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/sched/signal.h>
33 #include <linux/export.h>
34 #include <linux/swap.h>
35 #include <linux/uio.h>
36 #include <linux/khugepaged.h>
37 #include <linux/hugetlb.h>
39 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
41 static struct vfsmount *shm_mnt;
45 * This virtual memory filesystem is heavily based on the ramfs. It
46 * extends ramfs by the ability to use swap and honor resource limits
47 * which makes it a completely usable filesystem.
50 #include <linux/xattr.h>
51 #include <linux/exportfs.h>
52 #include <linux/posix_acl.h>
53 #include <linux/posix_acl_xattr.h>
54 #include <linux/mman.h>
55 #include <linux/string.h>
56 #include <linux/slab.h>
57 #include <linux/backing-dev.h>
58 #include <linux/shmem_fs.h>
59 #include <linux/writeback.h>
60 #include <linux/blkdev.h>
61 #include <linux/pagevec.h>
62 #include <linux/percpu_counter.h>
63 #include <linux/falloc.h>
64 #include <linux/splice.h>
65 #include <linux/security.h>
66 #include <linux/swapops.h>
67 #include <linux/mempolicy.h>
68 #include <linux/namei.h>
69 #include <linux/ctype.h>
70 #include <linux/migrate.h>
71 #include <linux/highmem.h>
72 #include <linux/seq_file.h>
73 #include <linux/magic.h>
74 #include <linux/syscalls.h>
75 #include <linux/fcntl.h>
76 #include <uapi/linux/memfd.h>
77 #include <linux/userfaultfd_k.h>
78 #include <linux/rmap.h>
79 #include <linux/uuid.h>
81 #include <linux/uaccess.h>
82 #include <asm/pgtable.h>
86 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 /* Pretend that each entry is of this size in directory's i_size */
90 #define BOGO_DIRENT_SIZE 20
92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93 #define SHORT_SYMLINK_LEN 128
96 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_mutex making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
100 struct shmem_falloc {
101 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
102 pgoff_t start; /* start of range currently being fallocated */
103 pgoff_t next; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
109 static unsigned long shmem_default_max_blocks(void)
111 return totalram_pages / 2;
114 static unsigned long shmem_default_max_inodes(void)
116 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
120 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122 struct shmem_inode_info *info, pgoff_t index);
123 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
124 struct page **pagep, enum sgp_type sgp,
125 gfp_t gfp, struct vm_area_struct *vma,
126 struct vm_fault *vmf, int *fault_type);
128 int shmem_getpage(struct inode *inode, pgoff_t index,
129 struct page **pagep, enum sgp_type sgp)
131 return shmem_getpage_gfp(inode, index, pagep, sgp,
132 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
135 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
137 return sb->s_fs_info;
141 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
142 * for shared memory and for shared anonymous (/dev/zero) mappings
143 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
144 * consistent with the pre-accounting of private mappings ...
146 static inline int shmem_acct_size(unsigned long flags, loff_t size)
148 return (flags & VM_NORESERVE) ?
149 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
152 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
154 if (!(flags & VM_NORESERVE))
155 vm_unacct_memory(VM_ACCT(size));
158 static inline int shmem_reacct_size(unsigned long flags,
159 loff_t oldsize, loff_t newsize)
161 if (!(flags & VM_NORESERVE)) {
162 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
163 return security_vm_enough_memory_mm(current->mm,
164 VM_ACCT(newsize) - VM_ACCT(oldsize));
165 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
166 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
172 * ... whereas tmpfs objects are accounted incrementally as
173 * pages are allocated, in order to allow large sparse files.
174 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
175 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
177 static inline int shmem_acct_block(unsigned long flags, long pages)
179 if (!(flags & VM_NORESERVE))
182 return security_vm_enough_memory_mm(current->mm,
183 pages * VM_ACCT(PAGE_SIZE));
186 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
188 if (flags & VM_NORESERVE)
189 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
192 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
194 struct shmem_inode_info *info = SHMEM_I(inode);
195 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
197 if (shmem_acct_block(info->flags, pages))
200 if (sbinfo->max_blocks) {
201 if (percpu_counter_compare(&sbinfo->used_blocks,
202 sbinfo->max_blocks - pages) > 0)
204 percpu_counter_add(&sbinfo->used_blocks, pages);
210 shmem_unacct_blocks(info->flags, pages);
214 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
216 struct shmem_inode_info *info = SHMEM_I(inode);
217 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
219 if (sbinfo->max_blocks)
220 percpu_counter_sub(&sbinfo->used_blocks, pages);
221 shmem_unacct_blocks(info->flags, pages);
224 static const struct super_operations shmem_ops;
225 static const struct address_space_operations shmem_aops;
226 static const struct file_operations shmem_file_operations;
227 static const struct inode_operations shmem_inode_operations;
228 static const struct inode_operations shmem_dir_inode_operations;
229 static const struct inode_operations shmem_special_inode_operations;
230 static const struct vm_operations_struct shmem_vm_ops;
231 static struct file_system_type shmem_fs_type;
233 bool vma_is_shmem(struct vm_area_struct *vma)
235 return vma->vm_ops == &shmem_vm_ops;
238 static LIST_HEAD(shmem_swaplist);
239 static DEFINE_MUTEX(shmem_swaplist_mutex);
241 static int shmem_reserve_inode(struct super_block *sb)
243 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
244 if (sbinfo->max_inodes) {
245 spin_lock(&sbinfo->stat_lock);
246 if (!sbinfo->free_inodes) {
247 spin_unlock(&sbinfo->stat_lock);
250 sbinfo->free_inodes--;
251 spin_unlock(&sbinfo->stat_lock);
256 static void shmem_free_inode(struct super_block *sb)
258 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
259 if (sbinfo->max_inodes) {
260 spin_lock(&sbinfo->stat_lock);
261 sbinfo->free_inodes++;
262 spin_unlock(&sbinfo->stat_lock);
267 * shmem_recalc_inode - recalculate the block usage of an inode
268 * @inode: inode to recalc
270 * We have to calculate the free blocks since the mm can drop
271 * undirtied hole pages behind our back.
273 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
274 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
276 * It has to be called with the spinlock held.
278 static void shmem_recalc_inode(struct inode *inode)
280 struct shmem_inode_info *info = SHMEM_I(inode);
283 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
285 info->alloced -= freed;
286 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
287 shmem_inode_unacct_blocks(inode, freed);
291 bool shmem_charge(struct inode *inode, long pages)
293 struct shmem_inode_info *info = SHMEM_I(inode);
296 if (!shmem_inode_acct_block(inode, pages))
299 spin_lock_irqsave(&info->lock, flags);
300 info->alloced += pages;
301 inode->i_blocks += pages * BLOCKS_PER_PAGE;
302 shmem_recalc_inode(inode);
303 spin_unlock_irqrestore(&info->lock, flags);
304 inode->i_mapping->nrpages += pages;
309 void shmem_uncharge(struct inode *inode, long pages)
311 struct shmem_inode_info *info = SHMEM_I(inode);
314 spin_lock_irqsave(&info->lock, flags);
315 info->alloced -= pages;
316 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
317 shmem_recalc_inode(inode);
318 spin_unlock_irqrestore(&info->lock, flags);
320 shmem_inode_unacct_blocks(inode, pages);
324 * Replace item expected in radix tree by a new item, while holding tree lock.
326 static int shmem_radix_tree_replace(struct address_space *mapping,
327 pgoff_t index, void *expected, void *replacement)
329 struct radix_tree_node *node;
333 VM_BUG_ON(!expected);
334 VM_BUG_ON(!replacement);
335 item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot);
338 if (item != expected)
340 __radix_tree_replace(&mapping->page_tree, node, pslot,
346 * Sometimes, before we decide whether to proceed or to fail, we must check
347 * that an entry was not already brought back from swap by a racing thread.
349 * Checking page is not enough: by the time a SwapCache page is locked, it
350 * might be reused, and again be SwapCache, using the same swap as before.
352 static bool shmem_confirm_swap(struct address_space *mapping,
353 pgoff_t index, swp_entry_t swap)
358 item = radix_tree_lookup(&mapping->page_tree, index);
360 return item == swp_to_radix_entry(swap);
364 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
367 * disables huge pages for the mount;
369 * enables huge pages for the mount;
370 * SHMEM_HUGE_WITHIN_SIZE:
371 * only allocate huge pages if the page will be fully within i_size,
372 * also respect fadvise()/madvise() hints;
374 * only allocate huge pages if requested with fadvise()/madvise();
377 #define SHMEM_HUGE_NEVER 0
378 #define SHMEM_HUGE_ALWAYS 1
379 #define SHMEM_HUGE_WITHIN_SIZE 2
380 #define SHMEM_HUGE_ADVISE 3
384 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
387 * disables huge on shm_mnt and all mounts, for emergency use;
389 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
392 #define SHMEM_HUGE_DENY (-1)
393 #define SHMEM_HUGE_FORCE (-2)
395 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
396 /* ifdef here to avoid bloating shmem.o when not necessary */
398 int shmem_huge __read_mostly;
400 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
401 static int shmem_parse_huge(const char *str)
403 if (!strcmp(str, "never"))
404 return SHMEM_HUGE_NEVER;
405 if (!strcmp(str, "always"))
406 return SHMEM_HUGE_ALWAYS;
407 if (!strcmp(str, "within_size"))
408 return SHMEM_HUGE_WITHIN_SIZE;
409 if (!strcmp(str, "advise"))
410 return SHMEM_HUGE_ADVISE;
411 if (!strcmp(str, "deny"))
412 return SHMEM_HUGE_DENY;
413 if (!strcmp(str, "force"))
414 return SHMEM_HUGE_FORCE;
418 static const char *shmem_format_huge(int huge)
421 case SHMEM_HUGE_NEVER:
423 case SHMEM_HUGE_ALWAYS:
425 case SHMEM_HUGE_WITHIN_SIZE:
426 return "within_size";
427 case SHMEM_HUGE_ADVISE:
429 case SHMEM_HUGE_DENY:
431 case SHMEM_HUGE_FORCE:
440 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
441 struct shrink_control *sc, unsigned long nr_to_split)
443 LIST_HEAD(list), *pos, *next;
444 LIST_HEAD(to_remove);
446 struct shmem_inode_info *info;
448 unsigned long batch = sc ? sc->nr_to_scan : 128;
449 int removed = 0, split = 0;
451 if (list_empty(&sbinfo->shrinklist))
454 spin_lock(&sbinfo->shrinklist_lock);
455 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
456 info = list_entry(pos, struct shmem_inode_info, shrinklist);
459 inode = igrab(&info->vfs_inode);
461 /* inode is about to be evicted */
463 list_del_init(&info->shrinklist);
468 /* Check if there's anything to gain */
469 if (round_up(inode->i_size, PAGE_SIZE) ==
470 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
471 list_move(&info->shrinklist, &to_remove);
476 list_move(&info->shrinklist, &list);
481 spin_unlock(&sbinfo->shrinklist_lock);
483 list_for_each_safe(pos, next, &to_remove) {
484 info = list_entry(pos, struct shmem_inode_info, shrinklist);
485 inode = &info->vfs_inode;
486 list_del_init(&info->shrinklist);
490 list_for_each_safe(pos, next, &list) {
493 info = list_entry(pos, struct shmem_inode_info, shrinklist);
494 inode = &info->vfs_inode;
496 if (nr_to_split && split >= nr_to_split) {
501 page = find_lock_page(inode->i_mapping,
502 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
506 if (!PageTransHuge(page)) {
512 ret = split_huge_page(page);
517 /* split failed: leave it on the list */
524 list_del_init(&info->shrinklist);
529 spin_lock(&sbinfo->shrinklist_lock);
530 list_splice_tail(&list, &sbinfo->shrinklist);
531 sbinfo->shrinklist_len -= removed;
532 spin_unlock(&sbinfo->shrinklist_lock);
537 static long shmem_unused_huge_scan(struct super_block *sb,
538 struct shrink_control *sc)
540 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
542 if (!READ_ONCE(sbinfo->shrinklist_len))
545 return shmem_unused_huge_shrink(sbinfo, sc, 0);
548 static long shmem_unused_huge_count(struct super_block *sb,
549 struct shrink_control *sc)
551 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
552 return READ_ONCE(sbinfo->shrinklist_len);
554 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
556 #define shmem_huge SHMEM_HUGE_DENY
558 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
559 struct shrink_control *sc, unsigned long nr_to_split)
563 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
566 * Like add_to_page_cache_locked, but error if expected item has gone.
