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/pagevec.h>
63 #include <linux/percpu_counter.h>
64 #include <linux/falloc.h>
65 #include <linux/splice.h>
66 #include <linux/security.h>
67 #include <linux/swapops.h>
68 #include <linux/mempolicy.h>
69 #include <linux/namei.h>
70 #include <linux/ctype.h>
71 #include <linux/migrate.h>
72 #include <linux/highmem.h>
73 #include <linux/seq_file.h>
74 #include <linux/magic.h>
75 #include <linux/syscalls.h>
76 #include <linux/fcntl.h>
77 #include <uapi/linux/memfd.h>
78 #include <linux/userfaultfd_k.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
82 #include <linux/uaccess.h>
86 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
87 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 /* Pretend that each entry is of this size in directory's i_size */
90 #define BOGO_DIRENT_SIZE 20
92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
93 #define SHORT_SYMLINK_LEN 128
96 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
97 * inode->i_private (with i_rwsem making sure that it has only one user at
98 * a time): we would prefer not to enlarge the shmem inode just for that.
100 struct shmem_falloc {
101 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
102 pgoff_t start; /* start of range currently being fallocated */
103 pgoff_t next; /* the next page offset to be fallocated */
104 pgoff_t nr_falloced; /* how many new pages have been fallocated */
105 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
108 struct shmem_options {
109 unsigned long long blocks;
110 unsigned long long inodes;
111 struct mempolicy *mpol;
118 #define SHMEM_SEEN_BLOCKS 1
119 #define SHMEM_SEEN_INODES 2
120 #define SHMEM_SEEN_HUGE 4
121 #define SHMEM_SEEN_INUMS 8
125 static unsigned long shmem_default_max_blocks(void)
127 return totalram_pages() / 2;
130 static unsigned long shmem_default_max_inodes(void)
132 unsigned long nr_pages = totalram_pages();
134 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
138 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
139 struct page **pagep, enum sgp_type sgp,
140 gfp_t gfp, struct vm_area_struct *vma,
141 vm_fault_t *fault_type);
142 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
143 struct page **pagep, enum sgp_type sgp,
144 gfp_t gfp, struct vm_area_struct *vma,
145 struct vm_fault *vmf, vm_fault_t *fault_type);
147 int shmem_getpage(struct inode *inode, pgoff_t index,
148 struct page **pagep, enum sgp_type sgp)
150 return shmem_getpage_gfp(inode, index, pagep, sgp,
151 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
154 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
156 return sb->s_fs_info;
160 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
161 * for shared memory and for shared anonymous (/dev/zero) mappings
162 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
163 * consistent with the pre-accounting of private mappings ...
165 static inline int shmem_acct_size(unsigned long flags, loff_t size)
167 return (flags & VM_NORESERVE) ?
168 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
171 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
173 if (!(flags & VM_NORESERVE))
174 vm_unacct_memory(VM_ACCT(size));
177 static inline int shmem_reacct_size(unsigned long flags,
178 loff_t oldsize, loff_t newsize)
180 if (!(flags & VM_NORESERVE)) {
181 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
182 return security_vm_enough_memory_mm(current->mm,
183 VM_ACCT(newsize) - VM_ACCT(oldsize));
184 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
185 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
191 * ... whereas tmpfs objects are accounted incrementally as
192 * pages are allocated, in order to allow large sparse files.
193 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
194 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
196 static inline int shmem_acct_block(unsigned long flags, long pages)
198 if (!(flags & VM_NORESERVE))
201 return security_vm_enough_memory_mm(current->mm,
202 pages * VM_ACCT(PAGE_SIZE));
205 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
207 if (flags & VM_NORESERVE)
208 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
211 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
213 struct shmem_inode_info *info = SHMEM_I(inode);
214 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
216 if (shmem_acct_block(info->flags, pages))
219 if (sbinfo->max_blocks) {
220 if (percpu_counter_compare(&sbinfo->used_blocks,
221 sbinfo->max_blocks - pages) > 0)
223 percpu_counter_add(&sbinfo->used_blocks, pages);
229 shmem_unacct_blocks(info->flags, pages);
233 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
235 struct shmem_inode_info *info = SHMEM_I(inode);
236 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
238 if (sbinfo->max_blocks)
239 percpu_counter_sub(&sbinfo->used_blocks, pages);
240 shmem_unacct_blocks(info->flags, pages);
243 static const struct super_operations shmem_ops;
244 const struct address_space_operations shmem_aops;
245 static const struct file_operations shmem_file_operations;
246 static const struct inode_operations shmem_inode_operations;
247 static const struct inode_operations shmem_dir_inode_operations;
248 static const struct inode_operations shmem_special_inode_operations;
249 static const struct vm_operations_struct shmem_vm_ops;
250 static struct file_system_type shmem_fs_type;
252 bool vma_is_shmem(struct vm_area_struct *vma)
254 return vma->vm_ops == &shmem_vm_ops;
257 static LIST_HEAD(shmem_swaplist);
258 static DEFINE_MUTEX(shmem_swaplist_mutex);
261 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
262 * produces a novel ino for the newly allocated inode.
264 * It may also be called when making a hard link to permit the space needed by
265 * each dentry. However, in that case, no new inode number is needed since that
266 * internally draws from another pool of inode numbers (currently global
267 * get_next_ino()). This case is indicated by passing NULL as inop.
269 #define SHMEM_INO_BATCH 1024
270 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
272 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
275 if (!(sb->s_flags & SB_KERNMOUNT)) {
276 raw_spin_lock(&sbinfo->stat_lock);
277 if (sbinfo->max_inodes) {
278 if (!sbinfo->free_inodes) {
279 raw_spin_unlock(&sbinfo->stat_lock);
282 sbinfo->free_inodes--;
285 ino = sbinfo->next_ino++;
286 if (unlikely(is_zero_ino(ino)))
287 ino = sbinfo->next_ino++;
288 if (unlikely(!sbinfo->full_inums &&
291 * Emulate get_next_ino uint wraparound for
294 if (IS_ENABLED(CONFIG_64BIT))
295 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
296 __func__, MINOR(sb->s_dev));
297 sbinfo->next_ino = 1;
298 ino = sbinfo->next_ino++;
302 raw_spin_unlock(&sbinfo->stat_lock);
305 * __shmem_file_setup, one of our callers, is lock-free: it
306 * doesn't hold stat_lock in shmem_reserve_inode since
307 * max_inodes is always 0, and is called from potentially
308 * unknown contexts. As such, use a per-cpu batched allocator
309 * which doesn't require the per-sb stat_lock unless we are at
310 * the batch boundary.
312 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
313 * shmem mounts are not exposed to userspace, so we don't need
314 * to worry about things like glibc compatibility.
318 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
320 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
321 raw_spin_lock(&sbinfo->stat_lock);
322 ino = sbinfo->next_ino;
323 sbinfo->next_ino += SHMEM_INO_BATCH;
324 raw_spin_unlock(&sbinfo->stat_lock);
325 if (unlikely(is_zero_ino(ino)))
336 static void shmem_free_inode(struct super_block *sb)
338 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
339 if (sbinfo->max_inodes) {
340 raw_spin_lock(&sbinfo->stat_lock);
341 sbinfo->free_inodes++;
342 raw_spin_unlock(&sbinfo->stat_lock);
347 * shmem_recalc_inode - recalculate the block usage of an inode
348 * @inode: inode to recalc
350 * We have to calculate the free blocks since the mm can drop
351 * undirtied hole pages behind our back.
353 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
354 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
356 * It has to be called with the spinlock held.
358 static void shmem_recalc_inode(struct inode *inode)
360 struct shmem_inode_info *info = SHMEM_I(inode);
363 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
365 info->alloced -= freed;
366 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
367 shmem_inode_unacct_blocks(inode, freed);
371 bool shmem_charge(struct inode *inode, long pages)
373 struct shmem_inode_info *info = SHMEM_I(inode);
376 if (!shmem_inode_acct_block(inode, pages))
379 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
380 inode->i_mapping->nrpages += pages;
382 spin_lock_irqsave(&info->lock, flags);
383 info->alloced += pages;
384 inode->i_blocks += pages * BLOCKS_PER_PAGE;
385 shmem_recalc_inode(inode);
386 spin_unlock_irqrestore(&info->lock, flags);
391 void shmem_uncharge(struct inode *inode, long pages)
393 struct shmem_inode_info *info = SHMEM_I(inode);
396 /* nrpages adjustment done by __delete_from_page_cache() or caller */
398 spin_lock_irqsave(&info->lock, flags);
399 info->alloced -= pages;
400 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
401 shmem_recalc_inode(inode);
402 spin_unlock_irqrestore(&info->lock, flags);
404 shmem_inode_unacct_blocks(inode, pages);
408 * Replace item expected in xarray by a new item, while holding xa_lock.
410 static int shmem_replace_entry(struct address_space *mapping,
411 pgoff_t index, void *expected, void *replacement)
413 XA_STATE(xas, &mapping->i_pages, index);
416 VM_BUG_ON(!expected);
417 VM_BUG_ON(!replacement);
418 item = xas_load(&xas);
419 if (item != expected)
421 xas_store(&xas, replacement);
426 * Sometimes, before we decide whether to proceed or to fail, we must check
427 * that an entry was not already brought back from swap by a racing thread.
429 * Checking page is not enough: by the time a SwapCache page is locked, it
430 * might be reused, and again be SwapCache, using the same swap as before.
432 static bool shmem_confirm_swap(struct address_space *mapping,
433 pgoff_t index, swp_entry_t swap)
435 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
439 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
442 * disables huge pages for the mount;
444 * enables huge pages for the mount;
445 * SHMEM_HUGE_WITHIN_SIZE:
446 * only allocate huge pages if the page will be fully within i_size,
447 * also respect fadvise()/madvise() hints;
449 * only allocate huge pages if requested with fadvise()/madvise();
452 #define SHMEM_HUGE_NEVER 0
453 #define SHMEM_HUGE_ALWAYS 1
454 #define SHMEM_HUGE_WITHIN_SIZE 2
455 #define SHMEM_HUGE_ADVISE 3
459 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
462 * disables huge on shm_mnt and all mounts, for emergency use;
464 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
467 #define SHMEM_HUGE_DENY (-1)
468 #define SHMEM_HUGE_FORCE (-2)
470 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
471 /* ifdef here to avoid bloating shmem.o when not necessary */
473 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
475 bool shmem_is_huge(struct vm_area_struct *vma,
476 struct inode *inode, pgoff_t index)
480 if (shmem_huge == SHMEM_HUGE_DENY)
482 if (vma && ((vma->vm_flags & VM_NOHUGEPAGE) ||
483 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
485 if (shmem_huge == SHMEM_HUGE_FORCE)
488 switch (SHMEM_SB(inode->i_sb)->huge) {
489 case SHMEM_HUGE_ALWAYS:
491 case SHMEM_HUGE_WITHIN_SIZE:
492 index = round_up(index + 1, HPAGE_PMD_NR);
493 i_size = round_up(i_size_read(inode), PAGE_SIZE);
494 if (i_size >> PAGE_SHIFT >= index)
497 case SHMEM_HUGE_ADVISE:
498 if (vma && (vma->vm_flags & VM_HUGEPAGE))
506 #if defined(CONFIG_SYSFS)
507 static int shmem_parse_huge(const char *str)
509 if (!strcmp(str, "never"))
510 return SHMEM_HUGE_NEVER;
511 if (!strcmp(str, "always"))
512 return SHMEM_HUGE_ALWAYS;
513 if (!strcmp(str, "within_size"))
514 return SHMEM_HUGE_WITHIN_SIZE;
515 if (!strcmp(str, "advise"))
516 return SHMEM_HUGE_ADVISE;
517 if (!strcmp(str, "deny"))
518 return SHMEM_HUGE_DENY;
519 if (!strcmp(str, "force"))
520 return SHMEM_HUGE_FORCE;
525 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
526 static const char *shmem_format_huge(int huge)
529 case SHMEM_HUGE_NEVER:
531 case SHMEM_HUGE_ALWAYS:
533 case SHMEM_HUGE_WITHIN_SIZE:
534 return "within_size";
535 case SHMEM_HUGE_ADVISE:
537 case SHMEM_HUGE_DENY:
539 case SHMEM_HUGE_FORCE:
548 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
549 struct shrink_control *sc, unsigned long nr_to_split)
551 LIST_HEAD(list), *pos, *next;
552 LIST_HEAD(to_remove);
554 struct shmem_inode_info *info;
556 unsigned long batch = sc ? sc->nr_to_scan : 128;
557 int removed = 0, split = 0;
559 if (list_empty(&sbinfo->shrinklist))
562 spin_lock(&sbinfo->shrinklist_lock);
563 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
564 info = list_entry(pos, struct shmem_inode_info, shrinklist);
567 inode = igrab(&info->vfs_inode);
569 /* inode is about to be evicted */
571 list_del_init(&info->shrinklist);
576 /* Check if there's anything to gain */
577 if (round_up(inode->i_size, PAGE_SIZE) ==
578 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
579 list_move(&info->shrinklist, &to_remove);
584 list_move(&info->shrinklist, &list);
589 spin_unlock(&sbinfo->shrinklist_lock);
591 list_for_each_safe(pos, next, &to_remove) {
592 info = list_entry(pos, struct shmem_inode_info, shrinklist);
593 inode = &info->vfs_inode;
594 list_del_init(&info->shrinklist);
598 list_for_each_safe(pos, next, &list) {
601 info = list_entry(pos, struct shmem_inode_info, shrinklist);
602 inode = &info->vfs_inode;
604 if (nr_to_split && split >= nr_to_split)
607 page = find_get_page(inode->i_mapping,
608 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
612 /* No huge page at the end of the file: nothing to split */
613 if (!PageTransHuge(page)) {
619 * Leave the inode on the list if we failed to lock
620 * the page at this time.
