2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
31 #include <linux/fileattr.h>
33 #include <linux/random.h>
34 #include <linux/sched/signal.h>
35 #include <linux/export.h>
36 #include <linux/shmem_fs.h>
37 #include <linux/swap.h>
38 #include <linux/uio.h>
39 #include <linux/hugetlb.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
42 #include <linux/iversion.h>
45 static struct vfsmount *shm_mnt;
49 * This virtual memory filesystem is heavily based on the ramfs. It
50 * extends ramfs by the ability to use swap and honor resource limits
51 * which makes it a completely usable filesystem.
54 #include <linux/xattr.h>
55 #include <linux/exportfs.h>
56 #include <linux/posix_acl.h>
57 #include <linux/posix_acl_xattr.h>
58 #include <linux/mman.h>
59 #include <linux/string.h>
60 #include <linux/slab.h>
61 #include <linux/backing-dev.h>
62 #include <linux/writeback.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
81 #include <linux/quotaops.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Pretend that one inode + its dentry occupy this much memory */
94 #define BOGO_INODE_SIZE 1024
96 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
97 #define SHORT_SYMLINK_LEN 128
100 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
101 * inode->i_private (with i_rwsem making sure that it has only one user at
102 * a time): we would prefer not to enlarge the shmem inode just for that.
104 struct shmem_falloc {
105 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
106 pgoff_t start; /* start of range currently being fallocated */
107 pgoff_t next; /* the next page offset to be fallocated */
108 pgoff_t nr_falloced; /* how many new pages have been fallocated */
109 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
112 struct shmem_options {
113 unsigned long long blocks;
114 unsigned long long inodes;
115 struct mempolicy *mpol;
123 unsigned short quota_types;
124 struct shmem_quota_limits qlimits;
125 #define SHMEM_SEEN_BLOCKS 1
126 #define SHMEM_SEEN_INODES 2
127 #define SHMEM_SEEN_HUGE 4
128 #define SHMEM_SEEN_INUMS 8
129 #define SHMEM_SEEN_NOSWAP 16
130 #define SHMEM_SEEN_QUOTA 32
134 static unsigned long shmem_default_max_blocks(void)
136 return totalram_pages() / 2;
139 static unsigned long shmem_default_max_inodes(void)
141 unsigned long nr_pages = totalram_pages();
143 return min3(nr_pages - totalhigh_pages(), nr_pages / 2,
144 ULONG_MAX / BOGO_INODE_SIZE);
148 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
149 struct folio **foliop, enum sgp_type sgp,
150 gfp_t gfp, struct vm_area_struct *vma,
151 vm_fault_t *fault_type);
153 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
155 return sb->s_fs_info;
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
164 static inline int shmem_acct_size(unsigned long flags, loff_t size)
166 return (flags & VM_NORESERVE) ?
167 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
170 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
172 if (!(flags & VM_NORESERVE))
173 vm_unacct_memory(VM_ACCT(size));
176 static inline int shmem_reacct_size(unsigned long flags,
177 loff_t oldsize, loff_t newsize)
179 if (!(flags & VM_NORESERVE)) {
180 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
181 return security_vm_enough_memory_mm(current->mm,
182 VM_ACCT(newsize) - VM_ACCT(oldsize));
183 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
184 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
190 * ... whereas tmpfs objects are accounted incrementally as
191 * pages are allocated, in order to allow large sparse files.
192 * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
193 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
195 static inline int shmem_acct_block(unsigned long flags, long pages)
197 if (!(flags & VM_NORESERVE))
200 return security_vm_enough_memory_mm(current->mm,
201 pages * VM_ACCT(PAGE_SIZE));
204 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
206 if (flags & VM_NORESERVE)
207 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
210 static int shmem_inode_acct_block(struct inode *inode, long pages)
212 struct shmem_inode_info *info = SHMEM_I(inode);
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
216 if (shmem_acct_block(info->flags, pages))
219 might_sleep(); /* when quotas */
220 if (sbinfo->max_blocks) {
221 if (percpu_counter_compare(&sbinfo->used_blocks,
222 sbinfo->max_blocks - pages) > 0)
225 err = dquot_alloc_block_nodirty(inode, pages);
229 percpu_counter_add(&sbinfo->used_blocks, pages);
231 err = dquot_alloc_block_nodirty(inode, pages);
239 shmem_unacct_blocks(info->flags, pages);
243 static void shmem_inode_unacct_blocks(struct inode *inode, long pages)
245 struct shmem_inode_info *info = SHMEM_I(inode);
246 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
248 might_sleep(); /* when quotas */
249 dquot_free_block_nodirty(inode, pages);
251 if (sbinfo->max_blocks)
252 percpu_counter_sub(&sbinfo->used_blocks, pages);
253 shmem_unacct_blocks(info->flags, pages);
256 static const struct super_operations shmem_ops;
257 const struct address_space_operations shmem_aops;
258 static const struct file_operations shmem_file_operations;
259 static const struct inode_operations shmem_inode_operations;
260 static const struct inode_operations shmem_dir_inode_operations;
261 static const struct inode_operations shmem_special_inode_operations;
262 static const struct vm_operations_struct shmem_vm_ops;
263 static const struct vm_operations_struct shmem_anon_vm_ops;
264 static struct file_system_type shmem_fs_type;
266 bool vma_is_anon_shmem(struct vm_area_struct *vma)
268 return vma->vm_ops == &shmem_anon_vm_ops;
271 bool vma_is_shmem(struct vm_area_struct *vma)
273 return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
276 static LIST_HEAD(shmem_swaplist);
277 static DEFINE_MUTEX(shmem_swaplist_mutex);
279 #ifdef CONFIG_TMPFS_QUOTA
281 static int shmem_enable_quotas(struct super_block *sb,
282 unsigned short quota_types)
286 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
287 for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
288 if (!(quota_types & (1 << type)))
290 err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
291 DQUOT_USAGE_ENABLED |
292 DQUOT_LIMITS_ENABLED);
299 pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
301 for (type--; type >= 0; type--)
302 dquot_quota_off(sb, type);
306 static void shmem_disable_quotas(struct super_block *sb)
310 for (type = 0; type < SHMEM_MAXQUOTAS; type++)
311 dquot_quota_off(sb, type);
314 static struct dquot **shmem_get_dquots(struct inode *inode)
316 return SHMEM_I(inode)->i_dquot;
318 #endif /* CONFIG_TMPFS_QUOTA */
321 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
322 * produces a novel ino for the newly allocated inode.
324 * It may also be called when making a hard link to permit the space needed by
325 * each dentry. However, in that case, no new inode number is needed since that
326 * internally draws from another pool of inode numbers (currently global
327 * get_next_ino()). This case is indicated by passing NULL as inop.
329 #define SHMEM_INO_BATCH 1024
330 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
332 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
335 if (!(sb->s_flags & SB_KERNMOUNT)) {
336 raw_spin_lock(&sbinfo->stat_lock);
337 if (sbinfo->max_inodes) {
338 if (sbinfo->free_ispace < BOGO_INODE_SIZE) {
339 raw_spin_unlock(&sbinfo->stat_lock);
342 sbinfo->free_ispace -= BOGO_INODE_SIZE;
345 ino = sbinfo->next_ino++;
346 if (unlikely(is_zero_ino(ino)))
347 ino = sbinfo->next_ino++;
348 if (unlikely(!sbinfo->full_inums &&
351 * Emulate get_next_ino uint wraparound for
354 if (IS_ENABLED(CONFIG_64BIT))
355 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
356 __func__, MINOR(sb->s_dev));
357 sbinfo->next_ino = 1;
358 ino = sbinfo->next_ino++;
362 raw_spin_unlock(&sbinfo->stat_lock);
365 * __shmem_file_setup, one of our callers, is lock-free: it
366 * doesn't hold stat_lock in shmem_reserve_inode since
367 * max_inodes is always 0, and is called from potentially
368 * unknown contexts. As such, use a per-cpu batched allocator
369 * which doesn't require the per-sb stat_lock unless we are at
370 * the batch boundary.
372 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
373 * shmem mounts are not exposed to userspace, so we don't need
374 * to worry about things like glibc compatibility.
378 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
380 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
381 raw_spin_lock(&sbinfo->stat_lock);
382 ino = sbinfo->next_ino;
383 sbinfo->next_ino += SHMEM_INO_BATCH;
384 raw_spin_unlock(&sbinfo->stat_lock);
385 if (unlikely(is_zero_ino(ino)))
396 static void shmem_free_inode(struct super_block *sb)
398 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
399 if (sbinfo->max_inodes) {
400 raw_spin_lock(&sbinfo->stat_lock);
401 sbinfo->free_ispace += BOGO_INODE_SIZE;
402 raw_spin_unlock(&sbinfo->stat_lock);
407 * shmem_recalc_inode - recalculate the block usage of an inode
408 * @inode: inode to recalc
409 * @alloced: the change in number of pages allocated to inode
410 * @swapped: the change in number of pages swapped from inode
412 * We have to calculate the free blocks since the mm can drop
413 * undirtied hole pages behind our back.
415 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
416 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
418 static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped)
420 struct shmem_inode_info *info = SHMEM_I(inode);
423 spin_lock(&info->lock);
424 info->alloced += alloced;
425 info->swapped += swapped;
426 freed = info->alloced - info->swapped -
427 READ_ONCE(inode->i_mapping->nrpages);
429 * Special case: whereas normally shmem_recalc_inode() is called
430 * after i_mapping->nrpages has already been adjusted (up or down),
431 * shmem_writepage() has to raise swapped before nrpages is lowered -
432 * to stop a racing shmem_recalc_inode() from thinking that a page has
433 * been freed. Compensate here, to avoid the need for a followup call.
438 info->alloced -= freed;
439 spin_unlock(&info->lock);
441 /* The quota case may block */
443 shmem_inode_unacct_blocks(inode, freed);
446 bool shmem_charge(struct inode *inode, long pages)
448 struct address_space *mapping = inode->i_mapping;
450 if (shmem_inode_acct_block(inode, pages))
453 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
454 xa_lock_irq(&mapping->i_pages);
455 mapping->nrpages += pages;
456 xa_unlock_irq(&mapping->i_pages);
458 shmem_recalc_inode(inode, pages, 0);
462 void shmem_uncharge(struct inode *inode, long pages)
464 /* pages argument is currently unused: keep it to help debugging */
465 /* nrpages adjustment done by __filemap_remove_folio() or caller */
467 shmem_recalc_inode(inode, 0, 0);
471 * Replace item expected in xarray by a new item, while holding xa_lock.
473 static int shmem_replace_entry(struct address_space *mapping,
474 pgoff_t index, void *expected, void *replacement)
476 XA_STATE(xas, &mapping->i_pages, index);
479 VM_BUG_ON(!expected);
480 VM_BUG_ON(!replacement);
481 item = xas_load(&xas);
482 if (item != expected)
484 xas_store(&xas, replacement);
489 * Sometimes, before we decide whether to proceed or to fail, we must check
490 * that an entry was not already brought back from swap by a racing thread.
492 * Checking page is not enough: by the time a SwapCache page is locked, it
493 * might be reused, and again be SwapCache, using the same swap as before.
495 static bool shmem_confirm_swap(struct address_space *mapping,
496 pgoff_t index, swp_entry_t swap)
498 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
502 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
505 * disables huge pages for the mount;
507 * enables huge pages for the mount;
508 * SHMEM_HUGE_WITHIN_SIZE:
509 * only allocate huge pages if the page will be fully within i_size,
510 * also respect fadvise()/madvise() hints;
512 * only allocate huge pages if requested with fadvise()/madvise();
515 #define SHMEM_HUGE_NEVER 0
516 #define SHMEM_HUGE_ALWAYS 1
517 #define SHMEM_HUGE_WITHIN_SIZE 2
518 #define SHMEM_HUGE_ADVISE 3
522 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
525 * disables huge on shm_mnt and all mounts, for emergency use;
527 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
530 #define SHMEM_HUGE_DENY (-1)
531 #define SHMEM_HUGE_FORCE (-2)
533 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
534 /* ifdef here to avoid bloating shmem.o when not necessary */
536 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
538 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
539 struct mm_struct *mm, unsigned long vm_flags)
543 if (!S_ISREG(inode->i_mode))
545 if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
547 if (shmem_huge == SHMEM_HUGE_DENY)
549 if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
552 switch (SHMEM_SB(inode->i_sb)->huge) {
553 case SHMEM_HUGE_ALWAYS:
555 case SHMEM_HUGE_WITHIN_SIZE:
556 index = round_up(index + 1, HPAGE_PMD_NR);
557 i_size = round_up(i_size_read(inode), PAGE_SIZE);
558 if (i_size >> PAGE_SHIFT >= index)
561 case SHMEM_HUGE_ADVISE:
562 if (mm && (vm_flags & VM_HUGEPAGE))
570 #if defined(CONFIG_SYSFS)
571 static int shmem_parse_huge(const char *str)
573 if (!strcmp(str, "never"))
574 return SHMEM_HUGE_NEVER;
575 if (!strcmp(str, "always"))
576 return SHMEM_HUGE_ALWAYS;
577 if (!strcmp(str, "within_size"))
578 return SHMEM_HUGE_WITHIN_SIZE;
579 if (!strcmp(str, "advise"))
580 return SHMEM_HUGE_ADVISE;
581 if (!strcmp(str, "deny"))
582 return SHMEM_HUGE_DENY;
583 if (!strcmp(str, "force"))
584 return SHMEM_HUGE_FORCE;
589 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
590 static const char *shmem_format_huge(int huge)
593 case SHMEM_HUGE_NEVER:
595 case SHMEM_HUGE_ALWAYS:
597 case SHMEM_HUGE_WITHIN_SIZE:
598 return "within_size";
599 case SHMEM_HUGE_ADVISE:
601 case SHMEM_HUGE_DENY:
603 case SHMEM_HUGE_FORCE:
612 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
613 struct shrink_control *sc, unsigned long nr_to_split)
615 LIST_HEAD(list), *pos, *next;
616 LIST_HEAD(to_remove);
618 struct shmem_inode_info *info;
620 unsigned long batch = sc ? sc->nr_to_scan : 128;
623 if (list_empty(&sbinfo->shrinklist))
626 spin_lock(&sbinfo->shrinklist_lock);
627 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
628 info = list_entry(pos, struct shmem_inode_info, shrinklist);
631 inode = igrab(&info->vfs_inode);
633 /* inode is about to be evicted */
635 list_del_init(&info->shrinklist);
639 /* Check if there's anything to gain */
640 if (round_up(inode->i_size, PAGE_SIZE) ==
641 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
642 list_move(&info->shrinklist, &to_remove);
646 list_move(&info->shrinklist, &list);
648 sbinfo->shrinklist_len--;
652 spin_unlock(&sbinfo->shrinklist_lock);
654 list_for_each_safe(pos, next, &to_remove) {
655 info = list_entry(pos, struct shmem_inode_info, shrinklist);
656 inode = &info->vfs_inode;
657 list_del_init(&info->shrinklist);
661 list_for_each_safe(pos, next, &list) {
665 info = list_entry(pos, struct shmem_inode_info, shrinklist);
666 inode = &info->vfs_inode;
668 if (nr_to_split && split >= nr_to_split)
671 index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
672 folio = filemap_get_folio(inode->i_mapping, index);
676 /* No huge page at the end of the file: nothing to split */
677 if (!folio_test_large(folio)) {
683 * Move the inode on the list back to shrinklist if we failed
684 * to lock the page at this time.