568 static int shmem_add_to_page_cache(struct page *page,
569 struct address_space *mapping,
570 pgoff_t index, void *expected)
572 int error, nr = hpage_nr_pages(page);
574 VM_BUG_ON_PAGE(PageTail(page), page);
575 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
576 VM_BUG_ON_PAGE(!PageLocked(page), page);
577 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
578 VM_BUG_ON(expected && PageTransHuge(page));
580 page_ref_add(page, nr);
581 page->mapping = mapping;
584 spin_lock_irq(&mapping->tree_lock);
585 if (PageTransHuge(page)) {
586 void __rcu **results;
591 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
592 &results, &idx, index, 1) &&
593 idx < index + HPAGE_PMD_NR) {
598 for (i = 0; i < HPAGE_PMD_NR; i++) {
599 error = radix_tree_insert(&mapping->page_tree,
600 index + i, page + i);
603 count_vm_event(THP_FILE_ALLOC);
605 } else if (!expected) {
606 error = radix_tree_insert(&mapping->page_tree, index, page);
608 error = shmem_radix_tree_replace(mapping, index, expected,
613 mapping->nrpages += nr;
614 if (PageTransHuge(page))
615 __inc_node_page_state(page, NR_SHMEM_THPS);
616 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
617 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
618 spin_unlock_irq(&mapping->tree_lock);
620 page->mapping = NULL;
621 spin_unlock_irq(&mapping->tree_lock);
622 page_ref_sub(page, nr);
628 * Like delete_from_page_cache, but substitutes swap for page.
630 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
632 struct address_space *mapping = page->mapping;
635 VM_BUG_ON_PAGE(PageCompound(page), page);
637 spin_lock_irq(&mapping->tree_lock);
638 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
639 page->mapping = NULL;
641 __dec_node_page_state(page, NR_FILE_PAGES);
642 __dec_node_page_state(page, NR_SHMEM);
643 spin_unlock_irq(&mapping->tree_lock);
649 * Remove swap entry from radix tree, free the swap and its page cache.
651 static int shmem_free_swap(struct address_space *mapping,
652 pgoff_t index, void *radswap)
656 spin_lock_irq(&mapping->tree_lock);
657 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
658 spin_unlock_irq(&mapping->tree_lock);
661 free_swap_and_cache(radix_to_swp_entry(radswap));
666 * Determine (in bytes) how many of the shmem object's pages mapped by the
667 * given offsets are swapped out.
669 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
670 * as long as the inode doesn't go away and racy results are not a problem.
672 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
673 pgoff_t start, pgoff_t end)
675 struct radix_tree_iter iter;
678 unsigned long swapped = 0;
682 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
683 if (iter.index >= end)
686 page = radix_tree_deref_slot(slot);
688 if (radix_tree_deref_retry(page)) {
689 slot = radix_tree_iter_retry(&iter);
693 if (radix_tree_exceptional_entry(page))
696 if (need_resched()) {
697 slot = radix_tree_iter_resume(slot, &iter);
704 return swapped << PAGE_SHIFT;
708 * Determine (in bytes) how many of the shmem object's pages mapped by the
709 * given vma is swapped out.
711 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
712 * as long as the inode doesn't go away and racy results are not a problem.
714 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
716 struct inode *inode = file_inode(vma->vm_file);
717 struct shmem_inode_info *info = SHMEM_I(inode);
718 struct address_space *mapping = inode->i_mapping;
719 unsigned long swapped;
721 /* Be careful as we don't hold info->lock */
722 swapped = READ_ONCE(info->swapped);
725 * The easier cases are when the shmem object has nothing in swap, or
726 * the vma maps it whole. Then we can simply use the stats that we
732 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
733 return swapped << PAGE_SHIFT;
735 /* Here comes the more involved part */
736 return shmem_partial_swap_usage(mapping,
737 linear_page_index(vma, vma->vm_start),
738 linear_page_index(vma, vma->vm_end));
742 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
744 void shmem_unlock_mapping(struct address_space *mapping)
747 pgoff_t indices[PAGEVEC_SIZE];
752 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
754 while (!mapping_unevictable(mapping)) {
756 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
757 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
759 pvec.nr = find_get_entries(mapping, index,
760 PAGEVEC_SIZE, pvec.pages, indices);
763 index = indices[pvec.nr - 1] + 1;
764 pagevec_remove_exceptionals(&pvec);
765 check_move_unevictable_pages(pvec.pages, pvec.nr);
766 pagevec_release(&pvec);
772 * Remove range of pages and swap entries from radix tree, and free them.
773 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
775 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
778 struct address_space *mapping = inode->i_mapping;
779 struct shmem_inode_info *info = SHMEM_I(inode);
780 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
781 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
782 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
783 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
785 pgoff_t indices[PAGEVEC_SIZE];
786 long nr_swaps_freed = 0;
791 end = -1; /* unsigned, so actually very big */
795 while (index < end) {
796 pvec.nr = find_get_entries(mapping, index,
797 min(end - index, (pgoff_t)PAGEVEC_SIZE),
798 pvec.pages, indices);
801 for (i = 0; i < pagevec_count(&pvec); i++) {
802 struct page *page = pvec.pages[i];
808 if (radix_tree_exceptional_entry(page)) {
811 nr_swaps_freed += !shmem_free_swap(mapping,
816 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
818 if (!trylock_page(page))
821 if (PageTransTail(page)) {
822 /* Middle of THP: zero out the page */
823 clear_highpage(page);
826 } else if (PageTransHuge(page)) {
827 if (index == round_down(end, HPAGE_PMD_NR)) {
829 * Range ends in the middle of THP:
832 clear_highpage(page);
836 index += HPAGE_PMD_NR - 1;
837 i += HPAGE_PMD_NR - 1;
840 if (!unfalloc || !PageUptodate(page)) {
841 VM_BUG_ON_PAGE(PageTail(page), page);
842 if (page_mapping(page) == mapping) {
843 VM_BUG_ON_PAGE(PageWriteback(page), page);
844 truncate_inode_page(mapping, page);
849 pagevec_remove_exceptionals(&pvec);
850 pagevec_release(&pvec);
856 struct page *page = NULL;
857 shmem_getpage(inode, start - 1, &page, SGP_READ);
859 unsigned int top = PAGE_SIZE;
864 zero_user_segment(page, partial_start, top);
865 set_page_dirty(page);
871 struct page *page = NULL;
872 shmem_getpage(inode, end, &page, SGP_READ);
874 zero_user_segment(page, 0, partial_end);
875 set_page_dirty(page);
884 while (index < end) {
887 pvec.nr = find_get_entries(mapping, index,
888 min(end - index, (pgoff_t)PAGEVEC_SIZE),
889 pvec.pages, indices);
891 /* If all gone or hole-punch or unfalloc, we're done */
892 if (index == start || end != -1)
894 /* But if truncating, restart to make sure all gone */
898 for (i = 0; i < pagevec_count(&pvec); i++) {
899 struct page *page = pvec.pages[i];
905 if (radix_tree_exceptional_entry(page)) {
908 if (shmem_free_swap(mapping, index, page)) {
909 /* Swap was replaced by page: retry */
919 if (PageTransTail(page)) {
920 /* Middle of THP: zero out the page */
921 clear_highpage(page);
924 * Partial thp truncate due 'start' in middle
925 * of THP: don't need to look on these pages
926 * again on !pvec.nr restart.
928 if (index != round_down(end, HPAGE_PMD_NR))
931 } else if (PageTransHuge(page)) {
932 if (index == round_down(end, HPAGE_PMD_NR)) {
934 * Range ends in the middle of THP:
937 clear_highpage(page);
941 index += HPAGE_PMD_NR - 1;
942 i += HPAGE_PMD_NR - 1;
945 if (!unfalloc || !PageUptodate(page)) {
946 VM_BUG_ON_PAGE(PageTail(page), page);
947 if (page_mapping(page) == mapping) {
948 VM_BUG_ON_PAGE(PageWriteback(page), page);
949 truncate_inode_page(mapping, page);
951 /* Page was replaced by swap: retry */
959 pagevec_remove_exceptionals(&pvec);
960 pagevec_release(&pvec);
964 spin_lock_irq(&info->lock);
965 info->swapped -= nr_swaps_freed;
966 shmem_recalc_inode(inode);
967 spin_unlock_irq(&info->lock);
970 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
972 shmem_undo_range(inode, lstart, lend, false);
973 inode->i_ctime = inode->i_mtime = current_time(inode);
975 EXPORT_SYMBOL_GPL(shmem_truncate_range);
977 static int shmem_getattr(const struct path *path, struct kstat *stat,
978 u32 request_mask, unsigned int query_flags)
980 struct inode *inode = path->dentry->d_inode;
981 struct shmem_inode_info *info = SHMEM_I(inode);
983 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
984 spin_lock_irq(&info->lock);
985 shmem_recalc_inode(inode);
986 spin_unlock_irq(&info->lock);
988 generic_fillattr(inode, stat);
992 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
994 struct inode *inode = d_inode(dentry);
995 struct shmem_inode_info *info = SHMEM_I(inode);
996 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
999 error = setattr_prepare(dentry, attr);
1003 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1004 loff_t oldsize = inode->i_size;
1005 loff_t newsize = attr->ia_size;
1007 /* protected by i_mutex */
1008 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1009 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1012 if (newsize != oldsize) {
1013 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1017 i_size_write(inode, newsize);
1018 inode->i_ctime = inode->i_mtime = current_time(inode);
1020 if (newsize <= oldsize) {
1021 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1022 if (oldsize > holebegin)
1023 unmap_mapping_range(inode->i_mapping,
1026 shmem_truncate_range(inode,
1027 newsize, (loff_t)-1);
1028 /* unmap again to remove racily COWed private pages */
1029 if (oldsize > holebegin)
1030 unmap_mapping_range(inode->i_mapping,
1034 * Part of the huge page can be beyond i_size: subject
1035 * to shrink under memory pressure.
1037 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1038 spin_lock(&sbinfo->shrinklist_lock);
1040 * _careful to defend against unlocked access to
1041 * ->shrink_list in shmem_unused_huge_shrink()
1043 if (list_empty_careful(&info->shrinklist)) {
1044 list_add_tail(&info->shrinklist,
1045 &sbinfo->shrinklist);
1046 sbinfo->shrinklist_len++;
1048 spin_unlock(&sbinfo->shrinklist_lock);
1053 setattr_copy(inode, attr);
1054 if (attr->ia_valid & ATTR_MODE)
1055 error = posix_acl_chmod(inode, inode->i_mode);
1059 static void shmem_evict_inode(struct inode *inode)
1061 struct shmem_inode_info *info = SHMEM_I(inode);
1062 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1064 if (inode->i_mapping->a_ops == &shmem_aops) {
1065 shmem_unacct_size(info->flags, inode->i_size);
1067 shmem_truncate_range(inode, 0, (loff_t)-1);
1068 if (!list_empty(&info->shrinklist)) {
1069 spin_lock(&sbinfo->shrinklist_lock);
1070 if (!list_empty(&info->shrinklist)) {
1071 list_del_init(&info->shrinklist);
1072 sbinfo->shrinklist_len--;
1074 spin_unlock(&sbinfo->shrinklist_lock);
1076 if (!list_empty(&info->swaplist)) {
1077 mutex_lock(&shmem_swaplist_mutex);
1078 list_del_init(&info->swaplist);
1079 mutex_unlock(&shmem_swaplist_mutex);
1083 simple_xattrs_free(&info->xattrs);
1084 WARN_ON(inode->i_blocks);
1085 shmem_free_inode(inode->i_sb);
1089 static unsigned long find_swap_entry(struct radix_tree_root *root, void *item)
1091 struct radix_tree_iter iter;
1093 unsigned long found = -1;
1094 unsigned int checked = 0;
1097 radix_tree_for_each_slot(slot, root, &iter, 0) {
1098 if (*slot == item) {
1103 if ((checked % 4096) != 0)
1105 slot = radix_tree_iter_resume(slot, &iter);
1114 * If swap found in inode, free it and move page from swapcache to filecache.
