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>
42 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
44 static struct vfsmount *shm_mnt;
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/blkdev.h>
64 #include <linux/pagevec.h>
65 #include <linux/percpu_counter.h>
66 #include <linux/falloc.h>
67 #include <linux/splice.h>
68 #include <linux/security.h>
69 #include <linux/swapops.h>
70 #include <linux/mempolicy.h>
71 #include <linux/namei.h>
72 #include <linux/ctype.h>
73 #include <linux/migrate.h>
74 #include <linux/highmem.h>
75 #include <linux/seq_file.h>
76 #include <linux/magic.h>
77 #include <linux/syscalls.h>
78 #include <linux/fcntl.h>
79 #include <uapi/linux/memfd.h>
80 #include <linux/userfaultfd_k.h>
81 #include <linux/rmap.h>
82 #include <linux/uuid.h>
84 #include <linux/uaccess.h>
88 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
91 /* Pretend that each entry is of this size in directory's i_size */
92 #define BOGO_DIRENT_SIZE 20
94 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
95 #define SHORT_SYMLINK_LEN 128
98 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
99 * inode->i_private (with i_mutex making sure that it has only one user at
100 * a time): we would prefer not to enlarge the shmem inode just for that.
102 struct shmem_falloc {
103 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
104 pgoff_t start; /* start of range currently being fallocated */
105 pgoff_t next; /* the next page offset to be fallocated */
106 pgoff_t nr_falloced; /* how many new pages have been fallocated */
107 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
110 struct shmem_options {
111 unsigned long long blocks;
112 unsigned long long inodes;
113 struct mempolicy *mpol;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
123 #define SHMEM_SEEN_INUMS 8
127 static unsigned long shmem_default_max_blocks(void)
129 return totalram_pages() / 2;
132 static unsigned long shmem_default_max_inodes(void)
134 unsigned long nr_pages = totalram_pages();
136 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
140 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
141 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
142 struct shmem_inode_info *info, pgoff_t index);
143 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 vm_fault_t *fault_type);
147 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
148 struct page **pagep, enum sgp_type sgp,
149 gfp_t gfp, struct vm_area_struct *vma,
150 struct vm_fault *vmf, vm_fault_t *fault_type);
152 int shmem_getpage(struct inode *inode, pgoff_t index,
153 struct page **pagep, enum sgp_type sgp)
155 return shmem_getpage_gfp(inode, index, pagep, sgp,
156 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
159 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
161 return sb->s_fs_info;
165 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
166 * for shared memory and for shared anonymous (/dev/zero) mappings
167 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
168 * consistent with the pre-accounting of private mappings ...
170 static inline int shmem_acct_size(unsigned long flags, loff_t size)
172 return (flags & VM_NORESERVE) ?
173 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
176 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
178 if (!(flags & VM_NORESERVE))
179 vm_unacct_memory(VM_ACCT(size));
182 static inline int shmem_reacct_size(unsigned long flags,
183 loff_t oldsize, loff_t newsize)
185 if (!(flags & VM_NORESERVE)) {
186 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
187 return security_vm_enough_memory_mm(current->mm,
188 VM_ACCT(newsize) - VM_ACCT(oldsize));
189 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
190 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
196 * ... whereas tmpfs objects are accounted incrementally as
197 * pages are allocated, in order to allow large sparse files.
198 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
199 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201 static inline int shmem_acct_block(unsigned long flags, long pages)
203 if (!(flags & VM_NORESERVE))
206 return security_vm_enough_memory_mm(current->mm,
207 pages * VM_ACCT(PAGE_SIZE));
210 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
212 if (flags & VM_NORESERVE)
213 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
216 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
218 struct shmem_inode_info *info = SHMEM_I(inode);
219 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
221 if (shmem_acct_block(info->flags, pages))
224 if (sbinfo->max_blocks) {
225 if (percpu_counter_compare(&sbinfo->used_blocks,
226 sbinfo->max_blocks - pages) > 0)
228 percpu_counter_add(&sbinfo->used_blocks, pages);
234 shmem_unacct_blocks(info->flags, pages);
238 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
240 struct shmem_inode_info *info = SHMEM_I(inode);
241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
243 if (sbinfo->max_blocks)
244 percpu_counter_sub(&sbinfo->used_blocks, pages);
245 shmem_unacct_blocks(info->flags, pages);
248 static const struct super_operations shmem_ops;
249 const struct address_space_operations shmem_aops;
250 static const struct file_operations shmem_file_operations;
251 static const struct inode_operations shmem_inode_operations;
252 static const struct inode_operations shmem_dir_inode_operations;
253 static const struct inode_operations shmem_special_inode_operations;
254 static const struct vm_operations_struct shmem_vm_ops;
255 static struct file_system_type shmem_fs_type;
257 bool vma_is_shmem(struct vm_area_struct *vma)
259 return vma->vm_ops == &shmem_vm_ops;
262 static LIST_HEAD(shmem_swaplist);
263 static DEFINE_MUTEX(shmem_swaplist_mutex);
266 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
267 * produces a novel ino for the newly allocated inode.
269 * It may also be called when making a hard link to permit the space needed by
270 * each dentry. However, in that case, no new inode number is needed since that
271 * internally draws from another pool of inode numbers (currently global
272 * get_next_ino()). This case is indicated by passing NULL as inop.
274 #define SHMEM_INO_BATCH 1024
275 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
277 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
280 if (!(sb->s_flags & SB_KERNMOUNT)) {
281 spin_lock(&sbinfo->stat_lock);
282 if (sbinfo->max_inodes) {
283 if (!sbinfo->free_inodes) {
284 spin_unlock(&sbinfo->stat_lock);
287 sbinfo->free_inodes--;
290 ino = sbinfo->next_ino++;
291 if (unlikely(is_zero_ino(ino)))
292 ino = sbinfo->next_ino++;
293 if (unlikely(!sbinfo->full_inums &&
296 * Emulate get_next_ino uint wraparound for
299 if (IS_ENABLED(CONFIG_64BIT))
300 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
301 __func__, MINOR(sb->s_dev));
302 sbinfo->next_ino = 1;
303 ino = sbinfo->next_ino++;
307 spin_unlock(&sbinfo->stat_lock);
310 * __shmem_file_setup, one of our callers, is lock-free: it
311 * doesn't hold stat_lock in shmem_reserve_inode since
312 * max_inodes is always 0, and is called from potentially
313 * unknown contexts. As such, use a per-cpu batched allocator
314 * which doesn't require the per-sb stat_lock unless we are at
315 * the batch boundary.
317 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
318 * shmem mounts are not exposed to userspace, so we don't need
319 * to worry about things like glibc compatibility.
322 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
324 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
325 spin_lock(&sbinfo->stat_lock);
326 ino = sbinfo->next_ino;
327 sbinfo->next_ino += SHMEM_INO_BATCH;
328 spin_unlock(&sbinfo->stat_lock);
329 if (unlikely(is_zero_ino(ino)))
340 static void shmem_free_inode(struct super_block *sb)
342 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
343 if (sbinfo->max_inodes) {
344 spin_lock(&sbinfo->stat_lock);
345 sbinfo->free_inodes++;
346 spin_unlock(&sbinfo->stat_lock);
351 * shmem_recalc_inode - recalculate the block usage of an inode
352 * @inode: inode to recalc
354 * We have to calculate the free blocks since the mm can drop
355 * undirtied hole pages behind our back.
357 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
358 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
360 * It has to be called with the spinlock held.
362 static void shmem_recalc_inode(struct inode *inode)
364 struct shmem_inode_info *info = SHMEM_I(inode);
367 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
369 info->alloced -= freed;
370 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
371 shmem_inode_unacct_blocks(inode, freed);
375 bool shmem_charge(struct inode *inode, long pages)
377 struct shmem_inode_info *info = SHMEM_I(inode);
380 if (!shmem_inode_acct_block(inode, pages))
383 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
384 inode->i_mapping->nrpages += pages;
386 spin_lock_irqsave(&info->lock, flags);
387 info->alloced += pages;
388 inode->i_blocks += pages * BLOCKS_PER_PAGE;
389 shmem_recalc_inode(inode);
390 spin_unlock_irqrestore(&info->lock, flags);
395 void shmem_uncharge(struct inode *inode, long pages)
397 struct shmem_inode_info *info = SHMEM_I(inode);
400 /* nrpages adjustment done by __delete_from_page_cache() or caller */
402 spin_lock_irqsave(&info->lock, flags);
403 info->alloced -= pages;
404 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
405 shmem_recalc_inode(inode);
406 spin_unlock_irqrestore(&info->lock, flags);
408 shmem_inode_unacct_blocks(inode, pages);
412 * Replace item expected in xarray by a new item, while holding xa_lock.
414 static int shmem_replace_entry(struct address_space *mapping,
415 pgoff_t index, void *expected, void *replacement)
417 XA_STATE(xas, &mapping->i_pages, index);
420 VM_BUG_ON(!expected);
421 VM_BUG_ON(!replacement);
422 item = xas_load(&xas);
423 if (item != expected)
425 xas_store(&xas, replacement);
430 * Sometimes, before we decide whether to proceed or to fail, we must check
431 * that an entry was not already brought back from swap by a racing thread.
433 * Checking page is not enough: by the time a SwapCache page is locked, it
434 * might be reused, and again be SwapCache, using the same swap as before.
436 static bool shmem_confirm_swap(struct address_space *mapping,
437 pgoff_t index, swp_entry_t swap)
439 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
443 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
446 * disables huge pages for the mount;
448 * enables huge pages for the mount;
449 * SHMEM_HUGE_WITHIN_SIZE:
450 * only allocate huge pages if the page will be fully within i_size,
451 * also respect fadvise()/madvise() hints;
453 * only allocate huge pages if requested with fadvise()/madvise();
456 #define SHMEM_HUGE_NEVER 0
457 #define SHMEM_HUGE_ALWAYS 1
458 #define SHMEM_HUGE_WITHIN_SIZE 2
459 #define SHMEM_HUGE_ADVISE 3
463 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
466 * disables huge on shm_mnt and all mounts, for emergency use;
468 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
471 #define SHMEM_HUGE_DENY (-1)
472 #define SHMEM_HUGE_FORCE (-2)
474 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
475 /* ifdef here to avoid bloating shmem.o when not necessary */
477 static int shmem_huge __read_mostly;
479 #if defined(CONFIG_SYSFS)
480 static int shmem_parse_huge(const char *str)
482 if (!strcmp(str, "never"))
483 return SHMEM_HUGE_NEVER;
484 if (!strcmp(str, "always"))
485 return SHMEM_HUGE_ALWAYS;
486 if (!strcmp(str, "within_size"))
487 return SHMEM_HUGE_WITHIN_SIZE;
488 if (!strcmp(str, "advise"))
489 return SHMEM_HUGE_ADVISE;
490 if (!strcmp(str, "deny"))
491 return SHMEM_HUGE_DENY;
492 if (!strcmp(str, "force"))
493 return SHMEM_HUGE_FORCE;
498 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
499 static const char *shmem_format_huge(int huge)
502 case SHMEM_HUGE_NEVER:
504 case SHMEM_HUGE_ALWAYS:
506 case SHMEM_HUGE_WITHIN_SIZE:
507 return "within_size";
508 case SHMEM_HUGE_ADVISE:
510 case SHMEM_HUGE_DENY:
512 case SHMEM_HUGE_FORCE:
521 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
522 struct shrink_control *sc, unsigned long nr_to_split)
524 LIST_HEAD(list), *pos, *next;
525 LIST_HEAD(to_remove);
527 struct shmem_inode_info *info;
529 unsigned long batch = sc ? sc->nr_to_scan : 128;
530 int removed = 0, split = 0;
532 if (list_empty(&sbinfo->shrinklist))
535 spin_lock(&sbinfo->shrinklist_lock);
536 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
537 info = list_entry(pos, struct shmem_inode_info, shrinklist);
540 inode = igrab(&info->vfs_inode);
542 /* inode is about to be evicted */
544 list_del_init(&info->shrinklist);
549 /* Check if there's anything to gain */
550 if (round_up(inode->i_size, PAGE_SIZE) ==
551 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
552 list_move(&info->shrinklist, &to_remove);
557 list_move(&info->shrinklist, &list);
562 spin_unlock(&sbinfo->shrinklist_lock);
564 list_for_each_safe(pos, next, &to_remove) {
565 info = list_entry(pos, struct shmem_inode_info, shrinklist);
566 inode = &info->vfs_inode;
567 list_del_init(&info->shrinklist);
571 list_for_each_safe(pos, next, &list) {
574 info = list_entry(pos, struct shmem_inode_info, shrinklist);
575 inode = &info->vfs_inode;
577 if (nr_to_split && split >= nr_to_split)
580 page = find_get_page(inode->i_mapping,
581 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
585 /* No huge page at the end of the file: nothing to split */
586 if (!PageTransHuge(page)) {
592 * Leave the inode on the list if we failed to lock
593 * the page at this time.
