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/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
40 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
42 static struct vfsmount *shm_mnt;
46 * This virtual memory filesystem is heavily based on the ramfs. It
47 * extends ramfs by the ability to use swap and honor resource limits
48 * which makes it a completely usable filesystem.
51 #include <linux/xattr.h>
52 #include <linux/exportfs.h>
53 #include <linux/posix_acl.h>
54 #include <linux/posix_acl_xattr.h>
55 #include <linux/mman.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/backing-dev.h>
59 #include <linux/shmem_fs.h>
60 #include <linux/writeback.h>
61 #include <linux/blkdev.h>
62 #include <linux/pagevec.h>
63 #include <linux/percpu_counter.h>
64 #include <linux/falloc.h>
65 #include <linux/splice.h>
66 #include <linux/security.h>
67 #include <linux/swapops.h>
68 #include <linux/mempolicy.h>
69 #include <linux/namei.h>
70 #include <linux/ctype.h>
71 #include <linux/migrate.h>
72 #include <linux/highmem.h>
73 #include <linux/seq_file.h>
74 #include <linux/magic.h>
75 #include <linux/syscalls.h>
76 #include <linux/fcntl.h>
77 #include <uapi/linux/memfd.h>
78 #include <linux/userfaultfd_k.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
82 #include <linux/uaccess.h>
83 #include <asm/pgtable.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_mutex making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc {
102 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
103 pgoff_t start; /* start of range currently being fallocated */
104 pgoff_t next; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
110 static unsigned long shmem_default_max_blocks(void)
112 return totalram_pages / 2;
115 static unsigned long shmem_default_max_inodes(void)
117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
121 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
122 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
123 struct shmem_inode_info *info, pgoff_t index);
124 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
125 struct page **pagep, enum sgp_type sgp,
126 gfp_t gfp, struct vm_area_struct *vma,
127 struct vm_fault *vmf, vm_fault_t *fault_type);
129 int shmem_getpage(struct inode *inode, pgoff_t index,
130 struct page **pagep, enum sgp_type sgp)
132 return shmem_getpage_gfp(inode, index, pagep, sgp,
133 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
136 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
138 return sb->s_fs_info;
142 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
143 * for shared memory and for shared anonymous (/dev/zero) mappings
144 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
145 * consistent with the pre-accounting of private mappings ...
147 static inline int shmem_acct_size(unsigned long flags, loff_t size)
149 return (flags & VM_NORESERVE) ?
150 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
153 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
155 if (!(flags & VM_NORESERVE))
156 vm_unacct_memory(VM_ACCT(size));
159 static inline int shmem_reacct_size(unsigned long flags,
160 loff_t oldsize, loff_t newsize)
162 if (!(flags & VM_NORESERVE)) {
163 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
164 return security_vm_enough_memory_mm(current->mm,
165 VM_ACCT(newsize) - VM_ACCT(oldsize));
166 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
167 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
173 * ... whereas tmpfs objects are accounted incrementally as
174 * pages are allocated, in order to allow large sparse files.
175 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
176 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
178 static inline int shmem_acct_block(unsigned long flags, long pages)
180 if (!(flags & VM_NORESERVE))
183 return security_vm_enough_memory_mm(current->mm,
184 pages * VM_ACCT(PAGE_SIZE));
187 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
189 if (flags & VM_NORESERVE)
190 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
193 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
195 struct shmem_inode_info *info = SHMEM_I(inode);
196 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
198 if (shmem_acct_block(info->flags, pages))
201 if (sbinfo->max_blocks) {
202 if (percpu_counter_compare(&sbinfo->used_blocks,
203 sbinfo->max_blocks - pages) > 0)
205 percpu_counter_add(&sbinfo->used_blocks, pages);
211 shmem_unacct_blocks(info->flags, pages);
215 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
217 struct shmem_inode_info *info = SHMEM_I(inode);
218 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
220 if (sbinfo->max_blocks)
221 percpu_counter_sub(&sbinfo->used_blocks, pages);
222 shmem_unacct_blocks(info->flags, pages);
225 static const struct super_operations shmem_ops;
226 static const struct address_space_operations shmem_aops;
227 static const struct file_operations shmem_file_operations;
228 static const struct inode_operations shmem_inode_operations;
229 static const struct inode_operations shmem_dir_inode_operations;
230 static const struct inode_operations shmem_special_inode_operations;
231 static const struct vm_operations_struct shmem_vm_ops;
232 static struct file_system_type shmem_fs_type;
234 bool vma_is_shmem(struct vm_area_struct *vma)
236 return vma->vm_ops == &shmem_vm_ops;
239 static LIST_HEAD(shmem_swaplist);
240 static DEFINE_MUTEX(shmem_swaplist_mutex);
242 static int shmem_reserve_inode(struct super_block *sb)
244 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
245 if (sbinfo->max_inodes) {
246 spin_lock(&sbinfo->stat_lock);
247 if (!sbinfo->free_inodes) {
248 spin_unlock(&sbinfo->stat_lock);
251 sbinfo->free_inodes--;
252 spin_unlock(&sbinfo->stat_lock);
257 static void shmem_free_inode(struct super_block *sb)
259 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
260 if (sbinfo->max_inodes) {
261 spin_lock(&sbinfo->stat_lock);
262 sbinfo->free_inodes++;
263 spin_unlock(&sbinfo->stat_lock);
268 * shmem_recalc_inode - recalculate the block usage of an inode
269 * @inode: inode to recalc
271 * We have to calculate the free blocks since the mm can drop
272 * undirtied hole pages behind our back.
274 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
275 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
277 * It has to be called with the spinlock held.
279 static void shmem_recalc_inode(struct inode *inode)
281 struct shmem_inode_info *info = SHMEM_I(inode);
284 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
286 info->alloced -= freed;
287 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
288 shmem_inode_unacct_blocks(inode, freed);
292 bool shmem_charge(struct inode *inode, long pages)
294 struct shmem_inode_info *info = SHMEM_I(inode);
297 if (!shmem_inode_acct_block(inode, pages))
300 spin_lock_irqsave(&info->lock, flags);
301 info->alloced += pages;
302 inode->i_blocks += pages * BLOCKS_PER_PAGE;
303 shmem_recalc_inode(inode);
304 spin_unlock_irqrestore(&info->lock, flags);
305 inode->i_mapping->nrpages += pages;
310 void shmem_uncharge(struct inode *inode, long pages)
312 struct shmem_inode_info *info = SHMEM_I(inode);
315 spin_lock_irqsave(&info->lock, flags);
316 info->alloced -= pages;
317 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
318 shmem_recalc_inode(inode);
319 spin_unlock_irqrestore(&info->lock, flags);
321 shmem_inode_unacct_blocks(inode, pages);
325 * Replace item expected in radix tree by a new item, while holding tree lock.
327 static int shmem_radix_tree_replace(struct address_space *mapping,
328 pgoff_t index, void *expected, void *replacement)
330 struct radix_tree_node *node;
334 VM_BUG_ON(!expected);
335 VM_BUG_ON(!replacement);
336 item = __radix_tree_lookup(&mapping->i_pages, index, &node, &pslot);
339 if (item != expected)
341 __radix_tree_replace(&mapping->i_pages, node, pslot,
347 * Sometimes, before we decide whether to proceed or to fail, we must check
348 * that an entry was not already brought back from swap by a racing thread.
350 * Checking page is not enough: by the time a SwapCache page is locked, it
351 * might be reused, and again be SwapCache, using the same swap as before.
353 static bool shmem_confirm_swap(struct address_space *mapping,
354 pgoff_t index, swp_entry_t swap)
359 item = radix_tree_lookup(&mapping->i_pages, index);
361 return item == swp_to_radix_entry(swap);
365 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
368 * disables huge pages for the mount;
370 * enables huge pages for the mount;
371 * SHMEM_HUGE_WITHIN_SIZE:
372 * only allocate huge pages if the page will be fully within i_size,
373 * also respect fadvise()/madvise() hints;
375 * only allocate huge pages if requested with fadvise()/madvise();
378 #define SHMEM_HUGE_NEVER 0
379 #define SHMEM_HUGE_ALWAYS 1
380 #define SHMEM_HUGE_WITHIN_SIZE 2
381 #define SHMEM_HUGE_ADVISE 3
385 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
388 * disables huge on shm_mnt and all mounts, for emergency use;
390 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
393 #define SHMEM_HUGE_DENY (-1)
394 #define SHMEM_HUGE_FORCE (-2)
396 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
397 /* ifdef here to avoid bloating shmem.o when not necessary */
399 static int shmem_huge __read_mostly;
401 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
402 static int shmem_parse_huge(const char *str)
404 if (!strcmp(str, "never"))
405 return SHMEM_HUGE_NEVER;
406 if (!strcmp(str, "always"))
407 return SHMEM_HUGE_ALWAYS;
408 if (!strcmp(str, "within_size"))
409 return SHMEM_HUGE_WITHIN_SIZE;
410 if (!strcmp(str, "advise"))
411 return SHMEM_HUGE_ADVISE;
412 if (!strcmp(str, "deny"))
413 return SHMEM_HUGE_DENY;
414 if (!strcmp(str, "force"))
415 return SHMEM_HUGE_FORCE;
419 static const char *shmem_format_huge(int huge)
422 case SHMEM_HUGE_NEVER:
424 case SHMEM_HUGE_ALWAYS:
426 case SHMEM_HUGE_WITHIN_SIZE:
427 return "within_size";
428 case SHMEM_HUGE_ADVISE:
430 case SHMEM_HUGE_DENY:
432 case SHMEM_HUGE_FORCE:
441 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
442 struct shrink_control *sc, unsigned long nr_to_split)
444 LIST_HEAD(list), *pos, *next;
445 LIST_HEAD(to_remove);
447 struct shmem_inode_info *info;
449 unsigned long batch = sc ? sc->nr_to_scan : 128;
450 int removed = 0, split = 0;
452 if (list_empty(&sbinfo->shrinklist))
455 spin_lock(&sbinfo->shrinklist_lock);
456 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
457 info = list_entry(pos, struct shmem_inode_info, shrinklist);
460 inode = igrab(&info->vfs_inode);
462 /* inode is about to be evicted */
464 list_del_init(&info->shrinklist);
469 /* Check if there's anything to gain */
470 if (round_up(inode->i_size, PAGE_SIZE) ==
471 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
472 list_move(&info->shrinklist, &to_remove);
477 list_move(&info->shrinklist, &list);
482 spin_unlock(&sbinfo->shrinklist_lock);
484 list_for_each_safe(pos, next, &to_remove) {
485 info = list_entry(pos, struct shmem_inode_info, shrinklist);
486 inode = &info->vfs_inode;
487 list_del_init(&info->shrinklist);
491 list_for_each_safe(pos, next, &list) {
494 info = list_entry(pos, struct shmem_inode_info, shrinklist);
495 inode = &info->vfs_inode;
497 if (nr_to_split && split >= nr_to_split)
500 page = find_get_page(inode->i_mapping,
501 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
505 /* No huge page at the end of the file: nothing to split */
506 if (!PageTransHuge(page)) {
512 * Leave the inode on the list if we failed to lock
513 * the page at this time.