1116 static int shmem_unuse_inode(struct shmem_inode_info *info,
1117 swp_entry_t swap, struct page **pagep)
1119 struct address_space *mapping = info->vfs_inode.i_mapping;
1125 radswap = swp_to_radix_entry(swap);
1126 index = find_swap_entry(&mapping->page_tree, radswap);
1128 return -EAGAIN; /* tell shmem_unuse we found nothing */
1131 * Move _head_ to start search for next from here.
1132 * But be careful: shmem_evict_inode checks list_empty without taking
1133 * mutex, and there's an instant in list_move_tail when info->swaplist
1134 * would appear empty, if it were the only one on shmem_swaplist.
1136 if (shmem_swaplist.next != &info->swaplist)
1137 list_move_tail(&shmem_swaplist, &info->swaplist);
1139 gfp = mapping_gfp_mask(mapping);
1140 if (shmem_should_replace_page(*pagep, gfp)) {
1141 mutex_unlock(&shmem_swaplist_mutex);
1142 error = shmem_replace_page(pagep, gfp, info, index);
1143 mutex_lock(&shmem_swaplist_mutex);
1145 * We needed to drop mutex to make that restrictive page
1146 * allocation, but the inode might have been freed while we
1147 * dropped it: although a racing shmem_evict_inode() cannot
1148 * complete without emptying the radix_tree, our page lock
1149 * on this swapcache page is not enough to prevent that -
1150 * free_swap_and_cache() of our swap entry will only
1151 * trylock_page(), removing swap from radix_tree whatever.
1153 * We must not proceed to shmem_add_to_page_cache() if the
1154 * inode has been freed, but of course we cannot rely on
1155 * inode or mapping or info to check that. However, we can
1156 * safely check if our swap entry is still in use (and here
1157 * it can't have got reused for another page): if it's still
1158 * in use, then the inode cannot have been freed yet, and we
1159 * can safely proceed (if it's no longer in use, that tells
1160 * nothing about the inode, but we don't need to unuse swap).
1162 if (!page_swapcount(*pagep))
1167 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1168 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1169 * beneath us (pagelock doesn't help until the page is in pagecache).
1172 error = shmem_add_to_page_cache(*pagep, mapping, index,
1174 if (error != -ENOMEM) {
1176 * Truncation and eviction use free_swap_and_cache(), which
1177 * only does trylock page: if we raced, best clean up here.
1179 delete_from_swap_cache(*pagep);
1180 set_page_dirty(*pagep);
1182 spin_lock_irq(&info->lock);
1184 spin_unlock_irq(&info->lock);
1192 * Search through swapped inodes to find and replace swap by page.
1194 int shmem_unuse(swp_entry_t swap, struct page *page)
1196 struct list_head *this, *next;
1197 struct shmem_inode_info *info;
1198 struct mem_cgroup *memcg;
1202 * There's a faint possibility that swap page was replaced before
1203 * caller locked it: caller will come back later with the right page.
1205 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1209 * Charge page using GFP_KERNEL while we can wait, before taking
1210 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1211 * Charged back to the user (not to caller) when swap account is used.
1213 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1217 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1220 mutex_lock(&shmem_swaplist_mutex);
1221 list_for_each_safe(this, next, &shmem_swaplist) {
1222 info = list_entry(this, struct shmem_inode_info, swaplist);
1224 error = shmem_unuse_inode(info, swap, &page);
1226 list_del_init(&info->swaplist);
1228 if (error != -EAGAIN)
1230 /* found nothing in this: move on to search the next */
1232 mutex_unlock(&shmem_swaplist_mutex);
1235 if (error != -ENOMEM)
1237 mem_cgroup_cancel_charge(page, memcg, false);
1239 mem_cgroup_commit_charge(page, memcg, true, false);
1247 * Move the page from the page cache to the swap cache.
1249 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1251 struct shmem_inode_info *info;
1252 struct address_space *mapping;
1253 struct inode *inode;
1257 VM_BUG_ON_PAGE(PageCompound(page), page);
1258 BUG_ON(!PageLocked(page));
1259 mapping = page->mapping;
1260 index = page->index;
1261 inode = mapping->host;
1262 info = SHMEM_I(inode);
1263 if (info->flags & VM_LOCKED)
1265 if (!total_swap_pages)
1269 * Our capabilities prevent regular writeback or sync from ever calling
1270 * shmem_writepage; but a stacking filesystem might use ->writepage of
1271 * its underlying filesystem, in which case tmpfs should write out to
1272 * swap only in response to memory pressure, and not for the writeback
1275 if (!wbc->for_reclaim) {
1276 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1281 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1282 * value into swapfile.c, the only way we can correctly account for a
1283 * fallocated page arriving here is now to initialize it and write it.
1285 * That's okay for a page already fallocated earlier, but if we have
1286 * not yet completed the fallocation, then (a) we want to keep track
1287 * of this page in case we have to undo it, and (b) it may not be a
1288 * good idea to continue anyway, once we're pushing into swap. So
1289 * reactivate the page, and let shmem_fallocate() quit when too many.
1291 if (!PageUptodate(page)) {
1292 if (inode->i_private) {
1293 struct shmem_falloc *shmem_falloc;
1294 spin_lock(&inode->i_lock);
1295 shmem_falloc = inode->i_private;
1297 !shmem_falloc->waitq &&
1298 index >= shmem_falloc->start &&
1299 index < shmem_falloc->next)
1300 shmem_falloc->nr_unswapped++;
1302 shmem_falloc = NULL;
1303 spin_unlock(&inode->i_lock);
1307 clear_highpage(page);
1308 flush_dcache_page(page);
1309 SetPageUptodate(page);
1312 swap = get_swap_page(page);
1316 if (mem_cgroup_try_charge_swap(page, swap))
1320 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1321 * if it's not already there. Do it now before the page is
1322 * moved to swap cache, when its pagelock no longer protects
1323 * the inode from eviction. But don't unlock the mutex until
1324 * we've incremented swapped, because shmem_unuse_inode() will
1325 * prune a !swapped inode from the swaplist under this mutex.
1327 mutex_lock(&shmem_swaplist_mutex);
1328 if (list_empty(&info->swaplist))
1329 list_add_tail(&info->swaplist, &shmem_swaplist);
1331 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1332 spin_lock_irq(&info->lock);
1333 shmem_recalc_inode(inode);
1335 spin_unlock_irq(&info->lock);
1337 swap_shmem_alloc(swap);
1338 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1340 mutex_unlock(&shmem_swaplist_mutex);
1341 BUG_ON(page_mapped(page));
1342 swap_writepage(page, wbc);
1346 mutex_unlock(&shmem_swaplist_mutex);
1348 put_swap_page(page, swap);
1350 set_page_dirty(page);
1351 if (wbc->for_reclaim)
1352 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1357 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1358 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1362 if (!mpol || mpol->mode == MPOL_DEFAULT)
1363 return; /* show nothing */
1365 mpol_to_str(buffer, sizeof(buffer), mpol);
1367 seq_printf(seq, ",mpol=%s", buffer);
1370 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1372 struct mempolicy *mpol = NULL;
1374 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1375 mpol = sbinfo->mpol;
1377 spin_unlock(&sbinfo->stat_lock);
1381 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1382 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1385 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1389 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1391 #define vm_policy vm_private_data
1394 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1395 struct shmem_inode_info *info, pgoff_t index)
1397 /* Create a pseudo vma that just contains the policy */
1399 /* Bias interleave by inode number to distribute better across nodes */
1400 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1402 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1405 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1407 /* Drop reference taken by mpol_shared_policy_lookup() */
1408 mpol_cond_put(vma->vm_policy);
1411 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1412 struct shmem_inode_info *info, pgoff_t index)
1414 struct vm_area_struct pvma;
1417 shmem_pseudo_vma_init(&pvma, info, index);
1418 page = swapin_readahead(swap, gfp, &pvma, 0);
1419 shmem_pseudo_vma_destroy(&pvma);
1424 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1425 struct shmem_inode_info *info, pgoff_t index)
1427 struct vm_area_struct pvma;
1428 struct inode *inode = &info->vfs_inode;
1429 struct address_space *mapping = inode->i_mapping;
1430 pgoff_t idx, hindex;
1431 void __rcu **results;
1434 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1437 hindex = round_down(index, HPAGE_PMD_NR);
1439 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1440 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1446 shmem_pseudo_vma_init(&pvma, info, hindex);
1447 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1448 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1449 shmem_pseudo_vma_destroy(&pvma);
1451 prep_transhuge_page(page);
1455 static struct page *shmem_alloc_page(gfp_t gfp,
1456 struct shmem_inode_info *info, pgoff_t index)
1458 struct vm_area_struct pvma;
1461 shmem_pseudo_vma_init(&pvma, info, index);
1462 page = alloc_page_vma(gfp, &pvma, 0);
1463 shmem_pseudo_vma_destroy(&pvma);
1468 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1469 struct inode *inode,
1470 pgoff_t index, bool huge)
1472 struct shmem_inode_info *info = SHMEM_I(inode);
1477 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1479 nr = huge ? HPAGE_PMD_NR : 1;
1481 if (!shmem_inode_acct_block(inode, nr))
1485 page = shmem_alloc_hugepage(gfp, info, index);
1487 page = shmem_alloc_page(gfp, info, index);
1489 __SetPageLocked(page);
1490 __SetPageSwapBacked(page);
1495 shmem_inode_unacct_blocks(inode, nr);
1497 return ERR_PTR(err);
1501 * When a page is moved from swapcache to shmem filecache (either by the
1502 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1503 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1504 * ignorance of the mapping it belongs to. If that mapping has special
1505 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1506 * we may need to copy to a suitable page before moving to filecache.
1508 * In a future release, this may well be extended to respect cpuset and
1509 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1510 * but for now it is a simple matter of zone.
1512 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1514 return page_zonenum(page) > gfp_zone(gfp);
1517 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1518 struct shmem_inode_info *info, pgoff_t index)
1520 struct page *oldpage, *newpage;
1521 struct address_space *swap_mapping;
1526 swap_index = page_private(oldpage);
1527 swap_mapping = page_mapping(oldpage);
1530 * We have arrived here because our zones are constrained, so don't
1531 * limit chance of success by further cpuset and node constraints.
1533 gfp &= ~GFP_CONSTRAINT_MASK;
1534 newpage = shmem_alloc_page(gfp, info, index);
1539 copy_highpage(newpage, oldpage);
1540 flush_dcache_page(newpage);
1542 __SetPageLocked(newpage);
1543 __SetPageSwapBacked(newpage);
1544 SetPageUptodate(newpage);
1545 set_page_private(newpage, swap_index);
1546 SetPageSwapCache(newpage);
1549 * Our caller will very soon move newpage out of swapcache, but it's
1550 * a nice clean interface for us to replace oldpage by newpage there.
1552 spin_lock_irq(&swap_mapping->tree_lock);
1553 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1556 __inc_node_page_state(newpage, NR_FILE_PAGES);
1557 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1559 spin_unlock_irq(&swap_mapping->tree_lock);
1561 if (unlikely(error)) {
1563 * Is this possible? I think not, now that our callers check
1564 * both PageSwapCache and page_private after getting page lock;
1565 * but be defensive. Reverse old to newpage for clear and free.
1569 mem_cgroup_migrate(oldpage, newpage);
1570 lru_cache_add_anon(newpage);
1574 ClearPageSwapCache(oldpage);
1575 set_page_private(oldpage, 0);
1577 unlock_page(oldpage);
1584 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1586 * If we allocate a new one we do not mark it dirty. That's up to the
1587 * vm. If we swap it in we mark it dirty since we also free the swap
1588 * entry since a page cannot live in both the swap and page cache.
1590 * fault_mm and fault_type are only supplied by shmem_fault:
1591 * otherwise they are NULL.
