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/swap.h>
37 #include <linux/uio.h>
38 #include <linux/hugetlb.h>
39 #include <linux/fs_parser.h>
40 #include <linux/swapfile.h>
43 static struct vfsmount *shm_mnt;
47 * This virtual memory filesystem is heavily based on the ramfs. It
48 * extends ramfs by the ability to use swap and honor resource limits
49 * which makes it a completely usable filesystem.
52 #include <linux/xattr.h>
53 #include <linux/exportfs.h>
54 #include <linux/posix_acl.h>
55 #include <linux/posix_acl_xattr.h>
56 #include <linux/mman.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/backing-dev.h>
60 #include <linux/shmem_fs.h>
61 #include <linux/writeback.h>
62 #include <linux/pagevec.h>
63 #include <linux/percpu_counter.h>
64 #include <linux/falloc.h>
65 #include <linux/splice.h>
66 #include <linux/security.h>
67 #include <linux/swapops.h>
68 #include <linux/mempolicy.h>
69 #include <linux/namei.h>
70 #include <linux/ctype.h>
71 #include <linux/migrate.h>
72 #include <linux/highmem.h>
73 #include <linux/seq_file.h>
74 #include <linux/magic.h>
75 #include <linux/syscalls.h>
76 #include <linux/fcntl.h>
77 #include <uapi/linux/memfd.h>
78 #include <linux/userfaultfd_k.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
82 #include <linux/uaccess.h>
86 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 /* Pretend that each entry is of this size in directory's i_size */
90 #define BOGO_DIRENT_SIZE 20
92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93 #define SHORT_SYMLINK_LEN 128
96 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_rwsem making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
100 struct shmem_falloc {
101 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
102 pgoff_t start; /* start of range currently being fallocated */
103 pgoff_t next; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
108 struct shmem_options {
109 unsigned long long blocks;
110 unsigned long long inodes;
111 struct mempolicy *mpol;
118 #define SHMEM_SEEN_BLOCKS 1
119 #define SHMEM_SEEN_INODES 2
120 #define SHMEM_SEEN_HUGE 4
121 #define SHMEM_SEEN_INUMS 8
125 static unsigned long shmem_default_max_blocks(void)
127 return totalram_pages() / 2;
130 static unsigned long shmem_default_max_inodes(void)
132 unsigned long nr_pages = totalram_pages();
134 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
138 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
139 struct folio **foliop, enum sgp_type sgp,
140 gfp_t gfp, struct vm_area_struct *vma,
141 vm_fault_t *fault_type);
143 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
145 return sb->s_fs_info;
149 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
150 * for shared memory and for shared anonymous (/dev/zero) mappings
151 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
152 * consistent with the pre-accounting of private mappings ...
154 static inline int shmem_acct_size(unsigned long flags, loff_t size)
156 return (flags & VM_NORESERVE) ?
157 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
160 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
162 if (!(flags & VM_NORESERVE))
163 vm_unacct_memory(VM_ACCT(size));
166 static inline int shmem_reacct_size(unsigned long flags,
167 loff_t oldsize, loff_t newsize)
169 if (!(flags & VM_NORESERVE)) {
170 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
171 return security_vm_enough_memory_mm(current->mm,
172 VM_ACCT(newsize) - VM_ACCT(oldsize));
173 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
174 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
180 * ... whereas tmpfs objects are accounted incrementally as
181 * pages are allocated, in order to allow large sparse files.
182 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
183 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
185 static inline int shmem_acct_block(unsigned long flags, long pages)
187 if (!(flags & VM_NORESERVE))
190 return security_vm_enough_memory_mm(current->mm,
191 pages * VM_ACCT(PAGE_SIZE));
194 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
196 if (flags & VM_NORESERVE)
197 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
200 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
202 struct shmem_inode_info *info = SHMEM_I(inode);
203 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
205 if (shmem_acct_block(info->flags, pages))
208 if (sbinfo->max_blocks) {
209 if (percpu_counter_compare(&sbinfo->used_blocks,
210 sbinfo->max_blocks - pages) > 0)
212 percpu_counter_add(&sbinfo->used_blocks, pages);
218 shmem_unacct_blocks(info->flags, pages);
222 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
224 struct shmem_inode_info *info = SHMEM_I(inode);
225 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
227 if (sbinfo->max_blocks)
228 percpu_counter_sub(&sbinfo->used_blocks, pages);
229 shmem_unacct_blocks(info->flags, pages);
232 static const struct super_operations shmem_ops;
233 const struct address_space_operations shmem_aops;
234 static const struct file_operations shmem_file_operations;
235 static const struct inode_operations shmem_inode_operations;
236 static const struct inode_operations shmem_dir_inode_operations;
237 static const struct inode_operations shmem_special_inode_operations;
238 static const struct vm_operations_struct shmem_vm_ops;
239 static struct file_system_type shmem_fs_type;
241 bool vma_is_shmem(struct vm_area_struct *vma)
243 return vma->vm_ops == &shmem_vm_ops;
246 static LIST_HEAD(shmem_swaplist);
247 static DEFINE_MUTEX(shmem_swaplist_mutex);
250 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
251 * produces a novel ino for the newly allocated inode.
253 * It may also be called when making a hard link to permit the space needed by
254 * each dentry. However, in that case, no new inode number is needed since that
255 * internally draws from another pool of inode numbers (currently global
256 * get_next_ino()). This case is indicated by passing NULL as inop.
258 #define SHMEM_INO_BATCH 1024
259 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
261 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
264 if (!(sb->s_flags & SB_KERNMOUNT)) {
265 raw_spin_lock(&sbinfo->stat_lock);
266 if (sbinfo->max_inodes) {
267 if (!sbinfo->free_inodes) {
268 raw_spin_unlock(&sbinfo->stat_lock);
271 sbinfo->free_inodes--;
274 ino = sbinfo->next_ino++;
275 if (unlikely(is_zero_ino(ino)))
276 ino = sbinfo->next_ino++;
277 if (unlikely(!sbinfo->full_inums &&
280 * Emulate get_next_ino uint wraparound for
283 if (IS_ENABLED(CONFIG_64BIT))
284 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
285 __func__, MINOR(sb->s_dev));
286 sbinfo->next_ino = 1;
287 ino = sbinfo->next_ino++;
291 raw_spin_unlock(&sbinfo->stat_lock);
294 * __shmem_file_setup, one of our callers, is lock-free: it
295 * doesn't hold stat_lock in shmem_reserve_inode since
296 * max_inodes is always 0, and is called from potentially
297 * unknown contexts. As such, use a per-cpu batched allocator
298 * which doesn't require the per-sb stat_lock unless we are at
299 * the batch boundary.
301 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
302 * shmem mounts are not exposed to userspace, so we don't need
303 * to worry about things like glibc compatibility.
307 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
309 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
310 raw_spin_lock(&sbinfo->stat_lock);
311 ino = sbinfo->next_ino;
312 sbinfo->next_ino += SHMEM_INO_BATCH;
313 raw_spin_unlock(&sbinfo->stat_lock);
314 if (unlikely(is_zero_ino(ino)))
325 static void shmem_free_inode(struct super_block *sb)
327 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
328 if (sbinfo->max_inodes) {
329 raw_spin_lock(&sbinfo->stat_lock);
330 sbinfo->free_inodes++;
331 raw_spin_unlock(&sbinfo->stat_lock);
336 * shmem_recalc_inode - recalculate the block usage of an inode
337 * @inode: inode to recalc
339 * We have to calculate the free blocks since the mm can drop
340 * undirtied hole pages behind our back.
342 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
343 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
345 * It has to be called with the spinlock held.
347 static void shmem_recalc_inode(struct inode *inode)
349 struct shmem_inode_info *info = SHMEM_I(inode);
352 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
354 info->alloced -= freed;
355 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
356 shmem_inode_unacct_blocks(inode, freed);
360 bool shmem_charge(struct inode *inode, long pages)
362 struct shmem_inode_info *info = SHMEM_I(inode);
365 if (!shmem_inode_acct_block(inode, pages))
368 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
369 inode->i_mapping->nrpages += pages;
371 spin_lock_irqsave(&info->lock, flags);
372 info->alloced += pages;
373 inode->i_blocks += pages * BLOCKS_PER_PAGE;
374 shmem_recalc_inode(inode);
375 spin_unlock_irqrestore(&info->lock, flags);
380 void shmem_uncharge(struct inode *inode, long pages)
382 struct shmem_inode_info *info = SHMEM_I(inode);
385 /* nrpages adjustment done by __filemap_remove_folio() or caller */
387 spin_lock_irqsave(&info->lock, flags);
388 info->alloced -= pages;
389 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
390 shmem_recalc_inode(inode);
391 spin_unlock_irqrestore(&info->lock, flags);
393 shmem_inode_unacct_blocks(inode, pages);
397 * Replace item expected in xarray by a new item, while holding xa_lock.
399 static int shmem_replace_entry(struct address_space *mapping,
400 pgoff_t index, void *expected, void *replacement)
402 XA_STATE(xas, &mapping->i_pages, index);
405 VM_BUG_ON(!expected);
406 VM_BUG_ON(!replacement);
407 item = xas_load(&xas);
408 if (item != expected)
410 xas_store(&xas, replacement);
415 * Sometimes, before we decide whether to proceed or to fail, we must check
416 * that an entry was not already brought back from swap by a racing thread.
418 * Checking page is not enough: by the time a SwapCache page is locked, it
419 * might be reused, and again be SwapCache, using the same swap as before.
421 static bool shmem_confirm_swap(struct address_space *mapping,
422 pgoff_t index, swp_entry_t swap)
424 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
428 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
431 * disables huge pages for the mount;
433 * enables huge pages for the mount;
434 * SHMEM_HUGE_WITHIN_SIZE:
435 * only allocate huge pages if the page will be fully within i_size,
436 * also respect fadvise()/madvise() hints;
438 * only allocate huge pages if requested with fadvise()/madvise();
441 #define SHMEM_HUGE_NEVER 0
442 #define SHMEM_HUGE_ALWAYS 1
443 #define SHMEM_HUGE_WITHIN_SIZE 2
444 #define SHMEM_HUGE_ADVISE 3
448 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
451 * disables huge on shm_mnt and all mounts, for emergency use;
453 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
456 #define SHMEM_HUGE_DENY (-1)
457 #define SHMEM_HUGE_FORCE (-2)
459 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
460 /* ifdef here to avoid bloating shmem.o when not necessary */
462 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
464 bool shmem_is_huge(struct vm_area_struct *vma,
465 struct inode *inode, pgoff_t index)
469 if (!S_ISREG(inode->i_mode))
471 if (shmem_huge == SHMEM_HUGE_DENY)
473 if (vma && ((vma->vm_flags & VM_NOHUGEPAGE) ||
474 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
476 if (shmem_huge == SHMEM_HUGE_FORCE)
479 switch (SHMEM_SB(inode->i_sb)->huge) {
480 case SHMEM_HUGE_ALWAYS:
482 case SHMEM_HUGE_WITHIN_SIZE:
483 index = round_up(index + 1, HPAGE_PMD_NR);
484 i_size = round_up(i_size_read(inode), PAGE_SIZE);
485 if (i_size >> PAGE_SHIFT >= index)
488 case SHMEM_HUGE_ADVISE:
489 if (vma && (vma->vm_flags & VM_HUGEPAGE))
497 #if defined(CONFIG_SYSFS)
498 static int shmem_parse_huge(const char *str)
500 if (!strcmp(str, "never"))
501 return SHMEM_HUGE_NEVER;
502 if (!strcmp(str, "always"))
503 return SHMEM_HUGE_ALWAYS;
504 if (!strcmp(str, "within_size"))
505 return SHMEM_HUGE_WITHIN_SIZE;
506 if (!strcmp(str, "advise"))
507 return SHMEM_HUGE_ADVISE;
508 if (!strcmp(str, "deny"))
509 return SHMEM_HUGE_DENY;
510 if (!strcmp(str, "force"))
511 return SHMEM_HUGE_FORCE;
516 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
517 static const char *shmem_format_huge(int huge)
520 case SHMEM_HUGE_NEVER:
522 case SHMEM_HUGE_ALWAYS:
524 case SHMEM_HUGE_WITHIN_SIZE:
525 return "within_size";
526 case SHMEM_HUGE_ADVISE:
528 case SHMEM_HUGE_DENY:
530 case SHMEM_HUGE_FORCE:
539 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
540 struct shrink_control *sc, unsigned long nr_to_split)
542 LIST_HEAD(list), *pos, *next;
543 LIST_HEAD(to_remove);
545 struct shmem_inode_info *info;
547 unsigned long batch = sc ? sc->nr_to_scan : 128;
550 if (list_empty(&sbinfo->shrinklist))
553 spin_lock(&sbinfo->shrinklist_lock);
554 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
555 info = list_entry(pos, struct shmem_inode_info, shrinklist);
558 inode = igrab(&info->vfs_inode);
560 /* inode is about to be evicted */
562 list_del_init(&info->shrinklist);
566 /* Check if there's anything to gain */
567 if (round_up(inode->i_size, PAGE_SIZE) ==
568 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
569 list_move(&info->shrinklist, &to_remove);
573 list_move(&info->shrinklist, &list);
575 sbinfo->shrinklist_len--;
579 spin_unlock(&sbinfo->shrinklist_lock);
581 list_for_each_safe(pos, next, &to_remove) {
582 info = list_entry(pos, struct shmem_inode_info, shrinklist);
583 inode = &info->vfs_inode;
584 list_del_init(&info->shrinklist);
588 list_for_each_safe(pos, next, &list) {
592 info = list_entry(pos, struct shmem_inode_info, shrinklist);
593 inode = &info->vfs_inode;
595 if (nr_to_split && split >= nr_to_split)
598 index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
599 folio = filemap_get_folio(inode->i_mapping, index);
603 /* No huge page at the end of the file: nothing to split */
604 if (!folio_test_large(folio)) {
610 * Move the inode on the list back to shrinklist if we failed
611 * to lock the page at this time.