595 * Waiting for the lock may lead to deadlock in the
598 if (!trylock_page(page)) {
603 ret = split_huge_page(page);
607 /* If split failed leave the inode on the list */
613 list_del_init(&info->shrinklist);
619 spin_lock(&sbinfo->shrinklist_lock);
620 list_splice_tail(&list, &sbinfo->shrinklist);
621 sbinfo->shrinklist_len -= removed;
622 spin_unlock(&sbinfo->shrinklist_lock);
627 static long shmem_unused_huge_scan(struct super_block *sb,
628 struct shrink_control *sc)
630 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
632 if (!READ_ONCE(sbinfo->shrinklist_len))
635 return shmem_unused_huge_shrink(sbinfo, sc, 0);
638 static long shmem_unused_huge_count(struct super_block *sb,
639 struct shrink_control *sc)
641 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
642 return READ_ONCE(sbinfo->shrinklist_len);
644 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
646 #define shmem_huge SHMEM_HUGE_DENY
648 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
649 struct shrink_control *sc, unsigned long nr_to_split)
653 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
655 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
657 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
658 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
659 shmem_huge != SHMEM_HUGE_DENY)
665 * Like add_to_page_cache_locked, but error if expected item has gone.
667 static int shmem_add_to_page_cache(struct page *page,
668 struct address_space *mapping,
669 pgoff_t index, void *expected, gfp_t gfp,
670 struct mm_struct *charge_mm)
672 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
674 unsigned long nr = compound_nr(page);
677 VM_BUG_ON_PAGE(PageTail(page), page);
678 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
679 VM_BUG_ON_PAGE(!PageLocked(page), page);
680 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
681 VM_BUG_ON(expected && PageTransHuge(page));
683 page_ref_add(page, nr);
684 page->mapping = mapping;
687 if (!PageSwapCache(page)) {
688 error = mem_cgroup_charge(page, charge_mm, gfp);
690 if (PageTransHuge(page)) {
691 count_vm_event(THP_FILE_FALLBACK);
692 count_vm_event(THP_FILE_FALLBACK_CHARGE);
697 cgroup_throttle_swaprate(page, gfp);
702 entry = xas_find_conflict(&xas);
703 if (entry != expected)
704 xas_set_err(&xas, -EEXIST);
705 xas_create_range(&xas);
709 xas_store(&xas, page);
714 if (PageTransHuge(page)) {
715 count_vm_event(THP_FILE_ALLOC);
716 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
718 mapping->nrpages += nr;
719 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
720 __mod_lruvec_page_state(page, NR_SHMEM, nr);
722 xas_unlock_irq(&xas);
723 } while (xas_nomem(&xas, gfp));
725 if (xas_error(&xas)) {
726 error = xas_error(&xas);
732 page->mapping = NULL;
733 page_ref_sub(page, nr);
738 * Like delete_from_page_cache, but substitutes swap for page.
740 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
742 struct address_space *mapping = page->mapping;
745 VM_BUG_ON_PAGE(PageCompound(page), page);
747 xa_lock_irq(&mapping->i_pages);
748 error = shmem_replace_entry(mapping, page->index, page, radswap);
749 page->mapping = NULL;
751 __dec_lruvec_page_state(page, NR_FILE_PAGES);
752 __dec_lruvec_page_state(page, NR_SHMEM);
753 xa_unlock_irq(&mapping->i_pages);
759 * Remove swap entry from page cache, free the swap and its page cache.
761 static int shmem_free_swap(struct address_space *mapping,
762 pgoff_t index, void *radswap)
766 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
769 free_swap_and_cache(radix_to_swp_entry(radswap));
774 * Determine (in bytes) how many of the shmem object's pages mapped by the
775 * given offsets are swapped out.
777 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
778 * as long as the inode doesn't go away and racy results are not a problem.
780 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
781 pgoff_t start, pgoff_t end)
783 XA_STATE(xas, &mapping->i_pages, start);
785 unsigned long swapped = 0;
788 xas_for_each(&xas, page, end - 1) {
789 if (xas_retry(&xas, page))
791 if (xa_is_value(page))
794 if (need_resched()) {
802 return swapped << PAGE_SHIFT;
806 * Determine (in bytes) how many of the shmem object's pages mapped by the
807 * given vma is swapped out.
809 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
810 * as long as the inode doesn't go away and racy results are not a problem.
812 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
814 struct inode *inode = file_inode(vma->vm_file);
815 struct shmem_inode_info *info = SHMEM_I(inode);
816 struct address_space *mapping = inode->i_mapping;
817 unsigned long swapped;
819 /* Be careful as we don't hold info->lock */
820 swapped = READ_ONCE(info->swapped);
823 * The easier cases are when the shmem object has nothing in swap, or
824 * the vma maps it whole. Then we can simply use the stats that we
830 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
831 return swapped << PAGE_SHIFT;
833 /* Here comes the more involved part */
834 return shmem_partial_swap_usage(mapping,
835 linear_page_index(vma, vma->vm_start),
836 linear_page_index(vma, vma->vm_end));
840 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
842 void shmem_unlock_mapping(struct address_space *mapping)
849 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
851 while (!mapping_unevictable(mapping)) {
852 if (!pagevec_lookup(&pvec, mapping, &index))
854 check_move_unevictable_pages(&pvec);
855 pagevec_release(&pvec);
861 * Check whether a hole-punch or truncation needs to split a huge page,
862 * returning true if no split was required, or the split has been successful.
864 * Eviction (or truncation to 0 size) should never need to split a huge page;
865 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
866 * head, and then succeeded to trylock on tail.
868 * A split can only succeed when there are no additional references on the
869 * huge page: so the split below relies upon find_get_entries() having stopped
870 * when it found a subpage of the huge page, without getting further references.
872 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
874 if (!PageTransCompound(page))
877 /* Just proceed to delete a huge page wholly within the range punched */
878 if (PageHead(page) &&
879 page->index >= start && page->index + HPAGE_PMD_NR <= end)
882 /* Try to split huge page, so we can truly punch the hole or truncate */
883 return split_huge_page(page) >= 0;
887 * Remove range of pages and swap entries from page cache, and free them.
888 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
890 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
893 struct address_space *mapping = inode->i_mapping;
894 struct shmem_inode_info *info = SHMEM_I(inode);
895 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
896 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
897 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
898 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
900 pgoff_t indices[PAGEVEC_SIZE];
901 long nr_swaps_freed = 0;
906 end = -1; /* unsigned, so actually very big */
910 while (index < end) {
911 pvec.nr = find_get_entries(mapping, index,
912 min(end - index, (pgoff_t)PAGEVEC_SIZE),
913 pvec.pages, indices);
916 for (i = 0; i < pagevec_count(&pvec); i++) {
917 struct page *page = pvec.pages[i];
923 if (xa_is_value(page)) {
926 nr_swaps_freed += !shmem_free_swap(mapping,
931 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
933 if (!trylock_page(page))
936 if ((!unfalloc || !PageUptodate(page)) &&
937 page_mapping(page) == mapping) {
938 VM_BUG_ON_PAGE(PageWriteback(page), page);
939 if (shmem_punch_compound(page, start, end))
940 truncate_inode_page(mapping, page);
944 pagevec_remove_exceptionals(&pvec);
945 pagevec_release(&pvec);
951 struct page *page = NULL;
952 shmem_getpage(inode, start - 1, &page, SGP_READ);
954 unsigned int top = PAGE_SIZE;
959 zero_user_segment(page, partial_start, top);
960 set_page_dirty(page);
966 struct page *page = NULL;
967 shmem_getpage(inode, end, &page, SGP_READ);
969 zero_user_segment(page, 0, partial_end);
970 set_page_dirty(page);
979 while (index < end) {
982 pvec.nr = find_get_entries(mapping, index,
983 min(end - index, (pgoff_t)PAGEVEC_SIZE),
984 pvec.pages, indices);
986 /* If all gone or hole-punch or unfalloc, we're done */
987 if (index == start || end != -1)
989 /* But if truncating, restart to make sure all gone */
993 for (i = 0; i < pagevec_count(&pvec); i++) {
994 struct page *page = pvec.pages[i];
1000 if (xa_is_value(page)) {
1003 if (shmem_free_swap(mapping, index, page)) {
1004 /* Swap was replaced by page: retry */
1014 if (!unfalloc || !PageUptodate(page)) {
1015 if (page_mapping(page) != mapping) {
1016 /* Page was replaced by swap: retry */
1021 VM_BUG_ON_PAGE(PageWriteback(page), page);
1022 if (shmem_punch_compound(page, start, end))
1023 truncate_inode_page(mapping, page);
1024 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1025 /* Wipe the page and don't get stuck */
1026 clear_highpage(page);
1027 flush_dcache_page(page);
1028 set_page_dirty(page);
1030 round_up(start, HPAGE_PMD_NR))
1036 pagevec_remove_exceptionals(&pvec);
1037 pagevec_release(&pvec);
1041 spin_lock_irq(&info->lock);
1042 info->swapped -= nr_swaps_freed;
1043 shmem_recalc_inode(inode);
1044 spin_unlock_irq(&info->lock);
1047 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1049 shmem_undo_range(inode, lstart, lend, false);
1050 inode->i_ctime = inode->i_mtime = current_time(inode);
1052 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1054 static int shmem_getattr(struct user_namespace *mnt_userns,
1055 const struct path *path, struct kstat *stat,
1056 u32 request_mask, unsigned int query_flags)
1058 struct inode *inode = path->dentry->d_inode;
1059 struct shmem_inode_info *info = SHMEM_I(inode);
1060 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1062 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1063 spin_lock_irq(&info->lock);
1064 shmem_recalc_inode(inode);
1065 spin_unlock_irq(&info->lock);
1067 generic_fillattr(&init_user_ns, inode, stat);
1069 if (is_huge_enabled(sb_info))
1070 stat->blksize = HPAGE_PMD_SIZE;
1075 static int shmem_setattr(struct user_namespace *mnt_userns,
1076 struct dentry *dentry, struct iattr *attr)
1078 struct inode *inode = d_inode(dentry);
1079 struct shmem_inode_info *info = SHMEM_I(inode);
1080 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1083 error = setattr_prepare(&init_user_ns, dentry, attr);
1087 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1088 loff_t oldsize = inode->i_size;
1089 loff_t newsize = attr->ia_size;
1091 /* protected by i_mutex */
1092 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1093 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1096 if (newsize != oldsize) {
1097 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1101 i_size_write(inode, newsize);
1102 inode->i_ctime = inode->i_mtime = current_time(inode);
1104 if (newsize <= oldsize) {
1105 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1106 if (oldsize > holebegin)
1107 unmap_mapping_range(inode->i_mapping,
1110 shmem_truncate_range(inode,
1111 newsize, (loff_t)-1);
1112 /* unmap again to remove racily COWed private pages */
1113 if (oldsize > holebegin)
1114 unmap_mapping_range(inode->i_mapping,
1118 * Part of the huge page can be beyond i_size: subject
1119 * to shrink under memory pressure.
