2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
42 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
44 static struct vfsmount *shm_mnt;
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/blkdev.h>
64 #include <linux/pagevec.h>
65 #include <linux/percpu_counter.h>
66 #include <linux/falloc.h>
67 #include <linux/splice.h>
68 #include <linux/security.h>
69 #include <linux/swapops.h>
70 #include <linux/mempolicy.h>
71 #include <linux/namei.h>
72 #include <linux/ctype.h>
73 #include <linux/migrate.h>
74 #include <linux/highmem.h>
75 #include <linux/seq_file.h>
76 #include <linux/magic.h>
77 #include <linux/syscalls.h>
78 #include <linux/fcntl.h>
79 #include <uapi/linux/memfd.h>
80 #include <linux/userfaultfd_k.h>
81 #include <linux/rmap.h>
82 #include <linux/uuid.h>
84 #include <linux/uaccess.h>
88 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
91 /* Pretend that each entry is of this size in directory's i_size */
92 #define BOGO_DIRENT_SIZE 20
94 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
95 #define SHORT_SYMLINK_LEN 128
98 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
99 * inode->i_private (with i_mutex making sure that it has only one user at
100 * a time): we would prefer not to enlarge the shmem inode just for that.
102 struct shmem_falloc {
103 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
104 pgoff_t start; /* start of range currently being fallocated */
105 pgoff_t next; /* the next page offset to be fallocated */
106 pgoff_t nr_falloced; /* how many new pages have been fallocated */
107 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
110 struct shmem_options {
111 unsigned long long blocks;
112 unsigned long long inodes;
113 struct mempolicy *mpol;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
123 #define SHMEM_SEEN_INUMS 8
127 static unsigned long shmem_default_max_blocks(void)
129 return totalram_pages() / 2;
132 static unsigned long shmem_default_max_inodes(void)
134 unsigned long nr_pages = totalram_pages();
136 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
140 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
141 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
142 struct shmem_inode_info *info, pgoff_t index);
143 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 vm_fault_t *fault_type);
147 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
148 struct page **pagep, enum sgp_type sgp,
149 gfp_t gfp, struct vm_area_struct *vma,
150 struct vm_fault *vmf, vm_fault_t *fault_type);
152 int shmem_getpage(struct inode *inode, pgoff_t index,
153 struct page **pagep, enum sgp_type sgp)
155 return shmem_getpage_gfp(inode, index, pagep, sgp,
156 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
159 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
161 return sb->s_fs_info;
165 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
166 * for shared memory and for shared anonymous (/dev/zero) mappings
167 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
168 * consistent with the pre-accounting of private mappings ...
170 static inline int shmem_acct_size(unsigned long flags, loff_t size)
172 return (flags & VM_NORESERVE) ?
173 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
176 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
178 if (!(flags & VM_NORESERVE))
179 vm_unacct_memory(VM_ACCT(size));
182 static inline int shmem_reacct_size(unsigned long flags,
183 loff_t oldsize, loff_t newsize)
185 if (!(flags & VM_NORESERVE)) {
186 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
187 return security_vm_enough_memory_mm(current->mm,
188 VM_ACCT(newsize) - VM_ACCT(oldsize));
189 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
190 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
196 * ... whereas tmpfs objects are accounted incrementally as
197 * pages are allocated, in order to allow large sparse files.
198 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
199 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201 static inline int shmem_acct_block(unsigned long flags, long pages)
203 if (!(flags & VM_NORESERVE))
206 return security_vm_enough_memory_mm(current->mm,
207 pages * VM_ACCT(PAGE_SIZE));
210 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
212 if (flags & VM_NORESERVE)
213 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
216 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
218 struct shmem_inode_info *info = SHMEM_I(inode);
219 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
221 if (shmem_acct_block(info->flags, pages))
224 if (sbinfo->max_blocks) {
225 if (percpu_counter_compare(&sbinfo->used_blocks,
226 sbinfo->max_blocks - pages) > 0)
228 percpu_counter_add(&sbinfo->used_blocks, pages);
234 shmem_unacct_blocks(info->flags, pages);
238 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
240 struct shmem_inode_info *info = SHMEM_I(inode);
241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
243 if (sbinfo->max_blocks)
244 percpu_counter_sub(&sbinfo->used_blocks, pages);
245 shmem_unacct_blocks(info->flags, pages);
248 static const struct super_operations shmem_ops;
249 const struct address_space_operations shmem_aops;
250 static const struct file_operations shmem_file_operations;
251 static const struct inode_operations shmem_inode_operations;
252 static const struct inode_operations shmem_dir_inode_operations;
253 static const struct inode_operations shmem_special_inode_operations;
254 static const struct vm_operations_struct shmem_vm_ops;
255 static struct file_system_type shmem_fs_type;
257 bool vma_is_shmem(struct vm_area_struct *vma)
259 return vma->vm_ops == &shmem_vm_ops;
262 static LIST_HEAD(shmem_swaplist);
263 static DEFINE_MUTEX(shmem_swaplist_mutex);
266 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
267 * produces a novel ino for the newly allocated inode.
269 * It may also be called when making a hard link to permit the space needed by
270 * each dentry. However, in that case, no new inode number is needed since that
271 * internally draws from another pool of inode numbers (currently global
272 * get_next_ino()). This case is indicated by passing NULL as inop.
274 #define SHMEM_INO_BATCH 1024
275 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
277 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
280 if (!(sb->s_flags & SB_KERNMOUNT)) {
281 spin_lock(&sbinfo->stat_lock);
282 if (sbinfo->max_inodes) {
283 if (!sbinfo->free_inodes) {
284 spin_unlock(&sbinfo->stat_lock);
287 sbinfo->free_inodes--;
290 ino = sbinfo->next_ino++;
291 if (unlikely(is_zero_ino(ino)))
292 ino = sbinfo->next_ino++;
293 if (unlikely(!sbinfo->full_inums &&
296 * Emulate get_next_ino uint wraparound for
299 if (IS_ENABLED(CONFIG_64BIT))
300 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
301 __func__, MINOR(sb->s_dev));
302 sbinfo->next_ino = 1;
303 ino = sbinfo->next_ino++;
307 spin_unlock(&sbinfo->stat_lock);
310 * __shmem_file_setup, one of our callers, is lock-free: it
311 * doesn't hold stat_lock in shmem_reserve_inode since
312 * max_inodes is always 0, and is called from potentially
313 * unknown contexts. As such, use a per-cpu batched allocator
314 * which doesn't require the per-sb stat_lock unless we are at
315 * the batch boundary.
317 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
318 * shmem mounts are not exposed to userspace, so we don't need
319 * to worry about things like glibc compatibility.
322 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
324 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
325 spin_lock(&sbinfo->stat_lock);
326 ino = sbinfo->next_ino;
327 sbinfo->next_ino += SHMEM_INO_BATCH;
328 spin_unlock(&sbinfo->stat_lock);
329 if (unlikely(is_zero_ino(ino)))
340 static void shmem_free_inode(struct super_block *sb)
342 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
343 if (sbinfo->max_inodes) {
344 spin_lock(&sbinfo->stat_lock);
345 sbinfo->free_inodes++;
346 spin_unlock(&sbinfo->stat_lock);
351 * shmem_recalc_inode - recalculate the block usage of an inode
352 * @inode: inode to recalc
354 * We have to calculate the free blocks since the mm can drop
355 * undirtied hole pages behind our back.
357 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
358 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
360 * It has to be called with the spinlock held.
362 static void shmem_recalc_inode(struct inode *inode)
364 struct shmem_inode_info *info = SHMEM_I(inode);
367 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
369 info->alloced -= freed;
370 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
371 shmem_inode_unacct_blocks(inode, freed);
375 bool shmem_charge(struct inode *inode, long pages)
377 struct shmem_inode_info *info = SHMEM_I(inode);
380 if (!shmem_inode_acct_block(inode, pages))
383 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
384 inode->i_mapping->nrpages += pages;
386 spin_lock_irqsave(&info->lock, flags);
387 info->alloced += pages;
388 inode->i_blocks += pages * BLOCKS_PER_PAGE;
389 shmem_recalc_inode(inode);
390 spin_unlock_irqrestore(&info->lock, flags);
395 void shmem_uncharge(struct inode *inode, long pages)
397 struct shmem_inode_info *info = SHMEM_I(inode);
400 /* nrpages adjustment done by __delete_from_page_cache() or caller */
402 spin_lock_irqsave(&info->lock, flags);
403 info->alloced -= pages;
404 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
405 shmem_recalc_inode(inode);
406 spin_unlock_irqrestore(&info->lock, flags);
408 shmem_inode_unacct_blocks(inode, pages);
412 * Replace item expected in xarray by a new item, while holding xa_lock.
414 static int shmem_replace_entry(struct address_space *mapping,
415 pgoff_t index, void *expected, void *replacement)
417 XA_STATE(xas, &mapping->i_pages, index);
420 VM_BUG_ON(!expected);
421 VM_BUG_ON(!replacement);
422 item = xas_load(&xas);
423 if (item != expected)
425 xas_store(&xas, replacement);
430 * Sometimes, before we decide whether to proceed or to fail, we must check
431 * that an entry was not already brought back from swap by a racing thread.
433 * Checking page is not enough: by the time a SwapCache page is locked, it
434 * might be reused, and again be SwapCache, using the same swap as before.
436 static bool shmem_confirm_swap(struct address_space *mapping,
437 pgoff_t index, swp_entry_t swap)
439 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
443 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
446 * disables huge pages for the mount;
448 * enables huge pages for the mount;
449 * SHMEM_HUGE_WITHIN_SIZE:
450 * only allocate huge pages if the page will be fully within i_size,
451 * also respect fadvise()/madvise() hints;
453 * only allocate huge pages if requested with fadvise()/madvise();
456 #define SHMEM_HUGE_NEVER 0
457 #define SHMEM_HUGE_ALWAYS 1
458 #define SHMEM_HUGE_WITHIN_SIZE 2
459 #define SHMEM_HUGE_ADVISE 3
463 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
466 * disables huge on shm_mnt and all mounts, for emergency use;
468 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
471 #define SHMEM_HUGE_DENY (-1)
472 #define SHMEM_HUGE_FORCE (-2)
474 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
475 /* ifdef here to avoid bloating shmem.o when not necessary */
477 static int shmem_huge __read_mostly;
479 #if defined(CONFIG_SYSFS)
480 static int shmem_parse_huge(const char *str)
482 if (!strcmp(str, "never"))
483 return SHMEM_HUGE_NEVER;
484 if (!strcmp(str, "always"))
485 return SHMEM_HUGE_ALWAYS;
486 if (!strcmp(str, "within_size"))
487 return SHMEM_HUGE_WITHIN_SIZE;
488 if (!strcmp(str, "advise"))
489 return SHMEM_HUGE_ADVISE;
490 if (!strcmp(str, "deny"))
491 return SHMEM_HUGE_DENY;
492 if (!strcmp(str, "force"))
493 return SHMEM_HUGE_FORCE;
498 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
499 static const char *shmem_format_huge(int huge)
502 case SHMEM_HUGE_NEVER:
504 case SHMEM_HUGE_ALWAYS:
506 case SHMEM_HUGE_WITHIN_SIZE:
507 return "within_size";
508 case SHMEM_HUGE_ADVISE:
510 case SHMEM_HUGE_DENY:
512 case SHMEM_HUGE_FORCE:
521 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
522 struct shrink_control *sc, unsigned long nr_to_split)
524 LIST_HEAD(list), *pos, *next;
525 LIST_HEAD(to_remove);
527 struct shmem_inode_info *info;
529 unsigned long batch = sc ? sc->nr_to_scan : 128;
530 int removed = 0, split = 0;
532 if (list_empty(&sbinfo->shrinklist))
535 spin_lock(&sbinfo->shrinklist_lock);
536 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
537 info = list_entry(pos, struct shmem_inode_info, shrinklist);
540 inode = igrab(&info->vfs_inode);
542 /* inode is about to be evicted */
544 list_del_init(&info->shrinklist);
549 /* Check if there's anything to gain */
550 if (round_up(inode->i_size, PAGE_SIZE) ==
551 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
552 list_move(&info->shrinklist, &to_remove);
557 list_move(&info->shrinklist, &list);
562 spin_unlock(&sbinfo->shrinklist_lock);
564 list_for_each_safe(pos, next, &to_remove) {
565 info = list_entry(pos, struct shmem_inode_info, shrinklist);
566 inode = &info->vfs_inode;
567 list_del_init(&info->shrinklist);
571 list_for_each_safe(pos, next, &list) {
574 info = list_entry(pos, struct shmem_inode_info, shrinklist);
575 inode = &info->vfs_inode;
577 if (nr_to_split && split >= nr_to_split)
580 page = find_get_page(inode->i_mapping,
581 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
585 /* No huge page at the end of the file: nothing to split */
586 if (!PageTransHuge(page)) {
592 * Leave the inode on the list if we failed to lock
593 * the page at this time.
