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 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_rwsem making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc {
102 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
103 pgoff_t start; /* start of range currently being fallocated */
104 pgoff_t next; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
109 struct shmem_options {
110 unsigned long long blocks;
111 unsigned long long inodes;
112 struct mempolicy *mpol;
120 unsigned short quota_types;
121 struct shmem_quota_limits qlimits;
122 #define SHMEM_SEEN_BLOCKS 1
123 #define SHMEM_SEEN_INODES 2
124 #define SHMEM_SEEN_HUGE 4
125 #define SHMEM_SEEN_INUMS 8
126 #define SHMEM_SEEN_NOSWAP 16
127 #define SHMEM_SEEN_QUOTA 32
131 static unsigned long shmem_default_max_blocks(void)
133 return totalram_pages() / 2;
136 static unsigned long shmem_default_max_inodes(void)
138 unsigned long nr_pages = totalram_pages();
140 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
144 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
145 struct folio **foliop, enum sgp_type sgp,
146 gfp_t gfp, struct vm_area_struct *vma,
147 vm_fault_t *fault_type);
149 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
151 return sb->s_fs_info;
155 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
156 * for shared memory and for shared anonymous (/dev/zero) mappings
157 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
158 * consistent with the pre-accounting of private mappings ...
160 static inline int shmem_acct_size(unsigned long flags, loff_t size)
162 return (flags & VM_NORESERVE) ?
163 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
166 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
168 if (!(flags & VM_NORESERVE))
169 vm_unacct_memory(VM_ACCT(size));
172 static inline int shmem_reacct_size(unsigned long flags,
173 loff_t oldsize, loff_t newsize)
175 if (!(flags & VM_NORESERVE)) {
176 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
177 return security_vm_enough_memory_mm(current->mm,
178 VM_ACCT(newsize) - VM_ACCT(oldsize));
179 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
180 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
186 * ... whereas tmpfs objects are accounted incrementally as
187 * pages are allocated, in order to allow large sparse files.
188 * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
189 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
191 static inline int shmem_acct_block(unsigned long flags, long pages)
193 if (!(flags & VM_NORESERVE))
196 return security_vm_enough_memory_mm(current->mm,
197 pages * VM_ACCT(PAGE_SIZE));
200 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
202 if (flags & VM_NORESERVE)
203 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
206 static inline int shmem_inode_acct_block(struct inode *inode, long pages)
208 struct shmem_inode_info *info = SHMEM_I(inode);
209 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
212 if (shmem_acct_block(info->flags, pages))
215 if (sbinfo->max_blocks) {
216 if (percpu_counter_compare(&sbinfo->used_blocks,
217 sbinfo->max_blocks - pages) > 0)
220 err = dquot_alloc_block_nodirty(inode, pages);
224 percpu_counter_add(&sbinfo->used_blocks, pages);
226 err = dquot_alloc_block_nodirty(inode, pages);
234 shmem_unacct_blocks(info->flags, pages);
238 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
240 struct shmem_inode_info *info = SHMEM_I(inode);
241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
243 dquot_free_block_nodirty(inode, pages);
245 if (sbinfo->max_blocks)
246 percpu_counter_sub(&sbinfo->used_blocks, pages);
247 shmem_unacct_blocks(info->flags, pages);
250 static const struct super_operations shmem_ops;
251 const struct address_space_operations shmem_aops;
252 static const struct file_operations shmem_file_operations;
253 static const struct inode_operations shmem_inode_operations;
254 static const struct inode_operations shmem_dir_inode_operations;
255 static const struct inode_operations shmem_special_inode_operations;
256 static const struct vm_operations_struct shmem_vm_ops;
257 static const struct vm_operations_struct shmem_anon_vm_ops;
258 static struct file_system_type shmem_fs_type;
260 bool vma_is_anon_shmem(struct vm_area_struct *vma)
262 return vma->vm_ops == &shmem_anon_vm_ops;
265 bool vma_is_shmem(struct vm_area_struct *vma)
267 return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
270 static LIST_HEAD(shmem_swaplist);
271 static DEFINE_MUTEX(shmem_swaplist_mutex);
273 #ifdef CONFIG_TMPFS_QUOTA
275 static int shmem_enable_quotas(struct super_block *sb,
276 unsigned short quota_types)
280 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
281 for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
282 if (!(quota_types & (1 << type)))
284 err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
285 DQUOT_USAGE_ENABLED |
286 DQUOT_LIMITS_ENABLED);
293 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
295 for (type--; type >= 0; type--)
296 dquot_quota_off(sb, type);
300 static void shmem_disable_quotas(struct super_block *sb)
304 for (type = 0; type < SHMEM_MAXQUOTAS; type++)
305 dquot_quota_off(sb, type);
308 static struct dquot **shmem_get_dquots(struct inode *inode)
310 return SHMEM_I(inode)->i_dquot;
312 #endif /* CONFIG_TMPFS_QUOTA */
315 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
316 * produces a novel ino for the newly allocated inode.
318 * It may also be called when making a hard link to permit the space needed by
319 * each dentry. However, in that case, no new inode number is needed since that
320 * internally draws from another pool of inode numbers (currently global
321 * get_next_ino()). This case is indicated by passing NULL as inop.
323 #define SHMEM_INO_BATCH 1024
324 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
326 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
329 if (!(sb->s_flags & SB_KERNMOUNT)) {
330 raw_spin_lock(&sbinfo->stat_lock);
331 if (sbinfo->max_inodes) {
332 if (!sbinfo->free_inodes) {
333 raw_spin_unlock(&sbinfo->stat_lock);
336 sbinfo->free_inodes--;
339 ino = sbinfo->next_ino++;
340 if (unlikely(is_zero_ino(ino)))
341 ino = sbinfo->next_ino++;
342 if (unlikely(!sbinfo->full_inums &&
345 * Emulate get_next_ino uint wraparound for
348 if (IS_ENABLED(CONFIG_64BIT))
349 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
350 __func__, MINOR(sb->s_dev));
351 sbinfo->next_ino = 1;
352 ino = sbinfo->next_ino++;
356 raw_spin_unlock(&sbinfo->stat_lock);
359 * __shmem_file_setup, one of our callers, is lock-free: it
360 * doesn't hold stat_lock in shmem_reserve_inode since
361 * max_inodes is always 0, and is called from potentially
362 * unknown contexts. As such, use a per-cpu batched allocator
363 * which doesn't require the per-sb stat_lock unless we are at
364 * the batch boundary.
366 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
367 * shmem mounts are not exposed to userspace, so we don't need
368 * to worry about things like glibc compatibility.
372 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
374 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
375 raw_spin_lock(&sbinfo->stat_lock);
376 ino = sbinfo->next_ino;
377 sbinfo->next_ino += SHMEM_INO_BATCH;
378 raw_spin_unlock(&sbinfo->stat_lock);
379 if (unlikely(is_zero_ino(ino)))
390 static void shmem_free_inode(struct super_block *sb)
392 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
393 if (sbinfo->max_inodes) {
394 raw_spin_lock(&sbinfo->stat_lock);
395 sbinfo->free_inodes++;
396 raw_spin_unlock(&sbinfo->stat_lock);
401 * shmem_recalc_inode - recalculate the block usage of an inode
402 * @inode: inode to recalc
404 * We have to calculate the free blocks since the mm can drop
405 * undirtied hole pages behind our back.
407 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
408 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
410 * It has to be called with the spinlock held.
412 static void shmem_recalc_inode(struct inode *inode)
414 struct shmem_inode_info *info = SHMEM_I(inode);
417 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
419 info->alloced -= freed;
420 shmem_inode_unacct_blocks(inode, freed);
424 bool shmem_charge(struct inode *inode, long pages)
426 struct shmem_inode_info *info = SHMEM_I(inode);
427 struct address_space *mapping = inode->i_mapping;
429 if (shmem_inode_acct_block(inode, pages))
432 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
433 xa_lock_irq(&mapping->i_pages);
434 mapping->nrpages += pages;
435 xa_unlock_irq(&mapping->i_pages);
437 spin_lock_irq(&info->lock);
438 info->alloced += pages;
439 shmem_recalc_inode(inode);
440 spin_unlock_irq(&info->lock);
445 void shmem_uncharge(struct inode *inode, long pages)
447 struct shmem_inode_info *info = SHMEM_I(inode);
449 /* nrpages adjustment done by __filemap_remove_folio() or caller */
451 spin_lock_irq(&info->lock);
452 shmem_recalc_inode(inode);
453 /* which has called shmem_inode_unacct_blocks() if necessary */
454 spin_unlock_irq(&info->lock);
458 * Replace item expected in xarray by a new item, while holding xa_lock.
460 static int shmem_replace_entry(struct address_space *mapping,
461 pgoff_t index, void *expected, void *replacement)
463 XA_STATE(xas, &mapping->i_pages, index);
466 VM_BUG_ON(!expected);
467 VM_BUG_ON(!replacement);
468 item = xas_load(&xas);
469 if (item != expected)
471 xas_store(&xas, replacement);
476 * Sometimes, before we decide whether to proceed or to fail, we must check
477 * that an entry was not already brought back from swap by a racing thread.
479 * Checking page is not enough: by the time a SwapCache page is locked, it
480 * might be reused, and again be SwapCache, using the same swap as before.
482 static bool shmem_confirm_swap(struct address_space *mapping,
483 pgoff_t index, swp_entry_t swap)
485 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
489 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
492 * disables huge pages for the mount;
494 * enables huge pages for the mount;
495 * SHMEM_HUGE_WITHIN_SIZE:
496 * only allocate huge pages if the page will be fully within i_size,
497 * also respect fadvise()/madvise() hints;
499 * only allocate huge pages if requested with fadvise()/madvise();
502 #define SHMEM_HUGE_NEVER 0
503 #define SHMEM_HUGE_ALWAYS 1
504 #define SHMEM_HUGE_WITHIN_SIZE 2
505 #define SHMEM_HUGE_ADVISE 3
509 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
512 * disables huge on shm_mnt and all mounts, for emergency use;
514 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
517 #define SHMEM_HUGE_DENY (-1)
518 #define SHMEM_HUGE_FORCE (-2)
520 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
521 /* ifdef here to avoid bloating shmem.o when not necessary */
523 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
525 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
526 struct mm_struct *mm, unsigned long vm_flags)
530 if (!S_ISREG(inode->i_mode))
532 if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
534 if (shmem_huge == SHMEM_HUGE_DENY)
536 if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
539 switch (SHMEM_SB(inode->i_sb)->huge) {
540 case SHMEM_HUGE_ALWAYS:
542 case SHMEM_HUGE_WITHIN_SIZE:
543 index = round_up(index + 1, HPAGE_PMD_NR);
544 i_size = round_up(i_size_read(inode), PAGE_SIZE);
545 if (i_size >> PAGE_SHIFT >= index)
548 case SHMEM_HUGE_ADVISE:
549 if (mm && (vm_flags & VM_HUGEPAGE))
557 #if defined(CONFIG_SYSFS)
558 static int shmem_parse_huge(const char *str)
560 if (!strcmp(str, "never"))
561 return SHMEM_HUGE_NEVER;
562 if (!strcmp(str, "always"))
563 return SHMEM_HUGE_ALWAYS;
564 if (!strcmp(str, "within_size"))
565 return SHMEM_HUGE_WITHIN_SIZE;
566 if (!strcmp(str, "advise"))
567 return SHMEM_HUGE_ADVISE;
568 if (!strcmp(str, "deny"))
569 return SHMEM_HUGE_DENY;
570 if (!strcmp(str, "force"))
571 return SHMEM_HUGE_FORCE;
576 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
577 static const char *shmem_format_huge(int huge)
580 case SHMEM_HUGE_NEVER:
582 case SHMEM_HUGE_ALWAYS:
584 case SHMEM_HUGE_WITHIN_SIZE:
585 return "within_size";
586 case SHMEM_HUGE_ADVISE:
588 case SHMEM_HUGE_DENY:
590 case SHMEM_HUGE_FORCE:
599 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
600 struct shrink_control *sc, unsigned long nr_to_split)
602 LIST_HEAD(list), *pos, *next;
603 LIST_HEAD(to_remove);
605 struct shmem_inode_info *info;
607 unsigned long batch = sc ? sc->nr_to_scan : 128;
610 if (list_empty(&sbinfo->shrinklist))
613 spin_lock(&sbinfo->shrinklist_lock);
614 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
615 info = list_entry(pos, struct shmem_inode_info, shrinklist);
618 inode = igrab(&info->vfs_inode);
620 /* inode is about to be evicted */
622 list_del_init(&info->shrinklist);
626 /* Check if there's anything to gain */
627 if (round_up(inode->i_size, PAGE_SIZE) ==
628 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
629 list_move(&info->shrinklist, &to_remove);
633 list_move(&info->shrinklist, &list);
635 sbinfo->shrinklist_len--;
639 spin_unlock(&sbinfo->shrinklist_lock);
641 list_for_each_safe(pos, next, &to_remove) {
642 info = list_entry(pos, struct shmem_inode_info, shrinklist);
643 inode = &info->vfs_inode;
644 list_del_init(&info->shrinklist);
648 list_for_each_safe(pos, next, &list) {
652 info = list_entry(pos, struct shmem_inode_info, shrinklist);
653 inode = &info->vfs_inode;
655 if (nr_to_split && split >= nr_to_split)
658 index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
659 folio = filemap_get_folio(inode->i_mapping, index);
663 /* No huge page at the end of the file: nothing to split */
664 if (!folio_test_large(folio)) {
670 * Move the inode on the list back to shrinklist if we failed
671 * to lock the page at this time.
