2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
35 #include <linux/khugepaged.h>
37 static struct vfsmount *shm_mnt;
41 * This virtual memory filesystem is heavily based on the ramfs. It
42 * extends ramfs by the ability to use swap and honor resource limits
43 * which makes it a completely usable filesystem.
46 #include <linux/xattr.h>
47 #include <linux/exportfs.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/mman.h>
51 #include <linux/string.h>
52 #include <linux/slab.h>
53 #include <linux/backing-dev.h>
54 #include <linux/shmem_fs.h>
55 #include <linux/writeback.h>
56 #include <linux/blkdev.h>
57 #include <linux/pagevec.h>
58 #include <linux/percpu_counter.h>
59 #include <linux/falloc.h>
60 #include <linux/splice.h>
61 #include <linux/security.h>
62 #include <linux/swapops.h>
63 #include <linux/mempolicy.h>
64 #include <linux/namei.h>
65 #include <linux/ctype.h>
66 #include <linux/migrate.h>
67 #include <linux/highmem.h>
68 #include <linux/seq_file.h>
69 #include <linux/magic.h>
70 #include <linux/syscalls.h>
71 #include <linux/fcntl.h>
72 #include <uapi/linux/memfd.h>
74 #include <asm/uaccess.h>
75 #include <asm/pgtable.h>
79 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
80 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
82 /* Pretend that each entry is of this size in directory's i_size */
83 #define BOGO_DIRENT_SIZE 20
85 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
86 #define SHORT_SYMLINK_LEN 128
89 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
90 * inode->i_private (with i_mutex making sure that it has only one user at
91 * a time): we would prefer not to enlarge the shmem inode just for that.
94 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
95 pgoff_t start; /* start of range currently being fallocated */
96 pgoff_t next; /* the next page offset to be fallocated */
97 pgoff_t nr_falloced; /* how many new pages have been fallocated */
98 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
102 static unsigned long shmem_default_max_blocks(void)
104 return totalram_pages / 2;
107 static unsigned long shmem_default_max_inodes(void)
109 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
113 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
114 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
115 struct shmem_inode_info *info, pgoff_t index);
116 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
117 struct page **pagep, enum sgp_type sgp,
118 gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
120 int shmem_getpage(struct inode *inode, pgoff_t index,
121 struct page **pagep, enum sgp_type sgp)
123 return shmem_getpage_gfp(inode, index, pagep, sgp,
124 mapping_gfp_mask(inode->i_mapping), NULL, NULL);
127 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
129 return sb->s_fs_info;
133 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
134 * for shared memory and for shared anonymous (/dev/zero) mappings
135 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136 * consistent with the pre-accounting of private mappings ...
138 static inline int shmem_acct_size(unsigned long flags, loff_t size)
140 return (flags & VM_NORESERVE) ?
141 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
144 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
146 if (!(flags & VM_NORESERVE))
147 vm_unacct_memory(VM_ACCT(size));
150 static inline int shmem_reacct_size(unsigned long flags,
151 loff_t oldsize, loff_t newsize)
153 if (!(flags & VM_NORESERVE)) {
154 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
155 return security_vm_enough_memory_mm(current->mm,
156 VM_ACCT(newsize) - VM_ACCT(oldsize));
157 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
158 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
164 * ... whereas tmpfs objects are accounted incrementally as
165 * pages are allocated, in order to allow large sparse files.
166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169 static inline int shmem_acct_block(unsigned long flags, long pages)
171 if (!(flags & VM_NORESERVE))
174 return security_vm_enough_memory_mm(current->mm,
175 pages * VM_ACCT(PAGE_SIZE));
178 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
180 if (flags & VM_NORESERVE)
181 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
184 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
186 struct shmem_inode_info *info = SHMEM_I(inode);
187 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
189 if (shmem_acct_block(info->flags, pages))
192 if (sbinfo->max_blocks) {
193 if (percpu_counter_compare(&sbinfo->used_blocks,
194 sbinfo->max_blocks - pages) > 0)
196 percpu_counter_add(&sbinfo->used_blocks, pages);
202 shmem_unacct_blocks(info->flags, pages);
206 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
208 struct shmem_inode_info *info = SHMEM_I(inode);
209 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
211 if (sbinfo->max_blocks)
212 percpu_counter_sub(&sbinfo->used_blocks, pages);
213 shmem_unacct_blocks(info->flags, pages);
216 static const struct super_operations shmem_ops;
217 static const struct address_space_operations shmem_aops;
218 static const struct file_operations shmem_file_operations;
219 static const struct inode_operations shmem_inode_operations;
220 static const struct inode_operations shmem_dir_inode_operations;
221 static const struct inode_operations shmem_special_inode_operations;
222 static const struct vm_operations_struct shmem_vm_ops;
223 static struct file_system_type shmem_fs_type;
225 static LIST_HEAD(shmem_swaplist);
226 static DEFINE_MUTEX(shmem_swaplist_mutex);
228 static int shmem_reserve_inode(struct super_block *sb)
230 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
231 if (sbinfo->max_inodes) {
232 spin_lock(&sbinfo->stat_lock);
233 if (!sbinfo->free_inodes) {
234 spin_unlock(&sbinfo->stat_lock);
237 sbinfo->free_inodes--;
238 spin_unlock(&sbinfo->stat_lock);
243 static void shmem_free_inode(struct super_block *sb)
245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246 if (sbinfo->max_inodes) {
247 spin_lock(&sbinfo->stat_lock);
248 sbinfo->free_inodes++;
249 spin_unlock(&sbinfo->stat_lock);
254 * shmem_recalc_inode - recalculate the block usage of an inode
255 * @inode: inode to recalc
257 * We have to calculate the free blocks since the mm can drop
258 * undirtied hole pages behind our back.
260 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
261 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
263 * It has to be called with the spinlock held.
265 static void shmem_recalc_inode(struct inode *inode)
267 struct shmem_inode_info *info = SHMEM_I(inode);
270 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
272 info->alloced -= freed;
273 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
274 shmem_inode_unacct_blocks(inode, freed);
278 bool shmem_charge(struct inode *inode, long pages)
280 struct shmem_inode_info *info = SHMEM_I(inode);
283 if (!shmem_inode_acct_block(inode, pages))
286 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
287 inode->i_mapping->nrpages += pages;
289 spin_lock_irqsave(&info->lock, flags);
290 info->alloced += pages;
291 inode->i_blocks += pages * BLOCKS_PER_PAGE;
292 shmem_recalc_inode(inode);
293 spin_unlock_irqrestore(&info->lock, flags);
298 void shmem_uncharge(struct inode *inode, long pages)
300 struct shmem_inode_info *info = SHMEM_I(inode);
303 /* nrpages adjustment done by __delete_from_page_cache() or caller */
305 spin_lock_irqsave(&info->lock, flags);
306 info->alloced -= pages;
307 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
308 shmem_recalc_inode(inode);
309 spin_unlock_irqrestore(&info->lock, flags);
311 shmem_inode_unacct_blocks(inode, pages);
315 * Replace item expected in radix tree by a new item, while holding tree lock.
317 static int shmem_radix_tree_replace(struct address_space *mapping,
318 pgoff_t index, void *expected, void *replacement)
323 VM_BUG_ON(!expected);
324 VM_BUG_ON(!replacement);
325 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
328 item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
329 if (item != expected)
331 radix_tree_replace_slot(pslot, replacement);
336 * Sometimes, before we decide whether to proceed or to fail, we must check
337 * that an entry was not already brought back from swap by a racing thread.
339 * Checking page is not enough: by the time a SwapCache page is locked, it
340 * might be reused, and again be SwapCache, using the same swap as before.
342 static bool shmem_confirm_swap(struct address_space *mapping,
343 pgoff_t index, swp_entry_t swap)
348 item = radix_tree_lookup(&mapping->page_tree, index);
350 return item == swp_to_radix_entry(swap);
354 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
357 * disables huge pages for the mount;
359 * enables huge pages for the mount;
360 * SHMEM_HUGE_WITHIN_SIZE:
361 * only allocate huge pages if the page will be fully within i_size,
362 * also respect fadvise()/madvise() hints;
364 * only allocate huge pages if requested with fadvise()/madvise();
367 #define SHMEM_HUGE_NEVER 0
368 #define SHMEM_HUGE_ALWAYS 1
369 #define SHMEM_HUGE_WITHIN_SIZE 2
370 #define SHMEM_HUGE_ADVISE 3
374 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
377 * disables huge on shm_mnt and all mounts, for emergency use;
379 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
382 #define SHMEM_HUGE_DENY (-1)
383 #define SHMEM_HUGE_FORCE (-2)
385 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
386 /* ifdef here to avoid bloating shmem.o when not necessary */
388 int shmem_huge __read_mostly;
390 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
391 static int shmem_parse_huge(const char *str)
393 if (!strcmp(str, "never"))
394 return SHMEM_HUGE_NEVER;
395 if (!strcmp(str, "always"))
396 return SHMEM_HUGE_ALWAYS;
397 if (!strcmp(str, "within_size"))
398 return SHMEM_HUGE_WITHIN_SIZE;
399 if (!strcmp(str, "advise"))
400 return SHMEM_HUGE_ADVISE;
401 if (!strcmp(str, "deny"))
402 return SHMEM_HUGE_DENY;
403 if (!strcmp(str, "force"))
404 return SHMEM_HUGE_FORCE;
408 static const char *shmem_format_huge(int huge)
411 case SHMEM_HUGE_NEVER:
413 case SHMEM_HUGE_ALWAYS:
415 case SHMEM_HUGE_WITHIN_SIZE:
416 return "within_size";
417 case SHMEM_HUGE_ADVISE:
419 case SHMEM_HUGE_DENY:
421 case SHMEM_HUGE_FORCE:
430 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
431 struct shrink_control *sc, unsigned long nr_to_split)
433 LIST_HEAD(list), *pos, *next;
434 LIST_HEAD(to_remove);
436 struct shmem_inode_info *info;
438 unsigned long batch = sc ? sc->nr_to_scan : 128;
441 if (list_empty(&sbinfo->shrinklist))
444 spin_lock(&sbinfo->shrinklist_lock);
445 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
446 info = list_entry(pos, struct shmem_inode_info, shrinklist);
449 inode = igrab(&info->vfs_inode);
451 /* inode is about to be evicted */
453 list_del_init(&info->shrinklist);
457 /* Check if there's anything to gain */
458 if (round_up(inode->i_size, PAGE_SIZE) ==
459 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
460 list_move(&info->shrinklist, &to_remove);
464 list_move(&info->shrinklist, &list);
466 sbinfo->shrinklist_len--;
470 spin_unlock(&sbinfo->shrinklist_lock);
472 list_for_each_safe(pos, next, &to_remove) {
473 info = list_entry(pos, struct shmem_inode_info, shrinklist);
474 inode = &info->vfs_inode;
475 list_del_init(&info->shrinklist);
479 list_for_each_safe(pos, next, &list) {
482 info = list_entry(pos, struct shmem_inode_info, shrinklist);
483 inode = &info->vfs_inode;
485 if (nr_to_split && split >= nr_to_split)
488 page = find_get_page(inode->i_mapping,
489 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
493 /* No huge page at the end of the file: nothing to split */
494 if (!PageTransHuge(page)) {
500 * Move the inode on the list back to shrinklist if we failed
501 * to lock the page at this time.