1121 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1122 spin_lock(&sbinfo->shrinklist_lock);
1124 * _careful to defend against unlocked access to
1125 * ->shrink_list in shmem_unused_huge_shrink()
1127 if (list_empty_careful(&info->shrinklist)) {
1128 list_add_tail(&info->shrinklist,
1129 &sbinfo->shrinklist);
1130 sbinfo->shrinklist_len++;
1132 spin_unlock(&sbinfo->shrinklist_lock);
1137 setattr_copy(&init_user_ns, inode, attr);
1138 if (attr->ia_valid & ATTR_MODE)
1139 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1143 static void shmem_evict_inode(struct inode *inode)
1145 struct shmem_inode_info *info = SHMEM_I(inode);
1146 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1148 if (shmem_mapping(inode->i_mapping)) {
1149 shmem_unacct_size(info->flags, inode->i_size);
1151 shmem_truncate_range(inode, 0, (loff_t)-1);
1152 if (!list_empty(&info->shrinklist)) {
1153 spin_lock(&sbinfo->shrinklist_lock);
1154 if (!list_empty(&info->shrinklist)) {
1155 list_del_init(&info->shrinklist);
1156 sbinfo->shrinklist_len--;
1158 spin_unlock(&sbinfo->shrinklist_lock);
1160 while (!list_empty(&info->swaplist)) {
1161 /* Wait while shmem_unuse() is scanning this inode... */
1162 wait_var_event(&info->stop_eviction,
1163 !atomic_read(&info->stop_eviction));
1164 mutex_lock(&shmem_swaplist_mutex);
1165 /* ...but beware of the race if we peeked too early */
1166 if (!atomic_read(&info->stop_eviction))
1167 list_del_init(&info->swaplist);
1168 mutex_unlock(&shmem_swaplist_mutex);
1172 simple_xattrs_free(&info->xattrs);
1173 WARN_ON(inode->i_blocks);
1174 shmem_free_inode(inode->i_sb);
1178 extern struct swap_info_struct *swap_info[];
1180 static int shmem_find_swap_entries(struct address_space *mapping,
1181 pgoff_t start, unsigned int nr_entries,
1182 struct page **entries, pgoff_t *indices,
1183 unsigned int type, bool frontswap)
1185 XA_STATE(xas, &mapping->i_pages, start);
1188 unsigned int ret = 0;
1194 xas_for_each(&xas, page, ULONG_MAX) {
1195 if (xas_retry(&xas, page))
1198 if (!xa_is_value(page))
1201 entry = radix_to_swp_entry(page);
1202 if (swp_type(entry) != type)
1205 !frontswap_test(swap_info[type], swp_offset(entry)))
1208 indices[ret] = xas.xa_index;
1209 entries[ret] = page;
1211 if (need_resched()) {
1215 if (++ret == nr_entries)
1224 * Move the swapped pages for an inode to page cache. Returns the count
1225 * of pages swapped in, or the error in case of failure.
1227 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1233 struct address_space *mapping = inode->i_mapping;
1235 for (i = 0; i < pvec.nr; i++) {
1236 struct page *page = pvec.pages[i];
1238 if (!xa_is_value(page))
1240 error = shmem_swapin_page(inode, indices[i],
1242 mapping_gfp_mask(mapping),
1249 if (error == -ENOMEM)
1253 return error ? error : ret;
1257 * If swap found in inode, free it and move page from swapcache to filecache.
1259 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1260 bool frontswap, unsigned long *fs_pages_to_unuse)
1262 struct address_space *mapping = inode->i_mapping;
1264 struct pagevec pvec;
1265 pgoff_t indices[PAGEVEC_SIZE];
1266 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1269 pagevec_init(&pvec);
1271 unsigned int nr_entries = PAGEVEC_SIZE;
1273 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1274 nr_entries = *fs_pages_to_unuse;
1276 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1277 pvec.pages, indices,
1284 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1288 if (frontswap_partial) {
1289 *fs_pages_to_unuse -= ret;
1290 if (*fs_pages_to_unuse == 0) {
1291 ret = FRONTSWAP_PAGES_UNUSED;
1296 start = indices[pvec.nr - 1];
1303 * Read all the shared memory data that resides in the swap
1304 * device 'type' back into memory, so the swap device can be
1307 int shmem_unuse(unsigned int type, bool frontswap,
1308 unsigned long *fs_pages_to_unuse)
1310 struct shmem_inode_info *info, *next;
1313 if (list_empty(&shmem_swaplist))
1316 mutex_lock(&shmem_swaplist_mutex);
1317 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1318 if (!info->swapped) {
1319 list_del_init(&info->swaplist);
1323 * Drop the swaplist mutex while searching the inode for swap;
1324 * but before doing so, make sure shmem_evict_inode() will not
1325 * remove placeholder inode from swaplist, nor let it be freed
1326 * (igrab() would protect from unlink, but not from unmount).
1328 atomic_inc(&info->stop_eviction);
1329 mutex_unlock(&shmem_swaplist_mutex);
1331 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1335 mutex_lock(&shmem_swaplist_mutex);
1336 next = list_next_entry(info, swaplist);
1338 list_del_init(&info->swaplist);
1339 if (atomic_dec_and_test(&info->stop_eviction))
1340 wake_up_var(&info->stop_eviction);
1344 mutex_unlock(&shmem_swaplist_mutex);
1350 * Move the page from the page cache to the swap cache.
1352 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1354 struct shmem_inode_info *info;
1355 struct address_space *mapping;
1356 struct inode *inode;
1360 VM_BUG_ON_PAGE(PageCompound(page), page);
1361 BUG_ON(!PageLocked(page));
1362 mapping = page->mapping;
1363 index = page->index;
1364 inode = mapping->host;
1365 info = SHMEM_I(inode);
1366 if (info->flags & VM_LOCKED)
1368 if (!total_swap_pages)
1372 * Our capabilities prevent regular writeback or sync from ever calling
1373 * shmem_writepage; but a stacking filesystem might use ->writepage of
1374 * its underlying filesystem, in which case tmpfs should write out to
1375 * swap only in response to memory pressure, and not for the writeback
1378 if (!wbc->for_reclaim) {
1379 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1384 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1385 * value into swapfile.c, the only way we can correctly account for a
1386 * fallocated page arriving here is now to initialize it and write it.
1388 * That's okay for a page already fallocated earlier, but if we have
1389 * not yet completed the fallocation, then (a) we want to keep track
1390 * of this page in case we have to undo it, and (b) it may not be a
1391 * good idea to continue anyway, once we're pushing into swap. So
1392 * reactivate the page, and let shmem_fallocate() quit when too many.
1394 if (!PageUptodate(page)) {
1395 if (inode->i_private) {
1396 struct shmem_falloc *shmem_falloc;
1397 spin_lock(&inode->i_lock);
1398 shmem_falloc = inode->i_private;
1400 !shmem_falloc->waitq &&
1401 index >= shmem_falloc->start &&
1402 index < shmem_falloc->next)
1403 shmem_falloc->nr_unswapped++;
1405 shmem_falloc = NULL;
1406 spin_unlock(&inode->i_lock);
1410 clear_highpage(page);
1411 flush_dcache_page(page);
1412 SetPageUptodate(page);
1415 swap = get_swap_page(page);
1420 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1421 * if it's not already there. Do it now before the page is
1422 * moved to swap cache, when its pagelock no longer protects
1423 * the inode from eviction. But don't unlock the mutex until
1424 * we've incremented swapped, because shmem_unuse_inode() will
1425 * prune a !swapped inode from the swaplist under this mutex.
1427 mutex_lock(&shmem_swaplist_mutex);
1428 if (list_empty(&info->swaplist))
1429 list_add(&info->swaplist, &shmem_swaplist);
1431 if (add_to_swap_cache(page, swap,
1432 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1434 spin_lock_irq(&info->lock);
1435 shmem_recalc_inode(inode);
1437 spin_unlock_irq(&info->lock);
1439 swap_shmem_alloc(swap);
1440 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1442 mutex_unlock(&shmem_swaplist_mutex);
1443 BUG_ON(page_mapped(page));
1444 swap_writepage(page, wbc);
1448 mutex_unlock(&shmem_swaplist_mutex);
1449 put_swap_page(page, swap);
1451 set_page_dirty(page);
1452 if (wbc->for_reclaim)
1453 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1458 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1459 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1463 if (!mpol || mpol->mode == MPOL_DEFAULT)
1464 return; /* show nothing */
1466 mpol_to_str(buffer, sizeof(buffer), mpol);
1468 seq_printf(seq, ",mpol=%s", buffer);
1471 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1473 struct mempolicy *mpol = NULL;
1475 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1476 mpol = sbinfo->mpol;
1478 spin_unlock(&sbinfo->stat_lock);
1482 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1483 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1486 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1490 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1492 #define vm_policy vm_private_data
1495 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1496 struct shmem_inode_info *info, pgoff_t index)
1498 /* Create a pseudo vma that just contains the policy */
1499 vma_init(vma, NULL);
1500 /* Bias interleave by inode number to distribute better across nodes */
1501 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1502 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1505 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1507 /* Drop reference taken by mpol_shared_policy_lookup() */
1508 mpol_cond_put(vma->vm_policy);
1511 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1512 struct shmem_inode_info *info, pgoff_t index)
1514 struct vm_area_struct pvma;
1516 struct vm_fault vmf = {
1520 shmem_pseudo_vma_init(&pvma, info, index);
1521 page = swap_cluster_readahead(swap, gfp, &vmf);
1522 shmem_pseudo_vma_destroy(&pvma);
1527 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1528 struct shmem_inode_info *info, pgoff_t index)
1530 struct vm_area_struct pvma;
1531 struct address_space *mapping = info->vfs_inode.i_mapping;
1535 hindex = round_down(index, HPAGE_PMD_NR);
1536 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1540 shmem_pseudo_vma_init(&pvma, info, hindex);
1541 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1542 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1543 shmem_pseudo_vma_destroy(&pvma);
1545 prep_transhuge_page(page);
1547 count_vm_event(THP_FILE_FALLBACK);
1551 static struct page *shmem_alloc_page(gfp_t gfp,
1552 struct shmem_inode_info *info, pgoff_t index)
1554 struct vm_area_struct pvma;
1557 shmem_pseudo_vma_init(&pvma, info, index);
1558 page = alloc_page_vma(gfp, &pvma, 0);
1559 shmem_pseudo_vma_destroy(&pvma);
1564 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1565 struct inode *inode,
1566 pgoff_t index, bool huge)
1568 struct shmem_inode_info *info = SHMEM_I(inode);
1573 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1575 nr = huge ? HPAGE_PMD_NR : 1;
1577 if (!shmem_inode_acct_block(inode, nr))
1581 page = shmem_alloc_hugepage(gfp, info, index);
1583 page = shmem_alloc_page(gfp, info, index);
1585 __SetPageLocked(page);
1586 __SetPageSwapBacked(page);
1591 shmem_inode_unacct_blocks(inode, nr);
1593 return ERR_PTR(err);
1597 * When a page is moved from swapcache to shmem filecache (either by the
1598 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1599 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1600 * ignorance of the mapping it belongs to. If that mapping has special
1601 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1602 * we may need to copy to a suitable page before moving to filecache.
1604 * In a future release, this may well be extended to respect cpuset and
1605 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1606 * but for now it is a simple matter of zone.
1608 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1610 return page_zonenum(page) > gfp_zone(gfp);
1613 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1614 struct shmem_inode_info *info, pgoff_t index)
1616 struct page *oldpage, *newpage;
1617 struct address_space *swap_mapping;
1623 entry.val = page_private(oldpage);
1624 swap_index = swp_offset(entry);
1625 swap_mapping = page_mapping(oldpage);
1628 * We have arrived here because our zones are constrained, so don't
1629 * limit chance of success by further cpuset and node constraints.
1631 gfp &= ~GFP_CONSTRAINT_MASK;
1632 newpage = shmem_alloc_page(gfp, info, index);
1637 copy_highpage(newpage, oldpage);
1638 flush_dcache_page(newpage);
1640 __SetPageLocked(newpage);
1641 __SetPageSwapBacked(newpage);
1642 SetPageUptodate(newpage);
1643 set_page_private(newpage, entry.val);
1644 SetPageSwapCache(newpage);
1647 * Our caller will very soon move newpage out of swapcache, but it's
1648 * a nice clean interface for us to replace oldpage by newpage there.