595 * Waiting for the lock may lead to deadlock in the
598 if (!trylock_page(page)) {
603 ret = split_huge_page(page);
607 /* If split failed leave the inode on the list */
613 list_del_init(&info->shrinklist);
619 spin_lock(&sbinfo->shrinklist_lock);
620 list_splice_tail(&list, &sbinfo->shrinklist);
621 sbinfo->shrinklist_len -= removed;
622 spin_unlock(&sbinfo->shrinklist_lock);
627 static long shmem_unused_huge_scan(struct super_block *sb,
628 struct shrink_control *sc)
630 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
632 if (!READ_ONCE(sbinfo->shrinklist_len))
635 return shmem_unused_huge_shrink(sbinfo, sc, 0);
638 static long shmem_unused_huge_count(struct super_block *sb,
639 struct shrink_control *sc)
641 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
642 return READ_ONCE(sbinfo->shrinklist_len);
644 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
646 #define shmem_huge SHMEM_HUGE_DENY
648 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
649 struct shrink_control *sc, unsigned long nr_to_split)
653 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
655 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
657 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
658 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
659 shmem_huge != SHMEM_HUGE_DENY)
665 * Like add_to_page_cache_locked, but error if expected item has gone.
667 static int shmem_add_to_page_cache(struct page *page,
668 struct address_space *mapping,
669 pgoff_t index, void *expected, gfp_t gfp,
670 struct mm_struct *charge_mm)
672 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
674 unsigned long nr = compound_nr(page);
677 VM_BUG_ON_PAGE(PageTail(page), page);
678 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
679 VM_BUG_ON_PAGE(!PageLocked(page), page);
680 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
681 VM_BUG_ON(expected && PageTransHuge(page));
683 page_ref_add(page, nr);
684 page->mapping = mapping;
687 if (!PageSwapCache(page)) {
688 error = mem_cgroup_charge(page, charge_mm, gfp);
690 if (PageTransHuge(page)) {
691 count_vm_event(THP_FILE_FALLBACK);
692 count_vm_event(THP_FILE_FALLBACK_CHARGE);
697 cgroup_throttle_swaprate(page, gfp);
702 entry = xas_find_conflict(&xas);
703 if (entry != expected)
704 xas_set_err(&xas, -EEXIST);
705 xas_create_range(&xas);
709 xas_store(&xas, page);
714 if (PageTransHuge(page)) {
715 count_vm_event(THP_FILE_ALLOC);
716 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
718 mapping->nrpages += nr;
719 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
720 __mod_lruvec_page_state(page, NR_SHMEM, nr);
722 xas_unlock_irq(&xas);
723 } while (xas_nomem(&xas, gfp));
725 if (xas_error(&xas)) {
726 error = xas_error(&xas);
732 page->mapping = NULL;
733 page_ref_sub(page, nr);
738 * Like delete_from_page_cache, but substitutes swap for page.
740 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
742 struct address_space *mapping = page->mapping;
745 VM_BUG_ON_PAGE(PageCompound(page), page);
747 xa_lock_irq(&mapping->i_pages);
748 error = shmem_replace_entry(mapping, page->index, page, radswap);
749 page->mapping = NULL;
751 __dec_lruvec_page_state(page, NR_FILE_PAGES);
752 __dec_lruvec_page_state(page, NR_SHMEM);
753 xa_unlock_irq(&mapping->i_pages);
759 * Remove swap entry from page cache, free the swap and its page cache.
761 static int shmem_free_swap(struct address_space *mapping,
762 pgoff_t index, void *radswap)
766 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
769 free_swap_and_cache(radix_to_swp_entry(radswap));
774 * Determine (in bytes) how many of the shmem object's pages mapped by the
775 * given offsets are swapped out.
777 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
778 * as long as the inode doesn't go away and racy results are not a problem.
780 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
781 pgoff_t start, pgoff_t end)
783 XA_STATE(xas, &mapping->i_pages, start);
785 unsigned long swapped = 0;
788 xas_for_each(&xas, page, end - 1) {
789 if (xas_retry(&xas, page))
791 if (xa_is_value(page))
794 if (need_resched()) {
802 return swapped << PAGE_SHIFT;
806 * Determine (in bytes) how many of the shmem object's pages mapped by the
807 * given vma is swapped out.
809 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
810 * as long as the inode doesn't go away and racy results are not a problem.
812 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
814 struct inode *inode = file_inode(vma->vm_file);
815 struct shmem_inode_info *info = SHMEM_I(inode);
816 struct address_space *mapping = inode->i_mapping;
817 unsigned long swapped;
819 /* Be careful as we don't hold info->lock */
820 swapped = READ_ONCE(info->swapped);
823 * The easier cases are when the shmem object has nothing in swap, or
824 * the vma maps it whole. Then we can simply use the stats that we
830 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
831 return swapped << PAGE_SHIFT;
833 /* Here comes the more involved part */
834 return shmem_partial_swap_usage(mapping,
835 linear_page_index(vma, vma->vm_start),
836 linear_page_index(vma, vma->vm_end));
840 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
842 void shmem_unlock_mapping(struct address_space *mapping)
849 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
851 while (!mapping_unevictable(mapping)) {
852 if (!pagevec_lookup(&pvec, mapping, &index))
854 check_move_unevictable_pages(&pvec);
855 pagevec_release(&pvec);
861 * Check whether a hole-punch or truncation needs to split a huge page,
862 * returning true if no split was required, or the split has been successful.
864 * Eviction (or truncation to 0 size) should never need to split a huge page;
865 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
866 * head, and then succeeded to trylock on tail.
868 * A split can only succeed when there are no additional references on the
869 * huge page: so the split below relies upon find_get_entries() having stopped
870 * when it found a subpage of the huge page, without getting further references.
872 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
874 if (!PageTransCompound(page))
877 /* Just proceed to delete a huge page wholly within the range punched */
878 if (PageHead(page) &&
879 page->index >= start && page->index + HPAGE_PMD_NR <= end)
882 /* Try to split huge page, so we can truly punch the hole or truncate */
883 return split_huge_page(page) >= 0;
887 * Remove range of pages and swap entries from page cache, and free them.
888 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
890 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
893 struct address_space *mapping = inode->i_mapping;
894 struct shmem_inode_info *info = SHMEM_I(inode);
895 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
896 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
897 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
898 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
900 pgoff_t indices[PAGEVEC_SIZE];
901 long nr_swaps_freed = 0;
906 end = -1; /* unsigned, so actually very big */
910 while (index < end && find_lock_entries(mapping, index, end - 1,
912 for (i = 0; i < pagevec_count(&pvec); i++) {
913 struct page *page = pvec.pages[i];
919 if (xa_is_value(page)) {
922 nr_swaps_freed += !shmem_free_swap(mapping,
926 index += thp_nr_pages(page) - 1;
928 if (!unfalloc || !PageUptodate(page))
929 truncate_inode_page(mapping, page);
932 pagevec_remove_exceptionals(&pvec);
933 pagevec_release(&pvec);
939 struct page *page = NULL;
940 shmem_getpage(inode, start - 1, &page, SGP_READ);
942 unsigned int top = PAGE_SIZE;
947 zero_user_segment(page, partial_start, top);
948 set_page_dirty(page);
954 struct page *page = NULL;
955 shmem_getpage(inode, end, &page, SGP_READ);
957 zero_user_segment(page, 0, partial_end);
958 set_page_dirty(page);
967 while (index < end) {
970 pvec.nr = find_get_entries(mapping, index,
971 min(end - index, (pgoff_t)PAGEVEC_SIZE),
972 pvec.pages, indices);
974 /* If all gone or hole-punch or unfalloc, we're done */
975 if (index == start || end != -1)
977 /* But if truncating, restart to make sure all gone */
981 for (i = 0; i < pagevec_count(&pvec); i++) {
982 struct page *page = pvec.pages[i];
988 if (xa_is_value(page)) {
991 if (shmem_free_swap(mapping, index, page)) {
992 /* Swap was replaced by page: retry */
1002 if (!unfalloc || !PageUptodate(page)) {
1003 if (page_mapping(page) != mapping) {
1004 /* Page was replaced by swap: retry */
1009 VM_BUG_ON_PAGE(PageWriteback(page), page);
1010 if (shmem_punch_compound(page, start, end))
1011 truncate_inode_page(mapping, page);
1012 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1013 /* Wipe the page and don't get stuck */
1014 clear_highpage(page);
1015 flush_dcache_page(page);
1016 set_page_dirty(page);
1018 round_up(start, HPAGE_PMD_NR))
1024 pagevec_remove_exceptionals(&pvec);
1025 pagevec_release(&pvec);
1029 spin_lock_irq(&info->lock);
1030 info->swapped -= nr_swaps_freed;
1031 shmem_recalc_inode(inode);
1032 spin_unlock_irq(&info->lock);
1035 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1037 shmem_undo_range(inode, lstart, lend, false);
1038 inode->i_ctime = inode->i_mtime = current_time(inode);
1040 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1042 static int shmem_getattr(struct user_namespace *mnt_userns,
1043 const struct path *path, struct kstat *stat,
1044 u32 request_mask, unsigned int query_flags)
1046 struct inode *inode = path->dentry->d_inode;
1047 struct shmem_inode_info *info = SHMEM_I(inode);
1048 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1050 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1051 spin_lock_irq(&info->lock);
1052 shmem_recalc_inode(inode);
1053 spin_unlock_irq(&info->lock);
1055 generic_fillattr(&init_user_ns, inode, stat);
1057 if (is_huge_enabled(sb_info))
1058 stat->blksize = HPAGE_PMD_SIZE;
1063 static int shmem_setattr(struct user_namespace *mnt_userns,
1064 struct dentry *dentry, struct iattr *attr)
1066 struct inode *inode = d_inode(dentry);
1067 struct shmem_inode_info *info = SHMEM_I(inode);
1068 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1071 error = setattr_prepare(&init_user_ns, dentry, attr);
1075 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1076 loff_t oldsize = inode->i_size;
1077 loff_t newsize = attr->ia_size;
1079 /* protected by i_mutex */
1080 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1081 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1084 if (newsize != oldsize) {
1085 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1089 i_size_write(inode, newsize);
1090 inode->i_ctime = inode->i_mtime = current_time(inode);
1092 if (newsize <= oldsize) {
1093 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1094 if (oldsize > holebegin)
1095 unmap_mapping_range(inode->i_mapping,
1098 shmem_truncate_range(inode,
1099 newsize, (loff_t)-1);
1100 /* unmap again to remove racily COWed private pages */
1101 if (oldsize > holebegin)
1102 unmap_mapping_range(inode->i_mapping,
1106 * Part of the huge page can be beyond i_size: subject
1107 * to shrink under memory pressure.