622 * Waiting for the lock may lead to deadlock in the
625 if (!trylock_page(page)) {
630 ret = split_huge_page(page);
634 /* If split failed leave the inode on the list */
640 list_del_init(&info->shrinklist);
646 spin_lock(&sbinfo->shrinklist_lock);
647 list_splice_tail(&list, &sbinfo->shrinklist);
648 sbinfo->shrinklist_len -= removed;
649 spin_unlock(&sbinfo->shrinklist_lock);
654 static long shmem_unused_huge_scan(struct super_block *sb,
655 struct shrink_control *sc)
657 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
659 if (!READ_ONCE(sbinfo->shrinklist_len))
662 return shmem_unused_huge_shrink(sbinfo, sc, 0);
665 static long shmem_unused_huge_count(struct super_block *sb,
666 struct shrink_control *sc)
668 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
669 return READ_ONCE(sbinfo->shrinklist_len);
671 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
673 #define shmem_huge SHMEM_HUGE_DENY
675 bool shmem_is_huge(struct vm_area_struct *vma,
676 struct inode *inode, pgoff_t index)
681 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
682 struct shrink_control *sc, unsigned long nr_to_split)
686 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
689 * Like add_to_page_cache_locked, but error if expected item has gone.
691 static int shmem_add_to_page_cache(struct page *page,
692 struct address_space *mapping,
693 pgoff_t index, void *expected, gfp_t gfp,
694 struct mm_struct *charge_mm)
696 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
697 unsigned long nr = compound_nr(page);
700 VM_BUG_ON_PAGE(PageTail(page), page);
701 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
702 VM_BUG_ON_PAGE(!PageLocked(page), page);
703 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
704 VM_BUG_ON(expected && PageTransHuge(page));
706 page_ref_add(page, nr);
707 page->mapping = mapping;
710 if (!PageSwapCache(page)) {
711 error = mem_cgroup_charge(page_folio(page), charge_mm, gfp);
713 if (PageTransHuge(page)) {
714 count_vm_event(THP_FILE_FALLBACK);
715 count_vm_event(THP_FILE_FALLBACK_CHARGE);
720 cgroup_throttle_swaprate(page, gfp);
724 if (expected != xas_find_conflict(&xas)) {
725 xas_set_err(&xas, -EEXIST);
728 if (expected && xas_find_conflict(&xas)) {
729 xas_set_err(&xas, -EEXIST);
732 xas_store(&xas, page);
735 if (PageTransHuge(page)) {
736 count_vm_event(THP_FILE_ALLOC);
737 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
739 mapping->nrpages += nr;
740 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
741 __mod_lruvec_page_state(page, NR_SHMEM, nr);
743 xas_unlock_irq(&xas);
744 } while (xas_nomem(&xas, gfp));
746 if (xas_error(&xas)) {
747 error = xas_error(&xas);
753 page->mapping = NULL;
754 page_ref_sub(page, nr);
759 * Like delete_from_page_cache, but substitutes swap for page.
761 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
763 struct address_space *mapping = page->mapping;
766 VM_BUG_ON_PAGE(PageCompound(page), page);
768 xa_lock_irq(&mapping->i_pages);
769 error = shmem_replace_entry(mapping, page->index, page, radswap);
770 page->mapping = NULL;
772 __dec_lruvec_page_state(page, NR_FILE_PAGES);
773 __dec_lruvec_page_state(page, NR_SHMEM);
774 xa_unlock_irq(&mapping->i_pages);
780 * Remove swap entry from page cache, free the swap and its page cache.
782 static int shmem_free_swap(struct address_space *mapping,
783 pgoff_t index, void *radswap)
787 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
790 free_swap_and_cache(radix_to_swp_entry(radswap));
795 * Determine (in bytes) how many of the shmem object's pages mapped by the
796 * given offsets are swapped out.
798 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
799 * as long as the inode doesn't go away and racy results are not a problem.
801 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
802 pgoff_t start, pgoff_t end)
804 XA_STATE(xas, &mapping->i_pages, start);
806 unsigned long swapped = 0;
809 xas_for_each(&xas, page, end - 1) {
810 if (xas_retry(&xas, page))
812 if (xa_is_value(page))
815 if (need_resched()) {
823 return swapped << PAGE_SHIFT;
827 * Determine (in bytes) how many of the shmem object's pages mapped by the
828 * given vma is swapped out.
830 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
831 * as long as the inode doesn't go away and racy results are not a problem.
833 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
835 struct inode *inode = file_inode(vma->vm_file);
836 struct shmem_inode_info *info = SHMEM_I(inode);
837 struct address_space *mapping = inode->i_mapping;
838 unsigned long swapped;
840 /* Be careful as we don't hold info->lock */
841 swapped = READ_ONCE(info->swapped);
844 * The easier cases are when the shmem object has nothing in swap, or
845 * the vma maps it whole. Then we can simply use the stats that we
851 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
852 return swapped << PAGE_SHIFT;
854 /* Here comes the more involved part */
855 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
856 vma->vm_pgoff + vma_pages(vma));
860 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
862 void shmem_unlock_mapping(struct address_space *mapping)
869 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
871 while (!mapping_unevictable(mapping)) {
872 if (!pagevec_lookup(&pvec, mapping, &index))
874 check_move_unevictable_pages(&pvec);
875 pagevec_release(&pvec);
880 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
886 * At first avoid shmem_getpage(,,,SGP_READ): that fails
887 * beyond i_size, and reports fallocated pages as holes.
889 folio = __filemap_get_folio(inode->i_mapping, index,
890 FGP_ENTRY | FGP_LOCK, 0);
891 if (!xa_is_value(folio))
894 * But read a page back from swap if any of it is within i_size
895 * (although in some cases this is just a waste of time).
898 shmem_getpage(inode, index, &page, SGP_READ);
899 return page ? page_folio(page) : NULL;
903 * Remove range of pages and swap entries from page cache, and free them.
904 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
906 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
909 struct address_space *mapping = inode->i_mapping;
910 struct shmem_inode_info *info = SHMEM_I(inode);
911 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
912 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
913 struct folio_batch fbatch;
914 pgoff_t indices[PAGEVEC_SIZE];
917 long nr_swaps_freed = 0;
922 end = -1; /* unsigned, so actually very big */
924 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
925 info->fallocend = start;
927 folio_batch_init(&fbatch);
929 while (index < end && find_lock_entries(mapping, index, end - 1,
931 for (i = 0; i < folio_batch_count(&fbatch); i++) {
932 folio = fbatch.folios[i];
936 if (xa_is_value(folio)) {
939 nr_swaps_freed += !shmem_free_swap(mapping,
943 index += folio_nr_pages(folio) - 1;
945 if (!unfalloc || !folio_test_uptodate(folio))
946 truncate_inode_folio(mapping, folio);
949 folio_batch_remove_exceptionals(&fbatch);
950 folio_batch_release(&fbatch);
955 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
956 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
958 same_folio = lend < folio_pos(folio) + folio_size(folio);
959 folio_mark_dirty(folio);
960 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
961 start = folio->index + folio_nr_pages(folio);
971 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
973 folio_mark_dirty(folio);
974 if (!truncate_inode_partial_folio(folio, lstart, lend))
981 while (index < end) {
984 if (!find_get_entries(mapping, index, end - 1, &fbatch,
986 /* If all gone or hole-punch or unfalloc, we're done */
987 if (index == start || end != -1)
989 /* But if truncating, restart to make sure all gone */
993 for (i = 0; i < folio_batch_count(&fbatch); i++) {
994 folio = fbatch.folios[i];
997 if (xa_is_value(folio)) {
1000 if (shmem_free_swap(mapping, index, folio)) {
1001 /* Swap was replaced by page: retry */
1011 if (!unfalloc || !folio_test_uptodate(folio)) {
1012 if (folio_mapping(folio) != mapping) {
1013 /* Page was replaced by swap: retry */
1014 folio_unlock(folio);
1018 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1020 truncate_inode_folio(mapping, folio);
1022 index = folio->index + folio_nr_pages(folio) - 1;
1023 folio_unlock(folio);
1025 folio_batch_remove_exceptionals(&fbatch);
1026 folio_batch_release(&fbatch);
1030 spin_lock_irq(&info->lock);
1031 info->swapped -= nr_swaps_freed;
1032 shmem_recalc_inode(inode);
1033 spin_unlock_irq(&info->lock);
1036 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1038 shmem_undo_range(inode, lstart, lend, false);
1039 inode->i_ctime = inode->i_mtime = current_time(inode);
1041 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1043 static int shmem_getattr(struct user_namespace *mnt_userns,
1044 const struct path *path, struct kstat *stat,
1045 u32 request_mask, unsigned int query_flags)
1047 struct inode *inode = path->dentry->d_inode;
1048 struct shmem_inode_info *info = SHMEM_I(inode);
1050 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1051 spin_lock_irq(&info->lock);
1052 shmem_recalc_inode(inode);
1053 spin_unlock_irq(&info->lock);
1055 generic_fillattr(&init_user_ns, inode, stat);
1057 if (shmem_is_huge(NULL, inode, 0))
1058 stat->blksize = HPAGE_PMD_SIZE;
1063 static int shmem_setattr(struct user_namespace *mnt_userns,
1064 struct dentry *dentry, struct iattr *attr)
1066 struct inode *inode = d_inode(dentry);
1067 struct shmem_inode_info *info = SHMEM_I(inode);
1070 error = setattr_prepare(&init_user_ns, dentry, attr);
1074 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1075 loff_t oldsize = inode->i_size;
1076 loff_t newsize = attr->ia_size;
1078 /* protected by i_rwsem */
1079 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1080 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1083 if (newsize != oldsize) {
1084 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1088 i_size_write(inode, newsize);
1089 inode->i_ctime = inode->i_mtime = current_time(inode);
1091 if (newsize <= oldsize) {
1092 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1093 if (oldsize > holebegin)
1094 unmap_mapping_range(inode->i_mapping,
1097 shmem_truncate_range(inode,
1098 newsize, (loff_t)-1);
1099 /* unmap again to remove racily COWed private pages */
1100 if (oldsize > holebegin)
1101 unmap_mapping_range(inode->i_mapping,
1106 setattr_copy(&init_user_ns, inode, attr);
1107 if (attr->ia_valid & ATTR_MODE)
1108 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1112 static void shmem_evict_inode(struct inode *inode)
1114 struct shmem_inode_info *info = SHMEM_I(inode);
1115 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1117 if (shmem_mapping(inode->i_mapping)) {
1118 shmem_unacct_size(info->flags, inode->i_size);
1120 shmem_truncate_range(inode, 0, (loff_t)-1);
1121 if (!list_empty(&info->shrinklist)) {
1122 spin_lock(&sbinfo->shrinklist_lock);
1123 if (!list_empty(&info->shrinklist)) {
1124 list_del_init(&info->shrinklist);
1125 sbinfo->shrinklist_len--;
1127 spin_unlock(&sbinfo->shrinklist_lock);
1129 while (!list_empty(&info->swaplist)) {
1130 /* Wait while shmem_unuse() is scanning this inode... */
1131 wait_var_event(&info->stop_eviction,
1132 !atomic_read(&info->stop_eviction));
1133 mutex_lock(&shmem_swaplist_mutex);
1134 /* ...but beware of the race if we peeked too early */
1135 if (!atomic_read(&info->stop_eviction))
1136 list_del_init(&info->swaplist);
1137 mutex_unlock(&shmem_swaplist_mutex);
1141 simple_xattrs_free(&info->xattrs);
1142 WARN_ON(inode->i_blocks);
1143 shmem_free_inode(inode->i_sb);
1147 static int shmem_find_swap_entries(struct address_space *mapping,
1148 pgoff_t start, unsigned int nr_entries,
1149 struct page **entries, pgoff_t *indices,
1150 unsigned int type, bool frontswap)
1152 XA_STATE(xas, &mapping->i_pages, start);
1155 unsigned int ret = 0;
1161 xas_for_each(&xas, page, ULONG_MAX) {
1162 if (xas_retry(&xas, page))
1165 if (!xa_is_value(page))
1168 entry = radix_to_swp_entry(page);
1169 if (swp_type(entry) != type)
1172 !frontswap_test(swap_info[type], swp_offset(entry)))
1175 indices[ret] = xas.xa_index;
1176 entries[ret] = page;
1178 if (need_resched()) {
1182 if (++ret == nr_entries)
1191 * Move the swapped pages for an inode to page cache. Returns the count
1192 * of pages swapped in, or the error in case of failure.
