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 static 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;
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);
548 /* Check if there's anything to gain */
549 if (round_up(inode->i_size, PAGE_SIZE) ==
550 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
551 list_move(&info->shrinklist, &to_remove);
555 list_move(&info->shrinklist, &list);
557 sbinfo->shrinklist_len--;
561 spin_unlock(&sbinfo->shrinklist_lock);
563 list_for_each_safe(pos, next, &to_remove) {
564 info = list_entry(pos, struct shmem_inode_info, shrinklist);
565 inode = &info->vfs_inode;
566 list_del_init(&info->shrinklist);
570 list_for_each_safe(pos, next, &list) {
573 info = list_entry(pos, struct shmem_inode_info, shrinklist);
574 inode = &info->vfs_inode;
576 if (nr_to_split && split >= nr_to_split)
579 page = find_get_page(inode->i_mapping,
580 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
584 /* No huge page at the end of the file: nothing to split */
585 if (!PageTransHuge(page)) {
591 * Move the inode on the list back to shrinklist if we failed
592 * to lock the page at this time.
594 * Waiting for the lock may lead to deadlock in the
597 if (!trylock_page(page)) {
602 ret = split_huge_page(page);
606 /* If split failed move the inode on the list back to shrinklist */
612 list_del_init(&info->shrinklist);
616 * Make sure the inode is either on the global list or deleted
617 * from any local list before iput() since it could be deleted
618 * in another thread once we put the inode (then the local list
621 spin_lock(&sbinfo->shrinklist_lock);
622 list_move(&info->shrinklist, &sbinfo->shrinklist);
623 sbinfo->shrinklist_len++;
624 spin_unlock(&sbinfo->shrinklist_lock);
632 static long shmem_unused_huge_scan(struct super_block *sb,
633 struct shrink_control *sc)
635 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
637 if (!READ_ONCE(sbinfo->shrinklist_len))
640 return shmem_unused_huge_shrink(sbinfo, sc, 0);
643 static long shmem_unused_huge_count(struct super_block *sb,
644 struct shrink_control *sc)
646 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
647 return READ_ONCE(sbinfo->shrinklist_len);
649 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
651 #define shmem_huge SHMEM_HUGE_DENY
653 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
654 struct shrink_control *sc, unsigned long nr_to_split)
658 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
660 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
662 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
663 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
664 shmem_huge != SHMEM_HUGE_DENY)
670 * Like add_to_page_cache_locked, but error if expected item has gone.
672 static int shmem_add_to_page_cache(struct page *page,
673 struct address_space *mapping,
674 pgoff_t index, void *expected, gfp_t gfp,
675 struct mm_struct *charge_mm)
677 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
679 unsigned long nr = compound_nr(page);
682 VM_BUG_ON_PAGE(PageTail(page), page);
683 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
684 VM_BUG_ON_PAGE(!PageLocked(page), page);
685 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
686 VM_BUG_ON(expected && PageTransHuge(page));
688 page_ref_add(page, nr);
689 page->mapping = mapping;
692 if (!PageSwapCache(page)) {
693 error = mem_cgroup_charge(page, charge_mm, gfp);
695 if (PageTransHuge(page)) {
696 count_vm_event(THP_FILE_FALLBACK);
697 count_vm_event(THP_FILE_FALLBACK_CHARGE);
702 cgroup_throttle_swaprate(page, gfp);
707 entry = xas_find_conflict(&xas);
708 if (entry != expected)
709 xas_set_err(&xas, -EEXIST);
710 xas_create_range(&xas);
714 xas_store(&xas, page);
719 if (PageTransHuge(page)) {
720 count_vm_event(THP_FILE_ALLOC);
721 __inc_node_page_state(page, NR_SHMEM_THPS);
723 mapping->nrpages += nr;
724 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
725 __mod_lruvec_page_state(page, NR_SHMEM, nr);
727 xas_unlock_irq(&xas);
728 } while (xas_nomem(&xas, gfp));
730 if (xas_error(&xas)) {
731 error = xas_error(&xas);
737 page->mapping = NULL;
738 page_ref_sub(page, nr);
743 * Like delete_from_page_cache, but substitutes swap for page.
745 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
747 struct address_space *mapping = page->mapping;
750 VM_BUG_ON_PAGE(PageCompound(page), page);
752 xa_lock_irq(&mapping->i_pages);
753 error = shmem_replace_entry(mapping, page->index, page, radswap);
754 page->mapping = NULL;
756 __dec_lruvec_page_state(page, NR_FILE_PAGES);
757 __dec_lruvec_page_state(page, NR_SHMEM);
758 xa_unlock_irq(&mapping->i_pages);
764 * Remove swap entry from page cache, free the swap and its page cache.
766 static int shmem_free_swap(struct address_space *mapping,
767 pgoff_t index, void *radswap)
771 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
774 free_swap_and_cache(radix_to_swp_entry(radswap));
779 * Determine (in bytes) how many of the shmem object's pages mapped by the
780 * given offsets are swapped out.
782 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
783 * as long as the inode doesn't go away and racy results are not a problem.
785 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
786 pgoff_t start, pgoff_t end)
788 XA_STATE(xas, &mapping->i_pages, start);
790 unsigned long swapped = 0;
793 xas_for_each(&xas, page, end - 1) {
794 if (xas_retry(&xas, page))
796 if (xa_is_value(page))
799 if (need_resched()) {
807 return swapped << PAGE_SHIFT;
811 * Determine (in bytes) how many of the shmem object's pages mapped by the
812 * given vma is swapped out.
814 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
815 * as long as the inode doesn't go away and racy results are not a problem.
817 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
819 struct inode *inode = file_inode(vma->vm_file);
820 struct shmem_inode_info *info = SHMEM_I(inode);
821 struct address_space *mapping = inode->i_mapping;
822 unsigned long swapped;
824 /* Be careful as we don't hold info->lock */
825 swapped = READ_ONCE(info->swapped);
828 * The easier cases are when the shmem object has nothing in swap, or
829 * the vma maps it whole. Then we can simply use the stats that we
835 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
836 return swapped << PAGE_SHIFT;
838 /* Here comes the more involved part */
839 return shmem_partial_swap_usage(mapping,
840 linear_page_index(vma, vma->vm_start),
841 linear_page_index(vma, vma->vm_end));
845 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
847 void shmem_unlock_mapping(struct address_space *mapping)
850 pgoff_t indices[PAGEVEC_SIZE];
855 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
857 while (!mapping_unevictable(mapping)) {
859 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
860 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
862 pvec.nr = find_get_entries(mapping, index,
863 PAGEVEC_SIZE, pvec.pages, indices);
866 index = indices[pvec.nr - 1] + 1;
867 pagevec_remove_exceptionals(&pvec);
868 check_move_unevictable_pages(&pvec);
869 pagevec_release(&pvec);
875 * Check whether a hole-punch or truncation needs to split a huge page,
876 * returning true if no split was required, or the split has been successful.
878 * Eviction (or truncation to 0 size) should never need to split a huge page;
879 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
880 * head, and then succeeded to trylock on tail.
882 * A split can only succeed when there are no additional references on the
883 * huge page: so the split below relies upon find_get_entries() having stopped
884 * when it found a subpage of the huge page, without getting further references.
886 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
888 if (!PageTransCompound(page))
891 /* Just proceed to delete a huge page wholly within the range punched */
892 if (PageHead(page) &&
893 page->index >= start && page->index + HPAGE_PMD_NR <= end)
896 /* Try to split huge page, so we can truly punch the hole or truncate */
897 return split_huge_page(page) >= 0;
901 * Remove range of pages and swap entries from page cache, and free them.
902 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
904 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
907 struct address_space *mapping = inode->i_mapping;
908 struct shmem_inode_info *info = SHMEM_I(inode);
909 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
910 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
911 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
912 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
914 pgoff_t indices[PAGEVEC_SIZE];
915 long nr_swaps_freed = 0;
920 end = -1; /* unsigned, so actually very big */
924 while (index < end) {
925 pvec.nr = find_get_entries(mapping, index,
926 min(end - index, (pgoff_t)PAGEVEC_SIZE),
927 pvec.pages, indices);
930 for (i = 0; i < pagevec_count(&pvec); i++) {
931 struct page *page = pvec.pages[i];
937 if (xa_is_value(page)) {
940 nr_swaps_freed += !shmem_free_swap(mapping,
945 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
947 if (!trylock_page(page))
950 if ((!unfalloc || !PageUptodate(page)) &&
951 page_mapping(page) == mapping) {
952 VM_BUG_ON_PAGE(PageWriteback(page), page);
953 if (shmem_punch_compound(page, start, end))
954 truncate_inode_page(mapping, page);
958 pagevec_remove_exceptionals(&pvec);
959 pagevec_release(&pvec);
965 struct page *page = NULL;
966 shmem_getpage(inode, start - 1, &page, SGP_READ);
968 unsigned int top = PAGE_SIZE;
973 zero_user_segment(page, partial_start, top);
974 set_page_dirty(page);
980 struct page *page = NULL;
981 shmem_getpage(inode, end, &page, SGP_READ);
983 zero_user_segment(page, 0, partial_end);
984 set_page_dirty(page);
993 while (index < end) {
996 pvec.nr = find_get_entries(mapping, index,
997 min(end - index, (pgoff_t)PAGEVEC_SIZE),
998 pvec.pages, indices);
1000 /* If all gone or hole-punch or unfalloc, we're done */
1001 if (index == start || end != -1)
1003 /* But if truncating, restart to make sure all gone */
1007 for (i = 0; i < pagevec_count(&pvec); i++) {
1008 struct page *page = pvec.pages[i];
1014 if (xa_is_value(page)) {
1017 if (shmem_free_swap(mapping, index, page)) {
1018 /* Swap was replaced by page: retry */
1028 if (!unfalloc || !PageUptodate(page)) {
1029 if (page_mapping(page) != mapping) {
1030 /* Page was replaced by swap: retry */
1035 VM_BUG_ON_PAGE(PageWriteback(page), page);
1036 if (shmem_punch_compound(page, start, end))
1037 truncate_inode_page(mapping, page);
1038 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1039 /* Wipe the page and don't get stuck */
1040 clear_highpage(page);
1041 flush_dcache_page(page);
1042 set_page_dirty(page);
1044 round_up(start, HPAGE_PMD_NR))
1050 pagevec_remove_exceptionals(&pvec);
1051 pagevec_release(&pvec);
1055 spin_lock_irq(&info->lock);
1056 info->swapped -= nr_swaps_freed;
1057 shmem_recalc_inode(inode);
1058 spin_unlock_irq(&info->lock);
1061 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1063 shmem_undo_range(inode, lstart, lend, false);
1064 inode->i_ctime = inode->i_mtime = current_time(inode);
1066 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1068 static int shmem_getattr(const struct path *path, struct kstat *stat,
1069 u32 request_mask, unsigned int query_flags)
1071 struct inode *inode = path->dentry->d_inode;
1072 struct shmem_inode_info *info = SHMEM_I(inode);
1073 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1075 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1076 spin_lock_irq(&info->lock);
1077 shmem_recalc_inode(inode);
1078 spin_unlock_irq(&info->lock);
1080 generic_fillattr(inode, stat);
1082 if (is_huge_enabled(sb_info))
1083 stat->blksize = HPAGE_PMD_SIZE;
1088 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
1090 struct inode *inode = d_inode(dentry);
1091 struct shmem_inode_info *info = SHMEM_I(inode);
1092 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1095 error = setattr_prepare(dentry, attr);
1099 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1100 loff_t oldsize = inode->i_size;
1101 loff_t newsize = attr->ia_size;
1103 /* protected by i_mutex */
1104 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1105 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1108 if (newsize != oldsize) {
1109 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1113 i_size_write(inode, newsize);
1114 inode->i_ctime = inode->i_mtime = current_time(inode);
1116 if (newsize <= oldsize) {
1117 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1118 if (oldsize > holebegin)
1119 unmap_mapping_range(inode->i_mapping,
1122 shmem_truncate_range(inode,
1123 newsize, (loff_t)-1);
1124 /* unmap again to remove racily COWed private pages */
1125 if (oldsize > holebegin)
1126 unmap_mapping_range(inode->i_mapping,
1130 * Part of the huge page can be beyond i_size: subject
1131 * to shrink under memory pressure.
