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/frontswap.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
43 static struct vfsmount *shm_mnt;
47 * This virtual memory filesystem is heavily based on the ramfs. It
48 * extends ramfs by the ability to use swap and honor resource limits
49 * which makes it a completely usable filesystem.
52 #include <linux/xattr.h>
53 #include <linux/exportfs.h>
54 #include <linux/posix_acl.h>
55 #include <linux/posix_acl_xattr.h>
56 #include <linux/mman.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/backing-dev.h>
60 #include <linux/shmem_fs.h>
61 #include <linux/writeback.h>
62 #include <linux/blkdev.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/userfaultfd_k.h>
80 #include <linux/rmap.h>
81 #include <linux/uuid.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_mutex making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc {
102 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
103 pgoff_t start; /* start of range currently being fallocated */
104 pgoff_t next; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
109 struct shmem_options {
110 unsigned long long blocks;
111 unsigned long long inodes;
112 struct mempolicy *mpol;
119 #define SHMEM_SEEN_BLOCKS 1
120 #define SHMEM_SEEN_INODES 2
121 #define SHMEM_SEEN_HUGE 4
122 #define SHMEM_SEEN_INUMS 8
126 static unsigned long shmem_default_max_blocks(void)
128 return totalram_pages() / 2;
131 static unsigned long shmem_default_max_inodes(void)
133 unsigned long nr_pages = totalram_pages();
135 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
139 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
140 struct page **pagep, enum sgp_type sgp,
141 gfp_t gfp, struct vm_area_struct *vma,
142 vm_fault_t *fault_type);
143 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 struct vm_fault *vmf, vm_fault_t *fault_type);
148 int shmem_getpage(struct inode *inode, pgoff_t index,
149 struct page **pagep, enum sgp_type sgp)
151 return shmem_getpage_gfp(inode, index, pagep, sgp,
152 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
155 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
157 return sb->s_fs_info;
161 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
162 * for shared memory and for shared anonymous (/dev/zero) mappings
163 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
164 * consistent with the pre-accounting of private mappings ...
166 static inline int shmem_acct_size(unsigned long flags, loff_t size)
168 return (flags & VM_NORESERVE) ?
169 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
172 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
174 if (!(flags & VM_NORESERVE))
175 vm_unacct_memory(VM_ACCT(size));
178 static inline int shmem_reacct_size(unsigned long flags,
179 loff_t oldsize, loff_t newsize)
181 if (!(flags & VM_NORESERVE)) {
182 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
183 return security_vm_enough_memory_mm(current->mm,
184 VM_ACCT(newsize) - VM_ACCT(oldsize));
185 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
186 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
192 * ... whereas tmpfs objects are accounted incrementally as
193 * pages are allocated, in order to allow large sparse files.
194 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
195 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
197 static inline int shmem_acct_block(unsigned long flags, long pages)
199 if (!(flags & VM_NORESERVE))
202 return security_vm_enough_memory_mm(current->mm,
203 pages * VM_ACCT(PAGE_SIZE));
206 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
208 if (flags & VM_NORESERVE)
209 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
212 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
214 struct shmem_inode_info *info = SHMEM_I(inode);
215 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
217 if (shmem_acct_block(info->flags, pages))
220 if (sbinfo->max_blocks) {
221 if (percpu_counter_compare(&sbinfo->used_blocks,
222 sbinfo->max_blocks - pages) > 0)
224 percpu_counter_add(&sbinfo->used_blocks, pages);
230 shmem_unacct_blocks(info->flags, pages);
234 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
236 struct shmem_inode_info *info = SHMEM_I(inode);
237 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
239 if (sbinfo->max_blocks)
240 percpu_counter_sub(&sbinfo->used_blocks, pages);
241 shmem_unacct_blocks(info->flags, pages);
244 static const struct super_operations shmem_ops;
245 const struct address_space_operations shmem_aops;
246 static const struct file_operations shmem_file_operations;
247 static const struct inode_operations shmem_inode_operations;
248 static const struct inode_operations shmem_dir_inode_operations;
249 static const struct inode_operations shmem_special_inode_operations;
250 static const struct vm_operations_struct shmem_vm_ops;
251 static struct file_system_type shmem_fs_type;
253 bool vma_is_shmem(struct vm_area_struct *vma)
255 return vma->vm_ops == &shmem_vm_ops;
258 static LIST_HEAD(shmem_swaplist);
259 static DEFINE_MUTEX(shmem_swaplist_mutex);
262 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
263 * produces a novel ino for the newly allocated inode.
265 * It may also be called when making a hard link to permit the space needed by
266 * each dentry. However, in that case, no new inode number is needed since that
267 * internally draws from another pool of inode numbers (currently global
268 * get_next_ino()). This case is indicated by passing NULL as inop.
270 #define SHMEM_INO_BATCH 1024
271 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
273 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
276 if (!(sb->s_flags & SB_KERNMOUNT)) {
277 raw_spin_lock(&sbinfo->stat_lock);
278 if (sbinfo->max_inodes) {
279 if (!sbinfo->free_inodes) {
280 raw_spin_unlock(&sbinfo->stat_lock);
283 sbinfo->free_inodes--;
286 ino = sbinfo->next_ino++;
287 if (unlikely(is_zero_ino(ino)))
288 ino = sbinfo->next_ino++;
289 if (unlikely(!sbinfo->full_inums &&
292 * Emulate get_next_ino uint wraparound for
295 if (IS_ENABLED(CONFIG_64BIT))
296 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
297 __func__, MINOR(sb->s_dev));
298 sbinfo->next_ino = 1;
299 ino = sbinfo->next_ino++;
303 raw_spin_unlock(&sbinfo->stat_lock);
306 * __shmem_file_setup, one of our callers, is lock-free: it
307 * doesn't hold stat_lock in shmem_reserve_inode since
308 * max_inodes is always 0, and is called from potentially
309 * unknown contexts. As such, use a per-cpu batched allocator
310 * which doesn't require the per-sb stat_lock unless we are at
311 * the batch boundary.
313 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
314 * shmem mounts are not exposed to userspace, so we don't need
315 * to worry about things like glibc compatibility.
319 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
321 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
322 raw_spin_lock(&sbinfo->stat_lock);
323 ino = sbinfo->next_ino;
324 sbinfo->next_ino += SHMEM_INO_BATCH;
325 raw_spin_unlock(&sbinfo->stat_lock);
326 if (unlikely(is_zero_ino(ino)))
337 static void shmem_free_inode(struct super_block *sb)
339 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
340 if (sbinfo->max_inodes) {
341 raw_spin_lock(&sbinfo->stat_lock);
342 sbinfo->free_inodes++;
343 raw_spin_unlock(&sbinfo->stat_lock);
348 * shmem_recalc_inode - recalculate the block usage of an inode
349 * @inode: inode to recalc
351 * We have to calculate the free blocks since the mm can drop
352 * undirtied hole pages behind our back.
354 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
355 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
357 * It has to be called with the spinlock held.
359 static void shmem_recalc_inode(struct inode *inode)
361 struct shmem_inode_info *info = SHMEM_I(inode);
364 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
366 info->alloced -= freed;
367 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
368 shmem_inode_unacct_blocks(inode, freed);
372 bool shmem_charge(struct inode *inode, long pages)
374 struct shmem_inode_info *info = SHMEM_I(inode);
377 if (!shmem_inode_acct_block(inode, pages))
380 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
381 inode->i_mapping->nrpages += pages;
383 spin_lock_irqsave(&info->lock, flags);
384 info->alloced += pages;
385 inode->i_blocks += pages * BLOCKS_PER_PAGE;
386 shmem_recalc_inode(inode);
387 spin_unlock_irqrestore(&info->lock, flags);
392 void shmem_uncharge(struct inode *inode, long pages)
394 struct shmem_inode_info *info = SHMEM_I(inode);
397 /* nrpages adjustment done by __delete_from_page_cache() or caller */
399 spin_lock_irqsave(&info->lock, flags);
400 info->alloced -= pages;
401 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
402 shmem_recalc_inode(inode);
403 spin_unlock_irqrestore(&info->lock, flags);
405 shmem_inode_unacct_blocks(inode, pages);
409 * Replace item expected in xarray by a new item, while holding xa_lock.
411 static int shmem_replace_entry(struct address_space *mapping,
412 pgoff_t index, void *expected, void *replacement)
414 XA_STATE(xas, &mapping->i_pages, index);
417 VM_BUG_ON(!expected);
418 VM_BUG_ON(!replacement);
419 item = xas_load(&xas);
420 if (item != expected)
422 xas_store(&xas, replacement);
427 * Sometimes, before we decide whether to proceed or to fail, we must check
428 * that an entry was not already brought back from swap by a racing thread.
430 * Checking page is not enough: by the time a SwapCache page is locked, it
431 * might be reused, and again be SwapCache, using the same swap as before.
433 static bool shmem_confirm_swap(struct address_space *mapping,
434 pgoff_t index, swp_entry_t swap)
436 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
440 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
443 * disables huge pages for the mount;
445 * enables huge pages for the mount;
446 * SHMEM_HUGE_WITHIN_SIZE:
447 * only allocate huge pages if the page will be fully within i_size,
448 * also respect fadvise()/madvise() hints;
450 * only allocate huge pages if requested with fadvise()/madvise();
453 #define SHMEM_HUGE_NEVER 0
454 #define SHMEM_HUGE_ALWAYS 1
455 #define SHMEM_HUGE_WITHIN_SIZE 2
456 #define SHMEM_HUGE_ADVISE 3
460 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
463 * disables huge on shm_mnt and all mounts, for emergency use;
465 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
468 #define SHMEM_HUGE_DENY (-1)
469 #define SHMEM_HUGE_FORCE (-2)
471 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
472 /* ifdef here to avoid bloating shmem.o when not necessary */
474 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
476 bool shmem_is_huge(struct vm_area_struct *vma,
477 struct inode *inode, pgoff_t index)
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, HPAGE_PMD_NR);
494 i_size = round_up(i_size_read(inode), PAGE_SIZE);
495 if (i_size >= HPAGE_PMD_SIZE && (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;
558 int removed = 0, split = 0;
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);
577 /* Check if there's anything to gain */
578 if (round_up(inode->i_size, PAGE_SIZE) ==
579 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
580 list_move(&info->shrinklist, &to_remove);
585 list_move(&info->shrinklist, &list);
590 spin_unlock(&sbinfo->shrinklist_lock);
592 list_for_each_safe(pos, next, &to_remove) {
593 info = list_entry(pos, struct shmem_inode_info, shrinklist);
594 inode = &info->vfs_inode;
595 list_del_init(&info->shrinklist);
599 list_for_each_safe(pos, next, &list) {
602 info = list_entry(pos, struct shmem_inode_info, shrinklist);
603 inode = &info->vfs_inode;
605 if (nr_to_split && split >= nr_to_split)
608 page = find_get_page(inode->i_mapping,
609 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
613 /* No huge page at the end of the file: nothing to split */
614 if (!PageTransHuge(page)) {
620 * Leave the inode on the list if we failed to lock
621 * the page at this time.