673 * Waiting for the lock may lead to deadlock in the
676 if (!folio_trylock(folio)) {
681 ret = split_folio(folio);
685 /* If split failed move the inode on the list back to shrinklist */
691 list_del_init(&info->shrinklist);
695 * Make sure the inode is either on the global list or deleted
696 * from any local list before iput() since it could be deleted
697 * in another thread once we put the inode (then the local list
700 spin_lock(&sbinfo->shrinklist_lock);
701 list_move(&info->shrinklist, &sbinfo->shrinklist);
702 sbinfo->shrinklist_len++;
703 spin_unlock(&sbinfo->shrinklist_lock);
711 static long shmem_unused_huge_scan(struct super_block *sb,
712 struct shrink_control *sc)
714 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
716 if (!READ_ONCE(sbinfo->shrinklist_len))
719 return shmem_unused_huge_shrink(sbinfo, sc, 0);
722 static long shmem_unused_huge_count(struct super_block *sb,
723 struct shrink_control *sc)
725 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
726 return READ_ONCE(sbinfo->shrinklist_len);
728 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
730 #define shmem_huge SHMEM_HUGE_DENY
732 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
733 struct mm_struct *mm, unsigned long vm_flags)
738 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
739 struct shrink_control *sc, unsigned long nr_to_split)
743 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
746 * Like filemap_add_folio, but error if expected item has gone.
748 static int shmem_add_to_page_cache(struct folio *folio,
749 struct address_space *mapping,
750 pgoff_t index, void *expected, gfp_t gfp,
751 struct mm_struct *charge_mm)
753 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
754 long nr = folio_nr_pages(folio);
757 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
758 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
759 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
760 VM_BUG_ON(expected && folio_test_large(folio));
762 folio_ref_add(folio, nr);
763 folio->mapping = mapping;
764 folio->index = index;
766 if (!folio_test_swapcache(folio)) {
767 error = mem_cgroup_charge(folio, charge_mm, gfp);
769 if (folio_test_pmd_mappable(folio)) {
770 count_vm_event(THP_FILE_FALLBACK);
771 count_vm_event(THP_FILE_FALLBACK_CHARGE);
776 folio_throttle_swaprate(folio, gfp);
780 if (expected != xas_find_conflict(&xas)) {
781 xas_set_err(&xas, -EEXIST);
784 if (expected && xas_find_conflict(&xas)) {
785 xas_set_err(&xas, -EEXIST);
788 xas_store(&xas, folio);
791 if (folio_test_pmd_mappable(folio)) {
792 count_vm_event(THP_FILE_ALLOC);
793 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
795 mapping->nrpages += nr;
796 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
797 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
799 xas_unlock_irq(&xas);
800 } while (xas_nomem(&xas, gfp));
802 if (xas_error(&xas)) {
803 error = xas_error(&xas);
809 folio->mapping = NULL;
810 folio_ref_sub(folio, nr);
815 * Like delete_from_page_cache, but substitutes swap for @folio.
817 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
819 struct address_space *mapping = folio->mapping;
820 long nr = folio_nr_pages(folio);
823 xa_lock_irq(&mapping->i_pages);
824 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
825 folio->mapping = NULL;
826 mapping->nrpages -= nr;
827 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
828 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
829 xa_unlock_irq(&mapping->i_pages);
835 * Remove swap entry from page cache, free the swap and its page cache.
837 static int shmem_free_swap(struct address_space *mapping,
838 pgoff_t index, void *radswap)
842 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
845 free_swap_and_cache(radix_to_swp_entry(radswap));
850 * Determine (in bytes) how many of the shmem object's pages mapped by the
851 * given offsets are swapped out.
853 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
854 * as long as the inode doesn't go away and racy results are not a problem.
856 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
857 pgoff_t start, pgoff_t end)
859 XA_STATE(xas, &mapping->i_pages, start);
861 unsigned long swapped = 0;
864 xas_for_each(&xas, page, end - 1) {
865 if (xas_retry(&xas, page))
867 if (xa_is_value(page))
870 if (need_resched()) {
878 return swapped << PAGE_SHIFT;
882 * Determine (in bytes) how many of the shmem object's pages mapped by the
883 * given vma is swapped out.
885 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
886 * as long as the inode doesn't go away and racy results are not a problem.
888 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
890 struct inode *inode = file_inode(vma->vm_file);
891 struct shmem_inode_info *info = SHMEM_I(inode);
892 struct address_space *mapping = inode->i_mapping;
893 unsigned long swapped;
895 /* Be careful as we don't hold info->lock */
896 swapped = READ_ONCE(info->swapped);
899 * The easier cases are when the shmem object has nothing in swap, or
900 * the vma maps it whole. Then we can simply use the stats that we
906 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
907 return swapped << PAGE_SHIFT;
909 /* Here comes the more involved part */
910 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
911 vma->vm_pgoff + vma_pages(vma));
915 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
917 void shmem_unlock_mapping(struct address_space *mapping)
919 struct folio_batch fbatch;
922 folio_batch_init(&fbatch);
924 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
926 while (!mapping_unevictable(mapping) &&
927 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
928 check_move_unevictable_folios(&fbatch);
929 folio_batch_release(&fbatch);
934 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
939 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
940 * beyond i_size, and reports fallocated folios as holes.
942 folio = filemap_get_entry(inode->i_mapping, index);
945 if (!xa_is_value(folio)) {
947 if (folio->mapping == inode->i_mapping)
949 /* The folio has been swapped out */
954 * But read a folio back from swap if any of it is within i_size
955 * (although in some cases this is just a waste of time).
958 shmem_get_folio(inode, index, &folio, SGP_READ);
963 * Remove range of pages and swap entries from page cache, and free them.
964 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
966 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
969 struct address_space *mapping = inode->i_mapping;
970 struct shmem_inode_info *info = SHMEM_I(inode);
971 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
972 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
973 struct folio_batch fbatch;
974 pgoff_t indices[PAGEVEC_SIZE];
977 long nr_swaps_freed = 0;
982 end = -1; /* unsigned, so actually very big */
984 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
985 info->fallocend = start;
987 folio_batch_init(&fbatch);
989 while (index < end && find_lock_entries(mapping, &index, end - 1,
991 for (i = 0; i < folio_batch_count(&fbatch); i++) {
992 folio = fbatch.folios[i];
994 if (xa_is_value(folio)) {
997 nr_swaps_freed += !shmem_free_swap(mapping,
1002 if (!unfalloc || !folio_test_uptodate(folio))
1003 truncate_inode_folio(mapping, folio);
1004 folio_unlock(folio);
1006 folio_batch_remove_exceptionals(&fbatch);
1007 folio_batch_release(&fbatch);
1012 * When undoing a failed fallocate, we want none of the partial folio
1013 * zeroing and splitting below, but shall want to truncate the whole
1014 * folio when !uptodate indicates that it was added by this fallocate,
1015 * even when [lstart, lend] covers only a part of the folio.
1020 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
1021 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
1023 same_folio = lend < folio_pos(folio) + folio_size(folio);
1024 folio_mark_dirty(folio);
1025 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
1026 start = folio->index + folio_nr_pages(folio);
1030 folio_unlock(folio);
1036 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
1038 folio_mark_dirty(folio);
1039 if (!truncate_inode_partial_folio(folio, lstart, lend))
1041 folio_unlock(folio);
1048 while (index < end) {
1051 if (!find_get_entries(mapping, &index, end - 1, &fbatch,
1053 /* If all gone or hole-punch or unfalloc, we're done */
1054 if (index == start || end != -1)
1056 /* But if truncating, restart to make sure all gone */
1060 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1061 folio = fbatch.folios[i];
1063 if (xa_is_value(folio)) {
1066 if (shmem_free_swap(mapping, indices[i], folio)) {
1067 /* Swap was replaced by page: retry */
1077 if (!unfalloc || !folio_test_uptodate(folio)) {
1078 if (folio_mapping(folio) != mapping) {
1079 /* Page was replaced by swap: retry */
1080 folio_unlock(folio);
1084 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1086 truncate_inode_folio(mapping, folio);
1088 folio_unlock(folio);
1090 folio_batch_remove_exceptionals(&fbatch);
1091 folio_batch_release(&fbatch);
1094 spin_lock_irq(&info->lock);
1095 info->swapped -= nr_swaps_freed;
1096 shmem_recalc_inode(inode);
1097 spin_unlock_irq(&info->lock);
1100 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1102 shmem_undo_range(inode, lstart, lend, false);
1103 inode->i_ctime = inode->i_mtime = current_time(inode);
1104 inode_inc_iversion(inode);
1106 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1108 static int shmem_getattr(struct mnt_idmap *idmap,
1109 const struct path *path, struct kstat *stat,
1110 u32 request_mask, unsigned int query_flags)
1112 struct inode *inode = path->dentry->d_inode;
1113 struct shmem_inode_info *info = SHMEM_I(inode);
1115 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1116 spin_lock_irq(&info->lock);
1117 shmem_recalc_inode(inode);
1118 spin_unlock_irq(&info->lock);
1120 if (info->fsflags & FS_APPEND_FL)
1121 stat->attributes |= STATX_ATTR_APPEND;
1122 if (info->fsflags & FS_IMMUTABLE_FL)
1123 stat->attributes |= STATX_ATTR_IMMUTABLE;
1124 if (info->fsflags & FS_NODUMP_FL)
1125 stat->attributes |= STATX_ATTR_NODUMP;
1126 stat->attributes_mask |= (STATX_ATTR_APPEND |
1127 STATX_ATTR_IMMUTABLE |
1129 generic_fillattr(idmap, inode, stat);
1131 if (shmem_is_huge(inode, 0, false, NULL, 0))
1132 stat->blksize = HPAGE_PMD_SIZE;
1134 if (request_mask & STATX_BTIME) {
1135 stat->result_mask |= STATX_BTIME;
1136 stat->btime.tv_sec = info->i_crtime.tv_sec;
1137 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1143 static int shmem_setattr(struct mnt_idmap *idmap,
1144 struct dentry *dentry, struct iattr *attr)
1146 struct inode *inode = d_inode(dentry);
1147 struct shmem_inode_info *info = SHMEM_I(inode);
1149 bool update_mtime = false;
1150 bool update_ctime = true;
1152 error = setattr_prepare(idmap, dentry, attr);
1156 if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1157 if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1162 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1163 loff_t oldsize = inode->i_size;
1164 loff_t newsize = attr->ia_size;
1166 /* protected by i_rwsem */
1167 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1168 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1171 if (newsize != oldsize) {
1172 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1176 i_size_write(inode, newsize);
1177 update_mtime = true;
1179 update_ctime = false;
1181 if (newsize <= oldsize) {
1182 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1183 if (oldsize > holebegin)
1184 unmap_mapping_range(inode->i_mapping,
1187 shmem_truncate_range(inode,
1188 newsize, (loff_t)-1);
1189 /* unmap again to remove racily COWed private pages */
1190 if (oldsize > holebegin)
1191 unmap_mapping_range(inode->i_mapping,
1196 if (is_quota_modification(idmap, inode, attr)) {
1197 error = dquot_initialize(inode);
1202 /* Transfer quota accounting */
1203 if (i_uid_needs_update(idmap, attr, inode) ||
1204 i_gid_needs_update(idmap, attr, inode)) {
1205 error = dquot_transfer(idmap, inode, attr);
1211 setattr_copy(idmap, inode, attr);
1212 if (attr->ia_valid & ATTR_MODE)
1213 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1214 if (!error && update_ctime) {
1215 inode->i_ctime = current_time(inode);
1217 inode->i_mtime = inode->i_ctime;
1218 inode_inc_iversion(inode);
1223 static void shmem_evict_inode(struct inode *inode)
1225 struct shmem_inode_info *info = SHMEM_I(inode);
1226 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1228 if (shmem_mapping(inode->i_mapping)) {
1229 shmem_unacct_size(info->flags, inode->i_size);
1231 mapping_set_exiting(inode->i_mapping);
1232 shmem_truncate_range(inode, 0, (loff_t)-1);
1233 if (!list_empty(&info->shrinklist)) {
1234 spin_lock(&sbinfo->shrinklist_lock);
1235 if (!list_empty(&info->shrinklist)) {
1236 list_del_init(&info->shrinklist);
1237 sbinfo->shrinklist_len--;
1239 spin_unlock(&sbinfo->shrinklist_lock);
1241 while (!list_empty(&info->swaplist)) {
1242 /* Wait while shmem_unuse() is scanning this inode... */
1243 wait_var_event(&info->stop_eviction,
1244 !atomic_read(&info->stop_eviction));
1245 mutex_lock(&shmem_swaplist_mutex);
1246 /* ...but beware of the race if we peeked too early */
1247 if (!atomic_read(&info->stop_eviction))
1248 list_del_init(&info->swaplist);
1249 mutex_unlock(&shmem_swaplist_mutex);
1253 simple_xattrs_free(&info->xattrs);
1254 WARN_ON(inode->i_blocks);
1255 shmem_free_inode(inode->i_sb);
1257 #ifdef CONFIG_TMPFS_QUOTA
1258 dquot_free_inode(inode);
1263 static int shmem_find_swap_entries(struct address_space *mapping,
1264 pgoff_t start, struct folio_batch *fbatch,
1265 pgoff_t *indices, unsigned int type)
1267 XA_STATE(xas, &mapping->i_pages, start);
1268 struct folio *folio;
1272 xas_for_each(&xas, folio, ULONG_MAX) {
1273 if (xas_retry(&xas, folio))
1276 if (!xa_is_value(folio))
1279 entry = radix_to_swp_entry(folio);
1281 * swapin error entries can be found in the mapping. But they're
1282 * deliberately ignored here as we've done everything we can do.