1109 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1110 spin_lock(&sbinfo->shrinklist_lock);
1112 * _careful to defend against unlocked access to
1113 * ->shrink_list in shmem_unused_huge_shrink()
1115 if (list_empty_careful(&info->shrinklist)) {
1116 list_add_tail(&info->shrinklist,
1117 &sbinfo->shrinklist);
1118 sbinfo->shrinklist_len++;
1120 spin_unlock(&sbinfo->shrinklist_lock);
1125 setattr_copy(&init_user_ns, inode, attr);
1126 if (attr->ia_valid & ATTR_MODE)
1127 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1131 static void shmem_evict_inode(struct inode *inode)
1133 struct shmem_inode_info *info = SHMEM_I(inode);
1134 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1136 if (shmem_mapping(inode->i_mapping)) {
1137 shmem_unacct_size(info->flags, inode->i_size);
1139 shmem_truncate_range(inode, 0, (loff_t)-1);
1140 if (!list_empty(&info->shrinklist)) {
1141 spin_lock(&sbinfo->shrinklist_lock);
1142 if (!list_empty(&info->shrinklist)) {
1143 list_del_init(&info->shrinklist);
1144 sbinfo->shrinklist_len--;
1146 spin_unlock(&sbinfo->shrinklist_lock);
1148 while (!list_empty(&info->swaplist)) {
1149 /* Wait while shmem_unuse() is scanning this inode... */
1150 wait_var_event(&info->stop_eviction,
1151 !atomic_read(&info->stop_eviction));
1152 mutex_lock(&shmem_swaplist_mutex);
1153 /* ...but beware of the race if we peeked too early */
1154 if (!atomic_read(&info->stop_eviction))
1155 list_del_init(&info->swaplist);
1156 mutex_unlock(&shmem_swaplist_mutex);
1160 simple_xattrs_free(&info->xattrs);
1161 WARN_ON(inode->i_blocks);
1162 shmem_free_inode(inode->i_sb);
1166 extern struct swap_info_struct *swap_info[];
1168 static int shmem_find_swap_entries(struct address_space *mapping,
1169 pgoff_t start, unsigned int nr_entries,
1170 struct page **entries, pgoff_t *indices,
1171 unsigned int type, bool frontswap)
1173 XA_STATE(xas, &mapping->i_pages, start);
1176 unsigned int ret = 0;
1182 xas_for_each(&xas, page, ULONG_MAX) {
1183 if (xas_retry(&xas, page))
1186 if (!xa_is_value(page))
1189 entry = radix_to_swp_entry(page);
1190 if (swp_type(entry) != type)
1193 !frontswap_test(swap_info[type], swp_offset(entry)))
1196 indices[ret] = xas.xa_index;
1197 entries[ret] = page;
1199 if (need_resched()) {
1203 if (++ret == nr_entries)
1212 * Move the swapped pages for an inode to page cache. Returns the count
1213 * of pages swapped in, or the error in case of failure.
1215 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1221 struct address_space *mapping = inode->i_mapping;
1223 for (i = 0; i < pvec.nr; i++) {
1224 struct page *page = pvec.pages[i];
1226 if (!xa_is_value(page))
1228 error = shmem_swapin_page(inode, indices[i],
1230 mapping_gfp_mask(mapping),
1237 if (error == -ENOMEM)
1241 return error ? error : ret;
1245 * If swap found in inode, free it and move page from swapcache to filecache.
1247 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1248 bool frontswap, unsigned long *fs_pages_to_unuse)
1250 struct address_space *mapping = inode->i_mapping;
1252 struct pagevec pvec;
1253 pgoff_t indices[PAGEVEC_SIZE];
1254 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1257 pagevec_init(&pvec);
1259 unsigned int nr_entries = PAGEVEC_SIZE;
1261 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1262 nr_entries = *fs_pages_to_unuse;
1264 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1265 pvec.pages, indices,
1272 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1276 if (frontswap_partial) {
1277 *fs_pages_to_unuse -= ret;
1278 if (*fs_pages_to_unuse == 0) {
1279 ret = FRONTSWAP_PAGES_UNUSED;
1284 start = indices[pvec.nr - 1];
1291 * Read all the shared memory data that resides in the swap
1292 * device 'type' back into memory, so the swap device can be
1295 int shmem_unuse(unsigned int type, bool frontswap,
1296 unsigned long *fs_pages_to_unuse)
1298 struct shmem_inode_info *info, *next;
1301 if (list_empty(&shmem_swaplist))
1304 mutex_lock(&shmem_swaplist_mutex);
1305 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1306 if (!info->swapped) {
1307 list_del_init(&info->swaplist);
1311 * Drop the swaplist mutex while searching the inode for swap;
1312 * but before doing so, make sure shmem_evict_inode() will not
1313 * remove placeholder inode from swaplist, nor let it be freed
1314 * (igrab() would protect from unlink, but not from unmount).
1316 atomic_inc(&info->stop_eviction);
1317 mutex_unlock(&shmem_swaplist_mutex);
1319 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1323 mutex_lock(&shmem_swaplist_mutex);
1324 next = list_next_entry(info, swaplist);
1326 list_del_init(&info->swaplist);
1327 if (atomic_dec_and_test(&info->stop_eviction))
1328 wake_up_var(&info->stop_eviction);
1332 mutex_unlock(&shmem_swaplist_mutex);
1338 * Move the page from the page cache to the swap cache.
1340 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1342 struct shmem_inode_info *info;
1343 struct address_space *mapping;
1344 struct inode *inode;
1348 VM_BUG_ON_PAGE(PageCompound(page), page);
1349 BUG_ON(!PageLocked(page));
1350 mapping = page->mapping;
1351 index = page->index;
1352 inode = mapping->host;
1353 info = SHMEM_I(inode);
1354 if (info->flags & VM_LOCKED)
1356 if (!total_swap_pages)
1360 * Our capabilities prevent regular writeback or sync from ever calling
1361 * shmem_writepage; but a stacking filesystem might use ->writepage of
1362 * its underlying filesystem, in which case tmpfs should write out to
1363 * swap only in response to memory pressure, and not for the writeback
1366 if (!wbc->for_reclaim) {
1367 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1372 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1373 * value into swapfile.c, the only way we can correctly account for a
1374 * fallocated page arriving here is now to initialize it and write it.
1376 * That's okay for a page already fallocated earlier, but if we have
1377 * not yet completed the fallocation, then (a) we want to keep track
1378 * of this page in case we have to undo it, and (b) it may not be a
1379 * good idea to continue anyway, once we're pushing into swap. So
1380 * reactivate the page, and let shmem_fallocate() quit when too many.
1382 if (!PageUptodate(page)) {
1383 if (inode->i_private) {
1384 struct shmem_falloc *shmem_falloc;
1385 spin_lock(&inode->i_lock);
1386 shmem_falloc = inode->i_private;
1388 !shmem_falloc->waitq &&
1389 index >= shmem_falloc->start &&
1390 index < shmem_falloc->next)
1391 shmem_falloc->nr_unswapped++;
1393 shmem_falloc = NULL;
1394 spin_unlock(&inode->i_lock);
1398 clear_highpage(page);
1399 flush_dcache_page(page);
1400 SetPageUptodate(page);
1403 swap = get_swap_page(page);
1408 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1409 * if it's not already there. Do it now before the page is
1410 * moved to swap cache, when its pagelock no longer protects
1411 * the inode from eviction. But don't unlock the mutex until
1412 * we've incremented swapped, because shmem_unuse_inode() will
1413 * prune a !swapped inode from the swaplist under this mutex.
1415 mutex_lock(&shmem_swaplist_mutex);
1416 if (list_empty(&info->swaplist))
1417 list_add(&info->swaplist, &shmem_swaplist);
1419 if (add_to_swap_cache(page, swap,
1420 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1422 spin_lock_irq(&info->lock);
1423 shmem_recalc_inode(inode);
1425 spin_unlock_irq(&info->lock);
1427 swap_shmem_alloc(swap);
1428 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1430 mutex_unlock(&shmem_swaplist_mutex);
1431 BUG_ON(page_mapped(page));
1432 swap_writepage(page, wbc);
1436 mutex_unlock(&shmem_swaplist_mutex);
1437 put_swap_page(page, swap);
1439 set_page_dirty(page);
1440 if (wbc->for_reclaim)
1441 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1446 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1447 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1451 if (!mpol || mpol->mode == MPOL_DEFAULT)
1452 return; /* show nothing */
1454 mpol_to_str(buffer, sizeof(buffer), mpol);
1456 seq_printf(seq, ",mpol=%s", buffer);
1459 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1461 struct mempolicy *mpol = NULL;
1463 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1464 mpol = sbinfo->mpol;
1466 spin_unlock(&sbinfo->stat_lock);
1470 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1471 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1474 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1478 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1480 #define vm_policy vm_private_data
1483 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1484 struct shmem_inode_info *info, pgoff_t index)
1486 /* Create a pseudo vma that just contains the policy */
1487 vma_init(vma, NULL);
1488 /* Bias interleave by inode number to distribute better across nodes */
1489 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1490 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1493 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1495 /* Drop reference taken by mpol_shared_policy_lookup() */
1496 mpol_cond_put(vma->vm_policy);
1499 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1500 struct shmem_inode_info *info, pgoff_t index)
1502 struct vm_area_struct pvma;
1504 struct vm_fault vmf = {
1508 shmem_pseudo_vma_init(&pvma, info, index);
1509 page = swap_cluster_readahead(swap, gfp, &vmf);
1510 shmem_pseudo_vma_destroy(&pvma);
1515 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1516 struct shmem_inode_info *info, pgoff_t index)
1518 struct vm_area_struct pvma;
1519 struct address_space *mapping = info->vfs_inode.i_mapping;
1523 hindex = round_down(index, HPAGE_PMD_NR);
1524 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1528 shmem_pseudo_vma_init(&pvma, info, hindex);
1529 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1530 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1531 shmem_pseudo_vma_destroy(&pvma);
1533 prep_transhuge_page(page);
1535 count_vm_event(THP_FILE_FALLBACK);
1539 static struct page *shmem_alloc_page(gfp_t gfp,
1540 struct shmem_inode_info *info, pgoff_t index)
1542 struct vm_area_struct pvma;
1545 shmem_pseudo_vma_init(&pvma, info, index);
1546 page = alloc_page_vma(gfp, &pvma, 0);
1547 shmem_pseudo_vma_destroy(&pvma);
1552 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1553 struct inode *inode,
1554 pgoff_t index, bool huge)
1556 struct shmem_inode_info *info = SHMEM_I(inode);
1561 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1563 nr = huge ? HPAGE_PMD_NR : 1;
1565 if (!shmem_inode_acct_block(inode, nr))
1569 page = shmem_alloc_hugepage(gfp, info, index);
1571 page = shmem_alloc_page(gfp, info, index);
1573 __SetPageLocked(page);
1574 __SetPageSwapBacked(page);
1579 shmem_inode_unacct_blocks(inode, nr);
1581 return ERR_PTR(err);
1585 * When a page is moved from swapcache to shmem filecache (either by the
1586 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1587 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1588 * ignorance of the mapping it belongs to. If that mapping has special
1589 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1590 * we may need to copy to a suitable page before moving to filecache.
1592 * In a future release, this may well be extended to respect cpuset and
1593 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1594 * but for now it is a simple matter of zone.
1596 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1598 return page_zonenum(page) > gfp_zone(gfp);
1601 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1602 struct shmem_inode_info *info, pgoff_t index)
1604 struct page *oldpage, *newpage;
1605 struct address_space *swap_mapping;
1611 entry.val = page_private(oldpage);
1612 swap_index = swp_offset(entry);
1613 swap_mapping = page_mapping(oldpage);
1616 * We have arrived here because our zones are constrained, so don't
1617 * limit chance of success by further cpuset and node constraints.
1619 gfp &= ~GFP_CONSTRAINT_MASK;
1620 newpage = shmem_alloc_page(gfp, info, index);
1625 copy_highpage(newpage, oldpage);
1626 flush_dcache_page(newpage);
1628 __SetPageLocked(newpage);
1629 __SetPageSwapBacked(newpage);
1630 SetPageUptodate(newpage);
1631 set_page_private(newpage, entry.val);
1632 SetPageSwapCache(newpage);
1635 * Our caller will very soon move newpage out of swapcache, but it's
1636 * a nice clean interface for us to replace oldpage by newpage there.