623 * Waiting for the lock may lead to deadlock in the
626 if (!trylock_page(page)) {
631 ret = split_huge_page(page);
635 /* If split failed leave the inode on the list */
641 list_del_init(&info->shrinklist);
647 spin_lock(&sbinfo->shrinklist_lock);
648 list_splice_tail(&list, &sbinfo->shrinklist);
649 sbinfo->shrinklist_len -= removed;
650 spin_unlock(&sbinfo->shrinklist_lock);
655 static long shmem_unused_huge_scan(struct super_block *sb,
656 struct shrink_control *sc)
658 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
660 if (!READ_ONCE(sbinfo->shrinklist_len))
663 return shmem_unused_huge_shrink(sbinfo, sc, 0);
666 static long shmem_unused_huge_count(struct super_block *sb,
667 struct shrink_control *sc)
669 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
670 return READ_ONCE(sbinfo->shrinklist_len);
672 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
674 #define shmem_huge SHMEM_HUGE_DENY
676 bool shmem_is_huge(struct vm_area_struct *vma,
677 struct inode *inode, pgoff_t index)
682 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
683 struct shrink_control *sc, unsigned long nr_to_split)
687 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
689 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
691 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
692 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
693 shmem_huge != SHMEM_HUGE_DENY)
699 * Like add_to_page_cache_locked, but error if expected item has gone.
701 static int shmem_add_to_page_cache(struct page *page,
702 struct address_space *mapping,
703 pgoff_t index, void *expected, gfp_t gfp,
704 struct mm_struct *charge_mm)
706 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
708 unsigned long nr = compound_nr(page);
711 VM_BUG_ON_PAGE(PageTail(page), page);
712 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
713 VM_BUG_ON_PAGE(!PageLocked(page), page);
714 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
715 VM_BUG_ON(expected && PageTransHuge(page));
717 page_ref_add(page, nr);
718 page->mapping = mapping;
721 if (!PageSwapCache(page)) {
722 error = mem_cgroup_charge(page, charge_mm, gfp);
724 if (PageTransHuge(page)) {
725 count_vm_event(THP_FILE_FALLBACK);
726 count_vm_event(THP_FILE_FALLBACK_CHARGE);
731 cgroup_throttle_swaprate(page, gfp);
736 entry = xas_find_conflict(&xas);
737 if (entry != expected)
738 xas_set_err(&xas, -EEXIST);
739 xas_create_range(&xas);
743 xas_store(&xas, page);
748 if (PageTransHuge(page)) {
749 count_vm_event(THP_FILE_ALLOC);
750 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
752 mapping->nrpages += nr;
753 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
754 __mod_lruvec_page_state(page, NR_SHMEM, nr);
756 xas_unlock_irq(&xas);
757 } while (xas_nomem(&xas, gfp));
759 if (xas_error(&xas)) {
760 error = xas_error(&xas);
766 page->mapping = NULL;
767 page_ref_sub(page, nr);
772 * Like delete_from_page_cache, but substitutes swap for page.
774 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
776 struct address_space *mapping = page->mapping;
779 VM_BUG_ON_PAGE(PageCompound(page), page);
781 xa_lock_irq(&mapping->i_pages);
782 error = shmem_replace_entry(mapping, page->index, page, radswap);
783 page->mapping = NULL;
785 __dec_lruvec_page_state(page, NR_FILE_PAGES);
786 __dec_lruvec_page_state(page, NR_SHMEM);
787 xa_unlock_irq(&mapping->i_pages);
793 * Remove swap entry from page cache, free the swap and its page cache.
795 static int shmem_free_swap(struct address_space *mapping,
796 pgoff_t index, void *radswap)
800 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
803 free_swap_and_cache(radix_to_swp_entry(radswap));
808 * Determine (in bytes) how many of the shmem object's pages mapped by the
809 * given offsets are swapped out.
811 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
812 * as long as the inode doesn't go away and racy results are not a problem.
814 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
815 pgoff_t start, pgoff_t end)
817 XA_STATE(xas, &mapping->i_pages, start);
819 unsigned long swapped = 0;
822 xas_for_each(&xas, page, end - 1) {
823 if (xas_retry(&xas, page))
825 if (xa_is_value(page))
828 if (need_resched()) {
836 return swapped << PAGE_SHIFT;
840 * Determine (in bytes) how many of the shmem object's pages mapped by the
841 * given vma is swapped out.
843 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
844 * as long as the inode doesn't go away and racy results are not a problem.
846 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
848 struct inode *inode = file_inode(vma->vm_file);
849 struct shmem_inode_info *info = SHMEM_I(inode);
850 struct address_space *mapping = inode->i_mapping;
851 unsigned long swapped;
853 /* Be careful as we don't hold info->lock */
854 swapped = READ_ONCE(info->swapped);
857 * The easier cases are when the shmem object has nothing in swap, or
858 * the vma maps it whole. Then we can simply use the stats that we
864 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
865 return swapped << PAGE_SHIFT;
867 /* Here comes the more involved part */
868 return shmem_partial_swap_usage(mapping,
869 linear_page_index(vma, vma->vm_start),
870 linear_page_index(vma, vma->vm_end));
874 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
876 void shmem_unlock_mapping(struct address_space *mapping)
883 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
885 while (!mapping_unevictable(mapping)) {
886 if (!pagevec_lookup(&pvec, mapping, &index))
888 check_move_unevictable_pages(&pvec);
889 pagevec_release(&pvec);
895 * Check whether a hole-punch or truncation needs to split a huge page,
896 * returning true if no split was required, or the split has been successful.
898 * Eviction (or truncation to 0 size) should never need to split a huge page;
899 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
900 * head, and then succeeded to trylock on tail.
902 * A split can only succeed when there are no additional references on the
903 * huge page: so the split below relies upon find_get_entries() having stopped
904 * when it found a subpage of the huge page, without getting further references.
906 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
908 if (!PageTransCompound(page))
911 /* Just proceed to delete a huge page wholly within the range punched */
912 if (PageHead(page) &&
913 page->index >= start && page->index + HPAGE_PMD_NR <= end)
916 /* Try to split huge page, so we can truly punch the hole or truncate */
917 return split_huge_page(page) >= 0;
921 * Remove range of pages and swap entries from page cache, and free them.
922 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
924 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
927 struct address_space *mapping = inode->i_mapping;
928 struct shmem_inode_info *info = SHMEM_I(inode);
929 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
930 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
931 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
932 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
934 pgoff_t indices[PAGEVEC_SIZE];
935 long nr_swaps_freed = 0;
940 end = -1; /* unsigned, so actually very big */
942 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
943 info->fallocend = start;
947 while (index < end && find_lock_entries(mapping, index, end - 1,
949 for (i = 0; i < pagevec_count(&pvec); i++) {
950 struct page *page = pvec.pages[i];
954 if (xa_is_value(page)) {
957 nr_swaps_freed += !shmem_free_swap(mapping,
961 index += thp_nr_pages(page) - 1;
963 if (!unfalloc || !PageUptodate(page))
964 truncate_inode_page(mapping, page);
967 pagevec_remove_exceptionals(&pvec);
968 pagevec_release(&pvec);
974 struct page *page = NULL;
975 shmem_getpage(inode, start - 1, &page, SGP_READ);
977 unsigned int top = PAGE_SIZE;
982 zero_user_segment(page, partial_start, top);
983 set_page_dirty(page);
989 struct page *page = NULL;
990 shmem_getpage(inode, end, &page, SGP_READ);
992 zero_user_segment(page, 0, partial_end);
993 set_page_dirty(page);
1002 while (index < end) {
1005 if (!find_get_entries(mapping, index, end - 1, &pvec,
1007 /* If all gone or hole-punch or unfalloc, we're done */
1008 if (index == start || end != -1)
1010 /* But if truncating, restart to make sure all gone */
1014 for (i = 0; i < pagevec_count(&pvec); i++) {
1015 struct page *page = pvec.pages[i];
1018 if (xa_is_value(page)) {
1021 if (shmem_free_swap(mapping, index, page)) {
1022 /* Swap was replaced by page: retry */
1032 if (!unfalloc || !PageUptodate(page)) {
1033 if (page_mapping(page) != mapping) {
1034 /* Page was replaced by swap: retry */
1039 VM_BUG_ON_PAGE(PageWriteback(page), page);
1040 if (shmem_punch_compound(page, start, end))
1041 truncate_inode_page(mapping, page);
1042 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1043 /* Wipe the page and don't get stuck */
1044 clear_highpage(page);
1045 flush_dcache_page(page);
1046 set_page_dirty(page);
1048 round_up(start, HPAGE_PMD_NR))
1054 pagevec_remove_exceptionals(&pvec);
1055 pagevec_release(&pvec);
1059 spin_lock_irq(&info->lock);
1060 info->swapped -= nr_swaps_freed;
1061 shmem_recalc_inode(inode);
1062 spin_unlock_irq(&info->lock);
1065 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1067 shmem_undo_range(inode, lstart, lend, false);
1068 inode->i_ctime = inode->i_mtime = current_time(inode);
1070 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1072 static int shmem_getattr(struct user_namespace *mnt_userns,
1073 const struct path *path, struct kstat *stat,
1074 u32 request_mask, unsigned int query_flags)
1076 struct inode *inode = path->dentry->d_inode;
1077 struct shmem_inode_info *info = SHMEM_I(inode);
1078 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1080 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1081 spin_lock_irq(&info->lock);
1082 shmem_recalc_inode(inode);
1083 spin_unlock_irq(&info->lock);
1085 generic_fillattr(&init_user_ns, inode, stat);
1087 if (is_huge_enabled(sb_info))
1088 stat->blksize = HPAGE_PMD_SIZE;
1093 static int shmem_setattr(struct user_namespace *mnt_userns,
1094 struct dentry *dentry, struct iattr *attr)
1096 struct inode *inode = d_inode(dentry);
1097 struct shmem_inode_info *info = SHMEM_I(inode);
1100 error = setattr_prepare(&init_user_ns, dentry, attr);
1104 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1105 loff_t oldsize = inode->i_size;
1106 loff_t newsize = attr->ia_size;
1108 /* protected by i_mutex */
1109 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1110 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1113 if (newsize != oldsize) {
1114 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1118 i_size_write(inode, newsize);
1119 inode->i_ctime = inode->i_mtime = current_time(inode);
1121 if (newsize <= oldsize) {
1122 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1123 if (oldsize > holebegin)
1124 unmap_mapping_range(inode->i_mapping,
1127 shmem_truncate_range(inode,
1128 newsize, (loff_t)-1);
1129 /* unmap again to remove racily COWed private pages */
1130 if (oldsize > holebegin)
1131 unmap_mapping_range(inode->i_mapping,
1136 setattr_copy(&init_user_ns, inode, attr);
1137 if (attr->ia_valid & ATTR_MODE)
1138 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1142 static void shmem_evict_inode(struct inode *inode)
1144 struct shmem_inode_info *info = SHMEM_I(inode);
1145 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1147 if (shmem_mapping(inode->i_mapping)) {
1148 shmem_unacct_size(info->flags, inode->i_size);
1150 shmem_truncate_range(inode, 0, (loff_t)-1);
1151 if (!list_empty(&info->shrinklist)) {
1152 spin_lock(&sbinfo->shrinklist_lock);
1153 if (!list_empty(&info->shrinklist)) {
1154 list_del_init(&info->shrinklist);
1155 sbinfo->shrinklist_len--;
1157 spin_unlock(&sbinfo->shrinklist_lock);
1159 while (!list_empty(&info->swaplist)) {
1160 /* Wait while shmem_unuse() is scanning this inode... */
1161 wait_var_event(&info->stop_eviction,
1162 !atomic_read(&info->stop_eviction));
1163 mutex_lock(&shmem_swaplist_mutex);
1164 /* ...but beware of the race if we peeked too early */
1165 if (!atomic_read(&info->stop_eviction))
1166 list_del_init(&info->swaplist);
1167 mutex_unlock(&shmem_swaplist_mutex);
1171 simple_xattrs_free(&info->xattrs);
1172 WARN_ON(inode->i_blocks);
1173 shmem_free_inode(inode->i_sb);
1177 static int shmem_find_swap_entries(struct address_space *mapping,
1178 pgoff_t start, unsigned int nr_entries,
1179 struct page **entries, pgoff_t *indices,
1180 unsigned int type, bool frontswap)
1182 XA_STATE(xas, &mapping->i_pages, start);
1185 unsigned int ret = 0;
1191 xas_for_each(&xas, page, ULONG_MAX) {
1192 if (xas_retry(&xas, page))
1195 if (!xa_is_value(page))
1198 entry = radix_to_swp_entry(page);
1199 if (swp_type(entry) != type)
1202 !frontswap_test(swap_info[type], swp_offset(entry)))
1205 indices[ret] = xas.xa_index;
1206 entries[ret] = page;
1208 if (need_resched()) {
1212 if (++ret == nr_entries)
1221 * Move the swapped pages for an inode to page cache. Returns the count
1222 * of pages swapped in, or the error in case of failure.