1194 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1200 struct address_space *mapping = inode->i_mapping;
1202 for (i = 0; i < pvec.nr; i++) {
1203 struct page *page = pvec.pages[i];
1205 if (!xa_is_value(page))
1207 error = shmem_swapin_page(inode, indices[i],
1209 mapping_gfp_mask(mapping),
1216 if (error == -ENOMEM)
1220 return error ? error : ret;
1224 * If swap found in inode, free it and move page from swapcache to filecache.
1226 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1227 bool frontswap, unsigned long *fs_pages_to_unuse)
1229 struct address_space *mapping = inode->i_mapping;
1231 struct pagevec pvec;
1232 pgoff_t indices[PAGEVEC_SIZE];
1233 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1236 pagevec_init(&pvec);
1238 unsigned int nr_entries = PAGEVEC_SIZE;
1240 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1241 nr_entries = *fs_pages_to_unuse;
1243 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1244 pvec.pages, indices,
1251 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1255 if (frontswap_partial) {
1256 *fs_pages_to_unuse -= ret;
1257 if (*fs_pages_to_unuse == 0) {
1258 ret = FRONTSWAP_PAGES_UNUSED;
1263 start = indices[pvec.nr - 1];
1270 * Read all the shared memory data that resides in the swap
1271 * device 'type' back into memory, so the swap device can be
1274 int shmem_unuse(unsigned int type, bool frontswap,
1275 unsigned long *fs_pages_to_unuse)
1277 struct shmem_inode_info *info, *next;
1280 if (list_empty(&shmem_swaplist))
1283 mutex_lock(&shmem_swaplist_mutex);
1284 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1285 if (!info->swapped) {
1286 list_del_init(&info->swaplist);
1290 * Drop the swaplist mutex while searching the inode for swap;
1291 * but before doing so, make sure shmem_evict_inode() will not
1292 * remove placeholder inode from swaplist, nor let it be freed
1293 * (igrab() would protect from unlink, but not from unmount).
1295 atomic_inc(&info->stop_eviction);
1296 mutex_unlock(&shmem_swaplist_mutex);
1298 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1302 mutex_lock(&shmem_swaplist_mutex);
1303 next = list_next_entry(info, swaplist);
1305 list_del_init(&info->swaplist);
1306 if (atomic_dec_and_test(&info->stop_eviction))
1307 wake_up_var(&info->stop_eviction);
1311 mutex_unlock(&shmem_swaplist_mutex);
1317 * Move the page from the page cache to the swap cache.
1319 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1321 struct shmem_inode_info *info;
1322 struct address_space *mapping;
1323 struct inode *inode;
1328 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1329 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1330 * and its shmem_writeback() needs them to be split when swapping.
1332 if (PageTransCompound(page)) {
1333 /* Ensure the subpages are still dirty */
1335 if (split_huge_page(page) < 0)
1337 ClearPageDirty(page);
1340 BUG_ON(!PageLocked(page));
1341 mapping = page->mapping;
1342 index = page->index;
1343 inode = mapping->host;
1344 info = SHMEM_I(inode);
1345 if (info->flags & VM_LOCKED)
1347 if (!total_swap_pages)
1351 * Our capabilities prevent regular writeback or sync from ever calling
1352 * shmem_writepage; but a stacking filesystem might use ->writepage of
1353 * its underlying filesystem, in which case tmpfs should write out to
1354 * swap only in response to memory pressure, and not for the writeback
1357 if (!wbc->for_reclaim) {
1358 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1363 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1364 * value into swapfile.c, the only way we can correctly account for a
1365 * fallocated page arriving here is now to initialize it and write it.
1367 * That's okay for a page already fallocated earlier, but if we have
1368 * not yet completed the fallocation, then (a) we want to keep track
1369 * of this page in case we have to undo it, and (b) it may not be a
1370 * good idea to continue anyway, once we're pushing into swap. So
1371 * reactivate the page, and let shmem_fallocate() quit when too many.
1373 if (!PageUptodate(page)) {
1374 if (inode->i_private) {
1375 struct shmem_falloc *shmem_falloc;
1376 spin_lock(&inode->i_lock);
1377 shmem_falloc = inode->i_private;
1379 !shmem_falloc->waitq &&
1380 index >= shmem_falloc->start &&
1381 index < shmem_falloc->next)
1382 shmem_falloc->nr_unswapped++;
1384 shmem_falloc = NULL;
1385 spin_unlock(&inode->i_lock);
1389 clear_highpage(page);
1390 flush_dcache_page(page);
1391 SetPageUptodate(page);
1394 swap = get_swap_page(page);
1399 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1400 * if it's not already there. Do it now before the page is
1401 * moved to swap cache, when its pagelock no longer protects
1402 * the inode from eviction. But don't unlock the mutex until
1403 * we've incremented swapped, because shmem_unuse_inode() will
1404 * prune a !swapped inode from the swaplist under this mutex.
1406 mutex_lock(&shmem_swaplist_mutex);
1407 if (list_empty(&info->swaplist))
1408 list_add(&info->swaplist, &shmem_swaplist);
1410 if (add_to_swap_cache(page, swap,
1411 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1413 spin_lock_irq(&info->lock);
1414 shmem_recalc_inode(inode);
1416 spin_unlock_irq(&info->lock);
1418 swap_shmem_alloc(swap);
1419 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1421 mutex_unlock(&shmem_swaplist_mutex);
1422 BUG_ON(page_mapped(page));
1423 swap_writepage(page, wbc);
1427 mutex_unlock(&shmem_swaplist_mutex);
1428 put_swap_page(page, swap);
1430 set_page_dirty(page);
1431 if (wbc->for_reclaim)
1432 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1437 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1438 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1442 if (!mpol || mpol->mode == MPOL_DEFAULT)
1443 return; /* show nothing */
1445 mpol_to_str(buffer, sizeof(buffer), mpol);
1447 seq_printf(seq, ",mpol=%s", buffer);
1450 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1452 struct mempolicy *mpol = NULL;
1454 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1455 mpol = sbinfo->mpol;
1457 raw_spin_unlock(&sbinfo->stat_lock);
1461 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1462 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1465 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1469 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1471 #define vm_policy vm_private_data
1474 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1475 struct shmem_inode_info *info, pgoff_t index)
1477 /* Create a pseudo vma that just contains the policy */
1478 vma_init(vma, NULL);
1479 /* Bias interleave by inode number to distribute better across nodes */
1480 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1481 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1484 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1486 /* Drop reference taken by mpol_shared_policy_lookup() */
1487 mpol_cond_put(vma->vm_policy);
1490 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1491 struct shmem_inode_info *info, pgoff_t index)
1493 struct vm_area_struct pvma;
1495 struct vm_fault vmf = {
1499 shmem_pseudo_vma_init(&pvma, info, index);
1500 page = swap_cluster_readahead(swap, gfp, &vmf);
1501 shmem_pseudo_vma_destroy(&pvma);
1507 * Make sure huge_gfp is always more limited than limit_gfp.
1508 * Some of the flags set permissions, while others set limitations.
1510 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1512 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1513 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1514 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1515 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1517 /* Allow allocations only from the originally specified zones. */
1518 result |= zoneflags;
1521 * Minimize the result gfp by taking the union with the deny flags,
1522 * and the intersection of the allow flags.
1524 result |= (limit_gfp & denyflags);
1525 result |= (huge_gfp & limit_gfp) & allowflags;
1530 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1531 struct shmem_inode_info *info, pgoff_t index)
1533 struct vm_area_struct pvma;
1534 struct address_space *mapping = info->vfs_inode.i_mapping;
1538 hindex = round_down(index, HPAGE_PMD_NR);
1539 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1543 shmem_pseudo_vma_init(&pvma, info, hindex);
1544 page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(),
1546 shmem_pseudo_vma_destroy(&pvma);
1548 prep_transhuge_page(page);
1550 count_vm_event(THP_FILE_FALLBACK);
1554 static struct page *shmem_alloc_page(gfp_t gfp,
1555 struct shmem_inode_info *info, pgoff_t index)
1557 struct vm_area_struct pvma;
1560 shmem_pseudo_vma_init(&pvma, info, index);
1561 page = alloc_page_vma(gfp, &pvma, 0);
1562 shmem_pseudo_vma_destroy(&pvma);
1567 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1568 struct inode *inode,
1569 pgoff_t index, bool huge)
1571 struct shmem_inode_info *info = SHMEM_I(inode);
1576 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1578 nr = huge ? HPAGE_PMD_NR : 1;
1580 if (!shmem_inode_acct_block(inode, nr))
1584 page = shmem_alloc_hugepage(gfp, info, index);
1586 page = shmem_alloc_page(gfp, info, index);
1588 __SetPageLocked(page);
1589 __SetPageSwapBacked(page);
1594 shmem_inode_unacct_blocks(inode, nr);
1596 return ERR_PTR(err);
1600 * When a page is moved from swapcache to shmem filecache (either by the
1601 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1602 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1603 * ignorance of the mapping it belongs to. If that mapping has special
1604 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1605 * we may need to copy to a suitable page before moving to filecache.
