2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
43 static struct vfsmount *shm_mnt;
47 * This virtual memory filesystem is heavily based on the ramfs. It
48 * extends ramfs by the ability to use swap and honor resource limits
49 * which makes it a completely usable filesystem.
52 #include <linux/xattr.h>
53 #include <linux/exportfs.h>
54 #include <linux/posix_acl.h>
55 #include <linux/posix_acl_xattr.h>
56 #include <linux/mman.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/backing-dev.h>
60 #include <linux/shmem_fs.h>
61 #include <linux/writeback.h>
62 #include <linux/blkdev.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/userfaultfd_k.h>
80 #include <linux/rmap.h>
81 #include <linux/uuid.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_mutex making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc {
102 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
103 pgoff_t start; /* start of range currently being fallocated */
104 pgoff_t next; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
109 struct shmem_options {
110 unsigned long long blocks;
111 unsigned long long inodes;
112 struct mempolicy *mpol;
119 #define SHMEM_SEEN_BLOCKS 1
120 #define SHMEM_SEEN_INODES 2
121 #define SHMEM_SEEN_HUGE 4
122 #define SHMEM_SEEN_INUMS 8
126 static unsigned long shmem_default_max_blocks(void)
128 return totalram_pages() / 2;
131 static unsigned long shmem_default_max_inodes(void)
133 unsigned long nr_pages = totalram_pages();
135 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
139 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
140 struct page **pagep, enum sgp_type sgp,
141 gfp_t gfp, struct vm_area_struct *vma,
142 vm_fault_t *fault_type);
143 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 struct vm_fault *vmf, vm_fault_t *fault_type);
148 int shmem_getpage(struct inode *inode, pgoff_t index,
149 struct page **pagep, enum sgp_type sgp)
151 return shmem_getpage_gfp(inode, index, pagep, sgp,
152 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
155 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
157 return sb->s_fs_info;
161 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
162 * for shared memory and for shared anonymous (/dev/zero) mappings
163 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
164 * consistent with the pre-accounting of private mappings ...
166 static inline int shmem_acct_size(unsigned long flags, loff_t size)
168 return (flags & VM_NORESERVE) ?
169 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
172 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
174 if (!(flags & VM_NORESERVE))
175 vm_unacct_memory(VM_ACCT(size));
178 static inline int shmem_reacct_size(unsigned long flags,
179 loff_t oldsize, loff_t newsize)
181 if (!(flags & VM_NORESERVE)) {
182 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
183 return security_vm_enough_memory_mm(current->mm,
184 VM_ACCT(newsize) - VM_ACCT(oldsize));
185 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
186 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
192 * ... whereas tmpfs objects are accounted incrementally as
193 * pages are allocated, in order to allow large sparse files.
194 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
195 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
197 static inline int shmem_acct_block(unsigned long flags, long pages)
199 if (!(flags & VM_NORESERVE))
202 return security_vm_enough_memory_mm(current->mm,
203 pages * VM_ACCT(PAGE_SIZE));
206 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
208 if (flags & VM_NORESERVE)
209 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
212 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
214 struct shmem_inode_info *info = SHMEM_I(inode);
215 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
217 if (shmem_acct_block(info->flags, pages))
220 if (sbinfo->max_blocks) {
221 if (percpu_counter_compare(&sbinfo->used_blocks,
222 sbinfo->max_blocks - pages) > 0)
224 percpu_counter_add(&sbinfo->used_blocks, pages);
230 shmem_unacct_blocks(info->flags, pages);
234 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
236 struct shmem_inode_info *info = SHMEM_I(inode);
237 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
239 if (sbinfo->max_blocks)
240 percpu_counter_sub(&sbinfo->used_blocks, pages);
241 shmem_unacct_blocks(info->flags, pages);
244 static const struct super_operations shmem_ops;
245 const struct address_space_operations shmem_aops;
246 static const struct file_operations shmem_file_operations;
247 static const struct inode_operations shmem_inode_operations;
248 static const struct inode_operations shmem_dir_inode_operations;
249 static const struct inode_operations shmem_special_inode_operations;
250 static const struct vm_operations_struct shmem_vm_ops;
251 static struct file_system_type shmem_fs_type;
253 bool vma_is_shmem(struct vm_area_struct *vma)
255 return vma->vm_ops == &shmem_vm_ops;
258 static LIST_HEAD(shmem_swaplist);
259 static DEFINE_MUTEX(shmem_swaplist_mutex);
262 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
263 * produces a novel ino for the newly allocated inode.
265 * It may also be called when making a hard link to permit the space needed by
266 * each dentry. However, in that case, no new inode number is needed since that
267 * internally draws from another pool of inode numbers (currently global
268 * get_next_ino()). This case is indicated by passing NULL as inop.
270 #define SHMEM_INO_BATCH 1024
271 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
273 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
276 if (!(sb->s_flags & SB_KERNMOUNT)) {
277 raw_spin_lock(&sbinfo->stat_lock);
278 if (sbinfo->max_inodes) {
279 if (!sbinfo->free_inodes) {
280 raw_spin_unlock(&sbinfo->stat_lock);
283 sbinfo->free_inodes--;
286 ino = sbinfo->next_ino++;
287 if (unlikely(is_zero_ino(ino)))
288 ino = sbinfo->next_ino++;
289 if (unlikely(!sbinfo->full_inums &&
292 * Emulate get_next_ino uint wraparound for
295 if (IS_ENABLED(CONFIG_64BIT))
296 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
297 __func__, MINOR(sb->s_dev));
298 sbinfo->next_ino = 1;
299 ino = sbinfo->next_ino++;
303 raw_spin_unlock(&sbinfo->stat_lock);
306 * __shmem_file_setup, one of our callers, is lock-free: it
307 * doesn't hold stat_lock in shmem_reserve_inode since
308 * max_inodes is always 0, and is called from potentially
309 * unknown contexts. As such, use a per-cpu batched allocator
310 * which doesn't require the per-sb stat_lock unless we are at
311 * the batch boundary.