503 * Waiting for the lock may lead to deadlock in the
506 if (!trylock_page(page)) {
511 ret = split_huge_page(page);
515 /* If split failed move the inode on the list back to shrinklist */
521 list_del_init(&info->shrinklist);
525 * Make sure the inode is either on the global list or deleted
526 * from any local list before iput() since it could be deleted
527 * in another thread once we put the inode (then the local list
530 spin_lock(&sbinfo->shrinklist_lock);
531 list_move(&info->shrinklist, &sbinfo->shrinklist);
532 sbinfo->shrinklist_len++;
533 spin_unlock(&sbinfo->shrinklist_lock);
541 static long shmem_unused_huge_scan(struct super_block *sb,
542 struct shrink_control *sc)
544 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
546 if (!READ_ONCE(sbinfo->shrinklist_len))
549 return shmem_unused_huge_shrink(sbinfo, sc, 0);
552 static long shmem_unused_huge_count(struct super_block *sb,
553 struct shrink_control *sc)
555 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
556 return READ_ONCE(sbinfo->shrinklist_len);
558 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
560 #define shmem_huge SHMEM_HUGE_DENY
562 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
563 struct shrink_control *sc, unsigned long nr_to_split)
567 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
570 * Like add_to_page_cache_locked, but error if expected item has gone.
572 static int shmem_add_to_page_cache(struct page *page,
573 struct address_space *mapping,
574 pgoff_t index, void *expected)
576 int error, nr = hpage_nr_pages(page);
578 VM_BUG_ON_PAGE(PageTail(page), page);
579 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
580 VM_BUG_ON_PAGE(!PageLocked(page), page);
581 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
582 VM_BUG_ON(expected && PageTransHuge(page));
584 page_ref_add(page, nr);
585 page->mapping = mapping;
588 spin_lock_irq(&mapping->tree_lock);
589 if (PageTransHuge(page)) {
590 void __rcu **results;
595 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
596 &results, &idx, index, 1) &&
597 idx < index + HPAGE_PMD_NR) {
602 for (i = 0; i < HPAGE_PMD_NR; i++) {
603 error = radix_tree_insert(&mapping->page_tree,
604 index + i, page + i);
607 count_vm_event(THP_FILE_ALLOC);
609 } else if (!expected) {
610 error = radix_tree_insert(&mapping->page_tree, index, page);
612 error = shmem_radix_tree_replace(mapping, index, expected,
617 mapping->nrpages += nr;
618 if (PageTransHuge(page))
619 __inc_node_page_state(page, NR_SHMEM_THPS);
620 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
621 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
622 spin_unlock_irq(&mapping->tree_lock);
624 page->mapping = NULL;
625 spin_unlock_irq(&mapping->tree_lock);
626 page_ref_sub(page, nr);
632 * Like delete_from_page_cache, but substitutes swap for page.
634 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
636 struct address_space *mapping = page->mapping;
639 VM_BUG_ON_PAGE(PageCompound(page), page);
641 spin_lock_irq(&mapping->tree_lock);
642 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
643 page->mapping = NULL;
645 __dec_node_page_state(page, NR_FILE_PAGES);
646 __dec_node_page_state(page, NR_SHMEM);
647 spin_unlock_irq(&mapping->tree_lock);
653 * Remove swap entry from radix tree, free the swap and its page cache.
655 static int shmem_free_swap(struct address_space *mapping,
656 pgoff_t index, void *radswap)
660 spin_lock_irq(&mapping->tree_lock);
661 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
662 spin_unlock_irq(&mapping->tree_lock);
665 free_swap_and_cache(radix_to_swp_entry(radswap));
670 * Determine (in bytes) how many of the shmem object's pages mapped by the
671 * given offsets are swapped out.
673 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
674 * as long as the inode doesn't go away and racy results are not a problem.
676 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
677 pgoff_t start, pgoff_t end)
679 struct radix_tree_iter iter;
682 unsigned long swapped = 0;
686 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
687 if (iter.index >= end)
690 page = radix_tree_deref_slot(slot);
692 if (radix_tree_deref_retry(page)) {
693 slot = radix_tree_iter_retry(&iter);
697 if (radix_tree_exceptional_entry(page))
700 if (need_resched()) {
702 slot = radix_tree_iter_next(&iter);
708 return swapped << PAGE_SHIFT;
712 * Determine (in bytes) how many of the shmem object's pages mapped by the
713 * given vma is swapped out.
715 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
716 * as long as the inode doesn't go away and racy results are not a problem.
718 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
720 struct inode *inode = file_inode(vma->vm_file);
721 struct shmem_inode_info *info = SHMEM_I(inode);
722 struct address_space *mapping = inode->i_mapping;
723 unsigned long swapped;
725 /* Be careful as we don't hold info->lock */
726 swapped = READ_ONCE(info->swapped);
729 * The easier cases are when the shmem object has nothing in swap, or
730 * the vma maps it whole. Then we can simply use the stats that we
736 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
737 return swapped << PAGE_SHIFT;
739 /* Here comes the more involved part */
740 return shmem_partial_swap_usage(mapping,
741 linear_page_index(vma, vma->vm_start),
742 linear_page_index(vma, vma->vm_end));
746 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
748 void shmem_unlock_mapping(struct address_space *mapping)
751 pgoff_t indices[PAGEVEC_SIZE];
754 pagevec_init(&pvec, 0);
756 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
758 while (!mapping_unevictable(mapping)) {
760 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
761 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
763 pvec.nr = find_get_entries(mapping, index,
764 PAGEVEC_SIZE, pvec.pages, indices);
767 index = indices[pvec.nr - 1] + 1;
768 pagevec_remove_exceptionals(&pvec);
769 check_move_unevictable_pages(pvec.pages, pvec.nr);
770 pagevec_release(&pvec);
776 * Remove range of pages and swap entries from radix tree, and free them.
777 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
779 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
782 struct address_space *mapping = inode->i_mapping;
783 struct shmem_inode_info *info = SHMEM_I(inode);
784 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
785 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
786 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
787 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
789 pgoff_t indices[PAGEVEC_SIZE];
790 long nr_swaps_freed = 0;
795 end = -1; /* unsigned, so actually very big */
797 pagevec_init(&pvec, 0);
799 while (index < end) {
800 pvec.nr = find_get_entries(mapping, index,
801 min(end - index, (pgoff_t)PAGEVEC_SIZE),
802 pvec.pages, indices);
805 for (i = 0; i < pagevec_count(&pvec); i++) {
806 struct page *page = pvec.pages[i];
812 if (radix_tree_exceptional_entry(page)) {
815 nr_swaps_freed += !shmem_free_swap(mapping,
820 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
822 if (!trylock_page(page))
825 if (PageTransTail(page)) {
826 /* Middle of THP: zero out the page */
827 clear_highpage(page);
830 } else if (PageTransHuge(page)) {
831 if (index == round_down(end, HPAGE_PMD_NR)) {
833 * Range ends in the middle of THP:
836 clear_highpage(page);
840 index += HPAGE_PMD_NR - 1;
841 i += HPAGE_PMD_NR - 1;
844 if (!unfalloc || !PageUptodate(page)) {
845 VM_BUG_ON_PAGE(PageTail(page), page);
846 if (page_mapping(page) == mapping) {
847 VM_BUG_ON_PAGE(PageWriteback(page), page);
848 truncate_inode_page(mapping, page);
853 pagevec_remove_exceptionals(&pvec);
854 pagevec_release(&pvec);
860 struct page *page = NULL;
861 shmem_getpage(inode, start - 1, &page, SGP_READ);
863 unsigned int top = PAGE_SIZE;
868 zero_user_segment(page, partial_start, top);
869 set_page_dirty(page);
875 struct page *page = NULL;
876 shmem_getpage(inode, end, &page, SGP_READ);
878 zero_user_segment(page, 0, partial_end);
879 set_page_dirty(page);
888 while (index < end) {
891 pvec.nr = find_get_entries(mapping, index,
892 min(end - index, (pgoff_t)PAGEVEC_SIZE),
893 pvec.pages, indices);
895 /* If all gone or hole-punch or unfalloc, we're done */
896 if (index == start || end != -1)
898 /* But if truncating, restart to make sure all gone */
902 for (i = 0; i < pagevec_count(&pvec); i++) {
903 struct page *page = pvec.pages[i];
909 if (radix_tree_exceptional_entry(page)) {
912 if (shmem_free_swap(mapping, index, page)) {
913 /* Swap was replaced by page: retry */
923 if (PageTransTail(page)) {
924 /* Middle of THP: zero out the page */
925 clear_highpage(page);
928 * Partial thp truncate due 'start' in middle
929 * of THP: don't need to look on these pages
930 * again on !pvec.nr restart.
932 if (index != round_down(end, HPAGE_PMD_NR))
935 } else if (PageTransHuge(page)) {
936 if (index == round_down(end, HPAGE_PMD_NR)) {
938 * Range ends in the middle of THP:
941 clear_highpage(page);
945 index += HPAGE_PMD_NR - 1;
946 i += HPAGE_PMD_NR - 1;
949 if (!unfalloc || !PageUptodate(page)) {
950 VM_BUG_ON_PAGE(PageTail(page), page);
951 if (page_mapping(page) == mapping) {
952 VM_BUG_ON_PAGE(PageWriteback(page), page);
953 truncate_inode_page(mapping, page);
955 /* Page was replaced by swap: retry */
963 pagevec_remove_exceptionals(&pvec);
964 pagevec_release(&pvec);
968 spin_lock_irq(&info->lock);
969 info->swapped -= nr_swaps_freed;
970 shmem_recalc_inode(inode);
971 spin_unlock_irq(&info->lock);
974 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
976 shmem_undo_range(inode, lstart, lend, false);
977 inode->i_ctime = inode->i_mtime = current_time(inode);
979 EXPORT_SYMBOL_GPL(shmem_truncate_range);
981 static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
984 struct inode *inode = dentry->d_inode;
985 struct shmem_inode_info *info = SHMEM_I(inode);
987 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
988 spin_lock_irq(&info->lock);
989 shmem_recalc_inode(inode);
990 spin_unlock_irq(&info->lock);
992 generic_fillattr(inode, stat);
996 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
998 struct inode *inode = d_inode(dentry);
999 struct shmem_inode_info *info = SHMEM_I(inode);
1000 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1003 error = setattr_prepare(dentry, attr);
1007 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1008 loff_t oldsize = inode->i_size;
1009 loff_t newsize = attr->ia_size;
1011 /* protected by i_mutex */
1012 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1013 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1016 if (newsize != oldsize) {
1017 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1021 i_size_write(inode, newsize);
1022 inode->i_ctime = inode->i_mtime = current_time(inode);
1024 if (newsize <= oldsize) {
1025 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1026 if (oldsize > holebegin)
1027 unmap_mapping_range(inode->i_mapping,
1030 shmem_truncate_range(inode,
1031 newsize, (loff_t)-1);
1032 /* unmap again to remove racily COWed private pages */
1033 if (oldsize > holebegin)
1034 unmap_mapping_range(inode->i_mapping,
1038 * Part of the huge page can be beyond i_size: subject
1039 * to shrink under memory pressure.
1041 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1042 spin_lock(&sbinfo->shrinklist_lock);
1044 * _careful to defend against unlocked access to
1045 * ->shrink_list in shmem_unused_huge_shrink()
1047 if (list_empty_careful(&info->shrinklist)) {
1048 list_add_tail(&info->shrinklist,
1049 &sbinfo->shrinklist);
1050 sbinfo->shrinklist_len++;
1052 spin_unlock(&sbinfo->shrinklist_lock);
1057 setattr_copy(inode, attr);
1058 if (attr->ia_valid & ATTR_MODE)
1059 error = posix_acl_chmod(inode, inode->i_mode);
1063 static void shmem_evict_inode(struct inode *inode)
1065 struct shmem_inode_info *info = SHMEM_I(inode);
1066 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1068 if (inode->i_mapping->a_ops == &shmem_aops) {
1069 shmem_unacct_size(info->flags, inode->i_size);
1071 shmem_truncate_range(inode, 0, (loff_t)-1);
1072 if (!list_empty(&info->shrinklist)) {
1073 spin_lock(&sbinfo->shrinklist_lock);
1074 if (!list_empty(&info->shrinklist)) {
1075 list_del_init(&info->shrinklist);
1076 sbinfo->shrinklist_len--;
1078 spin_unlock(&sbinfo->shrinklist_lock);
1080 if (!list_empty(&info->swaplist)) {
1081 mutex_lock(&shmem_swaplist_mutex);
1082 list_del_init(&info->swaplist);
1083 mutex_unlock(&shmem_swaplist_mutex);
1087 simple_xattrs_free(&info->xattrs);
1088 WARN_ON(inode->i_blocks);
1089 shmem_free_inode(inode->i_sb);
1094 * If swap found in inode, free it and move page from swapcache to filecache.