1607 * In a future release, this may well be extended to respect cpuset and
1608 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1609 * but for now it is a simple matter of zone.
1611 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1613 return page_zonenum(page) > gfp_zone(gfp);
1616 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1617 struct shmem_inode_info *info, pgoff_t index)
1619 struct page *oldpage, *newpage;
1620 struct folio *old, *new;
1621 struct address_space *swap_mapping;
1627 entry.val = page_private(oldpage);
1628 swap_index = swp_offset(entry);
1629 swap_mapping = page_mapping(oldpage);
1632 * We have arrived here because our zones are constrained, so don't
1633 * limit chance of success by further cpuset and node constraints.
1635 gfp &= ~GFP_CONSTRAINT_MASK;
1636 newpage = shmem_alloc_page(gfp, info, index);
1641 copy_highpage(newpage, oldpage);
1642 flush_dcache_page(newpage);
1644 __SetPageLocked(newpage);
1645 __SetPageSwapBacked(newpage);
1646 SetPageUptodate(newpage);
1647 set_page_private(newpage, entry.val);
1648 SetPageSwapCache(newpage);
1651 * Our caller will very soon move newpage out of swapcache, but it's
1652 * a nice clean interface for us to replace oldpage by newpage there.
1654 xa_lock_irq(&swap_mapping->i_pages);
1655 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1657 old = page_folio(oldpage);
1658 new = page_folio(newpage);
1659 mem_cgroup_migrate(old, new);
1660 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1661 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1663 xa_unlock_irq(&swap_mapping->i_pages);
1665 if (unlikely(error)) {
1667 * Is this possible? I think not, now that our callers check
1668 * both PageSwapCache and page_private after getting page lock;
1669 * but be defensive. Reverse old to newpage for clear and free.
1673 lru_cache_add(newpage);
1677 ClearPageSwapCache(oldpage);
1678 set_page_private(oldpage, 0);
1680 unlock_page(oldpage);
1687 * Swap in the page pointed to by *pagep.
1688 * Caller has to make sure that *pagep contains a valid swapped page.
1689 * Returns 0 and the page in pagep if success. On failure, returns the
1690 * error code and NULL in *pagep.
1692 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1693 struct page **pagep, enum sgp_type sgp,
1694 gfp_t gfp, struct vm_area_struct *vma,
1695 vm_fault_t *fault_type)
1697 struct address_space *mapping = inode->i_mapping;
1698 struct shmem_inode_info *info = SHMEM_I(inode);
1699 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1704 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1705 swap = radix_to_swp_entry(*pagep);
1708 /* Look it up and read it in.. */
1709 page = lookup_swap_cache(swap, NULL, 0);
1711 /* Or update major stats only when swapin succeeds?? */
1713 *fault_type |= VM_FAULT_MAJOR;
1714 count_vm_event(PGMAJFAULT);
1715 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1717 /* Here we actually start the io */
1718 page = shmem_swapin(swap, gfp, info, index);
1725 /* We have to do this with page locked to prevent races */
1727 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1728 !shmem_confirm_swap(mapping, index, swap)) {
1732 if (!PageUptodate(page)) {
1736 wait_on_page_writeback(page);
1739 * Some architectures may have to restore extra metadata to the
1740 * physical page after reading from swap.
1742 arch_swap_restore(swap, page);
1744 if (shmem_should_replace_page(page, gfp)) {
1745 error = shmem_replace_page(&page, gfp, info, index);
1750 error = shmem_add_to_page_cache(page, mapping, index,
1751 swp_to_radix_entry(swap), gfp,
1756 spin_lock_irq(&info->lock);
1758 shmem_recalc_inode(inode);
1759 spin_unlock_irq(&info->lock);
1761 if (sgp == SGP_WRITE)
1762 mark_page_accessed(page);
1764 delete_from_swap_cache(page);
1765 set_page_dirty(page);
1771 if (!shmem_confirm_swap(mapping, index, swap))
1783 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1785 * If we allocate a new one we do not mark it dirty. That's up to the
1786 * vm. If we swap it in we mark it dirty since we also free the swap
1787 * entry since a page cannot live in both the swap and page cache.
1789 * vma, vmf, and fault_type are only supplied by shmem_fault:
1790 * otherwise they are NULL.
1792 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1793 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1794 struct vm_area_struct *vma, struct vm_fault *vmf,
1795 vm_fault_t *fault_type)
1797 struct address_space *mapping = inode->i_mapping;
1798 struct shmem_inode_info *info = SHMEM_I(inode);
1799 struct shmem_sb_info *sbinfo;
1800 struct mm_struct *charge_mm;
1802 pgoff_t hindex = index;
1808 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1811 if (sgp <= SGP_CACHE &&
1812 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1816 sbinfo = SHMEM_SB(inode->i_sb);
1817 charge_mm = vma ? vma->vm_mm : NULL;
1819 page = pagecache_get_page(mapping, index,
1820 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1822 if (page && vma && userfaultfd_minor(vma)) {
1823 if (!xa_is_value(page)) {
1827 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1831 if (xa_is_value(page)) {
1832 error = shmem_swapin_page(inode, index, &page,
1833 sgp, gfp, vma, fault_type);
1834 if (error == -EEXIST)
1842 hindex = page->index;
1843 if (sgp == SGP_WRITE)
1844 mark_page_accessed(page);
1845 if (PageUptodate(page))
1847 /* fallocated page */
1848 if (sgp != SGP_READ)
1855 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1856 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1859 if (sgp == SGP_READ)
1861 if (sgp == SGP_NOALLOC)
1865 * Fast cache lookup and swap lookup did not find it: allocate.
1868 if (vma && userfaultfd_missing(vma)) {
1869 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1873 /* Never use a huge page for shmem_symlink() */
1874 if (S_ISLNK(inode->i_mode))
1876 if (!shmem_is_huge(vma, inode, index))
1879 huge_gfp = vma_thp_gfp_mask(vma);
1880 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1881 page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true);
1884 page = shmem_alloc_and_acct_page(gfp, inode,
1890 error = PTR_ERR(page);
1892 if (error != -ENOSPC)
1895 * Try to reclaim some space by splitting a huge page
1896 * beyond i_size on the filesystem.
1901 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1902 if (ret == SHRINK_STOP)
1910 if (PageTransHuge(page))
1911 hindex = round_down(index, HPAGE_PMD_NR);
1915 if (sgp == SGP_WRITE)
1916 __SetPageReferenced(page);
1918 error = shmem_add_to_page_cache(page, mapping, hindex,
1919 NULL, gfp & GFP_RECLAIM_MASK,
1923 lru_cache_add(page);
1925 spin_lock_irq(&info->lock);
1926 info->alloced += compound_nr(page);
1927 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1928 shmem_recalc_inode(inode);
1929 spin_unlock_irq(&info->lock);
1932 if (PageTransHuge(page) &&
1933 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1934 hindex + HPAGE_PMD_NR - 1) {
1936 * Part of the huge page is beyond i_size: subject
1937 * to shrink under memory pressure.
1939 spin_lock(&sbinfo->shrinklist_lock);
1941 * _careful to defend against unlocked access to
1942 * ->shrink_list in shmem_unused_huge_shrink()
1944 if (list_empty_careful(&info->shrinklist)) {
1945 list_add_tail(&info->shrinklist,
1946 &sbinfo->shrinklist);
1947 sbinfo->shrinklist_len++;
1949 spin_unlock(&sbinfo->shrinklist_lock);
1953 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1955 if (sgp == SGP_FALLOC)
1959 * Let SGP_WRITE caller clear ends if write does not fill page;
1960 * but SGP_FALLOC on a page fallocated earlier must initialize
1961 * it now, lest undo on failure cancel our earlier guarantee.
1963 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1966 for (i = 0; i < compound_nr(page); i++) {
1967 clear_highpage(page + i);
1968 flush_dcache_page(page + i);
1970 SetPageUptodate(page);
1973 /* Perhaps the file has been truncated since we checked */
1974 if (sgp <= SGP_CACHE &&
1975 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1977 ClearPageDirty(page);
1978 delete_from_page_cache(page);
1979 spin_lock_irq(&info->lock);
1980 shmem_recalc_inode(inode);
1981 spin_unlock_irq(&info->lock);
1987 *pagep = page + index - hindex;
1994 shmem_inode_unacct_blocks(inode, compound_nr(page));
1996 if (PageTransHuge(page)) {
2006 if (error == -ENOSPC && !once++) {
2007 spin_lock_irq(&info->lock);
2008 shmem_recalc_inode(inode);
2009 spin_unlock_irq(&info->lock);
2012 if (error == -EEXIST)
2018 * This is like autoremove_wake_function, but it removes the wait queue
2019 * entry unconditionally - even if something else had already woken the
2022 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2024 int ret = default_wake_function(wait, mode, sync, key);
2025 list_del_init(&wait->entry);
2029 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2031 struct vm_area_struct *vma = vmf->vma;
2032 struct inode *inode = file_inode(vma->vm_file);
2033 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2035 vm_fault_t ret = VM_FAULT_LOCKED;
2038 * Trinity finds that probing a hole which tmpfs is punching can
2039 * prevent the hole-punch from ever completing: which in turn
2040 * locks writers out with its hold on i_rwsem. So refrain from
2041 * faulting pages into the hole while it's being punched. Although
2042 * shmem_undo_range() does remove the additions, it may be unable to
2043 * keep up, as each new page needs its own unmap_mapping_range() call,
2044 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2046 * It does not matter if we sometimes reach this check just before the
2047 * hole-punch begins, so that one fault then races with the punch:
2048 * we just need to make racing faults a rare case.
2050 * The implementation below would be much simpler if we just used a
2051 * standard mutex or completion: but we cannot take i_rwsem in fault,
2052 * and bloating every shmem inode for this unlikely case would be sad.
2054 if (unlikely(inode->i_private)) {
2055 struct shmem_falloc *shmem_falloc;
2057 spin_lock(&inode->i_lock);
2058 shmem_falloc = inode->i_private;
2060 shmem_falloc->waitq &&
2061 vmf->pgoff >= shmem_falloc->start &&
2062 vmf->pgoff < shmem_falloc->next) {
2064 wait_queue_head_t *shmem_falloc_waitq;
2065 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2067 ret = VM_FAULT_NOPAGE;
2068 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2070 ret = VM_FAULT_RETRY;
2072 shmem_falloc_waitq = shmem_falloc->waitq;
2073 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2074 TASK_UNINTERRUPTIBLE);
2075 spin_unlock(&inode->i_lock);
2079 * shmem_falloc_waitq points into the shmem_fallocate()
2080 * stack of the hole-punching task: shmem_falloc_waitq
2081 * is usually invalid by the time we reach here, but
2082 * finish_wait() does not dereference it in that case;
2083 * though i_lock needed lest racing with wake_up_all().
2085 spin_lock(&inode->i_lock);
2086 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2087 spin_unlock(&inode->i_lock);
2093 spin_unlock(&inode->i_lock);
2096 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE,
2097 gfp, vma, vmf, &ret);
2099 return vmf_error(err);
2103 unsigned long shmem_get_unmapped_area(struct file *file,
2104 unsigned long uaddr, unsigned long len,
2105 unsigned long pgoff, unsigned long flags)
2107 unsigned long (*get_area)(struct file *,
2108 unsigned long, unsigned long, unsigned long, unsigned long);
2110 unsigned long offset;
2111 unsigned long inflated_len;
2112 unsigned long inflated_addr;
2113 unsigned long inflated_offset;
2115 if (len > TASK_SIZE)
2118 get_area = current->mm->get_unmapped_area;
2119 addr = get_area(file, uaddr, len, pgoff, flags);
2121 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2123 if (IS_ERR_VALUE(addr))
2125 if (addr & ~PAGE_MASK)
2127 if (addr > TASK_SIZE - len)
2130 if (shmem_huge == SHMEM_HUGE_DENY)
2132 if (len < HPAGE_PMD_SIZE)
2134 if (flags & MAP_FIXED)
2137 * Our priority is to support MAP_SHARED mapped hugely;
2138 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2139 * But if caller specified an address hint and we allocated area there
2140 * successfully, respect that as before.
