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/hugetlb.h>
38 #include <linux/fs_parser.h>
39 #include <linux/swapfile.h>
42 static struct vfsmount *shm_mnt;
46 * This virtual memory filesystem is heavily based on the ramfs. It
47 * extends ramfs by the ability to use swap and honor resource limits
48 * which makes it a completely usable filesystem.
51 #include <linux/xattr.h>
52 #include <linux/exportfs.h>
53 #include <linux/posix_acl.h>
54 #include <linux/posix_acl_xattr.h>
55 #include <linux/mman.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/backing-dev.h>
59 #include <linux/shmem_fs.h>
60 #include <linux/writeback.h>
61 #include <linux/pagevec.h>
62 #include <linux/percpu_counter.h>
63 #include <linux/falloc.h>
64 #include <linux/splice.h>
65 #include <linux/security.h>
66 #include <linux/swapops.h>
67 #include <linux/mempolicy.h>
68 #include <linux/namei.h>
69 #include <linux/ctype.h>
70 #include <linux/migrate.h>
71 #include <linux/highmem.h>
72 #include <linux/seq_file.h>
73 #include <linux/magic.h>
74 #include <linux/syscalls.h>
75 #include <linux/fcntl.h>
76 #include <uapi/linux/memfd.h>
77 #include <linux/userfaultfd_k.h>
78 #include <linux/rmap.h>
79 #include <linux/uuid.h>
81 #include <linux/uaccess.h>
85 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
86 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
88 /* Pretend that each entry is of this size in directory's i_size */
89 #define BOGO_DIRENT_SIZE 20
91 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
92 #define SHORT_SYMLINK_LEN 128
95 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
96 * inode->i_private (with i_rwsem making sure that it has only one user at
97 * a time): we would prefer not to enlarge the shmem inode just for that.
100 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
101 pgoff_t start; /* start of range currently being fallocated */
102 pgoff_t next; /* the next page offset to be fallocated */
103 pgoff_t nr_falloced; /* how many new pages have been fallocated */
104 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
107 struct shmem_options {
108 unsigned long long blocks;
109 unsigned long long inodes;
110 struct mempolicy *mpol;
117 #define SHMEM_SEEN_BLOCKS 1
118 #define SHMEM_SEEN_INODES 2
119 #define SHMEM_SEEN_HUGE 4
120 #define SHMEM_SEEN_INUMS 8
124 static unsigned long shmem_default_max_blocks(void)
126 return totalram_pages() / 2;
129 static unsigned long shmem_default_max_inodes(void)
131 unsigned long nr_pages = totalram_pages();
133 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
137 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
138 struct folio **foliop, enum sgp_type sgp,
139 gfp_t gfp, struct vm_area_struct *vma,
140 vm_fault_t *fault_type);
141 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
142 struct page **pagep, enum sgp_type sgp,
143 gfp_t gfp, struct vm_area_struct *vma,
144 struct vm_fault *vmf, vm_fault_t *fault_type);
146 int shmem_getpage(struct inode *inode, pgoff_t index,
147 struct page **pagep, enum sgp_type sgp)
149 return shmem_getpage_gfp(inode, index, pagep, sgp,
150 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
153 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
155 return sb->s_fs_info;
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
164 static inline int shmem_acct_size(unsigned long flags, loff_t size)
166 return (flags & VM_NORESERVE) ?
167 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
170 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
172 if (!(flags & VM_NORESERVE))
173 vm_unacct_memory(VM_ACCT(size));
176 static inline int shmem_reacct_size(unsigned long flags,
177 loff_t oldsize, loff_t newsize)
179 if (!(flags & VM_NORESERVE)) {
180 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
181 return security_vm_enough_memory_mm(current->mm,
182 VM_ACCT(newsize) - VM_ACCT(oldsize));
183 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
184 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
190 * ... whereas tmpfs objects are accounted incrementally as
191 * pages are allocated, in order to allow large sparse files.
192 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
193 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
195 static inline int shmem_acct_block(unsigned long flags, long pages)
197 if (!(flags & VM_NORESERVE))
200 return security_vm_enough_memory_mm(current->mm,
201 pages * VM_ACCT(PAGE_SIZE));
204 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
206 if (flags & VM_NORESERVE)
207 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
210 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
212 struct shmem_inode_info *info = SHMEM_I(inode);
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
215 if (shmem_acct_block(info->flags, pages))
218 if (sbinfo->max_blocks) {
219 if (percpu_counter_compare(&sbinfo->used_blocks,
220 sbinfo->max_blocks - pages) > 0)
222 percpu_counter_add(&sbinfo->used_blocks, pages);
228 shmem_unacct_blocks(info->flags, pages);
232 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
234 struct shmem_inode_info *info = SHMEM_I(inode);
235 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
237 if (sbinfo->max_blocks)
238 percpu_counter_sub(&sbinfo->used_blocks, pages);
239 shmem_unacct_blocks(info->flags, pages);
242 static const struct super_operations shmem_ops;
243 const struct address_space_operations shmem_aops;
244 static const struct file_operations shmem_file_operations;
245 static const struct inode_operations shmem_inode_operations;
246 static const struct inode_operations shmem_dir_inode_operations;
247 static const struct inode_operations shmem_special_inode_operations;
248 static const struct vm_operations_struct shmem_vm_ops;
249 static struct file_system_type shmem_fs_type;
251 bool vma_is_shmem(struct vm_area_struct *vma)
253 return vma->vm_ops == &shmem_vm_ops;
256 static LIST_HEAD(shmem_swaplist);
257 static DEFINE_MUTEX(shmem_swaplist_mutex);
260 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
261 * produces a novel ino for the newly allocated inode.
263 * It may also be called when making a hard link to permit the space needed by
264 * each dentry. However, in that case, no new inode number is needed since that
265 * internally draws from another pool of inode numbers (currently global
266 * get_next_ino()). This case is indicated by passing NULL as inop.
268 #define SHMEM_INO_BATCH 1024
269 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
271 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
274 if (!(sb->s_flags & SB_KERNMOUNT)) {
275 raw_spin_lock(&sbinfo->stat_lock);
276 if (sbinfo->max_inodes) {
277 if (!sbinfo->free_inodes) {
278 raw_spin_unlock(&sbinfo->stat_lock);
281 sbinfo->free_inodes--;
284 ino = sbinfo->next_ino++;
285 if (unlikely(is_zero_ino(ino)))
286 ino = sbinfo->next_ino++;
287 if (unlikely(!sbinfo->full_inums &&
290 * Emulate get_next_ino uint wraparound for
293 if (IS_ENABLED(CONFIG_64BIT))
294 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
295 __func__, MINOR(sb->s_dev));
296 sbinfo->next_ino = 1;
297 ino = sbinfo->next_ino++;
301 raw_spin_unlock(&sbinfo->stat_lock);
304 * __shmem_file_setup, one of our callers, is lock-free: it
305 * doesn't hold stat_lock in shmem_reserve_inode since
306 * max_inodes is always 0, and is called from potentially
307 * unknown contexts. As such, use a per-cpu batched allocator
308 * which doesn't require the per-sb stat_lock unless we are at
309 * the batch boundary.
311 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
312 * shmem mounts are not exposed to userspace, so we don't need
313 * to worry about things like glibc compatibility.
317 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
319 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
320 raw_spin_lock(&sbinfo->stat_lock);
321 ino = sbinfo->next_ino;
322 sbinfo->next_ino += SHMEM_INO_BATCH;
323 raw_spin_unlock(&sbinfo->stat_lock);
324 if (unlikely(is_zero_ino(ino)))
335 static void shmem_free_inode(struct super_block *sb)
337 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
338 if (sbinfo->max_inodes) {
339 raw_spin_lock(&sbinfo->stat_lock);
340 sbinfo->free_inodes++;
341 raw_spin_unlock(&sbinfo->stat_lock);
346 * shmem_recalc_inode - recalculate the block usage of an inode
347 * @inode: inode to recalc
349 * We have to calculate the free blocks since the mm can drop
350 * undirtied hole pages behind our back.
352 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
353 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
355 * It has to be called with the spinlock held.
357 static void shmem_recalc_inode(struct inode *inode)
359 struct shmem_inode_info *info = SHMEM_I(inode);
362 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
364 info->alloced -= freed;
365 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
366 shmem_inode_unacct_blocks(inode, freed);
370 bool shmem_charge(struct inode *inode, long pages)
372 struct shmem_inode_info *info = SHMEM_I(inode);
375 if (!shmem_inode_acct_block(inode, pages))
378 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
379 inode->i_mapping->nrpages += pages;
381 spin_lock_irqsave(&info->lock, flags);
382 info->alloced += pages;
383 inode->i_blocks += pages * BLOCKS_PER_PAGE;
384 shmem_recalc_inode(inode);
385 spin_unlock_irqrestore(&info->lock, flags);
390 void shmem_uncharge(struct inode *inode, long pages)
392 struct shmem_inode_info *info = SHMEM_I(inode);
395 /* nrpages adjustment done by __delete_from_page_cache() or caller */
397 spin_lock_irqsave(&info->lock, flags);
398 info->alloced -= pages;
399 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
400 shmem_recalc_inode(inode);
401 spin_unlock_irqrestore(&info->lock, flags);
403 shmem_inode_unacct_blocks(inode, pages);
407 * Replace item expected in xarray by a new item, while holding xa_lock.
409 static int shmem_replace_entry(struct address_space *mapping,
410 pgoff_t index, void *expected, void *replacement)
412 XA_STATE(xas, &mapping->i_pages, index);
415 VM_BUG_ON(!expected);
416 VM_BUG_ON(!replacement);
417 item = xas_load(&xas);
418 if (item != expected)
420 xas_store(&xas, replacement);
425 * Sometimes, before we decide whether to proceed or to fail, we must check
426 * that an entry was not already brought back from swap by a racing thread.
428 * Checking page is not enough: by the time a SwapCache page is locked, it
429 * might be reused, and again be SwapCache, using the same swap as before.