613 * Waiting for the lock may lead to deadlock in the
616 if (!folio_trylock(folio)) {
621 ret = split_folio(folio);
625 /* If split failed move the inode on the list back to shrinklist */
631 list_del_init(&info->shrinklist);
635 * Make sure the inode is either on the global list or deleted
636 * from any local list before iput() since it could be deleted
637 * in another thread once we put the inode (then the local list
640 spin_lock(&sbinfo->shrinklist_lock);
641 list_move(&info->shrinklist, &sbinfo->shrinklist);
642 sbinfo->shrinklist_len++;
643 spin_unlock(&sbinfo->shrinklist_lock);
651 static long shmem_unused_huge_scan(struct super_block *sb,
652 struct shrink_control *sc)
654 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
656 if (!READ_ONCE(sbinfo->shrinklist_len))
659 return shmem_unused_huge_shrink(sbinfo, sc, 0);
662 static long shmem_unused_huge_count(struct super_block *sb,
663 struct shrink_control *sc)
665 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
666 return READ_ONCE(sbinfo->shrinklist_len);
668 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
670 #define shmem_huge SHMEM_HUGE_DENY
672 bool shmem_is_huge(struct vm_area_struct *vma,
673 struct inode *inode, pgoff_t index)
678 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
679 struct shrink_control *sc, unsigned long nr_to_split)
683 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
686 * Like filemap_add_folio, but error if expected item has gone.
688 static int shmem_add_to_page_cache(struct folio *folio,
689 struct address_space *mapping,
690 pgoff_t index, void *expected, gfp_t gfp,
691 struct mm_struct *charge_mm)
693 XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
694 long nr = folio_nr_pages(folio);
697 VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
698 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
699 VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
700 VM_BUG_ON(expected && folio_test_large(folio));
702 folio_ref_add(folio, nr);
703 folio->mapping = mapping;
704 folio->index = index;
706 if (!folio_test_swapcache(folio)) {
707 error = mem_cgroup_charge(folio, charge_mm, gfp);
709 if (folio_test_pmd_mappable(folio)) {
710 count_vm_event(THP_FILE_FALLBACK);
711 count_vm_event(THP_FILE_FALLBACK_CHARGE);
716 folio_throttle_swaprate(folio, gfp);
720 if (expected != xas_find_conflict(&xas)) {
721 xas_set_err(&xas, -EEXIST);
724 if (expected && xas_find_conflict(&xas)) {
725 xas_set_err(&xas, -EEXIST);
728 xas_store(&xas, folio);
731 if (folio_test_pmd_mappable(folio)) {
732 count_vm_event(THP_FILE_ALLOC);
733 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
735 mapping->nrpages += nr;
736 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
737 __lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
739 xas_unlock_irq(&xas);
740 } while (xas_nomem(&xas, gfp));
742 if (xas_error(&xas)) {
743 error = xas_error(&xas);
749 folio->mapping = NULL;
750 folio_ref_sub(folio, nr);
755 * Like delete_from_page_cache, but substitutes swap for @folio.
757 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
759 struct address_space *mapping = folio->mapping;
760 long nr = folio_nr_pages(folio);
763 xa_lock_irq(&mapping->i_pages);
764 error = shmem_replace_entry(mapping, folio->index, folio, radswap);
765 folio->mapping = NULL;
766 mapping->nrpages -= nr;
767 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
768 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
769 xa_unlock_irq(&mapping->i_pages);
775 * Remove swap entry from page cache, free the swap and its page cache.
777 static int shmem_free_swap(struct address_space *mapping,
778 pgoff_t index, void *radswap)
782 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
785 free_swap_and_cache(radix_to_swp_entry(radswap));
790 * Determine (in bytes) how many of the shmem object's pages mapped by the
791 * given offsets are swapped out.
793 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
794 * as long as the inode doesn't go away and racy results are not a problem.
796 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
797 pgoff_t start, pgoff_t end)
799 XA_STATE(xas, &mapping->i_pages, start);
801 unsigned long swapped = 0;
804 xas_for_each(&xas, page, end - 1) {
805 if (xas_retry(&xas, page))
807 if (xa_is_value(page))
810 if (need_resched()) {
818 return swapped << PAGE_SHIFT;
822 * Determine (in bytes) how many of the shmem object's pages mapped by the
823 * given vma is swapped out.
825 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
826 * as long as the inode doesn't go away and racy results are not a problem.
828 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
830 struct inode *inode = file_inode(vma->vm_file);
831 struct shmem_inode_info *info = SHMEM_I(inode);
832 struct address_space *mapping = inode->i_mapping;
833 unsigned long swapped;
835 /* Be careful as we don't hold info->lock */
836 swapped = READ_ONCE(info->swapped);
839 * The easier cases are when the shmem object has nothing in swap, or
840 * the vma maps it whole. Then we can simply use the stats that we
846 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
847 return swapped << PAGE_SHIFT;
849 /* Here comes the more involved part */
850 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
851 vma->vm_pgoff + vma_pages(vma));
855 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
857 void shmem_unlock_mapping(struct address_space *mapping)
859 struct folio_batch fbatch;
862 folio_batch_init(&fbatch);
864 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
866 while (!mapping_unevictable(mapping) &&
867 filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
868 check_move_unevictable_folios(&fbatch);
869 folio_batch_release(&fbatch);
874 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
880 * At first avoid shmem_getpage(,,,SGP_READ): that fails
881 * beyond i_size, and reports fallocated pages as holes.
883 folio = __filemap_get_folio(inode->i_mapping, index,
884 FGP_ENTRY | FGP_LOCK, 0);
885 if (!xa_is_value(folio))
888 * But read a page back from swap if any of it is within i_size
889 * (although in some cases this is just a waste of time).
892 shmem_getpage(inode, index, &page, SGP_READ);
893 return page ? page_folio(page) : NULL;
897 * Remove range of pages and swap entries from page cache, and free them.
898 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
900 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
903 struct address_space *mapping = inode->i_mapping;
904 struct shmem_inode_info *info = SHMEM_I(inode);
905 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
906 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
907 struct folio_batch fbatch;
908 pgoff_t indices[PAGEVEC_SIZE];
911 long nr_swaps_freed = 0;
916 end = -1; /* unsigned, so actually very big */
918 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
919 info->fallocend = start;
921 folio_batch_init(&fbatch);
923 while (index < end && find_lock_entries(mapping, index, end - 1,
925 for (i = 0; i < folio_batch_count(&fbatch); i++) {
926 folio = fbatch.folios[i];
930 if (xa_is_value(folio)) {
933 nr_swaps_freed += !shmem_free_swap(mapping,
937 index += folio_nr_pages(folio) - 1;
939 if (!unfalloc || !folio_test_uptodate(folio))
940 truncate_inode_folio(mapping, folio);
943 folio_batch_remove_exceptionals(&fbatch);
944 folio_batch_release(&fbatch);
949 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
950 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
952 same_folio = lend < folio_pos(folio) + folio_size(folio);
953 folio_mark_dirty(folio);
954 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
955 start = folio->index + folio_nr_pages(folio);
965 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
967 folio_mark_dirty(folio);
968 if (!truncate_inode_partial_folio(folio, lstart, lend))
975 while (index < end) {
978 if (!find_get_entries(mapping, index, end - 1, &fbatch,
980 /* If all gone or hole-punch or unfalloc, we're done */
981 if (index == start || end != -1)
983 /* But if truncating, restart to make sure all gone */
987 for (i = 0; i < folio_batch_count(&fbatch); i++) {
988 folio = fbatch.folios[i];
991 if (xa_is_value(folio)) {
994 if (shmem_free_swap(mapping, index, folio)) {
995 /* Swap was replaced by page: retry */
1005 if (!unfalloc || !folio_test_uptodate(folio)) {
1006 if (folio_mapping(folio) != mapping) {
1007 /* Page was replaced by swap: retry */
1008 folio_unlock(folio);
1012 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1014 truncate_inode_folio(mapping, folio);
1016 index = folio->index + folio_nr_pages(folio) - 1;
1017 folio_unlock(folio);
1019 folio_batch_remove_exceptionals(&fbatch);
1020 folio_batch_release(&fbatch);
1024 spin_lock_irq(&info->lock);
1025 info->swapped -= nr_swaps_freed;
1026 shmem_recalc_inode(inode);
1027 spin_unlock_irq(&info->lock);
1030 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1032 shmem_undo_range(inode, lstart, lend, false);
1033 inode->i_ctime = inode->i_mtime = current_time(inode);
1035 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1037 static int shmem_getattr(struct user_namespace *mnt_userns,
1038 const struct path *path, struct kstat *stat,
1039 u32 request_mask, unsigned int query_flags)
1041 struct inode *inode = path->dentry->d_inode;
1042 struct shmem_inode_info *info = SHMEM_I(inode);
1044 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1045 spin_lock_irq(&info->lock);
1046 shmem_recalc_inode(inode);
1047 spin_unlock_irq(&info->lock);
1049 if (info->fsflags & FS_APPEND_FL)
1050 stat->attributes |= STATX_ATTR_APPEND;
1051 if (info->fsflags & FS_IMMUTABLE_FL)
1052 stat->attributes |= STATX_ATTR_IMMUTABLE;
1053 if (info->fsflags & FS_NODUMP_FL)
1054 stat->attributes |= STATX_ATTR_NODUMP;
1055 stat->attributes_mask |= (STATX_ATTR_APPEND |
1056 STATX_ATTR_IMMUTABLE |
1058 generic_fillattr(&init_user_ns, inode, stat);
1060 if (shmem_is_huge(NULL, inode, 0))
1061 stat->blksize = HPAGE_PMD_SIZE;
1063 if (request_mask & STATX_BTIME) {
1064 stat->result_mask |= STATX_BTIME;
1065 stat->btime.tv_sec = info->i_crtime.tv_sec;
1066 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1072 static int shmem_setattr(struct user_namespace *mnt_userns,
1073 struct dentry *dentry, struct iattr *attr)
1075 struct inode *inode = d_inode(dentry);
1076 struct shmem_inode_info *info = SHMEM_I(inode);
1079 error = setattr_prepare(&init_user_ns, dentry, attr);
1083 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1084 loff_t oldsize = inode->i_size;
1085 loff_t newsize = attr->ia_size;
1087 /* protected by i_rwsem */
1088 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1089 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1092 if (newsize != oldsize) {
1093 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1097 i_size_write(inode, newsize);
1098 inode->i_ctime = inode->i_mtime = current_time(inode);
1100 if (newsize <= oldsize) {
1101 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1102 if (oldsize > holebegin)
1103 unmap_mapping_range(inode->i_mapping,
1106 shmem_truncate_range(inode,
1107 newsize, (loff_t)-1);
1108 /* unmap again to remove racily COWed private pages */
1109 if (oldsize > holebegin)
1110 unmap_mapping_range(inode->i_mapping,
1115 setattr_copy(&init_user_ns, inode, attr);
1116 if (attr->ia_valid & ATTR_MODE)
1117 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1121 static void shmem_evict_inode(struct inode *inode)
1123 struct shmem_inode_info *info = SHMEM_I(inode);
1124 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1126 if (shmem_mapping(inode->i_mapping)) {
1127 shmem_unacct_size(info->flags, inode->i_size);
1129 mapping_set_exiting(inode->i_mapping);
1130 shmem_truncate_range(inode, 0, (loff_t)-1);
1131 if (!list_empty(&info->shrinklist)) {
1132 spin_lock(&sbinfo->shrinklist_lock);
1133 if (!list_empty(&info->shrinklist)) {
1134 list_del_init(&info->shrinklist);
1135 sbinfo->shrinklist_len--;
1137 spin_unlock(&sbinfo->shrinklist_lock);
1139 while (!list_empty(&info->swaplist)) {
1140 /* Wait while shmem_unuse() is scanning this inode... */
1141 wait_var_event(&info->stop_eviction,
1142 !atomic_read(&info->stop_eviction));
1143 mutex_lock(&shmem_swaplist_mutex);
1144 /* ...but beware of the race if we peeked too early */
1145 if (!atomic_read(&info->stop_eviction))
1146 list_del_init(&info->swaplist);
1147 mutex_unlock(&shmem_swaplist_mutex);
1151 simple_xattrs_free(&info->xattrs);
1152 WARN_ON(inode->i_blocks);
1153 shmem_free_inode(inode->i_sb);
1157 static int shmem_find_swap_entries(struct address_space *mapping,
1158 pgoff_t start, struct folio_batch *fbatch,
1159 pgoff_t *indices, unsigned int type)
1161 XA_STATE(xas, &mapping->i_pages, start);
1162 struct folio *folio;
1166 xas_for_each(&xas, folio, ULONG_MAX) {
1167 if (xas_retry(&xas, folio))
1170 if (!xa_is_value(folio))
1173 entry = radix_to_swp_entry(folio);
1175 * swapin error entries can be found in the mapping. But they're
1176 * deliberately ignored here as we've done everything we can do.