2145 if (shmem_huge != SHMEM_HUGE_FORCE) {
2146 struct super_block *sb;
2149 VM_BUG_ON(file->f_op != &shmem_file_operations);
2150 sb = file_inode(file)->i_sb;
2153 * Called directly from mm/mmap.c, or drivers/char/mem.c
2154 * for "/dev/zero", to create a shared anonymous object.
2156 if (IS_ERR(shm_mnt))
2158 sb = shm_mnt->mnt_sb;
2160 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2164 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2165 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2167 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2170 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2171 if (inflated_len > TASK_SIZE)
2173 if (inflated_len < len)
2176 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2177 if (IS_ERR_VALUE(inflated_addr))
2179 if (inflated_addr & ~PAGE_MASK)
2182 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2183 inflated_addr += offset - inflated_offset;
2184 if (inflated_offset > offset)
2185 inflated_addr += HPAGE_PMD_SIZE;
2187 if (inflated_addr > TASK_SIZE - len)
2189 return inflated_addr;
2193 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2195 struct inode *inode = file_inode(vma->vm_file);
2196 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2199 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2202 struct inode *inode = file_inode(vma->vm_file);
2205 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2206 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2210 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2212 struct inode *inode = file_inode(file);
2213 struct shmem_inode_info *info = SHMEM_I(inode);
2214 int retval = -ENOMEM;
2217 * What serializes the accesses to info->flags?
2218 * ipc_lock_object() when called from shmctl_do_lock(),
2219 * no serialization needed when called from shm_destroy().
2221 if (lock && !(info->flags & VM_LOCKED)) {
2222 if (!user_shm_lock(inode->i_size, ucounts))
2224 info->flags |= VM_LOCKED;
2225 mapping_set_unevictable(file->f_mapping);
2227 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2228 user_shm_unlock(inode->i_size, ucounts);
2229 info->flags &= ~VM_LOCKED;
2230 mapping_clear_unevictable(file->f_mapping);
2238 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2240 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2243 ret = seal_check_future_write(info->seals, vma);
2247 /* arm64 - allow memory tagging on RAM-based files */
2248 vma->vm_flags |= VM_MTE_ALLOWED;
2250 file_accessed(file);
2251 vma->vm_ops = &shmem_vm_ops;
2252 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2253 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2254 (vma->vm_end & HPAGE_PMD_MASK)) {
2255 khugepaged_enter(vma, vma->vm_flags);
2260 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2261 umode_t mode, dev_t dev, unsigned long flags)
2263 struct inode *inode;
2264 struct shmem_inode_info *info;
2265 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2268 if (shmem_reserve_inode(sb, &ino))
2271 inode = new_inode(sb);
2274 inode_init_owner(&init_user_ns, inode, dir, mode);
2275 inode->i_blocks = 0;
2276 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2277 inode->i_generation = prandom_u32();
2278 info = SHMEM_I(inode);
2279 memset(info, 0, (char *)inode - (char *)info);
2280 spin_lock_init(&info->lock);
2281 atomic_set(&info->stop_eviction, 0);
2282 info->seals = F_SEAL_SEAL;
2283 info->flags = flags & VM_NORESERVE;
2284 INIT_LIST_HEAD(&info->shrinklist);
2285 INIT_LIST_HEAD(&info->swaplist);
2286 simple_xattrs_init(&info->xattrs);
2287 cache_no_acl(inode);
2288 mapping_set_large_folios(inode->i_mapping);
2290 switch (mode & S_IFMT) {
2292 inode->i_op = &shmem_special_inode_operations;
2293 init_special_inode(inode, mode, dev);
2296 inode->i_mapping->a_ops = &shmem_aops;
2297 inode->i_op = &shmem_inode_operations;
2298 inode->i_fop = &shmem_file_operations;
2299 mpol_shared_policy_init(&info->policy,
2300 shmem_get_sbmpol(sbinfo));
2304 /* Some things misbehave if size == 0 on a directory */
2305 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2306 inode->i_op = &shmem_dir_inode_operations;
2307 inode->i_fop = &simple_dir_operations;
2311 * Must not load anything in the rbtree,
2312 * mpol_free_shared_policy will not be called.
2314 mpol_shared_policy_init(&info->policy, NULL);
2318 lockdep_annotate_inode_mutex_key(inode);
2320 shmem_free_inode(sb);
2324 #ifdef CONFIG_USERFAULTFD
2325 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2327 struct vm_area_struct *dst_vma,
2328 unsigned long dst_addr,
2329 unsigned long src_addr,
2331 struct page **pagep)
2333 struct inode *inode = file_inode(dst_vma->vm_file);
2334 struct shmem_inode_info *info = SHMEM_I(inode);
2335 struct address_space *mapping = inode->i_mapping;
2336 gfp_t gfp = mapping_gfp_mask(mapping);
2337 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2343 if (!shmem_inode_acct_block(inode, 1)) {
2345 * We may have got a page, returned -ENOENT triggering a retry,
2346 * and now we find ourselves with -ENOMEM. Release the page, to
2347 * avoid a BUG_ON in our caller.
2349 if (unlikely(*pagep)) {
2358 page = shmem_alloc_page(gfp, info, pgoff);
2360 goto out_unacct_blocks;
2362 if (!zeropage) { /* COPY */
2363 page_kaddr = kmap_atomic(page);
2364 ret = copy_from_user(page_kaddr,
2365 (const void __user *)src_addr,
2367 kunmap_atomic(page_kaddr);
2369 /* fallback to copy_from_user outside mmap_lock */
2370 if (unlikely(ret)) {
2373 /* don't free the page */
2374 goto out_unacct_blocks;
2376 } else { /* ZEROPAGE */
2377 clear_highpage(page);
2384 VM_BUG_ON(PageLocked(page));
2385 VM_BUG_ON(PageSwapBacked(page));
2386 __SetPageLocked(page);
2387 __SetPageSwapBacked(page);
2388 __SetPageUptodate(page);
2391 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2392 if (unlikely(pgoff >= max_off))
2395 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2396 gfp & GFP_RECLAIM_MASK, dst_mm);
2400 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2403 goto out_delete_from_cache;
2405 spin_lock_irq(&info->lock);
2407 inode->i_blocks += BLOCKS_PER_PAGE;
2408 shmem_recalc_inode(inode);
2409 spin_unlock_irq(&info->lock);
2413 out_delete_from_cache:
2414 delete_from_page_cache(page);
2419 shmem_inode_unacct_blocks(inode, 1);
2422 #endif /* CONFIG_USERFAULTFD */
2425 static const struct inode_operations shmem_symlink_inode_operations;
2426 static const struct inode_operations shmem_short_symlink_operations;
2428 #ifdef CONFIG_TMPFS_XATTR
2429 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2431 #define shmem_initxattrs NULL
2435 shmem_write_begin(struct file *file, struct address_space *mapping,
2436 loff_t pos, unsigned len, unsigned flags,
2437 struct page **pagep, void **fsdata)
2439 struct inode *inode = mapping->host;
2440 struct shmem_inode_info *info = SHMEM_I(inode);
2441 pgoff_t index = pos >> PAGE_SHIFT;
2443 /* i_rwsem is held by caller */
2444 if (unlikely(info->seals & (F_SEAL_GROW |
2445 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2446 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2448 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2452 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2456 shmem_write_end(struct file *file, struct address_space *mapping,
2457 loff_t pos, unsigned len, unsigned copied,
2458 struct page *page, void *fsdata)
2460 struct inode *inode = mapping->host;
2462 if (pos + copied > inode->i_size)
2463 i_size_write(inode, pos + copied);
2465 if (!PageUptodate(page)) {
2466 struct page *head = compound_head(page);
2467 if (PageTransCompound(page)) {
2470 for (i = 0; i < HPAGE_PMD_NR; i++) {
2471 if (head + i == page)
2473 clear_highpage(head + i);
2474 flush_dcache_page(head + i);
2477 if (copied < PAGE_SIZE) {
2478 unsigned from = pos & (PAGE_SIZE - 1);
2479 zero_user_segments(page, 0, from,
2480 from + copied, PAGE_SIZE);
2482 SetPageUptodate(head);
2484 set_page_dirty(page);
2491 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2493 struct file *file = iocb->ki_filp;
2494 struct inode *inode = file_inode(file);
2495 struct address_space *mapping = inode->i_mapping;
2497 unsigned long offset;
2498 enum sgp_type sgp = SGP_READ;
2501 loff_t *ppos = &iocb->ki_pos;
2504 * Might this read be for a stacking filesystem? Then when reading
2505 * holes of a sparse file, we actually need to allocate those pages,
2506 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2508 if (!iter_is_iovec(to))
2511 index = *ppos >> PAGE_SHIFT;
2512 offset = *ppos & ~PAGE_MASK;
2515 struct page *page = NULL;
2517 unsigned long nr, ret;
2518 loff_t i_size = i_size_read(inode);
2520 end_index = i_size >> PAGE_SHIFT;
2521 if (index > end_index)
2523 if (index == end_index) {
2524 nr = i_size & ~PAGE_MASK;
2529 error = shmem_getpage(inode, index, &page, sgp);
2531 if (error == -EINVAL)
2536 if (sgp == SGP_CACHE)
2537 set_page_dirty(page);
2542 * We must evaluate after, since reads (unlike writes)
2543 * are called without i_rwsem protection against truncate
2546 i_size = i_size_read(inode);
2547 end_index = i_size >> PAGE_SHIFT;
2548 if (index == end_index) {
2549 nr = i_size & ~PAGE_MASK;
2560 * If users can be writing to this page using arbitrary
2561 * virtual addresses, take care about potential aliasing
2562 * before reading the page on the kernel side.
2564 if (mapping_writably_mapped(mapping))
2565 flush_dcache_page(page);
2567 * Mark the page accessed if we read the beginning.
2570 mark_page_accessed(page);
2572 page = ZERO_PAGE(0);
2577 * Ok, we have the page, and it's up-to-date, so
2578 * now we can copy it to user space...
