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>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
42 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
44 static struct vfsmount *shm_mnt;
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/blkdev.h>
64 #include <linux/pagevec.h>
65 #include <linux/percpu_counter.h>
66 #include <linux/falloc.h>
67 #include <linux/splice.h>
68 #include <linux/security.h>
69 #include <linux/swapops.h>
70 #include <linux/mempolicy.h>
71 #include <linux/namei.h>
72 #include <linux/ctype.h>
73 #include <linux/migrate.h>
74 #include <linux/highmem.h>
75 #include <linux/seq_file.h>
76 #include <linux/magic.h>
77 #include <linux/syscalls.h>
78 #include <linux/fcntl.h>
79 #include <uapi/linux/memfd.h>
80 #include <linux/userfaultfd_k.h>
81 #include <linux/rmap.h>
82 #include <linux/uuid.h>
84 #include <linux/uaccess.h>
88 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
91 /* Pretend that each entry is of this size in directory's i_size */
92 #define BOGO_DIRENT_SIZE 20
94 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
95 #define SHORT_SYMLINK_LEN 128
98 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
99 * inode->i_private (with i_mutex making sure that it has only one user at
100 * a time): we would prefer not to enlarge the shmem inode just for that.
102 struct shmem_falloc {
103 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
104 pgoff_t start; /* start of range currently being fallocated */
105 pgoff_t next; /* the next page offset to be fallocated */
106 pgoff_t nr_falloced; /* how many new pages have been fallocated */
107 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
110 struct shmem_options {
111 unsigned long long blocks;
112 unsigned long long inodes;
113 struct mempolicy *mpol;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
123 #define SHMEM_SEEN_INUMS 8
127 static unsigned long shmem_default_max_blocks(void)
129 return totalram_pages() / 2;
132 static unsigned long shmem_default_max_inodes(void)
134 unsigned long nr_pages = totalram_pages();
136 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
140 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
141 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
142 struct shmem_inode_info *info, pgoff_t index);
143 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 vm_fault_t *fault_type);
147 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
148 struct page **pagep, enum sgp_type sgp,
149 gfp_t gfp, struct vm_area_struct *vma,
150 struct vm_fault *vmf, vm_fault_t *fault_type);
152 int shmem_getpage(struct inode *inode, pgoff_t index,
153 struct page **pagep, enum sgp_type sgp)
155 return shmem_getpage_gfp(inode, index, pagep, sgp,
156 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
159 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
161 return sb->s_fs_info;
165 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
166 * for shared memory and for shared anonymous (/dev/zero) mappings
167 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
168 * consistent with the pre-accounting of private mappings ...
170 static inline int shmem_acct_size(unsigned long flags, loff_t size)
172 return (flags & VM_NORESERVE) ?
173 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
176 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
178 if (!(flags & VM_NORESERVE))
179 vm_unacct_memory(VM_ACCT(size));
182 static inline int shmem_reacct_size(unsigned long flags,
183 loff_t oldsize, loff_t newsize)
185 if (!(flags & VM_NORESERVE)) {
186 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
187 return security_vm_enough_memory_mm(current->mm,
188 VM_ACCT(newsize) - VM_ACCT(oldsize));
189 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
190 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
196 * ... whereas tmpfs objects are accounted incrementally as
197 * pages are allocated, in order to allow large sparse files.
198 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
199 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201 static inline int shmem_acct_block(unsigned long flags, long pages)
203 if (!(flags & VM_NORESERVE))
206 return security_vm_enough_memory_mm(current->mm,
207 pages * VM_ACCT(PAGE_SIZE));
210 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
212 if (flags & VM_NORESERVE)
213 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
216 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
218 struct shmem_inode_info *info = SHMEM_I(inode);
219 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
221 if (shmem_acct_block(info->flags, pages))
224 if (sbinfo->max_blocks) {
225 if (percpu_counter_compare(&sbinfo->used_blocks,
226 sbinfo->max_blocks - pages) > 0)
228 percpu_counter_add(&sbinfo->used_blocks, pages);
234 shmem_unacct_blocks(info->flags, pages);
238 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
240 struct shmem_inode_info *info = SHMEM_I(inode);
241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
243 if (sbinfo->max_blocks)
244 percpu_counter_sub(&sbinfo->used_blocks, pages);
245 shmem_unacct_blocks(info->flags, pages);
248 static const struct super_operations shmem_ops;
249 static const struct address_space_operations shmem_aops;
250 static const struct file_operations shmem_file_operations;
251 static const struct inode_operations shmem_inode_operations;
252 static const struct inode_operations shmem_dir_inode_operations;
253 static const struct inode_operations shmem_special_inode_operations;
254 static const struct vm_operations_struct shmem_vm_ops;
255 static struct file_system_type shmem_fs_type;
257 bool vma_is_shmem(struct vm_area_struct *vma)
259 return vma->vm_ops == &shmem_vm_ops;
262 static LIST_HEAD(shmem_swaplist);
263 static DEFINE_MUTEX(shmem_swaplist_mutex);
266 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
267 * produces a novel ino for the newly allocated inode.
269 * It may also be called when making a hard link to permit the space needed by
270 * each dentry. However, in that case, no new inode number is needed since that
271 * internally draws from another pool of inode numbers (currently global
272 * get_next_ino()). This case is indicated by passing NULL as inop.
274 #define SHMEM_INO_BATCH 1024
275 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
277 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
280 if (!(sb->s_flags & SB_KERNMOUNT)) {
281 raw_spin_lock(&sbinfo->stat_lock);
282 if (sbinfo->max_inodes) {
283 if (!sbinfo->free_inodes) {
284 raw_spin_unlock(&sbinfo->stat_lock);
287 sbinfo->free_inodes--;
290 ino = sbinfo->next_ino++;
291 if (unlikely(is_zero_ino(ino)))
292 ino = sbinfo->next_ino++;
293 if (unlikely(!sbinfo->full_inums &&
296 * Emulate get_next_ino uint wraparound for
299 if (IS_ENABLED(CONFIG_64BIT))
300 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
301 __func__, MINOR(sb->s_dev));
302 sbinfo->next_ino = 1;
303 ino = sbinfo->next_ino++;
307 raw_spin_unlock(&sbinfo->stat_lock);
310 * __shmem_file_setup, one of our callers, is lock-free: it
311 * doesn't hold stat_lock in shmem_reserve_inode since
312 * max_inodes is always 0, and is called from potentially
313 * unknown contexts. As such, use a per-cpu batched allocator
314 * which doesn't require the per-sb stat_lock unless we are at
315 * the batch boundary.
317 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
318 * shmem mounts are not exposed to userspace, so we don't need
319 * to worry about things like glibc compatibility.
323 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
325 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
326 raw_spin_lock(&sbinfo->stat_lock);
327 ino = sbinfo->next_ino;
328 sbinfo->next_ino += SHMEM_INO_BATCH;
329 raw_spin_unlock(&sbinfo->stat_lock);
330 if (unlikely(is_zero_ino(ino)))
341 static void shmem_free_inode(struct super_block *sb)
343 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
344 if (sbinfo->max_inodes) {
345 raw_spin_lock(&sbinfo->stat_lock);
346 sbinfo->free_inodes++;
347 raw_spin_unlock(&sbinfo->stat_lock);
352 * shmem_recalc_inode - recalculate the block usage of an inode
353 * @inode: inode to recalc
355 * We have to calculate the free blocks since the mm can drop
356 * undirtied hole pages behind our back.
358 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
359 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
361 * It has to be called with the spinlock held.
363 static void shmem_recalc_inode(struct inode *inode)
365 struct shmem_inode_info *info = SHMEM_I(inode);
368 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
370 info->alloced -= freed;
371 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
372 shmem_inode_unacct_blocks(inode, freed);
376 bool shmem_charge(struct inode *inode, long pages)
378 struct shmem_inode_info *info = SHMEM_I(inode);
381 if (!shmem_inode_acct_block(inode, pages))
384 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
385 inode->i_mapping->nrpages += pages;
387 spin_lock_irqsave(&info->lock, flags);
388 info->alloced += pages;
389 inode->i_blocks += pages * BLOCKS_PER_PAGE;
390 shmem_recalc_inode(inode);
391 spin_unlock_irqrestore(&info->lock, flags);
396 void shmem_uncharge(struct inode *inode, long pages)
398 struct shmem_inode_info *info = SHMEM_I(inode);
401 /* nrpages adjustment done by __delete_from_page_cache() or caller */
403 spin_lock_irqsave(&info->lock, flags);
404 info->alloced -= pages;
405 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
406 shmem_recalc_inode(inode);
407 spin_unlock_irqrestore(&info->lock, flags);
409 shmem_inode_unacct_blocks(inode, pages);
413 * Replace item expected in xarray by a new item, while holding xa_lock.
415 static int shmem_replace_entry(struct address_space *mapping,
416 pgoff_t index, void *expected, void *replacement)
418 XA_STATE(xas, &mapping->i_pages, index);
421 VM_BUG_ON(!expected);
422 VM_BUG_ON(!replacement);
423 item = xas_load(&xas);
424 if (item != expected)
426 xas_store(&xas, replacement);
431 * Sometimes, before we decide whether to proceed or to fail, we must check
432 * that an entry was not already brought back from swap by a racing thread.
434 * Checking page is not enough: by the time a SwapCache page is locked, it
435 * might be reused, and again be SwapCache, using the same swap as before.
437 static bool shmem_confirm_swap(struct address_space *mapping,
438 pgoff_t index, swp_entry_t swap)
440 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
444 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
447 * disables huge pages for the mount;
449 * enables huge pages for the mount;
450 * SHMEM_HUGE_WITHIN_SIZE:
451 * only allocate huge pages if the page will be fully within i_size,
452 * also respect fadvise()/madvise() hints;
454 * only allocate huge pages if requested with fadvise()/madvise();
457 #define SHMEM_HUGE_NEVER 0
458 #define SHMEM_HUGE_ALWAYS 1
459 #define SHMEM_HUGE_WITHIN_SIZE 2
460 #define SHMEM_HUGE_ADVISE 3
464 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
467 * disables huge on shm_mnt and all mounts, for emergency use;
469 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
472 #define SHMEM_HUGE_DENY (-1)
473 #define SHMEM_HUGE_FORCE (-2)
475 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
476 /* ifdef here to avoid bloating shmem.o when not necessary */
478 static int shmem_huge __read_mostly;
480 #if defined(CONFIG_SYSFS)
481 static int shmem_parse_huge(const char *str)
483 if (!strcmp(str, "never"))
484 return SHMEM_HUGE_NEVER;
485 if (!strcmp(str, "always"))
486 return SHMEM_HUGE_ALWAYS;
487 if (!strcmp(str, "within_size"))
488 return SHMEM_HUGE_WITHIN_SIZE;
489 if (!strcmp(str, "advise"))
490 return SHMEM_HUGE_ADVISE;
491 if (!strcmp(str, "deny"))
492 return SHMEM_HUGE_DENY;
493 if (!strcmp(str, "force"))
494 return SHMEM_HUGE_FORCE;
499 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
500 static const char *shmem_format_huge(int huge)
503 case SHMEM_HUGE_NEVER:
505 case SHMEM_HUGE_ALWAYS:
507 case SHMEM_HUGE_WITHIN_SIZE:
508 return "within_size";
509 case SHMEM_HUGE_ADVISE:
511 case SHMEM_HUGE_DENY:
513 case SHMEM_HUGE_FORCE:
522 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
523 struct shrink_control *sc, unsigned long nr_to_split)
525 LIST_HEAD(list), *pos, *next;
526 LIST_HEAD(to_remove);
528 struct shmem_inode_info *info;
530 unsigned long batch = sc ? sc->nr_to_scan : 128;
533 if (list_empty(&sbinfo->shrinklist))
536 spin_lock(&sbinfo->shrinklist_lock);
537 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
538 info = list_entry(pos, struct shmem_inode_info, shrinklist);
541 inode = igrab(&info->vfs_inode);
543 /* inode is about to be evicted */
545 list_del_init(&info->shrinklist);
549 /* Check if there's anything to gain */
550 if (round_up(inode->i_size, PAGE_SIZE) ==
551 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
552 list_move(&info->shrinklist, &to_remove);
556 list_move(&info->shrinklist, &list);
558 sbinfo->shrinklist_len--;
562 spin_unlock(&sbinfo->shrinklist_lock);
564 list_for_each_safe(pos, next, &to_remove) {
565 info = list_entry(pos, struct shmem_inode_info, shrinklist);
566 inode = &info->vfs_inode;
567 list_del_init(&info->shrinklist);
571 list_for_each_safe(pos, next, &list) {
574 info = list_entry(pos, struct shmem_inode_info, shrinklist);
575 inode = &info->vfs_inode;
577 if (nr_to_split && split >= nr_to_split)
580 page = find_get_page(inode->i_mapping,
581 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
585 /* No huge page at the end of the file: nothing to split */
586 if (!PageTransHuge(page)) {
592 * Move the inode on the list back to shrinklist if we failed
593 * to lock the page at this time.
