode(dentry));
2911 return shmem_unlink(dir, dentry);
2914 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2916 bool old_is_dir = d_is_dir(old_dentry);
2917 bool new_is_dir = d_is_dir(new_dentry);
2919 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2921 drop_nlink(old_dir);
2924 drop_nlink(new_dir);
2928 old_dir->i_ctime = old_dir->i_mtime =
2929 new_dir->i_ctime = new_dir->i_mtime =
2930 d_inode(old_dentry)->i_ctime =
2931 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2936 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2938 struct dentry *whiteout;
2941 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2945 error = shmem_mknod(old_dir, whiteout,
2946 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2952 * Cheat and hash the whiteout while the old dentry is still in
2953 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2955 * d_lookup() will consistently find one of them at this point,
2956 * not sure which one, but that isn't even important.
2963 * The VFS layer already does all the dentry stuff for rename,
2964 * we just have to decrement the usage count for the target if
2965 * it exists so that the VFS layer correctly free's it when it
2968 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
2970 struct inode *inode = d_inode(old_dentry);
2971 int they_are_dirs = S_ISDIR(inode->i_mode);
2973 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2976 if (flags & RENAME_EXCHANGE)
2977 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2979 if (!simple_empty(new_dentry))
2982 if (flags & RENAME_WHITEOUT) {
2985 error = shmem_whiteout(old_dir, old_dentry);
2990 if (d_really_is_positive(new_dentry)) {
2991 (void) shmem_unlink(new_dir, new_dentry);
2992 if (they_are_dirs) {
2993 drop_nlink(d_inode(new_dentry));
2994 drop_nlink(old_dir);
2996 } else if (they_are_dirs) {
2997 drop_nlink(old_dir);
3001 old_dir->i_size -= BOGO_DIRENT_SIZE;
3002 new_dir->i_size += BOGO_DIRENT_SIZE;
3003 old_dir->i_ctime = old_dir->i_mtime =
3004 new_dir->i_ctime = new_dir->i_mtime =
3005 inode->i_ctime = current_time(old_dir);
3009 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3013 struct inode *inode;
3015 struct shmem_inode_info *info;
3017 len = strlen(symname) + 1;
3018 if (len > PAGE_SIZE)
3019 return -ENAMETOOLONG;
3021 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3025 error = security_inode_init_security(inode, dir, &dentry->d_name,
3026 shmem_initxattrs, NULL);
3028 if (error != -EOPNOTSUPP) {
3035 info = SHMEM_I(inode);
3036 inode->i_size = len-1;
3037 if (len <= SHORT_SYMLINK_LEN) {
3038 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3039 if (!inode->i_link) {
3043 inode->i_op = &shmem_short_symlink_operations;
3045 inode_nohighmem(inode);
3046 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3051 inode->i_mapping->a_ops = &shmem_aops;
3052 inode->i_op = &shmem_symlink_inode_operations;
3053 memcpy(page_address(page), symname, len);
3054 SetPageUptodate(page);
3055 set_page_dirty(page);
3059 dir->i_size += BOGO_DIRENT_SIZE;
3060 dir->i_ctime = dir->i_mtime = current_time(dir);
3061 d_instantiate(dentry, inode);
3066 static void shmem_put_link(void *arg)
3068 mark_page_accessed(arg);
3072 static const char *shmem_get_link(struct dentry *dentry,
3073 struct inode *inode,
3074 struct delayed_call *done)
3076 struct page *page = NULL;
3079 page = find_get_page(inode->i_mapping, 0);
3081 return ERR_PTR(-ECHILD);
3082 if (!PageUptodate(page)) {
3084 return ERR_PTR(-ECHILD);
3087 error = shmem_getpage(inode, 0, &page, SGP_READ);
3089 return ERR_PTR(error);
3092 set_delayed_call(done, shmem_put_link, page);
3093 return page_address(page);
3096 #ifdef CONFIG_TMPFS_XATTR
3098 * Superblocks without xattr inode operations may get some security.* xattr
3099 * support from the LSM "for free". As soon as we have any other xattrs
3100 * like ACLs, we also need to implement the security.* handlers at
3101 * filesystem level, though.