1638 xa_lock_irq(&swap_mapping->i_pages);
1639 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1641 mem_cgroup_migrate(oldpage, newpage);
1642 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1643 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1645 xa_unlock_irq(&swap_mapping->i_pages);
1647 if (unlikely(error)) {
1649 * Is this possible? I think not, now that our callers check
1650 * both PageSwapCache and page_private after getting page lock;
1651 * but be defensive. Reverse old to newpage for clear and free.
1655 lru_cache_add(newpage);
1659 ClearPageSwapCache(oldpage);
1660 set_page_private(oldpage, 0);
1662 unlock_page(oldpage);
1669 * Swap in the page pointed to by *pagep.
1670 * Caller has to make sure that *pagep contains a valid swapped page.
1671 * Returns 0 and the page in pagep if success. On failure, returns the
1672 * error code and NULL in *pagep.
1674 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1675 struct page **pagep, enum sgp_type sgp,
1676 gfp_t gfp, struct vm_area_struct *vma,
1677 vm_fault_t *fault_type)
1679 struct address_space *mapping = inode->i_mapping;
1680 struct shmem_inode_info *info = SHMEM_I(inode);
1681 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
1686 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1687 swap = radix_to_swp_entry(*pagep);
1690 /* Look it up and read it in.. */
1691 page = lookup_swap_cache(swap, NULL, 0);
1693 /* Or update major stats only when swapin succeeds?? */
1695 *fault_type |= VM_FAULT_MAJOR;
1696 count_vm_event(PGMAJFAULT);
1697 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1699 /* Here we actually start the io */
1700 page = shmem_swapin(swap, gfp, info, index);
1707 /* We have to do this with page locked to prevent races */
1709 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1710 !shmem_confirm_swap(mapping, index, swap)) {
1714 if (!PageUptodate(page)) {
1718 wait_on_page_writeback(page);
1721 * Some architectures may have to restore extra metadata to the
1722 * physical page after reading from swap.
1724 arch_swap_restore(swap, page);
1726 if (shmem_should_replace_page(page, gfp)) {
1727 error = shmem_replace_page(&page, gfp, info, index);
1732 error = shmem_add_to_page_cache(page, mapping, index,
1733 swp_to_radix_entry(swap), gfp,
1738 spin_lock_irq(&info->lock);
1740 shmem_recalc_inode(inode);
1741 spin_unlock_irq(&info->lock);
1743 if (sgp == SGP_WRITE)
1744 mark_page_accessed(page);
1746 delete_from_swap_cache(page);
1747 set_page_dirty(page);
1753 if (!shmem_confirm_swap(mapping, index, swap))
1765 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1767 * If we allocate a new one we do not mark it dirty. That's up to the
1768 * vm. If we swap it in we mark it dirty since we also free the swap
1769 * entry since a page cannot live in both the swap and page cache.
1771 * vmf and fault_type are only supplied by shmem_fault:
1772 * otherwise they are NULL.
1774 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1775 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1776 struct vm_area_struct *vma, struct vm_fault *vmf,
1777 vm_fault_t *fault_type)
1779 struct address_space *mapping = inode->i_mapping;
1780 struct shmem_inode_info *info = SHMEM_I(inode);
1781 struct shmem_sb_info *sbinfo;
1782 struct mm_struct *charge_mm;
1784 enum sgp_type sgp_huge = sgp;
1785 pgoff_t hindex = index;
1790 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1792 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1795 if (sgp <= SGP_CACHE &&
1796 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1800 sbinfo = SHMEM_SB(inode->i_sb);
1801 charge_mm = vma ? vma->vm_mm : current->mm;
1803 page = pagecache_get_page(mapping, index,
1804 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1805 if (xa_is_value(page)) {
1806 error = shmem_swapin_page(inode, index, &page,
1807 sgp, gfp, vma, fault_type);
1808 if (error == -EEXIST)
1816 hindex = page->index;
1817 if (page && sgp == SGP_WRITE)
1818 mark_page_accessed(page);
1820 /* fallocated page? */
1821 if (page && !PageUptodate(page)) {
1822 if (sgp != SGP_READ)
1829 if (page || sgp == SGP_READ)
1833 * Fast cache lookup did not find it:
1834 * bring it back from swap or allocate.
1837 if (vma && userfaultfd_missing(vma)) {
1838 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1842 /* shmem_symlink() */
1843 if (!shmem_mapping(mapping))
1845 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1847 if (shmem_huge == SHMEM_HUGE_FORCE)
1849 switch (sbinfo->huge) {
1850 case SHMEM_HUGE_NEVER:
1852 case SHMEM_HUGE_WITHIN_SIZE: {
1856 off = round_up(index, HPAGE_PMD_NR);
1857 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1858 if (i_size >= HPAGE_PMD_SIZE &&
1859 i_size >> PAGE_SHIFT >= off)
1864 case SHMEM_HUGE_ADVISE:
1865 if (sgp_huge == SGP_HUGE)
1867 /* TODO: implement fadvise() hints */
1872 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1875 page = shmem_alloc_and_acct_page(gfp, inode,
1881 error = PTR_ERR(page);
1883 if (error != -ENOSPC)
1886 * Try to reclaim some space by splitting a huge page
1887 * beyond i_size on the filesystem.
1892 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1893 if (ret == SHRINK_STOP)
1901 if (PageTransHuge(page))
1902 hindex = round_down(index, HPAGE_PMD_NR);
1906 if (sgp == SGP_WRITE)
1907 __SetPageReferenced(page);
1909 error = shmem_add_to_page_cache(page, mapping, hindex,
1910 NULL, gfp & GFP_RECLAIM_MASK,
1914 lru_cache_add(page);
1916 spin_lock_irq(&info->lock);
1917 info->alloced += compound_nr(page);
1918 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1919 shmem_recalc_inode(inode);
1920 spin_unlock_irq(&info->lock);
1923 if (PageTransHuge(page) &&
1924 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1925 hindex + HPAGE_PMD_NR - 1) {
1927 * Part of the huge page is beyond i_size: subject
1928 * to shrink under memory pressure.
1930 spin_lock(&sbinfo->shrinklist_lock);
1932 * _careful to defend against unlocked access to
1933 * ->shrink_list in shmem_unused_huge_shrink()
1935 if (list_empty_careful(&info->shrinklist)) {
1936 list_add_tail(&info->shrinklist,
1937 &sbinfo->shrinklist);
1938 sbinfo->shrinklist_len++;
1940 spin_unlock(&sbinfo->shrinklist_lock);
1944 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1946 if (sgp == SGP_FALLOC)
1950 * Let SGP_WRITE caller clear ends if write does not fill page;
1951 * but SGP_FALLOC on a page fallocated earlier must initialize
1952 * it now, lest undo on failure cancel our earlier guarantee.
1954 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1957 for (i = 0; i < compound_nr(page); i++) {
1958 clear_highpage(page + i);
1959 flush_dcache_page(page + i);
1961 SetPageUptodate(page);
1964 /* Perhaps the file has been truncated since we checked */
1965 if (sgp <= SGP_CACHE &&
1966 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1968 ClearPageDirty(page);
1969 delete_from_page_cache(page);
1970 spin_lock_irq(&info->lock);
1971 shmem_recalc_inode(inode);
1972 spin_unlock_irq(&info->lock);
1978 *pagep = page + index - hindex;
1985 shmem_inode_unacct_blocks(inode, compound_nr(page));
1987 if (PageTransHuge(page)) {
1997 if (error == -ENOSPC && !once++) {
1998 spin_lock_irq(&info->lock);
1999 shmem_recalc_inode(inode);
2000 spin_unlock_irq(&info->lock);
2003 if (error == -EEXIST)
2009 * This is like autoremove_wake_function, but it removes the wait queue
2010 * entry unconditionally - even if something else had already woken the
2013 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2015 int ret = default_wake_function(wait, mode, sync, key);
2016 list_del_init(&wait->entry);
2020 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2022 struct vm_area_struct *vma = vmf->vma;
2023 struct inode *inode = file_inode(vma->vm_file);
2024 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2027 vm_fault_t ret = VM_FAULT_LOCKED;
2030 * Trinity finds that probing a hole which tmpfs is punching can
2031 * prevent the hole-punch from ever completing: which in turn
2032 * locks writers out with its hold on i_mutex. So refrain from
2033 * faulting pages into the hole while it's being punched. Although
2034 * shmem_undo_range() does remove the additions, it may be unable to
2035 * keep up, as each new page needs its own unmap_mapping_range() call,
2036 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2038 * It does not matter if we sometimes reach this check just before the
2039 * hole-punch begins, so that one fault then races with the punch:
2040 * we just need to make racing faults a rare case.
2042 * The implementation below would be much simpler if we just used a
2043 * standard mutex or completion: but we cannot take i_mutex in fault,
2044 * and bloating every shmem inode for this unlikely case would be sad.
2046 if (unlikely(inode->i_private)) {
2047 struct shmem_falloc *shmem_falloc;
2049 spin_lock(&inode->i_lock);
2050 shmem_falloc = inode->i_private;
2052 shmem_falloc->waitq &&
2053 vmf->pgoff >= shmem_falloc->start &&
2054 vmf->pgoff < shmem_falloc->next) {
2056 wait_queue_head_t *shmem_falloc_waitq;
2057 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2059 ret = VM_FAULT_NOPAGE;
2060 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2062 ret = VM_FAULT_RETRY;
2064 shmem_falloc_waitq = shmem_falloc->waitq;
2065 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2066 TASK_UNINTERRUPTIBLE);
2067 spin_unlock(&inode->i_lock);
2071 * shmem_falloc_waitq points into the shmem_fallocate()
2072 * stack of the hole-punching task: shmem_falloc_waitq
2073 * is usually invalid by the time we reach here, but
2074 * finish_wait() does not dereference it in that case;
2075 * though i_lock needed lest racing with wake_up_all().
2077 spin_lock(&inode->i_lock);
2078 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2079 spin_unlock(&inode->i_lock);
2085 spin_unlock(&inode->i_lock);
2090 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2091 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2093 else if (vma->vm_flags & VM_HUGEPAGE)
2096 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2097 gfp, vma, vmf, &ret);
2099 return vmf_error(err);
2103 unsigned long shmem_get_unmapped_area(struct file *file,
2104 unsigned long uaddr, unsigned long len,
2105 unsigned long pgoff, unsigned long flags)
2107 unsigned long (*get_area)(struct file *,
2108 unsigned long, unsigned long, unsigned long, unsigned long);
2110 unsigned long offset;
2111 unsigned long inflated_len;
2112 unsigned long inflated_addr;
2113 unsigned long inflated_offset;
2115 if (len > TASK_SIZE)
2118 get_area = current->mm->get_unmapped_area;
2119 addr = get_area(file, uaddr, len, pgoff, flags);
2121 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2123 if (IS_ERR_VALUE(addr))
2125 if (addr & ~PAGE_MASK)
2127 if (addr > TASK_SIZE - len)
2130 if (shmem_huge == SHMEM_HUGE_DENY)
2132 if (len < HPAGE_PMD_SIZE)
2134 if (flags & MAP_FIXED)
2137 * Our priority is to support MAP_SHARED mapped hugely;
2138 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2139 * But if caller specified an address hint and we allocated area there
2140 * successfully, respect that as before.
2145 if (shmem_huge != SHMEM_HUGE_FORCE) {
2146 struct super_block *sb;
2149 VM_BUG_ON(file->f_op != &shmem_file_operations);
2150 sb = file_inode(file)->i_sb;
2153 * Called directly from mm/mmap.c, or drivers/char/mem.c
2154 * for "/dev/zero", to create a shared anonymous object.
