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>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/fs_parser.h>
40 #include <linux/swapfile.h>
42 static struct vfsmount *shm_mnt;
46 * This virtual memory filesystem is heavily based on the ramfs. It
47 * extends ramfs by the ability to use swap and honor resource limits
48 * which makes it a completely usable filesystem.
51 #include <linux/xattr.h>
52 #include <linux/exportfs.h>
53 #include <linux/posix_acl.h>
54 #include <linux/posix_acl_xattr.h>
55 #include <linux/mman.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/backing-dev.h>
59 #include <linux/shmem_fs.h>
60 #include <linux/writeback.h>
61 #include <linux/pagevec.h>
62 #include <linux/percpu_counter.h>
63 #include <linux/falloc.h>
64 #include <linux/splice.h>
65 #include <linux/security.h>
66 #include <linux/swapops.h>
67 #include <linux/mempolicy.h>
68 #include <linux/namei.h>
69 #include <linux/ctype.h>
70 #include <linux/migrate.h>
71 #include <linux/highmem.h>
72 #include <linux/seq_file.h>
73 #include <linux/magic.h>
74 #include <linux/syscalls.h>
75 #include <linux/fcntl.h>
76 #include <uapi/linux/memfd.h>
77 #include <linux/userfaultfd_k.h>
78 #include <linux/rmap.h>
79 #include <linux/uuid.h>
81 #include <linux/uaccess.h>
85 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
86 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
88 /* Pretend that each entry is of this size in directory's i_size */
89 #define BOGO_DIRENT_SIZE 20
91 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
92 #define SHORT_SYMLINK_LEN 128
95 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
96 * inode->i_private (with i_rwsem making sure that it has only one user at
97 * a time): we would prefer not to enlarge the shmem inode just for that.
100 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
101 pgoff_t start; /* start of range currently being fallocated */
102 pgoff_t next; /* the next page offset to be fallocated */
103 pgoff_t nr_falloced; /* how many new pages have been fallocated */
104 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
107 struct shmem_options {
108 unsigned long long blocks;
109 unsigned long long inodes;
110 struct mempolicy *mpol;
117 #define SHMEM_SEEN_BLOCKS 1
118 #define SHMEM_SEEN_INODES 2
119 #define SHMEM_SEEN_HUGE 4
120 #define SHMEM_SEEN_INUMS 8
124 static unsigned long shmem_default_max_blocks(void)
126 return totalram_pages() / 2;
129 static unsigned long shmem_default_max_inodes(void)
131 unsigned long nr_pages = totalram_pages();
133 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
137 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
138 struct page **pagep, enum sgp_type sgp,
139 gfp_t gfp, struct vm_area_struct *vma,
140 vm_fault_t *fault_type);
141 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
142 struct page **pagep, enum sgp_type sgp,
143 gfp_t gfp, struct vm_area_struct *vma,
144 struct vm_fault *vmf, vm_fault_t *fault_type);
146 int shmem_getpage(struct inode *inode, pgoff_t index,
147 struct page **pagep, enum sgp_type sgp)
149 return shmem_getpage_gfp(inode, index, pagep, sgp,
150 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
153 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
155 return sb->s_fs_info;
159 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
160 * for shared memory and for shared anonymous (/dev/zero) mappings
161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
162 * consistent with the pre-accounting of private mappings ...
164 static inline int shmem_acct_size(unsigned long flags, loff_t size)
166 return (flags & VM_NORESERVE) ?
167 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
170 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
172 if (!(flags & VM_NORESERVE))
173 vm_unacct_memory(VM_ACCT(size));
176 static inline int shmem_reacct_size(unsigned long flags,
177 loff_t oldsize, loff_t newsize)
179 if (!(flags & VM_NORESERVE)) {
180 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
181 return security_vm_enough_memory_mm(current->mm,
182 VM_ACCT(newsize) - VM_ACCT(oldsize));
183 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
184 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
190 * ... whereas tmpfs objects are accounted incrementally as
191 * pages are allocated, in order to allow large sparse files.
192 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
193 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
195 static inline int shmem_acct_block(unsigned long flags, long pages)
197 if (!(flags & VM_NORESERVE))
200 return security_vm_enough_memory_mm(current->mm,
201 pages * VM_ACCT(PAGE_SIZE));
204 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
206 if (flags & VM_NORESERVE)
207 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
210 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
212 struct shmem_inode_info *info = SHMEM_I(inode);
213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
215 if (shmem_acct_block(info->flags, pages))
218 if (sbinfo->max_blocks) {
219 if (percpu_counter_compare(&sbinfo->used_blocks,
220 sbinfo->max_blocks - pages) > 0)
222 percpu_counter_add(&sbinfo->used_blocks, pages);
228 shmem_unacct_blocks(info->flags, pages);
232 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
234 struct shmem_inode_info *info = SHMEM_I(inode);
235 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
237 if (sbinfo->max_blocks)
238 percpu_counter_sub(&sbinfo->used_blocks, pages);
239 shmem_unacct_blocks(info->flags, pages);
242 static const struct super_operations shmem_ops;
243 const struct address_space_operations shmem_aops;
244 static const struct file_operations shmem_file_operations;
245 static const struct inode_operations shmem_inode_operations;
246 static const struct inode_operations shmem_dir_inode_operations;
247 static const struct inode_operations shmem_special_inode_operations;
248 static const struct vm_operations_struct shmem_vm_ops;
249 static struct file_system_type shmem_fs_type;
251 bool vma_is_shmem(struct vm_area_struct *vma)
253 return vma->vm_ops == &shmem_vm_ops;
256 static LIST_HEAD(shmem_swaplist);
257 static DEFINE_MUTEX(shmem_swaplist_mutex);
260 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
261 * produces a novel ino for the newly allocated inode.
263 * It may also be called when making a hard link to permit the space needed by
264 * each dentry. However, in that case, no new inode number is needed since that
265 * internally draws from another pool of inode numbers (currently global
266 * get_next_ino()). This case is indicated by passing NULL as inop.
268 #define SHMEM_INO_BATCH 1024
269 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
271 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
274 if (!(sb->s_flags & SB_KERNMOUNT)) {
275 raw_spin_lock(&sbinfo->stat_lock);
276 if (sbinfo->max_inodes) {
277 if (!sbinfo->free_inodes) {
278 raw_spin_unlock(&sbinfo->stat_lock);
281 sbinfo->free_inodes--;
284 ino = sbinfo->next_ino++;
285 if (unlikely(is_zero_ino(ino)))
286 ino = sbinfo->next_ino++;
287 if (unlikely(!sbinfo->full_inums &&
290 * Emulate get_next_ino uint wraparound for
293 if (IS_ENABLED(CONFIG_64BIT))
294 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
295 __func__, MINOR(sb->s_dev));
296 sbinfo->next_ino = 1;
297 ino = sbinfo->next_ino++;
301 raw_spin_unlock(&sbinfo->stat_lock);
304 * __shmem_file_setup, one of our callers, is lock-free: it
305 * doesn't hold stat_lock in shmem_reserve_inode since
306 * max_inodes is always 0, and is called from potentially
307 * unknown contexts. As such, use a per-cpu batched allocator
308 * which doesn't require the per-sb stat_lock unless we are at
309 * the batch boundary.
311 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
312 * shmem mounts are not exposed to userspace, so we don't need
313 * to worry about things like glibc compatibility.
317 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
319 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
320 raw_spin_lock(&sbinfo->stat_lock);
321 ino = sbinfo->next_ino;
322 sbinfo->next_ino += SHMEM_INO_BATCH;
323 raw_spin_unlock(&sbinfo->stat_lock);
324 if (unlikely(is_zero_ino(ino)))
335 static void shmem_free_inode(struct super_block *sb)
337 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
338 if (sbinfo->max_inodes) {
339 raw_spin_lock(&sbinfo->stat_lock);
340 sbinfo->free_inodes++;
341 raw_spin_unlock(&sbinfo->stat_lock);
346 * shmem_recalc_inode - recalculate the block usage of an inode
347 * @inode: inode to recalc
349 * We have to calculate the free blocks since the mm can drop
350 * undirtied hole pages behind our back.
352 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
353 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
355 * It has to be called with the spinlock held.
357 static void shmem_recalc_inode(struct inode *inode)
359 struct shmem_inode_info *info = SHMEM_I(inode);
362 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
364 info->alloced -= freed;
365 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
366 shmem_inode_unacct_blocks(inode, freed);
370 bool shmem_charge(struct inode *inode, long pages)
372 struct shmem_inode_info *info = SHMEM_I(inode);
375 if (!shmem_inode_acct_block(inode, pages))
378 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
379 inode->i_mapping->nrpages += pages;
381 spin_lock_irqsave(&info->lock, flags);
382 info->alloced += pages;
383 inode->i_blocks += pages * BLOCKS_PER_PAGE;
384 shmem_recalc_inode(inode);
385 spin_unlock_irqrestore(&info->lock, flags);
390 void shmem_uncharge(struct inode *inode, long pages)
392 struct shmem_inode_info *info = SHMEM_I(inode);
395 /* nrpages adjustment done by __delete_from_page_cache() or caller */
397 spin_lock_irqsave(&info->lock, flags);
398 info->alloced -= pages;
399 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
400 shmem_recalc_inode(inode);
401 spin_unlock_irqrestore(&info->lock, flags);
403 shmem_inode_unacct_blocks(inode, pages);
407 * Replace item expected in xarray by a new item, while holding xa_lock.
409 static int shmem_replace_entry(struct address_space *mapping,
410 pgoff_t index, void *expected, void *replacement)
412 XA_STATE(xas, &mapping->i_pages, index);
415 VM_BUG_ON(!expected);
416 VM_BUG_ON(!replacement);
417 item = xas_load(&xas);
418 if (item != expected)
420 xas_store(&xas, replacement);
425 * Sometimes, before we decide whether to proceed or to fail, we must check
426 * that an entry was not already brought back from swap by a racing thread.
428 * Checking page is not enough: by the time a SwapCache page is locked, it
429 * might be reused, and again be SwapCache, using the same swap as before.