1096 static int shmem_unuse_inode(struct shmem_inode_info *info,
1097 swp_entry_t swap, struct page **pagep)
1099 struct address_space *mapping = info->vfs_inode.i_mapping;
1105 radswap = swp_to_radix_entry(swap);
1106 index = radix_tree_locate_item(&mapping->page_tree, radswap);
1108 return -EAGAIN; /* tell shmem_unuse we found nothing */
1111 * Move _head_ to start search for next from here.
1112 * But be careful: shmem_evict_inode checks list_empty without taking
1113 * mutex, and there's an instant in list_move_tail when info->swaplist
1114 * would appear empty, if it were the only one on shmem_swaplist.
1116 if (shmem_swaplist.next != &info->swaplist)
1117 list_move_tail(&shmem_swaplist, &info->swaplist);
1119 gfp = mapping_gfp_mask(mapping);
1120 if (shmem_should_replace_page(*pagep, gfp)) {
1121 mutex_unlock(&shmem_swaplist_mutex);
1122 error = shmem_replace_page(pagep, gfp, info, index);
1123 mutex_lock(&shmem_swaplist_mutex);
1125 * We needed to drop mutex to make that restrictive page
1126 * allocation, but the inode might have been freed while we
1127 * dropped it: although a racing shmem_evict_inode() cannot
1128 * complete without emptying the radix_tree, our page lock
1129 * on this swapcache page is not enough to prevent that -
1130 * free_swap_and_cache() of our swap entry will only
1131 * trylock_page(), removing swap from radix_tree whatever.
1133 * We must not proceed to shmem_add_to_page_cache() if the
1134 * inode has been freed, but of course we cannot rely on
1135 * inode or mapping or info to check that. However, we can
1136 * safely check if our swap entry is still in use (and here
1137 * it can't have got reused for another page): if it's still
1138 * in use, then the inode cannot have been freed yet, and we
1139 * can safely proceed (if it's no longer in use, that tells
1140 * nothing about the inode, but we don't need to unuse swap).
1142 if (!page_swapcount(*pagep))
1147 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1148 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1149 * beneath us (pagelock doesn't help until the page is in pagecache).
1152 error = shmem_add_to_page_cache(*pagep, mapping, index,
1154 if (error != -ENOMEM) {
1156 * Truncation and eviction use free_swap_and_cache(), which
1157 * only does trylock page: if we raced, best clean up here.
1159 delete_from_swap_cache(*pagep);
1160 set_page_dirty(*pagep);
1162 spin_lock_irq(&info->lock);
1164 spin_unlock_irq(&info->lock);
1172 * Search through swapped inodes to find and replace swap by page.
1174 int shmem_unuse(swp_entry_t swap, struct page *page)
1176 struct list_head *this, *next;
1177 struct shmem_inode_info *info;
1178 struct mem_cgroup *memcg;
1182 * There's a faint possibility that swap page was replaced before
1183 * caller locked it: caller will come back later with the right page.
1185 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1189 * Charge page using GFP_KERNEL while we can wait, before taking
1190 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1191 * Charged back to the user (not to caller) when swap account is used.
1193 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1197 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1200 mutex_lock(&shmem_swaplist_mutex);
1201 list_for_each_safe(this, next, &shmem_swaplist) {
1202 info = list_entry(this, struct shmem_inode_info, swaplist);
1204 error = shmem_unuse_inode(info, swap, &page);
1206 list_del_init(&info->swaplist);
1208 if (error != -EAGAIN)
1210 /* found nothing in this: move on to search the next */
1212 mutex_unlock(&shmem_swaplist_mutex);
1215 if (error != -ENOMEM)
1217 mem_cgroup_cancel_charge(page, memcg, false);
1219 mem_cgroup_commit_charge(page, memcg, true, false);
1227 * Move the page from the page cache to the swap cache.
1229 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1231 struct shmem_inode_info *info;
1232 struct address_space *mapping;
1233 struct inode *inode;
1237 VM_BUG_ON_PAGE(PageCompound(page), page);
1238 BUG_ON(!PageLocked(page));
1239 mapping = page->mapping;
1240 index = page->index;
1241 inode = mapping->host;
1242 info = SHMEM_I(inode);
1243 if (info->flags & VM_LOCKED)
1245 if (!total_swap_pages)
1249 * Our capabilities prevent regular writeback or sync from ever calling
1250 * shmem_writepage; but a stacking filesystem might use ->writepage of
1251 * its underlying filesystem, in which case tmpfs should write out to
1252 * swap only in response to memory pressure, and not for the writeback
1255 if (!wbc->for_reclaim) {
1256 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1261 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1262 * value into swapfile.c, the only way we can correctly account for a
1263 * fallocated page arriving here is now to initialize it and write it.
1265 * That's okay for a page already fallocated earlier, but if we have
1266 * not yet completed the fallocation, then (a) we want to keep track
1267 * of this page in case we have to undo it, and (b) it may not be a
1268 * good idea to continue anyway, once we're pushing into swap. So
1269 * reactivate the page, and let shmem_fallocate() quit when too many.
1271 if (!PageUptodate(page)) {
1272 if (inode->i_private) {
1273 struct shmem_falloc *shmem_falloc;
1274 spin_lock(&inode->i_lock);
1275 shmem_falloc = inode->i_private;
1277 !shmem_falloc->waitq &&
1278 index >= shmem_falloc->start &&
1279 index < shmem_falloc->next)
1280 shmem_falloc->nr_unswapped++;
1282 shmem_falloc = NULL;
1283 spin_unlock(&inode->i_lock);
1287 clear_highpage(page);
1288 flush_dcache_page(page);
1289 SetPageUptodate(page);
1292 swap = get_swap_page();
1296 if (mem_cgroup_try_charge_swap(page, swap))
1300 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1301 * if it's not already there. Do it now before the page is
1302 * moved to swap cache, when its pagelock no longer protects
1303 * the inode from eviction. But don't unlock the mutex until
1304 * we've incremented swapped, because shmem_unuse_inode() will
1305 * prune a !swapped inode from the swaplist under this mutex.
1307 mutex_lock(&shmem_swaplist_mutex);
1308 if (list_empty(&info->swaplist))
1309 list_add_tail(&info->swaplist, &shmem_swaplist);
1311 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1312 spin_lock_irq(&info->lock);
1313 shmem_recalc_inode(inode);
1315 spin_unlock_irq(&info->lock);
1317 swap_shmem_alloc(swap);
1318 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1320 mutex_unlock(&shmem_swaplist_mutex);
1321 BUG_ON(page_mapped(page));
1322 swap_writepage(page, wbc);
1326 mutex_unlock(&shmem_swaplist_mutex);
1328 swapcache_free(swap);
1330 set_page_dirty(page);
1331 if (wbc->for_reclaim)
1332 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1337 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1338 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1342 if (!mpol || mpol->mode == MPOL_DEFAULT)
1343 return; /* show nothing */
1345 mpol_to_str(buffer, sizeof(buffer), mpol);
1347 seq_printf(seq, ",mpol=%s", buffer);
1350 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1352 struct mempolicy *mpol = NULL;
1354 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1355 mpol = sbinfo->mpol;
1357 spin_unlock(&sbinfo->stat_lock);
1361 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1362 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1365 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1369 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1371 #define vm_policy vm_private_data
1374 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1375 struct shmem_inode_info *info, pgoff_t index)
1377 /* Create a pseudo vma that just contains the policy */
1379 /* Bias interleave by inode number to distribute better across nodes */
1380 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1382 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1385 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1387 /* Drop reference taken by mpol_shared_policy_lookup() */
1388 mpol_cond_put(vma->vm_policy);
1391 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1392 struct shmem_inode_info *info, pgoff_t index)
1394 struct vm_area_struct pvma;
1397 shmem_pseudo_vma_init(&pvma, info, index);
1398 page = swapin_readahead(swap, gfp, &pvma, 0);
1399 shmem_pseudo_vma_destroy(&pvma);
1404 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1405 struct shmem_inode_info *info, pgoff_t index)
1407 struct vm_area_struct pvma;
1408 struct inode *inode = &info->vfs_inode;
1409 struct address_space *mapping = inode->i_mapping;
1410 pgoff_t idx, hindex;
1411 void __rcu **results;
1414 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1417 hindex = round_down(index, HPAGE_PMD_NR);
1419 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1420 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1426 shmem_pseudo_vma_init(&pvma, info, hindex);
1427 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1428 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1429 shmem_pseudo_vma_destroy(&pvma);
1431 prep_transhuge_page(page);
1435 static struct page *shmem_alloc_page(gfp_t gfp,
1436 struct shmem_inode_info *info, pgoff_t index)
1438 struct vm_area_struct pvma;
1441 shmem_pseudo_vma_init(&pvma, info, index);
1442 page = alloc_page_vma(gfp, &pvma, 0);
1443 shmem_pseudo_vma_destroy(&pvma);
1448 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1449 struct inode *inode,
1450 pgoff_t index, bool huge)
1452 struct shmem_inode_info *info = SHMEM_I(inode);
1457 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1459 nr = huge ? HPAGE_PMD_NR : 1;
1461 if (!shmem_inode_acct_block(inode, nr))
1465 page = shmem_alloc_hugepage(gfp, info, index);
1467 page = shmem_alloc_page(gfp, info, index);
1469 __SetPageLocked(page);
1470 __SetPageSwapBacked(page);
1475 shmem_inode_unacct_blocks(inode, nr);
1477 return ERR_PTR(err);
1481 * When a page is moved from swapcache to shmem filecache (either by the
1482 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1483 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1484 * ignorance of the mapping it belongs to. If that mapping has special
1485 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1486 * we may need to copy to a suitable page before moving to filecache.
1488 * In a future release, this may well be extended to respect cpuset and
1489 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1490 * but for now it is a simple matter of zone.
1492 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1494 return page_zonenum(page) > gfp_zone(gfp);
1497 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1498 struct shmem_inode_info *info, pgoff_t index)
1500 struct page *oldpage, *newpage;
1501 struct address_space *swap_mapping;
1507 entry.val = page_private(oldpage);
1508 swap_index = swp_offset(entry);
1509 swap_mapping = page_mapping(oldpage);
1512 * We have arrived here because our zones are constrained, so don't
1513 * limit chance of success by further cpuset and node constraints.
1515 gfp &= ~GFP_CONSTRAINT_MASK;
1516 newpage = shmem_alloc_page(gfp, info, index);
1521 copy_highpage(newpage, oldpage);
1522 flush_dcache_page(newpage);
1524 __SetPageLocked(newpage);
1525 __SetPageSwapBacked(newpage);
1526 SetPageUptodate(newpage);
1527 set_page_private(newpage, entry.val);
1528 SetPageSwapCache(newpage);
1531 * Our caller will very soon move newpage out of swapcache, but it's
1532 * a nice clean interface for us to replace oldpage by newpage there.