431 static bool shmem_confirm_swap(struct address_space *mapping,
432 pgoff_t index, swp_entry_t swap)
434 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
438 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
441 * disables huge pages for the mount;
443 * enables huge pages for the mount;
444 * SHMEM_HUGE_WITHIN_SIZE:
445 * only allocate huge pages if the page will be fully within i_size,
446 * also respect fadvise()/madvise() hints;
448 * only allocate huge pages if requested with fadvise()/madvise();
451 #define SHMEM_HUGE_NEVER 0
452 #define SHMEM_HUGE_ALWAYS 1
453 #define SHMEM_HUGE_WITHIN_SIZE 2
454 #define SHMEM_HUGE_ADVISE 3
458 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
461 * disables huge on shm_mnt and all mounts, for emergency use;
463 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
466 #define SHMEM_HUGE_DENY (-1)
467 #define SHMEM_HUGE_FORCE (-2)
469 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
470 /* ifdef here to avoid bloating shmem.o when not necessary */
472 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
474 bool shmem_is_huge(struct vm_area_struct *vma,
475 struct inode *inode, pgoff_t index)
479 if (!S_ISREG(inode->i_mode))
481 if (shmem_huge == SHMEM_HUGE_DENY)
483 if (vma && ((vma->vm_flags & VM_NOHUGEPAGE) ||
484 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
486 if (shmem_huge == SHMEM_HUGE_FORCE)
489 switch (SHMEM_SB(inode->i_sb)->huge) {
490 case SHMEM_HUGE_ALWAYS:
492 case SHMEM_HUGE_WITHIN_SIZE:
493 index = round_up(index + 1, HPAGE_PMD_NR);
494 i_size = round_up(i_size_read(inode), PAGE_SIZE);
495 if (i_size >> PAGE_SHIFT >= index)
498 case SHMEM_HUGE_ADVISE:
499 if (vma && (vma->vm_flags & VM_HUGEPAGE))
507 #if defined(CONFIG_SYSFS)
508 static int shmem_parse_huge(const char *str)
510 if (!strcmp(str, "never"))
511 return SHMEM_HUGE_NEVER;
512 if (!strcmp(str, "always"))
513 return SHMEM_HUGE_ALWAYS;
514 if (!strcmp(str, "within_size"))
515 return SHMEM_HUGE_WITHIN_SIZE;
516 if (!strcmp(str, "advise"))
517 return SHMEM_HUGE_ADVISE;
518 if (!strcmp(str, "deny"))
519 return SHMEM_HUGE_DENY;
520 if (!strcmp(str, "force"))
521 return SHMEM_HUGE_FORCE;
526 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
527 static const char *shmem_format_huge(int huge)
530 case SHMEM_HUGE_NEVER:
532 case SHMEM_HUGE_ALWAYS:
534 case SHMEM_HUGE_WITHIN_SIZE:
535 return "within_size";
536 case SHMEM_HUGE_ADVISE:
538 case SHMEM_HUGE_DENY:
540 case SHMEM_HUGE_FORCE:
549 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
550 struct shrink_control *sc, unsigned long nr_to_split)
552 LIST_HEAD(list), *pos, *next;
553 LIST_HEAD(to_remove);
555 struct shmem_inode_info *info;
557 unsigned long batch = sc ? sc->nr_to_scan : 128;
560 if (list_empty(&sbinfo->shrinklist))
563 spin_lock(&sbinfo->shrinklist_lock);
564 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
565 info = list_entry(pos, struct shmem_inode_info, shrinklist);
568 inode = igrab(&info->vfs_inode);
570 /* inode is about to be evicted */
572 list_del_init(&info->shrinklist);
576 /* Check if there's anything to gain */
577 if (round_up(inode->i_size, PAGE_SIZE) ==
578 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
579 list_move(&info->shrinklist, &to_remove);
583 list_move(&info->shrinklist, &list);
585 sbinfo->shrinklist_len--;
589 spin_unlock(&sbinfo->shrinklist_lock);
591 list_for_each_safe(pos, next, &to_remove) {
592 info = list_entry(pos, struct shmem_inode_info, shrinklist);
593 inode = &info->vfs_inode;
594 list_del_init(&info->shrinklist);
598 list_for_each_safe(pos, next, &list) {
602 info = list_entry(pos, struct shmem_inode_info, shrinklist);
603 inode = &info->vfs_inode;
605 if (nr_to_split && split >= nr_to_split)
608 index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
609 folio = filemap_get_folio(inode->i_mapping, index);
613 /* No huge page at the end of the file: nothing to split */
614 if (!folio_test_large(folio)) {
620 * Move the inode on the list back to shrinklist if we failed
621 * to lock the page at this time.
623 * Waiting for the lock may lead to deadlock in the
626 if (!folio_trylock(folio)) {
631 ret = split_huge_page(&folio->page);
635 /* If split failed move the inode on the list back to shrinklist */
641 list_del_init(&info->shrinklist);
645 * Make sure the inode is either on the global list or deleted
646 * from any local list before iput() since it could be deleted
647 * in another thread once we put the inode (then the local list
650 spin_lock(&sbinfo->shrinklist_lock);
651 list_move(&info->shrinklist, &sbinfo->shrinklist);
652 sbinfo->shrinklist_len++;
653 spin_unlock(&sbinfo->shrinklist_lock);
661 static long shmem_unused_huge_scan(struct super_block *sb,
662 struct shrink_control *sc)
664 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
666 if (!READ_ONCE(sbinfo->shrinklist_len))
669 return shmem_unused_huge_shrink(sbinfo, sc, 0);
672 static long shmem_unused_huge_count(struct super_block *sb,
673 struct shrink_control *sc)
675 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
676 return READ_ONCE(sbinfo->shrinklist_len);
678 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
680 #define shmem_huge SHMEM_HUGE_DENY
682 bool shmem_is_huge(struct vm_area_struct *vma,
683 struct inode *inode, pgoff_t index)
688 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
689 struct shrink_control *sc, unsigned long nr_to_split)
693 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
696 * Like add_to_page_cache_locked, but error if expected item has gone.
698 static int shmem_add_to_page_cache(struct folio *folio,
699 struct address_space *mapping,
700 pgoff_t index, void *expected, gfp_t gfp,
701 struct mm_struct *charge_mm)
703 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
704 long nr = folio_nr_pages(folio);
707 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
708 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
709 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
710 VM_BUG_ON(expected && folio_test_large(folio));
712 folio_ref_add(folio, nr);
713 folio->mapping = mapping;
714 folio->index = index;
716 if (!folio_test_swapcache(folio)) {
717 error = mem_cgroup_charge(folio, charge_mm, gfp);
719 if (folio_test_pmd_mappable(folio)) {
720 count_vm_event(THP_FILE_FALLBACK);
721 count_vm_event(THP_FILE_FALLBACK_CHARGE);
726 folio_throttle_swaprate(folio, gfp);
730 if (expected != xas_find_conflict(&xas)) {
731 xas_set_err(&xas, -EEXIST);
734 if (expected && xas_find_conflict(&xas)) {
735 xas_set_err(&xas, -EEXIST);
738 xas_store(&xas, folio);
741 if (folio_test_pmd_mappable(folio)) {
742 count_vm_event(THP_FILE_ALLOC);
743 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
745 mapping->nrpages += nr;
746 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
747 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
749 xas_unlock_irq(&xas);
750 } while (xas_nomem(&xas, gfp));
752 if (xas_error(&xas)) {
753 error = xas_error(&xas);
759 folio->mapping = NULL;
760 folio_ref_sub(folio, nr);
765 * Like delete_from_page_cache, but substitutes swap for page.
767 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
769 struct address_space *mapping = page->mapping;
772 VM_BUG_ON_PAGE(PageCompound(page), page);
774 xa_lock_irq(&mapping->i_pages);
775 error = shmem_replace_entry(mapping, page->index, page, radswap);
776 page->mapping = NULL;
778 __dec_lruvec_page_state(page, NR_FILE_PAGES);
779 __dec_lruvec_page_state(page, NR_SHMEM);
780 xa_unlock_irq(&mapping->i_pages);
786 * Remove swap entry from page cache, free the swap and its page cache.
788 static int shmem_free_swap(struct address_space *mapping,
789 pgoff_t index, void *radswap)
793 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
796 free_swap_and_cache(radix_to_swp_entry(radswap));
801 * Determine (in bytes) how many of the shmem object's pages mapped by the
802 * given offsets are swapped out.
804 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
805 * as long as the inode doesn't go away and racy results are not a problem.
807 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
808 pgoff_t start, pgoff_t end)
810 XA_STATE(xas, &mapping->i_pages, start);
812 unsigned long swapped = 0;
815 xas_for_each(&xas, page, end - 1) {
816 if (xas_retry(&xas, page))
818 if (xa_is_value(page))
821 if (need_resched()) {
829 return swapped << PAGE_SHIFT;
833 * Determine (in bytes) how many of the shmem object's pages mapped by the
834 * given vma is swapped out.
836 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
837 * as long as the inode doesn't go away and racy results are not a problem.
839 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
841 struct inode *inode = file_inode(vma->vm_file);
842 struct shmem_inode_info *info = SHMEM_I(inode);
843 struct address_space *mapping = inode->i_mapping;
844 unsigned long swapped;
846 /* Be careful as we don't hold info->lock */
847 swapped = READ_ONCE(info->swapped);
850 * The easier cases are when the shmem object has nothing in swap, or
851 * the vma maps it whole. Then we can simply use the stats that we
857 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
858 return swapped << PAGE_SHIFT;
860 /* Here comes the more involved part */
861 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
862 vma->vm_pgoff + vma_pages(vma));
866 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
868 void shmem_unlock_mapping(struct address_space *mapping)
875 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
877 while (!mapping_unevictable(mapping)) {
878 if (!pagevec_lookup(&pvec, mapping, &index))
880 check_move_unevictable_pages(&pvec);
881 pagevec_release(&pvec);
886 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
892 * At first avoid shmem_getpage(,,,SGP_READ): that fails
893 * beyond i_size, and reports fallocated pages as holes.
895 folio = __filemap_get_folio(inode->i_mapping, index,
896 FGP_ENTRY | FGP_LOCK, 0);
897 if (!xa_is_value(folio))
900 * But read a page back from swap if any of it is within i_size
901 * (although in some cases this is just a waste of time).
904 shmem_getpage(inode, index, &page, SGP_READ);
905 return page ? page_folio(page) : NULL;
909 * Remove range of pages and swap entries from page cache, and free them.
910 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
912 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
915 struct address_space *mapping = inode->i_mapping;
916 struct shmem_inode_info *info = SHMEM_I(inode);
917 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
918 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
919 struct folio_batch fbatch;
920 pgoff_t indices[PAGEVEC_SIZE];
923 long nr_swaps_freed = 0;
928 end = -1; /* unsigned, so actually very big */
930 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
931 info->fallocend = start;
933 folio_batch_init(&fbatch);
935 while (index < end && find_lock_entries(mapping, index, end - 1,
937 for (i = 0; i < folio_batch_count(&fbatch); i++) {
938 folio = fbatch.folios[i];
942 if (xa_is_value(folio)) {
945 nr_swaps_freed += !shmem_free_swap(mapping,
949 index += folio_nr_pages(folio) - 1;
951 if (!unfalloc || !folio_test_uptodate(folio))
952 truncate_inode_folio(mapping, folio);
955 folio_batch_remove_exceptionals(&fbatch);
956 folio_batch_release(&fbatch);
961 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
962 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
964 same_folio = lend < folio_pos(folio) + folio_size(folio);
965 folio_mark_dirty(folio);
966 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
967 start = folio->index + folio_nr_pages(folio);
977 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
979 folio_mark_dirty(folio);
980 if (!truncate_inode_partial_folio(folio, lstart, lend))
987 while (index < end) {
990 if (!find_get_entries(mapping, index, end - 1, &fbatch,
992 /* If all gone or hole-punch or unfalloc, we're done */
993 if (index == start || end != -1)
995 /* But if truncating, restart to make sure all gone */
999 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1000 folio = fbatch.folios[i];
1003 if (xa_is_value(folio)) {
1006 if (shmem_free_swap(mapping, index, folio)) {
1007 /* Swap was replaced by page: retry */
1017 if (!unfalloc || !folio_test_uptodate(folio)) {
1018 if (folio_mapping(folio) != mapping) {
1019 /* Page was replaced by swap: retry */
1020 folio_unlock(folio);
1024 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1026 truncate_inode_folio(mapping, folio);
1028 index = folio->index + folio_nr_pages(folio) - 1;
1029 folio_unlock(folio);
1031 folio_batch_remove_exceptionals(&fbatch);
1032 folio_batch_release(&fbatch);
1036 spin_lock_irq(&info->lock);
1037 info->swapped -= nr_swaps_freed;
1038 shmem_recalc_inode(inode);
1039 spin_unlock_irq(&info->lock);
1042 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1044 shmem_undo_range(inode, lstart, lend, false);
1045 inode->i_ctime = inode->i_mtime = current_time(inode);
1047 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1049 static int shmem_getattr(struct user_namespace *mnt_userns,
1050 const struct path *path, struct kstat *stat,
1051 u32 request_mask, unsigned int query_flags)
1053 struct inode *inode = path->dentry->d_inode;
1054 struct shmem_inode_info *info = SHMEM_I(inode);
1056 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1057 spin_lock_irq(&info->lock);
1058 shmem_recalc_inode(inode);
1059 spin_unlock_irq(&info->lock);
1061 generic_fillattr(&init_user_ns, inode, stat);
1063 if (shmem_is_huge(NULL, inode, 0))
1064 stat->blksize = HPAGE_PMD_SIZE;
1066 if (request_mask & STATX_BTIME) {
1067 stat->result_mask |= STATX_BTIME;
1068 stat->btime.tv_sec = info->i_crtime.tv_sec;
1069 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1075 static int shmem_setattr(struct user_namespace *mnt_userns,
1076 struct dentry *dentry, struct iattr *attr)
1078 struct inode *inode = d_inode(dentry);
1079 struct shmem_inode_info *info = SHMEM_I(inode);
1082 error = setattr_prepare(&init_user_ns, dentry, attr);
1086 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1087 loff_t oldsize = inode->i_size;
1088 loff_t newsize = attr->ia_size;
1090 /* protected by i_rwsem */
1091 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1092 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1095 if (newsize != oldsize) {
1096 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1100 i_size_write(inode, newsize);
1101 inode->i_ctime = inode->i_mtime = current_time(inode);
1103 if (newsize <= oldsize) {
1104 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1105 if (oldsize > holebegin)
1106 unmap_mapping_range(inode->i_mapping,
1109 shmem_truncate_range(inode,
1110 newsize, (loff_t)-1);
1111 /* unmap again to remove racily COWed private pages */
1112 if (oldsize > holebegin)
1113 unmap_mapping_range(inode->i_mapping,
1118 setattr_copy(&init_user_ns, inode, attr);
1119 if (attr->ia_valid & ATTR_MODE)
1120 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1124 static void shmem_evict_inode(struct inode *inode)
1126 struct shmem_inode_info *info = SHMEM_I(inode);
1127 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1129 if (shmem_mapping(inode->i_mapping)) {
1130 shmem_unacct_size(info->flags, inode->i_size);
1132 mapping_set_exiting(inode->i_mapping);
1133 shmem_truncate_range(inode, 0, (loff_t)-1);
1134 if (!list_empty(&info->shrinklist)) {
1135 spin_lock(&sbinfo->shrinklist_lock);
1136 if (!list_empty(&info->shrinklist)) {
1137 list_del_init(&info->shrinklist);
1138 sbinfo->shrinklist_len--;
1140 spin_unlock(&sbinfo->shrinklist_lock);
1142 while (!list_empty(&info->swaplist)) {
1143 /* Wait while shmem_unuse() is scanning this inode... */
1144 wait_var_event(&info->stop_eviction,
1145 !atomic_read(&info->stop_eviction));
1146 mutex_lock(&shmem_swaplist_mutex);
1147 /* ...but beware of the race if we peeked too early */
1148 if (!atomic_read(&info->stop_eviction))
1149 list_del_init(&info->swaplist);
1150 mutex_unlock(&shmem_swaplist_mutex);
1154 simple_xattrs_free(&info->xattrs);
1155 WARN_ON(inode->i_blocks);
1156 shmem_free_inode(inode->i_sb);
1160 static int shmem_find_swap_entries(struct address_space *mapping,
1161 pgoff_t start, struct folio_batch *fbatch,
1162 pgoff_t *indices, unsigned int type)
1164 XA_STATE(xas, &mapping->i_pages, start);
1165 struct folio *folio;
1169 xas_for_each(&xas, folio, ULONG_MAX) {
1170 if (xas_retry(&xas, folio))
1173 if (!xa_is_value(folio))
1176 entry = radix_to_swp_entry(folio);
1177 if (swp_type(entry) != type)
1180 indices[folio_batch_count(fbatch)] = xas.xa_index;
1181 if (!folio_batch_add(fbatch, folio))
1184 if (need_resched()) {
1191 return xas.xa_index;
1195 * Move the swapped pages for an inode to page cache. Returns the count
1196 * of pages swapped in, or the error in case of failure.