2156 if (IS_ERR(shm_mnt))
2158 sb = shm_mnt->mnt_sb;
2160 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2164 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2165 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2167 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2170 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2171 if (inflated_len > TASK_SIZE)
2173 if (inflated_len < len)
2176 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2177 if (IS_ERR_VALUE(inflated_addr))
2179 if (inflated_addr & ~PAGE_MASK)
2182 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2183 inflated_addr += offset - inflated_offset;
2184 if (inflated_offset > offset)
2185 inflated_addr += HPAGE_PMD_SIZE;
2187 if (inflated_addr > TASK_SIZE - len)
2189 return inflated_addr;
2193 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2195 struct inode *inode = file_inode(vma->vm_file);
2196 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2199 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2202 struct inode *inode = file_inode(vma->vm_file);
2205 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2206 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2210 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2212 struct inode *inode = file_inode(file);
2213 struct shmem_inode_info *info = SHMEM_I(inode);
2214 int retval = -ENOMEM;
2217 * What serializes the accesses to info->flags?
2218 * ipc_lock_object() when called from shmctl_do_lock(),
2219 * no serialization needed when called from shm_destroy().
2221 if (lock && !(info->flags & VM_LOCKED)) {
2222 if (!user_shm_lock(inode->i_size, user))
2224 info->flags |= VM_LOCKED;
2225 mapping_set_unevictable(file->f_mapping);
2227 if (!lock && (info->flags & VM_LOCKED) && user) {
2228 user_shm_unlock(inode->i_size, user);
2229 info->flags &= ~VM_LOCKED;
2230 mapping_clear_unevictable(file->f_mapping);
2238 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2240 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2242 if (info->seals & F_SEAL_FUTURE_WRITE) {
2244 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2245 * "future write" seal active.
2247 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2251 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2252 * MAP_SHARED and read-only, take care to not allow mprotect to
2253 * revert protections on such mappings. Do this only for shared
2254 * mappings. For private mappings, don't need to mask
2255 * VM_MAYWRITE as we still want them to be COW-writable.
2257 if (vma->vm_flags & VM_SHARED)
2258 vma->vm_flags &= ~(VM_MAYWRITE);
2261 /* arm64 - allow memory tagging on RAM-based files */
2262 vma->vm_flags |= VM_MTE_ALLOWED;
2264 file_accessed(file);
2265 vma->vm_ops = &shmem_vm_ops;
2266 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2267 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2268 (vma->vm_end & HPAGE_PMD_MASK)) {
2269 khugepaged_enter(vma, vma->vm_flags);
2274 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2275 umode_t mode, dev_t dev, unsigned long flags)
2277 struct inode *inode;
2278 struct shmem_inode_info *info;
2279 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2282 if (shmem_reserve_inode(sb, &ino))
2285 inode = new_inode(sb);
2288 inode_init_owner(&init_user_ns, inode, dir, mode);
2289 inode->i_blocks = 0;
2290 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2291 inode->i_generation = prandom_u32();
2292 info = SHMEM_I(inode);
2293 memset(info, 0, (char *)inode - (char *)info);
2294 spin_lock_init(&info->lock);
2295 atomic_set(&info->stop_eviction, 0);
2296 info->seals = F_SEAL_SEAL;
2297 info->flags = flags & VM_NORESERVE;
2298 INIT_LIST_HEAD(&info->shrinklist);
2299 INIT_LIST_HEAD(&info->swaplist);
2300 simple_xattrs_init(&info->xattrs);
2301 cache_no_acl(inode);
2303 switch (mode & S_IFMT) {
2305 inode->i_op = &shmem_special_inode_operations;
2306 init_special_inode(inode, mode, dev);
2309 inode->i_mapping->a_ops = &shmem_aops;
2310 inode->i_op = &shmem_inode_operations;
2311 inode->i_fop = &shmem_file_operations;
2312 mpol_shared_policy_init(&info->policy,
2313 shmem_get_sbmpol(sbinfo));
2317 /* Some things misbehave if size == 0 on a directory */
2318 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2319 inode->i_op = &shmem_dir_inode_operations;
2320 inode->i_fop = &simple_dir_operations;
2324 * Must not load anything in the rbtree,
2325 * mpol_free_shared_policy will not be called.
2327 mpol_shared_policy_init(&info->policy, NULL);
2331 lockdep_annotate_inode_mutex_key(inode);
2333 shmem_free_inode(sb);
2337 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2339 struct vm_area_struct *dst_vma,
2340 unsigned long dst_addr,
2341 unsigned long src_addr,
2343 struct page **pagep)
2345 struct inode *inode = file_inode(dst_vma->vm_file);
2346 struct shmem_inode_info *info = SHMEM_I(inode);
2347 struct address_space *mapping = inode->i_mapping;
2348 gfp_t gfp = mapping_gfp_mask(mapping);
2349 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2353 pte_t _dst_pte, *dst_pte;
2355 pgoff_t offset, max_off;
2358 if (!shmem_inode_acct_block(inode, 1))
2362 page = shmem_alloc_page(gfp, info, pgoff);
2364 goto out_unacct_blocks;
2366 if (!zeropage) { /* mcopy_atomic */
2367 page_kaddr = kmap_atomic(page);
2368 ret = copy_from_user(page_kaddr,
2369 (const void __user *)src_addr,
2371 kunmap_atomic(page_kaddr);
2373 /* fallback to copy_from_user outside mmap_lock */
2374 if (unlikely(ret)) {
2376 shmem_inode_unacct_blocks(inode, 1);
2377 /* don't free the page */
2380 } else { /* mfill_zeropage_atomic */
2381 clear_highpage(page);
2388 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2389 __SetPageLocked(page);
2390 __SetPageSwapBacked(page);
2391 __SetPageUptodate(page);
2394 offset = linear_page_index(dst_vma, dst_addr);
2395 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2396 if (unlikely(offset >= max_off))
2399 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2400 gfp & GFP_RECLAIM_MASK, dst_mm);
2404 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2405 if (dst_vma->vm_flags & VM_WRITE)
2406 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2409 * We don't set the pte dirty if the vma has no
2410 * VM_WRITE permission, so mark the page dirty or it
2411 * could be freed from under us. We could do it
2412 * unconditionally before unlock_page(), but doing it
2413 * only if VM_WRITE is not set is faster.
2415 set_page_dirty(page);
2418 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2421 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2422 if (unlikely(offset >= max_off))
2423 goto out_release_unlock;
2426 if (!pte_none(*dst_pte))
2427 goto out_release_unlock;
2429 lru_cache_add(page);
2431 spin_lock_irq(&info->lock);
2433 inode->i_blocks += BLOCKS_PER_PAGE;
2434 shmem_recalc_inode(inode);
2435 spin_unlock_irq(&info->lock);
2437 inc_mm_counter(dst_mm, mm_counter_file(page));
2438 page_add_file_rmap(page, false);
2439 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2441 /* No need to invalidate - it was non-present before */
2442 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2443 pte_unmap_unlock(dst_pte, ptl);
2449 pte_unmap_unlock(dst_pte, ptl);
2450 ClearPageDirty(page);
2451 delete_from_page_cache(page);
2456 shmem_inode_unacct_blocks(inode, 1);
2460 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2462 struct vm_area_struct *dst_vma,
2463 unsigned long dst_addr,
2464 unsigned long src_addr,
2465 struct page **pagep)
2467 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2468 dst_addr, src_addr, false, pagep);
2471 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2473 struct vm_area_struct *dst_vma,
2474 unsigned long dst_addr)
2476 struct page *page = NULL;
2478 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2479 dst_addr, 0, true, &page);
2483 static const struct inode_operations shmem_symlink_inode_operations;
2484 static const struct inode_operations shmem_short_symlink_operations;
2486 #ifdef CONFIG_TMPFS_XATTR
2487 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2489 #define shmem_initxattrs NULL
2493 shmem_write_begin(struct file *file, struct address_space *mapping,
2494 loff_t pos, unsigned len, unsigned flags,
2495 struct page **pagep, void **fsdata)
2497 struct inode *inode = mapping->host;
2498 struct shmem_inode_info *info = SHMEM_I(inode);
2499 pgoff_t index = pos >> PAGE_SHIFT;
2501 /* i_mutex is held by caller */
2502 if (unlikely(info->seals & (F_SEAL_GROW |
2503 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2504 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2506 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2510 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2514 shmem_write_end(struct file *file, struct address_space *mapping,
2515 loff_t pos, unsigned len, unsigned copied,
2516 struct page *page, void *fsdata)
2518 struct inode *inode = mapping->host;
2520 if (pos + copied > inode->i_size)
2521 i_size_write(inode, pos + copied);
2523 if (!PageUptodate(page)) {
2524 struct page *head = compound_head(page);
2525 if (PageTransCompound(page)) {
2528 for (i = 0; i < HPAGE_PMD_NR; i++) {
2529 if (head + i == page)
2531 clear_highpage(head + i);
2532 flush_dcache_page(head + i);
2535 if (copied < PAGE_SIZE) {
2536 unsigned from = pos & (PAGE_SIZE - 1);
2537 zero_user_segments(page, 0, from,
2538 from + copied, PAGE_SIZE);
2540 SetPageUptodate(head);
2542 set_page_dirty(page);
2549 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2551 struct file *file = iocb->ki_filp;
2552 struct inode *inode = file_inode(file);
2553 struct address_space *mapping = inode->i_mapping;
2555 unsigned long offset;
2556 enum sgp_type sgp = SGP_READ;
2559 loff_t *ppos = &iocb->ki_pos;
2562 * Might this read be for a stacking filesystem? Then when reading
2563 * holes of a sparse file, we actually need to allocate those pages,
2564 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2566 if (!iter_is_iovec(to))
2569 index = *ppos >> PAGE_SHIFT;
2570 offset = *ppos & ~PAGE_MASK;
2573 struct page *page = NULL;
2575 unsigned long nr, ret;
2576 loff_t i_size = i_size_read(inode);
2578 end_index = i_size >> PAGE_SHIFT;
2579 if (index > end_index)
2581 if (index == end_index) {
2582 nr = i_size & ~PAGE_MASK;
2587 error = shmem_getpage(inode, index, &page, sgp);
2589 if (error == -EINVAL)
2594 if (sgp == SGP_CACHE)
2595 set_page_dirty(page);
2600 * We must evaluate after, since reads (unlike writes)
2601 * are called without i_mutex protection against truncate
2604 i_size = i_size_read(inode);
2605 end_index = i_size >> PAGE_SHIFT;
2606 if (index == end_index) {
2607 nr = i_size & ~PAGE_MASK;
2618 * If users can be writing to this page using arbitrary
2619 * virtual addresses, take care about potential aliasing
2620 * before reading the page on the kernel side.
2622 if (mapping_writably_mapped(mapping))
2623 flush_dcache_page(page);
2625 * Mark the page accessed if we read the beginning.
2628 mark_page_accessed(page);
2630 page = ZERO_PAGE(0);
2635 * Ok, we have the page, and it's up-to-date, so
2636 * now we can copy it to user space...