1534 spin_lock_irq(&swap_mapping->tree_lock);
1535 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1538 __inc_node_page_state(newpage, NR_FILE_PAGES);
1539 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1541 spin_unlock_irq(&swap_mapping->tree_lock);
1543 if (unlikely(error)) {
1545 * Is this possible? I think not, now that our callers check
1546 * both PageSwapCache and page_private after getting page lock;
1547 * but be defensive. Reverse old to newpage for clear and free.
1551 mem_cgroup_migrate(oldpage, newpage);
1552 lru_cache_add_anon(newpage);
1556 ClearPageSwapCache(oldpage);
1557 set_page_private(oldpage, 0);
1559 unlock_page(oldpage);
1566 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1568 * If we allocate a new one we do not mark it dirty. That's up to the
1569 * vm. If we swap it in we mark it dirty since we also free the swap
1570 * entry since a page cannot live in both the swap and page cache.
1572 * fault_mm and fault_type are only supplied by shmem_fault:
1573 * otherwise they are NULL.
1575 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1576 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1577 struct mm_struct *fault_mm, int *fault_type)
1579 struct address_space *mapping = inode->i_mapping;
1580 struct shmem_inode_info *info = SHMEM_I(inode);
1581 struct shmem_sb_info *sbinfo;
1582 struct mm_struct *charge_mm;
1583 struct mem_cgroup *memcg;
1586 enum sgp_type sgp_huge = sgp;
1587 pgoff_t hindex = index;
1592 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1594 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1598 page = find_lock_entry(mapping, index);
1599 if (radix_tree_exceptional_entry(page)) {
1600 swap = radix_to_swp_entry(page);
1604 if (sgp <= SGP_CACHE &&
1605 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1610 if (page && sgp == SGP_WRITE)
1611 mark_page_accessed(page);
1613 /* fallocated page? */
1614 if (page && !PageUptodate(page)) {
1615 if (sgp != SGP_READ)
1621 if (page || (sgp == SGP_READ && !swap.val)) {
1627 * Fast cache lookup did not find it:
1628 * bring it back from swap or allocate.
1630 sbinfo = SHMEM_SB(inode->i_sb);
1631 charge_mm = fault_mm ? : current->mm;
1634 /* Look it up and read it in.. */
1635 page = lookup_swap_cache(swap);
1637 /* Or update major stats only when swapin succeeds?? */
1639 *fault_type |= VM_FAULT_MAJOR;
1640 count_vm_event(PGMAJFAULT);
1641 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1643 /* Here we actually start the io */
1644 page = shmem_swapin(swap, gfp, info, index);
1651 /* We have to do this with page locked to prevent races */
1653 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1654 !shmem_confirm_swap(mapping, index, swap)) {
1655 error = -EEXIST; /* try again */
1658 if (!PageUptodate(page)) {
1662 wait_on_page_writeback(page);
1664 if (shmem_should_replace_page(page, gfp)) {
1665 error = shmem_replace_page(&page, gfp, info, index);
1670 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1673 error = shmem_add_to_page_cache(page, mapping, index,
1674 swp_to_radix_entry(swap));
1676 * We already confirmed swap under page lock, and make
1677 * no memory allocation here, so usually no possibility
1678 * of error; but free_swap_and_cache() only trylocks a
1679 * page, so it is just possible that the entry has been
1680 * truncated or holepunched since swap was confirmed.
1681 * shmem_undo_range() will have done some of the
1682 * unaccounting, now delete_from_swap_cache() will do
1684 * Reset swap.val? No, leave it so "failed" goes back to
1685 * "repeat": reading a hole and writing should succeed.
1688 mem_cgroup_cancel_charge(page, memcg, false);
1689 delete_from_swap_cache(page);
1695 mem_cgroup_commit_charge(page, memcg, true, false);
1697 spin_lock_irq(&info->lock);
1699 shmem_recalc_inode(inode);
1700 spin_unlock_irq(&info->lock);
1702 if (sgp == SGP_WRITE)
1703 mark_page_accessed(page);
1705 delete_from_swap_cache(page);
1706 set_page_dirty(page);
1710 /* shmem_symlink() */
1711 if (mapping->a_ops != &shmem_aops)
1713 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1715 if (shmem_huge == SHMEM_HUGE_FORCE)
1717 switch (sbinfo->huge) {
1720 case SHMEM_HUGE_NEVER:
1722 case SHMEM_HUGE_WITHIN_SIZE:
1723 off = round_up(index, HPAGE_PMD_NR);
1724 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1725 if (i_size >= HPAGE_PMD_SIZE &&
1726 i_size >> PAGE_SHIFT >= off)
1729 case SHMEM_HUGE_ADVISE:
1730 if (sgp_huge == SGP_HUGE)
1732 /* TODO: implement fadvise() hints */
1737 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1739 alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode,
1744 error = PTR_ERR(page);
1746 if (error != -ENOSPC)
1749 * Try to reclaim some spece by splitting a huge page
1750 * beyond i_size on the filesystem.
1754 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1755 if (ret == SHRINK_STOP)
1763 if (PageTransHuge(page))
1764 hindex = round_down(index, HPAGE_PMD_NR);
1768 if (sgp == SGP_WRITE)
1769 __SetPageReferenced(page);
1771 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1772 PageTransHuge(page));
1775 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1776 compound_order(page));
1778 error = shmem_add_to_page_cache(page, mapping, hindex,
1780 radix_tree_preload_end();
1783 mem_cgroup_cancel_charge(page, memcg,
1784 PageTransHuge(page));
1787 mem_cgroup_commit_charge(page, memcg, false,
1788 PageTransHuge(page));
1789 lru_cache_add_anon(page);
1791 spin_lock_irq(&info->lock);
1792 info->alloced += 1 << compound_order(page);
1793 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1794 shmem_recalc_inode(inode);
1795 spin_unlock_irq(&info->lock);
1798 if (PageTransHuge(page) &&
1799 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1800 hindex + HPAGE_PMD_NR - 1) {
1802 * Part of the huge page is beyond i_size: subject
1803 * to shrink under memory pressure.
1805 spin_lock(&sbinfo->shrinklist_lock);
1807 * _careful to defend against unlocked access to
1808 * ->shrink_list in shmem_unused_huge_shrink()
1810 if (list_empty_careful(&info->shrinklist)) {
1811 list_add_tail(&info->shrinklist,
1812 &sbinfo->shrinklist);
1813 sbinfo->shrinklist_len++;
1815 spin_unlock(&sbinfo->shrinklist_lock);
1819 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1821 if (sgp == SGP_FALLOC)
1825 * Let SGP_WRITE caller clear ends if write does not fill page;
1826 * but SGP_FALLOC on a page fallocated earlier must initialize
1827 * it now, lest undo on failure cancel our earlier guarantee.
1829 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1830 struct page *head = compound_head(page);
1833 for (i = 0; i < (1 << compound_order(head)); i++) {
1834 clear_highpage(head + i);
1835 flush_dcache_page(head + i);
1837 SetPageUptodate(head);
1841 /* Perhaps the file has been truncated since we checked */
1842 if (sgp <= SGP_CACHE &&
1843 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1845 ClearPageDirty(page);
1846 delete_from_page_cache(page);
1847 spin_lock_irq(&info->lock);
1848 shmem_recalc_inode(inode);
1849 spin_unlock_irq(&info->lock);
1854 *pagep = page + index - hindex;
1861 shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
1863 if (PageTransHuge(page)) {
1869 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1876 if (error == -ENOSPC && !once++) {
1877 spin_lock_irq(&info->lock);
1878 shmem_recalc_inode(inode);
1879 spin_unlock_irq(&info->lock);
1882 if (error == -EEXIST) /* from above or from radix_tree_insert */
1888 * This is like autoremove_wake_function, but it removes the wait queue
1889 * entry unconditionally - even if something else had already woken the
1892 static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1894 int ret = default_wake_function(wait, mode, sync, key);
1895 list_del_init(&wait->task_list);
1899 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1901 struct inode *inode = file_inode(vma->vm_file);
1902 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1905 int ret = VM_FAULT_LOCKED;
1908 * Trinity finds that probing a hole which tmpfs is punching can
1909 * prevent the hole-punch from ever completing: which in turn
1910 * locks writers out with its hold on i_mutex. So refrain from
1911 * faulting pages into the hole while it's being punched. Although
1912 * shmem_undo_range() does remove the additions, it may be unable to
1913 * keep up, as each new page needs its own unmap_mapping_range() call,
1914 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1916 * It does not matter if we sometimes reach this check just before the
1917 * hole-punch begins, so that one fault then races with the punch:
1918 * we just need to make racing faults a rare case.
1920 * The implementation below would be much simpler if we just used a
1921 * standard mutex or completion: but we cannot take i_mutex in fault,
1922 * and bloating every shmem inode for this unlikely case would be sad.
1924 if (unlikely(inode->i_private)) {
1925 struct shmem_falloc *shmem_falloc;
1927 spin_lock(&inode->i_lock);
1928 shmem_falloc = inode->i_private;
1930 shmem_falloc->waitq &&
1931 vmf->pgoff >= shmem_falloc->start &&
1932 vmf->pgoff < shmem_falloc->next) {
1933 wait_queue_head_t *shmem_falloc_waitq;
1934 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1936 ret = VM_FAULT_NOPAGE;
1937 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1938 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1939 /* It's polite to up mmap_sem if we can */
1940 up_read(&vma->vm_mm->mmap_sem);
1941 ret = VM_FAULT_RETRY;
1944 shmem_falloc_waitq = shmem_falloc->waitq;
1945 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1946 TASK_UNINTERRUPTIBLE);
1947 spin_unlock(&inode->i_lock);
1951 * shmem_falloc_waitq points into the shmem_fallocate()
1952 * stack of the hole-punching task: shmem_falloc_waitq
1953 * is usually invalid by the time we reach here, but
1954 * finish_wait() does not dereference it in that case;
1955 * though i_lock needed lest racing with wake_up_all().
1957 spin_lock(&inode->i_lock);
1958 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1959 spin_unlock(&inode->i_lock);
1962 spin_unlock(&inode->i_lock);
1966 if (vma->vm_flags & VM_HUGEPAGE)
1968 else if (vma->vm_flags & VM_NOHUGEPAGE)
1971 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1972 gfp, vma->vm_mm, &ret);
1974 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1978 unsigned long shmem_get_unmapped_area(struct file *file,
1979 unsigned long uaddr, unsigned long len,
1980 unsigned long pgoff, unsigned long flags)
1982 unsigned long (*get_area)(struct file *,
1983 unsigned long, unsigned long, unsigned long, unsigned long);
1985 unsigned long offset;
1986 unsigned long inflated_len;
1987 unsigned long inflated_addr;
1988 unsigned long inflated_offset;
1990 if (len > TASK_SIZE)
1993 get_area = current->mm->get_unmapped_area;
1994 addr = get_area(file, uaddr, len, pgoff, flags);
1996 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1998 if (IS_ERR_VALUE(addr))
2000 if (addr & ~PAGE_MASK)
2002 if (addr > TASK_SIZE - len)
2005 if (shmem_huge == SHMEM_HUGE_DENY)
2007 if (len < HPAGE_PMD_SIZE)
2009 if (flags & MAP_FIXED)
2012 * Our priority is to support MAP_SHARED mapped hugely;
2013 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2014 * But if caller specified an address hint, respect that as before.
2019 if (shmem_huge != SHMEM_HUGE_FORCE) {
2020 struct super_block *sb;
2023 VM_BUG_ON(file->f_op != &shmem_file_operations);
2024 sb = file_inode(file)->i_sb;
2027 * Called directly from mm/mmap.c, or drivers/char/mem.c
2028 * for "/dev/zero", to create a shared anonymous object.