431 static bool shmem_confirm_swap(struct address_space *mapping,
432 pgoff_t index, swp_entry_t swap)
434 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
438 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
441 * disables huge pages for the mount;
443 * enables huge pages for the mount;
444 * SHMEM_HUGE_WITHIN_SIZE:
445 * only allocate huge pages if the page will be fully within i_size,
446 * also respect fadvise()/madvise() hints;
448 * only allocate huge pages if requested with fadvise()/madvise();
451 #define SHMEM_HUGE_NEVER 0
452 #define SHMEM_HUGE_ALWAYS 1
453 #define SHMEM_HUGE_WITHIN_SIZE 2
454 #define SHMEM_HUGE_ADVISE 3
458 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
461 * disables huge on shm_mnt and all mounts, for emergency use;
463 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
466 #define SHMEM_HUGE_DENY (-1)
467 #define SHMEM_HUGE_FORCE (-2)
469 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
470 /* ifdef here to avoid bloating shmem.o when not necessary */
472 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
474 bool shmem_is_huge(struct vm_area_struct *vma,
475 struct inode *inode, pgoff_t index)
479 if (!S_ISREG(inode->i_mode))
481 if (shmem_huge == SHMEM_HUGE_DENY)
483 if (vma && ((vma->vm_flags & VM_NOHUGEPAGE) ||
484 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
486 if (shmem_huge == SHMEM_HUGE_FORCE)
489 switch (SHMEM_SB(inode->i_sb)->huge) {
490 case SHMEM_HUGE_ALWAYS:
492 case SHMEM_HUGE_WITHIN_SIZE:
493 index = round_up(index + 1, HPAGE_PMD_NR);
494 i_size = round_up(i_size_read(inode), PAGE_SIZE);
495 if (i_size >> PAGE_SHIFT >= index)
498 case SHMEM_HUGE_ADVISE:
499 if (vma && (vma->vm_flags & VM_HUGEPAGE))
507 #if defined(CONFIG_SYSFS)
508 static int shmem_parse_huge(const char *str)
510 if (!strcmp(str, "never"))
511 return SHMEM_HUGE_NEVER;
512 if (!strcmp(str, "always"))
513 return SHMEM_HUGE_ALWAYS;
514 if (!strcmp(str, "within_size"))
515 return SHMEM_HUGE_WITHIN_SIZE;
516 if (!strcmp(str, "advise"))
517 return SHMEM_HUGE_ADVISE;
518 if (!strcmp(str, "deny"))
519 return SHMEM_HUGE_DENY;
520 if (!strcmp(str, "force"))
521 return SHMEM_HUGE_FORCE;
526 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
527 static const char *shmem_format_huge(int huge)
530 case SHMEM_HUGE_NEVER:
532 case SHMEM_HUGE_ALWAYS:
534 case SHMEM_HUGE_WITHIN_SIZE:
535 return "within_size";
536 case SHMEM_HUGE_ADVISE:
538 case SHMEM_HUGE_DENY:
540 case SHMEM_HUGE_FORCE:
549 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
550 struct shrink_control *sc, unsigned long nr_to_split)
552 LIST_HEAD(list), *pos, *next;
553 LIST_HEAD(to_remove);
555 struct shmem_inode_info *info;
557 unsigned long batch = sc ? sc->nr_to_scan : 128;
560 if (list_empty(&sbinfo->shrinklist))
563 spin_lock(&sbinfo->shrinklist_lock);
564 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
565 info = list_entry(pos, struct shmem_inode_info, shrinklist);
568 inode = igrab(&info->vfs_inode);
570 /* inode is about to be evicted */
572 list_del_init(&info->shrinklist);
576 /* Check if there's anything to gain */
577 if (round_up(inode->i_size, PAGE_SIZE) ==
578 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
579 list_move(&info->shrinklist, &to_remove);
583 list_move(&info->shrinklist, &list);
585 sbinfo->shrinklist_len--;
589 spin_unlock(&sbinfo->shrinklist_lock);
591 list_for_each_safe(pos, next, &to_remove) {
592 info = list_entry(pos, struct shmem_inode_info, shrinklist);
593 inode = &info->vfs_inode;
594 list_del_init(&info->shrinklist);
598 list_for_each_safe(pos, next, &list) {
601 info = list_entry(pos, struct shmem_inode_info, shrinklist);
602 inode = &info->vfs_inode;
604 if (nr_to_split && split >= nr_to_split)
607 page = find_get_page(inode->i_mapping,
608 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
612 /* No huge page at the end of the file: nothing to split */
613 if (!PageTransHuge(page)) {
619 * Move the inode on the list back to shrinklist if we failed
620 * to lock the page at this time.
622 * Waiting for the lock may lead to deadlock in the
625 if (!trylock_page(page)) {
630 ret = split_huge_page(page);
634 /* If split failed move the inode on the list back to shrinklist */
640 list_del_init(&info->shrinklist);
644 * Make sure the inode is either on the global list or deleted
645 * from any local list before iput() since it could be deleted
646 * in another thread once we put the inode (then the local list
649 spin_lock(&sbinfo->shrinklist_lock);
650 list_move(&info->shrinklist, &sbinfo->shrinklist);
651 sbinfo->shrinklist_len++;
652 spin_unlock(&sbinfo->shrinklist_lock);
660 static long shmem_unused_huge_scan(struct super_block *sb,
661 struct shrink_control *sc)
663 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
665 if (!READ_ONCE(sbinfo->shrinklist_len))
668 return shmem_unused_huge_shrink(sbinfo, sc, 0);
671 static long shmem_unused_huge_count(struct super_block *sb,
672 struct shrink_control *sc)
674 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
675 return READ_ONCE(sbinfo->shrinklist_len);
677 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
679 #define shmem_huge SHMEM_HUGE_DENY
681 bool shmem_is_huge(struct vm_area_struct *vma,
682 struct inode *inode, pgoff_t index)
687 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
688 struct shrink_control *sc, unsigned long nr_to_split)
692 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
695 * Like add_to_page_cache_locked, but error if expected item has gone.
697 static int shmem_add_to_page_cache(struct page *page,
698 struct address_space *mapping,
699 pgoff_t index, void *expected, gfp_t gfp,
700 struct mm_struct *charge_mm)
702 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
703 unsigned long nr = compound_nr(page);
706 VM_BUG_ON_PAGE(PageTail(page), page);
707 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
708 VM_BUG_ON_PAGE(!PageLocked(page), page);
709 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
710 VM_BUG_ON(expected && PageTransHuge(page));
712 page_ref_add(page, nr);
713 page->mapping = mapping;
716 if (!PageSwapCache(page)) {
717 error = mem_cgroup_charge(page_folio(page), charge_mm, gfp);
719 if (PageTransHuge(page)) {
720 count_vm_event(THP_FILE_FALLBACK);
721 count_vm_event(THP_FILE_FALLBACK_CHARGE);
726 cgroup_throttle_swaprate(page, gfp);
730 if (expected != xas_find_conflict(&xas)) {
731 xas_set_err(&xas, -EEXIST);
734 if (expected && xas_find_conflict(&xas)) {
735 xas_set_err(&xas, -EEXIST);
738 xas_store(&xas, page);
741 if (PageTransHuge(page)) {
742 count_vm_event(THP_FILE_ALLOC);
743 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
745 mapping->nrpages += nr;
746 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
747 __mod_lruvec_page_state(page, NR_SHMEM, nr);
749 xas_unlock_irq(&xas);
750 } while (xas_nomem(&xas, gfp));
752 if (xas_error(&xas)) {
753 error = xas_error(&xas);
759 page->mapping = NULL;
760 page_ref_sub(page, nr);
765 * Like delete_from_page_cache, but substitutes swap for page.
767 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
769 struct address_space *mapping = page->mapping;
772 VM_BUG_ON_PAGE(PageCompound(page), page);
774 xa_lock_irq(&mapping->i_pages);
775 error = shmem_replace_entry(mapping, page->index, page, radswap);
776 page->mapping = NULL;
778 __dec_lruvec_page_state(page, NR_FILE_PAGES);
779 __dec_lruvec_page_state(page, NR_SHMEM);
780 xa_unlock_irq(&mapping->i_pages);
786 * Remove swap entry from page cache, free the swap and its page cache.
788 static int shmem_free_swap(struct address_space *mapping,
789 pgoff_t index, void *radswap)
793 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
796 free_swap_and_cache(radix_to_swp_entry(radswap));
801 * Determine (in bytes) how many of the shmem object's pages mapped by the
802 * given offsets are swapped out.
804 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
805 * as long as the inode doesn't go away and racy results are not a problem.
807 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
808 pgoff_t start, pgoff_t end)
810 XA_STATE(xas, &mapping->i_pages, start);
812 unsigned long swapped = 0;
815 xas_for_each(&xas, page, end - 1) {
816 if (xas_retry(&xas, page))
818 if (xa_is_value(page))
821 if (need_resched()) {
829 return swapped << PAGE_SHIFT;
833 * Determine (in bytes) how many of the shmem object's pages mapped by the
834 * given vma is swapped out.
836 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
837 * as long as the inode doesn't go away and racy results are not a problem.
839 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
841 struct inode *inode = file_inode(vma->vm_file);
842 struct shmem_inode_info *info = SHMEM_I(inode);
843 struct address_space *mapping = inode->i_mapping;
844 unsigned long swapped;
846 /* Be careful as we don't hold info->lock */
847 swapped = READ_ONCE(info->swapped);
850 * The easier cases are when the shmem object has nothing in swap, or
851 * the vma maps it whole. Then we can simply use the stats that we
857 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
858 return swapped << PAGE_SHIFT;
860 /* Here comes the more involved part */
861 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
862 vma->vm_pgoff + vma_pages(vma));
866 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
868 void shmem_unlock_mapping(struct address_space *mapping)
875 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
877 while (!mapping_unevictable(mapping)) {
878 if (!pagevec_lookup(&pvec, mapping, &index))
880 check_move_unevictable_pages(&pvec);
881 pagevec_release(&pvec);
886 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
892 * At first avoid shmem_getpage(,,,SGP_READ): that fails
893 * beyond i_size, and reports fallocated pages as holes.
895 folio = __filemap_get_folio(inode->i_mapping, index,
896 FGP_ENTRY | FGP_LOCK, 0);
897 if (!xa_is_value(folio))
900 * But read a page back from swap if any of it is within i_size
901 * (although in some cases this is just a waste of time).
904 shmem_getpage(inode, index, &page, SGP_READ);
905 return page ? page_folio(page) : NULL;
909 * Remove range of pages and swap entries from page cache, and free them.