1650 xa_lock_irq(&swap_mapping->i_pages);
1651 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1653 mem_cgroup_migrate(oldpage, newpage);
1654 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1655 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1657 xa_unlock_irq(&swap_mapping->i_pages);
1659 if (unlikely(error)) {
1661 * Is this possible? I think not, now that our callers check
1662 * both PageSwapCache and page_private after getting page lock;
1663 * but be defensive. Reverse old to newpage for clear and free.
1667 lru_cache_add(newpage);
1671 ClearPageSwapCache(oldpage);
1672 set_page_private(oldpage, 0);
1674 unlock_page(oldpage);
1681 * Swap in the page pointed to by *pagep.
1682 * Caller has to make sure that *pagep contains a valid swapped page.
1683 * Returns 0 and the page in pagep if success. On failure, returns the
1684 * error code and NULL in *pagep.
1686 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1687 struct page **pagep, enum sgp_type sgp,
1688 gfp_t gfp, struct vm_area_struct *vma,
1689 vm_fault_t *fault_type)
1691 struct address_space *mapping = inode->i_mapping;
1692 struct shmem_inode_info *info = SHMEM_I(inode);
1693 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
1698 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1699 swap = radix_to_swp_entry(*pagep);
1702 /* Look it up and read it in.. */
1703 page = lookup_swap_cache(swap, NULL, 0);
1705 /* Or update major stats only when swapin succeeds?? */
1707 *fault_type |= VM_FAULT_MAJOR;
1708 count_vm_event(PGMAJFAULT);
1709 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1711 /* Here we actually start the io */
1712 page = shmem_swapin(swap, gfp, info, index);
1719 /* We have to do this with page locked to prevent races */
1721 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1722 !shmem_confirm_swap(mapping, index, swap)) {
1726 if (!PageUptodate(page)) {
1730 wait_on_page_writeback(page);
1733 * Some architectures may have to restore extra metadata to the
1734 * physical page after reading from swap.
1736 arch_swap_restore(swap, page);
1738 if (shmem_should_replace_page(page, gfp)) {
1739 error = shmem_replace_page(&page, gfp, info, index);
1744 error = shmem_add_to_page_cache(page, mapping, index,
1745 swp_to_radix_entry(swap), gfp,
1750 spin_lock_irq(&info->lock);
1752 shmem_recalc_inode(inode);
1753 spin_unlock_irq(&info->lock);
1755 if (sgp == SGP_WRITE)
1756 mark_page_accessed(page);
1758 delete_from_swap_cache(page);
1759 set_page_dirty(page);
1765 if (!shmem_confirm_swap(mapping, index, swap))
1777 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1779 * If we allocate a new one we do not mark it dirty. That's up to the
1780 * vm. If we swap it in we mark it dirty since we also free the swap
1781 * entry since a page cannot live in both the swap and page cache.
1783 * vmf and fault_type are only supplied by shmem_fault:
1784 * otherwise they are NULL.
1786 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1787 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1788 struct vm_area_struct *vma, struct vm_fault *vmf,
1789 vm_fault_t *fault_type)
1791 struct address_space *mapping = inode->i_mapping;
1792 struct shmem_inode_info *info = SHMEM_I(inode);
1793 struct shmem_sb_info *sbinfo;
1794 struct mm_struct *charge_mm;
1796 enum sgp_type sgp_huge = sgp;
1797 pgoff_t hindex = index;
1802 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1804 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1807 if (sgp <= SGP_CACHE &&
1808 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1812 sbinfo = SHMEM_SB(inode->i_sb);
1813 charge_mm = vma ? vma->vm_mm : current->mm;
1815 page = find_lock_entry(mapping, index);
1816 if (xa_is_value(page)) {
1817 error = shmem_swapin_page(inode, index, &page,
1818 sgp, gfp, vma, fault_type);
1819 if (error == -EEXIST)
1827 hindex = page->index;
1828 if (page && sgp == SGP_WRITE)
1829 mark_page_accessed(page);
1831 /* fallocated page? */
1832 if (page && !PageUptodate(page)) {
1833 if (sgp != SGP_READ)
1840 if (page || sgp == SGP_READ)
1844 * Fast cache lookup did not find it:
1845 * bring it back from swap or allocate.
1848 if (vma && userfaultfd_missing(vma)) {
1849 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1853 /* shmem_symlink() */
1854 if (!shmem_mapping(mapping))
1856 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1858 if (shmem_huge == SHMEM_HUGE_FORCE)
1860 switch (sbinfo->huge) {
1861 case SHMEM_HUGE_NEVER:
1863 case SHMEM_HUGE_WITHIN_SIZE: {
1867 off = round_up(index, HPAGE_PMD_NR);
1868 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1869 if (i_size >= HPAGE_PMD_SIZE &&
1870 i_size >> PAGE_SHIFT >= off)
1875 case SHMEM_HUGE_ADVISE:
1876 if (sgp_huge == SGP_HUGE)
1878 /* TODO: implement fadvise() hints */
1883 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1886 page = shmem_alloc_and_acct_page(gfp, inode,
1892 error = PTR_ERR(page);
1894 if (error != -ENOSPC)
1897 * Try to reclaim some space by splitting a huge page
1898 * beyond i_size on the filesystem.
1903 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1904 if (ret == SHRINK_STOP)
1912 if (PageTransHuge(page))
1913 hindex = round_down(index, HPAGE_PMD_NR);
1917 if (sgp == SGP_WRITE)
1918 __SetPageReferenced(page);
1920 error = shmem_add_to_page_cache(page, mapping, hindex,
1921 NULL, gfp & GFP_RECLAIM_MASK,
1925 lru_cache_add(page);
1927 spin_lock_irq(&info->lock);
1928 info->alloced += compound_nr(page);
1929 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1930 shmem_recalc_inode(inode);
1931 spin_unlock_irq(&info->lock);
1934 if (PageTransHuge(page) &&
1935 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1936 hindex + HPAGE_PMD_NR - 1) {
1938 * Part of the huge page is beyond i_size: subject
1939 * to shrink under memory pressure.
1941 spin_lock(&sbinfo->shrinklist_lock);
1943 * _careful to defend against unlocked access to
1944 * ->shrink_list in shmem_unused_huge_shrink()
1946 if (list_empty_careful(&info->shrinklist)) {
1947 list_add_tail(&info->shrinklist,
1948 &sbinfo->shrinklist);
1949 sbinfo->shrinklist_len++;
1951 spin_unlock(&sbinfo->shrinklist_lock);
1955 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1957 if (sgp == SGP_FALLOC)
1961 * Let SGP_WRITE caller clear ends if write does not fill page;
1962 * but SGP_FALLOC on a page fallocated earlier must initialize
1963 * it now, lest undo on failure cancel our earlier guarantee.
1965 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1968 for (i = 0; i < compound_nr(page); i++) {
1969 clear_highpage(page + i);
1970 flush_dcache_page(page + i);
1972 SetPageUptodate(page);
1975 /* Perhaps the file has been truncated since we checked */
1976 if (sgp <= SGP_CACHE &&
1977 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1979 ClearPageDirty(page);
1980 delete_from_page_cache(page);
1981 spin_lock_irq(&info->lock);
1982 shmem_recalc_inode(inode);
1983 spin_unlock_irq(&info->lock);
1989 *pagep = page + index - hindex;
1996 shmem_inode_unacct_blocks(inode, compound_nr(page));
1998 if (PageTransHuge(page)) {
2008 if (error == -ENOSPC && !once++) {
2009 spin_lock_irq(&info->lock);
2010 shmem_recalc_inode(inode);
2011 spin_unlock_irq(&info->lock);
2014 if (error == -EEXIST)
2020 * This is like autoremove_wake_function, but it removes the wait queue
2021 * entry unconditionally - even if something else had already woken the
2024 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2026 int ret = default_wake_function(wait, mode, sync, key);
2027 list_del_init(&wait->entry);
2031 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2033 struct vm_area_struct *vma = vmf->vma;
2034 struct inode *inode = file_inode(vma->vm_file);
2035 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2038 vm_fault_t ret = VM_FAULT_LOCKED;
2041 * Trinity finds that probing a hole which tmpfs is punching can
2042 * prevent the hole-punch from ever completing: which in turn
2043 * locks writers out with its hold on i_mutex. So refrain from
2044 * faulting pages into the hole while it's being punched. Although
2045 * shmem_undo_range() does remove the additions, it may be unable to
2046 * keep up, as each new page needs its own unmap_mapping_range() call,
2047 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2049 * It does not matter if we sometimes reach this check just before the
2050 * hole-punch begins, so that one fault then races with the punch:
2051 * we just need to make racing faults a rare case.
2053 * The implementation below would be much simpler if we just used a
2054 * standard mutex or completion: but we cannot take i_mutex in fault,
2055 * and bloating every shmem inode for this unlikely case would be sad.
2057 if (unlikely(inode->i_private)) {
2058 struct shmem_falloc *shmem_falloc;
2060 spin_lock(&inode->i_lock);
2061 shmem_falloc = inode->i_private;
2063 shmem_falloc->waitq &&
2064 vmf->pgoff >= shmem_falloc->start &&
2065 vmf->pgoff < shmem_falloc->next) {
2067 wait_queue_head_t *shmem_falloc_waitq;
2068 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2070 ret = VM_FAULT_NOPAGE;
2071 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2073 ret = VM_FAULT_RETRY;
2075 shmem_falloc_waitq = shmem_falloc->waitq;
2076 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2077 TASK_UNINTERRUPTIBLE);
2078 spin_unlock(&inode->i_lock);
2082 * shmem_falloc_waitq points into the shmem_fallocate()
2083 * stack of the hole-punching task: shmem_falloc_waitq
2084 * is usually invalid by the time we reach here, but
2085 * finish_wait() does not dereference it in that case;
2086 * though i_lock needed lest racing with wake_up_all().
2088 spin_lock(&inode->i_lock);
2089 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2090 spin_unlock(&inode->i_lock);
2096 spin_unlock(&inode->i_lock);
2101 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2102 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2104 else if (vma->vm_flags & VM_HUGEPAGE)
2107 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2108 gfp, vma, vmf, &ret);
2110 return vmf_error(err);
2114 unsigned long shmem_get_unmapped_area(struct file *file,
2115 unsigned long uaddr, unsigned long len,
2116 unsigned long pgoff, unsigned long flags)
2118 unsigned long (*get_area)(struct file *,
2119 unsigned long, unsigned long, unsigned long, unsigned long);
2121 unsigned long offset;
2122 unsigned long inflated_len;
2123 unsigned long inflated_addr;
2124 unsigned long inflated_offset;
2126 if (len > TASK_SIZE)
2129 get_area = current->mm->get_unmapped_area;
2130 addr = get_area(file, uaddr, len, pgoff, flags);
2132 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2134 if (IS_ERR_VALUE(addr))
2136 if (addr & ~PAGE_MASK)
2138 if (addr > TASK_SIZE - len)
2141 if (shmem_huge == SHMEM_HUGE_DENY)
2143 if (len < HPAGE_PMD_SIZE)
2145 if (flags & MAP_FIXED)
2148 * Our priority is to support MAP_SHARED mapped hugely;
2149 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2150 * But if caller specified an address hint and we allocated area there
2151 * successfully, respect that as before.
2156 if (shmem_huge != SHMEM_HUGE_FORCE) {
2157 struct super_block *sb;
2160 VM_BUG_ON(file->f_op != &shmem_file_operations);
2161 sb = file_inode(file)->i_sb;
2164 * Called directly from mm/mmap.c, or drivers/char/mem.c
2165 * for "/dev/zero", to create a shared anonymous object.
2167 if (IS_ERR(shm_mnt))
2169 sb = shm_mnt->mnt_sb;
2171 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2175 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2176 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2178 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2181 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2182 if (inflated_len > TASK_SIZE)
2184 if (inflated_len < len)
2187 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2188 if (IS_ERR_VALUE(inflated_addr))
2190 if (inflated_addr & ~PAGE_MASK)
2193 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2194 inflated_addr += offset - inflated_offset;
2195 if (inflated_offset > offset)
2196 inflated_addr += HPAGE_PMD_SIZE;
2198 if (inflated_addr > TASK_SIZE - len)
2200 return inflated_addr;
2204 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2206 struct inode *inode = file_inode(vma->vm_file);
2207 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2210 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2213 struct inode *inode = file_inode(vma->vm_file);
2216 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2217 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2221 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2223 struct inode *inode = file_inode(file);
2224 struct shmem_inode_info *info = SHMEM_I(inode);
2225 int retval = -ENOMEM;
2228 * What serializes the accesses to info->flags?