1284 if (swp_type(entry) != type)
1287 indices[folio_batch_count(fbatch)] = xas.xa_index;
1288 if (!folio_batch_add(fbatch, folio))
1291 if (need_resched()) {
1298 return xas.xa_index;
1302 * Move the swapped pages for an inode to page cache. Returns the count
1303 * of pages swapped in, or the error in case of failure.
1305 static int shmem_unuse_swap_entries(struct inode *inode,
1306 struct folio_batch *fbatch, pgoff_t *indices)
1311 struct address_space *mapping = inode->i_mapping;
1313 for (i = 0; i < folio_batch_count(fbatch); i++) {
1314 struct folio *folio = fbatch->folios[i];
1316 if (!xa_is_value(folio))
1318 error = shmem_swapin_folio(inode, indices[i],
1320 mapping_gfp_mask(mapping),
1323 folio_unlock(folio);
1327 if (error == -ENOMEM)
1331 return error ? error : ret;
1335 * If swap found in inode, free it and move page from swapcache to filecache.
1337 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1339 struct address_space *mapping = inode->i_mapping;
1341 struct folio_batch fbatch;
1342 pgoff_t indices[PAGEVEC_SIZE];
1346 folio_batch_init(&fbatch);
1347 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1348 if (folio_batch_count(&fbatch) == 0) {
1353 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1357 start = indices[folio_batch_count(&fbatch) - 1];
1364 * Read all the shared memory data that resides in the swap
1365 * device 'type' back into memory, so the swap device can be
1368 int shmem_unuse(unsigned int type)
1370 struct shmem_inode_info *info, *next;
1373 if (list_empty(&shmem_swaplist))
1376 mutex_lock(&shmem_swaplist_mutex);
1377 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1378 if (!info->swapped) {
1379 list_del_init(&info->swaplist);
1383 * Drop the swaplist mutex while searching the inode for swap;
1384 * but before doing so, make sure shmem_evict_inode() will not
1385 * remove placeholder inode from swaplist, nor let it be freed
1386 * (igrab() would protect from unlink, but not from unmount).
1388 atomic_inc(&info->stop_eviction);
1389 mutex_unlock(&shmem_swaplist_mutex);
1391 error = shmem_unuse_inode(&info->vfs_inode, type);
1394 mutex_lock(&shmem_swaplist_mutex);
1395 next = list_next_entry(info, swaplist);
1397 list_del_init(&info->swaplist);
1398 if (atomic_dec_and_test(&info->stop_eviction))
1399 wake_up_var(&info->stop_eviction);
1403 mutex_unlock(&shmem_swaplist_mutex);
1409 * Move the page from the page cache to the swap cache.
1411 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1413 struct folio *folio = page_folio(page);
1414 struct address_space *mapping = folio->mapping;
1415 struct inode *inode = mapping->host;
1416 struct shmem_inode_info *info = SHMEM_I(inode);
1417 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1422 * Our capabilities prevent regular writeback or sync from ever calling
1423 * shmem_writepage; but a stacking filesystem might use ->writepage of
1424 * its underlying filesystem, in which case tmpfs should write out to
1425 * swap only in response to memory pressure, and not for the writeback
1428 if (WARN_ON_ONCE(!wbc->for_reclaim))
1431 if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
1434 if (!total_swap_pages)
1438 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1439 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1440 * and its shmem_writeback() needs them to be split when swapping.
1442 if (folio_test_large(folio)) {
1443 /* Ensure the subpages are still dirty */
1444 folio_test_set_dirty(folio);
1445 if (split_huge_page(page) < 0)
1447 folio = page_folio(page);
1448 folio_clear_dirty(folio);
1451 index = folio->index;
1454 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1455 * value into swapfile.c, the only way we can correctly account for a
1456 * fallocated folio arriving here is now to initialize it and write it.
1458 * That's okay for a folio already fallocated earlier, but if we have
1459 * not yet completed the fallocation, then (a) we want to keep track
1460 * of this folio in case we have to undo it, and (b) it may not be a
1461 * good idea to continue anyway, once we're pushing into swap. So
1462 * reactivate the folio, and let shmem_fallocate() quit when too many.
1464 if (!folio_test_uptodate(folio)) {
1465 if (inode->i_private) {
1466 struct shmem_falloc *shmem_falloc;
1467 spin_lock(&inode->i_lock);
1468 shmem_falloc = inode->i_private;
1470 !shmem_falloc->waitq &&
1471 index >= shmem_falloc->start &&
1472 index < shmem_falloc->next)
1473 shmem_falloc->nr_unswapped++;
1475 shmem_falloc = NULL;
1476 spin_unlock(&inode->i_lock);
1480 folio_zero_range(folio, 0, folio_size(folio));
1481 flush_dcache_folio(folio);
1482 folio_mark_uptodate(folio);
1485 swap = folio_alloc_swap(folio);
1490 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1491 * if it's not already there. Do it now before the folio is
1492 * moved to swap cache, when its pagelock no longer protects
1493 * the inode from eviction. But don't unlock the mutex until
1494 * we've incremented swapped, because shmem_unuse_inode() will
1495 * prune a !swapped inode from the swaplist under this mutex.
1497 mutex_lock(&shmem_swaplist_mutex);
1498 if (list_empty(&info->swaplist))
1499 list_add(&info->swaplist, &shmem_swaplist);
1501 if (add_to_swap_cache(folio, swap,
1502 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1504 spin_lock_irq(&info->lock);
1505 shmem_recalc_inode(inode);
1507 spin_unlock_irq(&info->lock);
1509 swap_shmem_alloc(swap);
1510 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1512 mutex_unlock(&shmem_swaplist_mutex);
1513 BUG_ON(folio_mapped(folio));
1514 swap_writepage(&folio->page, wbc);
1518 mutex_unlock(&shmem_swaplist_mutex);
1519 put_swap_folio(folio, swap);
1521 folio_mark_dirty(folio);
1522 if (wbc->for_reclaim)
1523 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1524 folio_unlock(folio);
1528 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1529 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1533 if (!mpol || mpol->mode == MPOL_DEFAULT)
1534 return; /* show nothing */
1536 mpol_to_str(buffer, sizeof(buffer), mpol);
1538 seq_printf(seq, ",mpol=%s", buffer);
1541 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1543 struct mempolicy *mpol = NULL;
1545 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1546 mpol = sbinfo->mpol;
1548 raw_spin_unlock(&sbinfo->stat_lock);
1552 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1553 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1556 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1560 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1562 #define vm_policy vm_private_data
1565 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1566 struct shmem_inode_info *info, pgoff_t index)
1568 /* Create a pseudo vma that just contains the policy */
1569 vma_init(vma, NULL);
1570 /* Bias interleave by inode number to distribute better across nodes */
1571 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1572 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1575 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1577 /* Drop reference taken by mpol_shared_policy_lookup() */
1578 mpol_cond_put(vma->vm_policy);
1581 static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1582 struct shmem_inode_info *info, pgoff_t index)
1584 struct vm_area_struct pvma;
1586 struct vm_fault vmf = {
1590 shmem_pseudo_vma_init(&pvma, info, index);
1591 page = swap_cluster_readahead(swap, gfp, &vmf);
1592 shmem_pseudo_vma_destroy(&pvma);
1596 return page_folio(page);
1600 * Make sure huge_gfp is always more limited than limit_gfp.
1601 * Some of the flags set permissions, while others set limitations.
1603 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1605 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1606 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1607 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1608 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1610 /* Allow allocations only from the originally specified zones. */
1611 result |= zoneflags;
1614 * Minimize the result gfp by taking the union with the deny flags,
1615 * and the intersection of the allow flags.
1617 result |= (limit_gfp & denyflags);
1618 result |= (huge_gfp & limit_gfp) & allowflags;
1623 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1624 struct shmem_inode_info *info, pgoff_t index)
1626 struct vm_area_struct pvma;
1627 struct address_space *mapping = info->vfs_inode.i_mapping;
1629 struct folio *folio;
1631 hindex = round_down(index, HPAGE_PMD_NR);
1632 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1636 shmem_pseudo_vma_init(&pvma, info, hindex);
1637 folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1638 shmem_pseudo_vma_destroy(&pvma);
1640 count_vm_event(THP_FILE_FALLBACK);
1644 static struct folio *shmem_alloc_folio(gfp_t gfp,
1645 struct shmem_inode_info *info, pgoff_t index)
1647 struct vm_area_struct pvma;
1648 struct folio *folio;
1650 shmem_pseudo_vma_init(&pvma, info, index);
1651 folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1652 shmem_pseudo_vma_destroy(&pvma);
1657 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1658 pgoff_t index, bool huge)
1660 struct shmem_inode_info *info = SHMEM_I(inode);
1661 struct folio *folio;
1665 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1667 nr = huge ? HPAGE_PMD_NR : 1;
1669 err = shmem_inode_acct_block(inode, nr);
1674 folio = shmem_alloc_hugefolio(gfp, info, index);
1676 folio = shmem_alloc_folio(gfp, info, index);
1678 __folio_set_locked(folio);
1679 __folio_set_swapbacked(folio);
1684 shmem_inode_unacct_blocks(inode, nr);
1686 return ERR_PTR(err);
1690 * When a page is moved from swapcache to shmem filecache (either by the
1691 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1692 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1693 * ignorance of the mapping it belongs to. If that mapping has special
1694 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1695 * we may need to copy to a suitable page before moving to filecache.
1697 * In a future release, this may well be extended to respect cpuset and
1698 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1699 * but for now it is a simple matter of zone.
1701 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1703 return folio_zonenum(folio) > gfp_zone(gfp);
1706 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1707 struct shmem_inode_info *info, pgoff_t index)
1709 struct folio *old, *new;
1710 struct address_space *swap_mapping;
1716 entry = folio_swap_entry(old);
1717 swap_index = swp_offset(entry);
1718 swap_mapping = swap_address_space(entry);
1721 * We have arrived here because our zones are constrained, so don't
1722 * limit chance of success by further cpuset and node constraints.
1724 gfp &= ~GFP_CONSTRAINT_MASK;
1725 VM_BUG_ON_FOLIO(folio_test_large(old), old);
1726 new = shmem_alloc_folio(gfp, info, index);
1731 folio_copy(new, old);
1732 flush_dcache_folio(new);
1734 __folio_set_locked(new);
1735 __folio_set_swapbacked(new);
1736 folio_mark_uptodate(new);
1737 folio_set_swap_entry(new, entry);
1738 folio_set_swapcache(new);
1741 * Our caller will very soon move newpage out of swapcache, but it's
1742 * a nice clean interface for us to replace oldpage by newpage there.
1744 xa_lock_irq(&swap_mapping->i_pages);
1745 error = shmem_replace_entry(swap_mapping, swap_index, old, new);
1747 mem_cgroup_migrate(old, new);
1748 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1749 __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1750 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1751 __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
1753 xa_unlock_irq(&swap_mapping->i_pages);
1755 if (unlikely(error)) {
1757 * Is this possible? I think not, now that our callers check
1758 * both PageSwapCache and page_private after getting page lock;
1759 * but be defensive. Reverse old to newpage for clear and free.
1767 folio_clear_swapcache(old);
1768 old->private = NULL;
1771 folio_put_refs(old, 2);
1775 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1776 struct folio *folio, swp_entry_t swap)
1778 struct address_space *mapping = inode->i_mapping;
1779 struct shmem_inode_info *info = SHMEM_I(inode);
1780 swp_entry_t swapin_error;
1783 swapin_error = make_swapin_error_entry();
1784 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1785 swp_to_radix_entry(swap),
1786 swp_to_radix_entry(swapin_error), 0);
1787 if (old != swp_to_radix_entry(swap))
1790 folio_wait_writeback(folio);
1791 delete_from_swap_cache(folio);
1792 spin_lock_irq(&info->lock);
1794 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks won't
1795 * be 0 when inode is released and thus trigger WARN_ON(inode->i_blocks) in
1796 * shmem_evict_inode.
1800 shmem_recalc_inode(inode);
1801 spin_unlock_irq(&info->lock);
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 spin_lock_irq(&info->lock);
1890 shmem_recalc_inode(inode);
1891 spin_unlock_irq(&info->lock);
1893 if (sgp == SGP_WRITE)
1894 folio_mark_accessed(folio);
1896 delete_from_swap_cache(folio);
1897 folio_mark_dirty(folio);
1899 put_swap_device(si);
1904 if (!shmem_confirm_swap(mapping, index, swap))
1907 shmem_set_folio_swapin_error(inode, index, folio, swap);
1910 folio_unlock(folio);
1913 put_swap_device(si);
1919 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1921 * If we allocate a new one we do not mark it dirty. That's up to the
1922 * vm. If we swap it in we mark it dirty since we also free the swap
1923 * entry since a page cannot live in both the swap and page cache.
1925 * vma, vmf, and fault_type are only supplied by shmem_fault:
1926 * otherwise they are NULL.