313 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
314 * shmem mounts are not exposed to userspace, so we don't need
315 * to worry about things like glibc compatibility.
319 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
321 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
322 raw_spin_lock(&sbinfo->stat_lock);
323 ino = sbinfo->next_ino;
324 sbinfo->next_ino += SHMEM_INO_BATCH;
325 raw_spin_unlock(&sbinfo->stat_lock);
326 if (unlikely(is_zero_ino(ino)))
337 static void shmem_free_inode(struct super_block *sb)
339 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
340 if (sbinfo->max_inodes) {
341 raw_spin_lock(&sbinfo->stat_lock);
342 sbinfo->free_inodes++;
343 raw_spin_unlock(&sbinfo->stat_lock);
348 * shmem_recalc_inode - recalculate the block usage of an inode
349 * @inode: inode to recalc
351 * We have to calculate the free blocks since the mm can drop
352 * undirtied hole pages behind our back.
354 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
355 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
357 * It has to be called with the spinlock held.
359 static void shmem_recalc_inode(struct inode *inode)
361 struct shmem_inode_info *info = SHMEM_I(inode);
364 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
366 info->alloced -= freed;
367 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
368 shmem_inode_unacct_blocks(inode, freed);
372 bool shmem_charge(struct inode *inode, long pages)
374 struct shmem_inode_info *info = SHMEM_I(inode);
377 if (!shmem_inode_acct_block(inode, pages))
380 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
381 inode->i_mapping->nrpages += pages;
383 spin_lock_irqsave(&info->lock, flags);
384 info->alloced += pages;
385 inode->i_blocks += pages * BLOCKS_PER_PAGE;
386 shmem_recalc_inode(inode);
387 spin_unlock_irqrestore(&info->lock, flags);
392 void shmem_uncharge(struct inode *inode, long pages)
394 struct shmem_inode_info *info = SHMEM_I(inode);
397 /* nrpages adjustment done by __delete_from_page_cache() or caller */
399 spin_lock_irqsave(&info->lock, flags);
400 info->alloced -= pages;
401 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
402 shmem_recalc_inode(inode);
403 spin_unlock_irqrestore(&info->lock, flags);
405 shmem_inode_unacct_blocks(inode, pages);
409 * Replace item expected in xarray by a new item, while holding xa_lock.
411 static int shmem_replace_entry(struct address_space *mapping,
412 pgoff_t index, void *expected, void *replacement)
414 XA_STATE(xas, &mapping->i_pages, index);
417 VM_BUG_ON(!expected);
418 VM_BUG_ON(!replacement);
419 item = xas_load(&xas);
420 if (item != expected)
422 xas_store(&xas, replacement);
427 * Sometimes, before we decide whether to proceed or to fail, we must check
428 * that an entry was not already brought back from swap by a racing thread.
430 * Checking page is not enough: by the time a SwapCache page is locked, it
431 * might be reused, and again be SwapCache, using the same swap as before.
433 static bool shmem_confirm_swap(struct address_space *mapping,
434 pgoff_t index, swp_entry_t swap)
436 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
440 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
443 * disables huge pages for the mount;
445 * enables huge pages for the mount;
446 * SHMEM_HUGE_WITHIN_SIZE:
447 * only allocate huge pages if the page will be fully within i_size,
448 * also respect fadvise()/madvise() hints;
450 * only allocate huge pages if requested with fadvise()/madvise();
453 #define SHMEM_HUGE_NEVER 0
454 #define SHMEM_HUGE_ALWAYS 1
455 #define SHMEM_HUGE_WITHIN_SIZE 2
456 #define SHMEM_HUGE_ADVISE 3
460 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
463 * disables huge on shm_mnt and all mounts, for emergency use;
465 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
468 #define SHMEM_HUGE_DENY (-1)
469 #define SHMEM_HUGE_FORCE (-2)
471 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
472 /* ifdef here to avoid bloating shmem.o when not necessary */
474 static int shmem_huge __read_mostly;
476 #if defined(CONFIG_SYSFS)
477 static int shmem_parse_huge(const char *str)
479 if (!strcmp(str, "never"))
480 return SHMEM_HUGE_NEVER;
481 if (!strcmp(str, "always"))
482 return SHMEM_HUGE_ALWAYS;
483 if (!strcmp(str, "within_size"))
484 return SHMEM_HUGE_WITHIN_SIZE;
485 if (!strcmp(str, "advise"))
486 return SHMEM_HUGE_ADVISE;
487 if (!strcmp(str, "deny"))
488 return SHMEM_HUGE_DENY;
489 if (!strcmp(str, "force"))
490 return SHMEM_HUGE_FORCE;
495 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
496 static const char *shmem_format_huge(int huge)
499 case SHMEM_HUGE_NEVER:
501 case SHMEM_HUGE_ALWAYS:
503 case SHMEM_HUGE_WITHIN_SIZE:
504 return "within_size";
505 case SHMEM_HUGE_ADVISE:
507 case SHMEM_HUGE_DENY:
509 case SHMEM_HUGE_FORCE:
518 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
519 struct shrink_control *sc, unsigned long nr_to_split)
521 LIST_HEAD(list), *pos, *next;
522 LIST_HEAD(to_remove);
524 struct shmem_inode_info *info;
526 unsigned long batch = sc ? sc->nr_to_scan : 128;
527 int removed = 0, split = 0;
529 if (list_empty(&sbinfo->shrinklist))
532 spin_lock(&sbinfo->shrinklist_lock);
533 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
534 info = list_entry(pos, struct shmem_inode_info, shrinklist);
537 inode = igrab(&info->vfs_inode);
539 /* inode is about to be evicted */
541 list_del_init(&info->shrinklist);
546 /* Check if there's anything to gain */
547 if (round_up(inode->i_size, PAGE_SIZE) ==
548 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
549 list_move(&info->shrinklist, &to_remove);
554 list_move(&info->shrinklist, &list);
559 spin_unlock(&sbinfo->shrinklist_lock);
561 list_for_each_safe(pos, next, &to_remove) {
562 info = list_entry(pos, struct shmem_inode_info, shrinklist);
563 inode = &info->vfs_inode;
564 list_del_init(&info->shrinklist);
568 list_for_each_safe(pos, next, &list) {
571 info = list_entry(pos, struct shmem_inode_info, shrinklist);
572 inode = &info->vfs_inode;
574 if (nr_to_split && split >= nr_to_split)
577 page = find_get_page(inode->i_mapping,
578 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
582 /* No huge page at the end of the file: nothing to split */
583 if (!PageTransHuge(page)) {
589 * Leave the inode on the list if we failed to lock
590 * the page at this time.
