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>
81 #include <linux/quotaops.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Pretend that one inode + its dentry occupy this much memory */
94 #define BOGO_INODE_SIZE 1024
96 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
97 #define SHORT_SYMLINK_LEN 128
100 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
101 * inode->i_private (with i_rwsem making sure that it has only one user at
102 * a time): we would prefer not to enlarge the shmem inode just for that.
104 struct shmem_falloc {
105 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
106 pgoff_t start; /* start of range currently being fallocated */
107 pgoff_t next; /* the next page offset to be fallocated */
108 pgoff_t nr_falloced; /* how many new pages have been fallocated */
109 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
112 struct shmem_options {
113 unsigned long long blocks;
114 unsigned long long inodes;
115 struct mempolicy *mpol;
123 unsigned short quota_types;
124 struct shmem_quota_limits qlimits;
125 #define SHMEM_SEEN_BLOCKS 1
126 #define SHMEM_SEEN_INODES 2
127 #define SHMEM_SEEN_HUGE 4
128 #define SHMEM_SEEN_INUMS 8
129 #define SHMEM_SEEN_NOSWAP 16
130 #define SHMEM_SEEN_QUOTA 32
134 static unsigned long shmem_default_max_blocks(void)
136 return totalram_pages() / 2;
139 static unsigned long shmem_default_max_inodes(void)
141 unsigned long nr_pages = totalram_pages();
143 return min3(nr_pages - totalhigh_pages(), nr_pages / 2,
144 ULONG_MAX / BOGO_INODE_SIZE);
148 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
149 struct folio **foliop, enum sgp_type sgp,
150 gfp_t gfp, struct vm_area_struct *vma,
151 vm_fault_t *fault_type);
153 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
155 return sb->s_fs_info;
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
164 static inline int shmem_acct_size(unsigned long flags, loff_t size)
166 return (flags & VM_NORESERVE) ?
167 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
170 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
172 if (!(flags & VM_NORESERVE))
173 vm_unacct_memory(VM_ACCT(size));
176 static inline int shmem_reacct_size(unsigned long flags,
177 loff_t oldsize, loff_t newsize)
179 if (!(flags & VM_NORESERVE)) {
180 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
181 return security_vm_enough_memory_mm(current->mm,
182 VM_ACCT(newsize) - VM_ACCT(oldsize));
183 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
184 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
190 * ... whereas tmpfs objects are accounted incrementally as
191 * pages are allocated, in order to allow large sparse files.
192 * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
193 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
195 static inline int shmem_acct_block(unsigned long flags, long pages)
197 if (!(flags & VM_NORESERVE))
200 return security_vm_enough_memory_mm(current->mm,
201 pages * VM_ACCT(PAGE_SIZE));
204 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
206 if (flags & VM_NORESERVE)
207 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
210 static int shmem_inode_acct_block(struct inode *inode, long pages)
212 struct shmem_inode_info *info = SHMEM_I(inode);
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
216 if (shmem_acct_block(info->flags, pages))
219 might_sleep(); /* when quotas */
220 if (sbinfo->max_blocks) {
221 if (percpu_counter_compare(&sbinfo->used_blocks,
222 sbinfo->max_blocks - pages) > 0)
225 err = dquot_alloc_block_nodirty(inode, pages);
229 percpu_counter_add(&sbinfo->used_blocks, pages);
231 err = dquot_alloc_block_nodirty(inode, pages);
239 shmem_unacct_blocks(info->flags, pages);
243 static void shmem_inode_unacct_blocks(struct inode *inode, long pages)
245 struct shmem_inode_info *info = SHMEM_I(inode);
246 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
248 might_sleep(); /* when quotas */
249 dquot_free_block_nodirty(inode, pages);
251 if (sbinfo->max_blocks)
252 percpu_counter_sub(&sbinfo->used_blocks, pages);
253 shmem_unacct_blocks(info->flags, pages);
256 static const struct super_operations shmem_ops;
257 const struct address_space_operations shmem_aops;
258 static const struct file_operations shmem_file_operations;
259 static const struct inode_operations shmem_inode_operations;
260 static const struct inode_operations shmem_dir_inode_operations;
261 static const struct inode_operations shmem_special_inode_operations;
262 static const struct vm_operations_struct shmem_vm_ops;
263 static const struct vm_operations_struct shmem_anon_vm_ops;
264 static struct file_system_type shmem_fs_type;
266 bool vma_is_anon_shmem(struct vm_area_struct *vma)
268 return vma->vm_ops == &shmem_anon_vm_ops;
271 bool vma_is_shmem(struct vm_area_struct *vma)
273 return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
276 static LIST_HEAD(shmem_swaplist);
277 static DEFINE_MUTEX(shmem_swaplist_mutex);
279 #ifdef CONFIG_TMPFS_QUOTA
281 static int shmem_enable_quotas(struct super_block *sb,
282 unsigned short quota_types)
286 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
287 for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
288 if (!(quota_types & (1 << type)))
290 err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
291 DQUOT_USAGE_ENABLED |
292 DQUOT_LIMITS_ENABLED);
299 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
301 for (type--; type >= 0; type--)
302 dquot_quota_off(sb, type);
306 static void shmem_disable_quotas(struct super_block *sb)
310 for (type = 0; type < SHMEM_MAXQUOTAS; type++)
311 dquot_quota_off(sb, type);
314 static struct dquot **shmem_get_dquots(struct inode *inode)
316 return SHMEM_I(inode)->i_dquot;
318 #endif /* CONFIG_TMPFS_QUOTA */
321 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
322 * produces a novel ino for the newly allocated inode.
324 * It may also be called when making a hard link to permit the space needed by
325 * each dentry. However, in that case, no new inode number is needed since that
326 * internally draws from another pool of inode numbers (currently global
327 * get_next_ino()). This case is indicated by passing NULL as inop.
329 #define SHMEM_INO_BATCH 1024
330 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
332 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
335 if (!(sb->s_flags & SB_KERNMOUNT)) {
336 raw_spin_lock(&sbinfo->stat_lock);
337 if (sbinfo->max_inodes) {
338 if (sbinfo->free_ispace < BOGO_INODE_SIZE) {
339 raw_spin_unlock(&sbinfo->stat_lock);
342 sbinfo->free_ispace -= BOGO_INODE_SIZE;
345 ino = sbinfo->next_ino++;
346 if (unlikely(is_zero_ino(ino)))
347 ino = sbinfo->next_ino++;
348 if (unlikely(!sbinfo->full_inums &&
351 * Emulate get_next_ino uint wraparound for
354 if (IS_ENABLED(CONFIG_64BIT))
355 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
356 __func__, MINOR(sb->s_dev));
357 sbinfo->next_ino = 1;
358 ino = sbinfo->next_ino++;
362 raw_spin_unlock(&sbinfo->stat_lock);
365 * __shmem_file_setup, one of our callers, is lock-free: it
366 * doesn't hold stat_lock in shmem_reserve_inode since
367 * max_inodes is always 0, and is called from potentially
368 * unknown contexts. As such, use a per-cpu batched allocator
369 * which doesn't require the per-sb stat_lock unless we are at
370 * the batch boundary.
372 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
373 * shmem mounts are not exposed to userspace, so we don't need
374 * to worry about things like glibc compatibility.
378 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
380 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
381 raw_spin_lock(&sbinfo->stat_lock);
382 ino = sbinfo->next_ino;
383 sbinfo->next_ino += SHMEM_INO_BATCH;
384 raw_spin_unlock(&sbinfo->stat_lock);
385 if (unlikely(is_zero_ino(ino)))
396 static void shmem_free_inode(struct super_block *sb, size_t freed_ispace)
398 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
399 if (sbinfo->max_inodes) {
400 raw_spin_lock(&sbinfo->stat_lock);
401 sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace;
402 raw_spin_unlock(&sbinfo->stat_lock);
407 * shmem_recalc_inode - recalculate the block usage of an inode
408 * @inode: inode to recalc
409 * @alloced: the change in number of pages allocated to inode
410 * @swapped: the change in number of pages swapped from inode
412 * We have to calculate the free blocks since the mm can drop
413 * undirtied hole pages behind our back.
415 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
416 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
418 static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped)
420 struct shmem_inode_info *info = SHMEM_I(inode);
423 spin_lock(&info->lock);
424 info->alloced += alloced;
425 info->swapped += swapped;
426 freed = info->alloced - info->swapped -
427 READ_ONCE(inode->i_mapping->nrpages);
429 * Special case: whereas normally shmem_recalc_inode() is called
430 * after i_mapping->nrpages has already been adjusted (up or down),
431 * shmem_writepage() has to raise swapped before nrpages is lowered -
432 * to stop a racing shmem_recalc_inode() from thinking that a page has
433 * been freed. Compensate here, to avoid the need for a followup call.
438 info->alloced -= freed;
439 spin_unlock(&info->lock);
441 /* The quota case may block */
443 shmem_inode_unacct_blocks(inode, freed);
446 bool shmem_charge(struct inode *inode, long pages)
448 struct address_space *mapping = inode->i_mapping;
450 if (shmem_inode_acct_block(inode, pages))
453 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
454 xa_lock_irq(&mapping->i_pages);
455 mapping->nrpages += pages;
456 xa_unlock_irq(&mapping->i_pages);
458 shmem_recalc_inode(inode, pages, 0);
462 void shmem_uncharge(struct inode *inode, long pages)
464 /* pages argument is currently unused: keep it to help debugging */
465 /* nrpages adjustment done by __filemap_remove_folio() or caller */
467 shmem_recalc_inode(inode, 0, 0);
471 * Replace item expected in xarray by a new item, while holding xa_lock.
473 static int shmem_replace_entry(struct address_space *mapping,
474 pgoff_t index, void *expected, void *replacement)
476 XA_STATE(xas, &mapping->i_pages, index);
479 VM_BUG_ON(!expected);
480 VM_BUG_ON(!replacement);
481 item = xas_load(&xas);
482 if (item != expected)
484 xas_store(&xas, replacement);
489 * Sometimes, before we decide whether to proceed or to fail, we must check
490 * that an entry was not already brought back from swap by a racing thread.
492 * Checking page is not enough: by the time a SwapCache page is locked, it
493 * might be reused, and again be SwapCache, using the same swap as before.
495 static bool shmem_confirm_swap(struct address_space *mapping,
496 pgoff_t index, swp_entry_t swap)
498 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
502 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
505 * disables huge pages for the mount;
507 * enables huge pages for the mount;
508 * SHMEM_HUGE_WITHIN_SIZE:
509 * only allocate huge pages if the page will be fully within i_size,
510 * also respect fadvise()/madvise() hints;
512 * only allocate huge pages if requested with fadvise()/madvise();
515 #define SHMEM_HUGE_NEVER 0
516 #define SHMEM_HUGE_ALWAYS 1
517 #define SHMEM_HUGE_WITHIN_SIZE 2
518 #define SHMEM_HUGE_ADVISE 3
522 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
525 * disables huge on shm_mnt and all mounts, for emergency use;
527 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
530 #define SHMEM_HUGE_DENY (-1)
531 #define SHMEM_HUGE_FORCE (-2)
533 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
534 /* ifdef here to avoid bloating shmem.o when not necessary */
536 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
538 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
539 struct mm_struct *mm, unsigned long vm_flags)
543 if (!S_ISREG(inode->i_mode))
545 if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
547 if (shmem_huge == SHMEM_HUGE_DENY)
549 if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
552 switch (SHMEM_SB(inode->i_sb)->huge) {
553 case SHMEM_HUGE_ALWAYS:
555 case SHMEM_HUGE_WITHIN_SIZE:
556 index = round_up(index + 1, HPAGE_PMD_NR);
557 i_size = round_up(i_size_read(inode), PAGE_SIZE);
558 if (i_size >> PAGE_SHIFT >= index)
561 case SHMEM_HUGE_ADVISE:
562 if (mm && (vm_flags & VM_HUGEPAGE))
570 #if defined(CONFIG_SYSFS)
571 static int shmem_parse_huge(const char *str)
573 if (!strcmp(str, "never"))
574 return SHMEM_HUGE_NEVER;
575 if (!strcmp(str, "always"))
576 return SHMEM_HUGE_ALWAYS;
577 if (!strcmp(str, "within_size"))
578 return SHMEM_HUGE_WITHIN_SIZE;
579 if (!strcmp(str, "advise"))
580 return SHMEM_HUGE_ADVISE;
581 if (!strcmp(str, "deny"))
582 return SHMEM_HUGE_DENY;
583 if (!strcmp(str, "force"))
584 return SHMEM_HUGE_FORCE;
589 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
590 static const char *shmem_format_huge(int huge)
593 case SHMEM_HUGE_NEVER:
595 case SHMEM_HUGE_ALWAYS:
597 case SHMEM_HUGE_WITHIN_SIZE:
598 return "within_size";
599 case SHMEM_HUGE_ADVISE:
601 case SHMEM_HUGE_DENY:
603 case SHMEM_HUGE_FORCE:
612 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
613 struct shrink_control *sc, unsigned long nr_to_split)
615 LIST_HEAD(list), *pos, *next;
616 LIST_HEAD(to_remove);
618 struct shmem_inode_info *info;
620 unsigned long batch = sc ? sc->nr_to_scan : 128;
623 if (list_empty(&sbinfo->shrinklist))
626 spin_lock(&sbinfo->shrinklist_lock);
627 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
628 info = list_entry(pos, struct shmem_inode_info, shrinklist);
631 inode = igrab(&info->vfs_inode);
633 /* inode is about to be evicted */
635 list_del_init(&info->shrinklist);
639 /* Check if there's anything to gain */
640 if (round_up(inode->i_size, PAGE_SIZE) ==
641 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
642 list_move(&info->shrinklist, &to_remove);
646 list_move(&info->shrinklist, &list);
648 sbinfo->shrinklist_len--;
652 spin_unlock(&sbinfo->shrinklist_lock);
654 list_for_each_safe(pos, next, &to_remove) {
655 info = list_entry(pos, struct shmem_inode_info, shrinklist);
656 inode = &info->vfs_inode;
657 list_del_init(&info->shrinklist);
661 list_for_each_safe(pos, next, &list) {
665 info = list_entry(pos, struct shmem_inode_info, shrinklist);
666 inode = &info->vfs_inode;
668 if (nr_to_split && split >= nr_to_split)
671 index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
672 folio = filemap_get_folio(inode->i_mapping, index);
676 /* No huge page at the end of the file: nothing to split */
677 if (!folio_test_large(folio)) {
683 * Move the inode on the list back to shrinklist if we failed
684 * to lock the page at this time.