1928 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1929 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
1930 struct vm_area_struct *vma, struct vm_fault *vmf,
1931 vm_fault_t *fault_type)
1933 struct address_space *mapping = inode->i_mapping;
1934 struct shmem_inode_info *info = SHMEM_I(inode);
1935 struct shmem_sb_info *sbinfo;
1936 struct mm_struct *charge_mm;
1937 struct folio *folio;
1944 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1947 if (sgp <= SGP_CACHE &&
1948 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1952 sbinfo = SHMEM_SB(inode->i_sb);
1953 charge_mm = vma ? vma->vm_mm : NULL;
1955 folio = filemap_get_entry(mapping, index);
1956 if (folio && vma && userfaultfd_minor(vma)) {
1957 if (!xa_is_value(folio))
1959 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1963 if (xa_is_value(folio)) {
1964 error = shmem_swapin_folio(inode, index, &folio,
1965 sgp, gfp, vma, fault_type);
1966 if (error == -EEXIST)
1976 /* Has the folio been truncated or swapped out? */
1977 if (unlikely(folio->mapping != mapping)) {
1978 folio_unlock(folio);
1982 if (sgp == SGP_WRITE)
1983 folio_mark_accessed(folio);
1984 if (folio_test_uptodate(folio))
1986 /* fallocated folio */
1987 if (sgp != SGP_READ)
1989 folio_unlock(folio);
1994 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1995 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1998 if (sgp == SGP_READ)
2000 if (sgp == SGP_NOALLOC)
2004 * Fast cache lookup and swap lookup did not find it: allocate.
2007 if (vma && userfaultfd_missing(vma)) {
2008 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
2012 if (!shmem_is_huge(inode, index, false,
2013 vma ? vma->vm_mm : NULL, vma ? vma->vm_flags : 0))
2016 huge_gfp = vma_thp_gfp_mask(vma);
2017 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
2018 folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
2019 if (IS_ERR(folio)) {
2021 folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
2023 if (IS_ERR(folio)) {
2026 error = PTR_ERR(folio);
2028 if (error != -ENOSPC)
2031 * Try to reclaim some space by splitting a large folio
2032 * beyond i_size on the filesystem.
2037 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
2038 if (ret == SHRINK_STOP)
2046 hindex = round_down(index, folio_nr_pages(folio));
2048 if (sgp == SGP_WRITE)
2049 __folio_set_referenced(folio);
2051 error = shmem_add_to_page_cache(folio, mapping, hindex,
2052 NULL, gfp & GFP_RECLAIM_MASK,
2056 folio_add_lru(folio);
2058 spin_lock_irq(&info->lock);
2059 info->alloced += folio_nr_pages(folio);
2060 shmem_recalc_inode(inode);
2061 spin_unlock_irq(&info->lock);
2064 if (folio_test_pmd_mappable(folio) &&
2065 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
2066 folio_next_index(folio) - 1) {
2068 * Part of the large folio is beyond i_size: subject
2069 * to shrink under memory pressure.
2071 spin_lock(&sbinfo->shrinklist_lock);
2073 * _careful to defend against unlocked access to
2074 * ->shrink_list in shmem_unused_huge_shrink()
2076 if (list_empty_careful(&info->shrinklist)) {
2077 list_add_tail(&info->shrinklist,
2078 &sbinfo->shrinklist);
2079 sbinfo->shrinklist_len++;
2081 spin_unlock(&sbinfo->shrinklist_lock);
2085 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2087 if (sgp == SGP_FALLOC)
2091 * Let SGP_WRITE caller clear ends if write does not fill folio;
2092 * but SGP_FALLOC on a folio fallocated earlier must initialize
2093 * it now, lest undo on failure cancel our earlier guarantee.
2095 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2096 long i, n = folio_nr_pages(folio);
2098 for (i = 0; i < n; i++)
2099 clear_highpage(folio_page(folio, i));
2100 flush_dcache_folio(folio);
2101 folio_mark_uptodate(folio);
2104 /* Perhaps the file has been truncated since we checked */
2105 if (sgp <= SGP_CACHE &&
2106 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2108 folio_clear_dirty(folio);
2109 filemap_remove_folio(folio);
2110 spin_lock_irq(&info->lock);
2111 shmem_recalc_inode(inode);
2112 spin_unlock_irq(&info->lock);
2125 shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
2127 if (folio_test_large(folio)) {
2128 folio_unlock(folio);
2134 folio_unlock(folio);
2137 if (error == -ENOSPC && !once++) {
2138 spin_lock_irq(&info->lock);
2139 shmem_recalc_inode(inode);
2140 spin_unlock_irq(&info->lock);
2143 if (error == -EEXIST)
2148 int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
2151 return shmem_get_folio_gfp(inode, index, foliop, sgp,
2152 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
2156 * This is like autoremove_wake_function, but it removes the wait queue
2157 * entry unconditionally - even if something else had already woken the
2160 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2162 int ret = default_wake_function(wait, mode, sync, key);
2163 list_del_init(&wait->entry);
2167 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2169 struct vm_area_struct *vma = vmf->vma;
2170 struct inode *inode = file_inode(vma->vm_file);
2171 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2172 struct folio *folio = NULL;
2174 vm_fault_t ret = VM_FAULT_LOCKED;
2177 * Trinity finds that probing a hole which tmpfs is punching can
2178 * prevent the hole-punch from ever completing: which in turn
2179 * locks writers out with its hold on i_rwsem. So refrain from
2180 * faulting pages into the hole while it's being punched. Although
2181 * shmem_undo_range() does remove the additions, it may be unable to
2182 * keep up, as each new page needs its own unmap_mapping_range() call,
2183 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2185 * It does not matter if we sometimes reach this check just before the
2186 * hole-punch begins, so that one fault then races with the punch:
2187 * we just need to make racing faults a rare case.
2189 * The implementation below would be much simpler if we just used a
2190 * standard mutex or completion: but we cannot take i_rwsem in fault,
2191 * and bloating every shmem inode for this unlikely case would be sad.
2193 if (unlikely(inode->i_private)) {
2194 struct shmem_falloc *shmem_falloc;
2196 spin_lock(&inode->i_lock);
2197 shmem_falloc = inode->i_private;
2199 shmem_falloc->waitq &&
2200 vmf->pgoff >= shmem_falloc->start &&
2201 vmf->pgoff < shmem_falloc->next) {
2203 wait_queue_head_t *shmem_falloc_waitq;
2204 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2206 ret = VM_FAULT_NOPAGE;
2207 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2209 ret = VM_FAULT_RETRY;
2211 shmem_falloc_waitq = shmem_falloc->waitq;
2212 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2213 TASK_UNINTERRUPTIBLE);
2214 spin_unlock(&inode->i_lock);
2218 * shmem_falloc_waitq points into the shmem_fallocate()
2219 * stack of the hole-punching task: shmem_falloc_waitq
2220 * is usually invalid by the time we reach here, but
2221 * finish_wait() does not dereference it in that case;
2222 * though i_lock needed lest racing with wake_up_all().
2224 spin_lock(&inode->i_lock);
2225 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2226 spin_unlock(&inode->i_lock);
2232 spin_unlock(&inode->i_lock);
2235 err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
2236 gfp, vma, vmf, &ret);
2238 return vmf_error(err);
2240 vmf->page = folio_file_page(folio, vmf->pgoff);
2244 unsigned long shmem_get_unmapped_area(struct file *file,
2245 unsigned long uaddr, unsigned long len,
2246 unsigned long pgoff, unsigned long flags)
2248 unsigned long (*get_area)(struct file *,
2249 unsigned long, unsigned long, unsigned long, unsigned long);
2251 unsigned long offset;
2252 unsigned long inflated_len;
2253 unsigned long inflated_addr;
2254 unsigned long inflated_offset;
2256 if (len > TASK_SIZE)
2259 get_area = current->mm->get_unmapped_area;
2260 addr = get_area(file, uaddr, len, pgoff, flags);
2262 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2264 if (IS_ERR_VALUE(addr))
2266 if (addr & ~PAGE_MASK)
2268 if (addr > TASK_SIZE - len)
2271 if (shmem_huge == SHMEM_HUGE_DENY)
2273 if (len < HPAGE_PMD_SIZE)
2275 if (flags & MAP_FIXED)
2278 * Our priority is to support MAP_SHARED mapped hugely;
2279 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2280 * But if caller specified an address hint and we allocated area there
2281 * successfully, respect that as before.
2286 if (shmem_huge != SHMEM_HUGE_FORCE) {
2287 struct super_block *sb;
2290 VM_BUG_ON(file->f_op != &shmem_file_operations);
2291 sb = file_inode(file)->i_sb;
2294 * Called directly from mm/mmap.c, or drivers/char/mem.c
2295 * for "/dev/zero", to create a shared anonymous object.
2297 if (IS_ERR(shm_mnt))
2299 sb = shm_mnt->mnt_sb;
2301 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2305 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2306 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2308 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2311 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2312 if (inflated_len > TASK_SIZE)
2314 if (inflated_len < len)
2317 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2318 if (IS_ERR_VALUE(inflated_addr))
2320 if (inflated_addr & ~PAGE_MASK)
2323 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2324 inflated_addr += offset - inflated_offset;
2325 if (inflated_offset > offset)
2326 inflated_addr += HPAGE_PMD_SIZE;
2328 if (inflated_addr > TASK_SIZE - len)
2330 return inflated_addr;
2334 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2336 struct inode *inode = file_inode(vma->vm_file);
2337 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2340 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2343 struct inode *inode = file_inode(vma->vm_file);
2346 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2347 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2351 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2353 struct inode *inode = file_inode(file);
2354 struct shmem_inode_info *info = SHMEM_I(inode);
2355 int retval = -ENOMEM;
2358 * What serializes the accesses to info->flags?
2359 * ipc_lock_object() when called from shmctl_do_lock(),
2360 * no serialization needed when called from shm_destroy().
2362 if (lock && !(info->flags & VM_LOCKED)) {
2363 if (!user_shm_lock(inode->i_size, ucounts))
2365 info->flags |= VM_LOCKED;
2366 mapping_set_unevictable(file->f_mapping);
2368 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2369 user_shm_unlock(inode->i_size, ucounts);
2370 info->flags &= ~VM_LOCKED;
2371 mapping_clear_unevictable(file->f_mapping);
2379 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2381 struct inode *inode = file_inode(file);
2382 struct shmem_inode_info *info = SHMEM_I(inode);
2385 ret = seal_check_future_write(info->seals, vma);
2389 /* arm64 - allow memory tagging on RAM-based files */
2390 vm_flags_set(vma, VM_MTE_ALLOWED);
2392 file_accessed(file);
2393 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2395 vma->vm_ops = &shmem_vm_ops;
2397 vma->vm_ops = &shmem_anon_vm_ops;
2401 #ifdef CONFIG_TMPFS_XATTR
2402 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2405 * chattr's fsflags are unrelated to extended attributes,
2406 * but tmpfs has chosen to enable them under the same config option.
2408 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2410 unsigned int i_flags = 0;
2412 if (fsflags & FS_NOATIME_FL)
2413 i_flags |= S_NOATIME;
2414 if (fsflags & FS_APPEND_FL)
2415 i_flags |= S_APPEND;
2416 if (fsflags & FS_IMMUTABLE_FL)
2417 i_flags |= S_IMMUTABLE;
2419 * But FS_NODUMP_FL does not require any action in i_flags.
2421 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2424 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2427 #define shmem_initxattrs NULL
2430 static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
2432 return &SHMEM_I(inode)->dir_offsets;
2435 static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
2436 struct super_block *sb,
2437 struct inode *dir, umode_t mode,
2438 dev_t dev, unsigned long flags)
2440 struct inode *inode;
2441 struct shmem_inode_info *info;
2442 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2446 err = shmem_reserve_inode(sb, &ino);
2448 return ERR_PTR(err);
2451 inode = new_inode(sb);
2454 shmem_free_inode(sb);
2455 return ERR_PTR(-ENOSPC);
2459 inode_init_owner(idmap, inode, dir, mode);
2460 inode->i_blocks = 0;
2461 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2462 inode->i_generation = get_random_u32();
2463 info = SHMEM_I(inode);
2464 memset(info, 0, (char *)inode - (char *)info);
2465 spin_lock_init(&info->lock);
2466 atomic_set(&info->stop_eviction, 0);
2467 info->seals = F_SEAL_SEAL;
2468 info->flags = flags & VM_NORESERVE;
2469 info->i_crtime = inode->i_mtime;
2470 info->fsflags = (dir == NULL) ? 0 :
2471 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2473 shmem_set_inode_flags(inode, info->fsflags);
2474 INIT_LIST_HEAD(&info->shrinklist);
2475 INIT_LIST_HEAD(&info->swaplist);
2476 INIT_LIST_HEAD(&info->swaplist);
2478 mapping_set_unevictable(inode->i_mapping);
2479 simple_xattrs_init(&info->xattrs);
2480 cache_no_acl(inode);
2481 mapping_set_large_folios(inode->i_mapping);
2483 switch (mode & S_IFMT) {
2485 inode->i_op = &shmem_special_inode_operations;
2486 init_special_inode(inode, mode, dev);
2489 inode->i_mapping->a_ops = &shmem_aops;
2490 inode->i_op = &shmem_inode_operations;
2491 inode->i_fop = &shmem_file_operations;
2492 mpol_shared_policy_init(&info->policy,
2493 shmem_get_sbmpol(sbinfo));
2497 /* Some things misbehave if size == 0 on a directory */
2498 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2499 inode->i_op = &shmem_dir_inode_operations;
2500 inode->i_fop = &simple_offset_dir_operations;
2501 simple_offset_init(shmem_get_offset_ctx(inode));
2505 * Must not load anything in the rbtree,
2506 * mpol_free_shared_policy will not be called.