1178 if (swp_type(entry) != type)
1181 indices[folio_batch_count(fbatch)] = xas.xa_index;
1182 if (!folio_batch_add(fbatch, folio))
1185 if (need_resched()) {
1192 return xas.xa_index;
1196 * Move the swapped pages for an inode to page cache. Returns the count
1197 * of pages swapped in, or the error in case of failure.
1199 static int shmem_unuse_swap_entries(struct inode *inode,
1200 struct folio_batch *fbatch, pgoff_t *indices)
1205 struct address_space *mapping = inode->i_mapping;
1207 for (i = 0; i < folio_batch_count(fbatch); i++) {
1208 struct folio *folio = fbatch->folios[i];
1210 if (!xa_is_value(folio))
1212 error = shmem_swapin_folio(inode, indices[i],
1214 mapping_gfp_mask(mapping),
1217 folio_unlock(folio);
1221 if (error == -ENOMEM)
1225 return error ? error : ret;
1229 * If swap found in inode, free it and move page from swapcache to filecache.
1231 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1233 struct address_space *mapping = inode->i_mapping;
1235 struct folio_batch fbatch;
1236 pgoff_t indices[PAGEVEC_SIZE];
1240 folio_batch_init(&fbatch);
1241 shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1242 if (folio_batch_count(&fbatch) == 0) {
1247 ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1251 start = indices[folio_batch_count(&fbatch) - 1];
1258 * Read all the shared memory data that resides in the swap
1259 * device 'type' back into memory, so the swap device can be
1262 int shmem_unuse(unsigned int type)
1264 struct shmem_inode_info *info, *next;
1267 if (list_empty(&shmem_swaplist))
1270 mutex_lock(&shmem_swaplist_mutex);
1271 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1272 if (!info->swapped) {
1273 list_del_init(&info->swaplist);
1277 * Drop the swaplist mutex while searching the inode for swap;
1278 * but before doing so, make sure shmem_evict_inode() will not
1279 * remove placeholder inode from swaplist, nor let it be freed
1280 * (igrab() would protect from unlink, but not from unmount).
1282 atomic_inc(&info->stop_eviction);
1283 mutex_unlock(&shmem_swaplist_mutex);
1285 error = shmem_unuse_inode(&info->vfs_inode, type);
1288 mutex_lock(&shmem_swaplist_mutex);
1289 next = list_next_entry(info, swaplist);
1291 list_del_init(&info->swaplist);
1292 if (atomic_dec_and_test(&info->stop_eviction))
1293 wake_up_var(&info->stop_eviction);
1297 mutex_unlock(&shmem_swaplist_mutex);
1303 * Move the page from the page cache to the swap cache.
1305 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1307 struct folio *folio = page_folio(page);
1308 struct shmem_inode_info *info;
1309 struct address_space *mapping;
1310 struct inode *inode;
1315 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1316 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1317 * and its shmem_writeback() needs them to be split when swapping.
1319 if (folio_test_large(folio)) {
1320 /* Ensure the subpages are still dirty */
1321 folio_test_set_dirty(folio);
1322 if (split_huge_page(page) < 0)
1324 folio = page_folio(page);
1325 folio_clear_dirty(folio);
1328 BUG_ON(!folio_test_locked(folio));
1329 mapping = folio->mapping;
1330 index = folio->index;
1331 inode = mapping->host;
1332 info = SHMEM_I(inode);
1333 if (info->flags & VM_LOCKED)
1335 if (!total_swap_pages)
1339 * Our capabilities prevent regular writeback or sync from ever calling
1340 * shmem_writepage; but a stacking filesystem might use ->writepage of
1341 * its underlying filesystem, in which case tmpfs should write out to
1342 * swap only in response to memory pressure, and not for the writeback
1345 if (!wbc->for_reclaim) {
1346 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1351 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1352 * value into swapfile.c, the only way we can correctly account for a
1353 * fallocated folio arriving here is now to initialize it and write it.
1355 * That's okay for a folio already fallocated earlier, but if we have
1356 * not yet completed the fallocation, then (a) we want to keep track
1357 * of this folio in case we have to undo it, and (b) it may not be a
1358 * good idea to continue anyway, once we're pushing into swap. So
1359 * reactivate the folio, and let shmem_fallocate() quit when too many.
1361 if (!folio_test_uptodate(folio)) {
1362 if (inode->i_private) {
1363 struct shmem_falloc *shmem_falloc;
1364 spin_lock(&inode->i_lock);
1365 shmem_falloc = inode->i_private;
1367 !shmem_falloc->waitq &&
1368 index >= shmem_falloc->start &&
1369 index < shmem_falloc->next)
1370 shmem_falloc->nr_unswapped++;
1372 shmem_falloc = NULL;
1373 spin_unlock(&inode->i_lock);
1377 folio_zero_range(folio, 0, folio_size(folio));
1378 flush_dcache_folio(folio);
1379 folio_mark_uptodate(folio);
1382 swap = folio_alloc_swap(folio);
1387 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1388 * if it's not already there. Do it now before the folio is
1389 * moved to swap cache, when its pagelock no longer protects
1390 * the inode from eviction. But don't unlock the mutex until
1391 * we've incremented swapped, because shmem_unuse_inode() will
1392 * prune a !swapped inode from the swaplist under this mutex.
1394 mutex_lock(&shmem_swaplist_mutex);
1395 if (list_empty(&info->swaplist))
1396 list_add(&info->swaplist, &shmem_swaplist);
1398 if (add_to_swap_cache(folio, swap,
1399 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1401 spin_lock_irq(&info->lock);
1402 shmem_recalc_inode(inode);
1404 spin_unlock_irq(&info->lock);
1406 swap_shmem_alloc(swap);
1407 shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1409 mutex_unlock(&shmem_swaplist_mutex);
1410 BUG_ON(folio_mapped(folio));
1411 swap_writepage(&folio->page, wbc);
1415 mutex_unlock(&shmem_swaplist_mutex);
1416 put_swap_folio(folio, swap);
1418 folio_mark_dirty(folio);
1419 if (wbc->for_reclaim)
1420 return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */
1421 folio_unlock(folio);
1425 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1426 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1430 if (!mpol || mpol->mode == MPOL_DEFAULT)
1431 return; /* show nothing */
1433 mpol_to_str(buffer, sizeof(buffer), mpol);
1435 seq_printf(seq, ",mpol=%s", buffer);
1438 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1440 struct mempolicy *mpol = NULL;
1442 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1443 mpol = sbinfo->mpol;
1445 raw_spin_unlock(&sbinfo->stat_lock);
1449 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1450 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1453 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1457 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1459 #define vm_policy vm_private_data
1462 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1463 struct shmem_inode_info *info, pgoff_t index)
1465 /* Create a pseudo vma that just contains the policy */
1466 vma_init(vma, NULL);
1467 /* Bias interleave by inode number to distribute better across nodes */
1468 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1469 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1472 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1474 /* Drop reference taken by mpol_shared_policy_lookup() */
1475 mpol_cond_put(vma->vm_policy);
1478 static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1479 struct shmem_inode_info *info, pgoff_t index)
1481 struct vm_area_struct pvma;
1483 struct vm_fault vmf = {
1487 shmem_pseudo_vma_init(&pvma, info, index);
1488 page = swap_cluster_readahead(swap, gfp, &vmf);
1489 shmem_pseudo_vma_destroy(&pvma);
1493 return page_folio(page);
1497 * Make sure huge_gfp is always more limited than limit_gfp.
1498 * Some of the flags set permissions, while others set limitations.
1500 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1502 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1503 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1504 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1505 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1507 /* Allow allocations only from the originally specified zones. */
1508 result |= zoneflags;
1511 * Minimize the result gfp by taking the union with the deny flags,
1512 * and the intersection of the allow flags.
1514 result |= (limit_gfp & denyflags);
1515 result |= (huge_gfp & limit_gfp) & allowflags;
1520 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1521 struct shmem_inode_info *info, pgoff_t index)
1523 struct vm_area_struct pvma;
1524 struct address_space *mapping = info->vfs_inode.i_mapping;
1526 struct folio *folio;
1528 hindex = round_down(index, HPAGE_PMD_NR);
1529 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1533 shmem_pseudo_vma_init(&pvma, info, hindex);
1534 folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1535 shmem_pseudo_vma_destroy(&pvma);
1537 count_vm_event(THP_FILE_FALLBACK);
1541 static struct folio *shmem_alloc_folio(gfp_t gfp,
1542 struct shmem_inode_info *info, pgoff_t index)
1544 struct vm_area_struct pvma;
1545 struct folio *folio;
1547 shmem_pseudo_vma_init(&pvma, info, index);
1548 folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1549 shmem_pseudo_vma_destroy(&pvma);
1554 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1555 pgoff_t index, bool huge)
1557 struct shmem_inode_info *info = SHMEM_I(inode);
1558 struct folio *folio;
1562 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1564 nr = huge ? HPAGE_PMD_NR : 1;
1566 if (!shmem_inode_acct_block(inode, nr))
1570 folio = shmem_alloc_hugefolio(gfp, info, index);
1572 folio = shmem_alloc_folio(gfp, info, index);
1574 __folio_set_locked(folio);
1575 __folio_set_swapbacked(folio);
1580 shmem_inode_unacct_blocks(inode, nr);
1582 return ERR_PTR(err);
1586 * When a page is moved from swapcache to shmem filecache (either by the
1587 * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1588 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1589 * ignorance of the mapping it belongs to. If that mapping has special
1590 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1591 * we may need to copy to a suitable page before moving to filecache.
1593 * In a future release, this may well be extended to respect cpuset and
1594 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1595 * but for now it is a simple matter of zone.
1597 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1599 return folio_zonenum(folio) > gfp_zone(gfp);
1602 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1603 struct shmem_inode_info *info, pgoff_t index)
1605 struct folio *old, *new;
1606 struct address_space *swap_mapping;
1612 entry = folio_swap_entry(old);
1613 swap_index = swp_offset(entry);
1614 swap_mapping = swap_address_space(entry);
1617 * We have arrived here because our zones are constrained, so don't
1618 * limit chance of success by further cpuset and node constraints.