1133 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1134 spin_lock(&sbinfo->shrinklist_lock);
1136 * _careful to defend against unlocked access to
1137 * ->shrink_list in shmem_unused_huge_shrink()
1139 if (list_empty_careful(&info->shrinklist)) {
1140 list_add_tail(&info->shrinklist,
1141 &sbinfo->shrinklist);
1142 sbinfo->shrinklist_len++;
1144 spin_unlock(&sbinfo->shrinklist_lock);
1149 setattr_copy(inode, attr);
1150 if (attr->ia_valid & ATTR_MODE)
1151 error = posix_acl_chmod(inode, inode->i_mode);
1155 static void shmem_evict_inode(struct inode *inode)
1157 struct shmem_inode_info *info = SHMEM_I(inode);
1158 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1160 if (inode->i_mapping->a_ops == &shmem_aops) {
1161 shmem_unacct_size(info->flags, inode->i_size);
1163 shmem_truncate_range(inode, 0, (loff_t)-1);
1164 if (!list_empty(&info->shrinklist)) {
1165 spin_lock(&sbinfo->shrinklist_lock);
1166 if (!list_empty(&info->shrinklist)) {
1167 list_del_init(&info->shrinklist);
1168 sbinfo->shrinklist_len--;
1170 spin_unlock(&sbinfo->shrinklist_lock);
1172 while (!list_empty(&info->swaplist)) {
1173 /* Wait while shmem_unuse() is scanning this inode... */
1174 wait_var_event(&info->stop_eviction,
1175 !atomic_read(&info->stop_eviction));
1176 mutex_lock(&shmem_swaplist_mutex);
1177 /* ...but beware of the race if we peeked too early */
1178 if (!atomic_read(&info->stop_eviction))
1179 list_del_init(&info->swaplist);
1180 mutex_unlock(&shmem_swaplist_mutex);
1184 simple_xattrs_free(&info->xattrs);
1185 WARN_ON(inode->i_blocks);
1186 shmem_free_inode(inode->i_sb);
1190 extern struct swap_info_struct *swap_info[];
1192 static int shmem_find_swap_entries(struct address_space *mapping,
1193 pgoff_t start, unsigned int nr_entries,
1194 struct page **entries, pgoff_t *indices,
1195 unsigned int type, bool frontswap)
1197 XA_STATE(xas, &mapping->i_pages, start);
1200 unsigned int ret = 0;
1206 xas_for_each(&xas, page, ULONG_MAX) {
1207 if (xas_retry(&xas, page))
1210 if (!xa_is_value(page))
1213 entry = radix_to_swp_entry(page);
1214 if (swp_type(entry) != type)
1217 !frontswap_test(swap_info[type], swp_offset(entry)))
1220 indices[ret] = xas.xa_index;
1221 entries[ret] = page;
1223 if (need_resched()) {
1227 if (++ret == nr_entries)
1236 * Move the swapped pages for an inode to page cache. Returns the count
1237 * of pages swapped in, or the error in case of failure.
1239 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1245 struct address_space *mapping = inode->i_mapping;
1247 for (i = 0; i < pvec.nr; i++) {
1248 struct page *page = pvec.pages[i];
1250 if (!xa_is_value(page))
1252 error = shmem_swapin_page(inode, indices[i],
1254 mapping_gfp_mask(mapping),
1261 if (error == -ENOMEM)
1265 return error ? error : ret;
1269 * If swap found in inode, free it and move page from swapcache to filecache.
1271 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1272 bool frontswap, unsigned long *fs_pages_to_unuse)
1274 struct address_space *mapping = inode->i_mapping;
1276 struct pagevec pvec;
1277 pgoff_t indices[PAGEVEC_SIZE];
1278 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1281 pagevec_init(&pvec);
1283 unsigned int nr_entries = PAGEVEC_SIZE;
1285 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1286 nr_entries = *fs_pages_to_unuse;
1288 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1289 pvec.pages, indices,
1296 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1300 if (frontswap_partial) {
1301 *fs_pages_to_unuse -= ret;
1302 if (*fs_pages_to_unuse == 0) {
1303 ret = FRONTSWAP_PAGES_UNUSED;
1308 start = indices[pvec.nr - 1];
1315 * Read all the shared memory data that resides in the swap
1316 * device 'type' back into memory, so the swap device can be
1319 int shmem_unuse(unsigned int type, bool frontswap,
1320 unsigned long *fs_pages_to_unuse)
1322 struct shmem_inode_info *info, *next;
1325 if (list_empty(&shmem_swaplist))
1328 mutex_lock(&shmem_swaplist_mutex);
1329 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1330 if (!info->swapped) {
1331 list_del_init(&info->swaplist);
1335 * Drop the swaplist mutex while searching the inode for swap;
1336 * but before doing so, make sure shmem_evict_inode() will not
1337 * remove placeholder inode from swaplist, nor let it be freed
1338 * (igrab() would protect from unlink, but not from unmount).
1340 atomic_inc(&info->stop_eviction);
1341 mutex_unlock(&shmem_swaplist_mutex);
1343 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1347 mutex_lock(&shmem_swaplist_mutex);
1348 next = list_next_entry(info, swaplist);
1350 list_del_init(&info->swaplist);
1351 if (atomic_dec_and_test(&info->stop_eviction))
1352 wake_up_var(&info->stop_eviction);
1356 mutex_unlock(&shmem_swaplist_mutex);
1362 * Move the page from the page cache to the swap cache.
1364 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1366 struct shmem_inode_info *info;
1367 struct address_space *mapping;
1368 struct inode *inode;
1372 VM_BUG_ON_PAGE(PageCompound(page), page);
1373 BUG_ON(!PageLocked(page));
1374 mapping = page->mapping;
1375 index = page->index;
1376 inode = mapping->host;
1377 info = SHMEM_I(inode);
1378 if (info->flags & VM_LOCKED)
1380 if (!total_swap_pages)
1384 * Our capabilities prevent regular writeback or sync from ever calling
1385 * shmem_writepage; but a stacking filesystem might use ->writepage of
1386 * its underlying filesystem, in which case tmpfs should write out to
1387 * swap only in response to memory pressure, and not for the writeback
1390 if (!wbc->for_reclaim) {
1391 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1396 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1397 * value into swapfile.c, the only way we can correctly account for a
1398 * fallocated page arriving here is now to initialize it and write it.
1400 * That's okay for a page already fallocated earlier, but if we have
1401 * not yet completed the fallocation, then (a) we want to keep track
1402 * of this page in case we have to undo it, and (b) it may not be a
1403 * good idea to continue anyway, once we're pushing into swap. So
1404 * reactivate the page, and let shmem_fallocate() quit when too many.
1406 if (!PageUptodate(page)) {
1407 if (inode->i_private) {
1408 struct shmem_falloc *shmem_falloc;
1409 spin_lock(&inode->i_lock);
1410 shmem_falloc = inode->i_private;
1412 !shmem_falloc->waitq &&
1413 index >= shmem_falloc->start &&
1414 index < shmem_falloc->next)
1415 shmem_falloc->nr_unswapped++;
1417 shmem_falloc = NULL;
1418 spin_unlock(&inode->i_lock);
1422 clear_highpage(page);
1423 flush_dcache_page(page);
1424 SetPageUptodate(page);
1427 swap = get_swap_page(page);
1432 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1433 * if it's not already there. Do it now before the page is
1434 * moved to swap cache, when its pagelock no longer protects
1435 * the inode from eviction. But don't unlock the mutex until
1436 * we've incremented swapped, because shmem_unuse_inode() will
1437 * prune a !swapped inode from the swaplist under this mutex.