686 * Waiting for the lock may lead to deadlock in the
689 if (!folio_trylock(folio)) {
694 ret = split_folio(folio);
698 /* If split failed move the inode on the list back to shrinklist */
704 list_del_init(&info->shrinklist);
708 * Make sure the inode is either on the global list or deleted
709 * from any local list before iput() since it could be deleted
710 * in another thread once we put the inode (then the local list
713 spin_lock(&sbinfo->shrinklist_lock);
714 list_move(&info->shrinklist, &sbinfo->shrinklist);
715 sbinfo->shrinklist_len++;
716 spin_unlock(&sbinfo->shrinklist_lock);
724 static long shmem_unused_huge_scan(struct super_block *sb,
725 struct shrink_control *sc)
727 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
729 if (!READ_ONCE(sbinfo->shrinklist_len))
732 return shmem_unused_huge_shrink(sbinfo, sc, 0);
735 static long shmem_unused_huge_count(struct super_block *sb,
736 struct shrink_control *sc)
738 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
739 return READ_ONCE(sbinfo->shrinklist_len);
741 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
743 #define shmem_huge SHMEM_HUGE_DENY
745 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
746 struct mm_struct *mm, unsigned long vm_flags)
751 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
752 struct shrink_control *sc, unsigned long nr_to_split)
756 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
759 * Like filemap_add_folio, but error if expected item has gone.
761 static int shmem_add_to_page_cache(struct folio *folio,
762 struct address_space *mapping,
763 pgoff_t index, void *expected, gfp_t gfp,
764 struct mm_struct *charge_mm)
766 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
767 long nr = folio_nr_pages(folio);
770 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
771 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
772 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
773 VM_BUG_ON(expected && folio_test_large(folio));
775 folio_ref_add(folio, nr);
776 folio->mapping = mapping;
777 folio->index = index;
779 if (!folio_test_swapcache(folio)) {
780 error = mem_cgroup_charge(folio, charge_mm, gfp);
782 if (folio_test_pmd_mappable(folio)) {
783 count_vm_event(THP_FILE_FALLBACK);
784 count_vm_event(THP_FILE_FALLBACK_CHARGE);
789 folio_throttle_swaprate(folio, gfp);
793 if (expected != xas_find_conflict(&xas)) {
794 xas_set_err(&xas, -EEXIST);
797 if (expected && xas_find_conflict(&xas)) {
798 xas_set_err(&xas, -EEXIST);
801 xas_store(&xas, folio);
804 if (folio_test_pmd_mappable(folio)) {
805 count_vm_event(THP_FILE_ALLOC);
806 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
808 mapping->nrpages += nr;
809 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
810 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
812 xas_unlock_irq(&xas);
813 } while (xas_nomem(&xas, gfp));
815 if (xas_error(&xas)) {
816 error = xas_error(&xas);
822 folio->mapping = NULL;
823 folio_ref_sub(folio, nr);
828 * Like delete_from_page_cache, but substitutes swap for @folio.
830 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
832 struct address_space *mapping = folio->mapping;
833 long nr = folio_nr_pages(folio);
836 xa_lock_irq(&mapping->i_pages);
837 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
838 folio->mapping = NULL;
839 mapping->nrpages -= nr;
840 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
841 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
842 xa_unlock_irq(&mapping->i_pages);
848 * Remove swap entry from page cache, free the swap and its page cache.
850 static int shmem_free_swap(struct address_space *mapping,
851 pgoff_t index, void *radswap)
855 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
858 free_swap_and_cache(radix_to_swp_entry(radswap));
863 * Determine (in bytes) how many of the shmem object's pages mapped by the
864 * given offsets are swapped out.
866 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
867 * as long as the inode doesn't go away and racy results are not a problem.
869 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
870 pgoff_t start, pgoff_t end)
872 XA_STATE(xas, &mapping->i_pages, start);
874 unsigned long swapped = 0;
877 xas_for_each(&xas, page, end - 1) {
878 if (xas_retry(&xas, page))
880 if (xa_is_value(page))
883 if (need_resched()) {
891 return swapped << PAGE_SHIFT;
895 * Determine (in bytes) how many of the shmem object's pages mapped by the
896 * given vma is swapped out.
898 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
899 * as long as the inode doesn't go away and racy results are not a problem.
901 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
903 struct inode *inode = file_inode(vma->vm_file);
904 struct shmem_inode_info *info = SHMEM_I(inode);
905 struct address_space *mapping = inode->i_mapping;
906 unsigned long swapped;
908 /* Be careful as we don't hold info->lock */
909 swapped = READ_ONCE(info->swapped);
912 * The easier cases are when the shmem object has nothing in swap, or
913 * the vma maps it whole. Then we can simply use the stats that we
919 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
920 return swapped << PAGE_SHIFT;
922 /* Here comes the more involved part */
923 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
924 vma->vm_pgoff + vma_pages(vma));
928 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
930 void shmem_unlock_mapping(struct address_space *mapping)
932 struct folio_batch fbatch;
935 folio_batch_init(&fbatch);
937 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
939 while (!mapping_unevictable(mapping) &&
940 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
941 check_move_unevictable_folios(&fbatch);
942 folio_batch_release(&fbatch);
947 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
952 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
953 * beyond i_size, and reports fallocated folios as holes.
955 folio = filemap_get_entry(inode->i_mapping, index);
958 if (!xa_is_value(folio)) {
960 if (folio->mapping == inode->i_mapping)
962 /* The folio has been swapped out */
967 * But read a folio back from swap if any of it is within i_size
968 * (although in some cases this is just a waste of time).
971 shmem_get_folio(inode, index, &folio, SGP_READ);
976 * Remove range of pages and swap entries from page cache, and free them.
977 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
979 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
982 struct address_space *mapping = inode->i_mapping;
983 struct shmem_inode_info *info = SHMEM_I(inode);
984 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
985 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
986 struct folio_batch fbatch;
987 pgoff_t indices[PAGEVEC_SIZE];
990 long nr_swaps_freed = 0;
995 end = -1; /* unsigned, so actually very big */
997 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
998 info->fallocend = start;
1000 folio_batch_init(&fbatch);
1002 while (index < end && find_lock_entries(mapping, &index, end - 1,
1003 &fbatch, indices)) {
1004 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1005 folio = fbatch.folios[i];
1007 if (xa_is_value(folio)) {
1010 nr_swaps_freed += !shmem_free_swap(mapping,
1015 if (!unfalloc || !folio_test_uptodate(folio))
1016 truncate_inode_folio(mapping, folio);
1017 folio_unlock(folio);
1019 folio_batch_remove_exceptionals(&fbatch);
1020 folio_batch_release(&fbatch);
1025 * When undoing a failed fallocate, we want none of the partial folio
1026 * zeroing and splitting below, but shall want to truncate the whole
1027 * folio when !uptodate indicates that it was added by this fallocate,
1028 * even when [lstart, lend] covers only a part of the folio.
1033 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
1034 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
1036 same_folio = lend < folio_pos(folio) + folio_size(folio);
1037 folio_mark_dirty(folio);
1038 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
1039 start = folio->index + folio_nr_pages(folio);
1043 folio_unlock(folio);
1049 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
1051 folio_mark_dirty(folio);
1052 if (!truncate_inode_partial_folio(folio, lstart, lend))
1054 folio_unlock(folio);
1061 while (index < end) {
1064 if (!find_get_entries(mapping, &index, end - 1, &fbatch,
1066 /* If all gone or hole-punch or unfalloc, we're done */
1067 if (index == start || end != -1)
1069 /* But if truncating, restart to make sure all gone */
1073 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1074 folio = fbatch.folios[i];
1076 if (xa_is_value(folio)) {
1079 if (shmem_free_swap(mapping, indices[i], folio)) {
1080 /* Swap was replaced by page: retry */
1090 if (!unfalloc || !folio_test_uptodate(folio)) {
1091 if (folio_mapping(folio) != mapping) {
1092 /* Page was replaced by swap: retry */
1093 folio_unlock(folio);
1097 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1099 truncate_inode_folio(mapping, folio);
1101 folio_unlock(folio);
1103 folio_batch_remove_exceptionals(&fbatch);
1104 folio_batch_release(&fbatch);
1107 shmem_recalc_inode(inode, 0, -nr_swaps_freed);
1110 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1112 shmem_undo_range(inode, lstart, lend, false);
1113 inode->i_ctime = inode->i_mtime = current_time(inode);
1114 inode_inc_iversion(inode);
1116 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1118 static int shmem_getattr(struct mnt_idmap *idmap,
1119 const struct path *path, struct kstat *stat,
1120 u32 request_mask, unsigned int query_flags)
1122 struct inode *inode = path->dentry->d_inode;
1123 struct shmem_inode_info *info = SHMEM_I(inode);
1125 if (info->alloced - info->swapped != inode->i_mapping->nrpages)
1126 shmem_recalc_inode(inode, 0, 0);
1128 if (info->fsflags & FS_APPEND_FL)
1129 stat->attributes |= STATX_ATTR_APPEND;
1130 if (info->fsflags & FS_IMMUTABLE_FL)
1131 stat->attributes |= STATX_ATTR_IMMUTABLE;
1132 if (info->fsflags & FS_NODUMP_FL)
1133 stat->attributes |= STATX_ATTR_NODUMP;
1134 stat->attributes_mask |= (STATX_ATTR_APPEND |
1135 STATX_ATTR_IMMUTABLE |
1137 generic_fillattr(idmap, inode, stat);
1139 if (shmem_is_huge(inode, 0, false, NULL, 0))
1140 stat->blksize = HPAGE_PMD_SIZE;
1142 if (request_mask & STATX_BTIME) {
1143 stat->result_mask |= STATX_BTIME;
1144 stat->btime.tv_sec = info->i_crtime.tv_sec;
1145 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1151 static int shmem_setattr(struct mnt_idmap *idmap,
1152 struct dentry *dentry, struct iattr *attr)
1154 struct inode *inode = d_inode(dentry);
1155 struct shmem_inode_info *info = SHMEM_I(inode);
1157 bool update_mtime = false;
1158 bool update_ctime = true;
1160 error = setattr_prepare(idmap, dentry, attr);
1164 if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1165 if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1170 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1171 loff_t oldsize = inode->i_size;
1172 loff_t newsize = attr->ia_size;
1174 /* protected by i_rwsem */
1175 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1176 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1179 if (newsize != oldsize) {
1180 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1184 i_size_write(inode, newsize);
1185 update_mtime = true;
1187 update_ctime = false;
1189 if (newsize <= oldsize) {
1190 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1191 if (oldsize > holebegin)
1192 unmap_mapping_range(inode->i_mapping,
1195 shmem_truncate_range(inode,
1196 newsize, (loff_t)-1);
1197 /* unmap again to remove racily COWed private pages */
1198 if (oldsize > holebegin)
1199 unmap_mapping_range(inode->i_mapping,
1204 if (is_quota_modification(idmap, inode, attr)) {
1205 error = dquot_initialize(inode);
1210 /* Transfer quota accounting */
1211 if (i_uid_needs_update(idmap, attr, inode) ||
1212 i_gid_needs_update(idmap, attr, inode)) {
1213 error = dquot_transfer(idmap, inode, attr);
1219 setattr_copy(idmap, inode, attr);
1220 if (attr->ia_valid & ATTR_MODE)
1221 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1222 if (!error && update_ctime) {
1223 inode->i_ctime = current_time(inode);
1225 inode->i_mtime = inode->i_ctime;
1226 inode_inc_iversion(inode);
1231 static void shmem_evict_inode(struct inode *inode)
1233 struct shmem_inode_info *info = SHMEM_I(inode);
1234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1237 if (shmem_mapping(inode->i_mapping)) {
1238 shmem_unacct_size(info->flags, inode->i_size);
1240 mapping_set_exiting(inode->i_mapping);
1241 shmem_truncate_range(inode, 0, (loff_t)-1);
1242 if (!list_empty(&info->shrinklist)) {
1243 spin_lock(&sbinfo->shrinklist_lock);
1244 if (!list_empty(&info->shrinklist)) {
1245 list_del_init(&info->shrinklist);
1246 sbinfo->shrinklist_len--;
1248 spin_unlock(&sbinfo->shrinklist_lock);
1250 while (!list_empty(&info->swaplist)) {
1251 /* Wait while shmem_unuse() is scanning this inode... */
1252 wait_var_event(&info->stop_eviction,
1253 !atomic_read(&info->stop_eviction));
1254 mutex_lock(&shmem_swaplist_mutex);
1255 /* ...but beware of the race if we peeked too early */
1256 if (!atomic_read(&info->stop_eviction))
1257 list_del_init(&info->swaplist);
1258 mutex_unlock(&shmem_swaplist_mutex);
1262 simple_xattrs_free(&info->xattrs, sbinfo->max_inodes ? &freed : NULL);
1263 shmem_free_inode(inode->i_sb, freed);
1264 WARN_ON(inode->i_blocks);
1266 #ifdef CONFIG_TMPFS_QUOTA
1267 dquot_free_inode(inode);
1272 static int shmem_find_swap_entries(struct address_space *mapping,
1273 pgoff_t start, struct folio_batch *fbatch,
1274 pgoff_t *indices, unsigned int type)
1276 XA_STATE(xas, &mapping->i_pages, start);
1277 struct folio *folio;
1281 xas_for_each(&xas, folio, ULONG_MAX) {
1282 if (xas_retry(&xas, folio))
1285 if (!xa_is_value(folio))
1288 entry = radix_to_swp_entry(folio);
1290 * swapin error entries can be found in the mapping. But they're
1291 * deliberately ignored here as we've done everything we can do.