592 * Waiting for the lock may lead to deadlock in the
595 if (!trylock_page(page)) {
600 ret = split_huge_page(page);
604 /* If split failed leave the inode on the list */
610 list_del_init(&info->shrinklist);
616 spin_lock(&sbinfo->shrinklist_lock);
617 list_splice_tail(&list, &sbinfo->shrinklist);
618 sbinfo->shrinklist_len -= removed;
619 spin_unlock(&sbinfo->shrinklist_lock);
624 static long shmem_unused_huge_scan(struct super_block *sb,
625 struct shrink_control *sc)
627 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
629 if (!READ_ONCE(sbinfo->shrinklist_len))
632 return shmem_unused_huge_shrink(sbinfo, sc, 0);
635 static long shmem_unused_huge_count(struct super_block *sb,
636 struct shrink_control *sc)
638 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
639 return READ_ONCE(sbinfo->shrinklist_len);
641 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
643 #define shmem_huge SHMEM_HUGE_DENY
645 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
646 struct shrink_control *sc, unsigned long nr_to_split)
650 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
652 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
654 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
655 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
656 shmem_huge != SHMEM_HUGE_DENY)
662 * Like add_to_page_cache_locked, but error if expected item has gone.
664 static int shmem_add_to_page_cache(struct page *page,
665 struct address_space *mapping,
666 pgoff_t index, void *expected, gfp_t gfp,
667 struct mm_struct *charge_mm)
669 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
671 unsigned long nr = compound_nr(page);
674 VM_BUG_ON_PAGE(PageTail(page), page);
675 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
676 VM_BUG_ON_PAGE(!PageLocked(page), page);
677 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
678 VM_BUG_ON(expected && PageTransHuge(page));
680 page_ref_add(page, nr);
681 page->mapping = mapping;
684 if (!PageSwapCache(page)) {
685 error = mem_cgroup_charge(page, charge_mm, gfp);
687 if (PageTransHuge(page)) {
688 count_vm_event(THP_FILE_FALLBACK);
689 count_vm_event(THP_FILE_FALLBACK_CHARGE);
694 cgroup_throttle_swaprate(page, gfp);
699 entry = xas_find_conflict(&xas);
700 if (entry != expected)
701 xas_set_err(&xas, -EEXIST);
702 xas_create_range(&xas);
706 xas_store(&xas, page);
711 if (PageTransHuge(page)) {
712 count_vm_event(THP_FILE_ALLOC);
713 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
715 mapping->nrpages += nr;
716 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
717 __mod_lruvec_page_state(page, NR_SHMEM, nr);
719 xas_unlock_irq(&xas);
720 } while (xas_nomem(&xas, gfp));
722 if (xas_error(&xas)) {
723 error = xas_error(&xas);
729 page->mapping = NULL;
730 page_ref_sub(page, nr);
735 * Like delete_from_page_cache, but substitutes swap for page.
737 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
739 struct address_space *mapping = page->mapping;
742 VM_BUG_ON_PAGE(PageCompound(page), page);
744 xa_lock_irq(&mapping->i_pages);
745 error = shmem_replace_entry(mapping, page->index, page, radswap);
746 page->mapping = NULL;
748 __dec_lruvec_page_state(page, NR_FILE_PAGES);
749 __dec_lruvec_page_state(page, NR_SHMEM);
750 xa_unlock_irq(&mapping->i_pages);
756 * Remove swap entry from page cache, free the swap and its page cache.
758 static int shmem_free_swap(struct address_space *mapping,
759 pgoff_t index, void *radswap)
763 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
766 free_swap_and_cache(radix_to_swp_entry(radswap));
771 * Determine (in bytes) how many of the shmem object's pages mapped by the
772 * given offsets are swapped out.
774 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
775 * as long as the inode doesn't go away and racy results are not a problem.
777 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
778 pgoff_t start, pgoff_t end)
780 XA_STATE(xas, &mapping->i_pages, start);
782 unsigned long swapped = 0;
785 xas_for_each(&xas, page, end - 1) {
786 if (xas_retry(&xas, page))
788 if (xa_is_value(page))
791 if (need_resched()) {
799 return swapped << PAGE_SHIFT;
803 * Determine (in bytes) how many of the shmem object's pages mapped by the
804 * given vma is swapped out.
806 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
807 * as long as the inode doesn't go away and racy results are not a problem.
809 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
811 struct inode *inode = file_inode(vma->vm_file);
812 struct shmem_inode_info *info = SHMEM_I(inode);
813 struct address_space *mapping = inode->i_mapping;
814 unsigned long swapped;
816 /* Be careful as we don't hold info->lock */
817 swapped = READ_ONCE(info->swapped);
820 * The easier cases are when the shmem object has nothing in swap, or
821 * the vma maps it whole. Then we can simply use the stats that we
827 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
828 return swapped << PAGE_SHIFT;
830 /* Here comes the more involved part */
831 return shmem_partial_swap_usage(mapping,
832 linear_page_index(vma, vma->vm_start),
833 linear_page_index(vma, vma->vm_end));
837 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
839 void shmem_unlock_mapping(struct address_space *mapping)
846 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
848 while (!mapping_unevictable(mapping)) {
849 if (!pagevec_lookup(&pvec, mapping, &index))
851 check_move_unevictable_pages(&pvec);
852 pagevec_release(&pvec);
858 * Check whether a hole-punch or truncation needs to split a huge page,
859 * returning true if no split was required, or the split has been successful.