1439 mutex_lock(&shmem_swaplist_mutex);
1440 if (list_empty(&info->swaplist))
1441 list_add(&info->swaplist, &shmem_swaplist);
1443 if (add_to_swap_cache(page, swap,
1444 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1446 spin_lock_irq(&info->lock);
1447 shmem_recalc_inode(inode);
1449 spin_unlock_irq(&info->lock);
1451 swap_shmem_alloc(swap);
1452 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1454 mutex_unlock(&shmem_swaplist_mutex);
1455 BUG_ON(page_mapped(page));
1456 swap_writepage(page, wbc);
1460 mutex_unlock(&shmem_swaplist_mutex);
1461 put_swap_page(page, swap);
1463 set_page_dirty(page);
1464 if (wbc->for_reclaim)
1465 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1470 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1471 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1475 if (!mpol || mpol->mode == MPOL_DEFAULT)
1476 return; /* show nothing */
1478 mpol_to_str(buffer, sizeof(buffer), mpol);
1480 seq_printf(seq, ",mpol=%s", buffer);
1483 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1485 struct mempolicy *mpol = NULL;
1487 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1488 mpol = sbinfo->mpol;
1490 spin_unlock(&sbinfo->stat_lock);
1494 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1495 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1498 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1502 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1504 #define vm_policy vm_private_data
1507 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1508 struct shmem_inode_info *info, pgoff_t index)
1510 /* Create a pseudo vma that just contains the policy */
1511 vma_init(vma, NULL);
1512 /* Bias interleave by inode number to distribute better across nodes */
1513 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1514 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1517 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1519 /* Drop reference taken by mpol_shared_policy_lookup() */
1520 mpol_cond_put(vma->vm_policy);
1523 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1524 struct shmem_inode_info *info, pgoff_t index)
1526 struct vm_area_struct pvma;
1528 struct vm_fault vmf;
1530 shmem_pseudo_vma_init(&pvma, info, index);
1533 page = swap_cluster_readahead(swap, gfp, &vmf);
1534 shmem_pseudo_vma_destroy(&pvma);
1539 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1540 struct shmem_inode_info *info, pgoff_t index)
1542 struct vm_area_struct pvma;
1543 struct address_space *mapping = info->vfs_inode.i_mapping;
1547 hindex = round_down(index, HPAGE_PMD_NR);
1548 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1552 shmem_pseudo_vma_init(&pvma, info, hindex);
1553 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1554 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1555 shmem_pseudo_vma_destroy(&pvma);
1557 prep_transhuge_page(page);
1559 count_vm_event(THP_FILE_FALLBACK);
1563 static struct page *shmem_alloc_page(gfp_t gfp,
1564 struct shmem_inode_info *info, pgoff_t index)
1566 struct vm_area_struct pvma;
1569 shmem_pseudo_vma_init(&pvma, info, index);
1570 page = alloc_page_vma(gfp, &pvma, 0);
1571 shmem_pseudo_vma_destroy(&pvma);
1576 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1577 struct inode *inode,
1578 pgoff_t index, bool huge)
1580 struct shmem_inode_info *info = SHMEM_I(inode);
1585 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1587 nr = huge ? HPAGE_PMD_NR : 1;
1589 if (!shmem_inode_acct_block(inode, nr))
1593 page = shmem_alloc_hugepage(gfp, info, index);
1595 page = shmem_alloc_page(gfp, info, index);
1597 __SetPageLocked(page);
1598 __SetPageSwapBacked(page);
1603 shmem_inode_unacct_blocks(inode, nr);
1605 return ERR_PTR(err);
1609 * When a page is moved from swapcache to shmem filecache (either by the
1610 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1611 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1612 * ignorance of the mapping it belongs to. If that mapping has special
1613 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1614 * we may need to copy to a suitable page before moving to filecache.
1616 * In a future release, this may well be extended to respect cpuset and
1617 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1618 * but for now it is a simple matter of zone.
1620 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1622 return page_zonenum(page) > gfp_zone(gfp);
1625 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1626 struct shmem_inode_info *info, pgoff_t index)
1628 struct page *oldpage, *newpage;
1629 struct address_space *swap_mapping;
1635 entry.val = page_private(oldpage);
1636 swap_index = swp_offset(entry);
1637 swap_mapping = page_mapping(oldpage);
1640 * We have arrived here because our zones are constrained, so don't
1641 * limit chance of success by further cpuset and node constraints.
1643 gfp &= ~GFP_CONSTRAINT_MASK;
1644 newpage = shmem_alloc_page(gfp, info, index);
1649 copy_highpage(newpage, oldpage);
1650 flush_dcache_page(newpage);
1652 __SetPageLocked(newpage);
1653 __SetPageSwapBacked(newpage);
1654 SetPageUptodate(newpage);
1655 set_page_private(newpage, entry.val);
1656 SetPageSwapCache(newpage);
1659 * Our caller will very soon move newpage out of swapcache, but it's
1660 * a nice clean interface for us to replace oldpage by newpage there.
1662 xa_lock_irq(&swap_mapping->i_pages);
1663 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1665 mem_cgroup_migrate(oldpage, newpage);
1666 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1667 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1669 xa_unlock_irq(&swap_mapping->i_pages);
1671 if (unlikely(error)) {
1673 * Is this possible? I think not, now that our callers check
1674 * both PageSwapCache and page_private after getting page lock;
1675 * but be defensive. Reverse old to newpage for clear and free.
1679 lru_cache_add(newpage);
1683 ClearPageSwapCache(oldpage);
1684 set_page_private(oldpage, 0);
1686 unlock_page(oldpage);
1693 * Swap in the page pointed to by *pagep.
1694 * Caller has to make sure that *pagep contains a valid swapped page.
1695 * Returns 0 and the page in pagep if success. On failure, returns the
1696 * error code and NULL in *pagep.
1698 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1699 struct page **pagep, enum sgp_type sgp,
1700 gfp_t gfp, struct vm_area_struct *vma,
1701 vm_fault_t *fault_type)
1703 struct address_space *mapping = inode->i_mapping;
1704 struct shmem_inode_info *info = SHMEM_I(inode);
1705 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
1710 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1711 swap = radix_to_swp_entry(*pagep);
1714 /* Look it up and read it in.. */
1715 page = lookup_swap_cache(swap, NULL, 0);
1717 /* Or update major stats only when swapin succeeds?? */
1719 *fault_type |= VM_FAULT_MAJOR;
1720 count_vm_event(PGMAJFAULT);
1721 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1723 /* Here we actually start the io */
1724 page = shmem_swapin(swap, gfp, info, index);
1731 /* We have to do this with page locked to prevent races */
1733 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1734 !shmem_confirm_swap(mapping, index, swap)) {
1738 if (!PageUptodate(page)) {
1742 wait_on_page_writeback(page);
1745 * Some architectures may have to restore extra metadata to the
1746 * physical page after reading from swap.
1748 arch_swap_restore(swap, page);
1750 if (shmem_should_replace_page(page, gfp)) {
1751 error = shmem_replace_page(&page, gfp, info, index);
1756 error = shmem_add_to_page_cache(page, mapping, index,
1757 swp_to_radix_entry(swap), gfp,
1762 spin_lock_irq(&info->lock);
1764 shmem_recalc_inode(inode);
1765 spin_unlock_irq(&info->lock);
1767 if (sgp == SGP_WRITE)
1768 mark_page_accessed(page);
1770 delete_from_swap_cache(page);
1771 set_page_dirty(page);
1777 if (!shmem_confirm_swap(mapping, index, swap))
1789 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1791 * If we allocate a new one we do not mark it dirty. That's up to the
1792 * vm. If we swap it in we mark it dirty since we also free the swap
1793 * entry since a page cannot live in both the swap and page cache.
1795 * vmf and fault_type are only supplied by shmem_fault:
1796 * otherwise they are NULL.
1798 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1799 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1800 struct vm_area_struct *vma, struct vm_fault *vmf,
1801 vm_fault_t *fault_type)
1803 struct address_space *mapping = inode->i_mapping;
1804 struct shmem_inode_info *info = SHMEM_I(inode);
1805 struct shmem_sb_info *sbinfo;
1806 struct mm_struct *charge_mm;
1808 enum sgp_type sgp_huge = sgp;
1809 pgoff_t hindex = index;
1814 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1816 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1819 if (sgp <= SGP_CACHE &&
1820 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1824 sbinfo = SHMEM_SB(inode->i_sb);
1825 charge_mm = vma ? vma->vm_mm : current->mm;
1827 page = find_lock_entry(mapping, index);
1828 if (xa_is_value(page)) {
1829 error = shmem_swapin_page(inode, index, &page,
1830 sgp, gfp, vma, fault_type);
1831 if (error == -EEXIST)
1839 hindex = page->index;
1840 if (page && sgp == SGP_WRITE)
1841 mark_page_accessed(page);
1843 /* fallocated page? */
1844 if (page && !PageUptodate(page)) {
1845 if (sgp != SGP_READ)
1852 if (page || sgp == SGP_READ)
1856 * Fast cache lookup did not find it:
1857 * bring it back from swap or allocate.
1860 if (vma && userfaultfd_missing(vma)) {
1861 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1865 /* shmem_symlink() */
1866 if (mapping->a_ops != &shmem_aops)
1868 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1870 if (shmem_huge == SHMEM_HUGE_FORCE)
1872 switch (sbinfo->huge) {
1873 case SHMEM_HUGE_NEVER:
1875 case SHMEM_HUGE_WITHIN_SIZE: {
1879 off = round_up(index, HPAGE_PMD_NR);
1880 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1881 if (i_size >= HPAGE_PMD_SIZE &&
1882 i_size >> PAGE_SHIFT >= off)
1887 case SHMEM_HUGE_ADVISE:
1888 if (sgp_huge == SGP_HUGE)
1890 /* TODO: implement fadvise() hints */
1895 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1898 page = shmem_alloc_and_acct_page(gfp, inode,
1904 error = PTR_ERR(page);
1906 if (error != -ENOSPC)
1909 * Try to reclaim some space by splitting a huge page
1910 * beyond i_size on the filesystem.
1915 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1916 if (ret == SHRINK_STOP)
1924 if (PageTransHuge(page))
1925 hindex = round_down(index, HPAGE_PMD_NR);
1929 if (sgp == SGP_WRITE)
1930 __SetPageReferenced(page);
1932 error = shmem_add_to_page_cache(page, mapping, hindex,
1933 NULL, gfp & GFP_RECLAIM_MASK,
1937 lru_cache_add(page);
1939 spin_lock_irq(&info->lock);
1940 info->alloced += compound_nr(page);
1941 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1942 shmem_recalc_inode(inode);
1943 spin_unlock_irq(&info->lock);
1946 if (PageTransHuge(page) &&
1947 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1948 hindex + HPAGE_PMD_NR - 1) {
1950 * Part of the huge page is beyond i_size: subject
1951 * to shrink under memory pressure.
1953 spin_lock(&sbinfo->shrinklist_lock);
1955 * _careful to defend against unlocked access to
1956 * ->shrink_list in shmem_unused_huge_shrink()
1958 if (list_empty_careful(&info->shrinklist)) {
1959 list_add_tail(&info->shrinklist,
1960 &sbinfo->shrinklist);
1961 sbinfo->shrinklist_len++;
1963 spin_unlock(&sbinfo->shrinklist_lock);
1967 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1969 if (sgp == SGP_FALLOC)
1973 * Let SGP_WRITE caller clear ends if write does not fill page;
1974 * but SGP_FALLOC on a page fallocated earlier must initialize
1975 * it now, lest undo on failure cancel our earlier guarantee.