1293 if (swp_type(entry) != type)
1296 indices[folio_batch_count(fbatch)] = xas.xa_index;
1297 if (!folio_batch_add(fbatch, folio))
1300 if (need_resched()) {
1307 return xas.xa_index;
1311 * Move the swapped pages for an inode to page cache. Returns the count
1312 * of pages swapped in, or the error in case of failure.
1314 static int shmem_unuse_swap_entries(struct inode *inode,
1315 struct folio_batch *fbatch, pgoff_t *indices)
1320 struct address_space *mapping = inode->i_mapping;
1322 for (i = 0; i < folio_batch_count(fbatch); i++) {
1323 struct folio *folio = fbatch->folios[i];
1325 if (!xa_is_value(folio))
1327 error = shmem_swapin_folio(inode, indices[i],
1329 mapping_gfp_mask(mapping),
1332 folio_unlock(folio);
1336 if (error == -ENOMEM)
1340 return error ? error : ret;
1344 * If swap found in inode, free it and move page from swapcache to filecache.
1346 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1348 struct address_space *mapping = inode->i_mapping;
1350 struct folio_batch fbatch;
1351 pgoff_t indices[PAGEVEC_SIZE];
1355 folio_batch_init(&fbatch);
1356 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1357 if (folio_batch_count(&fbatch) == 0) {
1362 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1366 start = indices[folio_batch_count(&fbatch) - 1];
1373 * Read all the shared memory data that resides in the swap
1374 * device 'type' back into memory, so the swap device can be
1377 int shmem_unuse(unsigned int type)
1379 struct shmem_inode_info *info, *next;
1382 if (list_empty(&shmem_swaplist))
1385 mutex_lock(&shmem_swaplist_mutex);
1386 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1387 if (!info->swapped) {
1388 list_del_init(&info->swaplist);
1392 * Drop the swaplist mutex while searching the inode for swap;
1393 * but before doing so, make sure shmem_evict_inode() will not
1394 * remove placeholder inode from swaplist, nor let it be freed
1395 * (igrab() would protect from unlink, but not from unmount).
1397 atomic_inc(&info->stop_eviction);
1398 mutex_unlock(&shmem_swaplist_mutex);
1400 error = shmem_unuse_inode(&info->vfs_inode, type);
1403 mutex_lock(&shmem_swaplist_mutex);
1404 next = list_next_entry(info, swaplist);
1406 list_del_init(&info->swaplist);
1407 if (atomic_dec_and_test(&info->stop_eviction))
1408 wake_up_var(&info->stop_eviction);
1412 mutex_unlock(&shmem_swaplist_mutex);
1418 * Move the page from the page cache to the swap cache.
1420 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1422 struct folio *folio = page_folio(page);
1423 struct address_space *mapping = folio->mapping;
1424 struct inode *inode = mapping->host;
1425 struct shmem_inode_info *info = SHMEM_I(inode);
1426 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1431 * Our capabilities prevent regular writeback or sync from ever calling
1432 * shmem_writepage; but a stacking filesystem might use ->writepage of
1433 * its underlying filesystem, in which case tmpfs should write out to
1434 * swap only in response to memory pressure, and not for the writeback
1437 if (WARN_ON_ONCE(!wbc->for_reclaim))
1440 if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
1443 if (!total_swap_pages)
1447 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1448 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1449 * and its shmem_writeback() needs them to be split when swapping.
1451 if (folio_test_large(folio)) {
1452 /* Ensure the subpages are still dirty */
1453 folio_test_set_dirty(folio);
1454 if (split_huge_page(page) < 0)
1456 folio = page_folio(page);
1457 folio_clear_dirty(folio);
1460 index = folio->index;
1463 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1464 * value into swapfile.c, the only way we can correctly account for a
1465 * fallocated folio arriving here is now to initialize it and write it.
1467 * That's okay for a folio already fallocated earlier, but if we have
1468 * not yet completed the fallocation, then (a) we want to keep track
1469 * of this folio in case we have to undo it, and (b) it may not be a
1470 * good idea to continue anyway, once we're pushing into swap. So
1471 * reactivate the folio, and let shmem_fallocate() quit when too many.
1473 if (!folio_test_uptodate(folio)) {
1474 if (inode->i_private) {
1475 struct shmem_falloc *shmem_falloc;
1476 spin_lock(&inode->i_lock);
1477 shmem_falloc = inode->i_private;
1479 !shmem_falloc->waitq &&
1480 index >= shmem_falloc->start &&
1481 index < shmem_falloc->next)
1482 shmem_falloc->nr_unswapped++;
1484 shmem_falloc = NULL;
1485 spin_unlock(&inode->i_lock);
1489 folio_zero_range(folio, 0, folio_size(folio));
1490 flush_dcache_folio(folio);
1491 folio_mark_uptodate(folio);
1494 swap = folio_alloc_swap(folio);
1499 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1500 * if it's not already there. Do it now before the folio is
1501 * moved to swap cache, when its pagelock no longer protects
1502 * the inode from eviction. But don't unlock the mutex until
1503 * we've incremented swapped, because shmem_unuse_inode() will
1504 * prune a !swapped inode from the swaplist under this mutex.
1506 mutex_lock(&shmem_swaplist_mutex);
1507 if (list_empty(&info->swaplist))
1508 list_add(&info->swaplist, &shmem_swaplist);
1510 if (add_to_swap_cache(folio, swap,
1511 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1513 shmem_recalc_inode(inode, 0, 1);
1514 swap_shmem_alloc(swap);
1515 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1517 mutex_unlock(&shmem_swaplist_mutex);
1518 BUG_ON(folio_mapped(folio));
1519 swap_writepage(&folio->page, wbc);
1523 mutex_unlock(&shmem_swaplist_mutex);
1524 put_swap_folio(folio, swap);
1526 folio_mark_dirty(folio);
1527 if (wbc->for_reclaim)
1528 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1529 folio_unlock(folio);
1533 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1534 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1538 if (!mpol || mpol->mode == MPOL_DEFAULT)
1539 return; /* show nothing */
1541 mpol_to_str(buffer, sizeof(buffer), mpol);
1543 seq_printf(seq, ",mpol=%s", buffer);
1546 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1548 struct mempolicy *mpol = NULL;
1550 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1551 mpol = sbinfo->mpol;
1553 raw_spin_unlock(&sbinfo->stat_lock);
1557 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1558 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1561 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1565 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1567 #define vm_policy vm_private_data
1570 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1571 struct shmem_inode_info *info, pgoff_t index)
1573 /* Create a pseudo vma that just contains the policy */
1574 vma_init(vma, NULL);
1575 /* Bias interleave by inode number to distribute better across nodes */
1576 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1577 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1580 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1582 /* Drop reference taken by mpol_shared_policy_lookup() */
1583 mpol_cond_put(vma->vm_policy);
1586 static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1587 struct shmem_inode_info *info, pgoff_t index)
1589 struct vm_area_struct pvma;
1591 struct vm_fault vmf = {
1595 shmem_pseudo_vma_init(&pvma, info, index);
1596 page = swap_cluster_readahead(swap, gfp, &vmf);
1597 shmem_pseudo_vma_destroy(&pvma);
1601 return page_folio(page);
1605 * Make sure huge_gfp is always more limited than limit_gfp.
1606 * Some of the flags set permissions, while others set limitations.
1608 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1610 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1611 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1612 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1613 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1615 /* Allow allocations only from the originally specified zones. */
1616 result |= zoneflags;
1619 * Minimize the result gfp by taking the union with the deny flags,
1620 * and the intersection of the allow flags.
1622 result |= (limit_gfp & denyflags);
1623 result |= (huge_gfp & limit_gfp) & allowflags;
1628 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1629 struct shmem_inode_info *info, pgoff_t index)
1631 struct vm_area_struct pvma;
1632 struct address_space *mapping = info->vfs_inode.i_mapping;
1634 struct folio *folio;
1636 hindex = round_down(index, HPAGE_PMD_NR);
1637 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1641 shmem_pseudo_vma_init(&pvma, info, hindex);
1642 folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1643 shmem_pseudo_vma_destroy(&pvma);
1645 count_vm_event(THP_FILE_FALLBACK);
1649 static struct folio *shmem_alloc_folio(gfp_t gfp,
1650 struct shmem_inode_info *info, pgoff_t index)
1652 struct vm_area_struct pvma;
1653 struct folio *folio;
1655 shmem_pseudo_vma_init(&pvma, info, index);
1656 folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1657 shmem_pseudo_vma_destroy(&pvma);
1662 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1663 pgoff_t index, bool huge)
1665 struct shmem_inode_info *info = SHMEM_I(inode);
1666 struct folio *folio;
1670 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1672 nr = huge ? HPAGE_PMD_NR : 1;
1674 err = shmem_inode_acct_block(inode, nr);
1679 folio = shmem_alloc_hugefolio(gfp, info, index);
1681 folio = shmem_alloc_folio(gfp, info, index);
1683 __folio_set_locked(folio);
1684 __folio_set_swapbacked(folio);
1689 shmem_inode_unacct_blocks(inode, nr);
1691 return ERR_PTR(err);
1695 * When a page is moved from swapcache to shmem filecache (either by the
1696 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1697 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1698 * ignorance of the mapping it belongs to. If that mapping has special
1699 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1700 * we may need to copy to a suitable page before moving to filecache.
1702 * In a future release, this may well be extended to respect cpuset and
1703 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1704 * but for now it is a simple matter of zone.
1706 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1708 return folio_zonenum(folio) > gfp_zone(gfp);
1711 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1712 struct shmem_inode_info *info, pgoff_t index)
1714 struct folio *old, *new;
1715 struct address_space *swap_mapping;
1721 entry = folio_swap_entry(old);
1722 swap_index = swp_offset(entry);
1723 swap_mapping = swap_address_space(entry);
1726 * We have arrived here because our zones are constrained, so don't
1727 * limit chance of success by further cpuset and node constraints.
1729 gfp &= ~GFP_CONSTRAINT_MASK;
1730 VM_BUG_ON_FOLIO(folio_test_large(old), old);
1731 new = shmem_alloc_folio(gfp, info, index);
1736 folio_copy(new, old);
1737 flush_dcache_folio(new);
1739 __folio_set_locked(new);
1740 __folio_set_swapbacked(new);
1741 folio_mark_uptodate(new);
1742 folio_set_swap_entry(new, entry);
1743 folio_set_swapcache(new);
1746 * Our caller will very soon move newpage out of swapcache, but it's
1747 * a nice clean interface for us to replace oldpage by newpage there.
1749 xa_lock_irq(&swap_mapping->i_pages);
1750 error = shmem_replace_entry(swap_mapping, swap_index, old, new);
1752 mem_cgroup_migrate(old, new);
1753 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1754 __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1755 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1756 __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
1758 xa_unlock_irq(&swap_mapping->i_pages);
1760 if (unlikely(error)) {
1762 * Is this possible? I think not, now that our callers check
1763 * both PageSwapCache and page_private after getting page lock;
1764 * but be defensive. Reverse old to newpage for clear and free.
1772 folio_clear_swapcache(old);
1773 old->private = NULL;
1776 folio_put_refs(old, 2);
1780 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1781 struct folio *folio, swp_entry_t swap)
1783 struct address_space *mapping = inode->i_mapping;
1784 swp_entry_t swapin_error;
1787 swapin_error = make_swapin_error_entry();
1788 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1789 swp_to_radix_entry(swap),
1790 swp_to_radix_entry(swapin_error), 0);
1791 if (old != swp_to_radix_entry(swap))
1794 folio_wait_writeback(folio);
1795 delete_from_swap_cache(folio);
1797 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
1798 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
1799 * in shmem_evict_inode().
1801 shmem_recalc_inode(inode, -1, -1);
1806 * Swap in the folio pointed to by *foliop.
1807 * Caller has to make sure that *foliop contains a valid swapped folio.
1808 * Returns 0 and the folio in foliop if success. On failure, returns the
1809 * error code and NULL in *foliop.
1811 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1812 struct folio **foliop, enum sgp_type sgp,
1813 gfp_t gfp, struct vm_area_struct *vma,
1814 vm_fault_t *fault_type)
1816 struct address_space *mapping = inode->i_mapping;
1817 struct shmem_inode_info *info = SHMEM_I(inode);
1818 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1819 struct swap_info_struct *si;
1820 struct folio *folio = NULL;
1824 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1825 swap = radix_to_swp_entry(*foliop);
1828 if (is_swapin_error_entry(swap))
1831 si = get_swap_device(swap);
1833 if (!shmem_confirm_swap(mapping, index, swap))
1839 /* Look it up and read it in.. */
1840 folio = swap_cache_get_folio(swap, NULL, 0);
1842 /* Or update major stats only when swapin succeeds?? */
1844 *fault_type |= VM_FAULT_MAJOR;
1845 count_vm_event(PGMAJFAULT);
1846 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1848 /* Here we actually start the io */
1849 folio = shmem_swapin(swap, gfp, info, index);
1856 /* We have to do this with folio locked to prevent races */
1858 if (!folio_test_swapcache(folio) ||
1859 folio_swap_entry(folio).val != swap.val ||
1860 !shmem_confirm_swap(mapping, index, swap)) {
1864 if (!folio_test_uptodate(folio)) {
1868 folio_wait_writeback(folio);
1871 * Some architectures may have to restore extra metadata to the
1872 * folio after reading from swap.
1874 arch_swap_restore(swap, folio);
1876 if (shmem_should_replace_folio(folio, gfp)) {
1877 error = shmem_replace_folio(&folio, gfp, info, index);
1882 error = shmem_add_to_page_cache(folio, mapping, index,
1883 swp_to_radix_entry(swap), gfp,
1888 shmem_recalc_inode(inode, 0, -1);
1890 if (sgp == SGP_WRITE)
1891 folio_mark_accessed(folio);
1893 delete_from_swap_cache(folio);
1894 folio_mark_dirty(folio);
1896 put_swap_device(si);
1901 if (!shmem_confirm_swap(mapping, index, swap))
1904 shmem_set_folio_swapin_error(inode, index, folio, swap);
1907 folio_unlock(folio);
1910 put_swap_device(si);
1916 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1918 * If we allocate a new one we do not mark it dirty. That's up to the
1919 * vm. If we swap it in we mark it dirty since we also free the swap
1920 * entry since a page cannot live in both the swap and page cache.