1593 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1594 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1595 struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type)
1597 struct address_space *mapping = inode->i_mapping;
1598 struct shmem_inode_info *info = SHMEM_I(inode);
1599 struct shmem_sb_info *sbinfo;
1600 struct mm_struct *charge_mm;
1601 struct mem_cgroup *memcg;
1604 enum sgp_type sgp_huge = sgp;
1605 pgoff_t hindex = index;
1610 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1612 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1616 page = find_lock_entry(mapping, index);
1617 if (radix_tree_exceptional_entry(page)) {
1618 swap = radix_to_swp_entry(page);
1622 if (sgp <= SGP_CACHE &&
1623 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1628 if (page && sgp == SGP_WRITE)
1629 mark_page_accessed(page);
1631 /* fallocated page? */
1632 if (page && !PageUptodate(page)) {
1633 if (sgp != SGP_READ)
1639 if (page || (sgp == SGP_READ && !swap.val)) {
1645 * Fast cache lookup did not find it:
1646 * bring it back from swap or allocate.
1648 sbinfo = SHMEM_SB(inode->i_sb);
1649 charge_mm = vma ? vma->vm_mm : current->mm;
1652 /* Look it up and read it in.. */
1653 page = lookup_swap_cache(swap, NULL, 0);
1655 /* Or update major stats only when swapin succeeds?? */
1657 *fault_type |= VM_FAULT_MAJOR;
1658 count_vm_event(PGMAJFAULT);
1659 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1661 /* Here we actually start the io */
1662 page = shmem_swapin(swap, gfp, info, index);
1669 /* We have to do this with page locked to prevent races */
1671 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1672 !shmem_confirm_swap(mapping, index, swap)) {
1673 error = -EEXIST; /* try again */
1676 if (!PageUptodate(page)) {
1680 wait_on_page_writeback(page);
1682 if (shmem_should_replace_page(page, gfp)) {
1683 error = shmem_replace_page(&page, gfp, info, index);
1688 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1691 error = shmem_add_to_page_cache(page, mapping, index,
1692 swp_to_radix_entry(swap));
1694 * We already confirmed swap under page lock, and make
1695 * no memory allocation here, so usually no possibility
1696 * of error; but free_swap_and_cache() only trylocks a
1697 * page, so it is just possible that the entry has been
1698 * truncated or holepunched since swap was confirmed.
1699 * shmem_undo_range() will have done some of the
1700 * unaccounting, now delete_from_swap_cache() will do
1702 * Reset swap.val? No, leave it so "failed" goes back to
1703 * "repeat": reading a hole and writing should succeed.
1706 mem_cgroup_cancel_charge(page, memcg, false);
1707 delete_from_swap_cache(page);
1713 mem_cgroup_commit_charge(page, memcg, true, false);
1715 spin_lock_irq(&info->lock);
1717 shmem_recalc_inode(inode);
1718 spin_unlock_irq(&info->lock);
1720 if (sgp == SGP_WRITE)
1721 mark_page_accessed(page);
1723 delete_from_swap_cache(page);
1724 set_page_dirty(page);
1728 if (vma && userfaultfd_missing(vma)) {
1729 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1733 /* shmem_symlink() */
1734 if (mapping->a_ops != &shmem_aops)
1736 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1738 if (shmem_huge == SHMEM_HUGE_FORCE)
1740 switch (sbinfo->huge) {
1743 case SHMEM_HUGE_NEVER:
1745 case SHMEM_HUGE_WITHIN_SIZE:
1746 off = round_up(index, HPAGE_PMD_NR);
1747 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1748 if (i_size >= HPAGE_PMD_SIZE &&
1749 i_size >> PAGE_SHIFT >= off)
1752 case SHMEM_HUGE_ADVISE:
1753 if (sgp_huge == SGP_HUGE)
1755 /* TODO: implement fadvise() hints */
1760 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1762 alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode,
1767 error = PTR_ERR(page);
1769 if (error != -ENOSPC)
1772 * Try to reclaim some spece by splitting a huge page
1773 * beyond i_size on the filesystem.
1777 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1778 if (ret == SHRINK_STOP)
1786 if (PageTransHuge(page))
1787 hindex = round_down(index, HPAGE_PMD_NR);
1791 if (sgp == SGP_WRITE)
1792 __SetPageReferenced(page);
1794 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1795 PageTransHuge(page));
1798 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1799 compound_order(page));
1801 error = shmem_add_to_page_cache(page, mapping, hindex,
1803 radix_tree_preload_end();
1806 mem_cgroup_cancel_charge(page, memcg,
1807 PageTransHuge(page));
1810 mem_cgroup_commit_charge(page, memcg, false,
1811 PageTransHuge(page));
1812 lru_cache_add_anon(page);
1814 spin_lock_irq(&info->lock);
1815 info->alloced += 1 << compound_order(page);
1816 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1817 shmem_recalc_inode(inode);
1818 spin_unlock_irq(&info->lock);
1821 if (PageTransHuge(page) &&
1822 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1823 hindex + HPAGE_PMD_NR - 1) {
1825 * Part of the huge page is beyond i_size: subject
1826 * to shrink under memory pressure.
1828 spin_lock(&sbinfo->shrinklist_lock);
1830 * _careful to defend against unlocked access to
1831 * ->shrink_list in shmem_unused_huge_shrink()
1833 if (list_empty_careful(&info->shrinklist)) {
1834 list_add_tail(&info->shrinklist,
1835 &sbinfo->shrinklist);
1836 sbinfo->shrinklist_len++;
1838 spin_unlock(&sbinfo->shrinklist_lock);
1842 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1844 if (sgp == SGP_FALLOC)
1848 * Let SGP_WRITE caller clear ends if write does not fill page;
1849 * but SGP_FALLOC on a page fallocated earlier must initialize
1850 * it now, lest undo on failure cancel our earlier guarantee.
1852 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1853 struct page *head = compound_head(page);
1856 for (i = 0; i < (1 << compound_order(head)); i++) {
1857 clear_highpage(head + i);
1858 flush_dcache_page(head + i);
1860 SetPageUptodate(head);
1864 /* Perhaps the file has been truncated since we checked */
1865 if (sgp <= SGP_CACHE &&
1866 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1868 ClearPageDirty(page);
1869 delete_from_page_cache(page);
1870 spin_lock_irq(&info->lock);
1871 shmem_recalc_inode(inode);
1872 spin_unlock_irq(&info->lock);
1877 *pagep = page + index - hindex;
1884 shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
1886 if (PageTransHuge(page)) {
1892 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1899 if (error == -ENOSPC && !once++) {
1900 spin_lock_irq(&info->lock);
1901 shmem_recalc_inode(inode);
1902 spin_unlock_irq(&info->lock);
1905 if (error == -EEXIST) /* from above or from radix_tree_insert */
1911 * This is like autoremove_wake_function, but it removes the wait queue
1912 * entry unconditionally - even if something else had already woken the
1915 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1917 int ret = default_wake_function(wait, mode, sync, key);
1918 list_del_init(&wait->entry);
1922 static int shmem_fault(struct vm_fault *vmf)
1924 struct vm_area_struct *vma = vmf->vma;
1925 struct inode *inode = file_inode(vma->vm_file);
1926 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1929 int ret = VM_FAULT_LOCKED;
1932 * Trinity finds that probing a hole which tmpfs is punching can
1933 * prevent the hole-punch from ever completing: which in turn
1934 * locks writers out with its hold on i_mutex. So refrain from
1935 * faulting pages into the hole while it's being punched. Although
1936 * shmem_undo_range() does remove the additions, it may be unable to
1937 * keep up, as each new page needs its own unmap_mapping_range() call,
1938 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1940 * It does not matter if we sometimes reach this check just before the
1941 * hole-punch begins, so that one fault then races with the punch:
1942 * we just need to make racing faults a rare case.
1944 * The implementation below would be much simpler if we just used a
1945 * standard mutex or completion: but we cannot take i_mutex in fault,
1946 * and bloating every shmem inode for this unlikely case would be sad.
1948 if (unlikely(inode->i_private)) {
1949 struct shmem_falloc *shmem_falloc;
1951 spin_lock(&inode->i_lock);
1952 shmem_falloc = inode->i_private;
1954 shmem_falloc->waitq &&
1955 vmf->pgoff >= shmem_falloc->start &&
1956 vmf->pgoff < shmem_falloc->next) {
1957 wait_queue_head_t *shmem_falloc_waitq;
1958 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1960 ret = VM_FAULT_NOPAGE;
1961 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1962 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1963 /* It's polite to up mmap_sem if we can */
1964 up_read(&vma->vm_mm->mmap_sem);
1965 ret = VM_FAULT_RETRY;
1968 shmem_falloc_waitq = shmem_falloc->waitq;
1969 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1970 TASK_UNINTERRUPTIBLE);
1971 spin_unlock(&inode->i_lock);
1975 * shmem_falloc_waitq points into the shmem_fallocate()
1976 * stack of the hole-punching task: shmem_falloc_waitq
1977 * is usually invalid by the time we reach here, but
1978 * finish_wait() does not dereference it in that case;
1979 * though i_lock needed lest racing with wake_up_all().
1981 spin_lock(&inode->i_lock);
1982 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1983 spin_unlock(&inode->i_lock);
1986 spin_unlock(&inode->i_lock);
1991 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
1992 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
1994 else if (vma->vm_flags & VM_HUGEPAGE)
1997 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1998 gfp, vma, vmf, &ret);
2000 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
2004 unsigned long shmem_get_unmapped_area(struct file *file,
2005 unsigned long uaddr, unsigned long len,
2006 unsigned long pgoff, unsigned long flags)
2008 unsigned long (*get_area)(struct file *,
2009 unsigned long, unsigned long, unsigned long, unsigned long);
2011 unsigned long offset;
2012 unsigned long inflated_len;
2013 unsigned long inflated_addr;
2014 unsigned long inflated_offset;
2016 if (len > TASK_SIZE)
2019 get_area = current->mm->get_unmapped_area;
2020 addr = get_area(file, uaddr, len, pgoff, flags);
2022 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2024 if (IS_ERR_VALUE(addr))
2026 if (addr & ~PAGE_MASK)
2028 if (addr > TASK_SIZE - len)
2031 if (shmem_huge == SHMEM_HUGE_DENY)
2033 if (len < HPAGE_PMD_SIZE)
2035 if (flags & MAP_FIXED)
2038 * Our priority is to support MAP_SHARED mapped hugely;
2039 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2040 * But if caller specified an address hint, respect that as before.
2045 if (shmem_huge != SHMEM_HUGE_FORCE) {
2046 struct super_block *sb;
2049 VM_BUG_ON(file->f_op != &shmem_file_operations);
2050 sb = file_inode(file)->i_sb;
2053 * Called directly from mm/mmap.c, or drivers/char/mem.c
2054 * for "/dev/zero", to create a shared anonymous object.