2508 mpol_shared_policy_init(&info->policy, NULL);
2512 lockdep_annotate_inode_mutex_key(inode);
2516 #ifdef CONFIG_TMPFS_QUOTA
2517 static struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2518 struct super_block *sb, struct inode *dir,
2519 umode_t mode, dev_t dev, unsigned long flags)
2522 struct inode *inode;
2524 inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2528 err = dquot_initialize(inode);
2532 err = dquot_alloc_inode(inode);
2540 inode->i_flags |= S_NOQUOTA;
2542 return ERR_PTR(err);
2545 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2546 struct super_block *sb, struct inode *dir,
2547 umode_t mode, dev_t dev, unsigned long flags)
2549 return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2551 #endif /* CONFIG_TMPFS_QUOTA */
2553 #ifdef CONFIG_USERFAULTFD
2554 int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
2555 struct vm_area_struct *dst_vma,
2556 unsigned long dst_addr,
2557 unsigned long src_addr,
2559 struct folio **foliop)
2561 struct inode *inode = file_inode(dst_vma->vm_file);
2562 struct shmem_inode_info *info = SHMEM_I(inode);
2563 struct address_space *mapping = inode->i_mapping;
2564 gfp_t gfp = mapping_gfp_mask(mapping);
2565 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2567 struct folio *folio;
2571 if (shmem_inode_acct_block(inode, 1)) {
2573 * We may have got a page, returned -ENOENT triggering a retry,
2574 * and now we find ourselves with -ENOMEM. Release the page, to
2575 * avoid a BUG_ON in our caller.
2577 if (unlikely(*foliop)) {
2586 folio = shmem_alloc_folio(gfp, info, pgoff);
2588 goto out_unacct_blocks;
2590 if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
2591 page_kaddr = kmap_local_folio(folio, 0);
2593 * The read mmap_lock is held here. Despite the
2594 * mmap_lock being read recursive a deadlock is still
2595 * possible if a writer has taken a lock. For example:
2597 * process A thread 1 takes read lock on own mmap_lock
2598 * process A thread 2 calls mmap, blocks taking write lock
2599 * process B thread 1 takes page fault, read lock on own mmap lock
2600 * process B thread 2 calls mmap, blocks taking write lock
2601 * process A thread 1 blocks taking read lock on process B
2602 * process B thread 1 blocks taking read lock on process A
2604 * Disable page faults to prevent potential deadlock
2605 * and retry the copy outside the mmap_lock.
2607 pagefault_disable();
2608 ret = copy_from_user(page_kaddr,
2609 (const void __user *)src_addr,
2612 kunmap_local(page_kaddr);
2614 /* fallback to copy_from_user outside mmap_lock */
2615 if (unlikely(ret)) {
2618 /* don't free the page */
2619 goto out_unacct_blocks;
2622 flush_dcache_folio(folio);
2623 } else { /* ZEROPAGE */
2624 clear_user_highpage(&folio->page, dst_addr);
2628 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2632 VM_BUG_ON(folio_test_locked(folio));
2633 VM_BUG_ON(folio_test_swapbacked(folio));
2634 __folio_set_locked(folio);
2635 __folio_set_swapbacked(folio);
2636 __folio_mark_uptodate(folio);
2639 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2640 if (unlikely(pgoff >= max_off))
2643 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2644 gfp & GFP_RECLAIM_MASK, dst_vma->vm_mm);
2648 ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
2649 &folio->page, true, flags);
2651 goto out_delete_from_cache;
2653 spin_lock_irq(&info->lock);
2655 shmem_recalc_inode(inode);
2656 spin_unlock_irq(&info->lock);
2658 folio_unlock(folio);
2660 out_delete_from_cache:
2661 filemap_remove_folio(folio);
2663 folio_unlock(folio);
2666 shmem_inode_unacct_blocks(inode, 1);
2669 #endif /* CONFIG_USERFAULTFD */
2672 static const struct inode_operations shmem_symlink_inode_operations;
2673 static const struct inode_operations shmem_short_symlink_operations;
2676 shmem_write_begin(struct file *file, struct address_space *mapping,
2677 loff_t pos, unsigned len,
2678 struct page **pagep, void **fsdata)
2680 struct inode *inode = mapping->host;
2681 struct shmem_inode_info *info = SHMEM_I(inode);
2682 pgoff_t index = pos >> PAGE_SHIFT;
2683 struct folio *folio;
2686 /* i_rwsem is held by caller */
2687 if (unlikely(info->seals & (F_SEAL_GROW |
2688 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2689 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2691 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2695 ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
2700 *pagep = folio_file_page(folio, index);
2701 if (PageHWPoison(*pagep)) {
2702 folio_unlock(folio);
2712 shmem_write_end(struct file *file, struct address_space *mapping,
2713 loff_t pos, unsigned len, unsigned copied,
2714 struct page *page, void *fsdata)
2716 struct folio *folio = page_folio(page);
2717 struct inode *inode = mapping->host;
2719 if (pos + copied > inode->i_size)
2720 i_size_write(inode, pos + copied);
2722 if (!folio_test_uptodate(folio)) {
2723 if (copied < folio_size(folio)) {
2724 size_t from = offset_in_folio(folio, pos);
2725 folio_zero_segments(folio, 0, from,
2726 from + copied, folio_size(folio));
2728 folio_mark_uptodate(folio);
2730 folio_mark_dirty(folio);
2731 folio_unlock(folio);
2737 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2739 struct file *file = iocb->ki_filp;
2740 struct inode *inode = file_inode(file);
2741 struct address_space *mapping = inode->i_mapping;
2743 unsigned long offset;
2746 loff_t *ppos = &iocb->ki_pos;
2748 index = *ppos >> PAGE_SHIFT;
2749 offset = *ppos & ~PAGE_MASK;
2752 struct folio *folio = NULL;
2753 struct page *page = NULL;
2755 unsigned long nr, ret;
2756 loff_t i_size = i_size_read(inode);
2758 end_index = i_size >> PAGE_SHIFT;
2759 if (index > end_index)
2761 if (index == end_index) {
2762 nr = i_size & ~PAGE_MASK;
2767 error = shmem_get_folio(inode, index, &folio, SGP_READ);
2769 if (error == -EINVAL)
2774 folio_unlock(folio);
2776 page = folio_file_page(folio, index);
2777 if (PageHWPoison(page)) {
2785 * We must evaluate after, since reads (unlike writes)
2786 * are called without i_rwsem protection against truncate
2789 i_size = i_size_read(inode);
2790 end_index = i_size >> PAGE_SHIFT;
2791 if (index == end_index) {
2792 nr = i_size & ~PAGE_MASK;
2803 * If users can be writing to this page using arbitrary
2804 * virtual addresses, take care about potential aliasing
2805 * before reading the page on the kernel side.
2807 if (mapping_writably_mapped(mapping))
2808 flush_dcache_page(page);
2810 * Mark the page accessed if we read the beginning.
2813 folio_mark_accessed(folio);
2815 * Ok, we have the page, and it's up-to-date, so
2816 * now we can copy it to user space...
2818 ret = copy_page_to_iter(page, offset, nr, to);
2821 } else if (user_backed_iter(to)) {
2823 * Copy to user tends to be so well optimized, but
2824 * clear_user() not so much, that it is noticeably
2825 * faster to copy the zero page instead of clearing.
2827 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2830 * But submitting the same page twice in a row to
2831 * splice() - or others? - can result in confusion:
2832 * so don't attempt that optimization on pipes etc.
2834 ret = iov_iter_zero(nr, to);
2839 index += offset >> PAGE_SHIFT;
2840 offset &= ~PAGE_MASK;
2842 if (!iov_iter_count(to))
2851 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2852 file_accessed(file);
2853 return retval ? retval : error;
2856 static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
2857 struct pipe_buffer *buf)
2862 static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
2863 struct pipe_buffer *buf)
2867 static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
2868 struct pipe_buffer *buf)
2873 static const struct pipe_buf_operations zero_pipe_buf_ops = {
2874 .release = zero_pipe_buf_release,
2875 .try_steal = zero_pipe_buf_try_steal,
2876 .get = zero_pipe_buf_get,
2879 static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
2880 loff_t fpos, size_t size)
2882 size_t offset = fpos & ~PAGE_MASK;
2884 size = min_t(size_t, size, PAGE_SIZE - offset);
2886 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2887 struct pipe_buffer *buf = pipe_head_buf(pipe);
2889 *buf = (struct pipe_buffer) {
2890 .ops = &zero_pipe_buf_ops,
2891 .page = ZERO_PAGE(0),
2901 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2902 struct pipe_inode_info *pipe,
2903 size_t len, unsigned int flags)
2905 struct inode *inode = file_inode(in);
2906 struct address_space *mapping = inode->i_mapping;
2907 struct folio *folio = NULL;
2908 size_t total_spliced = 0, used, npages, n, part;
2912 /* Work out how much data we can actually add into the pipe */
2913 used = pipe_occupancy(pipe->head, pipe->tail);
2914 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2915 len = min_t(size_t, len, npages * PAGE_SIZE);
2918 if (*ppos >= i_size_read(inode))
2921 error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio,
2924 if (error == -EINVAL)
2929 folio_unlock(folio);
2931 if (folio_test_hwpoison(folio) ||
2932 (folio_test_large(folio) &&
2933 folio_test_has_hwpoisoned(folio))) {
2940 * i_size must be checked after we know the pages are Uptodate.
2942 * Checking i_size after the check allows us to calculate
2943 * the correct value for "nr", which means the zero-filled
2944 * part of the page is not copied back to userspace (unless
2945 * another truncate extends the file - this is desired though).
2947 isize = i_size_read(inode);
2948 if (unlikely(*ppos >= isize))
2950 part = min_t(loff_t, isize - *ppos, len);
2954 * If users can be writing to this page using arbitrary
2955 * virtual addresses, take care about potential aliasing
2956 * before reading the page on the kernel side.
2958 if (mapping_writably_mapped(mapping))
2959 flush_dcache_folio(folio);
2960 folio_mark_accessed(folio);
2962 * Ok, we have the page, and it's up-to-date, so we can
2963 * now splice it into the pipe.
2965 n = splice_folio_into_pipe(pipe, folio, *ppos, part);
2969 n = splice_zeropage_into_pipe(pipe, *ppos, part);
2977 in->f_ra.prev_pos = *ppos;
2978 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2988 return total_spliced ? total_spliced : error;
2991 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2993 struct address_space *mapping = file->f_mapping;
2994 struct inode *inode = mapping->host;
2996 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2997 return generic_file_llseek_size(file, offset, whence,
2998 MAX_LFS_FILESIZE, i_size_read(inode));
3003 /* We're holding i_rwsem so we can access i_size directly */
3004 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
3006 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
3007 inode_unlock(inode);
3011 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
3014 struct inode *inode = file_inode(file);
3015 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3016 struct shmem_inode_info *info = SHMEM_I(inode);
3017 struct shmem_falloc shmem_falloc;
3018 pgoff_t start, index, end, undo_fallocend;
3021 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3026 if (mode & FALLOC_FL_PUNCH_HOLE) {
3027 struct address_space *mapping = file->f_mapping;
3028 loff_t unmap_start = round_up(offset, PAGE_SIZE);
3029 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
3030 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
3032 /* protected by i_rwsem */
3033 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
3038 shmem_falloc.waitq = &shmem_falloc_waitq;
3039 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
3040 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
3041 spin_lock(&inode->i_lock);
3042 inode->i_private = &shmem_falloc;
3043 spin_unlock(&inode->i_lock);
3045 if ((u64)unmap_end > (u64)unmap_start)
3046 unmap_mapping_range(mapping, unmap_start,
3047 1 + unmap_end - unmap_start, 0);
3048 shmem_truncate_range(inode, offset, offset + len - 1);
3049 /* No need to unmap again: hole-punching leaves COWed pages */
3051 spin_lock(&inode->i_lock);
3052 inode->i_private = NULL;
3053 wake_up_all(&shmem_falloc_waitq);
3054 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
3055 spin_unlock(&inode->i_lock);
3060 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3061 error = inode_newsize_ok(inode, offset + len);
3065 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
3070 start = offset >> PAGE_SHIFT;
3071 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
3072 /* Try to avoid a swapstorm if len is impossible to satisfy */
3073 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
3078 shmem_falloc.waitq = NULL;
3079 shmem_falloc.start = start;
3080 shmem_falloc.next = start;
3081 shmem_falloc.nr_falloced = 0;
3082 shmem_falloc.nr_unswapped = 0;
3083 spin_lock(&inode->i_lock);
3084 inode->i_private = &shmem_falloc;
3085 spin_unlock(&inode->i_lock);
3088 * info->fallocend is only relevant when huge pages might be
3089 * involved: to prevent split_huge_page() freeing fallocated
3090 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3092 undo_fallocend = info->fallocend;
3093 if (info->fallocend < end)
3094 info->fallocend = end;
3096 for (index = start; index < end; ) {
3097 struct folio *folio;
3100 * Good, the fallocate(2) manpage permits EINTR: we may have
3101 * been interrupted because we are using up too much memory.
3103 if (signal_pending(current))
3105 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
3108 error = shmem_get_folio(inode, index, &folio,
3111 info->fallocend = undo_fallocend;
3112 /* Remove the !uptodate folios we added */
3113 if (index > start) {
3114 shmem_undo_range(inode,
3115 (loff_t)start << PAGE_SHIFT,
3116 ((loff_t)index << PAGE_SHIFT) - 1, true);
3122 * Here is a more important optimization than it appears:
3123 * a second SGP_FALLOC on the same large folio will clear it,
3124 * making it uptodate and un-undoable if we fail later.
3126 index = folio_next_index(folio);
3127 /* Beware 32-bit wraparound */
3132 * Inform shmem_writepage() how far we have reached.
3133 * No need for lock or barrier: we have the page lock.