3105 * Callback for security_inode_init_security() for acquiring xattrs.
3107 static int shmem_initxattrs(struct inode *inode,
3108 const struct xattr *xattr_array,
3111 struct shmem_inode_info *info = SHMEM_I(inode);
3112 const struct xattr *xattr;
3113 struct simple_xattr *new_xattr;
3116 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3117 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3121 len = strlen(xattr->name) + 1;
3122 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3124 if (!new_xattr->name) {
3129 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3130 XATTR_SECURITY_PREFIX_LEN);
3131 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3134 simple_xattr_list_add(&info->xattrs, new_xattr);
3140 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3141 struct dentry *unused, struct inode *inode,
3142 const char *name, void *buffer, size_t size)
3144 struct shmem_inode_info *info = SHMEM_I(inode);
3146 name = xattr_full_name(handler, name);
3147 return simple_xattr_get(&info->xattrs, name, buffer, size);
3150 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3151 struct dentry *unused, struct inode *inode,
3152 const char *name, const void *value,
3153 size_t size, int flags)
3155 struct shmem_inode_info *info = SHMEM_I(inode);
3157 name = xattr_full_name(handler, name);
3158 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3161 static const struct xattr_handler shmem_security_xattr_handler = {
3162 .prefix = XATTR_SECURITY_PREFIX,
3163 .get = shmem_xattr_handler_get,
3164 .set = shmem_xattr_handler_set,
3167 static const struct xattr_handler shmem_trusted_xattr_handler = {
3168 .prefix = XATTR_TRUSTED_PREFIX,
3169 .get = shmem_xattr_handler_get,
3170 .set = shmem_xattr_handler_set,
3173 static const struct xattr_handler *shmem_xattr_handlers[] = {
3174 #ifdef CONFIG_TMPFS_POSIX_ACL
3175 &posix_acl_access_xattr_handler,
3176 &posix_acl_default_xattr_handler,
3178 &shmem_security_xattr_handler,
3179 &shmem_trusted_xattr_handler,
3183 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3185 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3186 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3188 #endif /* CONFIG_TMPFS_XATTR */
3190 static const struct inode_operations shmem_short_symlink_operations = {
3191 .readlink = generic_readlink,
3192 .get_link = simple_get_link,
3193 #ifdef CONFIG_TMPFS_XATTR
3194 .listxattr = shmem_listxattr,
3198 static const struct inode_operations shmem_symlink_inode_operations = {
3199 .readlink = generic_readlink,
3200 .get_link = shmem_get_link,
3201 #ifdef CONFIG_TMPFS_XATTR
3202 .listxattr = shmem_listxattr,
3206 static struct dentry *shmem_get_parent(struct dentry *child)
3208 return ERR_PTR(-ESTALE);
3211 static int shmem_match(struct inode *ino, void *vfh)
3215 inum = (inum << 32) | fh[1];
3216 return ino->i_ino == inum && fh[0] == ino->i_generation;
3219 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3220 struct fid *fid, int fh_len, int fh_type)
3222 struct inode *inode;
3223 struct dentry *dentry = NULL;
3230 inum = (inum << 32) | fid->raw[1];
3232 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3233 shmem_match, fid->raw);
3235 dentry = d_find_alias(inode);
3242 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3243 struct inode *parent)
3247 return FILEID_INVALID;
3250 if (inode_unhashed(inode)) {
3251 /* Unfortunately insert_inode_hash is not idempotent,
3252 * so as we hash inodes here rather than at creation
3253 * time, we need a lock to ensure we only try
3256 static DEFINE_SPINLOCK(lock);
3258 if (inode_unhashed(inode))
3259 __insert_inode_hash(inode,
3260 inode->i_ino + inode->i_generation);
3264 fh[0] = inode->i_generation;
3265 fh[1] = inode->i_ino;
3266 fh[2] = ((__u64)inode->i_ino) >> 32;
3272 static const struct export_operations shmem_export_ops = {
3273 .get_parent = shmem_get_parent,
3274 .encode_fh = shmem_encode_fh,
3275 .fh_to_dentry = shmem_fh_to_dentry,
3278 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3281 char *this_char, *value, *rest;
3282 struct mempolicy *mpol = NULL;
3286 while (options != NULL) {
3287 this_char = options;
3290 * NUL-terminate this option: unfortunately,
3291 * mount options form a comma-separated list,
3292 * but mpol's nodelist may also contain commas.