2030 if (IS_ERR(shm_mnt))
2032 sb = shm_mnt->mnt_sb;
2034 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2038 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2039 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2041 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2044 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2045 if (inflated_len > TASK_SIZE)
2047 if (inflated_len < len)
2050 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2051 if (IS_ERR_VALUE(inflated_addr))
2053 if (inflated_addr & ~PAGE_MASK)
2056 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2057 inflated_addr += offset - inflated_offset;
2058 if (inflated_offset > offset)
2059 inflated_addr += HPAGE_PMD_SIZE;
2061 if (inflated_addr > TASK_SIZE - len)
2063 return inflated_addr;
2067 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2069 struct inode *inode = file_inode(vma->vm_file);
2070 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2073 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2076 struct inode *inode = file_inode(vma->vm_file);
2079 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2080 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2084 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2086 struct inode *inode = file_inode(file);
2087 struct shmem_inode_info *info = SHMEM_I(inode);
2088 int retval = -ENOMEM;
2091 * What serializes the accesses to info->flags?
2092 * ipc_lock_object() when called from shmctl_do_lock(),
2093 * no serialization needed when called from shm_destroy().
2095 if (lock && !(info->flags & VM_LOCKED)) {
2096 if (!user_shm_lock(inode->i_size, user))
2098 info->flags |= VM_LOCKED;
2099 mapping_set_unevictable(file->f_mapping);
2101 if (!lock && (info->flags & VM_LOCKED) && user) {
2102 user_shm_unlock(inode->i_size, user);
2103 info->flags &= ~VM_LOCKED;
2104 mapping_clear_unevictable(file->f_mapping);
2112 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2114 file_accessed(file);
2115 vma->vm_ops = &shmem_vm_ops;
2116 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2117 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2118 (vma->vm_end & HPAGE_PMD_MASK)) {
2119 khugepaged_enter(vma, vma->vm_flags);
2124 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2125 umode_t mode, dev_t dev, unsigned long flags)
2127 struct inode *inode;
2128 struct shmem_inode_info *info;
2129 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2131 if (shmem_reserve_inode(sb))
2134 inode = new_inode(sb);
2136 inode->i_ino = get_next_ino();
2137 inode_init_owner(inode, dir, mode);
2138 inode->i_blocks = 0;
2139 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2140 inode->i_generation = get_seconds();
2141 info = SHMEM_I(inode);
2142 memset(info, 0, (char *)inode - (char *)info);
2143 spin_lock_init(&info->lock);
2144 info->seals = F_SEAL_SEAL;
2145 info->flags = flags & VM_NORESERVE;
2146 INIT_LIST_HEAD(&info->shrinklist);
2147 INIT_LIST_HEAD(&info->swaplist);
2148 simple_xattrs_init(&info->xattrs);
2149 cache_no_acl(inode);
2151 switch (mode & S_IFMT) {
2153 inode->i_op = &shmem_special_inode_operations;
2154 init_special_inode(inode, mode, dev);
2157 inode->i_mapping->a_ops = &shmem_aops;
2158 inode->i_op = &shmem_inode_operations;
2159 inode->i_fop = &shmem_file_operations;
2160 mpol_shared_policy_init(&info->policy,
2161 shmem_get_sbmpol(sbinfo));
2165 /* Some things misbehave if size == 0 on a directory */
2166 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2167 inode->i_op = &shmem_dir_inode_operations;
2168 inode->i_fop = &simple_dir_operations;
2172 * Must not load anything in the rbtree,
2173 * mpol_free_shared_policy will not be called.
2175 mpol_shared_policy_init(&info->policy, NULL);
2179 lockdep_annotate_inode_mutex_key(inode);
2181 shmem_free_inode(sb);
2185 bool shmem_mapping(struct address_space *mapping)
2190 return mapping->host->i_sb->s_op == &shmem_ops;
2194 static const struct inode_operations shmem_symlink_inode_operations;
2195 static const struct inode_operations shmem_short_symlink_operations;
2197 #ifdef CONFIG_TMPFS_XATTR
2198 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2200 #define shmem_initxattrs NULL
2204 shmem_write_begin(struct file *file, struct address_space *mapping,
2205 loff_t pos, unsigned len, unsigned flags,
2206 struct page **pagep, void **fsdata)
2208 struct inode *inode = mapping->host;
2209 struct shmem_inode_info *info = SHMEM_I(inode);
2210 pgoff_t index = pos >> PAGE_SHIFT;
2212 /* i_mutex is held by caller */
2213 if (unlikely(info->seals)) {
2214 if (info->seals & F_SEAL_WRITE)
2216 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2220 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2224 shmem_write_end(struct file *file, struct address_space *mapping,
2225 loff_t pos, unsigned len, unsigned copied,
2226 struct page *page, void *fsdata)
2228 struct inode *inode = mapping->host;
2230 if (pos + copied > inode->i_size)
2231 i_size_write(inode, pos + copied);
2233 if (!PageUptodate(page)) {
2234 struct page *head = compound_head(page);
2235 if (PageTransCompound(page)) {
2238 for (i = 0; i < HPAGE_PMD_NR; i++) {
2239 if (head + i == page)
2241 clear_highpage(head + i);
2242 flush_dcache_page(head + i);
2245 if (copied < PAGE_SIZE) {
2246 unsigned from = pos & (PAGE_SIZE - 1);
2247 zero_user_segments(page, 0, from,
2248 from + copied, PAGE_SIZE);
2250 SetPageUptodate(head);
2252 set_page_dirty(page);
2259 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2261 struct file *file = iocb->ki_filp;
2262 struct inode *inode = file_inode(file);
2263 struct address_space *mapping = inode->i_mapping;
2265 unsigned long offset;
2266 enum sgp_type sgp = SGP_READ;
2269 loff_t *ppos = &iocb->ki_pos;
2272 * Might this read be for a stacking filesystem? Then when reading
2273 * holes of a sparse file, we actually need to allocate those pages,
2274 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2276 if (!iter_is_iovec(to))
2279 index = *ppos >> PAGE_SHIFT;
2280 offset = *ppos & ~PAGE_MASK;
2283 struct page *page = NULL;
2285 unsigned long nr, ret;
2286 loff_t i_size = i_size_read(inode);
2288 end_index = i_size >> PAGE_SHIFT;
2289 if (index > end_index)
2291 if (index == end_index) {
2292 nr = i_size & ~PAGE_MASK;
2297 error = shmem_getpage(inode, index, &page, sgp);
2299 if (error == -EINVAL)
2304 if (sgp == SGP_CACHE)
2305 set_page_dirty(page);
2310 * We must evaluate after, since reads (unlike writes)
2311 * are called without i_mutex protection against truncate
2314 i_size = i_size_read(inode);
2315 end_index = i_size >> PAGE_SHIFT;
2316 if (index == end_index) {
2317 nr = i_size & ~PAGE_MASK;
2328 * If users can be writing to this page using arbitrary
2329 * virtual addresses, take care about potential aliasing
2330 * before reading the page on the kernel side.
2332 if (mapping_writably_mapped(mapping))
2333 flush_dcache_page(page);
2335 * Mark the page accessed if we read the beginning.
2338 mark_page_accessed(page);
2340 page = ZERO_PAGE(0);
2345 * Ok, we have the page, and it's up-to-date, so
2346 * now we can copy it to user space...
2348 ret = copy_page_to_iter(page, offset, nr, to);
2351 index += offset >> PAGE_SHIFT;
2352 offset &= ~PAGE_MASK;
2355 if (!iov_iter_count(to))
2364 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2365 file_accessed(file);
2366 return retval ? retval : error;
2370 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2372 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2373 pgoff_t index, pgoff_t end, int whence)
2376 struct pagevec pvec;
2377 pgoff_t indices[PAGEVEC_SIZE];
2381 pagevec_init(&pvec, 0);
2382 pvec.nr = 1; /* start small: we may be there already */
2384 pvec.nr = find_get_entries(mapping, index,
2385 pvec.nr, pvec.pages, indices);
2387 if (whence == SEEK_DATA)
2391 for (i = 0; i < pvec.nr; i++, index++) {
2392 if (index < indices[i]) {
2393 if (whence == SEEK_HOLE) {
2399 page = pvec.pages[i];
2400 if (page && !radix_tree_exceptional_entry(page)) {
2401 if (!PageUptodate(page))
2405 (page && whence == SEEK_DATA) ||
2406 (!page && whence == SEEK_HOLE)) {
2411 pagevec_remove_exceptionals(&pvec);
2412 pagevec_release(&pvec);
2413 pvec.nr = PAGEVEC_SIZE;
2419 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2421 struct address_space *mapping = file->f_mapping;
2422 struct inode *inode = mapping->host;
2426 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2427 return generic_file_llseek_size(file, offset, whence,
2428 MAX_LFS_FILESIZE, i_size_read(inode));
2430 /* We're holding i_mutex so we can access i_size directly */
2432 if (offset < 0 || offset >= inode->i_size)
2435 start = offset >> PAGE_SHIFT;
2436 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2437 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2438 new_offset <<= PAGE_SHIFT;
2439 if (new_offset > offset) {
2440 if (new_offset < inode->i_size)
2441 offset = new_offset;
2442 else if (whence == SEEK_DATA)
2445 offset = inode->i_size;
2450 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2451 inode_unlock(inode);
2456 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2457 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2459 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2460 #define LAST_SCAN 4 /* about 150ms max */
2462 static void shmem_tag_pins(struct address_space *mapping)
2464 struct radix_tree_iter iter;
2468 unsigned int tagged = 0;
2473 spin_lock_irq(&mapping->tree_lock);
2474 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2475 page = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
2476 if (!page || radix_tree_exception(page)) {
2477 if (radix_tree_deref_retry(page)) {
2478 slot = radix_tree_iter_retry(&iter);
2481 } else if (page_count(page) - page_mapcount(page) > 1) {
2482 radix_tree_tag_set(&mapping->page_tree, iter.index,
2486 if (++tagged % 1024)
2489 slot = radix_tree_iter_next(&iter);
2490 spin_unlock_irq(&mapping->tree_lock);
2492 spin_lock_irq(&mapping->tree_lock);
2494 spin_unlock_irq(&mapping->tree_lock);
2498 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2499 * via get_user_pages(), drivers might have some pending I/O without any active
2500 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2501 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2502 * them to be dropped.
2503 * The caller must guarantee that no new user will acquire writable references
2504 * to those pages to avoid races.
2506 static int shmem_wait_for_pins(struct address_space *mapping)
2508 struct radix_tree_iter iter;
2514 shmem_tag_pins(mapping);
2517 for (scan = 0; scan <= LAST_SCAN; scan++) {
2518 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2522 lru_add_drain_all();
2523 else if (schedule_timeout_killable((HZ << scan) / 200))
2528 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2529 start, SHMEM_TAG_PINNED) {
2531 page = radix_tree_deref_slot(slot);
2532 if (radix_tree_exception(page)) {
2533 if (radix_tree_deref_retry(page)) {
2534 slot = radix_tree_iter_retry(&iter);
2542 page_count(page) - page_mapcount(page) != 1) {
2543 if (scan < LAST_SCAN)
2544 goto continue_resched;
2547 * On the last scan, we clean up all those tags
2548 * we inserted; but make a note that we still
2549 * found pages pinned.
2554 spin_lock_irq(&mapping->tree_lock);
2555 radix_tree_tag_clear(&mapping->page_tree,
2556 iter.index, SHMEM_TAG_PINNED);
2557 spin_unlock_irq(&mapping->tree_lock);
2559 if (need_resched()) {
2561 slot = radix_tree_iter_next(&iter);
2570 #define F_ALL_SEALS (F_SEAL_SEAL | \
2575 int shmem_add_seals(struct file *file, unsigned int seals)
2577 struct inode *inode = file_inode(file);
2578 struct shmem_inode_info *info = SHMEM_I(inode);
2583 * Sealing allows multiple parties to share a shmem-file but restrict
2584 * access to a specific subset of file operations. Seals can only be
2585 * added, but never removed. This way, mutually untrusted parties can
2586 * share common memory regions with a well-defined policy. A malicious
2587 * peer can thus never perform unwanted operations on a shared object.