2056 if (IS_ERR(shm_mnt))
2058 sb = shm_mnt->mnt_sb;
2060 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2064 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2065 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2067 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2070 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2071 if (inflated_len > TASK_SIZE)
2073 if (inflated_len < len)
2076 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2077 if (IS_ERR_VALUE(inflated_addr))
2079 if (inflated_addr & ~PAGE_MASK)
2082 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2083 inflated_addr += offset - inflated_offset;
2084 if (inflated_offset > offset)
2085 inflated_addr += HPAGE_PMD_SIZE;
2087 if (inflated_addr > TASK_SIZE - len)
2089 return inflated_addr;
2093 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2095 struct inode *inode = file_inode(vma->vm_file);
2096 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2099 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2102 struct inode *inode = file_inode(vma->vm_file);
2105 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2106 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2110 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2112 struct inode *inode = file_inode(file);
2113 struct shmem_inode_info *info = SHMEM_I(inode);
2114 int retval = -ENOMEM;
2116 spin_lock_irq(&info->lock);
2117 if (lock && !(info->flags & VM_LOCKED)) {
2118 if (!user_shm_lock(inode->i_size, user))
2120 info->flags |= VM_LOCKED;
2121 mapping_set_unevictable(file->f_mapping);
2123 if (!lock && (info->flags & VM_LOCKED) && user) {
2124 user_shm_unlock(inode->i_size, user);
2125 info->flags &= ~VM_LOCKED;
2126 mapping_clear_unevictable(file->f_mapping);
2131 spin_unlock_irq(&info->lock);
2135 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2137 file_accessed(file);
2138 vma->vm_ops = &shmem_vm_ops;
2139 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2140 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2141 (vma->vm_end & HPAGE_PMD_MASK)) {
2142 khugepaged_enter(vma, vma->vm_flags);
2147 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2148 umode_t mode, dev_t dev, unsigned long flags)
2150 struct inode *inode;
2151 struct shmem_inode_info *info;
2152 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2154 if (shmem_reserve_inode(sb))
2157 inode = new_inode(sb);
2159 inode->i_ino = get_next_ino();
2160 inode_init_owner(inode, dir, mode);
2161 inode->i_blocks = 0;
2162 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2163 inode->i_generation = get_seconds();
2164 info = SHMEM_I(inode);
2165 memset(info, 0, (char *)inode - (char *)info);
2166 spin_lock_init(&info->lock);
2167 info->seals = F_SEAL_SEAL;
2168 info->flags = flags & VM_NORESERVE;
2169 INIT_LIST_HEAD(&info->shrinklist);
2170 INIT_LIST_HEAD(&info->swaplist);
2171 simple_xattrs_init(&info->xattrs);
2172 cache_no_acl(inode);
2174 switch (mode & S_IFMT) {
2176 inode->i_op = &shmem_special_inode_operations;
2177 init_special_inode(inode, mode, dev);
2180 inode->i_mapping->a_ops = &shmem_aops;
2181 inode->i_op = &shmem_inode_operations;
2182 inode->i_fop = &shmem_file_operations;
2183 mpol_shared_policy_init(&info->policy,
2184 shmem_get_sbmpol(sbinfo));
2188 /* Some things misbehave if size == 0 on a directory */
2189 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2190 inode->i_op = &shmem_dir_inode_operations;
2191 inode->i_fop = &simple_dir_operations;
2195 * Must not load anything in the rbtree,
2196 * mpol_free_shared_policy will not be called.
2198 mpol_shared_policy_init(&info->policy, NULL);
2202 shmem_free_inode(sb);
2206 bool shmem_mapping(struct address_space *mapping)
2208 return mapping->a_ops == &shmem_aops;
2211 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2213 struct vm_area_struct *dst_vma,
2214 unsigned long dst_addr,
2215 unsigned long src_addr,
2217 struct page **pagep)
2219 struct inode *inode = file_inode(dst_vma->vm_file);
2220 struct shmem_inode_info *info = SHMEM_I(inode);
2221 struct address_space *mapping = inode->i_mapping;
2222 gfp_t gfp = mapping_gfp_mask(mapping);
2223 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2224 struct mem_cgroup *memcg;
2228 pte_t _dst_pte, *dst_pte;
2232 if (!shmem_inode_acct_block(inode, 1))
2236 page = shmem_alloc_page(gfp, info, pgoff);
2238 goto out_unacct_blocks;
2240 if (!zeropage) { /* mcopy_atomic */
2241 page_kaddr = kmap_atomic(page);
2242 ret = copy_from_user(page_kaddr,
2243 (const void __user *)src_addr,
2245 kunmap_atomic(page_kaddr);
2247 /* fallback to copy_from_user outside mmap_sem */
2248 if (unlikely(ret)) {
2250 shmem_inode_unacct_blocks(inode, 1);
2251 /* don't free the page */
2254 } else { /* mfill_zeropage_atomic */
2255 clear_highpage(page);
2262 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2263 __SetPageLocked(page);
2264 __SetPageSwapBacked(page);
2265 __SetPageUptodate(page);
2267 ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false);
2271 ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
2273 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL);
2274 radix_tree_preload_end();
2277 goto out_release_uncharge;
2279 mem_cgroup_commit_charge(page, memcg, false, false);
2281 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2282 if (dst_vma->vm_flags & VM_WRITE)
2283 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2286 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2287 if (!pte_none(*dst_pte))
2288 goto out_release_uncharge_unlock;
2290 lru_cache_add_anon(page);
2292 spin_lock(&info->lock);
2294 inode->i_blocks += BLOCKS_PER_PAGE;
2295 shmem_recalc_inode(inode);
2296 spin_unlock(&info->lock);
2298 inc_mm_counter(dst_mm, mm_counter_file(page));
2299 page_add_file_rmap(page, false);
2300 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2302 /* No need to invalidate - it was non-present before */
2303 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2305 pte_unmap_unlock(dst_pte, ptl);
2309 out_release_uncharge_unlock:
2310 pte_unmap_unlock(dst_pte, ptl);
2311 out_release_uncharge:
2312 mem_cgroup_cancel_charge(page, memcg, false);
2317 shmem_inode_unacct_blocks(inode, 1);
2321 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2323 struct vm_area_struct *dst_vma,
2324 unsigned long dst_addr,
2325 unsigned long src_addr,
2326 struct page **pagep)
2328 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2329 dst_addr, src_addr, false, pagep);
2332 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2334 struct vm_area_struct *dst_vma,
2335 unsigned long dst_addr)
2337 struct page *page = NULL;
2339 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2340 dst_addr, 0, true, &page);
2344 static const struct inode_operations shmem_symlink_inode_operations;
2345 static const struct inode_operations shmem_short_symlink_operations;
2347 #ifdef CONFIG_TMPFS_XATTR
2348 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2350 #define shmem_initxattrs NULL
2354 shmem_write_begin(struct file *file, struct address_space *mapping,
2355 loff_t pos, unsigned len, unsigned flags,
2356 struct page **pagep, void **fsdata)
2358 struct inode *inode = mapping->host;
2359 struct shmem_inode_info *info = SHMEM_I(inode);
2360 pgoff_t index = pos >> PAGE_SHIFT;
2362 /* i_mutex is held by caller */
2363 if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) {
2364 if (info->seals & F_SEAL_WRITE)
2366 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2370 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2374 shmem_write_end(struct file *file, struct address_space *mapping,
2375 loff_t pos, unsigned len, unsigned copied,
2376 struct page *page, void *fsdata)
2378 struct inode *inode = mapping->host;
2380 if (pos + copied > inode->i_size)
2381 i_size_write(inode, pos + copied);
2383 if (!PageUptodate(page)) {
2384 struct page *head = compound_head(page);
2385 if (PageTransCompound(page)) {
2388 for (i = 0; i < HPAGE_PMD_NR; i++) {
2389 if (head + i == page)
2391 clear_highpage(head + i);
2392 flush_dcache_page(head + i);
2395 if (copied < PAGE_SIZE) {
2396 unsigned from = pos & (PAGE_SIZE - 1);
2397 zero_user_segments(page, 0, from,
2398 from + copied, PAGE_SIZE);
2400 SetPageUptodate(head);
2402 set_page_dirty(page);
2409 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2411 struct file *file = iocb->ki_filp;
2412 struct inode *inode = file_inode(file);
2413 struct address_space *mapping = inode->i_mapping;
2415 unsigned long offset;
2416 enum sgp_type sgp = SGP_READ;
2419 loff_t *ppos = &iocb->ki_pos;
2422 * Might this read be for a stacking filesystem? Then when reading
2423 * holes of a sparse file, we actually need to allocate those pages,
2424 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2426 if (!iter_is_iovec(to))
2429 index = *ppos >> PAGE_SHIFT;
2430 offset = *ppos & ~PAGE_MASK;
2433 struct page *page = NULL;
2435 unsigned long nr, ret;
2436 loff_t i_size = i_size_read(inode);
2438 end_index = i_size >> PAGE_SHIFT;
2439 if (index > end_index)
2441 if (index == end_index) {
2442 nr = i_size & ~PAGE_MASK;
2447 error = shmem_getpage(inode, index, &page, sgp);
2449 if (error == -EINVAL)
2454 if (sgp == SGP_CACHE)
2455 set_page_dirty(page);
2460 * We must evaluate after, since reads (unlike writes)
2461 * are called without i_mutex protection against truncate
2464 i_size = i_size_read(inode);
2465 end_index = i_size >> PAGE_SHIFT;
2466 if (index == end_index) {
2467 nr = i_size & ~PAGE_MASK;
2478 * If users can be writing to this page using arbitrary
2479 * virtual addresses, take care about potential aliasing
2480 * before reading the page on the kernel side.
2482 if (mapping_writably_mapped(mapping))
2483 flush_dcache_page(page);
2485 * Mark the page accessed if we read the beginning.
2488 mark_page_accessed(page);
2490 page = ZERO_PAGE(0);
2495 * Ok, we have the page, and it's up-to-date, so
2496 * now we can copy it to user space...
2498 ret = copy_page_to_iter(page, offset, nr, to);
2501 index += offset >> PAGE_SHIFT;
2502 offset &= ~PAGE_MASK;
2505 if (!iov_iter_count(to))
2514 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2515 file_accessed(file);
2516 return retval ? retval : error;
2520 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2522 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2523 pgoff_t index, pgoff_t end, int whence)
2526 struct pagevec pvec;
2527 pgoff_t indices[PAGEVEC_SIZE];
2531 pagevec_init(&pvec);
2532 pvec.nr = 1; /* start small: we may be there already */
2534 pvec.nr = find_get_entries(mapping, index,
2535 pvec.nr, pvec.pages, indices);
2537 if (whence == SEEK_DATA)
2541 for (i = 0; i < pvec.nr; i++, index++) {
2542 if (index < indices[i]) {
2543 if (whence == SEEK_HOLE) {
2549 page = pvec.pages[i];
2550 if (page && !radix_tree_exceptional_entry(page)) {
2551 if (!PageUptodate(page))
2555 (page && whence == SEEK_DATA) ||
2556 (!page && whence == SEEK_HOLE)) {
2561 pagevec_remove_exceptionals(&pvec);
2562 pagevec_release(&pvec);
2563 pvec.nr = PAGEVEC_SIZE;
2569 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2571 struct address_space *mapping = file->f_mapping;
2572 struct inode *inode = mapping->host;
2576 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2577 return generic_file_llseek_size(file, offset, whence,
2578 MAX_LFS_FILESIZE, i_size_read(inode));
2580 /* We're holding i_mutex so we can access i_size directly */
2584 else if (offset >= inode->i_size)
2587 start = offset >> PAGE_SHIFT;
2588 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2589 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2590 new_offset <<= PAGE_SHIFT;
2591 if (new_offset > offset) {
2592 if (new_offset < inode->i_size)
2593 offset = new_offset;
2594 else if (whence == SEEK_DATA)
2597 offset = inode->i_size;
2602 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2603 inode_unlock(inode);
2608 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2609 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2611 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2612 #define LAST_SCAN 4 /* about 150ms max */
2614 static void shmem_tag_pins(struct address_space *mapping)
2616 struct radix_tree_iter iter;
2625 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2626 page = radix_tree_deref_slot(slot);
2627 if (!page || radix_tree_exception(page)) {
2628 if (radix_tree_deref_retry(page)) {
2629 slot = radix_tree_iter_retry(&iter);
2632 } else if (page_count(page) - page_mapcount(page) > 1) {
2633 spin_lock_irq(&mapping->tree_lock);
2634 radix_tree_tag_set(&mapping->page_tree, iter.index,
2636 spin_unlock_irq(&mapping->tree_lock);
2639 if (need_resched()) {
2640 slot = radix_tree_iter_resume(slot, &iter);
2648 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2649 * via get_user_pages(), drivers might have some pending I/O without any active
2650 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2651 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2652 * them to be dropped.
2653 * The caller must guarantee that no new user will acquire writable references
2654 * to those pages to avoid races.
2656 static int shmem_wait_for_pins(struct address_space *mapping)
2658 struct radix_tree_iter iter;
2664 shmem_tag_pins(mapping);
2667 for (scan = 0; scan <= LAST_SCAN; scan++) {
2668 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2672 lru_add_drain_all();
2673 else if (schedule_timeout_killable((HZ << scan) / 200))
2678 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2679 start, SHMEM_TAG_PINNED) {
2681 page = radix_tree_deref_slot(slot);
2682 if (radix_tree_exception(page)) {
2683 if (radix_tree_deref_retry(page)) {
2684 slot = radix_tree_iter_retry(&iter);
2692 page_count(page) - page_mapcount(page) != 1) {
2693 if (scan < LAST_SCAN)
2694 goto continue_resched;
2697 * On the last scan, we clean up all those tags
2698 * we inserted; but make a note that we still
2699 * found pages pinned.