686 * Waiting for the lock may lead to deadlock in the
689 if (!folio_trylock(folio)) {
694 ret = split_folio(folio);
698 /* If split failed move the inode on the list back to shrinklist */
704 list_del_init(&info->shrinklist);
708 * Make sure the inode is either on the global list or deleted
709 * from any local list before iput() since it could be deleted
710 * in another thread once we put the inode (then the local list
713 spin_lock(&sbinfo->shrinklist_lock);
714 list_move(&info->shrinklist, &sbinfo->shrinklist);
715 sbinfo->shrinklist_len++;
716 spin_unlock(&sbinfo->shrinklist_lock);
724 static long shmem_unused_huge_scan(struct super_block *sb,
725 struct shrink_control *sc)
727 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
729 if (!READ_ONCE(sbinfo->shrinklist_len))
732 return shmem_unused_huge_shrink(sbinfo, sc, 0);
735 static long shmem_unused_huge_count(struct super_block *sb,
736 struct shrink_control *sc)
738 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
739 return READ_ONCE(sbinfo->shrinklist_len);
741 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
743 #define shmem_huge SHMEM_HUGE_DENY
745 bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
746 struct mm_struct *mm, unsigned long vm_flags)
751 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
752 struct shrink_control *sc, unsigned long nr_to_split)
756 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
759 * Like filemap_add_folio, but error if expected item has gone.
761 static int shmem_add_to_page_cache(struct folio *folio,
762 struct address_space *mapping,
763 pgoff_t index, void *expected, gfp_t gfp,
764 struct mm_struct *charge_mm)
766 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
767 long nr = folio_nr_pages(folio);
770 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
771 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
772 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
773 VM_BUG_ON(expected && folio_test_large(folio));
775 folio_ref_add(folio, nr);
776 folio->mapping = mapping;
777 folio->index = index;
779 if (!folio_test_swapcache(folio)) {
780 error = mem_cgroup_charge(folio, charge_mm, gfp);
782 if (folio_test_pmd_mappable(folio)) {
783 count_vm_event(THP_FILE_FALLBACK);
784 count_vm_event(THP_FILE_FALLBACK_CHARGE);
789 folio_throttle_swaprate(folio, gfp);
793 if (expected != xas_find_conflict(&xas)) {
794 xas_set_err(&xas, -EEXIST);
797 if (expected && xas_find_conflict(&xas)) {
798 xas_set_err(&xas, -EEXIST);
801 xas_store(&xas, folio);
804 if (folio_test_pmd_mappable(folio)) {
805 count_vm_event(THP_FILE_ALLOC);
806 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
808 mapping->nrpages += nr;
809 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
810 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
812 xas_unlock_irq(&xas);
813 } while (xas_nomem(&xas, gfp));
815 if (xas_error(&xas)) {
816 error = xas_error(&xas);
822 folio->mapping = NULL;
823 folio_ref_sub(folio, nr);
828 * Like delete_from_page_cache, but substitutes swap for @folio.
830 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
832 struct address_space *mapping = folio->mapping;
833 long nr = folio_nr_pages(folio);
836 xa_lock_irq(&mapping->i_pages);
837 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
838 folio->mapping = NULL;
839 mapping->nrpages -= nr;
840 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
841 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
842 xa_unlock_irq(&mapping->i_pages);
848 * Remove swap entry from page cache, free the swap and its page cache.
850 static int shmem_free_swap(struct address_space *mapping,
851 pgoff_t index, void *radswap)
855 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
858 free_swap_and_cache(radix_to_swp_entry(radswap));
863 * Determine (in bytes) how many of the shmem object's pages mapped by the
864 * given offsets are swapped out.
866 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
867 * as long as the inode doesn't go away and racy results are not a problem.
869 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
870 pgoff_t start, pgoff_t end)
872 XA_STATE(xas, &mapping->i_pages, start);
874 unsigned long swapped = 0;
877 xas_for_each(&xas, page, end - 1) {
878 if (xas_retry(&xas, page))
880 if (xa_is_value(page))
883 if (need_resched()) {
891 return swapped << PAGE_SHIFT;
895 * Determine (in bytes) how many of the shmem object's pages mapped by the
896 * given vma is swapped out.
898 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
899 * as long as the inode doesn't go away and racy results are not a problem.
901 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
903 struct inode *inode = file_inode(vma->vm_file);
904 struct shmem_inode_info *info = SHMEM_I(inode);
905 struct address_space *mapping = inode->i_mapping;
906 unsigned long swapped;
908 /* Be careful as we don't hold info->lock */
909 swapped = READ_ONCE(info->swapped);
912 * The easier cases are when the shmem object has nothing in swap, or
913 * the vma maps it whole. Then we can simply use the stats that we
919 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
920 return swapped << PAGE_SHIFT;
922 /* Here comes the more involved part */
923 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
924 vma->vm_pgoff + vma_pages(vma));
928 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
930 void shmem_unlock_mapping(struct address_space *mapping)
932 struct folio_batch fbatch;
935 folio_batch_init(&fbatch);
937 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
939 while (!mapping_unevictable(mapping) &&
940 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
941 check_move_unevictable_folios(&fbatch);
942 folio_batch_release(&fbatch);
947 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
952 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
953 * beyond i_size, and reports fallocated folios as holes.
955 folio = filemap_get_entry(inode->i_mapping, index);
958 if (!xa_is_value(folio)) {
960 if (folio->mapping == inode->i_mapping)
962 /* The folio has been swapped out */
967 * But read a folio back from swap if any of it is within i_size
968 * (although in some cases this is just a waste of time).
971 shmem_get_folio(inode, index, &folio, SGP_READ);
976 * Remove range of pages and swap entries from page cache, and free them.
977 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
979 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
982 struct address_space *mapping = inode->i_mapping;
983 struct shmem_inode_info *info = SHMEM_I(inode);
984 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
985 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
986 struct folio_batch fbatch;
987 pgoff_t indices[PAGEVEC_SIZE];
990 long nr_swaps_freed = 0;
995 end = -1; /* unsigned, so actually very big */
997 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
998 info->fallocend = start;
1000 folio_batch_init(&fbatch);
1002 while (index < end && find_lock_entries(mapping, &index, end - 1,
1003 &fbatch, indices)) {
1004 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1005 folio = fbatch.folios[i];
1007 if (xa_is_value(folio)) {
1010 nr_swaps_freed += !shmem_free_swap(mapping,
1015 if (!unfalloc || !folio_test_uptodate(folio))
1016 truncate_inode_folio(mapping, folio);
1017 folio_unlock(folio);
1019 folio_batch_remove_exceptionals(&fbatch);
1020 folio_batch_release(&fbatch);
1025 * When undoing a failed fallocate, we want none of the partial folio
1026 * zeroing and splitting below, but shall want to truncate the whole
1027 * folio when !uptodate indicates that it was added by this fallocate,
1028 * even when [lstart, lend] covers only a part of the folio.
1033 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
1034 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
1036 same_folio = lend < folio_pos(folio) + folio_size(folio);
1037 folio_mark_dirty(folio);
1038 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
1039 start = folio->index + folio_nr_pages(folio);
1043 folio_unlock(folio);
1049 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
1051 folio_mark_dirty(folio);
1052 if (!truncate_inode_partial_folio(folio, lstart, lend))
1054 folio_unlock(folio);
1061 while (index < end) {
1064 if (!find_get_entries(mapping, &index, end - 1, &fbatch,
1066 /* If all gone or hole-punch or unfalloc, we're done */
1067 if (index == start || end != -1)
1069 /* But if truncating, restart to make sure all gone */
1073 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1074 folio = fbatch.folios[i];
1076 if (xa_is_value(folio)) {
1079 if (shmem_free_swap(mapping, indices[i], folio)) {
1080 /* Swap was replaced by page: retry */
1090 if (!unfalloc || !folio_test_uptodate(folio)) {
1091 if (folio_mapping(folio) != mapping) {
1092 /* Page was replaced by swap: retry */
1093 folio_unlock(folio);
1097 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1099 truncate_inode_folio(mapping, folio);
1101 folio_unlock(folio);
1103 folio_batch_remove_exceptionals(&fbatch);
1104 folio_batch_release(&fbatch);
1107 shmem_recalc_inode(inode, 0, -nr_swaps_freed);
1110 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1112 shmem_undo_range(inode, lstart, lend, false);
1113 inode->i_ctime = inode->i_mtime = current_time(inode);
1114 inode_inc_iversion(inode);
1116 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1118 static int shmem_getattr(struct mnt_idmap *idmap,
1119 const struct path *path, struct kstat *stat,
1120 u32 request_mask, unsigned int query_flags)
1122 struct inode *inode = path->dentry->d_inode;
1123 struct shmem_inode_info *info = SHMEM_I(inode);
1125 if (info->alloced - info->swapped != inode->i_mapping->nrpages)
1126 shmem_recalc_inode(inode, 0, 0);
1128 if (info->fsflags & FS_APPEND_FL)
1129 stat->attributes |= STATX_ATTR_APPEND;
1130 if (info->fsflags & FS_IMMUTABLE_FL)
1131 stat->attributes |= STATX_ATTR_IMMUTABLE;
1132 if (info->fsflags & FS_NODUMP_FL)
1133 stat->attributes |= STATX_ATTR_NODUMP;
1134 stat->attributes_mask |= (STATX_ATTR_APPEND |
1135 STATX_ATTR_IMMUTABLE |
1137 generic_fillattr(idmap, inode, stat);
1139 if (shmem_is_huge(inode, 0, false, NULL, 0))
1140 stat->blksize = HPAGE_PMD_SIZE;
1142 if (request_mask & STATX_BTIME) {
1143 stat->result_mask |= STATX_BTIME;
1144 stat->btime.tv_sec = info->i_crtime.tv_sec;
1145 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1151 static int shmem_setattr(struct mnt_idmap *idmap,
1152 struct dentry *dentry, struct iattr *attr)
1154 struct inode *inode = d_inode(dentry);
1155 struct shmem_inode_info *info = SHMEM_I(inode);
1157 bool update_mtime = false;
1158 bool update_ctime = true;
1160 error = setattr_prepare(idmap, dentry, attr);
1164 if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1165 if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1170 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1171 loff_t oldsize = inode->i_size;
1172 loff_t newsize = attr->ia_size;
1174 /* protected by i_rwsem */
1175 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1176 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1179 if (newsize != oldsize) {
1180 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1184 i_size_write(inode, newsize);
1185 update_mtime = true;
1187 update_ctime = false;
1189 if (newsize <= oldsize) {
1190 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1191 if (oldsize > holebegin)
1192 unmap_mapping_range(inode->i_mapping,
1195 shmem_truncate_range(inode,
1196 newsize, (loff_t)-1);
1197 /* unmap again to remove racily COWed private pages */
1198 if (oldsize > holebegin)
1199 unmap_mapping_range(inode->i_mapping,
1204 if (is_quota_modification(idmap, inode, attr)) {
1205 error = dquot_initialize(inode);
1210 /* Transfer quota accounting */
1211 if (i_uid_needs_update(idmap, attr, inode) ||
1212 i_gid_needs_update(idmap, attr, inode)) {
1213 error = dquot_transfer(idmap, inode, attr);
1219 setattr_copy(idmap, inode, attr);
1220 if (attr->ia_valid & ATTR_MODE)
1221 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1222 if (!error && update_ctime) {
1223 inode->i_ctime = current_time(inode);
1225 inode->i_mtime = inode->i_ctime;
1226 inode_inc_iversion(inode);
1231 static void shmem_evict_inode(struct inode *inode)
1233 struct shmem_inode_info *info = SHMEM_I(inode);
1234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1236 if (shmem_mapping(inode->i_mapping)) {
1237 shmem_unacct_size(info->flags, inode->i_size);
1239 mapping_set_exiting(inode->i_mapping);
1240 shmem_truncate_range(inode, 0, (loff_t)-1);
1241 if (!list_empty(&info->shrinklist)) {
1242 spin_lock(&sbinfo->shrinklist_lock);
1243 if (!list_empty(&info->shrinklist)) {
1244 list_del_init(&info->shrinklist);
1245 sbinfo->shrinklist_len--;
1247 spin_unlock(&sbinfo->shrinklist_lock);
1249 while (!list_empty(&info->swaplist)) {
1250 /* Wait while shmem_unuse() is scanning this inode... */
1251 wait_var_event(&info->stop_eviction,
1252 !atomic_read(&info->stop_eviction));
1253 mutex_lock(&shmem_swaplist_mutex);
1254 /* ...but beware of the race if we peeked too early */
1255 if (!atomic_read(&info->stop_eviction))
1256 list_del_init(&info->swaplist);
1257 mutex_unlock(&shmem_swaplist_mutex);
1261 simple_xattrs_free(&info->xattrs);
1262 WARN_ON(inode->i_blocks);
1263 shmem_free_inode(inode->i_sb);
1265 #ifdef CONFIG_TMPFS_QUOTA
1266 dquot_free_inode(inode);
1271 static int shmem_find_swap_entries(struct address_space *mapping,
1272 pgoff_t start, struct folio_batch *fbatch,
1273 pgoff_t *indices, unsigned int type)
1275 XA_STATE(xas, &mapping->i_pages, start);
1276 struct folio *folio;
1280 xas_for_each(&xas, folio, ULONG_MAX) {
1281 if (xas_retry(&xas, folio))
1284 if (!xa_is_value(folio))
1287 entry = radix_to_swp_entry(folio);
1289 * swapin error entries can be found in the mapping. But they're
1290 * deliberately ignored here as we've done everything we can do.