515 * Waiting for the lock may lead to deadlock in the
518 if (!trylock_page(page)) {
523 ret = split_huge_page(page);
527 /* If split failed leave the inode on the list */
533 list_del_init(&info->shrinklist);
539 spin_lock(&sbinfo->shrinklist_lock);
540 list_splice_tail(&list, &sbinfo->shrinklist);
541 sbinfo->shrinklist_len -= removed;
542 spin_unlock(&sbinfo->shrinklist_lock);
547 static long shmem_unused_huge_scan(struct super_block *sb,
548 struct shrink_control *sc)
550 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
552 if (!READ_ONCE(sbinfo->shrinklist_len))
555 return shmem_unused_huge_shrink(sbinfo, sc, 0);
558 static long shmem_unused_huge_count(struct super_block *sb,
559 struct shrink_control *sc)
561 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
562 return READ_ONCE(sbinfo->shrinklist_len);
564 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
566 #define shmem_huge SHMEM_HUGE_DENY
568 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
569 struct shrink_control *sc, unsigned long nr_to_split)
573 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
575 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
577 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
578 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
579 shmem_huge != SHMEM_HUGE_DENY)
585 * Like add_to_page_cache_locked, but error if expected item has gone.
587 static int shmem_add_to_page_cache(struct page *page,
588 struct address_space *mapping,
589 pgoff_t index, void *expected)
591 int error, nr = hpage_nr_pages(page);
593 VM_BUG_ON_PAGE(PageTail(page), page);
594 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
595 VM_BUG_ON_PAGE(!PageLocked(page), page);
596 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
597 VM_BUG_ON(expected && PageTransHuge(page));
599 page_ref_add(page, nr);
600 page->mapping = mapping;
603 xa_lock_irq(&mapping->i_pages);
604 if (PageTransHuge(page)) {
605 void __rcu **results;
610 if (radix_tree_gang_lookup_slot(&mapping->i_pages,
611 &results, &idx, index, 1) &&
612 idx < index + HPAGE_PMD_NR) {
617 for (i = 0; i < HPAGE_PMD_NR; i++) {
618 error = radix_tree_insert(&mapping->i_pages,
619 index + i, page + i);
622 count_vm_event(THP_FILE_ALLOC);
624 } else if (!expected) {
625 error = radix_tree_insert(&mapping->i_pages, index, page);
627 error = shmem_radix_tree_replace(mapping, index, expected,
632 mapping->nrpages += nr;
633 if (PageTransHuge(page))
634 __inc_node_page_state(page, NR_SHMEM_THPS);
635 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
636 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
637 xa_unlock_irq(&mapping->i_pages);
639 page->mapping = NULL;
640 xa_unlock_irq(&mapping->i_pages);
641 page_ref_sub(page, nr);
647 * Like delete_from_page_cache, but substitutes swap for page.
649 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
651 struct address_space *mapping = page->mapping;
654 VM_BUG_ON_PAGE(PageCompound(page), page);
656 xa_lock_irq(&mapping->i_pages);
657 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
658 page->mapping = NULL;
660 __dec_node_page_state(page, NR_FILE_PAGES);
661 __dec_node_page_state(page, NR_SHMEM);
662 xa_unlock_irq(&mapping->i_pages);
668 * Remove swap entry from radix tree, free the swap and its page cache.
670 static int shmem_free_swap(struct address_space *mapping,
671 pgoff_t index, void *radswap)
675 xa_lock_irq(&mapping->i_pages);
676 old = radix_tree_delete_item(&mapping->i_pages, index, radswap);
677 xa_unlock_irq(&mapping->i_pages);
680 free_swap_and_cache(radix_to_swp_entry(radswap));
685 * Determine (in bytes) how many of the shmem object's pages mapped by the
686 * given offsets are swapped out.
688 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
689 * as long as the inode doesn't go away and racy results are not a problem.
691 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
692 pgoff_t start, pgoff_t end)
694 struct radix_tree_iter iter;
697 unsigned long swapped = 0;
701 radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
702 if (iter.index >= end)
705 page = radix_tree_deref_slot(slot);
707 if (radix_tree_deref_retry(page)) {
708 slot = radix_tree_iter_retry(&iter);
712 if (radix_tree_exceptional_entry(page))
715 if (need_resched()) {
716 slot = radix_tree_iter_resume(slot, &iter);
723 return swapped << PAGE_SHIFT;
727 * Determine (in bytes) how many of the shmem object's pages mapped by the
728 * given vma is swapped out.
730 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
731 * as long as the inode doesn't go away and racy results are not a problem.
733 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
735 struct inode *inode = file_inode(vma->vm_file);
736 struct shmem_inode_info *info = SHMEM_I(inode);
737 struct address_space *mapping = inode->i_mapping;
738 unsigned long swapped;
740 /* Be careful as we don't hold info->lock */
741 swapped = READ_ONCE(info->swapped);
744 * The easier cases are when the shmem object has nothing in swap, or
745 * the vma maps it whole. Then we can simply use the stats that we
751 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
752 return swapped << PAGE_SHIFT;
754 /* Here comes the more involved part */
755 return shmem_partial_swap_usage(mapping,
756 linear_page_index(vma, vma->vm_start),
757 linear_page_index(vma, vma->vm_end));
761 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
763 void shmem_unlock_mapping(struct address_space *mapping)
766 pgoff_t indices[PAGEVEC_SIZE];
771 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
773 while (!mapping_unevictable(mapping)) {
775 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
776 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
778 pvec.nr = find_get_entries(mapping, index,
779 PAGEVEC_SIZE, pvec.pages, indices);
782 index = indices[pvec.nr - 1] + 1;
783 pagevec_remove_exceptionals(&pvec);
784 check_move_unevictable_pages(pvec.pages, pvec.nr);
785 pagevec_release(&pvec);
791 * Remove range of pages and swap entries from radix tree, and free them.
792 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
794 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
797 struct address_space *mapping = inode->i_mapping;
798 struct shmem_inode_info *info = SHMEM_I(inode);
799 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
800 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
801 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
802 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
804 pgoff_t indices[PAGEVEC_SIZE];
805 long nr_swaps_freed = 0;
810 end = -1; /* unsigned, so actually very big */
814 while (index < end) {
815 pvec.nr = find_get_entries(mapping, index,
816 min(end - index, (pgoff_t)PAGEVEC_SIZE),
817 pvec.pages, indices);
820 for (i = 0; i < pagevec_count(&pvec); i++) {
821 struct page *page = pvec.pages[i];
827 if (radix_tree_exceptional_entry(page)) {
830 nr_swaps_freed += !shmem_free_swap(mapping,
835 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
837 if (!trylock_page(page))
840 if (PageTransTail(page)) {
841 /* Middle of THP: zero out the page */
842 clear_highpage(page);
845 } else if (PageTransHuge(page)) {
846 if (index == round_down(end, HPAGE_PMD_NR)) {
848 * Range ends in the middle of THP:
851 clear_highpage(page);
855 index += HPAGE_PMD_NR - 1;
856 i += HPAGE_PMD_NR - 1;
859 if (!unfalloc || !PageUptodate(page)) {
860 VM_BUG_ON_PAGE(PageTail(page), page);
861 if (page_mapping(page) == mapping) {
862 VM_BUG_ON_PAGE(PageWriteback(page), page);
863 truncate_inode_page(mapping, page);
868 pagevec_remove_exceptionals(&pvec);
869 pagevec_release(&pvec);
875 struct page *page = NULL;
876 shmem_getpage(inode, start - 1, &page, SGP_READ);
878 unsigned int top = PAGE_SIZE;
883 zero_user_segment(page, partial_start, top);
884 set_page_dirty(page);
890 struct page *page = NULL;
891 shmem_getpage(inode, end, &page, SGP_READ);
893 zero_user_segment(page, 0, partial_end);
894 set_page_dirty(page);
903 while (index < end) {
906 pvec.nr = find_get_entries(mapping, index,
907 min(end - index, (pgoff_t)PAGEVEC_SIZE),
908 pvec.pages, indices);
910 /* If all gone or hole-punch or unfalloc, we're done */
911 if (index == start || end != -1)
913 /* But if truncating, restart to make sure all gone */
917 for (i = 0; i < pagevec_count(&pvec); i++) {
918 struct page *page = pvec.pages[i];
924 if (radix_tree_exceptional_entry(page)) {
927 if (shmem_free_swap(mapping, index, page)) {
928 /* Swap was replaced by page: retry */
938 if (PageTransTail(page)) {
939 /* Middle of THP: zero out the page */
940 clear_highpage(page);
943 * Partial thp truncate due 'start' in middle
944 * of THP: don't need to look on these pages
945 * again on !pvec.nr restart.
947 if (index != round_down(end, HPAGE_PMD_NR))
950 } else if (PageTransHuge(page)) {
951 if (index == round_down(end, HPAGE_PMD_NR)) {
953 * Range ends in the middle of THP:
956 clear_highpage(page);
960 index += HPAGE_PMD_NR - 1;
961 i += HPAGE_PMD_NR - 1;
964 if (!unfalloc || !PageUptodate(page)) {
965 VM_BUG_ON_PAGE(PageTail(page), page);
966 if (page_mapping(page) == mapping) {
967 VM_BUG_ON_PAGE(PageWriteback(page), page);
968 truncate_inode_page(mapping, page);
970 /* Page was replaced by swap: retry */
978 pagevec_remove_exceptionals(&pvec);
979 pagevec_release(&pvec);
983 spin_lock_irq(&info->lock);
984 info->swapped -= nr_swaps_freed;
985 shmem_recalc_inode(inode);
986 spin_unlock_irq(&info->lock);
989 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
991 shmem_undo_range(inode, lstart, lend, false);
992 inode->i_ctime = inode->i_mtime = current_time(inode);
994 EXPORT_SYMBOL_GPL(shmem_truncate_range);
996 static int shmem_getattr(const struct path *path, struct kstat *stat,
997 u32 request_mask, unsigned int query_flags)
999 struct inode *inode = path->dentry->d_inode;
1000 struct shmem_inode_info *info = SHMEM_I(inode);
1001 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1003 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1004 spin_lock_irq(&info->lock);
1005 shmem_recalc_inode(inode);
1006 spin_unlock_irq(&info->lock);
1008 generic_fillattr(inode, stat);
1010 if (is_huge_enabled(sb_info))
1011 stat->blksize = HPAGE_PMD_SIZE;
1016 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
1018 struct inode *inode = d_inode(dentry);
1019 struct shmem_inode_info *info = SHMEM_I(inode);
1020 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1023 error = setattr_prepare(dentry, attr);
1027 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1028 loff_t oldsize = inode->i_size;
1029 loff_t newsize = attr->ia_size;
1031 /* protected by i_mutex */
1032 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1033 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1036 if (newsize != oldsize) {
1037 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1041 i_size_write(inode, newsize);
1042 inode->i_ctime = inode->i_mtime = current_time(inode);
1044 if (newsize <= oldsize) {
1045 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1046 if (oldsize > holebegin)
1047 unmap_mapping_range(inode->i_mapping,
1050 shmem_truncate_range(inode,
1051 newsize, (loff_t)-1);
1052 /* unmap again to remove racily COWed private pages */
1053 if (oldsize > holebegin)
1054 unmap_mapping_range(inode->i_mapping,
1058 * Part of the huge page can be beyond i_size: subject
1059 * to shrink under memory pressure.
