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;
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);
575 /* Check if there's anything to gain */
576 if (round_up(inode->i_size, PAGE_SIZE) ==
577 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
578 list_move(&info->shrinklist, &to_remove);
582 list_move(&info->shrinklist, &list);
584 sbinfo->shrinklist_len--;
588 spin_unlock(&sbinfo->shrinklist_lock);
590 list_for_each_safe(pos, next, &to_remove) {
591 info = list_entry(pos, struct shmem_inode_info, shrinklist);
592 inode = &info->vfs_inode;
593 list_del_init(&info->shrinklist);
597 list_for_each_safe(pos, next, &list) {
600 info = list_entry(pos, struct shmem_inode_info, shrinklist);
601 inode = &info->vfs_inode;
603 if (nr_to_split && split >= nr_to_split)
606 page = find_get_page(inode->i_mapping,
607 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
611 /* No huge page at the end of the file: nothing to split */
612 if (!PageTransHuge(page)) {
618 * Move the inode on the list back to shrinklist if we failed
619 * to lock the page at this time.
621 * Waiting for the lock may lead to deadlock in the
624 if (!trylock_page(page)) {
629 ret = split_huge_page(page);
633 /* If split failed move the inode on the list back to shrinklist */
639 list_del_init(&info->shrinklist);
643 * Make sure the inode is either on the global list or deleted
644 * from any local list before iput() since it could be deleted
645 * in another thread once we put the inode (then the local list
648 spin_lock(&sbinfo->shrinklist_lock);
649 list_move(&info->shrinklist, &sbinfo->shrinklist);
650 sbinfo->shrinklist_len++;
651 spin_unlock(&sbinfo->shrinklist_lock);
659 static long shmem_unused_huge_scan(struct super_block *sb,
660 struct shrink_control *sc)
662 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
664 if (!READ_ONCE(sbinfo->shrinklist_len))
667 return shmem_unused_huge_shrink(sbinfo, sc, 0);
670 static long shmem_unused_huge_count(struct super_block *sb,
671 struct shrink_control *sc)
673 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
674 return READ_ONCE(sbinfo->shrinklist_len);
676 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
678 #define shmem_huge SHMEM_HUGE_DENY
680 bool shmem_is_huge(struct vm_area_struct *vma,
681 struct inode *inode, pgoff_t index)
686 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
687 struct shrink_control *sc, unsigned long nr_to_split)
691 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
694 * Like add_to_page_cache_locked, but error if expected item has gone.
696 static int shmem_add_to_page_cache(struct page *page,
697 struct address_space *mapping,
698 pgoff_t index, void *expected, gfp_t gfp,
699 struct mm_struct *charge_mm)
701 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
702 unsigned long nr = compound_nr(page);
705 VM_BUG_ON_PAGE(PageTail(page), page);
706 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
707 VM_BUG_ON_PAGE(!PageLocked(page), page);
708 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
709 VM_BUG_ON(expected && PageTransHuge(page));
711 page_ref_add(page, nr);
712 page->mapping = mapping;
715 if (!PageSwapCache(page)) {
716 error = mem_cgroup_charge(page_folio(page), charge_mm, gfp);
718 if (PageTransHuge(page)) {
719 count_vm_event(THP_FILE_FALLBACK);
720 count_vm_event(THP_FILE_FALLBACK_CHARGE);
725 cgroup_throttle_swaprate(page, gfp);
729 if (expected != xas_find_conflict(&xas)) {
730 xas_set_err(&xas, -EEXIST);
733 if (expected && xas_find_conflict(&xas)) {
734 xas_set_err(&xas, -EEXIST);
737 xas_store(&xas, page);
740 if (PageTransHuge(page)) {
741 count_vm_event(THP_FILE_ALLOC);
742 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
744 mapping->nrpages += nr;
745 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
746 __mod_lruvec_page_state(page, NR_SHMEM, nr);
748 xas_unlock_irq(&xas);
749 } while (xas_nomem(&xas, gfp));
751 if (xas_error(&xas)) {
752 error = xas_error(&xas);
758 page->mapping = NULL;
759 page_ref_sub(page, nr);
764 * Like delete_from_page_cache, but substitutes swap for page.
766 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
768 struct address_space *mapping = page->mapping;
771 VM_BUG_ON_PAGE(PageCompound(page), page);
773 xa_lock_irq(&mapping->i_pages);
774 error = shmem_replace_entry(mapping, page->index, page, radswap);
775 page->mapping = NULL;
777 __dec_lruvec_page_state(page, NR_FILE_PAGES);
778 __dec_lruvec_page_state(page, NR_SHMEM);
779 xa_unlock_irq(&mapping->i_pages);
785 * Remove swap entry from page cache, free the swap and its page cache.
787 static int shmem_free_swap(struct address_space *mapping,
788 pgoff_t index, void *radswap)
792 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
795 free_swap_and_cache(radix_to_swp_entry(radswap));
800 * Determine (in bytes) how many of the shmem object's pages mapped by the
801 * given offsets are swapped out.
803 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
804 * as long as the inode doesn't go away and racy results are not a problem.
806 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
807 pgoff_t start, pgoff_t end)
809 XA_STATE(xas, &mapping->i_pages, start);
811 unsigned long swapped = 0;
814 xas_for_each(&xas, page, end - 1) {
815 if (xas_retry(&xas, page))
817 if (xa_is_value(page))
820 if (need_resched()) {
828 return swapped << PAGE_SHIFT;
832 * Determine (in bytes) how many of the shmem object's pages mapped by the
833 * given vma is swapped out.
835 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
836 * as long as the inode doesn't go away and racy results are not a problem.
838 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
840 struct inode *inode = file_inode(vma->vm_file);
841 struct shmem_inode_info *info = SHMEM_I(inode);
842 struct address_space *mapping = inode->i_mapping;
843 unsigned long swapped;
845 /* Be careful as we don't hold info->lock */
846 swapped = READ_ONCE(info->swapped);
849 * The easier cases are when the shmem object has nothing in swap, or
850 * the vma maps it whole. Then we can simply use the stats that we
856 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
857 return swapped << PAGE_SHIFT;
859 /* Here comes the more involved part */
860 return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
861 vma->vm_pgoff + vma_pages(vma));
865 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
867 void shmem_unlock_mapping(struct address_space *mapping)
874 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
876 while (!mapping_unevictable(mapping)) {
877 if (!pagevec_lookup(&pvec, mapping, &index))
879 check_move_unevictable_pages(&pvec);
880 pagevec_release(&pvec);
885 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
891 * At first avoid shmem_getpage(,,,SGP_READ): that fails
892 * beyond i_size, and reports fallocated pages as holes.
894 folio = __filemap_get_folio(inode->i_mapping, index,
895 FGP_ENTRY | FGP_LOCK, 0);
896 if (!xa_is_value(folio))
899 * But read a page back from swap if any of it is within i_size
900 * (although in some cases this is just a waste of time).
903 shmem_getpage(inode, index, &page, SGP_READ);
904 return page ? page_folio(page) : NULL;
908 * Remove range of pages and swap entries from page cache, and free them.
909 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
911 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
914 struct address_space *mapping = inode->i_mapping;
915 struct shmem_inode_info *info = SHMEM_I(inode);
916 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
917 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
918 struct folio_batch fbatch;
919 pgoff_t indices[PAGEVEC_SIZE];
922 long nr_swaps_freed = 0;
927 end = -1; /* unsigned, so actually very big */
929 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
930 info->fallocend = start;
932 folio_batch_init(&fbatch);
934 while (index < end && find_lock_entries(mapping, index, end - 1,
936 for (i = 0; i < folio_batch_count(&fbatch); i++) {
937 folio = fbatch.folios[i];
941 if (xa_is_value(folio)) {
944 nr_swaps_freed += !shmem_free_swap(mapping,
948 index += folio_nr_pages(folio) - 1;
950 if (!unfalloc || !folio_test_uptodate(folio))
951 truncate_inode_folio(mapping, folio);
954 folio_batch_remove_exceptionals(&fbatch);
955 folio_batch_release(&fbatch);
960 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
961 folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
963 same_folio = lend < folio_pos(folio) + folio_size(folio);
964 folio_mark_dirty(folio);
965 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
966 start = folio->index + folio_nr_pages(folio);
976 folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
978 folio_mark_dirty(folio);
979 if (!truncate_inode_partial_folio(folio, lstart, lend))
986 while (index < end) {
989 if (!find_get_entries(mapping, index, end - 1, &fbatch,
991 /* If all gone or hole-punch or unfalloc, we're done */
992 if (index == start || end != -1)
994 /* But if truncating, restart to make sure all gone */
998 for (i = 0; i < folio_batch_count(&fbatch); i++) {
999 folio = fbatch.folios[i];
1002 if (xa_is_value(folio)) {
1005 if (shmem_free_swap(mapping, index, folio)) {
1006 /* Swap was replaced by page: retry */
1016 if (!unfalloc || !folio_test_uptodate(folio)) {
1017 if (folio_mapping(folio) != mapping) {
1018 /* Page was replaced by swap: retry */
1019 folio_unlock(folio);
1023 VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1025 truncate_inode_folio(mapping, folio);
1027 index = folio->index + folio_nr_pages(folio) - 1;
1028 folio_unlock(folio);
1030 folio_batch_remove_exceptionals(&fbatch);
1031 folio_batch_release(&fbatch);
1035 spin_lock_irq(&info->lock);
1036 info->swapped -= nr_swaps_freed;
1037 shmem_recalc_inode(inode);
1038 spin_unlock_irq(&info->lock);
1041 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1043 shmem_undo_range(inode, lstart, lend, false);
1044 inode->i_ctime = inode->i_mtime = current_time(inode);
1046 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1048 static int shmem_getattr(struct user_namespace *mnt_userns,
1049 const struct path *path, struct kstat *stat,
1050 u32 request_mask, unsigned int query_flags)
1052 struct inode *inode = path->dentry->d_inode;
1053 struct shmem_inode_info *info = SHMEM_I(inode);
1055 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1056 spin_lock_irq(&info->lock);
1057 shmem_recalc_inode(inode);
1058 spin_unlock_irq(&info->lock);
1060 generic_fillattr(&init_user_ns, inode, stat);