595 * Waiting for the lock may lead to deadlock in the
598 if (!trylock_page(page)) {
603 ret = split_huge_page(page);
607 /* If split failed move the inode on the list back to shrinklist */
613 list_del_init(&info->shrinklist);
617 * Make sure the inode is either on the global list or deleted
618 * from any local list before iput() since it could be deleted
619 * in another thread once we put the inode (then the local list
622 spin_lock(&sbinfo->shrinklist_lock);
623 list_move(&info->shrinklist, &sbinfo->shrinklist);
624 sbinfo->shrinklist_len++;
625 spin_unlock(&sbinfo->shrinklist_lock);
633 static long shmem_unused_huge_scan(struct super_block *sb,
634 struct shrink_control *sc)
636 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
638 if (!READ_ONCE(sbinfo->shrinklist_len))
641 return shmem_unused_huge_shrink(sbinfo, sc, 0);
644 static long shmem_unused_huge_count(struct super_block *sb,
645 struct shrink_control *sc)
647 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
648 return READ_ONCE(sbinfo->shrinklist_len);
650 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
652 #define shmem_huge SHMEM_HUGE_DENY
654 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
655 struct shrink_control *sc, unsigned long nr_to_split)
659 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
661 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
663 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
664 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
665 shmem_huge != SHMEM_HUGE_DENY)
671 * Like add_to_page_cache_locked, but error if expected item has gone.
673 static int shmem_add_to_page_cache(struct page *page,
674 struct address_space *mapping,
675 pgoff_t index, void *expected, gfp_t gfp,
676 struct mm_struct *charge_mm)
678 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
680 unsigned long nr = compound_nr(page);
683 VM_BUG_ON_PAGE(PageTail(page), page);
684 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
685 VM_BUG_ON_PAGE(!PageLocked(page), page);
686 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
687 VM_BUG_ON(expected && PageTransHuge(page));
689 page_ref_add(page, nr);
690 page->mapping = mapping;
693 if (!PageSwapCache(page)) {
694 error = mem_cgroup_charge(page, charge_mm, gfp);
696 if (PageTransHuge(page)) {
697 count_vm_event(THP_FILE_FALLBACK);
698 count_vm_event(THP_FILE_FALLBACK_CHARGE);
703 cgroup_throttle_swaprate(page, gfp);
708 entry = xas_find_conflict(&xas);
709 if (entry != expected)
710 xas_set_err(&xas, -EEXIST);
711 xas_create_range(&xas);
715 xas_store(&xas, page);
720 if (PageTransHuge(page)) {
721 count_vm_event(THP_FILE_ALLOC);
722 #ifdef CONFIG_FINEGRAINED_THP
723 if (thp_nr_pages(page) == HPAGE_CONT_PTE_NR)
724 __inc_node_page_state(page, NR_SHMEM_64KB_THPS);
726 #endif /* CONFIG_FINEGRAINED_THP */
727 __inc_node_page_state(page, NR_SHMEM_THPS);
729 mapping->nrpages += nr;
730 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
731 __mod_lruvec_page_state(page, NR_SHMEM, nr);
733 xas_unlock_irq(&xas);
734 } while (xas_nomem(&xas, gfp));
736 if (xas_error(&xas)) {
737 error = xas_error(&xas);
743 page->mapping = NULL;
744 page_ref_sub(page, nr);
749 * Like delete_from_page_cache, but substitutes swap for page.
751 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
753 struct address_space *mapping = page->mapping;
756 VM_BUG_ON_PAGE(PageCompound(page), page);
758 xa_lock_irq(&mapping->i_pages);
759 error = shmem_replace_entry(mapping, page->index, page, radswap);
760 page->mapping = NULL;
762 __dec_lruvec_page_state(page, NR_FILE_PAGES);
763 __dec_lruvec_page_state(page, NR_SHMEM);
764 xa_unlock_irq(&mapping->i_pages);
770 * Remove swap entry from page cache, free the swap and its page cache.
772 static int shmem_free_swap(struct address_space *mapping,
773 pgoff_t index, void *radswap)
777 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
780 free_swap_and_cache(radix_to_swp_entry(radswap));
785 * Determine (in bytes) how many of the shmem object's pages mapped by the
786 * given offsets are swapped out.
788 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
789 * as long as the inode doesn't go away and racy results are not a problem.
791 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
792 pgoff_t start, pgoff_t end)
794 XA_STATE(xas, &mapping->i_pages, start);
796 unsigned long swapped = 0;
799 xas_for_each(&xas, page, end - 1) {
800 if (xas_retry(&xas, page))
802 if (xa_is_value(page))
805 if (need_resched()) {
813 return swapped << PAGE_SHIFT;
817 * Determine (in bytes) how many of the shmem object's pages mapped by the
818 * given vma is swapped out.
820 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
821 * as long as the inode doesn't go away and racy results are not a problem.
823 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
825 struct inode *inode = file_inode(vma->vm_file);
826 struct shmem_inode_info *info = SHMEM_I(inode);
827 struct address_space *mapping = inode->i_mapping;
828 unsigned long swapped;
830 /* Be careful as we don't hold info->lock */
831 swapped = READ_ONCE(info->swapped);
834 * The easier cases are when the shmem object has nothing in swap, or
835 * the vma maps it whole. Then we can simply use the stats that we
841 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
842 return swapped << PAGE_SHIFT;
844 /* Here comes the more involved part */
845 return shmem_partial_swap_usage(mapping,
846 linear_page_index(vma, vma->vm_start),
847 linear_page_index(vma, vma->vm_end));
851 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
853 void shmem_unlock_mapping(struct address_space *mapping)
856 pgoff_t indices[PAGEVEC_SIZE];
861 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
863 while (!mapping_unevictable(mapping)) {
865 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
866 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
868 pvec.nr = find_get_entries(mapping, index,
869 PAGEVEC_SIZE, pvec.pages, indices);
872 index = indices[pvec.nr - 1] + 1;
873 pagevec_remove_exceptionals(&pvec);
874 check_move_unevictable_pages(&pvec);
875 pagevec_release(&pvec);
881 * Check whether a hole-punch or truncation needs to split a huge page,
882 * returning true if no split was required, or the split has been successful.
884 * Eviction (or truncation to 0 size) should never need to split a huge page;
885 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
886 * head, and then succeeded to trylock on tail.
888 * A split can only succeed when there are no additional references on the
889 * huge page: so the split below relies upon find_get_entries() having stopped
890 * when it found a subpage of the huge page, without getting further references.
892 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
894 if (!PageTransCompound(page))
897 /* Just proceed to delete a huge page wholly within the range punched */
898 #ifdef CONFIG_FINEGRAINED_THP
899 if (PageHead(page) &&
900 page->index >= start && page->index + thp_nr_pages(page) <= end)
903 if (PageHead(page) &&
904 page->index >= start && page->index + HPAGE_PMD_NR <= end)
906 #endif /* CONFIG_FINEGRAINED_THP */
908 /* Try to split huge page, so we can truly punch the hole or truncate */
909 return split_huge_page(page) >= 0;
913 * Remove range of pages and swap entries from page cache, and free them.
914 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
916 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
919 struct address_space *mapping = inode->i_mapping;
920 struct shmem_inode_info *info = SHMEM_I(inode);
921 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
922 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
923 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
924 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
926 pgoff_t indices[PAGEVEC_SIZE];
927 long nr_swaps_freed = 0;
932 end = -1; /* unsigned, so actually very big */
936 while (index < end) {
937 pvec.nr = find_get_entries(mapping, index,
938 min(end - index, (pgoff_t)PAGEVEC_SIZE),
939 pvec.pages, indices);
942 for (i = 0; i < pagevec_count(&pvec); i++) {
943 struct page *page = pvec.pages[i];
949 if (xa_is_value(page)) {
952 nr_swaps_freed += !shmem_free_swap(mapping,
957 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
959 if (!trylock_page(page))
962 if ((!unfalloc || !PageUptodate(page)) &&
963 page_mapping(page) == mapping) {
964 VM_BUG_ON_PAGE(PageWriteback(page), page);
965 if (shmem_punch_compound(page, start, end))
966 truncate_inode_page(mapping, page);
970 pagevec_remove_exceptionals(&pvec);
971 pagevec_release(&pvec);
977 struct page *page = NULL;
978 shmem_getpage(inode, start - 1, &page, SGP_READ);
980 unsigned int top = PAGE_SIZE;
985 zero_user_segment(page, partial_start, top);
986 set_page_dirty(page);
992 struct page *page = NULL;
993 shmem_getpage(inode, end, &page, SGP_READ);
995 zero_user_segment(page, 0, partial_end);
996 set_page_dirty(page);
1005 while (index < end) {
1008 pvec.nr = find_get_entries(mapping, index,
1009 min(end - index, (pgoff_t)PAGEVEC_SIZE),
1010 pvec.pages, indices);
1012 /* If all gone or hole-punch or unfalloc, we're done */
1013 if (index == start || end != -1)
1015 /* But if truncating, restart to make sure all gone */
1019 for (i = 0; i < pagevec_count(&pvec); i++) {
1020 struct page *page = pvec.pages[i];
1026 if (xa_is_value(page)) {
1029 if (shmem_free_swap(mapping, index, page)) {
1030 /* Swap was replaced by page: retry */
1040 if (!unfalloc || !PageUptodate(page)) {
1041 if (page_mapping(page) != mapping) {
1042 /* Page was replaced by swap: retry */
1047 VM_BUG_ON_PAGE(PageWriteback(page), page);
1048 if (shmem_punch_compound(page, start, end))
1049 truncate_inode_page(mapping, page);
1050 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1051 /* Wipe the page and don't get stuck */
1052 clear_highpage(page);
1053 flush_dcache_page(page);
1054 set_page_dirty(page);
1055 #ifdef CONFIG_FINEGRAINED_THP
1057 round_up(start, thp_nr_pages(page)))
1059 #else /* CONFIG_FINEGRAINED_THP */
1061 round_up(start, HPAGE_PMD_NR))
1063 #endif /* CONFIG_FINEGRAINED_THP */
1068 pagevec_remove_exceptionals(&pvec);
1069 pagevec_release(&pvec);
1073 spin_lock_irq(&info->lock);
1074 info->swapped -= nr_swaps_freed;
1075 shmem_recalc_inode(inode);
1076 spin_unlock_irq(&info->lock);
1079 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1081 shmem_undo_range(inode, lstart, lend, false);
1082 inode->i_ctime = inode->i_mtime = current_time(inode);
1084 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1086 static int shmem_getattr(const struct path *path, struct kstat *stat,
1087 u32 request_mask, unsigned int query_flags)
1089 struct inode *inode = path->dentry->d_inode;
1090 struct shmem_inode_info *info = SHMEM_I(inode);
1091 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1093 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1094 spin_lock_irq(&info->lock);
1095 shmem_recalc_inode(inode);
1096 spin_unlock_irq(&info->lock);
1098 generic_fillattr(inode, stat);
1100 if (is_huge_enabled(sb_info))
1101 stat->blksize = HPAGE_PMD_SIZE;
1106 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
1108 struct inode *inode = d_inode(dentry);
1109 struct shmem_inode_info *info = SHMEM_I(inode);
1110 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1113 error = setattr_prepare(dentry, attr);
1117 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1118 loff_t oldsize = inode->i_size;
1119 loff_t newsize = attr->ia_size;
1121 /* protected by i_mutex */
1122 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1123 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1126 if (newsize != oldsize) {
1127 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1131 i_size_write(inode, newsize);
1132 inode->i_ctime = inode->i_mtime = current_time(inode);
1134 if (newsize <= oldsize) {
1135 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1136 if (oldsize > holebegin)
1137 unmap_mapping_range(inode->i_mapping,
1140 shmem_truncate_range(inode,
1141 newsize, (loff_t)-1);
1142 /* unmap again to remove racily COWed private pages */
1143 if (oldsize > holebegin)
1144 unmap_mapping_range(inode->i_mapping,
1148 * Part of the huge page can be beyond i_size: subject
1149 * to shrink under memory pressure.