1292 if (swp_type(entry) != type)
1295 indices[folio_batch_count(fbatch)] = xas.xa_index;
1296 if (!folio_batch_add(fbatch, folio))
1299 if (need_resched()) {
1306 return xas.xa_index;
1310 * Move the swapped pages for an inode to page cache. Returns the count
1311 * of pages swapped in, or the error in case of failure.
1313 static int shmem_unuse_swap_entries(struct inode *inode,
1314 struct folio_batch *fbatch, pgoff_t *indices)
1319 struct address_space *mapping = inode->i_mapping;
1321 for (i = 0; i < folio_batch_count(fbatch); i++) {
1322 struct folio *folio = fbatch->folios[i];
1324 if (!xa_is_value(folio))
1326 error = shmem_swapin_folio(inode, indices[i],
1328 mapping_gfp_mask(mapping),
1331 folio_unlock(folio);
1335 if (error == -ENOMEM)
1339 return error ? error : ret;
1343 * If swap found in inode, free it and move page from swapcache to filecache.
1345 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1347 struct address_space *mapping = inode->i_mapping;
1349 struct folio_batch fbatch;
1350 pgoff_t indices[PAGEVEC_SIZE];
1354 folio_batch_init(&fbatch);
1355 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1356 if (folio_batch_count(&fbatch) == 0) {
1361 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1365 start = indices[folio_batch_count(&fbatch) - 1];
1372 * Read all the shared memory data that resides in the swap
1373 * device 'type' back into memory, so the swap device can be
1376 int shmem_unuse(unsigned int type)
1378 struct shmem_inode_info *info, *next;
1381 if (list_empty(&shmem_swaplist))
1384 mutex_lock(&shmem_swaplist_mutex);
1385 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1386 if (!info->swapped) {
1387 list_del_init(&info->swaplist);
1391 * Drop the swaplist mutex while searching the inode for swap;
1392 * but before doing so, make sure shmem_evict_inode() will not
1393 * remove placeholder inode from swaplist, nor let it be freed
1394 * (igrab() would protect from unlink, but not from unmount).
1396 atomic_inc(&info->stop_eviction);
1397 mutex_unlock(&shmem_swaplist_mutex);
1399 error = shmem_unuse_inode(&info->vfs_inode, type);
1402 mutex_lock(&shmem_swaplist_mutex);
1403 next = list_next_entry(info, swaplist);
1405 list_del_init(&info->swaplist);
1406 if (atomic_dec_and_test(&info->stop_eviction))
1407 wake_up_var(&info->stop_eviction);
1411 mutex_unlock(&shmem_swaplist_mutex);
1417 * Move the page from the page cache to the swap cache.
1419 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1421 struct folio *folio = page_folio(page);
1422 struct address_space *mapping = folio->mapping;
1423 struct inode *inode = mapping->host;
1424 struct shmem_inode_info *info = SHMEM_I(inode);
1425 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1430 * Our capabilities prevent regular writeback or sync from ever calling
1431 * shmem_writepage; but a stacking filesystem might use ->writepage of
1432 * its underlying filesystem, in which case tmpfs should write out to
1433 * swap only in response to memory pressure, and not for the writeback
1436 if (WARN_ON_ONCE(!wbc->for_reclaim))
1439 if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
1442 if (!total_swap_pages)
1446 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1447 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1448 * and its shmem_writeback() needs them to be split when swapping.
1450 if (folio_test_large(folio)) {
1451 /* Ensure the subpages are still dirty */
1452 folio_test_set_dirty(folio);
1453 if (split_huge_page(page) < 0)
1455 folio = page_folio(page);
1456 folio_clear_dirty(folio);
1459 index = folio->index;
1462 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1463 * value into swapfile.c, the only way we can correctly account for a
1464 * fallocated folio arriving here is now to initialize it and write it.
1466 * That's okay for a folio already fallocated earlier, but if we have
1467 * not yet completed the fallocation, then (a) we want to keep track
1468 * of this folio in case we have to undo it, and (b) it may not be a
1469 * good idea to continue anyway, once we're pushing into swap. So
1470 * reactivate the folio, and let shmem_fallocate() quit when too many.
1472 if (!folio_test_uptodate(folio)) {
1473 if (inode->i_private) {
1474 struct shmem_falloc *shmem_falloc;
1475 spin_lock(&inode->i_lock);
1476 shmem_falloc = inode->i_private;
1478 !shmem_falloc->waitq &&
1479 index >= shmem_falloc->start &&
1480 index < shmem_falloc->next)
1481 shmem_falloc->nr_unswapped++;
1483 shmem_falloc = NULL;
1484 spin_unlock(&inode->i_lock);
1488 folio_zero_range(folio, 0, folio_size(folio));
1489 flush_dcache_folio(folio);
1490 folio_mark_uptodate(folio);
1493 swap = folio_alloc_swap(folio);
1498 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1499 * if it's not already there. Do it now before the folio is
1500 * moved to swap cache, when its pagelock no longer protects
1501 * the inode from eviction. But don't unlock the mutex until
1502 * we've incremented swapped, because shmem_unuse_inode() will
1503 * prune a !swapped inode from the swaplist under this mutex.
1505 mutex_lock(&shmem_swaplist_mutex);
1506 if (list_empty(&info->swaplist))
1507 list_add(&info->swaplist, &shmem_swaplist);
1509 if (add_to_swap_cache(folio, swap,
1510 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1512 shmem_recalc_inode(inode, 0, 1);
1513 swap_shmem_alloc(swap);
1514 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1516 mutex_unlock(&shmem_swaplist_mutex);
1517 BUG_ON(folio_mapped(folio));
1518 swap_writepage(&folio->page, wbc);
1522 mutex_unlock(&shmem_swaplist_mutex);
1523 put_swap_folio(folio, swap);
1525 folio_mark_dirty(folio);
1526 if (wbc->for_reclaim)
1527 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1528 folio_unlock(folio);
1532 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1533 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1537 if (!mpol || mpol->mode == MPOL_DEFAULT)
1538 return; /* show nothing */
1540 mpol_to_str(buffer, sizeof(buffer), mpol);
1542 seq_printf(seq, ",mpol=%s", buffer);
1545 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1547 struct mempolicy *mpol = NULL;
1549 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1550 mpol = sbinfo->mpol;
1552 raw_spin_unlock(&sbinfo->stat_lock);
1556 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1557 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1560 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1564 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1566 #define vm_policy vm_private_data
1569 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1570 struct shmem_inode_info *info, pgoff_t index)
1572 /* Create a pseudo vma that just contains the policy */
1573 vma_init(vma, NULL);
1574 /* Bias interleave by inode number to distribute better across nodes */
1575 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1576 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1579 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1581 /* Drop reference taken by mpol_shared_policy_lookup() */
1582 mpol_cond_put(vma->vm_policy);
1585 static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1586 struct shmem_inode_info *info, pgoff_t index)
1588 struct vm_area_struct pvma;
1590 struct vm_fault vmf = {
1594 shmem_pseudo_vma_init(&pvma, info, index);
1595 page = swap_cluster_readahead(swap, gfp, &vmf);
1596 shmem_pseudo_vma_destroy(&pvma);
1600 return page_folio(page);
1604 * Make sure huge_gfp is always more limited than limit_gfp.
1605 * Some of the flags set permissions, while others set limitations.
1607 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1609 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1610 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1611 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1612 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1614 /* Allow allocations only from the originally specified zones. */
1615 result |= zoneflags;
1618 * Minimize the result gfp by taking the union with the deny flags,
1619 * and the intersection of the allow flags.
1621 result |= (limit_gfp & denyflags);
1622 result |= (huge_gfp & limit_gfp) & allowflags;
1627 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1628 struct shmem_inode_info *info, pgoff_t index)
1630 struct vm_area_struct pvma;
1631 struct address_space *mapping = info->vfs_inode.i_mapping;
1633 struct folio *folio;
1635 hindex = round_down(index, HPAGE_PMD_NR);
1636 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1640 shmem_pseudo_vma_init(&pvma, info, hindex);
1641 folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1642 shmem_pseudo_vma_destroy(&pvma);
1644 count_vm_event(THP_FILE_FALLBACK);
1648 static struct folio *shmem_alloc_folio(gfp_t gfp,
1649 struct shmem_inode_info *info, pgoff_t index)
1651 struct vm_area_struct pvma;
1652 struct folio *folio;
1654 shmem_pseudo_vma_init(&pvma, info, index);
1655 folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1656 shmem_pseudo_vma_destroy(&pvma);
1661 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1662 pgoff_t index, bool huge)
1664 struct shmem_inode_info *info = SHMEM_I(inode);
1665 struct folio *folio;
1669 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1671 nr = huge ? HPAGE_PMD_NR : 1;
1673 err = shmem_inode_acct_block(inode, nr);
1678 folio = shmem_alloc_hugefolio(gfp, info, index);
1680 folio = shmem_alloc_folio(gfp, info, index);
1682 __folio_set_locked(folio);
1683 __folio_set_swapbacked(folio);
1688 shmem_inode_unacct_blocks(inode, nr);
1690 return ERR_PTR(err);
1694 * When a page is moved from swapcache to shmem filecache (either by the
1695 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1696 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1697 * ignorance of the mapping it belongs to. If that mapping has special
1698 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1699 * we may need to copy to a suitable page before moving to filecache.
1701 * In a future release, this may well be extended to respect cpuset and
1702 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1703 * but for now it is a simple matter of zone.
1705 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1707 return folio_zonenum(folio) > gfp_zone(gfp);
1710 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1711 struct shmem_inode_info *info, pgoff_t index)
1713 struct folio *old, *new;
1714 struct address_space *swap_mapping;
1720 entry = folio_swap_entry(old);
1721 swap_index = swp_offset(entry);
1722 swap_mapping = swap_address_space(entry);
1725 * We have arrived here because our zones are constrained, so don't
1726 * limit chance of success by further cpuset and node constraints.
1728 gfp &= ~GFP_CONSTRAINT_MASK;
1729 VM_BUG_ON_FOLIO(folio_test_large(old), old);
1730 new = shmem_alloc_folio(gfp, info, index);
1735 folio_copy(new, old);
1736 flush_dcache_folio(new);
1738 __folio_set_locked(new);
1739 __folio_set_swapbacked(new);
1740 folio_mark_uptodate(new);
1741 folio_set_swap_entry(new, entry);
1742 folio_set_swapcache(new);
1745 * Our caller will very soon move newpage out of swapcache, but it's
1746 * a nice clean interface for us to replace oldpage by newpage there.
1748 xa_lock_irq(&swap_mapping->i_pages);
1749 error = shmem_replace_entry(swap_mapping, swap_index, old, new);
1751 mem_cgroup_migrate(old, new);
1752 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1753 __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1754 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1755 __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
1757 xa_unlock_irq(&swap_mapping->i_pages);
1759 if (unlikely(error)) {
1761 * Is this possible? I think not, now that our callers check
1762 * both PageSwapCache and page_private after getting page lock;
1763 * but be defensive. Reverse old to newpage for clear and free.
1771 folio_clear_swapcache(old);
1772 old->private = NULL;
1775 folio_put_refs(old, 2);
1779 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1780 struct folio *folio, swp_entry_t swap)
1782 struct address_space *mapping = inode->i_mapping;
1783 swp_entry_t swapin_error;
1786 swapin_error = make_swapin_error_entry();
1787 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1788 swp_to_radix_entry(swap),
1789 swp_to_radix_entry(swapin_error), 0);
1790 if (old != swp_to_radix_entry(swap))
1793 folio_wait_writeback(folio);
1794 delete_from_swap_cache(folio);
1796 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
1797 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
1798 * in shmem_evict_inode().
1800 shmem_recalc_inode(inode, -1, -1);
1805 * Swap in the folio pointed to by *foliop.
1806 * Caller has to make sure that *foliop contains a valid swapped folio.
1807 * Returns 0 and the folio in foliop if success. On failure, returns the
1808 * error code and NULL in *foliop.
1810 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1811 struct folio **foliop, enum sgp_type sgp,
1812 gfp_t gfp, struct vm_area_struct *vma,
1813 vm_fault_t *fault_type)
1815 struct address_space *mapping = inode->i_mapping;
1816 struct shmem_inode_info *info = SHMEM_I(inode);
1817 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1818 struct swap_info_struct *si;
1819 struct folio *folio = NULL;
1823 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1824 swap = radix_to_swp_entry(*foliop);
1827 if (is_swapin_error_entry(swap))
1830 si = get_swap_device(swap);
1832 if (!shmem_confirm_swap(mapping, index, swap))
1838 /* Look it up and read it in.. */
1839 folio = swap_cache_get_folio(swap, NULL, 0);
1841 /* Or update major stats only when swapin succeeds?? */
1843 *fault_type |= VM_FAULT_MAJOR;
1844 count_vm_event(PGMAJFAULT);
1845 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1847 /* Here we actually start the io */
1848 folio = shmem_swapin(swap, gfp, info, index);
1855 /* We have to do this with folio locked to prevent races */
1857 if (!folio_test_swapcache(folio) ||
1858 folio_swap_entry(folio).val != swap.val ||
1859 !shmem_confirm_swap(mapping, index, swap)) {
1863 if (!folio_test_uptodate(folio)) {
1867 folio_wait_writeback(folio);
1870 * Some architectures may have to restore extra metadata to the
1871 * folio after reading from swap.