1062 if (shmem_is_huge(NULL, inode, 0))
1063 stat->blksize = HPAGE_PMD_SIZE;
1068 static int shmem_setattr(struct user_namespace *mnt_userns,
1069 struct dentry *dentry, struct iattr *attr)
1071 struct inode *inode = d_inode(dentry);
1072 struct shmem_inode_info *info = SHMEM_I(inode);
1075 error = setattr_prepare(&init_user_ns, dentry, attr);
1079 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1080 loff_t oldsize = inode->i_size;
1081 loff_t newsize = attr->ia_size;
1083 /* protected by i_rwsem */
1084 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1085 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1088 if (newsize != oldsize) {
1089 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1093 i_size_write(inode, newsize);
1094 inode->i_ctime = inode->i_mtime = current_time(inode);
1096 if (newsize <= oldsize) {
1097 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1098 if (oldsize > holebegin)
1099 unmap_mapping_range(inode->i_mapping,
1102 shmem_truncate_range(inode,
1103 newsize, (loff_t)-1);
1104 /* unmap again to remove racily COWed private pages */
1105 if (oldsize > holebegin)
1106 unmap_mapping_range(inode->i_mapping,
1111 setattr_copy(&init_user_ns, inode, attr);
1112 if (attr->ia_valid & ATTR_MODE)
1113 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1117 static void shmem_evict_inode(struct inode *inode)
1119 struct shmem_inode_info *info = SHMEM_I(inode);
1120 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1122 if (shmem_mapping(inode->i_mapping)) {
1123 shmem_unacct_size(info->flags, inode->i_size);
1125 shmem_truncate_range(inode, 0, (loff_t)-1);
1126 if (!list_empty(&info->shrinklist)) {
1127 spin_lock(&sbinfo->shrinklist_lock);
1128 if (!list_empty(&info->shrinklist)) {
1129 list_del_init(&info->shrinklist);
1130 sbinfo->shrinklist_len--;
1132 spin_unlock(&sbinfo->shrinklist_lock);
1134 while (!list_empty(&info->swaplist)) {
1135 /* Wait while shmem_unuse() is scanning this inode... */
1136 wait_var_event(&info->stop_eviction,
1137 !atomic_read(&info->stop_eviction));
1138 mutex_lock(&shmem_swaplist_mutex);
1139 /* ...but beware of the race if we peeked too early */
1140 if (!atomic_read(&info->stop_eviction))
1141 list_del_init(&info->swaplist);
1142 mutex_unlock(&shmem_swaplist_mutex);
1146 simple_xattrs_free(&info->xattrs);
1147 WARN_ON(inode->i_blocks);
1148 shmem_free_inode(inode->i_sb);
1152 static int shmem_find_swap_entries(struct address_space *mapping,
1153 pgoff_t start, unsigned int nr_entries,
1154 struct page **entries, pgoff_t *indices,
1155 unsigned int type, bool frontswap)
1157 XA_STATE(xas, &mapping->i_pages, start);
1160 unsigned int ret = 0;
1166 xas_for_each(&xas, page, ULONG_MAX) {
1167 if (xas_retry(&xas, page))
1170 if (!xa_is_value(page))
1173 entry = radix_to_swp_entry(page);
1174 if (swp_type(entry) != type)
1177 !frontswap_test(swap_info[type], swp_offset(entry)))
1180 indices[ret] = xas.xa_index;
1181 entries[ret] = page;
1183 if (need_resched()) {
1187 if (++ret == nr_entries)
1196 * Move the swapped pages for an inode to page cache. Returns the count
1197 * of pages swapped in, or the error in case of failure.
1199 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1205 struct address_space *mapping = inode->i_mapping;
1207 for (i = 0; i < pvec.nr; i++) {
1208 struct page *page = pvec.pages[i];
1210 if (!xa_is_value(page))
1212 error = shmem_swapin_page(inode, indices[i],
1214 mapping_gfp_mask(mapping),
1221 if (error == -ENOMEM)
1225 return error ? error : ret;
1229 * If swap found in inode, free it and move page from swapcache to filecache.
1231 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1232 bool frontswap, unsigned long *fs_pages_to_unuse)
1234 struct address_space *mapping = inode->i_mapping;
1236 struct pagevec pvec;
1237 pgoff_t indices[PAGEVEC_SIZE];
1238 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1241 pagevec_init(&pvec);
1243 unsigned int nr_entries = PAGEVEC_SIZE;
1245 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1246 nr_entries = *fs_pages_to_unuse;
1248 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1249 pvec.pages, indices,
1256 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1260 if (frontswap_partial) {
1261 *fs_pages_to_unuse -= ret;
1262 if (*fs_pages_to_unuse == 0) {
1263 ret = FRONTSWAP_PAGES_UNUSED;
1268 start = indices[pvec.nr - 1];
1275 * Read all the shared memory data that resides in the swap
1276 * device 'type' back into memory, so the swap device can be
1279 int shmem_unuse(unsigned int type, bool frontswap,
1280 unsigned long *fs_pages_to_unuse)
1282 struct shmem_inode_info *info, *next;
1285 if (list_empty(&shmem_swaplist))
1288 mutex_lock(&shmem_swaplist_mutex);
1289 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1290 if (!info->swapped) {
1291 list_del_init(&info->swaplist);
1295 * Drop the swaplist mutex while searching the inode for swap;
1296 * but before doing so, make sure shmem_evict_inode() will not
1297 * remove placeholder inode from swaplist, nor let it be freed
1298 * (igrab() would protect from unlink, but not from unmount).
1300 atomic_inc(&info->stop_eviction);
1301 mutex_unlock(&shmem_swaplist_mutex);
1303 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1307 mutex_lock(&shmem_swaplist_mutex);
1308 next = list_next_entry(info, swaplist);
1310 list_del_init(&info->swaplist);
1311 if (atomic_dec_and_test(&info->stop_eviction))
1312 wake_up_var(&info->stop_eviction);
1316 mutex_unlock(&shmem_swaplist_mutex);
1322 * Move the page from the page cache to the swap cache.
1324 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1326 struct shmem_inode_info *info;
1327 struct address_space *mapping;
1328 struct inode *inode;
1333 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1334 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1335 * and its shmem_writeback() needs them to be split when swapping.
1337 if (PageTransCompound(page)) {
1338 /* Ensure the subpages are still dirty */
1340 if (split_huge_page(page) < 0)
1342 ClearPageDirty(page);
1345 BUG_ON(!PageLocked(page));
1346 mapping = page->mapping;
1347 index = page->index;
1348 inode = mapping->host;
1349 info = SHMEM_I(inode);
1350 if (info->flags & VM_LOCKED)
1352 if (!total_swap_pages)
1356 * Our capabilities prevent regular writeback or sync from ever calling
1357 * shmem_writepage; but a stacking filesystem might use ->writepage of
1358 * its underlying filesystem, in which case tmpfs should write out to
1359 * swap only in response to memory pressure, and not for the writeback
1362 if (!wbc->for_reclaim) {
1363 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1368 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1369 * value into swapfile.c, the only way we can correctly account for a
1370 * fallocated page arriving here is now to initialize it and write it.
1372 * That's okay for a page already fallocated earlier, but if we have
1373 * not yet completed the fallocation, then (a) we want to keep track
1374 * of this page in case we have to undo it, and (b) it may not be a
1375 * good idea to continue anyway, once we're pushing into swap. So
1376 * reactivate the page, and let shmem_fallocate() quit when too many.
1378 if (!PageUptodate(page)) {
1379 if (inode->i_private) {
1380 struct shmem_falloc *shmem_falloc;
1381 spin_lock(&inode->i_lock);
1382 shmem_falloc = inode->i_private;
1384 !shmem_falloc->waitq &&
1385 index >= shmem_falloc->start &&
1386 index < shmem_falloc->next)
1387 shmem_falloc->nr_unswapped++;
1389 shmem_falloc = NULL;
1390 spin_unlock(&inode->i_lock);
1394 clear_highpage(page);
1395 flush_dcache_page(page);
1396 SetPageUptodate(page);
1399 swap = get_swap_page(page);
1404 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1405 * if it's not already there. Do it now before the page is
1406 * moved to swap cache, when its pagelock no longer protects
1407 * the inode from eviction. But don't unlock the mutex until
1408 * we've incremented swapped, because shmem_unuse_inode() will
1409 * prune a !swapped inode from the swaplist under this mutex.
1411 mutex_lock(&shmem_swaplist_mutex);
1412 if (list_empty(&info->swaplist))
1413 list_add(&info->swaplist, &shmem_swaplist);
1415 if (add_to_swap_cache(page, swap,
1416 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1418 spin_lock_irq(&info->lock);
1419 shmem_recalc_inode(inode);
1421 spin_unlock_irq(&info->lock);
1423 swap_shmem_alloc(swap);
1424 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1426 mutex_unlock(&shmem_swaplist_mutex);
1427 BUG_ON(page_mapped(page));
1428 swap_writepage(page, wbc);
1432 mutex_unlock(&shmem_swaplist_mutex);
1433 put_swap_page(page, swap);
1435 set_page_dirty(page);
1436 if (wbc->for_reclaim)
1437 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1442 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1443 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1447 if (!mpol || mpol->mode == MPOL_DEFAULT)
1448 return; /* show nothing */
1450 mpol_to_str(buffer, sizeof(buffer), mpol);
1452 seq_printf(seq, ",mpol=%s", buffer);
1455 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1457 struct mempolicy *mpol = NULL;
1459 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1460 mpol = sbinfo->mpol;
1462 raw_spin_unlock(&sbinfo->stat_lock);
1466 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1467 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1470 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1474 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1476 #define vm_policy vm_private_data
1479 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1480 struct shmem_inode_info *info, pgoff_t index)
1482 /* Create a pseudo vma that just contains the policy */
1483 vma_init(vma, NULL);
1484 /* Bias interleave by inode number to distribute better across nodes */
1485 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1486 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1489 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1491 /* Drop reference taken by mpol_shared_policy_lookup() */
1492 mpol_cond_put(vma->vm_policy);
1495 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1496 struct shmem_inode_info *info, pgoff_t index)
1498 struct vm_area_struct pvma;
1500 struct vm_fault vmf = {
1504 shmem_pseudo_vma_init(&pvma, info, index);
1505 page = swap_cluster_readahead(swap, gfp, &vmf);
1506 shmem_pseudo_vma_destroy(&pvma);
1512 * Make sure huge_gfp is always more limited than limit_gfp.