2704 spin_lock_irq(&mapping->tree_lock);
2705 radix_tree_tag_clear(&mapping->page_tree,
2706 iter.index, SHMEM_TAG_PINNED);
2707 spin_unlock_irq(&mapping->tree_lock);
2709 if (need_resched()) {
2710 slot = radix_tree_iter_resume(slot, &iter);
2720 #define F_ALL_SEALS (F_SEAL_SEAL | \
2725 static int memfd_add_seals(struct file *file, unsigned int seals)
2727 struct inode *inode = file_inode(file);
2728 struct shmem_inode_info *info = SHMEM_I(inode);
2733 * Sealing allows multiple parties to share a shmem-file but restrict
2734 * access to a specific subset of file operations. Seals can only be
2735 * added, but never removed. This way, mutually untrusted parties can
2736 * share common memory regions with a well-defined policy. A malicious
2737 * peer can thus never perform unwanted operations on a shared object.
2739 * Seals are only supported on special shmem-files and always affect
2740 * the whole underlying inode. Once a seal is set, it may prevent some
2741 * kinds of access to the file. Currently, the following seals are
2743 * SEAL_SEAL: Prevent further seals from being set on this file
2744 * SEAL_SHRINK: Prevent the file from shrinking
2745 * SEAL_GROW: Prevent the file from growing
2746 * SEAL_WRITE: Prevent write access to the file
2748 * As we don't require any trust relationship between two parties, we
2749 * must prevent seals from being removed. Therefore, sealing a file
2750 * only adds a given set of seals to the file, it never touches
2751 * existing seals. Furthermore, the "setting seals"-operation can be
2752 * sealed itself, which basically prevents any further seal from being
2755 * Semantics of sealing are only defined on volatile files. Only
2756 * anonymous shmem files support sealing. More importantly, seals are
2757 * never written to disk. Therefore, there's no plan to support it on
2761 if (file->f_op != &shmem_file_operations)
2763 if (!(file->f_mode & FMODE_WRITE))
2765 if (seals & ~(unsigned int)F_ALL_SEALS)
2770 if (info->seals & F_SEAL_SEAL) {
2775 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2776 error = mapping_deny_writable(file->f_mapping);
2780 error = shmem_wait_for_pins(file->f_mapping);
2782 mapping_allow_writable(file->f_mapping);
2787 info->seals |= seals;
2791 inode_unlock(inode);
2795 static int memfd_get_seals(struct file *file)
2797 if (file->f_op != &shmem_file_operations)
2800 return SHMEM_I(file_inode(file))->seals;
2803 long memfd_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2809 /* disallow upper 32bit */
2813 error = memfd_add_seals(file, arg);
2816 error = memfd_get_seals(file);
2826 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2829 struct inode *inode = file_inode(file);
2830 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2831 struct shmem_inode_info *info = SHMEM_I(inode);
2832 struct shmem_falloc shmem_falloc;
2833 pgoff_t start, index, end;
2836 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2841 if (mode & FALLOC_FL_PUNCH_HOLE) {
2842 struct address_space *mapping = file->f_mapping;
2843 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2844 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2845 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2847 /* protected by i_mutex */
2848 if (info->seals & F_SEAL_WRITE) {
2853 shmem_falloc.waitq = &shmem_falloc_waitq;
2854 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2855 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2856 spin_lock(&inode->i_lock);
2857 inode->i_private = &shmem_falloc;
2858 spin_unlock(&inode->i_lock);
2860 if ((u64)unmap_end > (u64)unmap_start)
2861 unmap_mapping_range(mapping, unmap_start,
2862 1 + unmap_end - unmap_start, 0);
2863 shmem_truncate_range(inode, offset, offset + len - 1);
2864 /* No need to unmap again: hole-punching leaves COWed pages */
2866 spin_lock(&inode->i_lock);
2867 inode->i_private = NULL;
2868 wake_up_all(&shmem_falloc_waitq);
2869 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2870 spin_unlock(&inode->i_lock);
2875 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2876 error = inode_newsize_ok(inode, offset + len);
2880 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2885 start = offset >> PAGE_SHIFT;
2886 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2887 /* Try to avoid a swapstorm if len is impossible to satisfy */
2888 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2893 shmem_falloc.waitq = NULL;
2894 shmem_falloc.start = start;
2895 shmem_falloc.next = start;
2896 shmem_falloc.nr_falloced = 0;
2897 shmem_falloc.nr_unswapped = 0;
2898 spin_lock(&inode->i_lock);
2899 inode->i_private = &shmem_falloc;
2900 spin_unlock(&inode->i_lock);
2902 for (index = start; index < end; index++) {
2906 * Good, the fallocate(2) manpage permits EINTR: we may have
2907 * been interrupted because we are using up too much memory.
2909 if (signal_pending(current))
2911 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2914 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2916 /* Remove the !PageUptodate pages we added */
2917 if (index > start) {
2918 shmem_undo_range(inode,
2919 (loff_t)start << PAGE_SHIFT,
2920 ((loff_t)index << PAGE_SHIFT) - 1, true);
2926 * Inform shmem_writepage() how far we have reached.
2927 * No need for lock or barrier: we have the page lock.
2929 shmem_falloc.next++;
2930 if (!PageUptodate(page))
2931 shmem_falloc.nr_falloced++;
2934 * If !PageUptodate, leave it that way so that freeable pages
2935 * can be recognized if we need to rollback on error later.
2936 * But set_page_dirty so that memory pressure will swap rather
2937 * than free the pages we are allocating (and SGP_CACHE pages
2938 * might still be clean: we now need to mark those dirty too).
2940 set_page_dirty(page);
2946 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2947 i_size_write(inode, offset + len);
2948 inode->i_ctime = current_time(inode);
2950 spin_lock(&inode->i_lock);
2951 inode->i_private = NULL;
2952 spin_unlock(&inode->i_lock);
2954 inode_unlock(inode);
2958 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2960 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2962 buf->f_type = TMPFS_MAGIC;
2963 buf->f_bsize = PAGE_SIZE;
2964 buf->f_namelen = NAME_MAX;
2965 if (sbinfo->max_blocks) {
2966 buf->f_blocks = sbinfo->max_blocks;
2968 buf->f_bfree = sbinfo->max_blocks -
2969 percpu_counter_sum(&sbinfo->used_blocks);
2971 if (sbinfo->max_inodes) {
2972 buf->f_files = sbinfo->max_inodes;
2973 buf->f_ffree = sbinfo->free_inodes;
2975 /* else leave those fields 0 like simple_statfs */
2980 * File creation. Allocate an inode, and we're done..
2983 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2985 struct inode *inode;
2986 int error = -ENOSPC;
2988 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2990 error = simple_acl_create(dir, inode);
2993 error = security_inode_init_security(inode, dir,
2995 shmem_initxattrs, NULL);
2996 if (error && error != -EOPNOTSUPP)
3000 dir->i_size += BOGO_DIRENT_SIZE;
3001 dir->i_ctime = dir->i_mtime = current_time(dir);
3002 d_instantiate(dentry, inode);
3003 dget(dentry); /* Extra count - pin the dentry in core */
3012 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
3014 struct inode *inode;
3015 int error = -ENOSPC;
3017 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
3019 error = security_inode_init_security(inode, dir,
3021 shmem_initxattrs, NULL);
3022 if (error && error != -EOPNOTSUPP)
3024 error = simple_acl_create(dir, inode);
3027 d_tmpfile(dentry, inode);
3035 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3039 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
3045 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
3048 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
3054 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3056 struct inode *inode = d_inode(old_dentry);
3060 * No ordinary (disk based) filesystem counts links as inodes;
3061 * but each new link needs a new dentry, pinning lowmem, and
3062 * tmpfs dentries cannot be pruned until they are unlinked.
3064 ret = shmem_reserve_inode(inode->i_sb);
3068 dir->i_size += BOGO_DIRENT_SIZE;
3069 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3071 ihold(inode); /* New dentry reference */
3072 dget(dentry); /* Extra pinning count for the created dentry */
3073 d_instantiate(dentry, inode);
3078 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3080 struct inode *inode = d_inode(dentry);
3082 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3083 shmem_free_inode(inode->i_sb);
3085 dir->i_size -= BOGO_DIRENT_SIZE;
3086 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3088 dput(dentry); /* Undo the count from "create" - this does all the work */
3092 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3094 if (!simple_empty(dentry))
3097 drop_nlink(d_inode(dentry));
3099 return shmem_unlink(dir, dentry);
3102 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3104 bool old_is_dir = d_is_dir(old_dentry);
3105 bool new_is_dir = d_is_dir(new_dentry);
3107 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3109 drop_nlink(old_dir);
3112 drop_nlink(new_dir);
3116 old_dir->i_ctime = old_dir->i_mtime =
3117 new_dir->i_ctime = new_dir->i_mtime =
3118 d_inode(old_dentry)->i_ctime =
3119 d_inode(new_dentry)->i_ctime = current_time(old_dir);
3124 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3126 struct dentry *whiteout;
3129 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3133 error = shmem_mknod(old_dir, whiteout,
3134 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3140 * Cheat and hash the whiteout while the old dentry is still in
3141 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3143 * d_lookup() will consistently find one of them at this point,
3144 * not sure which one, but that isn't even important.
3151 * The VFS layer already does all the dentry stuff for rename,
3152 * we just have to decrement the usage count for the target if
3153 * it exists so that the VFS layer correctly free's it when it
3156 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3158 struct inode *inode = d_inode(old_dentry);
3159 int they_are_dirs = S_ISDIR(inode->i_mode);
3161 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3164 if (flags & RENAME_EXCHANGE)
3165 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3167 if (!simple_empty(new_dentry))
3170 if (flags & RENAME_WHITEOUT) {
3173 error = shmem_whiteout(old_dir, old_dentry);
3178 if (d_really_is_positive(new_dentry)) {
3179 (void) shmem_unlink(new_dir, new_dentry);
3180 if (they_are_dirs) {
3181 drop_nlink(d_inode(new_dentry));
3182 drop_nlink(old_dir);
3184 } else if (they_are_dirs) {
3185 drop_nlink(old_dir);
3189 old_dir->i_size -= BOGO_DIRENT_SIZE;
3190 new_dir->i_size += BOGO_DIRENT_SIZE;
3191 old_dir->i_ctime = old_dir->i_mtime =
3192 new_dir->i_ctime = new_dir->i_mtime =
3193 inode->i_ctime = current_time(old_dir);
3197 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3201 struct inode *inode;
3204 len = strlen(symname) + 1;
3205 if (len > PAGE_SIZE)
3206 return -ENAMETOOLONG;
3208 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3212 error = security_inode_init_security(inode, dir, &dentry->d_name,
3213 shmem_initxattrs, NULL);
3215 if (error != -EOPNOTSUPP) {
3222 inode->i_size = len-1;
3223 if (len <= SHORT_SYMLINK_LEN) {
3224 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3225 if (!inode->i_link) {
3229 inode->i_op = &shmem_short_symlink_operations;
3231 inode_nohighmem(inode);
3232 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3237 inode->i_mapping->a_ops = &shmem_aops;
3238 inode->i_op = &shmem_symlink_inode_operations;
3239 memcpy(page_address(page), symname, len);
3240 SetPageUptodate(page);
3241 set_page_dirty(page);
3245 dir->i_size += BOGO_DIRENT_SIZE;
3246 dir->i_ctime = dir->i_mtime = current_time(dir);
3247 d_instantiate(dentry, inode);
3252 static void shmem_put_link(void *arg)
3254 mark_page_accessed(arg);
3258 static const char *shmem_get_link(struct dentry *dentry,
3259 struct inode *inode,
3260 struct delayed_call *done)
3262 struct page *page = NULL;
3265 page = find_get_page(inode->i_mapping, 0);
3267 return ERR_PTR(-ECHILD);
3268 if (!PageUptodate(page)) {
3270 return ERR_PTR(-ECHILD);
3273 error = shmem_getpage(inode, 0, &page, SGP_READ);
3275 return ERR_PTR(error);
3278 set_delayed_call(done, shmem_put_link, page);
3279 return page_address(page);
3282 #ifdef CONFIG_TMPFS_XATTR
3284 * Superblocks without xattr inode operations may get some security.* xattr
3285 * support from the LSM "for free". As soon as we have any other xattrs
3286 * like ACLs, we also need to implement the security.* handlers at
3287 * filesystem level, though.