1620 gfp &= ~GFP_CONSTRAINT_MASK;
1621 VM_BUG_ON_FOLIO(folio_test_large(old), old);
1622 new = shmem_alloc_folio(gfp, info, index);
1627 folio_copy(new, old);
1628 flush_dcache_folio(new);
1630 __folio_set_locked(new);
1631 __folio_set_swapbacked(new);
1632 folio_mark_uptodate(new);
1633 folio_set_swap_entry(new, entry);
1634 folio_set_swapcache(new);
1637 * Our caller will very soon move newpage out of swapcache, but it's
1638 * a nice clean interface for us to replace oldpage by newpage there.
1640 xa_lock_irq(&swap_mapping->i_pages);
1641 error = shmem_replace_entry(swap_mapping, swap_index, old, new);
1643 mem_cgroup_migrate(old, new);
1644 __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1645 __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1646 __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1647 __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
1649 xa_unlock_irq(&swap_mapping->i_pages);
1651 if (unlikely(error)) {
1653 * Is this possible? I think not, now that our callers check
1654 * both PageSwapCache and page_private after getting page lock;
1655 * but be defensive. Reverse old to newpage for clear and free.
1663 folio_clear_swapcache(old);
1664 old->private = NULL;
1667 folio_put_refs(old, 2);
1671 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1672 struct folio *folio, swp_entry_t swap)
1674 struct address_space *mapping = inode->i_mapping;
1675 struct shmem_inode_info *info = SHMEM_I(inode);
1676 swp_entry_t swapin_error;
1679 swapin_error = make_swapin_error_entry(&folio->page);
1680 old = xa_cmpxchg_irq(&mapping->i_pages, index,
1681 swp_to_radix_entry(swap),
1682 swp_to_radix_entry(swapin_error), 0);
1683 if (old != swp_to_radix_entry(swap))
1686 folio_wait_writeback(folio);
1687 delete_from_swap_cache(folio);
1688 spin_lock_irq(&info->lock);
1690 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks won't
1691 * be 0 when inode is released and thus trigger WARN_ON(inode->i_blocks) in
1692 * shmem_evict_inode.
1696 shmem_recalc_inode(inode);
1697 spin_unlock_irq(&info->lock);
1702 * Swap in the folio pointed to by *foliop.
1703 * Caller has to make sure that *foliop contains a valid swapped folio.
1704 * Returns 0 and the folio in foliop if success. On failure, returns the
1705 * error code and NULL in *foliop.
1707 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1708 struct folio **foliop, enum sgp_type sgp,
1709 gfp_t gfp, struct vm_area_struct *vma,
1710 vm_fault_t *fault_type)
1712 struct address_space *mapping = inode->i_mapping;
1713 struct shmem_inode_info *info = SHMEM_I(inode);
1714 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1715 struct folio *folio = NULL;
1719 VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1720 swap = radix_to_swp_entry(*foliop);
1723 if (is_swapin_error_entry(swap))
1726 /* Look it up and read it in.. */
1727 folio = swap_cache_get_folio(swap, NULL, 0);
1729 /* Or update major stats only when swapin succeeds?? */
1731 *fault_type |= VM_FAULT_MAJOR;
1732 count_vm_event(PGMAJFAULT);
1733 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1735 /* Here we actually start the io */
1736 folio = shmem_swapin(swap, gfp, info, index);
1743 /* We have to do this with folio locked to prevent races */
1745 if (!folio_test_swapcache(folio) ||
1746 folio_swap_entry(folio).val != swap.val ||
1747 !shmem_confirm_swap(mapping, index, swap)) {
1751 if (!folio_test_uptodate(folio)) {
1755 folio_wait_writeback(folio);
1758 * Some architectures may have to restore extra metadata to the
1759 * folio after reading from swap.
1761 arch_swap_restore(swap, folio);
1763 if (shmem_should_replace_folio(folio, gfp)) {
1764 error = shmem_replace_folio(&folio, gfp, info, index);
1769 error = shmem_add_to_page_cache(folio, mapping, index,
1770 swp_to_radix_entry(swap), gfp,
1775 spin_lock_irq(&info->lock);
1777 shmem_recalc_inode(inode);
1778 spin_unlock_irq(&info->lock);
1780 if (sgp == SGP_WRITE)
1781 folio_mark_accessed(folio);
1783 delete_from_swap_cache(folio);
1784 folio_mark_dirty(folio);
1790 if (!shmem_confirm_swap(mapping, index, swap))
1793 shmem_set_folio_swapin_error(inode, index, folio, swap);
1796 folio_unlock(folio);
1804 * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1806 * If we allocate a new one we do not mark it dirty. That's up to the
1807 * vm. If we swap it in we mark it dirty since we also free the swap
1808 * entry since a page cannot live in both the swap and page cache.
1810 * vma, vmf, and fault_type are only supplied by shmem_fault:
1811 * otherwise they are NULL.
1813 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1814 struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
1815 struct vm_area_struct *vma, struct vm_fault *vmf,
1816 vm_fault_t *fault_type)
1818 struct address_space *mapping = inode->i_mapping;
1819 struct shmem_inode_info *info = SHMEM_I(inode);
1820 struct shmem_sb_info *sbinfo;
1821 struct mm_struct *charge_mm;
1822 struct folio *folio;
1823 pgoff_t hindex = index;
1829 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1832 if (sgp <= SGP_CACHE &&
1833 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1837 sbinfo = SHMEM_SB(inode->i_sb);
1838 charge_mm = vma ? vma->vm_mm : NULL;
1840 folio = __filemap_get_folio(mapping, index, FGP_ENTRY | FGP_LOCK, 0);
1841 if (folio && vma && userfaultfd_minor(vma)) {
1842 if (!xa_is_value(folio)) {
1843 folio_unlock(folio);
1846 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1850 if (xa_is_value(folio)) {
1851 error = shmem_swapin_folio(inode, index, &folio,
1852 sgp, gfp, vma, fault_type);
1853 if (error == -EEXIST)
1861 hindex = folio->index;
1862 if (sgp == SGP_WRITE)
1863 folio_mark_accessed(folio);
1864 if (folio_test_uptodate(folio))
1866 /* fallocated folio */
1867 if (sgp != SGP_READ)
1869 folio_unlock(folio);
1874 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1875 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1878 if (sgp == SGP_READ)
1880 if (sgp == SGP_NOALLOC)
1884 * Fast cache lookup and swap lookup did not find it: allocate.
1887 if (vma && userfaultfd_missing(vma)) {
1888 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1892 if (!shmem_is_huge(vma, inode, index))
1895 huge_gfp = vma_thp_gfp_mask(vma);
1896 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1897 folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
1898 if (IS_ERR(folio)) {
1900 folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
1902 if (IS_ERR(folio)) {
1905 error = PTR_ERR(folio);
1907 if (error != -ENOSPC)
1910 * Try to reclaim some space by splitting a large folio
1911 * beyond i_size on the filesystem.
1916 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1917 if (ret == SHRINK_STOP)
1925 hindex = round_down(index, folio_nr_pages(folio));
1927 if (sgp == SGP_WRITE)
1928 __folio_set_referenced(folio);
1930 error = shmem_add_to_page_cache(folio, mapping, hindex,
1931 NULL, gfp & GFP_RECLAIM_MASK,
1935 folio_add_lru(folio);
1937 spin_lock_irq(&info->lock);
1938 info->alloced += folio_nr_pages(folio);
1939 inode->i_blocks += (blkcnt_t)BLOCKS_PER_PAGE << folio_order(folio);
1940 shmem_recalc_inode(inode);
1941 spin_unlock_irq(&info->lock);
1944 if (folio_test_pmd_mappable(folio) &&
1945 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1946 folio_next_index(folio) - 1) {
1948 * Part of the large folio is beyond i_size: subject
1949 * to shrink under memory pressure.
1951 spin_lock(&sbinfo->shrinklist_lock);
1953 * _careful to defend against unlocked access to
1954 * ->shrink_list in shmem_unused_huge_shrink()
1956 if (list_empty_careful(&info->shrinklist)) {
1957 list_add_tail(&info->shrinklist,
1958 &sbinfo->shrinklist);
1959 sbinfo->shrinklist_len++;
1961 spin_unlock(&sbinfo->shrinklist_lock);
1965 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
1967 if (sgp == SGP_FALLOC)
1971 * Let SGP_WRITE caller clear ends if write does not fill folio;
1972 * but SGP_FALLOC on a folio fallocated earlier must initialize
1973 * it now, lest undo on failure cancel our earlier guarantee.
1975 if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
1976 long i, n = folio_nr_pages(folio);
1978 for (i = 0; i < n; i++)
1979 clear_highpage(folio_page(folio, i));
1980 flush_dcache_folio(folio);
1981 folio_mark_uptodate(folio);
1984 /* Perhaps the file has been truncated since we checked */
1985 if (sgp <= SGP_CACHE &&
1986 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1988 folio_clear_dirty(folio);
1989 filemap_remove_folio(folio);
1990 spin_lock_irq(&info->lock);
1991 shmem_recalc_inode(inode);
1992 spin_unlock_irq(&info->lock);
2005 shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
2007 if (folio_test_large(folio)) {
2008 folio_unlock(folio);
2014 folio_unlock(folio);
2017 if (error == -ENOSPC && !once++) {
2018 spin_lock_irq(&info->lock);
2019 shmem_recalc_inode(inode);
2020 spin_unlock_irq(&info->lock);
2023 if (error == -EEXIST)
2028 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
2029 struct page **pagep, enum sgp_type sgp,
2030 gfp_t gfp, struct vm_area_struct *vma,
2031 struct vm_fault *vmf, vm_fault_t *fault_type)
2033 struct folio *folio = NULL;
2034 int ret = shmem_get_folio_gfp(inode, index, &folio, sgp, gfp, vma,
2038 *pagep = folio_file_page(folio, index);
2044 int shmem_getpage(struct inode *inode, pgoff_t index,
2045 struct page **pagep, enum sgp_type sgp)
2047 return shmem_getpage_gfp(inode, index, pagep, sgp,
2048 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
2052 * This is like autoremove_wake_function, but it removes the wait queue
2053 * entry unconditionally - even if something else had already woken the
2056 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2058 int ret = default_wake_function(wait, mode, sync, key);
2059 list_del_init(&wait->entry);
2063 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2065 struct vm_area_struct *vma = vmf->vma;
2066 struct inode *inode = file_inode(vma->vm_file);
2067 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2068 struct folio *folio = NULL;
2070 vm_fault_t ret = VM_FAULT_LOCKED;
2073 * Trinity finds that probing a hole which tmpfs is punching can
2074 * prevent the hole-punch from ever completing: which in turn
2075 * locks writers out with its hold on i_rwsem. So refrain from
2076 * faulting pages into the hole while it's being punched. Although
2077 * shmem_undo_range() does remove the additions, it may be unable to
2078 * keep up, as each new page needs its own unmap_mapping_range() call,
2079 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2081 * It does not matter if we sometimes reach this check just before the
2082 * hole-punch begins, so that one fault then races with the punch:
2083 * we just need to make racing faults a rare case.
2085 * The implementation below would be much simpler if we just used a
2086 * standard mutex or completion: but we cannot take i_rwsem in fault,
2087 * and bloating every shmem inode for this unlikely case would be sad.
2089 if (unlikely(inode->i_private)) {
2090 struct shmem_falloc *shmem_falloc;
2092 spin_lock(&inode->i_lock);
2093 shmem_falloc = inode->i_private;
2095 shmem_falloc->waitq &&
2096 vmf->pgoff >= shmem_falloc->start &&
2097 vmf->pgoff < shmem_falloc->next) {
2099 wait_queue_head_t *shmem_falloc_waitq;
2100 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2102 ret = VM_FAULT_NOPAGE;
2103 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2105 ret = VM_FAULT_RETRY;
2107 shmem_falloc_waitq = shmem_falloc->waitq;
2108 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2109 TASK_UNINTERRUPTIBLE);
2110 spin_unlock(&inode->i_lock);
2114 * shmem_falloc_waitq points into the shmem_fallocate()
2115 * stack of the hole-punching task: shmem_falloc_waitq
2116 * is usually invalid by the time we reach here, but
2117 * finish_wait() does not dereference it in that case;
2118 * though i_lock needed lest racing with wake_up_all().