910 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
912 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
915 struct address_space *mapping = inode->i_mapping;
916 struct shmem_inode_info *info = SHMEM_I(inode);
917 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
918 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
919 struct folio_batch fbatch;
920 pgoff_t indices[PAGEVEC_SIZE];
923 long nr_swaps_freed = 0;
928 end = -1; /* unsigned, so actually very big */
930 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
931 info->fallocend = start;
933 folio_batch_init(&fbatch);
935 while (index < end && find_lock_entries(mapping, index, end - 1,
937 for (i = 0; i < folio_batch_count(&fbatch); i++) {
938 folio = fbatch.folios[i];
942 if (xa_is_value(folio)) {
945 nr_swaps_freed += !shmem_free_swap(mapping,
949 index += folio_nr_pages(folio) - 1;
951 if (!unfalloc || !folio_test_uptodate(folio))
952 truncate_inode_folio(mapping, folio);
955 folio_batch_remove_exceptionals(&fbatch);
956 folio_batch_release(&fbatch);
961 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
962 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
964 same_folio = lend < folio_pos(folio) + folio_size(folio);
965 folio_mark_dirty(folio);
966 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
967 start = folio->index + folio_nr_pages(folio);
977 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
979 folio_mark_dirty(folio);
980 if (!truncate_inode_partial_folio(folio, lstart, lend))
987 while (index < end) {
990 if (!find_get_entries(mapping, index, end - 1, &fbatch,
992 /* If all gone or hole-punch or unfalloc, we're done */
993 if (index == start || end != -1)
995 /* But if truncating, restart to make sure all gone */
999 for (i = 0; i < folio_batch_count(&fbatch); i++) {
1000 folio = fbatch.folios[i];
1003 if (xa_is_value(folio)) {
1006 if (shmem_free_swap(mapping, index, folio)) {
1007 /* Swap was replaced by page: retry */
1017 if (!unfalloc || !folio_test_uptodate(folio)) {
1018 if (folio_mapping(folio) != mapping) {
1019 /* Page was replaced by swap: retry */
1020 folio_unlock(folio);
1024 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1026 truncate_inode_folio(mapping, folio);
1028 index = folio->index + folio_nr_pages(folio) - 1;
1029 folio_unlock(folio);
1031 folio_batch_remove_exceptionals(&fbatch);
1032 folio_batch_release(&fbatch);
1036 spin_lock_irq(&info->lock);
1037 info->swapped -= nr_swaps_freed;
1038 shmem_recalc_inode(inode);
1039 spin_unlock_irq(&info->lock);
1042 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1044 shmem_undo_range(inode, lstart, lend, false);
1045 inode->i_ctime = inode->i_mtime = current_time(inode);
1047 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1049 static int shmem_getattr(struct user_namespace *mnt_userns,
1050 const struct path *path, struct kstat *stat,
1051 u32 request_mask, unsigned int query_flags)
1053 struct inode *inode = path->dentry->d_inode;
1054 struct shmem_inode_info *info = SHMEM_I(inode);
1056 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1057 spin_lock_irq(&info->lock);
1058 shmem_recalc_inode(inode);
1059 spin_unlock_irq(&info->lock);
1061 generic_fillattr(&init_user_ns, inode, stat);
1063 if (shmem_is_huge(NULL, inode, 0))
1064 stat->blksize = HPAGE_PMD_SIZE;
1066 if (request_mask & STATX_BTIME) {
1067 stat->result_mask |= STATX_BTIME;
1068 stat->btime.tv_sec = info->i_crtime.tv_sec;
1069 stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1075 static int shmem_setattr(struct user_namespace *mnt_userns,
1076 struct dentry *dentry, struct iattr *attr)
1078 struct inode *inode = d_inode(dentry);
1079 struct shmem_inode_info *info = SHMEM_I(inode);
1082 error = setattr_prepare(&init_user_ns, dentry, attr);
1086 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1087 loff_t oldsize = inode->i_size;
1088 loff_t newsize = attr->ia_size;
1090 /* protected by i_rwsem */
1091 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1092 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1095 if (newsize != oldsize) {
1096 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1100 i_size_write(inode, newsize);
1101 inode->i_ctime = inode->i_mtime = current_time(inode);
1103 if (newsize <= oldsize) {
1104 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1105 if (oldsize > holebegin)
1106 unmap_mapping_range(inode->i_mapping,
1109 shmem_truncate_range(inode,
1110 newsize, (loff_t)-1);
1111 /* unmap again to remove racily COWed private pages */
1112 if (oldsize > holebegin)
1113 unmap_mapping_range(inode->i_mapping,
1118 setattr_copy(&init_user_ns, inode, attr);
1119 if (attr->ia_valid & ATTR_MODE)
1120 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1124 static void shmem_evict_inode(struct inode *inode)
1126 struct shmem_inode_info *info = SHMEM_I(inode);
1127 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1129 if (shmem_mapping(inode->i_mapping)) {
1130 shmem_unacct_size(info->flags, inode->i_size);
1132 mapping_set_exiting(inode->i_mapping);
1133 shmem_truncate_range(inode, 0, (loff_t)-1);
1134 if (!list_empty(&info->shrinklist)) {
1135 spin_lock(&sbinfo->shrinklist_lock);
1136 if (!list_empty(&info->shrinklist)) {
1137 list_del_init(&info->shrinklist);
1138 sbinfo->shrinklist_len--;
1140 spin_unlock(&sbinfo->shrinklist_lock);
1142 while (!list_empty(&info->swaplist)) {
1143 /* Wait while shmem_unuse() is scanning this inode... */
1144 wait_var_event(&info->stop_eviction,
1145 !atomic_read(&info->stop_eviction));
1146 mutex_lock(&shmem_swaplist_mutex);
1147 /* ...but beware of the race if we peeked too early */
1148 if (!atomic_read(&info->stop_eviction))
1149 list_del_init(&info->swaplist);
1150 mutex_unlock(&shmem_swaplist_mutex);
1154 simple_xattrs_free(&info->xattrs);
1155 WARN_ON(inode->i_blocks);
1156 shmem_free_inode(inode->i_sb);
1160 static int shmem_find_swap_entries(struct address_space *mapping,
1161 pgoff_t start, unsigned int nr_entries,
1162 struct page **entries, pgoff_t *indices,
1165 XA_STATE(xas, &mapping->i_pages, start);
1168 unsigned int ret = 0;
1174 xas_for_each(&xas, page, ULONG_MAX) {
1175 if (xas_retry(&xas, page))
1178 if (!xa_is_value(page))
1181 entry = radix_to_swp_entry(page);
1182 if (swp_type(entry) != type)
1185 indices[ret] = xas.xa_index;
1186 entries[ret] = page;
1188 if (need_resched()) {
1192 if (++ret == nr_entries)
1201 * Move the swapped pages for an inode to page cache. Returns the count
1202 * of pages swapped in, or the error in case of failure.
1204 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1210 struct address_space *mapping = inode->i_mapping;
1212 for (i = 0; i < pvec.nr; i++) {
1213 struct page *page = pvec.pages[i];
1215 if (!xa_is_value(page))
1217 error = shmem_swapin_page(inode, indices[i],
1219 mapping_gfp_mask(mapping),
1226 if (error == -ENOMEM)
1230 return error ? error : ret;
1234 * If swap found in inode, free it and move page from swapcache to filecache.
1236 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1238 struct address_space *mapping = inode->i_mapping;
1240 struct pagevec pvec;
1241 pgoff_t indices[PAGEVEC_SIZE];
1244 pagevec_init(&pvec);
1246 unsigned int nr_entries = PAGEVEC_SIZE;
1248 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1249 pvec.pages, indices, type);
1255 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1259 start = indices[pvec.nr - 1];
1266 * Read all the shared memory data that resides in the swap
1267 * device 'type' back into memory, so the swap device can be
1270 int shmem_unuse(unsigned int type)
1272 struct shmem_inode_info *info, *next;
1275 if (list_empty(&shmem_swaplist))
1278 mutex_lock(&shmem_swaplist_mutex);
1279 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1280 if (!info->swapped) {
1281 list_del_init(&info->swaplist);
1285 * Drop the swaplist mutex while searching the inode for swap;
1286 * but before doing so, make sure shmem_evict_inode() will not
1287 * remove placeholder inode from swaplist, nor let it be freed
1288 * (igrab() would protect from unlink, but not from unmount).
1290 atomic_inc(&info->stop_eviction);
1291 mutex_unlock(&shmem_swaplist_mutex);
1293 error = shmem_unuse_inode(&info->vfs_inode, type);
1296 mutex_lock(&shmem_swaplist_mutex);
1297 next = list_next_entry(info, swaplist);
1299 list_del_init(&info->swaplist);
1300 if (atomic_dec_and_test(&info->stop_eviction))
1301 wake_up_var(&info->stop_eviction);
1305 mutex_unlock(&shmem_swaplist_mutex);
1311 * Move the page from the page cache to the swap cache.
1313 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1315 struct shmem_inode_info *info;
1316 struct address_space *mapping;
1317 struct inode *inode;
1322 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1323 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1324 * and its shmem_writeback() needs them to be split when swapping.
1326 if (PageTransCompound(page)) {
1327 /* Ensure the subpages are still dirty */
1329 if (split_huge_page(page) < 0)
1331 ClearPageDirty(page);
1334 BUG_ON(!PageLocked(page));
1335 mapping = page->mapping;
1336 index = page->index;
1337 inode = mapping->host;
1338 info = SHMEM_I(inode);
1339 if (info->flags & VM_LOCKED)
1341 if (!total_swap_pages)
1345 * Our capabilities prevent regular writeback or sync from ever calling
1346 * shmem_writepage; but a stacking filesystem might use ->writepage of
1347 * its underlying filesystem, in which case tmpfs should write out to
1348 * swap only in response to memory pressure, and not for the writeback
1351 if (!wbc->for_reclaim) {
1352 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1357 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1358 * value into swapfile.c, the only way we can correctly account for a
1359 * fallocated page arriving here is now to initialize it and write it.
1361 * That's okay for a page already fallocated earlier, but if we have
1362 * not yet completed the fallocation, then (a) we want to keep track
1363 * of this page in case we have to undo it, and (b) it may not be a
1364 * good idea to continue anyway, once we're pushing into swap. So
1365 * reactivate the page, and let shmem_fallocate() quit when too many.
1367 if (!PageUptodate(page)) {
1368 if (inode->i_private) {
1369 struct shmem_falloc *shmem_falloc;
1370 spin_lock(&inode->i_lock);
1371 shmem_falloc = inode->i_private;
1373 !shmem_falloc->waitq &&
1374 index >= shmem_falloc->start &&
1375 index < shmem_falloc->next)
1376 shmem_falloc->nr_unswapped++;
1378 shmem_falloc = NULL;
1379 spin_unlock(&inode->i_lock);
1383 clear_highpage(page);
1384 flush_dcache_page(page);
1385 SetPageUptodate(page);
1388 swap = get_swap_page(page);
1393 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1394 * if it's not already there. Do it now before the page is
1395 * moved to swap cache, when its pagelock no longer protects
1396 * the inode from eviction. But don't unlock the mutex until
1397 * we've incremented swapped, because shmem_unuse_inode() will
1398 * prune a !swapped inode from the swaplist under this mutex.
1400 mutex_lock(&shmem_swaplist_mutex);
1401 if (list_empty(&info->swaplist))
1402 list_add(&info->swaplist, &shmem_swaplist);
1404 if (add_to_swap_cache(page, swap,
1405 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1407 spin_lock_irq(&info->lock);
1408 shmem_recalc_inode(inode);
1410 spin_unlock_irq(&info->lock);
1412 swap_shmem_alloc(swap);
1413 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1415 mutex_unlock(&shmem_swaplist_mutex);
1416 BUG_ON(page_mapped(page));
1417 swap_writepage(page, wbc);
1421 mutex_unlock(&shmem_swaplist_mutex);
1422 put_swap_page(page, swap);
1424 set_page_dirty(page);
1425 if (wbc->for_reclaim)
1426 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1431 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1432 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1436 if (!mpol || mpol->mode == MPOL_DEFAULT)
1437 return; /* show nothing */
1439 mpol_to_str(buffer, sizeof(buffer), mpol);
1441 seq_printf(seq, ",mpol=%s", buffer);
1444 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1446 struct mempolicy *mpol = NULL;
1448 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1449 mpol = sbinfo->mpol;
1451 raw_spin_unlock(&sbinfo->stat_lock);
1455 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1456 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1459 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1463 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1465 #define vm_policy vm_private_data
1468 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1469 struct shmem_inode_info *info, pgoff_t index)
1471 /* Create a pseudo vma that just contains the policy */
1472 vma_init(vma, NULL);
1473 /* Bias interleave by inode number to distribute better across nodes */
1474 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1475 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1478 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1480 /* Drop reference taken by mpol_shared_policy_lookup() */
1481 mpol_cond_put(vma->vm_policy);
1484 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1485 struct shmem_inode_info *info, pgoff_t index)
1487 struct vm_area_struct pvma;
1489 struct vm_fault vmf = {
1493 shmem_pseudo_vma_init(&pvma, info, index);
1494 page = swap_cluster_readahead(swap, gfp, &vmf);
1495 shmem_pseudo_vma_destroy(&pvma);
1501 * Make sure huge_gfp is always more limited than limit_gfp.
1502 * Some of the flags set permissions, while others set limitations.
1504 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1506 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1507 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1508 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1509 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1511 /* Allow allocations only from the originally specified zones. */
1512 result |= zoneflags;
1515 * Minimize the result gfp by taking the union with the deny flags,
1516 * and the intersection of the allow flags.
1518 result |= (limit_gfp & denyflags);
1519 result |= (huge_gfp & limit_gfp) & allowflags;
1524 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1525 struct shmem_inode_info *info, pgoff_t index)
1527 struct vm_area_struct pvma;
1528 struct address_space *mapping = info->vfs_inode.i_mapping;
1532 hindex = round_down(index, HPAGE_PMD_NR);
1533 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1537 shmem_pseudo_vma_init(&pvma, info, hindex);
1538 page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1539 shmem_pseudo_vma_destroy(&pvma);
1541 prep_transhuge_page(page);
1543 count_vm_event(THP_FILE_FALLBACK);
1547 static struct page *shmem_alloc_page(gfp_t gfp,
1548 struct shmem_inode_info *info, pgoff_t index)
1550 struct vm_area_struct pvma;
1553 shmem_pseudo_vma_init(&pvma, info, index);
1554 page = alloc_page_vma(gfp, &pvma, 0);
1555 shmem_pseudo_vma_destroy(&pvma);
1560 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1561 struct inode *inode,
1562 pgoff_t index, bool huge)
1564 struct shmem_inode_info *info = SHMEM_I(inode);
1569 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1571 nr = huge ? HPAGE_PMD_NR : 1;
1573 if (!shmem_inode_acct_block(inode, nr))
1577 page = shmem_alloc_hugepage(gfp, info, index);
1579 page = shmem_alloc_page(gfp, info, index);
1581 __SetPageLocked(page);
1582 __SetPageSwapBacked(page);
1587 shmem_inode_unacct_blocks(inode, nr);
1589 return ERR_PTR(err);
1593 * When a page is moved from swapcache to shmem filecache (either by the
1594 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1595 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1596 * ignorance of the mapping it belongs to. If that mapping has special
1597 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1598 * we may need to copy to a suitable page before moving to filecache.