1224 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1230 struct address_space *mapping = inode->i_mapping;
1232 for (i = 0; i < pvec.nr; i++) {
1233 struct page *page = pvec.pages[i];
1235 if (!xa_is_value(page))
1237 error = shmem_swapin_page(inode, indices[i],
1239 mapping_gfp_mask(mapping),
1246 if (error == -ENOMEM)
1250 return error ? error : ret;
1254 * If swap found in inode, free it and move page from swapcache to filecache.
1256 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1257 bool frontswap, unsigned long *fs_pages_to_unuse)
1259 struct address_space *mapping = inode->i_mapping;
1261 struct pagevec pvec;
1262 pgoff_t indices[PAGEVEC_SIZE];
1263 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1266 pagevec_init(&pvec);
1268 unsigned int nr_entries = PAGEVEC_SIZE;
1270 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1271 nr_entries = *fs_pages_to_unuse;
1273 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1274 pvec.pages, indices,
1281 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1285 if (frontswap_partial) {
1286 *fs_pages_to_unuse -= ret;
1287 if (*fs_pages_to_unuse == 0) {
1288 ret = FRONTSWAP_PAGES_UNUSED;
1293 start = indices[pvec.nr - 1];
1300 * Read all the shared memory data that resides in the swap
1301 * device 'type' back into memory, so the swap device can be
1304 int shmem_unuse(unsigned int type, bool frontswap,
1305 unsigned long *fs_pages_to_unuse)
1307 struct shmem_inode_info *info, *next;
1310 if (list_empty(&shmem_swaplist))
1313 mutex_lock(&shmem_swaplist_mutex);
1314 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1315 if (!info->swapped) {
1316 list_del_init(&info->swaplist);
1320 * Drop the swaplist mutex while searching the inode for swap;
1321 * but before doing so, make sure shmem_evict_inode() will not
1322 * remove placeholder inode from swaplist, nor let it be freed
1323 * (igrab() would protect from unlink, but not from unmount).
1325 atomic_inc(&info->stop_eviction);
1326 mutex_unlock(&shmem_swaplist_mutex);
1328 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1332 mutex_lock(&shmem_swaplist_mutex);
1333 next = list_next_entry(info, swaplist);
1335 list_del_init(&info->swaplist);
1336 if (atomic_dec_and_test(&info->stop_eviction))
1337 wake_up_var(&info->stop_eviction);
1341 mutex_unlock(&shmem_swaplist_mutex);
1347 * Move the page from the page cache to the swap cache.
1349 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1351 struct shmem_inode_info *info;
1352 struct address_space *mapping;
1353 struct inode *inode;
1357 VM_BUG_ON_PAGE(PageCompound(page), page);
1358 BUG_ON(!PageLocked(page));
1359 mapping = page->mapping;
1360 index = page->index;
1361 inode = mapping->host;
1362 info = SHMEM_I(inode);
1363 if (info->flags & VM_LOCKED)
1365 if (!total_swap_pages)
1369 * Our capabilities prevent regular writeback or sync from ever calling
1370 * shmem_writepage; but a stacking filesystem might use ->writepage of
1371 * its underlying filesystem, in which case tmpfs should write out to
1372 * swap only in response to memory pressure, and not for the writeback
1375 if (!wbc->for_reclaim) {
1376 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1381 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1382 * value into swapfile.c, the only way we can correctly account for a
1383 * fallocated page arriving here is now to initialize it and write it.
1385 * That's okay for a page already fallocated earlier, but if we have
1386 * not yet completed the fallocation, then (a) we want to keep track
1387 * of this page in case we have to undo it, and (b) it may not be a
1388 * good idea to continue anyway, once we're pushing into swap. So
1389 * reactivate the page, and let shmem_fallocate() quit when too many.
1391 if (!PageUptodate(page)) {
1392 if (inode->i_private) {
1393 struct shmem_falloc *shmem_falloc;
1394 spin_lock(&inode->i_lock);
1395 shmem_falloc = inode->i_private;
1397 !shmem_falloc->waitq &&
1398 index >= shmem_falloc->start &&
1399 index < shmem_falloc->next)
1400 shmem_falloc->nr_unswapped++;
1402 shmem_falloc = NULL;
1403 spin_unlock(&inode->i_lock);
1407 clear_highpage(page);
1408 flush_dcache_page(page);
1409 SetPageUptodate(page);
1412 swap = get_swap_page(page);
1417 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1418 * if it's not already there. Do it now before the page is
1419 * moved to swap cache, when its pagelock no longer protects
1420 * the inode from eviction. But don't unlock the mutex until
1421 * we've incremented swapped, because shmem_unuse_inode() will
1422 * prune a !swapped inode from the swaplist under this mutex.
1424 mutex_lock(&shmem_swaplist_mutex);
1425 if (list_empty(&info->swaplist))
1426 list_add(&info->swaplist, &shmem_swaplist);
1428 if (add_to_swap_cache(page, swap,
1429 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1431 spin_lock_irq(&info->lock);
1432 shmem_recalc_inode(inode);
1434 spin_unlock_irq(&info->lock);
1436 swap_shmem_alloc(swap);
1437 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1439 mutex_unlock(&shmem_swaplist_mutex);
1440 BUG_ON(page_mapped(page));
1441 swap_writepage(page, wbc);
1445 mutex_unlock(&shmem_swaplist_mutex);
1446 put_swap_page(page, swap);
1448 set_page_dirty(page);
1449 if (wbc->for_reclaim)
1450 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1455 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1456 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1460 if (!mpol || mpol->mode == MPOL_DEFAULT)
1461 return; /* show nothing */
1463 mpol_to_str(buffer, sizeof(buffer), mpol);
1465 seq_printf(seq, ",mpol=%s", buffer);
1468 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1470 struct mempolicy *mpol = NULL;
1472 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1473 mpol = sbinfo->mpol;
1475 raw_spin_unlock(&sbinfo->stat_lock);
1479 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1480 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1483 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1487 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1489 #define vm_policy vm_private_data
1492 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1493 struct shmem_inode_info *info, pgoff_t index)
1495 /* Create a pseudo vma that just contains the policy */
1496 vma_init(vma, NULL);
1497 /* Bias interleave by inode number to distribute better across nodes */
1498 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1499 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1502 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1504 /* Drop reference taken by mpol_shared_policy_lookup() */
1505 mpol_cond_put(vma->vm_policy);
1508 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1509 struct shmem_inode_info *info, pgoff_t index)
1511 struct vm_area_struct pvma;
1513 struct vm_fault vmf = {
1517 shmem_pseudo_vma_init(&pvma, info, index);
1518 page = swap_cluster_readahead(swap, gfp, &vmf);
1519 shmem_pseudo_vma_destroy(&pvma);
1525 * Make sure huge_gfp is always more limited than limit_gfp.
1526 * Some of the flags set permissions, while others set limitations.
1528 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1530 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1531 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1532 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1533 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1535 /* Allow allocations only from the originally specified zones. */
1536 result |= zoneflags;
1539 * Minimize the result gfp by taking the union with the deny flags,
1540 * and the intersection of the allow flags.
1542 result |= (limit_gfp & denyflags);
1543 result |= (huge_gfp & limit_gfp) & allowflags;
1548 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1549 struct shmem_inode_info *info, pgoff_t index)
1551 struct vm_area_struct pvma;
1552 struct address_space *mapping = info->vfs_inode.i_mapping;
1556 hindex = round_down(index, HPAGE_PMD_NR);
1557 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1561 shmem_pseudo_vma_init(&pvma, info, hindex);
1562 page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(),
1564 shmem_pseudo_vma_destroy(&pvma);
1566 prep_transhuge_page(page);
1568 count_vm_event(THP_FILE_FALLBACK);
1572 static struct page *shmem_alloc_page(gfp_t gfp,
1573 struct shmem_inode_info *info, pgoff_t index)
1575 struct vm_area_struct pvma;
1578 shmem_pseudo_vma_init(&pvma, info, index);
1579 page = alloc_page_vma(gfp, &pvma, 0);
1580 shmem_pseudo_vma_destroy(&pvma);
1585 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1586 struct inode *inode,
1587 pgoff_t index, bool huge)
1589 struct shmem_inode_info *info = SHMEM_I(inode);
1594 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1596 nr = huge ? HPAGE_PMD_NR : 1;
1598 if (!shmem_inode_acct_block(inode, nr))
1602 page = shmem_alloc_hugepage(gfp, info, index);
1604 page = shmem_alloc_page(gfp, info, index);
1606 __SetPageLocked(page);
1607 __SetPageSwapBacked(page);
1612 shmem_inode_unacct_blocks(inode, nr);
1614 return ERR_PTR(err);
1618 * When a page is moved from swapcache to shmem filecache (either by the
1619 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1620 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1621 * ignorance of the mapping it belongs to. If that mapping has special
1622 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1623 * we may need to copy to a suitable page before moving to filecache.
1625 * In a future release, this may well be extended to respect cpuset and
1626 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1627 * but for now it is a simple matter of zone.
1629 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1631 return page_zonenum(page) > gfp_zone(gfp);
1634 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1635 struct shmem_inode_info *info, pgoff_t index)
1637 struct page *oldpage, *newpage;
1638 struct address_space *swap_mapping;
1644 entry.val = page_private(oldpage);
1645 swap_index = swp_offset(entry);
1646 swap_mapping = page_mapping(oldpage);
1649 * We have arrived here because our zones are constrained, so don't
1650 * limit chance of success by further cpuset and node constraints.
1652 gfp &= ~GFP_CONSTRAINT_MASK;
1653 newpage = shmem_alloc_page(gfp, info, index);
1658 copy_highpage(newpage, oldpage);
1659 flush_dcache_page(newpage);
1661 __SetPageLocked(newpage);
1662 __SetPageSwapBacked(newpage);
1663 SetPageUptodate(newpage);
1664 set_page_private(newpage, entry.val);
1665 SetPageSwapCache(newpage);
1668 * Our caller will very soon move newpage out of swapcache, but it's
1669 * a nice clean interface for us to replace oldpage by newpage there.
1671 xa_lock_irq(&swap_mapping->i_pages);
1672 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1674 mem_cgroup_migrate(oldpage, newpage);
1675 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1676 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1678 xa_unlock_irq(&swap_mapping->i_pages);
1680 if (unlikely(error)) {
1682 * Is this possible? I think not, now that our callers check
1683 * both PageSwapCache and page_private after getting page lock;
1684 * but be defensive. Reverse old to newpage for clear and free.