1151 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1152 spin_lock(&sbinfo->shrinklist_lock);
1154 * _careful to defend against unlocked access to
1155 * ->shrink_list in shmem_unused_huge_shrink()
1157 if (list_empty_careful(&info->shrinklist)) {
1158 list_add_tail(&info->shrinklist,
1159 &sbinfo->shrinklist);
1160 sbinfo->shrinklist_len++;
1162 spin_unlock(&sbinfo->shrinklist_lock);
1167 setattr_copy(inode, attr);
1168 if (attr->ia_valid & ATTR_MODE)
1169 error = posix_acl_chmod(inode, inode->i_mode);
1173 static void shmem_evict_inode(struct inode *inode)
1175 struct shmem_inode_info *info = SHMEM_I(inode);
1176 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1178 if (inode->i_mapping->a_ops == &shmem_aops) {
1179 shmem_unacct_size(info->flags, inode->i_size);
1181 shmem_truncate_range(inode, 0, (loff_t)-1);
1182 if (!list_empty(&info->shrinklist)) {
1183 spin_lock(&sbinfo->shrinklist_lock);
1184 if (!list_empty(&info->shrinklist)) {
1185 list_del_init(&info->shrinklist);
1186 sbinfo->shrinklist_len--;
1188 spin_unlock(&sbinfo->shrinklist_lock);
1190 while (!list_empty(&info->swaplist)) {
1191 /* Wait while shmem_unuse() is scanning this inode... */
1192 wait_var_event(&info->stop_eviction,
1193 !atomic_read(&info->stop_eviction));
1194 mutex_lock(&shmem_swaplist_mutex);
1195 /* ...but beware of the race if we peeked too early */
1196 if (!atomic_read(&info->stop_eviction))
1197 list_del_init(&info->swaplist);
1198 mutex_unlock(&shmem_swaplist_mutex);
1202 simple_xattrs_free(&info->xattrs);
1203 WARN_ON(inode->i_blocks);
1204 shmem_free_inode(inode->i_sb);
1208 extern struct swap_info_struct *swap_info[];
1210 static int shmem_find_swap_entries(struct address_space *mapping,
1211 pgoff_t start, unsigned int nr_entries,
1212 struct page **entries, pgoff_t *indices,
1213 unsigned int type, bool frontswap)
1215 XA_STATE(xas, &mapping->i_pages, start);
1218 unsigned int ret = 0;
1224 xas_for_each(&xas, page, ULONG_MAX) {
1225 if (xas_retry(&xas, page))
1228 if (!xa_is_value(page))
1231 entry = radix_to_swp_entry(page);
1232 if (swp_type(entry) != type)
1235 !frontswap_test(swap_info[type], swp_offset(entry)))
1238 indices[ret] = xas.xa_index;
1239 entries[ret] = page;
1241 if (need_resched()) {
1245 if (++ret == nr_entries)
1254 * Move the swapped pages for an inode to page cache. Returns the count
1255 * of pages swapped in, or the error in case of failure.
1257 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1263 struct address_space *mapping = inode->i_mapping;
1265 for (i = 0; i < pvec.nr; i++) {
1266 struct page *page = pvec.pages[i];
1268 if (!xa_is_value(page))
1270 error = shmem_swapin_page(inode, indices[i],
1272 mapping_gfp_mask(mapping),
1279 if (error == -ENOMEM)
1283 return error ? error : ret;
1287 * If swap found in inode, free it and move page from swapcache to filecache.
1289 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1290 bool frontswap, unsigned long *fs_pages_to_unuse)
1292 struct address_space *mapping = inode->i_mapping;
1294 struct pagevec pvec;
1295 pgoff_t indices[PAGEVEC_SIZE];
1296 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1299 pagevec_init(&pvec);
1301 unsigned int nr_entries = PAGEVEC_SIZE;
1303 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1304 nr_entries = *fs_pages_to_unuse;
1306 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1307 pvec.pages, indices,
1314 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1318 if (frontswap_partial) {
1319 *fs_pages_to_unuse -= ret;
1320 if (*fs_pages_to_unuse == 0) {
1321 ret = FRONTSWAP_PAGES_UNUSED;
1326 start = indices[pvec.nr - 1];
1333 * Read all the shared memory data that resides in the swap
1334 * device 'type' back into memory, so the swap device can be
1337 int shmem_unuse(unsigned int type, bool frontswap,
1338 unsigned long *fs_pages_to_unuse)
1340 struct shmem_inode_info *info, *next;
1343 if (list_empty(&shmem_swaplist))
1346 mutex_lock(&shmem_swaplist_mutex);
1347 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1348 if (!info->swapped) {
1349 list_del_init(&info->swaplist);
1353 * Drop the swaplist mutex while searching the inode for swap;
1354 * but before doing so, make sure shmem_evict_inode() will not
1355 * remove placeholder inode from swaplist, nor let it be freed
1356 * (igrab() would protect from unlink, but not from unmount).
1358 atomic_inc(&info->stop_eviction);
1359 mutex_unlock(&shmem_swaplist_mutex);
1361 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1365 mutex_lock(&shmem_swaplist_mutex);
1366 next = list_next_entry(info, swaplist);
1368 list_del_init(&info->swaplist);
1369 if (atomic_dec_and_test(&info->stop_eviction))
1370 wake_up_var(&info->stop_eviction);
1374 mutex_unlock(&shmem_swaplist_mutex);
1380 * Move the page from the page cache to the swap cache.
1382 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1384 struct shmem_inode_info *info;
1385 struct address_space *mapping;
1386 struct inode *inode;
1390 VM_BUG_ON_PAGE(PageCompound(page), page);
1391 BUG_ON(!PageLocked(page));
1392 mapping = page->mapping;
1393 index = page->index;
1394 inode = mapping->host;
1395 info = SHMEM_I(inode);
1396 if (info->flags & VM_LOCKED)
1398 if (!total_swap_pages)
1402 * Our capabilities prevent regular writeback or sync from ever calling
1403 * shmem_writepage; but a stacking filesystem might use ->writepage of
1404 * its underlying filesystem, in which case tmpfs should write out to
1405 * swap only in response to memory pressure, and not for the writeback
1408 if (!wbc->for_reclaim) {
1409 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1414 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1415 * value into swapfile.c, the only way we can correctly account for a
1416 * fallocated page arriving here is now to initialize it and write it.
1418 * That's okay for a page already fallocated earlier, but if we have
1419 * not yet completed the fallocation, then (a) we want to keep track
1420 * of this page in case we have to undo it, and (b) it may not be a
1421 * good idea to continue anyway, once we're pushing into swap. So
1422 * reactivate the page, and let shmem_fallocate() quit when too many.
1424 if (!PageUptodate(page)) {
1425 if (inode->i_private) {
1426 struct shmem_falloc *shmem_falloc;
1427 spin_lock(&inode->i_lock);
1428 shmem_falloc = inode->i_private;
1430 !shmem_falloc->waitq &&
1431 index >= shmem_falloc->start &&
1432 index < shmem_falloc->next)
1433 shmem_falloc->nr_unswapped++;
1435 shmem_falloc = NULL;
1436 spin_unlock(&inode->i_lock);
1440 clear_highpage(page);
1441 flush_dcache_page(page);
1442 SetPageUptodate(page);
1445 swap = get_swap_page(page);
1450 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1451 * if it's not already there. Do it now before the page is
1452 * moved to swap cache, when its pagelock no longer protects
1453 * the inode from eviction. But don't unlock the mutex until
1454 * we've incremented swapped, because shmem_unuse_inode() will
1455 * prune a !swapped inode from the swaplist under this mutex.
1457 mutex_lock(&shmem_swaplist_mutex);
1458 if (list_empty(&info->swaplist))
1459 list_add(&info->swaplist, &shmem_swaplist);
1461 if (add_to_swap_cache(page, swap,
1462 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1464 spin_lock_irq(&info->lock);
1465 shmem_recalc_inode(inode);
1467 spin_unlock_irq(&info->lock);
1469 swap_shmem_alloc(swap);
1470 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1472 mutex_unlock(&shmem_swaplist_mutex);
1473 BUG_ON(page_mapped(page));
1474 swap_writepage(page, wbc);
1478 mutex_unlock(&shmem_swaplist_mutex);
1479 put_swap_page(page, swap);
1481 set_page_dirty(page);
1482 if (wbc->for_reclaim)
1483 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1488 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1489 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1493 if (!mpol || mpol->mode == MPOL_DEFAULT)
1494 return; /* show nothing */
1496 mpol_to_str(buffer, sizeof(buffer), mpol);
1498 seq_printf(seq, ",mpol=%s", buffer);
1501 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1503 struct mempolicy *mpol = NULL;
1505 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1506 mpol = sbinfo->mpol;
1508 raw_spin_unlock(&sbinfo->stat_lock);
1512 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1513 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1516 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1520 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1522 #define vm_policy vm_private_data
1525 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1526 struct shmem_inode_info *info, pgoff_t index)
1528 /* Create a pseudo vma that just contains the policy */
1529 vma_init(vma, NULL);
1530 /* Bias interleave by inode number to distribute better across nodes */
1531 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1532 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1535 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1537 /* Drop reference taken by mpol_shared_policy_lookup() */
1538 mpol_cond_put(vma->vm_policy);
1541 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1542 struct shmem_inode_info *info, pgoff_t index)
1544 struct vm_area_struct pvma;
1546 struct vm_fault vmf;
1548 shmem_pseudo_vma_init(&pvma, info, index);
1551 page = swap_cluster_readahead(swap, gfp, &vmf);
1552 shmem_pseudo_vma_destroy(&pvma);
1557 #ifdef CONFIG_FINEGRAINED_THP
1558 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1559 struct shmem_inode_info *info, pgoff_t index, int page_nr)
1560 #else /* CONFIG_FINEGRAINED_THP */
1561 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1562 struct shmem_inode_info *info, pgoff_t index)
1563 #endif/* CONFIG_FINEGRAINED_THP */
1565 struct vm_area_struct pvma;
1566 struct address_space *mapping = info->vfs_inode.i_mapping;
1570 #ifdef CONFIG_FINEGRAINED_THP
1571 hindex = round_down(index, page_nr);
1572 if (xa_find(&mapping->i_pages, &hindex, hindex + page_nr - 1,
1575 #else /* CONFIG_FINEGRAINED_THP */
1576 hindex = round_down(index, HPAGE_PMD_NR);
1577 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1580 #endif /* CONFIG_FINEGRAINED_THP */
1582 shmem_pseudo_vma_init(&pvma, info, hindex);
1583 #ifdef CONFIG_FINEGRAINED_THP
1584 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1585 page_nr == HPAGE_PMD_NR ? HPAGE_PMD_ORDER : HPAGE_CONT_PTE_ORDER,
1586 &pvma, 0, numa_node_id(), true);
1587 #else /* CONFIG_FINEGRAINED_THP */
1588 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1589 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1590 #endif /* CONFIG_FINEGRAINED_THP */
1591 shmem_pseudo_vma_destroy(&pvma);
1593 prep_transhuge_page(page);
1595 count_vm_event(THP_FILE_FALLBACK);
1599 static struct page *shmem_alloc_page(gfp_t gfp,
1600 struct shmem_inode_info *info, pgoff_t index)
1602 struct vm_area_struct pvma;
1605 shmem_pseudo_vma_init(&pvma, info, index);
1606 page = alloc_page_vma(gfp, &pvma, 0);
1607 shmem_pseudo_vma_destroy(&pvma);
1612 #ifdef CONFIG_FINEGRAINED_THP
1613 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1614 struct inode *inode,
1615 pgoff_t index, bool huge, int page_nr)
1616 #else /* CONFIG_FINEGRAINED_THP */
1617 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1618 struct inode *inode,
1619 pgoff_t index, bool huge)
1620 #endif /* CONFIG_FINEGRAINED_THP */
1622 struct shmem_inode_info *info = SHMEM_I(inode);
1627 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1629 #ifdef CONFIG_FINEGRAINED_THP
1630 nr = huge ? page_nr : 1;
1632 nr = huge ? HPAGE_PMD_NR : 1;
1635 if (!shmem_inode_acct_block(inode, nr))
1639 #ifdef CONFIG_FINEGRAINED_THP
1640 page = shmem_alloc_hugepage(gfp, info, index, nr);
1642 page = shmem_alloc_hugepage(gfp, info, index);
1645 page = shmem_alloc_page(gfp, info, index);
1647 __SetPageLocked(page);
1648 __SetPageSwapBacked(page);
1653 shmem_inode_unacct_blocks(inode, nr);
1655 return ERR_PTR(err);
1659 * When a page is moved from swapcache to shmem filecache (either by the
1660 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1661 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1662 * ignorance of the mapping it belongs to. If that mapping has special
1663 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1664 * we may need to copy to a suitable page before moving to filecache.
1666 * In a future release, this may well be extended to respect cpuset and
1667 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1668 * but for now it is a simple matter of zone.