1873 arch_swap_restore(swap, folio);
1875 if (shmem_should_replace_folio(folio, gfp)) {
1876 error = shmem_replace_folio(&folio, gfp, info, index);
1881 error = shmem_add_to_page_cache(folio, mapping, index,
1882 swp_to_radix_entry(swap), gfp,
1887 shmem_recalc_inode(inode, 0, -1);
1889 if (sgp == SGP_WRITE)
1890 folio_mark_accessed(folio);
1892 delete_from_swap_cache(folio);
1893 folio_mark_dirty(folio);
1895 put_swap_device(si);
1900 if (!shmem_confirm_swap(mapping, index, swap))
1903 shmem_set_folio_swapin_error(inode, index, folio, swap);
1906 folio_unlock(folio);
1909 put_swap_device(si);
1915 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1917 * If we allocate a new one we do not mark it dirty. That's up to the
1918 * vm. If we swap it in we mark it dirty since we also free the swap
1919 * entry since a page cannot live in both the swap and page cache.
1921 * vma, vmf, and fault_type are only supplied by shmem_fault:
1922 * otherwise they are NULL.
1924 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1925 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
1926 struct vm_area_struct *vma, struct vm_fault *vmf,
1927 vm_fault_t *fault_type)
1929 struct address_space *mapping = inode->i_mapping;
1930 struct shmem_inode_info *info = SHMEM_I(inode);
1931 struct shmem_sb_info *sbinfo;
1932 struct mm_struct *charge_mm;
1933 struct folio *folio;
1940 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1943 if (sgp <= SGP_CACHE &&
1944 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1948 sbinfo = SHMEM_SB(inode->i_sb);
1949 charge_mm = vma ? vma->vm_mm : NULL;
1951 folio = filemap_get_entry(mapping, index);
1952 if (folio && vma && userfaultfd_minor(vma)) {
1953 if (!xa_is_value(folio))
1955 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1959 if (xa_is_value(folio)) {
1960 error = shmem_swapin_folio(inode, index, &folio,
1961 sgp, gfp, vma, fault_type);
1962 if (error == -EEXIST)
1972 /* Has the folio been truncated or swapped out? */
1973 if (unlikely(folio->mapping != mapping)) {
1974 folio_unlock(folio);
1978 if (sgp == SGP_WRITE)
1979 folio_mark_accessed(folio);
1980 if (folio_test_uptodate(folio))
1982 /* fallocated folio */
1983 if (sgp != SGP_READ)
1985 folio_unlock(folio);
1990 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1991 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1994 if (sgp == SGP_READ)
1996 if (sgp == SGP_NOALLOC)
2000 * Fast cache lookup and swap lookup did not find it: allocate.
2003 if (vma && userfaultfd_missing(vma)) {
2004 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
2008 if (!shmem_is_huge(inode, index, false,
2009 vma ? vma->vm_mm : NULL, vma ? vma->vm_flags : 0))
2012 huge_gfp = vma_thp_gfp_mask(vma);
2013 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
2014 folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
2015 if (IS_ERR(folio)) {
2017 folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
2019 if (IS_ERR(folio)) {
2022 error = PTR_ERR(folio);
2024 if (error != -ENOSPC)
2027 * Try to reclaim some space by splitting a large folio
2028 * beyond i_size on the filesystem.
2033 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
2034 if (ret == SHRINK_STOP)
2042 hindex = round_down(index, folio_nr_pages(folio));
2044 if (sgp == SGP_WRITE)
2045 __folio_set_referenced(folio);
2047 error = shmem_add_to_page_cache(folio, mapping, hindex,
2048 NULL, gfp & GFP_RECLAIM_MASK,
2053 folio_add_lru(folio);
2054 shmem_recalc_inode(inode, folio_nr_pages(folio), 0);
2057 if (folio_test_pmd_mappable(folio) &&
2058 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
2059 folio_next_index(folio) - 1) {
2061 * Part of the large folio is beyond i_size: subject
2062 * to shrink under memory pressure.
2064 spin_lock(&sbinfo->shrinklist_lock);
2066 * _careful to defend against unlocked access to
2067 * ->shrink_list in shmem_unused_huge_shrink()
2069 if (list_empty_careful(&info->shrinklist)) {
2070 list_add_tail(&info->shrinklist,
2071 &sbinfo->shrinklist);
2072 sbinfo->shrinklist_len++;
2074 spin_unlock(&sbinfo->shrinklist_lock);
2078 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2080 if (sgp == SGP_FALLOC)
2084 * Let SGP_WRITE caller clear ends if write does not fill folio;
2085 * but SGP_FALLOC on a folio fallocated earlier must initialize
2086 * it now, lest undo on failure cancel our earlier guarantee.
2088 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2089 long i, n = folio_nr_pages(folio);
2091 for (i = 0; i < n; i++)
2092 clear_highpage(folio_page(folio, i));
2093 flush_dcache_folio(folio);
2094 folio_mark_uptodate(folio);
2097 /* Perhaps the file has been truncated since we checked */
2098 if (sgp <= SGP_CACHE &&
2099 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2101 folio_clear_dirty(folio);
2102 filemap_remove_folio(folio);
2103 shmem_recalc_inode(inode, 0, 0);
2116 shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
2118 if (folio_test_large(folio)) {
2119 folio_unlock(folio);
2125 folio_unlock(folio);
2128 if (error == -ENOSPC && !once++) {
2129 shmem_recalc_inode(inode, 0, 0);
2132 if (error == -EEXIST)
2137 int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
2140 return shmem_get_folio_gfp(inode, index, foliop, sgp,
2141 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
2145 * This is like autoremove_wake_function, but it removes the wait queue
2146 * entry unconditionally - even if something else had already woken the
2149 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2151 int ret = default_wake_function(wait, mode, sync, key);
2152 list_del_init(&wait->entry);
2156 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2158 struct vm_area_struct *vma = vmf->vma;
2159 struct inode *inode = file_inode(vma->vm_file);
2160 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2161 struct folio *folio = NULL;
2163 vm_fault_t ret = VM_FAULT_LOCKED;
2166 * Trinity finds that probing a hole which tmpfs is punching can
2167 * prevent the hole-punch from ever completing: which in turn
2168 * locks writers out with its hold on i_rwsem. So refrain from
2169 * faulting pages into the hole while it's being punched. Although
2170 * shmem_undo_range() does remove the additions, it may be unable to
2171 * keep up, as each new page needs its own unmap_mapping_range() call,
2172 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2174 * It does not matter if we sometimes reach this check just before the
2175 * hole-punch begins, so that one fault then races with the punch:
2176 * we just need to make racing faults a rare case.
2178 * The implementation below would be much simpler if we just used a
2179 * standard mutex or completion: but we cannot take i_rwsem in fault,
2180 * and bloating every shmem inode for this unlikely case would be sad.
2182 if (unlikely(inode->i_private)) {
2183 struct shmem_falloc *shmem_falloc;
2185 spin_lock(&inode->i_lock);
2186 shmem_falloc = inode->i_private;
2188 shmem_falloc->waitq &&
2189 vmf->pgoff >= shmem_falloc->start &&
2190 vmf->pgoff < shmem_falloc->next) {
2192 wait_queue_head_t *shmem_falloc_waitq;
2193 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2195 ret = VM_FAULT_NOPAGE;
2196 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2198 ret = VM_FAULT_RETRY;
2200 shmem_falloc_waitq = shmem_falloc->waitq;
2201 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2202 TASK_UNINTERRUPTIBLE);
2203 spin_unlock(&inode->i_lock);
2207 * shmem_falloc_waitq points into the shmem_fallocate()
2208 * stack of the hole-punching task: shmem_falloc_waitq
2209 * is usually invalid by the time we reach here, but
2210 * finish_wait() does not dereference it in that case;
2211 * though i_lock needed lest racing with wake_up_all().
2213 spin_lock(&inode->i_lock);
2214 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2215 spin_unlock(&inode->i_lock);
2221 spin_unlock(&inode->i_lock);
2224 err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
2225 gfp, vma, vmf, &ret);
2227 return vmf_error(err);
2229 vmf->page = folio_file_page(folio, vmf->pgoff);
2233 unsigned long shmem_get_unmapped_area(struct file *file,
2234 unsigned long uaddr, unsigned long len,
2235 unsigned long pgoff, unsigned long flags)
2237 unsigned long (*get_area)(struct file *,
2238 unsigned long, unsigned long, unsigned long, unsigned long);
2240 unsigned long offset;
2241 unsigned long inflated_len;
2242 unsigned long inflated_addr;
2243 unsigned long inflated_offset;
2245 if (len > TASK_SIZE)
2248 get_area = current->mm->get_unmapped_area;
2249 addr = get_area(file, uaddr, len, pgoff, flags);
2251 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2253 if (IS_ERR_VALUE(addr))
2255 if (addr & ~PAGE_MASK)
2257 if (addr > TASK_SIZE - len)
2260 if (shmem_huge == SHMEM_HUGE_DENY)
2262 if (len < HPAGE_PMD_SIZE)
2264 if (flags & MAP_FIXED)
2267 * Our priority is to support MAP_SHARED mapped hugely;
2268 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2269 * But if caller specified an address hint and we allocated area there
2270 * successfully, respect that as before.
2275 if (shmem_huge != SHMEM_HUGE_FORCE) {
2276 struct super_block *sb;
2279 VM_BUG_ON(file->f_op != &shmem_file_operations);
2280 sb = file_inode(file)->i_sb;
2283 * Called directly from mm/mmap.c, or drivers/char/mem.c
2284 * for "/dev/zero", to create a shared anonymous object.
2286 if (IS_ERR(shm_mnt))
2288 sb = shm_mnt->mnt_sb;
2290 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2294 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2295 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2297 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2300 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2301 if (inflated_len > TASK_SIZE)
2303 if (inflated_len < len)
2306 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2307 if (IS_ERR_VALUE(inflated_addr))
2309 if (inflated_addr & ~PAGE_MASK)
2312 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2313 inflated_addr += offset - inflated_offset;
2314 if (inflated_offset > offset)
2315 inflated_addr += HPAGE_PMD_SIZE;
2317 if (inflated_addr > TASK_SIZE - len)
2319 return inflated_addr;
2323 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2325 struct inode *inode = file_inode(vma->vm_file);
2326 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2329 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2332 struct inode *inode = file_inode(vma->vm_file);
2335 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2336 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2340 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2342 struct inode *inode = file_inode(file);
2343 struct shmem_inode_info *info = SHMEM_I(inode);
2344 int retval = -ENOMEM;
2347 * What serializes the accesses to info->flags?
2348 * ipc_lock_object() when called from shmctl_do_lock(),
2349 * no serialization needed when called from shm_destroy().
2351 if (lock && !(info->flags & VM_LOCKED)) {
2352 if (!user_shm_lock(inode->i_size, ucounts))
2354 info->flags |= VM_LOCKED;
2355 mapping_set_unevictable(file->f_mapping);
2357 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2358 user_shm_unlock(inode->i_size, ucounts);
2359 info->flags &= ~VM_LOCKED;
2360 mapping_clear_unevictable(file->f_mapping);
2368 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2370 struct inode *inode = file_inode(file);
2371 struct shmem_inode_info *info = SHMEM_I(inode);
2374 ret = seal_check_future_write(info->seals, vma);
2378 /* arm64 - allow memory tagging on RAM-based files */
2379 vm_flags_set(vma, VM_MTE_ALLOWED);
2381 file_accessed(file);
2382 /* This is anonymous shared memory if it is unlinked at the time of mmap */
2384 vma->vm_ops = &shmem_vm_ops;
2386 vma->vm_ops = &shmem_anon_vm_ops;
2390 #ifdef CONFIG_TMPFS_XATTR
2391 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2394 * chattr's fsflags are unrelated to extended attributes,
2395 * but tmpfs has chosen to enable them under the same config option.
2397 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2399 unsigned int i_flags = 0;
2401 if (fsflags & FS_NOATIME_FL)
2402 i_flags |= S_NOATIME;
2403 if (fsflags & FS_APPEND_FL)
2404 i_flags |= S_APPEND;
2405 if (fsflags & FS_IMMUTABLE_FL)
2406 i_flags |= S_IMMUTABLE;
2408 * But FS_NODUMP_FL does not require any action in i_flags.
2410 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2413 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2416 #define shmem_initxattrs NULL
2419 static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
2421 return &SHMEM_I(inode)->dir_offsets;
2424 static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
2425 struct super_block *sb,
2426 struct inode *dir, umode_t mode,
2427 dev_t dev, unsigned long flags)
2429 struct inode *inode;
2430 struct shmem_inode_info *info;
2431 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2435 err = shmem_reserve_inode(sb, &ino);
2437 return ERR_PTR(err);
2440 inode = new_inode(sb);
2443 shmem_free_inode(sb);
2444 return ERR_PTR(-ENOSPC);
2448 inode_init_owner(idmap, inode, dir, mode);
2449 inode->i_blocks = 0;
2450 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2451 inode->i_generation = get_random_u32();
2452 info = SHMEM_I(inode);
2453 memset(info, 0, (char *)inode - (char *)info);
2454 spin_lock_init(&info->lock);
2455 atomic_set(&info->stop_eviction, 0);
2456 info->seals = F_SEAL_SEAL;
2457 info->flags = flags & VM_NORESERVE;
2458 info->i_crtime = inode->i_mtime;
2459 info->fsflags = (dir == NULL) ? 0 :
2460 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2462 shmem_set_inode_flags(inode, info->fsflags);
2463 INIT_LIST_HEAD(&info->shrinklist);
2464 INIT_LIST_HEAD(&info->swaplist);
2465 INIT_LIST_HEAD(&info->swaplist);
2467 mapping_set_unevictable(inode->i_mapping);
2468 simple_xattrs_init(&info->xattrs);
2469 cache_no_acl(inode);
2470 mapping_set_large_folios(inode->i_mapping);
2472 switch (mode & S_IFMT) {
2474 inode->i_op = &shmem_special_inode_operations;
2475 init_special_inode(inode, mode, dev);
2478 inode->i_mapping->a_ops = &shmem_aops;
2479 inode->i_op = &shmem_inode_operations;
2480 inode->i_fop = &shmem_file_operations;
2481 mpol_shared_policy_init(&info->policy,
2482 shmem_get_sbmpol(sbinfo));
2486 /* Some things misbehave if size == 0 on a directory */
2487 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2488 inode->i_op = &shmem_dir_inode_operations;
2489 inode->i_fop = &simple_offset_dir_operations;
2490 simple_offset_init(shmem_get_offset_ctx(inode));
2494 * Must not load anything in the rbtree,
2495 * mpol_free_shared_policy will not be called.