1922 * vma, vmf, and fault_type are only supplied by shmem_fault:
1923 * otherwise they are NULL.
1925 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1926 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
1927 struct vm_area_struct *vma, struct vm_fault *vmf,
1928 vm_fault_t *fault_type)
1930 struct address_space *mapping = inode->i_mapping;
1931 struct shmem_inode_info *info = SHMEM_I(inode);
1932 struct shmem_sb_info *sbinfo;
1933 struct mm_struct *charge_mm;
1934 struct folio *folio;
1941 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1944 if (sgp <= SGP_CACHE &&
1945 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1949 sbinfo = SHMEM_SB(inode->i_sb);
1950 charge_mm = vma ? vma->vm_mm : NULL;
1952 folio = filemap_get_entry(mapping, index);
1953 if (folio && vma && userfaultfd_minor(vma)) {
1954 if (!xa_is_value(folio))
1956 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1960 if (xa_is_value(folio)) {
1961 error = shmem_swapin_folio(inode, index, &folio,
1962 sgp, gfp, vma, fault_type);
1963 if (error == -EEXIST)
1973 /* Has the folio been truncated or swapped out? */
1974 if (unlikely(folio->mapping != mapping)) {
1975 folio_unlock(folio);
1979 if (sgp == SGP_WRITE)
1980 folio_mark_accessed(folio);
1981 if (folio_test_uptodate(folio))
1983 /* fallocated folio */
1984 if (sgp != SGP_READ)
1986 folio_unlock(folio);
1991 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1992 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1995 if (sgp == SGP_READ)
1997 if (sgp == SGP_NOALLOC)
2001 * Fast cache lookup and swap lookup did not find it: allocate.
2004 if (vma && userfaultfd_missing(vma)) {
2005 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
2009 if (!shmem_is_huge(inode, index, false,
2010 vma ? vma->vm_mm : NULL, vma ? vma->vm_flags : 0))
2013 huge_gfp = vma_thp_gfp_mask(vma);
2014 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
2015 folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
2016 if (IS_ERR(folio)) {
2018 folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
2020 if (IS_ERR(folio)) {
2023 error = PTR_ERR(folio);
2025 if (error != -ENOSPC)
2028 * Try to reclaim some space by splitting a large folio
2029 * beyond i_size on the filesystem.
2034 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
2035 if (ret == SHRINK_STOP)
2043 hindex = round_down(index, folio_nr_pages(folio));
2045 if (sgp == SGP_WRITE)
2046 __folio_set_referenced(folio);
2048 error = shmem_add_to_page_cache(folio, mapping, hindex,
2049 NULL, gfp & GFP_RECLAIM_MASK,
2054 folio_add_lru(folio);
2055 shmem_recalc_inode(inode, folio_nr_pages(folio), 0);
2058 if (folio_test_pmd_mappable(folio) &&
2059 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
2060 folio_next_index(folio) - 1) {
2062 * Part of the large folio is beyond i_size: subject
2063 * to shrink under memory pressure.
2065 spin_lock(&sbinfo->shrinklist_lock);
2067 * _careful to defend against unlocked access to
2068 * ->shrink_list in shmem_unused_huge_shrink()
2070 if (list_empty_careful(&info->shrinklist)) {
2071 list_add_tail(&info->shrinklist,
2072 &sbinfo->shrinklist);
2073 sbinfo->shrinklist_len++;
2075 spin_unlock(&sbinfo->shrinklist_lock);
2079 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2081 if (sgp == SGP_FALLOC)
2085 * Let SGP_WRITE caller clear ends if write does not fill folio;
2086 * but SGP_FALLOC on a folio fallocated earlier must initialize
2087 * it now, lest undo on failure cancel our earlier guarantee.
2089 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2090 long i, n = folio_nr_pages(folio);
2092 for (i = 0; i < n; i++)
2093 clear_highpage(folio_page(folio, i));
2094 flush_dcache_folio(folio);
2095 folio_mark_uptodate(folio);
2098 /* Perhaps the file has been truncated since we checked */
2099 if (sgp <= SGP_CACHE &&
2100 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2102 folio_clear_dirty(folio);
2103 filemap_remove_folio(folio);
2104 shmem_recalc_inode(inode, 0, 0);
2117 shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
2119 if (folio_test_large(folio)) {
2120 folio_unlock(folio);
2126 folio_unlock(folio);
2129 if (error == -ENOSPC && !once++) {
2130 shmem_recalc_inode(inode, 0, 0);
2133 if (error == -EEXIST)
2138 int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
2141 return shmem_get_folio_gfp(inode, index, foliop, sgp,
2142 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
2146 * This is like autoremove_wake_function, but it removes the wait queue
2147 * entry unconditionally - even if something else had already woken the
2150 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2152 int ret = default_wake_function(wait, mode, sync, key);
2153 list_del_init(&wait->entry);
2157 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2159 struct vm_area_struct *vma = vmf->vma;
2160 struct inode *inode = file_inode(vma->vm_file);
2161 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2162 struct folio *folio = NULL;
2164 vm_fault_t ret = VM_FAULT_LOCKED;
2167 * Trinity finds that probing a hole which tmpfs is punching can
2168 * prevent the hole-punch from ever completing: which in turn
2169 * locks writers out with its hold on i_rwsem. So refrain from
2170 * faulting pages into the hole while it's being punched. Although
2171 * shmem_undo_range() does remove the additions, it may be unable to
2172 * keep up, as each new page needs its own unmap_mapping_range() call,
2173 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2175 * It does not matter if we sometimes reach this check just before the
2176 * hole-punch begins, so that one fault then races with the punch:
2177 * we just need to make racing faults a rare case.
2179 * The implementation below would be much simpler if we just used a
2180 * standard mutex or completion: but we cannot take i_rwsem in fault,
2181 * and bloating every shmem inode for this unlikely case would be sad.
2183 if (unlikely(inode->i_private)) {
2184 struct shmem_falloc *shmem_falloc;
2186 spin_lock(&inode->i_lock);
2187 shmem_falloc = inode->i_private;
2189 shmem_falloc->waitq &&
2190 vmf->pgoff >= shmem_falloc->start &&
2191 vmf->pgoff < shmem_falloc->next) {
2193 wait_queue_head_t *shmem_falloc_waitq;
2194 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2196 ret = VM_FAULT_NOPAGE;
2197 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2199 ret = VM_FAULT_RETRY;
2201 shmem_falloc_waitq = shmem_falloc->waitq;
2202 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2203 TASK_UNINTERRUPTIBLE);
2204 spin_unlock(&inode->i_lock);
2208 * shmem_falloc_waitq points into the shmem_fallocate()
2209 * stack of the hole-punching task: shmem_falloc_waitq
2210 * is usually invalid by the time we reach here, but
2211 * finish_wait() does not dereference it in that case;
2212 * though i_lock needed lest racing with wake_up_all().
2214 spin_lock(&inode->i_lock);
2215 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2216 spin_unlock(&inode->i_lock);
2222 spin_unlock(&inode->i_lock);
2225 err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
2226 gfp, vma, vmf, &ret);
2228 return vmf_error(err);
2230 vmf->page = folio_file_page(folio, vmf->pgoff);
2234 unsigned long shmem_get_unmapped_area(struct file *file,
2235 unsigned long uaddr, unsigned long len,
2236 unsigned long pgoff, unsigned long flags)
2238 unsigned long (*get_area)(struct file *,
2239 unsigned long, unsigned long, unsigned long, unsigned long);
2241 unsigned long offset;
2242 unsigned long inflated_len;
2243 unsigned long inflated_addr;
2244 unsigned long inflated_offset;
2246 if (len > TASK_SIZE)
2249 get_area = current->mm->get_unmapped_area;
2250 addr = get_area(file, uaddr, len, pgoff, flags);
2252 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2254 if (IS_ERR_VALUE(addr))
2256 if (addr & ~PAGE_MASK)
2258 if (addr > TASK_SIZE - len)
2261 if (shmem_huge == SHMEM_HUGE_DENY)
2263 if (len < HPAGE_PMD_SIZE)
2265 if (flags & MAP_FIXED)
2268 * Our priority is to support MAP_SHARED mapped hugely;
2269 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2270 * But if caller specified an address hint and we allocated area there
2271 * successfully, respect that as before.
2276 if (shmem_huge != SHMEM_HUGE_FORCE) {
2277 struct super_block *sb;
2280 VM_BUG_ON(file->f_op != &shmem_file_operations);
2281 sb = file_inode(file)->i_sb;
2284 * Called directly from mm/mmap.c, or drivers/char/mem.c
2285 * for "/dev/zero", to create a shared anonymous object.
2287 if (IS_ERR(shm_mnt))
2289 sb = shm_mnt->mnt_sb;
2291 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2295 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2296 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2298 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2301 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2302 if (inflated_len > TASK_SIZE)
2304 if (inflated_len < len)
2307 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2308 if (IS_ERR_VALUE(inflated_addr))
2310 if (inflated_addr & ~PAGE_MASK)
2313 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2314 inflated_addr += offset - inflated_offset;
2315 if (inflated_offset > offset)
2316 inflated_addr += HPAGE_PMD_SIZE;
2318 if (inflated_addr > TASK_SIZE - len)
2320 return inflated_addr;
2324 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2326 struct inode *inode = file_inode(vma->vm_file);
2327 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2330 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2333 struct inode *inode = file_inode(vma->vm_file);
2336 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2337 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2341 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2343 struct inode *inode = file_inode(file);
2344 struct shmem_inode_info *info = SHMEM_I(inode);
2345 int retval = -ENOMEM;
2348 * What serializes the accesses to info->flags?
2349 * ipc_lock_object() when called from shmctl_do_lock(),
2350 * no serialization needed when called from shm_destroy().
2352 if (lock && !(info->flags & VM_LOCKED)) {
2353 if (!user_shm_lock(inode->i_size, ucounts))
2355 info->flags |= VM_LOCKED;
2356 mapping_set_unevictable(file->f_mapping);
2358 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2359 user_shm_unlock(inode->i_size, ucounts);
2360 info->flags &= ~VM_LOCKED;
2361 mapping_clear_unevictable(file->f_mapping);
2369 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2371 struct inode *inode = file_inode(file);
2372 struct shmem_inode_info *info = SHMEM_I(inode);
2375 ret = seal_check_future_write(info->seals, vma);
2379 /* arm64 - allow memory tagging on RAM-based files */
2380 vm_flags_set(vma, VM_MTE_ALLOWED);
2382 file_accessed(file);
2383 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2385 vma->vm_ops = &shmem_vm_ops;
2387 vma->vm_ops = &shmem_anon_vm_ops;
2391 #ifdef CONFIG_TMPFS_XATTR
2392 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2395 * chattr's fsflags are unrelated to extended attributes,
2396 * but tmpfs has chosen to enable them under the same config option.
2398 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2400 unsigned int i_flags = 0;
2402 if (fsflags & FS_NOATIME_FL)
2403 i_flags |= S_NOATIME;
2404 if (fsflags & FS_APPEND_FL)
2405 i_flags |= S_APPEND;
2406 if (fsflags & FS_IMMUTABLE_FL)
2407 i_flags |= S_IMMUTABLE;
2409 * But FS_NODUMP_FL does not require any action in i_flags.
2411 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2414 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2417 #define shmem_initxattrs NULL
2420 static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
2422 return &SHMEM_I(inode)->dir_offsets;
2425 static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
2426 struct super_block *sb,
2427 struct inode *dir, umode_t mode,
2428 dev_t dev, unsigned long flags)
2430 struct inode *inode;
2431 struct shmem_inode_info *info;
2432 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2436 err = shmem_reserve_inode(sb, &ino);
2438 return ERR_PTR(err);
2441 inode = new_inode(sb);
2444 shmem_free_inode(sb, 0);
2445 return ERR_PTR(-ENOSPC);
2449 inode_init_owner(idmap, inode, dir, mode);
2450 inode->i_blocks = 0;
2451 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2452 inode->i_generation = get_random_u32();
2453 info = SHMEM_I(inode);
2454 memset(info, 0, (char *)inode - (char *)info);
2455 spin_lock_init(&info->lock);
2456 atomic_set(&info->stop_eviction, 0);
2457 info->seals = F_SEAL_SEAL;
2458 info->flags = flags & VM_NORESERVE;
2459 info->i_crtime = inode->i_mtime;
2460 info->fsflags = (dir == NULL) ? 0 :
2461 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2463 shmem_set_inode_flags(inode, info->fsflags);
2464 INIT_LIST_HEAD(&info->shrinklist);
2465 INIT_LIST_HEAD(&info->swaplist);
2466 INIT_LIST_HEAD(&info->swaplist);
2468 mapping_set_unevictable(inode->i_mapping);
2469 simple_xattrs_init(&info->xattrs);
2470 cache_no_acl(inode);
2471 mapping_set_large_folios(inode->i_mapping);
2473 switch (mode & S_IFMT) {
2475 inode->i_op = &shmem_special_inode_operations;
2476 init_special_inode(inode, mode, dev);
2479 inode->i_mapping->a_ops = &shmem_aops;
2480 inode->i_op = &shmem_inode_operations;
2481 inode->i_fop = &shmem_file_operations;
2482 mpol_shared_policy_init(&info->policy,
2483 shmem_get_sbmpol(sbinfo));
2487 /* Some things misbehave if size == 0 on a directory */
2488 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2489 inode->i_op = &shmem_dir_inode_operations;
2490 inode->i_fop = &simple_offset_dir_operations;
2491 simple_offset_init(shmem_get_offset_ctx(inode));
2495 * Must not load anything in the rbtree,
2496 * mpol_free_shared_policy will not be called.