2638 ret = copy_page_to_iter(page, offset, nr, to);
2641 index += offset >> PAGE_SHIFT;
2642 offset &= ~PAGE_MASK;
2645 if (!iov_iter_count(to))
2654 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2655 file_accessed(file);
2656 return retval ? retval : error;
2659 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2661 struct address_space *mapping = file->f_mapping;
2662 struct inode *inode = mapping->host;
2664 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2665 return generic_file_llseek_size(file, offset, whence,
2666 MAX_LFS_FILESIZE, i_size_read(inode));
2671 /* We're holding i_mutex so we can access i_size directly */
2672 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2674 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2675 inode_unlock(inode);
2679 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2682 struct inode *inode = file_inode(file);
2683 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2684 struct shmem_inode_info *info = SHMEM_I(inode);
2685 struct shmem_falloc shmem_falloc;
2686 pgoff_t start, index, end;
2689 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2694 if (mode & FALLOC_FL_PUNCH_HOLE) {
2695 struct address_space *mapping = file->f_mapping;
2696 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2697 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2698 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2700 /* protected by i_mutex */
2701 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2706 shmem_falloc.waitq = &shmem_falloc_waitq;
2707 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2708 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2709 spin_lock(&inode->i_lock);
2710 inode->i_private = &shmem_falloc;
2711 spin_unlock(&inode->i_lock);
2713 if ((u64)unmap_end > (u64)unmap_start)
2714 unmap_mapping_range(mapping, unmap_start,
2715 1 + unmap_end - unmap_start, 0);
2716 shmem_truncate_range(inode, offset, offset + len - 1);
2717 /* No need to unmap again: hole-punching leaves COWed pages */
2719 spin_lock(&inode->i_lock);
2720 inode->i_private = NULL;
2721 wake_up_all(&shmem_falloc_waitq);
2722 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2723 spin_unlock(&inode->i_lock);
2728 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2729 error = inode_newsize_ok(inode, offset + len);
2733 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2738 start = offset >> PAGE_SHIFT;
2739 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2740 /* Try to avoid a swapstorm if len is impossible to satisfy */
2741 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2746 shmem_falloc.waitq = NULL;
2747 shmem_falloc.start = start;
2748 shmem_falloc.next = start;
2749 shmem_falloc.nr_falloced = 0;
2750 shmem_falloc.nr_unswapped = 0;
2751 spin_lock(&inode->i_lock);
2752 inode->i_private = &shmem_falloc;
2753 spin_unlock(&inode->i_lock);
2755 for (index = start; index < end; index++) {
2759 * Good, the fallocate(2) manpage permits EINTR: we may have
2760 * been interrupted because we are using up too much memory.
2762 if (signal_pending(current))
2764 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2767 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2769 /* Remove the !PageUptodate pages we added */
2770 if (index > start) {
2771 shmem_undo_range(inode,
2772 (loff_t)start << PAGE_SHIFT,
2773 ((loff_t)index << PAGE_SHIFT) - 1, true);
2779 * Inform shmem_writepage() how far we have reached.
2780 * No need for lock or barrier: we have the page lock.
2782 shmem_falloc.next++;
2783 if (!PageUptodate(page))
2784 shmem_falloc.nr_falloced++;
2787 * If !PageUptodate, leave it that way so that freeable pages
2788 * can be recognized if we need to rollback on error later.
2789 * But set_page_dirty so that memory pressure will swap rather
2790 * than free the pages we are allocating (and SGP_CACHE pages
2791 * might still be clean: we now need to mark those dirty too).
2793 set_page_dirty(page);
2799 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2800 i_size_write(inode, offset + len);
2801 inode->i_ctime = current_time(inode);
2803 spin_lock(&inode->i_lock);
2804 inode->i_private = NULL;
2805 spin_unlock(&inode->i_lock);
2807 inode_unlock(inode);
2811 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2813 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2815 buf->f_type = TMPFS_MAGIC;
2816 buf->f_bsize = PAGE_SIZE;
2817 buf->f_namelen = NAME_MAX;
2818 if (sbinfo->max_blocks) {
2819 buf->f_blocks = sbinfo->max_blocks;
2821 buf->f_bfree = sbinfo->max_blocks -
2822 percpu_counter_sum(&sbinfo->used_blocks);
2824 if (sbinfo->max_inodes) {
2825 buf->f_files = sbinfo->max_inodes;
2826 buf->f_ffree = sbinfo->free_inodes;
2828 /* else leave those fields 0 like simple_statfs */
2833 * File creation. Allocate an inode, and we're done..
2836 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2837 struct dentry *dentry, umode_t mode, dev_t dev)
2839 struct inode *inode;
2840 int error = -ENOSPC;
2842 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2844 error = simple_acl_create(dir, inode);
2847 error = security_inode_init_security(inode, dir,
2849 shmem_initxattrs, NULL);
2850 if (error && error != -EOPNOTSUPP)
2854 dir->i_size += BOGO_DIRENT_SIZE;
2855 dir->i_ctime = dir->i_mtime = current_time(dir);
2856 d_instantiate(dentry, inode);
2857 dget(dentry); /* Extra count - pin the dentry in core */
2866 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2867 struct dentry *dentry, umode_t mode)
2869 struct inode *inode;
2870 int error = -ENOSPC;
2872 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2874 error = security_inode_init_security(inode, dir,
2876 shmem_initxattrs, NULL);
2877 if (error && error != -EOPNOTSUPP)
2879 error = simple_acl_create(dir, inode);
2882 d_tmpfile(dentry, inode);
2890 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2891 struct dentry *dentry, umode_t mode)
2895 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2896 mode | S_IFDIR, 0)))
2902 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2903 struct dentry *dentry, umode_t mode, bool excl)
2905 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2911 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2913 struct inode *inode = d_inode(old_dentry);
2917 * No ordinary (disk based) filesystem counts links as inodes;
2918 * but each new link needs a new dentry, pinning lowmem, and
2919 * tmpfs dentries cannot be pruned until they are unlinked.
2920 * But if an O_TMPFILE file is linked into the tmpfs, the
2921 * first link must skip that, to get the accounting right.
2923 if (inode->i_nlink) {
2924 ret = shmem_reserve_inode(inode->i_sb, NULL);
2929 dir->i_size += BOGO_DIRENT_SIZE;
2930 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2932 ihold(inode); /* New dentry reference */
2933 dget(dentry); /* Extra pinning count for the created dentry */
2934 d_instantiate(dentry, inode);
2939 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2941 struct inode *inode = d_inode(dentry);
2943 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2944 shmem_free_inode(inode->i_sb);
2946 dir->i_size -= BOGO_DIRENT_SIZE;
2947 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2949 dput(dentry); /* Undo the count from "create" - this does all the work */
2953 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2955 if (!simple_empty(dentry))
2958 drop_nlink(d_inode(dentry));
2960 return shmem_unlink(dir, dentry);
2963 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2965 bool old_is_dir = d_is_dir(old_dentry);
2966 bool new_is_dir = d_is_dir(new_dentry);
2968 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2970 drop_nlink(old_dir);
2973 drop_nlink(new_dir);
2977 old_dir->i_ctime = old_dir->i_mtime =
2978 new_dir->i_ctime = new_dir->i_mtime =
2979 d_inode(old_dentry)->i_ctime =
2980 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2985 static int shmem_whiteout(struct user_namespace *mnt_userns,
2986 struct inode *old_dir, struct dentry *old_dentry)
2988 struct dentry *whiteout;
2991 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2995 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2996 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3002 * Cheat and hash the whiteout while the old dentry is still in
3003 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3005 * d_lookup() will consistently find one of them at this point,
3006 * not sure which one, but that isn't even important.
3013 * The VFS layer already does all the dentry stuff for rename,
3014 * we just have to decrement the usage count for the target if
3015 * it exists so that the VFS layer correctly free's it when it
3018 static int shmem_rename2(struct user_namespace *mnt_userns,
3019 struct inode *old_dir, struct dentry *old_dentry,
3020 struct inode *new_dir, struct dentry *new_dentry,
3023 struct inode *inode = d_inode(old_dentry);
3024 int they_are_dirs = S_ISDIR(inode->i_mode);
3026 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3029 if (flags & RENAME_EXCHANGE)
3030 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3032 if (!simple_empty(new_dentry))
3035 if (flags & RENAME_WHITEOUT) {
3038 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3043 if (d_really_is_positive(new_dentry)) {
3044 (void) shmem_unlink(new_dir, new_dentry);
3045 if (they_are_dirs) {
3046 drop_nlink(d_inode(new_dentry));
3047 drop_nlink(old_dir);
3049 } else if (they_are_dirs) {
3050 drop_nlink(old_dir);
3054 old_dir->i_size -= BOGO_DIRENT_SIZE;
3055 new_dir->i_size += BOGO_DIRENT_SIZE;
3056 old_dir->i_ctime = old_dir->i_mtime =
3057 new_dir->i_ctime = new_dir->i_mtime =
3058 inode->i_ctime = current_time(old_dir);
3062 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3063 struct dentry *dentry, const char *symname)
3067 struct inode *inode;
3070 len = strlen(symname) + 1;
3071 if (len > PAGE_SIZE)
3072 return -ENAMETOOLONG;
3074 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3079 error = security_inode_init_security(inode, dir, &dentry->d_name,
3080 shmem_initxattrs, NULL);
3081 if (error && error != -EOPNOTSUPP) {
3086 inode->i_size = len-1;
3087 if (len <= SHORT_SYMLINK_LEN) {
3088 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3089 if (!inode->i_link) {
3093 inode->i_op = &shmem_short_symlink_operations;
3095 inode_nohighmem(inode);
3096 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3101 inode->i_mapping->a_ops = &shmem_aops;
3102 inode->i_op = &shmem_symlink_inode_operations;
3103 memcpy(page_address(page), symname, len);
3104 SetPageUptodate(page);
3105 set_page_dirty(page);
3109 dir->i_size += BOGO_DIRENT_SIZE;
3110 dir->i_ctime = dir->i_mtime = current_time(dir);
3111 d_instantiate(dentry, inode);
3116 static void shmem_put_link(void *arg)
3118 mark_page_accessed(arg);
3122 static const char *shmem_get_link(struct dentry *dentry,
3123 struct inode *inode,
3124 struct delayed_call *done)
3126 struct page *page = NULL;
3129 page = find_get_page(inode->i_mapping, 0);
3131 return ERR_PTR(-ECHILD);
3132 if (!PageUptodate(page)) {
3134 return ERR_PTR(-ECHILD);
3137 error = shmem_getpage(inode, 0, &page, SGP_READ);
3139 return ERR_PTR(error);
3142 set_delayed_call(done, shmem_put_link, page);
3143 return page_address(page);
3146 #ifdef CONFIG_TMPFS_XATTR
3148 * Superblocks without xattr inode operations may get some security.* xattr
3149 * support from the LSM "for free". As soon as we have any other xattrs
3150 * like ACLs, we also need to implement the security.* handlers at
3151 * filesystem level, though.
3155 * Callback for security_inode_init_security() for acquiring xattrs.