1061 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1062 spin_lock(&sbinfo->shrinklist_lock);
1064 * _careful to defend against unlocked access to
1065 * ->shrink_list in shmem_unused_huge_shrink()
1067 if (list_empty_careful(&info->shrinklist)) {
1068 list_add_tail(&info->shrinklist,
1069 &sbinfo->shrinklist);
1070 sbinfo->shrinklist_len++;
1072 spin_unlock(&sbinfo->shrinklist_lock);
1077 setattr_copy(inode, attr);
1078 if (attr->ia_valid & ATTR_MODE)
1079 error = posix_acl_chmod(inode, inode->i_mode);
1083 static void shmem_evict_inode(struct inode *inode)
1085 struct shmem_inode_info *info = SHMEM_I(inode);
1086 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1088 if (inode->i_mapping->a_ops == &shmem_aops) {
1089 shmem_unacct_size(info->flags, inode->i_size);
1091 shmem_truncate_range(inode, 0, (loff_t)-1);
1092 if (!list_empty(&info->shrinklist)) {
1093 spin_lock(&sbinfo->shrinklist_lock);
1094 if (!list_empty(&info->shrinklist)) {
1095 list_del_init(&info->shrinklist);
1096 sbinfo->shrinklist_len--;
1098 spin_unlock(&sbinfo->shrinklist_lock);
1100 if (!list_empty(&info->swaplist)) {
1101 mutex_lock(&shmem_swaplist_mutex);
1102 list_del_init(&info->swaplist);
1103 mutex_unlock(&shmem_swaplist_mutex);
1107 simple_xattrs_free(&info->xattrs);
1108 WARN_ON(inode->i_blocks);
1109 shmem_free_inode(inode->i_sb);
1113 static unsigned long find_swap_entry(struct radix_tree_root *root, void *item)
1115 struct radix_tree_iter iter;
1117 unsigned long found = -1;
1118 unsigned int checked = 0;
1121 radix_tree_for_each_slot(slot, root, &iter, 0) {
1122 void *entry = radix_tree_deref_slot(slot);
1124 if (radix_tree_deref_retry(entry)) {
1125 slot = radix_tree_iter_retry(&iter);
1128 if (entry == item) {
1133 if ((checked % 4096) != 0)
1135 slot = radix_tree_iter_resume(slot, &iter);
1144 * If swap found in inode, free it and move page from swapcache to filecache.
1146 static int shmem_unuse_inode(struct shmem_inode_info *info,
1147 swp_entry_t swap, struct page **pagep)
1149 struct address_space *mapping = info->vfs_inode.i_mapping;
1155 radswap = swp_to_radix_entry(swap);
1156 index = find_swap_entry(&mapping->i_pages, radswap);
1158 return -EAGAIN; /* tell shmem_unuse we found nothing */
1161 * Move _head_ to start search for next from here.
1162 * But be careful: shmem_evict_inode checks list_empty without taking
1163 * mutex, and there's an instant in list_move_tail when info->swaplist
1164 * would appear empty, if it were the only one on shmem_swaplist.
1166 if (shmem_swaplist.next != &info->swaplist)
1167 list_move_tail(&shmem_swaplist, &info->swaplist);
1169 gfp = mapping_gfp_mask(mapping);
1170 if (shmem_should_replace_page(*pagep, gfp)) {
1171 mutex_unlock(&shmem_swaplist_mutex);
1172 error = shmem_replace_page(pagep, gfp, info, index);
1173 mutex_lock(&shmem_swaplist_mutex);
1175 * We needed to drop mutex to make that restrictive page
1176 * allocation, but the inode might have been freed while we
1177 * dropped it: although a racing shmem_evict_inode() cannot
1178 * complete without emptying the radix_tree, our page lock
1179 * on this swapcache page is not enough to prevent that -
1180 * free_swap_and_cache() of our swap entry will only
1181 * trylock_page(), removing swap from radix_tree whatever.
1183 * We must not proceed to shmem_add_to_page_cache() if the
1184 * inode has been freed, but of course we cannot rely on
1185 * inode or mapping or info to check that. However, we can
1186 * safely check if our swap entry is still in use (and here
1187 * it can't have got reused for another page): if it's still
1188 * in use, then the inode cannot have been freed yet, and we
1189 * can safely proceed (if it's no longer in use, that tells
1190 * nothing about the inode, but we don't need to unuse swap).
1192 if (!page_swapcount(*pagep))
1197 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1198 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1199 * beneath us (pagelock doesn't help until the page is in pagecache).
1202 error = shmem_add_to_page_cache(*pagep, mapping, index,
1204 if (error != -ENOMEM) {
1206 * Truncation and eviction use free_swap_and_cache(), which
1207 * only does trylock page: if we raced, best clean up here.
1209 delete_from_swap_cache(*pagep);
1210 set_page_dirty(*pagep);
1212 spin_lock_irq(&info->lock);
1214 spin_unlock_irq(&info->lock);
1222 * Search through swapped inodes to find and replace swap by page.
1224 int shmem_unuse(swp_entry_t swap, struct page *page)
1226 struct list_head *this, *next;
1227 struct shmem_inode_info *info;
1228 struct mem_cgroup *memcg;
1232 * There's a faint possibility that swap page was replaced before
1233 * caller locked it: caller will come back later with the right page.
1235 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1239 * Charge page using GFP_KERNEL while we can wait, before taking
1240 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1241 * Charged back to the user (not to caller) when swap account is used.
1243 error = mem_cgroup_try_charge_delay(page, current->mm, GFP_KERNEL,
1247 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1250 mutex_lock(&shmem_swaplist_mutex);
1251 list_for_each_safe(this, next, &shmem_swaplist) {
1252 info = list_entry(this, struct shmem_inode_info, swaplist);
1254 error = shmem_unuse_inode(info, swap, &page);
1256 list_del_init(&info->swaplist);
1258 if (error != -EAGAIN)
1260 /* found nothing in this: move on to search the next */
1262 mutex_unlock(&shmem_swaplist_mutex);
1265 if (error != -ENOMEM)
1267 mem_cgroup_cancel_charge(page, memcg, false);
1269 mem_cgroup_commit_charge(page, memcg, true, false);
1277 * Move the page from the page cache to the swap cache.
1279 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1281 struct shmem_inode_info *info;
1282 struct address_space *mapping;
1283 struct inode *inode;
1287 VM_BUG_ON_PAGE(PageCompound(page), page);
1288 BUG_ON(!PageLocked(page));
1289 mapping = page->mapping;
1290 index = page->index;
1291 inode = mapping->host;
1292 info = SHMEM_I(inode);
1293 if (info->flags & VM_LOCKED)
1295 if (!total_swap_pages)
1299 * Our capabilities prevent regular writeback or sync from ever calling
1300 * shmem_writepage; but a stacking filesystem might use ->writepage of
1301 * its underlying filesystem, in which case tmpfs should write out to
1302 * swap only in response to memory pressure, and not for the writeback
1305 if (!wbc->for_reclaim) {
1306 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1311 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1312 * value into swapfile.c, the only way we can correctly account for a
1313 * fallocated page arriving here is now to initialize it and write it.
1315 * That's okay for a page already fallocated earlier, but if we have
1316 * not yet completed the fallocation, then (a) we want to keep track
1317 * of this page in case we have to undo it, and (b) it may not be a
1318 * good idea to continue anyway, once we're pushing into swap. So
1319 * reactivate the page, and let shmem_fallocate() quit when too many.
1321 if (!PageUptodate(page)) {
1322 if (inode->i_private) {
1323 struct shmem_falloc *shmem_falloc;
1324 spin_lock(&inode->i_lock);
1325 shmem_falloc = inode->i_private;
1327 !shmem_falloc->waitq &&
1328 index >= shmem_falloc->start &&
1329 index < shmem_falloc->next)
1330 shmem_falloc->nr_unswapped++;
1332 shmem_falloc = NULL;
1333 spin_unlock(&inode->i_lock);
1337 clear_highpage(page);
1338 flush_dcache_page(page);
1339 SetPageUptodate(page);
1342 swap = get_swap_page(page);
1347 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1348 * if it's not already there. Do it now before the page is
1349 * moved to swap cache, when its pagelock no longer protects
1350 * the inode from eviction. But don't unlock the mutex until
1351 * we've incremented swapped, because shmem_unuse_inode() will
1352 * prune a !swapped inode from the swaplist under this mutex.
1354 mutex_lock(&shmem_swaplist_mutex);
1355 if (list_empty(&info->swaplist))
1356 list_add_tail(&info->swaplist, &shmem_swaplist);
1358 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1359 spin_lock_irq(&info->lock);
1360 shmem_recalc_inode(inode);
1362 spin_unlock_irq(&info->lock);
1364 swap_shmem_alloc(swap);
1365 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1367 mutex_unlock(&shmem_swaplist_mutex);
1368 BUG_ON(page_mapped(page));
1369 swap_writepage(page, wbc);
1373 mutex_unlock(&shmem_swaplist_mutex);
1374 put_swap_page(page, swap);
1376 set_page_dirty(page);
1377 if (wbc->for_reclaim)
1378 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1383 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1384 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1388 if (!mpol || mpol->mode == MPOL_DEFAULT)
1389 return; /* show nothing */
1391 mpol_to_str(buffer, sizeof(buffer), mpol);
1393 seq_printf(seq, ",mpol=%s", buffer);
1396 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1398 struct mempolicy *mpol = NULL;
1400 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1401 mpol = sbinfo->mpol;
1403 spin_unlock(&sbinfo->stat_lock);
1407 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1408 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1411 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1415 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1417 #define vm_policy vm_private_data
1420 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1421 struct shmem_inode_info *info, pgoff_t index)
1423 /* Create a pseudo vma that just contains the policy */
1424 vma_init(vma, NULL);
1425 /* Bias interleave by inode number to distribute better across nodes */
1426 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1427 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1430 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1432 /* Drop reference taken by mpol_shared_policy_lookup() */
1433 mpol_cond_put(vma->vm_policy);
1436 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1437 struct shmem_inode_info *info, pgoff_t index)
1439 struct vm_area_struct pvma;
1441 struct vm_fault vmf;
1443 shmem_pseudo_vma_init(&pvma, info, index);
1446 page = swap_cluster_readahead(swap, gfp, &vmf);
1447 shmem_pseudo_vma_destroy(&pvma);
1452 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1453 struct shmem_inode_info *info, pgoff_t index)
1455 struct vm_area_struct pvma;
1456 struct inode *inode = &info->vfs_inode;
1457 struct address_space *mapping = inode->i_mapping;
1458 pgoff_t idx, hindex;
1459 void __rcu **results;
1462 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1465 hindex = round_down(index, HPAGE_PMD_NR);
1467 if (radix_tree_gang_lookup_slot(&mapping->i_pages, &results, &idx,
1468 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1474 shmem_pseudo_vma_init(&pvma, info, hindex);
1475 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1476 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1477 shmem_pseudo_vma_destroy(&pvma);
1479 prep_transhuge_page(page);
1483 static struct page *shmem_alloc_page(gfp_t gfp,
1484 struct shmem_inode_info *info, pgoff_t index)
1486 struct vm_area_struct pvma;
1489 shmem_pseudo_vma_init(&pvma, info, index);
1490 page = alloc_page_vma(gfp, &pvma, 0);
1491 shmem_pseudo_vma_destroy(&pvma);
1496 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1497 struct inode *inode,
1498 pgoff_t index, bool huge)
1500 struct shmem_inode_info *info = SHMEM_I(inode);
1505 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1507 nr = huge ? HPAGE_PMD_NR : 1;
1509 if (!shmem_inode_acct_block(inode, nr))
1513 page = shmem_alloc_hugepage(gfp, info, index);
1515 page = shmem_alloc_page(gfp, info, index);
1517 __SetPageLocked(page);
1518 __SetPageSwapBacked(page);
1523 shmem_inode_unacct_blocks(inode, nr);
1525 return ERR_PTR(err);
1529 * When a page is moved from swapcache to shmem filecache (either by the
1530 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1531 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1532 * ignorance of the mapping it belongs to. If that mapping has special
1533 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1534 * we may need to copy to a suitable page before moving to filecache.