2498 mpol_shared_policy_init(&info->policy, NULL);
2502 lockdep_annotate_inode_mutex_key(inode);
2506 #ifdef CONFIG_TMPFS_QUOTA
2507 static struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2508 struct super_block *sb, struct inode *dir,
2509 umode_t mode, dev_t dev, unsigned long flags)
2512 struct inode *inode;
2514 inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2518 err = dquot_initialize(inode);
2522 err = dquot_alloc_inode(inode);
2530 inode->i_flags |= S_NOQUOTA;
2532 return ERR_PTR(err);
2535 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2536 struct super_block *sb, struct inode *dir,
2537 umode_t mode, dev_t dev, unsigned long flags)
2539 return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2541 #endif /* CONFIG_TMPFS_QUOTA */
2543 #ifdef CONFIG_USERFAULTFD
2544 int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
2545 struct vm_area_struct *dst_vma,
2546 unsigned long dst_addr,
2547 unsigned long src_addr,
2549 struct folio **foliop)
2551 struct inode *inode = file_inode(dst_vma->vm_file);
2552 struct shmem_inode_info *info = SHMEM_I(inode);
2553 struct address_space *mapping = inode->i_mapping;
2554 gfp_t gfp = mapping_gfp_mask(mapping);
2555 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2557 struct folio *folio;
2561 if (shmem_inode_acct_block(inode, 1)) {
2563 * We may have got a page, returned -ENOENT triggering a retry,
2564 * and now we find ourselves with -ENOMEM. Release the page, to
2565 * avoid a BUG_ON in our caller.
2567 if (unlikely(*foliop)) {
2576 folio = shmem_alloc_folio(gfp, info, pgoff);
2578 goto out_unacct_blocks;
2580 if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
2581 page_kaddr = kmap_local_folio(folio, 0);
2583 * The read mmap_lock is held here. Despite the
2584 * mmap_lock being read recursive a deadlock is still
2585 * possible if a writer has taken a lock. For example:
2587 * process A thread 1 takes read lock on own mmap_lock
2588 * process A thread 2 calls mmap, blocks taking write lock
2589 * process B thread 1 takes page fault, read lock on own mmap lock
2590 * process B thread 2 calls mmap, blocks taking write lock
2591 * process A thread 1 blocks taking read lock on process B
2592 * process B thread 1 blocks taking read lock on process A
2594 * Disable page faults to prevent potential deadlock
2595 * and retry the copy outside the mmap_lock.
2597 pagefault_disable();
2598 ret = copy_from_user(page_kaddr,
2599 (const void __user *)src_addr,
2602 kunmap_local(page_kaddr);
2604 /* fallback to copy_from_user outside mmap_lock */
2605 if (unlikely(ret)) {
2608 /* don't free the page */
2609 goto out_unacct_blocks;
2612 flush_dcache_folio(folio);
2613 } else { /* ZEROPAGE */
2614 clear_user_highpage(&folio->page, dst_addr);
2618 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2622 VM_BUG_ON(folio_test_locked(folio));
2623 VM_BUG_ON(folio_test_swapbacked(folio));
2624 __folio_set_locked(folio);
2625 __folio_set_swapbacked(folio);
2626 __folio_mark_uptodate(folio);
2629 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2630 if (unlikely(pgoff >= max_off))
2633 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2634 gfp & GFP_RECLAIM_MASK, dst_vma->vm_mm);
2638 ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
2639 &folio->page, true, flags);
2641 goto out_delete_from_cache;
2643 shmem_recalc_inode(inode, 1, 0);
2644 folio_unlock(folio);
2646 out_delete_from_cache:
2647 filemap_remove_folio(folio);
2649 folio_unlock(folio);
2652 shmem_inode_unacct_blocks(inode, 1);
2655 #endif /* CONFIG_USERFAULTFD */
2658 static const struct inode_operations shmem_symlink_inode_operations;
2659 static const struct inode_operations shmem_short_symlink_operations;
2662 shmem_write_begin(struct file *file, struct address_space *mapping,
2663 loff_t pos, unsigned len,
2664 struct page **pagep, void **fsdata)
2666 struct inode *inode = mapping->host;
2667 struct shmem_inode_info *info = SHMEM_I(inode);
2668 pgoff_t index = pos >> PAGE_SHIFT;
2669 struct folio *folio;
2672 /* i_rwsem is held by caller */
2673 if (unlikely(info->seals & (F_SEAL_GROW |
2674 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2675 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2677 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2681 ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
2686 *pagep = folio_file_page(folio, index);
2687 if (PageHWPoison(*pagep)) {
2688 folio_unlock(folio);
2698 shmem_write_end(struct file *file, struct address_space *mapping,
2699 loff_t pos, unsigned len, unsigned copied,
2700 struct page *page, void *fsdata)
2702 struct folio *folio = page_folio(page);
2703 struct inode *inode = mapping->host;
2705 if (pos + copied > inode->i_size)
2706 i_size_write(inode, pos + copied);
2708 if (!folio_test_uptodate(folio)) {
2709 if (copied < folio_size(folio)) {
2710 size_t from = offset_in_folio(folio, pos);
2711 folio_zero_segments(folio, 0, from,
2712 from + copied, folio_size(folio));
2714 folio_mark_uptodate(folio);
2716 folio_mark_dirty(folio);
2717 folio_unlock(folio);
2723 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2725 struct file *file = iocb->ki_filp;
2726 struct inode *inode = file_inode(file);
2727 struct address_space *mapping = inode->i_mapping;
2729 unsigned long offset;
2732 loff_t *ppos = &iocb->ki_pos;
2734 index = *ppos >> PAGE_SHIFT;
2735 offset = *ppos & ~PAGE_MASK;
2738 struct folio *folio = NULL;
2739 struct page *page = NULL;
2741 unsigned long nr, ret;
2742 loff_t i_size = i_size_read(inode);
2744 end_index = i_size >> PAGE_SHIFT;
2745 if (index > end_index)
2747 if (index == end_index) {
2748 nr = i_size & ~PAGE_MASK;
2753 error = shmem_get_folio(inode, index, &folio, SGP_READ);
2755 if (error == -EINVAL)
2760 folio_unlock(folio);
2762 page = folio_file_page(folio, index);
2763 if (PageHWPoison(page)) {
2771 * We must evaluate after, since reads (unlike writes)
2772 * are called without i_rwsem protection against truncate
2775 i_size = i_size_read(inode);
2776 end_index = i_size >> PAGE_SHIFT;
2777 if (index == end_index) {
2778 nr = i_size & ~PAGE_MASK;
2789 * If users can be writing to this page using arbitrary
2790 * virtual addresses, take care about potential aliasing
2791 * before reading the page on the kernel side.
2793 if (mapping_writably_mapped(mapping))
2794 flush_dcache_page(page);
2796 * Mark the page accessed if we read the beginning.
2799 folio_mark_accessed(folio);
2801 * Ok, we have the page, and it's up-to-date, so
2802 * now we can copy it to user space...
2804 ret = copy_page_to_iter(page, offset, nr, to);
2807 } else if (user_backed_iter(to)) {
2809 * Copy to user tends to be so well optimized, but
2810 * clear_user() not so much, that it is noticeably
2811 * faster to copy the zero page instead of clearing.
2813 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2816 * But submitting the same page twice in a row to
2817 * splice() - or others? - can result in confusion:
2818 * so don't attempt that optimization on pipes etc.
2820 ret = iov_iter_zero(nr, to);
2825 index += offset >> PAGE_SHIFT;
2826 offset &= ~PAGE_MASK;
2828 if (!iov_iter_count(to))
2837 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2838 file_accessed(file);
2839 return retval ? retval : error;
2842 static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
2843 struct pipe_buffer *buf)
2848 static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
2849 struct pipe_buffer *buf)
2853 static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
2854 struct pipe_buffer *buf)
2859 static const struct pipe_buf_operations zero_pipe_buf_ops = {
2860 .release = zero_pipe_buf_release,
2861 .try_steal = zero_pipe_buf_try_steal,
2862 .get = zero_pipe_buf_get,
2865 static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
2866 loff_t fpos, size_t size)
2868 size_t offset = fpos & ~PAGE_MASK;
2870 size = min_t(size_t, size, PAGE_SIZE - offset);
2872 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2873 struct pipe_buffer *buf = pipe_head_buf(pipe);
2875 *buf = (struct pipe_buffer) {
2876 .ops = &zero_pipe_buf_ops,
2877 .page = ZERO_PAGE(0),
2887 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2888 struct pipe_inode_info *pipe,
2889 size_t len, unsigned int flags)
2891 struct inode *inode = file_inode(in);
2892 struct address_space *mapping = inode->i_mapping;
2893 struct folio *folio = NULL;
2894 size_t total_spliced = 0, used, npages, n, part;
2898 /* Work out how much data we can actually add into the pipe */
2899 used = pipe_occupancy(pipe->head, pipe->tail);
2900 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2901 len = min_t(size_t, len, npages * PAGE_SIZE);
2904 if (*ppos >= i_size_read(inode))
2907 error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio,
2910 if (error == -EINVAL)
2915 folio_unlock(folio);
2917 if (folio_test_hwpoison(folio) ||
2918 (folio_test_large(folio) &&
2919 folio_test_has_hwpoisoned(folio))) {
2926 * i_size must be checked after we know the pages are Uptodate.
2928 * Checking i_size after the check allows us to calculate
2929 * the correct value for "nr", which means the zero-filled
2930 * part of the page is not copied back to userspace (unless
2931 * another truncate extends the file - this is desired though).
2933 isize = i_size_read(inode);
2934 if (unlikely(*ppos >= isize))
2936 part = min_t(loff_t, isize - *ppos, len);
2940 * If users can be writing to this page using arbitrary
2941 * virtual addresses, take care about potential aliasing
2942 * before reading the page on the kernel side.
2944 if (mapping_writably_mapped(mapping))
2945 flush_dcache_folio(folio);
2946 folio_mark_accessed(folio);
2948 * Ok, we have the page, and it's up-to-date, so we can
2949 * now splice it into the pipe.
2951 n = splice_folio_into_pipe(pipe, folio, *ppos, part);
2955 n = splice_zeropage_into_pipe(pipe, *ppos, part);
2963 in->f_ra.prev_pos = *ppos;
2964 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2974 return total_spliced ? total_spliced : error;
2977 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2979 struct address_space *mapping = file->f_mapping;
2980 struct inode *inode = mapping->host;
2982 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2983 return generic_file_llseek_size(file, offset, whence,
2984 MAX_LFS_FILESIZE, i_size_read(inode));
2989 /* We're holding i_rwsem so we can access i_size directly */
2990 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2992 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2993 inode_unlock(inode);
2997 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
3000 struct inode *inode = file_inode(file);
3001 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3002 struct shmem_inode_info *info = SHMEM_I(inode);
3003 struct shmem_falloc shmem_falloc;
3004 pgoff_t start, index, end, undo_fallocend;
3007 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3012 if (mode & FALLOC_FL_PUNCH_HOLE) {
3013 struct address_space *mapping = file->f_mapping;
3014 loff_t unmap_start = round_up(offset, PAGE_SIZE);
3015 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
3016 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
3018 /* protected by i_rwsem */
3019 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
3024 shmem_falloc.waitq = &shmem_falloc_waitq;
3025 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
3026 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
3027 spin_lock(&inode->i_lock);
3028 inode->i_private = &shmem_falloc;
3029 spin_unlock(&inode->i_lock);
3031 if ((u64)unmap_end > (u64)unmap_start)
3032 unmap_mapping_range(mapping, unmap_start,
3033 1 + unmap_end - unmap_start, 0);
3034 shmem_truncate_range(inode, offset, offset + len - 1);
3035 /* No need to unmap again: hole-punching leaves COWed pages */
3037 spin_lock(&inode->i_lock);
3038 inode->i_private = NULL;
3039 wake_up_all(&shmem_falloc_waitq);
3040 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
3041 spin_unlock(&inode->i_lock);
3046 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3047 error = inode_newsize_ok(inode, offset + len);
3051 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
3056 start = offset >> PAGE_SHIFT;
3057 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
3058 /* Try to avoid a swapstorm if len is impossible to satisfy */
3059 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
3064 shmem_falloc.waitq = NULL;
3065 shmem_falloc.start = start;
3066 shmem_falloc.next = start;
3067 shmem_falloc.nr_falloced = 0;
3068 shmem_falloc.nr_unswapped = 0;
3069 spin_lock(&inode->i_lock);
3070 inode->i_private = &shmem_falloc;
3071 spin_unlock(&inode->i_lock);
3074 * info->fallocend is only relevant when huge pages might be
3075 * involved: to prevent split_huge_page() freeing fallocated
3076 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3078 undo_fallocend = info->fallocend;
3079 if (info->fallocend < end)
3080 info->fallocend = end;
3082 for (index = start; index < end; ) {
3083 struct folio *folio;
3086 * Good, the fallocate(2) manpage permits EINTR: we may have
3087 * been interrupted because we are using up too much memory.
3089 if (signal_pending(current))
3091 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
3094 error = shmem_get_folio(inode, index, &folio,
3097 info->fallocend = undo_fallocend;
3098 /* Remove the !uptodate folios we added */
3099 if (index > start) {
3100 shmem_undo_range(inode,
3101 (loff_t)start << PAGE_SHIFT,
3102 ((loff_t)index << PAGE_SHIFT) - 1, true);
3108 * Here is a more important optimization than it appears:
3109 * a second SGP_FALLOC on the same large folio will clear it,
3110 * making it uptodate and un-undoable if we fail later.
3112 index = folio_next_index(folio);
3113 /* Beware 32-bit wraparound */
3118 * Inform shmem_writepage() how far we have reached.
3119 * No need for lock or barrier: we have the page lock.
3121 if (!folio_test_uptodate(folio))
3122 shmem_falloc.nr_falloced += index - shmem_falloc.next;
3123 shmem_falloc.next = index;
3126 * If !uptodate, leave it that way so that freeable folios
3127 * can be recognized if we need to rollback on error later.
3128 * But mark it dirty so that memory pressure will swap rather
3129 * than free the folios we are allocating (and SGP_CACHE folios
3130 * might still be clean: we now need to mark those dirty too).