2580 ret = copy_page_to_iter(page, offset, nr, to);
2583 index += offset >> PAGE_SHIFT;
2584 offset &= ~PAGE_MASK;
2587 if (!iov_iter_count(to))
2596 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2597 file_accessed(file);
2598 return retval ? retval : error;
2601 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2603 struct address_space *mapping = file->f_mapping;
2604 struct inode *inode = mapping->host;
2606 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2607 return generic_file_llseek_size(file, offset, whence,
2608 MAX_LFS_FILESIZE, i_size_read(inode));
2613 /* We're holding i_rwsem so we can access i_size directly */
2614 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2616 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2617 inode_unlock(inode);
2621 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2624 struct inode *inode = file_inode(file);
2625 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2626 struct shmem_inode_info *info = SHMEM_I(inode);
2627 struct shmem_falloc shmem_falloc;
2628 pgoff_t start, index, end, undo_fallocend;
2631 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2636 if (mode & FALLOC_FL_PUNCH_HOLE) {
2637 struct address_space *mapping = file->f_mapping;
2638 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2639 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2640 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2642 /* protected by i_rwsem */
2643 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2648 shmem_falloc.waitq = &shmem_falloc_waitq;
2649 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2650 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2651 spin_lock(&inode->i_lock);
2652 inode->i_private = &shmem_falloc;
2653 spin_unlock(&inode->i_lock);
2655 if ((u64)unmap_end > (u64)unmap_start)
2656 unmap_mapping_range(mapping, unmap_start,
2657 1 + unmap_end - unmap_start, 0);
2658 shmem_truncate_range(inode, offset, offset + len - 1);
2659 /* No need to unmap again: hole-punching leaves COWed pages */
2661 spin_lock(&inode->i_lock);
2662 inode->i_private = NULL;
2663 wake_up_all(&shmem_falloc_waitq);
2664 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2665 spin_unlock(&inode->i_lock);
2670 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2671 error = inode_newsize_ok(inode, offset + len);
2675 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2680 start = offset >> PAGE_SHIFT;
2681 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2682 /* Try to avoid a swapstorm if len is impossible to satisfy */
2683 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2688 shmem_falloc.waitq = NULL;
2689 shmem_falloc.start = start;
2690 shmem_falloc.next = start;
2691 shmem_falloc.nr_falloced = 0;
2692 shmem_falloc.nr_unswapped = 0;
2693 spin_lock(&inode->i_lock);
2694 inode->i_private = &shmem_falloc;
2695 spin_unlock(&inode->i_lock);
2698 * info->fallocend is only relevant when huge pages might be
2699 * involved: to prevent split_huge_page() freeing fallocated
2700 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2702 undo_fallocend = info->fallocend;
2703 if (info->fallocend < end)
2704 info->fallocend = end;
2706 for (index = start; index < end; ) {
2710 * Good, the fallocate(2) manpage permits EINTR: we may have
2711 * been interrupted because we are using up too much memory.
2713 if (signal_pending(current))
2715 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2718 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2720 info->fallocend = undo_fallocend;
2721 /* Remove the !PageUptodate pages we added */
2722 if (index > start) {
2723 shmem_undo_range(inode,
2724 (loff_t)start << PAGE_SHIFT,
2725 ((loff_t)index << PAGE_SHIFT) - 1, true);
2732 * Here is a more important optimization than it appears:
2733 * a second SGP_FALLOC on the same huge page will clear it,
2734 * making it PageUptodate and un-undoable if we fail later.
2736 if (PageTransCompound(page)) {
2737 index = round_up(index, HPAGE_PMD_NR);
2738 /* Beware 32-bit wraparound */
2744 * Inform shmem_writepage() how far we have reached.
2745 * No need for lock or barrier: we have the page lock.
2747 if (!PageUptodate(page))
2748 shmem_falloc.nr_falloced += index - shmem_falloc.next;
2749 shmem_falloc.next = index;
2752 * If !PageUptodate, leave it that way so that freeable pages
2753 * can be recognized if we need to rollback on error later.
2754 * But set_page_dirty so that memory pressure will swap rather
2755 * than free the pages we are allocating (and SGP_CACHE pages
2756 * might still be clean: we now need to mark those dirty too).
2758 set_page_dirty(page);
2764 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2765 i_size_write(inode, offset + len);
2766 inode->i_ctime = current_time(inode);
2768 spin_lock(&inode->i_lock);
2769 inode->i_private = NULL;
2770 spin_unlock(&inode->i_lock);
2772 inode_unlock(inode);
2776 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2778 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2780 buf->f_type = TMPFS_MAGIC;
2781 buf->f_bsize = PAGE_SIZE;
2782 buf->f_namelen = NAME_MAX;
2783 if (sbinfo->max_blocks) {
2784 buf->f_blocks = sbinfo->max_blocks;
2786 buf->f_bfree = sbinfo->max_blocks -
2787 percpu_counter_sum(&sbinfo->used_blocks);
2789 if (sbinfo->max_inodes) {
2790 buf->f_files = sbinfo->max_inodes;
2791 buf->f_ffree = sbinfo->free_inodes;
2793 /* else leave those fields 0 like simple_statfs */
2795 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2801 * File creation. Allocate an inode, and we're done..
2804 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2805 struct dentry *dentry, umode_t mode, dev_t dev)
2807 struct inode *inode;
2808 int error = -ENOSPC;
2810 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2812 error = simple_acl_create(dir, inode);
2815 error = security_inode_init_security(inode, dir,
2817 shmem_initxattrs, NULL);
2818 if (error && error != -EOPNOTSUPP)
2822 dir->i_size += BOGO_DIRENT_SIZE;
2823 dir->i_ctime = dir->i_mtime = current_time(dir);
2824 d_instantiate(dentry, inode);
2825 dget(dentry); /* Extra count - pin the dentry in core */
2834 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2835 struct dentry *dentry, umode_t mode)
2837 struct inode *inode;
2838 int error = -ENOSPC;
2840 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2842 error = security_inode_init_security(inode, dir,
2844 shmem_initxattrs, NULL);
2845 if (error && error != -EOPNOTSUPP)
2847 error = simple_acl_create(dir, inode);
2850 d_tmpfile(dentry, inode);
2858 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2859 struct dentry *dentry, umode_t mode)
2863 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2864 mode | S_IFDIR, 0)))
2870 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2871 struct dentry *dentry, umode_t mode, bool excl)
2873 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2879 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2881 struct inode *inode = d_inode(old_dentry);
2885 * No ordinary (disk based) filesystem counts links as inodes;
2886 * but each new link needs a new dentry, pinning lowmem, and
2887 * tmpfs dentries cannot be pruned until they are unlinked.
2888 * But if an O_TMPFILE file is linked into the tmpfs, the
2889 * first link must skip that, to get the accounting right.
2891 if (inode->i_nlink) {
2892 ret = shmem_reserve_inode(inode->i_sb, NULL);
2897 dir->i_size += BOGO_DIRENT_SIZE;
2898 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2900 ihold(inode); /* New dentry reference */
2901 dget(dentry); /* Extra pinning count for the created dentry */
2902 d_instantiate(dentry, inode);
2907 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2909 struct inode *inode = d_inode(dentry);
2911 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2912 shmem_free_inode(inode->i_sb);
2914 dir->i_size -= BOGO_DIRENT_SIZE;
2915 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2917 dput(dentry); /* Undo the count from "create" - this does all the work */
2921 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2923 if (!simple_empty(dentry))
2926 drop_nlink(d_inode(dentry));
2928 return shmem_unlink(dir, dentry);
2931 static int shmem_whiteout(struct user_namespace *mnt_userns,
2932 struct inode *old_dir, struct dentry *old_dentry)
2934 struct dentry *whiteout;
2937 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2941 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2942 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2948 * Cheat and hash the whiteout while the old dentry is still in
2949 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2951 * d_lookup() will consistently find one of them at this point,
2952 * not sure which one, but that isn't even important.
2959 * The VFS layer already does all the dentry stuff for rename,
2960 * we just have to decrement the usage count for the target if
2961 * it exists so that the VFS layer correctly free's it when it
2964 static int shmem_rename2(struct user_namespace *mnt_userns,
2965 struct inode *old_dir, struct dentry *old_dentry,
2966 struct inode *new_dir, struct dentry *new_dentry,
2969 struct inode *inode = d_inode(old_dentry);
2970 int they_are_dirs = S_ISDIR(inode->i_mode);
2972 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2975 if (flags & RENAME_EXCHANGE)
2976 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
2978 if (!simple_empty(new_dentry))
2981 if (flags & RENAME_WHITEOUT) {
2984 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
2989 if (d_really_is_positive(new_dentry)) {
2990 (void) shmem_unlink(new_dir, new_dentry);
2991 if (they_are_dirs) {
2992 drop_nlink(d_inode(new_dentry));
2993 drop_nlink(old_dir);
2995 } else if (they_are_dirs) {
2996 drop_nlink(old_dir);
3000 old_dir->i_size -= BOGO_DIRENT_SIZE;
3001 new_dir->i_size += BOGO_DIRENT_SIZE;
3002 old_dir->i_ctime = old_dir->i_mtime =
3003 new_dir->i_ctime = new_dir->i_mtime =
3004 inode->i_ctime = current_time(old_dir);
3008 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3009 struct dentry *dentry, const char *symname)
3013 struct inode *inode;
3016 len = strlen(symname) + 1;
3017 if (len > PAGE_SIZE)
3018 return -ENAMETOOLONG;
3020 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3025 error = security_inode_init_security(inode, dir, &dentry->d_name,
3026 shmem_initxattrs, NULL);
3027 if (error && error != -EOPNOTSUPP) {
3032 inode->i_size = len-1;
3033 if (len <= SHORT_SYMLINK_LEN) {
3034 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3035 if (!inode->i_link) {
3039 inode->i_op = &shmem_short_symlink_operations;
3041 inode_nohighmem(inode);
3042 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3047 inode->i_mapping->a_ops = &shmem_aops;
3048 inode->i_op = &shmem_symlink_inode_operations;
3049 memcpy(page_address(page), symname, len);
3050 SetPageUptodate(page);
3051 set_page_dirty(page);
3055 dir->i_size += BOGO_DIRENT_SIZE;
3056 dir->i_ctime = dir->i_mtime = current_time(dir);
3057 d_instantiate(dentry, inode);
3062 static void shmem_put_link(void *arg)
3064 mark_page_accessed(arg);
3068 static const char *shmem_get_link(struct dentry *dentry,
3069 struct inode *inode,
3070 struct delayed_call *done)
3072 struct page *page = NULL;
3075 page = find_get_page(inode->i_mapping, 0);
3077 return ERR_PTR(-ECHILD);
3078 if (!PageUptodate(page)) {
3080 return ERR_PTR(-ECHILD);
3083 error = shmem_getpage(inode, 0, &page, SGP_READ);
3085 return ERR_PTR(error);
3088 set_delayed_call(done, shmem_put_link, page);
3089 return page_address(page);
3092 #ifdef CONFIG_TMPFS_XATTR
3094 * Superblocks without xattr inode operations may get some security.* xattr
3095 * support from the LSM "for free". As soon as we have any other xattrs
3096 * like ACLs, we also need to implement the security.* handlers at
3097 * filesystem level, though.
3101 * Callback for security_inode_init_security() for acquiring xattrs.