1688 lru_cache_add(newpage);
1692 ClearPageSwapCache(oldpage);
1693 set_page_private(oldpage, 0);
1695 unlock_page(oldpage);
1702 * Swap in the page pointed to by *pagep.
1703 * Caller has to make sure that *pagep contains a valid swapped page.
1704 * Returns 0 and the page in pagep if success. On failure, returns the
1705 * error code and NULL in *pagep.
1707 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1708 struct page **pagep, enum sgp_type sgp,
1709 gfp_t gfp, struct vm_area_struct *vma,
1710 vm_fault_t *fault_type)
1712 struct address_space *mapping = inode->i_mapping;
1713 struct shmem_inode_info *info = SHMEM_I(inode);
1714 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1719 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1720 swap = radix_to_swp_entry(*pagep);
1723 /* Look it up and read it in.. */
1724 page = lookup_swap_cache(swap, NULL, 0);
1726 /* Or update major stats only when swapin succeeds?? */
1728 *fault_type |= VM_FAULT_MAJOR;
1729 count_vm_event(PGMAJFAULT);
1730 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1732 /* Here we actually start the io */
1733 page = shmem_swapin(swap, gfp, info, index);
1740 /* We have to do this with page locked to prevent races */
1742 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1743 !shmem_confirm_swap(mapping, index, swap)) {
1747 if (!PageUptodate(page)) {
1751 wait_on_page_writeback(page);
1754 * Some architectures may have to restore extra metadata to the
1755 * physical page after reading from swap.
1757 arch_swap_restore(swap, page);
1759 if (shmem_should_replace_page(page, gfp)) {
1760 error = shmem_replace_page(&page, gfp, info, index);
1765 error = shmem_add_to_page_cache(page, mapping, index,
1766 swp_to_radix_entry(swap), gfp,
1771 spin_lock_irq(&info->lock);
1773 shmem_recalc_inode(inode);
1774 spin_unlock_irq(&info->lock);
1776 if (sgp == SGP_WRITE)
1777 mark_page_accessed(page);
1779 delete_from_swap_cache(page);
1780 set_page_dirty(page);
1786 if (!shmem_confirm_swap(mapping, index, swap))
1798 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1800 * If we allocate a new one we do not mark it dirty. That's up to the
1801 * vm. If we swap it in we mark it dirty since we also free the swap
1802 * entry since a page cannot live in both the swap and page cache.
1804 * vma, vmf, and fault_type are only supplied by shmem_fault:
1805 * otherwise they are NULL.
1807 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1808 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1809 struct vm_area_struct *vma, struct vm_fault *vmf,
1810 vm_fault_t *fault_type)
1812 struct address_space *mapping = inode->i_mapping;
1813 struct shmem_inode_info *info = SHMEM_I(inode);
1814 struct shmem_sb_info *sbinfo;
1815 struct mm_struct *charge_mm;
1817 pgoff_t hindex = index;
1823 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1826 if (sgp <= SGP_CACHE &&
1827 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1831 sbinfo = SHMEM_SB(inode->i_sb);
1832 charge_mm = vma ? vma->vm_mm : NULL;
1834 page = pagecache_get_page(mapping, index,
1835 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1837 if (page && vma && userfaultfd_minor(vma)) {
1838 if (!xa_is_value(page)) {
1842 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1846 if (xa_is_value(page)) {
1847 error = shmem_swapin_page(inode, index, &page,
1848 sgp, gfp, vma, fault_type);
1849 if (error == -EEXIST)
1857 hindex = page->index;
1858 if (sgp == SGP_WRITE)
1859 mark_page_accessed(page);
1860 if (PageUptodate(page))
1862 /* fallocated page */
1863 if (sgp != SGP_READ)
1870 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1871 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1874 if (sgp == SGP_READ)
1876 if (sgp == SGP_NOALLOC)
1880 * Fast cache lookup and swap lookup did not find it: allocate.
1883 if (vma && userfaultfd_missing(vma)) {
1884 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1888 /* Never use a huge page for shmem_symlink() */
1889 if (S_ISLNK(inode->i_mode))
1891 if (!shmem_is_huge(vma, inode, index))
1894 huge_gfp = vma_thp_gfp_mask(vma);
1895 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1896 page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true);
1899 page = shmem_alloc_and_acct_page(gfp, inode,
1905 error = PTR_ERR(page);
1907 if (error != -ENOSPC)
1910 * Try to reclaim some space by splitting a huge page
1911 * beyond i_size on the filesystem.
1916 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1917 if (ret == SHRINK_STOP)
1925 if (PageTransHuge(page))
1926 hindex = round_down(index, HPAGE_PMD_NR);
1930 if (sgp == SGP_WRITE)
1931 __SetPageReferenced(page);
1933 error = shmem_add_to_page_cache(page, mapping, hindex,
1934 NULL, gfp & GFP_RECLAIM_MASK,
1938 lru_cache_add(page);
1940 spin_lock_irq(&info->lock);
1941 info->alloced += compound_nr(page);
1942 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1943 shmem_recalc_inode(inode);
1944 spin_unlock_irq(&info->lock);
1947 if (PageTransHuge(page) &&
1948 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1949 hindex + HPAGE_PMD_NR - 1) {
1951 * Part of the huge page is beyond i_size: subject
1952 * to shrink under memory pressure.
1954 spin_lock(&sbinfo->shrinklist_lock);
1956 * _careful to defend against unlocked access to
1957 * ->shrink_list in shmem_unused_huge_shrink()
1959 if (list_empty_careful(&info->shrinklist)) {
1960 list_add_tail(&info->shrinklist,
1961 &sbinfo->shrinklist);
1962 sbinfo->shrinklist_len++;
1964 spin_unlock(&sbinfo->shrinklist_lock);
1968 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1970 if (sgp == SGP_FALLOC)
1974 * Let SGP_WRITE caller clear ends if write does not fill page;
1975 * but SGP_FALLOC on a page fallocated earlier must initialize
1976 * it now, lest undo on failure cancel our earlier guarantee.
1978 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1981 for (i = 0; i < compound_nr(page); i++) {
1982 clear_highpage(page + i);
1983 flush_dcache_page(page + i);
1985 SetPageUptodate(page);
1988 /* Perhaps the file has been truncated since we checked */
1989 if (sgp <= SGP_CACHE &&
1990 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1992 ClearPageDirty(page);
1993 delete_from_page_cache(page);
1994 spin_lock_irq(&info->lock);
1995 shmem_recalc_inode(inode);
1996 spin_unlock_irq(&info->lock);
2002 *pagep = page + index - hindex;
2009 shmem_inode_unacct_blocks(inode, compound_nr(page));
2011 if (PageTransHuge(page)) {
2021 if (error == -ENOSPC && !once++) {
2022 spin_lock_irq(&info->lock);
2023 shmem_recalc_inode(inode);
2024 spin_unlock_irq(&info->lock);
2027 if (error == -EEXIST)
2033 * This is like autoremove_wake_function, but it removes the wait queue
2034 * entry unconditionally - even if something else had already woken the
2037 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2039 int ret = default_wake_function(wait, mode, sync, key);
2040 list_del_init(&wait->entry);
2044 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2046 struct vm_area_struct *vma = vmf->vma;
2047 struct inode *inode = file_inode(vma->vm_file);
2048 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2050 vm_fault_t ret = VM_FAULT_LOCKED;
2053 * Trinity finds that probing a hole which tmpfs is punching can
2054 * prevent the hole-punch from ever completing: which in turn
2055 * locks writers out with its hold on i_mutex. So refrain from
2056 * faulting pages into the hole while it's being punched. Although
2057 * shmem_undo_range() does remove the additions, it may be unable to
2058 * keep up, as each new page needs its own unmap_mapping_range() call,
2059 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2061 * It does not matter if we sometimes reach this check just before the
2062 * hole-punch begins, so that one fault then races with the punch:
2063 * we just need to make racing faults a rare case.
2065 * The implementation below would be much simpler if we just used a
2066 * standard mutex or completion: but we cannot take i_mutex in fault,
2067 * and bloating every shmem inode for this unlikely case would be sad.
2069 if (unlikely(inode->i_private)) {
2070 struct shmem_falloc *shmem_falloc;
2072 spin_lock(&inode->i_lock);
2073 shmem_falloc = inode->i_private;
2075 shmem_falloc->waitq &&
2076 vmf->pgoff >= shmem_falloc->start &&
2077 vmf->pgoff < shmem_falloc->next) {
2079 wait_queue_head_t *shmem_falloc_waitq;
2080 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2082 ret = VM_FAULT_NOPAGE;
2083 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2085 ret = VM_FAULT_RETRY;
2087 shmem_falloc_waitq = shmem_falloc->waitq;
2088 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2089 TASK_UNINTERRUPTIBLE);
2090 spin_unlock(&inode->i_lock);
2094 * shmem_falloc_waitq points into the shmem_fallocate()
2095 * stack of the hole-punching task: shmem_falloc_waitq
2096 * is usually invalid by the time we reach here, but
2097 * finish_wait() does not dereference it in that case;
2098 * though i_lock needed lest racing with wake_up_all().
2100 spin_lock(&inode->i_lock);
2101 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2102 spin_unlock(&inode->i_lock);
2108 spin_unlock(&inode->i_lock);
2111 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE,
2112 gfp, vma, vmf, &ret);
2114 return vmf_error(err);
2118 unsigned long shmem_get_unmapped_area(struct file *file,
2119 unsigned long uaddr, unsigned long len,
2120 unsigned long pgoff, unsigned long flags)
2122 unsigned long (*get_area)(struct file *,
2123 unsigned long, unsigned long, unsigned long, unsigned long);
2125 unsigned long offset;
2126 unsigned long inflated_len;
2127 unsigned long inflated_addr;
2128 unsigned long inflated_offset;
2130 if (len > TASK_SIZE)
2133 get_area = current->mm->get_unmapped_area;
2134 addr = get_area(file, uaddr, len, pgoff, flags);
2136 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2138 if (IS_ERR_VALUE(addr))
2140 if (addr & ~PAGE_MASK)
2142 if (addr > TASK_SIZE - len)
2145 if (shmem_huge == SHMEM_HUGE_DENY)
2147 if (len < HPAGE_PMD_SIZE)
2149 if (flags & MAP_FIXED)
2152 * Our priority is to support MAP_SHARED mapped hugely;
2153 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2154 * But if caller specified an address hint and we allocated area there
2155 * successfully, respect that as before.
2160 if (shmem_huge != SHMEM_HUGE_FORCE) {
2161 struct super_block *sb;
2164 VM_BUG_ON(file->f_op != &shmem_file_operations);
2165 sb = file_inode(file)->i_sb;
2168 * Called directly from mm/mmap.c, or drivers/char/mem.c
2169 * for "/dev/zero", to create a shared anonymous object.