2589 * Seals are only supported on special shmem-files and always affect
2590 * the whole underlying inode. Once a seal is set, it may prevent some
2591 * kinds of access to the file. Currently, the following seals are
2593 * SEAL_SEAL: Prevent further seals from being set on this file
2594 * SEAL_SHRINK: Prevent the file from shrinking
2595 * SEAL_GROW: Prevent the file from growing
2596 * SEAL_WRITE: Prevent write access to the file
2598 * As we don't require any trust relationship between two parties, we
2599 * must prevent seals from being removed. Therefore, sealing a file
2600 * only adds a given set of seals to the file, it never touches
2601 * existing seals. Furthermore, the "setting seals"-operation can be
2602 * sealed itself, which basically prevents any further seal from being
2605 * Semantics of sealing are only defined on volatile files. Only
2606 * anonymous shmem files support sealing. More importantly, seals are
2607 * never written to disk. Therefore, there's no plan to support it on
2611 if (file->f_op != &shmem_file_operations)
2613 if (!(file->f_mode & FMODE_WRITE))
2615 if (seals & ~(unsigned int)F_ALL_SEALS)
2620 if (info->seals & F_SEAL_SEAL) {
2625 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2626 error = mapping_deny_writable(file->f_mapping);
2630 error = shmem_wait_for_pins(file->f_mapping);
2632 mapping_allow_writable(file->f_mapping);
2637 info->seals |= seals;
2641 inode_unlock(inode);
2644 EXPORT_SYMBOL_GPL(shmem_add_seals);
2646 int shmem_get_seals(struct file *file)
2648 if (file->f_op != &shmem_file_operations)
2651 return SHMEM_I(file_inode(file))->seals;
2653 EXPORT_SYMBOL_GPL(shmem_get_seals);
2655 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2661 /* disallow upper 32bit */
2665 error = shmem_add_seals(file, arg);
2668 error = shmem_get_seals(file);
2678 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2681 struct inode *inode = file_inode(file);
2682 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2683 struct shmem_inode_info *info = SHMEM_I(inode);
2684 struct shmem_falloc shmem_falloc;
2685 pgoff_t start, index, end;
2688 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2693 if (mode & FALLOC_FL_PUNCH_HOLE) {
2694 struct address_space *mapping = file->f_mapping;
2695 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2696 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2697 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2699 /* protected by i_mutex */
2700 if (info->seals & F_SEAL_WRITE) {
2705 shmem_falloc.waitq = &shmem_falloc_waitq;
2706 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2707 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2708 spin_lock(&inode->i_lock);
2709 inode->i_private = &shmem_falloc;
2710 spin_unlock(&inode->i_lock);
2712 if ((u64)unmap_end > (u64)unmap_start)
2713 unmap_mapping_range(mapping, unmap_start,
2714 1 + unmap_end - unmap_start, 0);
2715 shmem_truncate_range(inode, offset, offset + len - 1);
2716 /* No need to unmap again: hole-punching leaves COWed pages */
2718 spin_lock(&inode->i_lock);
2719 inode->i_private = NULL;
2720 wake_up_all(&shmem_falloc_waitq);
2721 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
2722 spin_unlock(&inode->i_lock);
2727 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2728 error = inode_newsize_ok(inode, offset + len);
2732 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2737 start = offset >> PAGE_SHIFT;
2738 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2739 /* Try to avoid a swapstorm if len is impossible to satisfy */
2740 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2745 shmem_falloc.waitq = NULL;
2746 shmem_falloc.start = start;
2747 shmem_falloc.next = start;
2748 shmem_falloc.nr_falloced = 0;
2749 shmem_falloc.nr_unswapped = 0;
2750 spin_lock(&inode->i_lock);
2751 inode->i_private = &shmem_falloc;
2752 spin_unlock(&inode->i_lock);
2754 for (index = start; index < end; index++) {
2758 * Good, the fallocate(2) manpage permits EINTR: we may have
2759 * been interrupted because we are using up too much memory.
2761 if (signal_pending(current))
2763 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2766 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2768 /* Remove the !PageUptodate pages we added */
2769 if (index > start) {
2770 shmem_undo_range(inode,
2771 (loff_t)start << PAGE_SHIFT,
2772 ((loff_t)index << PAGE_SHIFT) - 1, true);
2778 * Inform shmem_writepage() how far we have reached.
2779 * No need for lock or barrier: we have the page lock.
2781 shmem_falloc.next++;
2782 if (!PageUptodate(page))
2783 shmem_falloc.nr_falloced++;
2786 * If !PageUptodate, leave it that way so that freeable pages
2787 * can be recognized if we need to rollback on error later.
2788 * But set_page_dirty so that memory pressure will swap rather
2789 * than free the pages we are allocating (and SGP_CACHE pages
2790 * might still be clean: we now need to mark those dirty too).
2792 set_page_dirty(page);
2798 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2799 i_size_write(inode, offset + len);
2800 inode->i_ctime = current_time(inode);
2802 spin_lock(&inode->i_lock);
2803 inode->i_private = NULL;
2804 spin_unlock(&inode->i_lock);
2806 inode_unlock(inode);
2810 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2812 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2814 buf->f_type = TMPFS_MAGIC;
2815 buf->f_bsize = PAGE_SIZE;
2816 buf->f_namelen = NAME_MAX;
2817 if (sbinfo->max_blocks) {
2818 buf->f_blocks = sbinfo->max_blocks;
2820 buf->f_bfree = sbinfo->max_blocks -
2821 percpu_counter_sum(&sbinfo->used_blocks);
2823 if (sbinfo->max_inodes) {
2824 buf->f_files = sbinfo->max_inodes;
2825 buf->f_ffree = sbinfo->free_inodes;
2827 /* else leave those fields 0 like simple_statfs */
2832 * File creation. Allocate an inode, and we're done..
2835 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2837 struct inode *inode;
2838 int error = -ENOSPC;
2840 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2842 error = simple_acl_create(dir, inode);
2845 error = security_inode_init_security(inode, dir,
2847 shmem_initxattrs, NULL);
2848 if (error && error != -EOPNOTSUPP)
2852 dir->i_size += BOGO_DIRENT_SIZE;
2853 dir->i_ctime = dir->i_mtime = current_time(dir);
2854 d_instantiate(dentry, inode);
2855 dget(dentry); /* Extra count - pin the dentry in core */
2864 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2866 struct inode *inode;
2867 int error = -ENOSPC;
2869 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2871 error = security_inode_init_security(inode, dir,
2873 shmem_initxattrs, NULL);
2874 if (error && error != -EOPNOTSUPP)
2876 error = simple_acl_create(dir, inode);
2879 d_tmpfile(dentry, inode);
2887 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2891 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2897 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2900 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2906 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2908 struct inode *inode = d_inode(old_dentry);
2912 * No ordinary (disk based) filesystem counts links as inodes;
2913 * but each new link needs a new dentry, pinning lowmem, and
2914 * tmpfs dentries cannot be pruned until they are unlinked.
2915 * But if an O_TMPFILE file is linked into the tmpfs, the
2916 * first link must skip that, to get the accounting right.
2918 if (inode->i_nlink) {
2919 ret = shmem_reserve_inode(inode->i_sb);
2924 dir->i_size += BOGO_DIRENT_SIZE;
2925 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2927 ihold(inode); /* New dentry reference */
2928 dget(dentry); /* Extra pinning count for the created dentry */
2929 d_instantiate(dentry, inode);
2934 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2936 struct inode *inode = d_inode(dentry);
2938 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2939 shmem_free_inode(inode->i_sb);
2941 dir->i_size -= BOGO_DIRENT_SIZE;
2942 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2944 dput(dentry); /* Undo the count from "create" - this does all the work */
2948 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2950 if (!simple_empty(dentry))
2953 drop_nlink(d_inode(dentry));
2955 return shmem_unlink(dir, dentry);
2958 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2960 bool old_is_dir = d_is_dir(old_dentry);
2961 bool new_is_dir = d_is_dir(new_dentry);
2963 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2965 drop_nlink(old_dir);
2968 drop_nlink(new_dir);
2972 old_dir->i_ctime = old_dir->i_mtime =
2973 new_dir->i_ctime = new_dir->i_mtime =
2974 d_inode(old_dentry)->i_ctime =
2975 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2980 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2982 struct dentry *whiteout;
2985 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2989 error = shmem_mknod(old_dir, whiteout,
2990 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2996 * Cheat and hash the whiteout while the old dentry is still in
2997 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2999 * d_lookup() will consistently find one of them at this point,
3000 * not sure which one, but that isn't even important.
3007 * The VFS layer already does all the dentry stuff for rename,
3008 * we just have to decrement the usage count for the target if
3009 * it exists so that the VFS layer correctly free's it when it
3012 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3014 struct inode *inode = d_inode(old_dentry);
3015 int they_are_dirs = S_ISDIR(inode->i_mode);
3017 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3020 if (flags & RENAME_EXCHANGE)
3021 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3023 if (!simple_empty(new_dentry))
3026 if (flags & RENAME_WHITEOUT) {
3029 error = shmem_whiteout(old_dir, old_dentry);
3034 if (d_really_is_positive(new_dentry)) {
3035 (void) shmem_unlink(new_dir, new_dentry);
3036 if (they_are_dirs) {
3037 drop_nlink(d_inode(new_dentry));
3038 drop_nlink(old_dir);
3040 } else if (they_are_dirs) {
3041 drop_nlink(old_dir);
3045 old_dir->i_size -= BOGO_DIRENT_SIZE;
3046 new_dir->i_size += BOGO_DIRENT_SIZE;
3047 old_dir->i_ctime = old_dir->i_mtime =
3048 new_dir->i_ctime = new_dir->i_mtime =
3049 inode->i_ctime = current_time(old_dir);
3053 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3057 struct inode *inode;
3059 struct shmem_inode_info *info;
3061 len = strlen(symname) + 1;
3062 if (len > PAGE_SIZE)
3063 return -ENAMETOOLONG;
3065 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3069 error = security_inode_init_security(inode, dir, &dentry->d_name,
3070 shmem_initxattrs, NULL);
3072 if (error != -EOPNOTSUPP) {
3079 info = SHMEM_I(inode);
3080 inode->i_size = len-1;
3081 if (len <= SHORT_SYMLINK_LEN) {
3082 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3083 if (!inode->i_link) {
3087 inode->i_op = &shmem_short_symlink_operations;
3089 inode_nohighmem(inode);
3090 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3095 inode->i_mapping->a_ops = &shmem_aops;
3096 inode->i_op = &shmem_symlink_inode_operations;
3097 memcpy(page_address(page), symname, len);
3098 SetPageUptodate(page);
3099 set_page_dirty(page);
3103 dir->i_size += BOGO_DIRENT_SIZE;
3104 dir->i_ctime = dir->i_mtime = current_time(dir);
3105 d_instantiate(dentry, inode);
3110 static void shmem_put_link(void *arg)
3112 mark_page_accessed(arg);
3116 static const char *shmem_get_link(struct dentry *dentry,
3117 struct inode *inode,
3118 struct delayed_call *done)
3120 struct page *page = NULL;
3123 page = find_get_page(inode->i_mapping, 0);
3125 return ERR_PTR(-ECHILD);
3126 if (!PageUptodate(page)) {
3128 return ERR_PTR(-ECHILD);
3131 error = shmem_getpage(inode, 0, &page, SGP_READ);
3133 return ERR_PTR(error);
3136 set_delayed_call(done, shmem_put_link, page);
3137 return page_address(page);
3140 #ifdef CONFIG_TMPFS_XATTR
3142 * Superblocks without xattr inode operations may get some security.* xattr
3143 * support from the LSM "for free". As soon as we have any other xattrs
3144 * like ACLs, we also need to implement the security.* handlers at
3145 * filesystem level, though.