1513 * Some of the flags set permissions, while others set limitations.
1515 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1517 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1518 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1519 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1520 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1522 /* Allow allocations only from the originally specified zones. */
1523 result |= zoneflags;
1526 * Minimize the result gfp by taking the union with the deny flags,
1527 * and the intersection of the allow flags.
1529 result |= (limit_gfp & denyflags);
1530 result |= (huge_gfp & limit_gfp) & allowflags;
1535 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1536 struct shmem_inode_info *info, pgoff_t index)
1538 struct vm_area_struct pvma;
1539 struct address_space *mapping = info->vfs_inode.i_mapping;
1543 hindex = round_down(index, HPAGE_PMD_NR);
1544 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1548 shmem_pseudo_vma_init(&pvma, info, hindex);
1549 page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1550 shmem_pseudo_vma_destroy(&pvma);
1552 prep_transhuge_page(page);
1554 count_vm_event(THP_FILE_FALLBACK);
1558 static struct page *shmem_alloc_page(gfp_t gfp,
1559 struct shmem_inode_info *info, pgoff_t index)
1561 struct vm_area_struct pvma;
1564 shmem_pseudo_vma_init(&pvma, info, index);
1565 page = alloc_page_vma(gfp, &pvma, 0);
1566 shmem_pseudo_vma_destroy(&pvma);
1571 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1572 struct inode *inode,
1573 pgoff_t index, bool huge)
1575 struct shmem_inode_info *info = SHMEM_I(inode);
1580 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1582 nr = huge ? HPAGE_PMD_NR : 1;
1584 if (!shmem_inode_acct_block(inode, nr))
1588 page = shmem_alloc_hugepage(gfp, info, index);
1590 page = shmem_alloc_page(gfp, info, index);
1592 __SetPageLocked(page);
1593 __SetPageSwapBacked(page);
1598 shmem_inode_unacct_blocks(inode, nr);
1600 return ERR_PTR(err);
1604 * When a page is moved from swapcache to shmem filecache (either by the
1605 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1606 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1607 * ignorance of the mapping it belongs to. If that mapping has special
1608 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1609 * we may need to copy to a suitable page before moving to filecache.
1611 * In a future release, this may well be extended to respect cpuset and
1612 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1613 * but for now it is a simple matter of zone.
1615 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1617 return page_zonenum(page) > gfp_zone(gfp);
1620 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1621 struct shmem_inode_info *info, pgoff_t index)
1623 struct page *oldpage, *newpage;
1624 struct folio *old, *new;
1625 struct address_space *swap_mapping;
1631 entry.val = page_private(oldpage);
1632 swap_index = swp_offset(entry);
1633 swap_mapping = page_mapping(oldpage);
1636 * We have arrived here because our zones are constrained, so don't
1637 * limit chance of success by further cpuset and node constraints.
1639 gfp &= ~GFP_CONSTRAINT_MASK;
1640 newpage = shmem_alloc_page(gfp, info, index);
1645 copy_highpage(newpage, oldpage);
1646 flush_dcache_page(newpage);
1648 __SetPageLocked(newpage);
1649 __SetPageSwapBacked(newpage);
1650 SetPageUptodate(newpage);
1651 set_page_private(newpage, entry.val);
1652 SetPageSwapCache(newpage);
1655 * Our caller will very soon move newpage out of swapcache, but it's
1656 * a nice clean interface for us to replace oldpage by newpage there.
1658 xa_lock_irq(&swap_mapping->i_pages);
1659 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1661 old = page_folio(oldpage);
1662 new = page_folio(newpage);
1663 mem_cgroup_migrate(old, new);
1664 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1665 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1667 xa_unlock_irq(&swap_mapping->i_pages);
1669 if (unlikely(error)) {
1671 * Is this possible? I think not, now that our callers check
1672 * both PageSwapCache and page_private after getting page lock;
1673 * but be defensive. Reverse old to newpage for clear and free.
1677 lru_cache_add(newpage);
1681 ClearPageSwapCache(oldpage);
1682 set_page_private(oldpage, 0);
1684 unlock_page(oldpage);
1691 * Swap in the page pointed to by *pagep.
1692 * Caller has to make sure that *pagep contains a valid swapped page.
1693 * Returns 0 and the page in pagep if success. On failure, returns the
1694 * error code and NULL in *pagep.
1696 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1697 struct page **pagep, enum sgp_type sgp,
1698 gfp_t gfp, struct vm_area_struct *vma,
1699 vm_fault_t *fault_type)
1701 struct address_space *mapping = inode->i_mapping;
1702 struct shmem_inode_info *info = SHMEM_I(inode);
1703 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1708 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1709 swap = radix_to_swp_entry(*pagep);
1712 /* Look it up and read it in.. */
1713 page = lookup_swap_cache(swap, NULL, 0);
1715 /* Or update major stats only when swapin succeeds?? */
1717 *fault_type |= VM_FAULT_MAJOR;
1718 count_vm_event(PGMAJFAULT);
1719 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1721 /* Here we actually start the io */
1722 page = shmem_swapin(swap, gfp, info, index);
1729 /* We have to do this with page locked to prevent races */
1731 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1732 !shmem_confirm_swap(mapping, index, swap)) {
1736 if (!PageUptodate(page)) {
1740 wait_on_page_writeback(page);
1743 * Some architectures may have to restore extra metadata to the
1744 * physical page after reading from swap.
1746 arch_swap_restore(swap, page);
1748 if (shmem_should_replace_page(page, gfp)) {
1749 error = shmem_replace_page(&page, gfp, info, index);
1754 error = shmem_add_to_page_cache(page, mapping, index,
1755 swp_to_radix_entry(swap), gfp,
1760 spin_lock_irq(&info->lock);
1762 shmem_recalc_inode(inode);
1763 spin_unlock_irq(&info->lock);
1765 if (sgp == SGP_WRITE)
1766 mark_page_accessed(page);
1768 delete_from_swap_cache(page);
1769 set_page_dirty(page);
1775 if (!shmem_confirm_swap(mapping, index, swap))
1787 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1789 * If we allocate a new one we do not mark it dirty. That's up to the
1790 * vm. If we swap it in we mark it dirty since we also free the swap
1791 * entry since a page cannot live in both the swap and page cache.
1793 * vma, vmf, and fault_type are only supplied by shmem_fault:
1794 * otherwise they are NULL.
1796 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1797 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1798 struct vm_area_struct *vma, struct vm_fault *vmf,
1799 vm_fault_t *fault_type)
1801 struct address_space *mapping = inode->i_mapping;
1802 struct shmem_inode_info *info = SHMEM_I(inode);
1803 struct shmem_sb_info *sbinfo;
1804 struct mm_struct *charge_mm;
1806 pgoff_t hindex = index;
1812 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1815 if (sgp <= SGP_CACHE &&
1816 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1820 sbinfo = SHMEM_SB(inode->i_sb);
1821 charge_mm = vma ? vma->vm_mm : NULL;
1823 page = pagecache_get_page(mapping, index,
1824 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1826 if (page && vma && userfaultfd_minor(vma)) {
1827 if (!xa_is_value(page)) {
1831 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1835 if (xa_is_value(page)) {
1836 error = shmem_swapin_page(inode, index, &page,
1837 sgp, gfp, vma, fault_type);
1838 if (error == -EEXIST)
1846 hindex = page->index;
1847 if (sgp == SGP_WRITE)
1848 mark_page_accessed(page);
1849 if (PageUptodate(page))
1851 /* fallocated page */
1852 if (sgp != SGP_READ)
1859 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1860 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1863 if (sgp == SGP_READ)
1865 if (sgp == SGP_NOALLOC)
1869 * Fast cache lookup and swap lookup did not find it: allocate.
1872 if (vma && userfaultfd_missing(vma)) {
1873 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1877 /* Never use a huge page for shmem_symlink() */
1878 if (S_ISLNK(inode->i_mode))
1880 if (!shmem_is_huge(vma, inode, index))
1883 huge_gfp = vma_thp_gfp_mask(vma);
1884 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1885 page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true);
1888 page = shmem_alloc_and_acct_page(gfp, inode,
1894 error = PTR_ERR(page);
1896 if (error != -ENOSPC)
1899 * Try to reclaim some space by splitting a huge page
1900 * beyond i_size on the filesystem.
1905 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1906 if (ret == SHRINK_STOP)
1914 if (PageTransHuge(page))
1915 hindex = round_down(index, HPAGE_PMD_NR);
1919 if (sgp == SGP_WRITE)
1920 __SetPageReferenced(page);
1922 error = shmem_add_to_page_cache(page, mapping, hindex,
1923 NULL, gfp & GFP_RECLAIM_MASK,
1927 lru_cache_add(page);
1929 spin_lock_irq(&info->lock);
1930 info->alloced += compound_nr(page);
1931 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1932 shmem_recalc_inode(inode);
1933 spin_unlock_irq(&info->lock);
1936 if (PageTransHuge(page) &&
1937 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1938 hindex + HPAGE_PMD_NR - 1) {
1940 * Part of the huge page is beyond i_size: subject
1941 * to shrink under memory pressure.
1943 spin_lock(&sbinfo->shrinklist_lock);
1945 * _careful to defend against unlocked access to
1946 * ->shrink_list in shmem_unused_huge_shrink()
1948 if (list_empty_careful(&info->shrinklist)) {
1949 list_add_tail(&info->shrinklist,
1950 &sbinfo->shrinklist);
1951 sbinfo->shrinklist_len++;
1953 spin_unlock(&sbinfo->shrinklist_lock);
1957 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1959 if (sgp == SGP_FALLOC)
1963 * Let SGP_WRITE caller clear ends if write does not fill page;
1964 * but SGP_FALLOC on a page fallocated earlier must initialize
1965 * it now, lest undo on failure cancel our earlier guarantee.