861 * Eviction (or truncation to 0 size) should never need to split a huge page;
862 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
863 * head, and then succeeded to trylock on tail.
865 * A split can only succeed when there are no additional references on the
866 * huge page: so the split below relies upon find_get_entries() having stopped
867 * when it found a subpage of the huge page, without getting further references.
869 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
871 if (!PageTransCompound(page))
874 /* Just proceed to delete a huge page wholly within the range punched */
875 if (PageHead(page) &&
876 page->index >= start && page->index + HPAGE_PMD_NR <= end)
879 /* Try to split huge page, so we can truly punch the hole or truncate */
880 return split_huge_page(page) >= 0;
884 * Remove range of pages and swap entries from page cache, and free them.
885 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
887 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
890 struct address_space *mapping = inode->i_mapping;
891 struct shmem_inode_info *info = SHMEM_I(inode);
892 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
893 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
894 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
895 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
897 pgoff_t indices[PAGEVEC_SIZE];
898 long nr_swaps_freed = 0;
903 end = -1; /* unsigned, so actually very big */
907 while (index < end && find_lock_entries(mapping, index, end - 1,
909 for (i = 0; i < pagevec_count(&pvec); i++) {
910 struct page *page = pvec.pages[i];
914 if (xa_is_value(page)) {
917 nr_swaps_freed += !shmem_free_swap(mapping,
921 index += thp_nr_pages(page) - 1;
923 if (!unfalloc || !PageUptodate(page))
924 truncate_inode_page(mapping, page);
927 pagevec_remove_exceptionals(&pvec);
928 pagevec_release(&pvec);
934 struct page *page = NULL;
935 shmem_getpage(inode, start - 1, &page, SGP_READ);
937 unsigned int top = PAGE_SIZE;
942 zero_user_segment(page, partial_start, top);
943 set_page_dirty(page);
949 struct page *page = NULL;
950 shmem_getpage(inode, end, &page, SGP_READ);
952 zero_user_segment(page, 0, partial_end);
953 set_page_dirty(page);
962 while (index < end) {
965 if (!find_get_entries(mapping, index, end - 1, &pvec,
967 /* If all gone or hole-punch or unfalloc, we're done */
968 if (index == start || end != -1)
970 /* But if truncating, restart to make sure all gone */
974 for (i = 0; i < pagevec_count(&pvec); i++) {
975 struct page *page = pvec.pages[i];
978 if (xa_is_value(page)) {
981 if (shmem_free_swap(mapping, index, page)) {
982 /* Swap was replaced by page: retry */
992 if (!unfalloc || !PageUptodate(page)) {
993 if (page_mapping(page) != mapping) {
994 /* Page was replaced by swap: retry */
999 VM_BUG_ON_PAGE(PageWriteback(page), page);
1000 if (shmem_punch_compound(page, start, end))
1001 truncate_inode_page(mapping, page);
1002 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1003 /* Wipe the page and don't get stuck */
1004 clear_highpage(page);
1005 flush_dcache_page(page);
1006 set_page_dirty(page);
1008 round_up(start, HPAGE_PMD_NR))
1014 pagevec_remove_exceptionals(&pvec);
1015 pagevec_release(&pvec);
1019 spin_lock_irq(&info->lock);
1020 info->swapped -= nr_swaps_freed;
1021 shmem_recalc_inode(inode);
1022 spin_unlock_irq(&info->lock);
1025 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1027 shmem_undo_range(inode, lstart, lend, false);
1028 inode->i_ctime = inode->i_mtime = current_time(inode);
1030 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1032 static int shmem_getattr(struct user_namespace *mnt_userns,
1033 const struct path *path, struct kstat *stat,
1034 u32 request_mask, unsigned int query_flags)
1036 struct inode *inode = path->dentry->d_inode;
1037 struct shmem_inode_info *info = SHMEM_I(inode);
1038 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1040 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1041 spin_lock_irq(&info->lock);
1042 shmem_recalc_inode(inode);
1043 spin_unlock_irq(&info->lock);
1045 generic_fillattr(&init_user_ns, inode, stat);
1047 if (is_huge_enabled(sb_info))
1048 stat->blksize = HPAGE_PMD_SIZE;
1053 static int shmem_setattr(struct user_namespace *mnt_userns,
1054 struct dentry *dentry, struct iattr *attr)
1056 struct inode *inode = d_inode(dentry);
1057 struct shmem_inode_info *info = SHMEM_I(inode);
1058 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1061 error = setattr_prepare(&init_user_ns, dentry, attr);
1065 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1066 loff_t oldsize = inode->i_size;
1067 loff_t newsize = attr->ia_size;
1069 /* protected by i_mutex */
1070 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1071 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1074 if (newsize != oldsize) {
1075 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1079 i_size_write(inode, newsize);
1080 inode->i_ctime = inode->i_mtime = current_time(inode);
1082 if (newsize <= oldsize) {
1083 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1084 if (oldsize > holebegin)
1085 unmap_mapping_range(inode->i_mapping,
1088 shmem_truncate_range(inode,
1089 newsize, (loff_t)-1);
1090 /* unmap again to remove racily COWed private pages */
1091 if (oldsize > holebegin)
1092 unmap_mapping_range(inode->i_mapping,
1096 * Part of the huge page can be beyond i_size: subject
1097 * to shrink under memory pressure.
