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>
31 #include <linux/fileattr.h>
33 #include <linux/random.h>
34 #include <linux/sched/signal.h>
35 #include <linux/export.h>
36 #include <linux/shmem_fs.h>
37 #include <linux/swap.h>
38 #include <linux/uio.h>
39 #include <linux/hugetlb.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
42 #include <linux/iversion.h>
45 static struct vfsmount *shm_mnt;
49 * This virtual memory filesystem is heavily based on the ramfs. It
50 * extends ramfs by the ability to use swap and honor resource limits
51 * which makes it a completely usable filesystem.
54 #include <linux/xattr.h>
55 #include <linux/exportfs.h>
56 #include <linux/posix_acl.h>
57 #include <linux/posix_acl_xattr.h>
58 #include <linux/mman.h>
59 #include <linux/string.h>
60 #include <linux/slab.h>
61 #include <linux/backing-dev.h>
62 #include <linux/writeback.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
82 #include <linux/uaccess.h>
86 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 /* Pretend that each entry is of this size in directory's i_size */
90 #define BOGO_DIRENT_SIZE 20
92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93 #define SHORT_SYMLINK_LEN 128
96 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_rwsem making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
100 struct shmem_falloc {
101 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
102 pgoff_t start; /* start of range currently being fallocated */
103 pgoff_t next; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
108 struct shmem_options {
109 unsigned long long blocks;
110 unsigned long long inodes;
111 struct mempolicy *mpol;
119 #define SHMEM_SEEN_BLOCKS 1
120 #define SHMEM_SEEN_INODES 2
121 #define SHMEM_SEEN_HUGE 4
122 #define SHMEM_SEEN_INUMS 8
123 #define SHMEM_SEEN_NOSWAP 16
127 static unsigned long shmem_default_max_blocks(void)
129 return totalram_pages() / 2;
132 static unsigned long shmem_default_max_inodes(void)
134 unsigned long nr_pages = totalram_pages();
136 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
140 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
141 struct folio **foliop, enum sgp_type sgp,
142 gfp_t gfp, struct vm_area_struct *vma,
143 vm_fault_t *fault_type);
145 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
147 return sb->s_fs_info;
151 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
152 * for shared memory and for shared anonymous (/dev/zero) mappings
153 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
154 * consistent with the pre-accounting of private mappings ...
156 static inline int shmem_acct_size(unsigned long flags, loff_t size)
158 return (flags & VM_NORESERVE) ?
159 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
162 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
164 if (!(flags & VM_NORESERVE))
165 vm_unacct_memory(VM_ACCT(size));
168 static inline int shmem_reacct_size(unsigned long flags,
169 loff_t oldsize, loff_t newsize)
171 if (!(flags & VM_NORESERVE)) {
172 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
173 return security_vm_enough_memory_mm(current->mm,
174 VM_ACCT(newsize) - VM_ACCT(oldsize));
175 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
176 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
182 * ... whereas tmpfs objects are accounted incrementally as
183 * pages are allocated, in order to allow large sparse files.
184 * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
185 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
187 static inline int shmem_acct_block(unsigned long flags, long pages)
189 if (!(flags & VM_NORESERVE))
192 return security_vm_enough_memory_mm(current->mm,
193 pages * VM_ACCT(PAGE_SIZE));
196 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
198 if (flags & VM_NORESERVE)
199 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
202 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
204 struct shmem_inode_info *info = SHMEM_I(inode);
205 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
207 if (shmem_acct_block(info->flags, pages))
210 if (sbinfo->max_blocks) {
211 if (percpu_counter_compare(&sbinfo->used_blocks,
212 sbinfo->max_blocks - pages) > 0)
214 percpu_counter_add(&sbinfo->used_blocks, pages);
220 shmem_unacct_blocks(info->flags, pages);
224 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
226 struct shmem_inode_info *info = SHMEM_I(inode);
227 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
229 if (sbinfo->max_blocks)
230 percpu_counter_sub(&sbinfo->used_blocks, pages);
231 shmem_unacct_blocks(info->flags, pages);
234 static const struct super_operations shmem_ops;
235 const struct address_space_operations shmem_aops;
236 static const struct file_operations shmem_file_operations;
237 static const struct inode_operations shmem_inode_operations;
238 static const struct inode_operations shmem_dir_inode_operations;
239 static const struct inode_operations shmem_special_inode_operations;
240 static const struct vm_operations_struct shmem_vm_ops;
241 static const struct vm_operations_struct shmem_anon_vm_ops;
242 static struct file_system_type shmem_fs_type;
244 bool vma_is_anon_shmem(struct vm_area_struct *vma)
246 return vma->vm_ops == &shmem_anon_vm_ops;
249 bool vma_is_shmem(struct vm_area_struct *vma)
251 return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
254 static LIST_HEAD(shmem_swaplist);
255 static DEFINE_MUTEX(shmem_swaplist_mutex);
258 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
259 * produces a novel ino for the newly allocated inode.
261 * It may also be called when making a hard link to permit the space needed by
262 * each dentry. However, in that case, no new inode number is needed since that
263 * internally draws from another pool of inode numbers (currently global
264 * get_next_ino()). This case is indicated by passing NULL as inop.
266 #define SHMEM_INO_BATCH 1024
267 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
269 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
272 if (!(sb->s_flags & SB_KERNMOUNT)) {
273 raw_spin_lock(&sbinfo->stat_lock);
274 if (sbinfo->max_inodes) {
275 if (!sbinfo->free_inodes) {
276 raw_spin_unlock(&sbinfo->stat_lock);
279 sbinfo->free_inodes--;
282 ino = sbinfo->next_ino++;
283 if (unlikely(is_zero_ino(ino)))
284 ino = sbinfo->next_ino++;
285 if (unlikely(!sbinfo->full_inums &&
288 * Emulate get_next_ino uint wraparound for
291 if (IS_ENABLED(CONFIG_64BIT))
292 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
293 __func__, MINOR(sb->s_dev));
294 sbinfo->next_ino = 1;
295 ino = sbinfo->next_ino++;
299 raw_spin_unlock(&sbinfo->stat_lock);
302 * __shmem_file_setup, one of our callers, is lock-free: it
303 * doesn't hold stat_lock in shmem_reserve_inode since
304 * max_inodes is always 0, and is called from potentially
305 * unknown contexts. As such, use a per-cpu batched allocator
306 * which doesn't require the per-sb stat_lock unless we are at
307 * the batch boundary.
309 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
310 * shmem mounts are not exposed to userspace, so we don't need
311 * to worry about things like glibc compatibility.
315 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
317 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
318 raw_spin_lock(&sbinfo->stat_lock);
319 ino = sbinfo->next_ino;
320 sbinfo->next_ino += SHMEM_INO_BATCH;
321 raw_spin_unlock(&sbinfo->stat_lock);
322 if (unlikely(is_zero_ino(ino)))
333 static void shmem_free_inode(struct super_block *sb)
335 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
336 if (sbinfo->max_inodes) {
337 raw_spin_lock(&sbinfo->stat_lock);
338 sbinfo->free_inodes++;
339 raw_spin_unlock(&sbinfo->stat_lock);
344 * shmem_recalc_inode - recalculate the block usage of an inode
345 * @inode: inode to recalc
347 * We have to calculate the free blocks since the mm can drop
348 * undirtied hole pages behind our back.
350 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
351 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
353 * It has to be called with the spinlock held.
355 static void shmem_recalc_inode(struct inode *inode)
357 struct shmem_inode_info *info = SHMEM_I(inode);
360 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
362 info->alloced -= freed;
363 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
364 shmem_inode_unacct_blocks(inode, freed);
368 bool shmem_charge(struct inode *inode, long pages)
370 struct shmem_inode_info *info = SHMEM_I(inode);
373 if (!shmem_inode_acct_block(inode, pages))
376 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
377 inode->i_mapping->nrpages += pages;
379 spin_lock_irqsave(&info->lock, flags);
380 info->alloced += pages;
381 inode->i_blocks += pages * BLOCKS_PER_PAGE;
382 shmem_recalc_inode(inode);
383 spin_unlock_irqrestore(&info->lock, flags);
388 void shmem_uncharge(struct inode *inode, long pages)
390 struct shmem_inode_info *info = SHMEM_I(inode);
393 /* nrpages adjustment done by __filemap_remove_folio() or caller */
395 spin_lock_irqsave(&info->lock, flags);
396 info->alloced -= pages;
397 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
398 shmem_recalc_inode(inode);
399 spin_unlock_irqrestore(&info->lock, flags);
401 shmem_inode_unacct_blocks(inode, pages);
405 * Replace item expected in xarray by a new item, while holding xa_lock.
407 static int shmem_replace_entry(struct address_space *mapping,
408 pgoff_t index, void *expected, void *replacement)
410 XA_STATE(xas, &mapping->i_pages, index);
413 VM_BUG_ON(!expected);
414 VM_BUG_ON(!replacement);
415 item = xas_load(&xas);
416 if (item != expected)
418 xas_store(&xas, replacement);
423 * Sometimes, before we decide whether to proceed or to fail, we must check
424 * that an entry was not already brought back from swap by a racing thread.
426 * Checking page is not enough: by the time a SwapCache page is locked, it
427 * might be reused, and again be SwapCache, using the same swap as before.
429 static bool shmem_confirm_swap(struct address_space *mapping,
430 pgoff_t index, swp_entry_t swap)
432 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
436 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
439 * disables huge pages for the mount;
441 * enables huge pages for the mount;
442 * SHMEM_HUGE_WITHIN_SIZE:
443 * only allocate huge pages if the page will be fully within i_size,
444 * also respect fadvise()/madvise() hints;
446 * only allocate huge pages if requested with fadvise()/madvise();
449 #define SHMEM_HUGE_NEVER 0
450 #define SHMEM_HUGE_ALWAYS 1
451 #define SHMEM_HUGE_WITHIN_SIZE 2
452 #define SHMEM_HUGE_ADVISE 3
456 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
459 * disables huge on shm_mnt and all mounts, for emergency use;
461 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
464 #define SHMEM_HUGE_DENY (-1)
465 #define SHMEM_HUGE_FORCE (-2)
467 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
468 /* ifdef here to avoid bloating shmem.o when not necessary */
470 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
472 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
473 struct mm_struct *mm, unsigned long vm_flags)
477 if (!S_ISREG(inode->i_mode))
479 if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
481 if (shmem_huge == SHMEM_HUGE_DENY)
483 if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
486 switch (SHMEM_SB(inode->i_sb)->huge) {
487 case SHMEM_HUGE_ALWAYS:
489 case SHMEM_HUGE_WITHIN_SIZE:
490 index = round_up(index + 1, HPAGE_PMD_NR);
491 i_size = round_up(i_size_read(inode), PAGE_SIZE);
492 if (i_size >> PAGE_SHIFT >= index)
495 case SHMEM_HUGE_ADVISE:
496 if (mm && (vm_flags & VM_HUGEPAGE))
504 #if defined(CONFIG_SYSFS)
505 static int shmem_parse_huge(const char *str)
507 if (!strcmp(str, "never"))
508 return SHMEM_HUGE_NEVER;
509 if (!strcmp(str, "always"))
510 return SHMEM_HUGE_ALWAYS;
511 if (!strcmp(str, "within_size"))
512 return SHMEM_HUGE_WITHIN_SIZE;
513 if (!strcmp(str, "advise"))
514 return SHMEM_HUGE_ADVISE;
515 if (!strcmp(str, "deny"))
516 return SHMEM_HUGE_DENY;
517 if (!strcmp(str, "force"))
518 return SHMEM_HUGE_FORCE;
523 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
524 static const char *shmem_format_huge(int huge)
527 case SHMEM_HUGE_NEVER:
529 case SHMEM_HUGE_ALWAYS:
531 case SHMEM_HUGE_WITHIN_SIZE:
532 return "within_size";
533 case SHMEM_HUGE_ADVISE:
535 case SHMEM_HUGE_DENY:
537 case SHMEM_HUGE_FORCE:
546 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
547 struct shrink_control *sc, unsigned long nr_to_split)
549 LIST_HEAD(list), *pos, *next;
550 LIST_HEAD(to_remove);
552 struct shmem_inode_info *info;
554 unsigned long batch = sc ? sc->nr_to_scan : 128;
557 if (list_empty(&sbinfo->shrinklist))
560 spin_lock(&sbinfo->shrinklist_lock);
561 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
562 info = list_entry(pos, struct shmem_inode_info, shrinklist);
565 inode = igrab(&info->vfs_inode);
567 /* inode is about to be evicted */
569 list_del_init(&info->shrinklist);
573 /* Check if there's anything to gain */
574 if (round_up(inode->i_size, PAGE_SIZE) ==
575 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
576 list_move(&info->shrinklist, &to_remove);
580 list_move(&info->shrinklist, &list);
582 sbinfo->shrinklist_len--;
586 spin_unlock(&sbinfo->shrinklist_lock);
588 list_for_each_safe(pos, next, &to_remove) {
589 info = list_entry(pos, struct shmem_inode_info, shrinklist);
590 inode = &info->vfs_inode;
591 list_del_init(&info->shrinklist);
595 list_for_each_safe(pos, next, &list) {
599 info = list_entry(pos, struct shmem_inode_info, shrinklist);
600 inode = &info->vfs_inode;
602 if (nr_to_split && split >= nr_to_split)
605 index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
606 folio = filemap_get_folio(inode->i_mapping, index);
610 /* No huge page at the end of the file: nothing to split */
611 if (!folio_test_large(folio)) {
617 * Move the inode on the list back to shrinklist if we failed
618 * to lock the page at this time.