1536 * In a future release, this may well be extended to respect cpuset and
1537 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1538 * but for now it is a simple matter of zone.
1540 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1542 return page_zonenum(page) > gfp_zone(gfp);
1545 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1546 struct shmem_inode_info *info, pgoff_t index)
1548 struct page *oldpage, *newpage;
1549 struct address_space *swap_mapping;
1554 swap_index = page_private(oldpage);
1555 swap_mapping = page_mapping(oldpage);
1558 * We have arrived here because our zones are constrained, so don't
1559 * limit chance of success by further cpuset and node constraints.
1561 gfp &= ~GFP_CONSTRAINT_MASK;
1562 newpage = shmem_alloc_page(gfp, info, index);
1567 copy_highpage(newpage, oldpage);
1568 flush_dcache_page(newpage);
1570 __SetPageLocked(newpage);
1571 __SetPageSwapBacked(newpage);
1572 SetPageUptodate(newpage);
1573 set_page_private(newpage, swap_index);
1574 SetPageSwapCache(newpage);
1577 * Our caller will very soon move newpage out of swapcache, but it's
1578 * a nice clean interface for us to replace oldpage by newpage there.
1580 xa_lock_irq(&swap_mapping->i_pages);
1581 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1584 __inc_node_page_state(newpage, NR_FILE_PAGES);
1585 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1587 xa_unlock_irq(&swap_mapping->i_pages);
1589 if (unlikely(error)) {
1591 * Is this possible? I think not, now that our callers check
1592 * both PageSwapCache and page_private after getting page lock;
1593 * but be defensive. Reverse old to newpage for clear and free.
1597 mem_cgroup_migrate(oldpage, newpage);
1598 lru_cache_add_anon(newpage);
1602 ClearPageSwapCache(oldpage);
1603 set_page_private(oldpage, 0);
1605 unlock_page(oldpage);
1612 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1614 * If we allocate a new one we do not mark it dirty. That's up to the
1615 * vm. If we swap it in we mark it dirty since we also free the swap
1616 * entry since a page cannot live in both the swap and page cache.
1618 * fault_mm and fault_type are only supplied by shmem_fault:
1619 * otherwise they are NULL.
1621 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1622 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1623 struct vm_area_struct *vma, struct vm_fault *vmf,
1624 vm_fault_t *fault_type)
1626 struct address_space *mapping = inode->i_mapping;
1627 struct shmem_inode_info *info = SHMEM_I(inode);
1628 struct shmem_sb_info *sbinfo;
1629 struct mm_struct *charge_mm;
1630 struct mem_cgroup *memcg;
1633 enum sgp_type sgp_huge = sgp;
1634 pgoff_t hindex = index;
1639 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1641 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1645 page = find_lock_entry(mapping, index);
1646 if (radix_tree_exceptional_entry(page)) {
1647 swap = radix_to_swp_entry(page);
1651 if (sgp <= SGP_CACHE &&
1652 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1657 if (page && sgp == SGP_WRITE)
1658 mark_page_accessed(page);
1660 /* fallocated page? */
1661 if (page && !PageUptodate(page)) {
1662 if (sgp != SGP_READ)
1668 if (page || (sgp == SGP_READ && !swap.val)) {
1674 * Fast cache lookup did not find it:
1675 * bring it back from swap or allocate.
1677 sbinfo = SHMEM_SB(inode->i_sb);
1678 charge_mm = vma ? vma->vm_mm : current->mm;
1681 /* Look it up and read it in.. */
1682 page = lookup_swap_cache(swap, NULL, 0);
1684 /* Or update major stats only when swapin succeeds?? */
1686 *fault_type |= VM_FAULT_MAJOR;
1687 count_vm_event(PGMAJFAULT);
1688 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1690 /* Here we actually start the io */
1691 page = shmem_swapin(swap, gfp, info, index);
1698 /* We have to do this with page locked to prevent races */
1700 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1701 !shmem_confirm_swap(mapping, index, swap)) {
1702 error = -EEXIST; /* try again */
1705 if (!PageUptodate(page)) {
1709 wait_on_page_writeback(page);
1711 if (shmem_should_replace_page(page, gfp)) {
1712 error = shmem_replace_page(&page, gfp, info, index);
1717 error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1720 error = shmem_add_to_page_cache(page, mapping, index,
1721 swp_to_radix_entry(swap));
1723 * We already confirmed swap under page lock, and make
1724 * no memory allocation here, so usually no possibility
1725 * of error; but free_swap_and_cache() only trylocks a
1726 * page, so it is just possible that the entry has been
1727 * truncated or holepunched since swap was confirmed.
1728 * shmem_undo_range() will have done some of the
1729 * unaccounting, now delete_from_swap_cache() will do
1731 * Reset swap.val? No, leave it so "failed" goes back to
1732 * "repeat": reading a hole and writing should succeed.
1735 mem_cgroup_cancel_charge(page, memcg, false);
1736 delete_from_swap_cache(page);
1742 mem_cgroup_commit_charge(page, memcg, true, false);
1744 spin_lock_irq(&info->lock);
1746 shmem_recalc_inode(inode);
1747 spin_unlock_irq(&info->lock);
1749 if (sgp == SGP_WRITE)
1750 mark_page_accessed(page);
1752 delete_from_swap_cache(page);
1753 set_page_dirty(page);
1757 if (vma && userfaultfd_missing(vma)) {
1758 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1762 /* shmem_symlink() */
1763 if (mapping->a_ops != &shmem_aops)
1765 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1767 if (shmem_huge == SHMEM_HUGE_FORCE)
1769 switch (sbinfo->huge) {
1772 case SHMEM_HUGE_NEVER:
1774 case SHMEM_HUGE_WITHIN_SIZE:
1775 off = round_up(index, HPAGE_PMD_NR);
1776 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1777 if (i_size >= HPAGE_PMD_SIZE &&
1778 i_size >> PAGE_SHIFT >= off)
1781 case SHMEM_HUGE_ADVISE:
1782 if (sgp_huge == SGP_HUGE)
1784 /* TODO: implement fadvise() hints */
1789 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1791 alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode,
1796 error = PTR_ERR(page);
1798 if (error != -ENOSPC)
1801 * Try to reclaim some spece by splitting a huge page
1802 * beyond i_size on the filesystem.
1806 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1807 if (ret == SHRINK_STOP)
1815 if (PageTransHuge(page))
1816 hindex = round_down(index, HPAGE_PMD_NR);
1820 if (sgp == SGP_WRITE)
1821 __SetPageReferenced(page);
1823 error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1824 PageTransHuge(page));
1827 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1828 compound_order(page));
1830 error = shmem_add_to_page_cache(page, mapping, hindex,
1832 radix_tree_preload_end();
1835 mem_cgroup_cancel_charge(page, memcg,
1836 PageTransHuge(page));
1839 mem_cgroup_commit_charge(page, memcg, false,
1840 PageTransHuge(page));
1841 lru_cache_add_anon(page);
1843 spin_lock_irq(&info->lock);
1844 info->alloced += 1 << compound_order(page);
1845 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1846 shmem_recalc_inode(inode);
1847 spin_unlock_irq(&info->lock);
1850 if (PageTransHuge(page) &&
1851 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1852 hindex + HPAGE_PMD_NR - 1) {
1854 * Part of the huge page is beyond i_size: subject
1855 * to shrink under memory pressure.
1857 spin_lock(&sbinfo->shrinklist_lock);
1859 * _careful to defend against unlocked access to
1860 * ->shrink_list in shmem_unused_huge_shrink()
1862 if (list_empty_careful(&info->shrinklist)) {
1863 list_add_tail(&info->shrinklist,
1864 &sbinfo->shrinklist);
1865 sbinfo->shrinklist_len++;
1867 spin_unlock(&sbinfo->shrinklist_lock);
1871 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1873 if (sgp == SGP_FALLOC)
1877 * Let SGP_WRITE caller clear ends if write does not fill page;
1878 * but SGP_FALLOC on a page fallocated earlier must initialize
1879 * it now, lest undo on failure cancel our earlier guarantee.
1881 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1882 struct page *head = compound_head(page);
1885 for (i = 0; i < (1 << compound_order(head)); i++) {
1886 clear_highpage(head + i);
1887 flush_dcache_page(head + i);
1889 SetPageUptodate(head);
1893 /* Perhaps the file has been truncated since we checked */
1894 if (sgp <= SGP_CACHE &&
1895 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1897 ClearPageDirty(page);
1898 delete_from_page_cache(page);
1899 spin_lock_irq(&info->lock);
1900 shmem_recalc_inode(inode);
1901 spin_unlock_irq(&info->lock);
1906 *pagep = page + index - hindex;
1913 shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
1915 if (PageTransHuge(page)) {
1921 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1928 if (error == -ENOSPC && !once++) {
1929 spin_lock_irq(&info->lock);
1930 shmem_recalc_inode(inode);
1931 spin_unlock_irq(&info->lock);
1934 if (error == -EEXIST) /* from above or from radix_tree_insert */
1940 * This is like autoremove_wake_function, but it removes the wait queue
1941 * entry unconditionally - even if something else had already woken the
1944 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1946 int ret = default_wake_function(wait, mode, sync, key);
1947 list_del_init(&wait->entry);
1951 static vm_fault_t shmem_fault(struct vm_fault *vmf)
1953 struct vm_area_struct *vma = vmf->vma;
1954 struct inode *inode = file_inode(vma->vm_file);
1955 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1958 vm_fault_t ret = VM_FAULT_LOCKED;
1961 * Trinity finds that probing a hole which tmpfs is punching can
1962 * prevent the hole-punch from ever completing: which in turn
1963 * locks writers out with its hold on i_mutex. So refrain from
1964 * faulting pages into the hole while it's being punched. Although
1965 * shmem_undo_range() does remove the additions, it may be unable to
1966 * keep up, as each new page needs its own unmap_mapping_range() call,
1967 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1969 * It does not matter if we sometimes reach this check just before the
1970 * hole-punch begins, so that one fault then races with the punch:
1971 * we just need to make racing faults a rare case.
1973 * The implementation below would be much simpler if we just used a
1974 * standard mutex or completion: but we cannot take i_mutex in fault,
1975 * and bloating every shmem inode for this unlikely case would be sad.
1977 if (unlikely(inode->i_private)) {
1978 struct shmem_falloc *shmem_falloc;
1980 spin_lock(&inode->i_lock);
1981 shmem_falloc = inode->i_private;
1983 shmem_falloc->waitq &&
1984 vmf->pgoff >= shmem_falloc->start &&
1985 vmf->pgoff < shmem_falloc->next) {
1986 wait_queue_head_t *shmem_falloc_waitq;
1987 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1989 ret = VM_FAULT_NOPAGE;
1990 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1991 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1992 /* It's polite to up mmap_sem if we can */
1993 up_read(&vma->vm_mm->mmap_sem);
1994 ret = VM_FAULT_RETRY;
1997 shmem_falloc_waitq = shmem_falloc->waitq;
1998 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1999 TASK_UNINTERRUPTIBLE);
2000 spin_unlock(&inode->i_lock);
2004 * shmem_falloc_waitq points into the shmem_fallocate()
2005 * stack of the hole-punching task: shmem_falloc_waitq
2006 * is usually invalid by the time we reach here, but
2007 * finish_wait() does not dereference it in that case;
2008 * though i_lock needed lest racing with wake_up_all().