1600 * In a future release, this may well be extended to respect cpuset and
1601 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1602 * but for now it is a simple matter of zone.
1604 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1606 return page_zonenum(page) > gfp_zone(gfp);
1609 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1610 struct shmem_inode_info *info, pgoff_t index)
1612 struct page *oldpage, *newpage;
1613 struct folio *old, *new;
1614 struct address_space *swap_mapping;
1620 entry.val = page_private(oldpage);
1621 swap_index = swp_offset(entry);
1622 swap_mapping = page_mapping(oldpage);
1625 * We have arrived here because our zones are constrained, so don't
1626 * limit chance of success by further cpuset and node constraints.
1628 gfp &= ~GFP_CONSTRAINT_MASK;
1629 newpage = shmem_alloc_page(gfp, info, index);
1634 copy_highpage(newpage, oldpage);
1635 flush_dcache_page(newpage);
1637 __SetPageLocked(newpage);
1638 __SetPageSwapBacked(newpage);
1639 SetPageUptodate(newpage);
1640 set_page_private(newpage, entry.val);
1641 SetPageSwapCache(newpage);
1644 * Our caller will very soon move newpage out of swapcache, but it's
1645 * a nice clean interface for us to replace oldpage by newpage there.
1647 xa_lock_irq(&swap_mapping->i_pages);
1648 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1650 old = page_folio(oldpage);
1651 new = page_folio(newpage);
1652 mem_cgroup_migrate(old, new);
1653 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1654 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1656 xa_unlock_irq(&swap_mapping->i_pages);
1658 if (unlikely(error)) {
1660 * Is this possible? I think not, now that our callers check
1661 * both PageSwapCache and page_private after getting page lock;
1662 * but be defensive. Reverse old to newpage for clear and free.
1666 lru_cache_add(newpage);
1670 ClearPageSwapCache(oldpage);
1671 set_page_private(oldpage, 0);
1673 unlock_page(oldpage);
1680 * Swap in the page pointed to by *pagep.
1681 * Caller has to make sure that *pagep contains a valid swapped page.
1682 * Returns 0 and the page in pagep if success. On failure, returns the
1683 * error code and NULL in *pagep.
1685 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1686 struct page **pagep, enum sgp_type sgp,
1687 gfp_t gfp, struct vm_area_struct *vma,
1688 vm_fault_t *fault_type)
1690 struct address_space *mapping = inode->i_mapping;
1691 struct shmem_inode_info *info = SHMEM_I(inode);
1692 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1697 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1698 swap = radix_to_swp_entry(*pagep);
1701 /* Look it up and read it in.. */
1702 page = lookup_swap_cache(swap, NULL, 0);
1704 /* Or update major stats only when swapin succeeds?? */
1706 *fault_type |= VM_FAULT_MAJOR;
1707 count_vm_event(PGMAJFAULT);
1708 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1710 /* Here we actually start the io */
1711 page = shmem_swapin(swap, gfp, info, index);
1718 /* We have to do this with page locked to prevent races */
1720 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1721 !shmem_confirm_swap(mapping, index, swap)) {
1725 if (!PageUptodate(page)) {
1729 wait_on_page_writeback(page);
1732 * Some architectures may have to restore extra metadata to the
1733 * physical page after reading from swap.
1735 arch_swap_restore(swap, page);
1737 if (shmem_should_replace_page(page, gfp)) {
1738 error = shmem_replace_page(&page, gfp, info, index);
1743 error = shmem_add_to_page_cache(page, mapping, index,
1744 swp_to_radix_entry(swap), gfp,
1749 spin_lock_irq(&info->lock);
1751 shmem_recalc_inode(inode);
1752 spin_unlock_irq(&info->lock);
1754 if (sgp == SGP_WRITE)
1755 mark_page_accessed(page);
1757 delete_from_swap_cache(page);
1758 set_page_dirty(page);
1764 if (!shmem_confirm_swap(mapping, index, swap))
1776 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1778 * If we allocate a new one we do not mark it dirty. That's up to the
1779 * vm. If we swap it in we mark it dirty since we also free the swap
1780 * entry since a page cannot live in both the swap and page cache.
1782 * vma, vmf, and fault_type are only supplied by shmem_fault:
1783 * otherwise they are NULL.
1785 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1786 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1787 struct vm_area_struct *vma, struct vm_fault *vmf,
1788 vm_fault_t *fault_type)
1790 struct address_space *mapping = inode->i_mapping;
1791 struct shmem_inode_info *info = SHMEM_I(inode);
1792 struct shmem_sb_info *sbinfo;
1793 struct mm_struct *charge_mm;
1795 pgoff_t hindex = index;
1801 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1804 if (sgp <= SGP_CACHE &&
1805 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1809 sbinfo = SHMEM_SB(inode->i_sb);
1810 charge_mm = vma ? vma->vm_mm : NULL;
1812 page = pagecache_get_page(mapping, index,
1813 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1815 if (page && vma && userfaultfd_minor(vma)) {
1816 if (!xa_is_value(page)) {
1820 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1824 if (xa_is_value(page)) {
1825 error = shmem_swapin_page(inode, index, &page,
1826 sgp, gfp, vma, fault_type);
1827 if (error == -EEXIST)
1835 hindex = page->index;
1836 if (sgp == SGP_WRITE)
1837 mark_page_accessed(page);
1838 if (PageUptodate(page))
1840 /* fallocated page */
1841 if (sgp != SGP_READ)
1848 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1849 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1852 if (sgp == SGP_READ)
1854 if (sgp == SGP_NOALLOC)
1858 * Fast cache lookup and swap lookup did not find it: allocate.
1861 if (vma && userfaultfd_missing(vma)) {
1862 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1866 if (!shmem_is_huge(vma, inode, index))
1869 huge_gfp = vma_thp_gfp_mask(vma);
1870 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1871 page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true);
1874 page = shmem_alloc_and_acct_page(gfp, inode,
1880 error = PTR_ERR(page);
1882 if (error != -ENOSPC)
1885 * Try to reclaim some space by splitting a huge page
1886 * beyond i_size on the filesystem.
1891 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1892 if (ret == SHRINK_STOP)
1900 if (PageTransHuge(page))
1901 hindex = round_down(index, HPAGE_PMD_NR);
1905 if (sgp == SGP_WRITE)
1906 __SetPageReferenced(page);
1908 error = shmem_add_to_page_cache(page, mapping, hindex,
1909 NULL, gfp & GFP_RECLAIM_MASK,
1913 lru_cache_add(page);
1915 spin_lock_irq(&info->lock);
1916 info->alloced += compound_nr(page);
1917 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1918 shmem_recalc_inode(inode);
1919 spin_unlock_irq(&info->lock);
1922 if (PageTransHuge(page) &&
1923 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1924 hindex + HPAGE_PMD_NR - 1) {
1926 * Part of the huge page is beyond i_size: subject
1927 * to shrink under memory pressure.
1929 spin_lock(&sbinfo->shrinklist_lock);
1931 * _careful to defend against unlocked access to
1932 * ->shrink_list in shmem_unused_huge_shrink()
1934 if (list_empty_careful(&info->shrinklist)) {
1935 list_add_tail(&info->shrinklist,
1936 &sbinfo->shrinklist);
1937 sbinfo->shrinklist_len++;
1939 spin_unlock(&sbinfo->shrinklist_lock);
1943 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1945 if (sgp == SGP_FALLOC)
1949 * Let SGP_WRITE caller clear ends if write does not fill page;
1950 * but SGP_FALLOC on a page fallocated earlier must initialize
1951 * it now, lest undo on failure cancel our earlier guarantee.
1953 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1956 for (i = 0; i < compound_nr(page); i++) {
1957 clear_highpage(page + i);
1958 flush_dcache_page(page + i);
1960 SetPageUptodate(page);
1963 /* Perhaps the file has been truncated since we checked */
1964 if (sgp <= SGP_CACHE &&
1965 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1967 ClearPageDirty(page);
1968 delete_from_page_cache(page);
1969 spin_lock_irq(&info->lock);
1970 shmem_recalc_inode(inode);
1971 spin_unlock_irq(&info->lock);
1977 *pagep = page + index - hindex;
1984 shmem_inode_unacct_blocks(inode, compound_nr(page));
1986 if (PageTransHuge(page)) {
1996 if (error == -ENOSPC && !once++) {
1997 spin_lock_irq(&info->lock);
1998 shmem_recalc_inode(inode);
1999 spin_unlock_irq(&info->lock);
2002 if (error == -EEXIST)
2008 * This is like autoremove_wake_function, but it removes the wait queue
2009 * entry unconditionally - even if something else had already woken the
2012 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2014 int ret = default_wake_function(wait, mode, sync, key);
2015 list_del_init(&wait->entry);
2019 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2021 struct vm_area_struct *vma = vmf->vma;
2022 struct inode *inode = file_inode(vma->vm_file);
2023 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2025 vm_fault_t ret = VM_FAULT_LOCKED;
2028 * Trinity finds that probing a hole which tmpfs is punching can
2029 * prevent the hole-punch from ever completing: which in turn
2030 * locks writers out with its hold on i_rwsem. So refrain from
2031 * faulting pages into the hole while it's being punched. Although
2032 * shmem_undo_range() does remove the additions, it may be unable to
2033 * keep up, as each new page needs its own unmap_mapping_range() call,
2034 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2036 * It does not matter if we sometimes reach this check just before the
2037 * hole-punch begins, so that one fault then races with the punch:
2038 * we just need to make racing faults a rare case.
2040 * The implementation below would be much simpler if we just used a
2041 * standard mutex or completion: but we cannot take i_rwsem in fault,
2042 * and bloating every shmem inode for this unlikely case would be sad.
2044 if (unlikely(inode->i_private)) {
2045 struct shmem_falloc *shmem_falloc;
2047 spin_lock(&inode->i_lock);
2048 shmem_falloc = inode->i_private;
2050 shmem_falloc->waitq &&
2051 vmf->pgoff >= shmem_falloc->start &&
2052 vmf->pgoff < shmem_falloc->next) {
2054 wait_queue_head_t *shmem_falloc_waitq;
2055 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2057 ret = VM_FAULT_NOPAGE;
2058 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2060 ret = VM_FAULT_RETRY;
2062 shmem_falloc_waitq = shmem_falloc->waitq;
2063 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2064 TASK_UNINTERRUPTIBLE);
2065 spin_unlock(&inode->i_lock);
2069 * shmem_falloc_waitq points into the shmem_fallocate()
2070 * stack of the hole-punching task: shmem_falloc_waitq
2071 * is usually invalid by the time we reach here, but
2072 * finish_wait() does not dereference it in that case;
2073 * though i_lock needed lest racing with wake_up_all().
2075 spin_lock(&inode->i_lock);
2076 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2077 spin_unlock(&inode->i_lock);
2083 spin_unlock(&inode->i_lock);
2086 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE,
2087 gfp, vma, vmf, &ret);
2089 return vmf_error(err);
2093 unsigned long shmem_get_unmapped_area(struct file *file,
2094 unsigned long uaddr, unsigned long len,
2095 unsigned long pgoff, unsigned long flags)
2097 unsigned long (*get_area)(struct file *,
2098 unsigned long, unsigned long, unsigned long, unsigned long);
2100 unsigned long offset;
2101 unsigned long inflated_len;
2102 unsigned long inflated_addr;
2103 unsigned long inflated_offset;
2105 if (len > TASK_SIZE)
2108 get_area = current->mm->get_unmapped_area;
2109 addr = get_area(file, uaddr, len, pgoff, flags);
2111 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2113 if (IS_ERR_VALUE(addr))
2115 if (addr & ~PAGE_MASK)
2117 if (addr > TASK_SIZE - len)
2120 if (shmem_huge == SHMEM_HUGE_DENY)
2122 if (len < HPAGE_PMD_SIZE)
2124 if (flags & MAP_FIXED)
2127 * Our priority is to support MAP_SHARED mapped hugely;
2128 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2129 * But if caller specified an address hint and we allocated area there
2130 * successfully, respect that as before.