1099 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1100 spin_lock(&sbinfo->shrinklist_lock);
1102 * _careful to defend against unlocked access to
1103 * ->shrink_list in shmem_unused_huge_shrink()
1105 if (list_empty_careful(&info->shrinklist)) {
1106 list_add_tail(&info->shrinklist,
1107 &sbinfo->shrinklist);
1108 sbinfo->shrinklist_len++;
1110 spin_unlock(&sbinfo->shrinklist_lock);
1115 setattr_copy(&init_user_ns, inode, attr);
1116 if (attr->ia_valid & ATTR_MODE)
1117 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1121 static void shmem_evict_inode(struct inode *inode)
1123 struct shmem_inode_info *info = SHMEM_I(inode);
1124 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1126 if (shmem_mapping(inode->i_mapping)) {
1127 shmem_unacct_size(info->flags, inode->i_size);
1129 shmem_truncate_range(inode, 0, (loff_t)-1);
1130 if (!list_empty(&info->shrinklist)) {
1131 spin_lock(&sbinfo->shrinklist_lock);
1132 if (!list_empty(&info->shrinklist)) {
1133 list_del_init(&info->shrinklist);
1134 sbinfo->shrinklist_len--;
1136 spin_unlock(&sbinfo->shrinklist_lock);
1138 while (!list_empty(&info->swaplist)) {
1139 /* Wait while shmem_unuse() is scanning this inode... */
1140 wait_var_event(&info->stop_eviction,
1141 !atomic_read(&info->stop_eviction));
1142 mutex_lock(&shmem_swaplist_mutex);
1143 /* ...but beware of the race if we peeked too early */
1144 if (!atomic_read(&info->stop_eviction))
1145 list_del_init(&info->swaplist);
1146 mutex_unlock(&shmem_swaplist_mutex);
1150 simple_xattrs_free(&info->xattrs);
1151 WARN_ON(inode->i_blocks);
1152 shmem_free_inode(inode->i_sb);
1156 static int shmem_find_swap_entries(struct address_space *mapping,
1157 pgoff_t start, unsigned int nr_entries,
1158 struct page **entries, pgoff_t *indices,
1159 unsigned int type, bool frontswap)
1161 XA_STATE(xas, &mapping->i_pages, start);
1164 unsigned int ret = 0;
1170 xas_for_each(&xas, page, ULONG_MAX) {
1171 if (xas_retry(&xas, page))
1174 if (!xa_is_value(page))
1177 entry = radix_to_swp_entry(page);
1178 if (swp_type(entry) != type)
1181 !frontswap_test(swap_info[type], swp_offset(entry)))
1184 indices[ret] = xas.xa_index;
1185 entries[ret] = page;
1187 if (need_resched()) {
1191 if (++ret == nr_entries)
1200 * Move the swapped pages for an inode to page cache. Returns the count
1201 * of pages swapped in, or the error in case of failure.
1203 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1209 struct address_space *mapping = inode->i_mapping;
1211 for (i = 0; i < pvec.nr; i++) {
1212 struct page *page = pvec.pages[i];
1214 if (!xa_is_value(page))
1216 error = shmem_swapin_page(inode, indices[i],
1218 mapping_gfp_mask(mapping),
1225 if (error == -ENOMEM)
1229 return error ? error : ret;
1233 * If swap found in inode, free it and move page from swapcache to filecache.
1235 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1236 bool frontswap, unsigned long *fs_pages_to_unuse)
1238 struct address_space *mapping = inode->i_mapping;
1240 struct pagevec pvec;
1241 pgoff_t indices[PAGEVEC_SIZE];
1242 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1245 pagevec_init(&pvec);
1247 unsigned int nr_entries = PAGEVEC_SIZE;
1249 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1250 nr_entries = *fs_pages_to_unuse;
1252 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1253 pvec.pages, indices,
1260 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1264 if (frontswap_partial) {
1265 *fs_pages_to_unuse -= ret;
1266 if (*fs_pages_to_unuse == 0) {
1267 ret = FRONTSWAP_PAGES_UNUSED;
1272 start = indices[pvec.nr - 1];
1279 * Read all the shared memory data that resides in the swap
1280 * device 'type' back into memory, so the swap device can be
1283 int shmem_unuse(unsigned int type, bool frontswap,
1284 unsigned long *fs_pages_to_unuse)
1286 struct shmem_inode_info *info, *next;
1289 if (list_empty(&shmem_swaplist))
1292 mutex_lock(&shmem_swaplist_mutex);
1293 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1294 if (!info->swapped) {
1295 list_del_init(&info->swaplist);
1299 * Drop the swaplist mutex while searching the inode for swap;
1300 * but before doing so, make sure shmem_evict_inode() will not
1301 * remove placeholder inode from swaplist, nor let it be freed
1302 * (igrab() would protect from unlink, but not from unmount).
1304 atomic_inc(&info->stop_eviction);
1305 mutex_unlock(&shmem_swaplist_mutex);
1307 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1311 mutex_lock(&shmem_swaplist_mutex);
1312 next = list_next_entry(info, swaplist);
1314 list_del_init(&info->swaplist);
1315 if (atomic_dec_and_test(&info->stop_eviction))
1316 wake_up_var(&info->stop_eviction);
1320 mutex_unlock(&shmem_swaplist_mutex);
1326 * Move the page from the page cache to the swap cache.
1328 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1330 struct shmem_inode_info *info;
1331 struct address_space *mapping;
1332 struct inode *inode;
1336 VM_BUG_ON_PAGE(PageCompound(page), page);
1337 BUG_ON(!PageLocked(page));
1338 mapping = page->mapping;
1339 index = page->index;
1340 inode = mapping->host;
1341 info = SHMEM_I(inode);
1342 if (info->flags & VM_LOCKED)
1344 if (!total_swap_pages)
1348 * Our capabilities prevent regular writeback or sync from ever calling
1349 * shmem_writepage; but a stacking filesystem might use ->writepage of
1350 * its underlying filesystem, in which case tmpfs should write out to
1351 * swap only in response to memory pressure, and not for the writeback
1354 if (!wbc->for_reclaim) {
1355 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1360 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1361 * value into swapfile.c, the only way we can correctly account for a
1362 * fallocated page arriving here is now to initialize it and write it.
1364 * That's okay for a page already fallocated earlier, but if we have
1365 * not yet completed the fallocation, then (a) we want to keep track
1366 * of this page in case we have to undo it, and (b) it may not be a
1367 * good idea to continue anyway, once we're pushing into swap. So
1368 * reactivate the page, and let shmem_fallocate() quit when too many.