1670 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1672 return page_zonenum(page) > gfp_zone(gfp);
1675 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1676 struct shmem_inode_info *info, pgoff_t index)
1678 struct page *oldpage, *newpage;
1679 struct address_space *swap_mapping;
1685 entry.val = page_private(oldpage);
1686 swap_index = swp_offset(entry);
1687 swap_mapping = page_mapping(oldpage);
1690 * We have arrived here because our zones are constrained, so don't
1691 * limit chance of success by further cpuset and node constraints.
1693 gfp &= ~GFP_CONSTRAINT_MASK;
1694 newpage = shmem_alloc_page(gfp, info, index);
1699 copy_highpage(newpage, oldpage);
1700 flush_dcache_page(newpage);
1702 __SetPageLocked(newpage);
1703 __SetPageSwapBacked(newpage);
1704 SetPageUptodate(newpage);
1705 set_page_private(newpage, entry.val);
1706 SetPageSwapCache(newpage);
1709 * Our caller will very soon move newpage out of swapcache, but it's
1710 * a nice clean interface for us to replace oldpage by newpage there.
1712 xa_lock_irq(&swap_mapping->i_pages);
1713 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1715 mem_cgroup_migrate(oldpage, newpage);
1716 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1717 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1719 xa_unlock_irq(&swap_mapping->i_pages);
1721 if (unlikely(error)) {
1723 * Is this possible? I think not, now that our callers check
1724 * both PageSwapCache and page_private after getting page lock;
1725 * but be defensive. Reverse old to newpage for clear and free.
1729 lru_cache_add(newpage);
1733 ClearPageSwapCache(oldpage);
1734 set_page_private(oldpage, 0);
1736 unlock_page(oldpage);
1743 * Swap in the page pointed to by *pagep.
1744 * Caller has to make sure that *pagep contains a valid swapped page.
1745 * Returns 0 and the page in pagep if success. On failure, returns the
1746 * error code and NULL in *pagep.
1748 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1749 struct page **pagep, enum sgp_type sgp,
1750 gfp_t gfp, struct vm_area_struct *vma,
1751 vm_fault_t *fault_type)
1753 struct address_space *mapping = inode->i_mapping;
1754 struct shmem_inode_info *info = SHMEM_I(inode);
1755 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
1760 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1761 swap = radix_to_swp_entry(*pagep);
1764 /* Look it up and read it in.. */
1765 page = lookup_swap_cache(swap, NULL, 0);
1767 /* Or update major stats only when swapin succeeds?? */
1769 *fault_type |= VM_FAULT_MAJOR;
1770 count_vm_event(PGMAJFAULT);
1771 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1773 /* Here we actually start the io */
1774 page = shmem_swapin(swap, gfp, info, index);
1781 /* We have to do this with page locked to prevent races */
1783 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1784 !shmem_confirm_swap(mapping, index, swap)) {
1788 if (!PageUptodate(page)) {
1792 wait_on_page_writeback(page);
1795 * Some architectures may have to restore extra metadata to the
1796 * physical page after reading from swap.
1798 arch_swap_restore(swap, page);
1800 if (shmem_should_replace_page(page, gfp)) {
1801 error = shmem_replace_page(&page, gfp, info, index);
1806 error = shmem_add_to_page_cache(page, mapping, index,
1807 swp_to_radix_entry(swap), gfp,
1812 spin_lock_irq(&info->lock);
1814 shmem_recalc_inode(inode);
1815 spin_unlock_irq(&info->lock);
1817 if (sgp == SGP_WRITE)
1818 mark_page_accessed(page);
1820 delete_from_swap_cache(page);
1821 set_page_dirty(page);
1827 if (!shmem_confirm_swap(mapping, index, swap))
1839 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1841 * If we allocate a new one we do not mark it dirty. That's up to the
1842 * vm. If we swap it in we mark it dirty since we also free the swap
1843 * entry since a page cannot live in both the swap and page cache.
1845 * vmf and fault_type are only supplied by shmem_fault:
1846 * otherwise they are NULL.
1848 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1849 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1850 struct vm_area_struct *vma, struct vm_fault *vmf,
1851 vm_fault_t *fault_type)
1853 struct address_space *mapping = inode->i_mapping;
1854 struct shmem_inode_info *info = SHMEM_I(inode);
1855 struct shmem_sb_info *sbinfo;
1856 struct mm_struct *charge_mm;
1858 enum sgp_type sgp_huge = sgp;
1859 pgoff_t hindex = index;
1863 #ifdef CONFIG_FINEGRAINED_THP
1864 int nr_pages = HPAGE_PMD_NR;
1867 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1869 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1872 if (sgp <= SGP_CACHE &&
1873 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1877 sbinfo = SHMEM_SB(inode->i_sb);
1878 charge_mm = vma ? vma->vm_mm : current->mm;
1880 page = find_lock_entry(mapping, index);
1881 if (xa_is_value(page)) {
1882 error = shmem_swapin_page(inode, index, &page,
1883 sgp, gfp, vma, fault_type);
1884 if (error == -EEXIST)
1892 hindex = page->index;
1893 if (page && sgp == SGP_WRITE)
1894 mark_page_accessed(page);
1896 #ifdef CONFIG_FINEGRAINED_THP
1898 nr_pages = thp_nr_pages(page);
1901 /* fallocated page? */
1902 if (page && !PageUptodate(page)) {
1903 if (sgp != SGP_READ)
1910 if (page || sgp == SGP_READ)
1914 * Fast cache lookup did not find it:
1915 * bring it back from swap or allocate.
1918 if (vma && userfaultfd_missing(vma)) {
1919 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1923 /* shmem_symlink() */
1924 if (mapping->a_ops != &shmem_aops)
1926 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1928 if (shmem_huge == SHMEM_HUGE_FORCE)
1930 switch (sbinfo->huge) {
1931 case SHMEM_HUGE_NEVER:
1933 case SHMEM_HUGE_WITHIN_SIZE: {
1936 #ifdef CONFIG_FINEGRAINED_THP
1937 off = round_up(index, nr_pages);
1939 off = round_up(index, HPAGE_PMD_NR);
1941 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1942 #ifdef CONFIG_FINEGRAINED_THP
1943 if (i_size >= nr_pages * PAGE_SIZE &&
1944 i_size >> PAGE_SHIFT >= off)
1947 if (i_size >= HPAGE_PMD_SIZE &&
1948 i_size >> PAGE_SHIFT >= off)
1954 case SHMEM_HUGE_ADVISE:
1955 if (sgp_huge == SGP_HUGE)
1957 /* TODO: implement fadvise() hints */
1962 #ifdef CONFIG_FINEGRAINED_THP
1963 page = shmem_alloc_and_acct_page(gfp, inode, index, true, nr_pages);
1965 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1969 #ifdef CONFIG_FINEGRAINED_THP
1970 page = shmem_alloc_and_acct_page(gfp, inode,
1973 page = shmem_alloc_and_acct_page(gfp, inode,
1980 error = PTR_ERR(page);
1982 if (error != -ENOSPC)
1985 * Try to reclaim some space by splitting a huge page
1986 * beyond i_size on the filesystem.
1991 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1992 if (ret == SHRINK_STOP)
2000 if (PageTransHuge(page))
2001 #ifdef CONFIG_FINEGRAINED_THP
2002 hindex = round_down(index, nr_pages);
2004 hindex = round_down(index, HPAGE_PMD_NR);
2009 if (sgp == SGP_WRITE)
2010 __SetPageReferenced(page);
2012 error = shmem_add_to_page_cache(page, mapping, hindex,
2013 NULL, gfp & GFP_RECLAIM_MASK,
2017 lru_cache_add(page);
2019 spin_lock_irq(&info->lock);
2020 info->alloced += compound_nr(page);
2021 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
2022 shmem_recalc_inode(inode);
2023 spin_unlock_irq(&info->lock);
2026 #ifdef CONFIG_FINEGRAINED_THP
2027 if (PageTransHuge(page) &&
2028 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
2029 hindex + nr_pages - 1) {
2031 * Part of the huge page is beyond i_size: subject
2032 * to shrink under memory pressure.
2034 spin_lock(&sbinfo->shrinklist_lock);
2036 * _careful to defend against unlocked access to
2037 * ->shrink_list in shmem_unused_huge_shrink()
2039 if (list_empty_careful(&info->shrinklist)) {
2040 list_add_tail(&info->shrinklist,
2041 &sbinfo->shrinklist);
2042 sbinfo->shrinklist_len++;
2044 spin_unlock(&sbinfo->shrinklist_lock);
2046 #else /* CONFIG_FINEGRAINED_THP */
2047 if (PageTransHuge(page) &&
2048 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
2049 hindex + HPAGE_PMD_NR - 1) {
2051 * Part of the huge page is beyond i_size: subject
2052 * to shrink under memory pressure.
2054 spin_lock(&sbinfo->shrinklist_lock);
2056 * _careful to defend against unlocked access to
2057 * ->shrink_list in shmem_unused_huge_shrink()
2059 if (list_empty_careful(&info->shrinklist)) {
2060 list_add_tail(&info->shrinklist,
2061 &sbinfo->shrinklist);
2062 sbinfo->shrinklist_len++;
2064 spin_unlock(&sbinfo->shrinklist_lock);
2066 #endif /* CONFIG_FINEGRAINED_THP */
2068 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
2070 if (sgp == SGP_FALLOC)
2074 * Let SGP_WRITE caller clear ends if write does not fill page;
2075 * but SGP_FALLOC on a page fallocated earlier must initialize
2076 * it now, lest undo on failure cancel our earlier guarantee.
2078 if (sgp != SGP_WRITE && !PageUptodate(page)) {
2081 for (i = 0; i < compound_nr(page); i++) {
2082 clear_highpage(page + i);
2083 flush_dcache_page(page + i);
2085 SetPageUptodate(page);
2088 /* Perhaps the file has been truncated since we checked */
2089 if (sgp <= SGP_CACHE &&
2090 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2092 ClearPageDirty(page);
2093 delete_from_page_cache(page);
2094 spin_lock_irq(&info->lock);
2095 shmem_recalc_inode(inode);
2096 spin_unlock_irq(&info->lock);
2102 *pagep = page + index - hindex;
2109 shmem_inode_unacct_blocks(inode, compound_nr(page));
2111 if (PageTransHuge(page)) {
2121 if (error == -ENOSPC && !once++) {
2122 spin_lock_irq(&info->lock);
2123 shmem_recalc_inode(inode);
2124 spin_unlock_irq(&info->lock);
2127 if (error == -EEXIST)
2133 * This is like autoremove_wake_function, but it removes the wait queue
2134 * entry unconditionally - even if something else had already woken the
2137 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2139 int ret = default_wake_function(wait, mode, sync, key);
2140 list_del_init(&wait->entry);
2144 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2146 struct vm_area_struct *vma = vmf->vma;
2147 struct inode *inode = file_inode(vma->vm_file);
2148 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2151 vm_fault_t ret = VM_FAULT_LOCKED;
2154 * Trinity finds that probing a hole which tmpfs is punching can
2155 * prevent the hole-punch from ever completing: which in turn
2156 * locks writers out with its hold on i_mutex. So refrain from
2157 * faulting pages into the hole while it's being punched. Although
2158 * shmem_undo_range() does remove the additions, it may be unable to
2159 * keep up, as each new page needs its own unmap_mapping_range() call,
2160 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2162 * It does not matter if we sometimes reach this check just before the
2163 * hole-punch begins, so that one fault then races with the punch:
2164 * we just need to make racing faults a rare case.
2166 * The implementation below would be much simpler if we just used a
2167 * standard mutex or completion: but we cannot take i_mutex in fault,
2168 * and bloating every shmem inode for this unlikely case would be sad.
2170 if (unlikely(inode->i_private)) {
2171 struct shmem_falloc *shmem_falloc;
2173 spin_lock(&inode->i_lock);
2174 shmem_falloc = inode->i_private;
2176 shmem_falloc->waitq &&
2177 vmf->pgoff >= shmem_falloc->start &&
2178 vmf->pgoff < shmem_falloc->next) {
2180 wait_queue_head_t *shmem_falloc_waitq;
2181 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2183 ret = VM_FAULT_NOPAGE;
2184 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2186 ret = VM_FAULT_RETRY;
2188 shmem_falloc_waitq = shmem_falloc->waitq;
2189 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2190 TASK_UNINTERRUPTIBLE);
2191 spin_unlock(&inode->i_lock);
2195 * shmem_falloc_waitq points into the shmem_fallocate()
2196 * stack of the hole-punching task: shmem_falloc_waitq
2197 * is usually invalid by the time we reach here, but
2198 * finish_wait() does not dereference it in that case;
2199 * though i_lock needed lest racing with wake_up_all().