1198 static int shmem_unuse_swap_entries(struct inode *inode,
1199 struct folio_batch *fbatch, pgoff_t *indices)
1204 struct address_space *mapping = inode->i_mapping;
1206 for (i = 0; i < folio_batch_count(fbatch); i++) {
1207 struct folio *folio = fbatch->folios[i];
1209 if (!xa_is_value(folio))
1211 error = shmem_swapin_folio(inode, indices[i],
1213 mapping_gfp_mask(mapping),
1216 folio_unlock(folio);
1220 if (error == -ENOMEM)
1224 return error ? error : ret;
1228 * If swap found in inode, free it and move page from swapcache to filecache.
1230 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1232 struct address_space *mapping = inode->i_mapping;
1234 struct folio_batch fbatch;
1235 pgoff_t indices[PAGEVEC_SIZE];
1239 folio_batch_init(&fbatch);
1240 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1241 if (folio_batch_count(&fbatch) == 0) {
1246 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1250 start = indices[folio_batch_count(&fbatch) - 1];
1257 * Read all the shared memory data that resides in the swap
1258 * device 'type' back into memory, so the swap device can be
1261 int shmem_unuse(unsigned int type)
1263 struct shmem_inode_info *info, *next;
1266 if (list_empty(&shmem_swaplist))
1269 mutex_lock(&shmem_swaplist_mutex);
1270 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1271 if (!info->swapped) {
1272 list_del_init(&info->swaplist);
1276 * Drop the swaplist mutex while searching the inode for swap;
1277 * but before doing so, make sure shmem_evict_inode() will not
1278 * remove placeholder inode from swaplist, nor let it be freed
1279 * (igrab() would protect from unlink, but not from unmount).
1281 atomic_inc(&info->stop_eviction);
1282 mutex_unlock(&shmem_swaplist_mutex);
1284 error = shmem_unuse_inode(&info->vfs_inode, type);
1287 mutex_lock(&shmem_swaplist_mutex);
1288 next = list_next_entry(info, swaplist);
1290 list_del_init(&info->swaplist);
1291 if (atomic_dec_and_test(&info->stop_eviction))
1292 wake_up_var(&info->stop_eviction);
1296 mutex_unlock(&shmem_swaplist_mutex);
1302 * Move the page from the page cache to the swap cache.
1304 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1306 struct folio *folio = page_folio(page);
1307 struct shmem_inode_info *info;
1308 struct address_space *mapping;
1309 struct inode *inode;
1314 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1315 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1316 * and its shmem_writeback() needs them to be split when swapping.
1318 if (PageTransCompound(page)) {
1319 /* Ensure the subpages are still dirty */
1321 if (split_huge_page(page) < 0)
1323 ClearPageDirty(page);
1326 BUG_ON(!PageLocked(page));
1327 mapping = page->mapping;
1328 index = page->index;
1329 inode = mapping->host;
1330 info = SHMEM_I(inode);
1331 if (info->flags & VM_LOCKED)
1333 if (!total_swap_pages)
1337 * Our capabilities prevent regular writeback or sync from ever calling
1338 * shmem_writepage; but a stacking filesystem might use ->writepage of
1339 * its underlying filesystem, in which case tmpfs should write out to
1340 * swap only in response to memory pressure, and not for the writeback
1343 if (!wbc->for_reclaim) {
1344 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1349 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1350 * value into swapfile.c, the only way we can correctly account for a
1351 * fallocated page arriving here is now to initialize it and write it.
1353 * That's okay for a page already fallocated earlier, but if we have
1354 * not yet completed the fallocation, then (a) we want to keep track
1355 * of this page in case we have to undo it, and (b) it may not be a
1356 * good idea to continue anyway, once we're pushing into swap. So
1357 * reactivate the page, and let shmem_fallocate() quit when too many.
1359 if (!PageUptodate(page)) {
1360 if (inode->i_private) {
1361 struct shmem_falloc *shmem_falloc;
1362 spin_lock(&inode->i_lock);
1363 shmem_falloc = inode->i_private;
1365 !shmem_falloc->waitq &&
1366 index >= shmem_falloc->start &&
1367 index < shmem_falloc->next)
1368 shmem_falloc->nr_unswapped++;
1370 shmem_falloc = NULL;
1371 spin_unlock(&inode->i_lock);
1375 clear_highpage(page);
1376 flush_dcache_page(page);
1377 SetPageUptodate(page);
1380 swap = folio_alloc_swap(folio);
1385 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1386 * if it's not already there. Do it now before the page is
1387 * moved to swap cache, when its pagelock no longer protects
1388 * the inode from eviction. But don't unlock the mutex until
1389 * we've incremented swapped, because shmem_unuse_inode() will
1390 * prune a !swapped inode from the swaplist under this mutex.
1392 mutex_lock(&shmem_swaplist_mutex);
1393 if (list_empty(&info->swaplist))
1394 list_add(&info->swaplist, &shmem_swaplist);
1396 if (add_to_swap_cache(page, swap,
1397 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1399 spin_lock_irq(&info->lock);
1400 shmem_recalc_inode(inode);
1402 spin_unlock_irq(&info->lock);
1404 swap_shmem_alloc(swap);
1405 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1407 mutex_unlock(&shmem_swaplist_mutex);
1408 BUG_ON(page_mapped(page));
1409 swap_writepage(page, wbc);
1413 mutex_unlock(&shmem_swaplist_mutex);
1414 put_swap_page(page, swap);
1416 set_page_dirty(page);
1417 if (wbc->for_reclaim)
1418 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1423 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1424 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1428 if (!mpol || mpol->mode == MPOL_DEFAULT)
1429 return; /* show nothing */
1431 mpol_to_str(buffer, sizeof(buffer), mpol);
1433 seq_printf(seq, ",mpol=%s", buffer);
1436 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1438 struct mempolicy *mpol = NULL;
1440 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1441 mpol = sbinfo->mpol;
1443 raw_spin_unlock(&sbinfo->stat_lock);
1447 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1448 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1451 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1455 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1457 #define vm_policy vm_private_data
1460 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1461 struct shmem_inode_info *info, pgoff_t index)
1463 /* Create a pseudo vma that just contains the policy */
1464 vma_init(vma, NULL);
1465 /* Bias interleave by inode number to distribute better across nodes */
1466 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1467 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1470 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1472 /* Drop reference taken by mpol_shared_policy_lookup() */
1473 mpol_cond_put(vma->vm_policy);
1476 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1477 struct shmem_inode_info *info, pgoff_t index)
1479 struct vm_area_struct pvma;
1481 struct vm_fault vmf = {
1485 shmem_pseudo_vma_init(&pvma, info, index);
1486 page = swap_cluster_readahead(swap, gfp, &vmf);
1487 shmem_pseudo_vma_destroy(&pvma);
1493 * Make sure huge_gfp is always more limited than limit_gfp.
1494 * Some of the flags set permissions, while others set limitations.
1496 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1498 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1499 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1500 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1501 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1503 /* Allow allocations only from the originally specified zones. */
1504 result |= zoneflags;
1507 * Minimize the result gfp by taking the union with the deny flags,
1508 * and the intersection of the allow flags.
1510 result |= (limit_gfp & denyflags);
1511 result |= (huge_gfp & limit_gfp) & allowflags;
1516 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1517 struct shmem_inode_info *info, pgoff_t index)
1519 struct vm_area_struct pvma;
1520 struct address_space *mapping = info->vfs_inode.i_mapping;
1522 struct folio *folio;
1524 hindex = round_down(index, HPAGE_PMD_NR);
1525 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1529 shmem_pseudo_vma_init(&pvma, info, hindex);
1530 folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1531 shmem_pseudo_vma_destroy(&pvma);
1533 count_vm_event(THP_FILE_FALLBACK);
1537 static struct folio *shmem_alloc_folio(gfp_t gfp,
1538 struct shmem_inode_info *info, pgoff_t index)
1540 struct vm_area_struct pvma;
1541 struct folio *folio;
1543 shmem_pseudo_vma_init(&pvma, info, index);
1544 folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1545 shmem_pseudo_vma_destroy(&pvma);
1550 static struct page *shmem_alloc_page(gfp_t gfp,
1551 struct shmem_inode_info *info, pgoff_t index)
1553 return &shmem_alloc_folio(gfp, info, index)->page;
1556 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1557 pgoff_t index, bool huge)
1559 struct shmem_inode_info *info = SHMEM_I(inode);
1560 struct folio *folio;
1564 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1566 nr = huge ? HPAGE_PMD_NR : 1;
1568 if (!shmem_inode_acct_block(inode, nr))
1572 folio = shmem_alloc_hugefolio(gfp, info, index);
1574 folio = shmem_alloc_folio(gfp, info, index);
1576 __folio_set_locked(folio);
1577 __folio_set_swapbacked(folio);
1582 shmem_inode_unacct_blocks(inode, nr);
1584 return ERR_PTR(err);
1588 * When a page is moved from swapcache to shmem filecache (either by the
1589 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1590 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1591 * ignorance of the mapping it belongs to. If that mapping has special
1592 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1593 * we may need to copy to a suitable page before moving to filecache.