3291 * Callback for security_inode_init_security() for acquiring xattrs.
3293 static int shmem_initxattrs(struct inode *inode,
3294 const struct xattr *xattr_array,
3297 struct shmem_inode_info *info = SHMEM_I(inode);
3298 const struct xattr *xattr;
3299 struct simple_xattr *new_xattr;
3302 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3303 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3307 len = strlen(xattr->name) + 1;
3308 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3310 if (!new_xattr->name) {
3315 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3316 XATTR_SECURITY_PREFIX_LEN);
3317 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3320 simple_xattr_list_add(&info->xattrs, new_xattr);
3326 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3327 struct dentry *unused, struct inode *inode,
3328 const char *name, void *buffer, size_t size)
3330 struct shmem_inode_info *info = SHMEM_I(inode);
3332 name = xattr_full_name(handler, name);
3333 return simple_xattr_get(&info->xattrs, name, buffer, size);
3336 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3337 struct dentry *unused, struct inode *inode,
3338 const char *name, const void *value,
3339 size_t size, int flags)
3341 struct shmem_inode_info *info = SHMEM_I(inode);
3343 name = xattr_full_name(handler, name);
3344 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3347 static const struct xattr_handler shmem_security_xattr_handler = {
3348 .prefix = XATTR_SECURITY_PREFIX,
3349 .get = shmem_xattr_handler_get,
3350 .set = shmem_xattr_handler_set,
3353 static const struct xattr_handler shmem_trusted_xattr_handler = {
3354 .prefix = XATTR_TRUSTED_PREFIX,
3355 .get = shmem_xattr_handler_get,
3356 .set = shmem_xattr_handler_set,
3359 static const struct xattr_handler *shmem_xattr_handlers[] = {
3360 #ifdef CONFIG_TMPFS_POSIX_ACL
3361 &posix_acl_access_xattr_handler,
3362 &posix_acl_default_xattr_handler,
3364 &shmem_security_xattr_handler,
3365 &shmem_trusted_xattr_handler,
3369 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3371 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3372 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3374 #endif /* CONFIG_TMPFS_XATTR */
3376 static const struct inode_operations shmem_short_symlink_operations = {
3377 .get_link = simple_get_link,
3378 #ifdef CONFIG_TMPFS_XATTR
3379 .listxattr = shmem_listxattr,
3383 static const struct inode_operations shmem_symlink_inode_operations = {
3384 .get_link = shmem_get_link,
3385 #ifdef CONFIG_TMPFS_XATTR
3386 .listxattr = shmem_listxattr,
3390 static struct dentry *shmem_get_parent(struct dentry *child)
3392 return ERR_PTR(-ESTALE);
3395 static int shmem_match(struct inode *ino, void *vfh)
3399 inum = (inum << 32) | fh[1];
3400 return ino->i_ino == inum && fh[0] == ino->i_generation;
3403 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3404 struct fid *fid, int fh_len, int fh_type)
3406 struct inode *inode;
3407 struct dentry *dentry = NULL;
3414 inum = (inum << 32) | fid->raw[1];
3416 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3417 shmem_match, fid->raw);
3419 dentry = d_find_alias(inode);
3426 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3427 struct inode *parent)
3431 return FILEID_INVALID;
3434 if (inode_unhashed(inode)) {
3435 /* Unfortunately insert_inode_hash is not idempotent,
3436 * so as we hash inodes here rather than at creation
3437 * time, we need a lock to ensure we only try
3440 static DEFINE_SPINLOCK(lock);
3442 if (inode_unhashed(inode))
3443 __insert_inode_hash(inode,
3444 inode->i_ino + inode->i_generation);
3448 fh[0] = inode->i_generation;
3449 fh[1] = inode->i_ino;
3450 fh[2] = ((__u64)inode->i_ino) >> 32;
3456 static const struct export_operations shmem_export_ops = {
3457 .get_parent = shmem_get_parent,
3458 .encode_fh = shmem_encode_fh,
3459 .fh_to_dentry = shmem_fh_to_dentry,
3462 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3465 char *this_char, *value, *rest;
3466 struct mempolicy *mpol = NULL;
3470 while (options != NULL) {
3471 this_char = options;
3474 * NUL-terminate this option: unfortunately,
3475 * mount options form a comma-separated list,
3476 * but mpol's nodelist may also contain commas.
3478 options = strchr(options, ',');
3479 if (options == NULL)
3482 if (!isdigit(*options)) {
3489 if ((value = strchr(this_char,'=')) != NULL) {
3492 pr_err("tmpfs: No value for mount option '%s'\n",
3497 if (!strcmp(this_char,"size")) {
3498 unsigned long long size;
3499 size = memparse(value,&rest);
3501 size <<= PAGE_SHIFT;
3502 size *= totalram_pages;
3508 sbinfo->max_blocks =
3509 DIV_ROUND_UP(size, PAGE_SIZE);
3510 } else if (!strcmp(this_char,"nr_blocks")) {
3511 sbinfo->max_blocks = memparse(value, &rest);
3514 } else if (!strcmp(this_char,"nr_inodes")) {
3515 sbinfo->max_inodes = memparse(value, &rest);
3518 } else if (!strcmp(this_char,"mode")) {
3521 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3524 } else if (!strcmp(this_char,"uid")) {
3527 uid = simple_strtoul(value, &rest, 0);
3530 sbinfo->uid = make_kuid(current_user_ns(), uid);
3531 if (!uid_valid(sbinfo->uid))
3533 } else if (!strcmp(this_char,"gid")) {
3536 gid = simple_strtoul(value, &rest, 0);
3539 sbinfo->gid = make_kgid(current_user_ns(), gid);
3540 if (!gid_valid(sbinfo->gid))
3542 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3543 } else if (!strcmp(this_char, "huge")) {
3545 huge = shmem_parse_huge(value);
3548 if (!has_transparent_hugepage() &&
3549 huge != SHMEM_HUGE_NEVER)
3551 sbinfo->huge = huge;
3554 } else if (!strcmp(this_char,"mpol")) {
3557 if (mpol_parse_str(value, &mpol))
3561 pr_err("tmpfs: Bad mount option %s\n", this_char);
3565 sbinfo->mpol = mpol;
3569 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3577 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3579 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3580 struct shmem_sb_info config = *sbinfo;
3581 unsigned long inodes;
3582 int error = -EINVAL;
3585 if (shmem_parse_options(data, &config, true))
3588 spin_lock(&sbinfo->stat_lock);
3589 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3590 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3592 if (config.max_inodes < inodes)
3595 * Those tests disallow limited->unlimited while any are in use;
3596 * but we must separately disallow unlimited->limited, because
3597 * in that case we have no record of how much is already in use.
3599 if (config.max_blocks && !sbinfo->max_blocks)
3601 if (config.max_inodes && !sbinfo->max_inodes)
3605 sbinfo->huge = config.huge;
3606 sbinfo->max_blocks = config.max_blocks;
3607 sbinfo->max_inodes = config.max_inodes;
3608 sbinfo->free_inodes = config.max_inodes - inodes;
3611 * Preserve previous mempolicy unless mpol remount option was specified.
3614 mpol_put(sbinfo->mpol);
3615 sbinfo->mpol = config.mpol; /* transfers initial ref */
3618 spin_unlock(&sbinfo->stat_lock);
3622 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3624 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3626 if (sbinfo->max_blocks != shmem_default_max_blocks())
3627 seq_printf(seq, ",size=%luk",
3628 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3629 if (sbinfo->max_inodes != shmem_default_max_inodes())
3630 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3631 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3632 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3633 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3634 seq_printf(seq, ",uid=%u",
3635 from_kuid_munged(&init_user_ns, sbinfo->uid));
3636 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3637 seq_printf(seq, ",gid=%u",
3638 from_kgid_munged(&init_user_ns, sbinfo->gid));
3639 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3640 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3642 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3644 shmem_show_mpol(seq, sbinfo->mpol);
3648 #define MFD_NAME_PREFIX "memfd:"
3649 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3650 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3652 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
3654 SYSCALL_DEFINE2(memfd_create,
3655 const char __user *, uname,
3656 unsigned int, flags)
3658 struct shmem_inode_info *info;
3664 if (!(flags & MFD_HUGETLB)) {
3665 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3668 /* Sealing not supported in hugetlbfs (MFD_HUGETLB) */
3669 if (flags & MFD_ALLOW_SEALING)
3671 /* Allow huge page size encoding in flags. */
3672 if (flags & ~(unsigned int)(MFD_ALL_FLAGS |
3673 (MFD_HUGE_MASK << MFD_HUGE_SHIFT)))
3677 /* length includes terminating zero */
3678 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3681 if (len > MFD_NAME_MAX_LEN + 1)
3684 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL);
3688 strcpy(name, MFD_NAME_PREFIX);
3689 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3694 /* terminating-zero may have changed after strnlen_user() returned */
3695 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3700 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3706 if (flags & MFD_HUGETLB) {
3707 struct user_struct *user = NULL;
3709 file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user,
3710 HUGETLB_ANONHUGE_INODE,
3711 (flags >> MFD_HUGE_SHIFT) &
3714 file = shmem_file_setup(name, 0, VM_NORESERVE);
3716 error = PTR_ERR(file);
3719 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3720 file->f_flags |= O_RDWR | O_LARGEFILE;
3722 if (flags & MFD_ALLOW_SEALING) {
3724 * flags check at beginning of function ensures
3725 * this is not a hugetlbfs (MFD_HUGETLB) file.
3727 info = SHMEM_I(file_inode(file));
3728 info->seals &= ~F_SEAL_SEAL;
3731 fd_install(fd, file);
3742 #endif /* CONFIG_TMPFS */
3744 static void shmem_put_super(struct super_block *sb)
3746 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3748 percpu_counter_destroy(&sbinfo->used_blocks);
3749 mpol_put(sbinfo->mpol);
3751 sb->s_fs_info = NULL;
3754 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3756 struct inode *inode;
3757 struct shmem_sb_info *sbinfo;
3760 /* Round up to L1_CACHE_BYTES to resist false sharing */
3761 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3762 L1_CACHE_BYTES), GFP_KERNEL);
3766 sbinfo->mode = S_IRWXUGO | S_ISVTX;
3767 sbinfo->uid = current_fsuid();
3768 sbinfo->gid = current_fsgid();
3769 sb->s_fs_info = sbinfo;
3773 * Per default we only allow half of the physical ram per
3774 * tmpfs instance, limiting inodes to one per page of lowmem;
3775 * but the internal instance is left unlimited.