2229 * ipc_lock_object() when called from shmctl_do_lock(),
2230 * no serialization needed when called from shm_destroy().
2232 if (lock && !(info->flags & VM_LOCKED)) {
2233 if (!user_shm_lock(inode->i_size, user))
2235 info->flags |= VM_LOCKED;
2236 mapping_set_unevictable(file->f_mapping);
2238 if (!lock && (info->flags & VM_LOCKED) && user) {
2239 user_shm_unlock(inode->i_size, user);
2240 info->flags &= ~VM_LOCKED;
2241 mapping_clear_unevictable(file->f_mapping);
2249 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2251 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2253 if (info->seals & F_SEAL_FUTURE_WRITE) {
2255 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2256 * "future write" seal active.
2258 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2262 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2263 * MAP_SHARED and read-only, take care to not allow mprotect to
2264 * revert protections on such mappings. Do this only for shared
2265 * mappings. For private mappings, don't need to mask
2266 * VM_MAYWRITE as we still want them to be COW-writable.
2268 if (vma->vm_flags & VM_SHARED)
2269 vma->vm_flags &= ~(VM_MAYWRITE);
2272 /* arm64 - allow memory tagging on RAM-based files */
2273 vma->vm_flags |= VM_MTE_ALLOWED;
2275 file_accessed(file);
2276 vma->vm_ops = &shmem_vm_ops;
2277 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2278 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2279 (vma->vm_end & HPAGE_PMD_MASK)) {
2280 khugepaged_enter(vma, vma->vm_flags);
2285 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2286 umode_t mode, dev_t dev, unsigned long flags)
2288 struct inode *inode;
2289 struct shmem_inode_info *info;
2290 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2293 if (shmem_reserve_inode(sb, &ino))
2296 inode = new_inode(sb);
2299 inode_init_owner(&init_user_ns, inode, dir, mode);
2300 inode->i_blocks = 0;
2301 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2302 inode->i_generation = prandom_u32();
2303 info = SHMEM_I(inode);
2304 memset(info, 0, (char *)inode - (char *)info);
2305 spin_lock_init(&info->lock);
2306 atomic_set(&info->stop_eviction, 0);
2307 info->seals = F_SEAL_SEAL;
2308 info->flags = flags & VM_NORESERVE;
2309 INIT_LIST_HEAD(&info->shrinklist);
2310 INIT_LIST_HEAD(&info->swaplist);
2311 simple_xattrs_init(&info->xattrs);
2312 cache_no_acl(inode);
2314 switch (mode & S_IFMT) {
2316 inode->i_op = &shmem_special_inode_operations;
2317 init_special_inode(inode, mode, dev);
2320 inode->i_mapping->a_ops = &shmem_aops;
2321 inode->i_op = &shmem_inode_operations;
2322 inode->i_fop = &shmem_file_operations;
2323 mpol_shared_policy_init(&info->policy,
2324 shmem_get_sbmpol(sbinfo));
2328 /* Some things misbehave if size == 0 on a directory */
2329 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2330 inode->i_op = &shmem_dir_inode_operations;
2331 inode->i_fop = &simple_dir_operations;
2335 * Must not load anything in the rbtree,
2336 * mpol_free_shared_policy will not be called.
2338 mpol_shared_policy_init(&info->policy, NULL);
2342 lockdep_annotate_inode_mutex_key(inode);
2344 shmem_free_inode(sb);
2348 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2350 struct vm_area_struct *dst_vma,
2351 unsigned long dst_addr,
2352 unsigned long src_addr,
2354 struct page **pagep)
2356 struct inode *inode = file_inode(dst_vma->vm_file);
2357 struct shmem_inode_info *info = SHMEM_I(inode);
2358 struct address_space *mapping = inode->i_mapping;
2359 gfp_t gfp = mapping_gfp_mask(mapping);
2360 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2364 pte_t _dst_pte, *dst_pte;
2366 pgoff_t offset, max_off;
2369 if (!shmem_inode_acct_block(inode, 1))
2373 page = shmem_alloc_page(gfp, info, pgoff);
2375 goto out_unacct_blocks;
2377 if (!zeropage) { /* mcopy_atomic */
2378 page_kaddr = kmap_atomic(page);
2379 ret = copy_from_user(page_kaddr,
2380 (const void __user *)src_addr,
2382 kunmap_atomic(page_kaddr);
2384 /* fallback to copy_from_user outside mmap_lock */
2385 if (unlikely(ret)) {
2387 shmem_inode_unacct_blocks(inode, 1);
2388 /* don't free the page */
2391 } else { /* mfill_zeropage_atomic */
2392 clear_highpage(page);
2399 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2400 __SetPageLocked(page);
2401 __SetPageSwapBacked(page);
2402 __SetPageUptodate(page);
2405 offset = linear_page_index(dst_vma, dst_addr);
2406 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2407 if (unlikely(offset >= max_off))
2410 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2411 gfp & GFP_RECLAIM_MASK, dst_mm);
2415 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2416 if (dst_vma->vm_flags & VM_WRITE)
2417 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2420 * We don't set the pte dirty if the vma has no
2421 * VM_WRITE permission, so mark the page dirty or it
2422 * could be freed from under us. We could do it
2423 * unconditionally before unlock_page(), but doing it
2424 * only if VM_WRITE is not set is faster.
2426 set_page_dirty(page);
2429 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2432 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2433 if (unlikely(offset >= max_off))
2434 goto out_release_unlock;
2437 if (!pte_none(*dst_pte))
2438 goto out_release_unlock;
2440 lru_cache_add(page);
2442 spin_lock_irq(&info->lock);
2444 inode->i_blocks += BLOCKS_PER_PAGE;
2445 shmem_recalc_inode(inode);
2446 spin_unlock_irq(&info->lock);
2448 inc_mm_counter(dst_mm, mm_counter_file(page));
2449 page_add_file_rmap(page, false);
2450 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2452 /* No need to invalidate - it was non-present before */
2453 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2454 pte_unmap_unlock(dst_pte, ptl);
2460 pte_unmap_unlock(dst_pte, ptl);
2461 ClearPageDirty(page);
2462 delete_from_page_cache(page);
2467 shmem_inode_unacct_blocks(inode, 1);
2471 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2473 struct vm_area_struct *dst_vma,
2474 unsigned long dst_addr,
2475 unsigned long src_addr,
2476 struct page **pagep)
2478 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2479 dst_addr, src_addr, false, pagep);
2482 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2484 struct vm_area_struct *dst_vma,
2485 unsigned long dst_addr)
2487 struct page *page = NULL;
2489 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2490 dst_addr, 0, true, &page);
2494 static const struct inode_operations shmem_symlink_inode_operations;
2495 static const struct inode_operations shmem_short_symlink_operations;
2497 #ifdef CONFIG_TMPFS_XATTR
2498 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2500 #define shmem_initxattrs NULL
2504 shmem_write_begin(struct file *file, struct address_space *mapping,
2505 loff_t pos, unsigned len, unsigned flags,
2506 struct page **pagep, void **fsdata)
2508 struct inode *inode = mapping->host;
2509 struct shmem_inode_info *info = SHMEM_I(inode);
2510 pgoff_t index = pos >> PAGE_SHIFT;
2512 /* i_mutex is held by caller */
2513 if (unlikely(info->seals & (F_SEAL_GROW |
2514 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2515 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2517 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2521 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2525 shmem_write_end(struct file *file, struct address_space *mapping,
2526 loff_t pos, unsigned len, unsigned copied,
2527 struct page *page, void *fsdata)
2529 struct inode *inode = mapping->host;
2531 if (pos + copied > inode->i_size)
2532 i_size_write(inode, pos + copied);
2534 if (!PageUptodate(page)) {
2535 struct page *head = compound_head(page);
2536 if (PageTransCompound(page)) {
2539 for (i = 0; i < HPAGE_PMD_NR; i++) {
2540 if (head + i == page)
2542 clear_highpage(head + i);
2543 flush_dcache_page(head + i);
2546 if (copied < PAGE_SIZE) {
2547 unsigned from = pos & (PAGE_SIZE - 1);
2548 zero_user_segments(page, 0, from,
2549 from + copied, PAGE_SIZE);
2551 SetPageUptodate(head);
2553 set_page_dirty(page);
2560 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2562 struct file *file = iocb->ki_filp;
2563 struct inode *inode = file_inode(file);
2564 struct address_space *mapping = inode->i_mapping;
2566 unsigned long offset;
2567 enum sgp_type sgp = SGP_READ;
2570 loff_t *ppos = &iocb->ki_pos;
2573 * Might this read be for a stacking filesystem? Then when reading
2574 * holes of a sparse file, we actually need to allocate those pages,
2575 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2577 if (!iter_is_iovec(to))
2580 index = *ppos >> PAGE_SHIFT;
2581 offset = *ppos & ~PAGE_MASK;
2584 struct page *page = NULL;
2586 unsigned long nr, ret;
2587 loff_t i_size = i_size_read(inode);
2589 end_index = i_size >> PAGE_SHIFT;
2590 if (index > end_index)
2592 if (index == end_index) {
2593 nr = i_size & ~PAGE_MASK;
2598 error = shmem_getpage(inode, index, &page, sgp);
2600 if (error == -EINVAL)
2605 if (sgp == SGP_CACHE)
2606 set_page_dirty(page);
2611 * We must evaluate after, since reads (unlike writes)
2612 * are called without i_mutex protection against truncate
2615 i_size = i_size_read(inode);
2616 end_index = i_size >> PAGE_SHIFT;
2617 if (index == end_index) {
2618 nr = i_size & ~PAGE_MASK;
2629 * If users can be writing to this page using arbitrary
2630 * virtual addresses, take care about potential aliasing
2631 * before reading the page on the kernel side.
2633 if (mapping_writably_mapped(mapping))
2634 flush_dcache_page(page);
2636 * Mark the page accessed if we read the beginning.
2639 mark_page_accessed(page);
2641 page = ZERO_PAGE(0);
2646 * Ok, we have the page, and it's up-to-date, so
2647 * now we can copy it to user space...