3132 folio_mark_dirty(folio);
3133 folio_unlock(folio);
3138 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3139 i_size_write(inode, offset + len);
3141 spin_lock(&inode->i_lock);
3142 inode->i_private = NULL;
3143 spin_unlock(&inode->i_lock);
3146 file_modified(file);
3147 inode_unlock(inode);
3151 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
3153 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
3155 buf->f_type = TMPFS_MAGIC;
3156 buf->f_bsize = PAGE_SIZE;
3157 buf->f_namelen = NAME_MAX;
3158 if (sbinfo->max_blocks) {
3159 buf->f_blocks = sbinfo->max_blocks;
3161 buf->f_bfree = sbinfo->max_blocks -
3162 percpu_counter_sum(&sbinfo->used_blocks);
3164 if (sbinfo->max_inodes) {
3165 buf->f_files = sbinfo->max_inodes;
3166 buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE;
3168 /* else leave those fields 0 like simple_statfs */
3170 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3176 * File creation. Allocate an inode, and we're done..
3179 shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
3180 struct dentry *dentry, umode_t mode, dev_t dev)
3182 struct inode *inode;
3185 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
3188 return PTR_ERR(inode);
3190 error = simple_acl_create(dir, inode);
3193 error = security_inode_init_security(inode, dir,
3195 shmem_initxattrs, NULL);
3196 if (error && error != -EOPNOTSUPP)
3199 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3203 dir->i_size += BOGO_DIRENT_SIZE;
3204 dir->i_ctime = dir->i_mtime = current_time(dir);
3205 inode_inc_iversion(dir);
3206 d_instantiate(dentry, inode);
3207 dget(dentry); /* Extra count - pin the dentry in core */
3216 shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
3217 struct file *file, umode_t mode)
3219 struct inode *inode;
3222 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
3224 if (IS_ERR(inode)) {
3225 error = PTR_ERR(inode);
3229 error = security_inode_init_security(inode, dir,
3231 shmem_initxattrs, NULL);
3232 if (error && error != -EOPNOTSUPP)
3234 error = simple_acl_create(dir, inode);
3237 d_tmpfile(file, inode);
3240 return finish_open_simple(file, error);
3246 static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3247 struct dentry *dentry, umode_t mode)
3251 error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3258 static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
3259 struct dentry *dentry, umode_t mode, bool excl)
3261 return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
3267 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3269 struct inode *inode = d_inode(old_dentry);
3273 * No ordinary (disk based) filesystem counts links as inodes;
3274 * but each new link needs a new dentry, pinning lowmem, and
3275 * tmpfs dentries cannot be pruned until they are unlinked.
3276 * But if an O_TMPFILE file is linked into the tmpfs, the
3277 * first link must skip that, to get the accounting right.
3279 if (inode->i_nlink) {
3280 ret = shmem_reserve_inode(inode->i_sb, NULL);
3285 ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3288 shmem_free_inode(inode->i_sb, 0);
3292 dir->i_size += BOGO_DIRENT_SIZE;
3293 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3294 inode_inc_iversion(dir);
3296 ihold(inode); /* New dentry reference */
3297 dget(dentry); /* Extra pinning count for the created dentry */
3298 d_instantiate(dentry, inode);
3303 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3305 struct inode *inode = d_inode(dentry);
3307 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3308 shmem_free_inode(inode->i_sb, 0);
3310 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3312 dir->i_size -= BOGO_DIRENT_SIZE;
3313 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3314 inode_inc_iversion(dir);
3316 dput(dentry); /* Undo the count from "create" - this does all the work */
3320 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3322 if (!simple_empty(dentry))
3325 drop_nlink(d_inode(dentry));
3327 return shmem_unlink(dir, dentry);
3330 static int shmem_whiteout(struct mnt_idmap *idmap,
3331 struct inode *old_dir, struct dentry *old_dentry)
3333 struct dentry *whiteout;
3336 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3340 error = shmem_mknod(idmap, old_dir, whiteout,
3341 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3347 * Cheat and hash the whiteout while the old dentry is still in
3348 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3350 * d_lookup() will consistently find one of them at this point,
3351 * not sure which one, but that isn't even important.
3358 * The VFS layer already does all the dentry stuff for rename,
3359 * we just have to decrement the usage count for the target if
3360 * it exists so that the VFS layer correctly free's it when it
3363 static int shmem_rename2(struct mnt_idmap *idmap,
3364 struct inode *old_dir, struct dentry *old_dentry,
3365 struct inode *new_dir, struct dentry *new_dentry,
3368 struct inode *inode = d_inode(old_dentry);
3369 int they_are_dirs = S_ISDIR(inode->i_mode);
3372 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3375 if (flags & RENAME_EXCHANGE)
3376 return simple_offset_rename_exchange(old_dir, old_dentry,
3377 new_dir, new_dentry);
3379 if (!simple_empty(new_dentry))
3382 if (flags & RENAME_WHITEOUT) {
3383 error = shmem_whiteout(idmap, old_dir, old_dentry);
3388 simple_offset_remove(shmem_get_offset_ctx(old_dir), old_dentry);
3389 error = simple_offset_add(shmem_get_offset_ctx(new_dir), old_dentry);
3393 if (d_really_is_positive(new_dentry)) {
3394 (void) shmem_unlink(new_dir, new_dentry);
3395 if (they_are_dirs) {
3396 drop_nlink(d_inode(new_dentry));
3397 drop_nlink(old_dir);
3399 } else if (they_are_dirs) {
3400 drop_nlink(old_dir);
3404 old_dir->i_size -= BOGO_DIRENT_SIZE;
3405 new_dir->i_size += BOGO_DIRENT_SIZE;
3406 old_dir->i_ctime = old_dir->i_mtime =
3407 new_dir->i_ctime = new_dir->i_mtime =
3408 inode->i_ctime = current_time(old_dir);
3409 inode_inc_iversion(old_dir);
3410 inode_inc_iversion(new_dir);
3414 static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3415 struct dentry *dentry, const char *symname)
3419 struct inode *inode;
3420 struct folio *folio;
3422 len = strlen(symname) + 1;
3423 if (len > PAGE_SIZE)
3424 return -ENAMETOOLONG;
3426 inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3430 return PTR_ERR(inode);
3432 error = security_inode_init_security(inode, dir, &dentry->d_name,
3433 shmem_initxattrs, NULL);
3434 if (error && error != -EOPNOTSUPP)
3437 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3441 inode->i_size = len-1;
3442 if (len <= SHORT_SYMLINK_LEN) {
3443 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3444 if (!inode->i_link) {
3446 goto out_remove_offset;
3448 inode->i_op = &shmem_short_symlink_operations;
3450 inode_nohighmem(inode);
3451 error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
3453 goto out_remove_offset;
3454 inode->i_mapping->a_ops = &shmem_aops;
3455 inode->i_op = &shmem_symlink_inode_operations;
3456 memcpy(folio_address(folio), symname, len);
3457 folio_mark_uptodate(folio);
3458 folio_mark_dirty(folio);
3459 folio_unlock(folio);
3462 dir->i_size += BOGO_DIRENT_SIZE;
3463 dir->i_ctime = dir->i_mtime = current_time(dir);
3464 inode_inc_iversion(dir);
3465 d_instantiate(dentry, inode);
3470 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3476 static void shmem_put_link(void *arg)
3478 folio_mark_accessed(arg);
3482 static const char *shmem_get_link(struct dentry *dentry,
3483 struct inode *inode,
3484 struct delayed_call *done)
3486 struct folio *folio = NULL;
3490 folio = filemap_get_folio(inode->i_mapping, 0);
3492 return ERR_PTR(-ECHILD);
3493 if (PageHWPoison(folio_page(folio, 0)) ||
3494 !folio_test_uptodate(folio)) {
3496 return ERR_PTR(-ECHILD);
3499 error = shmem_get_folio(inode, 0, &folio, SGP_READ);
3501 return ERR_PTR(error);
3503 return ERR_PTR(-ECHILD);
3504 if (PageHWPoison(folio_page(folio, 0))) {
3505 folio_unlock(folio);
3507 return ERR_PTR(-ECHILD);
3509 folio_unlock(folio);
3511 set_delayed_call(done, shmem_put_link, folio);
3512 return folio_address(folio);
3515 #ifdef CONFIG_TMPFS_XATTR
3517 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3519 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3521 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3526 static int shmem_fileattr_set(struct mnt_idmap *idmap,
3527 struct dentry *dentry, struct fileattr *fa)
3529 struct inode *inode = d_inode(dentry);
3530 struct shmem_inode_info *info = SHMEM_I(inode);
3532 if (fileattr_has_fsx(fa))
3534 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3537 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3538 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3540 shmem_set_inode_flags(inode, info->fsflags);
3541 inode->i_ctime = current_time(inode);
3542 inode_inc_iversion(inode);
3547 * Superblocks without xattr inode operations may get some security.* xattr
3548 * support from the LSM "for free". As soon as we have any other xattrs
3549 * like ACLs, we also need to implement the security.* handlers at
3550 * filesystem level, though.
3554 * Callback for security_inode_init_security() for acquiring xattrs.
3556 static int shmem_initxattrs(struct inode *inode,
3557 const struct xattr *xattr_array,
3560 struct shmem_inode_info *info = SHMEM_I(inode);
3561 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3562 const struct xattr *xattr;
3563 struct simple_xattr *new_xattr;
3567 if (sbinfo->max_inodes) {
3568 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3569 ispace += simple_xattr_space(xattr->name,
3570 xattr->value_len + XATTR_SECURITY_PREFIX_LEN);
3573 raw_spin_lock(&sbinfo->stat_lock);
3574 if (sbinfo->free_ispace < ispace)
3577 sbinfo->free_ispace -= ispace;
3578 raw_spin_unlock(&sbinfo->stat_lock);
3584 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3585 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3589 len = strlen(xattr->name) + 1;
3590 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3592 if (!new_xattr->name) {
3597 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3598 XATTR_SECURITY_PREFIX_LEN);
3599 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3602 simple_xattr_add(&info->xattrs, new_xattr);
3605 if (xattr->name != NULL) {
3607 raw_spin_lock(&sbinfo->stat_lock);
3608 sbinfo->free_ispace += ispace;
3609 raw_spin_unlock(&sbinfo->stat_lock);
3611 simple_xattrs_free(&info->xattrs, NULL);
3618 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3619 struct dentry *unused, struct inode *inode,
3620 const char *name, void *buffer, size_t size)
3622 struct shmem_inode_info *info = SHMEM_I(inode);
3624 name = xattr_full_name(handler, name);
3625 return simple_xattr_get(&info->xattrs, name, buffer, size);
3628 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3629 struct mnt_idmap *idmap,
3630 struct dentry *unused, struct inode *inode,
3631 const char *name, const void *value,
3632 size_t size, int flags)
3634 struct shmem_inode_info *info = SHMEM_I(inode);
3635 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3636 struct simple_xattr *old_xattr;
3639 name = xattr_full_name(handler, name);
3640 if (value && sbinfo->max_inodes) {
3641 ispace = simple_xattr_space(name, size);
3642 raw_spin_lock(&sbinfo->stat_lock);
3643 if (sbinfo->free_ispace < ispace)
3646 sbinfo->free_ispace -= ispace;
3647 raw_spin_unlock(&sbinfo->stat_lock);
3652 old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags);
3653 if (!IS_ERR(old_xattr)) {
3655 if (old_xattr && sbinfo->max_inodes)
3656 ispace = simple_xattr_space(old_xattr->name,
3658 simple_xattr_free(old_xattr);
3660 inode->i_ctime = current_time(inode);
3661 inode_inc_iversion(inode);
3664 raw_spin_lock(&sbinfo->stat_lock);
3665 sbinfo->free_ispace += ispace;
3666 raw_spin_unlock(&sbinfo->stat_lock);
3668 return PTR_ERR(old_xattr);
3671 static const struct xattr_handler shmem_security_xattr_handler = {
3672 .prefix = XATTR_SECURITY_PREFIX,
3673 .get = shmem_xattr_handler_get,
3674 .set = shmem_xattr_handler_set,
3677 static const struct xattr_handler shmem_trusted_xattr_handler = {
3678 .prefix = XATTR_TRUSTED_PREFIX,
3679 .get = shmem_xattr_handler_get,
3680 .set = shmem_xattr_handler_set,
3683 static const struct xattr_handler shmem_user_xattr_handler = {
3684 .prefix = XATTR_USER_PREFIX,
3685 .get = shmem_xattr_handler_get,
3686 .set = shmem_xattr_handler_set,
3689 static const struct xattr_handler *shmem_xattr_handlers[] = {
3690 &shmem_security_xattr_handler,
3691 &shmem_trusted_xattr_handler,
3692 &shmem_user_xattr_handler,
3696 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3698 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3699 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3701 #endif /* CONFIG_TMPFS_XATTR */
3703 static const struct inode_operations shmem_short_symlink_operations = {
3704 .getattr = shmem_getattr,
3705 .setattr = shmem_setattr,
3706 .get_link = simple_get_link,
3707 #ifdef CONFIG_TMPFS_XATTR
3708 .listxattr = shmem_listxattr,
3712 static const struct inode_operations shmem_symlink_inode_operations = {
3713 .getattr = shmem_getattr,
3714 .setattr = shmem_setattr,
3715 .get_link = shmem_get_link,
3716 #ifdef CONFIG_TMPFS_XATTR
3717 .listxattr = shmem_listxattr,
3721 static struct dentry *shmem_get_parent(struct dentry *child)
3723 return ERR_PTR(-ESTALE);
3726 static int shmem_match(struct inode *ino, void *vfh)
3730 inum = (inum << 32) | fh[1];
3731 return ino->i_ino == inum && fh[0] == ino->i_generation;
3734 /* Find any alias of inode, but prefer a hashed alias */
3735 static struct dentry *shmem_find_alias(struct inode *inode)
3737 struct dentry *alias = d_find_alias(inode);
3739 return alias ?: d_find_any_alias(inode);
3743 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3744 struct fid *fid, int fh_len, int fh_type)
3746 struct inode *inode;
3747 struct dentry *dentry = NULL;
3754 inum = (inum << 32) | fid->raw[1];
3756 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3757 shmem_match, fid->raw);
3759 dentry = shmem_find_alias(inode);
3766 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3767 struct inode *parent)
3771 return FILEID_INVALID;
3774 if (inode_unhashed(inode)) {
3775 /* Unfortunately insert_inode_hash is not idempotent,
3776 * so as we hash inodes here rather than at creation
3777 * time, we need a lock to ensure we only try
3780 static DEFINE_SPINLOCK(lock);
3782 if (inode_unhashed(inode))
3783 __insert_inode_hash(inode,
3784 inode->i_ino + inode->i_generation);
3788 fh[0] = inode->i_generation;
3789 fh[1] = inode->i_ino;
3790 fh[2] = ((__u64)inode->i_ino) >> 32;
3796 static const struct export_operations shmem_export_ops = {
3797 .get_parent = shmem_get_parent,
3798 .encode_fh = shmem_encode_fh,
3799 .fh_to_dentry = shmem_fh_to_dentry,
3817 Opt_usrquota_block_hardlimit,
3818 Opt_usrquota_inode_hardlimit,
3819 Opt_grpquota_block_hardlimit,
3820 Opt_grpquota_inode_hardlimit,
3823 static const struct constant_table shmem_param_enums_huge[] = {
3824 {"never", SHMEM_HUGE_NEVER },
3825 {"always", SHMEM_HUGE_ALWAYS },
3826 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3827 {"advise", SHMEM_HUGE_ADVISE },
3831 const struct fs_parameter_spec shmem_fs_parameters[] = {
3832 fsparam_u32 ("gid", Opt_gid),
3833 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3834 fsparam_u32oct("mode", Opt_mode),
3835 fsparam_string("mpol", Opt_mpol),
3836 fsparam_string("nr_blocks", Opt_nr_blocks),
3837 fsparam_string("nr_inodes", Opt_nr_inodes),
3838 fsparam_string("size", Opt_size),
3839 fsparam_u32 ("uid", Opt_uid),
3840 fsparam_flag ("inode32", Opt_inode32),
3841 fsparam_flag ("inode64", Opt_inode64),
3842 fsparam_flag ("noswap", Opt_noswap),
3843 #ifdef CONFIG_TMPFS_QUOTA
3844 fsparam_flag ("quota", Opt_quota),
3845 fsparam_flag ("usrquota", Opt_usrquota),
3846 fsparam_flag ("grpquota", Opt_grpquota),
3847 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
3848 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
3849 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
3850 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
3855 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3857 struct shmem_options *ctx = fc->fs_private;
3858 struct fs_parse_result result;
3859 unsigned long long size;
3865 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3871 size = memparse(param->string, &rest);
3873 size <<= PAGE_SHIFT;
3874 size *= totalram_pages();
3880 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3881 ctx->seen |= SHMEM_SEEN_BLOCKS;
3884 ctx->blocks = memparse(param->string, &rest);
3885 if (*rest || ctx->blocks > LONG_MAX)
3887 ctx->seen |= SHMEM_SEEN_BLOCKS;
3890 ctx->inodes = memparse(param->string, &rest);
3891 if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE)
3893 ctx->seen |= SHMEM_SEEN_INODES;
3896 ctx->mode = result.uint_32 & 07777;
3899 kuid = make_kuid(current_user_ns(), result.uint_32);
3900 if (!uid_valid(kuid))
3904 * The requested uid must be representable in the
3905 * filesystem's idmapping.