3294 options = strchr(options, ',');
3295 if (options == NULL)
3298 if (!isdigit(*options)) {
3305 if ((value = strchr(this_char,'=')) != NULL) {
3308 pr_err("tmpfs: No value for mount option '%s'\n",
3313 if (!strcmp(this_char,"size")) {
3314 unsigned long long size;
3315 size = memparse(value,&rest);
3317 size <<= PAGE_SHIFT;
3318 size *= totalram_pages;
3324 sbinfo->max_blocks =
3325 DIV_ROUND_UP(size, PAGE_SIZE);
3326 } else if (!strcmp(this_char,"nr_blocks")) {
3327 sbinfo->max_blocks = memparse(value, &rest);
3330 } else if (!strcmp(this_char,"nr_inodes")) {
3331 sbinfo->max_inodes = memparse(value, &rest);
3334 } else if (!strcmp(this_char,"mode")) {
3337 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3340 } else if (!strcmp(this_char,"uid")) {
3343 uid = simple_strtoul(value, &rest, 0);
3346 sbinfo->uid = make_kuid(current_user_ns(), uid);
3347 if (!uid_valid(sbinfo->uid))
3349 } else if (!strcmp(this_char,"gid")) {
3352 gid = simple_strtoul(value, &rest, 0);
3355 sbinfo->gid = make_kgid(current_user_ns(), gid);
3356 if (!gid_valid(sbinfo->gid))
3358 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3359 } else if (!strcmp(this_char, "huge")) {
3361 huge = shmem_parse_huge(value);
3364 if (!has_transparent_hugepage() &&
3365 huge != SHMEM_HUGE_NEVER)
3367 sbinfo->huge = huge;
3370 } else if (!strcmp(this_char,"mpol")) {
3373 if (mpol_parse_str(value, &mpol))
3377 pr_err("tmpfs: Bad mount option %s\n", this_char);
3381 sbinfo->mpol = mpol;
3385 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3393 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3395 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3396 struct shmem_sb_info config = *sbinfo;
3397 unsigned long inodes;
3398 int error = -EINVAL;
3401 if (shmem_parse_options(data, &config, true))
3404 spin_lock(&sbinfo->stat_lock);
3405 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3406 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3408 if (config.max_inodes < inodes)
3411 * Those tests disallow limited->unlimited while any are in use;
3412 * but we must separately disallow unlimited->limited, because
3413 * in that case we have no record of how much is already in use.
3415 if (config.max_blocks && !sbinfo->max_blocks)
3417 if (config.max_inodes && !sbinfo->max_inodes)
3421 sbinfo->huge = config.huge;
3422 sbinfo->max_blocks = config.max_blocks;
3423 sbinfo->max_inodes = config.max_inodes;
3424 sbinfo->free_inodes = config.max_inodes - inodes;
3427 * Preserve previous mempolicy unless mpol remount option was specified.