1977 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1980 for (i = 0; i < compound_nr(page); i++) {
1981 clear_highpage(page + i);
1982 flush_dcache_page(page + i);
1984 SetPageUptodate(page);
1987 /* Perhaps the file has been truncated since we checked */
1988 if (sgp <= SGP_CACHE &&
1989 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1991 ClearPageDirty(page);
1992 delete_from_page_cache(page);
1993 spin_lock_irq(&info->lock);
1994 shmem_recalc_inode(inode);
1995 spin_unlock_irq(&info->lock);
2001 *pagep = page + index - hindex;
2008 shmem_inode_unacct_blocks(inode, compound_nr(page));
2010 if (PageTransHuge(page)) {
2020 if (error == -ENOSPC && !once++) {
2021 spin_lock_irq(&info->lock);
2022 shmem_recalc_inode(inode);
2023 spin_unlock_irq(&info->lock);
2026 if (error == -EEXIST)
2032 * This is like autoremove_wake_function, but it removes the wait queue
2033 * entry unconditionally - even if something else had already woken the
2036 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2038 int ret = default_wake_function(wait, mode, sync, key);
2039 list_del_init(&wait->entry);
2043 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2045 struct vm_area_struct *vma = vmf->vma;
2046 struct inode *inode = file_inode(vma->vm_file);
2047 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2050 vm_fault_t ret = VM_FAULT_LOCKED;
2053 * Trinity finds that probing a hole which tmpfs is punching can
2054 * prevent the hole-punch from ever completing: which in turn
2055 * locks writers out with its hold on i_mutex. So refrain from
2056 * faulting pages into the hole while it's being punched. Although
2057 * shmem_undo_range() does remove the additions, it may be unable to
2058 * keep up, as each new page needs its own unmap_mapping_range() call,
2059 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2061 * It does not matter if we sometimes reach this check just before the
2062 * hole-punch begins, so that one fault then races with the punch:
2063 * we just need to make racing faults a rare case.
2065 * The implementation below would be much simpler if we just used a
2066 * standard mutex or completion: but we cannot take i_mutex in fault,
2067 * and bloating every shmem inode for this unlikely case would be sad.
2069 if (unlikely(inode->i_private)) {
2070 struct shmem_falloc *shmem_falloc;
2072 spin_lock(&inode->i_lock);
2073 shmem_falloc = inode->i_private;
2075 shmem_falloc->waitq &&
2076 vmf->pgoff >= shmem_falloc->start &&
2077 vmf->pgoff < shmem_falloc->next) {
2079 wait_queue_head_t *shmem_falloc_waitq;
2080 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2082 ret = VM_FAULT_NOPAGE;
2083 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2085 ret = VM_FAULT_RETRY;
2087 shmem_falloc_waitq = shmem_falloc->waitq;
2088 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2089 TASK_UNINTERRUPTIBLE);
2090 spin_unlock(&inode->i_lock);
2094 * shmem_falloc_waitq points into the shmem_fallocate()
2095 * stack of the hole-punching task: shmem_falloc_waitq
2096 * is usually invalid by the time we reach here, but
2097 * finish_wait() does not dereference it in that case;
2098 * though i_lock needed lest racing with wake_up_all().
2100 spin_lock(&inode->i_lock);
2101 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2102 spin_unlock(&inode->i_lock);
2108 spin_unlock(&inode->i_lock);
2113 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2114 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2116 else if (vma->vm_flags & VM_HUGEPAGE)
2119 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2120 gfp, vma, vmf, &ret);
2122 return vmf_error(err);
2126 unsigned long shmem_get_unmapped_area(struct file *file,
2127 unsigned long uaddr, unsigned long len,
2128 unsigned long pgoff, unsigned long flags)
2130 unsigned long (*get_area)(struct file *,
2131 unsigned long, unsigned long, unsigned long, unsigned long);
2133 unsigned long offset;
2134 unsigned long inflated_len;
2135 unsigned long inflated_addr;
2136 unsigned long inflated_offset;
2138 if (len > TASK_SIZE)
2141 get_area = current->mm->get_unmapped_area;
2142 addr = get_area(file, uaddr, len, pgoff, flags);
2144 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2146 if (IS_ERR_VALUE(addr))
2148 if (addr & ~PAGE_MASK)
2150 if (addr > TASK_SIZE - len)
2153 if (shmem_huge == SHMEM_HUGE_DENY)
2155 if (len < HPAGE_PMD_SIZE)
2157 if (flags & MAP_FIXED)
2160 * Our priority is to support MAP_SHARED mapped hugely;
2161 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2162 * But if caller specified an address hint and we allocated area there
2163 * successfully, respect that as before.
2168 if (shmem_huge != SHMEM_HUGE_FORCE) {
2169 struct super_block *sb;
2172 VM_BUG_ON(file->f_op != &shmem_file_operations);
2173 sb = file_inode(file)->i_sb;
2176 * Called directly from mm/mmap.c, or drivers/char/mem.c
2177 * for "/dev/zero", to create a shared anonymous object.
2179 if (IS_ERR(shm_mnt))
2181 sb = shm_mnt->mnt_sb;
2183 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2187 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2188 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2190 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2193 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2194 if (inflated_len > TASK_SIZE)
2196 if (inflated_len < len)
2199 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2200 if (IS_ERR_VALUE(inflated_addr))
2202 if (inflated_addr & ~PAGE_MASK)
2205 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2206 inflated_addr += offset - inflated_offset;
2207 if (inflated_offset > offset)
2208 inflated_addr += HPAGE_PMD_SIZE;
2210 if (inflated_addr > TASK_SIZE - len)
2212 return inflated_addr;
2216 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2218 struct inode *inode = file_inode(vma->vm_file);
2219 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2222 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2225 struct inode *inode = file_inode(vma->vm_file);
2228 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2229 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2233 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2235 struct inode *inode = file_inode(file);
2236 struct shmem_inode_info *info = SHMEM_I(inode);
2237 int retval = -ENOMEM;
2240 * What serializes the accesses to info->flags?
2241 * ipc_lock_object() when called from shmctl_do_lock(),
2242 * no serialization needed when called from shm_destroy().
2244 if (lock && !(info->flags & VM_LOCKED)) {
2245 if (!user_shm_lock(inode->i_size, user))
2247 info->flags |= VM_LOCKED;
2248 mapping_set_unevictable(file->f_mapping);
2250 if (!lock && (info->flags & VM_LOCKED) && user) {
2251 user_shm_unlock(inode->i_size, user);
2252 info->flags &= ~VM_LOCKED;
2253 mapping_clear_unevictable(file->f_mapping);
2261 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2263 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2266 ret = seal_check_future_write(info->seals, vma);
2270 /* arm64 - allow memory tagging on RAM-based files */
2271 vma->vm_flags |= VM_MTE_ALLOWED;
2273 file_accessed(file);
2274 vma->vm_ops = &shmem_vm_ops;
2275 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2276 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2277 (vma->vm_end & HPAGE_PMD_MASK)) {
2278 khugepaged_enter(vma, vma->vm_flags);
2283 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2284 umode_t mode, dev_t dev, unsigned long flags)
2286 struct inode *inode;
2287 struct shmem_inode_info *info;
2288 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2291 if (shmem_reserve_inode(sb, &ino))
2294 inode = new_inode(sb);
2297 inode_init_owner(inode, dir, mode);
2298 inode->i_blocks = 0;
2299 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2300 inode->i_generation = prandom_u32();
2301 info = SHMEM_I(inode);
2302 memset(info, 0, (char *)inode - (char *)info);
2303 spin_lock_init(&info->lock);
2304 atomic_set(&info->stop_eviction, 0);
2305 info->seals = F_SEAL_SEAL;
2306 info->flags = flags & VM_NORESERVE;
2307 INIT_LIST_HEAD(&info->shrinklist);
2308 INIT_LIST_HEAD(&info->swaplist);
2309 simple_xattrs_init(&info->xattrs);
2310 cache_no_acl(inode);
2312 switch (mode & S_IFMT) {
2314 inode->i_op = &shmem_special_inode_operations;
2315 init_special_inode(inode, mode, dev);
2318 inode->i_mapping->a_ops = &shmem_aops;
2319 inode->i_op = &shmem_inode_operations;
2320 inode->i_fop = &shmem_file_operations;
2321 mpol_shared_policy_init(&info->policy,
2322 shmem_get_sbmpol(sbinfo));
2326 /* Some things misbehave if size == 0 on a directory */
2327 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2328 inode->i_op = &shmem_dir_inode_operations;
2329 inode->i_fop = &simple_dir_operations;
2333 * Must not load anything in the rbtree,
2334 * mpol_free_shared_policy will not be called.
2336 mpol_shared_policy_init(&info->policy, NULL);
2340 lockdep_annotate_inode_mutex_key(inode);
2342 shmem_free_inode(sb);
2346 bool shmem_mapping(struct address_space *mapping)
2348 return mapping->a_ops == &shmem_aops;
2351 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2353 struct vm_area_struct *dst_vma,
2354 unsigned long dst_addr,
2355 unsigned long src_addr,
2357 struct page **pagep)
2359 struct inode *inode = file_inode(dst_vma->vm_file);
2360 struct shmem_inode_info *info = SHMEM_I(inode);
2361 struct address_space *mapping = inode->i_mapping;
2362 gfp_t gfp = mapping_gfp_mask(mapping);
2363 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2367 pte_t _dst_pte, *dst_pte;
2369 pgoff_t offset, max_off;
2372 if (!shmem_inode_acct_block(inode, 1)) {
2374 * We may have got a page, returned -ENOENT triggering a retry,
2375 * and now we find ourselves with -ENOMEM. Release the page, to
2376 * avoid a BUG_ON in our caller.
2378 if (unlikely(*pagep)) {
2386 page = shmem_alloc_page(gfp, info, pgoff);
2388 goto out_unacct_blocks;
2390 if (!zeropage) { /* mcopy_atomic */
2391 page_kaddr = kmap_atomic(page);
2392 ret = copy_from_user(page_kaddr,
2393 (const void __user *)src_addr,
2395 kunmap_atomic(page_kaddr);
2397 /* fallback to copy_from_user outside mmap_lock */
2398 if (unlikely(ret)) {
2400 shmem_inode_unacct_blocks(inode, 1);
2401 /* don't free the page */
2404 } else { /* mfill_zeropage_atomic */
2405 clear_highpage(page);
2412 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2413 __SetPageLocked(page);
2414 __SetPageSwapBacked(page);
2415 __SetPageUptodate(page);
2418 offset = linear_page_index(dst_vma, dst_addr);
2419 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2420 if (unlikely(offset >= max_off))
2423 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2424 gfp & GFP_RECLAIM_MASK, dst_mm);
2428 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2429 if (dst_vma->vm_flags & VM_WRITE)
2430 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2433 * We don't set the pte dirty if the vma has no
2434 * VM_WRITE permission, so mark the page dirty or it
2435 * could be freed from under us. We could do it
2436 * unconditionally before unlock_page(), but doing it
2437 * only if VM_WRITE is not set is faster.