2497 mpol_shared_policy_init(&info->policy, NULL);
2501 lockdep_annotate_inode_mutex_key(inode);
2505 #ifdef CONFIG_TMPFS_QUOTA
2506 static struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2507 struct super_block *sb, struct inode *dir,
2508 umode_t mode, dev_t dev, unsigned long flags)
2511 struct inode *inode;
2513 inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2517 err = dquot_initialize(inode);
2521 err = dquot_alloc_inode(inode);
2529 inode->i_flags |= S_NOQUOTA;
2531 return ERR_PTR(err);
2534 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2535 struct super_block *sb, struct inode *dir,
2536 umode_t mode, dev_t dev, unsigned long flags)
2538 return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2540 #endif /* CONFIG_TMPFS_QUOTA */
2542 #ifdef CONFIG_USERFAULTFD
2543 int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
2544 struct vm_area_struct *dst_vma,
2545 unsigned long dst_addr,
2546 unsigned long src_addr,
2548 struct folio **foliop)
2550 struct inode *inode = file_inode(dst_vma->vm_file);
2551 struct shmem_inode_info *info = SHMEM_I(inode);
2552 struct address_space *mapping = inode->i_mapping;
2553 gfp_t gfp = mapping_gfp_mask(mapping);
2554 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2556 struct folio *folio;
2560 if (shmem_inode_acct_block(inode, 1)) {
2562 * We may have got a page, returned -ENOENT triggering a retry,
2563 * and now we find ourselves with -ENOMEM. Release the page, to
2564 * avoid a BUG_ON in our caller.
2566 if (unlikely(*foliop)) {
2575 folio = shmem_alloc_folio(gfp, info, pgoff);
2577 goto out_unacct_blocks;
2579 if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
2580 page_kaddr = kmap_local_folio(folio, 0);
2582 * The read mmap_lock is held here. Despite the
2583 * mmap_lock being read recursive a deadlock is still
2584 * possible if a writer has taken a lock. For example:
2586 * process A thread 1 takes read lock on own mmap_lock
2587 * process A thread 2 calls mmap, blocks taking write lock
2588 * process B thread 1 takes page fault, read lock on own mmap lock
2589 * process B thread 2 calls mmap, blocks taking write lock
2590 * process A thread 1 blocks taking read lock on process B
2591 * process B thread 1 blocks taking read lock on process A
2593 * Disable page faults to prevent potential deadlock
2594 * and retry the copy outside the mmap_lock.
2596 pagefault_disable();
2597 ret = copy_from_user(page_kaddr,
2598 (const void __user *)src_addr,
2601 kunmap_local(page_kaddr);
2603 /* fallback to copy_from_user outside mmap_lock */
2604 if (unlikely(ret)) {
2607 /* don't free the page */
2608 goto out_unacct_blocks;
2611 flush_dcache_folio(folio);
2612 } else { /* ZEROPAGE */
2613 clear_user_highpage(&folio->page, dst_addr);
2617 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2621 VM_BUG_ON(folio_test_locked(folio));
2622 VM_BUG_ON(folio_test_swapbacked(folio));
2623 __folio_set_locked(folio);
2624 __folio_set_swapbacked(folio);
2625 __folio_mark_uptodate(folio);
2628 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2629 if (unlikely(pgoff >= max_off))
2632 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2633 gfp & GFP_RECLAIM_MASK, dst_vma->vm_mm);
2637 ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
2638 &folio->page, true, flags);
2640 goto out_delete_from_cache;
2642 shmem_recalc_inode(inode, 1, 0);
2643 folio_unlock(folio);
2645 out_delete_from_cache:
2646 filemap_remove_folio(folio);
2648 folio_unlock(folio);
2651 shmem_inode_unacct_blocks(inode, 1);
2654 #endif /* CONFIG_USERFAULTFD */
2657 static const struct inode_operations shmem_symlink_inode_operations;
2658 static const struct inode_operations shmem_short_symlink_operations;
2661 shmem_write_begin(struct file *file, struct address_space *mapping,
2662 loff_t pos, unsigned len,
2663 struct page **pagep, void **fsdata)
2665 struct inode *inode = mapping->host;
2666 struct shmem_inode_info *info = SHMEM_I(inode);
2667 pgoff_t index = pos >> PAGE_SHIFT;
2668 struct folio *folio;
2671 /* i_rwsem is held by caller */
2672 if (unlikely(info->seals & (F_SEAL_GROW |
2673 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2674 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2676 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2680 ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
2685 *pagep = folio_file_page(folio, index);
2686 if (PageHWPoison(*pagep)) {
2687 folio_unlock(folio);
2697 shmem_write_end(struct file *file, struct address_space *mapping,
2698 loff_t pos, unsigned len, unsigned copied,
2699 struct page *page, void *fsdata)
2701 struct folio *folio = page_folio(page);
2702 struct inode *inode = mapping->host;
2704 if (pos + copied > inode->i_size)
2705 i_size_write(inode, pos + copied);
2707 if (!folio_test_uptodate(folio)) {
2708 if (copied < folio_size(folio)) {
2709 size_t from = offset_in_folio(folio, pos);
2710 folio_zero_segments(folio, 0, from,
2711 from + copied, folio_size(folio));
2713 folio_mark_uptodate(folio);
2715 folio_mark_dirty(folio);
2716 folio_unlock(folio);
2722 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2724 struct file *file = iocb->ki_filp;
2725 struct inode *inode = file_inode(file);
2726 struct address_space *mapping = inode->i_mapping;
2728 unsigned long offset;
2731 loff_t *ppos = &iocb->ki_pos;
2733 index = *ppos >> PAGE_SHIFT;
2734 offset = *ppos & ~PAGE_MASK;
2737 struct folio *folio = NULL;
2738 struct page *page = NULL;
2740 unsigned long nr, ret;
2741 loff_t i_size = i_size_read(inode);
2743 end_index = i_size >> PAGE_SHIFT;
2744 if (index > end_index)
2746 if (index == end_index) {
2747 nr = i_size & ~PAGE_MASK;
2752 error = shmem_get_folio(inode, index, &folio, SGP_READ);
2754 if (error == -EINVAL)
2759 folio_unlock(folio);
2761 page = folio_file_page(folio, index);
2762 if (PageHWPoison(page)) {
2770 * We must evaluate after, since reads (unlike writes)
2771 * are called without i_rwsem protection against truncate
2774 i_size = i_size_read(inode);
2775 end_index = i_size >> PAGE_SHIFT;
2776 if (index == end_index) {
2777 nr = i_size & ~PAGE_MASK;
2788 * If users can be writing to this page using arbitrary
2789 * virtual addresses, take care about potential aliasing
2790 * before reading the page on the kernel side.
2792 if (mapping_writably_mapped(mapping))
2793 flush_dcache_page(page);
2795 * Mark the page accessed if we read the beginning.
2798 folio_mark_accessed(folio);
2800 * Ok, we have the page, and it's up-to-date, so
2801 * now we can copy it to user space...
2803 ret = copy_page_to_iter(page, offset, nr, to);
2806 } else if (user_backed_iter(to)) {
2808 * Copy to user tends to be so well optimized, but
2809 * clear_user() not so much, that it is noticeably
2810 * faster to copy the zero page instead of clearing.
2812 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2815 * But submitting the same page twice in a row to
2816 * splice() - or others? - can result in confusion:
2817 * so don't attempt that optimization on pipes etc.
2819 ret = iov_iter_zero(nr, to);
2824 index += offset >> PAGE_SHIFT;
2825 offset &= ~PAGE_MASK;
2827 if (!iov_iter_count(to))
2836 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2837 file_accessed(file);
2838 return retval ? retval : error;
2841 static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
2842 struct pipe_buffer *buf)
2847 static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
2848 struct pipe_buffer *buf)
2852 static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
2853 struct pipe_buffer *buf)
2858 static const struct pipe_buf_operations zero_pipe_buf_ops = {
2859 .release = zero_pipe_buf_release,
2860 .try_steal = zero_pipe_buf_try_steal,
2861 .get = zero_pipe_buf_get,
2864 static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
2865 loff_t fpos, size_t size)
2867 size_t offset = fpos & ~PAGE_MASK;
2869 size = min_t(size_t, size, PAGE_SIZE - offset);
2871 if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2872 struct pipe_buffer *buf = pipe_head_buf(pipe);
2874 *buf = (struct pipe_buffer) {
2875 .ops = &zero_pipe_buf_ops,
2876 .page = ZERO_PAGE(0),
2886 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2887 struct pipe_inode_info *pipe,
2888 size_t len, unsigned int flags)
2890 struct inode *inode = file_inode(in);
2891 struct address_space *mapping = inode->i_mapping;
2892 struct folio *folio = NULL;
2893 size_t total_spliced = 0, used, npages, n, part;
2897 /* Work out how much data we can actually add into the pipe */
2898 used = pipe_occupancy(pipe->head, pipe->tail);
2899 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2900 len = min_t(size_t, len, npages * PAGE_SIZE);
2903 if (*ppos >= i_size_read(inode))
2906 error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio,
2909 if (error == -EINVAL)
2914 folio_unlock(folio);
2916 if (folio_test_hwpoison(folio) ||
2917 (folio_test_large(folio) &&
2918 folio_test_has_hwpoisoned(folio))) {
2925 * i_size must be checked after we know the pages are Uptodate.
2927 * Checking i_size after the check allows us to calculate
2928 * the correct value for "nr", which means the zero-filled
2929 * part of the page is not copied back to userspace (unless
2930 * another truncate extends the file - this is desired though).
2932 isize = i_size_read(inode);
2933 if (unlikely(*ppos >= isize))
2935 part = min_t(loff_t, isize - *ppos, len);
2939 * If users can be writing to this page using arbitrary
2940 * virtual addresses, take care about potential aliasing
2941 * before reading the page on the kernel side.
2943 if (mapping_writably_mapped(mapping))
2944 flush_dcache_folio(folio);
2945 folio_mark_accessed(folio);
2947 * Ok, we have the page, and it's up-to-date, so we can
2948 * now splice it into the pipe.
2950 n = splice_folio_into_pipe(pipe, folio, *ppos, part);
2954 n = splice_zeropage_into_pipe(pipe, *ppos, part);
2962 in->f_ra.prev_pos = *ppos;
2963 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2973 return total_spliced ? total_spliced : error;
2976 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2978 struct address_space *mapping = file->f_mapping;
2979 struct inode *inode = mapping->host;
2981 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2982 return generic_file_llseek_size(file, offset, whence,
2983 MAX_LFS_FILESIZE, i_size_read(inode));
2988 /* We're holding i_rwsem so we can access i_size directly */
2989 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2991 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2992 inode_unlock(inode);
2996 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2999 struct inode *inode = file_inode(file);
3000 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3001 struct shmem_inode_info *info = SHMEM_I(inode);
3002 struct shmem_falloc shmem_falloc;
3003 pgoff_t start, index, end, undo_fallocend;
3006 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3011 if (mode & FALLOC_FL_PUNCH_HOLE) {
3012 struct address_space *mapping = file->f_mapping;
3013 loff_t unmap_start = round_up(offset, PAGE_SIZE);
3014 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
3015 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
3017 /* protected by i_rwsem */
3018 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
3023 shmem_falloc.waitq = &shmem_falloc_waitq;
3024 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
3025 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
3026 spin_lock(&inode->i_lock);
3027 inode->i_private = &shmem_falloc;
3028 spin_unlock(&inode->i_lock);
3030 if ((u64)unmap_end > (u64)unmap_start)
3031 unmap_mapping_range(mapping, unmap_start,
3032 1 + unmap_end - unmap_start, 0);
3033 shmem_truncate_range(inode, offset, offset + len - 1);
3034 /* No need to unmap again: hole-punching leaves COWed pages */
3036 spin_lock(&inode->i_lock);
3037 inode->i_private = NULL;
3038 wake_up_all(&shmem_falloc_waitq);
3039 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
3040 spin_unlock(&inode->i_lock);
3045 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3046 error = inode_newsize_ok(inode, offset + len);
3050 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
3055 start = offset >> PAGE_SHIFT;
3056 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
3057 /* Try to avoid a swapstorm if len is impossible to satisfy */
3058 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
3063 shmem_falloc.waitq = NULL;
3064 shmem_falloc.start = start;
3065 shmem_falloc.next = start;
3066 shmem_falloc.nr_falloced = 0;
3067 shmem_falloc.nr_unswapped = 0;
3068 spin_lock(&inode->i_lock);
3069 inode->i_private = &shmem_falloc;
3070 spin_unlock(&inode->i_lock);
3073 * info->fallocend is only relevant when huge pages might be
3074 * involved: to prevent split_huge_page() freeing fallocated
3075 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3077 undo_fallocend = info->fallocend;
3078 if (info->fallocend < end)
3079 info->fallocend = end;
3081 for (index = start; index < end; ) {
3082 struct folio *folio;
3085 * Good, the fallocate(2) manpage permits EINTR: we may have
3086 * been interrupted because we are using up too much memory.