2649 ret = copy_page_to_iter(page, offset, nr, to);
2652 index += offset >> PAGE_SHIFT;
2653 offset &= ~PAGE_MASK;
2656 if (!iov_iter_count(to))
2665 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2666 file_accessed(file);
2667 return retval ? retval : error;
2671 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2673 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2674 pgoff_t index, pgoff_t end, int whence)
2677 struct pagevec pvec;
2678 pgoff_t indices[PAGEVEC_SIZE];
2682 pagevec_init(&pvec);
2683 pvec.nr = 1; /* start small: we may be there already */
2685 pvec.nr = find_get_entries(mapping, index,
2686 pvec.nr, pvec.pages, indices);
2688 if (whence == SEEK_DATA)
2692 for (i = 0; i < pvec.nr; i++, index++) {
2693 if (index < indices[i]) {
2694 if (whence == SEEK_HOLE) {
2700 page = pvec.pages[i];
2701 if (page && !xa_is_value(page)) {
2702 if (!PageUptodate(page))
2706 (page && whence == SEEK_DATA) ||
2707 (!page && whence == SEEK_HOLE)) {
2712 pagevec_remove_exceptionals(&pvec);
2713 pagevec_release(&pvec);
2714 pvec.nr = PAGEVEC_SIZE;
2720 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2722 struct address_space *mapping = file->f_mapping;
2723 struct inode *inode = mapping->host;
2727 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2728 return generic_file_llseek_size(file, offset, whence,
2729 MAX_LFS_FILESIZE, i_size_read(inode));
2731 /* We're holding i_mutex so we can access i_size directly */
2733 if (offset < 0 || offset >= inode->i_size)
2736 start = offset >> PAGE_SHIFT;
2737 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2738 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2739 new_offset <<= PAGE_SHIFT;
2740 if (new_offset > offset) {
2741 if (new_offset < inode->i_size)
2742 offset = new_offset;
2743 else if (whence == SEEK_DATA)
2746 offset = inode->i_size;
2751 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2752 inode_unlock(inode);
2756 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2759 struct inode *inode = file_inode(file);
2760 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2761 struct shmem_inode_info *info = SHMEM_I(inode);
2762 struct shmem_falloc shmem_falloc;
2763 pgoff_t start, index, end;
2766 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2771 if (mode & FALLOC_FL_PUNCH_HOLE) {
2772 struct address_space *mapping = file->f_mapping;
2773 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2774 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2775 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2777 /* protected by i_mutex */
2778 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2783 shmem_falloc.waitq = &shmem_falloc_waitq;
2784 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2785 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2786 spin_lock(&inode->i_lock);
2787 inode->i_private = &shmem_falloc;
2788 spin_unlock(&inode->i_lock);
2790 if ((u64)unmap_end > (u64)unmap_start)
2791 unmap_mapping_range(mapping, unmap_start,
2792 1 + unmap_end - unmap_start, 0);
2793 shmem_truncate_range(inode, offset, offset + len - 1);
2794 /* No need to unmap again: hole-punching leaves COWed pages */
2796 spin_lock(&inode->i_lock);
2797 inode->i_private = NULL;
2798 wake_up_all(&shmem_falloc_waitq);
2799 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2800 spin_unlock(&inode->i_lock);
2805 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2806 error = inode_newsize_ok(inode, offset + len);
2810 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2815 start = offset >> PAGE_SHIFT;
2816 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2817 /* Try to avoid a swapstorm if len is impossible to satisfy */
2818 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2823 shmem_falloc.waitq = NULL;
2824 shmem_falloc.start = start;
2825 shmem_falloc.next = start;
2826 shmem_falloc.nr_falloced = 0;
2827 shmem_falloc.nr_unswapped = 0;
2828 spin_lock(&inode->i_lock);
2829 inode->i_private = &shmem_falloc;
2830 spin_unlock(&inode->i_lock);
2832 for (index = start; index < end; index++) {
2836 * Good, the fallocate(2) manpage permits EINTR: we may have
2837 * been interrupted because we are using up too much memory.
2839 if (signal_pending(current))
2841 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2844 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2846 /* Remove the !PageUptodate pages we added */
2847 if (index > start) {
2848 shmem_undo_range(inode,
2849 (loff_t)start << PAGE_SHIFT,
2850 ((loff_t)index << PAGE_SHIFT) - 1, true);
2856 * Inform shmem_writepage() how far we have reached.
2857 * No need for lock or barrier: we have the page lock.
2859 shmem_falloc.next++;
2860 if (!PageUptodate(page))
2861 shmem_falloc.nr_falloced++;
2864 * If !PageUptodate, leave it that way so that freeable pages
2865 * can be recognized if we need to rollback on error later.
2866 * But set_page_dirty so that memory pressure will swap rather
2867 * than free the pages we are allocating (and SGP_CACHE pages
2868 * might still be clean: we now need to mark those dirty too).
2870 set_page_dirty(page);
2876 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2877 i_size_write(inode, offset + len);
2878 inode->i_ctime = current_time(inode);
2880 spin_lock(&inode->i_lock);
2881 inode->i_private = NULL;
2882 spin_unlock(&inode->i_lock);
2884 inode_unlock(inode);
2888 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2890 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2892 buf->f_type = TMPFS_MAGIC;
2893 buf->f_bsize = PAGE_SIZE;
2894 buf->f_namelen = NAME_MAX;
2895 if (sbinfo->max_blocks) {
2896 buf->f_blocks = sbinfo->max_blocks;
2898 buf->f_bfree = sbinfo->max_blocks -
2899 percpu_counter_sum(&sbinfo->used_blocks);
2901 if (sbinfo->max_inodes) {
2902 buf->f_files = sbinfo->max_inodes;
2903 buf->f_ffree = sbinfo->free_inodes;
2905 /* else leave those fields 0 like simple_statfs */
2910 * File creation. Allocate an inode, and we're done..
2913 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2914 struct dentry *dentry, umode_t mode, dev_t dev)
2916 struct inode *inode;
2917 int error = -ENOSPC;
2919 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2921 error = simple_acl_create(dir, inode);
2924 error = security_inode_init_security(inode, dir,
2926 shmem_initxattrs, NULL);
2927 if (error && error != -EOPNOTSUPP)
2931 dir->i_size += BOGO_DIRENT_SIZE;
2932 dir->i_ctime = dir->i_mtime = current_time(dir);
2933 d_instantiate(dentry, inode);
2934 dget(dentry); /* Extra count - pin the dentry in core */
2943 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2944 struct dentry *dentry, umode_t mode)
2946 struct inode *inode;
2947 int error = -ENOSPC;
2949 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2951 error = security_inode_init_security(inode, dir,
2953 shmem_initxattrs, NULL);
2954 if (error && error != -EOPNOTSUPP)
2956 error = simple_acl_create(dir, inode);
2959 d_tmpfile(dentry, inode);
2967 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2968 struct dentry *dentry, umode_t mode)
2972 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2973 mode | S_IFDIR, 0)))
2979 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2980 struct dentry *dentry, umode_t mode, bool excl)
2982 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2988 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2990 struct inode *inode = d_inode(old_dentry);
2994 * No ordinary (disk based) filesystem counts links as inodes;
2995 * but each new link needs a new dentry, pinning lowmem, and
2996 * tmpfs dentries cannot be pruned until they are unlinked.
2997 * But if an O_TMPFILE file is linked into the tmpfs, the
2998 * first link must skip that, to get the accounting right.
3000 if (inode->i_nlink) {
3001 ret = shmem_reserve_inode(inode->i_sb, NULL);
3006 dir->i_size += BOGO_DIRENT_SIZE;
3007 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3009 ihold(inode); /* New dentry reference */
3010 dget(dentry); /* Extra pinning count for the created dentry */
3011 d_instantiate(dentry, inode);
3016 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3018 struct inode *inode = d_inode(dentry);
3020 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3021 shmem_free_inode(inode->i_sb);
3023 dir->i_size -= BOGO_DIRENT_SIZE;
3024 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3026 dput(dentry); /* Undo the count from "create" - this does all the work */
3030 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3032 if (!simple_empty(dentry))
3035 drop_nlink(d_inode(dentry));
3037 return shmem_unlink(dir, dentry);
3040 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3042 bool old_is_dir = d_is_dir(old_dentry);
3043 bool new_is_dir = d_is_dir(new_dentry);
3045 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3047 drop_nlink(old_dir);
3050 drop_nlink(new_dir);
3054 old_dir->i_ctime = old_dir->i_mtime =
3055 new_dir->i_ctime = new_dir->i_mtime =
3056 d_inode(old_dentry)->i_ctime =
3057 d_inode(new_dentry)->i_ctime = current_time(old_dir);
3062 static int shmem_whiteout(struct user_namespace *mnt_userns,
3063 struct inode *old_dir, struct dentry *old_dentry)
3065 struct dentry *whiteout;
3068 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3072 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
3073 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3079 * Cheat and hash the whiteout while the old dentry is still in
3080 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3082 * d_lookup() will consistently find one of them at this point,
3083 * not sure which one, but that isn't even important.
3090 * The VFS layer already does all the dentry stuff for rename,
3091 * we just have to decrement the usage count for the target if
3092 * it exists so that the VFS layer correctly free's it when it
3095 static int shmem_rename2(struct user_namespace *mnt_userns,
3096 struct inode *old_dir, struct dentry *old_dentry,
3097 struct inode *new_dir, struct dentry *new_dentry,
3100 struct inode *inode = d_inode(old_dentry);
3101 int they_are_dirs = S_ISDIR(inode->i_mode);
3103 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3106 if (flags & RENAME_EXCHANGE)
3107 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3109 if (!simple_empty(new_dentry))
3112 if (flags & RENAME_WHITEOUT) {
3115 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3120 if (d_really_is_positive(new_dentry)) {
3121 (void) shmem_unlink(new_dir, new_dentry);
3122 if (they_are_dirs) {
3123 drop_nlink(d_inode(new_dentry));
3124 drop_nlink(old_dir);
3126 } else if (they_are_dirs) {
3127 drop_nlink(old_dir);
3131 old_dir->i_size -= BOGO_DIRENT_SIZE;
3132 new_dir->i_size += BOGO_DIRENT_SIZE;
3133 old_dir->i_ctime = old_dir->i_mtime =
3134 new_dir->i_ctime = new_dir->i_mtime =
3135 inode->i_ctime = current_time(old_dir);
3139 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3140 struct dentry *dentry, const char *symname)
3144 struct inode *inode;
3147 len = strlen(symname) + 1;
3148 if (len > PAGE_SIZE)
3149 return -ENAMETOOLONG;
3151 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3156 error = security_inode_init_security(inode, dir, &dentry->d_name,
3157 shmem_initxattrs, NULL);
3158 if (error && error != -EOPNOTSUPP) {
3163 inode->i_size = len-1;
3164 if (len <= SHORT_SYMLINK_LEN) {
3165 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3166 if (!inode->i_link) {
3170 inode->i_op = &shmem_short_symlink_operations;
3172 inode_nohighmem(inode);
3173 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3178 inode->i_mapping->a_ops = &shmem_aops;
3179 inode->i_op = &shmem_symlink_inode_operations;
3180 memcpy(page_address(page), symname, len);
3181 SetPageUptodate(page);
3182 set_page_dirty(page);
3186 dir->i_size += BOGO_DIRENT_SIZE;
3187 dir->i_ctime = dir->i_mtime = current_time(dir);
3188 d_instantiate(dentry, inode);
3193 static void shmem_put_link(void *arg)
3195 mark_page_accessed(arg);
3199 static const char *shmem_get_link(struct dentry *dentry,
3200 struct inode *inode,
3201 struct delayed_call *done)
3203 struct page *page = NULL;
3206 page = find_get_page(inode->i_mapping, 0);
3208 return ERR_PTR(-ECHILD);
3209 if (!PageUptodate(page)) {
3211 return ERR_PTR(-ECHILD);
3214 error = shmem_getpage(inode, 0, &page, SGP_READ);
3216 return ERR_PTR(error);
3219 set_delayed_call(done, shmem_put_link, page);
3220 return page_address(page);
3223 #ifdef CONFIG_TMPFS_XATTR
3225 * Superblocks without xattr inode operations may get some security.* xattr
3226 * support from the LSM "for free". As soon as we have any other xattrs
3227 * like ACLs, we also need to implement the security.* handlers at
3228 * filesystem level, though.
3232 * Callback for security_inode_init_security() for acquiring xattrs.