3135 if (!folio_test_uptodate(folio))
3136 shmem_falloc.nr_falloced += index - shmem_falloc.next;
3137 shmem_falloc.next = index;
3140 * If !uptodate, leave it that way so that freeable folios
3141 * can be recognized if we need to rollback on error later.
3142 * But mark it dirty so that memory pressure will swap rather
3143 * than free the folios we are allocating (and SGP_CACHE folios
3144 * might still be clean: we now need to mark those dirty too).
3146 folio_mark_dirty(folio);
3147 folio_unlock(folio);
3152 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3153 i_size_write(inode, offset + len);
3155 spin_lock(&inode->i_lock);
3156 inode->i_private = NULL;
3157 spin_unlock(&inode->i_lock);
3160 file_modified(file);
3161 inode_unlock(inode);
3165 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
3167 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
3169 buf->f_type = TMPFS_MAGIC;
3170 buf->f_bsize = PAGE_SIZE;
3171 buf->f_namelen = NAME_MAX;
3172 if (sbinfo->max_blocks) {
3173 buf->f_blocks = sbinfo->max_blocks;
3175 buf->f_bfree = sbinfo->max_blocks -
3176 percpu_counter_sum(&sbinfo->used_blocks);
3178 if (sbinfo->max_inodes) {
3179 buf->f_files = sbinfo->max_inodes;
3180 buf->f_ffree = sbinfo->free_inodes;
3182 /* else leave those fields 0 like simple_statfs */
3184 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3190 * File creation. Allocate an inode, and we're done..
3193 shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
3194 struct dentry *dentry, umode_t mode, dev_t dev)
3196 struct inode *inode;
3199 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
3202 return PTR_ERR(inode);
3204 error = simple_acl_create(dir, inode);
3207 error = security_inode_init_security(inode, dir,
3209 shmem_initxattrs, NULL);
3210 if (error && error != -EOPNOTSUPP)
3213 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3217 dir->i_size += BOGO_DIRENT_SIZE;
3218 dir->i_ctime = dir->i_mtime = current_time(dir);
3219 inode_inc_iversion(dir);
3220 d_instantiate(dentry, inode);
3221 dget(dentry); /* Extra count - pin the dentry in core */
3230 shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
3231 struct file *file, umode_t mode)
3233 struct inode *inode;
3236 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
3238 if (IS_ERR(inode)) {
3239 error = PTR_ERR(inode);
3243 error = security_inode_init_security(inode, dir,
3245 shmem_initxattrs, NULL);
3246 if (error && error != -EOPNOTSUPP)
3248 error = simple_acl_create(dir, inode);
3251 d_tmpfile(file, inode);
3254 return finish_open_simple(file, error);
3260 static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3261 struct dentry *dentry, umode_t mode)
3265 error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3272 static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
3273 struct dentry *dentry, umode_t mode, bool excl)
3275 return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
3281 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3283 struct inode *inode = d_inode(old_dentry);
3287 * No ordinary (disk based) filesystem counts links as inodes;
3288 * but each new link needs a new dentry, pinning lowmem, and
3289 * tmpfs dentries cannot be pruned until they are unlinked.
3290 * But if an O_TMPFILE file is linked into the tmpfs, the
3291 * first link must skip that, to get the accounting right.
3293 if (inode->i_nlink) {
3294 ret = shmem_reserve_inode(inode->i_sb, NULL);
3299 ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3302 shmem_free_inode(inode->i_sb);
3306 dir->i_size += BOGO_DIRENT_SIZE;
3307 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3308 inode_inc_iversion(dir);
3310 ihold(inode); /* New dentry reference */
3311 dget(dentry); /* Extra pinning count for the created dentry */
3312 d_instantiate(dentry, inode);
3317 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3319 struct inode *inode = d_inode(dentry);
3321 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3322 shmem_free_inode(inode->i_sb);
3324 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3326 dir->i_size -= BOGO_DIRENT_SIZE;
3327 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3328 inode_inc_iversion(dir);
3330 dput(dentry); /* Undo the count from "create" - this does all the work */
3334 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3336 if (!simple_empty(dentry))
3339 drop_nlink(d_inode(dentry));
3341 return shmem_unlink(dir, dentry);
3344 static int shmem_whiteout(struct mnt_idmap *idmap,
3345 struct inode *old_dir, struct dentry *old_dentry)
3347 struct dentry *whiteout;
3350 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3354 error = shmem_mknod(idmap, old_dir, whiteout,
3355 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3361 * Cheat and hash the whiteout while the old dentry is still in
3362 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3364 * d_lookup() will consistently find one of them at this point,
3365 * not sure which one, but that isn't even important.
3372 * The VFS layer already does all the dentry stuff for rename,
3373 * we just have to decrement the usage count for the target if
3374 * it exists so that the VFS layer correctly free's it when it
3377 static int shmem_rename2(struct mnt_idmap *idmap,
3378 struct inode *old_dir, struct dentry *old_dentry,
3379 struct inode *new_dir, struct dentry *new_dentry,
3382 struct inode *inode = d_inode(old_dentry);
3383 int they_are_dirs = S_ISDIR(inode->i_mode);
3386 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3389 if (flags & RENAME_EXCHANGE)
3390 return simple_offset_rename_exchange(old_dir, old_dentry,
3391 new_dir, new_dentry);
3393 if (!simple_empty(new_dentry))
3396 if (flags & RENAME_WHITEOUT) {
3397 error = shmem_whiteout(idmap, old_dir, old_dentry);
3402 simple_offset_remove(shmem_get_offset_ctx(old_dir), old_dentry);
3403 error = simple_offset_add(shmem_get_offset_ctx(new_dir), old_dentry);
3407 if (d_really_is_positive(new_dentry)) {
3408 (void) shmem_unlink(new_dir, new_dentry);
3409 if (they_are_dirs) {
3410 drop_nlink(d_inode(new_dentry));
3411 drop_nlink(old_dir);
3413 } else if (they_are_dirs) {
3414 drop_nlink(old_dir);
3418 old_dir->i_size -= BOGO_DIRENT_SIZE;
3419 new_dir->i_size += BOGO_DIRENT_SIZE;
3420 old_dir->i_ctime = old_dir->i_mtime =
3421 new_dir->i_ctime = new_dir->i_mtime =
3422 inode->i_ctime = current_time(old_dir);
3423 inode_inc_iversion(old_dir);
3424 inode_inc_iversion(new_dir);
3428 static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3429 struct dentry *dentry, const char *symname)
3433 struct inode *inode;
3434 struct folio *folio;
3436 len = strlen(symname) + 1;
3437 if (len > PAGE_SIZE)
3438 return -ENAMETOOLONG;
3440 inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3444 return PTR_ERR(inode);
3446 error = security_inode_init_security(inode, dir, &dentry->d_name,
3447 shmem_initxattrs, NULL);
3448 if (error && error != -EOPNOTSUPP)
3451 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3455 inode->i_size = len-1;
3456 if (len <= SHORT_SYMLINK_LEN) {
3457 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3458 if (!inode->i_link) {
3460 goto out_remove_offset;
3462 inode->i_op = &shmem_short_symlink_operations;
3464 inode_nohighmem(inode);
3465 error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
3467 goto out_remove_offset;
3468 inode->i_mapping->a_ops = &shmem_aops;
3469 inode->i_op = &shmem_symlink_inode_operations;
3470 memcpy(folio_address(folio), symname, len);
3471 folio_mark_uptodate(folio);
3472 folio_mark_dirty(folio);
3473 folio_unlock(folio);
3476 dir->i_size += BOGO_DIRENT_SIZE;
3477 dir->i_ctime = dir->i_mtime = current_time(dir);
3478 inode_inc_iversion(dir);
3479 d_instantiate(dentry, inode);
3484 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3490 static void shmem_put_link(void *arg)
3492 folio_mark_accessed(arg);
3496 static const char *shmem_get_link(struct dentry *dentry,
3497 struct inode *inode,
3498 struct delayed_call *done)
3500 struct folio *folio = NULL;
3504 folio = filemap_get_folio(inode->i_mapping, 0);
3506 return ERR_PTR(-ECHILD);
3507 if (PageHWPoison(folio_page(folio, 0)) ||
3508 !folio_test_uptodate(folio)) {
3510 return ERR_PTR(-ECHILD);
3513 error = shmem_get_folio(inode, 0, &folio, SGP_READ);
3515 return ERR_PTR(error);
3517 return ERR_PTR(-ECHILD);
3518 if (PageHWPoison(folio_page(folio, 0))) {
3519 folio_unlock(folio);
3521 return ERR_PTR(-ECHILD);
3523 folio_unlock(folio);
3525 set_delayed_call(done, shmem_put_link, folio);
3526 return folio_address(folio);
3529 #ifdef CONFIG_TMPFS_XATTR
3531 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3533 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3535 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3540 static int shmem_fileattr_set(struct mnt_idmap *idmap,
3541 struct dentry *dentry, struct fileattr *fa)
3543 struct inode *inode = d_inode(dentry);
3544 struct shmem_inode_info *info = SHMEM_I(inode);
3546 if (fileattr_has_fsx(fa))
3548 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3551 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3552 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3554 shmem_set_inode_flags(inode, info->fsflags);
3555 inode->i_ctime = current_time(inode);
3556 inode_inc_iversion(inode);
3561 * Superblocks without xattr inode operations may get some security.* xattr
3562 * support from the LSM "for free". As soon as we have any other xattrs
3563 * like ACLs, we also need to implement the security.* handlers at
3564 * filesystem level, though.
3568 * Callback for security_inode_init_security() for acquiring xattrs.
3570 static int shmem_initxattrs(struct inode *inode,
3571 const struct xattr *xattr_array,
3574 struct shmem_inode_info *info = SHMEM_I(inode);
3575 const struct xattr *xattr;
3576 struct simple_xattr *new_xattr;
3579 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3580 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3584 len = strlen(xattr->name) + 1;
3585 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3587 if (!new_xattr->name) {
3592 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3593 XATTR_SECURITY_PREFIX_LEN);
3594 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3597 simple_xattr_add(&info->xattrs, new_xattr);
3603 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3604 struct dentry *unused, struct inode *inode,
3605 const char *name, void *buffer, size_t size)
3607 struct shmem_inode_info *info = SHMEM_I(inode);
3609 name = xattr_full_name(handler, name);
3610 return simple_xattr_get(&info->xattrs, name, buffer, size);
3613 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3614 struct mnt_idmap *idmap,
3615 struct dentry *unused, struct inode *inode,
3616 const char *name, const void *value,
3617 size_t size, int flags)
3619 struct shmem_inode_info *info = SHMEM_I(inode);
3622 name = xattr_full_name(handler, name);
3623 err = simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3625 inode->i_ctime = current_time(inode);
3626 inode_inc_iversion(inode);
3631 static const struct xattr_handler shmem_security_xattr_handler = {
3632 .prefix = XATTR_SECURITY_PREFIX,
3633 .get = shmem_xattr_handler_get,
3634 .set = shmem_xattr_handler_set,
3637 static const struct xattr_handler shmem_trusted_xattr_handler = {
3638 .prefix = XATTR_TRUSTED_PREFIX,
3639 .get = shmem_xattr_handler_get,
3640 .set = shmem_xattr_handler_set,
3643 static const struct xattr_handler *shmem_xattr_handlers[] = {
3644 &shmem_security_xattr_handler,
3645 &shmem_trusted_xattr_handler,
3649 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3651 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3652 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3654 #endif /* CONFIG_TMPFS_XATTR */
3656 static const struct inode_operations shmem_short_symlink_operations = {
3657 .getattr = shmem_getattr,
3658 .setattr = shmem_setattr,
3659 .get_link = simple_get_link,
3660 #ifdef CONFIG_TMPFS_XATTR
3661 .listxattr = shmem_listxattr,
3665 static const struct inode_operations shmem_symlink_inode_operations = {
3666 .getattr = shmem_getattr,
3667 .setattr = shmem_setattr,
3668 .get_link = shmem_get_link,
3669 #ifdef CONFIG_TMPFS_XATTR
3670 .listxattr = shmem_listxattr,