3088 if (signal_pending(current))
3090 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
3093 error = shmem_get_folio(inode, index, &folio,
3096 info->fallocend = undo_fallocend;
3097 /* Remove the !uptodate folios we added */
3098 if (index > start) {
3099 shmem_undo_range(inode,
3100 (loff_t)start << PAGE_SHIFT,
3101 ((loff_t)index << PAGE_SHIFT) - 1, true);
3107 * Here is a more important optimization than it appears:
3108 * a second SGP_FALLOC on the same large folio will clear it,
3109 * making it uptodate and un-undoable if we fail later.
3111 index = folio_next_index(folio);
3112 /* Beware 32-bit wraparound */
3117 * Inform shmem_writepage() how far we have reached.
3118 * No need for lock or barrier: we have the page lock.
3120 if (!folio_test_uptodate(folio))
3121 shmem_falloc.nr_falloced += index - shmem_falloc.next;
3122 shmem_falloc.next = index;
3125 * If !uptodate, leave it that way so that freeable folios
3126 * can be recognized if we need to rollback on error later.
3127 * But mark it dirty so that memory pressure will swap rather
3128 * than free the folios we are allocating (and SGP_CACHE folios
3129 * might still be clean: we now need to mark those dirty too).
3131 folio_mark_dirty(folio);
3132 folio_unlock(folio);
3137 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3138 i_size_write(inode, offset + len);
3140 spin_lock(&inode->i_lock);
3141 inode->i_private = NULL;
3142 spin_unlock(&inode->i_lock);
3145 file_modified(file);
3146 inode_unlock(inode);
3150 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
3152 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
3154 buf->f_type = TMPFS_MAGIC;
3155 buf->f_bsize = PAGE_SIZE;
3156 buf->f_namelen = NAME_MAX;
3157 if (sbinfo->max_blocks) {
3158 buf->f_blocks = sbinfo->max_blocks;
3160 buf->f_bfree = sbinfo->max_blocks -
3161 percpu_counter_sum(&sbinfo->used_blocks);
3163 if (sbinfo->max_inodes) {
3164 buf->f_files = sbinfo->max_inodes;
3165 buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE;
3167 /* else leave those fields 0 like simple_statfs */
3169 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3175 * File creation. Allocate an inode, and we're done..
3178 shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
3179 struct dentry *dentry, umode_t mode, dev_t dev)
3181 struct inode *inode;
3184 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
3187 return PTR_ERR(inode);
3189 error = simple_acl_create(dir, inode);
3192 error = security_inode_init_security(inode, dir,
3194 shmem_initxattrs, NULL);
3195 if (error && error != -EOPNOTSUPP)
3198 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3202 dir->i_size += BOGO_DIRENT_SIZE;
3203 dir->i_ctime = dir->i_mtime = current_time(dir);
3204 inode_inc_iversion(dir);
3205 d_instantiate(dentry, inode);
3206 dget(dentry); /* Extra count - pin the dentry in core */
3215 shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
3216 struct file *file, umode_t mode)
3218 struct inode *inode;
3221 inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
3223 if (IS_ERR(inode)) {
3224 error = PTR_ERR(inode);
3228 error = security_inode_init_security(inode, dir,
3230 shmem_initxattrs, NULL);
3231 if (error && error != -EOPNOTSUPP)
3233 error = simple_acl_create(dir, inode);
3236 d_tmpfile(file, inode);
3239 return finish_open_simple(file, error);
3245 static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3246 struct dentry *dentry, umode_t mode)
3250 error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3257 static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
3258 struct dentry *dentry, umode_t mode, bool excl)
3260 return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
3266 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3268 struct inode *inode = d_inode(old_dentry);
3272 * No ordinary (disk based) filesystem counts links as inodes;
3273 * but each new link needs a new dentry, pinning lowmem, and
3274 * tmpfs dentries cannot be pruned until they are unlinked.
3275 * But if an O_TMPFILE file is linked into the tmpfs, the
3276 * first link must skip that, to get the accounting right.
3278 if (inode->i_nlink) {
3279 ret = shmem_reserve_inode(inode->i_sb, NULL);
3284 ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3287 shmem_free_inode(inode->i_sb);
3291 dir->i_size += BOGO_DIRENT_SIZE;
3292 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3293 inode_inc_iversion(dir);
3295 ihold(inode); /* New dentry reference */
3296 dget(dentry); /* Extra pinning count for the created dentry */
3297 d_instantiate(dentry, inode);
3302 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3304 struct inode *inode = d_inode(dentry);
3306 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3307 shmem_free_inode(inode->i_sb);
3309 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3311 dir->i_size -= BOGO_DIRENT_SIZE;
3312 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3313 inode_inc_iversion(dir);
3315 dput(dentry); /* Undo the count from "create" - this does all the work */
3319 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3321 if (!simple_empty(dentry))
3324 drop_nlink(d_inode(dentry));
3326 return shmem_unlink(dir, dentry);
3329 static int shmem_whiteout(struct mnt_idmap *idmap,
3330 struct inode *old_dir, struct dentry *old_dentry)
3332 struct dentry *whiteout;
3335 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3339 error = shmem_mknod(idmap, old_dir, whiteout,
3340 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3346 * Cheat and hash the whiteout while the old dentry is still in
3347 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3349 * d_lookup() will consistently find one of them at this point,
3350 * not sure which one, but that isn't even important.
3357 * The VFS layer already does all the dentry stuff for rename,
3358 * we just have to decrement the usage count for the target if
3359 * it exists so that the VFS layer correctly free's it when it
3362 static int shmem_rename2(struct mnt_idmap *idmap,
3363 struct inode *old_dir, struct dentry *old_dentry,
3364 struct inode *new_dir, struct dentry *new_dentry,
3367 struct inode *inode = d_inode(old_dentry);
3368 int they_are_dirs = S_ISDIR(inode->i_mode);
3371 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3374 if (flags & RENAME_EXCHANGE)
3375 return simple_offset_rename_exchange(old_dir, old_dentry,
3376 new_dir, new_dentry);
3378 if (!simple_empty(new_dentry))
3381 if (flags & RENAME_WHITEOUT) {
3382 error = shmem_whiteout(idmap, old_dir, old_dentry);
3387 simple_offset_remove(shmem_get_offset_ctx(old_dir), old_dentry);
3388 error = simple_offset_add(shmem_get_offset_ctx(new_dir), old_dentry);
3392 if (d_really_is_positive(new_dentry)) {
3393 (void) shmem_unlink(new_dir, new_dentry);
3394 if (they_are_dirs) {
3395 drop_nlink(d_inode(new_dentry));
3396 drop_nlink(old_dir);
3398 } else if (they_are_dirs) {
3399 drop_nlink(old_dir);
3403 old_dir->i_size -= BOGO_DIRENT_SIZE;
3404 new_dir->i_size += BOGO_DIRENT_SIZE;
3405 old_dir->i_ctime = old_dir->i_mtime =
3406 new_dir->i_ctime = new_dir->i_mtime =
3407 inode->i_ctime = current_time(old_dir);
3408 inode_inc_iversion(old_dir);
3409 inode_inc_iversion(new_dir);
3413 static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3414 struct dentry *dentry, const char *symname)
3418 struct inode *inode;
3419 struct folio *folio;
3421 len = strlen(symname) + 1;
3422 if (len > PAGE_SIZE)
3423 return -ENAMETOOLONG;
3425 inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3429 return PTR_ERR(inode);
3431 error = security_inode_init_security(inode, dir, &dentry->d_name,
3432 shmem_initxattrs, NULL);
3433 if (error && error != -EOPNOTSUPP)
3436 error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3440 inode->i_size = len-1;
3441 if (len <= SHORT_SYMLINK_LEN) {
3442 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3443 if (!inode->i_link) {
3445 goto out_remove_offset;
3447 inode->i_op = &shmem_short_symlink_operations;
3449 inode_nohighmem(inode);
3450 error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
3452 goto out_remove_offset;
3453 inode->i_mapping->a_ops = &shmem_aops;
3454 inode->i_op = &shmem_symlink_inode_operations;
3455 memcpy(folio_address(folio), symname, len);
3456 folio_mark_uptodate(folio);
3457 folio_mark_dirty(folio);
3458 folio_unlock(folio);
3461 dir->i_size += BOGO_DIRENT_SIZE;
3462 dir->i_ctime = dir->i_mtime = current_time(dir);
3463 inode_inc_iversion(dir);
3464 d_instantiate(dentry, inode);
3469 simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3475 static void shmem_put_link(void *arg)
3477 folio_mark_accessed(arg);
3481 static const char *shmem_get_link(struct dentry *dentry,
3482 struct inode *inode,
3483 struct delayed_call *done)
3485 struct folio *folio = NULL;
3489 folio = filemap_get_folio(inode->i_mapping, 0);
3491 return ERR_PTR(-ECHILD);
3492 if (PageHWPoison(folio_page(folio, 0)) ||
3493 !folio_test_uptodate(folio)) {
3495 return ERR_PTR(-ECHILD);
3498 error = shmem_get_folio(inode, 0, &folio, SGP_READ);
3500 return ERR_PTR(error);
3502 return ERR_PTR(-ECHILD);
3503 if (PageHWPoison(folio_page(folio, 0))) {
3504 folio_unlock(folio);
3506 return ERR_PTR(-ECHILD);
3508 folio_unlock(folio);
3510 set_delayed_call(done, shmem_put_link, folio);
3511 return folio_address(folio);
3514 #ifdef CONFIG_TMPFS_XATTR
3516 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3518 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3520 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3525 static int shmem_fileattr_set(struct mnt_idmap *idmap,
3526 struct dentry *dentry, struct fileattr *fa)
3528 struct inode *inode = d_inode(dentry);
3529 struct shmem_inode_info *info = SHMEM_I(inode);
3531 if (fileattr_has_fsx(fa))
3533 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3536 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3537 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3539 shmem_set_inode_flags(inode, info->fsflags);
3540 inode->i_ctime = current_time(inode);
3541 inode_inc_iversion(inode);
3546 * Superblocks without xattr inode operations may get some security.* xattr
3547 * support from the LSM "for free". As soon as we have any other xattrs
3548 * like ACLs, we also need to implement the security.* handlers at
3549 * filesystem level, though.
3553 * Callback for security_inode_init_security() for acquiring xattrs.
3555 static int shmem_initxattrs(struct inode *inode,
3556 const struct xattr *xattr_array,
3559 struct shmem_inode_info *info = SHMEM_I(inode);
3560 const struct xattr *xattr;
3561 struct simple_xattr *new_xattr;
3564 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3565 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3569 len = strlen(xattr->name) + 1;
3570 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3572 if (!new_xattr->name) {
3577 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3578 XATTR_SECURITY_PREFIX_LEN);
3579 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3582 simple_xattr_add(&info->xattrs, new_xattr);
3588 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3589 struct dentry *unused, struct inode *inode,
3590 const char *name, void *buffer, size_t size)
3592 struct shmem_inode_info *info = SHMEM_I(inode);
3594 name = xattr_full_name(handler, name);
3595 return simple_xattr_get(&info->xattrs, name, buffer, size);
3598 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3599 struct mnt_idmap *idmap,
3600 struct dentry *unused, struct inode *inode,
3601 const char *name, const void *value,
3602 size_t size, int flags)
3604 struct shmem_inode_info *info = SHMEM_I(inode);
3605 struct simple_xattr *old_xattr;
3607 name = xattr_full_name(handler, name);
3608 old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags);
3609 if (!IS_ERR(old_xattr)) {
3610 simple_xattr_free(old_xattr);
3612 inode->i_ctime = current_time(inode);
3613 inode_inc_iversion(inode);
3615 return PTR_ERR(old_xattr);
3618 static const struct xattr_handler shmem_security_xattr_handler = {
3619 .prefix = XATTR_SECURITY_PREFIX,
3620 .get = shmem_xattr_handler_get,
3621 .set = shmem_xattr_handler_set,
3624 static const struct xattr_handler shmem_trusted_xattr_handler = {
3625 .prefix = XATTR_TRUSTED_PREFIX,
3626 .get = shmem_xattr_handler_get,
3627 .set = shmem_xattr_handler_set,
3630 static const struct xattr_handler *shmem_xattr_handlers[] = {
3631 &shmem_security_xattr_handler,
3632 &shmem_trusted_xattr_handler,
3636 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3638 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3639 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3641 #endif /* CONFIG_TMPFS_XATTR */
3643 static const struct inode_operations shmem_short_symlink_operations = {
3644 .getattr = shmem_getattr,
3645 .setattr = shmem_setattr,
3646 .get_link = simple_get_link,
3647 #ifdef CONFIG_TMPFS_XATTR
3648 .listxattr = shmem_listxattr,
3652 static const struct inode_operations shmem_symlink_inode_operations = {
3653 .getattr = shmem_getattr,
3654 .setattr = shmem_setattr,
3655 .get_link = shmem_get_link,
3656 #ifdef CONFIG_TMPFS_XATTR
3657 .listxattr = shmem_listxattr,
3661 static struct dentry *shmem_get_parent(struct dentry *child)
3663 return ERR_PTR(-ESTALE);
3666 static int shmem_match(struct inode *ino, void *vfh)
3670 inum = (inum << 32) | fh[1];
3671 return ino->i_ino == inum && fh[0] == ino->i_generation;
3674 /* Find any alias of inode, but prefer a hashed alias */
3675 static struct dentry *shmem_find_alias(struct inode *inode)
3677 struct dentry *alias = d_find_alias(inode);
3679 return alias ?: d_find_any_alias(inode);
3683 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3684 struct fid *fid, int fh_len, int fh_type)
3686 struct inode *inode;
3687 struct dentry *dentry = NULL;
3694 inum = (inum << 32) | fid->raw[1];
3696 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3697 shmem_match, fid->raw);
3699 dentry = shmem_find_alias(inode);
3706 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3707 struct inode *parent)
3711 return FILEID_INVALID;
3714 if (inode_unhashed(inode)) {
3715 /* Unfortunately insert_inode_hash is not idempotent,
3716 * so as we hash inodes here rather than at creation
3717 * time, we need a lock to ensure we only try
3720 static DEFINE_SPINLOCK(lock);
3722 if (inode_unhashed(inode))
3723 __insert_inode_hash(inode,
3724 inode->i_ino + inode->i_generation);
3728 fh[0] = inode->i_generation;
3729 fh[1] = inode->i_ino;
3730 fh[2] = ((__u64)inode->i_ino) >> 32;
3736 static const struct export_operations shmem_export_ops = {
3737 .get_parent = shmem_get_parent,
3738 .encode_fh = shmem_encode_fh,
3739 .fh_to_dentry = shmem_fh_to_dentry,
3757 Opt_usrquota_block_hardlimit,
3758 Opt_usrquota_inode_hardlimit,
3759 Opt_grpquota_block_hardlimit,
3760 Opt_grpquota_inode_hardlimit,
3763 static const struct constant_table shmem_param_enums_huge[] = {
3764 {"never", SHMEM_HUGE_NEVER },
3765 {"always", SHMEM_HUGE_ALWAYS },
3766 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3767 {"advise", SHMEM_HUGE_ADVISE },
3771 const struct fs_parameter_spec shmem_fs_parameters[] = {
3772 fsparam_u32 ("gid", Opt_gid),
3773 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3774 fsparam_u32oct("mode", Opt_mode),
3775 fsparam_string("mpol", Opt_mpol),
3776 fsparam_string("nr_blocks", Opt_nr_blocks),
3777 fsparam_string("nr_inodes", Opt_nr_inodes),
3778 fsparam_string("size", Opt_size),
3779 fsparam_u32 ("uid", Opt_uid),
3780 fsparam_flag ("inode32", Opt_inode32),
3781 fsparam_flag ("inode64", Opt_inode64),
3782 fsparam_flag ("noswap", Opt_noswap),
3783 #ifdef CONFIG_TMPFS_QUOTA
3784 fsparam_flag ("quota", Opt_quota),
3785 fsparam_flag ("usrquota", Opt_usrquota),
3786 fsparam_flag ("grpquota", Opt_grpquota),
3787 fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
3788 fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
3789 fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
3790 fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
3795 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3797 struct shmem_options *ctx = fc->fs_private;
3798 struct fs_parse_result result;
3799 unsigned long long size;
3805 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3811 size = memparse(param->string, &rest);
3813 size <<= PAGE_SHIFT;
3814 size *= totalram_pages();
3820 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3821 ctx->seen |= SHMEM_SEEN_BLOCKS;
3824 ctx->blocks = memparse(param->string, &rest);
3825 if (*rest || ctx->blocks > LONG_MAX)
3827 ctx->seen |= SHMEM_SEEN_BLOCKS;
3830 ctx->inodes = memparse(param->string, &rest);
3831 if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE)
3833 ctx->seen |= SHMEM_SEEN_INODES;
3836 ctx->mode = result.uint_32 & 07777;
3839 kuid = make_kuid(current_user_ns(), result.uint_32);
3840 if (!uid_valid(kuid))
3844 * The requested uid must be representable in the
3845 * filesystem's idmapping.