1370 if (!PageUptodate(page)) {
1371 if (inode->i_private) {
1372 struct shmem_falloc *shmem_falloc;
1373 spin_lock(&inode->i_lock);
1374 shmem_falloc = inode->i_private;
1376 !shmem_falloc->waitq &&
1377 index >= shmem_falloc->start &&
1378 index < shmem_falloc->next)
1379 shmem_falloc->nr_unswapped++;
1381 shmem_falloc = NULL;
1382 spin_unlock(&inode->i_lock);
1386 clear_highpage(page);
1387 flush_dcache_page(page);
1388 SetPageUptodate(page);
1391 swap = get_swap_page(page);
1396 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1397 * if it's not already there. Do it now before the page is
1398 * moved to swap cache, when its pagelock no longer protects
1399 * the inode from eviction. But don't unlock the mutex until
1400 * we've incremented swapped, because shmem_unuse_inode() will
1401 * prune a !swapped inode from the swaplist under this mutex.
1403 mutex_lock(&shmem_swaplist_mutex);
1404 if (list_empty(&info->swaplist))
1405 list_add(&info->swaplist, &shmem_swaplist);
1407 if (add_to_swap_cache(page, swap,
1408 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1410 spin_lock_irq(&info->lock);
1411 shmem_recalc_inode(inode);
1413 spin_unlock_irq(&info->lock);
1415 swap_shmem_alloc(swap);
1416 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1418 mutex_unlock(&shmem_swaplist_mutex);
1419 BUG_ON(page_mapped(page));
1420 swap_writepage(page, wbc);
1424 mutex_unlock(&shmem_swaplist_mutex);
1425 put_swap_page(page, swap);
1427 set_page_dirty(page);
1428 if (wbc->for_reclaim)
1429 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1434 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1435 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1439 if (!mpol || mpol->mode == MPOL_DEFAULT)
1440 return; /* show nothing */
1442 mpol_to_str(buffer, sizeof(buffer), mpol);
1444 seq_printf(seq, ",mpol=%s", buffer);
1447 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1449 struct mempolicy *mpol = NULL;
1451 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1452 mpol = sbinfo->mpol;
1454 raw_spin_unlock(&sbinfo->stat_lock);
1458 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1459 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1462 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1466 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1468 #define vm_policy vm_private_data
1471 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1472 struct shmem_inode_info *info, pgoff_t index)
1474 /* Create a pseudo vma that just contains the policy */
1475 vma_init(vma, NULL);
1476 /* Bias interleave by inode number to distribute better across nodes */
1477 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1478 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1481 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1483 /* Drop reference taken by mpol_shared_policy_lookup() */
1484 mpol_cond_put(vma->vm_policy);
1487 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1488 struct shmem_inode_info *info, pgoff_t index)
1490 struct vm_area_struct pvma;
1492 struct vm_fault vmf = {
1496 shmem_pseudo_vma_init(&pvma, info, index);
1497 page = swap_cluster_readahead(swap, gfp, &vmf);
1498 shmem_pseudo_vma_destroy(&pvma);
1504 * Make sure huge_gfp is always more limited than limit_gfp.
1505 * Some of the flags set permissions, while others set limitations.
1507 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1509 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1510 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1511 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1512 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1514 /* Allow allocations only from the originally specified zones. */
1515 result |= zoneflags;
1518 * Minimize the result gfp by taking the union with the deny flags,
1519 * and the intersection of the allow flags.
1521 result |= (limit_gfp & denyflags);
1522 result |= (huge_gfp & limit_gfp) & allowflags;
1527 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1528 struct shmem_inode_info *info, pgoff_t index)
1530 struct vm_area_struct pvma;
1531 struct address_space *mapping = info->vfs_inode.i_mapping;
1535 hindex = round_down(index, HPAGE_PMD_NR);
1536 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1540 shmem_pseudo_vma_init(&pvma, info, hindex);
1541 page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(),
1543 shmem_pseudo_vma_destroy(&pvma);
1545 prep_transhuge_page(page);
1547 count_vm_event(THP_FILE_FALLBACK);
1551 static struct page *shmem_alloc_page(gfp_t gfp,
1552 struct shmem_inode_info *info, pgoff_t index)
1554 struct vm_area_struct pvma;
1557 shmem_pseudo_vma_init(&pvma, info, index);
1558 page = alloc_page_vma(gfp, &pvma, 0);
1559 shmem_pseudo_vma_destroy(&pvma);
1564 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1565 struct inode *inode,
1566 pgoff_t index, bool huge)
1568 struct shmem_inode_info *info = SHMEM_I(inode);
1573 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1575 nr = huge ? HPAGE_PMD_NR : 1;
1577 if (!shmem_inode_acct_block(inode, nr))
1581 page = shmem_alloc_hugepage(gfp, info, index);
1583 page = shmem_alloc_page(gfp, info, index);
1585 __SetPageLocked(page);
1586 __SetPageSwapBacked(page);
1591 shmem_inode_unacct_blocks(inode, nr);
1593 return ERR_PTR(err);
1597 * When a page is moved from swapcache to shmem filecache (either by the
1598 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1599 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1600 * ignorance of the mapping it belongs to. If that mapping has special
1601 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1602 * we may need to copy to a suitable page before moving to filecache.
1604 * In a future release, this may well be extended to respect cpuset and
1605 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1606 * but for now it is a simple matter of zone.
1608 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1610 return page_zonenum(page) > gfp_zone(gfp);
1613 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1614 struct shmem_inode_info *info, pgoff_t index)
1616 struct page *oldpage, *newpage;
1617 struct address_space *swap_mapping;
1623 entry.val = page_private(oldpage);
1624 swap_index = swp_offset(entry);
1625 swap_mapping = page_mapping(oldpage);
1628 * We have arrived here because our zones are constrained, so don't
1629 * limit chance of success by further cpuset and node constraints.
1631 gfp &= ~GFP_CONSTRAINT_MASK;
1632 newpage = shmem_alloc_page(gfp, info, index);
1637 copy_highpage(newpage, oldpage);
1638 flush_dcache_page(newpage);
1640 __SetPageLocked(newpage);
1641 __SetPageSwapBacked(newpage);
1642 SetPageUptodate(newpage);
1643 set_page_private(newpage, entry.val);
1644 SetPageSwapCache(newpage);
1647 * Our caller will very soon move newpage out of swapcache, but it's
1648 * a nice clean interface for us to replace oldpage by newpage there.