1967 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1970 for (i = 0; i < compound_nr(page); i++) {
1971 clear_highpage(page + i);
1972 flush_dcache_page(page + i);
1974 SetPageUptodate(page);
1977 /* Perhaps the file has been truncated since we checked */
1978 if (sgp <= SGP_CACHE &&
1979 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1981 ClearPageDirty(page);
1982 delete_from_page_cache(page);
1983 spin_lock_irq(&info->lock);
1984 shmem_recalc_inode(inode);
1985 spin_unlock_irq(&info->lock);
1991 *pagep = page + index - hindex;
1998 shmem_inode_unacct_blocks(inode, compound_nr(page));
2000 if (PageTransHuge(page)) {
2010 if (error == -ENOSPC && !once++) {
2011 spin_lock_irq(&info->lock);
2012 shmem_recalc_inode(inode);
2013 spin_unlock_irq(&info->lock);
2016 if (error == -EEXIST)
2022 * This is like autoremove_wake_function, but it removes the wait queue
2023 * entry unconditionally - even if something else had already woken the
2026 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2028 int ret = default_wake_function(wait, mode, sync, key);
2029 list_del_init(&wait->entry);
2033 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2035 struct vm_area_struct *vma = vmf->vma;
2036 struct inode *inode = file_inode(vma->vm_file);
2037 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2039 vm_fault_t ret = VM_FAULT_LOCKED;
2042 * Trinity finds that probing a hole which tmpfs is punching can
2043 * prevent the hole-punch from ever completing: which in turn
2044 * locks writers out with its hold on i_rwsem. So refrain from
2045 * faulting pages into the hole while it's being punched. Although
2046 * shmem_undo_range() does remove the additions, it may be unable to
2047 * keep up, as each new page needs its own unmap_mapping_range() call,
2048 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2050 * It does not matter if we sometimes reach this check just before the
2051 * hole-punch begins, so that one fault then races with the punch:
2052 * we just need to make racing faults a rare case.
2054 * The implementation below would be much simpler if we just used a
2055 * standard mutex or completion: but we cannot take i_rwsem in fault,
2056 * and bloating every shmem inode for this unlikely case would be sad.
2058 if (unlikely(inode->i_private)) {
2059 struct shmem_falloc *shmem_falloc;
2061 spin_lock(&inode->i_lock);
2062 shmem_falloc = inode->i_private;
2064 shmem_falloc->waitq &&
2065 vmf->pgoff >= shmem_falloc->start &&
2066 vmf->pgoff < shmem_falloc->next) {
2068 wait_queue_head_t *shmem_falloc_waitq;
2069 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2071 ret = VM_FAULT_NOPAGE;
2072 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2074 ret = VM_FAULT_RETRY;
2076 shmem_falloc_waitq = shmem_falloc->waitq;
2077 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2078 TASK_UNINTERRUPTIBLE);
2079 spin_unlock(&inode->i_lock);
2083 * shmem_falloc_waitq points into the shmem_fallocate()
2084 * stack of the hole-punching task: shmem_falloc_waitq
2085 * is usually invalid by the time we reach here, but
2086 * finish_wait() does not dereference it in that case;
2087 * though i_lock needed lest racing with wake_up_all().
2089 spin_lock(&inode->i_lock);
2090 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2091 spin_unlock(&inode->i_lock);
2097 spin_unlock(&inode->i_lock);
2100 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE,
2101 gfp, vma, vmf, &ret);
2103 return vmf_error(err);
2107 unsigned long shmem_get_unmapped_area(struct file *file,
2108 unsigned long uaddr, unsigned long len,
2109 unsigned long pgoff, unsigned long flags)
2111 unsigned long (*get_area)(struct file *,
2112 unsigned long, unsigned long, unsigned long, unsigned long);
2114 unsigned long offset;
2115 unsigned long inflated_len;
2116 unsigned long inflated_addr;
2117 unsigned long inflated_offset;
2119 if (len > TASK_SIZE)
2122 get_area = current->mm->get_unmapped_area;
2123 addr = get_area(file, uaddr, len, pgoff, flags);
2125 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2127 if (IS_ERR_VALUE(addr))
2129 if (addr & ~PAGE_MASK)
2131 if (addr > TASK_SIZE - len)
2134 if (shmem_huge == SHMEM_HUGE_DENY)
2136 if (len < HPAGE_PMD_SIZE)
2138 if (flags & MAP_FIXED)
2141 * Our priority is to support MAP_SHARED mapped hugely;
2142 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2143 * But if caller specified an address hint and we allocated area there
2144 * successfully, respect that as before.
2149 if (shmem_huge != SHMEM_HUGE_FORCE) {
2150 struct super_block *sb;
2153 VM_BUG_ON(file->f_op != &shmem_file_operations);
2154 sb = file_inode(file)->i_sb;
2157 * Called directly from mm/mmap.c, or drivers/char/mem.c
2158 * for "/dev/zero", to create a shared anonymous object.
2160 if (IS_ERR(shm_mnt))
2162 sb = shm_mnt->mnt_sb;
2164 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2168 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2169 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2171 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2174 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2175 if (inflated_len > TASK_SIZE)
2177 if (inflated_len < len)
2180 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2181 if (IS_ERR_VALUE(inflated_addr))
2183 if (inflated_addr & ~PAGE_MASK)
2186 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2187 inflated_addr += offset - inflated_offset;
2188 if (inflated_offset > offset)
2189 inflated_addr += HPAGE_PMD_SIZE;
2191 if (inflated_addr > TASK_SIZE - len)
2193 return inflated_addr;
2197 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2199 struct inode *inode = file_inode(vma->vm_file);
2200 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2203 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2206 struct inode *inode = file_inode(vma->vm_file);
2209 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2210 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2214 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2216 struct inode *inode = file_inode(file);
2217 struct shmem_inode_info *info = SHMEM_I(inode);
2218 int retval = -ENOMEM;
2221 * What serializes the accesses to info->flags?
2222 * ipc_lock_object() when called from shmctl_do_lock(),
2223 * no serialization needed when called from shm_destroy().
2225 if (lock && !(info->flags & VM_LOCKED)) {
2226 if (!user_shm_lock(inode->i_size, ucounts))
2228 info->flags |= VM_LOCKED;
2229 mapping_set_unevictable(file->f_mapping);
2231 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2232 user_shm_unlock(inode->i_size, ucounts);
2233 info->flags &= ~VM_LOCKED;
2234 mapping_clear_unevictable(file->f_mapping);
2242 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2244 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2247 ret = seal_check_future_write(info->seals, vma);
2251 /* arm64 - allow memory tagging on RAM-based files */
2252 vma->vm_flags |= VM_MTE_ALLOWED;
2254 file_accessed(file);
2255 vma->vm_ops = &shmem_vm_ops;
2256 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2257 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2258 (vma->vm_end & HPAGE_PMD_MASK)) {
2259 khugepaged_enter(vma, vma->vm_flags);
2264 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2265 umode_t mode, dev_t dev, unsigned long flags)
2267 struct inode *inode;
2268 struct shmem_inode_info *info;
2269 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2272 if (shmem_reserve_inode(sb, &ino))
2275 inode = new_inode(sb);
2278 inode_init_owner(&init_user_ns, inode, dir, mode);
2279 inode->i_blocks = 0;
2280 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2281 inode->i_generation = prandom_u32();
2282 info = SHMEM_I(inode);
2283 memset(info, 0, (char *)inode - (char *)info);
2284 spin_lock_init(&info->lock);
2285 atomic_set(&info->stop_eviction, 0);
2286 info->seals = F_SEAL_SEAL;
2287 info->flags = flags & VM_NORESERVE;
2288 INIT_LIST_HEAD(&info->shrinklist);
2289 INIT_LIST_HEAD(&info->swaplist);
2290 simple_xattrs_init(&info->xattrs);
2291 cache_no_acl(inode);
2292 mapping_set_large_folios(inode->i_mapping);
2294 switch (mode & S_IFMT) {
2296 inode->i_op = &shmem_special_inode_operations;
2297 init_special_inode(inode, mode, dev);
2300 inode->i_mapping->a_ops = &shmem_aops;
2301 inode->i_op = &shmem_inode_operations;
2302 inode->i_fop = &shmem_file_operations;
2303 mpol_shared_policy_init(&info->policy,
2304 shmem_get_sbmpol(sbinfo));
2308 /* Some things misbehave if size == 0 on a directory */
2309 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2310 inode->i_op = &shmem_dir_inode_operations;
2311 inode->i_fop = &simple_dir_operations;
2315 * Must not load anything in the rbtree,
2316 * mpol_free_shared_policy will not be called.
2318 mpol_shared_policy_init(&info->policy, NULL);
2322 lockdep_annotate_inode_mutex_key(inode);
2324 shmem_free_inode(sb);
2328 #ifdef CONFIG_USERFAULTFD
2329 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2331 struct vm_area_struct *dst_vma,
2332 unsigned long dst_addr,
2333 unsigned long src_addr,
2335 struct page **pagep)
2337 struct inode *inode = file_inode(dst_vma->vm_file);
2338 struct shmem_inode_info *info = SHMEM_I(inode);
2339 struct address_space *mapping = inode->i_mapping;
2340 gfp_t gfp = mapping_gfp_mask(mapping);
2341 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2347 if (!shmem_inode_acct_block(inode, 1)) {
2349 * We may have got a page, returned -ENOENT triggering a retry,
2350 * and now we find ourselves with -ENOMEM. Release the page, to
2351 * avoid a BUG_ON in our caller.