3907 if (!kuid_has_mapping(fc->user_ns, kuid))
3913 kgid = make_kgid(current_user_ns(), result.uint_32);
3914 if (!gid_valid(kgid))
3918 * The requested gid must be representable in the
3919 * filesystem's idmapping.
3921 if (!kgid_has_mapping(fc->user_ns, kgid))
3927 ctx->huge = result.uint_32;
3928 if (ctx->huge != SHMEM_HUGE_NEVER &&
3929 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3930 has_transparent_hugepage()))
3931 goto unsupported_parameter;
3932 ctx->seen |= SHMEM_SEEN_HUGE;
3935 if (IS_ENABLED(CONFIG_NUMA)) {
3936 mpol_put(ctx->mpol);
3938 if (mpol_parse_str(param->string, &ctx->mpol))
3942 goto unsupported_parameter;
3944 ctx->full_inums = false;
3945 ctx->seen |= SHMEM_SEEN_INUMS;
3948 if (sizeof(ino_t) < 8) {
3950 "Cannot use inode64 with <64bit inums in kernel\n");
3952 ctx->full_inums = true;
3953 ctx->seen |= SHMEM_SEEN_INUMS;
3956 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
3958 "Turning off swap in unprivileged tmpfs mounts unsupported");
3961 ctx->seen |= SHMEM_SEEN_NOSWAP;
3964 if (fc->user_ns != &init_user_ns)
3965 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3966 ctx->seen |= SHMEM_SEEN_QUOTA;
3967 ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
3970 if (fc->user_ns != &init_user_ns)
3971 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3972 ctx->seen |= SHMEM_SEEN_QUOTA;
3973 ctx->quota_types |= QTYPE_MASK_USR;
3976 if (fc->user_ns != &init_user_ns)
3977 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3978 ctx->seen |= SHMEM_SEEN_QUOTA;
3979 ctx->quota_types |= QTYPE_MASK_GRP;
3981 case Opt_usrquota_block_hardlimit:
3982 size = memparse(param->string, &rest);
3985 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
3987 "User quota block hardlimit too large.");
3988 ctx->qlimits.usrquota_bhardlimit = size;
3990 case Opt_grpquota_block_hardlimit:
3991 size = memparse(param->string, &rest);
3994 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
3996 "Group quota block hardlimit too large.");
3997 ctx->qlimits.grpquota_bhardlimit = size;
3999 case Opt_usrquota_inode_hardlimit:
4000 size = memparse(param->string, &rest);
4003 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4005 "User quota inode hardlimit too large.");
4006 ctx->qlimits.usrquota_ihardlimit = size;
4008 case Opt_grpquota_inode_hardlimit:
4009 size = memparse(param->string, &rest);
4012 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4014 "Group quota inode hardlimit too large.");
4015 ctx->qlimits.grpquota_ihardlimit = size;
4020 unsupported_parameter:
4021 return invalfc(fc, "Unsupported parameter '%s'", param->key);
4023 return invalfc(fc, "Bad value for '%s'", param->key);
4026 static int shmem_parse_options(struct fs_context *fc, void *data)
4028 char *options = data;
4031 int err = security_sb_eat_lsm_opts(options, &fc->security);
4036 while (options != NULL) {
4037 char *this_char = options;
4040 * NUL-terminate this option: unfortunately,
4041 * mount options form a comma-separated list,
4042 * but mpol's nodelist may also contain commas.
4044 options = strchr(options, ',');
4045 if (options == NULL)
4048 if (!isdigit(*options)) {
4054 char *value = strchr(this_char, '=');
4060 len = strlen(value);
4062 err = vfs_parse_fs_string(fc, this_char, value, len);
4071 * Reconfigure a shmem filesystem.
4073 static int shmem_reconfigure(struct fs_context *fc)
4075 struct shmem_options *ctx = fc->fs_private;
4076 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
4077 unsigned long used_isp;
4078 struct mempolicy *mpol = NULL;
4081 raw_spin_lock(&sbinfo->stat_lock);
4082 used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace;
4084 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
4085 if (!sbinfo->max_blocks) {
4086 err = "Cannot retroactively limit size";
4089 if (percpu_counter_compare(&sbinfo->used_blocks,
4091 err = "Too small a size for current use";
4095 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
4096 if (!sbinfo->max_inodes) {
4097 err = "Cannot retroactively limit inodes";
4100 if (ctx->inodes * BOGO_INODE_SIZE < used_isp) {
4101 err = "Too few inodes for current use";
4106 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
4107 sbinfo->next_ino > UINT_MAX) {
4108 err = "Current inum too high to switch to 32-bit inums";
4111 if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
4112 err = "Cannot disable swap on remount";
4115 if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
4116 err = "Cannot enable swap on remount if it was disabled on first mount";
4120 if (ctx->seen & SHMEM_SEEN_QUOTA &&
4121 !sb_any_quota_loaded(fc->root->d_sb)) {
4122 err = "Cannot enable quota on remount";
4126 #ifdef CONFIG_TMPFS_QUOTA
4127 #define CHANGED_LIMIT(name) \
4128 (ctx->qlimits.name## hardlimit && \
4129 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4131 if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
4132 CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
4133 err = "Cannot change global quota limit on remount";
4136 #endif /* CONFIG_TMPFS_QUOTA */
4138 if (ctx->seen & SHMEM_SEEN_HUGE)
4139 sbinfo->huge = ctx->huge;
4140 if (ctx->seen & SHMEM_SEEN_INUMS)
4141 sbinfo->full_inums = ctx->full_inums;
4142 if (ctx->seen & SHMEM_SEEN_BLOCKS)
4143 sbinfo->max_blocks = ctx->blocks;
4144 if (ctx->seen & SHMEM_SEEN_INODES) {
4145 sbinfo->max_inodes = ctx->inodes;
4146 sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp;
4150 * Preserve previous mempolicy unless mpol remount option was specified.
4153 mpol = sbinfo->mpol;
4154 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
4159 sbinfo->noswap = true;
4161 raw_spin_unlock(&sbinfo->stat_lock);
4165 raw_spin_unlock(&sbinfo->stat_lock);
4166 return invalfc(fc, "%s", err);
4169 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
4171 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
4172 struct mempolicy *mpol;
4174 if (sbinfo->max_blocks != shmem_default_max_blocks())
4175 seq_printf(seq, ",size=%luk",
4176 sbinfo->max_blocks << (PAGE_SHIFT - 10));
4177 if (sbinfo->max_inodes != shmem_default_max_inodes())
4178 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
4179 if (sbinfo->mode != (0777 | S_ISVTX))
4180 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
4181 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
4182 seq_printf(seq, ",uid=%u",
4183 from_kuid_munged(&init_user_ns, sbinfo->uid));
4184 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
4185 seq_printf(seq, ",gid=%u",
4186 from_kgid_munged(&init_user_ns, sbinfo->gid));
4189 * Showing inode{64,32} might be useful even if it's the system default,
4190 * since then people don't have to resort to checking both here and
4191 * /proc/config.gz to confirm 64-bit inums were successfully applied
4192 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4194 * We hide it when inode64 isn't the default and we are using 32-bit
4195 * inodes, since that probably just means the feature isn't even under
4200 * +-----------------+-----------------+
4201 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4202 * +------------------+-----------------+-----------------+
4203 * | full_inums=true | show | show |
4204 * | full_inums=false | show | hide |
4205 * +------------------+-----------------+-----------------+
4208 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
4209 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
4210 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4211 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4213 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
4215 mpol = shmem_get_sbmpol(sbinfo);
4216 shmem_show_mpol(seq, mpol);
4219 seq_printf(seq, ",noswap");
4223 #endif /* CONFIG_TMPFS */
4225 static void shmem_put_super(struct super_block *sb)
4227 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
4229 #ifdef CONFIG_TMPFS_QUOTA
4230 shmem_disable_quotas(sb);
4232 free_percpu(sbinfo->ino_batch);
4233 percpu_counter_destroy(&sbinfo->used_blocks);
4234 mpol_put(sbinfo->mpol);
4236 sb->s_fs_info = NULL;
4239 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
4241 struct shmem_options *ctx = fc->fs_private;
4242 struct inode *inode;
4243 struct shmem_sb_info *sbinfo;
4244 int error = -ENOMEM;
4246 /* Round up to L1_CACHE_BYTES to resist false sharing */
4247 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
4248 L1_CACHE_BYTES), GFP_KERNEL);
4252 sb->s_fs_info = sbinfo;
4256 * Per default we only allow half of the physical ram per
4257 * tmpfs instance, limiting inodes to one per page of lowmem;
4258 * but the internal instance is left unlimited.