2439 set_page_dirty(page);
2442 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2445 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2446 if (unlikely(offset >= max_off))
2447 goto out_release_unlock;
2450 if (!pte_none(*dst_pte))
2451 goto out_release_unlock;
2453 lru_cache_add(page);
2455 spin_lock_irq(&info->lock);
2457 inode->i_blocks += BLOCKS_PER_PAGE;
2458 shmem_recalc_inode(inode);
2459 spin_unlock_irq(&info->lock);
2461 inc_mm_counter(dst_mm, mm_counter_file(page));
2462 page_add_file_rmap(page, false);
2463 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2465 /* No need to invalidate - it was non-present before */
2466 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2467 pte_unmap_unlock(dst_pte, ptl);
2473 pte_unmap_unlock(dst_pte, ptl);
2474 ClearPageDirty(page);
2475 delete_from_page_cache(page);
2480 shmem_inode_unacct_blocks(inode, 1);
2484 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2486 struct vm_area_struct *dst_vma,
2487 unsigned long dst_addr,
2488 unsigned long src_addr,
2489 struct page **pagep)
2491 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2492 dst_addr, src_addr, false, pagep);
2495 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2497 struct vm_area_struct *dst_vma,
2498 unsigned long dst_addr)
2500 struct page *page = NULL;
2502 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2503 dst_addr, 0, true, &page);
2507 static const struct inode_operations shmem_symlink_inode_operations;
2508 static const struct inode_operations shmem_short_symlink_operations;
2510 #ifdef CONFIG_TMPFS_XATTR
2511 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2513 #define shmem_initxattrs NULL
2517 shmem_write_begin(struct file *file, struct address_space *mapping,
2518 loff_t pos, unsigned len, unsigned flags,
2519 struct page **pagep, void **fsdata)
2521 struct inode *inode = mapping->host;
2522 struct shmem_inode_info *info = SHMEM_I(inode);
2523 pgoff_t index = pos >> PAGE_SHIFT;
2525 /* i_mutex is held by caller */
2526 if (unlikely(info->seals & (F_SEAL_GROW |
2527 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2528 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2530 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2534 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2538 shmem_write_end(struct file *file, struct address_space *mapping,
2539 loff_t pos, unsigned len, unsigned copied,
2540 struct page *page, void *fsdata)
2542 struct inode *inode = mapping->host;
2544 if (pos + copied > inode->i_size)
2545 i_size_write(inode, pos + copied);
2547 if (!PageUptodate(page)) {
2548 struct page *head = compound_head(page);
2549 if (PageTransCompound(page)) {
2552 for (i = 0; i < HPAGE_PMD_NR; i++) {
2553 if (head + i == page)
2555 clear_highpage(head + i);
2556 flush_dcache_page(head + i);
2559 if (copied < PAGE_SIZE) {
2560 unsigned from = pos & (PAGE_SIZE - 1);
2561 zero_user_segments(page, 0, from,
2562 from + copied, PAGE_SIZE);
2564 SetPageUptodate(head);
2566 set_page_dirty(page);
2573 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2575 struct file *file = iocb->ki_filp;
2576 struct inode *inode = file_inode(file);
2577 struct address_space *mapping = inode->i_mapping;
2579 unsigned long offset;
2580 enum sgp_type sgp = SGP_READ;
2583 loff_t *ppos = &iocb->ki_pos;
2586 * Might this read be for a stacking filesystem? Then when reading
2587 * holes of a sparse file, we actually need to allocate those pages,
2588 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2590 if (!iter_is_iovec(to))
2593 index = *ppos >> PAGE_SHIFT;
2594 offset = *ppos & ~PAGE_MASK;
2597 struct page *page = NULL;
2599 unsigned long nr, ret;
2600 loff_t i_size = i_size_read(inode);
2602 end_index = i_size >> PAGE_SHIFT;
2603 if (index > end_index)
2605 if (index == end_index) {
2606 nr = i_size & ~PAGE_MASK;
2611 error = shmem_getpage(inode, index, &page, sgp);
2613 if (error == -EINVAL)
2618 if (sgp == SGP_CACHE)
2619 set_page_dirty(page);
2624 * We must evaluate after, since reads (unlike writes)
2625 * are called without i_mutex protection against truncate
2628 i_size = i_size_read(inode);
2629 end_index = i_size >> PAGE_SHIFT;
2630 if (index == end_index) {
2631 nr = i_size & ~PAGE_MASK;
2642 * If users can be writing to this page using arbitrary
2643 * virtual addresses, take care about potential aliasing
2644 * before reading the page on the kernel side.
2646 if (mapping_writably_mapped(mapping))
2647 flush_dcache_page(page);
2649 * Mark the page accessed if we read the beginning.
2652 mark_page_accessed(page);
2654 page = ZERO_PAGE(0);
2659 * Ok, we have the page, and it's up-to-date, so
2660 * now we can copy it to user space...
2662 ret = copy_page_to_iter(page, offset, nr, to);
2665 index += offset >> PAGE_SHIFT;
2666 offset &= ~PAGE_MASK;
2669 if (!iov_iter_count(to))
2678 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2679 file_accessed(file);
2680 return retval ? retval : error;
2684 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2686 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2687 pgoff_t index, pgoff_t end, int whence)
2690 struct pagevec pvec;
2691 pgoff_t indices[PAGEVEC_SIZE];
2695 pagevec_init(&pvec);
2696 pvec.nr = 1; /* start small: we may be there already */
2698 pvec.nr = find_get_entries(mapping, index,
2699 pvec.nr, pvec.pages, indices);
2701 if (whence == SEEK_DATA)
2705 for (i = 0; i < pvec.nr; i++, index++) {
2706 if (index < indices[i]) {
2707 if (whence == SEEK_HOLE) {
2713 page = pvec.pages[i];
2714 if (page && !xa_is_value(page)) {
2715 if (!PageUptodate(page))
2719 (page && whence == SEEK_DATA) ||
2720 (!page && whence == SEEK_HOLE)) {
2725 pagevec_remove_exceptionals(&pvec);
2726 pagevec_release(&pvec);
2727 pvec.nr = PAGEVEC_SIZE;
2733 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2735 struct address_space *mapping = file->f_mapping;
2736 struct inode *inode = mapping->host;
2740 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2741 return generic_file_llseek_size(file, offset, whence,
2742 MAX_LFS_FILESIZE, i_size_read(inode));
2744 /* We're holding i_mutex so we can access i_size directly */
2746 if (offset < 0 || offset >= inode->i_size)
2749 start = offset >> PAGE_SHIFT;
2750 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2751 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2752 new_offset <<= PAGE_SHIFT;
2753 if (new_offset > offset) {
2754 if (new_offset < inode->i_size)
2755 offset = new_offset;
2756 else if (whence == SEEK_DATA)
2759 offset = inode->i_size;
2764 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2765 inode_unlock(inode);
2769 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2772 struct inode *inode = file_inode(file);
2773 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2774 struct shmem_inode_info *info = SHMEM_I(inode);
2775 struct shmem_falloc shmem_falloc;
2776 pgoff_t start, index, end;
2779 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2784 if (mode & FALLOC_FL_PUNCH_HOLE) {
2785 struct address_space *mapping = file->f_mapping;
2786 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2787 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2788 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2790 /* protected by i_mutex */
2791 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2796 shmem_falloc.waitq = &shmem_falloc_waitq;
2797 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2798 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2799 spin_lock(&inode->i_lock);
2800 inode->i_private = &shmem_falloc;
2801 spin_unlock(&inode->i_lock);
2803 if ((u64)unmap_end > (u64)unmap_start)
2804 unmap_mapping_range(mapping, unmap_start,
2805 1 + unmap_end - unmap_start, 0);
2806 shmem_truncate_range(inode, offset, offset + len - 1);
2807 /* No need to unmap again: hole-punching leaves COWed pages */
2809 spin_lock(&inode->i_lock);
2810 inode->i_private = NULL;
2811 wake_up_all(&shmem_falloc_waitq);
2812 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2813 spin_unlock(&inode->i_lock);
2818 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2819 error = inode_newsize_ok(inode, offset + len);
2823 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2828 start = offset >> PAGE_SHIFT;
2829 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2830 /* Try to avoid a swapstorm if len is impossible to satisfy */
2831 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2836 shmem_falloc.waitq = NULL;
2837 shmem_falloc.start = start;
2838 shmem_falloc.next = start;
2839 shmem_falloc.nr_falloced = 0;
2840 shmem_falloc.nr_unswapped = 0;
2841 spin_lock(&inode->i_lock);
2842 inode->i_private = &shmem_falloc;
2843 spin_unlock(&inode->i_lock);
2845 for (index = start; index < end; index++) {
2849 * Good, the fallocate(2) manpage permits EINTR: we may have
2850 * been interrupted because we are using up too much memory.
2852 if (signal_pending(current))
2854 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2857 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2859 /* Remove the !PageUptodate pages we added */
2860 if (index > start) {
2861 shmem_undo_range(inode,
2862 (loff_t)start << PAGE_SHIFT,
2863 ((loff_t)index << PAGE_SHIFT) - 1, true);
2869 * Inform shmem_writepage() how far we have reached.
2870 * No need for lock or barrier: we have the page lock.
2872 shmem_falloc.next++;
2873 if (!PageUptodate(page))
2874 shmem_falloc.nr_falloced++;
2877 * If !PageUptodate, leave it that way so that freeable pages
2878 * can be recognized if we need to rollback on error later.
2879 * But set_page_dirty so that memory pressure will swap rather
2880 * than free the pages we are allocating (and SGP_CACHE pages
2881 * might still be clean: we now need to mark those dirty too).
2883 set_page_dirty(page);
2889 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2890 i_size_write(inode, offset + len);
2891 inode->i_ctime = current_time(inode);
2893 spin_lock(&inode->i_lock);
2894 inode->i_private = NULL;
2895 spin_unlock(&inode->i_lock);
2897 inode_unlock(inode);
2901 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2903 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2905 buf->f_type = TMPFS_MAGIC;
2906 buf->f_bsize = PAGE_SIZE;
2907 buf->f_namelen = NAME_MAX;
2908 if (sbinfo->max_blocks) {
2909 buf->f_blocks = sbinfo->max_blocks;
2911 buf->f_bfree = sbinfo->max_blocks -
2912 percpu_counter_sum(&sbinfo->used_blocks);
2914 if (sbinfo->max_inodes) {
2915 buf->f_files = sbinfo->max_inodes;
2916 buf->f_ffree = sbinfo->free_inodes;
2918 /* else leave those fields 0 like simple_statfs */
2923 * File creation. Allocate an inode, and we're done..