2353 if (unlikely(*pagep)) {
2362 page = shmem_alloc_page(gfp, info, pgoff);
2364 goto out_unacct_blocks;
2366 if (!zeropage) { /* COPY */
2367 page_kaddr = kmap_atomic(page);
2368 ret = copy_from_user(page_kaddr,
2369 (const void __user *)src_addr,
2371 kunmap_atomic(page_kaddr);
2373 /* fallback to copy_from_user outside mmap_lock */
2374 if (unlikely(ret)) {
2377 /* don't free the page */
2378 goto out_unacct_blocks;
2380 } else { /* ZEROPAGE */
2381 clear_highpage(page);
2388 VM_BUG_ON(PageLocked(page));
2389 VM_BUG_ON(PageSwapBacked(page));
2390 __SetPageLocked(page);
2391 __SetPageSwapBacked(page);
2392 __SetPageUptodate(page);
2395 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2396 if (unlikely(pgoff >= max_off))
2399 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2400 gfp & GFP_RECLAIM_MASK, dst_mm);
2404 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2407 goto out_delete_from_cache;
2409 spin_lock_irq(&info->lock);
2411 inode->i_blocks += BLOCKS_PER_PAGE;
2412 shmem_recalc_inode(inode);
2413 spin_unlock_irq(&info->lock);
2417 out_delete_from_cache:
2418 delete_from_page_cache(page);
2423 shmem_inode_unacct_blocks(inode, 1);
2426 #endif /* CONFIG_USERFAULTFD */
2429 static const struct inode_operations shmem_symlink_inode_operations;
2430 static const struct inode_operations shmem_short_symlink_operations;
2432 #ifdef CONFIG_TMPFS_XATTR
2433 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2435 #define shmem_initxattrs NULL
2439 shmem_write_begin(struct file *file, struct address_space *mapping,
2440 loff_t pos, unsigned len, unsigned flags,
2441 struct page **pagep, void **fsdata)
2443 struct inode *inode = mapping->host;
2444 struct shmem_inode_info *info = SHMEM_I(inode);
2445 pgoff_t index = pos >> PAGE_SHIFT;
2448 /* i_rwsem is held by caller */
2449 if (unlikely(info->seals & (F_SEAL_GROW |
2450 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2451 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2453 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2457 ret = shmem_getpage(inode, index, pagep, SGP_WRITE);
2462 if (PageHWPoison(*pagep)) {
2463 unlock_page(*pagep);
2473 shmem_write_end(struct file *file, struct address_space *mapping,
2474 loff_t pos, unsigned len, unsigned copied,
2475 struct page *page, void *fsdata)
2477 struct inode *inode = mapping->host;
2479 if (pos + copied > inode->i_size)
2480 i_size_write(inode, pos + copied);
2482 if (!PageUptodate(page)) {
2483 struct page *head = compound_head(page);
2484 if (PageTransCompound(page)) {
2487 for (i = 0; i < HPAGE_PMD_NR; i++) {
2488 if (head + i == page)
2490 clear_highpage(head + i);
2491 flush_dcache_page(head + i);
2494 if (copied < PAGE_SIZE) {
2495 unsigned from = pos & (PAGE_SIZE - 1);
2496 zero_user_segments(page, 0, from,
2497 from + copied, PAGE_SIZE);
2499 SetPageUptodate(head);
2501 set_page_dirty(page);
2508 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2510 struct file *file = iocb->ki_filp;
2511 struct inode *inode = file_inode(file);
2512 struct address_space *mapping = inode->i_mapping;
2514 unsigned long offset;
2515 enum sgp_type sgp = SGP_READ;
2518 loff_t *ppos = &iocb->ki_pos;
2521 * Might this read be for a stacking filesystem? Then when reading
2522 * holes of a sparse file, we actually need to allocate those pages,
2523 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2525 if (!iter_is_iovec(to))
2528 index = *ppos >> PAGE_SHIFT;
2529 offset = *ppos & ~PAGE_MASK;
2532 struct page *page = NULL;
2534 unsigned long nr, ret;
2535 loff_t i_size = i_size_read(inode);
2537 end_index = i_size >> PAGE_SHIFT;
2538 if (index > end_index)
2540 if (index == end_index) {
2541 nr = i_size & ~PAGE_MASK;
2546 error = shmem_getpage(inode, index, &page, sgp);
2548 if (error == -EINVAL)
2553 if (sgp == SGP_CACHE)
2554 set_page_dirty(page);
2557 if (PageHWPoison(page)) {
2565 * We must evaluate after, since reads (unlike writes)
2566 * are called without i_rwsem protection against truncate
2569 i_size = i_size_read(inode);
2570 end_index = i_size >> PAGE_SHIFT;
2571 if (index == end_index) {
2572 nr = i_size & ~PAGE_MASK;
2583 * If users can be writing to this page using arbitrary
2584 * virtual addresses, take care about potential aliasing
2585 * before reading the page on the kernel side.
2587 if (mapping_writably_mapped(mapping))
2588 flush_dcache_page(page);
2590 * Mark the page accessed if we read the beginning.
2593 mark_page_accessed(page);
2595 page = ZERO_PAGE(0);
2600 * Ok, we have the page, and it's up-to-date, so
2601 * now we can copy it to user space...
2603 ret = copy_page_to_iter(page, offset, nr, to);
2606 index += offset >> PAGE_SHIFT;
2607 offset &= ~PAGE_MASK;
2610 if (!iov_iter_count(to))
2619 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2620 file_accessed(file);
2621 return retval ? retval : error;
2624 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2626 struct address_space *mapping = file->f_mapping;
2627 struct inode *inode = mapping->host;
2629 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2630 return generic_file_llseek_size(file, offset, whence,
2631 MAX_LFS_FILESIZE, i_size_read(inode));
2636 /* We're holding i_rwsem so we can access i_size directly */
2637 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2639 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2640 inode_unlock(inode);
2644 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2647 struct inode *inode = file_inode(file);
2648 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2649 struct shmem_inode_info *info = SHMEM_I(inode);
2650 struct shmem_falloc shmem_falloc;
2651 pgoff_t start, index, end, undo_fallocend;
2654 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2659 if (mode & FALLOC_FL_PUNCH_HOLE) {
2660 struct address_space *mapping = file->f_mapping;
2661 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2662 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2663 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2665 /* protected by i_rwsem */
2666 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2671 shmem_falloc.waitq = &shmem_falloc_waitq;
2672 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2673 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2674 spin_lock(&inode->i_lock);
2675 inode->i_private = &shmem_falloc;
2676 spin_unlock(&inode->i_lock);
2678 if ((u64)unmap_end > (u64)unmap_start)
2679 unmap_mapping_range(mapping, unmap_start,
2680 1 + unmap_end - unmap_start, 0);
2681 shmem_truncate_range(inode, offset, offset + len - 1);
2682 /* No need to unmap again: hole-punching leaves COWed pages */
2684 spin_lock(&inode->i_lock);
2685 inode->i_private = NULL;
2686 wake_up_all(&shmem_falloc_waitq);
2687 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2688 spin_unlock(&inode->i_lock);
2693 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2694 error = inode_newsize_ok(inode, offset + len);
2698 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2703 start = offset >> PAGE_SHIFT;
2704 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2705 /* Try to avoid a swapstorm if len is impossible to satisfy */
2706 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2711 shmem_falloc.waitq = NULL;
2712 shmem_falloc.start = start;
2713 shmem_falloc.next = start;
2714 shmem_falloc.nr_falloced = 0;
2715 shmem_falloc.nr_unswapped = 0;
2716 spin_lock(&inode->i_lock);
2717 inode->i_private = &shmem_falloc;
2718 spin_unlock(&inode->i_lock);
2721 * info->fallocend is only relevant when huge pages might be
2722 * involved: to prevent split_huge_page() freeing fallocated
2723 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2725 undo_fallocend = info->fallocend;
2726 if (info->fallocend < end)
2727 info->fallocend = end;
2729 for (index = start; index < end; ) {
2733 * Good, the fallocate(2) manpage permits EINTR: we may have
2734 * been interrupted because we are using up too much memory.
2736 if (signal_pending(current))
2738 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2741 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2743 info->fallocend = undo_fallocend;
2744 /* Remove the !PageUptodate pages we added */
2745 if (index > start) {
2746 shmem_undo_range(inode,
2747 (loff_t)start << PAGE_SHIFT,
2748 ((loff_t)index << PAGE_SHIFT) - 1, true);
2755 * Here is a more important optimization than it appears:
2756 * a second SGP_FALLOC on the same huge page will clear it,
2757 * making it PageUptodate and un-undoable if we fail later.
2759 if (PageTransCompound(page)) {
2760 index = round_up(index, HPAGE_PMD_NR);
2761 /* Beware 32-bit wraparound */
2767 * Inform shmem_writepage() how far we have reached.
2768 * No need for lock or barrier: we have the page lock.
2770 if (!PageUptodate(page))
2771 shmem_falloc.nr_falloced += index - shmem_falloc.next;
2772 shmem_falloc.next = index;
2775 * If !PageUptodate, leave it that way so that freeable pages
2776 * can be recognized if we need to rollback on error later.
2777 * But set_page_dirty so that memory pressure will swap rather
2778 * than free the pages we are allocating (and SGP_CACHE pages
2779 * might still be clean: we now need to mark those dirty too).
2781 set_page_dirty(page);
2787 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2788 i_size_write(inode, offset + len);
2789 inode->i_ctime = current_time(inode);
2791 spin_lock(&inode->i_lock);
2792 inode->i_private = NULL;
2793 spin_unlock(&inode->i_lock);
2795 inode_unlock(inode);
2799 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2801 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2803 buf->f_type = TMPFS_MAGIC;
2804 buf->f_bsize = PAGE_SIZE;
2805 buf->f_namelen = NAME_MAX;
2806 if (sbinfo->max_blocks) {
2807 buf->f_blocks = sbinfo->max_blocks;
2809 buf->f_bfree = sbinfo->max_blocks -
2810 percpu_counter_sum(&sbinfo->used_blocks);
2812 if (sbinfo->max_inodes) {
2813 buf->f_files = sbinfo->max_inodes;
2814 buf->f_ffree = sbinfo->free_inodes;
2816 /* else leave those fields 0 like simple_statfs */
2818 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2824 * File creation. Allocate an inode, and we're done..
2827 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2828 struct dentry *dentry, umode_t mode, dev_t dev)
2830 struct inode *inode;
2831 int error = -ENOSPC;
2833 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2835 error = simple_acl_create(dir, inode);
2838 error = security_inode_init_security(inode, dir,
2840 shmem_initxattrs, NULL);
2841 if (error && error != -EOPNOTSUPP)
2845 dir->i_size += BOGO_DIRENT_SIZE;
2846 dir->i_ctime = dir->i_mtime = current_time(dir);
2847 d_instantiate(dentry, inode);
2848 dget(dentry); /* Extra count - pin the dentry in core */
2857 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2858 struct dentry *dentry, umode_t mode)
2860 struct inode *inode;
2861 int error = -ENOSPC;
2863 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2865 error = security_inode_init_security(inode, dir,
2867 shmem_initxattrs, NULL);
2868 if (error && error != -EOPNOTSUPP)
2870 error = simple_acl_create(dir, inode);
2873 d_tmpfile(dentry, inode);
2881 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2882 struct dentry *dentry, umode_t mode)
2886 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2887 mode | S_IFDIR, 0)))
2893 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2894 struct dentry *dentry, umode_t mode, bool excl)
2896 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2902 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2904 struct inode *inode = d_inode(old_dentry);
2908 * No ordinary (disk based) filesystem counts links as inodes;
2909 * but each new link needs a new dentry, pinning lowmem, and
2910 * tmpfs dentries cannot be pruned until they are unlinked.