620 * Waiting for the lock may lead to deadlock in the
623 if (!folio_trylock(folio)) {
628 ret = split_folio(folio);
632 /* If split failed move the inode on the list back to shrinklist */
638 list_del_init(&info->shrinklist);
642 * Make sure the inode is either on the global list or deleted
643 * from any local list before iput() since it could be deleted
644 * in another thread once we put the inode (then the local list
647 spin_lock(&sbinfo->shrinklist_lock);
648 list_move(&info->shrinklist, &sbinfo->shrinklist);
649 sbinfo->shrinklist_len++;
650 spin_unlock(&sbinfo->shrinklist_lock);
658 static long shmem_unused_huge_scan(struct super_block *sb,
659 struct shrink_control *sc)
661 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
663 if (!READ_ONCE(sbinfo->shrinklist_len))
666 return shmem_unused_huge_shrink(sbinfo, sc, 0);
669 static long shmem_unused_huge_count(struct super_block *sb,
670 struct shrink_control *sc)
672 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
673 return READ_ONCE(sbinfo->shrinklist_len);
675 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
677 #define shmem_huge SHMEM_HUGE_DENY
679 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
680 struct mm_struct *mm, unsigned long vm_flags)
685 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
686 struct shrink_control *sc, unsigned long nr_to_split)
690 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
693 * Like filemap_add_folio, but error if expected item has gone.
695 static int shmem_add_to_page_cache(struct folio *folio,
696 struct address_space *mapping,
697 pgoff_t index, void *expected, gfp_t gfp,
698 struct mm_struct *charge_mm)
700 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
701 long nr = folio_nr_pages(folio);
704 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
705 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
706 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
707 VM_BUG_ON(expected && folio_test_large(folio));
709 folio_ref_add(folio, nr);
710 folio->mapping = mapping;
711 folio->index = index;
713 if (!folio_test_swapcache(folio)) {
714 error = mem_cgroup_charge(folio, charge_mm, gfp);
716 if (folio_test_pmd_mappable(folio)) {
717 count_vm_event(THP_FILE_FALLBACK);
718 count_vm_event(THP_FILE_FALLBACK_CHARGE);
723 folio_throttle_swaprate(folio, gfp);
727 if (expected != xas_find_conflict(&xas)) {
728 xas_set_err(&xas, -EEXIST);
731 if (expected && xas_find_conflict(&xas)) {
732 xas_set_err(&xas, -EEXIST);
735 xas_store(&xas, folio);
738 if (folio_test_pmd_mappable(folio)) {
739 count_vm_event(THP_FILE_ALLOC);
740 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
742 mapping->nrpages += nr;
743 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
744 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
746 xas_unlock_irq(&xas);
747 } while (xas_nomem(&xas, gfp));
749 if (xas_error(&xas)) {
750 error = xas_error(&xas);
756 folio->mapping = NULL;
757 folio_ref_sub(folio, nr);
762 * Like delete_from_page_cache, but substitutes swap for @folio.
764 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
766 struct address_space *mapping = folio->mapping;
767 long nr = folio_nr_pages(folio);
770 xa_lock_irq(&mapping->i_pages);
771 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
772 folio->mapping = NULL;
773 mapping->nrpages -= nr;
774 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
775 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
776 xa_unlock_irq(&mapping->i_pages);
782 * Remove swap entry from page cache, free the swap and its page cache.
784 static int shmem_free_swap(struct address_space *mapping,
785 pgoff_t index, void *radswap)
789 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
792 free_swap_and_cache(radix_to_swp_entry(radswap));
797 * Determine (in bytes) how many of the shmem object's pages mapped by the
798 * given offsets are swapped out.
800 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
801 * as long as the inode doesn't go away and racy results are not a problem.
803 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
804 pgoff_t start, pgoff_t end)
806 XA_STATE(xas, &mapping->i_pages, start);
808 unsigned long swapped = 0;
809 unsigned long max = end - 1;
812 xas_for_each(&xas, page, max) {
813 if (xas_retry(&xas, page))
815 if (xa_is_value(page))
817 if (xas.xa_index == max)
819 if (need_resched()) {
827 return swapped << PAGE_SHIFT;
831 * Determine (in bytes) how many of the shmem object's pages mapped by the
832 * given vma is swapped out.
834 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
835 * as long as the inode doesn't go away and racy results are not a problem.
837 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
839 struct inode *inode = file_inode(vma->vm_file);
840 struct shmem_inode_info *info = SHMEM_I(inode);
841 struct address_space *mapping = inode->i_mapping;
842 unsigned long swapped;
844 /* Be careful as we don't hold info->lock */
845 swapped = READ_ONCE(info->swapped);
848 * The easier cases are when the shmem object has nothing in swap, or
849 * the vma maps it whole. Then we can simply use the stats that we
855 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
856 return swapped << PAGE_SHIFT;
858 /* Here comes the more involved part */
859 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
860 vma->vm_pgoff + vma_pages(vma));
864 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
866 void shmem_unlock_mapping(struct address_space *mapping)
868 struct folio_batch fbatch;
871 folio_batch_init(&fbatch);
873 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
875 while (!mapping_unevictable(mapping) &&
876 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
877 check_move_unevictable_folios(&fbatch);
878 folio_batch_release(&fbatch);
883 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
888 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
889 * beyond i_size, and reports fallocated folios as holes.
891 folio = filemap_get_entry(inode->i_mapping, index);
894 if (!xa_is_value(folio)) {
896 if (folio->mapping == inode->i_mapping)
898 /* The folio has been swapped out */
903 * But read a folio back from swap if any of it is within i_size
904 * (although in some cases this is just a waste of time).
907 shmem_get_folio(inode, index, &folio, SGP_READ);
912 * Remove range of pages and swap entries from page cache, and free them.
913 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
915 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
918 struct address_space *mapping = inode->i_mapping;
919 struct shmem_inode_info *info = SHMEM_I(inode);
920 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
921 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
922 struct folio_batch fbatch;
923 pgoff_t indices[PAGEVEC_SIZE];
926 long nr_swaps_freed = 0;
931 end = -1; /* unsigned, so actually very big */
933 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
934 info->fallocend = start;
936 folio_batch_init(&fbatch);
938 while (index < end && find_lock_entries(mapping, &index, end - 1,
940 for (i = 0; i < folio_batch_count(&fbatch); i++) {
941 folio = fbatch.folios[i];
943 if (xa_is_value(folio)) {
946 nr_swaps_freed += !shmem_free_swap(mapping,
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 * When undoing a failed fallocate, we want none of the partial folio
962 * zeroing and splitting below, but shall want to truncate the whole
963 * folio when !uptodate indicates that it was added by this fallocate,
964 * even when [lstart, lend] covers only a part of the folio.
969 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
970 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
972 same_folio = lend < folio_pos(folio) + folio_size(folio);
973 folio_mark_dirty(folio);
974 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
975 start = folio->index + folio_nr_pages(folio);
985 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
987 folio_mark_dirty(folio);
988 if (!truncate_inode_partial_folio(folio, lstart, lend))
997 while (index < end) {
1000 if (!find_get_entries(mapping, &index, end - 1, &fbatch,
1002 /* If all gone or hole-punch or unfalloc, we're done */
1003 if (index == start || end != -1)
1005 /* But if truncating, restart to make sure all gone */
1009 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1010 folio = fbatch.folios[i];
1012 if (xa_is_value(folio)) {
1015 if (shmem_free_swap(mapping, indices[i], folio)) {
1016 /* Swap was replaced by page: retry */
1026 if (!unfalloc || !folio_test_uptodate(folio)) {
1027 if (folio_mapping(folio) != mapping) {
1028 /* Page was replaced by swap: retry */
1029 folio_unlock(folio);
1033 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1035 truncate_inode_folio(mapping, folio);
1037 folio_unlock(folio);
1039 folio_batch_remove_exceptionals(&fbatch);
1040 folio_batch_release(&fbatch);
1043 spin_lock_irq(&info->lock);
1044 info->swapped -= nr_swaps_freed;
1045 shmem_recalc_inode(inode);
1046 spin_unlock_irq(&info->lock);
1049 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1051 shmem_undo_range(inode, lstart, lend, false);
1052 inode->i_ctime = inode->i_mtime = current_time(inode);
1053 inode_inc_iversion(inode);
1055 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1057 static int shmem_getattr(struct mnt_idmap *idmap,
1058 const struct path *path, struct kstat *stat,
1059 u32 request_mask, unsigned int query_flags)
1061 struct inode *inode = path->dentry->d_inode;
1062 struct shmem_inode_info *info = SHMEM_I(inode);
1064 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1065 spin_lock_irq(&info->lock);
1066 shmem_recalc_inode(inode);
1067 spin_unlock_irq(&info->lock);
1069 if (info->fsflags & FS_APPEND_FL)
1070 stat->attributes |= STATX_ATTR_APPEND;
1071 if (info->fsflags & FS_IMMUTABLE_FL)
1072 stat->attributes |= STATX_ATTR_IMMUTABLE;
1073 if (info->fsflags & FS_NODUMP_FL)
1074 stat->attributes |= STATX_ATTR_NODUMP;
1075 stat->attributes_mask |= (STATX_ATTR_APPEND |
1076 STATX_ATTR_IMMUTABLE |
1078 generic_fillattr(idmap, inode, stat);
1080 if (shmem_is_huge(inode, 0, false, NULL, 0))
1081 stat->blksize = HPAGE_PMD_SIZE;
1083 if (request_mask & STATX_BTIME) {
1084 stat->result_mask |= STATX_BTIME;
1085 stat->btime.tv_sec = info->i_crtime.tv_sec;
1086 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1092 static int shmem_setattr(struct mnt_idmap *idmap,
1093 struct dentry *dentry, struct iattr *attr)
1095 struct inode *inode = d_inode(dentry);
1096 struct shmem_inode_info *info = SHMEM_I(inode);
1098 bool update_mtime = false;
1099 bool update_ctime = true;
1101 error = setattr_prepare(idmap, dentry, attr);
1105 if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1106 if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1111 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1112 loff_t oldsize = inode->i_size;
1113 loff_t newsize = attr->ia_size;
1115 /* protected by i_rwsem */
1116 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1117 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1120 if (newsize != oldsize) {
1121 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1125 i_size_write(inode, newsize);
1126 update_mtime = true;
1128 update_ctime = false;
1130 if (newsize <= oldsize) {
1131 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1132 if (oldsize > holebegin)
1133 unmap_mapping_range(inode->i_mapping,
1136 shmem_truncate_range(inode,
1137 newsize, (loff_t)-1);
1138 /* unmap again to remove racily COWed private pages */
1139 if (oldsize > holebegin)
1140 unmap_mapping_range(inode->i_mapping,
1145 setattr_copy(idmap, inode, attr);
1146 if (attr->ia_valid & ATTR_MODE)
1147 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1148 if (!error && update_ctime) {
1149 inode->i_ctime = current_time(inode);
1151 inode->i_mtime = inode->i_ctime;
1152 inode_inc_iversion(inode);
1157 static void shmem_evict_inode(struct inode *inode)
1159 struct shmem_inode_info *info = SHMEM_I(inode);
1160 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1162 if (shmem_mapping(inode->i_mapping)) {
1163 shmem_unacct_size(info->flags, inode->i_size);
1165 mapping_set_exiting(inode->i_mapping);
1166 shmem_truncate_range(inode, 0, (loff_t)-1);
1167 if (!list_empty(&info->shrinklist)) {
1168 spin_lock(&sbinfo->shrinklist_lock);
1169 if (!list_empty(&info->shrinklist)) {
1170 list_del_init(&info->shrinklist);
1171 sbinfo->shrinklist_len--;
1173 spin_unlock(&sbinfo->shrinklist_lock);
1175 while (!list_empty(&info->swaplist)) {
1176 /* Wait while shmem_unuse() is scanning this inode... */
1177 wait_var_event(&info->stop_eviction,
1178 !atomic_read(&info->stop_eviction));
1179 mutex_lock(&shmem_swaplist_mutex);
1180 /* ...but beware of the race if we peeked too early */
1181 if (!atomic_read(&info->stop_eviction))
1182 list_del_init(&info->swaplist);
1183 mutex_unlock(&shmem_swaplist_mutex);
1187 simple_xattrs_free(&info->xattrs);
1188 WARN_ON(inode->i_blocks);
1189 shmem_free_inode(inode->i_sb);
1193 static int shmem_find_swap_entries(struct address_space *mapping,
1194 pgoff_t start, struct folio_batch *fbatch,
1195 pgoff_t *indices, unsigned int type)
1197 XA_STATE(xas, &mapping->i_pages, start);
1198 struct folio *folio;
1202 xas_for_each(&xas, folio, ULONG_MAX) {
1203 if (xas_retry(&xas, folio))
1206 if (!xa_is_value(folio))
1209 entry = radix_to_swp_entry(folio);
1211 * swapin error entries can be found in the mapping. But they're
1212 * deliberately ignored here as we've done everything we can do.