2171 if (IS_ERR(shm_mnt))
2173 sb = shm_mnt->mnt_sb;
2175 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2179 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2180 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2182 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2185 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2186 if (inflated_len > TASK_SIZE)
2188 if (inflated_len < len)
2191 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2192 if (IS_ERR_VALUE(inflated_addr))
2194 if (inflated_addr & ~PAGE_MASK)
2197 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2198 inflated_addr += offset - inflated_offset;
2199 if (inflated_offset > offset)
2200 inflated_addr += HPAGE_PMD_SIZE;
2202 if (inflated_addr > TASK_SIZE - len)
2204 return inflated_addr;
2208 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2210 struct inode *inode = file_inode(vma->vm_file);
2211 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2214 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2217 struct inode *inode = file_inode(vma->vm_file);
2220 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2221 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2225 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2227 struct inode *inode = file_inode(file);
2228 struct shmem_inode_info *info = SHMEM_I(inode);
2229 int retval = -ENOMEM;
2232 * What serializes the accesses to info->flags?
2233 * ipc_lock_object() when called from shmctl_do_lock(),
2234 * no serialization needed when called from shm_destroy().
2236 if (lock && !(info->flags & VM_LOCKED)) {
2237 if (!user_shm_lock(inode->i_size, ucounts))
2239 info->flags |= VM_LOCKED;
2240 mapping_set_unevictable(file->f_mapping);
2242 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2243 user_shm_unlock(inode->i_size, ucounts);
2244 info->flags &= ~VM_LOCKED;
2245 mapping_clear_unevictable(file->f_mapping);
2253 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2255 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2258 ret = seal_check_future_write(info->seals, vma);
2262 /* arm64 - allow memory tagging on RAM-based files */
2263 vma->vm_flags |= VM_MTE_ALLOWED;
2265 file_accessed(file);
2266 vma->vm_ops = &shmem_vm_ops;
2267 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2268 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2269 (vma->vm_end & HPAGE_PMD_MASK)) {
2270 khugepaged_enter(vma, vma->vm_flags);
2275 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2276 umode_t mode, dev_t dev, unsigned long flags)
2278 struct inode *inode;
2279 struct shmem_inode_info *info;
2280 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2283 if (shmem_reserve_inode(sb, &ino))
2286 inode = new_inode(sb);
2289 inode_init_owner(&init_user_ns, inode, dir, mode);
2290 inode->i_blocks = 0;
2291 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2292 inode->i_generation = prandom_u32();
2293 info = SHMEM_I(inode);
2294 memset(info, 0, (char *)inode - (char *)info);
2295 spin_lock_init(&info->lock);
2296 atomic_set(&info->stop_eviction, 0);
2297 info->seals = F_SEAL_SEAL;
2298 info->flags = flags & VM_NORESERVE;
2299 INIT_LIST_HEAD(&info->shrinklist);
2300 INIT_LIST_HEAD(&info->swaplist);
2301 simple_xattrs_init(&info->xattrs);
2302 cache_no_acl(inode);
2304 switch (mode & S_IFMT) {
2306 inode->i_op = &shmem_special_inode_operations;
2307 init_special_inode(inode, mode, dev);
2310 inode->i_mapping->a_ops = &shmem_aops;
2311 inode->i_op = &shmem_inode_operations;
2312 inode->i_fop = &shmem_file_operations;
2313 mpol_shared_policy_init(&info->policy,
2314 shmem_get_sbmpol(sbinfo));
2318 /* Some things misbehave if size == 0 on a directory */
2319 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2320 inode->i_op = &shmem_dir_inode_operations;
2321 inode->i_fop = &simple_dir_operations;
2325 * Must not load anything in the rbtree,
2326 * mpol_free_shared_policy will not be called.
2328 mpol_shared_policy_init(&info->policy, NULL);
2332 lockdep_annotate_inode_mutex_key(inode);
2334 shmem_free_inode(sb);
2338 #ifdef CONFIG_USERFAULTFD
2339 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2341 struct vm_area_struct *dst_vma,
2342 unsigned long dst_addr,
2343 unsigned long src_addr,
2345 struct page **pagep)
2347 struct inode *inode = file_inode(dst_vma->vm_file);
2348 struct shmem_inode_info *info = SHMEM_I(inode);
2349 struct address_space *mapping = inode->i_mapping;
2350 gfp_t gfp = mapping_gfp_mask(mapping);
2351 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2357 if (!shmem_inode_acct_block(inode, 1)) {
2359 * We may have got a page, returned -ENOENT triggering a retry,
2360 * and now we find ourselves with -ENOMEM. Release the page, to
2361 * avoid a BUG_ON in our caller.
2363 if (unlikely(*pagep)) {
2372 page = shmem_alloc_page(gfp, info, pgoff);
2374 goto out_unacct_blocks;
2376 if (!zeropage) { /* COPY */
2377 page_kaddr = kmap_atomic(page);
2378 ret = copy_from_user(page_kaddr,
2379 (const void __user *)src_addr,
2381 kunmap_atomic(page_kaddr);
2383 /* fallback to copy_from_user outside mmap_lock */
2384 if (unlikely(ret)) {
2387 /* don't free the page */
2388 goto out_unacct_blocks;
2390 } else { /* ZEROPAGE */
2391 clear_highpage(page);
2398 VM_BUG_ON(PageLocked(page));
2399 VM_BUG_ON(PageSwapBacked(page));
2400 __SetPageLocked(page);
2401 __SetPageSwapBacked(page);
2402 __SetPageUptodate(page);
2405 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2406 if (unlikely(pgoff >= max_off))
2409 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2410 gfp & GFP_RECLAIM_MASK, dst_mm);
2414 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2417 goto out_delete_from_cache;
2419 spin_lock_irq(&info->lock);
2421 inode->i_blocks += BLOCKS_PER_PAGE;
2422 shmem_recalc_inode(inode);
2423 spin_unlock_irq(&info->lock);
2428 out_delete_from_cache:
2429 delete_from_page_cache(page);
2434 shmem_inode_unacct_blocks(inode, 1);
2437 #endif /* CONFIG_USERFAULTFD */
2440 static const struct inode_operations shmem_symlink_inode_operations;
2441 static const struct inode_operations shmem_short_symlink_operations;
2443 #ifdef CONFIG_TMPFS_XATTR
2444 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2446 #define shmem_initxattrs NULL
2450 shmem_write_begin(struct file *file, struct address_space *mapping,
2451 loff_t pos, unsigned len, unsigned flags,
2452 struct page **pagep, void **fsdata)
2454 struct inode *inode = mapping->host;
2455 struct shmem_inode_info *info = SHMEM_I(inode);
2456 pgoff_t index = pos >> PAGE_SHIFT;
2458 /* i_mutex is held by caller */
2459 if (unlikely(info->seals & (F_SEAL_GROW |
2460 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2461 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2463 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2467 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2471 shmem_write_end(struct file *file, struct address_space *mapping,
2472 loff_t pos, unsigned len, unsigned copied,
2473 struct page *page, void *fsdata)
2475 struct inode *inode = mapping->host;
2477 if (pos + copied > inode->i_size)
2478 i_size_write(inode, pos + copied);
2480 if (!PageUptodate(page)) {
2481 struct page *head = compound_head(page);
2482 if (PageTransCompound(page)) {
2485 for (i = 0; i < HPAGE_PMD_NR; i++) {
2486 if (head + i == page)
2488 clear_highpage(head + i);
2489 flush_dcache_page(head + i);
2492 if (copied < PAGE_SIZE) {
2493 unsigned from = pos & (PAGE_SIZE - 1);
2494 zero_user_segments(page, 0, from,
2495 from + copied, PAGE_SIZE);
2497 SetPageUptodate(head);
2499 set_page_dirty(page);
2506 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2508 struct file *file = iocb->ki_filp;
2509 struct inode *inode = file_inode(file);
2510 struct address_space *mapping = inode->i_mapping;
2512 unsigned long offset;
2513 enum sgp_type sgp = SGP_READ;
2516 loff_t *ppos = &iocb->ki_pos;
2519 * Might this read be for a stacking filesystem? Then when reading
2520 * holes of a sparse file, we actually need to allocate those pages,
2521 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2523 if (!iter_is_iovec(to))
2526 index = *ppos >> PAGE_SHIFT;
2527 offset = *ppos & ~PAGE_MASK;
2530 struct page *page = NULL;
2532 unsigned long nr, ret;
2533 loff_t i_size = i_size_read(inode);
2535 end_index = i_size >> PAGE_SHIFT;
2536 if (index > end_index)
2538 if (index == end_index) {
2539 nr = i_size & ~PAGE_MASK;
2544 error = shmem_getpage(inode, index, &page, sgp);
2546 if (error == -EINVAL)
2551 if (sgp == SGP_CACHE)
2552 set_page_dirty(page);
2557 * We must evaluate after, since reads (unlike writes)
2558 * are called without i_mutex protection against truncate
2561 i_size = i_size_read(inode);
2562 end_index = i_size >> PAGE_SHIFT;
2563 if (index == end_index) {
2564 nr = i_size & ~PAGE_MASK;
2575 * If users can be writing to this page using arbitrary
2576 * virtual addresses, take care about potential aliasing
2577 * before reading the page on the kernel side.
2579 if (mapping_writably_mapped(mapping))
2580 flush_dcache_page(page);
2582 * Mark the page accessed if we read the beginning.
2585 mark_page_accessed(page);
2587 page = ZERO_PAGE(0);
2592 * Ok, we have the page, and it's up-to-date, so
2593 * now we can copy it to user space...
2595 ret = copy_page_to_iter(page, offset, nr, to);
2598 index += offset >> PAGE_SHIFT;
2599 offset &= ~PAGE_MASK;
2602 if (!iov_iter_count(to))
2611 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2612 file_accessed(file);
2613 return retval ? retval : error;
2616 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2618 struct address_space *mapping = file->f_mapping;
2619 struct inode *inode = mapping->host;
2621 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2622 return generic_file_llseek_size(file, offset, whence,
2623 MAX_LFS_FILESIZE, i_size_read(inode));
2628 /* We're holding i_mutex so we can access i_size directly */
2629 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2631 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2632 inode_unlock(inode);
2636 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2639 struct inode *inode = file_inode(file);
2640 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2641 struct shmem_inode_info *info = SHMEM_I(inode);
2642 struct shmem_falloc shmem_falloc;
2643 pgoff_t start, index, end, undo_fallocend;
2646 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2651 if (mode & FALLOC_FL_PUNCH_HOLE) {
2652 struct address_space *mapping = file->f_mapping;
2653 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2654 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2655 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2657 /* protected by i_mutex */
2658 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2663 shmem_falloc.waitq = &shmem_falloc_waitq;
2664 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2665 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2666 spin_lock(&inode->i_lock);
2667 inode->i_private = &shmem_falloc;
2668 spin_unlock(&inode->i_lock);
2670 if ((u64)unmap_end > (u64)unmap_start)
2671 unmap_mapping_range(mapping, unmap_start,
2672 1 + unmap_end - unmap_start, 0);
2673 shmem_truncate_range(inode, offset, offset + len - 1);
2674 /* No need to unmap again: hole-punching leaves COWed pages */
2676 spin_lock(&inode->i_lock);
2677 inode->i_private = NULL;
2678 wake_up_all(&shmem_falloc_waitq);
2679 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2680 spin_unlock(&inode->i_lock);
2685 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2686 error = inode_newsize_ok(inode, offset + len);
2690 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2695 start = offset >> PAGE_SHIFT;
2696 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2697 /* Try to avoid a swapstorm if len is impossible to satisfy */
2698 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2703 shmem_falloc.waitq = NULL;
2704 shmem_falloc.start = start;
2705 shmem_falloc.next = start;
2706 shmem_falloc.nr_falloced = 0;
2707 shmem_falloc.nr_unswapped = 0;
2708 spin_lock(&inode->i_lock);
2709 inode->i_private = &shmem_falloc;
2710 spin_unlock(&inode->i_lock);
2713 * info->fallocend is only relevant when huge pages might be
2714 * involved: to prevent split_huge_page() freeing fallocated
2715 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2717 undo_fallocend = info->fallocend;
2718 if (info->fallocend < end)
2719 info->fallocend = end;
2721 for (index = start; index < end; ) {
2725 * Good, the fallocate(2) manpage permits EINTR: we may have
2726 * been interrupted because we are using up too much memory.