2201 spin_lock(&inode->i_lock);
2202 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2203 spin_unlock(&inode->i_lock);
2209 spin_unlock(&inode->i_lock);
2214 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2215 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2217 else if (vma->vm_flags & VM_HUGEPAGE)
2220 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2221 gfp, vma, vmf, &ret);
2223 return vmf_error(err);
2227 unsigned long shmem_get_unmapped_area(struct file *file,
2228 unsigned long uaddr, unsigned long len,
2229 unsigned long pgoff, unsigned long flags)
2231 unsigned long (*get_area)(struct file *,
2232 unsigned long, unsigned long, unsigned long, unsigned long);
2234 unsigned long offset;
2235 unsigned long inflated_len;
2236 unsigned long inflated_addr;
2237 unsigned long inflated_offset;
2239 if (len > TASK_SIZE)
2242 get_area = current->mm->get_unmapped_area;
2243 addr = get_area(file, uaddr, len, pgoff, flags);
2245 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2247 if (IS_ERR_VALUE(addr))
2249 if (addr & ~PAGE_MASK)
2251 if (addr > TASK_SIZE - len)
2254 if (shmem_huge == SHMEM_HUGE_DENY)
2256 if (len < HPAGE_PMD_SIZE)
2258 if (flags & MAP_FIXED)
2261 * Our priority is to support MAP_SHARED mapped hugely;
2262 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2263 * But if caller specified an address hint and we allocated area there
2264 * successfully, respect that as before.
2269 if (shmem_huge != SHMEM_HUGE_FORCE) {
2270 struct super_block *sb;
2273 VM_BUG_ON(file->f_op != &shmem_file_operations);
2274 sb = file_inode(file)->i_sb;
2277 * Called directly from mm/mmap.c, or drivers/char/mem.c
2278 * for "/dev/zero", to create a shared anonymous object.
2280 if (IS_ERR(shm_mnt))
2282 sb = shm_mnt->mnt_sb;
2284 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2288 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2289 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2291 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2294 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2295 if (inflated_len > TASK_SIZE)
2297 if (inflated_len < len)
2300 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2301 if (IS_ERR_VALUE(inflated_addr))
2303 if (inflated_addr & ~PAGE_MASK)
2306 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2307 inflated_addr += offset - inflated_offset;
2308 if (inflated_offset > offset)
2309 inflated_addr += HPAGE_PMD_SIZE;
2311 if (inflated_addr > TASK_SIZE - len)
2313 return inflated_addr;
2317 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2319 struct inode *inode = file_inode(vma->vm_file);
2320 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2323 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2326 struct inode *inode = file_inode(vma->vm_file);
2329 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2330 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2334 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2336 struct inode *inode = file_inode(file);
2337 struct shmem_inode_info *info = SHMEM_I(inode);
2338 int retval = -ENOMEM;
2341 * What serializes the accesses to info->flags?
2342 * ipc_lock_object() when called from shmctl_do_lock(),
2343 * no serialization needed when called from shm_destroy().
2345 if (lock && !(info->flags & VM_LOCKED)) {
2346 if (!user_shm_lock(inode->i_size, user))
2348 info->flags |= VM_LOCKED;
2349 mapping_set_unevictable(file->f_mapping);
2351 if (!lock && (info->flags & VM_LOCKED) && user) {
2352 user_shm_unlock(inode->i_size, user);
2353 info->flags &= ~VM_LOCKED;
2354 mapping_clear_unevictable(file->f_mapping);
2362 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2364 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2367 ret = seal_check_future_write(info->seals, vma);
2371 /* arm64 - allow memory tagging on RAM-based files */
2372 vma->vm_flags |= VM_MTE_ALLOWED;
2374 file_accessed(file);
2375 vma->vm_ops = &shmem_vm_ops;
2376 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2377 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2378 (vma->vm_end & HPAGE_PMD_MASK)) {
2379 khugepaged_enter(vma, vma->vm_flags);
2384 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2385 umode_t mode, dev_t dev, unsigned long flags)
2387 struct inode *inode;
2388 struct shmem_inode_info *info;
2389 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2392 if (shmem_reserve_inode(sb, &ino))
2395 inode = new_inode(sb);
2398 inode_init_owner(inode, dir, mode);
2399 inode->i_blocks = 0;
2400 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2401 inode->i_generation = prandom_u32();
2402 info = SHMEM_I(inode);
2403 memset(info, 0, (char *)inode - (char *)info);
2404 spin_lock_init(&info->lock);
2405 atomic_set(&info->stop_eviction, 0);
2406 info->seals = F_SEAL_SEAL;
2407 info->flags = flags & VM_NORESERVE;
2408 INIT_LIST_HEAD(&info->shrinklist);
2409 INIT_LIST_HEAD(&info->swaplist);
2410 simple_xattrs_init(&info->xattrs);
2411 cache_no_acl(inode);
2413 switch (mode & S_IFMT) {
2415 inode->i_op = &shmem_special_inode_operations;
2416 init_special_inode(inode, mode, dev);
2419 inode->i_mapping->a_ops = &shmem_aops;
2420 inode->i_op = &shmem_inode_operations;
2421 inode->i_fop = &shmem_file_operations;
2422 mpol_shared_policy_init(&info->policy,
2423 shmem_get_sbmpol(sbinfo));
2427 /* Some things misbehave if size == 0 on a directory */
2428 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2429 inode->i_op = &shmem_dir_inode_operations;
2430 inode->i_fop = &simple_dir_operations;
2434 * Must not load anything in the rbtree,
2435 * mpol_free_shared_policy will not be called.
2437 mpol_shared_policy_init(&info->policy, NULL);
2441 lockdep_annotate_inode_mutex_key(inode);
2443 shmem_free_inode(sb);
2447 bool shmem_mapping(struct address_space *mapping)
2449 return mapping->a_ops == &shmem_aops;
2452 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2454 struct vm_area_struct *dst_vma,
2455 unsigned long dst_addr,
2456 unsigned long src_addr,
2458 struct page **pagep)
2460 struct inode *inode = file_inode(dst_vma->vm_file);
2461 struct shmem_inode_info *info = SHMEM_I(inode);
2462 struct address_space *mapping = inode->i_mapping;
2463 gfp_t gfp = mapping_gfp_mask(mapping);
2464 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2468 pte_t _dst_pte, *dst_pte;
2470 pgoff_t offset, max_off;
2473 if (!shmem_inode_acct_block(inode, 1)) {
2475 * We may have got a page, returned -ENOENT triggering a retry,
2476 * and now we find ourselves with -ENOMEM. Release the page, to
2477 * avoid a BUG_ON in our caller.
2479 if (unlikely(*pagep)) {
2487 page = shmem_alloc_page(gfp, info, pgoff);
2489 goto out_unacct_blocks;
2491 if (!zeropage) { /* mcopy_atomic */
2492 page_kaddr = kmap_atomic(page);
2493 ret = copy_from_user(page_kaddr,
2494 (const void __user *)src_addr,
2496 kunmap_atomic(page_kaddr);
2498 /* fallback to copy_from_user outside mmap_lock */
2499 if (unlikely(ret)) {
2501 shmem_inode_unacct_blocks(inode, 1);
2502 /* don't free the page */
2505 } else { /* mfill_zeropage_atomic */
2506 clear_highpage(page);
2513 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2514 __SetPageLocked(page);
2515 __SetPageSwapBacked(page);
2516 __SetPageUptodate(page);
2519 offset = linear_page_index(dst_vma, dst_addr);
2520 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2521 if (unlikely(offset >= max_off))
2524 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2525 gfp & GFP_RECLAIM_MASK, dst_mm);
2529 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2530 if (dst_vma->vm_flags & VM_WRITE)
2531 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2534 * We don't set the pte dirty if the vma has no
2535 * VM_WRITE permission, so mark the page dirty or it
2536 * could be freed from under us. We could do it
2537 * unconditionally before unlock_page(), but doing it
2538 * only if VM_WRITE is not set is faster.
2540 set_page_dirty(page);
2543 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2546 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2547 if (unlikely(offset >= max_off))
2548 goto out_release_unlock;
2551 if (!pte_none(*dst_pte))
2552 goto out_release_unlock;
2554 lru_cache_add(page);
2556 spin_lock_irq(&info->lock);
2558 inode->i_blocks += BLOCKS_PER_PAGE;
2559 shmem_recalc_inode(inode);
2560 spin_unlock_irq(&info->lock);
2562 inc_mm_counter(dst_mm, mm_counter_file(page));
2563 page_add_file_rmap(page, false);
2564 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2566 /* No need to invalidate - it was non-present before */
2567 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2568 pte_unmap_unlock(dst_pte, ptl);
2574 pte_unmap_unlock(dst_pte, ptl);
2575 ClearPageDirty(page);
2576 delete_from_page_cache(page);
2581 shmem_inode_unacct_blocks(inode, 1);
2585 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2587 struct vm_area_struct *dst_vma,
2588 unsigned long dst_addr,
2589 unsigned long src_addr,
2590 struct page **pagep)
2592 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2593 dst_addr, src_addr, false, pagep);
2596 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2598 struct vm_area_struct *dst_vma,
2599 unsigned long dst_addr)
2601 struct page *page = NULL;
2603 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2604 dst_addr, 0, true, &page);
2608 static const struct inode_operations shmem_symlink_inode_operations;
2609 static const struct inode_operations shmem_short_symlink_operations;
2611 #ifdef CONFIG_TMPFS_XATTR
2612 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2614 #define shmem_initxattrs NULL
2618 shmem_write_begin(struct file *file, struct address_space *mapping,
2619 loff_t pos, unsigned len, unsigned flags,
2620 struct page **pagep, void **fsdata)
2622 struct inode *inode = mapping->host;
2623 struct shmem_inode_info *info = SHMEM_I(inode);
2624 pgoff_t index = pos >> PAGE_SHIFT;
2626 /* i_mutex is held by caller */
2627 if (unlikely(info->seals & (F_SEAL_GROW |
2628 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2629 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2631 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2635 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2639 shmem_write_end(struct file *file, struct address_space *mapping,
2640 loff_t pos, unsigned len, unsigned copied,
2641 struct page *page, void *fsdata)
2643 struct inode *inode = mapping->host;
2645 if (pos + copied > inode->i_size)
2646 i_size_write(inode, pos + copied);
2648 if (!PageUptodate(page)) {
2649 struct page *head = compound_head(page);
2650 if (PageTransCompound(page)) {
2652 #ifdef CONFIG_FINEGRAINED_THP
2653 for (i = 0; i < thp_nr_pages(page); i++) {
2654 if (head + i == page)
2656 clear_highpage(head + i);
2657 flush_dcache_page(head + i);
2659 #else /* CONFIG_FINEGRAINED_THP */
2660 for (i = 0; i < HPAGE_PMD_NR; i++) {
2661 if (head + i == page)
2663 clear_highpage(head + i);
2664 flush_dcache_page(head + i);
2666 #endif /* CONFIG_FINEGRAINED_THP */
2668 if (copied < PAGE_SIZE) {
2669 unsigned from = pos & (PAGE_SIZE - 1);
2670 zero_user_segments(page, 0, from,
2671 from + copied, PAGE_SIZE);
2673 SetPageUptodate(head);
2675 set_page_dirty(page);
2682 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2684 struct file *file = iocb->ki_filp;
2685 struct inode *inode = file_inode(file);
2686 struct address_space *mapping = inode->i_mapping;
2688 unsigned long offset;
2689 enum sgp_type sgp = SGP_READ;
2692 loff_t *ppos = &iocb->ki_pos;
2695 * Might this read be for a stacking filesystem? Then when reading
2696 * holes of a sparse file, we actually need to allocate those pages,
2697 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2699 if (!iter_is_iovec(to))
2702 index = *ppos >> PAGE_SHIFT;
2703 offset = *ppos & ~PAGE_MASK;
2706 struct page *page = NULL;
2708 unsigned long nr, ret;
2709 loff_t i_size = i_size_read(inode);
2711 end_index = i_size >> PAGE_SHIFT;
2712 if (index > end_index)
2714 if (index == end_index) {
2715 nr = i_size & ~PAGE_MASK;
2720 error = shmem_getpage(inode, index, &page, sgp);
2722 if (error == -EINVAL)
2727 if (sgp == SGP_CACHE)
2728 set_page_dirty(page);
2733 * We must evaluate after, since reads (unlike writes)
2734 * are called without i_mutex protection against truncate
2737 i_size = i_size_read(inode);
2738 end_index = i_size >> PAGE_SHIFT;
2739 if (index == end_index) {
2740 nr = i_size & ~PAGE_MASK;
2751 * If users can be writing to this page using arbitrary
2752 * virtual addresses, take care about potential aliasing
2753 * before reading the page on the kernel side.
2755 if (mapping_writably_mapped(mapping))
2756 flush_dcache_page(page);
2758 * Mark the page accessed if we read the beginning.