1214 if (swp_type(entry) != type)
1217 indices[folio_batch_count(fbatch)] = xas.xa_index;
1218 if (!folio_batch_add(fbatch, folio))
1221 if (need_resched()) {
1228 return xas.xa_index;
1232 * Move the swapped pages for an inode to page cache. Returns the count
1233 * of pages swapped in, or the error in case of failure.
1235 static int shmem_unuse_swap_entries(struct inode *inode,
1236 struct folio_batch *fbatch, pgoff_t *indices)
1241 struct address_space *mapping = inode->i_mapping;
1243 for (i = 0; i < folio_batch_count(fbatch); i++) {
1244 struct folio *folio = fbatch->folios[i];
1246 if (!xa_is_value(folio))
1248 error = shmem_swapin_folio(inode, indices[i],
1250 mapping_gfp_mask(mapping),
1253 folio_unlock(folio);
1257 if (error == -ENOMEM)
1261 return error ? error : ret;
1265 * If swap found in inode, free it and move page from swapcache to filecache.
1267 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1269 struct address_space *mapping = inode->i_mapping;
1271 struct folio_batch fbatch;
1272 pgoff_t indices[PAGEVEC_SIZE];
1276 folio_batch_init(&fbatch);
1277 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1278 if (folio_batch_count(&fbatch) == 0) {
1283 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1287 start = indices[folio_batch_count(&fbatch) - 1];
1294 * Read all the shared memory data that resides in the swap
1295 * device 'type' back into memory, so the swap device can be
1298 int shmem_unuse(unsigned int type)
1300 struct shmem_inode_info *info, *next;
1303 if (list_empty(&shmem_swaplist))
1306 mutex_lock(&shmem_swaplist_mutex);
1307 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1308 if (!info->swapped) {
1309 list_del_init(&info->swaplist);
1313 * Drop the swaplist mutex while searching the inode for swap;
1314 * but before doing so, make sure shmem_evict_inode() will not
1315 * remove placeholder inode from swaplist, nor let it be freed
1316 * (igrab() would protect from unlink, but not from unmount).
1318 atomic_inc(&info->stop_eviction);
1319 mutex_unlock(&shmem_swaplist_mutex);
1321 error = shmem_unuse_inode(&info->vfs_inode, type);
1324 mutex_lock(&shmem_swaplist_mutex);
1325 next = list_next_entry(info, swaplist);
1327 list_del_init(&info->swaplist);
1328 if (atomic_dec_and_test(&info->stop_eviction))
1329 wake_up_var(&info->stop_eviction);
1333 mutex_unlock(&shmem_swaplist_mutex);
1339 * Move the page from the page cache to the swap cache.
1341 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1343 struct folio *folio = page_folio(page);
1344 struct address_space *mapping = folio->mapping;
1345 struct inode *inode = mapping->host;
1346 struct shmem_inode_info *info = SHMEM_I(inode);
1347 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1352 * Our capabilities prevent regular writeback or sync from ever calling
1353 * shmem_writepage; but a stacking filesystem might use ->writepage of
1354 * its underlying filesystem, in which case tmpfs should write out to
1355 * swap only in response to memory pressure, and not for the writeback
1358 if (WARN_ON_ONCE(!wbc->for_reclaim))
1361 if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
1364 if (!total_swap_pages)
1368 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1369 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1370 * and its shmem_writeback() needs them to be split when swapping.
1372 if (folio_test_large(folio)) {
1373 /* Ensure the subpages are still dirty */
1374 folio_test_set_dirty(folio);
1375 if (split_huge_page(page) < 0)
1377 folio = page_folio(page);
1378 folio_clear_dirty(folio);
1381 index = folio->index;
1384 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1385 * value into swapfile.c, the only way we can correctly account for a
1386 * fallocated folio arriving here is now to initialize it and write it.
1388 * That's okay for a folio already fallocated earlier, but if we have
1389 * not yet completed the fallocation, then (a) we want to keep track
1390 * of this folio in case we have to undo it, and (b) it may not be a
1391 * good idea to continue anyway, once we're pushing into swap. So
1392 * reactivate the folio, and let shmem_fallocate() quit when too many.
1394 if (!folio_test_uptodate(folio)) {
1395 if (inode->i_private) {
1396 struct shmem_falloc *shmem_falloc;
1397 spin_lock(&inode->i_lock);
1398 shmem_falloc = inode->i_private;
1400 !shmem_falloc->waitq &&
1401 index >= shmem_falloc->start &&
1402 index < shmem_falloc->next)
1403 shmem_falloc->nr_unswapped++;
1405 shmem_falloc = NULL;
1406 spin_unlock(&inode->i_lock);
1410 folio_zero_range(folio, 0, folio_size(folio));
1411 flush_dcache_folio(folio);
1412 folio_mark_uptodate(folio);
1415 swap = folio_alloc_swap(folio);
1420 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1421 * if it's not already there. Do it now before the folio is
1422 * moved to swap cache, when its pagelock no longer protects
1423 * the inode from eviction. But don't unlock the mutex until
1424 * we've incremented swapped, because shmem_unuse_inode() will
1425 * prune a !swapped inode from the swaplist under this mutex.
1427 mutex_lock(&shmem_swaplist_mutex);
1428 if (list_empty(&info->swaplist))
1429 list_add(&info->swaplist, &shmem_swaplist);
1431 if (add_to_swap_cache(folio, swap,
1432 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1434 spin_lock_irq(&info->lock);
1435 shmem_recalc_inode(inode);
1437 spin_unlock_irq(&info->lock);
1439 swap_shmem_alloc(swap);
1440 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1442 mutex_unlock(&shmem_swaplist_mutex);
1443 BUG_ON(folio_mapped(folio));
1444 swap_writepage(&folio->page, wbc);
1448 mutex_unlock(&shmem_swaplist_mutex);
1449 put_swap_folio(folio, swap);
1451 folio_mark_dirty(folio);
1452 if (wbc->for_reclaim)
1453 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1454 folio_unlock(folio);
1458 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1459 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1463 if (!mpol || mpol->mode == MPOL_DEFAULT)
1464 return; /* show nothing */
1466 mpol_to_str(buffer, sizeof(buffer), mpol);
1468 seq_printf(seq, ",mpol=%s", buffer);
1471 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1473 struct mempolicy *mpol = NULL;
1475 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1476 mpol = sbinfo->mpol;
1478 raw_spin_unlock(&sbinfo->stat_lock);
1482 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1483 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1486 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1490 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1492 #define vm_policy vm_private_data
1495 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1496 struct shmem_inode_info *info, pgoff_t index)
1498 /* Create a pseudo vma that just contains the policy */
1499 vma_init(vma, NULL);
1500 /* Bias interleave by inode number to distribute better across nodes */
1501 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1502 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1505 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1507 /* Drop reference taken by mpol_shared_policy_lookup() */
1508 mpol_cond_put(vma->vm_policy);
1511 static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1512 struct shmem_inode_info *info, pgoff_t index)
1514 struct vm_area_struct pvma;
1516 struct vm_fault vmf = {
1520 shmem_pseudo_vma_init(&pvma, info, index);
1521 page = swap_cluster_readahead(swap, gfp, &vmf);
1522 shmem_pseudo_vma_destroy(&pvma);
1526 return page_folio(page);
1530 * Make sure huge_gfp is always more limited than limit_gfp.
1531 * Some of the flags set permissions, while others set limitations.
1533 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1535 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1536 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1537 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1538 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1540 /* Allow allocations only from the originally specified zones. */
1541 result |= zoneflags;
1544 * Minimize the result gfp by taking the union with the deny flags,
1545 * and the intersection of the allow flags.
1547 result |= (limit_gfp & denyflags);
1548 result |= (huge_gfp & limit_gfp) & allowflags;
1553 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1554 struct shmem_inode_info *info, pgoff_t index)
1556 struct vm_area_struct pvma;
1557 struct address_space *mapping = info->vfs_inode.i_mapping;
1559 struct folio *folio;
1561 hindex = round_down(index, HPAGE_PMD_NR);
1562 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1566 shmem_pseudo_vma_init(&pvma, info, hindex);
1567 folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1568 shmem_pseudo_vma_destroy(&pvma);
1570 count_vm_event(THP_FILE_FALLBACK);
1574 static struct folio *shmem_alloc_folio(gfp_t gfp,
1575 struct shmem_inode_info *info, pgoff_t index)
1577 struct vm_area_struct pvma;
1578 struct folio *folio;
1580 shmem_pseudo_vma_init(&pvma, info, index);
1581 folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1582 shmem_pseudo_vma_destroy(&pvma);
1587 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1588 pgoff_t index, bool huge)
1590 struct shmem_inode_info *info = SHMEM_I(inode);
1591 struct folio *folio;
1595 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1597 nr = huge ? HPAGE_PMD_NR : 1;
1599 if (!shmem_inode_acct_block(inode, nr))
1603 folio = shmem_alloc_hugefolio(gfp, info, index);
1605 folio = shmem_alloc_folio(gfp, info, index);
1607 __folio_set_locked(folio);
1608 __folio_set_swapbacked(folio);
1613 shmem_inode_unacct_blocks(inode, nr);
1615 return ERR_PTR(err);
1619 * When a page is moved from swapcache to shmem filecache (either by the
1620 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1621 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1622 * ignorance of the mapping it belongs to. If that mapping has special
1623 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1624 * we may need to copy to a suitable page before moving to filecache.
1626 * In a future release, this may well be extended to respect cpuset and
1627 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1628 * but for now it is a simple matter of zone.
1630 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1632 return folio_zonenum(folio) > gfp_zone(gfp);
1635 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1636 struct shmem_inode_info *info, pgoff_t index)
1638 struct folio *old, *new;
1639 struct address_space *swap_mapping;
1645 entry = folio_swap_entry(old);
1646 swap_index = swp_offset(entry);
1647 swap_mapping = swap_address_space(entry);
1650 * We have arrived here because our zones are constrained, so don't
1651 * limit chance of success by further cpuset and node constraints.
1653 gfp &= ~GFP_CONSTRAINT_MASK;
1654 VM_BUG_ON_FOLIO(folio_test_large(old), old);
1655 new = shmem_alloc_folio(gfp, info, index);
1660 folio_copy(new, old);
1661 flush_dcache_folio(new);
1663 __folio_set_locked(new);
1664 __folio_set_swapbacked(new);
1665 folio_mark_uptodate(new);
1666 folio_set_swap_entry(new, entry);
1667 folio_set_swapcache(new);
1670 * Our caller will very soon move newpage out of swapcache, but it's
1671 * a nice clean interface for us to replace oldpage by newpage there.
1673 xa_lock_irq(&swap_mapping->i_pages);
1674 error = shmem_replace_entry(swap_mapping, swap_index, old, new);
1676 mem_cgroup_migrate(old, new);
1677 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1678 __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1679 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1680 __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
1682 xa_unlock_irq(&swap_mapping->i_pages);
1684 if (unlikely(error)) {
1686 * Is this possible? I think not, now that our callers check
1687 * both PageSwapCache and page_private after getting page lock;
1688 * but be defensive. Reverse old to newpage for clear and free.
1696 folio_clear_swapcache(old);
1697 old->private = NULL;
1700 folio_put_refs(old, 2);
1704 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1705 struct folio *folio, swp_entry_t swap)
1707 struct address_space *mapping = inode->i_mapping;
1708 struct shmem_inode_info *info = SHMEM_I(inode);
1709 swp_entry_t swapin_error;
1712 swapin_error = make_swapin_error_entry();
1713 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1714 swp_to_radix_entry(swap),
1715 swp_to_radix_entry(swapin_error), 0);
1716 if (old != swp_to_radix_entry(swap))
1719 folio_wait_writeback(folio);
1720 delete_from_swap_cache(folio);
1721 spin_lock_irq(&info->lock);
1723 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks won't
1724 * be 0 when inode is released and thus trigger WARN_ON(inode->i_blocks) in
1725 * shmem_evict_inode.
1729 shmem_recalc_inode(inode);
1730 spin_unlock_irq(&info->lock);
1735 * Swap in the folio pointed to by *foliop.
1736 * Caller has to make sure that *foliop contains a valid swapped folio.
1737 * Returns 0 and the folio in foliop if success. On failure, returns the
1738 * error code and NULL in *foliop.