2728 if (signal_pending(current))
2730 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2733 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2735 info->fallocend = undo_fallocend;
2736 /* Remove the !PageUptodate pages we added */
2737 if (index > start) {
2738 shmem_undo_range(inode,
2739 (loff_t)start << PAGE_SHIFT,
2740 ((loff_t)index << PAGE_SHIFT) - 1, true);
2747 * Here is a more important optimization than it appears:
2748 * a second SGP_FALLOC on the same huge page will clear it,
2749 * making it PageUptodate and un-undoable if we fail later.
2751 if (PageTransCompound(page)) {
2752 index = round_up(index, HPAGE_PMD_NR);
2753 /* Beware 32-bit wraparound */
2759 * Inform shmem_writepage() how far we have reached.
2760 * No need for lock or barrier: we have the page lock.
2762 if (!PageUptodate(page))
2763 shmem_falloc.nr_falloced += index - shmem_falloc.next;
2764 shmem_falloc.next = index;
2767 * If !PageUptodate, leave it that way so that freeable pages
2768 * can be recognized if we need to rollback on error later.
2769 * But set_page_dirty so that memory pressure will swap rather
2770 * than free the pages we are allocating (and SGP_CACHE pages
2771 * might still be clean: we now need to mark those dirty too).
2773 set_page_dirty(page);
2779 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2780 i_size_write(inode, offset + len);
2781 inode->i_ctime = current_time(inode);
2783 spin_lock(&inode->i_lock);
2784 inode->i_private = NULL;
2785 spin_unlock(&inode->i_lock);
2787 inode_unlock(inode);
2791 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2793 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2795 buf->f_type = TMPFS_MAGIC;
2796 buf->f_bsize = PAGE_SIZE;
2797 buf->f_namelen = NAME_MAX;
2798 if (sbinfo->max_blocks) {
2799 buf->f_blocks = sbinfo->max_blocks;
2801 buf->f_bfree = sbinfo->max_blocks -
2802 percpu_counter_sum(&sbinfo->used_blocks);
2804 if (sbinfo->max_inodes) {
2805 buf->f_files = sbinfo->max_inodes;
2806 buf->f_ffree = sbinfo->free_inodes;
2808 /* else leave those fields 0 like simple_statfs */
2810 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2816 * File creation. Allocate an inode, and we're done..
2819 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2820 struct dentry *dentry, umode_t mode, dev_t dev)
2822 struct inode *inode;
2823 int error = -ENOSPC;
2825 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2827 error = simple_acl_create(dir, inode);
2830 error = security_inode_init_security(inode, dir,
2832 shmem_initxattrs, NULL);
2833 if (error && error != -EOPNOTSUPP)
2837 dir->i_size += BOGO_DIRENT_SIZE;
2838 dir->i_ctime = dir->i_mtime = current_time(dir);
2839 d_instantiate(dentry, inode);
2840 dget(dentry); /* Extra count - pin the dentry in core */
2849 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2850 struct dentry *dentry, umode_t mode)
2852 struct inode *inode;
2853 int error = -ENOSPC;
2855 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2857 error = security_inode_init_security(inode, dir,
2859 shmem_initxattrs, NULL);
2860 if (error && error != -EOPNOTSUPP)
2862 error = simple_acl_create(dir, inode);
2865 d_tmpfile(dentry, inode);
2873 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2874 struct dentry *dentry, umode_t mode)
2878 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2879 mode | S_IFDIR, 0)))
2885 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2886 struct dentry *dentry, umode_t mode, bool excl)
2888 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2894 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2896 struct inode *inode = d_inode(old_dentry);
2900 * No ordinary (disk based) filesystem counts links as inodes;
2901 * but each new link needs a new dentry, pinning lowmem, and
2902 * tmpfs dentries cannot be pruned until they are unlinked.
2903 * But if an O_TMPFILE file is linked into the tmpfs, the
2904 * first link must skip that, to get the accounting right.
2906 if (inode->i_nlink) {
2907 ret = shmem_reserve_inode(inode->i_sb, NULL);
2912 dir->i_size += BOGO_DIRENT_SIZE;
2913 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2915 ihold(inode); /* New dentry reference */
2916 dget(dentry); /* Extra pinning count for the created dentry */
2917 d_instantiate(dentry, inode);
2922 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2924 struct inode *inode = d_inode(dentry);
2926 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2927 shmem_free_inode(inode->i_sb);
2929 dir->i_size -= BOGO_DIRENT_SIZE;
2930 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2932 dput(dentry); /* Undo the count from "create" - this does all the work */
2936 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2938 if (!simple_empty(dentry))
2941 drop_nlink(d_inode(dentry));
2943 return shmem_unlink(dir, dentry);
2946 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2948 bool old_is_dir = d_is_dir(old_dentry);
2949 bool new_is_dir = d_is_dir(new_dentry);
2951 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2953 drop_nlink(old_dir);
2956 drop_nlink(new_dir);
2960 old_dir->i_ctime = old_dir->i_mtime =
2961 new_dir->i_ctime = new_dir->i_mtime =
2962 d_inode(old_dentry)->i_ctime =
2963 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2968 static int shmem_whiteout(struct user_namespace *mnt_userns,
2969 struct inode *old_dir, struct dentry *old_dentry)
2971 struct dentry *whiteout;
2974 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2978 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2979 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2985 * Cheat and hash the whiteout while the old dentry is still in
2986 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2988 * d_lookup() will consistently find one of them at this point,
2989 * not sure which one, but that isn't even important.
2996 * The VFS layer already does all the dentry stuff for rename,
2997 * we just have to decrement the usage count for the target if
2998 * it exists so that the VFS layer correctly free's it when it
3001 static int shmem_rename2(struct user_namespace *mnt_userns,
3002 struct inode *old_dir, struct dentry *old_dentry,
3003 struct inode *new_dir, struct dentry *new_dentry,
3006 struct inode *inode = d_inode(old_dentry);
3007 int they_are_dirs = S_ISDIR(inode->i_mode);
3009 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3012 if (flags & RENAME_EXCHANGE)
3013 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3015 if (!simple_empty(new_dentry))
3018 if (flags & RENAME_WHITEOUT) {
3021 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3026 if (d_really_is_positive(new_dentry)) {
3027 (void) shmem_unlink(new_dir, new_dentry);
3028 if (they_are_dirs) {
3029 drop_nlink(d_inode(new_dentry));
3030 drop_nlink(old_dir);
3032 } else if (they_are_dirs) {
3033 drop_nlink(old_dir);
3037 old_dir->i_size -= BOGO_DIRENT_SIZE;
3038 new_dir->i_size += BOGO_DIRENT_SIZE;
3039 old_dir->i_ctime = old_dir->i_mtime =
3040 new_dir->i_ctime = new_dir->i_mtime =
3041 inode->i_ctime = current_time(old_dir);
3045 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3046 struct dentry *dentry, const char *symname)
3050 struct inode *inode;
3053 len = strlen(symname) + 1;
3054 if (len > PAGE_SIZE)
3055 return -ENAMETOOLONG;
3057 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3062 error = security_inode_init_security(inode, dir, &dentry->d_name,
3063 shmem_initxattrs, NULL);
3064 if (error && error != -EOPNOTSUPP) {
3069 inode->i_size = len-1;
3070 if (len <= SHORT_SYMLINK_LEN) {
3071 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3072 if (!inode->i_link) {
3076 inode->i_op = &shmem_short_symlink_operations;
3078 inode_nohighmem(inode);
3079 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3084 inode->i_mapping->a_ops = &shmem_aops;
3085 inode->i_op = &shmem_symlink_inode_operations;
3086 memcpy(page_address(page), symname, len);
3087 SetPageUptodate(page);
3088 set_page_dirty(page);
3092 dir->i_size += BOGO_DIRENT_SIZE;
3093 dir->i_ctime = dir->i_mtime = current_time(dir);
3094 d_instantiate(dentry, inode);
3099 static void shmem_put_link(void *arg)
3101 mark_page_accessed(arg);
3105 static const char *shmem_get_link(struct dentry *dentry,
3106 struct inode *inode,
3107 struct delayed_call *done)
3109 struct page *page = NULL;
3112 page = find_get_page(inode->i_mapping, 0);
3114 return ERR_PTR(-ECHILD);
3115 if (!PageUptodate(page)) {
3117 return ERR_PTR(-ECHILD);
3120 error = shmem_getpage(inode, 0, &page, SGP_READ);
3122 return ERR_PTR(error);
3125 set_delayed_call(done, shmem_put_link, page);
3126 return page_address(page);
3129 #ifdef CONFIG_TMPFS_XATTR
3131 * Superblocks without xattr inode operations may get some security.* xattr
3132 * support from the LSM "for free". As soon as we have any other xattrs
3133 * like ACLs, we also need to implement the security.* handlers at
3134 * filesystem level, though.
3138 * Callback for security_inode_init_security() for acquiring xattrs.