3847 if (!kuid_has_mapping(fc->user_ns, kuid))
3853 kgid = make_kgid(current_user_ns(), result.uint_32);
3854 if (!gid_valid(kgid))
3858 * The requested gid must be representable in the
3859 * filesystem's idmapping.
3861 if (!kgid_has_mapping(fc->user_ns, kgid))
3867 ctx->huge = result.uint_32;
3868 if (ctx->huge != SHMEM_HUGE_NEVER &&
3869 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3870 has_transparent_hugepage()))
3871 goto unsupported_parameter;
3872 ctx->seen |= SHMEM_SEEN_HUGE;
3875 if (IS_ENABLED(CONFIG_NUMA)) {
3876 mpol_put(ctx->mpol);
3878 if (mpol_parse_str(param->string, &ctx->mpol))
3882 goto unsupported_parameter;
3884 ctx->full_inums = false;
3885 ctx->seen |= SHMEM_SEEN_INUMS;
3888 if (sizeof(ino_t) < 8) {
3890 "Cannot use inode64 with <64bit inums in kernel\n");
3892 ctx->full_inums = true;
3893 ctx->seen |= SHMEM_SEEN_INUMS;
3896 if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
3898 "Turning off swap in unprivileged tmpfs mounts unsupported");
3901 ctx->seen |= SHMEM_SEEN_NOSWAP;
3904 if (fc->user_ns != &init_user_ns)
3905 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3906 ctx->seen |= SHMEM_SEEN_QUOTA;
3907 ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
3910 if (fc->user_ns != &init_user_ns)
3911 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3912 ctx->seen |= SHMEM_SEEN_QUOTA;
3913 ctx->quota_types |= QTYPE_MASK_USR;
3916 if (fc->user_ns != &init_user_ns)
3917 return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
3918 ctx->seen |= SHMEM_SEEN_QUOTA;
3919 ctx->quota_types |= QTYPE_MASK_GRP;
3921 case Opt_usrquota_block_hardlimit:
3922 size = memparse(param->string, &rest);
3925 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
3927 "User quota block hardlimit too large.");
3928 ctx->qlimits.usrquota_bhardlimit = size;
3930 case Opt_grpquota_block_hardlimit:
3931 size = memparse(param->string, &rest);
3934 if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
3936 "Group quota block hardlimit too large.");
3937 ctx->qlimits.grpquota_bhardlimit = size;
3939 case Opt_usrquota_inode_hardlimit:
3940 size = memparse(param->string, &rest);
3943 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
3945 "User quota inode hardlimit too large.");
3946 ctx->qlimits.usrquota_ihardlimit = size;
3948 case Opt_grpquota_inode_hardlimit:
3949 size = memparse(param->string, &rest);
3952 if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
3954 "Group quota inode hardlimit too large.");
3955 ctx->qlimits.grpquota_ihardlimit = size;
3960 unsupported_parameter:
3961 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3963 return invalfc(fc, "Bad value for '%s'", param->key);
3966 static int shmem_parse_options(struct fs_context *fc, void *data)
3968 char *options = data;
3971 int err = security_sb_eat_lsm_opts(options, &fc->security);
3976 while (options != NULL) {
3977 char *this_char = options;
3980 * NUL-terminate this option: unfortunately,
3981 * mount options form a comma-separated list,
3982 * but mpol's nodelist may also contain commas.
3984 options = strchr(options, ',');
3985 if (options == NULL)
3988 if (!isdigit(*options)) {
3994 char *value = strchr(this_char, '=');
4000 len = strlen(value);
4002 err = vfs_parse_fs_string(fc, this_char, value, len);
4011 * Reconfigure a shmem filesystem.
4013 static int shmem_reconfigure(struct fs_context *fc)
4015 struct shmem_options *ctx = fc->fs_private;
4016 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
4017 unsigned long used_isp;
4018 struct mempolicy *mpol = NULL;
4021 raw_spin_lock(&sbinfo->stat_lock);
4022 used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace;
4024 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
4025 if (!sbinfo->max_blocks) {
4026 err = "Cannot retroactively limit size";
4029 if (percpu_counter_compare(&sbinfo->used_blocks,
4031 err = "Too small a size for current use";
4035 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
4036 if (!sbinfo->max_inodes) {
4037 err = "Cannot retroactively limit inodes";
4040 if (ctx->inodes * BOGO_INODE_SIZE < used_isp) {
4041 err = "Too few inodes for current use";
4046 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
4047 sbinfo->next_ino > UINT_MAX) {
4048 err = "Current inum too high to switch to 32-bit inums";
4051 if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
4052 err = "Cannot disable swap on remount";
4055 if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
4056 err = "Cannot enable swap on remount if it was disabled on first mount";
4060 if (ctx->seen & SHMEM_SEEN_QUOTA &&
4061 !sb_any_quota_loaded(fc->root->d_sb)) {
4062 err = "Cannot enable quota on remount";
4066 #ifdef CONFIG_TMPFS_QUOTA
4067 #define CHANGED_LIMIT(name) \
4068 (ctx->qlimits.name## hardlimit && \
4069 (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4071 if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
4072 CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
4073 err = "Cannot change global quota limit on remount";
4076 #endif /* CONFIG_TMPFS_QUOTA */
4078 if (ctx->seen & SHMEM_SEEN_HUGE)
4079 sbinfo->huge = ctx->huge;
4080 if (ctx->seen & SHMEM_SEEN_INUMS)
4081 sbinfo->full_inums = ctx->full_inums;
4082 if (ctx->seen & SHMEM_SEEN_BLOCKS)
4083 sbinfo->max_blocks = ctx->blocks;
4084 if (ctx->seen & SHMEM_SEEN_INODES) {
4085 sbinfo->max_inodes = ctx->inodes;
4086 sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp;
4090 * Preserve previous mempolicy unless mpol remount option was specified.
4093 mpol = sbinfo->mpol;
4094 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
4099 sbinfo->noswap = true;
4101 raw_spin_unlock(&sbinfo->stat_lock);
4105 raw_spin_unlock(&sbinfo->stat_lock);
4106 return invalfc(fc, "%s", err);
4109 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
4111 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
4112 struct mempolicy *mpol;
4114 if (sbinfo->max_blocks != shmem_default_max_blocks())
4115 seq_printf(seq, ",size=%luk",
4116 sbinfo->max_blocks << (PAGE_SHIFT - 10));
4117 if (sbinfo->max_inodes != shmem_default_max_inodes())
4118 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
4119 if (sbinfo->mode != (0777 | S_ISVTX))
4120 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
4121 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
4122 seq_printf(seq, ",uid=%u",
4123 from_kuid_munged(&init_user_ns, sbinfo->uid));
4124 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
4125 seq_printf(seq, ",gid=%u",
4126 from_kgid_munged(&init_user_ns, sbinfo->gid));
4129 * Showing inode{64,32} might be useful even if it's the system default,
4130 * since then people don't have to resort to checking both here and
4131 * /proc/config.gz to confirm 64-bit inums were successfully applied
4132 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4134 * We hide it when inode64 isn't the default and we are using 32-bit
4135 * inodes, since that probably just means the feature isn't even under
4140 * +-----------------+-----------------+
4141 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
4142 * +------------------+-----------------+-----------------+
4143 * | full_inums=true | show | show |
4144 * | full_inums=false | show | hide |
4145 * +------------------+-----------------+-----------------+
4148 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
4149 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
4150 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4151 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4153 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
4155 mpol = shmem_get_sbmpol(sbinfo);
4156 shmem_show_mpol(seq, mpol);
4159 seq_printf(seq, ",noswap");
4163 #endif /* CONFIG_TMPFS */
4165 static void shmem_put_super(struct super_block *sb)
4167 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
4169 #ifdef CONFIG_TMPFS_QUOTA
4170 shmem_disable_quotas(sb);
4172 free_percpu(sbinfo->ino_batch);
4173 percpu_counter_destroy(&sbinfo->used_blocks);
4174 mpol_put(sbinfo->mpol);
4176 sb->s_fs_info = NULL;
4179 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
4181 struct shmem_options *ctx = fc->fs_private;
4182 struct inode *inode;
4183 struct shmem_sb_info *sbinfo;
4184 int error = -ENOMEM;
4186 /* Round up to L1_CACHE_BYTES to resist false sharing */
4187 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
4188 L1_CACHE_BYTES), GFP_KERNEL);
4192 sb->s_fs_info = sbinfo;
4196 * Per default we only allow half of the physical ram per
4197 * tmpfs instance, limiting inodes to one per page of lowmem;
4198 * but the internal instance is left unlimited.