1740 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1741 struct folio **foliop, enum sgp_type sgp,
1742 gfp_t gfp, struct vm_area_struct *vma,
1743 vm_fault_t *fault_type)
1745 struct address_space *mapping = inode->i_mapping;
1746 struct shmem_inode_info *info = SHMEM_I(inode);
1747 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1748 struct swap_info_struct *si;
1749 struct folio *folio = NULL;
1753 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1754 swap = radix_to_swp_entry(*foliop);
1757 if (is_swapin_error_entry(swap))
1760 si = get_swap_device(swap);
1762 if (!shmem_confirm_swap(mapping, index, swap))
1768 /* Look it up and read it in.. */
1769 folio = swap_cache_get_folio(swap, NULL, 0);
1771 /* Or update major stats only when swapin succeeds?? */
1773 *fault_type |= VM_FAULT_MAJOR;
1774 count_vm_event(PGMAJFAULT);
1775 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1777 /* Here we actually start the io */
1778 folio = shmem_swapin(swap, gfp, info, index);
1785 /* We have to do this with folio locked to prevent races */
1787 if (!folio_test_swapcache(folio) ||
1788 folio_swap_entry(folio).val != swap.val ||
1789 !shmem_confirm_swap(mapping, index, swap)) {
1793 if (!folio_test_uptodate(folio)) {
1797 folio_wait_writeback(folio);
1800 * Some architectures may have to restore extra metadata to the
1801 * folio after reading from swap.
1803 arch_swap_restore(swap, folio);
1805 if (shmem_should_replace_folio(folio, gfp)) {
1806 error = shmem_replace_folio(&folio, gfp, info, index);
1811 error = shmem_add_to_page_cache(folio, mapping, index,
1812 swp_to_radix_entry(swap), gfp,
1817 spin_lock_irq(&info->lock);
1819 shmem_recalc_inode(inode);
1820 spin_unlock_irq(&info->lock);
1822 if (sgp == SGP_WRITE)
1823 folio_mark_accessed(folio);
1825 delete_from_swap_cache(folio);
1826 folio_mark_dirty(folio);
1828 put_swap_device(si);
1833 if (!shmem_confirm_swap(mapping, index, swap))
1836 shmem_set_folio_swapin_error(inode, index, folio, swap);
1839 folio_unlock(folio);
1842 put_swap_device(si);
1848 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1850 * If we allocate a new one we do not mark it dirty. That's up to the
1851 * vm. If we swap it in we mark it dirty since we also free the swap
1852 * entry since a page cannot live in both the swap and page cache.
1854 * vma, vmf, and fault_type are only supplied by shmem_fault:
1855 * otherwise they are NULL.
1857 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1858 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
1859 struct vm_area_struct *vma, struct vm_fault *vmf,
1860 vm_fault_t *fault_type)
1862 struct address_space *mapping = inode->i_mapping;
1863 struct shmem_inode_info *info = SHMEM_I(inode);
1864 struct shmem_sb_info *sbinfo;
1865 struct mm_struct *charge_mm;
1866 struct folio *folio;
1873 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1876 if (sgp <= SGP_CACHE &&
1877 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1881 sbinfo = SHMEM_SB(inode->i_sb);
1882 charge_mm = vma ? vma->vm_mm : NULL;
1884 folio = filemap_get_entry(mapping, index);
1885 if (folio && vma && userfaultfd_minor(vma)) {
1886 if (!xa_is_value(folio))
1888 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1892 if (xa_is_value(folio)) {
1893 error = shmem_swapin_folio(inode, index, &folio,
1894 sgp, gfp, vma, fault_type);
1895 if (error == -EEXIST)
1905 /* Has the folio been truncated or swapped out? */
1906 if (unlikely(folio->mapping != mapping)) {
1907 folio_unlock(folio);
1911 if (sgp == SGP_WRITE)
1912 folio_mark_accessed(folio);
1913 if (folio_test_uptodate(folio))
1915 /* fallocated folio */
1916 if (sgp != SGP_READ)
1918 folio_unlock(folio);
1923 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1924 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1927 if (sgp == SGP_READ)
1929 if (sgp == SGP_NOALLOC)
1933 * Fast cache lookup and swap lookup did not find it: allocate.
1936 if (vma && userfaultfd_missing(vma)) {
1937 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1941 if (!shmem_is_huge(inode, index, false,
1942 vma ? vma->vm_mm : NULL, vma ? vma->vm_flags : 0))
1945 huge_gfp = vma_thp_gfp_mask(vma);
1946 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1947 folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
1948 if (IS_ERR(folio)) {
1950 folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
1952 if (IS_ERR(folio)) {
1955 error = PTR_ERR(folio);
1957 if (error != -ENOSPC)
1960 * Try to reclaim some space by splitting a large folio
1961 * beyond i_size on the filesystem.
1966 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1967 if (ret == SHRINK_STOP)
1975 hindex = round_down(index, folio_nr_pages(folio));
1977 if (sgp == SGP_WRITE)
1978 __folio_set_referenced(folio);
1980 error = shmem_add_to_page_cache(folio, mapping, hindex,
1981 NULL, gfp & GFP_RECLAIM_MASK,
1985 folio_add_lru(folio);
1987 spin_lock_irq(&info->lock);
1988 info->alloced += folio_nr_pages(folio);
1989 inode->i_blocks += (blkcnt_t)BLOCKS_PER_PAGE << folio_order(folio);
1990 shmem_recalc_inode(inode);
1991 spin_unlock_irq(&info->lock);
1994 if (folio_test_pmd_mappable(folio) &&
1995 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1996 folio_next_index(folio) - 1) {
1998 * Part of the large folio is beyond i_size: subject
1999 * to shrink under memory pressure.
2001 spin_lock(&sbinfo->shrinklist_lock);
2003 * _careful to defend against unlocked access to
2004 * ->shrink_list in shmem_unused_huge_shrink()
2006 if (list_empty_careful(&info->shrinklist)) {
2007 list_add_tail(&info->shrinklist,
2008 &sbinfo->shrinklist);
2009 sbinfo->shrinklist_len++;
2011 spin_unlock(&sbinfo->shrinklist_lock);
2015 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2017 if (sgp == SGP_FALLOC)
2021 * Let SGP_WRITE caller clear ends if write does not fill folio;
2022 * but SGP_FALLOC on a folio fallocated earlier must initialize
2023 * it now, lest undo on failure cancel our earlier guarantee.
2025 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2026 long i, n = folio_nr_pages(folio);
2028 for (i = 0; i < n; i++)
2029 clear_highpage(folio_page(folio, i));
2030 flush_dcache_folio(folio);
2031 folio_mark_uptodate(folio);
2034 /* Perhaps the file has been truncated since we checked */
2035 if (sgp <= SGP_CACHE &&
2036 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2038 folio_clear_dirty(folio);
2039 filemap_remove_folio(folio);
2040 spin_lock_irq(&info->lock);
2041 shmem_recalc_inode(inode);
2042 spin_unlock_irq(&info->lock);
2055 shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
2057 if (folio_test_large(folio)) {
2058 folio_unlock(folio);
2064 folio_unlock(folio);
2067 if (error == -ENOSPC && !once++) {
2068 spin_lock_irq(&info->lock);
2069 shmem_recalc_inode(inode);
2070 spin_unlock_irq(&info->lock);
2073 if (error == -EEXIST)
2078 int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
2081 return shmem_get_folio_gfp(inode, index, foliop, sgp,
2082 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
2086 * This is like autoremove_wake_function, but it removes the wait queue
2087 * entry unconditionally - even if something else had already woken the
2090 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2092 int ret = default_wake_function(wait, mode, sync, key);
2093 list_del_init(&wait->entry);
2097 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2099 struct vm_area_struct *vma = vmf->vma;
2100 struct inode *inode = file_inode(vma->vm_file);
2101 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2102 struct folio *folio = NULL;
2104 vm_fault_t ret = VM_FAULT_LOCKED;
2107 * Trinity finds that probing a hole which tmpfs is punching can
2108 * prevent the hole-punch from ever completing: which in turn
2109 * locks writers out with its hold on i_rwsem. So refrain from
2110 * faulting pages into the hole while it's being punched. Although
2111 * shmem_undo_range() does remove the additions, it may be unable to
2112 * keep up, as each new page needs its own unmap_mapping_range() call,
2113 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2115 * It does not matter if we sometimes reach this check just before the
2116 * hole-punch begins, so that one fault then races with the punch:
2117 * we just need to make racing faults a rare case.
2119 * The implementation below would be much simpler if we just used a
2120 * standard mutex or completion: but we cannot take i_rwsem in fault,
2121 * and bloating every shmem inode for this unlikely case would be sad.
2123 if (unlikely(inode->i_private)) {
2124 struct shmem_falloc *shmem_falloc;
2126 spin_lock(&inode->i_lock);
2127 shmem_falloc = inode->i_private;
2129 shmem_falloc->waitq &&
2130 vmf->pgoff >= shmem_falloc->start &&
2131 vmf->pgoff < shmem_falloc->next) {
2133 wait_queue_head_t *shmem_falloc_waitq;
2134 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2136 ret = VM_FAULT_NOPAGE;
2137 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2139 ret = VM_FAULT_RETRY;
2141 shmem_falloc_waitq = shmem_falloc->waitq;
2142 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2143 TASK_UNINTERRUPTIBLE);
2144 spin_unlock(&inode->i_lock);
2148 * shmem_falloc_waitq points into the shmem_fallocate()
2149 * stack of the hole-punching task: shmem_falloc_waitq
2150 * is usually invalid by the time we reach here, but
2151 * finish_wait() does not dereference it in that case;
2152 * though i_lock needed lest racing with wake_up_all().
2154 spin_lock(&inode->i_lock);
2155 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2156 spin_unlock(&inode->i_lock);
2162 spin_unlock(&inode->i_lock);
2165 err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
2166 gfp, vma, vmf, &ret);
2168 return vmf_error(err);
2170 vmf->page = folio_file_page(folio, vmf->pgoff);
2174 unsigned long shmem_get_unmapped_area(struct file *file,
2175 unsigned long uaddr, unsigned long len,
2176 unsigned long pgoff, unsigned long flags)
2178 unsigned long (*get_area)(struct file *,
2179 unsigned long, unsigned long, unsigned long, unsigned long);
2181 unsigned long offset;
2182 unsigned long inflated_len;
2183 unsigned long inflated_addr;
2184 unsigned long inflated_offset;
2186 if (len > TASK_SIZE)
2189 get_area = current->mm->get_unmapped_area;
2190 addr = get_area(file, uaddr, len, pgoff, flags);
2192 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2194 if (IS_ERR_VALUE(addr))
2196 if (addr & ~PAGE_MASK)
2198 if (addr > TASK_SIZE - len)
2201 if (shmem_huge == SHMEM_HUGE_DENY)
2203 if (len < HPAGE_PMD_SIZE)
2205 if (flags & MAP_FIXED)
2208 * Our priority is to support MAP_SHARED mapped hugely;
2209 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2210 * But if caller specified an address hint and we allocated area there
2211 * successfully, respect that as before.
2216 if (shmem_huge != SHMEM_HUGE_FORCE) {
2217 struct super_block *sb;
2220 VM_BUG_ON(file->f_op != &shmem_file_operations);
2221 sb = file_inode(file)->i_sb;
2224 * Called directly from mm/mmap.c, or drivers/char/mem.c
2225 * for "/dev/zero", to create a shared anonymous object.
2227 if (IS_ERR(shm_mnt))
2229 sb = shm_mnt->mnt_sb;
2231 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2235 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2236 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2238 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2241 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2242 if (inflated_len > TASK_SIZE)
2244 if (inflated_len < len)
2247 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2248 if (IS_ERR_VALUE(inflated_addr))
2250 if (inflated_addr & ~PAGE_MASK)
2253 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2254 inflated_addr += offset - inflated_offset;
2255 if (inflated_offset > offset)
2256 inflated_addr += HPAGE_PMD_SIZE;
2258 if (inflated_addr > TASK_SIZE - len)
2260 return inflated_addr;
2264 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2266 struct inode *inode = file_inode(vma->vm_file);
2267 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2270 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2273 struct inode *inode = file_inode(vma->vm_file);
2276 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2277 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2281 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2283 struct inode *inode = file_inode(file);
2284 struct shmem_inode_info *info = SHMEM_I(inode);
2285 int retval = -ENOMEM;
2288 * What serializes the accesses to info->flags?
2289 * ipc_lock_object() when called from shmctl_do_lock(),
2290 * no serialization needed when called from shm_destroy().
2292 if (lock && !(info->flags & VM_LOCKED)) {
2293 if (!user_shm_lock(inode->i_size, ucounts))
2295 info->flags |= VM_LOCKED;
2296 mapping_set_unevictable(file->f_mapping);
2298 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2299 user_shm_unlock(inode->i_size, ucounts);
2300 info->flags &= ~VM_LOCKED;
2301 mapping_clear_unevictable(file->f_mapping);
2309 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2311 struct inode *inode = file_inode(file);
2312 struct shmem_inode_info *info = SHMEM_I(inode);
2315 ret = seal_check_future_write(info->seals, vma);
2319 /* arm64 - allow memory tagging on RAM-based files */
2320 vm_flags_set(vma, VM_MTE_ALLOWED);
2322 file_accessed(file);
2323 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2325 vma->vm_ops = &shmem_vm_ops;
2327 vma->vm_ops = &shmem_anon_vm_ops;
2331 #ifdef CONFIG_TMPFS_XATTR
2332 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2335 * chattr's fsflags are unrelated to extended attributes,
2336 * but tmpfs has chosen to enable them under the same config option.