2926 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2928 struct inode *inode;
2929 int error = -ENOSPC;
2931 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2933 error = simple_acl_create(dir, inode);
2936 error = security_inode_init_security(inode, dir,
2938 shmem_initxattrs, NULL);
2939 if (error && error != -EOPNOTSUPP)
2943 dir->i_size += BOGO_DIRENT_SIZE;
2944 dir->i_ctime = dir->i_mtime = current_time(dir);
2945 d_instantiate(dentry, inode);
2946 dget(dentry); /* Extra count - pin the dentry in core */
2955 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2957 struct inode *inode;
2958 int error = -ENOSPC;
2960 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2962 error = security_inode_init_security(inode, dir,
2964 shmem_initxattrs, NULL);
2965 if (error && error != -EOPNOTSUPP)
2967 error = simple_acl_create(dir, inode);
2970 d_tmpfile(dentry, inode);
2978 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2982 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2988 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2991 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2997 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2999 struct inode *inode = d_inode(old_dentry);
3003 * No ordinary (disk based) filesystem counts links as inodes;
3004 * but each new link needs a new dentry, pinning lowmem, and
3005 * tmpfs dentries cannot be pruned until they are unlinked.
3006 * But if an O_TMPFILE file is linked into the tmpfs, the
3007 * first link must skip that, to get the accounting right.
3009 if (inode->i_nlink) {
3010 ret = shmem_reserve_inode(inode->i_sb, NULL);
3015 dir->i_size += BOGO_DIRENT_SIZE;
3016 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3018 ihold(inode); /* New dentry reference */
3019 dget(dentry); /* Extra pinning count for the created dentry */
3020 d_instantiate(dentry, inode);
3025 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3027 struct inode *inode = d_inode(dentry);
3029 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3030 shmem_free_inode(inode->i_sb);
3032 dir->i_size -= BOGO_DIRENT_SIZE;
3033 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3035 dput(dentry); /* Undo the count from "create" - this does all the work */
3039 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3041 if (!simple_empty(dentry))
3044 drop_nlink(d_inode(dentry));
3046 return shmem_unlink(dir, dentry);
3049 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3051 bool old_is_dir = d_is_dir(old_dentry);
3052 bool new_is_dir = d_is_dir(new_dentry);
3054 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3056 drop_nlink(old_dir);
3059 drop_nlink(new_dir);
3063 old_dir->i_ctime = old_dir->i_mtime =
3064 new_dir->i_ctime = new_dir->i_mtime =
3065 d_inode(old_dentry)->i_ctime =
3066 d_inode(new_dentry)->i_ctime = current_time(old_dir);
3071 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3073 struct dentry *whiteout;
3076 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3080 error = shmem_mknod(old_dir, whiteout,
3081 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3087 * Cheat and hash the whiteout while the old dentry is still in
3088 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3090 * d_lookup() will consistently find one of them at this point,
3091 * not sure which one, but that isn't even important.
3098 * The VFS layer already does all the dentry stuff for rename,
3099 * we just have to decrement the usage count for the target if
3100 * it exists so that the VFS layer correctly free's it when it
3103 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3105 struct inode *inode = d_inode(old_dentry);
3106 int they_are_dirs = S_ISDIR(inode->i_mode);
3108 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3111 if (flags & RENAME_EXCHANGE)
3112 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3114 if (!simple_empty(new_dentry))
3117 if (flags & RENAME_WHITEOUT) {
3120 error = shmem_whiteout(old_dir, old_dentry);
3125 if (d_really_is_positive(new_dentry)) {
3126 (void) shmem_unlink(new_dir, new_dentry);
3127 if (they_are_dirs) {
3128 drop_nlink(d_inode(new_dentry));
3129 drop_nlink(old_dir);
3131 } else if (they_are_dirs) {
3132 drop_nlink(old_dir);
3136 old_dir->i_size -= BOGO_DIRENT_SIZE;
3137 new_dir->i_size += BOGO_DIRENT_SIZE;
3138 old_dir->i_ctime = old_dir->i_mtime =
3139 new_dir->i_ctime = new_dir->i_mtime =
3140 inode->i_ctime = current_time(old_dir);
3144 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3148 struct inode *inode;
3151 len = strlen(symname) + 1;
3152 if (len > PAGE_SIZE)
3153 return -ENAMETOOLONG;
3155 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3160 error = security_inode_init_security(inode, dir, &dentry->d_name,
3161 shmem_initxattrs, NULL);
3162 if (error && error != -EOPNOTSUPP) {
3167 inode->i_size = len-1;
3168 if (len <= SHORT_SYMLINK_LEN) {
3169 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3170 if (!inode->i_link) {
3174 inode->i_op = &shmem_short_symlink_operations;
3176 inode_nohighmem(inode);
3177 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3182 inode->i_mapping->a_ops = &shmem_aops;
3183 inode->i_op = &shmem_symlink_inode_operations;
3184 memcpy(page_address(page), symname, len);
3185 SetPageUptodate(page);
3186 set_page_dirty(page);
3190 dir->i_size += BOGO_DIRENT_SIZE;
3191 dir->i_ctime = dir->i_mtime = current_time(dir);
3192 d_instantiate(dentry, inode);
3197 static void shmem_put_link(void *arg)
3199 mark_page_accessed(arg);
3203 static const char *shmem_get_link(struct dentry *dentry,
3204 struct inode *inode,
3205 struct delayed_call *done)
3207 struct page *page = NULL;
3210 page = find_get_page(inode->i_mapping, 0);
3212 return ERR_PTR(-ECHILD);
3213 if (!PageUptodate(page)) {
3215 return ERR_PTR(-ECHILD);
3218 error = shmem_getpage(inode, 0, &page, SGP_READ);
3220 return ERR_PTR(error);
3223 set_delayed_call(done, shmem_put_link, page);
3224 return page_address(page);
3227 #ifdef CONFIG_TMPFS_XATTR
3229 * Superblocks without xattr inode operations may get some security.* xattr
3230 * support from the LSM "for free". As soon as we have any other xattrs
3231 * like ACLs, we also need to implement the security.* handlers at
3232 * filesystem level, though.
3236 * Callback for security_inode_init_security() for acquiring xattrs.
3238 static int shmem_initxattrs(struct inode *inode,
3239 const struct xattr *xattr_array,
3242 struct shmem_inode_info *info = SHMEM_I(inode);
3243 const struct xattr *xattr;
3244 struct simple_xattr *new_xattr;
3247 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3248 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3252 len = strlen(xattr->name) + 1;
3253 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3255 if (!new_xattr->name) {
3260 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3261 XATTR_SECURITY_PREFIX_LEN);
3262 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3265 simple_xattr_list_add(&info->xattrs, new_xattr);
3271 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3272 struct dentry *unused, struct inode *inode,
3273 const char *name, void *buffer, size_t size)
3275 struct shmem_inode_info *info = SHMEM_I(inode);
3277 name = xattr_full_name(handler, name);
3278 return simple_xattr_get(&info->xattrs, name, buffer, size);
3281 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3282 struct dentry *unused, struct inode *inode,
3283 const char *name, const void *value,
3284 size_t size, int flags)
3286 struct shmem_inode_info *info = SHMEM_I(inode);
3288 name = xattr_full_name(handler, name);
3289 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3292 static const struct xattr_handler shmem_security_xattr_handler = {
3293 .prefix = XATTR_SECURITY_PREFIX,
3294 .get = shmem_xattr_handler_get,
3295 .set = shmem_xattr_handler_set,
3298 static const struct xattr_handler shmem_trusted_xattr_handler = {
3299 .prefix = XATTR_TRUSTED_PREFIX,
3300 .get = shmem_xattr_handler_get,
3301 .set = shmem_xattr_handler_set,
3304 static const struct xattr_handler *shmem_xattr_handlers[] = {
3305 #ifdef CONFIG_TMPFS_POSIX_ACL
3306 &posix_acl_access_xattr_handler,
3307 &posix_acl_default_xattr_handler,
3309 &shmem_security_xattr_handler,
3310 &shmem_trusted_xattr_handler,
3314 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3316 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3317 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3319 #endif /* CONFIG_TMPFS_XATTR */
3321 static const struct inode_operations shmem_short_symlink_operations = {
3322 .get_link = simple_get_link,
3323 #ifdef CONFIG_TMPFS_XATTR
3324 .listxattr = shmem_listxattr,
3328 static const struct inode_operations shmem_symlink_inode_operations = {
3329 .get_link = shmem_get_link,
3330 #ifdef CONFIG_TMPFS_XATTR
3331 .listxattr = shmem_listxattr,
3335 static struct dentry *shmem_get_parent(struct dentry *child)
3337 return ERR_PTR(-ESTALE);
3340 static int shmem_match(struct inode *ino, void *vfh)
3344 inum = (inum << 32) | fh[1];
3345 return ino->i_ino == inum && fh[0] == ino->i_generation;
3348 /* Find any alias of inode, but prefer a hashed alias */
3349 static struct dentry *shmem_find_alias(struct inode *inode)
3351 struct dentry *alias = d_find_alias(inode);
3353 return alias ?: d_find_any_alias(inode);
3357 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3358 struct fid *fid, int fh_len, int fh_type)
3360 struct inode *inode;
3361 struct dentry *dentry = NULL;
3368 inum = (inum << 32) | fid->raw[1];
3370 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3371 shmem_match, fid->raw);
3373 dentry = shmem_find_alias(inode);
3380 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3381 struct inode *parent)
3385 return FILEID_INVALID;
3388 if (inode_unhashed(inode)) {
3389 /* Unfortunately insert_inode_hash is not idempotent,
3390 * so as we hash inodes here rather than at creation
3391 * time, we need a lock to ensure we only try
3394 static DEFINE_SPINLOCK(lock);
3396 if (inode_unhashed(inode))
3397 __insert_inode_hash(inode,
3398 inode->i_ino + inode->i_generation);
3402 fh[0] = inode->i_generation;
3403 fh[1] = inode->i_ino;
3404 fh[2] = ((__u64)inode->i_ino) >> 32;
3410 static const struct export_operations shmem_export_ops = {
3411 .get_parent = shmem_get_parent,
3412 .encode_fh = shmem_encode_fh,
3413 .fh_to_dentry = shmem_fh_to_dentry,
3429 static const struct constant_table shmem_param_enums_huge[] = {
3430 {"never", SHMEM_HUGE_NEVER },
3431 {"always", SHMEM_HUGE_ALWAYS },
3432 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3433 {"advise", SHMEM_HUGE_ADVISE },
3437 const struct fs_parameter_spec shmem_fs_parameters[] = {
3438 fsparam_u32 ("gid", Opt_gid),
3439 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3440 fsparam_u32oct("mode", Opt_mode),
3441 fsparam_string("mpol", Opt_mpol),
3442 fsparam_string("nr_blocks", Opt_nr_blocks),
3443 fsparam_string("nr_inodes", Opt_nr_inodes),
3444 fsparam_string("size", Opt_size),
3445 fsparam_u32 ("uid", Opt_uid),
3446 fsparam_flag ("inode32", Opt_inode32),
3447 fsparam_flag ("inode64", Opt_inode64),
3451 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3453 struct shmem_options *ctx = fc->fs_private;
3454 struct fs_parse_result result;
3455 unsigned long long size;
3461 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3467 size = memparse(param->string, &rest);
3469 size <<= PAGE_SHIFT;
3470 size *= totalram_pages();
3476 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3477 ctx->seen |= SHMEM_SEEN_BLOCKS;
3480 ctx->blocks = memparse(param->string, &rest);
3483 ctx->seen |= SHMEM_SEEN_BLOCKS;
3486 ctx->inodes = memparse(param->string, &rest);
3489 ctx->seen |= SHMEM_SEEN_INODES;
3492 ctx->mode = result.uint_32 & 07777;
3495 kuid = make_kuid(current_user_ns(), result.uint_32);
3496 if (!uid_valid(kuid))
3500 * The requested uid must be representable in the
3501 * filesystem's idmapping.