3149 * Callback for security_inode_init_security() for acquiring xattrs.
3151 static int shmem_initxattrs(struct inode *inode,
3152 const struct xattr *xattr_array,
3155 struct shmem_inode_info *info = SHMEM_I(inode);
3156 const struct xattr *xattr;
3157 struct simple_xattr *new_xattr;
3160 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3161 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3165 len = strlen(xattr->name) + 1;
3166 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3168 if (!new_xattr->name) {
3173 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3174 XATTR_SECURITY_PREFIX_LEN);
3175 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3178 simple_xattr_list_add(&info->xattrs, new_xattr);
3184 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3185 struct dentry *unused, struct inode *inode,
3186 const char *name, void *buffer, size_t size)
3188 struct shmem_inode_info *info = SHMEM_I(inode);
3190 name = xattr_full_name(handler, name);
3191 return simple_xattr_get(&info->xattrs, name, buffer, size);
3194 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3195 struct dentry *unused, struct inode *inode,
3196 const char *name, const void *value,
3197 size_t size, int flags)
3199 struct shmem_inode_info *info = SHMEM_I(inode);
3201 name = xattr_full_name(handler, name);
3202 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3205 static const struct xattr_handler shmem_security_xattr_handler = {
3206 .prefix = XATTR_SECURITY_PREFIX,
3207 .get = shmem_xattr_handler_get,
3208 .set = shmem_xattr_handler_set,
3211 static const struct xattr_handler shmem_trusted_xattr_handler = {
3212 .prefix = XATTR_TRUSTED_PREFIX,
3213 .get = shmem_xattr_handler_get,
3214 .set = shmem_xattr_handler_set,
3217 static const struct xattr_handler *shmem_xattr_handlers[] = {
3218 #ifdef CONFIG_TMPFS_POSIX_ACL
3219 &posix_acl_access_xattr_handler,
3220 &posix_acl_default_xattr_handler,
3222 &shmem_security_xattr_handler,
3223 &shmem_trusted_xattr_handler,
3227 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3229 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3230 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3232 #endif /* CONFIG_TMPFS_XATTR */
3234 static const struct inode_operations shmem_short_symlink_operations = {
3235 .readlink = generic_readlink,
3236 .get_link = simple_get_link,
3237 #ifdef CONFIG_TMPFS_XATTR
3238 .listxattr = shmem_listxattr,
3242 static const struct inode_operations shmem_symlink_inode_operations = {
3243 .readlink = generic_readlink,
3244 .get_link = shmem_get_link,
3245 #ifdef CONFIG_TMPFS_XATTR
3246 .listxattr = shmem_listxattr,
3250 static struct dentry *shmem_get_parent(struct dentry *child)
3252 return ERR_PTR(-ESTALE);
3255 static int shmem_match(struct inode *ino, void *vfh)
3259 inum = (inum << 32) | fh[1];
3260 return ino->i_ino == inum && fh[0] == ino->i_generation;
3263 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3264 struct fid *fid, int fh_len, int fh_type)
3266 struct inode *inode;
3267 struct dentry *dentry = NULL;
3274 inum = (inum << 32) | fid->raw[1];
3276 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3277 shmem_match, fid->raw);
3279 dentry = d_find_alias(inode);
3286 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3287 struct inode *parent)
3291 return FILEID_INVALID;
3294 if (inode_unhashed(inode)) {
3295 /* Unfortunately insert_inode_hash is not idempotent,
3296 * so as we hash inodes here rather than at creation
3297 * time, we need a lock to ensure we only try
3300 static DEFINE_SPINLOCK(lock);
3302 if (inode_unhashed(inode))
3303 __insert_inode_hash(inode,
3304 inode->i_ino + inode->i_generation);
3308 fh[0] = inode->i_generation;
3309 fh[1] = inode->i_ino;
3310 fh[2] = ((__u64)inode->i_ino) >> 32;
3316 static const struct export_operations shmem_export_ops = {
3317 .get_parent = shmem_get_parent,
3318 .encode_fh = shmem_encode_fh,
3319 .fh_to_dentry = shmem_fh_to_dentry,
3322 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3325 char *this_char, *value, *rest;
3326 struct mempolicy *mpol = NULL;
3330 while (options != NULL) {
3331 this_char = options;
3334 * NUL-terminate this option: unfortunately,
3335 * mount options form a comma-separated list,
3336 * but mpol's nodelist may also contain commas.
3338 options = strchr(options, ',');
3339 if (options == NULL)
3342 if (!isdigit(*options)) {
3349 if ((value = strchr(this_char,'=')) != NULL) {
3352 pr_err("tmpfs: No value for mount option '%s'\n",
3357 if (!strcmp(this_char,"size")) {
3358 unsigned long long size;
3359 size = memparse(value,&rest);
3361 size <<= PAGE_SHIFT;
3362 size *= totalram_pages;
3368 sbinfo->max_blocks =
3369 DIV_ROUND_UP(size, PAGE_SIZE);
3370 } else if (!strcmp(this_char,"nr_blocks")) {
3371 sbinfo->max_blocks = memparse(value, &rest);
3374 } else if (!strcmp(this_char,"nr_inodes")) {
3375 sbinfo->max_inodes = memparse(value, &rest);
3378 } else if (!strcmp(this_char,"mode")) {
3381 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3384 } else if (!strcmp(this_char,"uid")) {
3387 uid = simple_strtoul(value, &rest, 0);
3390 sbinfo->uid = make_kuid(current_user_ns(), uid);
3391 if (!uid_valid(sbinfo->uid))
3393 } else if (!strcmp(this_char,"gid")) {
3396 gid = simple_strtoul(value, &rest, 0);
3399 sbinfo->gid = make_kgid(current_user_ns(), gid);
3400 if (!gid_valid(sbinfo->gid))
3402 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3403 } else if (!strcmp(this_char, "huge")) {
3405 huge = shmem_parse_huge(value);
3408 if (!has_transparent_hugepage() &&
3409 huge != SHMEM_HUGE_NEVER)
3411 sbinfo->huge = huge;
3414 } else if (!strcmp(this_char,"mpol")) {
3417 if (mpol_parse_str(value, &mpol))
3421 pr_err("tmpfs: Bad mount option %s\n", this_char);
3425 sbinfo->mpol = mpol;
3429 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3437 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3439 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3440 struct shmem_sb_info config = *sbinfo;
3441 unsigned long inodes;
3442 int error = -EINVAL;
3445 if (shmem_parse_options(data, &config, true))
3448 spin_lock(&sbinfo->stat_lock);
3449 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3450 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3452 if (config.max_inodes < inodes)
3455 * Those tests disallow limited->unlimited while any are in use;
3456 * but we must separately disallow unlimited->limited, because
3457 * in that case we have no record of how much is already in use.
3459 if (config.max_blocks && !sbinfo->max_blocks)
3461 if (config.max_inodes && !sbinfo->max_inodes)
3465 sbinfo->huge = config.huge;
3466 sbinfo->max_blocks = config.max_blocks;
3467 sbinfo->max_inodes = config.max_inodes;
3468 sbinfo->free_inodes = config.max_inodes - inodes;
3471 * Preserve previous mempolicy unless mpol remount option was specified.
3474 mpol_put(sbinfo->mpol);
3475 sbinfo->mpol = config.mpol; /* transfers initial ref */
3478 spin_unlock(&sbinfo->stat_lock);
3482 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3484 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3486 if (sbinfo->max_blocks != shmem_default_max_blocks())
3487 seq_printf(seq, ",size=%luk",
3488 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3489 if (sbinfo->max_inodes != shmem_default_max_inodes())
3490 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3491 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3492 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3493 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3494 seq_printf(seq, ",uid=%u",
3495 from_kuid_munged(&init_user_ns, sbinfo->uid));
3496 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3497 seq_printf(seq, ",gid=%u",
3498 from_kgid_munged(&init_user_ns, sbinfo->gid));
3499 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3500 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3502 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3504 shmem_show_mpol(seq, sbinfo->mpol);
3508 #define MFD_NAME_PREFIX "memfd:"
3509 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3510 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3512 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3514 SYSCALL_DEFINE2(memfd_create,
3515 const char __user *, uname,
3516 unsigned int, flags)
3518 struct shmem_inode_info *info;
3524 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3527 /* length includes terminating zero */
3528 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3531 if (len > MFD_NAME_MAX_LEN + 1)
3534 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3538 strcpy(name, MFD_NAME_PREFIX);
3539 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3544 /* terminating-zero may have changed after strnlen_user() returned */
3545 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3550 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3556 file = shmem_file_setup(name, 0, VM_NORESERVE);
3558 error = PTR_ERR(file);
3561 info = SHMEM_I(file_inode(file));
3562 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3563 file->f_flags |= O_RDWR | O_LARGEFILE;
3564 if (flags & MFD_ALLOW_SEALING)
3565 info->seals &= ~F_SEAL_SEAL;
3567 fd_install(fd, file);
3578 #endif /* CONFIG_TMPFS */
3580 static void shmem_put_super(struct super_block *sb)
3582 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3584 percpu_counter_destroy(&sbinfo->used_blocks);
3585 mpol_put(sbinfo->mpol);
3587 sb->s_fs_info = NULL;
3590 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3592 struct inode *inode;
3593 struct shmem_sb_info *sbinfo;
3596 /* Round up to L1_CACHE_BYTES to resist false sharing */
3597 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3598 L1_CACHE_BYTES), GFP_KERNEL);
3602 sbinfo->mode = S_IRWXUGO | S_ISVTX;
3603 sbinfo->uid = current_fsuid();
3604 sbinfo->gid = current_fsgid();
3605 sb->s_fs_info = sbinfo;
3609 * Per default we only allow half of the physical ram per
3610 * tmpfs instance, limiting inodes to one per page of lowmem;
3611 * but the internal instance is left unlimited.