2338 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2340 unsigned int i_flags = 0;
2342 if (fsflags & FS_NOATIME_FL)
2343 i_flags |= S_NOATIME;
2344 if (fsflags & FS_APPEND_FL)
2345 i_flags |= S_APPEND;
2346 if (fsflags & FS_IMMUTABLE_FL)
2347 i_flags |= S_IMMUTABLE;
2349 * But FS_NODUMP_FL does not require any action in i_flags.
2351 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2354 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2357 #define shmem_initxattrs NULL
2360 static struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block *sb,
2361 struct inode *dir, umode_t mode, dev_t dev,
2362 unsigned long flags)
2364 struct inode *inode;
2365 struct shmem_inode_info *info;
2366 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2369 if (shmem_reserve_inode(sb, &ino))
2372 inode = new_inode(sb);
2375 inode_init_owner(idmap, inode, dir, mode);
2376 inode->i_blocks = 0;
2377 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2378 inode->i_generation = get_random_u32();
2379 info = SHMEM_I(inode);
2380 memset(info, 0, (char *)inode - (char *)info);
2381 spin_lock_init(&info->lock);
2382 atomic_set(&info->stop_eviction, 0);
2383 info->seals = F_SEAL_SEAL;
2384 info->flags = flags & VM_NORESERVE;
2385 info->i_crtime = inode->i_mtime;
2386 info->fsflags = (dir == NULL) ? 0 :
2387 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2389 shmem_set_inode_flags(inode, info->fsflags);
2390 INIT_LIST_HEAD(&info->shrinklist);
2391 INIT_LIST_HEAD(&info->swaplist);
2393 mapping_set_unevictable(inode->i_mapping);
2394 simple_xattrs_init(&info->xattrs);
2395 cache_no_acl(inode);
2396 mapping_set_large_folios(inode->i_mapping);
2398 switch (mode & S_IFMT) {
2400 inode->i_op = &shmem_special_inode_operations;
2401 init_special_inode(inode, mode, dev);
2404 inode->i_mapping->a_ops = &shmem_aops;
2405 inode->i_op = &shmem_inode_operations;
2406 inode->i_fop = &shmem_file_operations;
2407 mpol_shared_policy_init(&info->policy,
2408 shmem_get_sbmpol(sbinfo));
2412 /* Some things misbehave if size == 0 on a directory */
2413 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2414 inode->i_op = &shmem_dir_inode_operations;
2415 inode->i_fop = &simple_dir_operations;
2419 * Must not load anything in the rbtree,
2420 * mpol_free_shared_policy will not be called.
2422 mpol_shared_policy_init(&info->policy, NULL);
2426 lockdep_annotate_inode_mutex_key(inode);
2428 shmem_free_inode(sb);
2432 #ifdef CONFIG_USERFAULTFD
2433 int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
2434 struct vm_area_struct *dst_vma,
2435 unsigned long dst_addr,
2436 unsigned long src_addr,
2438 struct folio **foliop)
2440 struct inode *inode = file_inode(dst_vma->vm_file);
2441 struct shmem_inode_info *info = SHMEM_I(inode);
2442 struct address_space *mapping = inode->i_mapping;
2443 gfp_t gfp = mapping_gfp_mask(mapping);
2444 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2446 struct folio *folio;
2450 if (!shmem_inode_acct_block(inode, 1)) {
2452 * We may have got a page, returned -ENOENT triggering a retry,
2453 * and now we find ourselves with -ENOMEM. Release the page, to
2454 * avoid a BUG_ON in our caller.
2456 if (unlikely(*foliop)) {
2465 folio = shmem_alloc_folio(gfp, info, pgoff);
2467 goto out_unacct_blocks;
2469 if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
2470 page_kaddr = kmap_local_folio(folio, 0);
2472 * The read mmap_lock is held here. Despite the
2473 * mmap_lock being read recursive a deadlock is still
2474 * possible if a writer has taken a lock. For example:
2476 * process A thread 1 takes read lock on own mmap_lock
2477 * process A thread 2 calls mmap, blocks taking write lock
2478 * process B thread 1 takes page fault, read lock on own mmap lock
2479 * process B thread 2 calls mmap, blocks taking write lock
2480 * process A thread 1 blocks taking read lock on process B
2481 * process B thread 1 blocks taking read lock on process A
2483 * Disable page faults to prevent potential deadlock
2484 * and retry the copy outside the mmap_lock.
2486 pagefault_disable();
2487 ret = copy_from_user(page_kaddr,
2488 (const void __user *)src_addr,
2491 kunmap_local(page_kaddr);
2493 /* fallback to copy_from_user outside mmap_lock */
2494 if (unlikely(ret)) {
2497 /* don't free the page */
2498 goto out_unacct_blocks;
2501 flush_dcache_folio(folio);
2502 } else { /* ZEROPAGE */
2503 clear_user_highpage(&folio->page, dst_addr);
2507 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2511 VM_BUG_ON(folio_test_locked(folio));
2512 VM_BUG_ON(folio_test_swapbacked(folio));
2513 __folio_set_locked(folio);
2514 __folio_set_swapbacked(folio);
2515 __folio_mark_uptodate(folio);
2518 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2519 if (unlikely(pgoff >= max_off))
2522 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2523 gfp & GFP_RECLAIM_MASK, dst_vma->vm_mm);
2527 ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
2528 &folio->page, true, flags);
2530 goto out_delete_from_cache;
2532 spin_lock_irq(&info->lock);
2534 inode->i_blocks += BLOCKS_PER_PAGE;
2535 shmem_recalc_inode(inode);
2536 spin_unlock_irq(&info->lock);
2538 folio_unlock(folio);
2540 out_delete_from_cache:
2541 filemap_remove_folio(folio);
2543 folio_unlock(folio);
2546 shmem_inode_unacct_blocks(inode, 1);
2549 #endif /* CONFIG_USERFAULTFD */
2552 static const struct inode_operations shmem_symlink_inode_operations;
2553 static const struct inode_operations shmem_short_symlink_operations;
2556 shmem_write_begin(struct file *file, struct address_space *mapping,
2557 loff_t pos, unsigned len,
2558 struct page **pagep, void **fsdata)
2560 struct inode *inode = mapping->host;
2561 struct shmem_inode_info *info = SHMEM_I(inode);
2562 pgoff_t index = pos >> PAGE_SHIFT;
2563 struct folio *folio;
2566 /* i_rwsem is held by caller */
2567 if (unlikely(info->seals & (F_SEAL_GROW |
2568 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2569 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2571 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2575 ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
2580 *pagep = folio_file_page(folio, index);
2581 if (PageHWPoison(*pagep)) {
2582 folio_unlock(folio);
2592 shmem_write_end(struct file *file, struct address_space *mapping,
2593 loff_t pos, unsigned len, unsigned copied,
2594 struct page *page, void *fsdata)
2596 struct folio *folio = page_folio(page);
2597 struct inode *inode = mapping->host;
2599 if (pos + copied > inode->i_size)
2600 i_size_write(inode, pos + copied);
2602 if (!folio_test_uptodate(folio)) {
2603 if (copied < folio_size(folio)) {
2604 size_t from = offset_in_folio(folio, pos);
2605 folio_zero_segments(folio, 0, from,
2606 from + copied, folio_size(folio));
2608 folio_mark_uptodate(folio);
2610 folio_mark_dirty(folio);
2611 folio_unlock(folio);
2617 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2619 struct file *file = iocb->ki_filp;
2620 struct inode *inode = file_inode(file);
2621 struct address_space *mapping = inode->i_mapping;
2623 unsigned long offset;
2626 loff_t *ppos = &iocb->ki_pos;
2628 index = *ppos >> PAGE_SHIFT;
2629 offset = *ppos & ~PAGE_MASK;
2632 struct folio *folio = NULL;
2633 struct page *page = NULL;
2635 unsigned long nr, ret;
2636 loff_t i_size = i_size_read(inode);
2638 end_index = i_size >> PAGE_SHIFT;
2639 if (index > end_index)
2641 if (index == end_index) {
2642 nr = i_size & ~PAGE_MASK;
2647 error = shmem_get_folio(inode, index, &folio, SGP_READ);
2649 if (error == -EINVAL)
2654 folio_unlock(folio);
2656 page = folio_file_page(folio, index);
2657 if (PageHWPoison(page)) {
2665 * We must evaluate after, since reads (unlike writes)
2666 * are called without i_rwsem protection against truncate
2669 i_size = i_size_read(inode);
2670 end_index = i_size >> PAGE_SHIFT;
2671 if (index == end_index) {
2672 nr = i_size & ~PAGE_MASK;
2683 * If users can be writing to this page using arbitrary
2684 * virtual addresses, take care about potential aliasing
2685 * before reading the page on the kernel side.
2687 if (mapping_writably_mapped(mapping))
2688 flush_dcache_page(page);
2690 * Mark the page accessed if we read the beginning.
2693 folio_mark_accessed(folio);
2695 * Ok, we have the page, and it's up-to-date, so
2696 * now we can copy it to user space...
2698 ret = copy_page_to_iter(page, offset, nr, to);
2701 } else if (user_backed_iter(to)) {
2703 * Copy to user tends to be so well optimized, but
2704 * clear_user() not so much, that it is noticeably
2705 * faster to copy the zero page instead of clearing.
2707 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2710 * But submitting the same page twice in a row to
2711 * splice() - or others? - can result in confusion:
2712 * so don't attempt that optimization on pipes etc.
2714 ret = iov_iter_zero(nr, to);
2719 index += offset >> PAGE_SHIFT;
2720 offset &= ~PAGE_MASK;
2722 if (!iov_iter_count(to))
2731 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2732 file_accessed(file);
2733 return retval ? retval : error;
2736 static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
2737 struct pipe_buffer *buf)
2742 static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
2743 struct pipe_buffer *buf)
2747 static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
2748 struct pipe_buffer *buf)
2753 static const struct pipe_buf_operations zero_pipe_buf_ops = {
2754 .release = zero_pipe_buf_release,
2755 .try_steal = zero_pipe_buf_try_steal,
2756 .get = zero_pipe_buf_get,
2759 static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
2760 loff_t fpos, size_t size)
2762 size_t offset = fpos & ~PAGE_MASK;
2764 size = min_t(size_t, size, PAGE_SIZE - offset);
2766 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2767 struct pipe_buffer *buf = pipe_head_buf(pipe);
2769 *buf = (struct pipe_buffer) {
2770 .ops = &zero_pipe_buf_ops,
2771 .page = ZERO_PAGE(0),
2781 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2782 struct pipe_inode_info *pipe,
2783 size_t len, unsigned int flags)
2785 struct inode *inode = file_inode(in);
2786 struct address_space *mapping = inode->i_mapping;
2787 struct folio *folio = NULL;
2788 size_t total_spliced = 0, used, npages, n, part;
2792 /* Work out how much data we can actually add into the pipe */
2793 used = pipe_occupancy(pipe->head, pipe->tail);
2794 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2795 len = min_t(size_t, len, npages * PAGE_SIZE);
2798 if (*ppos >= i_size_read(inode))
2801 error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio,
2804 if (error == -EINVAL)
2809 folio_unlock(folio);
2811 if (folio_test_hwpoison(folio) ||
2812 (folio_test_large(folio) &&
2813 folio_test_has_hwpoisoned(folio))) {
2820 * i_size must be checked after we know the pages are Uptodate.
2822 * Checking i_size after the check allows us to calculate
2823 * the correct value for "nr", which means the zero-filled
2824 * part of the page is not copied back to userspace (unless
2825 * another truncate extends the file - this is desired though).
2827 isize = i_size_read(inode);
2828 if (unlikely(*ppos >= isize))
2830 part = min_t(loff_t, isize - *ppos, len);
2834 * If users can be writing to this page using arbitrary
2835 * virtual addresses, take care about potential aliasing
2836 * before reading the page on the kernel side.
2838 if (mapping_writably_mapped(mapping))
2839 flush_dcache_folio(folio);
2840 folio_mark_accessed(folio);
2842 * Ok, we have the page, and it's up-to-date, so we can
2843 * now splice it into the pipe.