2911 * But if an O_TMPFILE file is linked into the tmpfs, the
2912 * first link must skip that, to get the accounting right.
2914 if (inode->i_nlink) {
2915 ret = shmem_reserve_inode(inode->i_sb, NULL);
2920 dir->i_size += BOGO_DIRENT_SIZE;
2921 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2923 ihold(inode); /* New dentry reference */
2924 dget(dentry); /* Extra pinning count for the created dentry */
2925 d_instantiate(dentry, inode);
2930 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2932 struct inode *inode = d_inode(dentry);
2934 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2935 shmem_free_inode(inode->i_sb);
2937 dir->i_size -= BOGO_DIRENT_SIZE;
2938 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2940 dput(dentry); /* Undo the count from "create" - this does all the work */
2944 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2946 if (!simple_empty(dentry))
2949 drop_nlink(d_inode(dentry));
2951 return shmem_unlink(dir, dentry);
2954 static int shmem_whiteout(struct user_namespace *mnt_userns,
2955 struct inode *old_dir, struct dentry *old_dentry)
2957 struct dentry *whiteout;
2960 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2964 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2965 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2971 * Cheat and hash the whiteout while the old dentry is still in
2972 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2974 * d_lookup() will consistently find one of them at this point,
2975 * not sure which one, but that isn't even important.
2982 * The VFS layer already does all the dentry stuff for rename,
2983 * we just have to decrement the usage count for the target if
2984 * it exists so that the VFS layer correctly free's it when it
2987 static int shmem_rename2(struct user_namespace *mnt_userns,
2988 struct inode *old_dir, struct dentry *old_dentry,
2989 struct inode *new_dir, struct dentry *new_dentry,
2992 struct inode *inode = d_inode(old_dentry);
2993 int they_are_dirs = S_ISDIR(inode->i_mode);
2995 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2998 if (flags & RENAME_EXCHANGE)
2999 return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
3001 if (!simple_empty(new_dentry))
3004 if (flags & RENAME_WHITEOUT) {
3007 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3012 if (d_really_is_positive(new_dentry)) {
3013 (void) shmem_unlink(new_dir, new_dentry);
3014 if (they_are_dirs) {
3015 drop_nlink(d_inode(new_dentry));
3016 drop_nlink(old_dir);
3018 } else if (they_are_dirs) {
3019 drop_nlink(old_dir);
3023 old_dir->i_size -= BOGO_DIRENT_SIZE;
3024 new_dir->i_size += BOGO_DIRENT_SIZE;
3025 old_dir->i_ctime = old_dir->i_mtime =
3026 new_dir->i_ctime = new_dir->i_mtime =
3027 inode->i_ctime = current_time(old_dir);
3031 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3032 struct dentry *dentry, const char *symname)
3036 struct inode *inode;
3039 len = strlen(symname) + 1;
3040 if (len > PAGE_SIZE)
3041 return -ENAMETOOLONG;
3043 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3048 error = security_inode_init_security(inode, dir, &dentry->d_name,
3049 shmem_initxattrs, NULL);
3050 if (error && error != -EOPNOTSUPP) {
3055 inode->i_size = len-1;
3056 if (len <= SHORT_SYMLINK_LEN) {
3057 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3058 if (!inode->i_link) {
3062 inode->i_op = &shmem_short_symlink_operations;
3064 inode_nohighmem(inode);
3065 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3070 inode->i_mapping->a_ops = &shmem_aops;
3071 inode->i_op = &shmem_symlink_inode_operations;
3072 memcpy(page_address(page), symname, len);
3073 SetPageUptodate(page);
3074 set_page_dirty(page);
3078 dir->i_size += BOGO_DIRENT_SIZE;
3079 dir->i_ctime = dir->i_mtime = current_time(dir);
3080 d_instantiate(dentry, inode);
3085 static void shmem_put_link(void *arg)
3087 mark_page_accessed(arg);
3091 static const char *shmem_get_link(struct dentry *dentry,
3092 struct inode *inode,
3093 struct delayed_call *done)
3095 struct page *page = NULL;
3098 page = find_get_page(inode->i_mapping, 0);
3100 return ERR_PTR(-ECHILD);
3101 if (PageHWPoison(page) ||
3102 !PageUptodate(page)) {
3104 return ERR_PTR(-ECHILD);
3107 error = shmem_getpage(inode, 0, &page, SGP_READ);
3109 return ERR_PTR(error);
3111 return ERR_PTR(-ECHILD);
3112 if (PageHWPoison(page)) {
3115 return ERR_PTR(-ECHILD);
3119 set_delayed_call(done, shmem_put_link, page);
3120 return page_address(page);
3123 #ifdef CONFIG_TMPFS_XATTR
3125 * Superblocks without xattr inode operations may get some security.* xattr
3126 * support from the LSM "for free". As soon as we have any other xattrs
3127 * like ACLs, we also need to implement the security.* handlers at
3128 * filesystem level, though.
3132 * Callback for security_inode_init_security() for acquiring xattrs.
3134 static int shmem_initxattrs(struct inode *inode,
3135 const struct xattr *xattr_array,
3138 struct shmem_inode_info *info = SHMEM_I(inode);
3139 const struct xattr *xattr;
3140 struct simple_xattr *new_xattr;
3143 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3144 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3148 len = strlen(xattr->name) + 1;
3149 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3151 if (!new_xattr->name) {
3156 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3157 XATTR_SECURITY_PREFIX_LEN);
3158 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3161 simple_xattr_list_add(&info->xattrs, new_xattr);
3167 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3168 struct dentry *unused, struct inode *inode,
3169 const char *name, void *buffer, size_t size)
3171 struct shmem_inode_info *info = SHMEM_I(inode);
3173 name = xattr_full_name(handler, name);
3174 return simple_xattr_get(&info->xattrs, name, buffer, size);
3177 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3178 struct user_namespace *mnt_userns,
3179 struct dentry *unused, struct inode *inode,
3180 const char *name, const void *value,
3181 size_t size, int flags)
3183 struct shmem_inode_info *info = SHMEM_I(inode);
3185 name = xattr_full_name(handler, name);
3186 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3189 static const struct xattr_handler shmem_security_xattr_handler = {
3190 .prefix = XATTR_SECURITY_PREFIX,
3191 .get = shmem_xattr_handler_get,
3192 .set = shmem_xattr_handler_set,
3195 static const struct xattr_handler shmem_trusted_xattr_handler = {
3196 .prefix = XATTR_TRUSTED_PREFIX,
3197 .get = shmem_xattr_handler_get,
3198 .set = shmem_xattr_handler_set,
3201 static const struct xattr_handler *shmem_xattr_handlers[] = {
3202 #ifdef CONFIG_TMPFS_POSIX_ACL
3203 &posix_acl_access_xattr_handler,
3204 &posix_acl_default_xattr_handler,
3206 &shmem_security_xattr_handler,
3207 &shmem_trusted_xattr_handler,
3211 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3213 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3214 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3216 #endif /* CONFIG_TMPFS_XATTR */
3218 static const struct inode_operations shmem_short_symlink_operations = {
3219 .get_link = simple_get_link,
3220 #ifdef CONFIG_TMPFS_XATTR
3221 .listxattr = shmem_listxattr,
3225 static const struct inode_operations shmem_symlink_inode_operations = {
3226 .get_link = shmem_get_link,
3227 #ifdef CONFIG_TMPFS_XATTR
3228 .listxattr = shmem_listxattr,
3232 static struct dentry *shmem_get_parent(struct dentry *child)
3234 return ERR_PTR(-ESTALE);
3237 static int shmem_match(struct inode *ino, void *vfh)
3241 inum = (inum << 32) | fh[1];
3242 return ino->i_ino == inum && fh[0] == ino->i_generation;
3245 /* Find any alias of inode, but prefer a hashed alias */
3246 static struct dentry *shmem_find_alias(struct inode *inode)
3248 struct dentry *alias = d_find_alias(inode);
3250 return alias ?: d_find_any_alias(inode);
3254 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3255 struct fid *fid, int fh_len, int fh_type)
3257 struct inode *inode;
3258 struct dentry *dentry = NULL;
3265 inum = (inum << 32) | fid->raw[1];
3267 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3268 shmem_match, fid->raw);
3270 dentry = shmem_find_alias(inode);
3277 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3278 struct inode *parent)
3282 return FILEID_INVALID;
3285 if (inode_unhashed(inode)) {
3286 /* Unfortunately insert_inode_hash is not idempotent,
3287 * so as we hash inodes here rather than at creation
3288 * time, we need a lock to ensure we only try
3291 static DEFINE_SPINLOCK(lock);
3293 if (inode_unhashed(inode))
3294 __insert_inode_hash(inode,
3295 inode->i_ino + inode->i_generation);
3299 fh[0] = inode->i_generation;
3300 fh[1] = inode->i_ino;
3301 fh[2] = ((__u64)inode->i_ino) >> 32;
3307 static const struct export_operations shmem_export_ops = {
3308 .get_parent = shmem_get_parent,
3309 .encode_fh = shmem_encode_fh,
3310 .fh_to_dentry = shmem_fh_to_dentry,
3326 static const struct constant_table shmem_param_enums_huge[] = {
3327 {"never", SHMEM_HUGE_NEVER },
3328 {"always", SHMEM_HUGE_ALWAYS },
3329 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3330 {"advise", SHMEM_HUGE_ADVISE },
3334 const struct fs_parameter_spec shmem_fs_parameters[] = {
3335 fsparam_u32 ("gid", Opt_gid),
3336 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3337 fsparam_u32oct("mode", Opt_mode),
3338 fsparam_string("mpol", Opt_mpol),
3339 fsparam_string("nr_blocks", Opt_nr_blocks),
3340 fsparam_string("nr_inodes", Opt_nr_inodes),
3341 fsparam_string("size", Opt_size),
3342 fsparam_u32 ("uid", Opt_uid),
3343 fsparam_flag ("inode32", Opt_inode32),
3344 fsparam_flag ("inode64", Opt_inode64),
3348 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3350 struct shmem_options *ctx = fc->fs_private;
3351 struct fs_parse_result result;
3352 unsigned long long size;
3356 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3362 size = memparse(param->string, &rest);
3364 size <<= PAGE_SHIFT;
3365 size *= totalram_pages();
3371 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3372 ctx->seen |= SHMEM_SEEN_BLOCKS;
3375 ctx->blocks = memparse(param->string, &rest);
3378 ctx->seen |= SHMEM_SEEN_BLOCKS;
3381 ctx->inodes = memparse(param->string, &rest);
3384 ctx->seen |= SHMEM_SEEN_INODES;
3387 ctx->mode = result.uint_32 & 07777;
3390 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3391 if (!uid_valid(ctx->uid))
3395 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3396 if (!gid_valid(ctx->gid))
3400 ctx->huge = result.uint_32;
3401 if (ctx->huge != SHMEM_HUGE_NEVER &&
3402 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3403 has_transparent_hugepage()))
3404 goto unsupported_parameter;
3405 ctx->seen |= SHMEM_SEEN_HUGE;
3408 if (IS_ENABLED(CONFIG_NUMA)) {
3409 mpol_put(ctx->mpol);
3411 if (mpol_parse_str(param->string, &ctx->mpol))
3415 goto unsupported_parameter;
3417 ctx->full_inums = false;
3418 ctx->seen |= SHMEM_SEEN_INUMS;
3421 if (sizeof(ino_t) < 8) {
3423 "Cannot use inode64 with <64bit inums in kernel\n");
3425 ctx->full_inums = true;
3426 ctx->seen |= SHMEM_SEEN_INUMS;
3431 unsupported_parameter:
3432 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3434 return invalfc(fc, "Bad value for '%s'", param->key);
3437 static int shmem_parse_options(struct fs_context *fc, void *data)
3439 char *options = data;
3442 int err = security_sb_eat_lsm_opts(options, &fc->security);
3447 while (options != NULL) {
3448 char *this_char = options;
3451 * NUL-terminate this option: unfortunately,
3452 * mount options form a comma-separated list,
3453 * but mpol's nodelist may also contain commas.