2010 spin_lock(&inode->i_lock);
2011 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2012 spin_unlock(&inode->i_lock);
2015 spin_unlock(&inode->i_lock);
2020 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2021 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2023 else if (vma->vm_flags & VM_HUGEPAGE)
2026 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2027 gfp, vma, vmf, &ret);
2029 return vmf_error(err);
2033 unsigned long shmem_get_unmapped_area(struct file *file,
2034 unsigned long uaddr, unsigned long len,
2035 unsigned long pgoff, unsigned long flags)
2037 unsigned long (*get_area)(struct file *,
2038 unsigned long, unsigned long, unsigned long, unsigned long);
2040 unsigned long offset;
2041 unsigned long inflated_len;
2042 unsigned long inflated_addr;
2043 unsigned long inflated_offset;
2045 if (len > TASK_SIZE)
2048 get_area = current->mm->get_unmapped_area;
2049 addr = get_area(file, uaddr, len, pgoff, flags);
2051 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2053 if (IS_ERR_VALUE(addr))
2055 if (addr & ~PAGE_MASK)
2057 if (addr > TASK_SIZE - len)
2060 if (shmem_huge == SHMEM_HUGE_DENY)
2062 if (len < HPAGE_PMD_SIZE)
2064 if (flags & MAP_FIXED)
2067 * Our priority is to support MAP_SHARED mapped hugely;
2068 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2069 * But if caller specified an address hint, respect that as before.
2074 if (shmem_huge != SHMEM_HUGE_FORCE) {
2075 struct super_block *sb;
2078 VM_BUG_ON(file->f_op != &shmem_file_operations);
2079 sb = file_inode(file)->i_sb;
2082 * Called directly from mm/mmap.c, or drivers/char/mem.c
2083 * for "/dev/zero", to create a shared anonymous object.
2085 if (IS_ERR(shm_mnt))
2087 sb = shm_mnt->mnt_sb;
2089 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2093 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2094 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2096 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2099 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2100 if (inflated_len > TASK_SIZE)
2102 if (inflated_len < len)
2105 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2106 if (IS_ERR_VALUE(inflated_addr))
2108 if (inflated_addr & ~PAGE_MASK)
2111 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2112 inflated_addr += offset - inflated_offset;
2113 if (inflated_offset > offset)
2114 inflated_addr += HPAGE_PMD_SIZE;
2116 if (inflated_addr > TASK_SIZE - len)
2118 return inflated_addr;
2122 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2124 struct inode *inode = file_inode(vma->vm_file);
2125 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2128 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2131 struct inode *inode = file_inode(vma->vm_file);
2134 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2135 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2139 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2141 struct inode *inode = file_inode(file);
2142 struct shmem_inode_info *info = SHMEM_I(inode);
2143 int retval = -ENOMEM;
2145 spin_lock_irq(&info->lock);
2146 if (lock && !(info->flags & VM_LOCKED)) {
2147 if (!user_shm_lock(inode->i_size, user))
2149 info->flags |= VM_LOCKED;
2150 mapping_set_unevictable(file->f_mapping);
2152 if (!lock && (info->flags & VM_LOCKED) && user) {
2153 user_shm_unlock(inode->i_size, user);
2154 info->flags &= ~VM_LOCKED;
2155 mapping_clear_unevictable(file->f_mapping);
2160 spin_unlock_irq(&info->lock);
2164 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2166 file_accessed(file);
2167 vma->vm_ops = &shmem_vm_ops;
2168 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2169 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2170 (vma->vm_end & HPAGE_PMD_MASK)) {
2171 khugepaged_enter(vma, vma->vm_flags);
2176 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2177 umode_t mode, dev_t dev, unsigned long flags)
2179 struct inode *inode;
2180 struct shmem_inode_info *info;
2181 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2183 if (shmem_reserve_inode(sb))
2186 inode = new_inode(sb);
2188 inode->i_ino = get_next_ino();
2189 inode_init_owner(inode, dir, mode);
2190 inode->i_blocks = 0;
2191 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2192 inode->i_generation = prandom_u32();
2193 info = SHMEM_I(inode);
2194 memset(info, 0, (char *)inode - (char *)info);
2195 spin_lock_init(&info->lock);
2196 info->seals = F_SEAL_SEAL;
2197 info->flags = flags & VM_NORESERVE;
2198 INIT_LIST_HEAD(&info->shrinklist);
2199 INIT_LIST_HEAD(&info->swaplist);
2200 simple_xattrs_init(&info->xattrs);
2201 cache_no_acl(inode);
2203 switch (mode & S_IFMT) {
2205 inode->i_op = &shmem_special_inode_operations;
2206 init_special_inode(inode, mode, dev);
2209 inode->i_mapping->a_ops = &shmem_aops;
2210 inode->i_op = &shmem_inode_operations;
2211 inode->i_fop = &shmem_file_operations;
2212 mpol_shared_policy_init(&info->policy,
2213 shmem_get_sbmpol(sbinfo));
2217 /* Some things misbehave if size == 0 on a directory */
2218 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2219 inode->i_op = &shmem_dir_inode_operations;
2220 inode->i_fop = &simple_dir_operations;
2224 * Must not load anything in the rbtree,
2225 * mpol_free_shared_policy will not be called.
2227 mpol_shared_policy_init(&info->policy, NULL);
2231 shmem_free_inode(sb);
2235 bool shmem_mapping(struct address_space *mapping)
2237 return mapping->a_ops == &shmem_aops;
2240 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2242 struct vm_area_struct *dst_vma,
2243 unsigned long dst_addr,
2244 unsigned long src_addr,
2246 struct page **pagep)
2248 struct inode *inode = file_inode(dst_vma->vm_file);
2249 struct shmem_inode_info *info = SHMEM_I(inode);
2250 struct address_space *mapping = inode->i_mapping;
2251 gfp_t gfp = mapping_gfp_mask(mapping);
2252 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2253 struct mem_cgroup *memcg;
2257 pte_t _dst_pte, *dst_pte;
2261 if (!shmem_inode_acct_block(inode, 1))
2265 page = shmem_alloc_page(gfp, info, pgoff);
2267 goto out_unacct_blocks;
2269 if (!zeropage) { /* mcopy_atomic */
2270 page_kaddr = kmap_atomic(page);
2271 ret = copy_from_user(page_kaddr,
2272 (const void __user *)src_addr,
2274 kunmap_atomic(page_kaddr);
2276 /* fallback to copy_from_user outside mmap_sem */
2277 if (unlikely(ret)) {
2279 shmem_inode_unacct_blocks(inode, 1);
2280 /* don't free the page */
2283 } else { /* mfill_zeropage_atomic */
2284 clear_highpage(page);
2291 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2292 __SetPageLocked(page);
2293 __SetPageSwapBacked(page);
2294 __SetPageUptodate(page);
2296 ret = mem_cgroup_try_charge_delay(page, dst_mm, gfp, &memcg, false);
2300 ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
2302 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL);
2303 radix_tree_preload_end();
2306 goto out_release_uncharge;
2308 mem_cgroup_commit_charge(page, memcg, false, false);
2310 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2311 if (dst_vma->vm_flags & VM_WRITE)
2312 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2315 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2316 if (!pte_none(*dst_pte))
2317 goto out_release_uncharge_unlock;
2319 lru_cache_add_anon(page);
2321 spin_lock(&info->lock);
2323 inode->i_blocks += BLOCKS_PER_PAGE;
2324 shmem_recalc_inode(inode);
2325 spin_unlock(&info->lock);
2327 inc_mm_counter(dst_mm, mm_counter_file(page));
2328 page_add_file_rmap(page, false);
2329 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2331 /* No need to invalidate - it was non-present before */
2332 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2334 pte_unmap_unlock(dst_pte, ptl);
2338 out_release_uncharge_unlock:
2339 pte_unmap_unlock(dst_pte, ptl);
2340 out_release_uncharge:
2341 mem_cgroup_cancel_charge(page, memcg, false);
2346 shmem_inode_unacct_blocks(inode, 1);
2350 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2352 struct vm_area_struct *dst_vma,
2353 unsigned long dst_addr,
2354 unsigned long src_addr,
2355 struct page **pagep)
2357 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2358 dst_addr, src_addr, false, pagep);
2361 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2363 struct vm_area_struct *dst_vma,
2364 unsigned long dst_addr)
2366 struct page *page = NULL;
2368 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2369 dst_addr, 0, true, &page);
2373 static const struct inode_operations shmem_symlink_inode_operations;
2374 static const struct inode_operations shmem_short_symlink_operations;
2376 #ifdef CONFIG_TMPFS_XATTR
2377 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2379 #define shmem_initxattrs NULL
2383 shmem_write_begin(struct file *file, struct address_space *mapping,
2384 loff_t pos, unsigned len, unsigned flags,
2385 struct page **pagep, void **fsdata)
2387 struct inode *inode = mapping->host;
2388 struct shmem_inode_info *info = SHMEM_I(inode);
2389 pgoff_t index = pos >> PAGE_SHIFT;
2391 /* i_mutex is held by caller */
2392 if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) {
2393 if (info->seals & F_SEAL_WRITE)
2395 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2399 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2403 shmem_write_end(struct file *file, struct address_space *mapping,
2404 loff_t pos, unsigned len, unsigned copied,
2405 struct page *page, void *fsdata)
2407 struct inode *inode = mapping->host;
2409 if (pos + copied > inode->i_size)
2410 i_size_write(inode, pos + copied);
2412 if (!PageUptodate(page)) {
2413 struct page *head = compound_head(page);
2414 if (PageTransCompound(page)) {
2417 for (i = 0; i < HPAGE_PMD_NR; i++) {
2418 if (head + i == page)
2420 clear_highpage(head + i);
2421 flush_dcache_page(head + i);
2424 if (copied < PAGE_SIZE) {
2425 unsigned from = pos & (PAGE_SIZE - 1);
2426 zero_user_segments(page, 0, from,
2427 from + copied, PAGE_SIZE);
2429 SetPageUptodate(head);
2431 set_page_dirty(page);
2438 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2440 struct file *file = iocb->ki_filp;
2441 struct inode *inode = file_inode(file);
2442 struct address_space *mapping = inode->i_mapping;
2444 unsigned long offset;
2445 enum sgp_type sgp = SGP_READ;
2448 loff_t *ppos = &iocb->ki_pos;
2451 * Might this read be for a stacking filesystem? Then when reading
2452 * holes of a sparse file, we actually need to allocate those pages,
2453 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2455 if (!iter_is_iovec(to))
2458 index = *ppos >> PAGE_SHIFT;
2459 offset = *ppos & ~PAGE_MASK;
2462 struct page *page = NULL;
2464 unsigned long nr, ret;
2465 loff_t i_size = i_size_read(inode);
2467 end_index = i_size >> PAGE_SHIFT;
2468 if (index > end_index)
2470 if (index == end_index) {
2471 nr = i_size & ~PAGE_MASK;
2476 error = shmem_getpage(inode, index, &page, sgp);
2478 if (error == -EINVAL)
2483 if (sgp == SGP_CACHE)
2484 set_page_dirty(page);
2489 * We must evaluate after, since reads (unlike writes)
2490 * are called without i_mutex protection against truncate
2493 i_size = i_size_read(inode);
2494 end_index = i_size >> PAGE_SHIFT;
2495 if (index == end_index) {
2496 nr = i_size & ~PAGE_MASK;
2507 * If users can be writing to this page using arbitrary
2508 * virtual addresses, take care about potential aliasing
2509 * before reading the page on the kernel side.