3103 static int shmem_initxattrs(struct inode *inode,
3104 const struct xattr *xattr_array,
3107 struct shmem_inode_info *info = SHMEM_I(inode);
3108 const struct xattr *xattr;
3109 struct simple_xattr *new_xattr;
3112 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3113 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3117 len = strlen(xattr->name) + 1;
3118 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3120 if (!new_xattr->name) {
3125 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3126 XATTR_SECURITY_PREFIX_LEN);
3127 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3130 simple_xattr_list_add(&info->xattrs, new_xattr);
3136 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3137 struct dentry *unused, struct inode *inode,
3138 const char *name, void *buffer, size_t size)
3140 struct shmem_inode_info *info = SHMEM_I(inode);
3142 name = xattr_full_name(handler, name);
3143 return simple_xattr_get(&info->xattrs, name, buffer, size);
3146 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3147 struct user_namespace *mnt_userns,
3148 struct dentry *unused, struct inode *inode,
3149 const char *name, const void *value,
3150 size_t size, int flags)
3152 struct shmem_inode_info *info = SHMEM_I(inode);
3154 name = xattr_full_name(handler, name);
3155 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3158 static const struct xattr_handler shmem_security_xattr_handler = {
3159 .prefix = XATTR_SECURITY_PREFIX,
3160 .get = shmem_xattr_handler_get,
3161 .set = shmem_xattr_handler_set,
3164 static const struct xattr_handler shmem_trusted_xattr_handler = {
3165 .prefix = XATTR_TRUSTED_PREFIX,
3166 .get = shmem_xattr_handler_get,
3167 .set = shmem_xattr_handler_set,
3170 static const struct xattr_handler *shmem_xattr_handlers[] = {
3171 #ifdef CONFIG_TMPFS_POSIX_ACL
3172 &posix_acl_access_xattr_handler,
3173 &posix_acl_default_xattr_handler,
3175 &shmem_security_xattr_handler,
3176 &shmem_trusted_xattr_handler,
3180 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3182 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3183 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3185 #endif /* CONFIG_TMPFS_XATTR */
3187 static const struct inode_operations shmem_short_symlink_operations = {
3188 .get_link = simple_get_link,
3189 #ifdef CONFIG_TMPFS_XATTR
3190 .listxattr = shmem_listxattr,
3194 static const struct inode_operations shmem_symlink_inode_operations = {
3195 .get_link = shmem_get_link,
3196 #ifdef CONFIG_TMPFS_XATTR
3197 .listxattr = shmem_listxattr,
3201 static struct dentry *shmem_get_parent(struct dentry *child)
3203 return ERR_PTR(-ESTALE);
3206 static int shmem_match(struct inode *ino, void *vfh)
3210 inum = (inum << 32) | fh[1];
3211 return ino->i_ino == inum && fh[0] == ino->i_generation;
3214 /* Find any alias of inode, but prefer a hashed alias */
3215 static struct dentry *shmem_find_alias(struct inode *inode)
3217 struct dentry *alias = d_find_alias(inode);
3219 return alias ?: d_find_any_alias(inode);
3223 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3224 struct fid *fid, int fh_len, int fh_type)
3226 struct inode *inode;
3227 struct dentry *dentry = NULL;
3234 inum = (inum << 32) | fid->raw[1];
3236 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3237 shmem_match, fid->raw);
3239 dentry = shmem_find_alias(inode);
3246 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3247 struct inode *parent)
3251 return FILEID_INVALID;
3254 if (inode_unhashed(inode)) {
3255 /* Unfortunately insert_inode_hash is not idempotent,
3256 * so as we hash inodes here rather than at creation
3257 * time, we need a lock to ensure we only try
3260 static DEFINE_SPINLOCK(lock);
3262 if (inode_unhashed(inode))
3263 __insert_inode_hash(inode,
3264 inode->i_ino + inode->i_generation);
3268 fh[0] = inode->i_generation;
3269 fh[1] = inode->i_ino;
3270 fh[2] = ((__u64)inode->i_ino) >> 32;
3276 static const struct export_operations shmem_export_ops = {
3277 .get_parent = shmem_get_parent,
3278 .encode_fh = shmem_encode_fh,
3279 .fh_to_dentry = shmem_fh_to_dentry,
3295 static const struct constant_table shmem_param_enums_huge[] = {
3296 {"never", SHMEM_HUGE_NEVER },
3297 {"always", SHMEM_HUGE_ALWAYS },
3298 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3299 {"advise", SHMEM_HUGE_ADVISE },
3303 const struct fs_parameter_spec shmem_fs_parameters[] = {
3304 fsparam_u32 ("gid", Opt_gid),
3305 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3306 fsparam_u32oct("mode", Opt_mode),
3307 fsparam_string("mpol", Opt_mpol),
3308 fsparam_string("nr_blocks", Opt_nr_blocks),
3309 fsparam_string("nr_inodes", Opt_nr_inodes),
3310 fsparam_string("size", Opt_size),
3311 fsparam_u32 ("uid", Opt_uid),
3312 fsparam_flag ("inode32", Opt_inode32),
3313 fsparam_flag ("inode64", Opt_inode64),
3317 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3319 struct shmem_options *ctx = fc->fs_private;
3320 struct fs_parse_result result;
3321 unsigned long long size;
3325 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3331 size = memparse(param->string, &rest);
3333 size <<= PAGE_SHIFT;
3334 size *= totalram_pages();
3340 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3341 ctx->seen |= SHMEM_SEEN_BLOCKS;
3344 ctx->blocks = memparse(param->string, &rest);
3347 ctx->seen |= SHMEM_SEEN_BLOCKS;
3350 ctx->inodes = memparse(param->string, &rest);
3353 ctx->seen |= SHMEM_SEEN_INODES;
3356 ctx->mode = result.uint_32 & 07777;
3359 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3360 if (!uid_valid(ctx->uid))
3364 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3365 if (!gid_valid(ctx->gid))
3369 ctx->huge = result.uint_32;
3370 if (ctx->huge != SHMEM_HUGE_NEVER &&
3371 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3372 has_transparent_hugepage()))
3373 goto unsupported_parameter;
3374 ctx->seen |= SHMEM_SEEN_HUGE;
3377 if (IS_ENABLED(CONFIG_NUMA)) {
3378 mpol_put(ctx->mpol);
3380 if (mpol_parse_str(param->string, &ctx->mpol))
3384 goto unsupported_parameter;
3386 ctx->full_inums = false;
3387 ctx->seen |= SHMEM_SEEN_INUMS;
3390 if (sizeof(ino_t) < 8) {
3392 "Cannot use inode64 with <64bit inums in kernel\n");
3394 ctx->full_inums = true;
3395 ctx->seen |= SHMEM_SEEN_INUMS;
3400 unsupported_parameter:
3401 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3403 return invalfc(fc, "Bad value for '%s'", param->key);
3406 static int shmem_parse_options(struct fs_context *fc, void *data)
3408 char *options = data;
3411 int err = security_sb_eat_lsm_opts(options, &fc->security);
3416 while (options != NULL) {
3417 char *this_char = options;
3420 * NUL-terminate this option: unfortunately,
3421 * mount options form a comma-separated list,
3422 * but mpol's nodelist may also contain commas.
3424 options = strchr(options, ',');
3425 if (options == NULL)
3428 if (!isdigit(*options)) {
3434 char *value = strchr(this_char, '=');
3440 len = strlen(value);
3442 err = vfs_parse_fs_string(fc, this_char, value, len);
3451 * Reconfigure a shmem filesystem.
3453 * Note that we disallow change from limited->unlimited blocks/inodes while any
3454 * are in use; but we must separately disallow unlimited->limited, because in
3455 * that case we have no record of how much is already in use.
3457 static int shmem_reconfigure(struct fs_context *fc)
3459 struct shmem_options *ctx = fc->fs_private;
3460 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3461 unsigned long inodes;
3462 struct mempolicy *mpol = NULL;
3465 raw_spin_lock(&sbinfo->stat_lock);
3466 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3467 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3468 if (!sbinfo->max_blocks) {
3469 err = "Cannot retroactively limit size";
3472 if (percpu_counter_compare(&sbinfo->used_blocks,
3474 err = "Too small a size for current use";
3478 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3479 if (!sbinfo->max_inodes) {
3480 err = "Cannot retroactively limit inodes";
3483 if (ctx->inodes < inodes) {
3484 err = "Too few inodes for current use";
3489 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3490 sbinfo->next_ino > UINT_MAX) {
3491 err = "Current inum too high to switch to 32-bit inums";
3495 if (ctx->seen & SHMEM_SEEN_HUGE)
3496 sbinfo->huge = ctx->huge;
3497 if (ctx->seen & SHMEM_SEEN_INUMS)
3498 sbinfo->full_inums = ctx->full_inums;
3499 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3500 sbinfo->max_blocks = ctx->blocks;
3501 if (ctx->seen & SHMEM_SEEN_INODES) {
3502 sbinfo->max_inodes = ctx->inodes;
3503 sbinfo->free_inodes = ctx->inodes - inodes;
3507 * Preserve previous mempolicy unless mpol remount option was specified.
3510 mpol = sbinfo->mpol;
3511 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3514 raw_spin_unlock(&sbinfo->stat_lock);
3518 raw_spin_unlock(&sbinfo->stat_lock);
3519 return invalfc(fc, "%s", err);
3522 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3524 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3526 if (sbinfo->max_blocks != shmem_default_max_blocks())
3527 seq_printf(seq, ",size=%luk",
3528 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3529 if (sbinfo->max_inodes != shmem_default_max_inodes())
3530 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3531 if (sbinfo->mode != (0777 | S_ISVTX))
3532 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3533 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3534 seq_printf(seq, ",uid=%u",
3535 from_kuid_munged(&init_user_ns, sbinfo->uid));
3536 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3537 seq_printf(seq, ",gid=%u",
3538 from_kgid_munged(&init_user_ns, sbinfo->gid));
3541 * Showing inode{64,32} might be useful even if it's the system default,
3542 * since then people don't have to resort to checking both here and
3543 * /proc/config.gz to confirm 64-bit inums were successfully applied
3544 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3546 * We hide it when inode64 isn't the default and we are using 32-bit
3547 * inodes, since that probably just means the feature isn't even under
3552 * +-----------------+-----------------+
3553 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3554 * +------------------+-----------------+-----------------+
3555 * | full_inums=true | show | show |
3556 * | full_inums=false | show | hide |
3557 * +------------------+-----------------+-----------------+
3560 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3561 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3562 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3563 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3565 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3567 shmem_show_mpol(seq, sbinfo->mpol);
3571 #endif /* CONFIG_TMPFS */
3573 static void shmem_put_super(struct super_block *sb)
3575 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3577 free_percpu(sbinfo->ino_batch);
3578 percpu_counter_destroy(&sbinfo->used_blocks);
3579 mpol_put(sbinfo->mpol);
3581 sb->s_fs_info = NULL;
3584 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3586 struct shmem_options *ctx = fc->fs_private;
3587 struct inode *inode;
3588 struct shmem_sb_info *sbinfo;
3590 /* Round up to L1_CACHE_BYTES to resist false sharing */
3591 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3592 L1_CACHE_BYTES), GFP_KERNEL);
3596 sb->s_fs_info = sbinfo;
3600 * Per default we only allow half of the physical ram per
3601 * tmpfs instance, limiting inodes to one per page of lowmem;
3602 * but the internal instance is left unlimited.