3455 options = strchr(options, ',');
3456 if (options == NULL)
3459 if (!isdigit(*options)) {
3465 char *value = strchr(this_char, '=');
3471 len = strlen(value);
3473 err = vfs_parse_fs_string(fc, this_char, value, len);
3482 * Reconfigure a shmem filesystem.
3484 * Note that we disallow change from limited->unlimited blocks/inodes while any
3485 * are in use; but we must separately disallow unlimited->limited, because in
3486 * that case we have no record of how much is already in use.
3488 static int shmem_reconfigure(struct fs_context *fc)
3490 struct shmem_options *ctx = fc->fs_private;
3491 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3492 unsigned long inodes;
3493 struct mempolicy *mpol = NULL;
3496 raw_spin_lock(&sbinfo->stat_lock);
3497 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3498 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3499 if (!sbinfo->max_blocks) {
3500 err = "Cannot retroactively limit size";
3503 if (percpu_counter_compare(&sbinfo->used_blocks,
3505 err = "Too small a size for current use";
3509 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3510 if (!sbinfo->max_inodes) {
3511 err = "Cannot retroactively limit inodes";
3514 if (ctx->inodes < inodes) {
3515 err = "Too few inodes for current use";
3520 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3521 sbinfo->next_ino > UINT_MAX) {
3522 err = "Current inum too high to switch to 32-bit inums";
3526 if (ctx->seen & SHMEM_SEEN_HUGE)
3527 sbinfo->huge = ctx->huge;
3528 if (ctx->seen & SHMEM_SEEN_INUMS)
3529 sbinfo->full_inums = ctx->full_inums;
3530 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3531 sbinfo->max_blocks = ctx->blocks;
3532 if (ctx->seen & SHMEM_SEEN_INODES) {
3533 sbinfo->max_inodes = ctx->inodes;
3534 sbinfo->free_inodes = ctx->inodes - inodes;
3538 * Preserve previous mempolicy unless mpol remount option was specified.
3541 mpol = sbinfo->mpol;
3542 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3545 raw_spin_unlock(&sbinfo->stat_lock);
3549 raw_spin_unlock(&sbinfo->stat_lock);
3550 return invalfc(fc, "%s", err);
3553 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3555 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3557 if (sbinfo->max_blocks != shmem_default_max_blocks())
3558 seq_printf(seq, ",size=%luk",
3559 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3560 if (sbinfo->max_inodes != shmem_default_max_inodes())
3561 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3562 if (sbinfo->mode != (0777 | S_ISVTX))
3563 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3564 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3565 seq_printf(seq, ",uid=%u",
3566 from_kuid_munged(&init_user_ns, sbinfo->uid));
3567 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3568 seq_printf(seq, ",gid=%u",
3569 from_kgid_munged(&init_user_ns, sbinfo->gid));
3572 * Showing inode{64,32} might be useful even if it's the system default,
3573 * since then people don't have to resort to checking both here and
3574 * /proc/config.gz to confirm 64-bit inums were successfully applied
3575 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3577 * We hide it when inode64 isn't the default and we are using 32-bit
3578 * inodes, since that probably just means the feature isn't even under
3583 * +-----------------+-----------------+
3584 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3585 * +------------------+-----------------+-----------------+
3586 * | full_inums=true | show | show |
3587 * | full_inums=false | show | hide |
3588 * +------------------+-----------------+-----------------+
3591 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3592 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3593 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3594 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3596 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3598 shmem_show_mpol(seq, sbinfo->mpol);
3602 #endif /* CONFIG_TMPFS */
3604 static void shmem_put_super(struct super_block *sb)
3606 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3608 free_percpu(sbinfo->ino_batch);
3609 percpu_counter_destroy(&sbinfo->used_blocks);
3610 mpol_put(sbinfo->mpol);
3612 sb->s_fs_info = NULL;
3615 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3617 struct shmem_options *ctx = fc->fs_private;
3618 struct inode *inode;
3619 struct shmem_sb_info *sbinfo;
3621 /* Round up to L1_CACHE_BYTES to resist false sharing */
3622 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3623 L1_CACHE_BYTES), GFP_KERNEL);
3627 sb->s_fs_info = sbinfo;
3631 * Per default we only allow half of the physical ram per
3632 * tmpfs instance, limiting inodes to one per page of lowmem;
3633 * but the internal instance is left unlimited.
3635 if (!(sb->s_flags & SB_KERNMOUNT)) {
3636 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3637 ctx->blocks = shmem_default_max_blocks();
3638 if (!(ctx->seen & SHMEM_SEEN_INODES))
3639 ctx->inodes = shmem_default_max_inodes();
3640 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3641 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3643 sb->s_flags |= SB_NOUSER;
3645 sb->s_export_op = &shmem_export_ops;
3646 sb->s_flags |= SB_NOSEC;
3648 sb->s_flags |= SB_NOUSER;
3650 sbinfo->max_blocks = ctx->blocks;
3651 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3652 if (sb->s_flags & SB_KERNMOUNT) {
3653 sbinfo->ino_batch = alloc_percpu(ino_t);
3654 if (!sbinfo->ino_batch)
3657 sbinfo->uid = ctx->uid;
3658 sbinfo->gid = ctx->gid;
3659 sbinfo->full_inums = ctx->full_inums;
3660 sbinfo->mode = ctx->mode;
3661 sbinfo->huge = ctx->huge;
3662 sbinfo->mpol = ctx->mpol;
3665 raw_spin_lock_init(&sbinfo->stat_lock);
3666 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3668 spin_lock_init(&sbinfo->shrinklist_lock);
3669 INIT_LIST_HEAD(&sbinfo->shrinklist);
3671 sb->s_maxbytes = MAX_LFS_FILESIZE;
3672 sb->s_blocksize = PAGE_SIZE;
3673 sb->s_blocksize_bits = PAGE_SHIFT;
3674 sb->s_magic = TMPFS_MAGIC;
3675 sb->s_op = &shmem_ops;
3676 sb->s_time_gran = 1;
3677 #ifdef CONFIG_TMPFS_XATTR
3678 sb->s_xattr = shmem_xattr_handlers;
3680 #ifdef CONFIG_TMPFS_POSIX_ACL
3681 sb->s_flags |= SB_POSIXACL;
3683 uuid_gen(&sb->s_uuid);
3685 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3688 inode->i_uid = sbinfo->uid;
3689 inode->i_gid = sbinfo->gid;
3690 sb->s_root = d_make_root(inode);
3696 shmem_put_super(sb);
3700 static int shmem_get_tree(struct fs_context *fc)
3702 return get_tree_nodev(fc, shmem_fill_super);
3705 static void shmem_free_fc(struct fs_context *fc)
3707 struct shmem_options *ctx = fc->fs_private;
3710 mpol_put(ctx->mpol);
3715 static const struct fs_context_operations shmem_fs_context_ops = {
3716 .free = shmem_free_fc,
3717 .get_tree = shmem_get_tree,
3719 .parse_monolithic = shmem_parse_options,
3720 .parse_param = shmem_parse_one,
3721 .reconfigure = shmem_reconfigure,
3725 static struct kmem_cache *shmem_inode_cachep;
3727 static struct inode *shmem_alloc_inode(struct super_block *sb)
3729 struct shmem_inode_info *info;
3730 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3733 return &info->vfs_inode;
3736 static void shmem_free_in_core_inode(struct inode *inode)
3738 if (S_ISLNK(inode->i_mode))
3739 kfree(inode->i_link);
3740 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3743 static void shmem_destroy_inode(struct inode *inode)
3745 if (S_ISREG(inode->i_mode))
3746 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3749 static void shmem_init_inode(void *foo)
3751 struct shmem_inode_info *info = foo;
3752 inode_init_once(&info->vfs_inode);
3755 static void shmem_init_inodecache(void)
3757 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3758 sizeof(struct shmem_inode_info),
3759 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3762 static void shmem_destroy_inodecache(void)
3764 kmem_cache_destroy(shmem_inode_cachep);
3767 /* Keep the page in page cache instead of truncating it */
3768 static int shmem_error_remove_page(struct address_space *mapping,
3774 const struct address_space_operations shmem_aops = {
3775 .writepage = shmem_writepage,
3776 .set_page_dirty = __set_page_dirty_no_writeback,
3778 .write_begin = shmem_write_begin,
3779 .write_end = shmem_write_end,
3781 #ifdef CONFIG_MIGRATION
3782 .migratepage = migrate_page,
3784 .error_remove_page = shmem_error_remove_page,
3786 EXPORT_SYMBOL(shmem_aops);
3788 static const struct file_operations shmem_file_operations = {
3790 .get_unmapped_area = shmem_get_unmapped_area,
3792 .llseek = shmem_file_llseek,
3793 .read_iter = shmem_file_read_iter,
3794 .write_iter = generic_file_write_iter,
3795 .fsync = noop_fsync,
3796 .splice_read = generic_file_splice_read,
3797 .splice_write = iter_file_splice_write,
3798 .fallocate = shmem_fallocate,
3802 static const struct inode_operations shmem_inode_operations = {
3803 .getattr = shmem_getattr,
3804 .setattr = shmem_setattr,
3805 #ifdef CONFIG_TMPFS_XATTR
3806 .listxattr = shmem_listxattr,
3807 .set_acl = simple_set_acl,
3811 static const struct inode_operations shmem_dir_inode_operations = {
3813 .create = shmem_create,
3814 .lookup = simple_lookup,
3816 .unlink = shmem_unlink,
3817 .symlink = shmem_symlink,
3818 .mkdir = shmem_mkdir,
3819 .rmdir = shmem_rmdir,
3820 .mknod = shmem_mknod,
3821 .rename = shmem_rename2,