1595 * In a future release, this may well be extended to respect cpuset and
1596 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1597 * but for now it is a simple matter of zone.
1599 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1601 return folio_zonenum(folio) > gfp_zone(gfp);
1604 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1605 struct shmem_inode_info *info, pgoff_t index)
1607 struct page *oldpage, *newpage;
1608 struct folio *old, *new;
1609 struct address_space *swap_mapping;
1615 entry.val = page_private(oldpage);
1616 swap_index = swp_offset(entry);
1617 swap_mapping = page_mapping(oldpage);
1620 * We have arrived here because our zones are constrained, so don't
1621 * limit chance of success by further cpuset and node constraints.
1623 gfp &= ~GFP_CONSTRAINT_MASK;
1624 newpage = shmem_alloc_page(gfp, info, index);
1629 copy_highpage(newpage, oldpage);
1630 flush_dcache_page(newpage);
1632 __SetPageLocked(newpage);
1633 __SetPageSwapBacked(newpage);
1634 SetPageUptodate(newpage);
1635 set_page_private(newpage, entry.val);
1636 SetPageSwapCache(newpage);
1639 * Our caller will very soon move newpage out of swapcache, but it's
1640 * a nice clean interface for us to replace oldpage by newpage there.
1642 xa_lock_irq(&swap_mapping->i_pages);
1643 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1645 old = page_folio(oldpage);
1646 new = page_folio(newpage);
1647 mem_cgroup_migrate(old, new);
1648 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1649 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1651 xa_unlock_irq(&swap_mapping->i_pages);
1653 if (unlikely(error)) {
1655 * Is this possible? I think not, now that our callers check
1656 * both PageSwapCache and page_private after getting page lock;
1657 * but be defensive. Reverse old to newpage for clear and free.
1661 lru_cache_add(newpage);
1665 ClearPageSwapCache(oldpage);
1666 set_page_private(oldpage, 0);
1668 unlock_page(oldpage);
1675 * Swap in the page pointed to by *pagep.
1676 * Caller has to make sure that *pagep contains a valid swapped page.
1677 * Returns 0 and the page in pagep if success. On failure, returns the
1678 * error code and NULL in *pagep.
1680 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1681 struct folio **foliop, enum sgp_type sgp,
1682 gfp_t gfp, struct vm_area_struct *vma,
1683 vm_fault_t *fault_type)
1685 struct address_space *mapping = inode->i_mapping;
1686 struct shmem_inode_info *info = SHMEM_I(inode);
1687 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1689 struct folio *folio = NULL;
1693 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1694 swap = radix_to_swp_entry(*foliop);
1697 /* Look it up and read it in.. */
1698 page = lookup_swap_cache(swap, NULL, 0);
1700 /* Or update major stats only when swapin succeeds?? */
1702 *fault_type |= VM_FAULT_MAJOR;
1703 count_vm_event(PGMAJFAULT);
1704 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1706 /* Here we actually start the io */
1707 page = shmem_swapin(swap, gfp, info, index);
1713 folio = page_folio(page);
1715 /* We have to do this with page locked to prevent races */
1717 if (!folio_test_swapcache(folio) ||
1718 folio_swap_entry(folio).val != swap.val ||
1719 !shmem_confirm_swap(mapping, index, swap)) {
1723 if (!folio_test_uptodate(folio)) {
1727 folio_wait_writeback(folio);
1730 * Some architectures may have to restore extra metadata to the
1731 * folio after reading from swap.
1733 arch_swap_restore(swap, folio);
1735 if (shmem_should_replace_folio(folio, gfp)) {
1736 error = shmem_replace_page(&page, gfp, info, index);
1741 error = shmem_add_to_page_cache(folio, mapping, index,
1742 swp_to_radix_entry(swap), gfp,
1747 spin_lock_irq(&info->lock);
1749 shmem_recalc_inode(inode);
1750 spin_unlock_irq(&info->lock);
1752 if (sgp == SGP_WRITE)
1753 folio_mark_accessed(folio);
1755 delete_from_swap_cache(&folio->page);
1756 folio_mark_dirty(folio);
1762 if (!shmem_confirm_swap(mapping, index, swap))
1766 folio_unlock(folio);
1774 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1776 * If we allocate a new one we do not mark it dirty. That's up to the
1777 * vm. If we swap it in we mark it dirty since we also free the swap
1778 * entry since a page cannot live in both the swap and page cache.
1780 * vma, vmf, and fault_type are only supplied by shmem_fault:
1781 * otherwise they are NULL.
1783 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1784 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1785 struct vm_area_struct *vma, struct vm_fault *vmf,
1786 vm_fault_t *fault_type)
1788 struct address_space *mapping = inode->i_mapping;
1789 struct shmem_inode_info *info = SHMEM_I(inode);
1790 struct shmem_sb_info *sbinfo;
1791 struct mm_struct *charge_mm;
1792 struct folio *folio;
1793 pgoff_t hindex = index;
1799 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1802 if (sgp <= SGP_CACHE &&
1803 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1807 sbinfo = SHMEM_SB(inode->i_sb);
1808 charge_mm = vma ? vma->vm_mm : NULL;
1810 folio = __filemap_get_folio(mapping, index, FGP_ENTRY | FGP_LOCK, 0);
1811 if (folio && vma && userfaultfd_minor(vma)) {
1812 if (!xa_is_value(folio)) {
1813 folio_unlock(folio);
1816 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1820 if (xa_is_value(folio)) {
1821 error = shmem_swapin_folio(inode, index, &folio,
1822 sgp, gfp, vma, fault_type);
1823 if (error == -EEXIST)
1826 *pagep = &folio->page;
1831 hindex = folio->index;
1832 if (sgp == SGP_WRITE)
1833 folio_mark_accessed(folio);
1834 if (folio_test_uptodate(folio))
1836 /* fallocated page */
1837 if (sgp != SGP_READ)
1839 folio_unlock(folio);
1844 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1845 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1848 if (sgp == SGP_READ)
1850 if (sgp == SGP_NOALLOC)
1854 * Fast cache lookup and swap lookup did not find it: allocate.
1857 if (vma && userfaultfd_missing(vma)) {
1858 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1862 if (!shmem_is_huge(vma, inode, index))
1865 huge_gfp = vma_thp_gfp_mask(vma);
1866 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1867 folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
1868 if (IS_ERR(folio)) {
1870 folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
1872 if (IS_ERR(folio)) {
1875 error = PTR_ERR(folio);
1877 if (error != -ENOSPC)
1880 * Try to reclaim some space by splitting a huge page
1881 * beyond i_size on the filesystem.
1886 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1887 if (ret == SHRINK_STOP)
1895 hindex = round_down(index, folio_nr_pages(folio));
1897 if (sgp == SGP_WRITE)
1898 __folio_set_referenced(folio);
1900 error = shmem_add_to_page_cache(folio, mapping, hindex,
1901 NULL, gfp & GFP_RECLAIM_MASK,
1905 folio_add_lru(folio);
1907 spin_lock_irq(&info->lock);
1908 info->alloced += folio_nr_pages(folio);
1909 inode->i_blocks += BLOCKS_PER_PAGE << folio_order(folio);
1910 shmem_recalc_inode(inode);
1911 spin_unlock_irq(&info->lock);
1914 if (folio_test_pmd_mappable(folio) &&
1915 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1916 hindex + HPAGE_PMD_NR - 1) {
1918 * Part of the huge page is beyond i_size: subject
1919 * to shrink under memory pressure.
1921 spin_lock(&sbinfo->shrinklist_lock);
1923 * _careful to defend against unlocked access to
1924 * ->shrink_list in shmem_unused_huge_shrink()
1926 if (list_empty_careful(&info->shrinklist)) {
1927 list_add_tail(&info->shrinklist,
1928 &sbinfo->shrinklist);
1929 sbinfo->shrinklist_len++;
1931 spin_unlock(&sbinfo->shrinklist_lock);
1935 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1937 if (sgp == SGP_FALLOC)
1941 * Let SGP_WRITE caller clear ends if write does not fill page;
1942 * but SGP_FALLOC on a page fallocated earlier must initialize
1943 * it now, lest undo on failure cancel our earlier guarantee.
1945 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
1946 long i, n = folio_nr_pages(folio);
1948 for (i = 0; i < n; i++)
1949 clear_highpage(folio_page(folio, i));
1950 flush_dcache_folio(folio);
1951 folio_mark_uptodate(folio);
1954 /* Perhaps the file has been truncated since we checked */
1955 if (sgp <= SGP_CACHE &&
1956 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1958 folio_clear_dirty(folio);
1959 filemap_remove_folio(folio);
1960 spin_lock_irq(&info->lock);
1961 shmem_recalc_inode(inode);
1962 spin_unlock_irq(&info->lock);
1968 *pagep = folio_page(folio, index - hindex);
1975 shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
1977 if (folio_test_large(folio)) {
1978 folio_unlock(folio);
1984 folio_unlock(folio);
1987 if (error == -ENOSPC && !once++) {
1988 spin_lock_irq(&info->lock);
1989 shmem_recalc_inode(inode);
1990 spin_unlock_irq(&info->lock);
1993 if (error == -EEXIST)
1999 * This is like autoremove_wake_function, but it removes the wait queue
2000 * entry unconditionally - even if something else had already woken the
2003 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2005 int ret = default_wake_function(wait, mode, sync, key);
2006 list_del_init(&wait->entry);
2010 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2012 struct vm_area_struct *vma = vmf->vma;
2013 struct inode *inode = file_inode(vma->vm_file);
2014 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2016 vm_fault_t ret = VM_FAULT_LOCKED;
2019 * Trinity finds that probing a hole which tmpfs is punching can
2020 * prevent the hole-punch from ever completing: which in turn
2021 * locks writers out with its hold on i_rwsem. So refrain from
2022 * faulting pages into the hole while it's being punched. Although
2023 * shmem_undo_range() does remove the additions, it may be unable to
2024 * keep up, as each new page needs its own unmap_mapping_range() call,
2025 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2027 * It does not matter if we sometimes reach this check just before the
2028 * hole-punch begins, so that one fault then races with the punch:
2029 * we just need to make racing faults a rare case.
2031 * The implementation below would be much simpler if we just used a
2032 * standard mutex or completion: but we cannot take i_rwsem in fault,
2033 * and bloating every shmem inode for this unlikely case would be sad.
2035 if (unlikely(inode->i_private)) {
2036 struct shmem_falloc *shmem_falloc;
2038 spin_lock(&inode->i_lock);
2039 shmem_falloc = inode->i_private;
2041 shmem_falloc->waitq &&
2042 vmf->pgoff >= shmem_falloc->start &&
2043 vmf->pgoff < shmem_falloc->next) {
2045 wait_queue_head_t *shmem_falloc_waitq;
2046 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2048 ret = VM_FAULT_NOPAGE;
2049 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2051 ret = VM_FAULT_RETRY;
2053 shmem_falloc_waitq = shmem_falloc->waitq;
2054 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2055 TASK_UNINTERRUPTIBLE);
2056 spin_unlock(&inode->i_lock);
2060 * shmem_falloc_waitq points into the shmem_fallocate()
2061 * stack of the hole-punching task: shmem_falloc_waitq
2062 * is usually invalid by the time we reach here, but
2063 * finish_wait() does not dereference it in that case;
2064 * though i_lock needed lest racing with wake_up_all().