3604 if (!(sb->s_flags & SB_KERNMOUNT)) {
3605 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3606 ctx->blocks = shmem_default_max_blocks();
3607 if (!(ctx->seen & SHMEM_SEEN_INODES))
3608 ctx->inodes = shmem_default_max_inodes();
3609 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3610 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3612 sb->s_flags |= SB_NOUSER;
3614 sb->s_export_op = &shmem_export_ops;
3615 sb->s_flags |= SB_NOSEC;
3617 sb->s_flags |= SB_NOUSER;
3619 sbinfo->max_blocks = ctx->blocks;
3620 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3621 if (sb->s_flags & SB_KERNMOUNT) {
3622 sbinfo->ino_batch = alloc_percpu(ino_t);
3623 if (!sbinfo->ino_batch)
3626 sbinfo->uid = ctx->uid;
3627 sbinfo->gid = ctx->gid;
3628 sbinfo->full_inums = ctx->full_inums;
3629 sbinfo->mode = ctx->mode;
3630 sbinfo->huge = ctx->huge;
3631 sbinfo->mpol = ctx->mpol;
3634 raw_spin_lock_init(&sbinfo->stat_lock);
3635 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3637 spin_lock_init(&sbinfo->shrinklist_lock);
3638 INIT_LIST_HEAD(&sbinfo->shrinklist);
3640 sb->s_maxbytes = MAX_LFS_FILESIZE;
3641 sb->s_blocksize = PAGE_SIZE;
3642 sb->s_blocksize_bits = PAGE_SHIFT;
3643 sb->s_magic = TMPFS_MAGIC;
3644 sb->s_op = &shmem_ops;
3645 sb->s_time_gran = 1;
3646 #ifdef CONFIG_TMPFS_XATTR
3647 sb->s_xattr = shmem_xattr_handlers;
3649 #ifdef CONFIG_TMPFS_POSIX_ACL
3650 sb->s_flags |= SB_POSIXACL;
3652 uuid_gen(&sb->s_uuid);
3654 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3657 inode->i_uid = sbinfo->uid;
3658 inode->i_gid = sbinfo->gid;
3659 sb->s_root = d_make_root(inode);
3665 shmem_put_super(sb);
3669 static int shmem_get_tree(struct fs_context *fc)
3671 return get_tree_nodev(fc, shmem_fill_super);
3674 static void shmem_free_fc(struct fs_context *fc)
3676 struct shmem_options *ctx = fc->fs_private;
3679 mpol_put(ctx->mpol);
3684 static const struct fs_context_operations shmem_fs_context_ops = {
3685 .free = shmem_free_fc,
3686 .get_tree = shmem_get_tree,
3688 .parse_monolithic = shmem_parse_options,
3689 .parse_param = shmem_parse_one,
3690 .reconfigure = shmem_reconfigure,
3694 static struct kmem_cache *shmem_inode_cachep;
3696 static struct inode *shmem_alloc_inode(struct super_block *sb)
3698 struct shmem_inode_info *info;
3699 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3702 return &info->vfs_inode;
3705 static void shmem_free_in_core_inode(struct inode *inode)
3707 if (S_ISLNK(inode->i_mode))
3708 kfree(inode->i_link);
3709 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3712 static void shmem_destroy_inode(struct inode *inode)
3714 if (S_ISREG(inode->i_mode))
3715 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3718 static void shmem_init_inode(void *foo)
3720 struct shmem_inode_info *info = foo;
3721 inode_init_once(&info->vfs_inode);
3724 static void shmem_init_inodecache(void)
3726 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3727 sizeof(struct shmem_inode_info),
3728 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3731 static void shmem_destroy_inodecache(void)
3733 kmem_cache_destroy(shmem_inode_cachep);
3736 const struct address_space_operations shmem_aops = {
3737 .writepage = shmem_writepage,
3738 .set_page_dirty = __set_page_dirty_no_writeback,
3740 .write_begin = shmem_write_begin,
3741 .write_end = shmem_write_end,
3743 #ifdef CONFIG_MIGRATION
3744 .migratepage = migrate_page,
3746 .error_remove_page = generic_error_remove_page,
3748 EXPORT_SYMBOL(shmem_aops);
3750 static const struct file_operations shmem_file_operations = {
3752 .get_unmapped_area = shmem_get_unmapped_area,
3754 .llseek = shmem_file_llseek,
3755 .read_iter = shmem_file_read_iter,
3756 .write_iter = generic_file_write_iter,
3757 .fsync = noop_fsync,
3758 .splice_read = generic_file_splice_read,
3759 .splice_write = iter_file_splice_write,
3760 .fallocate = shmem_fallocate,
3764 static const struct inode_operations shmem_inode_operations = {
3765 .getattr = shmem_getattr,
3766 .setattr = shmem_setattr,
3767 #ifdef CONFIG_TMPFS_XATTR
3768 .listxattr = shmem_listxattr,
3769 .set_acl = simple_set_acl,
3773 static const struct inode_operations shmem_dir_inode_operations = {
3775 .create = shmem_create,
3776 .lookup = simple_lookup,
3778 .unlink = shmem_unlink,
3779 .symlink = shmem_symlink,
3780 .mkdir = shmem_mkdir,
3781 .rmdir = shmem_rmdir,
3782 .mknod = shmem_mknod,
3783 .rename = shmem_rename2,
3784 .tmpfile = shmem_tmpfile,
3786 #ifdef CONFIG_TMPFS_XATTR
3787 .listxattr = shmem_listxattr,
3789 #ifdef CONFIG_TMPFS_POSIX_ACL
3790 .setattr = shmem_setattr,
3791 .set_acl = simple_set_acl,
3795 static const struct inode_operations shmem_special_inode_operations = {
3796 #ifdef CONFIG_TMPFS_XATTR
3797 .listxattr = shmem_listxattr,
3799 #ifdef CONFIG_TMPFS_POSIX_ACL
3800 .setattr = shmem_setattr,
3801 .set_acl = simple_set_acl,
3805 static const struct super_operations shmem_ops = {
3806 .alloc_inode = shmem_alloc_inode,
3807 .free_inode = shmem_free_in_core_inode,
3808 .destroy_inode = shmem_destroy_inode,
3810 .statfs = shmem_statfs,
3811 .show_options = shmem_show_options,
3813 .evict_inode = shmem_evict_inode,
3814 .drop_inode = generic_delete_inode,
3815 .put_super = shmem_put_super,
3816 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3817 .nr_cached_objects = shmem_unused_huge_count,
3818 .free_cached_objects = shmem_unused_huge_scan,
3822 static const struct vm_operations_struct shmem_vm_ops = {
3823 .fault = shmem_fault,
3824 .map_pages = filemap_map_pages,
3826 .set_policy = shmem_set_policy,
3827 .get_policy = shmem_get_policy,
3831 int shmem_init_fs_context(struct fs_context *fc)
3833 struct shmem_options *ctx;
3835 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3839 ctx->mode = 0777 | S_ISVTX;
3840 ctx->uid = current_fsuid();
3841 ctx->gid = current_fsgid();
3843 fc->fs_private = ctx;
3844 fc->ops = &shmem_fs_context_ops;
3848 static struct file_system_type shmem_fs_type = {
3849 .owner = THIS_MODULE,
3851 .init_fs_context = shmem_init_fs_context,
3853 .parameters = shmem_fs_parameters,
3855 .kill_sb = kill_litter_super,
3856 .fs_flags = FS_USERNS_MOUNT,
3859 int __init shmem_init(void)
3863 shmem_init_inodecache();
3865 error = register_filesystem(&shmem_fs_type);
3867 pr_err("Could not register tmpfs\n");
3871 shm_mnt = kern_mount(&shmem_fs_type);
3872 if (IS_ERR(shm_mnt)) {
3873 error = PTR_ERR(shm_mnt);
3874 pr_err("Could not kern_mount tmpfs\n");
3878 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3879 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3880 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3882 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
3887 unregister_filesystem(&shmem_fs_type);
3889 shmem_destroy_inodecache();
3890 shm_mnt = ERR_PTR(error);
3894 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3895 static ssize_t shmem_enabled_show(struct kobject *kobj,
3896 struct kobj_attribute *attr, char *buf)
3898 static const int values[] = {
3900 SHMEM_HUGE_WITHIN_SIZE,
3909 for (i = 0; i < ARRAY_SIZE(values); i++) {
3910 len += sysfs_emit_at(buf, len,
3911 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3913 shmem_format_huge(values[i]));
3916 len += sysfs_emit_at(buf, len, "\n");
3921 static ssize_t shmem_enabled_store(struct kobject *kobj,
3922 struct kobj_attribute *attr, const char *buf, size_t count)
3927 if (count + 1 > sizeof(tmp))
3929 memcpy(tmp, buf, count);
3931 if (count && tmp[count - 1] == '\n')
3932 tmp[count - 1] = '\0';
3934 huge = shmem_parse_huge(tmp);
3935 if (huge == -EINVAL)
3937 if (!has_transparent_hugepage() &&
3938 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3942 if (shmem_huge > SHMEM_HUGE_DENY)
3943 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3947 struct kobj_attribute shmem_enabled_attr =
3948 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3949 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3951 #else /* !CONFIG_SHMEM */
3954 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3956 * This is intended for small system where the benefits of the full
3957 * shmem code (swap-backed and resource-limited) are outweighed by
3958 * their complexity. On systems without swap this code should be
3959 * effectively equivalent, but much lighter weight.
3962 static struct file_system_type shmem_fs_type = {
3964 .init_fs_context = ramfs_init_fs_context,
3965 .parameters = ramfs_fs_parameters,
3966 .kill_sb = kill_litter_super,
3967 .fs_flags = FS_USERNS_MOUNT,
3970 int __init shmem_init(void)
3972 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3974 shm_mnt = kern_mount(&shmem_fs_type);
3975 BUG_ON(IS_ERR(shm_mnt));
3980 int shmem_unuse(unsigned int type, bool frontswap,
3981 unsigned long *fs_pages_to_unuse)
3986 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
3991 void shmem_unlock_mapping(struct address_space *mapping)
3996 unsigned long shmem_get_unmapped_area(struct file *file,
3997 unsigned long addr, unsigned long len,
3998 unsigned long pgoff, unsigned long flags)
4000 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4004 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4006 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4008 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4010 #define shmem_vm_ops generic_file_vm_ops
4011 #define shmem_file_operations ramfs_file_operations
4012 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4013 #define shmem_acct_size(flags, size) 0
4014 #define shmem_unacct_size(flags, size) do {} while (0)
4016 #endif /* CONFIG_SHMEM */
4020 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4021 unsigned long flags, unsigned int i_flags)
4023 struct inode *inode;
4027 return ERR_CAST(mnt);
4029 if (size < 0 || size > MAX_LFS_FILESIZE)
4030 return ERR_PTR(-EINVAL);
4032 if (shmem_acct_size(flags, size))
4033 return ERR_PTR(-ENOMEM);
4035 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4037 if (unlikely(!inode)) {
4038 shmem_unacct_size(flags, size);
4039 return ERR_PTR(-ENOSPC);
4041 inode->i_flags |= i_flags;
4042 inode->i_size = size;
4043 clear_nlink(inode); /* It is unlinked */
4044 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4046 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4047 &shmem_file_operations);
4054 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4055 * kernel internal. There will be NO LSM permission checks against the
4056 * underlying inode. So users of this interface must do LSM checks at a
4057 * higher layer. The users are the big_key and shm implementations. LSM
4058 * checks are provided at the key or shm level rather than the inode.
4059 * @name: name for dentry (to be seen in /proc/<pid>/maps
4060 * @size: size to be set for the file
4061 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4063 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4065 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4069 * shmem_file_setup - get an unlinked file living in tmpfs
4070 * @name: name for dentry (to be seen in /proc/<pid>/maps
4071 * @size: size to be set for the file
4072 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4074 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4076 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4078 EXPORT_SYMBOL_GPL(shmem_file_setup);
4081 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4082 * @mnt: the tmpfs mount where the file will be created
4083 * @name: name for dentry (to be seen in /proc/<pid>/maps
4084 * @size: size to be set for the file
4085 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4087 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4088 loff_t size, unsigned long flags)
4090 return __shmem_file_setup(mnt, name, size, flags, 0);
4092 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4095 * shmem_zero_setup - setup a shared anonymous mapping
4096 * @vma: the vma to be mmapped is prepared by do_mmap
4098 int shmem_zero_setup(struct vm_area_struct *vma)
4101 loff_t size = vma->vm_end - vma->vm_start;
4104 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4105 * between XFS directory reading and selinux: since this file is only
4106 * accessible to the user through its mapping, use S_PRIVATE flag to
4107 * bypass file security, in the same way as shmem_kernel_file_setup().
4109 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4111 return PTR_ERR(file);
4115 vma->vm_file = file;
4116 vma->vm_ops = &shmem_vm_ops;
4118 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4119 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4120 (vma->vm_end & HPAGE_PMD_MASK)) {
4121 khugepaged_enter(vma, vma->vm_flags);
4128 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4129 * @mapping: the page's address_space
4130 * @index: the page index
4131 * @gfp: the page allocator flags to use if allocating
4133 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4134 * with any new page allocations done using the specified allocation flags.
4135 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4136 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4137 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4139 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4140 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4142 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4143 pgoff_t index, gfp_t gfp)
4146 struct inode *inode = mapping->host;
4150 BUG_ON(!shmem_mapping(mapping));
4151 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4152 gfp, NULL, NULL, NULL);
4154 page = ERR_PTR(error);
4160 * The tiny !SHMEM case uses ramfs without swap
4162 return read_cache_page_gfp(mapping, index, gfp);
4165 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);