3157 static int shmem_initxattrs(struct inode *inode,
3158 const struct xattr *xattr_array,
3161 struct shmem_inode_info *info = SHMEM_I(inode);
3162 const struct xattr *xattr;
3163 struct simple_xattr *new_xattr;
3166 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3167 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3171 len = strlen(xattr->name) + 1;
3172 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3174 if (!new_xattr->name) {
3179 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3180 XATTR_SECURITY_PREFIX_LEN);
3181 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3184 simple_xattr_list_add(&info->xattrs, new_xattr);
3190 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3191 struct dentry *unused, struct inode *inode,
3192 const char *name, void *buffer, size_t size)
3194 struct shmem_inode_info *info = SHMEM_I(inode);
3196 name = xattr_full_name(handler, name);
3197 return simple_xattr_get(&info->xattrs, name, buffer, size);
3200 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3201 struct user_namespace *mnt_userns,
3202 struct dentry *unused, struct inode *inode,
3203 const char *name, const void *value,
3204 size_t size, int flags)
3206 struct shmem_inode_info *info = SHMEM_I(inode);
3208 name = xattr_full_name(handler, name);
3209 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3212 static const struct xattr_handler shmem_security_xattr_handler = {
3213 .prefix = XATTR_SECURITY_PREFIX,
3214 .get = shmem_xattr_handler_get,
3215 .set = shmem_xattr_handler_set,
3218 static const struct xattr_handler shmem_trusted_xattr_handler = {
3219 .prefix = XATTR_TRUSTED_PREFIX,
3220 .get = shmem_xattr_handler_get,
3221 .set = shmem_xattr_handler_set,
3224 static const struct xattr_handler *shmem_xattr_handlers[] = {
3225 #ifdef CONFIG_TMPFS_POSIX_ACL
3226 &posix_acl_access_xattr_handler,
3227 &posix_acl_default_xattr_handler,
3229 &shmem_security_xattr_handler,
3230 &shmem_trusted_xattr_handler,
3234 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3236 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3237 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3239 #endif /* CONFIG_TMPFS_XATTR */
3241 static const struct inode_operations shmem_short_symlink_operations = {
3242 .get_link = simple_get_link,
3243 #ifdef CONFIG_TMPFS_XATTR
3244 .listxattr = shmem_listxattr,
3248 static const struct inode_operations shmem_symlink_inode_operations = {
3249 .get_link = shmem_get_link,
3250 #ifdef CONFIG_TMPFS_XATTR
3251 .listxattr = shmem_listxattr,
3255 static struct dentry *shmem_get_parent(struct dentry *child)
3257 return ERR_PTR(-ESTALE);
3260 static int shmem_match(struct inode *ino, void *vfh)
3264 inum = (inum << 32) | fh[1];
3265 return ino->i_ino == inum && fh[0] == ino->i_generation;
3268 /* Find any alias of inode, but prefer a hashed alias */
3269 static struct dentry *shmem_find_alias(struct inode *inode)
3271 struct dentry *alias = d_find_alias(inode);
3273 return alias ?: d_find_any_alias(inode);
3277 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3278 struct fid *fid, int fh_len, int fh_type)
3280 struct inode *inode;
3281 struct dentry *dentry = NULL;
3288 inum = (inum << 32) | fid->raw[1];
3290 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3291 shmem_match, fid->raw);
3293 dentry = shmem_find_alias(inode);
3300 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3301 struct inode *parent)
3305 return FILEID_INVALID;
3308 if (inode_unhashed(inode)) {
3309 /* Unfortunately insert_inode_hash is not idempotent,
3310 * so as we hash inodes here rather than at creation
3311 * time, we need a lock to ensure we only try
3314 static DEFINE_SPINLOCK(lock);
3316 if (inode_unhashed(inode))
3317 __insert_inode_hash(inode,
3318 inode->i_ino + inode->i_generation);
3322 fh[0] = inode->i_generation;
3323 fh[1] = inode->i_ino;
3324 fh[2] = ((__u64)inode->i_ino) >> 32;
3330 static const struct export_operations shmem_export_ops = {
3331 .get_parent = shmem_get_parent,
3332 .encode_fh = shmem_encode_fh,
3333 .fh_to_dentry = shmem_fh_to_dentry,
3349 static const struct constant_table shmem_param_enums_huge[] = {
3350 {"never", SHMEM_HUGE_NEVER },
3351 {"always", SHMEM_HUGE_ALWAYS },
3352 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3353 {"advise", SHMEM_HUGE_ADVISE },
3357 const struct fs_parameter_spec shmem_fs_parameters[] = {
3358 fsparam_u32 ("gid", Opt_gid),
3359 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3360 fsparam_u32oct("mode", Opt_mode),
3361 fsparam_string("mpol", Opt_mpol),
3362 fsparam_string("nr_blocks", Opt_nr_blocks),
3363 fsparam_string("nr_inodes", Opt_nr_inodes),
3364 fsparam_string("size", Opt_size),
3365 fsparam_u32 ("uid", Opt_uid),
3366 fsparam_flag ("inode32", Opt_inode32),
3367 fsparam_flag ("inode64", Opt_inode64),
3371 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3373 struct shmem_options *ctx = fc->fs_private;
3374 struct fs_parse_result result;
3375 unsigned long long size;
3379 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3385 size = memparse(param->string, &rest);
3387 size <<= PAGE_SHIFT;
3388 size *= totalram_pages();
3394 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3395 ctx->seen |= SHMEM_SEEN_BLOCKS;
3398 ctx->blocks = memparse(param->string, &rest);
3401 ctx->seen |= SHMEM_SEEN_BLOCKS;
3404 ctx->inodes = memparse(param->string, &rest);
3407 ctx->seen |= SHMEM_SEEN_INODES;
3410 ctx->mode = result.uint_32 & 07777;
3413 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3414 if (!uid_valid(ctx->uid))
3418 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3419 if (!gid_valid(ctx->gid))
3423 ctx->huge = result.uint_32;
3424 if (ctx->huge != SHMEM_HUGE_NEVER &&
3425 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3426 has_transparent_hugepage()))
3427 goto unsupported_parameter;
3428 ctx->seen |= SHMEM_SEEN_HUGE;
3431 if (IS_ENABLED(CONFIG_NUMA)) {
3432 mpol_put(ctx->mpol);
3434 if (mpol_parse_str(param->string, &ctx->mpol))
3438 goto unsupported_parameter;
3440 ctx->full_inums = false;
3441 ctx->seen |= SHMEM_SEEN_INUMS;
3444 if (sizeof(ino_t) < 8) {
3446 "Cannot use inode64 with <64bit inums in kernel\n");
3448 ctx->full_inums = true;
3449 ctx->seen |= SHMEM_SEEN_INUMS;
3454 unsupported_parameter:
3455 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3457 return invalfc(fc, "Bad value for '%s'", param->key);
3460 static int shmem_parse_options(struct fs_context *fc, void *data)
3462 char *options = data;
3465 int err = security_sb_eat_lsm_opts(options, &fc->security);
3470 while (options != NULL) {
3471 char *this_char = options;
3474 * NUL-terminate this option: unfortunately,
3475 * mount options form a comma-separated list,
3476 * but mpol's nodelist may also contain commas.
3478 options = strchr(options, ',');
3479 if (options == NULL)
3482 if (!isdigit(*options)) {
3488 char *value = strchr(this_char,'=');
3494 len = strlen(value);
3496 err = vfs_parse_fs_string(fc, this_char, value, len);
3505 * Reconfigure a shmem filesystem.
3507 * Note that we disallow change from limited->unlimited blocks/inodes while any
3508 * are in use; but we must separately disallow unlimited->limited, because in
3509 * that case we have no record of how much is already in use.
3511 static int shmem_reconfigure(struct fs_context *fc)
3513 struct shmem_options *ctx = fc->fs_private;
3514 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3515 unsigned long inodes;
3518 spin_lock(&sbinfo->stat_lock);
3519 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3520 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3521 if (!sbinfo->max_blocks) {
3522 err = "Cannot retroactively limit size";
3525 if (percpu_counter_compare(&sbinfo->used_blocks,
3527 err = "Too small a size for current use";
3531 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3532 if (!sbinfo->max_inodes) {
3533 err = "Cannot retroactively limit inodes";
3536 if (ctx->inodes < inodes) {
3537 err = "Too few inodes for current use";
3542 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3543 sbinfo->next_ino > UINT_MAX) {
3544 err = "Current inum too high to switch to 32-bit inums";
3548 if (ctx->seen & SHMEM_SEEN_HUGE)
3549 sbinfo->huge = ctx->huge;
3550 if (ctx->seen & SHMEM_SEEN_INUMS)
3551 sbinfo->full_inums = ctx->full_inums;
3552 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3553 sbinfo->max_blocks = ctx->blocks;
3554 if (ctx->seen & SHMEM_SEEN_INODES) {
3555 sbinfo->max_inodes = ctx->inodes;
3556 sbinfo->free_inodes = ctx->inodes - inodes;
3560 * Preserve previous mempolicy unless mpol remount option was specified.
3563 mpol_put(sbinfo->mpol);
3564 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3567 spin_unlock(&sbinfo->stat_lock);
3570 spin_unlock(&sbinfo->stat_lock);
3571 return invalfc(fc, "%s", err);
3574 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3576 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3578 if (sbinfo->max_blocks != shmem_default_max_blocks())
3579 seq_printf(seq, ",size=%luk",
3580 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3581 if (sbinfo->max_inodes != shmem_default_max_inodes())
3582 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3583 if (sbinfo->mode != (0777 | S_ISVTX))
3584 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3585 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3586 seq_printf(seq, ",uid=%u",
3587 from_kuid_munged(&init_user_ns, sbinfo->uid));
3588 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3589 seq_printf(seq, ",gid=%u",
3590 from_kgid_munged(&init_user_ns, sbinfo->gid));
3593 * Showing inode{64,32} might be useful even if it's the system default,
3594 * since then people don't have to resort to checking both here and
3595 * /proc/config.gz to confirm 64-bit inums were successfully applied
3596 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3598 * We hide it when inode64 isn't the default and we are using 32-bit
3599 * inodes, since that probably just means the feature isn't even under
3604 * +-----------------+-----------------+
3605 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3606 * +------------------+-----------------+-----------------+
3607 * | full_inums=true | show | show |
3608 * | full_inums=false | show | hide |
3609 * +------------------+-----------------+-----------------+
3612 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3613 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3614 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3615 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3617 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3619 shmem_show_mpol(seq, sbinfo->mpol);
3623 #endif /* CONFIG_TMPFS */
3625 static void shmem_put_super(struct super_block *sb)
3627 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3629 free_percpu(sbinfo->ino_batch);
3630 percpu_counter_destroy(&sbinfo->used_blocks);
3631 mpol_put(sbinfo->mpol);
3633 sb->s_fs_info = NULL;
3636 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3638 struct shmem_options *ctx = fc->fs_private;
3639 struct inode *inode;
3640 struct shmem_sb_info *sbinfo;
3643 /* Round up to L1_CACHE_BYTES to resist false sharing */
3644 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3645 L1_CACHE_BYTES), GFP_KERNEL);
3649 sb->s_fs_info = sbinfo;
3653 * Per default we only allow half of the physical ram per
3654 * tmpfs instance, limiting inodes to one per page of lowmem;
3655 * but the internal instance is left unlimited.