2761 mark_page_accessed(page);
2763 page = ZERO_PAGE(0);
2768 * Ok, we have the page, and it's up-to-date, so
2769 * now we can copy it to user space...
2771 ret = copy_page_to_iter(page, offset, nr, to);
2774 index += offset >> PAGE_SHIFT;
2775 offset &= ~PAGE_MASK;
2778 if (!iov_iter_count(to))
2787 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2788 file_accessed(file);
2789 return retval ? retval : error;
2793 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2795 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2796 pgoff_t index, pgoff_t end, int whence)
2799 struct pagevec pvec;
2800 pgoff_t indices[PAGEVEC_SIZE];
2804 pagevec_init(&pvec);
2805 pvec.nr = 1; /* start small: we may be there already */
2807 pvec.nr = find_get_entries(mapping, index,
2808 pvec.nr, pvec.pages, indices);
2810 if (whence == SEEK_DATA)
2814 for (i = 0; i < pvec.nr; i++, index++) {
2815 if (index < indices[i]) {
2816 if (whence == SEEK_HOLE) {
2822 page = pvec.pages[i];
2823 if (page && !xa_is_value(page)) {
2824 if (!PageUptodate(page))
2828 (page && whence == SEEK_DATA) ||
2829 (!page && whence == SEEK_HOLE)) {
2834 pagevec_remove_exceptionals(&pvec);
2835 pagevec_release(&pvec);
2836 pvec.nr = PAGEVEC_SIZE;
2842 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2844 struct address_space *mapping = file->f_mapping;
2845 struct inode *inode = mapping->host;
2849 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2850 return generic_file_llseek_size(file, offset, whence,
2851 MAX_LFS_FILESIZE, i_size_read(inode));
2853 /* We're holding i_mutex so we can access i_size directly */
2855 if (offset < 0 || offset >= inode->i_size)
2858 start = offset >> PAGE_SHIFT;
2859 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2860 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2861 new_offset <<= PAGE_SHIFT;
2862 if (new_offset > offset) {
2863 if (new_offset < inode->i_size)
2864 offset = new_offset;
2865 else if (whence == SEEK_DATA)
2868 offset = inode->i_size;
2873 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2874 inode_unlock(inode);
2878 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2881 struct inode *inode = file_inode(file);
2882 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2883 struct shmem_inode_info *info = SHMEM_I(inode);
2884 struct shmem_falloc shmem_falloc;
2885 pgoff_t start, index, end;
2888 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2893 if (mode & FALLOC_FL_PUNCH_HOLE) {
2894 struct address_space *mapping = file->f_mapping;
2895 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2896 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2897 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2899 /* protected by i_mutex */
2900 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2905 shmem_falloc.waitq = &shmem_falloc_waitq;
2906 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2907 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2908 spin_lock(&inode->i_lock);
2909 inode->i_private = &shmem_falloc;
2910 spin_unlock(&inode->i_lock);
2912 if ((u64)unmap_end > (u64)unmap_start)
2913 unmap_mapping_range(mapping, unmap_start,
2914 1 + unmap_end - unmap_start, 0);
2915 shmem_truncate_range(inode, offset, offset + len - 1);
2916 /* No need to unmap again: hole-punching leaves COWed pages */
2918 spin_lock(&inode->i_lock);
2919 inode->i_private = NULL;
2920 wake_up_all(&shmem_falloc_waitq);
2921 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2922 spin_unlock(&inode->i_lock);
2927 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2928 error = inode_newsize_ok(inode, offset + len);
2932 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2937 start = offset >> PAGE_SHIFT;
2938 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2939 /* Try to avoid a swapstorm if len is impossible to satisfy */
2940 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2945 shmem_falloc.waitq = NULL;
2946 shmem_falloc.start = start;
2947 shmem_falloc.next = start;
2948 shmem_falloc.nr_falloced = 0;
2949 shmem_falloc.nr_unswapped = 0;
2950 spin_lock(&inode->i_lock);
2951 inode->i_private = &shmem_falloc;
2952 spin_unlock(&inode->i_lock);
2954 for (index = start; index < end; index++) {
2958 * Good, the fallocate(2) manpage permits EINTR: we may have
2959 * been interrupted because we are using up too much memory.
2961 if (signal_pending(current))
2963 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2966 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2968 /* Remove the !PageUptodate pages we added */
2969 if (index > start) {
2970 shmem_undo_range(inode,
2971 (loff_t)start << PAGE_SHIFT,
2972 ((loff_t)index << PAGE_SHIFT) - 1, true);
2978 * Inform shmem_writepage() how far we have reached.
2979 * No need for lock or barrier: we have the page lock.
2981 shmem_falloc.next++;
2982 if (!PageUptodate(page))
2983 shmem_falloc.nr_falloced++;
2986 * If !PageUptodate, leave it that way so that freeable pages
2987 * can be recognized if we need to rollback on error later.
2988 * But set_page_dirty so that memory pressure will swap rather
2989 * than free the pages we are allocating (and SGP_CACHE pages
2990 * might still be clean: we now need to mark those dirty too).
2992 set_page_dirty(page);
2998 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2999 i_size_write(inode, offset + len);
3000 inode->i_ctime = current_time(inode);
3002 spin_lock(&inode->i_lock);
3003 inode->i_private = NULL;
3004 spin_unlock(&inode->i_lock);
3006 inode_unlock(inode);
3010 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
3012 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
3014 buf->f_type = TMPFS_MAGIC;
3015 buf->f_bsize = PAGE_SIZE;
3016 buf->f_namelen = NAME_MAX;
3017 if (sbinfo->max_blocks) {
3018 buf->f_blocks = sbinfo->max_blocks;
3020 buf->f_bfree = sbinfo->max_blocks -
3021 percpu_counter_sum(&sbinfo->used_blocks);
3023 if (sbinfo->max_inodes) {
3024 buf->f_files = sbinfo->max_inodes;
3025 buf->f_ffree = sbinfo->free_inodes;
3027 /* else leave those fields 0 like simple_statfs */
3032 * File creation. Allocate an inode, and we're done..
3035 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3037 struct inode *inode;
3038 int error = -ENOSPC;
3040 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
3042 error = simple_acl_create(dir, inode);
3045 error = security_inode_init_security(inode, dir,
3047 shmem_initxattrs, NULL);
3048 if (error && error != -EOPNOTSUPP)
3052 dir->i_size += BOGO_DIRENT_SIZE;
3053 dir->i_ctime = dir->i_mtime = current_time(dir);
3054 d_instantiate(dentry, inode);
3055 dget(dentry); /* Extra count - pin the dentry in core */
3064 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
3066 struct inode *inode;
3067 int error = -ENOSPC;
3069 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
3071 error = security_inode_init_security(inode, dir,
3073 shmem_initxattrs, NULL);
3074 if (error && error != -EOPNOTSUPP)
3076 error = simple_acl_create(dir, inode);
3079 d_tmpfile(dentry, inode);
3087 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3091 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
3097 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
3100 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
3106 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3108 struct inode *inode = d_inode(old_dentry);
3112 * No ordinary (disk based) filesystem counts links as inodes;
3113 * but each new link needs a new dentry, pinning lowmem, and
3114 * tmpfs dentries cannot be pruned until they are unlinked.
3115 * But if an O_TMPFILE file is linked into the tmpfs, the
3116 * first link must skip that, to get the accounting right.
3118 if (inode->i_nlink) {
3119 ret = shmem_reserve_inode(inode->i_sb, NULL);
3124 dir->i_size += BOGO_DIRENT_SIZE;
3125 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3127 ihold(inode); /* New dentry reference */
3128 dget(dentry); /* Extra pinning count for the created dentry */
3129 d_instantiate(dentry, inode);
3134 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3136 struct inode *inode = d_inode(dentry);
3138 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3139 shmem_free_inode(inode->i_sb);
3141 dir->i_size -= BOGO_DIRENT_SIZE;
3142 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3144 dput(dentry); /* Undo the count from "create" - this does all the work */
3148 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3150 if (!simple_empty(dentry))
3153 drop_nlink(d_inode(dentry));
3155 return shmem_unlink(dir, dentry);
3158 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3160 bool old_is_dir = d_is_dir(old_dentry);
3161 bool new_is_dir = d_is_dir(new_dentry);
3163 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3165 drop_nlink(old_dir);
3168 drop_nlink(new_dir);
3172 old_dir->i_ctime = old_dir->i_mtime =
3173 new_dir->i_ctime = new_dir->i_mtime =
3174 d_inode(old_dentry)->i_ctime =
3175 d_inode(new_dentry)->i_ctime = current_time(old_dir);
3180 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3182 struct dentry *whiteout;
3185 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3189 error = shmem_mknod(old_dir, whiteout,
3190 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3196 * Cheat and hash the whiteout while the old dentry is still in
3197 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3199 * d_lookup() will consistently find one of them at this point,
3200 * not sure which one, but that isn't even important.
3207 * The VFS layer already does all the dentry stuff for rename,
3208 * we just have to decrement the usage count for the target if
3209 * it exists so that the VFS layer correctly free's it when it
3212 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3214 struct inode *inode = d_inode(old_dentry);
3215 int they_are_dirs = S_ISDIR(inode->i_mode);
3217 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3220 if (flags & RENAME_EXCHANGE)
3221 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3223 if (!simple_empty(new_dentry))
3226 if (flags & RENAME_WHITEOUT) {
3229 error = shmem_whiteout(old_dir, old_dentry);
3234 if (d_really_is_positive(new_dentry)) {
3235 (void) shmem_unlink(new_dir, new_dentry);
3236 if (they_are_dirs) {
3237 drop_nlink(d_inode(new_dentry));
3238 drop_nlink(old_dir);
3240 } else if (they_are_dirs) {
3241 drop_nlink(old_dir);
3245 old_dir->i_size -= BOGO_DIRENT_SIZE;
3246 new_dir->i_size += BOGO_DIRENT_SIZE;
3247 old_dir->i_ctime = old_dir->i_mtime =
3248 new_dir->i_ctime = new_dir->i_mtime =
3249 inode->i_ctime = current_time(old_dir);
3253 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3257 struct inode *inode;
3260 len = strlen(symname) + 1;
3261 if (len > PAGE_SIZE)
3262 return -ENAMETOOLONG;
3264 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3269 error = security_inode_init_security(inode, dir, &dentry->d_name,
3270 shmem_initxattrs, NULL);
3271 if (error && error != -EOPNOTSUPP) {
3276 inode->i_size = len-1;
3277 if (len <= SHORT_SYMLINK_LEN) {
3278 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3279 if (!inode->i_link) {
3283 inode->i_op = &shmem_short_symlink_operations;
3285 inode_nohighmem(inode);
3286 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3291 inode->i_mapping->a_ops = &shmem_aops;
3292 inode->i_op = &shmem_symlink_inode_operations;
3293 memcpy(page_address(page), symname, len);
3294 SetPageUptodate(page);
3295 set_page_dirty(page);
3299 dir->i_size += BOGO_DIRENT_SIZE;
3300 dir->i_ctime = dir->i_mtime = current_time(dir);
3301 d_instantiate(dentry, inode);
3306 static void shmem_put_link(void *arg)
3308 mark_page_accessed(arg);
3312 static const char *shmem_get_link(struct dentry *dentry,
3313 struct inode *inode,
3314 struct delayed_call *done)
3316 struct page *page = NULL;
3319 page = find_get_page(inode->i_mapping, 0);
3321 return ERR_PTR(-ECHILD);
3322 if (!PageUptodate(page)) {
3324 return ERR_PTR(-ECHILD);
3327 error = shmem_getpage(inode, 0, &page, SGP_READ);
3329 return ERR_PTR(error);
3332 set_delayed_call(done, shmem_put_link, page);
3333 return page_address(page);
3336 #ifdef CONFIG_TMPFS_XATTR
3338 * Superblocks without xattr inode operations may get some security.* xattr
3339 * support from the LSM "for free". As soon as we have any other xattrs
3340 * like ACLs, we also need to implement the security.* handlers at
3341 * filesystem level, though.
3345 * Callback for security_inode_init_security() for acquiring xattrs.