2845 n = splice_folio_into_pipe(pipe, folio, *ppos, part);
2849 n = splice_zeropage_into_pipe(pipe, *ppos, part);
2857 in->f_ra.prev_pos = *ppos;
2858 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2868 return total_spliced ? total_spliced : error;
2871 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2873 struct address_space *mapping = file->f_mapping;
2874 struct inode *inode = mapping->host;
2876 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2877 return generic_file_llseek_size(file, offset, whence,
2878 MAX_LFS_FILESIZE, i_size_read(inode));
2883 /* We're holding i_rwsem so we can access i_size directly */
2884 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2886 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2887 inode_unlock(inode);
2891 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2894 struct inode *inode = file_inode(file);
2895 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2896 struct shmem_inode_info *info = SHMEM_I(inode);
2897 struct shmem_falloc shmem_falloc;
2898 pgoff_t start, index, end, undo_fallocend;
2901 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2906 if (mode & FALLOC_FL_PUNCH_HOLE) {
2907 struct address_space *mapping = file->f_mapping;
2908 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2909 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2910 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2912 /* protected by i_rwsem */
2913 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2918 shmem_falloc.waitq = &shmem_falloc_waitq;
2919 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2920 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2921 spin_lock(&inode->i_lock);
2922 inode->i_private = &shmem_falloc;
2923 spin_unlock(&inode->i_lock);
2925 if ((u64)unmap_end > (u64)unmap_start)
2926 unmap_mapping_range(mapping, unmap_start,
2927 1 + unmap_end - unmap_start, 0);
2928 shmem_truncate_range(inode, offset, offset + len - 1);
2929 /* No need to unmap again: hole-punching leaves COWed pages */
2931 spin_lock(&inode->i_lock);
2932 inode->i_private = NULL;
2933 wake_up_all(&shmem_falloc_waitq);
2934 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2935 spin_unlock(&inode->i_lock);
2940 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2941 error = inode_newsize_ok(inode, offset + len);
2945 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2950 start = offset >> PAGE_SHIFT;
2951 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2952 /* Try to avoid a swapstorm if len is impossible to satisfy */
2953 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2958 shmem_falloc.waitq = NULL;
2959 shmem_falloc.start = start;
2960 shmem_falloc.next = start;
2961 shmem_falloc.nr_falloced = 0;
2962 shmem_falloc.nr_unswapped = 0;
2963 spin_lock(&inode->i_lock);
2964 inode->i_private = &shmem_falloc;
2965 spin_unlock(&inode->i_lock);
2968 * info->fallocend is only relevant when huge pages might be
2969 * involved: to prevent split_huge_page() freeing fallocated
2970 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2972 undo_fallocend = info->fallocend;
2973 if (info->fallocend < end)
2974 info->fallocend = end;
2976 for (index = start; index < end; ) {
2977 struct folio *folio;
2980 * Good, the fallocate(2) manpage permits EINTR: we may have
2981 * been interrupted because we are using up too much memory.
2983 if (signal_pending(current))
2985 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2988 error = shmem_get_folio(inode, index, &folio,
2991 info->fallocend = undo_fallocend;
2992 /* Remove the !uptodate folios we added */
2993 if (index > start) {
2994 shmem_undo_range(inode,
2995 (loff_t)start << PAGE_SHIFT,
2996 ((loff_t)index << PAGE_SHIFT) - 1, true);
3002 * Here is a more important optimization than it appears:
3003 * a second SGP_FALLOC on the same large folio will clear it,
3004 * making it uptodate and un-undoable if we fail later.
3006 index = folio_next_index(folio);
3007 /* Beware 32-bit wraparound */
3012 * Inform shmem_writepage() how far we have reached.
3013 * No need for lock or barrier: we have the page lock.
3015 if (!folio_test_uptodate(folio))
3016 shmem_falloc.nr_falloced += index - shmem_falloc.next;
3017 shmem_falloc.next = index;
3020 * If !uptodate, leave it that way so that freeable folios
3021 * can be recognized if we need to rollback on error later.
3022 * But mark it dirty so that memory pressure will swap rather
3023 * than free the folios we are allocating (and SGP_CACHE folios
3024 * might still be clean: we now need to mark those dirty too).
3026 folio_mark_dirty(folio);
3027 folio_unlock(folio);
3032 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3033 i_size_write(inode, offset + len);
3035 spin_lock(&inode->i_lock);
3036 inode->i_private = NULL;
3037 spin_unlock(&inode->i_lock);
3040 file_modified(file);
3041 inode_unlock(inode);
3045 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
3047 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
3049 buf->f_type = TMPFS_MAGIC;
3050 buf->f_bsize = PAGE_SIZE;
3051 buf->f_namelen = NAME_MAX;
3052 if (sbinfo->max_blocks) {
3053 buf->f_blocks = sbinfo->max_blocks;
3055 buf->f_bfree = sbinfo->max_blocks -
3056 percpu_counter_sum(&sbinfo->used_blocks);
3058 if (sbinfo->max_inodes) {
3059 buf->f_files = sbinfo->max_inodes;
3060 buf->f_ffree = sbinfo->free_inodes;
3062 /* else leave those fields 0 like simple_statfs */
3064 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3070 * File creation. Allocate an inode, and we're done..
3073 shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
3074 struct dentry *dentry, umode_t mode, dev_t dev)
3076 struct inode *inode;
3077 int error = -ENOSPC;
3079 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
3081 error = simple_acl_create(dir, inode);
3084 error = security_inode_init_security(inode, dir,
3086 shmem_initxattrs, NULL);
3087 if (error && error != -EOPNOTSUPP)
3091 dir->i_size += BOGO_DIRENT_SIZE;
3092 dir->i_ctime = dir->i_mtime = current_time(dir);
3093 inode_inc_iversion(dir);
3094 d_instantiate(dentry, inode);
3095 dget(dentry); /* Extra count - pin the dentry in core */
3104 shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
3105 struct file *file, umode_t mode)
3107 struct inode *inode;
3108 int error = -ENOSPC;
3110 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
3112 error = security_inode_init_security(inode, dir,
3114 shmem_initxattrs, NULL);
3115 if (error && error != -EOPNOTSUPP)
3117 error = simple_acl_create(dir, inode);
3120 d_tmpfile(file, inode);
3122 return finish_open_simple(file, error);
3128 static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3129 struct dentry *dentry, umode_t mode)
3133 error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3140 static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
3141 struct dentry *dentry, umode_t mode, bool excl)
3143 return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
3149 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3151 struct inode *inode = d_inode(old_dentry);
3155 * No ordinary (disk based) filesystem counts links as inodes;
3156 * but each new link needs a new dentry, pinning lowmem, and
3157 * tmpfs dentries cannot be pruned until they are unlinked.
3158 * But if an O_TMPFILE file is linked into the tmpfs, the
3159 * first link must skip that, to get the accounting right.
3161 if (inode->i_nlink) {
3162 ret = shmem_reserve_inode(inode->i_sb, NULL);
3167 dir->i_size += BOGO_DIRENT_SIZE;
3168 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3169 inode_inc_iversion(dir);
3171 ihold(inode); /* New dentry reference */
3172 dget(dentry); /* Extra pinning count for the created dentry */
3173 d_instantiate(dentry, inode);
3178 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3180 struct inode *inode = d_inode(dentry);
3182 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3183 shmem_free_inode(inode->i_sb);
3185 dir->i_size -= BOGO_DIRENT_SIZE;
3186 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3187 inode_inc_iversion(dir);
3189 dput(dentry); /* Undo the count from "create" - this does all the work */
3193 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3195 if (!simple_empty(dentry))
3198 drop_nlink(d_inode(dentry));
3200 return shmem_unlink(dir, dentry);
3203 static int shmem_whiteout(struct mnt_idmap *idmap,
3204 struct inode *old_dir, struct dentry *old_dentry)
3206 struct dentry *whiteout;
3209 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3213 error = shmem_mknod(idmap, old_dir, whiteout,
3214 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3220 * Cheat and hash the whiteout while the old dentry is still in
3221 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3223 * d_lookup() will consistently find one of them at this point,
3224 * not sure which one, but that isn't even important.
3231 * The VFS layer already does all the dentry stuff for rename,
3232 * we just have to decrement the usage count for the target if
3233 * it exists so that the VFS layer correctly free's it when it
3236 static int shmem_rename2(struct mnt_idmap *idmap,
3237 struct inode *old_dir, struct dentry *old_dentry,
3238 struct inode *new_dir, struct dentry *new_dentry,
3241 struct inode *inode = d_inode(old_dentry);
3242 int they_are_dirs = S_ISDIR(inode->i_mode);
3244 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3247 if (flags & RENAME_EXCHANGE)
3248 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
3250 if (!simple_empty(new_dentry))
3253 if (flags & RENAME_WHITEOUT) {
3256 error = shmem_whiteout(idmap, old_dir, old_dentry);
3261 if (d_really_is_positive(new_dentry)) {
3262 (void) shmem_unlink(new_dir, new_dentry);
3263 if (they_are_dirs) {
3264 drop_nlink(d_inode(new_dentry));
3265 drop_nlink(old_dir);
3267 } else if (they_are_dirs) {
3268 drop_nlink(old_dir);
3272 old_dir->i_size -= BOGO_DIRENT_SIZE;
3273 new_dir->i_size += BOGO_DIRENT_SIZE;
3274 old_dir->i_ctime = old_dir->i_mtime =
3275 new_dir->i_ctime = new_dir->i_mtime =
3276 inode->i_ctime = current_time(old_dir);
3277 inode_inc_iversion(old_dir);
3278 inode_inc_iversion(new_dir);
3282 static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3283 struct dentry *dentry, const char *symname)
3287 struct inode *inode;
3288 struct folio *folio;
3290 len = strlen(symname) + 1;
3291 if (len > PAGE_SIZE)
3292 return -ENAMETOOLONG;
3294 inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3299 error = security_inode_init_security(inode, dir, &dentry->d_name,
3300 shmem_initxattrs, NULL);
3301 if (error && error != -EOPNOTSUPP) {
3306 inode->i_size = len-1;
3307 if (len <= SHORT_SYMLINK_LEN) {
3308 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3309 if (!inode->i_link) {
3313 inode->i_op = &shmem_short_symlink_operations;
3315 inode_nohighmem(inode);
3316 error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
3321 inode->i_mapping->a_ops = &shmem_aops;
3322 inode->i_op = &shmem_symlink_inode_operations;
3323 memcpy(folio_address(folio), symname, len);
3324 folio_mark_uptodate(folio);
3325 folio_mark_dirty(folio);
3326 folio_unlock(folio);
3329 dir->i_size += BOGO_DIRENT_SIZE;
3330 dir->i_ctime = dir->i_mtime = current_time(dir);
3331 inode_inc_iversion(dir);
3332 d_instantiate(dentry, inode);
3337 static void shmem_put_link(void *arg)
3339 folio_mark_accessed(arg);
3343 static const char *shmem_get_link(struct dentry *dentry,
3344 struct inode *inode,
3345 struct delayed_call *done)
3347 struct folio *folio = NULL;
3351 folio = filemap_get_folio(inode->i_mapping, 0);
3353 return ERR_PTR(-ECHILD);
3354 if (PageHWPoison(folio_page(folio, 0)) ||
3355 !folio_test_uptodate(folio)) {
3357 return ERR_PTR(-ECHILD);
3360 error = shmem_get_folio(inode, 0, &folio, SGP_READ);
3362 return ERR_PTR(error);
3364 return ERR_PTR(-ECHILD);
3365 if (PageHWPoison(folio_page(folio, 0))) {
3366 folio_unlock(folio);
3368 return ERR_PTR(-ECHILD);
3370 folio_unlock(folio);
3372 set_delayed_call(done, shmem_put_link, folio);
3373 return folio_address(folio);
3376 #ifdef CONFIG_TMPFS_XATTR
3378 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3380 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3382 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3387 static int shmem_fileattr_set(struct mnt_idmap *idmap,
3388 struct dentry *dentry, struct fileattr *fa)
3390 struct inode *inode = d_inode(dentry);
3391 struct shmem_inode_info *info = SHMEM_I(inode);
3393 if (fileattr_has_fsx(fa))
3395 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3398 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3399 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3401 shmem_set_inode_flags(inode, info->fsflags);
3402 inode->i_ctime = current_time(inode);
3403 inode_inc_iversion(inode);
3408 * Superblocks without xattr inode operations may get some security.* xattr
3409 * support from the LSM "for free". As soon as we have any other xattrs
3410 * like ACLs, we also need to implement the security.* handlers at
3411 * filesystem level, though.
3415 * Callback for security_inode_init_security() for acquiring xattrs.