2511 if (mapping_writably_mapped(mapping))
2512 flush_dcache_page(page);
2514 * Mark the page accessed if we read the beginning.
2517 mark_page_accessed(page);
2519 page = ZERO_PAGE(0);
2524 * Ok, we have the page, and it's up-to-date, so
2525 * now we can copy it to user space...
2527 ret = copy_page_to_iter(page, offset, nr, to);
2530 index += offset >> PAGE_SHIFT;
2531 offset &= ~PAGE_MASK;
2534 if (!iov_iter_count(to))
2543 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2544 file_accessed(file);
2545 return retval ? retval : error;
2549 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2551 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2552 pgoff_t index, pgoff_t end, int whence)
2555 struct pagevec pvec;
2556 pgoff_t indices[PAGEVEC_SIZE];
2560 pagevec_init(&pvec);
2561 pvec.nr = 1; /* start small: we may be there already */
2563 pvec.nr = find_get_entries(mapping, index,
2564 pvec.nr, pvec.pages, indices);
2566 if (whence == SEEK_DATA)
2570 for (i = 0; i < pvec.nr; i++, index++) {
2571 if (index < indices[i]) {
2572 if (whence == SEEK_HOLE) {
2578 page = pvec.pages[i];
2579 if (page && !radix_tree_exceptional_entry(page)) {
2580 if (!PageUptodate(page))
2584 (page && whence == SEEK_DATA) ||
2585 (!page && whence == SEEK_HOLE)) {
2590 pagevec_remove_exceptionals(&pvec);
2591 pagevec_release(&pvec);
2592 pvec.nr = PAGEVEC_SIZE;
2598 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2600 struct address_space *mapping = file->f_mapping;
2601 struct inode *inode = mapping->host;
2605 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2606 return generic_file_llseek_size(file, offset, whence,
2607 MAX_LFS_FILESIZE, i_size_read(inode));
2609 /* We're holding i_mutex so we can access i_size directly */
2613 else if (offset >= inode->i_size)
2616 start = offset >> PAGE_SHIFT;
2617 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2618 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2619 new_offset <<= PAGE_SHIFT;
2620 if (new_offset > offset) {
2621 if (new_offset < inode->i_size)
2622 offset = new_offset;
2623 else if (whence == SEEK_DATA)
2626 offset = inode->i_size;
2631 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2632 inode_unlock(inode);
2636 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2639 struct inode *inode = file_inode(file);
2640 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2641 struct shmem_inode_info *info = SHMEM_I(inode);
2642 struct shmem_falloc shmem_falloc;
2643 pgoff_t start, index, end;
2646 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2651 if (mode & FALLOC_FL_PUNCH_HOLE) {
2652 struct address_space *mapping = file->f_mapping;
2653 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2654 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2655 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2657 /* protected by i_mutex */
2658 if (info->seals & F_SEAL_WRITE) {
2663 shmem_falloc.waitq = &shmem_falloc_waitq;
2664 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2665 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2666 spin_lock(&inode->i_lock);
2667 inode->i_private = &shmem_falloc;
2668 spin_unlock(&inode->i_lock);
2670 if ((u64)unmap_end > (u64)unmap_start)
2671 unmap_mapping_range(mapping, unmap_start,
2672 1 + unmap_end - unmap_start, 0);
2673 shmem_truncate_range(inode, offset, offset + len - 1);
2674 /* No need to unmap again: hole-punching leaves COWed pages */
2676 spin_lock(&inode->i_lock);
2677 inode->i_private = NULL;
2678 wake_up_all(&shmem_falloc_waitq);
2679 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2680 spin_unlock(&inode->i_lock);
2685 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2686 error = inode_newsize_ok(inode, offset + len);
2690 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2695 start = offset >> PAGE_SHIFT;
2696 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2697 /* Try to avoid a swapstorm if len is impossible to satisfy */
2698 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2703 shmem_falloc.waitq = NULL;
2704 shmem_falloc.start = start;
2705 shmem_falloc.next = start;
2706 shmem_falloc.nr_falloced = 0;
2707 shmem_falloc.nr_unswapped = 0;
2708 spin_lock(&inode->i_lock);
2709 inode->i_private = &shmem_falloc;
2710 spin_unlock(&inode->i_lock);
2712 for (index = start; index < end; index++) {
2716 * Good, the fallocate(2) manpage permits EINTR: we may have
2717 * been interrupted because we are using up too much memory.
2719 if (signal_pending(current))
2721 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2724 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2726 /* Remove the !PageUptodate pages we added */
2727 if (index > start) {
2728 shmem_undo_range(inode,
2729 (loff_t)start << PAGE_SHIFT,
2730 ((loff_t)index << PAGE_SHIFT) - 1, true);
2736 * Inform shmem_writepage() how far we have reached.
2737 * No need for lock or barrier: we have the page lock.
2739 shmem_falloc.next++;
2740 if (!PageUptodate(page))
2741 shmem_falloc.nr_falloced++;
2744 * If !PageUptodate, leave it that way so that freeable pages
2745 * can be recognized if we need to rollback on error later.
2746 * But set_page_dirty so that memory pressure will swap rather
2747 * than free the pages we are allocating (and SGP_CACHE pages
2748 * might still be clean: we now need to mark those dirty too).
2750 set_page_dirty(page);
2756 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2757 i_size_write(inode, offset + len);
2758 inode->i_ctime = current_time(inode);
2760 spin_lock(&inode->i_lock);
2761 inode->i_private = NULL;
2762 spin_unlock(&inode->i_lock);
2764 inode_unlock(inode);
2768 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2770 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2772 buf->f_type = TMPFS_MAGIC;
2773 buf->f_bsize = PAGE_SIZE;
2774 buf->f_namelen = NAME_MAX;
2775 if (sbinfo->max_blocks) {
2776 buf->f_blocks = sbinfo->max_blocks;
2778 buf->f_bfree = sbinfo->max_blocks -
2779 percpu_counter_sum(&sbinfo->used_blocks);
2781 if (sbinfo->max_inodes) {
2782 buf->f_files = sbinfo->max_inodes;
2783 buf->f_ffree = sbinfo->free_inodes;
2785 /* else leave those fields 0 like simple_statfs */
2790 * File creation. Allocate an inode, and we're done..
2793 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2795 struct inode *inode;
2796 int error = -ENOSPC;
2798 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2800 error = simple_acl_create(dir, inode);
2803 error = security_inode_init_security(inode, dir,
2805 shmem_initxattrs, NULL);
2806 if (error && error != -EOPNOTSUPP)
2810 dir->i_size += BOGO_DIRENT_SIZE;
2811 dir->i_ctime = dir->i_mtime = current_time(dir);
2812 d_instantiate(dentry, inode);
2813 dget(dentry); /* Extra count - pin the dentry in core */
2822 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2824 struct inode *inode;
2825 int error = -ENOSPC;
2827 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2829 error = security_inode_init_security(inode, dir,
2831 shmem_initxattrs, NULL);
2832 if (error && error != -EOPNOTSUPP)
2834 error = simple_acl_create(dir, inode);
2837 d_tmpfile(dentry, inode);
2845 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2849 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2855 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2858 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2864 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2866 struct inode *inode = d_inode(old_dentry);
2870 * No ordinary (disk based) filesystem counts links as inodes;
2871 * but each new link needs a new dentry, pinning lowmem, and
2872 * tmpfs dentries cannot be pruned until they are unlinked.
2874 ret = shmem_reserve_inode(inode->i_sb);
2878 dir->i_size += BOGO_DIRENT_SIZE;
2879 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2881 ihold(inode); /* New dentry reference */
2882 dget(dentry); /* Extra pinning count for the created dentry */
2883 d_instantiate(dentry, inode);
2888 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2890 struct inode *inode = d_inode(dentry);
2892 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2893 shmem_free_inode(inode->i_sb);
2895 dir->i_size -= BOGO_DIRENT_SIZE;
2896 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2898 dput(dentry); /* Undo the count from "create" - this does all the work */
2902 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2904 if (!simple_empty(dentry))
2907 drop_nlink(d_inode(dentry));
2909 return shmem_unlink(dir, dentry);
2912 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2914 bool old_is_dir = d_is_dir(old_dentry);
2915 bool new_is_dir = d_is_dir(new_dentry);
2917 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2919 drop_nlink(old_dir);
2922 drop_nlink(new_dir);
2926 old_dir->i_ctime = old_dir->i_mtime =
2927 new_dir->i_ctime = new_dir->i_mtime =
2928 d_inode(old_dentry)->i_ctime =
2929 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2934 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2936 struct dentry *whiteout;
2939 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2943 error = shmem_mknod(old_dir, whiteout,
2944 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2950 * Cheat and hash the whiteout while the old dentry is still in
2951 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2953 * d_lookup() will consistently find one of them at this point,
2954 * not sure which one, but that isn't even important.
2961 * The VFS layer already does all the dentry stuff for rename,
2962 * we just have to decrement the usage count for the target if
2963 * it exists so that the VFS layer correctly free's it when it
2966 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
2968 struct inode *inode = d_inode(old_dentry);
2969 int they_are_dirs = S_ISDIR(inode->i_mode);
2971 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2974 if (flags & RENAME_EXCHANGE)
2975 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2977 if (!simple_empty(new_dentry))
2980 if (flags & RENAME_WHITEOUT) {
2983 error = shmem_whiteout(old_dir, old_dentry);
2988 if (d_really_is_positive(new_dentry)) {
2989 (void) shmem_unlink(new_dir, new_dentry);
2990 if (they_are_dirs) {
2991 drop_nlink(d_inode(new_dentry));
2992 drop_nlink(old_dir);
2994 } else if (they_are_dirs) {
2995 drop_nlink(old_dir);
2999 old_dir->i_size -= BOGO_DIRENT_SIZE;
3000 new_dir->i_size += BOGO_DIRENT_SIZE;
3001 old_dir->i_ctime = old_dir->i_mtime =
3002 new_dir->i_ctime = new_dir->i_mtime =
3003 inode->i_ctime = current_time(old_dir);
3007 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3011 struct inode *inode;
3014 len = strlen(symname) + 1;
3015 if (len > PAGE_SIZE)
3016 return -ENAMETOOLONG;
3018 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3023 error = security_inode_init_security(inode, dir, &dentry->d_name,
3024 shmem_initxattrs, NULL);
3026 if (error != -EOPNOTSUPP) {
3033 inode->i_size = len-1;
3034 if (len <= SHORT_SYMLINK_LEN) {
3035 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3036 if (!inode->i_link) {
3040 inode->i_op = &shmem_short_symlink_operations;
3042 inode_nohighmem(inode);
3043 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3048 inode->i_mapping->a_ops = &shmem_aops;
3049 inode->i_op = &shmem_symlink_inode_operations;
3050 memcpy(page_address(page), symname, len);
3051 SetPageUptodate(page);
3052 set_page_dirty(page);
3056 dir->i_size += BOGO_DIRENT_SIZE;
3057 dir->i_ctime = dir->i_mtime = current_time(dir);
3058 d_instantiate(dentry, inode);
3063 static void shmem_put_link(void *arg)
3065 mark_page_accessed(arg);
3069 static const char *shmem_get_link(struct dentry *dentry,
3070 struct inode *inode,
3071 struct delayed_call *done)
3073 struct page *page = NULL;
3076 page = find_get_page(inode->i_mapping, 0);
3078 return ERR_PTR(-ECHILD);
3079 if (!PageUptodate(page)) {
3081 return ERR_PTR(-ECHILD);
3084 error = shmem_getpage(inode, 0, &page, SGP_READ);
3086 return ERR_PTR(error);
3089 set_delayed_call(done, shmem_put_link, page);
3090 return page_address(page);
3093 #ifdef CONFIG_TMPFS_XATTR
3095 * Superblocks without xattr inode operations may get some security.* xattr
3096 * support from the LSM "for free". As soon as we have any other xattrs
3097 * like ACLs, we also need to implement the security.* handlers at
3098 * filesystem level, though.