2066 spin_lock(&inode->i_lock);
2067 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2068 spin_unlock(&inode->i_lock);
2074 spin_unlock(&inode->i_lock);
2077 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE,
2078 gfp, vma, vmf, &ret);
2080 return vmf_error(err);
2084 unsigned long shmem_get_unmapped_area(struct file *file,
2085 unsigned long uaddr, unsigned long len,
2086 unsigned long pgoff, unsigned long flags)
2088 unsigned long (*get_area)(struct file *,
2089 unsigned long, unsigned long, unsigned long, unsigned long);
2091 unsigned long offset;
2092 unsigned long inflated_len;
2093 unsigned long inflated_addr;
2094 unsigned long inflated_offset;
2096 if (len > TASK_SIZE)
2099 get_area = current->mm->get_unmapped_area;
2100 addr = get_area(file, uaddr, len, pgoff, flags);
2102 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2104 if (IS_ERR_VALUE(addr))
2106 if (addr & ~PAGE_MASK)
2108 if (addr > TASK_SIZE - len)
2111 if (shmem_huge == SHMEM_HUGE_DENY)
2113 if (len < HPAGE_PMD_SIZE)
2115 if (flags & MAP_FIXED)
2118 * Our priority is to support MAP_SHARED mapped hugely;
2119 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2120 * But if caller specified an address hint and we allocated area there
2121 * successfully, respect that as before.
2126 if (shmem_huge != SHMEM_HUGE_FORCE) {
2127 struct super_block *sb;
2130 VM_BUG_ON(file->f_op != &shmem_file_operations);
2131 sb = file_inode(file)->i_sb;
2134 * Called directly from mm/mmap.c, or drivers/char/mem.c
2135 * for "/dev/zero", to create a shared anonymous object.
2137 if (IS_ERR(shm_mnt))
2139 sb = shm_mnt->mnt_sb;
2141 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2145 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2146 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2148 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2151 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2152 if (inflated_len > TASK_SIZE)
2154 if (inflated_len < len)
2157 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2158 if (IS_ERR_VALUE(inflated_addr))
2160 if (inflated_addr & ~PAGE_MASK)
2163 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2164 inflated_addr += offset - inflated_offset;
2165 if (inflated_offset > offset)
2166 inflated_addr += HPAGE_PMD_SIZE;
2168 if (inflated_addr > TASK_SIZE - len)
2170 return inflated_addr;
2174 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2176 struct inode *inode = file_inode(vma->vm_file);
2177 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2180 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2183 struct inode *inode = file_inode(vma->vm_file);
2186 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2187 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2191 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2193 struct inode *inode = file_inode(file);
2194 struct shmem_inode_info *info = SHMEM_I(inode);
2195 int retval = -ENOMEM;
2198 * What serializes the accesses to info->flags?
2199 * ipc_lock_object() when called from shmctl_do_lock(),
2200 * no serialization needed when called from shm_destroy().
2202 if (lock && !(info->flags & VM_LOCKED)) {
2203 if (!user_shm_lock(inode->i_size, ucounts))
2205 info->flags |= VM_LOCKED;
2206 mapping_set_unevictable(file->f_mapping);
2208 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2209 user_shm_unlock(inode->i_size, ucounts);
2210 info->flags &= ~VM_LOCKED;
2211 mapping_clear_unevictable(file->f_mapping);
2219 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2221 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2224 ret = seal_check_future_write(info->seals, vma);
2228 /* arm64 - allow memory tagging on RAM-based files */
2229 vma->vm_flags |= VM_MTE_ALLOWED;
2231 file_accessed(file);
2232 vma->vm_ops = &shmem_vm_ops;
2236 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2237 umode_t mode, dev_t dev, unsigned long flags)
2239 struct inode *inode;
2240 struct shmem_inode_info *info;
2241 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2244 if (shmem_reserve_inode(sb, &ino))
2247 inode = new_inode(sb);
2250 inode_init_owner(&init_user_ns, inode, dir, mode);
2251 inode->i_blocks = 0;
2252 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2253 inode->i_generation = prandom_u32();
2254 info = SHMEM_I(inode);
2255 memset(info, 0, (char *)inode - (char *)info);
2256 spin_lock_init(&info->lock);
2257 atomic_set(&info->stop_eviction, 0);
2258 info->seals = F_SEAL_SEAL;
2259 info->flags = flags & VM_NORESERVE;
2260 info->i_crtime = inode->i_mtime;
2261 INIT_LIST_HEAD(&info->shrinklist);
2262 INIT_LIST_HEAD(&info->swaplist);
2263 simple_xattrs_init(&info->xattrs);
2264 cache_no_acl(inode);
2265 mapping_set_large_folios(inode->i_mapping);
2267 switch (mode & S_IFMT) {
2269 inode->i_op = &shmem_special_inode_operations;
2270 init_special_inode(inode, mode, dev);
2273 inode->i_mapping->a_ops = &shmem_aops;
2274 inode->i_op = &shmem_inode_operations;
2275 inode->i_fop = &shmem_file_operations;
2276 mpol_shared_policy_init(&info->policy,
2277 shmem_get_sbmpol(sbinfo));
2281 /* Some things misbehave if size == 0 on a directory */
2282 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2283 inode->i_op = &shmem_dir_inode_operations;
2284 inode->i_fop = &simple_dir_operations;
2288 * Must not load anything in the rbtree,
2289 * mpol_free_shared_policy will not be called.
2291 mpol_shared_policy_init(&info->policy, NULL);
2295 lockdep_annotate_inode_mutex_key(inode);
2297 shmem_free_inode(sb);
2301 #ifdef CONFIG_USERFAULTFD
2302 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2304 struct vm_area_struct *dst_vma,
2305 unsigned long dst_addr,
2306 unsigned long src_addr,
2307 bool zeropage, bool wp_copy,
2308 struct page **pagep)
2310 struct inode *inode = file_inode(dst_vma->vm_file);
2311 struct shmem_inode_info *info = SHMEM_I(inode);
2312 struct address_space *mapping = inode->i_mapping;
2313 gfp_t gfp = mapping_gfp_mask(mapping);
2314 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2316 struct folio *folio;
2321 if (!shmem_inode_acct_block(inode, 1)) {
2323 * We may have got a page, returned -ENOENT triggering a retry,
2324 * and now we find ourselves with -ENOMEM. Release the page, to
2325 * avoid a BUG_ON in our caller.
2327 if (unlikely(*pagep)) {
2336 page = shmem_alloc_page(gfp, info, pgoff);
2338 goto out_unacct_blocks;
2340 if (!zeropage) { /* COPY */
2341 page_kaddr = kmap_atomic(page);
2342 ret = copy_from_user(page_kaddr,
2343 (const void __user *)src_addr,
2345 kunmap_atomic(page_kaddr);
2347 /* fallback to copy_from_user outside mmap_lock */
2348 if (unlikely(ret)) {
2351 /* don't free the page */
2352 goto out_unacct_blocks;
2355 flush_dcache_page(page);
2356 } else { /* ZEROPAGE */
2357 clear_user_highpage(page, dst_addr);
2364 VM_BUG_ON(PageLocked(page));
2365 VM_BUG_ON(PageSwapBacked(page));
2366 __SetPageLocked(page);
2367 __SetPageSwapBacked(page);
2368 __SetPageUptodate(page);
2371 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2372 if (unlikely(pgoff >= max_off))
2375 folio = page_folio(page);
2376 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2377 gfp & GFP_RECLAIM_MASK, dst_mm);
2381 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2382 page, true, wp_copy);
2384 goto out_delete_from_cache;
2386 spin_lock_irq(&info->lock);
2388 inode->i_blocks += BLOCKS_PER_PAGE;
2389 shmem_recalc_inode(inode);
2390 spin_unlock_irq(&info->lock);
2394 out_delete_from_cache:
2395 delete_from_page_cache(page);
2400 shmem_inode_unacct_blocks(inode, 1);
2403 #endif /* CONFIG_USERFAULTFD */
2406 static const struct inode_operations shmem_symlink_inode_operations;
2407 static const struct inode_operations shmem_short_symlink_operations;
2409 #ifdef CONFIG_TMPFS_XATTR
2410 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2412 #define shmem_initxattrs NULL
2416 shmem_write_begin(struct file *file, struct address_space *mapping,
2417 loff_t pos, unsigned len, unsigned flags,
2418 struct page **pagep, void **fsdata)
2420 struct inode *inode = mapping->host;
2421 struct shmem_inode_info *info = SHMEM_I(inode);
2422 pgoff_t index = pos >> PAGE_SHIFT;
2425 /* i_rwsem is held by caller */
2426 if (unlikely(info->seals & (F_SEAL_GROW |
2427 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2428 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2430 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2434 ret = shmem_getpage(inode, index, pagep, SGP_WRITE);
2439 if (PageHWPoison(*pagep)) {
2440 unlock_page(*pagep);
2450 shmem_write_end(struct file *file, struct address_space *mapping,
2451 loff_t pos, unsigned len, unsigned copied,
2452 struct page *page, void *fsdata)
2454 struct inode *inode = mapping->host;
2456 if (pos + copied > inode->i_size)
2457 i_size_write(inode, pos + copied);
2459 if (!PageUptodate(page)) {
2460 struct page *head = compound_head(page);
2461 if (PageTransCompound(page)) {
2464 for (i = 0; i < HPAGE_PMD_NR; i++) {
2465 if (head + i == page)
2467 clear_highpage(head + i);
2468 flush_dcache_page(head + i);
2471 if (copied < PAGE_SIZE) {
2472 unsigned from = pos & (PAGE_SIZE - 1);
2473 zero_user_segments(page, 0, from,
2474 from + copied, PAGE_SIZE);
2476 SetPageUptodate(head);
2478 set_page_dirty(page);
2485 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2487 struct file *file = iocb->ki_filp;
2488 struct inode *inode = file_inode(file);
2489 struct address_space *mapping = inode->i_mapping;
2491 unsigned long offset;
2494 loff_t *ppos = &iocb->ki_pos;
2496 index = *ppos >> PAGE_SHIFT;
2497 offset = *ppos & ~PAGE_MASK;
2500 struct page *page = NULL;
2502 unsigned long nr, ret;
2503 loff_t i_size = i_size_read(inode);
2505 end_index = i_size >> PAGE_SHIFT;
2506 if (index > end_index)
2508 if (index == end_index) {
2509 nr = i_size & ~PAGE_MASK;
2514 error = shmem_getpage(inode, index, &page, SGP_READ);
2516 if (error == -EINVAL)
2523 if (PageHWPoison(page)) {
2531 * We must evaluate after, since reads (unlike writes)
2532 * are called without i_rwsem protection against truncate
2535 i_size = i_size_read(inode);
2536 end_index = i_size >> PAGE_SHIFT;
2537 if (index == end_index) {
2538 nr = i_size & ~PAGE_MASK;
2549 * If users can be writing to this page using arbitrary
2550 * virtual addresses, take care about potential aliasing
2551 * before reading the page on the kernel side.
2553 if (mapping_writably_mapped(mapping))
2554 flush_dcache_page(page);
2556 * Mark the page accessed if we read the beginning.
2559 mark_page_accessed(page);
2561 * Ok, we have the page, and it's up-to-date, so
2562 * now we can copy it to user space...
2564 ret = copy_page_to_iter(page, offset, nr, to);
2567 } else if (iter_is_iovec(to)) {
2569 * Copy to user tends to be so well optimized, but
2570 * clear_user() not so much, that it is noticeably
2571 * faster to copy the zero page instead of clearing.
2573 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2576 * But submitting the same page twice in a row to
2577 * splice() - or others? - can result in confusion:
2578 * so don't attempt that optimization on pipes etc.