3822 .tmpfile = shmem_tmpfile,
3824 #ifdef CONFIG_TMPFS_XATTR
3825 .listxattr = shmem_listxattr,
3827 #ifdef CONFIG_TMPFS_POSIX_ACL
3828 .setattr = shmem_setattr,
3829 .set_acl = simple_set_acl,
3833 static const struct inode_operations shmem_special_inode_operations = {
3834 #ifdef CONFIG_TMPFS_XATTR
3835 .listxattr = shmem_listxattr,
3837 #ifdef CONFIG_TMPFS_POSIX_ACL
3838 .setattr = shmem_setattr,
3839 .set_acl = simple_set_acl,
3843 static const struct super_operations shmem_ops = {
3844 .alloc_inode = shmem_alloc_inode,
3845 .free_inode = shmem_free_in_core_inode,
3846 .destroy_inode = shmem_destroy_inode,
3848 .statfs = shmem_statfs,
3849 .show_options = shmem_show_options,
3851 .evict_inode = shmem_evict_inode,
3852 .drop_inode = generic_delete_inode,
3853 .put_super = shmem_put_super,
3854 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3855 .nr_cached_objects = shmem_unused_huge_count,
3856 .free_cached_objects = shmem_unused_huge_scan,
3860 static const struct vm_operations_struct shmem_vm_ops = {
3861 .fault = shmem_fault,
3862 .map_pages = filemap_map_pages,
3864 .set_policy = shmem_set_policy,
3865 .get_policy = shmem_get_policy,
3869 int shmem_init_fs_context(struct fs_context *fc)
3871 struct shmem_options *ctx;
3873 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3877 ctx->mode = 0777 | S_ISVTX;
3878 ctx->uid = current_fsuid();
3879 ctx->gid = current_fsgid();
3881 fc->fs_private = ctx;
3882 fc->ops = &shmem_fs_context_ops;
3886 static struct file_system_type shmem_fs_type = {
3887 .owner = THIS_MODULE,
3889 .init_fs_context = shmem_init_fs_context,
3891 .parameters = shmem_fs_parameters,
3893 .kill_sb = kill_litter_super,
3894 .fs_flags = FS_USERNS_MOUNT,
3897 int __init shmem_init(void)
3901 shmem_init_inodecache();
3903 error = register_filesystem(&shmem_fs_type);
3905 pr_err("Could not register tmpfs\n");
3909 shm_mnt = kern_mount(&shmem_fs_type);
3910 if (IS_ERR(shm_mnt)) {
3911 error = PTR_ERR(shm_mnt);
3912 pr_err("Could not kern_mount tmpfs\n");
3916 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3917 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3918 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3920 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
3925 unregister_filesystem(&shmem_fs_type);
3927 shmem_destroy_inodecache();
3928 shm_mnt = ERR_PTR(error);
3932 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3933 static ssize_t shmem_enabled_show(struct kobject *kobj,
3934 struct kobj_attribute *attr, char *buf)
3936 static const int values[] = {
3938 SHMEM_HUGE_WITHIN_SIZE,
3947 for (i = 0; i < ARRAY_SIZE(values); i++) {
3948 len += sysfs_emit_at(buf, len,
3949 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3951 shmem_format_huge(values[i]));
3954 len += sysfs_emit_at(buf, len, "\n");
3959 static ssize_t shmem_enabled_store(struct kobject *kobj,
3960 struct kobj_attribute *attr, const char *buf, size_t count)
3965 if (count + 1 > sizeof(tmp))
3967 memcpy(tmp, buf, count);
3969 if (count && tmp[count - 1] == '\n')
3970 tmp[count - 1] = '\0';
3972 huge = shmem_parse_huge(tmp);
3973 if (huge == -EINVAL)
3975 if (!has_transparent_hugepage() &&
3976 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3980 if (shmem_huge > SHMEM_HUGE_DENY)
3981 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3985 struct kobj_attribute shmem_enabled_attr =
3986 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3987 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3989 #else /* !CONFIG_SHMEM */
3992 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3994 * This is intended for small system where the benefits of the full
3995 * shmem code (swap-backed and resource-limited) are outweighed by
3996 * their complexity. On systems without swap this code should be
3997 * effectively equivalent, but much lighter weight.
4000 static struct file_system_type shmem_fs_type = {
4002 .init_fs_context = ramfs_init_fs_context,
4003 .parameters = ramfs_fs_parameters,
4004 .kill_sb = kill_litter_super,
4005 .fs_flags = FS_USERNS_MOUNT,
4008 int __init shmem_init(void)
4010 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4012 shm_mnt = kern_mount(&shmem_fs_type);
4013 BUG_ON(IS_ERR(shm_mnt));
4018 int shmem_unuse(unsigned int type, bool frontswap,
4019 unsigned long *fs_pages_to_unuse)
4024 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4029 void shmem_unlock_mapping(struct address_space *mapping)
4034 unsigned long shmem_get_unmapped_area(struct file *file,
4035 unsigned long addr, unsigned long len,
4036 unsigned long pgoff, unsigned long flags)
4038 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4042 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4044 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4046 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4048 #define shmem_vm_ops generic_file_vm_ops
4049 #define shmem_file_operations ramfs_file_operations
4050 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4051 #define shmem_acct_size(flags, size) 0
4052 #define shmem_unacct_size(flags, size) do {} while (0)
4054 #endif /* CONFIG_SHMEM */
4058 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4059 unsigned long flags, unsigned int i_flags)
4061 struct inode *inode;
4065 return ERR_CAST(mnt);
4067 if (size < 0 || size > MAX_LFS_FILESIZE)
4068 return ERR_PTR(-EINVAL);
4070 if (shmem_acct_size(flags, size))
4071 return ERR_PTR(-ENOMEM);
4073 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4075 if (unlikely(!inode)) {
4076 shmem_unacct_size(flags, size);
4077 return ERR_PTR(-ENOSPC);
4079 inode->i_flags |= i_flags;
4080 inode->i_size = size;
4081 clear_nlink(inode); /* It is unlinked */
4082 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4084 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4085 &shmem_file_operations);
4092 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4093 * kernel internal. There will be NO LSM permission checks against the
4094 * underlying inode. So users of this interface must do LSM checks at a
4095 * higher layer. The users are the big_key and shm implementations. LSM
4096 * checks are provided at the key or shm level rather than the inode.
4097 * @name: name for dentry (to be seen in /proc/<pid>/maps
4098 * @size: size to be set for the file
4099 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4101 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4103 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4107 * shmem_file_setup - get an unlinked file living in tmpfs
4108 * @name: name for dentry (to be seen in /proc/<pid>/maps
4109 * @size: size to be set for the file
4110 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4112 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4114 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4116 EXPORT_SYMBOL_GPL(shmem_file_setup);
4119 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4120 * @mnt: the tmpfs mount where the file will be created
4121 * @name: name for dentry (to be seen in /proc/<pid>/maps
4122 * @size: size to be set for the file
4123 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4125 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4126 loff_t size, unsigned long flags)
4128 return __shmem_file_setup(mnt, name, size, flags, 0);
4130 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4133 * shmem_zero_setup - setup a shared anonymous mapping
4134 * @vma: the vma to be mmapped is prepared by do_mmap
4136 int shmem_zero_setup(struct vm_area_struct *vma)
4139 loff_t size = vma->vm_end - vma->vm_start;
4142 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4143 * between XFS directory reading and selinux: since this file is only
4144 * accessible to the user through its mapping, use S_PRIVATE flag to
4145 * bypass file security, in the same way as shmem_kernel_file_setup().
4147 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4149 return PTR_ERR(file);
4153 vma->vm_file = file;
4154 vma->vm_ops = &shmem_vm_ops;
4156 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4157 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4158 (vma->vm_end & HPAGE_PMD_MASK)) {
4159 khugepaged_enter(vma, vma->vm_flags);
4166 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4167 * @mapping: the page's address_space
4168 * @index: the page index
4169 * @gfp: the page allocator flags to use if allocating
4171 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4172 * with any new page allocations done using the specified allocation flags.
4173 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4174 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4175 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4177 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4178 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4180 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4181 pgoff_t index, gfp_t gfp)
4184 struct inode *inode = mapping->host;
4188 BUG_ON(!shmem_mapping(mapping));
4189 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4190 gfp, NULL, NULL, NULL);
4192 return ERR_PTR(error);
4195 if (PageHWPoison(page)) {
4197 return ERR_PTR(-EIO);
4203 * The tiny !SHMEM case uses ramfs without swap
4205 return read_cache_page_gfp(mapping, index, gfp);
4208 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);