3430 mpol_put(sbinfo->mpol);
3431 sbinfo->mpol = config.mpol; /* transfers initial ref */
3434 spin_unlock(&sbinfo->stat_lock);
3438 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3440 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3442 if (sbinfo->max_blocks != shmem_default_max_blocks())
3443 seq_printf(seq, ",size=%luk",
3444 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3445 if (sbinfo->max_inodes != shmem_default_max_inodes())
3446 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3447 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3448 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3449 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3450 seq_printf(seq, ",uid=%u",
3451 from_kuid_munged(&init_user_ns, sbinfo->uid));
3452 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3453 seq_printf(seq, ",gid=%u",
3454 from_kgid_munged(&init_user_ns, sbinfo->gid));
3455 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3456 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3458 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3460 shmem_show_mpol(seq, sbinfo->mpol);
3464 #define MFD_NAME_PREFIX "memfd:"
3465 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3466 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3468 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3470 SYSCALL_DEFINE2(memfd_create,
3471 const char __user *, uname,
3472 unsigned int, flags)
3474 struct shmem_inode_info *info;
3480 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3483 /* length includes terminating zero */
3484 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3487 if (len > MFD_NAME_MAX_LEN + 1)
3490 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3494 strcpy(name, MFD_NAME_PREFIX);
3495 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3500 /* terminating-zero may have changed after strnlen_user() returned */
3501 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3506 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3512 file = shmem_file_setup(name, 0, VM_NORESERVE);
3514 error = PTR_ERR(file);
3517 info = SHMEM_I(file_inode(file));
3518 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3519 file->f_flags |= O_RDWR | O_LARGEFILE;
3520 if (flags & MFD_ALLOW_SEALING)
3521 info->seals &= ~F_SEAL_SEAL;
3523 fd_install(fd, file);
3534 #endif /* CONFIG_TMPFS */
3536 static void shmem_put_super(struct super_block *sb)
3538 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3540 percpu_counter_destroy(&sbinfo->used_blocks);
3541 mpol_put(sbinfo->mpol);
3543 sb->s_fs_info = NULL;
3546 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3548 struct inode *inode;
3549 struct shmem_sb_info *sbinfo;
3552 /* Round up to L1_CACHE_BYTES to resist false sharing */
3553 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3554 L1_CACHE_BYTES), GFP_KERNEL);
3558 sbinfo->mode = S_IRWXUGO | S_ISVTX;
3559 sbinfo->uid = current_fsuid();
3560 sbinfo->gid = current_fsgid();
3561 sb->s_fs_info = sbinfo;
3565 * Per default we only allow half of the physical ram per
3566 * tmpfs instance, limiting inodes to one per page of lowmem;
3567 * but the internal instance is left unlimited.
3569 if (!(sb->s_flags & MS_KERNMOUNT)) {
3570 sbinfo->max_blocks = shmem_default_max_blocks();
3571 sbinfo->max_inodes = shmem_default_max_inodes();
3572 if (shmem_parse_options(data, sbinfo, false)) {
3577 sb->s_flags |= MS_NOUSER;
3579 sb->s_export_op = &shmem_export_ops;
3580 sb->s_flags |= MS_NOSEC;
3582 sb->s_flags |= MS_NOUSER;
3585 spin_lock_init(&sbinfo->stat_lock);
3586 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3588 sbinfo->free_inodes = sbinfo->max_inodes;
3589 spin_lock_init(&sbinfo->shrinklist_lock);