4260 if (!(sb->s_flags & SB_KERNMOUNT)) {
4261 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
4262 ctx->blocks = shmem_default_max_blocks();
4263 if (!(ctx->seen & SHMEM_SEEN_INODES))
4264 ctx->inodes = shmem_default_max_inodes();
4265 if (!(ctx->seen & SHMEM_SEEN_INUMS))
4266 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
4267 sbinfo->noswap = ctx->noswap;
4269 sb->s_flags |= SB_NOUSER;
4271 sb->s_export_op = &shmem_export_ops;
4272 sb->s_flags |= SB_NOSEC | SB_I_VERSION;
4274 sb->s_flags |= SB_NOUSER;
4276 sbinfo->max_blocks = ctx->blocks;
4277 sbinfo->max_inodes = ctx->inodes;
4278 sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE;
4279 if (sb->s_flags & SB_KERNMOUNT) {
4280 sbinfo->ino_batch = alloc_percpu(ino_t);
4281 if (!sbinfo->ino_batch)
4284 sbinfo->uid = ctx->uid;
4285 sbinfo->gid = ctx->gid;
4286 sbinfo->full_inums = ctx->full_inums;
4287 sbinfo->mode = ctx->mode;
4288 sbinfo->huge = ctx->huge;
4289 sbinfo->mpol = ctx->mpol;
4292 raw_spin_lock_init(&sbinfo->stat_lock);
4293 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
4295 spin_lock_init(&sbinfo->shrinklist_lock);
4296 INIT_LIST_HEAD(&sbinfo->shrinklist);
4298 sb->s_maxbytes = MAX_LFS_FILESIZE;
4299 sb->s_blocksize = PAGE_SIZE;
4300 sb->s_blocksize_bits = PAGE_SHIFT;
4301 sb->s_magic = TMPFS_MAGIC;
4302 sb->s_op = &shmem_ops;
4303 sb->s_time_gran = 1;
4304 #ifdef CONFIG_TMPFS_XATTR
4305 sb->s_xattr = shmem_xattr_handlers;
4307 #ifdef CONFIG_TMPFS_POSIX_ACL
4308 sb->s_flags |= SB_POSIXACL;
4310 uuid_gen(&sb->s_uuid);
4312 #ifdef CONFIG_TMPFS_QUOTA
4313 if (ctx->seen & SHMEM_SEEN_QUOTA) {
4314 sb->dq_op = &shmem_quota_operations;
4315 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4316 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4318 /* Copy the default limits from ctx into sbinfo */
4319 memcpy(&sbinfo->qlimits, &ctx->qlimits,
4320 sizeof(struct shmem_quota_limits));
4322 if (shmem_enable_quotas(sb, ctx->quota_types))
4325 #endif /* CONFIG_TMPFS_QUOTA */
4327 inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, S_IFDIR | sbinfo->mode, 0,
4329 if (IS_ERR(inode)) {
4330 error = PTR_ERR(inode);
4333 inode->i_uid = sbinfo->uid;
4334 inode->i_gid = sbinfo->gid;
4335 sb->s_root = d_make_root(inode);
4341 shmem_put_super(sb);
4345 static int shmem_get_tree(struct fs_context *fc)
4347 return get_tree_nodev(fc, shmem_fill_super);
4350 static void shmem_free_fc(struct fs_context *fc)
4352 struct shmem_options *ctx = fc->fs_private;
4355 mpol_put(ctx->mpol);
4360 static const struct fs_context_operations shmem_fs_context_ops = {
4361 .free = shmem_free_fc,
4362 .get_tree = shmem_get_tree,
4364 .parse_monolithic = shmem_parse_options,
4365 .parse_param = shmem_parse_one,
4366 .reconfigure = shmem_reconfigure,
4370 static struct kmem_cache *shmem_inode_cachep;
4372 static struct inode *shmem_alloc_inode(struct super_block *sb)
4374 struct shmem_inode_info *info;
4375 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
4378 return &info->vfs_inode;
4381 static void shmem_free_in_core_inode(struct inode *inode)
4383 if (S_ISLNK(inode->i_mode))
4384 kfree(inode->i_link);
4385 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4388 static void shmem_destroy_inode(struct inode *inode)
4390 if (S_ISREG(inode->i_mode))
4391 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
4392 if (S_ISDIR(inode->i_mode))
4393 simple_offset_destroy(shmem_get_offset_ctx(inode));
4396 static void shmem_init_inode(void *foo)
4398 struct shmem_inode_info *info = foo;
4399 inode_init_once(&info->vfs_inode);
4402 static void shmem_init_inodecache(void)
4404 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4405 sizeof(struct shmem_inode_info),
4406 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
4409 static void shmem_destroy_inodecache(void)
4411 kmem_cache_destroy(shmem_inode_cachep);
4414 /* Keep the page in page cache instead of truncating it */
4415 static int shmem_error_remove_page(struct address_space *mapping,
4421 const struct address_space_operations shmem_aops = {
4422 .writepage = shmem_writepage,
4423 .dirty_folio = noop_dirty_folio,
4425 .write_begin = shmem_write_begin,
4426 .write_end = shmem_write_end,
4428 #ifdef CONFIG_MIGRATION
4429 .migrate_folio = migrate_folio,
4431 .error_remove_page = shmem_error_remove_page,
4433 EXPORT_SYMBOL(shmem_aops);
4435 static const struct file_operations shmem_file_operations = {
4437 .open = generic_file_open,
4438 .get_unmapped_area = shmem_get_unmapped_area,
4440 .llseek = shmem_file_llseek,
4441 .read_iter = shmem_file_read_iter,
4442 .write_iter = generic_file_write_iter,
4443 .fsync = noop_fsync,
4444 .splice_read = shmem_file_splice_read,
4445 .splice_write = iter_file_splice_write,
4446 .fallocate = shmem_fallocate,
4450 static const struct inode_operations shmem_inode_operations = {
4451 .getattr = shmem_getattr,
4452 .setattr = shmem_setattr,
4453 #ifdef CONFIG_TMPFS_XATTR
4454 .listxattr = shmem_listxattr,
4455 .set_acl = simple_set_acl,
4456 .fileattr_get = shmem_fileattr_get,
4457 .fileattr_set = shmem_fileattr_set,
4461 static const struct inode_operations shmem_dir_inode_operations = {
4463 .getattr = shmem_getattr,
4464 .create = shmem_create,
4465 .lookup = simple_lookup,
4467 .unlink = shmem_unlink,
4468 .symlink = shmem_symlink,
4469 .mkdir = shmem_mkdir,
4470 .rmdir = shmem_rmdir,
4471 .mknod = shmem_mknod,
4472 .rename = shmem_rename2,
4473 .tmpfile = shmem_tmpfile,
4474 .get_offset_ctx = shmem_get_offset_ctx,
4476 #ifdef CONFIG_TMPFS_XATTR
4477 .listxattr = shmem_listxattr,
4478 .fileattr_get = shmem_fileattr_get,
4479 .fileattr_set = shmem_fileattr_set,
4481 #ifdef CONFIG_TMPFS_POSIX_ACL
4482 .setattr = shmem_setattr,
4483 .set_acl = simple_set_acl,
4487 static const struct inode_operations shmem_special_inode_operations = {
4488 .getattr = shmem_getattr,
4489 #ifdef CONFIG_TMPFS_XATTR
4490 .listxattr = shmem_listxattr,
4492 #ifdef CONFIG_TMPFS_POSIX_ACL
4493 .setattr = shmem_setattr,
4494 .set_acl = simple_set_acl,
4498 static const struct super_operations shmem_ops = {
4499 .alloc_inode = shmem_alloc_inode,
4500 .free_inode = shmem_free_in_core_inode,
4501 .destroy_inode = shmem_destroy_inode,
4503 .statfs = shmem_statfs,
4504 .show_options = shmem_show_options,
4506 #ifdef CONFIG_TMPFS_QUOTA
4507 .get_dquots = shmem_get_dquots,
4509 .evict_inode = shmem_evict_inode,
4510 .drop_inode = generic_delete_inode,
4511 .put_super = shmem_put_super,
4512 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4513 .nr_cached_objects = shmem_unused_huge_count,
4514 .free_cached_objects = shmem_unused_huge_scan,
4518 static const struct vm_operations_struct shmem_vm_ops = {
4519 .fault = shmem_fault,
4520 .map_pages = filemap_map_pages,
4522 .set_policy = shmem_set_policy,
4523 .get_policy = shmem_get_policy,
4527 static const struct vm_operations_struct shmem_anon_vm_ops = {
4528 .fault = shmem_fault,
4529 .map_pages = filemap_map_pages,
4531 .set_policy = shmem_set_policy,
4532 .get_policy = shmem_get_policy,
4536 int shmem_init_fs_context(struct fs_context *fc)
4538 struct shmem_options *ctx;
4540 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4544 ctx->mode = 0777 | S_ISVTX;
4545 ctx->uid = current_fsuid();
4546 ctx->gid = current_fsgid();
4548 fc->fs_private = ctx;
4549 fc->ops = &shmem_fs_context_ops;
4553 static struct file_system_type shmem_fs_type = {
4554 .owner = THIS_MODULE,
4556 .init_fs_context = shmem_init_fs_context,
4558 .parameters = shmem_fs_parameters,
4560 .kill_sb = kill_litter_super,
4562 .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4564 .fs_flags = FS_USERNS_MOUNT,
4568 void __init shmem_init(void)
4572 shmem_init_inodecache();
4574 #ifdef CONFIG_TMPFS_QUOTA
4575 error = register_quota_format(&shmem_quota_format);
4577 pr_err("Could not register quota format\n");
4582 error = register_filesystem(&shmem_fs_type);
4584 pr_err("Could not register tmpfs\n");
4588 shm_mnt = kern_mount(&shmem_fs_type);
4589 if (IS_ERR(shm_mnt)) {
4590 error = PTR_ERR(shm_mnt);
4591 pr_err("Could not kern_mount tmpfs\n");
4595 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4596 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4597 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4599 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4604 unregister_filesystem(&shmem_fs_type);
4606 #ifdef CONFIG_TMPFS_QUOTA
4607 unregister_quota_format(&shmem_quota_format);
4610 shmem_destroy_inodecache();
4611 shm_mnt = ERR_PTR(error);
4614 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4615 static ssize_t shmem_enabled_show(struct kobject *kobj,
4616 struct kobj_attribute *attr, char *buf)
4618 static const int values[] = {
4620 SHMEM_HUGE_WITHIN_SIZE,
4629 for (i = 0; i < ARRAY_SIZE(values); i++) {
4630 len += sysfs_emit_at(buf, len,
4631 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4633 shmem_format_huge(values[i]));
4636 len += sysfs_emit_at(buf, len, "\n");
4641 static ssize_t shmem_enabled_store(struct kobject *kobj,
4642 struct kobj_attribute *attr, const char *buf, size_t count)
4647 if (count + 1 > sizeof(tmp))
4649 memcpy(tmp, buf, count);
4651 if (count && tmp[count - 1] == '\n')
4652 tmp[count - 1] = '\0';
4654 huge = shmem_parse_huge(tmp);
4655 if (huge == -EINVAL)
4657 if (!has_transparent_hugepage() &&
4658 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4662 if (shmem_huge > SHMEM_HUGE_DENY)
4663 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4667 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
4668 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4670 #else /* !CONFIG_SHMEM */
4673 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4675 * This is intended for small system where the benefits of the full
4676 * shmem code (swap-backed and resource-limited) are outweighed by
4677 * their complexity. On systems without swap this code should be
4678 * effectively equivalent, but much lighter weight.
4681 static struct file_system_type shmem_fs_type = {
4683 .init_fs_context = ramfs_init_fs_context,
4684 .parameters = ramfs_fs_parameters,
4685 .kill_sb = ramfs_kill_sb,
4686 .fs_flags = FS_USERNS_MOUNT,
4689 void __init shmem_init(void)
4691 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4693 shm_mnt = kern_mount(&shmem_fs_type);
4694 BUG_ON(IS_ERR(shm_mnt));
4697 int shmem_unuse(unsigned int type)
4702 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4707 void shmem_unlock_mapping(struct address_space *mapping)
4712 unsigned long shmem_get_unmapped_area(struct file *file,
4713 unsigned long addr, unsigned long len,
4714 unsigned long pgoff, unsigned long flags)
4716 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4720 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4722 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4724 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4726 #define shmem_vm_ops generic_file_vm_ops
4727 #define shmem_anon_vm_ops generic_file_vm_ops
4728 #define shmem_file_operations ramfs_file_operations
4729 #define shmem_acct_size(flags, size) 0
4730 #define shmem_unacct_size(flags, size) do {} while (0)
4732 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block *sb, struct inode *dir,
4733 umode_t mode, dev_t dev, unsigned long flags)
4735 struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
4736 return inode ? inode : ERR_PTR(-ENOSPC);
4739 #endif /* CONFIG_SHMEM */
4743 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4744 unsigned long flags, unsigned int i_flags)
4746 struct inode *inode;
4750 return ERR_CAST(mnt);
4752 if (size < 0 || size > MAX_LFS_FILESIZE)
4753 return ERR_PTR(-EINVAL);
4755 if (shmem_acct_size(flags, size))
4756 return ERR_PTR(-ENOMEM);
4758 if (is_idmapped_mnt(mnt))
4759 return ERR_PTR(-EINVAL);
4761 inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
4762 S_IFREG | S_IRWXUGO, 0, flags);
4764 if (IS_ERR(inode)) {
4765 shmem_unacct_size(flags, size);
4766 return ERR_CAST(inode);
4768 inode->i_flags |= i_flags;
4769 inode->i_size = size;
4770 clear_nlink(inode); /* It is unlinked */
4771 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4773 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4774 &shmem_file_operations);
4781 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4782 * kernel internal. There will be NO LSM permission checks against the
4783 * underlying inode. So users of this interface must do LSM checks at a
4784 * higher layer. The users are the big_key and shm implementations. LSM
4785 * checks are provided at the key or shm level rather than the inode.
4786 * @name: name for dentry (to be seen in /proc/<pid>/maps
4787 * @size: size to be set for the file
4788 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4790 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4792 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4796 * shmem_file_setup - get an unlinked file living in tmpfs
4797 * @name: name for dentry (to be seen in /proc/<pid>/maps
4798 * @size: size to be set for the file
4799 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4801 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4803 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4805 EXPORT_SYMBOL_GPL(shmem_file_setup);
4808 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4809 * @mnt: the tmpfs mount where the file will be created
4810 * @name: name for dentry (to be seen in /proc/<pid>/maps
4811 * @size: size to be set for the file
4812 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4814 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4815 loff_t size, unsigned long flags)
4817 return __shmem_file_setup(mnt, name, size, flags, 0);
4819 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4822 * shmem_zero_setup - setup a shared anonymous mapping
4823 * @vma: the vma to be mmapped is prepared by do_mmap
4825 int shmem_zero_setup(struct vm_area_struct *vma)
4828 loff_t size = vma->vm_end - vma->vm_start;
4831 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4832 * between XFS directory reading and selinux: since this file is only
4833 * accessible to the user through its mapping, use S_PRIVATE flag to
4834 * bypass file security, in the same way as shmem_kernel_file_setup().
4836 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4838 return PTR_ERR(file);
4842 vma->vm_file = file;
4843 vma->vm_ops = &shmem_anon_vm_ops;
4849 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4850 * @mapping: the folio's address_space
4851 * @index: the folio index
4852 * @gfp: the page allocator flags to use if allocating
4854 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4855 * with any new page allocations done using the specified allocation flags.
4856 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4857 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4858 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4860 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4861 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4863 struct folio *shmem_read_folio_gfp(struct address_space *mapping,
4864 pgoff_t index, gfp_t gfp)
4867 struct inode *inode = mapping->host;
4868 struct folio *folio;
4871 BUG_ON(!shmem_mapping(mapping));
4872 error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
4873 gfp, NULL, NULL, NULL);
4875 return ERR_PTR(error);
4877 folio_unlock(folio);
4881 * The tiny !SHMEM case uses ramfs without swap
4883 return mapping_read_folio_gfp(mapping, index, gfp);
4886 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
4888 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4889 pgoff_t index, gfp_t gfp)
4891 struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
4895 return &folio->page;
4897 page = folio_file_page(folio, index);
4898 if (PageHWPoison(page)) {
4900 return ERR_PTR(-EIO);
4905 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);