4200 if (!(sb->s_flags & SB_KERNMOUNT)) {
4201 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
4202 ctx->blocks = shmem_default_max_blocks();
4203 if (!(ctx->seen & SHMEM_SEEN_INODES))
4204 ctx->inodes = shmem_default_max_inodes();
4205 if (!(ctx->seen & SHMEM_SEEN_INUMS))
4206 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
4207 sbinfo->noswap = ctx->noswap;
4209 sb->s_flags |= SB_NOUSER;
4211 sb->s_export_op = &shmem_export_ops;
4212 sb->s_flags |= SB_NOSEC | SB_I_VERSION;
4214 sb->s_flags |= SB_NOUSER;
4216 sbinfo->max_blocks = ctx->blocks;
4217 sbinfo->max_inodes = ctx->inodes;
4218 sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE;
4219 if (sb->s_flags & SB_KERNMOUNT) {
4220 sbinfo->ino_batch = alloc_percpu(ino_t);
4221 if (!sbinfo->ino_batch)
4224 sbinfo->uid = ctx->uid;
4225 sbinfo->gid = ctx->gid;
4226 sbinfo->full_inums = ctx->full_inums;
4227 sbinfo->mode = ctx->mode;
4228 sbinfo->huge = ctx->huge;
4229 sbinfo->mpol = ctx->mpol;
4232 raw_spin_lock_init(&sbinfo->stat_lock);
4233 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
4235 spin_lock_init(&sbinfo->shrinklist_lock);
4236 INIT_LIST_HEAD(&sbinfo->shrinklist);
4238 sb->s_maxbytes = MAX_LFS_FILESIZE;
4239 sb->s_blocksize = PAGE_SIZE;
4240 sb->s_blocksize_bits = PAGE_SHIFT;
4241 sb->s_magic = TMPFS_MAGIC;
4242 sb->s_op = &shmem_ops;
4243 sb->s_time_gran = 1;
4244 #ifdef CONFIG_TMPFS_XATTR
4245 sb->s_xattr = shmem_xattr_handlers;
4247 #ifdef CONFIG_TMPFS_POSIX_ACL
4248 sb->s_flags |= SB_POSIXACL;
4250 uuid_gen(&sb->s_uuid);
4252 #ifdef CONFIG_TMPFS_QUOTA
4253 if (ctx->seen & SHMEM_SEEN_QUOTA) {
4254 sb->dq_op = &shmem_quota_operations;
4255 sb->s_qcop = &dquot_quotactl_sysfile_ops;
4256 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4258 /* Copy the default limits from ctx into sbinfo */
4259 memcpy(&sbinfo->qlimits, &ctx->qlimits,
4260 sizeof(struct shmem_quota_limits));
4262 if (shmem_enable_quotas(sb, ctx->quota_types))
4265 #endif /* CONFIG_TMPFS_QUOTA */
4267 inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, S_IFDIR | sbinfo->mode, 0,
4269 if (IS_ERR(inode)) {
4270 error = PTR_ERR(inode);
4273 inode->i_uid = sbinfo->uid;
4274 inode->i_gid = sbinfo->gid;
4275 sb->s_root = d_make_root(inode);
4281 shmem_put_super(sb);
4285 static int shmem_get_tree(struct fs_context *fc)
4287 return get_tree_nodev(fc, shmem_fill_super);
4290 static void shmem_free_fc(struct fs_context *fc)
4292 struct shmem_options *ctx = fc->fs_private;
4295 mpol_put(ctx->mpol);
4300 static const struct fs_context_operations shmem_fs_context_ops = {
4301 .free = shmem_free_fc,
4302 .get_tree = shmem_get_tree,
4304 .parse_monolithic = shmem_parse_options,
4305 .parse_param = shmem_parse_one,
4306 .reconfigure = shmem_reconfigure,
4310 static struct kmem_cache *shmem_inode_cachep;
4312 static struct inode *shmem_alloc_inode(struct super_block *sb)
4314 struct shmem_inode_info *info;
4315 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
4318 return &info->vfs_inode;
4321 static void shmem_free_in_core_inode(struct inode *inode)
4323 if (S_ISLNK(inode->i_mode))
4324 kfree(inode->i_link);
4325 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4328 static void shmem_destroy_inode(struct inode *inode)
4330 if (S_ISREG(inode->i_mode))
4331 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
4332 if (S_ISDIR(inode->i_mode))
4333 simple_offset_destroy(shmem_get_offset_ctx(inode));
4336 static void shmem_init_inode(void *foo)
4338 struct shmem_inode_info *info = foo;
4339 inode_init_once(&info->vfs_inode);
4342 static void shmem_init_inodecache(void)
4344 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4345 sizeof(struct shmem_inode_info),
4346 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
4349 static void shmem_destroy_inodecache(void)
4351 kmem_cache_destroy(shmem_inode_cachep);
4354 /* Keep the page in page cache instead of truncating it */
4355 static int shmem_error_remove_page(struct address_space *mapping,
4361 const struct address_space_operations shmem_aops = {
4362 .writepage = shmem_writepage,
4363 .dirty_folio = noop_dirty_folio,
4365 .write_begin = shmem_write_begin,
4366 .write_end = shmem_write_end,
4368 #ifdef CONFIG_MIGRATION
4369 .migrate_folio = migrate_folio,
4371 .error_remove_page = shmem_error_remove_page,
4373 EXPORT_SYMBOL(shmem_aops);
4375 static const struct file_operations shmem_file_operations = {
4377 .open = generic_file_open,
4378 .get_unmapped_area = shmem_get_unmapped_area,
4380 .llseek = shmem_file_llseek,
4381 .read_iter = shmem_file_read_iter,
4382 .write_iter = generic_file_write_iter,
4383 .fsync = noop_fsync,
4384 .splice_read = shmem_file_splice_read,
4385 .splice_write = iter_file_splice_write,
4386 .fallocate = shmem_fallocate,
4390 static const struct inode_operations shmem_inode_operations = {
4391 .getattr = shmem_getattr,
4392 .setattr = shmem_setattr,
4393 #ifdef CONFIG_TMPFS_XATTR
4394 .listxattr = shmem_listxattr,
4395 .set_acl = simple_set_acl,
4396 .fileattr_get = shmem_fileattr_get,
4397 .fileattr_set = shmem_fileattr_set,
4401 static const struct inode_operations shmem_dir_inode_operations = {
4403 .getattr = shmem_getattr,
4404 .create = shmem_create,
4405 .lookup = simple_lookup,
4407 .unlink = shmem_unlink,
4408 .symlink = shmem_symlink,
4409 .mkdir = shmem_mkdir,
4410 .rmdir = shmem_rmdir,
4411 .mknod = shmem_mknod,
4412 .rename = shmem_rename2,
4413 .tmpfile = shmem_tmpfile,
4414 .get_offset_ctx = shmem_get_offset_ctx,
4416 #ifdef CONFIG_TMPFS_XATTR
4417 .listxattr = shmem_listxattr,
4418 .fileattr_get = shmem_fileattr_get,
4419 .fileattr_set = shmem_fileattr_set,
4421 #ifdef CONFIG_TMPFS_POSIX_ACL
4422 .setattr = shmem_setattr,
4423 .set_acl = simple_set_acl,
4427 static const struct inode_operations shmem_special_inode_operations = {
4428 .getattr = shmem_getattr,
4429 #ifdef CONFIG_TMPFS_XATTR
4430 .listxattr = shmem_listxattr,
4432 #ifdef CONFIG_TMPFS_POSIX_ACL
4433 .setattr = shmem_setattr,
4434 .set_acl = simple_set_acl,
4438 static const struct super_operations shmem_ops = {
4439 .alloc_inode = shmem_alloc_inode,
4440 .free_inode = shmem_free_in_core_inode,
4441 .destroy_inode = shmem_destroy_inode,
4443 .statfs = shmem_statfs,
4444 .show_options = shmem_show_options,
4446 #ifdef CONFIG_TMPFS_QUOTA
4447 .get_dquots = shmem_get_dquots,
4449 .evict_inode = shmem_evict_inode,
4450 .drop_inode = generic_delete_inode,
4451 .put_super = shmem_put_super,
4452 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4453 .nr_cached_objects = shmem_unused_huge_count,
4454 .free_cached_objects = shmem_unused_huge_scan,
4458 static const struct vm_operations_struct shmem_vm_ops = {
4459 .fault = shmem_fault,
4460 .map_pages = filemap_map_pages,
4462 .set_policy = shmem_set_policy,
4463 .get_policy = shmem_get_policy,
4467 static const struct vm_operations_struct shmem_anon_vm_ops = {
4468 .fault = shmem_fault,
4469 .map_pages = filemap_map_pages,
4471 .set_policy = shmem_set_policy,
4472 .get_policy = shmem_get_policy,
4476 int shmem_init_fs_context(struct fs_context *fc)
4478 struct shmem_options *ctx;
4480 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4484 ctx->mode = 0777 | S_ISVTX;
4485 ctx->uid = current_fsuid();
4486 ctx->gid = current_fsgid();
4488 fc->fs_private = ctx;
4489 fc->ops = &shmem_fs_context_ops;
4493 static struct file_system_type shmem_fs_type = {
4494 .owner = THIS_MODULE,
4496 .init_fs_context = shmem_init_fs_context,
4498 .parameters = shmem_fs_parameters,
4500 .kill_sb = kill_litter_super,
4502 .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4504 .fs_flags = FS_USERNS_MOUNT,
4508 void __init shmem_init(void)
4512 shmem_init_inodecache();
4514 #ifdef CONFIG_TMPFS_QUOTA
4515 error = register_quota_format(&shmem_quota_format);
4517 pr_err("Could not register quota format\n");
4522 error = register_filesystem(&shmem_fs_type);
4524 pr_err("Could not register tmpfs\n");
4528 shm_mnt = kern_mount(&shmem_fs_type);
4529 if (IS_ERR(shm_mnt)) {
4530 error = PTR_ERR(shm_mnt);
4531 pr_err("Could not kern_mount tmpfs\n");
4535 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4536 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4537 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4539 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4544 unregister_filesystem(&shmem_fs_type);
4546 #ifdef CONFIG_TMPFS_QUOTA
4547 unregister_quota_format(&shmem_quota_format);
4550 shmem_destroy_inodecache();
4551 shm_mnt = ERR_PTR(error);
4554 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4555 static ssize_t shmem_enabled_show(struct kobject *kobj,
4556 struct kobj_attribute *attr, char *buf)
4558 static const int values[] = {
4560 SHMEM_HUGE_WITHIN_SIZE,
4569 for (i = 0; i < ARRAY_SIZE(values); i++) {
4570 len += sysfs_emit_at(buf, len,
4571 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4573 shmem_format_huge(values[i]));
4576 len += sysfs_emit_at(buf, len, "\n");
4581 static ssize_t shmem_enabled_store(struct kobject *kobj,
4582 struct kobj_attribute *attr, const char *buf, size_t count)
4587 if (count + 1 > sizeof(tmp))
4589 memcpy(tmp, buf, count);
4591 if (count && tmp[count - 1] == '\n')
4592 tmp[count - 1] = '\0';
4594 huge = shmem_parse_huge(tmp);
4595 if (huge == -EINVAL)
4597 if (!has_transparent_hugepage() &&
4598 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4602 if (shmem_huge > SHMEM_HUGE_DENY)
4603 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4607 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
4608 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4610 #else /* !CONFIG_SHMEM */
4613 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4615 * This is intended for small system where the benefits of the full
4616 * shmem code (swap-backed and resource-limited) are outweighed by
4617 * their complexity. On systems without swap this code should be
4618 * effectively equivalent, but much lighter weight.
4621 static struct file_system_type shmem_fs_type = {
4623 .init_fs_context = ramfs_init_fs_context,
4624 .parameters = ramfs_fs_parameters,
4625 .kill_sb = ramfs_kill_sb,
4626 .fs_flags = FS_USERNS_MOUNT,
4629 void __init shmem_init(void)
4631 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4633 shm_mnt = kern_mount(&shmem_fs_type);
4634 BUG_ON(IS_ERR(shm_mnt));
4637 int shmem_unuse(unsigned int type)
4642 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4647 void shmem_unlock_mapping(struct address_space *mapping)
4652 unsigned long shmem_get_unmapped_area(struct file *file,
4653 unsigned long addr, unsigned long len,
4654 unsigned long pgoff, unsigned long flags)
4656 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4660 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4662 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4664 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4666 #define shmem_vm_ops generic_file_vm_ops
4667 #define shmem_anon_vm_ops generic_file_vm_ops
4668 #define shmem_file_operations ramfs_file_operations
4669 #define shmem_acct_size(flags, size) 0
4670 #define shmem_unacct_size(flags, size) do {} while (0)
4672 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block *sb, struct inode *dir,
4673 umode_t mode, dev_t dev, unsigned long flags)
4675 struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
4676 return inode ? inode : ERR_PTR(-ENOSPC);
4679 #endif /* CONFIG_SHMEM */
4683 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4684 unsigned long flags, unsigned int i_flags)
4686 struct inode *inode;
4690 return ERR_CAST(mnt);
4692 if (size < 0 || size > MAX_LFS_FILESIZE)
4693 return ERR_PTR(-EINVAL);
4695 if (shmem_acct_size(flags, size))
4696 return ERR_PTR(-ENOMEM);
4698 if (is_idmapped_mnt(mnt))
4699 return ERR_PTR(-EINVAL);
4701 inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
4702 S_IFREG | S_IRWXUGO, 0, flags);
4704 if (IS_ERR(inode)) {
4705 shmem_unacct_size(flags, size);
4706 return ERR_CAST(inode);
4708 inode->i_flags |= i_flags;
4709 inode->i_size = size;
4710 clear_nlink(inode); /* It is unlinked */
4711 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4713 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4714 &shmem_file_operations);
4721 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4722 * kernel internal. There will be NO LSM permission checks against the
4723 * underlying inode. So users of this interface must do LSM checks at a
4724 * higher layer. The users are the big_key and shm implementations. LSM
4725 * checks are provided at the key or shm level rather than the inode.
4726 * @name: name for dentry (to be seen in /proc/<pid>/maps
4727 * @size: size to be set for the file
4728 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4730 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4732 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4736 * shmem_file_setup - get an unlinked file living in tmpfs
4737 * @name: name for dentry (to be seen in /proc/<pid>/maps
4738 * @size: size to be set for the file
4739 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4741 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4743 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4745 EXPORT_SYMBOL_GPL(shmem_file_setup);
4748 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4749 * @mnt: the tmpfs mount where the file will be created
4750 * @name: name for dentry (to be seen in /proc/<pid>/maps
4751 * @size: size to be set for the file
4752 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4754 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4755 loff_t size, unsigned long flags)
4757 return __shmem_file_setup(mnt, name, size, flags, 0);
4759 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4762 * shmem_zero_setup - setup a shared anonymous mapping
4763 * @vma: the vma to be mmapped is prepared by do_mmap
4765 int shmem_zero_setup(struct vm_area_struct *vma)
4768 loff_t size = vma->vm_end - vma->vm_start;
4771 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4772 * between XFS directory reading and selinux: since this file is only
4773 * accessible to the user through its mapping, use S_PRIVATE flag to
4774 * bypass file security, in the same way as shmem_kernel_file_setup().
4776 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4778 return PTR_ERR(file);
4782 vma->vm_file = file;
4783 vma->vm_ops = &shmem_anon_vm_ops;
4789 * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
4790 * @mapping: the folio's address_space
4791 * @index: the folio index
4792 * @gfp: the page allocator flags to use if allocating
4794 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4795 * with any new page allocations done using the specified allocation flags.
4796 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4797 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4798 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4800 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4801 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4803 struct folio *shmem_read_folio_gfp(struct address_space *mapping,
4804 pgoff_t index, gfp_t gfp)
4807 struct inode *inode = mapping->host;
4808 struct folio *folio;
4811 BUG_ON(!shmem_mapping(mapping));
4812 error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
4813 gfp, NULL, NULL, NULL);
4815 return ERR_PTR(error);
4817 folio_unlock(folio);
4821 * The tiny !SHMEM case uses ramfs without swap
4823 return mapping_read_folio_gfp(mapping, index, gfp);
4826 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
4828 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4829 pgoff_t index, gfp_t gfp)
4831 struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
4835 return &folio->page;
4837 page = folio_file_page(folio, index);
4838 if (PageHWPoison(page)) {
4840 return ERR_PTR(-EIO);
4845 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);