2120 spin_lock(&inode->i_lock);
2121 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2122 spin_unlock(&inode->i_lock);
2128 spin_unlock(&inode->i_lock);
2131 err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
2132 gfp, vma, vmf, &ret);
2134 return vmf_error(err);
2136 vmf->page = folio_file_page(folio, vmf->pgoff);
2140 unsigned long shmem_get_unmapped_area(struct file *file,
2141 unsigned long uaddr, unsigned long len,
2142 unsigned long pgoff, unsigned long flags)
2144 unsigned long (*get_area)(struct file *,
2145 unsigned long, unsigned long, unsigned long, unsigned long);
2147 unsigned long offset;
2148 unsigned long inflated_len;
2149 unsigned long inflated_addr;
2150 unsigned long inflated_offset;
2152 if (len > TASK_SIZE)
2155 get_area = current->mm->get_unmapped_area;
2156 addr = get_area(file, uaddr, len, pgoff, flags);
2158 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2160 if (IS_ERR_VALUE(addr))
2162 if (addr & ~PAGE_MASK)
2164 if (addr > TASK_SIZE - len)
2167 if (shmem_huge == SHMEM_HUGE_DENY)
2169 if (len < HPAGE_PMD_SIZE)
2171 if (flags & MAP_FIXED)
2174 * Our priority is to support MAP_SHARED mapped hugely;
2175 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2176 * But if caller specified an address hint and we allocated area there
2177 * successfully, respect that as before.
2182 if (shmem_huge != SHMEM_HUGE_FORCE) {
2183 struct super_block *sb;
2186 VM_BUG_ON(file->f_op != &shmem_file_operations);
2187 sb = file_inode(file)->i_sb;
2190 * Called directly from mm/mmap.c, or drivers/char/mem.c
2191 * for "/dev/zero", to create a shared anonymous object.
2193 if (IS_ERR(shm_mnt))
2195 sb = shm_mnt->mnt_sb;
2197 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2201 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2202 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2204 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2207 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2208 if (inflated_len > TASK_SIZE)
2210 if (inflated_len < len)
2213 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2214 if (IS_ERR_VALUE(inflated_addr))
2216 if (inflated_addr & ~PAGE_MASK)
2219 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2220 inflated_addr += offset - inflated_offset;
2221 if (inflated_offset > offset)
2222 inflated_addr += HPAGE_PMD_SIZE;
2224 if (inflated_addr > TASK_SIZE - len)
2226 return inflated_addr;
2230 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2232 struct inode *inode = file_inode(vma->vm_file);
2233 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2236 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2239 struct inode *inode = file_inode(vma->vm_file);
2242 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2243 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2247 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2249 struct inode *inode = file_inode(file);
2250 struct shmem_inode_info *info = SHMEM_I(inode);
2251 int retval = -ENOMEM;
2254 * What serializes the accesses to info->flags?
2255 * ipc_lock_object() when called from shmctl_do_lock(),
2256 * no serialization needed when called from shm_destroy().
2258 if (lock && !(info->flags & VM_LOCKED)) {
2259 if (!user_shm_lock(inode->i_size, ucounts))
2261 info->flags |= VM_LOCKED;
2262 mapping_set_unevictable(file->f_mapping);
2264 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2265 user_shm_unlock(inode->i_size, ucounts);
2266 info->flags &= ~VM_LOCKED;
2267 mapping_clear_unevictable(file->f_mapping);
2275 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2277 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2280 ret = seal_check_future_write(info->seals, vma);
2284 /* arm64 - allow memory tagging on RAM-based files */
2285 vma->vm_flags |= VM_MTE_ALLOWED;
2287 file_accessed(file);
2288 vma->vm_ops = &shmem_vm_ops;
2292 #ifdef CONFIG_TMPFS_XATTR
2293 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2296 * chattr's fsflags are unrelated to extended attributes,
2297 * but tmpfs has chosen to enable them under the same config option.
2299 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2301 unsigned int i_flags = 0;
2303 if (fsflags & FS_NOATIME_FL)
2304 i_flags |= S_NOATIME;
2305 if (fsflags & FS_APPEND_FL)
2306 i_flags |= S_APPEND;
2307 if (fsflags & FS_IMMUTABLE_FL)
2308 i_flags |= S_IMMUTABLE;
2310 * But FS_NODUMP_FL does not require any action in i_flags.
2312 inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2315 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2318 #define shmem_initxattrs NULL
2321 static struct inode *shmem_get_inode(struct super_block *sb, struct inode *dir,
2322 umode_t mode, dev_t dev, unsigned long flags)
2324 struct inode *inode;
2325 struct shmem_inode_info *info;
2326 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2329 if (shmem_reserve_inode(sb, &ino))
2332 inode = new_inode(sb);
2335 inode_init_owner(&init_user_ns, inode, dir, mode);
2336 inode->i_blocks = 0;
2337 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2338 inode->i_generation = prandom_u32();
2339 info = SHMEM_I(inode);
2340 memset(info, 0, (char *)inode - (char *)info);
2341 spin_lock_init(&info->lock);
2342 atomic_set(&info->stop_eviction, 0);
2343 info->seals = F_SEAL_SEAL;
2344 info->flags = flags & VM_NORESERVE;
2345 info->i_crtime = inode->i_mtime;
2346 info->fsflags = (dir == NULL) ? 0 :
2347 SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2349 shmem_set_inode_flags(inode, info->fsflags);
2350 INIT_LIST_HEAD(&info->shrinklist);
2351 INIT_LIST_HEAD(&info->swaplist);
2352 simple_xattrs_init(&info->xattrs);
2353 cache_no_acl(inode);
2354 mapping_set_large_folios(inode->i_mapping);
2356 switch (mode & S_IFMT) {
2358 inode->i_op = &shmem_special_inode_operations;
2359 init_special_inode(inode, mode, dev);
2362 inode->i_mapping->a_ops = &shmem_aops;
2363 inode->i_op = &shmem_inode_operations;
2364 inode->i_fop = &shmem_file_operations;
2365 mpol_shared_policy_init(&info->policy,
2366 shmem_get_sbmpol(sbinfo));
2370 /* Some things misbehave if size == 0 on a directory */
2371 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2372 inode->i_op = &shmem_dir_inode_operations;
2373 inode->i_fop = &simple_dir_operations;
2377 * Must not load anything in the rbtree,
2378 * mpol_free_shared_policy will not be called.
2380 mpol_shared_policy_init(&info->policy, NULL);
2384 lockdep_annotate_inode_mutex_key(inode);
2386 shmem_free_inode(sb);
2390 #ifdef CONFIG_USERFAULTFD
2391 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2393 struct vm_area_struct *dst_vma,
2394 unsigned long dst_addr,
2395 unsigned long src_addr,
2396 bool zeropage, bool wp_copy,
2397 struct page **pagep)
2399 struct inode *inode = file_inode(dst_vma->vm_file);
2400 struct shmem_inode_info *info = SHMEM_I(inode);
2401 struct address_space *mapping = inode->i_mapping;
2402 gfp_t gfp = mapping_gfp_mask(mapping);
2403 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2405 struct folio *folio;
2409 if (!shmem_inode_acct_block(inode, 1)) {
2411 * We may have got a page, returned -ENOENT triggering a retry,
2412 * and now we find ourselves with -ENOMEM. Release the page, to
2413 * avoid a BUG_ON in our caller.
2415 if (unlikely(*pagep)) {
2424 folio = shmem_alloc_folio(gfp, info, pgoff);
2426 goto out_unacct_blocks;
2428 if (!zeropage) { /* COPY */
2429 page_kaddr = kmap_local_folio(folio, 0);
2430 ret = copy_from_user(page_kaddr,
2431 (const void __user *)src_addr,
2433 kunmap_local(page_kaddr);
2435 /* fallback to copy_from_user outside mmap_lock */
2436 if (unlikely(ret)) {
2437 *pagep = &folio->page;
2439 /* don't free the page */
2440 goto out_unacct_blocks;
2443 flush_dcache_folio(folio);
2444 } else { /* ZEROPAGE */
2445 clear_user_highpage(&folio->page, dst_addr);
2448 folio = page_folio(*pagep);
2449 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2453 VM_BUG_ON(folio_test_locked(folio));
2454 VM_BUG_ON(folio_test_swapbacked(folio));
2455 __folio_set_locked(folio);
2456 __folio_set_swapbacked(folio);
2457 __folio_mark_uptodate(folio);
2460 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2461 if (unlikely(pgoff >= max_off))
2464 ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2465 gfp & GFP_RECLAIM_MASK, dst_mm);
2469 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2470 &folio->page, true, wp_copy);
2472 goto out_delete_from_cache;
2474 spin_lock_irq(&info->lock);
2476 inode->i_blocks += BLOCKS_PER_PAGE;
2477 shmem_recalc_inode(inode);
2478 spin_unlock_irq(&info->lock);
2480 folio_unlock(folio);
2482 out_delete_from_cache:
2483 filemap_remove_folio(folio);
2485 folio_unlock(folio);
2488 shmem_inode_unacct_blocks(inode, 1);
2491 #endif /* CONFIG_USERFAULTFD */
2494 static const struct inode_operations shmem_symlink_inode_operations;
2495 static const struct inode_operations shmem_short_symlink_operations;
2498 shmem_write_begin(struct file *file, struct address_space *mapping,
2499 loff_t pos, unsigned len,
2500 struct page **pagep, void **fsdata)
2502 struct inode *inode = mapping->host;
2503 struct shmem_inode_info *info = SHMEM_I(inode);
2504 pgoff_t index = pos >> PAGE_SHIFT;
2507 /* i_rwsem is held by caller */
2508 if (unlikely(info->seals & (F_SEAL_GROW |
2509 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2510 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2512 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2516 ret = shmem_getpage(inode, index, pagep, SGP_WRITE);
2521 if (PageHWPoison(*pagep)) {
2522 unlock_page(*pagep);
2532 shmem_write_end(struct file *file, struct address_space *mapping,
2533 loff_t pos, unsigned len, unsigned copied,
2534 struct page *page, void *fsdata)
2536 struct inode *inode = mapping->host;
2538 if (pos + copied > inode->i_size)
2539 i_size_write(inode, pos + copied);
2541 if (!PageUptodate(page)) {
2542 struct page *head = compound_head(page);
2543 if (PageTransCompound(page)) {
2546 for (i = 0; i < HPAGE_PMD_NR; i++) {
2547 if (head + i == page)
2549 clear_highpage(head + i);
2550 flush_dcache_page(head + i);
2553 if (copied < PAGE_SIZE) {
2554 unsigned from = pos & (PAGE_SIZE - 1);
2555 zero_user_segments(page, 0, from,
2556 from + copied, PAGE_SIZE);
2558 SetPageUptodate(head);
2560 set_page_dirty(page);
2567 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2569 struct file *file = iocb->ki_filp;
2570 struct inode *inode = file_inode(file);
2571 struct address_space *mapping = inode->i_mapping;
2573 unsigned long offset;
2576 loff_t *ppos = &iocb->ki_pos;
2578 index = *ppos >> PAGE_SHIFT;
2579 offset = *ppos & ~PAGE_MASK;
2582 struct page *page = NULL;
2584 unsigned long nr, ret;
2585 loff_t i_size = i_size_read(inode);
2587 end_index = i_size >> PAGE_SHIFT;
2588 if (index > end_index)
2590 if (index == end_index) {
2591 nr = i_size & ~PAGE_MASK;
2596 error = shmem_getpage(inode, index, &page, SGP_READ);
2598 if (error == -EINVAL)
2605 if (PageHWPoison(page)) {
2613 * We must evaluate after, since reads (unlike writes)
2614 * are called without i_rwsem protection against truncate
2617 i_size = i_size_read(inode);
2618 end_index = i_size >> PAGE_SHIFT;
2619 if (index == end_index) {
2620 nr = i_size & ~PAGE_MASK;
2631 * If users can be writing to this page using arbitrary
2632 * virtual addresses, take care about potential aliasing
2633 * before reading the page on the kernel side.
2635 if (mapping_writably_mapped(mapping))
2636 flush_dcache_page(page);
2638 * Mark the page accessed if we read the beginning.
2641 mark_page_accessed(page);
2643 * Ok, we have the page, and it's up-to-date, so
2644 * now we can copy it to user space...
2646 ret = copy_page_to_iter(page, offset, nr, to);
2649 } else if (user_backed_iter(to)) {
2651 * Copy to user tends to be so well optimized, but
2652 * clear_user() not so much, that it is noticeably
2653 * faster to copy the zero page instead of clearing.
2655 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2658 * But submitting the same page twice in a row to
2659 * splice() - or others? - can result in confusion:
2660 * so don't attempt that optimization on pipes etc.