3777 if (!(sb->s_flags & SB_KERNMOUNT)) {
3778 sbinfo->max_blocks = shmem_default_max_blocks();
3779 sbinfo->max_inodes = shmem_default_max_inodes();
3780 if (shmem_parse_options(data, sbinfo, false)) {
3785 sb->s_flags |= SB_NOUSER;
3787 sb->s_export_op = &shmem_export_ops;
3788 sb->s_flags |= SB_NOSEC;
3790 sb->s_flags |= SB_NOUSER;
3793 spin_lock_init(&sbinfo->stat_lock);
3794 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3796 sbinfo->free_inodes = sbinfo->max_inodes;
3797 spin_lock_init(&sbinfo->shrinklist_lock);
3798 INIT_LIST_HEAD(&sbinfo->shrinklist);
3800 sb->s_maxbytes = MAX_LFS_FILESIZE;
3801 sb->s_blocksize = PAGE_SIZE;
3802 sb->s_blocksize_bits = PAGE_SHIFT;
3803 sb->s_magic = TMPFS_MAGIC;
3804 sb->s_op = &shmem_ops;
3805 sb->s_time_gran = 1;
3806 #ifdef CONFIG_TMPFS_XATTR
3807 sb->s_xattr = shmem_xattr_handlers;
3809 #ifdef CONFIG_TMPFS_POSIX_ACL
3810 sb->s_flags |= SB_POSIXACL;
3812 uuid_gen(&sb->s_uuid);
3814 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3817 inode->i_uid = sbinfo->uid;
3818 inode->i_gid = sbinfo->gid;
3819 sb->s_root = d_make_root(inode);
3825 shmem_put_super(sb);
3829 static struct kmem_cache *shmem_inode_cachep;
3831 static struct inode *shmem_alloc_inode(struct super_block *sb)
3833 struct shmem_inode_info *info;
3834 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3837 return &info->vfs_inode;
3840 static void shmem_destroy_callback(struct rcu_head *head)
3842 struct inode *inode = container_of(head, struct inode, i_rcu);
3843 if (S_ISLNK(inode->i_mode))
3844 kfree(inode->i_link);
3845 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3848 static void shmem_destroy_inode(struct inode *inode)
3850 if (S_ISREG(inode->i_mode))
3851 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3852 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3855 static void shmem_init_inode(void *foo)
3857 struct shmem_inode_info *info = foo;
3858 inode_init_once(&info->vfs_inode);
3861 static void shmem_init_inodecache(void)
3863 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3864 sizeof(struct shmem_inode_info),
3865 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3868 static void shmem_destroy_inodecache(void)
3870 kmem_cache_destroy(shmem_inode_cachep);
3873 static const struct address_space_operations shmem_aops = {
3874 .writepage = shmem_writepage,
3875 .set_page_dirty = __set_page_dirty_no_writeback,
3877 .write_begin = shmem_write_begin,
3878 .write_end = shmem_write_end,
3880 #ifdef CONFIG_MIGRATION
3881 .migratepage = migrate_page,
3883 .error_remove_page = generic_error_remove_page,
3886 static const struct file_operations shmem_file_operations = {
3888 .get_unmapped_area = shmem_get_unmapped_area,
3890 .llseek = shmem_file_llseek,
3891 .read_iter = shmem_file_read_iter,
3892 .write_iter = generic_file_write_iter,
3893 .fsync = noop_fsync,
3894 .splice_read = generic_file_splice_read,
3895 .splice_write = iter_file_splice_write,
3896 .fallocate = shmem_fallocate,
3900 static const struct inode_operations shmem_inode_operations = {
3901 .getattr = shmem_getattr,
3902 .setattr = shmem_setattr,
3903 #ifdef CONFIG_TMPFS_XATTR
3904 .listxattr = shmem_listxattr,
3905 .set_acl = simple_set_acl,
3909 static const struct inode_operations shmem_dir_inode_operations = {
3911 .create = shmem_create,
3912 .lookup = simple_lookup,
3914 .unlink = shmem_unlink,
3915 .symlink = shmem_symlink,
3916 .mkdir = shmem_mkdir,
3917 .rmdir = shmem_rmdir,
3918 .mknod = shmem_mknod,
3919 .rename = shmem_rename2,
3920 .tmpfile = shmem_tmpfile,
3922 #ifdef CONFIG_TMPFS_XATTR
3923 .listxattr = shmem_listxattr,
3925 #ifdef CONFIG_TMPFS_POSIX_ACL
3926 .setattr = shmem_setattr,
3927 .set_acl = simple_set_acl,
3931 static const struct inode_operations shmem_special_inode_operations = {
3932 #ifdef CONFIG_TMPFS_XATTR
3933 .listxattr = shmem_listxattr,
3935 #ifdef CONFIG_TMPFS_POSIX_ACL
3936 .setattr = shmem_setattr,
3937 .set_acl = simple_set_acl,
3941 static const struct super_operations shmem_ops = {
3942 .alloc_inode = shmem_alloc_inode,
3943 .destroy_inode = shmem_destroy_inode,
3945 .statfs = shmem_statfs,
3946 .remount_fs = shmem_remount_fs,
3947 .show_options = shmem_show_options,
3949 .evict_inode = shmem_evict_inode,
3950 .drop_inode = generic_delete_inode,
3951 .put_super = shmem_put_super,
3952 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3953 .nr_cached_objects = shmem_unused_huge_count,
3954 .free_cached_objects = shmem_unused_huge_scan,
3958 static const struct vm_operations_struct shmem_vm_ops = {
3959 .fault = shmem_fault,
3960 .map_pages = filemap_map_pages,
3962 .set_policy = shmem_set_policy,
3963 .get_policy = shmem_get_policy,
3967 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3968 int flags, const char *dev_name, void *data)
3970 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3973 static struct file_system_type shmem_fs_type = {
3974 .owner = THIS_MODULE,
3976 .mount = shmem_mount,
3977 .kill_sb = kill_litter_super,
3978 .fs_flags = FS_USERNS_MOUNT,
3981 int __init shmem_init(void)
3985 /* If rootfs called this, don't re-init */
3986 if (shmem_inode_cachep)
3989 shmem_init_inodecache();
3991 error = register_filesystem(&shmem_fs_type);
3993 pr_err("Could not register tmpfs\n");
3997 shm_mnt = kern_mount(&shmem_fs_type);
3998 if (IS_ERR(shm_mnt)) {
3999 error = PTR_ERR(shm_mnt);
4000 pr_err("Could not kern_mount tmpfs\n");
4004 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4005 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4006 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4008 shmem_huge = 0; /* just in case it was patched */
4013 unregister_filesystem(&shmem_fs_type);
4015 shmem_destroy_inodecache();
4016 shm_mnt = ERR_PTR(error);
4020 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
4021 static ssize_t shmem_enabled_show(struct kobject *kobj,
4022 struct kobj_attribute *attr, char *buf)
4026 SHMEM_HUGE_WITHIN_SIZE,
4034 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
4035 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
4037 count += sprintf(buf + count, fmt,
4038 shmem_format_huge(values[i]));
4040 buf[count - 1] = '\n';
4044 static ssize_t shmem_enabled_store(struct kobject *kobj,
4045 struct kobj_attribute *attr, const char *buf, size_t count)
4050 if (count + 1 > sizeof(tmp))
4052 memcpy(tmp, buf, count);
4054 if (count && tmp[count - 1] == '\n')
4055 tmp[count - 1] = '\0';
4057 huge = shmem_parse_huge(tmp);
4058 if (huge == -EINVAL)
4060 if (!has_transparent_hugepage() &&
4061 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4065 if (shmem_huge > SHMEM_HUGE_DENY)
4066 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4070 struct kobj_attribute shmem_enabled_attr =
4071 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4072 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4074 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4075 bool shmem_huge_enabled(struct vm_area_struct *vma)
4077 struct inode *inode = file_inode(vma->vm_file);
4078 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4082 if (shmem_huge == SHMEM_HUGE_FORCE)
4084 if (shmem_huge == SHMEM_HUGE_DENY)
4086 switch (sbinfo->huge) {
4087 case SHMEM_HUGE_NEVER:
4089 case SHMEM_HUGE_ALWAYS:
4091 case SHMEM_HUGE_WITHIN_SIZE:
4092 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4093 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4094 if (i_size >= HPAGE_PMD_SIZE &&
4095 i_size >> PAGE_SHIFT >= off)
4098 case SHMEM_HUGE_ADVISE:
4099 /* TODO: implement fadvise() hints */
4100 return (vma->vm_flags & VM_HUGEPAGE);
4106 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4108 #else /* !CONFIG_SHMEM */
4111 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4113 * This is intended for small system where the benefits of the full
4114 * shmem code (swap-backed and resource-limited) are outweighed by
4115 * their complexity. On systems without swap this code should be
4116 * effectively equivalent, but much lighter weight.
4119 static struct file_system_type shmem_fs_type = {
4121 .mount = ramfs_mount,
4122 .kill_sb = kill_litter_super,
4123 .fs_flags = FS_USERNS_MOUNT,
4126 int __init shmem_init(void)
4128 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4130 shm_mnt = kern_mount(&shmem_fs_type);
4131 BUG_ON(IS_ERR(shm_mnt));
4136 int shmem_unuse(swp_entry_t swap, struct page *page)
4141 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4146 void shmem_unlock_mapping(struct address_space *mapping)
4151 unsigned long shmem_get_unmapped_area(struct file *file,
4152 unsigned long addr, unsigned long len,
4153 unsigned long pgoff, unsigned long flags)
4155 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4159 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4161 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4163 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4165 #define shmem_vm_ops generic_file_vm_ops
4166 #define shmem_file_operations ramfs_file_operations
4167 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4168 #define shmem_acct_size(flags, size) 0
4169 #define shmem_unacct_size(flags, size) do {} while (0)
4171 #endif /* CONFIG_SHMEM */
4175 static const struct dentry_operations anon_ops = {
4176 .d_dname = simple_dname
4179 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4180 unsigned long flags, unsigned int i_flags)
4183 struct inode *inode;
4185 struct super_block *sb;
4189 return ERR_CAST(mnt);
4191 if (size < 0 || size > MAX_LFS_FILESIZE)
4192 return ERR_PTR(-EINVAL);
4194 if (shmem_acct_size(flags, size))
4195 return ERR_PTR(-ENOMEM);
4197 res = ERR_PTR(-ENOMEM);
4199 this.len = strlen(name);
4200 this.hash = 0; /* will go */
4202 path.mnt = mntget(mnt);
4203 path.dentry = d_alloc_pseudo(sb, &this);
4206 d_set_d_op(path.dentry, &anon_ops);
4208 res = ERR_PTR(-ENOSPC);
4209 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4213 inode->i_flags |= i_flags;
4214 d_instantiate(path.dentry, inode);
4215 inode->i_size = size;
4216 clear_nlink(inode); /* It is unlinked */
4217 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4221 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4222 &shmem_file_operations);
4229 shmem_unacct_size(flags, size);
4236 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4237 * kernel internal. There will be NO LSM permission checks against the
4238 * underlying inode. So users of this interface must do LSM checks at a
4239 * higher layer. The users are the big_key and shm implementations. LSM
4240 * checks are provided at the key or shm level rather than the inode.
4241 * @name: name for dentry (to be seen in /proc/<pid>/maps
4242 * @size: size to be set for the file
4243 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4245 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4247 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4251 * shmem_file_setup - get an unlinked file living in tmpfs
4252 * @name: name for dentry (to be seen in /proc/<pid>/maps
4253 * @size: size to be set for the file
4254 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4256 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4258 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4260 EXPORT_SYMBOL_GPL(shmem_file_setup);
4263 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4264 * @mnt: the tmpfs mount where the file will be created
4265 * @name: name for dentry (to be seen in /proc/<pid>/maps
4266 * @size: size to be set for the file
4267 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4269 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4270 loff_t size, unsigned long flags)
4272 return __shmem_file_setup(mnt, name, size, flags, 0);
4274 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4277 * shmem_zero_setup - setup a shared anonymous mapping
4278 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4280 int shmem_zero_setup(struct vm_area_struct *vma)
4283 loff_t size = vma->vm_end - vma->vm_start;
4286 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4287 * between XFS directory reading and selinux: since this file is only
4288 * accessible to the user through its mapping, use S_PRIVATE flag to
4289 * bypass file security, in the same way as shmem_kernel_file_setup().
4291 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4293 return PTR_ERR(file);
4297 vma->vm_file = file;
4298 vma->vm_ops = &shmem_vm_ops;
4300 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4301 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4302 (vma->vm_end & HPAGE_PMD_MASK)) {
4303 khugepaged_enter(vma, vma->vm_flags);
4310 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4311 * @mapping: the page's address_space
4312 * @index: the page index
4313 * @gfp: the page allocator flags to use if allocating
4315 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4316 * with any new page allocations done using the specified allocation flags.
4317 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4318 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4319 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4321 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4322 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4324 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4325 pgoff_t index, gfp_t gfp)
4328 struct inode *inode = mapping->host;
4332 BUG_ON(mapping->a_ops != &shmem_aops);
4333 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4334 gfp, NULL, NULL, NULL);
4336 page = ERR_PTR(error);
4342 * The tiny !SHMEM case uses ramfs without swap
4344 return read_cache_page_gfp(mapping, index, gfp);
4347 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);