3234 static int shmem_initxattrs(struct inode *inode,
3235 const struct xattr *xattr_array,
3238 struct shmem_inode_info *info = SHMEM_I(inode);
3239 const struct xattr *xattr;
3240 struct simple_xattr *new_xattr;
3243 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3244 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3248 len = strlen(xattr->name) + 1;
3249 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3251 if (!new_xattr->name) {
3256 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3257 XATTR_SECURITY_PREFIX_LEN);
3258 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3261 simple_xattr_list_add(&info->xattrs, new_xattr);
3267 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3268 struct dentry *unused, struct inode *inode,
3269 const char *name, void *buffer, size_t size)
3271 struct shmem_inode_info *info = SHMEM_I(inode);
3273 name = xattr_full_name(handler, name);
3274 return simple_xattr_get(&info->xattrs, name, buffer, size);
3277 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3278 struct user_namespace *mnt_userns,
3279 struct dentry *unused, struct inode *inode,
3280 const char *name, const void *value,
3281 size_t size, int flags)
3283 struct shmem_inode_info *info = SHMEM_I(inode);
3285 name = xattr_full_name(handler, name);
3286 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3289 static const struct xattr_handler shmem_security_xattr_handler = {
3290 .prefix = XATTR_SECURITY_PREFIX,
3291 .get = shmem_xattr_handler_get,
3292 .set = shmem_xattr_handler_set,
3295 static const struct xattr_handler shmem_trusted_xattr_handler = {
3296 .prefix = XATTR_TRUSTED_PREFIX,
3297 .get = shmem_xattr_handler_get,
3298 .set = shmem_xattr_handler_set,
3301 static const struct xattr_handler *shmem_xattr_handlers[] = {
3302 #ifdef CONFIG_TMPFS_POSIX_ACL
3303 &posix_acl_access_xattr_handler,
3304 &posix_acl_default_xattr_handler,
3306 &shmem_security_xattr_handler,
3307 &shmem_trusted_xattr_handler,
3311 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3313 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3314 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3316 #endif /* CONFIG_TMPFS_XATTR */
3318 static const struct inode_operations shmem_short_symlink_operations = {
3319 .get_link = simple_get_link,
3320 #ifdef CONFIG_TMPFS_XATTR
3321 .listxattr = shmem_listxattr,
3325 static const struct inode_operations shmem_symlink_inode_operations = {
3326 .get_link = shmem_get_link,
3327 #ifdef CONFIG_TMPFS_XATTR
3328 .listxattr = shmem_listxattr,
3332 static struct dentry *shmem_get_parent(struct dentry *child)
3334 return ERR_PTR(-ESTALE);
3337 static int shmem_match(struct inode *ino, void *vfh)
3341 inum = (inum << 32) | fh[1];
3342 return ino->i_ino == inum && fh[0] == ino->i_generation;
3345 /* Find any alias of inode, but prefer a hashed alias */
3346 static struct dentry *shmem_find_alias(struct inode *inode)
3348 struct dentry *alias = d_find_alias(inode);
3350 return alias ?: d_find_any_alias(inode);
3354 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3355 struct fid *fid, int fh_len, int fh_type)
3357 struct inode *inode;
3358 struct dentry *dentry = NULL;
3365 inum = (inum << 32) | fid->raw[1];
3367 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3368 shmem_match, fid->raw);
3370 dentry = shmem_find_alias(inode);
3377 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3378 struct inode *parent)
3382 return FILEID_INVALID;
3385 if (inode_unhashed(inode)) {
3386 /* Unfortunately insert_inode_hash is not idempotent,
3387 * so as we hash inodes here rather than at creation
3388 * time, we need a lock to ensure we only try
3391 static DEFINE_SPINLOCK(lock);
3393 if (inode_unhashed(inode))
3394 __insert_inode_hash(inode,
3395 inode->i_ino + inode->i_generation);
3399 fh[0] = inode->i_generation;
3400 fh[1] = inode->i_ino;
3401 fh[2] = ((__u64)inode->i_ino) >> 32;
3407 static const struct export_operations shmem_export_ops = {
3408 .get_parent = shmem_get_parent,
3409 .encode_fh = shmem_encode_fh,
3410 .fh_to_dentry = shmem_fh_to_dentry,
3426 static const struct constant_table shmem_param_enums_huge[] = {
3427 {"never", SHMEM_HUGE_NEVER },
3428 {"always", SHMEM_HUGE_ALWAYS },
3429 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3430 {"advise", SHMEM_HUGE_ADVISE },
3434 const struct fs_parameter_spec shmem_fs_parameters[] = {
3435 fsparam_u32 ("gid", Opt_gid),
3436 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3437 fsparam_u32oct("mode", Opt_mode),
3438 fsparam_string("mpol", Opt_mpol),
3439 fsparam_string("nr_blocks", Opt_nr_blocks),
3440 fsparam_string("nr_inodes", Opt_nr_inodes),
3441 fsparam_string("size", Opt_size),
3442 fsparam_u32 ("uid", Opt_uid),
3443 fsparam_flag ("inode32", Opt_inode32),
3444 fsparam_flag ("inode64", Opt_inode64),
3448 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3450 struct shmem_options *ctx = fc->fs_private;
3451 struct fs_parse_result result;
3452 unsigned long long size;
3456 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3462 size = memparse(param->string, &rest);
3464 size <<= PAGE_SHIFT;
3465 size *= totalram_pages();
3471 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3472 ctx->seen |= SHMEM_SEEN_BLOCKS;
3475 ctx->blocks = memparse(param->string, &rest);
3478 ctx->seen |= SHMEM_SEEN_BLOCKS;
3481 ctx->inodes = memparse(param->string, &rest);
3484 ctx->seen |= SHMEM_SEEN_INODES;
3487 ctx->mode = result.uint_32 & 07777;
3490 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3491 if (!uid_valid(ctx->uid))
3495 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3496 if (!gid_valid(ctx->gid))
3500 ctx->huge = result.uint_32;
3501 if (ctx->huge != SHMEM_HUGE_NEVER &&
3502 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3503 has_transparent_hugepage()))
3504 goto unsupported_parameter;
3505 ctx->seen |= SHMEM_SEEN_HUGE;
3508 if (IS_ENABLED(CONFIG_NUMA)) {
3509 mpol_put(ctx->mpol);
3511 if (mpol_parse_str(param->string, &ctx->mpol))
3515 goto unsupported_parameter;
3517 ctx->full_inums = false;
3518 ctx->seen |= SHMEM_SEEN_INUMS;
3521 if (sizeof(ino_t) < 8) {
3523 "Cannot use inode64 with <64bit inums in kernel\n");
3525 ctx->full_inums = true;
3526 ctx->seen |= SHMEM_SEEN_INUMS;
3531 unsupported_parameter:
3532 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3534 return invalfc(fc, "Bad value for '%s'", param->key);
3537 static int shmem_parse_options(struct fs_context *fc, void *data)
3539 char *options = data;
3542 int err = security_sb_eat_lsm_opts(options, &fc->security);
3547 while (options != NULL) {
3548 char *this_char = options;
3551 * NUL-terminate this option: unfortunately,
3552 * mount options form a comma-separated list,
3553 * but mpol's nodelist may also contain commas.
3555 options = strchr(options, ',');
3556 if (options == NULL)
3559 if (!isdigit(*options)) {
3565 char *value = strchr(this_char,'=');
3571 len = strlen(value);
3573 err = vfs_parse_fs_string(fc, this_char, value, len);
3582 * Reconfigure a shmem filesystem.
3584 * Note that we disallow change from limited->unlimited blocks/inodes while any
3585 * are in use; but we must separately disallow unlimited->limited, because in
3586 * that case we have no record of how much is already in use.
3588 static int shmem_reconfigure(struct fs_context *fc)
3590 struct shmem_options *ctx = fc->fs_private;
3591 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3592 unsigned long inodes;
3595 spin_lock(&sbinfo->stat_lock);
3596 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3597 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3598 if (!sbinfo->max_blocks) {
3599 err = "Cannot retroactively limit size";
3602 if (percpu_counter_compare(&sbinfo->used_blocks,
3604 err = "Too small a size for current use";
3608 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3609 if (!sbinfo->max_inodes) {
3610 err = "Cannot retroactively limit inodes";
3613 if (ctx->inodes < inodes) {
3614 err = "Too few inodes for current use";
3619 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3620 sbinfo->next_ino > UINT_MAX) {
3621 err = "Current inum too high to switch to 32-bit inums";
3625 if (ctx->seen & SHMEM_SEEN_HUGE)
3626 sbinfo->huge = ctx->huge;
3627 if (ctx->seen & SHMEM_SEEN_INUMS)
3628 sbinfo->full_inums = ctx->full_inums;
3629 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3630 sbinfo->max_blocks = ctx->blocks;
3631 if (ctx->seen & SHMEM_SEEN_INODES) {
3632 sbinfo->max_inodes = ctx->inodes;
3633 sbinfo->free_inodes = ctx->inodes - inodes;
3637 * Preserve previous mempolicy unless mpol remount option was specified.
3640 mpol_put(sbinfo->mpol);
3641 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3644 spin_unlock(&sbinfo->stat_lock);
3647 spin_unlock(&sbinfo->stat_lock);
3648 return invalfc(fc, "%s", err);
3651 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3653 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3655 if (sbinfo->max_blocks != shmem_default_max_blocks())
3656 seq_printf(seq, ",size=%luk",
3657 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3658 if (sbinfo->max_inodes != shmem_default_max_inodes())
3659 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3660 if (sbinfo->mode != (0777 | S_ISVTX))
3661 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3662 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3663 seq_printf(seq, ",uid=%u",
3664 from_kuid_munged(&init_user_ns, sbinfo->uid));
3665 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3666 seq_printf(seq, ",gid=%u",
3667 from_kgid_munged(&init_user_ns, sbinfo->gid));
3670 * Showing inode{64,32} might be useful even if it's the system default,
3671 * since then people don't have to resort to checking both here and
3672 * /proc/config.gz to confirm 64-bit inums were successfully applied
3673 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3675 * We hide it when inode64 isn't the default and we are using 32-bit
3676 * inodes, since that probably just means the feature isn't even under
3681 * +-----------------+-----------------+
3682 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3683 * +------------------+-----------------+-----------------+
3684 * | full_inums=true | show | show |
3685 * | full_inums=false | show | hide |
3686 * +------------------+-----------------+-----------------+
3689 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3690 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3691 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3692 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3694 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3696 shmem_show_mpol(seq, sbinfo->mpol);
3700 #endif /* CONFIG_TMPFS */
3702 static void shmem_put_super(struct super_block *sb)
3704 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3706 free_percpu(sbinfo->ino_batch);
3707 percpu_counter_destroy(&sbinfo->used_blocks);
3708 mpol_put(sbinfo->mpol);
3710 sb->s_fs_info = NULL;
3713 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3715 struct shmem_options *ctx = fc->fs_private;
3716 struct inode *inode;
3717 struct shmem_sb_info *sbinfo;
3720 /* Round up to L1_CACHE_BYTES to resist false sharing */
3721 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3722 L1_CACHE_BYTES), GFP_KERNEL);
3726 sb->s_fs_info = sbinfo;
3730 * Per default we only allow half of the physical ram per
3731 * tmpfs instance, limiting inodes to one per page of lowmem;
3732 * but the internal instance is left unlimited.