3674 static struct dentry *shmem_get_parent(struct dentry *child)
3676 return ERR_PTR(-ESTALE);
3679 static int shmem_match(struct inode *ino, void *vfh)
3683 inum = (inum << 32) | fh[1];
3684 return ino->i_ino == inum && fh[0] == ino->i_generation;
3687 /* Find any alias of inode, but prefer a hashed alias */
3688 static struct dentry *shmem_find_alias(struct inode *inode)
3690 struct dentry *alias = d_find_alias(inode);
3692 return alias ?: d_find_any_alias(inode);
3696 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3697 struct fid *fid, int fh_len, int fh_type)
3699 struct inode *inode;
3700 struct dentry *dentry = NULL;
3707 inum = (inum << 32) | fid->raw[1];
3709 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3710 shmem_match, fid->raw);
3712 dentry = shmem_find_alias(inode);
3719 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3720 struct inode *parent)
3724 return FILEID_INVALID;
3727 if (inode_unhashed(inode)) {
3728 /* Unfortunately insert_inode_hash is not idempotent,
3729 * so as we hash inodes here rather than at creation
3730 * time, we need a lock to ensure we only try
3733 static DEFINE_SPINLOCK(lock);
3735 if (inode_unhashed(inode))
3736 __insert_inode_hash(inode,
3737 inode->i_ino + inode->i_generation);
3741 fh[0] = inode->i_generation;
3742 fh[1] = inode->i_ino;
3743 fh[2] = ((__u64)inode->i_ino) >> 32;
3749 static const struct export_operations shmem_export_ops = {
3750 .get_parent = shmem_get_parent,
3751 .encode_fh = shmem_encode_fh,
3752 .fh_to_dentry = shmem_fh_to_dentry,
3770 Opt_usrquota_block_hardlimit,
3771 Opt_usrquota_inode_hardlimit,
3772 Opt_grpquota_block_hardlimit,
3773 Opt_grpquota_inode_hardlimit,
3776 static const struct constant_table shmem_param_enums_huge[] = {
3777 {"never", SHMEM_HUGE_NEVER },
3778 {"always", SHMEM_HUGE_ALWAYS },
3779 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3780 {"advise", SHMEM_HUGE_ADVISE },
3784 const struct fs_parameter_spec shmem_fs_parameters[] = {
3785 fsparam_u32 ("gid", Opt_gid),
3786 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3787 fsparam_u32oct("mode", Opt_mode),
3788 fsparam_string("mpol", Opt_mpol),
3789 fsparam_string("nr_blocks", Opt_nr_blocks),
3790 fsparam_string("nr_inodes", Opt_nr_inodes),
3791 fsparam_string("size", Opt_size),
3792 fsparam_u32 ("uid", Opt_uid),
3793 fsparam_flag ("inode32", Opt_inode32),
3794 fsparam_flag ("inode64", Opt_inode64),
3795 fsparam_flag ("noswap", Opt_noswap),
3796 #ifdef CONFIG_TMPFS_QUOTA
3797 fsparam_flag ("quota", Opt_quota),
3798 fsparam_flag ("usrquota", Opt_usrquota),
3799 fsparam_flag ("grpquota", Opt_grpquota),
3800 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
3801 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
3802 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
3803 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
3808 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3810 struct shmem_options *ctx = fc->fs_private;
3811 struct fs_parse_result result;
3812 unsigned long long size;
3816 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3822 size = memparse(param->string, &rest);
3824 size <<= PAGE_SHIFT;
3825 size *= totalram_pages();
3831 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3832 ctx->seen |= SHMEM_SEEN_BLOCKS;
3835 ctx->blocks = memparse(param->string, &rest);
3836 if (*rest || ctx->blocks > S64_MAX)
3838 ctx->seen |= SHMEM_SEEN_BLOCKS;
3841 ctx->inodes = memparse(param->string, &rest);
3844 ctx->seen |= SHMEM_SEEN_INODES;
3847 ctx->mode = result.uint_32 & 07777;
3850 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3851 if (!uid_valid(ctx->uid))
3855 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3856 if (!gid_valid(ctx->gid))
3860 ctx->huge = result.uint_32;
3861 if (ctx->huge != SHMEM_HUGE_NEVER &&
3862 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3863 has_transparent_hugepage()))
3864 goto unsupported_parameter;
3865 ctx->seen |= SHMEM_SEEN_HUGE;
3868 if (IS_ENABLED(CONFIG_NUMA)) {
3869 mpol_put(ctx->mpol);
3871 if (mpol_parse_str(param->string, &ctx->mpol))
3875 goto unsupported_parameter;
3877 ctx->full_inums = false;
3878 ctx->seen |= SHMEM_SEEN_INUMS;
3881 if (sizeof(ino_t) < 8) {
3883 "Cannot use inode64 with <64bit inums in kernel\n");
3885 ctx->full_inums = true;
3886 ctx->seen |= SHMEM_SEEN_INUMS;
3889 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
3891 "Turning off swap in unprivileged tmpfs mounts unsupported");
3894 ctx->seen |= SHMEM_SEEN_NOSWAP;
3897 if (fc->user_ns != &init_user_ns)
3898 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3899 ctx->seen |= SHMEM_SEEN_QUOTA;
3900 ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
3903 if (fc->user_ns != &init_user_ns)
3904 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3905 ctx->seen |= SHMEM_SEEN_QUOTA;
3906 ctx->quota_types |= QTYPE_MASK_USR;
3909 if (fc->user_ns != &init_user_ns)
3910 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3911 ctx->seen |= SHMEM_SEEN_QUOTA;
3912 ctx->quota_types |= QTYPE_MASK_GRP;
3914 case Opt_usrquota_block_hardlimit:
3915 size = memparse(param->string, &rest);
3918 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
3920 "User quota block hardlimit too large.");
3921 ctx->qlimits.usrquota_bhardlimit = size;
3923 case Opt_grpquota_block_hardlimit:
3924 size = memparse(param->string, &rest);
3927 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
3929 "Group quota block hardlimit too large.");
3930 ctx->qlimits.grpquota_bhardlimit = size;
3932 case Opt_usrquota_inode_hardlimit:
3933 size = memparse(param->string, &rest);
3936 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
3938 "User quota inode hardlimit too large.");
3939 ctx->qlimits.usrquota_ihardlimit = size;
3941 case Opt_grpquota_inode_hardlimit:
3942 size = memparse(param->string, &rest);
3945 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
3947 "Group quota inode hardlimit too large.");
3948 ctx->qlimits.grpquota_ihardlimit = size;
3953 unsupported_parameter:
3954 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3956 return invalfc(fc, "Bad value for '%s'", param->key);
3959 static int shmem_parse_options(struct fs_context *fc, void *data)
3961 char *options = data;
3964 int err = security_sb_eat_lsm_opts(options, &fc->security);
3969 while (options != NULL) {
3970 char *this_char = options;
3973 * NUL-terminate this option: unfortunately,
3974 * mount options form a comma-separated list,
3975 * but mpol's nodelist may also contain commas.
3977 options = strchr(options, ',');
3978 if (options == NULL)
3981 if (!isdigit(*options)) {
3987 char *value = strchr(this_char, '=');
3993 len = strlen(value);
3995 err = vfs_parse_fs_string(fc, this_char, value, len);
4004 * Reconfigure a shmem filesystem.
4006 * Note that we disallow change from limited->unlimited blocks/inodes while any
4007 * are in use; but we must separately disallow unlimited->limited, because in
4008 * that case we have no record of how much is already in use.
4010 static int shmem_reconfigure(struct fs_context *fc)
4012 struct shmem_options *ctx = fc->fs_private;
4013 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
4014 unsigned long inodes;
4015 struct mempolicy *mpol = NULL;
4018 raw_spin_lock(&sbinfo->stat_lock);
4019 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
4021 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
4022 if (!sbinfo->max_blocks) {
4023 err = "Cannot retroactively limit size";
4026 if (percpu_counter_compare(&sbinfo->used_blocks,
4028 err = "Too small a size for current use";
4032 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
4033 if (!sbinfo->max_inodes) {
4034 err = "Cannot retroactively limit inodes";
4037 if (ctx->inodes < inodes) {
4038 err = "Too few inodes for current use";
4043 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
4044 sbinfo->next_ino > UINT_MAX) {
4045 err = "Current inum too high to switch to 32-bit inums";
4048 if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
4049 err = "Cannot disable swap on remount";
4052 if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
4053 err = "Cannot enable swap on remount if it was disabled on first mount";
4057 if (ctx->seen & SHMEM_SEEN_QUOTA &&
4058 !sb_any_quota_loaded(fc->root->d_sb)) {
4059 err = "Cannot enable quota on remount";
4063 #ifdef CONFIG_TMPFS_QUOTA
4064 #define CHANGED_LIMIT(name) \
4065 (ctx->qlimits.name## hardlimit && \
4066 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4068 if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
4069 CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
4070 err = "Cannot change global quota limit on remount";
4073 #endif /* CONFIG_TMPFS_QUOTA */
4075 if (ctx->seen & SHMEM_SEEN_HUGE)
4076 sbinfo->huge = ctx->huge;
4077 if (ctx->seen & SHMEM_SEEN_INUMS)
4078 sbinfo->full_inums = ctx->full_inums;
4079 if (ctx->seen & SHMEM_SEEN_BLOCKS)
4080 sbinfo->max_blocks = ctx->blocks;
4081 if (ctx->seen & SHMEM_SEEN_INODES) {
4082 sbinfo->max_inodes = ctx->inodes;
4083 sbinfo->free_inodes = ctx->inodes - inodes;
4087 * Preserve previous mempolicy unless mpol remount option was specified.
4090 mpol = sbinfo->mpol;
4091 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
4096 sbinfo->noswap = true;
4098 raw_spin_unlock(&sbinfo->stat_lock);
4102 raw_spin_unlock(&sbinfo->stat_lock);
4103 return invalfc(fc, "%s", err);
4106 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
4108 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
4109 struct mempolicy *mpol;
4111 if (sbinfo->max_blocks != shmem_default_max_blocks())
4112 seq_printf(seq, ",size=%luk",
4113 sbinfo->max_blocks << (PAGE_SHIFT - 10));
4114 if (sbinfo->max_inodes != shmem_default_max_inodes())
4115 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
4116 if (sbinfo->mode != (0777 | S_ISVTX))
4117 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
4118 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
4119 seq_printf(seq, ",uid=%u",
4120 from_kuid_munged(&init_user_ns, sbinfo->uid));
4121 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
4122 seq_printf(seq, ",gid=%u",
4123 from_kgid_munged(&init_user_ns, sbinfo->gid));
4126 * Showing inode{64,32} might be useful even if it's the system default,
4127 * since then people don't have to resort to checking both here and
4128 * /proc/config.gz to confirm 64-bit inums were successfully applied
4129 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4131 * We hide it when inode64 isn't the default and we are using 32-bit
4132 * inodes, since that probably just means the feature isn't even under
4137 * +-----------------+-----------------+
4138 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4139 * +------------------+-----------------+-----------------+
4140 * | full_inums=true | show | show |
4141 * | full_inums=false | show | hide |
4142 * +------------------+-----------------+-----------------+
4145 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
4146 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
4147 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4148 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4150 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
4152 mpol = shmem_get_sbmpol(sbinfo);
4153 shmem_show_mpol(seq, mpol);
4156 seq_printf(seq, ",noswap");
4160 #endif /* CONFIG_TMPFS */
4162 static void shmem_put_super(struct super_block *sb)
4164 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
4166 #ifdef CONFIG_TMPFS_QUOTA
4167 shmem_disable_quotas(sb);
4169 free_percpu(sbinfo->ino_batch);
4170 percpu_counter_destroy(&sbinfo->used_blocks);
4171 mpol_put(sbinfo->mpol);
4173 sb->s_fs_info = NULL;
4176 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
4178 struct shmem_options *ctx = fc->fs_private;
4179 struct inode *inode;
4180 struct shmem_sb_info *sbinfo;
4181 int error = -ENOMEM;
4183 /* Round up to L1_CACHE_BYTES to resist false sharing */
4184 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
4185 L1_CACHE_BYTES), GFP_KERNEL);
4189 sb->s_fs_info = sbinfo;
4193 * Per default we only allow half of the physical ram per
4194 * tmpfs instance, limiting inodes to one per page of lowmem;
4195 * but the internal instance is left unlimited.