2135 if (shmem_huge != SHMEM_HUGE_FORCE) {
2136 struct super_block *sb;
2139 VM_BUG_ON(file->f_op != &shmem_file_operations);
2140 sb = file_inode(file)->i_sb;
2143 * Called directly from mm/mmap.c, or drivers/char/mem.c
2144 * for "/dev/zero", to create a shared anonymous object.
2146 if (IS_ERR(shm_mnt))
2148 sb = shm_mnt->mnt_sb;
2150 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2154 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2155 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2157 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2160 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2161 if (inflated_len > TASK_SIZE)
2163 if (inflated_len < len)
2166 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2167 if (IS_ERR_VALUE(inflated_addr))
2169 if (inflated_addr & ~PAGE_MASK)
2172 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2173 inflated_addr += offset - inflated_offset;
2174 if (inflated_offset > offset)
2175 inflated_addr += HPAGE_PMD_SIZE;
2177 if (inflated_addr > TASK_SIZE - len)
2179 return inflated_addr;
2183 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2185 struct inode *inode = file_inode(vma->vm_file);
2186 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2189 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2192 struct inode *inode = file_inode(vma->vm_file);
2195 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2196 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2200 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2202 struct inode *inode = file_inode(file);
2203 struct shmem_inode_info *info = SHMEM_I(inode);
2204 int retval = -ENOMEM;
2207 * What serializes the accesses to info->flags?
2208 * ipc_lock_object() when called from shmctl_do_lock(),
2209 * no serialization needed when called from shm_destroy().
2211 if (lock && !(info->flags & VM_LOCKED)) {
2212 if (!user_shm_lock(inode->i_size, ucounts))
2214 info->flags |= VM_LOCKED;
2215 mapping_set_unevictable(file->f_mapping);
2217 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2218 user_shm_unlock(inode->i_size, ucounts);
2219 info->flags &= ~VM_LOCKED;
2220 mapping_clear_unevictable(file->f_mapping);
2228 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2230 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2233 ret = seal_check_future_write(info->seals, vma);
2237 /* arm64 - allow memory tagging on RAM-based files */
2238 vma->vm_flags |= VM_MTE_ALLOWED;
2240 file_accessed(file);
2241 vma->vm_ops = &shmem_vm_ops;
2242 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2243 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2244 (vma->vm_end & HPAGE_PMD_MASK)) {
2245 khugepaged_enter(vma, vma->vm_flags);
2250 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2251 umode_t mode, dev_t dev, unsigned long flags)
2253 struct inode *inode;
2254 struct shmem_inode_info *info;
2255 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2258 if (shmem_reserve_inode(sb, &ino))
2261 inode = new_inode(sb);
2264 inode_init_owner(&init_user_ns, inode, dir, mode);
2265 inode->i_blocks = 0;
2266 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2267 inode->i_generation = prandom_u32();
2268 info = SHMEM_I(inode);
2269 memset(info, 0, (char *)inode - (char *)info);
2270 spin_lock_init(&info->lock);
2271 atomic_set(&info->stop_eviction, 0);
2272 info->seals = F_SEAL_SEAL;
2273 info->flags = flags & VM_NORESERVE;
2274 info->i_crtime = inode->i_mtime;
2275 INIT_LIST_HEAD(&info->shrinklist);
2276 INIT_LIST_HEAD(&info->swaplist);
2277 simple_xattrs_init(&info->xattrs);
2278 cache_no_acl(inode);
2279 mapping_set_large_folios(inode->i_mapping);
2281 switch (mode & S_IFMT) {
2283 inode->i_op = &shmem_special_inode_operations;
2284 init_special_inode(inode, mode, dev);
2287 inode->i_mapping->a_ops = &shmem_aops;
2288 inode->i_op = &shmem_inode_operations;
2289 inode->i_fop = &shmem_file_operations;
2290 mpol_shared_policy_init(&info->policy,
2291 shmem_get_sbmpol(sbinfo));
2295 /* Some things misbehave if size == 0 on a directory */
2296 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2297 inode->i_op = &shmem_dir_inode_operations;
2298 inode->i_fop = &simple_dir_operations;
2302 * Must not load anything in the rbtree,
2303 * mpol_free_shared_policy will not be called.
2305 mpol_shared_policy_init(&info->policy, NULL);
2309 lockdep_annotate_inode_mutex_key(inode);
2311 shmem_free_inode(sb);
2315 #ifdef CONFIG_USERFAULTFD
2316 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2318 struct vm_area_struct *dst_vma,
2319 unsigned long dst_addr,
2320 unsigned long src_addr,
2322 struct page **pagep)
2324 struct inode *inode = file_inode(dst_vma->vm_file);
2325 struct shmem_inode_info *info = SHMEM_I(inode);
2326 struct address_space *mapping = inode->i_mapping;
2327 gfp_t gfp = mapping_gfp_mask(mapping);
2328 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2334 if (!shmem_inode_acct_block(inode, 1)) {
2336 * We may have got a page, returned -ENOENT triggering a retry,
2337 * and now we find ourselves with -ENOMEM. Release the page, to
2338 * avoid a BUG_ON in our caller.
2340 if (unlikely(*pagep)) {
2349 page = shmem_alloc_page(gfp, info, pgoff);
2351 goto out_unacct_blocks;
2353 if (!zeropage) { /* COPY */
2354 page_kaddr = kmap_atomic(page);
2355 ret = copy_from_user(page_kaddr,
2356 (const void __user *)src_addr,
2358 kunmap_atomic(page_kaddr);
2360 /* fallback to copy_from_user outside mmap_lock */
2361 if (unlikely(ret)) {
2364 /* don't free the page */
2365 goto out_unacct_blocks;
2368 flush_dcache_page(page);
2369 } else { /* ZEROPAGE */
2370 clear_user_highpage(page, dst_addr);
2377 VM_BUG_ON(PageLocked(page));
2378 VM_BUG_ON(PageSwapBacked(page));
2379 __SetPageLocked(page);
2380 __SetPageSwapBacked(page);
2381 __SetPageUptodate(page);
2384 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2385 if (unlikely(pgoff >= max_off))
2388 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2389 gfp & GFP_RECLAIM_MASK, dst_mm);
2393 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2396 goto out_delete_from_cache;
2398 spin_lock_irq(&info->lock);
2400 inode->i_blocks += BLOCKS_PER_PAGE;
2401 shmem_recalc_inode(inode);
2402 spin_unlock_irq(&info->lock);
2406 out_delete_from_cache:
2407 delete_from_page_cache(page);
2412 shmem_inode_unacct_blocks(inode, 1);
2415 #endif /* CONFIG_USERFAULTFD */
2418 static const struct inode_operations shmem_symlink_inode_operations;
2419 static const struct inode_operations shmem_short_symlink_operations;
2421 #ifdef CONFIG_TMPFS_XATTR
2422 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2424 #define shmem_initxattrs NULL
2428 shmem_write_begin(struct file *file, struct address_space *mapping,
2429 loff_t pos, unsigned len, unsigned flags,
2430 struct page **pagep, void **fsdata)
2432 struct inode *inode = mapping->host;
2433 struct shmem_inode_info *info = SHMEM_I(inode);
2434 pgoff_t index = pos >> PAGE_SHIFT;
2437 /* i_rwsem is held by caller */
2438 if (unlikely(info->seals & (F_SEAL_GROW |
2439 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2440 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2442 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2446 ret = shmem_getpage(inode, index, pagep, SGP_WRITE);
2451 if (PageHWPoison(*pagep)) {
2452 unlock_page(*pagep);
2462 shmem_write_end(struct file *file, struct address_space *mapping,
2463 loff_t pos, unsigned len, unsigned copied,
2464 struct page *page, void *fsdata)
2466 struct inode *inode = mapping->host;
2468 if (pos + copied > inode->i_size)
2469 i_size_write(inode, pos + copied);
2471 if (!PageUptodate(page)) {
2472 struct page *head = compound_head(page);
2473 if (PageTransCompound(page)) {
2476 for (i = 0; i < HPAGE_PMD_NR; i++) {
2477 if (head + i == page)
2479 clear_highpage(head + i);
2480 flush_dcache_page(head + i);
2483 if (copied < PAGE_SIZE) {
2484 unsigned from = pos & (PAGE_SIZE - 1);
2485 zero_user_segments(page, 0, from,
2486 from + copied, PAGE_SIZE);
2488 SetPageUptodate(head);
2490 set_page_dirty(page);
2497 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2499 struct file *file = iocb->ki_filp;
2500 struct inode *inode = file_inode(file);
2501 struct address_space *mapping = inode->i_mapping;
2503 unsigned long offset;
2506 loff_t *ppos = &iocb->ki_pos;
2508 index = *ppos >> PAGE_SHIFT;
2509 offset = *ppos & ~PAGE_MASK;
2512 struct page *page = NULL;
2514 unsigned long nr, ret;
2515 loff_t i_size = i_size_read(inode);
2517 end_index = i_size >> PAGE_SHIFT;
2518 if (index > end_index)
2520 if (index == end_index) {
2521 nr = i_size & ~PAGE_MASK;
2526 error = shmem_getpage(inode, index, &page, SGP_READ);
2528 if (error == -EINVAL)
2535 if (PageHWPoison(page)) {
2543 * We must evaluate after, since reads (unlike writes)
2544 * are called without i_rwsem protection against truncate
2547 i_size = i_size_read(inode);
2548 end_index = i_size >> PAGE_SHIFT;
2549 if (index == end_index) {
2550 nr = i_size & ~PAGE_MASK;
2561 * If users can be writing to this page using arbitrary
2562 * virtual addresses, take care about potential aliasing
2563 * before reading the page on the kernel side.
2565 if (mapping_writably_mapped(mapping))
2566 flush_dcache_page(page);
2568 * Mark the page accessed if we read the beginning.
2571 mark_page_accessed(page);
2573 * Ok, we have the page, and it's up-to-date, so
2574 * now we can copy it to user space...
2576 ret = copy_page_to_iter(page, offset, nr, to);
2579 } else if (iter_is_iovec(to)) {
2581 * Copy to user tends to be so well optimized, but
2582 * clear_user() not so much, that it is noticeably
2583 * faster to copy the zero page instead of clearing.
2585 ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2588 * But submitting the same page twice in a row to
2589 * splice() - or others? - can result in confusion:
2590 * so don't attempt that optimization on pipes etc.