1650 xa_lock_irq(&swap_mapping->i_pages);
1651 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1653 mem_cgroup_migrate(oldpage, newpage);
1654 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1655 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1657 xa_unlock_irq(&swap_mapping->i_pages);
1659 if (unlikely(error)) {
1661 * Is this possible? I think not, now that our callers check
1662 * both PageSwapCache and page_private after getting page lock;
1663 * but be defensive. Reverse old to newpage for clear and free.
1667 lru_cache_add(newpage);
1671 ClearPageSwapCache(oldpage);
1672 set_page_private(oldpage, 0);
1674 unlock_page(oldpage);
1681 * Swap in the page pointed to by *pagep.
1682 * Caller has to make sure that *pagep contains a valid swapped page.
1683 * Returns 0 and the page in pagep if success. On failure, returns the
1684 * error code and NULL in *pagep.
1686 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1687 struct page **pagep, enum sgp_type sgp,
1688 gfp_t gfp, struct vm_area_struct *vma,
1689 vm_fault_t *fault_type)
1691 struct address_space *mapping = inode->i_mapping;
1692 struct shmem_inode_info *info = SHMEM_I(inode);
1693 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1698 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1699 swap = radix_to_swp_entry(*pagep);
1702 /* Look it up and read it in.. */
1703 page = lookup_swap_cache(swap, NULL, 0);
1705 /* Or update major stats only when swapin succeeds?? */
1707 *fault_type |= VM_FAULT_MAJOR;
1708 count_vm_event(PGMAJFAULT);
1709 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1711 /* Here we actually start the io */
1712 page = shmem_swapin(swap, gfp, info, index);
1719 /* We have to do this with page locked to prevent races */
1721 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1722 !shmem_confirm_swap(mapping, index, swap)) {
1726 if (!PageUptodate(page)) {
1730 wait_on_page_writeback(page);
1733 * Some architectures may have to restore extra metadata to the
1734 * physical page after reading from swap.
1736 arch_swap_restore(swap, page);
1738 if (shmem_should_replace_page(page, gfp)) {
1739 error = shmem_replace_page(&page, gfp, info, index);
1744 error = shmem_add_to_page_cache(page, mapping, index,
1745 swp_to_radix_entry(swap), gfp,
1750 spin_lock_irq(&info->lock);
1752 shmem_recalc_inode(inode);
1753 spin_unlock_irq(&info->lock);
1755 if (sgp == SGP_WRITE)
1756 mark_page_accessed(page);
1758 delete_from_swap_cache(page);
1759 set_page_dirty(page);
1765 if (!shmem_confirm_swap(mapping, index, swap))
1777 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1779 * If we allocate a new one we do not mark it dirty. That's up to the
1780 * vm. If we swap it in we mark it dirty since we also free the swap
1781 * entry since a page cannot live in both the swap and page cache.
1783 * vma, vmf, and fault_type are only supplied by shmem_fault:
1784 * otherwise they are NULL.
1786 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1787 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1788 struct vm_area_struct *vma, struct vm_fault *vmf,
1789 vm_fault_t *fault_type)
1791 struct address_space *mapping = inode->i_mapping;
1792 struct shmem_inode_info *info = SHMEM_I(inode);
1793 struct shmem_sb_info *sbinfo;
1794 struct mm_struct *charge_mm;
1796 enum sgp_type sgp_huge = sgp;
1797 pgoff_t hindex = index;
1803 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1805 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1808 if (sgp <= SGP_CACHE &&
1809 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1813 sbinfo = SHMEM_SB(inode->i_sb);
1814 charge_mm = vma ? vma->vm_mm : NULL;
1816 page = pagecache_get_page(mapping, index,
1817 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1819 if (page && vma && userfaultfd_minor(vma)) {
1820 if (!xa_is_value(page)) {
1824 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1828 if (xa_is_value(page)) {
1829 error = shmem_swapin_page(inode, index, &page,
1830 sgp, gfp, vma, fault_type);
1831 if (error == -EEXIST)
1839 hindex = page->index;
1840 if (page && sgp == SGP_WRITE)
1841 mark_page_accessed(page);
1843 /* fallocated page? */
1844 if (page && !PageUptodate(page)) {
1845 if (sgp != SGP_READ)
1852 if (page || sgp == SGP_READ)
1856 * Fast cache lookup did not find it:
1857 * bring it back from swap or allocate.
1860 if (vma && userfaultfd_missing(vma)) {
1861 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1865 /* shmem_symlink() */
1866 if (!shmem_mapping(mapping))
1868 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1870 if (shmem_huge == SHMEM_HUGE_FORCE)
1872 switch (sbinfo->huge) {
1873 case SHMEM_HUGE_NEVER:
1875 case SHMEM_HUGE_WITHIN_SIZE: {
1879 off = round_up(index, HPAGE_PMD_NR);
1880 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1881 if (i_size >= HPAGE_PMD_SIZE &&
1882 i_size >> PAGE_SHIFT >= off)
1887 case SHMEM_HUGE_ADVISE:
1888 if (sgp_huge == SGP_HUGE)
1890 /* TODO: implement fadvise() hints */
1895 huge_gfp = vma_thp_gfp_mask(vma);
1896 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1897 page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true);
1900 page = shmem_alloc_and_acct_page(gfp, inode,
1906 error = PTR_ERR(page);
1908 if (error != -ENOSPC)
1911 * Try to reclaim some space by splitting a huge page
1912 * beyond i_size on the filesystem.
1917 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1918 if (ret == SHRINK_STOP)
1926 if (PageTransHuge(page))
1927 hindex = round_down(index, HPAGE_PMD_NR);
1931 if (sgp == SGP_WRITE)
1932 __SetPageReferenced(page);
1934 error = shmem_add_to_page_cache(page, mapping, hindex,
1935 NULL, gfp & GFP_RECLAIM_MASK,
1939 lru_cache_add(page);
1941 spin_lock_irq(&info->lock);
1942 info->alloced += compound_nr(page);
1943 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1944 shmem_recalc_inode(inode);
1945 spin_unlock_irq(&info->lock);
1948 if (PageTransHuge(page) &&
1949 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1950 hindex + HPAGE_PMD_NR - 1) {
1952 * Part of the huge page is beyond i_size: subject
1953 * to shrink under memory pressure.