3347 static int shmem_initxattrs(struct inode *inode,
3348 const struct xattr *xattr_array,
3351 struct shmem_inode_info *info = SHMEM_I(inode);
3352 const struct xattr *xattr;
3353 struct simple_xattr *new_xattr;
3356 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3357 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3361 len = strlen(xattr->name) + 1;
3362 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3364 if (!new_xattr->name) {
3369 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3370 XATTR_SECURITY_PREFIX_LEN);
3371 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3374 simple_xattr_list_add(&info->xattrs, new_xattr);
3380 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3381 struct dentry *unused, struct inode *inode,
3382 const char *name, void *buffer, size_t size)
3384 struct shmem_inode_info *info = SHMEM_I(inode);
3386 name = xattr_full_name(handler, name);
3387 return simple_xattr_get(&info->xattrs, name, buffer, size);
3390 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3391 struct dentry *unused, struct inode *inode,
3392 const char *name, const void *value,
3393 size_t size, int flags)
3395 struct shmem_inode_info *info = SHMEM_I(inode);
3397 name = xattr_full_name(handler, name);
3398 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3401 static const struct xattr_handler shmem_security_xattr_handler = {
3402 .prefix = XATTR_SECURITY_PREFIX,
3403 .get = shmem_xattr_handler_get,
3404 .set = shmem_xattr_handler_set,
3407 static const struct xattr_handler shmem_trusted_xattr_handler = {
3408 .prefix = XATTR_TRUSTED_PREFIX,
3409 .get = shmem_xattr_handler_get,
3410 .set = shmem_xattr_handler_set,
3413 static const struct xattr_handler *shmem_xattr_handlers[] = {
3414 #ifdef CONFIG_TMPFS_POSIX_ACL
3415 &posix_acl_access_xattr_handler,
3416 &posix_acl_default_xattr_handler,
3418 &shmem_security_xattr_handler,
3419 &shmem_trusted_xattr_handler,
3423 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3425 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3426 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3428 #endif /* CONFIG_TMPFS_XATTR */
3430 static const struct inode_operations shmem_short_symlink_operations = {
3431 .get_link = simple_get_link,
3432 #ifdef CONFIG_TMPFS_XATTR
3433 .listxattr = shmem_listxattr,
3437 static const struct inode_operations shmem_symlink_inode_operations = {
3438 .get_link = shmem_get_link,
3439 #ifdef CONFIG_TMPFS_XATTR
3440 .listxattr = shmem_listxattr,
3444 static struct dentry *shmem_get_parent(struct dentry *child)
3446 return ERR_PTR(-ESTALE);
3449 static int shmem_match(struct inode *ino, void *vfh)
3453 inum = (inum << 32) | fh[1];
3454 return ino->i_ino == inum && fh[0] == ino->i_generation;
3457 /* Find any alias of inode, but prefer a hashed alias */
3458 static struct dentry *shmem_find_alias(struct inode *inode)
3460 struct dentry *alias = d_find_alias(inode);
3462 return alias ?: d_find_any_alias(inode);
3466 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3467 struct fid *fid, int fh_len, int fh_type)
3469 struct inode *inode;
3470 struct dentry *dentry = NULL;
3477 inum = (inum << 32) | fid->raw[1];
3479 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3480 shmem_match, fid->raw);
3482 dentry = shmem_find_alias(inode);
3489 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3490 struct inode *parent)
3494 return FILEID_INVALID;
3497 if (inode_unhashed(inode)) {
3498 /* Unfortunately insert_inode_hash is not idempotent,
3499 * so as we hash inodes here rather than at creation
3500 * time, we need a lock to ensure we only try
3503 static DEFINE_SPINLOCK(lock);
3505 if (inode_unhashed(inode))
3506 __insert_inode_hash(inode,
3507 inode->i_ino + inode->i_generation);
3511 fh[0] = inode->i_generation;
3512 fh[1] = inode->i_ino;
3513 fh[2] = ((__u64)inode->i_ino) >> 32;
3519 static const struct export_operations shmem_export_ops = {
3520 .get_parent = shmem_get_parent,
3521 .encode_fh = shmem_encode_fh,
3522 .fh_to_dentry = shmem_fh_to_dentry,
3538 static const struct constant_table shmem_param_enums_huge[] = {
3539 {"never", SHMEM_HUGE_NEVER },
3540 {"always", SHMEM_HUGE_ALWAYS },
3541 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3542 {"advise", SHMEM_HUGE_ADVISE },
3546 const struct fs_parameter_spec shmem_fs_parameters[] = {
3547 fsparam_u32 ("gid", Opt_gid),
3548 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3549 fsparam_u32oct("mode", Opt_mode),
3550 fsparam_string("mpol", Opt_mpol),
3551 fsparam_string("nr_blocks", Opt_nr_blocks),
3552 fsparam_string("nr_inodes", Opt_nr_inodes),
3553 fsparam_string("size", Opt_size),
3554 fsparam_u32 ("uid", Opt_uid),
3555 fsparam_flag ("inode32", Opt_inode32),
3556 fsparam_flag ("inode64", Opt_inode64),
3560 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3562 struct shmem_options *ctx = fc->fs_private;
3563 struct fs_parse_result result;
3564 unsigned long long size;
3568 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3574 size = memparse(param->string, &rest);
3576 size <<= PAGE_SHIFT;
3577 size *= totalram_pages();
3583 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3584 ctx->seen |= SHMEM_SEEN_BLOCKS;
3587 ctx->blocks = memparse(param->string, &rest);
3590 ctx->seen |= SHMEM_SEEN_BLOCKS;
3593 ctx->inodes = memparse(param->string, &rest);
3596 ctx->seen |= SHMEM_SEEN_INODES;
3599 ctx->mode = result.uint_32 & 07777;
3602 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3603 if (!uid_valid(ctx->uid))
3607 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3608 if (!gid_valid(ctx->gid))
3612 ctx->huge = result.uint_32;
3613 if (ctx->huge != SHMEM_HUGE_NEVER &&
3614 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3615 has_transparent_hugepage()))
3616 goto unsupported_parameter;
3617 ctx->seen |= SHMEM_SEEN_HUGE;
3620 if (IS_ENABLED(CONFIG_NUMA)) {
3621 mpol_put(ctx->mpol);
3623 if (mpol_parse_str(param->string, &ctx->mpol))
3627 goto unsupported_parameter;
3629 ctx->full_inums = false;
3630 ctx->seen |= SHMEM_SEEN_INUMS;
3633 if (sizeof(ino_t) < 8) {
3635 "Cannot use inode64 with <64bit inums in kernel\n");
3637 ctx->full_inums = true;
3638 ctx->seen |= SHMEM_SEEN_INUMS;
3643 unsupported_parameter:
3644 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3646 return invalfc(fc, "Bad value for '%s'", param->key);
3649 static int shmem_parse_options(struct fs_context *fc, void *data)
3651 char *options = data;
3654 int err = security_sb_eat_lsm_opts(options, &fc->security);
3659 while (options != NULL) {
3660 char *this_char = options;
3663 * NUL-terminate this option: unfortunately,
3664 * mount options form a comma-separated list,
3665 * but mpol's nodelist may also contain commas.
3667 options = strchr(options, ',');
3668 if (options == NULL)
3671 if (!isdigit(*options)) {
3677 char *value = strchr(this_char,'=');
3683 len = strlen(value);
3685 err = vfs_parse_fs_string(fc, this_char, value, len);
3694 * Reconfigure a shmem filesystem.
3696 * Note that we disallow change from limited->unlimited blocks/inodes while any
3697 * are in use; but we must separately disallow unlimited->limited, because in
3698 * that case we have no record of how much is already in use.
3700 static int shmem_reconfigure(struct fs_context *fc)
3702 struct shmem_options *ctx = fc->fs_private;
3703 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3704 unsigned long inodes;
3705 struct mempolicy *mpol = NULL;
3708 raw_spin_lock(&sbinfo->stat_lock);
3709 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3710 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3711 if (!sbinfo->max_blocks) {
3712 err = "Cannot retroactively limit size";
3715 if (percpu_counter_compare(&sbinfo->used_blocks,
3717 err = "Too small a size for current use";
3721 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3722 if (!sbinfo->max_inodes) {
3723 err = "Cannot retroactively limit inodes";
3726 if (ctx->inodes < inodes) {
3727 err = "Too few inodes for current use";
3732 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3733 sbinfo->next_ino > UINT_MAX) {
3734 err = "Current inum too high to switch to 32-bit inums";
3738 if (ctx->seen & SHMEM_SEEN_HUGE)
3739 sbinfo->huge = ctx->huge;
3740 if (ctx->seen & SHMEM_SEEN_INUMS)
3741 sbinfo->full_inums = ctx->full_inums;
3742 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3743 sbinfo->max_blocks = ctx->blocks;
3744 if (ctx->seen & SHMEM_SEEN_INODES) {
3745 sbinfo->max_inodes = ctx->inodes;
3746 sbinfo->free_inodes = ctx->inodes - inodes;
3750 * Preserve previous mempolicy unless mpol remount option was specified.
3753 mpol = sbinfo->mpol;
3754 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3757 raw_spin_unlock(&sbinfo->stat_lock);
3761 raw_spin_unlock(&sbinfo->stat_lock);
3762 return invalfc(fc, "%s", err);
3765 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3767 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3769 if (sbinfo->max_blocks != shmem_default_max_blocks())
3770 seq_printf(seq, ",size=%luk",
3771 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3772 if (sbinfo->max_inodes != shmem_default_max_inodes())
3773 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3774 if (sbinfo->mode != (0777 | S_ISVTX))
3775 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3776 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3777 seq_printf(seq, ",uid=%u",
3778 from_kuid_munged(&init_user_ns, sbinfo->uid));
3779 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3780 seq_printf(seq, ",gid=%u",
3781 from_kgid_munged(&init_user_ns, sbinfo->gid));
3784 * Showing inode{64,32} might be useful even if it's the system default,
3785 * since then people don't have to resort to checking both here and
3786 * /proc/config.gz to confirm 64-bit inums were successfully applied
3787 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3789 * We hide it when inode64 isn't the default and we are using 32-bit
3790 * inodes, since that probably just means the feature isn't even under
3795 * +-----------------+-----------------+
3796 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3797 * +------------------+-----------------+-----------------+
3798 * | full_inums=true | show | show |
3799 * | full_inums=false | show | hide |
3800 * +------------------+-----------------+-----------------+
3803 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3804 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3805 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3806 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3808 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3810 shmem_show_mpol(seq, sbinfo->mpol);
3814 #endif /* CONFIG_TMPFS */
3816 static void shmem_put_super(struct super_block *sb)
3818 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3820 free_percpu(sbinfo->ino_batch);
3821 percpu_counter_destroy(&sbinfo->used_blocks);
3822 mpol_put(sbinfo->mpol);
3824 sb->s_fs_info = NULL;
3827 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3829 struct shmem_options *ctx = fc->fs_private;
3830 struct inode *inode;
3831 struct shmem_sb_info *sbinfo;
3834 /* Round up to L1_CACHE_BYTES to resist false sharing */
3835 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3836 L1_CACHE_BYTES), GFP_KERNEL);
3840 sb->s_fs_info = sbinfo;
3844 * Per default we only allow half of the physical ram per
3845 * tmpfs instance, limiting inodes to one per page of lowmem;
3846 * but the internal instance is left unlimited.