2592 ret = iov_iter_zero(nr, to);
2597 index += offset >> PAGE_SHIFT;
2598 offset &= ~PAGE_MASK;
2600 if (!iov_iter_count(to))
2609 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2610 file_accessed(file);
2611 return retval ? retval : error;
2614 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2616 struct address_space *mapping = file->f_mapping;
2617 struct inode *inode = mapping->host;
2619 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2620 return generic_file_llseek_size(file, offset, whence,
2621 MAX_LFS_FILESIZE, i_size_read(inode));
2626 /* We're holding i_rwsem so we can access i_size directly */
2627 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2629 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2630 inode_unlock(inode);
2634 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2637 struct inode *inode = file_inode(file);
2638 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2639 struct shmem_inode_info *info = SHMEM_I(inode);
2640 struct shmem_falloc shmem_falloc;
2641 pgoff_t start, index, end, undo_fallocend;
2644 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2649 if (mode & FALLOC_FL_PUNCH_HOLE) {
2650 struct address_space *mapping = file->f_mapping;
2651 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2652 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2653 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2655 /* protected by i_rwsem */
2656 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2661 shmem_falloc.waitq = &shmem_falloc_waitq;
2662 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2663 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2664 spin_lock(&inode->i_lock);
2665 inode->i_private = &shmem_falloc;
2666 spin_unlock(&inode->i_lock);
2668 if ((u64)unmap_end > (u64)unmap_start)
2669 unmap_mapping_range(mapping, unmap_start,
2670 1 + unmap_end - unmap_start, 0);
2671 shmem_truncate_range(inode, offset, offset + len - 1);
2672 /* No need to unmap again: hole-punching leaves COWed pages */
2674 spin_lock(&inode->i_lock);
2675 inode->i_private = NULL;
2676 wake_up_all(&shmem_falloc_waitq);
2677 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2678 spin_unlock(&inode->i_lock);
2683 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2684 error = inode_newsize_ok(inode, offset + len);
2688 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2693 start = offset >> PAGE_SHIFT;
2694 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2695 /* Try to avoid a swapstorm if len is impossible to satisfy */
2696 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2701 shmem_falloc.waitq = NULL;
2702 shmem_falloc.start = start;
2703 shmem_falloc.next = start;
2704 shmem_falloc.nr_falloced = 0;
2705 shmem_falloc.nr_unswapped = 0;
2706 spin_lock(&inode->i_lock);
2707 inode->i_private = &shmem_falloc;
2708 spin_unlock(&inode->i_lock);
2711 * info->fallocend is only relevant when huge pages might be
2712 * involved: to prevent split_huge_page() freeing fallocated
2713 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2715 undo_fallocend = info->fallocend;
2716 if (info->fallocend < end)
2717 info->fallocend = end;
2719 for (index = start; index < end; ) {
2723 * Good, the fallocate(2) manpage permits EINTR: we may have
2724 * been interrupted because we are using up too much memory.
2726 if (signal_pending(current))
2728 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2731 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2733 info->fallocend = undo_fallocend;
2734 /* Remove the !PageUptodate pages we added */
2735 if (index > start) {
2736 shmem_undo_range(inode,
2737 (loff_t)start << PAGE_SHIFT,
2738 ((loff_t)index << PAGE_SHIFT) - 1, true);
2745 * Here is a more important optimization than it appears:
2746 * a second SGP_FALLOC on the same huge page will clear it,
2747 * making it PageUptodate and un-undoable if we fail later.
2749 if (PageTransCompound(page)) {
2750 index = round_up(index, HPAGE_PMD_NR);
2751 /* Beware 32-bit wraparound */
2757 * Inform shmem_writepage() how far we have reached.
2758 * No need for lock or barrier: we have the page lock.
2760 if (!PageUptodate(page))
2761 shmem_falloc.nr_falloced += index - shmem_falloc.next;
2762 shmem_falloc.next = index;
2765 * If !PageUptodate, leave it that way so that freeable pages
2766 * can be recognized if we need to rollback on error later.
2767 * But set_page_dirty so that memory pressure will swap rather
2768 * than free the pages we are allocating (and SGP_CACHE pages
2769 * might still be clean: we now need to mark those dirty too).
2771 set_page_dirty(page);
2777 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2778 i_size_write(inode, offset + len);
2779 inode->i_ctime = current_time(inode);
2781 spin_lock(&inode->i_lock);
2782 inode->i_private = NULL;
2783 spin_unlock(&inode->i_lock);
2785 inode_unlock(inode);
2789 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2791 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2793 buf->f_type = TMPFS_MAGIC;
2794 buf->f_bsize = PAGE_SIZE;
2795 buf->f_namelen = NAME_MAX;
2796 if (sbinfo->max_blocks) {
2797 buf->f_blocks = sbinfo->max_blocks;
2799 buf->f_bfree = sbinfo->max_blocks -
2800 percpu_counter_sum(&sbinfo->used_blocks);
2802 if (sbinfo->max_inodes) {
2803 buf->f_files = sbinfo->max_inodes;
2804 buf->f_ffree = sbinfo->free_inodes;
2806 /* else leave those fields 0 like simple_statfs */
2808 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2814 * File creation. Allocate an inode, and we're done..
2817 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2818 struct dentry *dentry, umode_t mode, dev_t dev)
2820 struct inode *inode;
2821 int error = -ENOSPC;
2823 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2825 error = simple_acl_create(dir, inode);
2828 error = security_inode_init_security(inode, dir,
2830 shmem_initxattrs, NULL);
2831 if (error && error != -EOPNOTSUPP)
2835 dir->i_size += BOGO_DIRENT_SIZE;
2836 dir->i_ctime = dir->i_mtime = current_time(dir);
2837 d_instantiate(dentry, inode);
2838 dget(dentry); /* Extra count - pin the dentry in core */
2847 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2848 struct dentry *dentry, umode_t mode)
2850 struct inode *inode;
2851 int error = -ENOSPC;
2853 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2855 error = security_inode_init_security(inode, dir,
2857 shmem_initxattrs, NULL);
2858 if (error && error != -EOPNOTSUPP)
2860 error = simple_acl_create(dir, inode);
2863 d_tmpfile(dentry, inode);
2871 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2872 struct dentry *dentry, umode_t mode)
2876 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2877 mode | S_IFDIR, 0)))
2883 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2884 struct dentry *dentry, umode_t mode, bool excl)
2886 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2892 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2894 struct inode *inode = d_inode(old_dentry);
2898 * No ordinary (disk based) filesystem counts links as inodes;
2899 * but each new link needs a new dentry, pinning lowmem, and
2900 * tmpfs dentries cannot be pruned until they are unlinked.
2901 * But if an O_TMPFILE file is linked into the tmpfs, the
2902 * first link must skip that, to get the accounting right.
2904 if (inode->i_nlink) {
2905 ret = shmem_reserve_inode(inode->i_sb, NULL);
2910 dir->i_size += BOGO_DIRENT_SIZE;
2911 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2913 ihold(inode); /* New dentry reference */
2914 dget(dentry); /* Extra pinning count for the created dentry */
2915 d_instantiate(dentry, inode);
2920 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2922 struct inode *inode = d_inode(dentry);
2924 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2925 shmem_free_inode(inode->i_sb);
2927 dir->i_size -= BOGO_DIRENT_SIZE;
2928 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2930 dput(dentry); /* Undo the count from "create" - this does all the work */
2934 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2936 if (!simple_empty(dentry))
2939 drop_nlink(d_inode(dentry));
2941 return shmem_unlink(dir, dentry);
2944 static int shmem_whiteout(struct user_namespace *mnt_userns,
2945 struct inode *old_dir, struct dentry *old_dentry)
2947 struct dentry *whiteout;
2950 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2954 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2955 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2961 * Cheat and hash the whiteout while the old dentry is still in
2962 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2964 * d_lookup() will consistently find one of them at this point,
2965 * not sure which one, but that isn't even important.
2972 * The VFS layer already does all the dentry stuff for rename,
2973 * we just have to decrement the usage count for the target if
2974 * it exists so that the VFS layer correctly free's it when it
2977 static int shmem_rename2(struct user_namespace *mnt_userns,
2978 struct inode *old_dir, struct dentry *old_dentry,
2979 struct inode *new_dir, struct dentry *new_dentry,
2982 struct inode *inode = d_inode(old_dentry);
2983 int they_are_dirs = S_ISDIR(inode->i_mode);
2985 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2988 if (flags & RENAME_EXCHANGE)
2989 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
2991 if (!simple_empty(new_dentry))
2994 if (flags & RENAME_WHITEOUT) {
2997 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3002 if (d_really_is_positive(new_dentry)) {
3003 (void) shmem_unlink(new_dir, new_dentry);
3004 if (they_are_dirs) {
3005 drop_nlink(d_inode(new_dentry));
3006 drop_nlink(old_dir);
3008 } else if (they_are_dirs) {
3009 drop_nlink(old_dir);
3013 old_dir->i_size -= BOGO_DIRENT_SIZE;
3014 new_dir->i_size += BOGO_DIRENT_SIZE;
3015 old_dir->i_ctime = old_dir->i_mtime =
3016 new_dir->i_ctime = new_dir->i_mtime =
3017 inode->i_ctime = current_time(old_dir);
3021 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3022 struct dentry *dentry, const char *symname)
3026 struct inode *inode;
3029 len = strlen(symname) + 1;
3030 if (len > PAGE_SIZE)
3031 return -ENAMETOOLONG;
3033 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3038 error = security_inode_init_security(inode, dir, &dentry->d_name,
3039 shmem_initxattrs, NULL);
3040 if (error && error != -EOPNOTSUPP) {
3045 inode->i_size = len-1;
3046 if (len <= SHORT_SYMLINK_LEN) {
3047 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3048 if (!inode->i_link) {
3052 inode->i_op = &shmem_short_symlink_operations;
3054 inode_nohighmem(inode);
3055 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3060 inode->i_mapping->a_ops = &shmem_aops;
3061 inode->i_op = &shmem_symlink_inode_operations;
3062 memcpy(page_address(page), symname, len);
3063 SetPageUptodate(page);
3064 set_page_dirty(page);
3068 dir->i_size += BOGO_DIRENT_SIZE;
3069 dir->i_ctime = dir->i_mtime = current_time(dir);
3070 d_instantiate(dentry, inode);
3075 static void shmem_put_link(void *arg)
3077 mark_page_accessed(arg);
3081 static const char *shmem_get_link(struct dentry *dentry,
3082 struct inode *inode,
3083 struct delayed_call *done)
3085 struct page *page = NULL;
3088 page = find_get_page(inode->i_mapping, 0);
3090 return ERR_PTR(-ECHILD);
3091 if (PageHWPoison(page) ||
3092 !PageUptodate(page)) {
3094 return ERR_PTR(-ECHILD);
3097 error = shmem_getpage(inode, 0, &page, SGP_READ);
3099 return ERR_PTR(error);
3101 return ERR_PTR(-ECHILD);
3102 if (PageHWPoison(page)) {
3105 return ERR_PTR(-ECHILD);
3109 set_delayed_call(done, shmem_put_link, page);
3110 return page_address(page);
3113 #ifdef CONFIG_TMPFS_XATTR
3115 * Superblocks without xattr inode operations may get some security.* xattr
3116 * support from the LSM "for free". As soon as we have any other xattrs
3117 * like ACLs, we also need to implement the security.* handlers at
3118 * filesystem level, though.
3122 * Callback for security_inode_init_security() for acquiring xattrs.