2580 ret = iov_iter_zero(nr, to);
2585 index += offset >> PAGE_SHIFT;
2586 offset &= ~PAGE_MASK;
2588 if (!iov_iter_count(to))
2597 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2598 file_accessed(file);
2599 return retval ? retval : error;
2602 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2604 struct address_space *mapping = file->f_mapping;
2605 struct inode *inode = mapping->host;
2607 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2608 return generic_file_llseek_size(file, offset, whence,
2609 MAX_LFS_FILESIZE, i_size_read(inode));
2614 /* We're holding i_rwsem so we can access i_size directly */
2615 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2617 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2618 inode_unlock(inode);
2622 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2625 struct inode *inode = file_inode(file);
2626 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2627 struct shmem_inode_info *info = SHMEM_I(inode);
2628 struct shmem_falloc shmem_falloc;
2629 pgoff_t start, index, end, undo_fallocend;
2632 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2637 if (mode & FALLOC_FL_PUNCH_HOLE) {
2638 struct address_space *mapping = file->f_mapping;
2639 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2640 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2641 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2643 /* protected by i_rwsem */
2644 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2649 shmem_falloc.waitq = &shmem_falloc_waitq;
2650 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2651 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2652 spin_lock(&inode->i_lock);
2653 inode->i_private = &shmem_falloc;
2654 spin_unlock(&inode->i_lock);
2656 if ((u64)unmap_end > (u64)unmap_start)
2657 unmap_mapping_range(mapping, unmap_start,
2658 1 + unmap_end - unmap_start, 0);
2659 shmem_truncate_range(inode, offset, offset + len - 1);
2660 /* No need to unmap again: hole-punching leaves COWed pages */
2662 spin_lock(&inode->i_lock);
2663 inode->i_private = NULL;
2664 wake_up_all(&shmem_falloc_waitq);
2665 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2666 spin_unlock(&inode->i_lock);
2671 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2672 error = inode_newsize_ok(inode, offset + len);
2676 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2681 start = offset >> PAGE_SHIFT;
2682 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2683 /* Try to avoid a swapstorm if len is impossible to satisfy */
2684 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2689 shmem_falloc.waitq = NULL;
2690 shmem_falloc.start = start;
2691 shmem_falloc.next = start;
2692 shmem_falloc.nr_falloced = 0;
2693 shmem_falloc.nr_unswapped = 0;
2694 spin_lock(&inode->i_lock);
2695 inode->i_private = &shmem_falloc;
2696 spin_unlock(&inode->i_lock);
2699 * info->fallocend is only relevant when huge pages might be
2700 * involved: to prevent split_huge_page() freeing fallocated
2701 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2703 undo_fallocend = info->fallocend;
2704 if (info->fallocend < end)
2705 info->fallocend = end;
2707 for (index = start; index < end; ) {
2711 * Good, the fallocate(2) manpage permits EINTR: we may have
2712 * been interrupted because we are using up too much memory.
2714 if (signal_pending(current))
2716 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2719 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2721 info->fallocend = undo_fallocend;
2722 /* Remove the !PageUptodate pages we added */
2723 if (index > start) {
2724 shmem_undo_range(inode,
2725 (loff_t)start << PAGE_SHIFT,
2726 ((loff_t)index << PAGE_SHIFT) - 1, true);
2733 * Here is a more important optimization than it appears:
2734 * a second SGP_FALLOC on the same huge page will clear it,
2735 * making it PageUptodate and un-undoable if we fail later.
2737 if (PageTransCompound(page)) {
2738 index = round_up(index, HPAGE_PMD_NR);
2739 /* Beware 32-bit wraparound */
2745 * Inform shmem_writepage() how far we have reached.
2746 * No need for lock or barrier: we have the page lock.
2748 if (!PageUptodate(page))
2749 shmem_falloc.nr_falloced += index - shmem_falloc.next;
2750 shmem_falloc.next = index;
2753 * If !PageUptodate, leave it that way so that freeable pages
2754 * can be recognized if we need to rollback on error later.
2755 * But set_page_dirty so that memory pressure will swap rather
2756 * than free the pages we are allocating (and SGP_CACHE pages
2757 * might still be clean: we now need to mark those dirty too).
2759 set_page_dirty(page);
2765 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2766 i_size_write(inode, offset + len);
2767 inode->i_ctime = current_time(inode);
2769 spin_lock(&inode->i_lock);
2770 inode->i_private = NULL;
2771 spin_unlock(&inode->i_lock);
2773 inode_unlock(inode);
2777 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2779 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2781 buf->f_type = TMPFS_MAGIC;
2782 buf->f_bsize = PAGE_SIZE;
2783 buf->f_namelen = NAME_MAX;
2784 if (sbinfo->max_blocks) {
2785 buf->f_blocks = sbinfo->max_blocks;
2787 buf->f_bfree = sbinfo->max_blocks -
2788 percpu_counter_sum(&sbinfo->used_blocks);
2790 if (sbinfo->max_inodes) {
2791 buf->f_files = sbinfo->max_inodes;
2792 buf->f_ffree = sbinfo->free_inodes;
2794 /* else leave those fields 0 like simple_statfs */
2796 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2802 * File creation. Allocate an inode, and we're done..
2805 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2806 struct dentry *dentry, umode_t mode, dev_t dev)
2808 struct inode *inode;
2809 int error = -ENOSPC;
2811 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2813 error = simple_acl_create(dir, inode);
2816 error = security_inode_init_security(inode, dir,
2818 shmem_initxattrs, NULL);
2819 if (error && error != -EOPNOTSUPP)
2823 dir->i_size += BOGO_DIRENT_SIZE;
2824 dir->i_ctime = dir->i_mtime = current_time(dir);
2825 d_instantiate(dentry, inode);
2826 dget(dentry); /* Extra count - pin the dentry in core */
2835 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2836 struct dentry *dentry, umode_t mode)
2838 struct inode *inode;
2839 int error = -ENOSPC;
2841 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2843 error = security_inode_init_security(inode, dir,
2845 shmem_initxattrs, NULL);
2846 if (error && error != -EOPNOTSUPP)
2848 error = simple_acl_create(dir, inode);
2851 d_tmpfile(dentry, inode);
2859 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2860 struct dentry *dentry, umode_t mode)
2864 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2865 mode | S_IFDIR, 0)))
2871 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2872 struct dentry *dentry, umode_t mode, bool excl)
2874 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2880 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2882 struct inode *inode = d_inode(old_dentry);
2886 * No ordinary (disk based) filesystem counts links as inodes;
2887 * but each new link needs a new dentry, pinning lowmem, and
2888 * tmpfs dentries cannot be pruned until they are unlinked.
2889 * But if an O_TMPFILE file is linked into the tmpfs, the
2890 * first link must skip that, to get the accounting right.
2892 if (inode->i_nlink) {
2893 ret = shmem_reserve_inode(inode->i_sb, NULL);
2898 dir->i_size += BOGO_DIRENT_SIZE;
2899 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2901 ihold(inode); /* New dentry reference */
2902 dget(dentry); /* Extra pinning count for the created dentry */
2903 d_instantiate(dentry, inode);
2908 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2910 struct inode *inode = d_inode(dentry);
2912 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2913 shmem_free_inode(inode->i_sb);
2915 dir->i_size -= BOGO_DIRENT_SIZE;
2916 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2918 dput(dentry); /* Undo the count from "create" - this does all the work */
2922 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2924 if (!simple_empty(dentry))
2927 drop_nlink(d_inode(dentry));
2929 return shmem_unlink(dir, dentry);
2932 static int shmem_whiteout(struct user_namespace *mnt_userns,
2933 struct inode *old_dir, struct dentry *old_dentry)
2935 struct dentry *whiteout;
2938 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2942 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2943 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2949 * Cheat and hash the whiteout while the old dentry is still in
2950 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2952 * d_lookup() will consistently find one of them at this point,
2953 * not sure which one, but that isn't even important.
2960 * The VFS layer already does all the dentry stuff for rename,
2961 * we just have to decrement the usage count for the target if
2962 * it exists so that the VFS layer correctly free's it when it
2965 static int shmem_rename2(struct user_namespace *mnt_userns,
2966 struct inode *old_dir, struct dentry *old_dentry,
2967 struct inode *new_dir, struct dentry *new_dentry,
2970 struct inode *inode = d_inode(old_dentry);
2971 int they_are_dirs = S_ISDIR(inode->i_mode);
2973 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2976 if (flags & RENAME_EXCHANGE)
2977 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
2979 if (!simple_empty(new_dentry))
2982 if (flags & RENAME_WHITEOUT) {
2985 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
2990 if (d_really_is_positive(new_dentry)) {
2991 (void) shmem_unlink(new_dir, new_dentry);
2992 if (they_are_dirs) {
2993 drop_nlink(d_inode(new_dentry));
2994 drop_nlink(old_dir);
2996 } else if (they_are_dirs) {
2997 drop_nlink(old_dir);
3001 old_dir->i_size -= BOGO_DIRENT_SIZE;
3002 new_dir->i_size += BOGO_DIRENT_SIZE;
3003 old_dir->i_ctime = old_dir->i_mtime =
3004 new_dir->i_ctime = new_dir->i_mtime =
3005 inode->i_ctime = current_time(old_dir);
3009 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3010 struct dentry *dentry, const char *symname)
3014 struct inode *inode;
3017 len = strlen(symname) + 1;
3018 if (len > PAGE_SIZE)
3019 return -ENAMETOOLONG;
3021 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3026 error = security_inode_init_security(inode, dir, &dentry->d_name,
3027 shmem_initxattrs, NULL);
3028 if (error && error != -EOPNOTSUPP) {
3033 inode->i_size = len-1;
3034 if (len <= SHORT_SYMLINK_LEN) {
3035 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3036 if (!inode->i_link) {
3040 inode->i_op = &shmem_short_symlink_operations;
3042 inode_nohighmem(inode);
3043 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3048 inode->i_mapping->a_ops = &shmem_aops;
3049 inode->i_op = &shmem_symlink_inode_operations;
3050 memcpy(page_address(page), symname, len);
3051 SetPageUptodate(page);
3052 set_page_dirty(page);
3056 dir->i_size += BOGO_DIRENT_SIZE;
3057 dir->i_ctime = dir->i_mtime = current_time(dir);
3058 d_instantiate(dentry, inode);
3063 static void shmem_put_link(void *arg)
3065 mark_page_accessed(arg);
3069 static const char *shmem_get_link(struct dentry *dentry,
3070 struct inode *inode,
3071 struct delayed_call *done)
3073 struct page *page = NULL;
3076 page = find_get_page(inode->i_mapping, 0);
3078 return ERR_PTR(-ECHILD);
3079 if (PageHWPoison(page) ||
3080 !PageUptodate(page)) {
3082 return ERR_PTR(-ECHILD);
3085 error = shmem_getpage(inode, 0, &page, SGP_READ);
3087 return ERR_PTR(error);
3089 return ERR_PTR(-ECHILD);
3090 if (PageHWPoison(page)) {
3093 return ERR_PTR(-ECHILD);
3097 set_delayed_call(done, shmem_put_link, page);
3098 return page_address(page);
3101 #ifdef CONFIG_TMPFS_XATTR
3103 * Superblocks without xattr inode operations may get some security.* xattr
3104 * support from the LSM "for free". As soon as we have any other xattrs
3105 * like ACLs, we also need to implement the security.* handlers at
3106 * filesystem level, though.
3110 * Callback for security_inode_init_security() for acquiring xattrs.