3102 * Callback for security_inode_init_security() for acquiring xattrs.
3104 static int shmem_initxattrs(struct inode *inode,
3105 const struct xattr *xattr_array,
3108 struct shmem_inode_info *info = SHMEM_I(inode);
3109 const struct xattr *xattr;
3110 struct simple_xattr *new_xattr;
3113 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3114 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3118 len = strlen(xattr->name) + 1;
3119 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3121 if (!new_xattr->name) {
3126 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3127 XATTR_SECURITY_PREFIX_LEN);
3128 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3131 simple_xattr_list_add(&info->xattrs, new_xattr);
3137 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3138 struct dentry *unused, struct inode *inode,
3139 const char *name, void *buffer, size_t size)
3141 struct shmem_inode_info *info = SHMEM_I(inode);
3143 name = xattr_full_name(handler, name);
3144 return simple_xattr_get(&info->xattrs, name, buffer, size);
3147 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3148 struct dentry *unused, struct inode *inode,
3149 const char *name, const void *value,
3150 size_t size, int flags)
3152 struct shmem_inode_info *info = SHMEM_I(inode);
3154 name = xattr_full_name(handler, name);
3155 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3158 static const struct xattr_handler shmem_security_xattr_handler = {
3159 .prefix = XATTR_SECURITY_PREFIX,
3160 .get = shmem_xattr_handler_get,
3161 .set = shmem_xattr_handler_set,
3164 static const struct xattr_handler shmem_trusted_xattr_handler = {
3165 .prefix = XATTR_TRUSTED_PREFIX,
3166 .get = shmem_xattr_handler_get,
3167 .set = shmem_xattr_handler_set,
3170 static const struct xattr_handler *shmem_xattr_handlers[] = {
3171 #ifdef CONFIG_TMPFS_POSIX_ACL
3172 &posix_acl_access_xattr_handler,
3173 &posix_acl_default_xattr_handler,
3175 &shmem_security_xattr_handler,
3176 &shmem_trusted_xattr_handler,
3180 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3182 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3183 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3185 #endif /* CONFIG_TMPFS_XATTR */
3187 static const struct inode_operations shmem_short_symlink_operations = {
3188 .get_link = simple_get_link,
3189 #ifdef CONFIG_TMPFS_XATTR
3190 .listxattr = shmem_listxattr,
3194 static const struct inode_operations shmem_symlink_inode_operations = {
3195 .get_link = shmem_get_link,
3196 #ifdef CONFIG_TMPFS_XATTR
3197 .listxattr = shmem_listxattr,
3201 static struct dentry *shmem_get_parent(struct dentry *child)
3203 return ERR_PTR(-ESTALE);
3206 static int shmem_match(struct inode *ino, void *vfh)
3210 inum = (inum << 32) | fh[1];
3211 return ino->i_ino == inum && fh[0] == ino->i_generation;
3214 /* Find any alias of inode, but prefer a hashed alias */
3215 static struct dentry *shmem_find_alias(struct inode *inode)
3217 struct dentry *alias = d_find_alias(inode);
3219 return alias ?: d_find_any_alias(inode);
3223 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3224 struct fid *fid, int fh_len, int fh_type)
3226 struct inode *inode;
3227 struct dentry *dentry = NULL;
3234 inum = (inum << 32) | fid->raw[1];
3236 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3237 shmem_match, fid->raw);
3239 dentry = shmem_find_alias(inode);
3246 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3247 struct inode *parent)
3251 return FILEID_INVALID;
3254 if (inode_unhashed(inode)) {
3255 /* Unfortunately insert_inode_hash is not idempotent,
3256 * so as we hash inodes here rather than at creation
3257 * time, we need a lock to ensure we only try
3260 static DEFINE_SPINLOCK(lock);
3262 if (inode_unhashed(inode))
3263 __insert_inode_hash(inode,
3264 inode->i_ino + inode->i_generation);
3268 fh[0] = inode->i_generation;
3269 fh[1] = inode->i_ino;
3270 fh[2] = ((__u64)inode->i_ino) >> 32;
3276 static const struct export_operations shmem_export_ops = {
3277 .get_parent = shmem_get_parent,
3278 .encode_fh = shmem_encode_fh,
3279 .fh_to_dentry = shmem_fh_to_dentry,
3282 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3285 char *this_char, *value, *rest;
3286 struct mempolicy *mpol = NULL;
3290 while (options != NULL) {
3291 this_char = options;
3294 * NUL-terminate this option: unfortunately,
3295 * mount options form a comma-separated list,
3296 * but mpol's nodelist may also contain commas.
3298 options = strchr(options, ',');
3299 if (options == NULL)
3302 if (!isdigit(*options)) {
3309 if ((value = strchr(this_char,'=')) != NULL) {
3312 pr_err("tmpfs: No value for mount option '%s'\n",
3317 if (!strcmp(this_char,"size")) {
3318 unsigned long long size;
3319 size = memparse(value,&rest);
3321 size <<= PAGE_SHIFT;
3322 size *= totalram_pages;
3328 sbinfo->max_blocks =
3329 DIV_ROUND_UP(size, PAGE_SIZE);
3330 } else if (!strcmp(this_char,"nr_blocks")) {
3331 sbinfo->max_blocks = memparse(value, &rest);
3334 } else if (!strcmp(this_char,"nr_inodes")) {
3335 sbinfo->max_inodes = memparse(value, &rest);
3338 } else if (!strcmp(this_char,"mode")) {
3341 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3344 } else if (!strcmp(this_char,"uid")) {
3347 uid = simple_strtoul(value, &rest, 0);
3350 sbinfo->uid = make_kuid(current_user_ns(), uid);
3351 if (!uid_valid(sbinfo->uid))
3353 } else if (!strcmp(this_char,"gid")) {
3356 gid = simple_strtoul(value, &rest, 0);
3359 sbinfo->gid = make_kgid(current_user_ns(), gid);
3360 if (!gid_valid(sbinfo->gid))
3362 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3363 } else if (!strcmp(this_char, "huge")) {
3365 huge = shmem_parse_huge(value);
3368 if (!has_transparent_hugepage() &&
3369 huge != SHMEM_HUGE_NEVER)
3371 sbinfo->huge = huge;
3374 } else if (!strcmp(this_char,"mpol")) {
3377 if (mpol_parse_str(value, &mpol))
3381 pr_err("tmpfs: Bad mount option %s\n", this_char);
3385 sbinfo->mpol = mpol;
3389 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3397 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3399 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3400 struct shmem_sb_info config = *sbinfo;
3401 unsigned long inodes;
3402 int error = -EINVAL;
3405 if (shmem_parse_options(data, &config, true))
3408 spin_lock(&sbinfo->stat_lock);
3409 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3410 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3412 if (config.max_inodes < inodes)
3415 * Those tests disallow limited->unlimited while any are in use;
3416 * but we must separately disallow unlimited->limited, because
3417 * in that case we have no record of how much is already in use.
3419 if (config.max_blocks && !sbinfo->max_blocks)
3421 if (config.max_inodes && !sbinfo->max_inodes)
3425 sbinfo->huge = config.huge;
3426 sbinfo->max_blocks = config.max_blocks;
3427 sbinfo->max_inodes = config.max_inodes;
3428 sbinfo->free_inodes = config.max_inodes - inodes;
3431 * Preserve previous mempolicy unless mpol remount option was specified.
3434 mpol_put(sbinfo->mpol);
3435 sbinfo->mpol = config.mpol; /* transfers initial ref */
3438 spin_unlock(&sbinfo->stat_lock);
3442 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3444 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3446 if (sbinfo->max_blocks != shmem_default_max_blocks())
3447 seq_printf(seq, ",size=%luk",
3448 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3449 if (sbinfo->max_inodes != shmem_default_max_inodes())
3450 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3451 if (sbinfo->mode != (0777 | S_ISVTX))
3452 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3453 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3454 seq_printf(seq, ",uid=%u",
3455 from_kuid_munged(&init_user_ns, sbinfo->uid));
3456 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3457 seq_printf(seq, ",gid=%u",
3458 from_kgid_munged(&init_user_ns, sbinfo->gid));
3459 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3460 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3462 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3464 shmem_show_mpol(seq, sbinfo->mpol);
3468 #endif /* CONFIG_TMPFS */
3470 static void shmem_put_super(struct super_block *sb)
3472 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3474 percpu_counter_destroy(&sbinfo->used_blocks);
3475 mpol_put(sbinfo->mpol);
3477 sb->s_fs_info = NULL;
3480 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3482 struct inode *inode;
3483 struct shmem_sb_info *sbinfo;
3486 /* Round up to L1_CACHE_BYTES to resist false sharing */
3487 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3488 L1_CACHE_BYTES), GFP_KERNEL);
3492 sbinfo->mode = 0777 | S_ISVTX;
3493 sbinfo->uid = current_fsuid();
3494 sbinfo->gid = current_fsgid();
3495 sb->s_fs_info = sbinfo;
3499 * Per default we only allow half of the physical ram per
3500 * tmpfs instance, limiting inodes to one per page of lowmem;
3501 * but the internal instance is left unlimited.