3124 static int shmem_initxattrs(struct inode *inode,
3125 const struct xattr *xattr_array,
3128 struct shmem_inode_info *info = SHMEM_I(inode);
3129 const struct xattr *xattr;
3130 struct simple_xattr *new_xattr;
3133 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3134 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3138 len = strlen(xattr->name) + 1;
3139 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3141 if (!new_xattr->name) {
3146 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3147 XATTR_SECURITY_PREFIX_LEN);
3148 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3151 simple_xattr_list_add(&info->xattrs, new_xattr);
3157 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3158 struct dentry *unused, struct inode *inode,
3159 const char *name, void *buffer, size_t size)
3161 struct shmem_inode_info *info = SHMEM_I(inode);
3163 name = xattr_full_name(handler, name);
3164 return simple_xattr_get(&info->xattrs, name, buffer, size);
3167 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3168 struct user_namespace *mnt_userns,
3169 struct dentry *unused, struct inode *inode,
3170 const char *name, const void *value,
3171 size_t size, int flags)
3173 struct shmem_inode_info *info = SHMEM_I(inode);
3175 name = xattr_full_name(handler, name);
3176 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3179 static const struct xattr_handler shmem_security_xattr_handler = {
3180 .prefix = XATTR_SECURITY_PREFIX,
3181 .get = shmem_xattr_handler_get,
3182 .set = shmem_xattr_handler_set,
3185 static const struct xattr_handler shmem_trusted_xattr_handler = {
3186 .prefix = XATTR_TRUSTED_PREFIX,
3187 .get = shmem_xattr_handler_get,
3188 .set = shmem_xattr_handler_set,
3191 static const struct xattr_handler *shmem_xattr_handlers[] = {
3192 #ifdef CONFIG_TMPFS_POSIX_ACL
3193 &posix_acl_access_xattr_handler,
3194 &posix_acl_default_xattr_handler,
3196 &shmem_security_xattr_handler,
3197 &shmem_trusted_xattr_handler,
3201 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3203 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3204 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3206 #endif /* CONFIG_TMPFS_XATTR */
3208 static const struct inode_operations shmem_short_symlink_operations = {
3209 .getattr = shmem_getattr,
3210 .get_link = simple_get_link,
3211 #ifdef CONFIG_TMPFS_XATTR
3212 .listxattr = shmem_listxattr,
3216 static const struct inode_operations shmem_symlink_inode_operations = {
3217 .getattr = shmem_getattr,
3218 .get_link = shmem_get_link,
3219 #ifdef CONFIG_TMPFS_XATTR
3220 .listxattr = shmem_listxattr,
3224 static struct dentry *shmem_get_parent(struct dentry *child)
3226 return ERR_PTR(-ESTALE);
3229 static int shmem_match(struct inode *ino, void *vfh)
3233 inum = (inum << 32) | fh[1];
3234 return ino->i_ino == inum && fh[0] == ino->i_generation;
3237 /* Find any alias of inode, but prefer a hashed alias */
3238 static struct dentry *shmem_find_alias(struct inode *inode)
3240 struct dentry *alias = d_find_alias(inode);
3242 return alias ?: d_find_any_alias(inode);
3246 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3247 struct fid *fid, int fh_len, int fh_type)
3249 struct inode *inode;
3250 struct dentry *dentry = NULL;
3257 inum = (inum << 32) | fid->raw[1];
3259 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3260 shmem_match, fid->raw);
3262 dentry = shmem_find_alias(inode);
3269 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3270 struct inode *parent)
3274 return FILEID_INVALID;
3277 if (inode_unhashed(inode)) {
3278 /* Unfortunately insert_inode_hash is not idempotent,
3279 * so as we hash inodes here rather than at creation
3280 * time, we need a lock to ensure we only try
3283 static DEFINE_SPINLOCK(lock);
3285 if (inode_unhashed(inode))
3286 __insert_inode_hash(inode,
3287 inode->i_ino + inode->i_generation);
3291 fh[0] = inode->i_generation;
3292 fh[1] = inode->i_ino;
3293 fh[2] = ((__u64)inode->i_ino) >> 32;
3299 static const struct export_operations shmem_export_ops = {
3300 .get_parent = shmem_get_parent,
3301 .encode_fh = shmem_encode_fh,
3302 .fh_to_dentry = shmem_fh_to_dentry,
3318 static const struct constant_table shmem_param_enums_huge[] = {
3319 {"never", SHMEM_HUGE_NEVER },
3320 {"always", SHMEM_HUGE_ALWAYS },
3321 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3322 {"advise", SHMEM_HUGE_ADVISE },
3326 const struct fs_parameter_spec shmem_fs_parameters[] = {
3327 fsparam_u32 ("gid", Opt_gid),
3328 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3329 fsparam_u32oct("mode", Opt_mode),
3330 fsparam_string("mpol", Opt_mpol),
3331 fsparam_string("nr_blocks", Opt_nr_blocks),
3332 fsparam_string("nr_inodes", Opt_nr_inodes),
3333 fsparam_string("size", Opt_size),
3334 fsparam_u32 ("uid", Opt_uid),
3335 fsparam_flag ("inode32", Opt_inode32),
3336 fsparam_flag ("inode64", Opt_inode64),
3340 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3342 struct shmem_options *ctx = fc->fs_private;
3343 struct fs_parse_result result;
3344 unsigned long long size;
3348 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3354 size = memparse(param->string, &rest);
3356 size <<= PAGE_SHIFT;
3357 size *= totalram_pages();
3363 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3364 ctx->seen |= SHMEM_SEEN_BLOCKS;
3367 ctx->blocks = memparse(param->string, &rest);
3370 ctx->seen |= SHMEM_SEEN_BLOCKS;
3373 ctx->inodes = memparse(param->string, &rest);
3376 ctx->seen |= SHMEM_SEEN_INODES;
3379 ctx->mode = result.uint_32 & 07777;
3382 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3383 if (!uid_valid(ctx->uid))
3387 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3388 if (!gid_valid(ctx->gid))
3392 ctx->huge = result.uint_32;
3393 if (ctx->huge != SHMEM_HUGE_NEVER &&
3394 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3395 has_transparent_hugepage()))
3396 goto unsupported_parameter;
3397 ctx->seen |= SHMEM_SEEN_HUGE;
3400 if (IS_ENABLED(CONFIG_NUMA)) {
3401 mpol_put(ctx->mpol);
3403 if (mpol_parse_str(param->string, &ctx->mpol))
3407 goto unsupported_parameter;
3409 ctx->full_inums = false;
3410 ctx->seen |= SHMEM_SEEN_INUMS;
3413 if (sizeof(ino_t) < 8) {
3415 "Cannot use inode64 with <64bit inums in kernel\n");
3417 ctx->full_inums = true;
3418 ctx->seen |= SHMEM_SEEN_INUMS;
3423 unsupported_parameter:
3424 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3426 return invalfc(fc, "Bad value for '%s'", param->key);
3429 static int shmem_parse_options(struct fs_context *fc, void *data)
3431 char *options = data;
3434 int err = security_sb_eat_lsm_opts(options, &fc->security);
3439 while (options != NULL) {
3440 char *this_char = options;
3443 * NUL-terminate this option: unfortunately,
3444 * mount options form a comma-separated list,
3445 * but mpol's nodelist may also contain commas.
3447 options = strchr(options, ',');
3448 if (options == NULL)
3451 if (!isdigit(*options)) {
3457 char *value = strchr(this_char, '=');
3463 len = strlen(value);
3465 err = vfs_parse_fs_string(fc, this_char, value, len);
3474 * Reconfigure a shmem filesystem.
3476 * Note that we disallow change from limited->unlimited blocks/inodes while any
3477 * are in use; but we must separately disallow unlimited->limited, because in
3478 * that case we have no record of how much is already in use.
3480 static int shmem_reconfigure(struct fs_context *fc)
3482 struct shmem_options *ctx = fc->fs_private;
3483 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3484 unsigned long inodes;
3485 struct mempolicy *mpol = NULL;
3488 raw_spin_lock(&sbinfo->stat_lock);
3489 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3490 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3491 if (!sbinfo->max_blocks) {
3492 err = "Cannot retroactively limit size";
3495 if (percpu_counter_compare(&sbinfo->used_blocks,
3497 err = "Too small a size for current use";
3501 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3502 if (!sbinfo->max_inodes) {
3503 err = "Cannot retroactively limit inodes";
3506 if (ctx->inodes < inodes) {
3507 err = "Too few inodes for current use";
3512 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3513 sbinfo->next_ino > UINT_MAX) {
3514 err = "Current inum too high to switch to 32-bit inums";
3518 if (ctx->seen & SHMEM_SEEN_HUGE)
3519 sbinfo->huge = ctx->huge;
3520 if (ctx->seen & SHMEM_SEEN_INUMS)
3521 sbinfo->full_inums = ctx->full_inums;
3522 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3523 sbinfo->max_blocks = ctx->blocks;
3524 if (ctx->seen & SHMEM_SEEN_INODES) {
3525 sbinfo->max_inodes = ctx->inodes;
3526 sbinfo->free_inodes = ctx->inodes - inodes;
3530 * Preserve previous mempolicy unless mpol remount option was specified.
3533 mpol = sbinfo->mpol;
3534 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3537 raw_spin_unlock(&sbinfo->stat_lock);
3541 raw_spin_unlock(&sbinfo->stat_lock);
3542 return invalfc(fc, "%s", err);
3545 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3547 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3549 if (sbinfo->max_blocks != shmem_default_max_blocks())
3550 seq_printf(seq, ",size=%luk",
3551 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3552 if (sbinfo->max_inodes != shmem_default_max_inodes())
3553 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3554 if (sbinfo->mode != (0777 | S_ISVTX))
3555 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3556 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3557 seq_printf(seq, ",uid=%u",
3558 from_kuid_munged(&init_user_ns, sbinfo->uid));
3559 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3560 seq_printf(seq, ",gid=%u",
3561 from_kgid_munged(&init_user_ns, sbinfo->gid));
3564 * Showing inode{64,32} might be useful even if it's the system default,
3565 * since then people don't have to resort to checking both here and
3566 * /proc/config.gz to confirm 64-bit inums were successfully applied
3567 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3569 * We hide it when inode64 isn't the default and we are using 32-bit
3570 * inodes, since that probably just means the feature isn't even under
3575 * +-----------------+-----------------+
3576 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3577 * +------------------+-----------------+-----------------+
3578 * | full_inums=true | show | show |
3579 * | full_inums=false | show | hide |
3580 * +------------------+-----------------+-----------------+
3583 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3584 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3585 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3586 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3588 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3590 shmem_show_mpol(seq, sbinfo->mpol);
3594 #endif /* CONFIG_TMPFS */
3596 static void shmem_put_super(struct super_block *sb)
3598 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3600 free_percpu(sbinfo->ino_batch);
3601 percpu_counter_destroy(&sbinfo->used_blocks);
3602 mpol_put(sbinfo->mpol);
3604 sb->s_fs_info = NULL;
3607 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3609 struct shmem_options *ctx = fc->fs_private;
3610 struct inode *inode;
3611 struct shmem_sb_info *sbinfo;
3613 /* Round up to L1_CACHE_BYTES to resist false sharing */
3614 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3615 L1_CACHE_BYTES), GFP_KERNEL);
3619 sb->s_fs_info = sbinfo;
3623 * Per default we only allow half of the physical ram per
3624 * tmpfs instance, limiting inodes to one per page of lowmem;
3625 * but the internal instance is left unlimited.