3112 static int shmem_initxattrs(struct inode *inode,
3113 const struct xattr *xattr_array,
3116 struct shmem_inode_info *info = SHMEM_I(inode);
3117 const struct xattr *xattr;
3118 struct simple_xattr *new_xattr;
3121 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3122 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3126 len = strlen(xattr->name) + 1;
3127 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3129 if (!new_xattr->name) {
3134 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3135 XATTR_SECURITY_PREFIX_LEN);
3136 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3139 simple_xattr_list_add(&info->xattrs, new_xattr);
3145 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3146 struct dentry *unused, struct inode *inode,
3147 const char *name, void *buffer, size_t size)
3149 struct shmem_inode_info *info = SHMEM_I(inode);
3151 name = xattr_full_name(handler, name);
3152 return simple_xattr_get(&info->xattrs, name, buffer, size);
3155 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3156 struct user_namespace *mnt_userns,
3157 struct dentry *unused, struct inode *inode,
3158 const char *name, const void *value,
3159 size_t size, int flags)
3161 struct shmem_inode_info *info = SHMEM_I(inode);
3163 name = xattr_full_name(handler, name);
3164 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3167 static const struct xattr_handler shmem_security_xattr_handler = {
3168 .prefix = XATTR_SECURITY_PREFIX,
3169 .get = shmem_xattr_handler_get,
3170 .set = shmem_xattr_handler_set,
3173 static const struct xattr_handler shmem_trusted_xattr_handler = {
3174 .prefix = XATTR_TRUSTED_PREFIX,
3175 .get = shmem_xattr_handler_get,
3176 .set = shmem_xattr_handler_set,
3179 static const struct xattr_handler *shmem_xattr_handlers[] = {
3180 #ifdef CONFIG_TMPFS_POSIX_ACL
3181 &posix_acl_access_xattr_handler,
3182 &posix_acl_default_xattr_handler,
3184 &shmem_security_xattr_handler,
3185 &shmem_trusted_xattr_handler,
3189 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3191 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3192 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3194 #endif /* CONFIG_TMPFS_XATTR */
3196 static const struct inode_operations shmem_short_symlink_operations = {
3197 .getattr = shmem_getattr,
3198 .get_link = simple_get_link,
3199 #ifdef CONFIG_TMPFS_XATTR
3200 .listxattr = shmem_listxattr,
3204 static const struct inode_operations shmem_symlink_inode_operations = {
3205 .getattr = shmem_getattr,
3206 .get_link = shmem_get_link,
3207 #ifdef CONFIG_TMPFS_XATTR
3208 .listxattr = shmem_listxattr,
3212 static struct dentry *shmem_get_parent(struct dentry *child)
3214 return ERR_PTR(-ESTALE);
3217 static int shmem_match(struct inode *ino, void *vfh)
3221 inum = (inum << 32) | fh[1];
3222 return ino->i_ino == inum && fh[0] == ino->i_generation;
3225 /* Find any alias of inode, but prefer a hashed alias */
3226 static struct dentry *shmem_find_alias(struct inode *inode)
3228 struct dentry *alias = d_find_alias(inode);
3230 return alias ?: d_find_any_alias(inode);
3234 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3235 struct fid *fid, int fh_len, int fh_type)
3237 struct inode *inode;
3238 struct dentry *dentry = NULL;
3245 inum = (inum << 32) | fid->raw[1];
3247 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3248 shmem_match, fid->raw);
3250 dentry = shmem_find_alias(inode);
3257 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3258 struct inode *parent)
3262 return FILEID_INVALID;
3265 if (inode_unhashed(inode)) {
3266 /* Unfortunately insert_inode_hash is not idempotent,
3267 * so as we hash inodes here rather than at creation
3268 * time, we need a lock to ensure we only try
3271 static DEFINE_SPINLOCK(lock);
3273 if (inode_unhashed(inode))
3274 __insert_inode_hash(inode,
3275 inode->i_ino + inode->i_generation);
3279 fh[0] = inode->i_generation;
3280 fh[1] = inode->i_ino;
3281 fh[2] = ((__u64)inode->i_ino) >> 32;
3287 static const struct export_operations shmem_export_ops = {
3288 .get_parent = shmem_get_parent,
3289 .encode_fh = shmem_encode_fh,
3290 .fh_to_dentry = shmem_fh_to_dentry,
3306 static const struct constant_table shmem_param_enums_huge[] = {
3307 {"never", SHMEM_HUGE_NEVER },
3308 {"always", SHMEM_HUGE_ALWAYS },
3309 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3310 {"advise", SHMEM_HUGE_ADVISE },
3314 const struct fs_parameter_spec shmem_fs_parameters[] = {
3315 fsparam_u32 ("gid", Opt_gid),
3316 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3317 fsparam_u32oct("mode", Opt_mode),
3318 fsparam_string("mpol", Opt_mpol),
3319 fsparam_string("nr_blocks", Opt_nr_blocks),
3320 fsparam_string("nr_inodes", Opt_nr_inodes),
3321 fsparam_string("size", Opt_size),
3322 fsparam_u32 ("uid", Opt_uid),
3323 fsparam_flag ("inode32", Opt_inode32),
3324 fsparam_flag ("inode64", Opt_inode64),
3328 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3330 struct shmem_options *ctx = fc->fs_private;
3331 struct fs_parse_result result;
3332 unsigned long long size;
3336 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3342 size = memparse(param->string, &rest);
3344 size <<= PAGE_SHIFT;
3345 size *= totalram_pages();
3351 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3352 ctx->seen |= SHMEM_SEEN_BLOCKS;
3355 ctx->blocks = memparse(param->string, &rest);
3358 ctx->seen |= SHMEM_SEEN_BLOCKS;
3361 ctx->inodes = memparse(param->string, &rest);
3364 ctx->seen |= SHMEM_SEEN_INODES;
3367 ctx->mode = result.uint_32 & 07777;
3370 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3371 if (!uid_valid(ctx->uid))
3375 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3376 if (!gid_valid(ctx->gid))
3380 ctx->huge = result.uint_32;
3381 if (ctx->huge != SHMEM_HUGE_NEVER &&
3382 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3383 has_transparent_hugepage()))
3384 goto unsupported_parameter;
3385 ctx->seen |= SHMEM_SEEN_HUGE;
3388 if (IS_ENABLED(CONFIG_NUMA)) {
3389 mpol_put(ctx->mpol);
3391 if (mpol_parse_str(param->string, &ctx->mpol))
3395 goto unsupported_parameter;
3397 ctx->full_inums = false;
3398 ctx->seen |= SHMEM_SEEN_INUMS;
3401 if (sizeof(ino_t) < 8) {
3403 "Cannot use inode64 with <64bit inums in kernel\n");
3405 ctx->full_inums = true;
3406 ctx->seen |= SHMEM_SEEN_INUMS;
3411 unsupported_parameter:
3412 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3414 return invalfc(fc, "Bad value for '%s'", param->key);
3417 static int shmem_parse_options(struct fs_context *fc, void *data)
3419 char *options = data;
3422 int err = security_sb_eat_lsm_opts(options, &fc->security);
3427 while (options != NULL) {
3428 char *this_char = options;
3431 * NUL-terminate this option: unfortunately,
3432 * mount options form a comma-separated list,
3433 * but mpol's nodelist may also contain commas.
3435 options = strchr(options, ',');
3436 if (options == NULL)
3439 if (!isdigit(*options)) {
3445 char *value = strchr(this_char, '=');
3451 len = strlen(value);
3453 err = vfs_parse_fs_string(fc, this_char, value, len);
3462 * Reconfigure a shmem filesystem.
3464 * Note that we disallow change from limited->unlimited blocks/inodes while any
3465 * are in use; but we must separately disallow unlimited->limited, because in
3466 * that case we have no record of how much is already in use.
3468 static int shmem_reconfigure(struct fs_context *fc)
3470 struct shmem_options *ctx = fc->fs_private;
3471 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3472 unsigned long inodes;
3473 struct mempolicy *mpol = NULL;
3476 raw_spin_lock(&sbinfo->stat_lock);
3477 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3478 if (ctx->blocks > S64_MAX) {
3479 err = "Number of blocks too large";
3482 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3483 if (!sbinfo->max_blocks) {
3484 err = "Cannot retroactively limit size";
3487 if (percpu_counter_compare(&sbinfo->used_blocks,
3489 err = "Too small a size for current use";
3493 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3494 if (!sbinfo->max_inodes) {
3495 err = "Cannot retroactively limit inodes";
3498 if (ctx->inodes < inodes) {
3499 err = "Too few inodes for current use";
3504 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3505 sbinfo->next_ino > UINT_MAX) {
3506 err = "Current inum too high to switch to 32-bit inums";
3510 if (ctx->seen & SHMEM_SEEN_HUGE)
3511 sbinfo->huge = ctx->huge;
3512 if (ctx->seen & SHMEM_SEEN_INUMS)
3513 sbinfo->full_inums = ctx->full_inums;
3514 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3515 sbinfo->max_blocks = ctx->blocks;
3516 if (ctx->seen & SHMEM_SEEN_INODES) {
3517 sbinfo->max_inodes = ctx->inodes;
3518 sbinfo->free_inodes = ctx->inodes - inodes;
3522 * Preserve previous mempolicy unless mpol remount option was specified.
3525 mpol = sbinfo->mpol;
3526 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3529 raw_spin_unlock(&sbinfo->stat_lock);
3533 raw_spin_unlock(&sbinfo->stat_lock);
3534 return invalfc(fc, "%s", err);
3537 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3539 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3541 if (sbinfo->max_blocks != shmem_default_max_blocks())
3542 seq_printf(seq, ",size=%luk",
3543 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3544 if (sbinfo->max_inodes != shmem_default_max_inodes())
3545 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3546 if (sbinfo->mode != (0777 | S_ISVTX))
3547 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3548 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3549 seq_printf(seq, ",uid=%u",
3550 from_kuid_munged(&init_user_ns, sbinfo->uid));
3551 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3552 seq_printf(seq, ",gid=%u",
3553 from_kgid_munged(&init_user_ns, sbinfo->gid));
3556 * Showing inode{64,32} might be useful even if it's the system default,
3557 * since then people don't have to resort to checking both here and
3558 * /proc/config.gz to confirm 64-bit inums were successfully applied
3559 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3561 * We hide it when inode64 isn't the default and we are using 32-bit
3562 * inodes, since that probably just means the feature isn't even under
3567 * +-----------------+-----------------+
3568 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3569 * +------------------+-----------------+-----------------+
3570 * | full_inums=true | show | show |
3571 * | full_inums=false | show | hide |
3572 * +------------------+-----------------+-----------------+
3575 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3576 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3577 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3578 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3580 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3582 shmem_show_mpol(seq, sbinfo->mpol);
3586 #endif /* CONFIG_TMPFS */
3588 static void shmem_put_super(struct super_block *sb)
3590 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3592 free_percpu(sbinfo->ino_batch);
3593 percpu_counter_destroy(&sbinfo->used_blocks);
3594 mpol_put(sbinfo->mpol);
3596 sb->s_fs_info = NULL;
3599 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3601 struct shmem_options *ctx = fc->fs_private;
3602 struct inode *inode;
3603 struct shmem_sb_info *sbinfo;
3605 /* Round up to L1_CACHE_BYTES to resist false sharing */
3606 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3607 L1_CACHE_BYTES), GFP_KERNEL);
3611 sb->s_fs_info = sbinfo;
3615 * Per default we only allow half of the physical ram per
3616 * tmpfs instance, limiting inodes to one per page of lowmem;
3617 * but the internal instance is left unlimited.