3734 if (!(sb->s_flags & SB_KERNMOUNT)) {
3735 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3736 ctx->blocks = shmem_default_max_blocks();
3737 if (!(ctx->seen & SHMEM_SEEN_INODES))
3738 ctx->inodes = shmem_default_max_inodes();
3739 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3740 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3742 sb->s_flags |= SB_NOUSER;
3744 sb->s_export_op = &shmem_export_ops;
3745 sb->s_flags |= SB_NOSEC;
3747 sb->s_flags |= SB_NOUSER;
3749 sbinfo->max_blocks = ctx->blocks;
3750 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3751 if (sb->s_flags & SB_KERNMOUNT) {
3752 sbinfo->ino_batch = alloc_percpu(ino_t);
3753 if (!sbinfo->ino_batch)
3756 sbinfo->uid = ctx->uid;
3757 sbinfo->gid = ctx->gid;
3758 sbinfo->full_inums = ctx->full_inums;
3759 sbinfo->mode = ctx->mode;
3760 sbinfo->huge = ctx->huge;
3761 sbinfo->mpol = ctx->mpol;
3764 spin_lock_init(&sbinfo->stat_lock);
3765 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3767 spin_lock_init(&sbinfo->shrinklist_lock);
3768 INIT_LIST_HEAD(&sbinfo->shrinklist);
3770 sb->s_maxbytes = MAX_LFS_FILESIZE;
3771 sb->s_blocksize = PAGE_SIZE;
3772 sb->s_blocksize_bits = PAGE_SHIFT;
3773 sb->s_magic = TMPFS_MAGIC;
3774 sb->s_op = &shmem_ops;
3775 sb->s_time_gran = 1;
3776 #ifdef CONFIG_TMPFS_XATTR
3777 sb->s_xattr = shmem_xattr_handlers;
3779 #ifdef CONFIG_TMPFS_POSIX_ACL
3780 sb->s_flags |= SB_POSIXACL;
3782 uuid_gen(&sb->s_uuid);
3784 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3787 inode->i_uid = sbinfo->uid;
3788 inode->i_gid = sbinfo->gid;
3789 sb->s_root = d_make_root(inode);
3795 shmem_put_super(sb);
3799 static int shmem_get_tree(struct fs_context *fc)
3801 return get_tree_nodev(fc, shmem_fill_super);
3804 static void shmem_free_fc(struct fs_context *fc)
3806 struct shmem_options *ctx = fc->fs_private;
3809 mpol_put(ctx->mpol);
3814 static const struct fs_context_operations shmem_fs_context_ops = {
3815 .free = shmem_free_fc,
3816 .get_tree = shmem_get_tree,
3818 .parse_monolithic = shmem_parse_options,
3819 .parse_param = shmem_parse_one,
3820 .reconfigure = shmem_reconfigure,
3824 static struct kmem_cache *shmem_inode_cachep;
3826 static struct inode *shmem_alloc_inode(struct super_block *sb)
3828 struct shmem_inode_info *info;
3829 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3832 return &info->vfs_inode;
3835 static void shmem_free_in_core_inode(struct inode *inode)
3837 if (S_ISLNK(inode->i_mode))
3838 kfree(inode->i_link);
3839 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3842 static void shmem_destroy_inode(struct inode *inode)
3844 if (S_ISREG(inode->i_mode))
3845 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3848 static void shmem_init_inode(void *foo)
3850 struct shmem_inode_info *info = foo;
3851 inode_init_once(&info->vfs_inode);
3854 static void shmem_init_inodecache(void)
3856 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3857 sizeof(struct shmem_inode_info),
3858 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3861 static void shmem_destroy_inodecache(void)
3863 kmem_cache_destroy(shmem_inode_cachep);
3866 const struct address_space_operations shmem_aops = {
3867 .writepage = shmem_writepage,
3868 .set_page_dirty = __set_page_dirty_no_writeback,
3870 .write_begin = shmem_write_begin,
3871 .write_end = shmem_write_end,
3873 #ifdef CONFIG_MIGRATION
3874 .migratepage = migrate_page,
3876 .error_remove_page = generic_error_remove_page,
3878 EXPORT_SYMBOL(shmem_aops);
3880 static const struct file_operations shmem_file_operations = {
3882 .get_unmapped_area = shmem_get_unmapped_area,
3884 .llseek = shmem_file_llseek,
3885 .read_iter = shmem_file_read_iter,
3886 .write_iter = generic_file_write_iter,
3887 .fsync = noop_fsync,
3888 .splice_read = generic_file_splice_read,
3889 .splice_write = iter_file_splice_write,
3890 .fallocate = shmem_fallocate,
3894 static const struct inode_operations shmem_inode_operations = {
3895 .getattr = shmem_getattr,
3896 .setattr = shmem_setattr,
3897 #ifdef CONFIG_TMPFS_XATTR
3898 .listxattr = shmem_listxattr,
3899 .set_acl = simple_set_acl,
3903 static const struct inode_operations shmem_dir_inode_operations = {
3905 .create = shmem_create,
3906 .lookup = simple_lookup,
3908 .unlink = shmem_unlink,
3909 .symlink = shmem_symlink,
3910 .mkdir = shmem_mkdir,
3911 .rmdir = shmem_rmdir,
3912 .mknod = shmem_mknod,
3913 .rename = shmem_rename2,
3914 .tmpfile = shmem_tmpfile,
3916 #ifdef CONFIG_TMPFS_XATTR
3917 .listxattr = shmem_listxattr,
3919 #ifdef CONFIG_TMPFS_POSIX_ACL
3920 .setattr = shmem_setattr,
3921 .set_acl = simple_set_acl,
3925 static const struct inode_operations shmem_special_inode_operations = {
3926 #ifdef CONFIG_TMPFS_XATTR
3927 .listxattr = shmem_listxattr,
3929 #ifdef CONFIG_TMPFS_POSIX_ACL
3930 .setattr = shmem_setattr,
3931 .set_acl = simple_set_acl,
3935 static const struct super_operations shmem_ops = {
3936 .alloc_inode = shmem_alloc_inode,
3937 .free_inode = shmem_free_in_core_inode,
3938 .destroy_inode = shmem_destroy_inode,
3940 .statfs = shmem_statfs,
3941 .show_options = shmem_show_options,
3943 .evict_inode = shmem_evict_inode,
3944 .drop_inode = generic_delete_inode,
3945 .put_super = shmem_put_super,
3946 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3947 .nr_cached_objects = shmem_unused_huge_count,
3948 .free_cached_objects = shmem_unused_huge_scan,
3952 static const struct vm_operations_struct shmem_vm_ops = {
3953 .fault = shmem_fault,
3954 .map_pages = filemap_map_pages,
3956 .set_policy = shmem_set_policy,
3957 .get_policy = shmem_get_policy,
3961 int shmem_init_fs_context(struct fs_context *fc)
3963 struct shmem_options *ctx;
3965 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3969 ctx->mode = 0777 | S_ISVTX;
3970 ctx->uid = current_fsuid();
3971 ctx->gid = current_fsgid();
3973 fc->fs_private = ctx;
3974 fc->ops = &shmem_fs_context_ops;
3978 static struct file_system_type shmem_fs_type = {
3979 .owner = THIS_MODULE,
3981 .init_fs_context = shmem_init_fs_context,
3983 .parameters = shmem_fs_parameters,
3985 .kill_sb = kill_litter_super,
3986 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
3989 int __init shmem_init(void)
3993 shmem_init_inodecache();
3995 error = register_filesystem(&shmem_fs_type);
3997 pr_err("Could not register tmpfs\n");
4001 shm_mnt = kern_mount(&shmem_fs_type);
4002 if (IS_ERR(shm_mnt)) {
4003 error = PTR_ERR(shm_mnt);
4004 pr_err("Could not kern_mount tmpfs\n");
4008 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4009 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4010 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4012 shmem_huge = 0; /* just in case it was patched */
4017 unregister_filesystem(&shmem_fs_type);
4019 shmem_destroy_inodecache();
4020 shm_mnt = ERR_PTR(error);
4024 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4025 static ssize_t shmem_enabled_show(struct kobject *kobj,
4026 struct kobj_attribute *attr, char *buf)
4028 static const int values[] = {
4030 SHMEM_HUGE_WITHIN_SIZE,
4039 for (i = 0; i < ARRAY_SIZE(values); i++) {
4040 len += sysfs_emit_at(buf, len,
4041 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4043 shmem_format_huge(values[i]));
4046 len += sysfs_emit_at(buf, len, "\n");
4051 static ssize_t shmem_enabled_store(struct kobject *kobj,
4052 struct kobj_attribute *attr, const char *buf, size_t count)
4057 if (count + 1 > sizeof(tmp))
4059 memcpy(tmp, buf, count);
4061 if (count && tmp[count - 1] == '\n')
4062 tmp[count - 1] = '\0';
4064 huge = shmem_parse_huge(tmp);
4065 if (huge == -EINVAL)
4067 if (!has_transparent_hugepage() &&
4068 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4072 if (shmem_huge > SHMEM_HUGE_DENY)
4073 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4077 struct kobj_attribute shmem_enabled_attr =
4078 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4079 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4081 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4082 bool shmem_huge_enabled(struct vm_area_struct *vma)
4084 struct inode *inode = file_inode(vma->vm_file);
4085 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4089 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
4090 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
4092 if (shmem_huge == SHMEM_HUGE_FORCE)
4094 if (shmem_huge == SHMEM_HUGE_DENY)
4096 switch (sbinfo->huge) {
4097 case SHMEM_HUGE_NEVER:
4099 case SHMEM_HUGE_ALWAYS:
4101 case SHMEM_HUGE_WITHIN_SIZE:
4102 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4103 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4104 if (i_size >= HPAGE_PMD_SIZE &&
4105 i_size >> PAGE_SHIFT >= off)
4108 case SHMEM_HUGE_ADVISE:
4109 /* TODO: implement fadvise() hints */
4110 return (vma->vm_flags & VM_HUGEPAGE);
4116 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4118 #else /* !CONFIG_SHMEM */
4121 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4123 * This is intended for small system where the benefits of the full
4124 * shmem code (swap-backed and resource-limited) are outweighed by
4125 * their complexity. On systems without swap this code should be
4126 * effectively equivalent, but much lighter weight.
4129 static struct file_system_type shmem_fs_type = {
4131 .init_fs_context = ramfs_init_fs_context,
4132 .parameters = ramfs_fs_parameters,
4133 .kill_sb = kill_litter_super,
4134 .fs_flags = FS_USERNS_MOUNT,
4137 int __init shmem_init(void)
4139 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4141 shm_mnt = kern_mount(&shmem_fs_type);
4142 BUG_ON(IS_ERR(shm_mnt));
4147 int shmem_unuse(unsigned int type, bool frontswap,
4148 unsigned long *fs_pages_to_unuse)
4153 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4158 void shmem_unlock_mapping(struct address_space *mapping)
4163 unsigned long shmem_get_unmapped_area(struct file *file,
4164 unsigned long addr, unsigned long len,
4165 unsigned long pgoff, unsigned long flags)
4167 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4171 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4173 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4175 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4177 #define shmem_vm_ops generic_file_vm_ops
4178 #define shmem_file_operations ramfs_file_operations
4179 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4180 #define shmem_acct_size(flags, size) 0
4181 #define shmem_unacct_size(flags, size) do {} while (0)
4183 #endif /* CONFIG_SHMEM */
4187 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4188 unsigned long flags, unsigned int i_flags)
4190 struct inode *inode;
4194 return ERR_CAST(mnt);
4196 if (size < 0 || size > MAX_LFS_FILESIZE)
4197 return ERR_PTR(-EINVAL);
4199 if (shmem_acct_size(flags, size))
4200 return ERR_PTR(-ENOMEM);
4202 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4204 if (unlikely(!inode)) {
4205 shmem_unacct_size(flags, size);
4206 return ERR_PTR(-ENOSPC);
4208 inode->i_flags |= i_flags;
4209 inode->i_size = size;
4210 clear_nlink(inode); /* It is unlinked */
4211 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4213 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4214 &shmem_file_operations);
4221 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4222 * kernel internal. There will be NO LSM permission checks against the
4223 * underlying inode. So users of this interface must do LSM checks at a
4224 * higher layer. The users are the big_key and shm implementations. LSM
4225 * checks are provided at the key or shm level rather than the inode.
4226 * @name: name for dentry (to be seen in /proc/<pid>/maps
4227 * @size: size to be set for the file
4228 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4230 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4232 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4236 * shmem_file_setup - get an unlinked file living in tmpfs
4237 * @name: name for dentry (to be seen in /proc/<pid>/maps
4238 * @size: size to be set for the file
4239 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4241 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4243 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4245 EXPORT_SYMBOL_GPL(shmem_file_setup);
4248 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4249 * @mnt: the tmpfs mount where the file will be created
4250 * @name: name for dentry (to be seen in /proc/<pid>/maps
4251 * @size: size to be set for the file
4252 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4254 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4255 loff_t size, unsigned long flags)
4257 return __shmem_file_setup(mnt, name, size, flags, 0);
4259 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4262 * shmem_zero_setup - setup a shared anonymous mapping
4263 * @vma: the vma to be mmapped is prepared by do_mmap
4265 int shmem_zero_setup(struct vm_area_struct *vma)
4268 loff_t size = vma->vm_end - vma->vm_start;
4271 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4272 * between XFS directory reading and selinux: since this file is only
4273 * accessible to the user through its mapping, use S_PRIVATE flag to
4274 * bypass file security, in the same way as shmem_kernel_file_setup().
4276 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4278 return PTR_ERR(file);
4282 vma->vm_file = file;
4283 vma->vm_ops = &shmem_vm_ops;
4285 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4286 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4287 (vma->vm_end & HPAGE_PMD_MASK)) {
4288 khugepaged_enter(vma, vma->vm_flags);
4295 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4296 * @mapping: the page's address_space
4297 * @index: the page index
4298 * @gfp: the page allocator flags to use if allocating
4300 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4301 * with any new page allocations done using the specified allocation flags.
4302 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4303 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4304 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4306 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4307 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4309 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4310 pgoff_t index, gfp_t gfp)
4313 struct inode *inode = mapping->host;
4317 BUG_ON(!shmem_mapping(mapping));
4318 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4319 gfp, NULL, NULL, NULL);
4321 page = ERR_PTR(error);
4327 * The tiny !SHMEM case uses ramfs without swap
4329 return read_cache_page_gfp(mapping, index, gfp);
4332 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);