3627 if (!(sb->s_flags & SB_KERNMOUNT)) {
3628 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3629 ctx->blocks = shmem_default_max_blocks();
3630 if (!(ctx->seen & SHMEM_SEEN_INODES))
3631 ctx->inodes = shmem_default_max_inodes();
3632 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3633 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3635 sb->s_flags |= SB_NOUSER;
3637 sb->s_export_op = &shmem_export_ops;
3638 sb->s_flags |= SB_NOSEC;
3640 sb->s_flags |= SB_NOUSER;
3642 sbinfo->max_blocks = ctx->blocks;
3643 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3644 if (sb->s_flags & SB_KERNMOUNT) {
3645 sbinfo->ino_batch = alloc_percpu(ino_t);
3646 if (!sbinfo->ino_batch)
3649 sbinfo->uid = ctx->uid;
3650 sbinfo->gid = ctx->gid;
3651 sbinfo->full_inums = ctx->full_inums;
3652 sbinfo->mode = ctx->mode;
3653 sbinfo->huge = ctx->huge;
3654 sbinfo->mpol = ctx->mpol;
3657 raw_spin_lock_init(&sbinfo->stat_lock);
3658 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3660 spin_lock_init(&sbinfo->shrinklist_lock);
3661 INIT_LIST_HEAD(&sbinfo->shrinklist);
3663 sb->s_maxbytes = MAX_LFS_FILESIZE;
3664 sb->s_blocksize = PAGE_SIZE;
3665 sb->s_blocksize_bits = PAGE_SHIFT;
3666 sb->s_magic = TMPFS_MAGIC;
3667 sb->s_op = &shmem_ops;
3668 sb->s_time_gran = 1;
3669 #ifdef CONFIG_TMPFS_XATTR
3670 sb->s_xattr = shmem_xattr_handlers;
3672 #ifdef CONFIG_TMPFS_POSIX_ACL
3673 sb->s_flags |= SB_POSIXACL;
3675 uuid_gen(&sb->s_uuid);
3677 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3680 inode->i_uid = sbinfo->uid;
3681 inode->i_gid = sbinfo->gid;
3682 sb->s_root = d_make_root(inode);
3688 shmem_put_super(sb);
3692 static int shmem_get_tree(struct fs_context *fc)
3694 return get_tree_nodev(fc, shmem_fill_super);
3697 static void shmem_free_fc(struct fs_context *fc)
3699 struct shmem_options *ctx = fc->fs_private;
3702 mpol_put(ctx->mpol);
3707 static const struct fs_context_operations shmem_fs_context_ops = {
3708 .free = shmem_free_fc,
3709 .get_tree = shmem_get_tree,
3711 .parse_monolithic = shmem_parse_options,
3712 .parse_param = shmem_parse_one,
3713 .reconfigure = shmem_reconfigure,
3717 static struct kmem_cache *shmem_inode_cachep;
3719 static struct inode *shmem_alloc_inode(struct super_block *sb)
3721 struct shmem_inode_info *info;
3722 info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
3725 return &info->vfs_inode;
3728 static void shmem_free_in_core_inode(struct inode *inode)
3730 if (S_ISLNK(inode->i_mode))
3731 kfree(inode->i_link);
3732 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3735 static void shmem_destroy_inode(struct inode *inode)
3737 if (S_ISREG(inode->i_mode))
3738 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3741 static void shmem_init_inode(void *foo)
3743 struct shmem_inode_info *info = foo;
3744 inode_init_once(&info->vfs_inode);
3747 static void shmem_init_inodecache(void)
3749 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3750 sizeof(struct shmem_inode_info),
3751 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3754 static void shmem_destroy_inodecache(void)
3756 kmem_cache_destroy(shmem_inode_cachep);
3759 /* Keep the page in page cache instead of truncating it */
3760 static int shmem_error_remove_page(struct address_space *mapping,
3766 const struct address_space_operations shmem_aops = {
3767 .writepage = shmem_writepage,
3768 .dirty_folio = noop_dirty_folio,
3770 .write_begin = shmem_write_begin,
3771 .write_end = shmem_write_end,
3773 #ifdef CONFIG_MIGRATION
3774 .migratepage = migrate_page,
3776 .error_remove_page = shmem_error_remove_page,
3778 EXPORT_SYMBOL(shmem_aops);
3780 static const struct file_operations shmem_file_operations = {
3782 .get_unmapped_area = shmem_get_unmapped_area,
3784 .llseek = shmem_file_llseek,
3785 .read_iter = shmem_file_read_iter,
3786 .write_iter = generic_file_write_iter,
3787 .fsync = noop_fsync,
3788 .splice_read = generic_file_splice_read,
3789 .splice_write = iter_file_splice_write,
3790 .fallocate = shmem_fallocate,
3794 static const struct inode_operations shmem_inode_operations = {
3795 .getattr = shmem_getattr,
3796 .setattr = shmem_setattr,
3797 #ifdef CONFIG_TMPFS_XATTR
3798 .listxattr = shmem_listxattr,
3799 .set_acl = simple_set_acl,
3803 static const struct inode_operations shmem_dir_inode_operations = {
3805 .getattr = shmem_getattr,
3806 .create = shmem_create,
3807 .lookup = simple_lookup,
3809 .unlink = shmem_unlink,
3810 .symlink = shmem_symlink,
3811 .mkdir = shmem_mkdir,
3812 .rmdir = shmem_rmdir,
3813 .mknod = shmem_mknod,
3814 .rename = shmem_rename2,
3815 .tmpfile = shmem_tmpfile,
3817 #ifdef CONFIG_TMPFS_XATTR
3818 .listxattr = shmem_listxattr,
3820 #ifdef CONFIG_TMPFS_POSIX_ACL
3821 .setattr = shmem_setattr,
3822 .set_acl = simple_set_acl,
3826 static const struct inode_operations shmem_special_inode_operations = {
3827 .getattr = shmem_getattr,
3828 #ifdef CONFIG_TMPFS_XATTR
3829 .listxattr = shmem_listxattr,
3831 #ifdef CONFIG_TMPFS_POSIX_ACL
3832 .setattr = shmem_setattr,
3833 .set_acl = simple_set_acl,
3837 static const struct super_operations shmem_ops = {
3838 .alloc_inode = shmem_alloc_inode,
3839 .free_inode = shmem_free_in_core_inode,
3840 .destroy_inode = shmem_destroy_inode,
3842 .statfs = shmem_statfs,
3843 .show_options = shmem_show_options,
3845 .evict_inode = shmem_evict_inode,
3846 .drop_inode = generic_delete_inode,
3847 .put_super = shmem_put_super,
3848 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3849 .nr_cached_objects = shmem_unused_huge_count,
3850 .free_cached_objects = shmem_unused_huge_scan,
3854 static const struct vm_operations_struct shmem_vm_ops = {
3855 .fault = shmem_fault,
3856 .map_pages = filemap_map_pages,
3858 .set_policy = shmem_set_policy,
3859 .get_policy = shmem_get_policy,
3863 int shmem_init_fs_context(struct fs_context *fc)
3865 struct shmem_options *ctx;
3867 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3871 ctx->mode = 0777 | S_ISVTX;
3872 ctx->uid = current_fsuid();
3873 ctx->gid = current_fsgid();
3875 fc->fs_private = ctx;
3876 fc->ops = &shmem_fs_context_ops;
3880 static struct file_system_type shmem_fs_type = {
3881 .owner = THIS_MODULE,
3883 .init_fs_context = shmem_init_fs_context,
3885 .parameters = shmem_fs_parameters,
3887 .kill_sb = kill_litter_super,
3888 .fs_flags = FS_USERNS_MOUNT,
3891 int __init shmem_init(void)
3895 shmem_init_inodecache();
3897 error = register_filesystem(&shmem_fs_type);
3899 pr_err("Could not register tmpfs\n");
3903 shm_mnt = kern_mount(&shmem_fs_type);
3904 if (IS_ERR(shm_mnt)) {
3905 error = PTR_ERR(shm_mnt);
3906 pr_err("Could not kern_mount tmpfs\n");
3910 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3911 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3912 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3914 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
3919 unregister_filesystem(&shmem_fs_type);
3921 shmem_destroy_inodecache();
3922 shm_mnt = ERR_PTR(error);
3926 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3927 static ssize_t shmem_enabled_show(struct kobject *kobj,
3928 struct kobj_attribute *attr, char *buf)
3930 static const int values[] = {
3932 SHMEM_HUGE_WITHIN_SIZE,
3941 for (i = 0; i < ARRAY_SIZE(values); i++) {
3942 len += sysfs_emit_at(buf, len,
3943 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3945 shmem_format_huge(values[i]));
3948 len += sysfs_emit_at(buf, len, "\n");
3953 static ssize_t shmem_enabled_store(struct kobject *kobj,
3954 struct kobj_attribute *attr, const char *buf, size_t count)
3959 if (count + 1 > sizeof(tmp))
3961 memcpy(tmp, buf, count);
3963 if (count && tmp[count - 1] == '\n')
3964 tmp[count - 1] = '\0';
3966 huge = shmem_parse_huge(tmp);
3967 if (huge == -EINVAL)
3969 if (!has_transparent_hugepage() &&
3970 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3974 if (shmem_huge > SHMEM_HUGE_DENY)
3975 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3979 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
3980 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3982 #else /* !CONFIG_SHMEM */
3985 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3987 * This is intended for small system where the benefits of the full
3988 * shmem code (swap-backed and resource-limited) are outweighed by
3989 * their complexity. On systems without swap this code should be
3990 * effectively equivalent, but much lighter weight.
3993 static struct file_system_type shmem_fs_type = {
3995 .init_fs_context = ramfs_init_fs_context,
3996 .parameters = ramfs_fs_parameters,
3997 .kill_sb = kill_litter_super,
3998 .fs_flags = FS_USERNS_MOUNT,
4001 int __init shmem_init(void)
4003 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4005 shm_mnt = kern_mount(&shmem_fs_type);
4006 BUG_ON(IS_ERR(shm_mnt));
4011 int shmem_unuse(unsigned int type)
4016 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4021 void shmem_unlock_mapping(struct address_space *mapping)
4026 unsigned long shmem_get_unmapped_area(struct file *file,
4027 unsigned long addr, unsigned long len,
4028 unsigned long pgoff, unsigned long flags)
4030 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4034 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4036 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4038 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4040 #define shmem_vm_ops generic_file_vm_ops
4041 #define shmem_file_operations ramfs_file_operations
4042 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4043 #define shmem_acct_size(flags, size) 0
4044 #define shmem_unacct_size(flags, size) do {} while (0)
4046 #endif /* CONFIG_SHMEM */
4050 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4051 unsigned long flags, unsigned int i_flags)
4053 struct inode *inode;
4057 return ERR_CAST(mnt);
4059 if (size < 0 || size > MAX_LFS_FILESIZE)
4060 return ERR_PTR(-EINVAL);
4062 if (shmem_acct_size(flags, size))
4063 return ERR_PTR(-ENOMEM);
4065 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4067 if (unlikely(!inode)) {
4068 shmem_unacct_size(flags, size);
4069 return ERR_PTR(-ENOSPC);
4071 inode->i_flags |= i_flags;
4072 inode->i_size = size;
4073 clear_nlink(inode); /* It is unlinked */
4074 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4076 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4077 &shmem_file_operations);
4084 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4085 * kernel internal. There will be NO LSM permission checks against the
4086 * underlying inode. So users of this interface must do LSM checks at a
4087 * higher layer. The users are the big_key and shm implementations. LSM
4088 * checks are provided at the key or shm level rather than the inode.
4089 * @name: name for dentry (to be seen in /proc/<pid>/maps
4090 * @size: size to be set for the file
4091 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4093 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4095 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4099 * shmem_file_setup - get an unlinked file living in tmpfs
4100 * @name: name for dentry (to be seen in /proc/<pid>/maps
4101 * @size: size to be set for the file
4102 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4104 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4106 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4108 EXPORT_SYMBOL_GPL(shmem_file_setup);
4111 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4112 * @mnt: the tmpfs mount where the file will be created
4113 * @name: name for dentry (to be seen in /proc/<pid>/maps
4114 * @size: size to be set for the file
4115 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4117 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4118 loff_t size, unsigned long flags)
4120 return __shmem_file_setup(mnt, name, size, flags, 0);
4122 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4125 * shmem_zero_setup - setup a shared anonymous mapping
4126 * @vma: the vma to be mmapped is prepared by do_mmap
4128 int shmem_zero_setup(struct vm_area_struct *vma)
4131 loff_t size = vma->vm_end - vma->vm_start;
4134 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4135 * between XFS directory reading and selinux: since this file is only
4136 * accessible to the user through its mapping, use S_PRIVATE flag to
4137 * bypass file security, in the same way as shmem_kernel_file_setup().
4139 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4141 return PTR_ERR(file);
4145 vma->vm_file = file;
4146 vma->vm_ops = &shmem_vm_ops;
4148 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4149 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4150 (vma->vm_end & HPAGE_PMD_MASK)) {
4151 khugepaged_enter(vma, vma->vm_flags);
4158 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4159 * @mapping: the page's address_space
4160 * @index: the page index
4161 * @gfp: the page allocator flags to use if allocating
4163 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4164 * with any new page allocations done using the specified allocation flags.
4165 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4166 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4167 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4169 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4170 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4172 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4173 pgoff_t index, gfp_t gfp)
4176 struct inode *inode = mapping->host;
4180 BUG_ON(!shmem_mapping(mapping));
4181 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4182 gfp, NULL, NULL, NULL);
4184 return ERR_PTR(error);
4187 if (PageHWPoison(page)) {
4189 return ERR_PTR(-EIO);
4195 * The tiny !SHMEM case uses ramfs without swap
4197 return read_cache_page_gfp(mapping, index, gfp);
4200 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);