2662 ret = iov_iter_zero(nr, to);
2667 index += offset >> PAGE_SHIFT;
2668 offset &= ~PAGE_MASK;
2670 if (!iov_iter_count(to))
2679 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2680 file_accessed(file);
2681 return retval ? retval : error;
2684 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2686 struct address_space *mapping = file->f_mapping;
2687 struct inode *inode = mapping->host;
2689 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2690 return generic_file_llseek_size(file, offset, whence,
2691 MAX_LFS_FILESIZE, i_size_read(inode));
2696 /* We're holding i_rwsem so we can access i_size directly */
2697 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2699 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2700 inode_unlock(inode);
2704 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2707 struct inode *inode = file_inode(file);
2708 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2709 struct shmem_inode_info *info = SHMEM_I(inode);
2710 struct shmem_falloc shmem_falloc;
2711 pgoff_t start, index, end, undo_fallocend;
2714 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2719 if (mode & FALLOC_FL_PUNCH_HOLE) {
2720 struct address_space *mapping = file->f_mapping;
2721 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2722 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2723 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2725 /* protected by i_rwsem */
2726 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2731 shmem_falloc.waitq = &shmem_falloc_waitq;
2732 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2733 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2734 spin_lock(&inode->i_lock);
2735 inode->i_private = &shmem_falloc;
2736 spin_unlock(&inode->i_lock);
2738 if ((u64)unmap_end > (u64)unmap_start)
2739 unmap_mapping_range(mapping, unmap_start,
2740 1 + unmap_end - unmap_start, 0);
2741 shmem_truncate_range(inode, offset, offset + len - 1);
2742 /* No need to unmap again: hole-punching leaves COWed pages */
2744 spin_lock(&inode->i_lock);
2745 inode->i_private = NULL;
2746 wake_up_all(&shmem_falloc_waitq);
2747 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2748 spin_unlock(&inode->i_lock);
2753 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2754 error = inode_newsize_ok(inode, offset + len);
2758 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2763 start = offset >> PAGE_SHIFT;
2764 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2765 /* Try to avoid a swapstorm if len is impossible to satisfy */
2766 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2771 shmem_falloc.waitq = NULL;
2772 shmem_falloc.start = start;
2773 shmem_falloc.next = start;
2774 shmem_falloc.nr_falloced = 0;
2775 shmem_falloc.nr_unswapped = 0;
2776 spin_lock(&inode->i_lock);
2777 inode->i_private = &shmem_falloc;
2778 spin_unlock(&inode->i_lock);
2781 * info->fallocend is only relevant when huge pages might be
2782 * involved: to prevent split_huge_page() freeing fallocated
2783 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2785 undo_fallocend = info->fallocend;
2786 if (info->fallocend < end)
2787 info->fallocend = end;
2789 for (index = start; index < end; ) {
2793 * Good, the fallocate(2) manpage permits EINTR: we may have
2794 * been interrupted because we are using up too much memory.
2796 if (signal_pending(current))
2798 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2801 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2803 info->fallocend = undo_fallocend;
2804 /* Remove the !PageUptodate pages we added */
2805 if (index > start) {
2806 shmem_undo_range(inode,
2807 (loff_t)start << PAGE_SHIFT,
2808 ((loff_t)index << PAGE_SHIFT) - 1, true);
2815 * Here is a more important optimization than it appears:
2816 * a second SGP_FALLOC on the same huge page will clear it,
2817 * making it PageUptodate and un-undoable if we fail later.
2819 if (PageTransCompound(page)) {
2820 index = round_up(index, HPAGE_PMD_NR);
2821 /* Beware 32-bit wraparound */
2827 * Inform shmem_writepage() how far we have reached.
2828 * No need for lock or barrier: we have the page lock.
2830 if (!PageUptodate(page))
2831 shmem_falloc.nr_falloced += index - shmem_falloc.next;
2832 shmem_falloc.next = index;
2835 * If !PageUptodate, leave it that way so that freeable pages
2836 * can be recognized if we need to rollback on error later.
2837 * But set_page_dirty so that memory pressure will swap rather
2838 * than free the pages we are allocating (and SGP_CACHE pages
2839 * might still be clean: we now need to mark those dirty too).
2841 set_page_dirty(page);
2847 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2848 i_size_write(inode, offset + len);
2850 spin_lock(&inode->i_lock);
2851 inode->i_private = NULL;
2852 spin_unlock(&inode->i_lock);
2855 file_modified(file);
2856 inode_unlock(inode);
2860 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2862 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2864 buf->f_type = TMPFS_MAGIC;
2865 buf->f_bsize = PAGE_SIZE;
2866 buf->f_namelen = NAME_MAX;
2867 if (sbinfo->max_blocks) {
2868 buf->f_blocks = sbinfo->max_blocks;
2870 buf->f_bfree = sbinfo->max_blocks -
2871 percpu_counter_sum(&sbinfo->used_blocks);
2873 if (sbinfo->max_inodes) {
2874 buf->f_files = sbinfo->max_inodes;
2875 buf->f_ffree = sbinfo->free_inodes;
2877 /* else leave those fields 0 like simple_statfs */
2879 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2885 * File creation. Allocate an inode, and we're done..
2888 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2889 struct dentry *dentry, umode_t mode, dev_t dev)
2891 struct inode *inode;
2892 int error = -ENOSPC;
2894 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2896 error = simple_acl_create(dir, inode);
2899 error = security_inode_init_security(inode, dir,
2901 shmem_initxattrs, NULL);
2902 if (error && error != -EOPNOTSUPP)
2906 dir->i_size += BOGO_DIRENT_SIZE;
2907 dir->i_ctime = dir->i_mtime = current_time(dir);
2908 d_instantiate(dentry, inode);
2909 dget(dentry); /* Extra count - pin the dentry in core */
2918 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2919 struct dentry *dentry, umode_t mode)
2921 struct inode *inode;
2922 int error = -ENOSPC;
2924 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2926 error = security_inode_init_security(inode, dir,
2928 shmem_initxattrs, NULL);
2929 if (error && error != -EOPNOTSUPP)
2931 error = simple_acl_create(dir, inode);
2934 d_tmpfile(dentry, inode);
2942 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2943 struct dentry *dentry, umode_t mode)
2947 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2948 mode | S_IFDIR, 0)))
2954 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2955 struct dentry *dentry, umode_t mode, bool excl)
2957 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2963 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2965 struct inode *inode = d_inode(old_dentry);
2969 * No ordinary (disk based) filesystem counts links as inodes;
2970 * but each new link needs a new dentry, pinning lowmem, and
2971 * tmpfs dentries cannot be pruned until they are unlinked.
2972 * But if an O_TMPFILE file is linked into the tmpfs, the
2973 * first link must skip that, to get the accounting right.
2975 if (inode->i_nlink) {
2976 ret = shmem_reserve_inode(inode->i_sb, NULL);
2981 dir->i_size += BOGO_DIRENT_SIZE;
2982 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2984 ihold(inode); /* New dentry reference */
2985 dget(dentry); /* Extra pinning count for the created dentry */
2986 d_instantiate(dentry, inode);
2991 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2993 struct inode *inode = d_inode(dentry);
2995 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2996 shmem_free_inode(inode->i_sb);
2998 dir->i_size -= BOGO_DIRENT_SIZE;
2999 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3001 dput(dentry); /* Undo the count from "create" - this does all the work */
3005 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3007 if (!simple_empty(dentry))
3010 drop_nlink(d_inode(dentry));
3012 return shmem_unlink(dir, dentry);
3015 static int shmem_whiteout(struct user_namespace *mnt_userns,
3016 struct inode *old_dir, struct dentry *old_dentry)
3018 struct dentry *whiteout;
3021 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3025 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
3026 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3032 * Cheat and hash the whiteout while the old dentry is still in
3033 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3035 * d_lookup() will consistently find one of them at this point,
3036 * not sure which one, but that isn't even important.
3043 * The VFS layer already does all the dentry stuff for rename,
3044 * we just have to decrement the usage count for the target if
3045 * it exists so that the VFS layer correctly free's it when it
3048 static int shmem_rename2(struct user_namespace *mnt_userns,
3049 struct inode *old_dir, struct dentry *old_dentry,
3050 struct inode *new_dir, struct dentry *new_dentry,
3053 struct inode *inode = d_inode(old_dentry);
3054 int they_are_dirs = S_ISDIR(inode->i_mode);
3056 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3059 if (flags & RENAME_EXCHANGE)
3060 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
3062 if (!simple_empty(new_dentry))
3065 if (flags & RENAME_WHITEOUT) {
3068 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3073 if (d_really_is_positive(new_dentry)) {
3074 (void) shmem_unlink(new_dir, new_dentry);
3075 if (they_are_dirs) {
3076 drop_nlink(d_inode(new_dentry));
3077 drop_nlink(old_dir);
3079 } else if (they_are_dirs) {
3080 drop_nlink(old_dir);
3084 old_dir->i_size -= BOGO_DIRENT_SIZE;
3085 new_dir->i_size += BOGO_DIRENT_SIZE;
3086 old_dir->i_ctime = old_dir->i_mtime =
3087 new_dir->i_ctime = new_dir->i_mtime =
3088 inode->i_ctime = current_time(old_dir);
3092 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3093 struct dentry *dentry, const char *symname)
3097 struct inode *inode;
3100 len = strlen(symname) + 1;
3101 if (len > PAGE_SIZE)
3102 return -ENAMETOOLONG;
3104 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3109 error = security_inode_init_security(inode, dir, &dentry->d_name,
3110 shmem_initxattrs, NULL);
3111 if (error && error != -EOPNOTSUPP) {
3116 inode->i_size = len-1;
3117 if (len <= SHORT_SYMLINK_LEN) {
3118 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3119 if (!inode->i_link) {
3123 inode->i_op = &shmem_short_symlink_operations;
3125 inode_nohighmem(inode);
3126 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3131 inode->i_mapping->a_ops = &shmem_aops;
3132 inode->i_op = &shmem_symlink_inode_operations;
3133 memcpy(page_address(page), symname, len);
3134 SetPageUptodate(page);
3135 set_page_dirty(page);
3139 dir->i_size += BOGO_DIRENT_SIZE;
3140 dir->i_ctime = dir->i_mtime = current_time(dir);
3141 d_instantiate(dentry, inode);
3146 static void shmem_put_link(void *arg)
3148 mark_page_accessed(arg);
3152 static const char *shmem_get_link(struct dentry *dentry,
3153 struct inode *inode,
3154 struct delayed_call *done)
3156 struct page *page = NULL;
3159 page = find_get_page(inode->i_mapping, 0);
3161 return ERR_PTR(-ECHILD);
3162 if (PageHWPoison(page) ||
3163 !PageUptodate(page)) {
3165 return ERR_PTR(-ECHILD);
3168 error = shmem_getpage(inode, 0, &page, SGP_READ);
3170 return ERR_PTR(error);
3172 return ERR_PTR(-ECHILD);
3173 if (PageHWPoison(page)) {
3176 return ERR_PTR(-ECHILD);
3180 set_delayed_call(done, shmem_put_link, page);
3181 return page_address(page);
3184 #ifdef CONFIG_TMPFS_XATTR
3186 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3188 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3190 fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3195 static int shmem_fileattr_set(struct user_namespace *mnt_userns,
3196 struct dentry *dentry, struct fileattr *fa)
3198 struct inode *inode = d_inode(dentry);
3199 struct shmem_inode_info *info = SHMEM_I(inode);
3201 if (fileattr_has_fsx(fa))
3203 if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3206 info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3207 (fa->flags & SHMEM_FL_USER_MODIFIABLE);
3209 shmem_set_inode_flags(inode, info->fsflags);
3210 inode->i_ctime = current_time(inode);
3215 * Superblocks without xattr inode operations may get some security.* xattr
3216 * support from the LSM "for free". As soon as we have any other xattrs
3217 * like ACLs, we also need to implement the security.* handlers at
3218 * filesystem level, though.
3222 * Callback for security_inode_init_security() for acquiring xattrs.