3590 INIT_LIST_HEAD(&sbinfo->shrinklist);
3592 sb->s_maxbytes = MAX_LFS_FILESIZE;
3593 sb->s_blocksize = PAGE_SIZE;
3594 sb->s_blocksize_bits = PAGE_SHIFT;
3595 sb->s_magic = TMPFS_MAGIC;
3596 sb->s_op = &shmem_ops;
3597 sb->s_time_gran = 1;
3598 #ifdef CONFIG_TMPFS_XATTR
3599 sb->s_xattr = shmem_xattr_handlers;
3601 #ifdef CONFIG_TMPFS_POSIX_ACL
3602 sb->s_flags |= MS_POSIXACL;
3605 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3608 inode->i_uid = sbinfo->uid;
3609 inode->i_gid = sbinfo->gid;
3610 sb->s_root = d_make_root(inode);
3616 shmem_put_super(sb);
3620 static struct kmem_cache *shmem_inode_cachep;
3622 static struct inode *shmem_alloc_inode(struct super_block *sb)
3624 struct shmem_inode_info *info;
3625 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3628 return &info->vfs_inode;
3631 static void shmem_destroy_callback(struct rcu_head *head)
3633 struct inode *inode = container_of(head, struct inode, i_rcu);
3634 if (S_ISLNK(inode->i_mode))
3635 kfree(inode->i_link);
3636 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3639 static void shmem_destroy_inode(struct inode *inode)
3641 if (S_ISREG(inode->i_mode))
3642 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3643 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3646 static void shmem_init_inode(void *foo)
3648 struct shmem_inode_info *info = foo;
3649 inode_init_once(&info->vfs_inode);
3652 static int shmem_init_inodecache(void)
3654 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3655 sizeof(struct shmem_inode_info),
3656 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3660 static void shmem_destroy_inodecache(void)
3662 kmem_cache_destroy(shmem_inode_cachep);
3665 static const struct address_space_operations shmem_aops = {
3666 .writepage = shmem_writepage,
3667 .set_page_dirty = __set_page_dirty_no_writeback,
3669 .write_begin = shmem_write_begin,
3670 .write_end = shmem_write_end,
3672 #ifdef CONFIG_MIGRATION
3673 .migratepage = migrate_page,
3675 .error_remove_page = generic_error_remove_page,
3678 static const struct file_operations shmem_file_operations = {
3680 .get_unmapped_area = shmem_get_unmapped_area,
3682 .llseek = shmem_file_llseek,
3683 .read_iter = shmem_file_read_iter,
3684 .write_iter = generic_file_write_iter,
3685 .fsync = noop_fsync,
3686 .splice_read = generic_file_splice_read,
3687 .splice_write = iter_file_splice_write,
3688 .fallocate = shmem_fallocate,
3692 static const struct inode_operations shmem_inode_operations = {
3693 .getattr = shmem_getattr,
3694 .setattr = shmem_setattr,
3695 #ifdef CONFIG_TMPFS_XATTR
3696 .listxattr = shmem_listxattr,
3697 .set_acl = simple_set_acl,
3701 static const struct inode_operations shmem_dir_inode_operations = {
3703 .create = shmem_create,
3704 .lookup = simple_lookup,
3706 .unlink = shmem_unlink,
3707 .symlink = shmem_symlink,
3708 .mkdir = shmem_mkdir,
3709 .rmdir = shmem_rmdir,
3710 .mknod = shmem_mknod,
3711 .rename = shmem_rename2,
3712 .tmpfile = shmem_tmpfile,
3714 #ifdef CONFIG_TMPFS_XATTR
3715 .listxattr = shmem_listxattr,
3717 #ifdef CONFIG_TMPFS_POSIX_ACL
3718 .setattr = shmem_setattr,
3719 .set_acl = simple_set_acl,
3723 static const struct inode_operations shmem_special_inode_operations = {
3724 #ifdef CONFIG_TMPFS_XATTR
3725 .listxattr = shmem_listxattr,
3727 #ifdef CONFIG_TMPFS_POSIX_ACL
3728 .setattr = shmem_setattr,
3729 .set_acl = simple_set_acl,
3733 static const struct super_operations shmem_ops = {
3734 .alloc_inode = shmem_alloc_inode,
3735 .destroy_inode = shmem_destroy_inode,
3737 .statfs = shmem_statfs,
3738 .remount_fs = shmem_remount_fs,
3739 .show_options = shmem_show_options,
3741 .evict_inode = shmem_evict_inode,
3742 .drop_inode = generic_delete_inode,