3613 if (!(sb->s_flags & MS_KERNMOUNT)) {
3614 sbinfo->max_blocks = shmem_default_max_blocks();
3615 sbinfo->max_inodes = shmem_default_max_inodes();
3616 if (shmem_parse_options(data, sbinfo, false)) {
3621 sb->s_flags |= MS_NOUSER;
3623 sb->s_export_op = &shmem_export_ops;
3624 sb->s_flags |= MS_NOSEC;
3626 sb->s_flags |= MS_NOUSER;
3629 spin_lock_init(&sbinfo->stat_lock);
3630 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3632 sbinfo->free_inodes = sbinfo->max_inodes;
3633 spin_lock_init(&sbinfo->shrinklist_lock);
3634 INIT_LIST_HEAD(&sbinfo->shrinklist);
3636 sb->s_maxbytes = MAX_LFS_FILESIZE;
3637 sb->s_blocksize = PAGE_SIZE;
3638 sb->s_blocksize_bits = PAGE_SHIFT;
3639 sb->s_magic = TMPFS_MAGIC;
3640 sb->s_op = &shmem_ops;
3641 sb->s_time_gran = 1;
3642 #ifdef CONFIG_TMPFS_XATTR
3643 sb->s_xattr = shmem_xattr_handlers;
3645 #ifdef CONFIG_TMPFS_POSIX_ACL
3646 sb->s_flags |= MS_POSIXACL;
3649 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3652 inode->i_uid = sbinfo->uid;
3653 inode->i_gid = sbinfo->gid;
3654 sb->s_root = d_make_root(inode);
3660 shmem_put_super(sb);
3664 static struct kmem_cache *shmem_inode_cachep;
3666 static struct inode *shmem_alloc_inode(struct super_block *sb)
3668 struct shmem_inode_info *info;
3669 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3672 return &info->vfs_inode;
3675 static void shmem_destroy_callback(struct rcu_head *head)
3677 struct inode *inode = container_of(head, struct inode, i_rcu);
3678 if (S_ISLNK(inode->i_mode))
3679 kfree(inode->i_link);
3680 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3683 static void shmem_destroy_inode(struct inode *inode)
3685 if (S_ISREG(inode->i_mode))
3686 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3687 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3690 static void shmem_init_inode(void *foo)
3692 struct shmem_inode_info *info = foo;
3693 inode_init_once(&info->vfs_inode);
3696 static int shmem_init_inodecache(void)
3698 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3699 sizeof(struct shmem_inode_info),
3700 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3704 static void shmem_destroy_inodecache(void)
3706 kmem_cache_destroy(shmem_inode_cachep);
3709 static const struct address_space_operations shmem_aops = {
3710 .writepage = shmem_writepage,
3711 .set_page_dirty = __set_page_dirty_no_writeback,
3713 .write_begin = shmem_write_begin,
3714 .write_end = shmem_write_end,
3716 #ifdef CONFIG_MIGRATION
3717 .migratepage = migrate_page,
3719 .error_remove_page = generic_error_remove_page,
3722 static const struct file_operations shmem_file_operations = {
3724 .get_unmapped_area = shmem_get_unmapped_area,
3726 .llseek = shmem_file_llseek,
3727 .read_iter = shmem_file_read_iter,
3728 .write_iter = generic_file_write_iter,
3729 .fsync = noop_fsync,
3730 .splice_read = generic_file_splice_read,
3731 .splice_write = iter_file_splice_write,
3732 .fallocate = shmem_fallocate,
3736 static const struct inode_operations shmem_inode_operations = {
3737 .getattr = shmem_getattr,
3738 .setattr = shmem_setattr,
3739 #ifdef CONFIG_TMPFS_XATTR
3740 .listxattr = shmem_listxattr,
3741 .set_acl = simple_set_acl,
3745 static const struct inode_operations shmem_dir_inode_operations = {
3747 .create = shmem_create,
3748 .lookup = simple_lookup,
3750 .unlink = shmem_unlink,
3751 .symlink = shmem_symlink,
3752 .mkdir = shmem_mkdir,
3753 .rmdir = shmem_rmdir,
3754 .mknod = shmem_mknod,
3755 .rename = shmem_rename2,
3756 .tmpfile = shmem_tmpfile,
3758 #ifdef CONFIG_TMPFS_XATTR
3759 .listxattr = shmem_listxattr,
3761 #ifdef CONFIG_TMPFS_POSIX_ACL
3762 .setattr = shmem_setattr,
3763 .set_acl = simple_set_acl,
3767 static const struct inode_operations shmem_special_inode_operations = {
3768 #ifdef CONFIG_TMPFS_XATTR
3769 .listxattr = shmem_listxattr,
3771 #ifdef CONFIG_TMPFS_POSIX_ACL
3772 .setattr = shmem_setattr,
3773 .set_acl = simple_set_acl,
3777 static const struct super_operations shmem_ops = {
3778 .alloc_inode = shmem_alloc_inode,
3779 .destroy_inode = shmem_destroy_inode,
3781 .statfs = shmem_statfs,
3782 .remount_fs = shmem_remount_fs,
3783 .show_options = shmem_show_options,
3785 .evict_inode = shmem_evict_inode,
3786 .drop_inode = generic_delete_inode,
3787 .put_super = shmem_put_super,
3788 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3789 .nr_cached_objects = shmem_unused_huge_count,
3790 .free_cached_objects = shmem_unused_huge_scan,
3794 static const struct vm_operations_struct shmem_vm_ops = {
3795 .fault = shmem_fault,
3796 .map_pages = filemap_map_pages,
3798 .set_policy = shmem_set_policy,
3799 .get_policy = shmem_get_policy,
3803 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3804 int flags, const char *dev_name, void *data)
3806 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3809 static struct file_system_type shmem_fs_type = {
3810 .owner = THIS_MODULE,
3812 .mount = shmem_mount,
3813 .kill_sb = kill_litter_super,
3814 .fs_flags = FS_USERNS_MOUNT,
3817 int __init shmem_init(void)
3821 /* If rootfs called this, don't re-init */
3822 if (shmem_inode_cachep)
3825 error = shmem_init_inodecache();
3829 error = register_filesystem(&shmem_fs_type);
3831 pr_err("Could not register tmpfs\n");
3835 shm_mnt = kern_mount(&shmem_fs_type);
3836 if (IS_ERR(shm_mnt)) {
3837 error = PTR_ERR(shm_mnt);
3838 pr_err("Could not kern_mount tmpfs\n");
3842 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3843 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3844 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3846 shmem_huge = 0; /* just in case it was patched */
3851 unregister_filesystem(&shmem_fs_type);
3853 shmem_destroy_inodecache();
3855 shm_mnt = ERR_PTR(error);
3859 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3860 static ssize_t shmem_enabled_show(struct kobject *kobj,
3861 struct kobj_attribute *attr, char *buf)
3865 SHMEM_HUGE_WITHIN_SIZE,
3873 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3874 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3876 count += sprintf(buf + count, fmt,
3877 shmem_format_huge(values[i]));
3879 buf[count - 1] = '\n';
3883 static ssize_t shmem_enabled_store(struct kobject *kobj,
3884 struct kobj_attribute *attr, const char *buf, size_t count)
3889 if (count + 1 > sizeof(tmp))
3891 memcpy(tmp, buf, count);
3893 if (count && tmp[count - 1] == '\n')
3894 tmp[count - 1] = '\0';
3896 huge = shmem_parse_huge(tmp);
3897 if (huge == -EINVAL)
3899 if (!has_transparent_hugepage() &&
3900 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3904 if (shmem_huge > SHMEM_HUGE_DENY)
3905 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3909 struct kobj_attribute shmem_enabled_attr =
3910 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3911 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3913 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3914 bool shmem_huge_enabled(struct vm_area_struct *vma)
3916 struct inode *inode = file_inode(vma->vm_file);
3917 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3921 if (shmem_huge == SHMEM_HUGE_FORCE)
3923 if (shmem_huge == SHMEM_HUGE_DENY)
3925 switch (sbinfo->huge) {
3926 case SHMEM_HUGE_NEVER:
3928 case SHMEM_HUGE_ALWAYS:
3930 case SHMEM_HUGE_WITHIN_SIZE:
3931 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3932 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3933 if (i_size >= HPAGE_PMD_SIZE &&
3934 i_size >> PAGE_SHIFT >= off)
3936 case SHMEM_HUGE_ADVISE:
3937 /* TODO: implement fadvise() hints */
3938 return (vma->vm_flags & VM_HUGEPAGE);
3944 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3946 #else /* !CONFIG_SHMEM */
3949 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3951 * This is intended for small system where the benefits of the full
3952 * shmem code (swap-backed and resource-limited) are outweighed by
3953 * their complexity. On systems without swap this code should be
3954 * effectively equivalent, but much lighter weight.
3957 static struct file_system_type shmem_fs_type = {
3959 .mount = ramfs_mount,
3960 .kill_sb = kill_litter_super,
3961 .fs_flags = FS_USERNS_MOUNT,
3964 int __init shmem_init(void)
3966 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3968 shm_mnt = kern_mount(&shmem_fs_type);
3969 BUG_ON(IS_ERR(shm_mnt));
3974 int shmem_unuse(swp_entry_t swap, struct page *page)
3979 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3984 void shmem_unlock_mapping(struct address_space *mapping)
3989 unsigned long shmem_get_unmapped_area(struct file *file,
3990 unsigned long addr, unsigned long len,
3991 unsigned long pgoff, unsigned long flags)
3993 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3997 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3999 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4001 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4003 #define shmem_vm_ops generic_file_vm_ops
4004 #define shmem_file_operations ramfs_file_operations
4005 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4006 #define shmem_acct_size(flags, size) 0
4007 #define shmem_unacct_size(flags, size) do {} while (0)
4009 #endif /* CONFIG_SHMEM */
4013 static const struct dentry_operations anon_ops = {
4014 .d_dname = simple_dname
4017 static struct file *__shmem_file_setup(const char *name, loff_t size,
4018 unsigned long flags, unsigned int i_flags)
4021 struct inode *inode;
4023 struct super_block *sb;
4026 if (IS_ERR(shm_mnt))
4027 return ERR_CAST(shm_mnt);
4029 if (size < 0 || size > MAX_LFS_FILESIZE)
4030 return ERR_PTR(-EINVAL);
4032 if (shmem_acct_size(flags, size))
4033 return ERR_PTR(-ENOMEM);
4035 res = ERR_PTR(-ENOMEM);
4037 this.len = strlen(name);
4038 this.hash = 0; /* will go */
4039 sb = shm_mnt->mnt_sb;
4040 path.mnt = mntget(shm_mnt);
4041 path.dentry = d_alloc_pseudo(sb, &this);
4044 d_set_d_op(path.dentry, &anon_ops);
4046 res = ERR_PTR(-ENOSPC);
4047 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4051 inode->i_flags |= i_flags;
4052 d_instantiate(path.dentry, inode);
4053 inode->i_size = size;
4054 clear_nlink(inode); /* It is unlinked */
4055 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4059 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4060 &shmem_file_operations);
4067 shmem_unacct_size(flags, size);
4074 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4075 * kernel internal. There will be NO LSM permission checks against the
4076 * underlying inode. So users of this interface must do LSM checks at a
4077 * higher layer. The users are the big_key and shm implementations. LSM
4078 * checks are provided at the key or shm level rather than the inode.
4079 * @name: name for dentry (to be seen in /proc/<pid>/maps
4080 * @size: size to be set for the file
4081 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4083 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4085 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4089 * shmem_file_setup - get an unlinked file living in tmpfs
4090 * @name: name for dentry (to be seen in /proc/<pid>/maps
4091 * @size: size to be set for the file
4092 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4094 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4096 return __shmem_file_setup(name, size, flags, 0);
4098 EXPORT_SYMBOL_GPL(shmem_file_setup);
4101 * shmem_zero_setup - setup a shared anonymous mapping
4102 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4104 int shmem_zero_setup(struct vm_area_struct *vma)
4107 loff_t size = vma->vm_end - vma->vm_start;
4110 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4111 * between XFS directory reading and selinux: since this file is only
4112 * accessible to the user through its mapping, use S_PRIVATE flag to
4113 * bypass file security, in the same way as shmem_kernel_file_setup().
4115 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4117 return PTR_ERR(file);
4121 vma->vm_file = file;
4122 vma->vm_ops = &shmem_vm_ops;
4124 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4125 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4126 (vma->vm_end & HPAGE_PMD_MASK)) {
4127 khugepaged_enter(vma, vma->vm_flags);
4134 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4135 * @mapping: the page's address_space
4136 * @index: the page index
4137 * @gfp: the page allocator flags to use if allocating
4139 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4140 * with any new page allocations done using the specified allocation flags.
4141 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4142 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4143 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4145 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4146 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4148 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4149 pgoff_t index, gfp_t gfp)
4152 struct inode *inode = mapping->host;
4156 BUG_ON(mapping->a_ops != &shmem_aops);
4157 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4160 page = ERR_PTR(error);
4166 * The tiny !SHMEM case uses ramfs without swap
4168 return read_cache_page_gfp(mapping, index, gfp);
4171 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);