3224 static int shmem_initxattrs(struct inode *inode,
3225 const struct xattr *xattr_array,
3228 struct shmem_inode_info *info = SHMEM_I(inode);
3229 const struct xattr *xattr;
3230 struct simple_xattr *new_xattr;
3233 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3234 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3238 len = strlen(xattr->name) + 1;
3239 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3241 if (!new_xattr->name) {
3246 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3247 XATTR_SECURITY_PREFIX_LEN);
3248 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3251 simple_xattr_list_add(&info->xattrs, new_xattr);
3257 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3258 struct dentry *unused, struct inode *inode,
3259 const char *name, void *buffer, size_t size)
3261 struct shmem_inode_info *info = SHMEM_I(inode);
3263 name = xattr_full_name(handler, name);
3264 return simple_xattr_get(&info->xattrs, name, buffer, size);
3267 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3268 struct user_namespace *mnt_userns,
3269 struct dentry *unused, struct inode *inode,
3270 const char *name, const void *value,
3271 size_t size, int flags)
3273 struct shmem_inode_info *info = SHMEM_I(inode);
3275 name = xattr_full_name(handler, name);
3276 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3279 static const struct xattr_handler shmem_security_xattr_handler = {
3280 .prefix = XATTR_SECURITY_PREFIX,
3281 .get = shmem_xattr_handler_get,
3282 .set = shmem_xattr_handler_set,
3285 static const struct xattr_handler shmem_trusted_xattr_handler = {
3286 .prefix = XATTR_TRUSTED_PREFIX,
3287 .get = shmem_xattr_handler_get,
3288 .set = shmem_xattr_handler_set,
3291 static const struct xattr_handler *shmem_xattr_handlers[] = {
3292 #ifdef CONFIG_TMPFS_POSIX_ACL
3293 &posix_acl_access_xattr_handler,
3294 &posix_acl_default_xattr_handler,
3296 &shmem_security_xattr_handler,
3297 &shmem_trusted_xattr_handler,
3301 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3303 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3304 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3306 #endif /* CONFIG_TMPFS_XATTR */
3308 static const struct inode_operations shmem_short_symlink_operations = {
3309 .getattr = shmem_getattr,
3310 .get_link = simple_get_link,
3311 #ifdef CONFIG_TMPFS_XATTR
3312 .listxattr = shmem_listxattr,
3316 static const struct inode_operations shmem_symlink_inode_operations = {
3317 .getattr = shmem_getattr,
3318 .get_link = shmem_get_link,
3319 #ifdef CONFIG_TMPFS_XATTR
3320 .listxattr = shmem_listxattr,
3324 static struct dentry *shmem_get_parent(struct dentry *child)
3326 return ERR_PTR(-ESTALE);
3329 static int shmem_match(struct inode *ino, void *vfh)
3333 inum = (inum << 32) | fh[1];
3334 return ino->i_ino == inum && fh[0] == ino->i_generation;
3337 /* Find any alias of inode, but prefer a hashed alias */
3338 static struct dentry *shmem_find_alias(struct inode *inode)
3340 struct dentry *alias = d_find_alias(inode);
3342 return alias ?: d_find_any_alias(inode);
3346 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3347 struct fid *fid, int fh_len, int fh_type)
3349 struct inode *inode;
3350 struct dentry *dentry = NULL;
3357 inum = (inum << 32) | fid->raw[1];
3359 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3360 shmem_match, fid->raw);
3362 dentry = shmem_find_alias(inode);
3369 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3370 struct inode *parent)
3374 return FILEID_INVALID;
3377 if (inode_unhashed(inode)) {
3378 /* Unfortunately insert_inode_hash is not idempotent,
3379 * so as we hash inodes here rather than at creation
3380 * time, we need a lock to ensure we only try
3383 static DEFINE_SPINLOCK(lock);
3385 if (inode_unhashed(inode))
3386 __insert_inode_hash(inode,
3387 inode->i_ino + inode->i_generation);
3391 fh[0] = inode->i_generation;
3392 fh[1] = inode->i_ino;
3393 fh[2] = ((__u64)inode->i_ino) >> 32;
3399 static const struct export_operations shmem_export_ops = {
3400 .get_parent = shmem_get_parent,
3401 .encode_fh = shmem_encode_fh,
3402 .fh_to_dentry = shmem_fh_to_dentry,
3418 static const struct constant_table shmem_param_enums_huge[] = {
3419 {"never", SHMEM_HUGE_NEVER },
3420 {"always", SHMEM_HUGE_ALWAYS },
3421 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3422 {"advise", SHMEM_HUGE_ADVISE },
3426 const struct fs_parameter_spec shmem_fs_parameters[] = {
3427 fsparam_u32 ("gid", Opt_gid),
3428 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3429 fsparam_u32oct("mode", Opt_mode),
3430 fsparam_string("mpol", Opt_mpol),
3431 fsparam_string("nr_blocks", Opt_nr_blocks),
3432 fsparam_string("nr_inodes", Opt_nr_inodes),
3433 fsparam_string("size", Opt_size),
3434 fsparam_u32 ("uid", Opt_uid),
3435 fsparam_flag ("inode32", Opt_inode32),
3436 fsparam_flag ("inode64", Opt_inode64),
3440 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3442 struct shmem_options *ctx = fc->fs_private;
3443 struct fs_parse_result result;
3444 unsigned long long size;
3448 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3454 size = memparse(param->string, &rest);
3456 size <<= PAGE_SHIFT;
3457 size *= totalram_pages();
3463 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3464 ctx->seen |= SHMEM_SEEN_BLOCKS;
3467 ctx->blocks = memparse(param->string, &rest);
3468 if (*rest || ctx->blocks > S64_MAX)
3470 ctx->seen |= SHMEM_SEEN_BLOCKS;
3473 ctx->inodes = memparse(param->string, &rest);
3476 ctx->seen |= SHMEM_SEEN_INODES;
3479 ctx->mode = result.uint_32 & 07777;
3482 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3483 if (!uid_valid(ctx->uid))
3487 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3488 if (!gid_valid(ctx->gid))
3492 ctx->huge = result.uint_32;
3493 if (ctx->huge != SHMEM_HUGE_NEVER &&
3494 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3495 has_transparent_hugepage()))
3496 goto unsupported_parameter;
3497 ctx->seen |= SHMEM_SEEN_HUGE;
3500 if (IS_ENABLED(CONFIG_NUMA)) {
3501 mpol_put(ctx->mpol);
3503 if (mpol_parse_str(param->string, &ctx->mpol))
3507 goto unsupported_parameter;
3509 ctx->full_inums = false;
3510 ctx->seen |= SHMEM_SEEN_INUMS;
3513 if (sizeof(ino_t) < 8) {
3515 "Cannot use inode64 with <64bit inums in kernel\n");
3517 ctx->full_inums = true;
3518 ctx->seen |= SHMEM_SEEN_INUMS;
3523 unsupported_parameter:
3524 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3526 return invalfc(fc, "Bad value for '%s'", param->key);
3529 static int shmem_parse_options(struct fs_context *fc, void *data)
3531 char *options = data;
3534 int err = security_sb_eat_lsm_opts(options, &fc->security);
3539 while (options != NULL) {
3540 char *this_char = options;
3543 * NUL-terminate this option: unfortunately,
3544 * mount options form a comma-separated list,
3545 * but mpol's nodelist may also contain commas.
3547 options = strchr(options, ',');
3548 if (options == NULL)
3551 if (!isdigit(*options)) {
3557 char *value = strchr(this_char, '=');
3563 len = strlen(value);
3565 err = vfs_parse_fs_string(fc, this_char, value, len);
3574 * Reconfigure a shmem filesystem.
3576 * Note that we disallow change from limited->unlimited blocks/inodes while any
3577 * are in use; but we must separately disallow unlimited->limited, because in
3578 * that case we have no record of how much is already in use.
3580 static int shmem_reconfigure(struct fs_context *fc)
3582 struct shmem_options *ctx = fc->fs_private;
3583 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3584 unsigned long inodes;
3585 struct mempolicy *mpol = NULL;
3588 raw_spin_lock(&sbinfo->stat_lock);
3589 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3591 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3592 if (!sbinfo->max_blocks) {
3593 err = "Cannot retroactively limit size";
3596 if (percpu_counter_compare(&sbinfo->used_blocks,
3598 err = "Too small a size for current use";
3602 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3603 if (!sbinfo->max_inodes) {
3604 err = "Cannot retroactively limit inodes";
3607 if (ctx->inodes < inodes) {
3608 err = "Too few inodes for current use";
3613 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3614 sbinfo->next_ino > UINT_MAX) {
3615 err = "Current inum too high to switch to 32-bit inums";
3619 if (ctx->seen & SHMEM_SEEN_HUGE)
3620 sbinfo->huge = ctx->huge;
3621 if (ctx->seen & SHMEM_SEEN_INUMS)
3622 sbinfo->full_inums = ctx->full_inums;
3623 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3624 sbinfo->max_blocks = ctx->blocks;
3625 if (ctx->seen & SHMEM_SEEN_INODES) {
3626 sbinfo->max_inodes = ctx->inodes;
3627 sbinfo->free_inodes = ctx->inodes - inodes;
3631 * Preserve previous mempolicy unless mpol remount option was specified.
3634 mpol = sbinfo->mpol;
3635 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3638 raw_spin_unlock(&sbinfo->stat_lock);
3642 raw_spin_unlock(&sbinfo->stat_lock);
3643 return invalfc(fc, "%s", err);
3646 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3648 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3650 if (sbinfo->max_blocks != shmem_default_max_blocks())
3651 seq_printf(seq, ",size=%luk",
3652 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3653 if (sbinfo->max_inodes != shmem_default_max_inodes())
3654 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3655 if (sbinfo->mode != (0777 | S_ISVTX))
3656 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3657 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3658 seq_printf(seq, ",uid=%u",
3659 from_kuid_munged(&init_user_ns, sbinfo->uid));
3660 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3661 seq_printf(seq, ",gid=%u",
3662 from_kgid_munged(&init_user_ns, sbinfo->gid));
3665 * Showing inode{64,32} might be useful even if it's the system default,
3666 * since then people don't have to resort to checking both here and
3667 * /proc/config.gz to confirm 64-bit inums were successfully applied
3668 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3670 * We hide it when inode64 isn't the default and we are using 32-bit
3671 * inodes, since that probably just means the feature isn't even under
3676 * +-----------------+-----------------+
3677 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3678 * +------------------+-----------------+-----------------+
3679 * | full_inums=true | show | show |
3680 * | full_inums=false | show | hide |
3681 * +------------------+-----------------+-----------------+
3684 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3685 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3686 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3687 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3689 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3691 shmem_show_mpol(seq, sbinfo->mpol);
3695 #endif /* CONFIG_TMPFS */
3697 static void shmem_put_super(struct super_block *sb)
3699 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3701 free_percpu(sbinfo->ino_batch);
3702 percpu_counter_destroy(&sbinfo->used_blocks);
3703 mpol_put(sbinfo->mpol);
3705 sb->s_fs_info = NULL;
3708 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3710 struct shmem_options *ctx = fc->fs_private;
3711 struct inode *inode;
3712 struct shmem_sb_info *sbinfo;
3714 /* Round up to L1_CACHE_BYTES to resist false sharing */
3715 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3716 L1_CACHE_BYTES), GFP_KERNEL);
3720 sb->s_fs_info = sbinfo;
3724 * Per default we only allow half of the physical ram per
3725 * tmpfs instance, limiting inodes to one per page of lowmem;
3726 * but the internal instance is left unlimited.