3743 .put_super = shmem_put_super,
3744 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3745 .nr_cached_objects = shmem_unused_huge_count,
3746 .free_cached_objects = shmem_unused_huge_scan,
3750 static const struct vm_operations_struct shmem_vm_ops = {
3751 .fault = shmem_fault,
3752 .map_pages = filemap_map_pages,
3754 .set_policy = shmem_set_policy,
3755 .get_policy = shmem_get_policy,
3759 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3760 int flags, const char *dev_name, void *data)
3762 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3765 static struct file_system_type shmem_fs_type = {
3766 .owner = THIS_MODULE,
3768 .mount = shmem_mount,
3769 .kill_sb = kill_litter_super,
3770 .fs_flags = FS_USERNS_MOUNT,
3773 int __init shmem_init(void)
3777 /* If rootfs called this, don't re-init */
3778 if (shmem_inode_cachep)
3781 error = shmem_init_inodecache();
3785 error = register_filesystem(&shmem_fs_type);
3787 pr_err("Could not register tmpfs\n");
3791 shm_mnt = kern_mount(&shmem_fs_type);
3792 if (IS_ERR(shm_mnt)) {
3793 error = PTR_ERR(shm_mnt);
3794 pr_err("Could not kern_mount tmpfs\n");
3798 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3799 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3800 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3802 shmem_huge = 0; /* just in case it was patched */
3807 unregister_filesystem(&shmem_fs_type);
3809 shmem_destroy_inodecache();
3811 shm_mnt = ERR_PTR(error);
3815 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3816 static ssize_t shmem_enabled_show(struct kobject *kobj,
3817 struct kobj_attribute *attr, char *buf)
3821 SHMEM_HUGE_WITHIN_SIZE,
3829 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3830 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3832 count += sprintf(buf + count, fmt,
3833 shmem_format_huge(values[i]));
3835 buf[count - 1] = '\n';
3839 static ssize_t shmem_enabled_store(struct kobject *kobj,
3840 struct kobj_attribute *attr, const char *buf, size_t count)
3845 if (count + 1 > sizeof(tmp))
3847 memcpy(tmp, buf, count);
3849 if (count && tmp[count - 1] == '\n')
3850 tmp[count - 1] = '\0';
3852 huge = shmem_parse_huge(tmp);
3853 if (huge == -EINVAL)
3855 if (!has_transparent_hugepage() &&
3856 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3860 if (shmem_huge < SHMEM_HUGE_DENY)
3861 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3865 struct kobj_attribute shmem_enabled_attr =
3866 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3867 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3869 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3870 bool shmem_huge_enabled(struct vm_area_struct *vma)
3872 struct inode *inode = file_inode(vma->vm_file);
3873 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3877 if (shmem_huge == SHMEM_HUGE_FORCE)
3879 if (shmem_huge == SHMEM_HUGE_DENY)
3881 switch (sbinfo->huge) {
3882 case SHMEM_HUGE_NEVER:
3884 case SHMEM_HUGE_ALWAYS:
3886 case SHMEM_HUGE_WITHIN_SIZE:
3887 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3888 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3889 if (i_size >= HPAGE_PMD_SIZE &&
3890 i_size >> PAGE_SHIFT >= off)
3892 case SHMEM_HUGE_ADVISE:
3893 /* TODO: implement fadvise() hints */
3894 return (vma->vm_flags & VM_HUGEPAGE);
3900 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3902 #else /* !CONFIG_SHMEM */
3905 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3907 * This is intended for small system where the benefits of the full
3908 * shmem code (swap-backed and resource-limited) are outweighed by
3909 * their complexity. On systems without swap this code should be
3910 * effectively equivalent, but much lighter weight.