3503 if (!kuid_has_mapping(fc->user_ns, kuid))
3509 kgid = make_kgid(current_user_ns(), result.uint_32);
3510 if (!gid_valid(kgid))
3514 * The requested gid must be representable in the
3515 * filesystem's idmapping.
3517 if (!kgid_has_mapping(fc->user_ns, kgid))
3523 ctx->huge = result.uint_32;
3524 if (ctx->huge != SHMEM_HUGE_NEVER &&
3525 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3526 has_transparent_hugepage()))
3527 goto unsupported_parameter;
3528 ctx->seen |= SHMEM_SEEN_HUGE;
3531 if (IS_ENABLED(CONFIG_NUMA)) {
3532 mpol_put(ctx->mpol);
3534 if (mpol_parse_str(param->string, &ctx->mpol))
3538 goto unsupported_parameter;
3540 ctx->full_inums = false;
3541 ctx->seen |= SHMEM_SEEN_INUMS;
3544 if (sizeof(ino_t) < 8) {
3546 "Cannot use inode64 with <64bit inums in kernel\n");
3548 ctx->full_inums = true;
3549 ctx->seen |= SHMEM_SEEN_INUMS;
3554 unsupported_parameter:
3555 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3557 return invalfc(fc, "Bad value for '%s'", param->key);
3560 static int shmem_parse_options(struct fs_context *fc, void *data)
3562 char *options = data;
3565 int err = security_sb_eat_lsm_opts(options, &fc->security);
3570 while (options != NULL) {
3571 char *this_char = options;
3574 * NUL-terminate this option: unfortunately,
3575 * mount options form a comma-separated list,
3576 * but mpol's nodelist may also contain commas.
3578 options = strchr(options, ',');
3579 if (options == NULL)
3582 if (!isdigit(*options)) {
3588 char *value = strchr(this_char,'=');
3594 len = strlen(value);
3596 err = vfs_parse_fs_string(fc, this_char, value, len);
3605 * Reconfigure a shmem filesystem.
3607 * Note that we disallow change from limited->unlimited blocks/inodes while any
3608 * are in use; but we must separately disallow unlimited->limited, because in
3609 * that case we have no record of how much is already in use.
3611 static int shmem_reconfigure(struct fs_context *fc)
3613 struct shmem_options *ctx = fc->fs_private;
3614 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3615 unsigned long inodes;
3618 spin_lock(&sbinfo->stat_lock);
3619 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3620 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3621 if (!sbinfo->max_blocks) {
3622 err = "Cannot retroactively limit size";
3625 if (percpu_counter_compare(&sbinfo->used_blocks,
3627 err = "Too small a size for current use";
3631 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3632 if (!sbinfo->max_inodes) {
3633 err = "Cannot retroactively limit inodes";
3636 if (ctx->inodes < inodes) {
3637 err = "Too few inodes for current use";
3642 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3643 sbinfo->next_ino > UINT_MAX) {
3644 err = "Current inum too high to switch to 32-bit inums";
3648 if (ctx->seen & SHMEM_SEEN_HUGE)
3649 sbinfo->huge = ctx->huge;
3650 if (ctx->seen & SHMEM_SEEN_INUMS)
3651 sbinfo->full_inums = ctx->full_inums;
3652 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3653 sbinfo->max_blocks = ctx->blocks;
3654 if (ctx->seen & SHMEM_SEEN_INODES) {
3655 sbinfo->max_inodes = ctx->inodes;
3656 sbinfo->free_inodes = ctx->inodes - inodes;
3660 * Preserve previous mempolicy unless mpol remount option was specified.
3663 mpol_put(sbinfo->mpol);
3664 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3667 spin_unlock(&sbinfo->stat_lock);
3670 spin_unlock(&sbinfo->stat_lock);
3671 return invalfc(fc, "%s", err);
3674 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3676 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3678 if (sbinfo->max_blocks != shmem_default_max_blocks())
3679 seq_printf(seq, ",size=%luk",
3680 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3681 if (sbinfo->max_inodes != shmem_default_max_inodes())
3682 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3683 if (sbinfo->mode != (0777 | S_ISVTX))
3684 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3685 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3686 seq_printf(seq, ",uid=%u",
3687 from_kuid_munged(&init_user_ns, sbinfo->uid));
3688 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3689 seq_printf(seq, ",gid=%u",
3690 from_kgid_munged(&init_user_ns, sbinfo->gid));
3693 * Showing inode{64,32} might be useful even if it's the system default,
3694 * since then people don't have to resort to checking both here and
3695 * /proc/config.gz to confirm 64-bit inums were successfully applied
3696 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3698 * We hide it when inode64 isn't the default and we are using 32-bit
3699 * inodes, since that probably just means the feature isn't even under
3704 * +-----------------+-----------------+
3705 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3706 * +------------------+-----------------+-----------------+
3707 * | full_inums=true | show | show |
3708 * | full_inums=false | show | hide |
3709 * +------------------+-----------------+-----------------+
3712 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3713 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3714 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3715 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3717 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3719 shmem_show_mpol(seq, sbinfo->mpol);
3723 #endif /* CONFIG_TMPFS */
3725 static void shmem_put_super(struct super_block *sb)
3727 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3729 free_percpu(sbinfo->ino_batch);
3730 percpu_counter_destroy(&sbinfo->used_blocks);
3731 mpol_put(sbinfo->mpol);
3733 sb->s_fs_info = NULL;
3736 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3738 struct shmem_options *ctx = fc->fs_private;
3739 struct inode *inode;
3740 struct shmem_sb_info *sbinfo;
3743 /* Round up to L1_CACHE_BYTES to resist false sharing */
3744 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3745 L1_CACHE_BYTES), GFP_KERNEL);
3749 sb->s_fs_info = sbinfo;
3753 * Per default we only allow half of the physical ram per
3754 * tmpfs instance, limiting inodes to one per page of lowmem;
3755 * but the internal instance is left unlimited.
3757 if (!(sb->s_flags & SB_KERNMOUNT)) {
3758 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3759 ctx->blocks = shmem_default_max_blocks();
3760 if (!(ctx->seen & SHMEM_SEEN_INODES))
3761 ctx->inodes = shmem_default_max_inodes();
3762 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3763 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3765 sb->s_flags |= SB_NOUSER;
3767 sb->s_export_op = &shmem_export_ops;
3768 sb->s_flags |= SB_NOSEC;
3770 sb->s_flags |= SB_NOUSER;
3772 sbinfo->max_blocks = ctx->blocks;
3773 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3774 if (sb->s_flags & SB_KERNMOUNT) {
3775 sbinfo->ino_batch = alloc_percpu(ino_t);
3776 if (!sbinfo->ino_batch)
3779 sbinfo->uid = ctx->uid;
3780 sbinfo->gid = ctx->gid;
3781 sbinfo->full_inums = ctx->full_inums;
3782 sbinfo->mode = ctx->mode;
3783 sbinfo->huge = ctx->huge;
3784 sbinfo->mpol = ctx->mpol;
3787 spin_lock_init(&sbinfo->stat_lock);
3788 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3790 spin_lock_init(&sbinfo->shrinklist_lock);
3791 INIT_LIST_HEAD(&sbinfo->shrinklist);
3793 sb->s_maxbytes = MAX_LFS_FILESIZE;
3794 sb->s_blocksize = PAGE_SIZE;
3795 sb->s_blocksize_bits = PAGE_SHIFT;
3796 sb->s_magic = TMPFS_MAGIC;
3797 sb->s_op = &shmem_ops;
3798 sb->s_time_gran = 1;
3799 #ifdef CONFIG_TMPFS_XATTR
3800 sb->s_xattr = shmem_xattr_handlers;
3802 #ifdef CONFIG_TMPFS_POSIX_ACL
3803 sb->s_flags |= SB_POSIXACL;
3805 uuid_gen(&sb->s_uuid);
3807 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3810 inode->i_uid = sbinfo->uid;
3811 inode->i_gid = sbinfo->gid;
3812 sb->s_root = d_make_root(inode);
3818 shmem_put_super(sb);
3822 static int shmem_get_tree(struct fs_context *fc)
3824 return get_tree_nodev(fc, shmem_fill_super);
3827 static void shmem_free_fc(struct fs_context *fc)
3829 struct shmem_options *ctx = fc->fs_private;
3832 mpol_put(ctx->mpol);
3837 static const struct fs_context_operations shmem_fs_context_ops = {
3838 .free = shmem_free_fc,
3839 .get_tree = shmem_get_tree,
3841 .parse_monolithic = shmem_parse_options,
3842 .parse_param = shmem_parse_one,
3843 .reconfigure = shmem_reconfigure,
3847 static struct kmem_cache *shmem_inode_cachep;
3849 static struct inode *shmem_alloc_inode(struct super_block *sb)
3851 struct shmem_inode_info *info;
3852 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3855 return &info->vfs_inode;
3858 static void shmem_free_in_core_inode(struct inode *inode)
3860 if (S_ISLNK(inode->i_mode))
3861 kfree(inode->i_link);
3862 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3865 static void shmem_destroy_inode(struct inode *inode)
3867 if (S_ISREG(inode->i_mode))
3868 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3871 static void shmem_init_inode(void *foo)
3873 struct shmem_inode_info *info = foo;
3874 inode_init_once(&info->vfs_inode);
3877 static void shmem_init_inodecache(void)
3879 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3880 sizeof(struct shmem_inode_info),
3881 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3884 static void shmem_destroy_inodecache(void)
3886 kmem_cache_destroy(shmem_inode_cachep);
3889 static const struct address_space_operations shmem_aops = {
3890 .writepage = shmem_writepage,
3891 .set_page_dirty = __set_page_dirty_no_writeback,
3893 .write_begin = shmem_write_begin,
3894 .write_end = shmem_write_end,
3896 #ifdef CONFIG_MIGRATION
3897 .migratepage = migrate_page,
3899 .error_remove_page = generic_error_remove_page,
3902 static const struct file_operations shmem_file_operations = {
3904 .get_unmapped_area = shmem_get_unmapped_area,
3906 .llseek = shmem_file_llseek,
3907 .read_iter = shmem_file_read_iter,
3908 .write_iter = generic_file_write_iter,
3909 .fsync = noop_fsync,
3910 .splice_read = generic_file_splice_read,
3911 .splice_write = iter_file_splice_write,
3912 .fallocate = shmem_fallocate,
3916 static const struct inode_operations shmem_inode_operations = {
3917 .getattr = shmem_getattr,
3918 .setattr = shmem_setattr,
3919 #ifdef CONFIG_TMPFS_XATTR
3920 .listxattr = shmem_listxattr,
3921 .set_acl = simple_set_acl,
3925 static const struct inode_operations shmem_dir_inode_operations = {
3927 .create = shmem_create,
3928 .lookup = simple_lookup,