3503 if (!(sb->s_flags & SB_KERNMOUNT)) {
3504 sbinfo->max_blocks = shmem_default_max_blocks();
3505 sbinfo->max_inodes = shmem_default_max_inodes();
3506 if (shmem_parse_options(data, sbinfo, false)) {
3511 sb->s_flags |= SB_NOUSER;
3513 sb->s_export_op = &shmem_export_ops;
3514 sb->s_flags |= SB_NOSEC;
3516 sb->s_flags |= SB_NOUSER;
3519 spin_lock_init(&sbinfo->stat_lock);
3520 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3522 sbinfo->free_inodes = sbinfo->max_inodes;
3523 spin_lock_init(&sbinfo->shrinklist_lock);
3524 INIT_LIST_HEAD(&sbinfo->shrinklist);
3526 sb->s_maxbytes = MAX_LFS_FILESIZE;
3527 sb->s_blocksize = PAGE_SIZE;
3528 sb->s_blocksize_bits = PAGE_SHIFT;
3529 sb->s_magic = TMPFS_MAGIC;
3530 sb->s_op = &shmem_ops;
3531 sb->s_time_gran = 1;
3532 #ifdef CONFIG_TMPFS_XATTR
3533 sb->s_xattr = shmem_xattr_handlers;
3535 #ifdef CONFIG_TMPFS_POSIX_ACL
3536 sb->s_flags |= SB_POSIXACL;
3538 uuid_gen(&sb->s_uuid);
3540 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3543 inode->i_uid = sbinfo->uid;
3544 inode->i_gid = sbinfo->gid;
3545 sb->s_root = d_make_root(inode);
3551 shmem_put_super(sb);
3555 static struct kmem_cache *shmem_inode_cachep;
3557 static struct inode *shmem_alloc_inode(struct super_block *sb)
3559 struct shmem_inode_info *info;
3560 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3563 return &info->vfs_inode;
3566 static void shmem_destroy_callback(struct rcu_head *head)
3568 struct inode *inode = container_of(head, struct inode, i_rcu);
3569 if (S_ISLNK(inode->i_mode))
3570 kfree(inode->i_link);
3571 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3574 static void shmem_destroy_inode(struct inode *inode)
3576 if (S_ISREG(inode->i_mode))
3577 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3578 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3581 static void shmem_init_inode(void *foo)
3583 struct shmem_inode_info *info = foo;
3584 inode_init_once(&info->vfs_inode);
3587 static void shmem_init_inodecache(void)
3589 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3590 sizeof(struct shmem_inode_info),
3591 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3594 static void shmem_destroy_inodecache(void)
3596 kmem_cache_destroy(shmem_inode_cachep);
3599 static const struct address_space_operations shmem_aops = {
3600 .writepage = shmem_writepage,
3601 .set_page_dirty = __set_page_dirty_no_writeback,
3603 .write_begin = shmem_write_begin,
3604 .write_end = shmem_write_end,
3606 #ifdef CONFIG_MIGRATION
3607 .migratepage = migrate_page,
3609 .error_remove_page = generic_error_remove_page,
3612 static const struct file_operations shmem_file_operations = {
3614 .get_unmapped_area = shmem_get_unmapped_area,
3616 .llseek = shmem_file_llseek,
3617 .read_iter = shmem_file_read_iter,
3618 .write_iter = generic_file_write_iter,
3619 .fsync = noop_fsync,
3620 .splice_read = generic_file_splice_read,
3621 .splice_write = iter_file_splice_write,
3622 .fallocate = shmem_fallocate,
3626 static const struct inode_operations shmem_inode_operations = {
3627 .getattr = shmem_getattr,
3628 .setattr = shmem_setattr,
3629 #ifdef CONFIG_TMPFS_XATTR
3630 .listxattr = shmem_listxattr,
3631 .set_acl = simple_set_acl,
3635 static const struct inode_operations shmem_dir_inode_operations = {
3637 .create = shmem_create,
3638 .lookup = simple_lookup,
3640 .unlink = shmem_unlink,
3641 .symlink = shmem_symlink,
3642 .mkdir = shmem_mkdir,
3643 .rmdir = shmem_rmdir,
3644 .mknod = shmem_mknod,
3645 .rename = shmem_rename2,
3646 .tmpfile = shmem_tmpfile,
3648 #ifdef CONFIG_TMPFS_XATTR
3649 .listxattr = shmem_listxattr,
3651 #ifdef CONFIG_TMPFS_POSIX_ACL
3652 .setattr = shmem_setattr,
3653 .set_acl = simple_set_acl,
3657 static const struct inode_operations shmem_special_inode_operations = {
3658 #ifdef CONFIG_TMPFS_XATTR
3659 .listxattr = shmem_listxattr,
3661 #ifdef CONFIG_TMPFS_POSIX_ACL
3662 .setattr = shmem_setattr,
3663 .set_acl = simple_set_acl,
3667 static const struct super_operations shmem_ops = {
3668 .alloc_inode = shmem_alloc_inode,
3669 .destroy_inode = shmem_destroy_inode,
3671 .statfs = shmem_statfs,
3672 .remount_fs = shmem_remount_fs,
3673 .show_options = shmem_show_options,
3675 .evict_inode = shmem_evict_inode,
3676 .drop_inode = generic_delete_inode,
3677 .put_super = shmem_put_super,
3678 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3679 .nr_cached_objects = shmem_unused_huge_count,
3680 .free_cached_objects = shmem_unused_huge_scan,
3684 static const struct vm_operations_struct shmem_vm_ops = {
3685 .fault = shmem_fault,
3686 .map_pages = filemap_map_pages,
3688 .set_policy = shmem_set_policy,
3689 .get_policy = shmem_get_policy,
3693 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3694 int flags, const char *dev_name, void *data)
3696 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3699 static struct file_system_type shmem_fs_type = {
3700 .owner = THIS_MODULE,
3702 .mount = shmem_mount,
3703 .kill_sb = kill_litter_super,
3704 .fs_flags = FS_USERNS_MOUNT,
3707 int __init shmem_init(void)
3711 /* If rootfs called this, don't re-init */
3712 if (shmem_inode_cachep)
3715 shmem_init_inodecache();
3717 error = register_filesystem(&shmem_fs_type);
3719 pr_err("Could not register tmpfs\n");
3723 shm_mnt = kern_mount(&shmem_fs_type);
3724 if (IS_ERR(shm_mnt)) {
3725 error = PTR_ERR(shm_mnt);
3726 pr_err("Could not kern_mount tmpfs\n");
3730 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3731 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3732 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3734 shmem_huge = 0; /* just in case it was patched */
3739 unregister_filesystem(&shmem_fs_type);
3741 shmem_destroy_inodecache();
3742 shm_mnt = ERR_PTR(error);
3746 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3747 static ssize_t shmem_enabled_show(struct kobject *kobj,
3748 struct kobj_attribute *attr, char *buf)
3752 SHMEM_HUGE_WITHIN_SIZE,
3760 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3761 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3763 count += sprintf(buf + count, fmt,
3764 shmem_format_huge(values[i]));
3766 buf[count - 1] = '\n';
3770 static ssize_t shmem_enabled_store(struct kobject *kobj,
3771 struct kobj_attribute *attr, const char *buf, size_t count)
3776 if (count + 1 > sizeof(tmp))
3778 memcpy(tmp, buf, count);
3780 if (count && tmp[count - 1] == '\n')
3781 tmp[count - 1] = '\0';
3783 huge = shmem_parse_huge(tmp);
3784 if (huge == -EINVAL)
3786 if (!has_transparent_hugepage() &&
3787 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3791 if (shmem_huge > SHMEM_HUGE_DENY)
3792 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3796 struct kobj_attribute shmem_enabled_attr =
3797 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3798 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3800 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3801 bool shmem_huge_enabled(struct vm_area_struct *vma)
3803 struct inode *inode = file_inode(vma->vm_file);
3804 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3808 if (shmem_huge == SHMEM_HUGE_FORCE)
3810 if (shmem_huge == SHMEM_HUGE_DENY)
3812 switch (sbinfo->huge) {
3813 case SHMEM_HUGE_NEVER:
3815 case SHMEM_HUGE_ALWAYS:
3817 case SHMEM_HUGE_WITHIN_SIZE:
3818 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3819 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3820 if (i_size >= HPAGE_PMD_SIZE &&
3821 i_size >> PAGE_SHIFT >= off)
3824 case SHMEM_HUGE_ADVISE:
3825 /* TODO: implement fadvise() hints */
3826 return (vma->vm_flags & VM_HUGEPAGE);
3832 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3834 #else /* !CONFIG_SHMEM */
3837 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3839 * This is intended for small system where the benefits of the full
3840 * shmem code (swap-backed and resource-limited) are outweighed by
3841 * their complexity. On systems without swap this code should be
3842 * effectively equivalent, but much lighter weight.
3845 static struct file_system_type shmem_fs_type = {
3847 .mount = ramfs_mount,
3848 .kill_sb = kill_litter_super,
3849 .fs_flags = FS_USERNS_MOUNT,
3852 int __init shmem_init(void)
3854 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3856 shm_mnt = kern_mount(&shmem_fs_type);
3857 BUG_ON(IS_ERR(shm_mnt));
3862 int shmem_unuse(swp_entry_t swap, struct page *page)
3867 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3872 void shmem_unlock_mapping(struct address_space *mapping)
3877 unsigned long shmem_get_unmapped_area(struct file *file,
3878 unsigned long addr, unsigned long len,
3879 unsigned long pgoff, unsigned long flags)
3881 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3885 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3887 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
3889 EXPORT_SYMBOL_GPL(shmem_truncate_range);
3891 #define shmem_vm_ops generic_file_vm_ops
3892 #define shmem_file_operations ramfs_file_operations
3893 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
3894 #define shmem_acct_size(flags, size) 0
3895 #define shmem_unacct_size(flags, size) do {} while (0)
3897 #endif /* CONFIG_SHMEM */
3901 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
3902 unsigned long flags, unsigned int i_flags)
3904 struct inode *inode;
3908 return ERR_CAST(mnt);
3910 if (size < 0 || size > MAX_LFS_FILESIZE)
3911 return ERR_PTR(-EINVAL);
3913 if (shmem_acct_size(flags, size))
3914 return ERR_PTR(-ENOMEM);
3916 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
3918 if (unlikely(!inode)) {
3919 shmem_unacct_size(flags, size);
3920 return ERR_PTR(-ENOSPC);
3922 inode->i_flags |= i_flags;
3923 inode->i_size = size;
3924 clear_nlink(inode); /* It is unlinked */
3925 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3927 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
3928 &shmem_file_operations);
3935 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3936 * kernel internal. There will be NO LSM permission checks against the
3937 * underlying inode. So users of this interface must do LSM checks at a
3938 * higher layer. The users are the big_key and shm implementations. LSM
3939 * checks are provided at the key or shm level rather than the inode.
3940 * @name: name for dentry (to be seen in /proc/<pid>/maps
3941 * @size: size to be set for the file
3942 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3944 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3946 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
3950 * shmem_file_setup - get an unlinked file living in tmpfs
3951 * @name: name for dentry (to be seen in /proc/<pid>/maps
3952 * @size: size to be set for the file
3953 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3955 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3957 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
3959 EXPORT_SYMBOL_GPL(shmem_file_setup);
3962 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
3963 * @mnt: the tmpfs mount where the file will be created
3964 * @name: name for dentry (to be seen in /proc/<pid>/maps
3965 * @size: size to be set for the file
3966 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3968 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
3969 loff_t size, unsigned long flags)
3971 return __shmem_file_setup(mnt, name, size, flags, 0);
3973 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
3976 * shmem_zero_setup - setup a shared anonymous mapping
3977 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3979 int shmem_zero_setup(struct vm_area_struct *vma)
3982 loff_t size = vma->vm_end - vma->vm_start;
3985 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3986 * between XFS directory reading and selinux: since this file is only
3987 * accessible to the user through its mapping, use S_PRIVATE flag to
3988 * bypass file security, in the same way as shmem_kernel_file_setup().
3990 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
3992 return PTR_ERR(file);
3996 vma->vm_file = file;
3997 vma->vm_ops = &shmem_vm_ops;
3999 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4000 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4001 (vma->vm_end & HPAGE_PMD_MASK)) {
4002 khugepaged_enter(vma, vma->vm_flags);
4009 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4010 * @mapping: the page's address_space
4011 * @index: the page index
4012 * @gfp: the page allocator flags to use if allocating
4014 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4015 * with any new page allocations done using the specified allocation flags.
4016 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4017 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4018 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4020 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4021 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4023 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4024 pgoff_t index, gfp_t gfp)
4027 struct inode *inode = mapping->host;
4031 BUG_ON(mapping->a_ops != &shmem_aops);
4032 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4033 gfp, NULL, NULL, NULL);
4035 page = ERR_PTR(error);
4041 * The tiny !SHMEM case uses ramfs without swap
4043 return read_cache_page_gfp(mapping, index, gfp);
4046 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);