3417 static int shmem_initxattrs(struct inode *inode,
3418 const struct xattr *xattr_array,
3421 struct shmem_inode_info *info = SHMEM_I(inode);
3422 const struct xattr *xattr;
3423 struct simple_xattr *new_xattr;
3426 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3427 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3431 len = strlen(xattr->name) + 1;
3432 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3434 if (!new_xattr->name) {
3439 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3440 XATTR_SECURITY_PREFIX_LEN);
3441 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3444 simple_xattr_add(&info->xattrs, new_xattr);
3450 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3451 struct dentry *unused, struct inode *inode,
3452 const char *name, void *buffer, size_t size)
3454 struct shmem_inode_info *info = SHMEM_I(inode);
3456 name = xattr_full_name(handler, name);
3457 return simple_xattr_get(&info->xattrs, name, buffer, size);
3460 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3461 struct mnt_idmap *idmap,
3462 struct dentry *unused, struct inode *inode,
3463 const char *name, const void *value,
3464 size_t size, int flags)
3466 struct shmem_inode_info *info = SHMEM_I(inode);
3469 name = xattr_full_name(handler, name);
3470 err = simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3472 inode->i_ctime = current_time(inode);
3473 inode_inc_iversion(inode);
3478 static const struct xattr_handler shmem_security_xattr_handler = {
3479 .prefix = XATTR_SECURITY_PREFIX,
3480 .get = shmem_xattr_handler_get,
3481 .set = shmem_xattr_handler_set,
3484 static const struct xattr_handler shmem_trusted_xattr_handler = {
3485 .prefix = XATTR_TRUSTED_PREFIX,
3486 .get = shmem_xattr_handler_get,
3487 .set = shmem_xattr_handler_set,
3490 static const struct xattr_handler *shmem_xattr_handlers[] = {
3491 &shmem_security_xattr_handler,
3492 &shmem_trusted_xattr_handler,
3496 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3498 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3499 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3501 #endif /* CONFIG_TMPFS_XATTR */
3503 static const struct inode_operations shmem_short_symlink_operations = {
3504 .getattr = shmem_getattr,
3505 .get_link = simple_get_link,
3506 #ifdef CONFIG_TMPFS_XATTR
3507 .listxattr = shmem_listxattr,
3511 static const struct inode_operations shmem_symlink_inode_operations = {
3512 .getattr = shmem_getattr,
3513 .get_link = shmem_get_link,
3514 #ifdef CONFIG_TMPFS_XATTR
3515 .listxattr = shmem_listxattr,
3519 static struct dentry *shmem_get_parent(struct dentry *child)
3521 return ERR_PTR(-ESTALE);
3524 static int shmem_match(struct inode *ino, void *vfh)
3528 inum = (inum << 32) | fh[1];
3529 return ino->i_ino == inum && fh[0] == ino->i_generation;
3532 /* Find any alias of inode, but prefer a hashed alias */
3533 static struct dentry *shmem_find_alias(struct inode *inode)
3535 struct dentry *alias = d_find_alias(inode);
3537 return alias ?: d_find_any_alias(inode);
3541 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3542 struct fid *fid, int fh_len, int fh_type)
3544 struct inode *inode;
3545 struct dentry *dentry = NULL;
3552 inum = (inum << 32) | fid->raw[1];
3554 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3555 shmem_match, fid->raw);
3557 dentry = shmem_find_alias(inode);
3564 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3565 struct inode *parent)
3569 return FILEID_INVALID;
3572 if (inode_unhashed(inode)) {
3573 /* Unfortunately insert_inode_hash is not idempotent,
3574 * so as we hash inodes here rather than at creation
3575 * time, we need a lock to ensure we only try
3578 static DEFINE_SPINLOCK(lock);
3580 if (inode_unhashed(inode))
3581 __insert_inode_hash(inode,
3582 inode->i_ino + inode->i_generation);
3586 fh[0] = inode->i_generation;
3587 fh[1] = inode->i_ino;
3588 fh[2] = ((__u64)inode->i_ino) >> 32;
3594 static const struct export_operations shmem_export_ops = {
3595 .get_parent = shmem_get_parent,
3596 .encode_fh = shmem_encode_fh,
3597 .fh_to_dentry = shmem_fh_to_dentry,
3614 static const struct constant_table shmem_param_enums_huge[] = {
3615 {"never", SHMEM_HUGE_NEVER },
3616 {"always", SHMEM_HUGE_ALWAYS },
3617 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3618 {"advise", SHMEM_HUGE_ADVISE },
3622 const struct fs_parameter_spec shmem_fs_parameters[] = {
3623 fsparam_u32 ("gid", Opt_gid),
3624 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3625 fsparam_u32oct("mode", Opt_mode),
3626 fsparam_string("mpol", Opt_mpol),
3627 fsparam_string("nr_blocks", Opt_nr_blocks),
3628 fsparam_string("nr_inodes", Opt_nr_inodes),
3629 fsparam_string("size", Opt_size),
3630 fsparam_u32 ("uid", Opt_uid),
3631 fsparam_flag ("inode32", Opt_inode32),
3632 fsparam_flag ("inode64", Opt_inode64),
3633 fsparam_flag ("noswap", Opt_noswap),
3637 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3639 struct shmem_options *ctx = fc->fs_private;
3640 struct fs_parse_result result;
3641 unsigned long long size;
3645 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3651 size = memparse(param->string, &rest);
3653 size <<= PAGE_SHIFT;
3654 size *= totalram_pages();
3660 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3661 ctx->seen |= SHMEM_SEEN_BLOCKS;
3664 ctx->blocks = memparse(param->string, &rest);
3665 if (*rest || ctx->blocks > S64_MAX)
3667 ctx->seen |= SHMEM_SEEN_BLOCKS;
3670 ctx->inodes = memparse(param->string, &rest);
3673 ctx->seen |= SHMEM_SEEN_INODES;
3676 ctx->mode = result.uint_32 & 07777;
3679 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3680 if (!uid_valid(ctx->uid))
3684 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3685 if (!gid_valid(ctx->gid))
3689 ctx->huge = result.uint_32;
3690 if (ctx->huge != SHMEM_HUGE_NEVER &&
3691 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3692 has_transparent_hugepage()))
3693 goto unsupported_parameter;
3694 ctx->seen |= SHMEM_SEEN_HUGE;
3697 if (IS_ENABLED(CONFIG_NUMA)) {
3698 mpol_put(ctx->mpol);
3700 if (mpol_parse_str(param->string, &ctx->mpol))
3704 goto unsupported_parameter;
3706 ctx->full_inums = false;
3707 ctx->seen |= SHMEM_SEEN_INUMS;
3710 if (sizeof(ino_t) < 8) {
3712 "Cannot use inode64 with <64bit inums in kernel\n");
3714 ctx->full_inums = true;
3715 ctx->seen |= SHMEM_SEEN_INUMS;
3718 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
3720 "Turning off swap in unprivileged tmpfs mounts unsupported");
3723 ctx->seen |= SHMEM_SEEN_NOSWAP;
3728 unsupported_parameter:
3729 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3731 return invalfc(fc, "Bad value for '%s'", param->key);
3734 static int shmem_parse_options(struct fs_context *fc, void *data)
3736 char *options = data;
3739 int err = security_sb_eat_lsm_opts(options, &fc->security);
3744 while (options != NULL) {
3745 char *this_char = options;
3748 * NUL-terminate this option: unfortunately,
3749 * mount options form a comma-separated list,
3750 * but mpol's nodelist may also contain commas.
3752 options = strchr(options, ',');
3753 if (options == NULL)
3756 if (!isdigit(*options)) {
3762 char *value = strchr(this_char, '=');
3768 len = strlen(value);
3770 err = vfs_parse_fs_string(fc, this_char, value, len);
3779 * Reconfigure a shmem filesystem.
3781 * Note that we disallow change from limited->unlimited blocks/inodes while any
3782 * are in use; but we must separately disallow unlimited->limited, because in
3783 * that case we have no record of how much is already in use.
3785 static int shmem_reconfigure(struct fs_context *fc)
3787 struct shmem_options *ctx = fc->fs_private;
3788 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3789 unsigned long inodes;
3790 struct mempolicy *mpol = NULL;
3793 raw_spin_lock(&sbinfo->stat_lock);
3794 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3796 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3797 if (!sbinfo->max_blocks) {
3798 err = "Cannot retroactively limit size";
3801 if (percpu_counter_compare(&sbinfo->used_blocks,
3803 err = "Too small a size for current use";
3807 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3808 if (!sbinfo->max_inodes) {
3809 err = "Cannot retroactively limit inodes";
3812 if (ctx->inodes < inodes) {
3813 err = "Too few inodes for current use";
3818 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3819 sbinfo->next_ino > UINT_MAX) {
3820 err = "Current inum too high to switch to 32-bit inums";
3823 if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
3824 err = "Cannot disable swap on remount";
3827 if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
3828 err = "Cannot enable swap on remount if it was disabled on first mount";
3832 if (ctx->seen & SHMEM_SEEN_HUGE)
3833 sbinfo->huge = ctx->huge;
3834 if (ctx->seen & SHMEM_SEEN_INUMS)
3835 sbinfo->full_inums = ctx->full_inums;
3836 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3837 sbinfo->max_blocks = ctx->blocks;
3838 if (ctx->seen & SHMEM_SEEN_INODES) {
3839 sbinfo->max_inodes = ctx->inodes;
3840 sbinfo->free_inodes = ctx->inodes - inodes;
3844 * Preserve previous mempolicy unless mpol remount option was specified.
3847 mpol = sbinfo->mpol;
3848 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3853 sbinfo->noswap = true;
3855 raw_spin_unlock(&sbinfo->stat_lock);
3859 raw_spin_unlock(&sbinfo->stat_lock);
3860 return invalfc(fc, "%s", err);
3863 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3865 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3866 struct mempolicy *mpol;
3868 if (sbinfo->max_blocks != shmem_default_max_blocks())
3869 seq_printf(seq, ",size=%luk",
3870 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3871 if (sbinfo->max_inodes != shmem_default_max_inodes())
3872 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3873 if (sbinfo->mode != (0777 | S_ISVTX))
3874 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3875 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3876 seq_printf(seq, ",uid=%u",
3877 from_kuid_munged(&init_user_ns, sbinfo->uid));
3878 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3879 seq_printf(seq, ",gid=%u",
3880 from_kgid_munged(&init_user_ns, sbinfo->gid));
3883 * Showing inode{64,32} might be useful even if it's the system default,
3884 * since then people don't have to resort to checking both here and
3885 * /proc/config.gz to confirm 64-bit inums were successfully applied
3886 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3888 * We hide it when inode64 isn't the default and we are using 32-bit
3889 * inodes, since that probably just means the feature isn't even under
3894 * +-----------------+-----------------+
3895 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3896 * +------------------+-----------------+-----------------+
3897 * | full_inums=true | show | show |
3898 * | full_inums=false | show | hide |
3899 * +------------------+-----------------+-----------------+
3902 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3903 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3904 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3905 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3907 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3909 mpol = shmem_get_sbmpol(sbinfo);
3910 shmem_show_mpol(seq, mpol);
3913 seq_printf(seq, ",noswap");
3917 #endif /* CONFIG_TMPFS */
3919 static void shmem_put_super(struct super_block *sb)
3921 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3923 free_percpu(sbinfo->ino_batch);
3924 percpu_counter_destroy(&sbinfo->used_blocks);
3925 mpol_put(sbinfo->mpol);
3927 sb->s_fs_info = NULL;
3930 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3932 struct shmem_options *ctx = fc->fs_private;
3933 struct inode *inode;
3934 struct shmem_sb_info *sbinfo;
3936 /* Round up to L1_CACHE_BYTES to resist false sharing */
3937 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3938 L1_CACHE_BYTES), GFP_KERNEL);
3942 sb->s_fs_info = sbinfo;
3946 * Per default we only allow half of the physical ram per
3947 * tmpfs instance, limiting inodes to one per page of lowmem;
3948 * but the internal instance is left unlimited.