3140 static int shmem_initxattrs(struct inode *inode,
3141 const struct xattr *xattr_array,
3144 struct shmem_inode_info *info = SHMEM_I(inode);
3145 const struct xattr *xattr;
3146 struct simple_xattr *new_xattr;
3149 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3150 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3154 len = strlen(xattr->name) + 1;
3155 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3157 if (!new_xattr->name) {
3162 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3163 XATTR_SECURITY_PREFIX_LEN);
3164 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3167 simple_xattr_list_add(&info->xattrs, new_xattr);
3173 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3174 struct dentry *unused, struct inode *inode,
3175 const char *name, void *buffer, size_t size)
3177 struct shmem_inode_info *info = SHMEM_I(inode);
3179 name = xattr_full_name(handler, name);
3180 return simple_xattr_get(&info->xattrs, name, buffer, size);
3183 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3184 struct user_namespace *mnt_userns,
3185 struct dentry *unused, struct inode *inode,
3186 const char *name, const void *value,
3187 size_t size, int flags)
3189 struct shmem_inode_info *info = SHMEM_I(inode);
3191 name = xattr_full_name(handler, name);
3192 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3195 static const struct xattr_handler shmem_security_xattr_handler = {
3196 .prefix = XATTR_SECURITY_PREFIX,
3197 .get = shmem_xattr_handler_get,
3198 .set = shmem_xattr_handler_set,
3201 static const struct xattr_handler shmem_trusted_xattr_handler = {
3202 .prefix = XATTR_TRUSTED_PREFIX,
3203 .get = shmem_xattr_handler_get,
3204 .set = shmem_xattr_handler_set,
3207 static const struct xattr_handler *shmem_xattr_handlers[] = {
3208 #ifdef CONFIG_TMPFS_POSIX_ACL
3209 &posix_acl_access_xattr_handler,
3210 &posix_acl_default_xattr_handler,
3212 &shmem_security_xattr_handler,
3213 &shmem_trusted_xattr_handler,
3217 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3219 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3220 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3222 #endif /* CONFIG_TMPFS_XATTR */
3224 static const struct inode_operations shmem_short_symlink_operations = {
3225 .get_link = simple_get_link,
3226 #ifdef CONFIG_TMPFS_XATTR
3227 .listxattr = shmem_listxattr,
3231 static const struct inode_operations shmem_symlink_inode_operations = {
3232 .get_link = shmem_get_link,
3233 #ifdef CONFIG_TMPFS_XATTR
3234 .listxattr = shmem_listxattr,
3238 static struct dentry *shmem_get_parent(struct dentry *child)
3240 return ERR_PTR(-ESTALE);
3243 static int shmem_match(struct inode *ino, void *vfh)
3247 inum = (inum << 32) | fh[1];
3248 return ino->i_ino == inum && fh[0] == ino->i_generation;
3251 /* Find any alias of inode, but prefer a hashed alias */
3252 static struct dentry *shmem_find_alias(struct inode *inode)
3254 struct dentry *alias = d_find_alias(inode);
3256 return alias ?: d_find_any_alias(inode);
3260 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3261 struct fid *fid, int fh_len, int fh_type)
3263 struct inode *inode;
3264 struct dentry *dentry = NULL;
3271 inum = (inum << 32) | fid->raw[1];
3273 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3274 shmem_match, fid->raw);
3276 dentry = shmem_find_alias(inode);
3283 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3284 struct inode *parent)
3288 return FILEID_INVALID;
3291 if (inode_unhashed(inode)) {
3292 /* Unfortunately insert_inode_hash is not idempotent,
3293 * so as we hash inodes here rather than at creation
3294 * time, we need a lock to ensure we only try
3297 static DEFINE_SPINLOCK(lock);
3299 if (inode_unhashed(inode))
3300 __insert_inode_hash(inode,
3301 inode->i_ino + inode->i_generation);
3305 fh[0] = inode->i_generation;
3306 fh[1] = inode->i_ino;
3307 fh[2] = ((__u64)inode->i_ino) >> 32;
3313 static const struct export_operations shmem_export_ops = {
3314 .get_parent = shmem_get_parent,
3315 .encode_fh = shmem_encode_fh,
3316 .fh_to_dentry = shmem_fh_to_dentry,
3332 static const struct constant_table shmem_param_enums_huge[] = {
3333 {"never", SHMEM_HUGE_NEVER },
3334 {"always", SHMEM_HUGE_ALWAYS },
3335 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3336 {"advise", SHMEM_HUGE_ADVISE },
3340 const struct fs_parameter_spec shmem_fs_parameters[] = {
3341 fsparam_u32 ("gid", Opt_gid),
3342 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3343 fsparam_u32oct("mode", Opt_mode),
3344 fsparam_string("mpol", Opt_mpol),
3345 fsparam_string("nr_blocks", Opt_nr_blocks),
3346 fsparam_string("nr_inodes", Opt_nr_inodes),
3347 fsparam_string("size", Opt_size),
3348 fsparam_u32 ("uid", Opt_uid),
3349 fsparam_flag ("inode32", Opt_inode32),
3350 fsparam_flag ("inode64", Opt_inode64),
3354 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3356 struct shmem_options *ctx = fc->fs_private;
3357 struct fs_parse_result result;
3358 unsigned long long size;
3362 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3368 size = memparse(param->string, &rest);
3370 size <<= PAGE_SHIFT;
3371 size *= totalram_pages();
3377 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3378 ctx->seen |= SHMEM_SEEN_BLOCKS;
3381 ctx->blocks = memparse(param->string, &rest);
3384 ctx->seen |= SHMEM_SEEN_BLOCKS;
3387 ctx->inodes = memparse(param->string, &rest);
3390 ctx->seen |= SHMEM_SEEN_INODES;
3393 ctx->mode = result.uint_32 & 07777;
3396 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3397 if (!uid_valid(ctx->uid))
3401 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3402 if (!gid_valid(ctx->gid))
3406 ctx->huge = result.uint_32;
3407 if (ctx->huge != SHMEM_HUGE_NEVER &&
3408 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3409 has_transparent_hugepage()))
3410 goto unsupported_parameter;
3411 ctx->seen |= SHMEM_SEEN_HUGE;
3414 if (IS_ENABLED(CONFIG_NUMA)) {
3415 mpol_put(ctx->mpol);
3417 if (mpol_parse_str(param->string, &ctx->mpol))
3421 goto unsupported_parameter;
3423 ctx->full_inums = false;
3424 ctx->seen |= SHMEM_SEEN_INUMS;
3427 if (sizeof(ino_t) < 8) {
3429 "Cannot use inode64 with <64bit inums in kernel\n");
3431 ctx->full_inums = true;
3432 ctx->seen |= SHMEM_SEEN_INUMS;
3437 unsupported_parameter:
3438 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3440 return invalfc(fc, "Bad value for '%s'", param->key);
3443 static int shmem_parse_options(struct fs_context *fc, void *data)
3445 char *options = data;
3448 int err = security_sb_eat_lsm_opts(options, &fc->security);
3453 while (options != NULL) {
3454 char *this_char = options;
3457 * NUL-terminate this option: unfortunately,
3458 * mount options form a comma-separated list,
3459 * but mpol's nodelist may also contain commas.
3461 options = strchr(options, ',');
3462 if (options == NULL)
3465 if (!isdigit(*options)) {
3471 char *value = strchr(this_char, '=');
3477 len = strlen(value);
3479 err = vfs_parse_fs_string(fc, this_char, value, len);
3488 * Reconfigure a shmem filesystem.
3490 * Note that we disallow change from limited->unlimited blocks/inodes while any
3491 * are in use; but we must separately disallow unlimited->limited, because in
3492 * that case we have no record of how much is already in use.
3494 static int shmem_reconfigure(struct fs_context *fc)
3496 struct shmem_options *ctx = fc->fs_private;
3497 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3498 unsigned long inodes;
3499 struct mempolicy *mpol = NULL;
3502 raw_spin_lock(&sbinfo->stat_lock);
3503 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3504 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3505 if (!sbinfo->max_blocks) {
3506 err = "Cannot retroactively limit size";
3509 if (percpu_counter_compare(&sbinfo->used_blocks,
3511 err = "Too small a size for current use";
3515 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3516 if (!sbinfo->max_inodes) {
3517 err = "Cannot retroactively limit inodes";
3520 if (ctx->inodes < inodes) {
3521 err = "Too few inodes for current use";
3526 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3527 sbinfo->next_ino > UINT_MAX) {
3528 err = "Current inum too high to switch to 32-bit inums";
3532 if (ctx->seen & SHMEM_SEEN_HUGE)
3533 sbinfo->huge = ctx->huge;
3534 if (ctx->seen & SHMEM_SEEN_INUMS)
3535 sbinfo->full_inums = ctx->full_inums;
3536 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3537 sbinfo->max_blocks = ctx->blocks;
3538 if (ctx->seen & SHMEM_SEEN_INODES) {
3539 sbinfo->max_inodes = ctx->inodes;
3540 sbinfo->free_inodes = ctx->inodes - inodes;
3544 * Preserve previous mempolicy unless mpol remount option was specified.
3547 mpol = sbinfo->mpol;
3548 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3551 raw_spin_unlock(&sbinfo->stat_lock);
3555 raw_spin_unlock(&sbinfo->stat_lock);
3556 return invalfc(fc, "%s", err);
3559 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3561 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3563 if (sbinfo->max_blocks != shmem_default_max_blocks())
3564 seq_printf(seq, ",size=%luk",
3565 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3566 if (sbinfo->max_inodes != shmem_default_max_inodes())
3567 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3568 if (sbinfo->mode != (0777 | S_ISVTX))
3569 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3570 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3571 seq_printf(seq, ",uid=%u",
3572 from_kuid_munged(&init_user_ns, sbinfo->uid));
3573 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3574 seq_printf(seq, ",gid=%u",
3575 from_kgid_munged(&init_user_ns, sbinfo->gid));
3578 * Showing inode{64,32} might be useful even if it's the system default,
3579 * since then people don't have to resort to checking both here and
3580 * /proc/config.gz to confirm 64-bit inums were successfully applied
3581 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3583 * We hide it when inode64 isn't the default and we are using 32-bit
3584 * inodes, since that probably just means the feature isn't even under
3589 * +-----------------+-----------------+
3590 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3591 * +------------------+-----------------+-----------------+
3592 * | full_inums=true | show | show |
3593 * | full_inums=false | show | hide |
3594 * +------------------+-----------------+-----------------+
3597 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3598 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3599 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3600 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3602 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3604 shmem_show_mpol(seq, sbinfo->mpol);
3608 #endif /* CONFIG_TMPFS */
3610 static void shmem_put_super(struct super_block *sb)
3612 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3614 free_percpu(sbinfo->ino_batch);
3615 percpu_counter_destroy(&sbinfo->used_blocks);
3616 mpol_put(sbinfo->mpol);
3618 sb->s_fs_info = NULL;
3621 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3623 struct shmem_options *ctx = fc->fs_private;
3624 struct inode *inode;
3625 struct shmem_sb_info *sbinfo;
3627 /* Round up to L1_CACHE_BYTES to resist false sharing */
3628 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3629 L1_CACHE_BYTES), GFP_KERNEL);
3633 sb->s_fs_info = sbinfo;
3637 * Per default we only allow half of the physical ram per
3638 * tmpfs instance, limiting inodes to one per page of lowmem;
3639 * but the internal instance is left unlimited.