3913 static struct file_system_type shmem_fs_type = {
3915 .mount = ramfs_mount,
3916 .kill_sb = kill_litter_super,
3917 .fs_flags = FS_USERNS_MOUNT,
3920 int __init shmem_init(void)
3922 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3924 shm_mnt = kern_mount(&shmem_fs_type);
3925 BUG_ON(IS_ERR(shm_mnt));
3930 int shmem_unuse(swp_entry_t swap, struct page *page)
3935 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3940 void shmem_unlock_mapping(struct address_space *mapping)
3945 unsigned long shmem_get_unmapped_area(struct file *file,
3946 unsigned long addr, unsigned long len,
3947 unsigned long pgoff, unsigned long flags)
3949 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3953 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3955 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
3957 EXPORT_SYMBOL_GPL(shmem_truncate_range);
3959 #define shmem_vm_ops generic_file_vm_ops
3960 #define shmem_file_operations ramfs_file_operations
3961 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3962 #define shmem_acct_size(flags, size) 0
3963 #define shmem_unacct_size(flags, size) do {} while (0)
3965 #endif /* CONFIG_SHMEM */
3969 static const struct dentry_operations anon_ops = {
3970 .d_dname = simple_dname
3973 static struct file *__shmem_file_setup(const char *name, loff_t size,
3974 unsigned long flags, unsigned int i_flags)
3977 struct inode *inode;
3979 struct super_block *sb;
3982 if (IS_ERR(shm_mnt))
3983 return ERR_CAST(shm_mnt);
3985 if (size < 0 || size > MAX_LFS_FILESIZE)
3986 return ERR_PTR(-EINVAL);
3988 if (shmem_acct_size(flags, size))
3989 return ERR_PTR(-ENOMEM);
3991 res = ERR_PTR(-ENOMEM);
3993 this.len = strlen(name);
3994 this.hash = 0; /* will go */
3995 sb = shm_mnt->mnt_sb;
3996 path.mnt = mntget(shm_mnt);
3997 path.dentry = d_alloc_pseudo(sb, &this);
4000 d_set_d_op(path.dentry, &anon_ops);
4002 res = ERR_PTR(-ENOSPC);
4003 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4007 inode->i_flags |= i_flags;
4008 d_instantiate(path.dentry, inode);
4009 inode->i_size = size;
4010 clear_nlink(inode); /* It is unlinked */
4011 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4015 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4016 &shmem_file_operations);
4023 shmem_unacct_size(flags, size);
4030 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4031 * kernel internal. There will be NO LSM permission checks against the
4032 * underlying inode. So users of this interface must do LSM checks at a
4033 * higher layer. The users are the big_key and shm implementations. LSM
4034 * checks are provided at the key or shm level rather than the inode.
4035 * @name: name for dentry (to be seen in /proc/<pid>/maps
4036 * @size: size to be set for the file
4037 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4039 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4041 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4045 * shmem_file_setup - get an unlinked file living in tmpfs
4046 * @name: name for dentry (to be seen in /proc/<pid>/maps
4047 * @size: size to be set for the file
4048 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4050 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4052 return __shmem_file_setup(name, size, flags, 0);
4054 EXPORT_SYMBOL_GPL(shmem_file_setup);
4057 * shmem_zero_setup - setup a shared anonymous mapping
4058 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4060 int shmem_zero_setup(struct vm_area_struct *vma)
4063 loff_t size = vma->vm_end - vma->vm_start;
4066 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4067 * between XFS directory reading and selinux: since this file is only
4068 * accessible to the user through its mapping, use S_PRIVATE flag to
4069 * bypass file security, in the same way as shmem_kernel_file_setup().
4071 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4073 return PTR_ERR(file);
4077 vma->vm_file = file;
4078 vma->vm_ops = &shmem_vm_ops;
4080 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4081 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4082 (vma->vm_end & HPAGE_PMD_MASK)) {
4083 khugepaged_enter(vma, vma->vm_flags);
4090 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4091 * @mapping: the page's address_space
4092 * @index: the page index
4093 * @gfp: the page allocator flags to use if allocating
4095 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4096 * with any new page allocations done using the specified allocation flags.
4097 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4098 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4099 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4101 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4102 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4104 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4105 pgoff_t index, gfp_t gfp)
4108 struct inode *inode = mapping->host;
4112 BUG_ON(mapping->a_ops != &shmem_aops);
4113 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4116 page = ERR_PTR(error);
4122 * The tiny !SHMEM case uses ramfs without swap
4124 return read_cache_page_gfp(mapping, index, gfp);
4127 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);