3657 if (!(sb->s_flags & SB_KERNMOUNT)) {
3658 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3659 ctx->blocks = shmem_default_max_blocks();
3660 if (!(ctx->seen & SHMEM_SEEN_INODES))
3661 ctx->inodes = shmem_default_max_inodes();
3662 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3663 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3665 sb->s_flags |= SB_NOUSER;
3667 sb->s_export_op = &shmem_export_ops;
3668 sb->s_flags |= SB_NOSEC;
3670 sb->s_flags |= SB_NOUSER;
3672 sbinfo->max_blocks = ctx->blocks;
3673 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3674 if (sb->s_flags & SB_KERNMOUNT) {
3675 sbinfo->ino_batch = alloc_percpu(ino_t);
3676 if (!sbinfo->ino_batch)
3679 sbinfo->uid = ctx->uid;
3680 sbinfo->gid = ctx->gid;
3681 sbinfo->full_inums = ctx->full_inums;
3682 sbinfo->mode = ctx->mode;
3683 sbinfo->huge = ctx->huge;
3684 sbinfo->mpol = ctx->mpol;
3687 spin_lock_init(&sbinfo->stat_lock);
3688 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3690 spin_lock_init(&sbinfo->shrinklist_lock);
3691 INIT_LIST_HEAD(&sbinfo->shrinklist);
3693 sb->s_maxbytes = MAX_LFS_FILESIZE;
3694 sb->s_blocksize = PAGE_SIZE;
3695 sb->s_blocksize_bits = PAGE_SHIFT;
3696 sb->s_magic = TMPFS_MAGIC;
3697 sb->s_op = &shmem_ops;
3698 sb->s_time_gran = 1;
3699 #ifdef CONFIG_TMPFS_XATTR
3700 sb->s_xattr = shmem_xattr_handlers;
3702 #ifdef CONFIG_TMPFS_POSIX_ACL
3703 sb->s_flags |= SB_POSIXACL;
3705 uuid_gen(&sb->s_uuid);
3707 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3710 inode->i_uid = sbinfo->uid;
3711 inode->i_gid = sbinfo->gid;
3712 sb->s_root = d_make_root(inode);
3718 shmem_put_super(sb);
3722 static int shmem_get_tree(struct fs_context *fc)
3724 return get_tree_nodev(fc, shmem_fill_super);
3727 static void shmem_free_fc(struct fs_context *fc)
3729 struct shmem_options *ctx = fc->fs_private;
3732 mpol_put(ctx->mpol);
3737 static const struct fs_context_operations shmem_fs_context_ops = {
3738 .free = shmem_free_fc,
3739 .get_tree = shmem_get_tree,
3741 .parse_monolithic = shmem_parse_options,
3742 .parse_param = shmem_parse_one,
3743 .reconfigure = shmem_reconfigure,
3747 static struct kmem_cache *shmem_inode_cachep;
3749 static struct inode *shmem_alloc_inode(struct super_block *sb)
3751 struct shmem_inode_info *info;
3752 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3755 return &info->vfs_inode;
3758 static void shmem_free_in_core_inode(struct inode *inode)
3760 if (S_ISLNK(inode->i_mode))
3761 kfree(inode->i_link);
3762 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3765 static void shmem_destroy_inode(struct inode *inode)
3767 if (S_ISREG(inode->i_mode))
3768 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3771 static void shmem_init_inode(void *foo)
3773 struct shmem_inode_info *info = foo;
3774 inode_init_once(&info->vfs_inode);
3777 static void shmem_init_inodecache(void)
3779 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3780 sizeof(struct shmem_inode_info),
3781 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3784 static void shmem_destroy_inodecache(void)
3786 kmem_cache_destroy(shmem_inode_cachep);
3789 const struct address_space_operations shmem_aops = {
3790 .writepage = shmem_writepage,
3791 .set_page_dirty = __set_page_dirty_no_writeback,
3793 .write_begin = shmem_write_begin,
3794 .write_end = shmem_write_end,
3796 #ifdef CONFIG_MIGRATION
3797 .migratepage = migrate_page,
3799 .error_remove_page = generic_error_remove_page,
3801 EXPORT_SYMBOL(shmem_aops);
3803 static const struct file_operations shmem_file_operations = {
3805 .get_unmapped_area = shmem_get_unmapped_area,
3807 .llseek = shmem_file_llseek,
3808 .read_iter = shmem_file_read_iter,
3809 .write_iter = generic_file_write_iter,
3810 .fsync = noop_fsync,
3811 .splice_read = generic_file_splice_read,
3812 .splice_write = iter_file_splice_write,
3813 .fallocate = shmem_fallocate,
3817 static const struct inode_operations shmem_inode_operations = {
3818 .getattr = shmem_getattr,
3819 .setattr = shmem_setattr,
3820 #ifdef CONFIG_TMPFS_XATTR
3821 .listxattr = shmem_listxattr,
3822 .set_acl = simple_set_acl,
3826 static const struct inode_operations shmem_dir_inode_operations = {
3828 .create = shmem_create,
3829 .lookup = simple_lookup,
3831 .unlink = shmem_unlink,
3832 .symlink = shmem_symlink,
3833 .mkdir = shmem_mkdir,
3834 .rmdir = shmem_rmdir,
3835 .mknod = shmem_mknod,
3836 .rename = shmem_rename2,
3837 .tmpfile = shmem_tmpfile,
3839 #ifdef CONFIG_TMPFS_XATTR
3840 .listxattr = shmem_listxattr,
3842 #ifdef CONFIG_TMPFS_POSIX_ACL
3843 .setattr = shmem_setattr,
3844 .set_acl = simple_set_acl,
3848 static const struct inode_operations shmem_special_inode_operations = {
3849 #ifdef CONFIG_TMPFS_XATTR
3850 .listxattr = shmem_listxattr,
3852 #ifdef CONFIG_TMPFS_POSIX_ACL
3853 .setattr = shmem_setattr,
3854 .set_acl = simple_set_acl,
3858 static const struct super_operations shmem_ops = {
3859 .alloc_inode = shmem_alloc_inode,
3860 .free_inode = shmem_free_in_core_inode,
3861 .destroy_inode = shmem_destroy_inode,
3863 .statfs = shmem_statfs,
3864 .show_options = shmem_show_options,
3866 .evict_inode = shmem_evict_inode,
3867 .drop_inode = generic_delete_inode,
3868 .put_super = shmem_put_super,
3869 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3870 .nr_cached_objects = shmem_unused_huge_count,
3871 .free_cached_objects = shmem_unused_huge_scan,
3875 static const struct vm_operations_struct shmem_vm_ops = {
3876 .fault = shmem_fault,
3877 .map_pages = filemap_map_pages,
3879 .set_policy = shmem_set_policy,
3880 .get_policy = shmem_get_policy,
3884 int shmem_init_fs_context(struct fs_context *fc)
3886 struct shmem_options *ctx;
3888 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3892 ctx->mode = 0777 | S_ISVTX;
3893 ctx->uid = current_fsuid();
3894 ctx->gid = current_fsgid();
3896 fc->fs_private = ctx;
3897 fc->ops = &shmem_fs_context_ops;
3901 static struct file_system_type shmem_fs_type = {
3902 .owner = THIS_MODULE,
3904 .init_fs_context = shmem_init_fs_context,
3906 .parameters = shmem_fs_parameters,
3908 .kill_sb = kill_litter_super,
3909 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
3912 int __init shmem_init(void)
3916 shmem_init_inodecache();
3918 error = register_filesystem(&shmem_fs_type);
3920 pr_err("Could not register tmpfs\n");
3924 shm_mnt = kern_mount(&shmem_fs_type);
3925 if (IS_ERR(shm_mnt)) {
3926 error = PTR_ERR(shm_mnt);
3927 pr_err("Could not kern_mount tmpfs\n");
3931 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3932 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3933 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3935 shmem_huge = 0; /* just in case it was patched */
3940 unregister_filesystem(&shmem_fs_type);
3942 shmem_destroy_inodecache();
3943 shm_mnt = ERR_PTR(error);
3947 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3948 static ssize_t shmem_enabled_show(struct kobject *kobj,
3949 struct kobj_attribute *attr, char *buf)
3951 static const int values[] = {
3953 SHMEM_HUGE_WITHIN_SIZE,
3962 for (i = 0; i < ARRAY_SIZE(values); i++) {
3963 len += sysfs_emit_at(buf, len,
3964 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3966 shmem_format_huge(values[i]));
3969 len += sysfs_emit_at(buf, len, "\n");
3974 static ssize_t shmem_enabled_store(struct kobject *kobj,
3975 struct kobj_attribute *attr, const char *buf, size_t count)
3980 if (count + 1 > sizeof(tmp))
3982 memcpy(tmp, buf, count);
3984 if (count && tmp[count - 1] == '\n')
3985 tmp[count - 1] = '\0';
3987 huge = shmem_parse_huge(tmp);
3988 if (huge == -EINVAL)
3990 if (!has_transparent_hugepage() &&
3991 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3995 if (shmem_huge > SHMEM_HUGE_DENY)
3996 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4000 struct kobj_attribute shmem_enabled_attr =
4001 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4002 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4004 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4005 bool shmem_huge_enabled(struct vm_area_struct *vma)
4007 struct inode *inode = file_inode(vma->vm_file);
4008 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4012 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
4013 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
4015 if (shmem_huge == SHMEM_HUGE_FORCE)
4017 if (shmem_huge == SHMEM_HUGE_DENY)
4019 switch (sbinfo->huge) {
4020 case SHMEM_HUGE_NEVER:
4022 case SHMEM_HUGE_ALWAYS:
4024 case SHMEM_HUGE_WITHIN_SIZE:
4025 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4026 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4027 if (i_size >= HPAGE_PMD_SIZE &&
4028 i_size >> PAGE_SHIFT >= off)
4031 case SHMEM_HUGE_ADVISE:
4032 /* TODO: implement fadvise() hints */
4033 return (vma->vm_flags & VM_HUGEPAGE);
4039 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4041 #else /* !CONFIG_SHMEM */
4044 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4046 * This is intended for small system where the benefits of the full
4047 * shmem code (swap-backed and resource-limited) are outweighed by
4048 * their complexity. On systems without swap this code should be
4049 * effectively equivalent, but much lighter weight.
4052 static struct file_system_type shmem_fs_type = {
4054 .init_fs_context = ramfs_init_fs_context,
4055 .parameters = ramfs_fs_parameters,
4056 .kill_sb = kill_litter_super,
4057 .fs_flags = FS_USERNS_MOUNT,
4060 int __init shmem_init(void)
4062 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4064 shm_mnt = kern_mount(&shmem_fs_type);
4065 BUG_ON(IS_ERR(shm_mnt));
4070 int shmem_unuse(unsigned int type, bool frontswap,
4071 unsigned long *fs_pages_to_unuse)
4076 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4081 void shmem_unlock_mapping(struct address_space *mapping)
4086 unsigned long shmem_get_unmapped_area(struct file *file,
4087 unsigned long addr, unsigned long len,
4088 unsigned long pgoff, unsigned long flags)
4090 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4094 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4096 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4098 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4100 #define shmem_vm_ops generic_file_vm_ops
4101 #define shmem_file_operations ramfs_file_operations
4102 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4103 #define shmem_acct_size(flags, size) 0
4104 #define shmem_unacct_size(flags, size) do {} while (0)
4106 #endif /* CONFIG_SHMEM */
4110 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4111 unsigned long flags, unsigned int i_flags)
4113 struct inode *inode;
4117 return ERR_CAST(mnt);
4119 if (size < 0 || size > MAX_LFS_FILESIZE)
4120 return ERR_PTR(-EINVAL);
4122 if (shmem_acct_size(flags, size))
4123 return ERR_PTR(-ENOMEM);
4125 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4127 if (unlikely(!inode)) {
4128 shmem_unacct_size(flags, size);
4129 return ERR_PTR(-ENOSPC);
4131 inode->i_flags |= i_flags;
4132 inode->i_size = size;
4133 clear_nlink(inode); /* It is unlinked */
4134 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4136 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4137 &shmem_file_operations);
4144 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4145 * kernel internal. There will be NO LSM permission checks against the
4146 * underlying inode. So users of this interface must do LSM checks at a
4147 * higher layer. The users are the big_key and shm implementations. LSM
4148 * checks are provided at the key or shm level rather than the inode.
4149 * @name: name for dentry (to be seen in /proc/<pid>/maps
4150 * @size: size to be set for the file
4151 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4153 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4155 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4159 * shmem_file_setup - get an unlinked file living in tmpfs
4160 * @name: name for dentry (to be seen in /proc/<pid>/maps
4161 * @size: size to be set for the file
4162 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4164 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4166 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4168 EXPORT_SYMBOL_GPL(shmem_file_setup);
4171 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4172 * @mnt: the tmpfs mount where the file will be created
4173 * @name: name for dentry (to be seen in /proc/<pid>/maps
4174 * @size: size to be set for the file
4175 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4177 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4178 loff_t size, unsigned long flags)
4180 return __shmem_file_setup(mnt, name, size, flags, 0);
4182 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4185 * shmem_zero_setup - setup a shared anonymous mapping
4186 * @vma: the vma to be mmapped is prepared by do_mmap
4188 int shmem_zero_setup(struct vm_area_struct *vma)
4191 loff_t size = vma->vm_end - vma->vm_start;
4194 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4195 * between XFS directory reading and selinux: since this file is only
4196 * accessible to the user through its mapping, use S_PRIVATE flag to
4197 * bypass file security, in the same way as shmem_kernel_file_setup().
4199 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4201 return PTR_ERR(file);
4205 vma->vm_file = file;
4206 vma->vm_ops = &shmem_vm_ops;
4208 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4209 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4210 (vma->vm_end & HPAGE_PMD_MASK)) {
4211 khugepaged_enter(vma, vma->vm_flags);
4218 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4219 * @mapping: the page's address_space
4220 * @index: the page index
4221 * @gfp: the page allocator flags to use if allocating
4223 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4224 * with any new page allocations done using the specified allocation flags.
4225 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4226 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4227 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4229 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4230 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4232 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4233 pgoff_t index, gfp_t gfp)
4236 struct inode *inode = mapping->host;
4240 BUG_ON(!shmem_mapping(mapping));
4241 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4242 gfp, NULL, NULL, NULL);
4244 page = ERR_PTR(error);
4250 * The tiny !SHMEM case uses ramfs without swap
4252 return read_cache_page_gfp(mapping, index, gfp);
4255 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);