3848 if (!(sb->s_flags & SB_KERNMOUNT)) {
3849 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3850 ctx->blocks = shmem_default_max_blocks();
3851 if (!(ctx->seen & SHMEM_SEEN_INODES))
3852 ctx->inodes = shmem_default_max_inodes();
3853 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3854 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3856 sb->s_flags |= SB_NOUSER;
3858 sb->s_export_op = &shmem_export_ops;
3859 sb->s_flags |= SB_NOSEC;
3861 sb->s_flags |= SB_NOUSER;
3863 sbinfo->max_blocks = ctx->blocks;
3864 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3865 if (sb->s_flags & SB_KERNMOUNT) {
3866 sbinfo->ino_batch = alloc_percpu(ino_t);
3867 if (!sbinfo->ino_batch)
3870 sbinfo->uid = ctx->uid;
3871 sbinfo->gid = ctx->gid;
3872 sbinfo->full_inums = ctx->full_inums;
3873 sbinfo->mode = ctx->mode;
3874 sbinfo->huge = ctx->huge;
3875 sbinfo->mpol = ctx->mpol;
3878 raw_spin_lock_init(&sbinfo->stat_lock);
3879 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3881 spin_lock_init(&sbinfo->shrinklist_lock);
3882 INIT_LIST_HEAD(&sbinfo->shrinklist);
3884 sb->s_maxbytes = MAX_LFS_FILESIZE;
3885 sb->s_blocksize = PAGE_SIZE;
3886 sb->s_blocksize_bits = PAGE_SHIFT;
3887 sb->s_magic = TMPFS_MAGIC;
3888 sb->s_op = &shmem_ops;
3889 sb->s_time_gran = 1;
3890 #ifdef CONFIG_TMPFS_XATTR
3891 sb->s_xattr = shmem_xattr_handlers;
3893 #ifdef CONFIG_TMPFS_POSIX_ACL
3894 sb->s_flags |= SB_POSIXACL;
3896 uuid_gen(&sb->s_uuid);
3898 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3901 inode->i_uid = sbinfo->uid;
3902 inode->i_gid = sbinfo->gid;
3903 sb->s_root = d_make_root(inode);
3909 shmem_put_super(sb);
3913 static int shmem_get_tree(struct fs_context *fc)
3915 return get_tree_nodev(fc, shmem_fill_super);
3918 static void shmem_free_fc(struct fs_context *fc)
3920 struct shmem_options *ctx = fc->fs_private;
3923 mpol_put(ctx->mpol);
3928 static const struct fs_context_operations shmem_fs_context_ops = {
3929 .free = shmem_free_fc,
3930 .get_tree = shmem_get_tree,
3932 .parse_monolithic = shmem_parse_options,
3933 .parse_param = shmem_parse_one,
3934 .reconfigure = shmem_reconfigure,
3938 static struct kmem_cache *shmem_inode_cachep;
3940 static struct inode *shmem_alloc_inode(struct super_block *sb)
3942 struct shmem_inode_info *info;
3943 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3946 return &info->vfs_inode;
3949 static void shmem_free_in_core_inode(struct inode *inode)
3951 if (S_ISLNK(inode->i_mode))
3952 kfree(inode->i_link);
3953 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3956 static void shmem_destroy_inode(struct inode *inode)
3958 if (S_ISREG(inode->i_mode))
3959 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3962 static void shmem_init_inode(void *foo)
3964 struct shmem_inode_info *info = foo;
3965 inode_init_once(&info->vfs_inode);
3968 static void shmem_init_inodecache(void)
3970 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3971 sizeof(struct shmem_inode_info),
3972 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3975 static void shmem_destroy_inodecache(void)
3977 kmem_cache_destroy(shmem_inode_cachep);
3980 static const struct address_space_operations shmem_aops = {
3981 .writepage = shmem_writepage,
3982 .set_page_dirty = __set_page_dirty_no_writeback,
3984 .write_begin = shmem_write_begin,
3985 .write_end = shmem_write_end,
3987 #ifdef CONFIG_MIGRATION
3988 .migratepage = migrate_page,
3990 .error_remove_page = generic_error_remove_page,
3993 static const struct file_operations shmem_file_operations = {
3995 .get_unmapped_area = shmem_get_unmapped_area,
3997 .llseek = shmem_file_llseek,
3998 .read_iter = shmem_file_read_iter,
3999 .write_iter = generic_file_write_iter,
4000 .fsync = noop_fsync,
4001 .splice_read = generic_file_splice_read,
4002 .splice_write = iter_file_splice_write,
4003 .fallocate = shmem_fallocate,
4007 static const struct inode_operations shmem_inode_operations = {
4008 .getattr = shmem_getattr,
4009 .setattr = shmem_setattr,
4010 #ifdef CONFIG_TMPFS_XATTR
4011 .listxattr = shmem_listxattr,
4012 .set_acl = simple_set_acl,
4016 static const struct inode_operations shmem_dir_inode_operations = {
4018 .create = shmem_create,
4019 .lookup = simple_lookup,
4021 .unlink = shmem_unlink,
4022 .symlink = shmem_symlink,
4023 .mkdir = shmem_mkdir,
4024 .rmdir = shmem_rmdir,
4025 .mknod = shmem_mknod,
4026 .rename = shmem_rename2,
4027 .tmpfile = shmem_tmpfile,
4029 #ifdef CONFIG_TMPFS_XATTR
4030 .listxattr = shmem_listxattr,
4032 #ifdef CONFIG_TMPFS_POSIX_ACL
4033 .setattr = shmem_setattr,
4034 .set_acl = simple_set_acl,
4038 static const struct inode_operations shmem_special_inode_operations = {
4039 #ifdef CONFIG_TMPFS_XATTR
4040 .listxattr = shmem_listxattr,
4042 #ifdef CONFIG_TMPFS_POSIX_ACL
4043 .setattr = shmem_setattr,
4044 .set_acl = simple_set_acl,
4048 static const struct super_operations shmem_ops = {
4049 .alloc_inode = shmem_alloc_inode,
4050 .free_inode = shmem_free_in_core_inode,
4051 .destroy_inode = shmem_destroy_inode,
4053 .statfs = shmem_statfs,
4054 .show_options = shmem_show_options,
4056 .evict_inode = shmem_evict_inode,
4057 .drop_inode = generic_delete_inode,
4058 .put_super = shmem_put_super,
4059 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4060 .nr_cached_objects = shmem_unused_huge_count,
4061 .free_cached_objects = shmem_unused_huge_scan,
4065 static const struct vm_operations_struct shmem_vm_ops = {
4066 .fault = shmem_fault,
4067 .map_pages = filemap_map_pages,
4069 .set_policy = shmem_set_policy,
4070 .get_policy = shmem_get_policy,
4074 int shmem_init_fs_context(struct fs_context *fc)
4076 struct shmem_options *ctx;
4078 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4082 ctx->mode = 0777 | S_ISVTX;
4083 ctx->uid = current_fsuid();
4084 ctx->gid = current_fsgid();
4086 fc->fs_private = ctx;
4087 fc->ops = &shmem_fs_context_ops;
4091 static struct file_system_type shmem_fs_type = {
4092 .owner = THIS_MODULE,
4094 .init_fs_context = shmem_init_fs_context,
4096 .parameters = shmem_fs_parameters,
4098 .kill_sb = kill_litter_super,
4099 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
4102 int __init shmem_init(void)
4106 shmem_init_inodecache();
4108 error = register_filesystem(&shmem_fs_type);
4110 pr_err("Could not register tmpfs\n");
4114 shm_mnt = kern_mount(&shmem_fs_type);
4115 if (IS_ERR(shm_mnt)) {
4116 error = PTR_ERR(shm_mnt);
4117 pr_err("Could not kern_mount tmpfs\n");
4121 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4122 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4123 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4125 shmem_huge = 0; /* just in case it was patched */
4130 unregister_filesystem(&shmem_fs_type);
4132 shmem_destroy_inodecache();
4133 shm_mnt = ERR_PTR(error);
4137 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4138 static ssize_t shmem_enabled_show(struct kobject *kobj,
4139 struct kobj_attribute *attr, char *buf)
4141 static const int values[] = {
4143 SHMEM_HUGE_WITHIN_SIZE,
4151 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
4152 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
4154 count += sprintf(buf + count, fmt,
4155 shmem_format_huge(values[i]));
4157 buf[count - 1] = '\n';
4161 static ssize_t shmem_enabled_store(struct kobject *kobj,
4162 struct kobj_attribute *attr, const char *buf, size_t count)
4167 if (count + 1 > sizeof(tmp))
4169 memcpy(tmp, buf, count);
4171 if (count && tmp[count - 1] == '\n')
4172 tmp[count - 1] = '\0';
4174 huge = shmem_parse_huge(tmp);
4175 if (huge == -EINVAL)
4177 if (!has_transparent_hugepage() &&
4178 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4182 if (shmem_huge > SHMEM_HUGE_DENY)
4183 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4187 struct kobj_attribute shmem_enabled_attr =
4188 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4189 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4191 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4192 bool shmem_huge_enabled(struct vm_area_struct *vma)
4194 struct inode *inode = file_inode(vma->vm_file);
4195 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4199 if (!transhuge_vma_enabled(vma, vma->vm_flags))
4201 if (shmem_huge == SHMEM_HUGE_FORCE)
4203 if (shmem_huge == SHMEM_HUGE_DENY)
4205 switch (sbinfo->huge) {
4206 case SHMEM_HUGE_NEVER:
4208 case SHMEM_HUGE_ALWAYS:
4210 case SHMEM_HUGE_WITHIN_SIZE:
4211 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4212 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4213 if (i_size >= HPAGE_PMD_SIZE &&
4214 i_size >> PAGE_SHIFT >= off)
4216 #ifdef CONFIG_FINEGRAINED_THP
4217 off = round_up(vma->vm_pgoff, HPAGE_CONT_PTE_NR);
4218 if (i_size >= HPAGE_CONT_PTE_SIZE &&
4219 i_size >> PAGE_SHIFT >= off)
4221 #endif /* CONFIG_FINEGRAINED_THP */
4223 case SHMEM_HUGE_ADVISE:
4224 /* TODO: implement fadvise() hints */
4225 return (vma->vm_flags & VM_HUGEPAGE);
4231 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4233 #else /* !CONFIG_SHMEM */
4236 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4238 * This is intended for small system where the benefits of the full
4239 * shmem code (swap-backed and resource-limited) are outweighed by
4240 * their complexity. On systems without swap this code should be
4241 * effectively equivalent, but much lighter weight.
4244 static struct file_system_type shmem_fs_type = {
4246 .init_fs_context = ramfs_init_fs_context,
4247 .parameters = ramfs_fs_parameters,
4248 .kill_sb = kill_litter_super,
4249 .fs_flags = FS_USERNS_MOUNT,
4252 int __init shmem_init(void)
4254 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4256 shm_mnt = kern_mount(&shmem_fs_type);
4257 BUG_ON(IS_ERR(shm_mnt));
4262 int shmem_unuse(unsigned int type, bool frontswap,
4263 unsigned long *fs_pages_to_unuse)
4268 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4273 void shmem_unlock_mapping(struct address_space *mapping)
4278 unsigned long shmem_get_unmapped_area(struct file *file,
4279 unsigned long addr, unsigned long len,
4280 unsigned long pgoff, unsigned long flags)
4282 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4286 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4288 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4290 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4292 #define shmem_vm_ops generic_file_vm_ops
4293 #define shmem_file_operations ramfs_file_operations
4294 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4295 #define shmem_acct_size(flags, size) 0
4296 #define shmem_unacct_size(flags, size) do {} while (0)
4298 #endif /* CONFIG_SHMEM */
4302 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4303 unsigned long flags, unsigned int i_flags)
4305 struct inode *inode;
4309 return ERR_CAST(mnt);
4311 if (size < 0 || size > MAX_LFS_FILESIZE)
4312 return ERR_PTR(-EINVAL);
4314 if (shmem_acct_size(flags, size))
4315 return ERR_PTR(-ENOMEM);
4317 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4319 if (unlikely(!inode)) {
4320 shmem_unacct_size(flags, size);
4321 return ERR_PTR(-ENOSPC);
4323 inode->i_flags |= i_flags;
4324 inode->i_size = size;
4325 clear_nlink(inode); /* It is unlinked */
4326 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4328 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4329 &shmem_file_operations);
4336 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4337 * kernel internal. There will be NO LSM permission checks against the
4338 * underlying inode. So users of this interface must do LSM checks at a
4339 * higher layer. The users are the big_key and shm implementations. LSM
4340 * checks are provided at the key or shm level rather than the inode.
4341 * @name: name for dentry (to be seen in /proc/<pid>/maps
4342 * @size: size to be set for the file
4343 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4345 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4347 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4351 * shmem_file_setup - get an unlinked file living in tmpfs
4352 * @name: name for dentry (to be seen in /proc/<pid>/maps
4353 * @size: size to be set for the file
4354 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4356 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4358 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4360 EXPORT_SYMBOL_GPL(shmem_file_setup);
4363 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4364 * @mnt: the tmpfs mount where the file will be created
4365 * @name: name for dentry (to be seen in /proc/<pid>/maps
4366 * @size: size to be set for the file
4367 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4369 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4370 loff_t size, unsigned long flags)
4372 return __shmem_file_setup(mnt, name, size, flags, 0);
4374 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4377 * shmem_zero_setup - setup a shared anonymous mapping
4378 * @vma: the vma to be mmapped is prepared by do_mmap
4380 int shmem_zero_setup(struct vm_area_struct *vma)
4383 loff_t size = vma->vm_end - vma->vm_start;
4386 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4387 * between XFS directory reading and selinux: since this file is only
4388 * accessible to the user through its mapping, use S_PRIVATE flag to
4389 * bypass file security, in the same way as shmem_kernel_file_setup().
4391 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4393 return PTR_ERR(file);
4397 vma->vm_file = file;
4398 vma->vm_ops = &shmem_vm_ops;
4400 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4401 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4402 (vma->vm_end & HPAGE_PMD_MASK)) {
4403 khugepaged_enter(vma, vma->vm_flags);
4410 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4411 * @mapping: the page's address_space
4412 * @index: the page index
4413 * @gfp: the page allocator flags to use if allocating
4415 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4416 * with any new page allocations done using the specified allocation flags.
4417 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4418 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4419 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4421 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4422 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4424 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4425 pgoff_t index, gfp_t gfp)
4428 struct inode *inode = mapping->host;
4432 BUG_ON(mapping->a_ops != &shmem_aops);
4433 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4434 gfp, NULL, NULL, NULL);
4436 page = ERR_PTR(error);
4442 * The tiny !SHMEM case uses ramfs without swap
4444 return read_cache_page_gfp(mapping, index, gfp);
4447 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);