3950 if (!(sb->s_flags & SB_KERNMOUNT)) {
3951 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3952 ctx->blocks = shmem_default_max_blocks();
3953 if (!(ctx->seen & SHMEM_SEEN_INODES))
3954 ctx->inodes = shmem_default_max_inodes();
3955 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3956 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3957 sbinfo->noswap = ctx->noswap;
3959 sb->s_flags |= SB_NOUSER;
3961 sb->s_export_op = &shmem_export_ops;
3962 sb->s_flags |= SB_NOSEC | SB_I_VERSION;
3964 sb->s_flags |= SB_NOUSER;
3966 sbinfo->max_blocks = ctx->blocks;
3967 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3968 if (sb->s_flags & SB_KERNMOUNT) {
3969 sbinfo->ino_batch = alloc_percpu(ino_t);
3970 if (!sbinfo->ino_batch)
3973 sbinfo->uid = ctx->uid;
3974 sbinfo->gid = ctx->gid;
3975 sbinfo->full_inums = ctx->full_inums;
3976 sbinfo->mode = ctx->mode;
3977 sbinfo->huge = ctx->huge;
3978 sbinfo->mpol = ctx->mpol;
3981 raw_spin_lock_init(&sbinfo->stat_lock);
3982 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3984 spin_lock_init(&sbinfo->shrinklist_lock);
3985 INIT_LIST_HEAD(&sbinfo->shrinklist);
3987 sb->s_maxbytes = MAX_LFS_FILESIZE;
3988 sb->s_blocksize = PAGE_SIZE;
3989 sb->s_blocksize_bits = PAGE_SHIFT;
3990 sb->s_magic = TMPFS_MAGIC;
3991 sb->s_op = &shmem_ops;
3992 sb->s_time_gran = 1;
3993 #ifdef CONFIG_TMPFS_XATTR
3994 sb->s_xattr = shmem_xattr_handlers;
3996 #ifdef CONFIG_TMPFS_POSIX_ACL
3997 sb->s_flags |= SB_POSIXACL;
3999 uuid_gen(&sb->s_uuid);
4001 inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, S_IFDIR | sbinfo->mode, 0,
4005 inode->i_uid = sbinfo->uid;
4006 inode->i_gid = sbinfo->gid;
4007 sb->s_root = d_make_root(inode);
4013 shmem_put_super(sb);
4017 static int shmem_get_tree(struct fs_context *fc)
4019 return get_tree_nodev(fc, shmem_fill_super);
4022 static void shmem_free_fc(struct fs_context *fc)
4024 struct shmem_options *ctx = fc->fs_private;
4027 mpol_put(ctx->mpol);
4032 static const struct fs_context_operations shmem_fs_context_ops = {
4033 .free = shmem_free_fc,
4034 .get_tree = shmem_get_tree,
4036 .parse_monolithic = shmem_parse_options,
4037 .parse_param = shmem_parse_one,
4038 .reconfigure = shmem_reconfigure,
4042 static struct kmem_cache *shmem_inode_cachep;
4044 static struct inode *shmem_alloc_inode(struct super_block *sb)
4046 struct shmem_inode_info *info;
4047 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
4050 return &info->vfs_inode;
4053 static void shmem_free_in_core_inode(struct inode *inode)
4055 if (S_ISLNK(inode->i_mode))
4056 kfree(inode->i_link);
4057 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4060 static void shmem_destroy_inode(struct inode *inode)
4062 if (S_ISREG(inode->i_mode))
4063 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
4066 static void shmem_init_inode(void *foo)
4068 struct shmem_inode_info *info = foo;
4069 inode_init_once(&info->vfs_inode);
4072 static void shmem_init_inodecache(void)
4074 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4075 sizeof(struct shmem_inode_info),
4076 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
4079 static void shmem_destroy_inodecache(void)
4081 kmem_cache_destroy(shmem_inode_cachep);
4084 /* Keep the page in page cache instead of truncating it */
4085 static int shmem_error_remove_page(struct address_space *mapping,
4091 const struct address_space_operations shmem_aops = {
4092 .writepage = shmem_writepage,
4093 .dirty_folio = noop_dirty_folio,
4095 .write_begin = shmem_write_begin,
4096 .write_end = shmem_write_end,
4098 #ifdef CONFIG_MIGRATION
4099 .migrate_folio = migrate_folio,
4101 .error_remove_page = shmem_error_remove_page,
4103 EXPORT_SYMBOL(shmem_aops);
4105 static const struct file_operations shmem_file_operations = {
4107 .open = generic_file_open,
4108 .get_unmapped_area = shmem_get_unmapped_area,
4110 .llseek = shmem_file_llseek,
4111 .read_iter = shmem_file_read_iter,
4112 .write_iter = generic_file_write_iter,
4113 .fsync = noop_fsync,
4114 .splice_read = shmem_file_splice_read,
4115 .splice_write = iter_file_splice_write,
4116 .fallocate = shmem_fallocate,
4120 static const struct inode_operations shmem_inode_operations = {
4121 .getattr = shmem_getattr,
4122 .setattr = shmem_setattr,
4123 #ifdef CONFIG_TMPFS_XATTR
4124 .listxattr = shmem_listxattr,
4125 .set_acl = simple_set_acl,
4126 .fileattr_get = shmem_fileattr_get,
4127 .fileattr_set = shmem_fileattr_set,
4131 static const struct inode_operations shmem_dir_inode_operations = {
4133 .getattr = shmem_getattr,
4134 .create = shmem_create,
4135 .lookup = simple_lookup,
4137 .unlink = shmem_unlink,
4138 .symlink = shmem_symlink,
4139 .mkdir = shmem_mkdir,
4140 .rmdir = shmem_rmdir,
4141 .mknod = shmem_mknod,
4142 .rename = shmem_rename2,
4143 .tmpfile = shmem_tmpfile,
4145 #ifdef CONFIG_TMPFS_XATTR
4146 .listxattr = shmem_listxattr,
4147 .fileattr_get = shmem_fileattr_get,
4148 .fileattr_set = shmem_fileattr_set,
4150 #ifdef CONFIG_TMPFS_POSIX_ACL
4151 .setattr = shmem_setattr,
4152 .set_acl = simple_set_acl,
4156 static const struct inode_operations shmem_special_inode_operations = {
4157 .getattr = shmem_getattr,
4158 #ifdef CONFIG_TMPFS_XATTR
4159 .listxattr = shmem_listxattr,
4161 #ifdef CONFIG_TMPFS_POSIX_ACL
4162 .setattr = shmem_setattr,
4163 .set_acl = simple_set_acl,
4167 static const struct super_operations shmem_ops = {
4168 .alloc_inode = shmem_alloc_inode,
4169 .free_inode = shmem_free_in_core_inode,
4170 .destroy_inode = shmem_destroy_inode,
4172 .statfs = shmem_statfs,
4173 .show_options = shmem_show_options,
4175 .evict_inode = shmem_evict_inode,
4176 .drop_inode = generic_delete_inode,
4177 .put_super = shmem_put_super,
4178 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4179 .nr_cached_objects = shmem_unused_huge_count,
4180 .free_cached_objects = shmem_unused_huge_scan,
4184 static const struct vm_operations_struct shmem_vm_ops = {
4185 .fault = shmem_fault,
4186 .map_pages = filemap_map_pages,
4188 .set_policy = shmem_set_policy,
4189 .get_policy = shmem_get_policy,
4193 static const struct vm_operations_struct shmem_anon_vm_ops = {
4194 .fault = shmem_fault,
4195 .map_pages = filemap_map_pages,
4197 .set_policy = shmem_set_policy,
4198 .get_policy = shmem_get_policy,
4202 int shmem_init_fs_context(struct fs_context *fc)
4204 struct shmem_options *ctx;
4206 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4210 ctx->mode = 0777 | S_ISVTX;
4211 ctx->uid = current_fsuid();
4212 ctx->gid = current_fsgid();
4214 fc->fs_private = ctx;
4215 fc->ops = &shmem_fs_context_ops;
4219 static struct file_system_type shmem_fs_type = {
4220 .owner = THIS_MODULE,
4222 .init_fs_context = shmem_init_fs_context,
4224 .parameters = shmem_fs_parameters,
4226 .kill_sb = kill_litter_super,
4228 .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4230 .fs_flags = FS_USERNS_MOUNT,
4234 void __init shmem_init(void)
4238 shmem_init_inodecache();
4240 error = register_filesystem(&shmem_fs_type);
4242 pr_err("Could not register tmpfs\n");
4246 shm_mnt = kern_mount(&shmem_fs_type);
4247 if (IS_ERR(shm_mnt)) {
4248 error = PTR_ERR(shm_mnt);
4249 pr_err("Could not kern_mount tmpfs\n");
4253 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4254 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4255 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4257 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4262 unregister_filesystem(&shmem_fs_type);
4264 shmem_destroy_inodecache();
4265 shm_mnt = ERR_PTR(error);
4268 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4269 static ssize_t shmem_enabled_show(struct kobject *kobj,
4270 struct kobj_attribute *attr, char *buf)
4272 static const int values[] = {
4274 SHMEM_HUGE_WITHIN_SIZE,
4283 for (i = 0; i < ARRAY_SIZE(values); i++) {
4284 len += sysfs_emit_at(buf, len,
4285 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4287 shmem_format_huge(values[i]));
4290 len += sysfs_emit_at(buf, len, "\n");
4295 static ssize_t shmem_enabled_store(struct kobject *kobj,
4296 struct kobj_attribute *attr, const char *buf, size_t count)
4301 if (count + 1 > sizeof(tmp))
4303 memcpy(tmp, buf, count);
4305 if (count && tmp[count - 1] == '\n')
4306 tmp[count - 1] = '\0';
4308 huge = shmem_parse_huge(tmp);
4309 if (huge == -EINVAL)
4311 if (!has_transparent_hugepage() &&
4312 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4316 if (shmem_huge > SHMEM_HUGE_DENY)
4317 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4321 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
4322 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4324 #else /* !CONFIG_SHMEM */
4327 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4329 * This is intended for small system where the benefits of the full
4330 * shmem code (swap-backed and resource-limited) are outweighed by
4331 * their complexity. On systems without swap this code should be
4332 * effectively equivalent, but much lighter weight.
4335 static struct file_system_type shmem_fs_type = {
4337 .init_fs_context = ramfs_init_fs_context,
4338 .parameters = ramfs_fs_parameters,
4339 .kill_sb = ramfs_kill_sb,
4340 .fs_flags = FS_USERNS_MOUNT,
4343 void __init shmem_init(void)
4345 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4347 shm_mnt = kern_mount(&shmem_fs_type);
4348 BUG_ON(IS_ERR(shm_mnt));
4351 int shmem_unuse(unsigned int type)
4356 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4361 void shmem_unlock_mapping(struct address_space *mapping)
4366 unsigned long shmem_get_unmapped_area(struct file *file,
4367 unsigned long addr, unsigned long len,
4368 unsigned long pgoff, unsigned long flags)
4370 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4374 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4376 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4378 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4380 #define shmem_vm_ops generic_file_vm_ops
4381 #define shmem_anon_vm_ops generic_file_vm_ops
4382 #define shmem_file_operations ramfs_file_operations
4383 #define shmem_get_inode(idmap, sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4384 #define shmem_acct_size(flags, size) 0
4385 #define shmem_unacct_size(flags, size) do {} while (0)
4387 #endif /* CONFIG_SHMEM */
4391 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4392 unsigned long flags, unsigned int i_flags)
4394 struct inode *inode;
4398 return ERR_CAST(mnt);
4400 if (size < 0 || size > MAX_LFS_FILESIZE)
4401 return ERR_PTR(-EINVAL);
4403 if (shmem_acct_size(flags, size))
4404 return ERR_PTR(-ENOMEM);
4406 if (is_idmapped_mnt(mnt))
4407 return ERR_PTR(-EINVAL);
4409 inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
4410 S_IFREG | S_IRWXUGO, 0, flags);
4411 if (unlikely(!inode)) {
4412 shmem_unacct_size(flags, size);
4413 return ERR_PTR(-ENOSPC);
4415 inode->i_flags |= i_flags;
4416 inode->i_size = size;
4417 clear_nlink(inode); /* It is unlinked */
4418 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4420 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4421 &shmem_file_operations);
4428 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4429 * kernel internal. There will be NO LSM permission checks against the
4430 * underlying inode. So users of this interface must do LSM checks at a
4431 * higher layer. The users are the big_key and shm implementations. LSM
4432 * checks are provided at the key or shm level rather than the inode.
4433 * @name: name for dentry (to be seen in /proc/<pid>/maps
4434 * @size: size to be set for the file
4435 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4437 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4439 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4443 * shmem_file_setup - get an unlinked file living in tmpfs
4444 * @name: name for dentry (to be seen in /proc/<pid>/maps
4445 * @size: size to be set for the file
4446 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4448 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4450 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4452 EXPORT_SYMBOL_GPL(shmem_file_setup);
4455 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4456 * @mnt: the tmpfs mount where the file will be created
4457 * @name: name for dentry (to be seen in /proc/<pid>/maps
4458 * @size: size to be set for the file
4459 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4461 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4462 loff_t size, unsigned long flags)
4464 return __shmem_file_setup(mnt, name, size, flags, 0);
4466 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4469 * shmem_zero_setup - setup a shared anonymous mapping
4470 * @vma: the vma to be mmapped is prepared by do_mmap
4472 int shmem_zero_setup(struct vm_area_struct *vma)
4475 loff_t size = vma->vm_end - vma->vm_start;
4478 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4479 * between XFS directory reading and selinux: since this file is only
4480 * accessible to the user through its mapping, use S_PRIVATE flag to
4481 * bypass file security, in the same way as shmem_kernel_file_setup().
4483 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4485 return PTR_ERR(file);
4489 vma->vm_file = file;
4490 vma->vm_ops = &shmem_anon_vm_ops;
4496 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4497 * @mapping: the folio's address_space
4498 * @index: the folio index
4499 * @gfp: the page allocator flags to use if allocating
4501 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4502 * with any new page allocations done using the specified allocation flags.
4503 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4504 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4505 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4507 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4508 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4510 struct folio *shmem_read_folio_gfp(struct address_space *mapping,
4511 pgoff_t index, gfp_t gfp)
4514 struct inode *inode = mapping->host;
4515 struct folio *folio;
4518 BUG_ON(!shmem_mapping(mapping));
4519 error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
4520 gfp, NULL, NULL, NULL);
4522 return ERR_PTR(error);
4524 folio_unlock(folio);
4528 * The tiny !SHMEM case uses ramfs without swap
4530 return mapping_read_folio_gfp(mapping, index, gfp);
4533 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
4535 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4536 pgoff_t index, gfp_t gfp)
4538 struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
4542 return &folio->page;
4544 page = folio_file_page(folio, index);
4545 if (PageHWPoison(page)) {
4547 return ERR_PTR(-EIO);
4552 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);