3641 if (!(sb->s_flags & SB_KERNMOUNT)) {
3642 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3643 ctx->blocks = shmem_default_max_blocks();
3644 if (!(ctx->seen & SHMEM_SEEN_INODES))
3645 ctx->inodes = shmem_default_max_inodes();
3646 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3647 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3649 sb->s_flags |= SB_NOUSER;
3651 sb->s_export_op = &shmem_export_ops;
3652 sb->s_flags |= SB_NOSEC;
3654 sb->s_flags |= SB_NOUSER;
3656 sbinfo->max_blocks = ctx->blocks;
3657 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3658 if (sb->s_flags & SB_KERNMOUNT) {
3659 sbinfo->ino_batch = alloc_percpu(ino_t);
3660 if (!sbinfo->ino_batch)
3663 sbinfo->uid = ctx->uid;
3664 sbinfo->gid = ctx->gid;
3665 sbinfo->full_inums = ctx->full_inums;
3666 sbinfo->mode = ctx->mode;
3667 sbinfo->huge = ctx->huge;
3668 sbinfo->mpol = ctx->mpol;
3671 raw_spin_lock_init(&sbinfo->stat_lock);
3672 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3674 spin_lock_init(&sbinfo->shrinklist_lock);
3675 INIT_LIST_HEAD(&sbinfo->shrinklist);
3677 sb->s_maxbytes = MAX_LFS_FILESIZE;
3678 sb->s_blocksize = PAGE_SIZE;
3679 sb->s_blocksize_bits = PAGE_SHIFT;
3680 sb->s_magic = TMPFS_MAGIC;
3681 sb->s_op = &shmem_ops;
3682 sb->s_time_gran = 1;
3683 #ifdef CONFIG_TMPFS_XATTR
3684 sb->s_xattr = shmem_xattr_handlers;
3686 #ifdef CONFIG_TMPFS_POSIX_ACL
3687 sb->s_flags |= SB_POSIXACL;
3689 uuid_gen(&sb->s_uuid);
3691 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3694 inode->i_uid = sbinfo->uid;
3695 inode->i_gid = sbinfo->gid;
3696 sb->s_root = d_make_root(inode);
3702 shmem_put_super(sb);
3706 static int shmem_get_tree(struct fs_context *fc)
3708 return get_tree_nodev(fc, shmem_fill_super);
3711 static void shmem_free_fc(struct fs_context *fc)
3713 struct shmem_options *ctx = fc->fs_private;
3716 mpol_put(ctx->mpol);
3721 static const struct fs_context_operations shmem_fs_context_ops = {
3722 .free = shmem_free_fc,
3723 .get_tree = shmem_get_tree,
3725 .parse_monolithic = shmem_parse_options,
3726 .parse_param = shmem_parse_one,
3727 .reconfigure = shmem_reconfigure,
3731 static struct kmem_cache *shmem_inode_cachep;
3733 static struct inode *shmem_alloc_inode(struct super_block *sb)
3735 struct shmem_inode_info *info;
3736 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3739 return &info->vfs_inode;
3742 static void shmem_free_in_core_inode(struct inode *inode)
3744 if (S_ISLNK(inode->i_mode))
3745 kfree(inode->i_link);
3746 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3749 static void shmem_destroy_inode(struct inode *inode)
3751 if (S_ISREG(inode->i_mode))
3752 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3755 static void shmem_init_inode(void *foo)
3757 struct shmem_inode_info *info = foo;
3758 inode_init_once(&info->vfs_inode);
3761 static void shmem_init_inodecache(void)
3763 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3764 sizeof(struct shmem_inode_info),
3765 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3768 static void shmem_destroy_inodecache(void)
3770 kmem_cache_destroy(shmem_inode_cachep);
3773 const struct address_space_operations shmem_aops = {
3774 .writepage = shmem_writepage,
3775 .set_page_dirty = __set_page_dirty_no_writeback,
3777 .write_begin = shmem_write_begin,
3778 .write_end = shmem_write_end,
3780 #ifdef CONFIG_MIGRATION
3781 .migratepage = migrate_page,
3783 .error_remove_page = generic_error_remove_page,
3785 EXPORT_SYMBOL(shmem_aops);
3787 static const struct file_operations shmem_file_operations = {
3789 .get_unmapped_area = shmem_get_unmapped_area,
3791 .llseek = shmem_file_llseek,
3792 .read_iter = shmem_file_read_iter,
3793 .write_iter = generic_file_write_iter,
3794 .fsync = noop_fsync,
3795 .splice_read = generic_file_splice_read,
3796 .splice_write = iter_file_splice_write,
3797 .fallocate = shmem_fallocate,
3801 static const struct inode_operations shmem_inode_operations = {
3802 .getattr = shmem_getattr,
3803 .setattr = shmem_setattr,
3804 #ifdef CONFIG_TMPFS_XATTR
3805 .listxattr = shmem_listxattr,
3806 .set_acl = simple_set_acl,
3810 static const struct inode_operations shmem_dir_inode_operations = {
3812 .create = shmem_create,
3813 .lookup = simple_lookup,
3815 .unlink = shmem_unlink,
3816 .symlink = shmem_symlink,
3817 .mkdir = shmem_mkdir,
3818 .rmdir = shmem_rmdir,
3819 .mknod = shmem_mknod,
3820 .rename = shmem_rename2,
3821 .tmpfile = shmem_tmpfile,
3823 #ifdef CONFIG_TMPFS_XATTR
3824 .listxattr = shmem_listxattr,
3826 #ifdef CONFIG_TMPFS_POSIX_ACL
3827 .setattr = shmem_setattr,
3828 .set_acl = simple_set_acl,
3832 static const struct inode_operations shmem_special_inode_operations = {
3833 #ifdef CONFIG_TMPFS_XATTR
3834 .listxattr = shmem_listxattr,
3836 #ifdef CONFIG_TMPFS_POSIX_ACL
3837 .setattr = shmem_setattr,
3838 .set_acl = simple_set_acl,
3842 static const struct super_operations shmem_ops = {
3843 .alloc_inode = shmem_alloc_inode,
3844 .free_inode = shmem_free_in_core_inode,
3845 .destroy_inode = shmem_destroy_inode,
3847 .statfs = shmem_statfs,
3848 .show_options = shmem_show_options,
3850 .evict_inode = shmem_evict_inode,
3851 .drop_inode = generic_delete_inode,
3852 .put_super = shmem_put_super,
3853 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3854 .nr_cached_objects = shmem_unused_huge_count,
3855 .free_cached_objects = shmem_unused_huge_scan,
3859 static const struct vm_operations_struct shmem_vm_ops = {
3860 .fault = shmem_fault,
3861 .map_pages = filemap_map_pages,
3863 .set_policy = shmem_set_policy,
3864 .get_policy = shmem_get_policy,
3868 int shmem_init_fs_context(struct fs_context *fc)
3870 struct shmem_options *ctx;
3872 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3876 ctx->mode = 0777 | S_ISVTX;
3877 ctx->uid = current_fsuid();
3878 ctx->gid = current_fsgid();
3880 fc->fs_private = ctx;
3881 fc->ops = &shmem_fs_context_ops;
3885 static struct file_system_type shmem_fs_type = {
3886 .owner = THIS_MODULE,
3888 .init_fs_context = shmem_init_fs_context,
3890 .parameters = shmem_fs_parameters,
3892 .kill_sb = kill_litter_super,
3893 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
3896 int __init shmem_init(void)
3900 shmem_init_inodecache();
3902 error = register_filesystem(&shmem_fs_type);
3904 pr_err("Could not register tmpfs\n");
3908 shm_mnt = kern_mount(&shmem_fs_type);
3909 if (IS_ERR(shm_mnt)) {
3910 error = PTR_ERR(shm_mnt);
3911 pr_err("Could not kern_mount tmpfs\n");
3915 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3916 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3917 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3919 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
3924 unregister_filesystem(&shmem_fs_type);
3926 shmem_destroy_inodecache();
3927 shm_mnt = ERR_PTR(error);
3931 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3932 static ssize_t shmem_enabled_show(struct kobject *kobj,
3933 struct kobj_attribute *attr, char *buf)
3935 static const int values[] = {
3937 SHMEM_HUGE_WITHIN_SIZE,
3946 for (i = 0; i < ARRAY_SIZE(values); i++) {
3947 len += sysfs_emit_at(buf, len,
3948 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3950 shmem_format_huge(values[i]));
3953 len += sysfs_emit_at(buf, len, "\n");
3958 static ssize_t shmem_enabled_store(struct kobject *kobj,
3959 struct kobj_attribute *attr, const char *buf, size_t count)
3964 if (count + 1 > sizeof(tmp))
3966 memcpy(tmp, buf, count);
3968 if (count && tmp[count - 1] == '\n')
3969 tmp[count - 1] = '\0';
3971 huge = shmem_parse_huge(tmp);
3972 if (huge == -EINVAL)
3974 if (!has_transparent_hugepage() &&
3975 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3979 if (shmem_huge > SHMEM_HUGE_DENY)
3980 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3984 struct kobj_attribute shmem_enabled_attr =
3985 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3986 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3988 #else /* !CONFIG_SHMEM */
3991 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3993 * This is intended for small system where the benefits of the full
3994 * shmem code (swap-backed and resource-limited) are outweighed by
3995 * their complexity. On systems without swap this code should be
3996 * effectively equivalent, but much lighter weight.
3999 static struct file_system_type shmem_fs_type = {
4001 .init_fs_context = ramfs_init_fs_context,
4002 .parameters = ramfs_fs_parameters,
4003 .kill_sb = kill_litter_super,
4004 .fs_flags = FS_USERNS_MOUNT,
4007 int __init shmem_init(void)
4009 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4011 shm_mnt = kern_mount(&shmem_fs_type);
4012 BUG_ON(IS_ERR(shm_mnt));
4017 int shmem_unuse(unsigned int type, bool frontswap,
4018 unsigned long *fs_pages_to_unuse)
4023 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4028 void shmem_unlock_mapping(struct address_space *mapping)
4033 unsigned long shmem_get_unmapped_area(struct file *file,
4034 unsigned long addr, unsigned long len,
4035 unsigned long pgoff, unsigned long flags)
4037 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4041 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4043 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4045 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4047 #define shmem_vm_ops generic_file_vm_ops
4048 #define shmem_file_operations ramfs_file_operations
4049 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4050 #define shmem_acct_size(flags, size) 0
4051 #define shmem_unacct_size(flags, size) do {} while (0)
4053 #endif /* CONFIG_SHMEM */
4057 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4058 unsigned long flags, unsigned int i_flags)
4060 struct inode *inode;
4064 return ERR_CAST(mnt);
4066 if (size < 0 || size > MAX_LFS_FILESIZE)
4067 return ERR_PTR(-EINVAL);
4069 if (shmem_acct_size(flags, size))
4070 return ERR_PTR(-ENOMEM);
4072 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4074 if (unlikely(!inode)) {
4075 shmem_unacct_size(flags, size);
4076 return ERR_PTR(-ENOSPC);
4078 inode->i_flags |= i_flags;
4079 inode->i_size = size;
4080 clear_nlink(inode); /* It is unlinked */
4081 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4083 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4084 &shmem_file_operations);
4091 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4092 * kernel internal. There will be NO LSM permission checks against the
4093 * underlying inode. So users of this interface must do LSM checks at a
4094 * higher layer. The users are the big_key and shm implementations. LSM
4095 * checks are provided at the key or shm level rather than the inode.
4096 * @name: name for dentry (to be seen in /proc/<pid>/maps
4097 * @size: size to be set for the file
4098 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4100 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4102 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4106 * shmem_file_setup - get an unlinked file living in tmpfs
4107 * @name: name for dentry (to be seen in /proc/<pid>/maps
4108 * @size: size to be set for the file
4109 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4111 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4113 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4115 EXPORT_SYMBOL_GPL(shmem_file_setup);
4118 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4119 * @mnt: the tmpfs mount where the file will be created
4120 * @name: name for dentry (to be seen in /proc/<pid>/maps
4121 * @size: size to be set for the file
4122 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4124 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4125 loff_t size, unsigned long flags)
4127 return __shmem_file_setup(mnt, name, size, flags, 0);
4129 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4132 * shmem_zero_setup - setup a shared anonymous mapping
4133 * @vma: the vma to be mmapped is prepared by do_mmap
4135 int shmem_zero_setup(struct vm_area_struct *vma)
4138 loff_t size = vma->vm_end - vma->vm_start;
4141 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4142 * between XFS directory reading and selinux: since this file is only
4143 * accessible to the user through its mapping, use S_PRIVATE flag to
4144 * bypass file security, in the same way as shmem_kernel_file_setup().
4146 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4148 return PTR_ERR(file);
4152 vma->vm_file = file;
4153 vma->vm_ops = &shmem_vm_ops;
4155 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4156 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4157 (vma->vm_end & HPAGE_PMD_MASK)) {
4158 khugepaged_enter(vma, vma->vm_flags);
4165 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4166 * @mapping: the page's address_space
4167 * @index: the page index
4168 * @gfp: the page allocator flags to use if allocating
4170 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4171 * with any new page allocations done using the specified allocation flags.
4172 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4173 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4174 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4176 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4177 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4179 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4180 pgoff_t index, gfp_t gfp)
4183 struct inode *inode = mapping->host;
4187 BUG_ON(!shmem_mapping(mapping));
4188 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4189 gfp, NULL, NULL, NULL);
4191 page = ERR_PTR(error);
4197 * The tiny !SHMEM case uses ramfs without swap
4199 return read_cache_page_gfp(mapping, index, gfp);
4202 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);