3728 if (!(sb->s_flags & SB_KERNMOUNT)) {
3729 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3730 ctx->blocks = shmem_default_max_blocks();
3731 if (!(ctx->seen & SHMEM_SEEN_INODES))
3732 ctx->inodes = shmem_default_max_inodes();
3733 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3734 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3736 sb->s_flags |= SB_NOUSER;
3738 sb->s_export_op = &shmem_export_ops;
3739 sb->s_flags |= SB_NOSEC;
3741 sb->s_flags |= SB_NOUSER;
3743 sbinfo->max_blocks = ctx->blocks;
3744 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3745 if (sb->s_flags & SB_KERNMOUNT) {
3746 sbinfo->ino_batch = alloc_percpu(ino_t);
3747 if (!sbinfo->ino_batch)
3750 sbinfo->uid = ctx->uid;
3751 sbinfo->gid = ctx->gid;
3752 sbinfo->full_inums = ctx->full_inums;
3753 sbinfo->mode = ctx->mode;
3754 sbinfo->huge = ctx->huge;
3755 sbinfo->mpol = ctx->mpol;
3758 raw_spin_lock_init(&sbinfo->stat_lock);
3759 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3761 spin_lock_init(&sbinfo->shrinklist_lock);
3762 INIT_LIST_HEAD(&sbinfo->shrinklist);
3764 sb->s_maxbytes = MAX_LFS_FILESIZE;
3765 sb->s_blocksize = PAGE_SIZE;
3766 sb->s_blocksize_bits = PAGE_SHIFT;
3767 sb->s_magic = TMPFS_MAGIC;
3768 sb->s_op = &shmem_ops;
3769 sb->s_time_gran = 1;
3770 #ifdef CONFIG_TMPFS_XATTR
3771 sb->s_xattr = shmem_xattr_handlers;
3773 #ifdef CONFIG_TMPFS_POSIX_ACL
3774 sb->s_flags |= SB_POSIXACL;
3776 uuid_gen(&sb->s_uuid);
3778 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3781 inode->i_uid = sbinfo->uid;
3782 inode->i_gid = sbinfo->gid;
3783 sb->s_root = d_make_root(inode);
3789 shmem_put_super(sb);
3793 static int shmem_get_tree(struct fs_context *fc)
3795 return get_tree_nodev(fc, shmem_fill_super);
3798 static void shmem_free_fc(struct fs_context *fc)
3800 struct shmem_options *ctx = fc->fs_private;
3803 mpol_put(ctx->mpol);
3808 static const struct fs_context_operations shmem_fs_context_ops = {
3809 .free = shmem_free_fc,
3810 .get_tree = shmem_get_tree,
3812 .parse_monolithic = shmem_parse_options,
3813 .parse_param = shmem_parse_one,
3814 .reconfigure = shmem_reconfigure,
3818 static struct kmem_cache *shmem_inode_cachep;
3820 static struct inode *shmem_alloc_inode(struct super_block *sb)
3822 struct shmem_inode_info *info;
3823 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
3826 return &info->vfs_inode;
3829 static void shmem_free_in_core_inode(struct inode *inode)
3831 if (S_ISLNK(inode->i_mode))
3832 kfree(inode->i_link);
3833 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3836 static void shmem_destroy_inode(struct inode *inode)
3838 if (S_ISREG(inode->i_mode))
3839 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3842 static void shmem_init_inode(void *foo)
3844 struct shmem_inode_info *info = foo;
3845 inode_init_once(&info->vfs_inode);
3848 static void shmem_init_inodecache(void)
3850 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3851 sizeof(struct shmem_inode_info),
3852 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3855 static void shmem_destroy_inodecache(void)
3857 kmem_cache_destroy(shmem_inode_cachep);
3860 /* Keep the page in page cache instead of truncating it */
3861 static int shmem_error_remove_page(struct address_space *mapping,
3867 const struct address_space_operations shmem_aops = {
3868 .writepage = shmem_writepage,
3869 .dirty_folio = noop_dirty_folio,
3871 .write_begin = shmem_write_begin,
3872 .write_end = shmem_write_end,
3874 #ifdef CONFIG_MIGRATION
3875 .migrate_folio = migrate_folio,
3877 .error_remove_page = shmem_error_remove_page,
3879 EXPORT_SYMBOL(shmem_aops);
3881 static const struct file_operations shmem_file_operations = {
3883 .get_unmapped_area = shmem_get_unmapped_area,
3885 .llseek = shmem_file_llseek,
3886 .read_iter = shmem_file_read_iter,
3887 .write_iter = generic_file_write_iter,
3888 .fsync = noop_fsync,
3889 .splice_read = generic_file_splice_read,
3890 .splice_write = iter_file_splice_write,
3891 .fallocate = shmem_fallocate,
3895 static const struct inode_operations shmem_inode_operations = {
3896 .getattr = shmem_getattr,
3897 .setattr = shmem_setattr,
3898 #ifdef CONFIG_TMPFS_XATTR
3899 .listxattr = shmem_listxattr,
3900 .set_acl = simple_set_acl,
3901 .fileattr_get = shmem_fileattr_get,
3902 .fileattr_set = shmem_fileattr_set,
3906 static const struct inode_operations shmem_dir_inode_operations = {
3908 .getattr = shmem_getattr,
3909 .create = shmem_create,
3910 .lookup = simple_lookup,
3912 .unlink = shmem_unlink,
3913 .symlink = shmem_symlink,
3914 .mkdir = shmem_mkdir,
3915 .rmdir = shmem_rmdir,
3916 .mknod = shmem_mknod,
3917 .rename = shmem_rename2,
3918 .tmpfile = shmem_tmpfile,
3920 #ifdef CONFIG_TMPFS_XATTR
3921 .listxattr = shmem_listxattr,
3922 .fileattr_get = shmem_fileattr_get,
3923 .fileattr_set = shmem_fileattr_set,
3925 #ifdef CONFIG_TMPFS_POSIX_ACL
3926 .setattr = shmem_setattr,
3927 .set_acl = simple_set_acl,
3931 static const struct inode_operations shmem_special_inode_operations = {
3932 .getattr = shmem_getattr,
3933 #ifdef CONFIG_TMPFS_XATTR
3934 .listxattr = shmem_listxattr,
3936 #ifdef CONFIG_TMPFS_POSIX_ACL
3937 .setattr = shmem_setattr,
3938 .set_acl = simple_set_acl,
3942 static const struct super_operations shmem_ops = {
3943 .alloc_inode = shmem_alloc_inode,
3944 .free_inode = shmem_free_in_core_inode,
3945 .destroy_inode = shmem_destroy_inode,
3947 .statfs = shmem_statfs,
3948 .show_options = shmem_show_options,
3950 .evict_inode = shmem_evict_inode,
3951 .drop_inode = generic_delete_inode,
3952 .put_super = shmem_put_super,
3953 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3954 .nr_cached_objects = shmem_unused_huge_count,
3955 .free_cached_objects = shmem_unused_huge_scan,
3959 static const struct vm_operations_struct shmem_vm_ops = {
3960 .fault = shmem_fault,
3961 .map_pages = filemap_map_pages,
3963 .set_policy = shmem_set_policy,
3964 .get_policy = shmem_get_policy,
3968 int shmem_init_fs_context(struct fs_context *fc)
3970 struct shmem_options *ctx;
3972 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3976 ctx->mode = 0777 | S_ISVTX;
3977 ctx->uid = current_fsuid();
3978 ctx->gid = current_fsgid();
3980 fc->fs_private = ctx;
3981 fc->ops = &shmem_fs_context_ops;
3985 static struct file_system_type shmem_fs_type = {
3986 .owner = THIS_MODULE,
3988 .init_fs_context = shmem_init_fs_context,
3990 .parameters = shmem_fs_parameters,
3992 .kill_sb = kill_litter_super,
3993 .fs_flags = FS_USERNS_MOUNT,
3996 void __init shmem_init(void)
4000 shmem_init_inodecache();
4002 error = register_filesystem(&shmem_fs_type);
4004 pr_err("Could not register tmpfs\n");
4008 shm_mnt = kern_mount(&shmem_fs_type);
4009 if (IS_ERR(shm_mnt)) {
4010 error = PTR_ERR(shm_mnt);
4011 pr_err("Could not kern_mount tmpfs\n");
4015 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4016 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4017 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4019 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4024 unregister_filesystem(&shmem_fs_type);
4026 shmem_destroy_inodecache();
4027 shm_mnt = ERR_PTR(error);
4030 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4031 static ssize_t shmem_enabled_show(struct kobject *kobj,
4032 struct kobj_attribute *attr, char *buf)
4034 static const int values[] = {
4036 SHMEM_HUGE_WITHIN_SIZE,
4045 for (i = 0; i < ARRAY_SIZE(values); i++) {
4046 len += sysfs_emit_at(buf, len,
4047 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4049 shmem_format_huge(values[i]));
4052 len += sysfs_emit_at(buf, len, "\n");
4057 static ssize_t shmem_enabled_store(struct kobject *kobj,
4058 struct kobj_attribute *attr, const char *buf, size_t count)
4063 if (count + 1 > sizeof(tmp))
4065 memcpy(tmp, buf, count);
4067 if (count && tmp[count - 1] == '\n')
4068 tmp[count - 1] = '\0';
4070 huge = shmem_parse_huge(tmp);
4071 if (huge == -EINVAL)
4073 if (!has_transparent_hugepage() &&
4074 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4078 if (shmem_huge > SHMEM_HUGE_DENY)
4079 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4083 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
4084 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4086 #else /* !CONFIG_SHMEM */
4089 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4091 * This is intended for small system where the benefits of the full
4092 * shmem code (swap-backed and resource-limited) are outweighed by
4093 * their complexity. On systems without swap this code should be
4094 * effectively equivalent, but much lighter weight.
4097 static struct file_system_type shmem_fs_type = {
4099 .init_fs_context = ramfs_init_fs_context,
4100 .parameters = ramfs_fs_parameters,
4101 .kill_sb = kill_litter_super,
4102 .fs_flags = FS_USERNS_MOUNT,
4105 void __init shmem_init(void)
4107 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4109 shm_mnt = kern_mount(&shmem_fs_type);
4110 BUG_ON(IS_ERR(shm_mnt));
4113 int shmem_unuse(unsigned int type)
4118 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4123 void shmem_unlock_mapping(struct address_space *mapping)
4128 unsigned long shmem_get_unmapped_area(struct file *file,
4129 unsigned long addr, unsigned long len,
4130 unsigned long pgoff, unsigned long flags)
4132 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4136 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4138 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4140 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4142 #define shmem_vm_ops generic_file_vm_ops
4143 #define shmem_file_operations ramfs_file_operations
4144 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4145 #define shmem_acct_size(flags, size) 0
4146 #define shmem_unacct_size(flags, size) do {} while (0)
4148 #endif /* CONFIG_SHMEM */
4152 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4153 unsigned long flags, unsigned int i_flags)
4155 struct inode *inode;
4159 return ERR_CAST(mnt);
4161 if (size < 0 || size > MAX_LFS_FILESIZE)
4162 return ERR_PTR(-EINVAL);
4164 if (shmem_acct_size(flags, size))
4165 return ERR_PTR(-ENOMEM);
4167 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4169 if (unlikely(!inode)) {
4170 shmem_unacct_size(flags, size);
4171 return ERR_PTR(-ENOSPC);
4173 inode->i_flags |= i_flags;
4174 inode->i_size = size;
4175 clear_nlink(inode); /* It is unlinked */
4176 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4178 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4179 &shmem_file_operations);
4186 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4187 * kernel internal. There will be NO LSM permission checks against the
4188 * underlying inode. So users of this interface must do LSM checks at a
4189 * higher layer. The users are the big_key and shm implementations. LSM
4190 * checks are provided at the key or shm level rather than the inode.
4191 * @name: name for dentry (to be seen in /proc/<pid>/maps
4192 * @size: size to be set for the file
4193 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4195 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4197 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4201 * shmem_file_setup - get an unlinked file living in tmpfs
4202 * @name: name for dentry (to be seen in /proc/<pid>/maps
4203 * @size: size to be set for the file
4204 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4206 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4208 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4210 EXPORT_SYMBOL_GPL(shmem_file_setup);
4213 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4214 * @mnt: the tmpfs mount where the file will be created
4215 * @name: name for dentry (to be seen in /proc/<pid>/maps
4216 * @size: size to be set for the file
4217 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4219 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4220 loff_t size, unsigned long flags)
4222 return __shmem_file_setup(mnt, name, size, flags, 0);
4224 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4227 * shmem_zero_setup - setup a shared anonymous mapping
4228 * @vma: the vma to be mmapped is prepared by do_mmap
4230 int shmem_zero_setup(struct vm_area_struct *vma)
4233 loff_t size = vma->vm_end - vma->vm_start;
4236 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4237 * between XFS directory reading and selinux: since this file is only
4238 * accessible to the user through its mapping, use S_PRIVATE flag to
4239 * bypass file security, in the same way as shmem_kernel_file_setup().
4241 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4243 return PTR_ERR(file);
4247 vma->vm_file = file;
4248 vma->vm_ops = &shmem_vm_ops;
4254 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4255 * @mapping: the page's address_space
4256 * @index: the page index
4257 * @gfp: the page allocator flags to use if allocating
4259 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4260 * with any new page allocations done using the specified allocation flags.
4261 * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4262 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4263 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4265 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4266 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4268 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4269 pgoff_t index, gfp_t gfp)
4272 struct inode *inode = mapping->host;
4276 BUG_ON(!shmem_mapping(mapping));
4277 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4278 gfp, NULL, NULL, NULL);
4280 return ERR_PTR(error);
4283 if (PageHWPoison(page)) {
4285 return ERR_PTR(-EIO);
4291 * The tiny !SHMEM case uses ramfs without swap
4293 return read_cache_page_gfp(mapping, index, gfp);
4296 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);