4197 if (!(sb->s_flags & SB_KERNMOUNT)) {
4198 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
4199 ctx->blocks = shmem_default_max_blocks();
4200 if (!(ctx->seen & SHMEM_SEEN_INODES))
4201 ctx->inodes = shmem_default_max_inodes();
4202 if (!(ctx->seen & SHMEM_SEEN_INUMS))
4203 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
4204 sbinfo->noswap = ctx->noswap;
4206 sb->s_flags |= SB_NOUSER;
4208 sb->s_export_op = &shmem_export_ops;
4209 sb->s_flags |= SB_NOSEC | SB_I_VERSION;
4211 sb->s_flags |= SB_NOUSER;
4213 sbinfo->max_blocks = ctx->blocks;
4214 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
4215 if (sb->s_flags & SB_KERNMOUNT) {
4216 sbinfo->ino_batch = alloc_percpu(ino_t);
4217 if (!sbinfo->ino_batch)
4220 sbinfo->uid = ctx->uid;
4221 sbinfo->gid = ctx->gid;
4222 sbinfo->full_inums = ctx->full_inums;
4223 sbinfo->mode = ctx->mode;
4224 sbinfo->huge = ctx->huge;
4225 sbinfo->mpol = ctx->mpol;
4228 raw_spin_lock_init(&sbinfo->stat_lock);
4229 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
4231 spin_lock_init(&sbinfo->shrinklist_lock);
4232 INIT_LIST_HEAD(&sbinfo->shrinklist);
4234 sb->s_maxbytes = MAX_LFS_FILESIZE;
4235 sb->s_blocksize = PAGE_SIZE;
4236 sb->s_blocksize_bits = PAGE_SHIFT;
4237 sb->s_magic = TMPFS_MAGIC;
4238 sb->s_op = &shmem_ops;
4239 sb->s_time_gran = 1;
4240 #ifdef CONFIG_TMPFS_XATTR
4241 sb->s_xattr = shmem_xattr_handlers;
4243 #ifdef CONFIG_TMPFS_POSIX_ACL
4244 sb->s_flags |= SB_POSIXACL;
4246 uuid_gen(&sb->s_uuid);
4248 #ifdef CONFIG_TMPFS_QUOTA
4249 if (ctx->seen & SHMEM_SEEN_QUOTA) {
4250 sb->dq_op = &shmem_quota_operations;
4251 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4252 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4254 /* Copy the default limits from ctx into sbinfo */
4255 memcpy(&sbinfo->qlimits, &ctx->qlimits,
4256 sizeof(struct shmem_quota_limits));
4258 if (shmem_enable_quotas(sb, ctx->quota_types))
4261 #endif /* CONFIG_TMPFS_QUOTA */
4263 inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, S_IFDIR | sbinfo->mode, 0,
4265 if (IS_ERR(inode)) {
4266 error = PTR_ERR(inode);
4269 inode->i_uid = sbinfo->uid;
4270 inode->i_gid = sbinfo->gid;
4271 sb->s_root = d_make_root(inode);
4277 shmem_put_super(sb);
4281 static int shmem_get_tree(struct fs_context *fc)
4283 return get_tree_nodev(fc, shmem_fill_super);
4286 static void shmem_free_fc(struct fs_context *fc)
4288 struct shmem_options *ctx = fc->fs_private;
4291 mpol_put(ctx->mpol);
4296 static const struct fs_context_operations shmem_fs_context_ops = {
4297 .free = shmem_free_fc,
4298 .get_tree = shmem_get_tree,
4300 .parse_monolithic = shmem_parse_options,
4301 .parse_param = shmem_parse_one,
4302 .reconfigure = shmem_reconfigure,
4306 static struct kmem_cache *shmem_inode_cachep;
4308 static struct inode *shmem_alloc_inode(struct super_block *sb)
4310 struct shmem_inode_info *info;
4311 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
4314 return &info->vfs_inode;
4317 static void shmem_free_in_core_inode(struct inode *inode)
4319 if (S_ISLNK(inode->i_mode))
4320 kfree(inode->i_link);
4321 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4324 static void shmem_destroy_inode(struct inode *inode)
4326 if (S_ISREG(inode->i_mode))
4327 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
4328 if (S_ISDIR(inode->i_mode))
4329 simple_offset_destroy(shmem_get_offset_ctx(inode));
4332 static void shmem_init_inode(void *foo)
4334 struct shmem_inode_info *info = foo;
4335 inode_init_once(&info->vfs_inode);
4338 static void shmem_init_inodecache(void)
4340 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4341 sizeof(struct shmem_inode_info),
4342 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
4345 static void shmem_destroy_inodecache(void)
4347 kmem_cache_destroy(shmem_inode_cachep);
4350 /* Keep the page in page cache instead of truncating it */
4351 static int shmem_error_remove_page(struct address_space *mapping,
4357 const struct address_space_operations shmem_aops = {
4358 .writepage = shmem_writepage,
4359 .dirty_folio = noop_dirty_folio,
4361 .write_begin = shmem_write_begin,
4362 .write_end = shmem_write_end,
4364 #ifdef CONFIG_MIGRATION
4365 .migrate_folio = migrate_folio,
4367 .error_remove_page = shmem_error_remove_page,
4369 EXPORT_SYMBOL(shmem_aops);
4371 static const struct file_operations shmem_file_operations = {
4373 .open = generic_file_open,
4374 .get_unmapped_area = shmem_get_unmapped_area,
4376 .llseek = shmem_file_llseek,
4377 .read_iter = shmem_file_read_iter,
4378 .write_iter = generic_file_write_iter,
4379 .fsync = noop_fsync,
4380 .splice_read = shmem_file_splice_read,
4381 .splice_write = iter_file_splice_write,
4382 .fallocate = shmem_fallocate,
4386 static const struct inode_operations shmem_inode_operations = {
4387 .getattr = shmem_getattr,
4388 .setattr = shmem_setattr,
4389 #ifdef CONFIG_TMPFS_XATTR
4390 .listxattr = shmem_listxattr,
4391 .set_acl = simple_set_acl,
4392 .fileattr_get = shmem_fileattr_get,
4393 .fileattr_set = shmem_fileattr_set,
4397 static const struct inode_operations shmem_dir_inode_operations = {
4399 .getattr = shmem_getattr,
4400 .create = shmem_create,
4401 .lookup = simple_lookup,
4403 .unlink = shmem_unlink,
4404 .symlink = shmem_symlink,
4405 .mkdir = shmem_mkdir,
4406 .rmdir = shmem_rmdir,
4407 .mknod = shmem_mknod,
4408 .rename = shmem_rename2,
4409 .tmpfile = shmem_tmpfile,
4410 .get_offset_ctx = shmem_get_offset_ctx,
4412 #ifdef CONFIG_TMPFS_XATTR
4413 .listxattr = shmem_listxattr,
4414 .fileattr_get = shmem_fileattr_get,
4415 .fileattr_set = shmem_fileattr_set,
4417 #ifdef CONFIG_TMPFS_POSIX_ACL
4418 .setattr = shmem_setattr,
4419 .set_acl = simple_set_acl,
4423 static const struct inode_operations shmem_special_inode_operations = {
4424 .getattr = shmem_getattr,
4425 #ifdef CONFIG_TMPFS_XATTR
4426 .listxattr = shmem_listxattr,
4428 #ifdef CONFIG_TMPFS_POSIX_ACL
4429 .setattr = shmem_setattr,
4430 .set_acl = simple_set_acl,
4434 static const struct super_operations shmem_ops = {
4435 .alloc_inode = shmem_alloc_inode,
4436 .free_inode = shmem_free_in_core_inode,
4437 .destroy_inode = shmem_destroy_inode,
4439 .statfs = shmem_statfs,
4440 .show_options = shmem_show_options,
4442 #ifdef CONFIG_TMPFS_QUOTA
4443 .get_dquots = shmem_get_dquots,
4445 .evict_inode = shmem_evict_inode,
4446 .drop_inode = generic_delete_inode,
4447 .put_super = shmem_put_super,
4448 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4449 .nr_cached_objects = shmem_unused_huge_count,
4450 .free_cached_objects = shmem_unused_huge_scan,
4454 static const struct vm_operations_struct shmem_vm_ops = {
4455 .fault = shmem_fault,
4456 .map_pages = filemap_map_pages,
4458 .set_policy = shmem_set_policy,
4459 .get_policy = shmem_get_policy,
4463 static const struct vm_operations_struct shmem_anon_vm_ops = {
4464 .fault = shmem_fault,
4465 .map_pages = filemap_map_pages,
4467 .set_policy = shmem_set_policy,
4468 .get_policy = shmem_get_policy,
4472 int shmem_init_fs_context(struct fs_context *fc)
4474 struct shmem_options *ctx;
4476 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4480 ctx->mode = 0777 | S_ISVTX;
4481 ctx->uid = current_fsuid();
4482 ctx->gid = current_fsgid();
4484 fc->fs_private = ctx;
4485 fc->ops = &shmem_fs_context_ops;
4489 static struct file_system_type shmem_fs_type = {
4490 .owner = THIS_MODULE,
4492 .init_fs_context = shmem_init_fs_context,
4494 .parameters = shmem_fs_parameters,
4496 .kill_sb = kill_litter_super,
4498 .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4500 .fs_flags = FS_USERNS_MOUNT,
4504 void __init shmem_init(void)
4508 shmem_init_inodecache();
4510 #ifdef CONFIG_TMPFS_QUOTA
4511 error = register_quota_format(&shmem_quota_format);
4513 pr_err("Could not register quota format\n");
4518 error = register_filesystem(&shmem_fs_type);
4520 pr_err("Could not register tmpfs\n");
4524 shm_mnt = kern_mount(&shmem_fs_type);
4525 if (IS_ERR(shm_mnt)) {
4526 error = PTR_ERR(shm_mnt);
4527 pr_err("Could not kern_mount tmpfs\n");
4531 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4532 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4533 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4535 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4540 unregister_filesystem(&shmem_fs_type);
4542 #ifdef CONFIG_TMPFS_QUOTA
4543 unregister_quota_format(&shmem_quota_format);
4546 shmem_destroy_inodecache();
4547 shm_mnt = ERR_PTR(error);
4550 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4551 static ssize_t shmem_enabled_show(struct kobject *kobj,
4552 struct kobj_attribute *attr, char *buf)
4554 static const int values[] = {
4556 SHMEM_HUGE_WITHIN_SIZE,
4565 for (i = 0; i < ARRAY_SIZE(values); i++) {
4566 len += sysfs_emit_at(buf, len,
4567 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4569 shmem_format_huge(values[i]));
4572 len += sysfs_emit_at(buf, len, "\n");
4577 static ssize_t shmem_enabled_store(struct kobject *kobj,
4578 struct kobj_attribute *attr, const char *buf, size_t count)
4583 if (count + 1 > sizeof(tmp))
4585 memcpy(tmp, buf, count);
4587 if (count && tmp[count - 1] == '\n')
4588 tmp[count - 1] = '\0';
4590 huge = shmem_parse_huge(tmp);
4591 if (huge == -EINVAL)
4593 if (!has_transparent_hugepage() &&
4594 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4598 if (shmem_huge > SHMEM_HUGE_DENY)
4599 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4603 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
4604 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4606 #else /* !CONFIG_SHMEM */
4609 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4611 * This is intended for small system where the benefits of the full
4612 * shmem code (swap-backed and resource-limited) are outweighed by
4613 * their complexity. On systems without swap this code should be
4614 * effectively equivalent, but much lighter weight.
4617 static struct file_system_type shmem_fs_type = {
4619 .init_fs_context = ramfs_init_fs_context,
4620 .parameters = ramfs_fs_parameters,
4621 .kill_sb = ramfs_kill_sb,
4622 .fs_flags = FS_USERNS_MOUNT,
4625 void __init shmem_init(void)
4627 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4629 shm_mnt = kern_mount(&shmem_fs_type);
4630 BUG_ON(IS_ERR(shm_mnt));
4633 int shmem_unuse(unsigned int type)
4638 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4643 void shmem_unlock_mapping(struct address_space *mapping)
4648 unsigned long shmem_get_unmapped_area(struct file *file,
4649 unsigned long addr, unsigned long len,
4650 unsigned long pgoff, unsigned long flags)
4652 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4656 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4658 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4660 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4662 #define shmem_vm_ops generic_file_vm_ops
4663 #define shmem_anon_vm_ops generic_file_vm_ops
4664 #define shmem_file_operations ramfs_file_operations
4665 #define shmem_acct_size(flags, size) 0
4666 #define shmem_unacct_size(flags, size) do {} while (0)
4668 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block *sb, struct inode *dir,
4669 umode_t mode, dev_t dev, unsigned long flags)
4671 struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
4672 return inode ? inode : ERR_PTR(-ENOSPC);
4675 #endif /* CONFIG_SHMEM */
4679 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4680 unsigned long flags, unsigned int i_flags)
4682 struct inode *inode;
4686 return ERR_CAST(mnt);
4688 if (size < 0 || size > MAX_LFS_FILESIZE)
4689 return ERR_PTR(-EINVAL);
4691 if (shmem_acct_size(flags, size))
4692 return ERR_PTR(-ENOMEM);
4694 if (is_idmapped_mnt(mnt))
4695 return ERR_PTR(-EINVAL);
4697 inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
4698 S_IFREG | S_IRWXUGO, 0, flags);
4700 if (IS_ERR(inode)) {
4701 shmem_unacct_size(flags, size);
4702 return ERR_CAST(inode);
4704 inode->i_flags |= i_flags;
4705 inode->i_size = size;
4706 clear_nlink(inode); /* It is unlinked */
4707 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4709 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4710 &shmem_file_operations);
4717 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4718 * kernel internal. There will be NO LSM permission checks against the
4719 * underlying inode. So users of this interface must do LSM checks at a
4720 * higher layer. The users are the big_key and shm implementations. LSM
4721 * checks are provided at the key or shm level rather than the inode.
4722 * @name: name for dentry (to be seen in /proc/<pid>/maps
4723 * @size: size to be set for the file
4724 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4726 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4728 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4732 * shmem_file_setup - get an unlinked file living in tmpfs
4733 * @name: name for dentry (to be seen in /proc/<pid>/maps
4734 * @size: size to be set for the file
4735 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4737 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4739 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4741 EXPORT_SYMBOL_GPL(shmem_file_setup);
4744 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4745 * @mnt: the tmpfs mount where the file will be created
4746 * @name: name for dentry (to be seen in /proc/<pid>/maps
4747 * @size: size to be set for the file
4748 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4750 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4751 loff_t size, unsigned long flags)
4753 return __shmem_file_setup(mnt, name, size, flags, 0);
4755 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4758 * shmem_zero_setup - setup a shared anonymous mapping
4759 * @vma: the vma to be mmapped is prepared by do_mmap
4761 int shmem_zero_setup(struct vm_area_struct *vma)
4764 loff_t size = vma->vm_end - vma->vm_start;
4767 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4768 * between XFS directory reading and selinux: since this file is only
4769 * accessible to the user through its mapping, use S_PRIVATE flag to
4770 * bypass file security, in the same way as shmem_kernel_file_setup().
4772 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4774 return PTR_ERR(file);
4778 vma->vm_file = file;
4779 vma->vm_ops = &shmem_anon_vm_ops;
4785 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4786 * @mapping: the folio's address_space
4787 * @index: the folio index
4788 * @gfp: the page allocator flags to use if allocating
4790 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4791 * with any new page allocations done using the specified allocation flags.
4792 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4793 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4794 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4796 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4797 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4799 struct folio *shmem_read_folio_gfp(struct address_space *mapping,
4800 pgoff_t index, gfp_t gfp)
4803 struct inode *inode = mapping->host;
4804 struct folio *folio;
4807 BUG_ON(!shmem_mapping(mapping));
4808 error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
4809 gfp, NULL, NULL, NULL);
4811 return ERR_PTR(error);
4813 folio_unlock(folio);
4817 * The tiny !SHMEM case uses ramfs without swap
4819 return mapping_read_folio_gfp(mapping, index, gfp);
4822 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
4824 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4825 pgoff_t index, gfp_t gfp)
4827 struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
4831 return &folio->page;
4833 page = folio_file_page(folio, index);
4834 if (PageHWPoison(page)) {
4836 return ERR_PTR(-EIO);
4841 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);