3930 .unlink = shmem_unlink,
3931 .symlink = shmem_symlink,
3932 .mkdir = shmem_mkdir,
3933 .rmdir = shmem_rmdir,
3934 .mknod = shmem_mknod,
3935 .rename = shmem_rename2,
3936 .tmpfile = shmem_tmpfile,
3938 #ifdef CONFIG_TMPFS_XATTR
3939 .listxattr = shmem_listxattr,
3941 #ifdef CONFIG_TMPFS_POSIX_ACL
3942 .setattr = shmem_setattr,
3943 .set_acl = simple_set_acl,
3947 static const struct inode_operations shmem_special_inode_operations = {
3948 #ifdef CONFIG_TMPFS_XATTR
3949 .listxattr = shmem_listxattr,
3951 #ifdef CONFIG_TMPFS_POSIX_ACL
3952 .setattr = shmem_setattr,
3953 .set_acl = simple_set_acl,
3957 static const struct super_operations shmem_ops = {
3958 .alloc_inode = shmem_alloc_inode,
3959 .free_inode = shmem_free_in_core_inode,
3960 .destroy_inode = shmem_destroy_inode,
3962 .statfs = shmem_statfs,
3963 .show_options = shmem_show_options,
3965 .evict_inode = shmem_evict_inode,
3966 .drop_inode = generic_delete_inode,
3967 .put_super = shmem_put_super,
3968 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3969 .nr_cached_objects = shmem_unused_huge_count,
3970 .free_cached_objects = shmem_unused_huge_scan,
3974 static const struct vm_operations_struct shmem_vm_ops = {
3975 .fault = shmem_fault,
3976 .map_pages = filemap_map_pages,
3978 .set_policy = shmem_set_policy,
3979 .get_policy = shmem_get_policy,
3983 int shmem_init_fs_context(struct fs_context *fc)
3985 struct shmem_options *ctx;
3987 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3991 ctx->mode = 0777 | S_ISVTX;
3992 ctx->uid = current_fsuid();
3993 ctx->gid = current_fsgid();
3995 fc->fs_private = ctx;
3996 fc->ops = &shmem_fs_context_ops;
4000 static struct file_system_type shmem_fs_type = {
4001 .owner = THIS_MODULE,
4003 .init_fs_context = shmem_init_fs_context,
4005 .parameters = shmem_fs_parameters,
4007 .kill_sb = kill_litter_super,
4008 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
4011 int __init shmem_init(void)
4015 shmem_init_inodecache();
4017 error = register_filesystem(&shmem_fs_type);
4019 pr_err("Could not register tmpfs\n");
4023 shm_mnt = kern_mount(&shmem_fs_type);
4024 if (IS_ERR(shm_mnt)) {
4025 error = PTR_ERR(shm_mnt);
4026 pr_err("Could not kern_mount tmpfs\n");
4030 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4031 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4032 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4034 shmem_huge = 0; /* just in case it was patched */
4039 unregister_filesystem(&shmem_fs_type);
4041 shmem_destroy_inodecache();
4042 shm_mnt = ERR_PTR(error);
4046 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4047 static ssize_t shmem_enabled_show(struct kobject *kobj,
4048 struct kobj_attribute *attr, char *buf)
4050 static const int values[] = {
4052 SHMEM_HUGE_WITHIN_SIZE,
4060 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
4061 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
4063 count += sprintf(buf + count, fmt,
4064 shmem_format_huge(values[i]));
4066 buf[count - 1] = '\n';
4070 static ssize_t shmem_enabled_store(struct kobject *kobj,
4071 struct kobj_attribute *attr, const char *buf, size_t count)
4076 if (count + 1 > sizeof(tmp))
4078 memcpy(tmp, buf, count);
4080 if (count && tmp[count - 1] == '\n')
4081 tmp[count - 1] = '\0';
4083 huge = shmem_parse_huge(tmp);
4084 if (huge == -EINVAL)
4086 if (!has_transparent_hugepage() &&
4087 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4091 if (shmem_huge > SHMEM_HUGE_DENY)
4092 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4096 struct kobj_attribute shmem_enabled_attr =
4097 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4098 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4100 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4101 bool shmem_huge_enabled(struct vm_area_struct *vma)
4103 struct inode *inode = file_inode(vma->vm_file);
4104 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4108 if (!transhuge_vma_enabled(vma, vma->vm_flags))
4110 if (shmem_huge == SHMEM_HUGE_FORCE)
4112 if (shmem_huge == SHMEM_HUGE_DENY)
4114 switch (sbinfo->huge) {
4115 case SHMEM_HUGE_NEVER:
4117 case SHMEM_HUGE_ALWAYS:
4119 case SHMEM_HUGE_WITHIN_SIZE:
4120 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4121 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4122 if (i_size >= HPAGE_PMD_SIZE &&
4123 i_size >> PAGE_SHIFT >= off)
4126 case SHMEM_HUGE_ADVISE:
4127 /* TODO: implement fadvise() hints */
4128 return (vma->vm_flags & VM_HUGEPAGE);
4134 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4136 #else /* !CONFIG_SHMEM */
4139 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4141 * This is intended for small system where the benefits of the full
4142 * shmem code (swap-backed and resource-limited) are outweighed by
4143 * their complexity. On systems without swap this code should be
4144 * effectively equivalent, but much lighter weight.
4147 static struct file_system_type shmem_fs_type = {
4149 .init_fs_context = ramfs_init_fs_context,
4150 .parameters = ramfs_fs_parameters,
4151 .kill_sb = ramfs_kill_sb,
4152 .fs_flags = FS_USERNS_MOUNT,
4155 int __init shmem_init(void)
4157 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4159 shm_mnt = kern_mount(&shmem_fs_type);
4160 BUG_ON(IS_ERR(shm_mnt));
4165 int shmem_unuse(unsigned int type, bool frontswap,
4166 unsigned long *fs_pages_to_unuse)
4171 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4176 void shmem_unlock_mapping(struct address_space *mapping)
4181 unsigned long shmem_get_unmapped_area(struct file *file,
4182 unsigned long addr, unsigned long len,
4183 unsigned long pgoff, unsigned long flags)
4185 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4189 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4191 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4193 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4195 #define shmem_vm_ops generic_file_vm_ops
4196 #define shmem_file_operations ramfs_file_operations
4197 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4198 #define shmem_acct_size(flags, size) 0
4199 #define shmem_unacct_size(flags, size) do {} while (0)
4201 #endif /* CONFIG_SHMEM */
4205 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4206 unsigned long flags, unsigned int i_flags)
4208 struct inode *inode;
4212 return ERR_CAST(mnt);
4214 if (size < 0 || size > MAX_LFS_FILESIZE)
4215 return ERR_PTR(-EINVAL);
4217 if (shmem_acct_size(flags, size))
4218 return ERR_PTR(-ENOMEM);
4220 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4222 if (unlikely(!inode)) {
4223 shmem_unacct_size(flags, size);
4224 return ERR_PTR(-ENOSPC);
4226 inode->i_flags |= i_flags;
4227 inode->i_size = size;
4228 clear_nlink(inode); /* It is unlinked */
4229 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4231 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4232 &shmem_file_operations);
4239 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4240 * kernel internal. There will be NO LSM permission checks against the
4241 * underlying inode. So users of this interface must do LSM checks at a
4242 * higher layer. The users are the big_key and shm implementations. LSM
4243 * checks are provided at the key or shm level rather than the inode.
4244 * @name: name for dentry (to be seen in /proc/<pid>/maps
4245 * @size: size to be set for the file
4246 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4248 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4250 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4254 * shmem_file_setup - get an unlinked file living in tmpfs
4255 * @name: name for dentry (to be seen in /proc/<pid>/maps
4256 * @size: size to be set for the file
4257 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4259 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4261 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4263 EXPORT_SYMBOL_GPL(shmem_file_setup);
4266 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4267 * @mnt: the tmpfs mount where the file will be created
4268 * @name: name for dentry (to be seen in /proc/<pid>/maps
4269 * @size: size to be set for the file
4270 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4272 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4273 loff_t size, unsigned long flags)
4275 return __shmem_file_setup(mnt, name, size, flags, 0);
4277 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4280 * shmem_zero_setup - setup a shared anonymous mapping
4281 * @vma: the vma to be mmapped is prepared by do_mmap
4283 int shmem_zero_setup(struct vm_area_struct *vma)
4286 loff_t size = vma->vm_end - vma->vm_start;
4289 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4290 * between XFS directory reading and selinux: since this file is only
4291 * accessible to the user through its mapping, use S_PRIVATE flag to
4292 * bypass file security, in the same way as shmem_kernel_file_setup().
4294 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4296 return PTR_ERR(file);
4300 vma->vm_file = file;
4301 vma->vm_ops = &shmem_vm_ops;
4303 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4304 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4305 (vma->vm_end & HPAGE_PMD_MASK)) {
4306 khugepaged_enter(vma, vma->vm_flags);
4313 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4314 * @mapping: the page's address_space
4315 * @index: the page index
4316 * @gfp: the page allocator flags to use if allocating
4318 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4319 * with any new page allocations done using the specified allocation flags.
4320 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4321 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4322 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4324 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4325 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4327 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4328 pgoff_t index, gfp_t gfp)
4331 struct inode *inode = mapping->host;
4335 BUG_ON(mapping->a_ops != &shmem_aops);
4336 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4337 gfp, NULL, NULL, NULL);
4339 page = ERR_PTR(error);
4345 * The tiny !SHMEM case uses ramfs without swap
4347 return read_cache_page_gfp(mapping, index, gfp);
4350 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);