3619 if (!(sb->s_flags & SB_KERNMOUNT)) {
3620 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3621 ctx->blocks = shmem_default_max_blocks();
3622 if (!(ctx->seen & SHMEM_SEEN_INODES))
3623 ctx->inodes = shmem_default_max_inodes();
3624 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3625 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3627 sb->s_flags |= SB_NOUSER;
3629 sb->s_export_op = &shmem_export_ops;
3630 sb->s_flags |= SB_NOSEC;
3632 sb->s_flags |= SB_NOUSER;
3634 sbinfo->max_blocks = ctx->blocks;
3635 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3636 if (sb->s_flags & SB_KERNMOUNT) {
3637 sbinfo->ino_batch = alloc_percpu(ino_t);
3638 if (!sbinfo->ino_batch)
3641 sbinfo->uid = ctx->uid;
3642 sbinfo->gid = ctx->gid;
3643 sbinfo->full_inums = ctx->full_inums;
3644 sbinfo->mode = ctx->mode;
3645 sbinfo->huge = ctx->huge;
3646 sbinfo->mpol = ctx->mpol;
3649 raw_spin_lock_init(&sbinfo->stat_lock);
3650 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3652 spin_lock_init(&sbinfo->shrinklist_lock);
3653 INIT_LIST_HEAD(&sbinfo->shrinklist);
3655 sb->s_maxbytes = MAX_LFS_FILESIZE;
3656 sb->s_blocksize = PAGE_SIZE;
3657 sb->s_blocksize_bits = PAGE_SHIFT;
3658 sb->s_magic = TMPFS_MAGIC;
3659 sb->s_op = &shmem_ops;
3660 sb->s_time_gran = 1;
3661 #ifdef CONFIG_TMPFS_XATTR
3662 sb->s_xattr = shmem_xattr_handlers;
3664 #ifdef CONFIG_TMPFS_POSIX_ACL
3665 sb->s_flags |= SB_POSIXACL;
3667 uuid_gen(&sb->s_uuid);
3669 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3672 inode->i_uid = sbinfo->uid;
3673 inode->i_gid = sbinfo->gid;
3674 sb->s_root = d_make_root(inode);
3680 shmem_put_super(sb);
3684 static int shmem_get_tree(struct fs_context *fc)
3686 return get_tree_nodev(fc, shmem_fill_super);
3689 static void shmem_free_fc(struct fs_context *fc)
3691 struct shmem_options *ctx = fc->fs_private;
3694 mpol_put(ctx->mpol);
3699 static const struct fs_context_operations shmem_fs_context_ops = {
3700 .free = shmem_free_fc,
3701 .get_tree = shmem_get_tree,
3703 .parse_monolithic = shmem_parse_options,
3704 .parse_param = shmem_parse_one,
3705 .reconfigure = shmem_reconfigure,
3709 static struct kmem_cache *shmem_inode_cachep;
3711 static struct inode *shmem_alloc_inode(struct super_block *sb)
3713 struct shmem_inode_info *info;
3714 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
3717 return &info->vfs_inode;
3720 static void shmem_free_in_core_inode(struct inode *inode)
3722 if (S_ISLNK(inode->i_mode))
3723 kfree(inode->i_link);
3724 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3727 static void shmem_destroy_inode(struct inode *inode)
3729 if (S_ISREG(inode->i_mode))
3730 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3733 static void shmem_init_inode(void *foo)
3735 struct shmem_inode_info *info = foo;
3736 inode_init_once(&info->vfs_inode);
3739 static void shmem_init_inodecache(void)
3741 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3742 sizeof(struct shmem_inode_info),
3743 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3746 static void shmem_destroy_inodecache(void)
3748 kmem_cache_destroy(shmem_inode_cachep);
3751 /* Keep the page in page cache instead of truncating it */
3752 static int shmem_error_remove_page(struct address_space *mapping,
3758 const struct address_space_operations shmem_aops = {
3759 .writepage = shmem_writepage,
3760 .dirty_folio = noop_dirty_folio,
3762 .write_begin = shmem_write_begin,
3763 .write_end = shmem_write_end,
3765 #ifdef CONFIG_MIGRATION
3766 .migratepage = migrate_page,
3768 .error_remove_page = shmem_error_remove_page,
3770 EXPORT_SYMBOL(shmem_aops);
3772 static const struct file_operations shmem_file_operations = {
3774 .get_unmapped_area = shmem_get_unmapped_area,
3776 .llseek = shmem_file_llseek,
3777 .read_iter = shmem_file_read_iter,
3778 .write_iter = generic_file_write_iter,
3779 .fsync = noop_fsync,
3780 .splice_read = generic_file_splice_read,
3781 .splice_write = iter_file_splice_write,
3782 .fallocate = shmem_fallocate,
3786 static const struct inode_operations shmem_inode_operations = {
3787 .getattr = shmem_getattr,
3788 .setattr = shmem_setattr,
3789 #ifdef CONFIG_TMPFS_XATTR
3790 .listxattr = shmem_listxattr,
3791 .set_acl = simple_set_acl,
3795 static const struct inode_operations shmem_dir_inode_operations = {
3797 .getattr = shmem_getattr,
3798 .create = shmem_create,
3799 .lookup = simple_lookup,
3801 .unlink = shmem_unlink,
3802 .symlink = shmem_symlink,
3803 .mkdir = shmem_mkdir,
3804 .rmdir = shmem_rmdir,
3805 .mknod = shmem_mknod,
3806 .rename = shmem_rename2,
3807 .tmpfile = shmem_tmpfile,
3809 #ifdef CONFIG_TMPFS_XATTR
3810 .listxattr = shmem_listxattr,
3812 #ifdef CONFIG_TMPFS_POSIX_ACL
3813 .setattr = shmem_setattr,
3814 .set_acl = simple_set_acl,
3818 static const struct inode_operations shmem_special_inode_operations = {
3819 .getattr = shmem_getattr,
3820 #ifdef CONFIG_TMPFS_XATTR
3821 .listxattr = shmem_listxattr,
3823 #ifdef CONFIG_TMPFS_POSIX_ACL
3824 .setattr = shmem_setattr,
3825 .set_acl = simple_set_acl,
3829 static const struct super_operations shmem_ops = {
3830 .alloc_inode = shmem_alloc_inode,
3831 .free_inode = shmem_free_in_core_inode,
3832 .destroy_inode = shmem_destroy_inode,
3834 .statfs = shmem_statfs,
3835 .show_options = shmem_show_options,
3837 .evict_inode = shmem_evict_inode,
3838 .drop_inode = generic_delete_inode,
3839 .put_super = shmem_put_super,
3840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3841 .nr_cached_objects = shmem_unused_huge_count,
3842 .free_cached_objects = shmem_unused_huge_scan,
3846 static const struct vm_operations_struct shmem_vm_ops = {
3847 .fault = shmem_fault,
3848 .map_pages = filemap_map_pages,
3850 .set_policy = shmem_set_policy,
3851 .get_policy = shmem_get_policy,
3855 int shmem_init_fs_context(struct fs_context *fc)
3857 struct shmem_options *ctx;
3859 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3863 ctx->mode = 0777 | S_ISVTX;
3864 ctx->uid = current_fsuid();
3865 ctx->gid = current_fsgid();
3867 fc->fs_private = ctx;
3868 fc->ops = &shmem_fs_context_ops;
3872 static struct file_system_type shmem_fs_type = {
3873 .owner = THIS_MODULE,
3875 .init_fs_context = shmem_init_fs_context,
3877 .parameters = shmem_fs_parameters,
3879 .kill_sb = kill_litter_super,
3880 .fs_flags = FS_USERNS_MOUNT,
3883 void __init shmem_init(void)
3887 shmem_init_inodecache();
3889 error = register_filesystem(&shmem_fs_type);
3891 pr_err("Could not register tmpfs\n");
3895 shm_mnt = kern_mount(&shmem_fs_type);
3896 if (IS_ERR(shm_mnt)) {
3897 error = PTR_ERR(shm_mnt);
3898 pr_err("Could not kern_mount tmpfs\n");
3902 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3903 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3904 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3906 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
3911 unregister_filesystem(&shmem_fs_type);
3913 shmem_destroy_inodecache();
3914 shm_mnt = ERR_PTR(error);
3917 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3918 static ssize_t shmem_enabled_show(struct kobject *kobj,
3919 struct kobj_attribute *attr, char *buf)
3921 static const int values[] = {
3923 SHMEM_HUGE_WITHIN_SIZE,
3932 for (i = 0; i < ARRAY_SIZE(values); i++) {
3933 len += sysfs_emit_at(buf, len,
3934 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3936 shmem_format_huge(values[i]));
3939 len += sysfs_emit_at(buf, len, "\n");
3944 static ssize_t shmem_enabled_store(struct kobject *kobj,
3945 struct kobj_attribute *attr, const char *buf, size_t count)
3950 if (count + 1 > sizeof(tmp))
3952 memcpy(tmp, buf, count);
3954 if (count && tmp[count - 1] == '\n')
3955 tmp[count - 1] = '\0';
3957 huge = shmem_parse_huge(tmp);
3958 if (huge == -EINVAL)
3960 if (!has_transparent_hugepage() &&
3961 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3965 if (shmem_huge > SHMEM_HUGE_DENY)
3966 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3970 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
3971 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3973 #else /* !CONFIG_SHMEM */
3976 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3978 * This is intended for small system where the benefits of the full
3979 * shmem code (swap-backed and resource-limited) are outweighed by
3980 * their complexity. On systems without swap this code should be
3981 * effectively equivalent, but much lighter weight.
3984 static struct file_system_type shmem_fs_type = {
3986 .init_fs_context = ramfs_init_fs_context,
3987 .parameters = ramfs_fs_parameters,
3988 .kill_sb = kill_litter_super,
3989 .fs_flags = FS_USERNS_MOUNT,
3992 void __init shmem_init(void)
3994 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3996 shm_mnt = kern_mount(&shmem_fs_type);
3997 BUG_ON(IS_ERR(shm_mnt));
4000 int shmem_unuse(unsigned int type)
4005 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4010 void shmem_unlock_mapping(struct address_space *mapping)
4015 unsigned long shmem_get_unmapped_area(struct file *file,
4016 unsigned long addr, unsigned long len,
4017 unsigned long pgoff, unsigned long flags)
4019 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4023 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4025 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4027 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4029 #define shmem_vm_ops generic_file_vm_ops
4030 #define shmem_file_operations ramfs_file_operations
4031 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4032 #define shmem_acct_size(flags, size) 0
4033 #define shmem_unacct_size(flags, size) do {} while (0)
4035 #endif /* CONFIG_SHMEM */
4039 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4040 unsigned long flags, unsigned int i_flags)
4042 struct inode *inode;
4046 return ERR_CAST(mnt);
4048 if (size < 0 || size > MAX_LFS_FILESIZE)
4049 return ERR_PTR(-EINVAL);
4051 if (shmem_acct_size(flags, size))
4052 return ERR_PTR(-ENOMEM);
4054 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4056 if (unlikely(!inode)) {
4057 shmem_unacct_size(flags, size);
4058 return ERR_PTR(-ENOSPC);
4060 inode->i_flags |= i_flags;
4061 inode->i_size = size;
4062 clear_nlink(inode); /* It is unlinked */
4063 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4065 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4066 &shmem_file_operations);
4073 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4074 * kernel internal. There will be NO LSM permission checks against the
4075 * underlying inode. So users of this interface must do LSM checks at a
4076 * higher layer. The users are the big_key and shm implementations. LSM
4077 * checks are provided at the key or shm level rather than the inode.
4078 * @name: name for dentry (to be seen in /proc/<pid>/maps
4079 * @size: size to be set for the file
4080 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4082 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4084 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4088 * shmem_file_setup - get an unlinked file living in tmpfs
4089 * @name: name for dentry (to be seen in /proc/<pid>/maps
4090 * @size: size to be set for the file
4091 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4093 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4095 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4097 EXPORT_SYMBOL_GPL(shmem_file_setup);
4100 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4101 * @mnt: the tmpfs mount where the file will be created
4102 * @name: name for dentry (to be seen in /proc/<pid>/maps
4103 * @size: size to be set for the file
4104 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4106 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4107 loff_t size, unsigned long flags)
4109 return __shmem_file_setup(mnt, name, size, flags, 0);
4111 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4114 * shmem_zero_setup - setup a shared anonymous mapping
4115 * @vma: the vma to be mmapped is prepared by do_mmap
4117 int shmem_zero_setup(struct vm_area_struct *vma)
4120 loff_t size = vma->vm_end - vma->vm_start;
4123 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4124 * between XFS directory reading and selinux: since this file is only
4125 * accessible to the user through its mapping, use S_PRIVATE flag to
4126 * bypass file security, in the same way as shmem_kernel_file_setup().
4128 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4130 return PTR_ERR(file);
4134 vma->vm_file = file;
4135 vma->vm_ops = &shmem_vm_ops;
4141 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4142 * @mapping: the page's address_space
4143 * @index: the page index
4144 * @gfp: the page allocator flags to use if allocating
4146 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4147 * with any new page allocations done using the specified allocation flags.
4148 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4149 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4150 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4152 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4153 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4155 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4156 pgoff_t index, gfp_t gfp)
4159 struct inode *inode = mapping->host;
4163 BUG_ON(!shmem_mapping(mapping));
4164 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4165 gfp, NULL, NULL, NULL);
4167 return ERR_PTR(error);
4170 if (PageHWPoison(page)) {
4172 return ERR_PTR(-EIO);
4178 * The tiny !SHMEM case uses ramfs without swap
4180 return read_cache_page_gfp(mapping, index, gfp);
4183 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);