1955 spin_lock(&sbinfo->shrinklist_lock);
1957 * _careful to defend against unlocked access to
1958 * ->shrink_list in shmem_unused_huge_shrink()
1960 if (list_empty_careful(&info->shrinklist)) {
1961 list_add_tail(&info->shrinklist,
1962 &sbinfo->shrinklist);
1963 sbinfo->shrinklist_len++;
1965 spin_unlock(&sbinfo->shrinklist_lock);
1969 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1971 if (sgp == SGP_FALLOC)
1975 * Let SGP_WRITE caller clear ends if write does not fill page;
1976 * but SGP_FALLOC on a page fallocated earlier must initialize
1977 * it now, lest undo on failure cancel our earlier guarantee.
1979 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1982 for (i = 0; i < compound_nr(page); i++) {
1983 clear_highpage(page + i);
1984 flush_dcache_page(page + i);
1986 SetPageUptodate(page);
1989 /* Perhaps the file has been truncated since we checked */
1990 if (sgp <= SGP_CACHE &&
1991 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1993 ClearPageDirty(page);
1994 delete_from_page_cache(page);
1995 spin_lock_irq(&info->lock);
1996 shmem_recalc_inode(inode);
1997 spin_unlock_irq(&info->lock);
2003 *pagep = page + index - hindex;
2010 shmem_inode_unacct_blocks(inode, compound_nr(page));
2012 if (PageTransHuge(page)) {
2022 if (error == -ENOSPC && !once++) {
2023 spin_lock_irq(&info->lock);
2024 shmem_recalc_inode(inode);
2025 spin_unlock_irq(&info->lock);
2028 if (error == -EEXIST)
2034 * This is like autoremove_wake_function, but it removes the wait queue
2035 * entry unconditionally - even if something else had already woken the
2038 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2040 int ret = default_wake_function(wait, mode, sync, key);
2041 list_del_init(&wait->entry);
2045 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2047 struct vm_area_struct *vma = vmf->vma;
2048 struct inode *inode = file_inode(vma->vm_file);
2049 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2052 vm_fault_t ret = VM_FAULT_LOCKED;
2055 * Trinity finds that probing a hole which tmpfs is punching can
2056 * prevent the hole-punch from ever completing: which in turn
2057 * locks writers out with its hold on i_mutex. So refrain from
2058 * faulting pages into the hole while it's being punched. Although
2059 * shmem_undo_range() does remove the additions, it may be unable to
2060 * keep up, as each new page needs its own unmap_mapping_range() call,
2061 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2063 * It does not matter if we sometimes reach this check just before the
2064 * hole-punch begins, so that one fault then races with the punch:
2065 * we just need to make racing faults a rare case.
2067 * The implementation below would be much simpler if we just used a
2068 * standard mutex or completion: but we cannot take i_mutex in fault,
2069 * and bloating every shmem inode for this unlikely case would be sad.
2071 if (unlikely(inode->i_private)) {
2072 struct shmem_falloc *shmem_falloc;
2074 spin_lock(&inode->i_lock);
2075 shmem_falloc = inode->i_private;
2077 shmem_falloc->waitq &&
2078 vmf->pgoff >= shmem_falloc->start &&
2079 vmf->pgoff < shmem_falloc->next) {
2081 wait_queue_head_t *shmem_falloc_waitq;
2082 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2084 ret = VM_FAULT_NOPAGE;
2085 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2087 ret = VM_FAULT_RETRY;
2089 shmem_falloc_waitq = shmem_falloc->waitq;
2090 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2091 TASK_UNINTERRUPTIBLE);
2092 spin_unlock(&inode->i_lock);
2096 * shmem_falloc_waitq points into the shmem_fallocate()
2097 * stack of the hole-punching task: shmem_falloc_waitq
2098 * is usually invalid by the time we reach here, but
2099 * finish_wait() does not dereference it in that case;
2100 * though i_lock needed lest racing with wake_up_all().
2102 spin_lock(&inode->i_lock);
2103 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2104 spin_unlock(&inode->i_lock);
2110 spin_unlock(&inode->i_lock);
2115 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2116 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2118 else if (vma->vm_flags & VM_HUGEPAGE)
2121 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2122 gfp, vma, vmf, &ret);
2124 return vmf_error(err);
2128 unsigned long shmem_get_unmapped_area(struct file *file,
2129 unsigned long uaddr, unsigned long len,
2130 unsigned long pgoff, unsigned long flags)
2132 unsigned long (*get_area)(struct file *,
2133 unsigned long, unsigned long, unsigned long, unsigned long);
2135 unsigned long offset;
2136 unsigned long inflated_len;
2137 unsigned long inflated_addr;
2138 unsigned long inflated_offset;
2140 if (len > TASK_SIZE)
2143 get_area = current->mm->get_unmapped_area;
2144 addr = get_area(file, uaddr, len, pgoff, flags);
2146 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2148 if (IS_ERR_VALUE(addr))
2150 if (addr & ~PAGE_MASK)
2152 if (addr > TASK_SIZE - len)
2155 if (shmem_huge == SHMEM_HUGE_DENY)
2157 if (len < HPAGE_PMD_SIZE)
2159 if (flags & MAP_FIXED)
2162 * Our priority is to support MAP_SHARED mapped hugely;
2163 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2164 * But if caller specified an address hint and we allocated area there
2165 * successfully, respect that as before.
2170 if (shmem_huge != SHMEM_HUGE_FORCE) {
2171 struct super_block *sb;
2174 VM_BUG_ON(file->f_op != &shmem_file_operations);
2175 sb = file_inode(file)->i_sb;
2178 * Called directly from mm/mmap.c, or drivers/char/mem.c
2179 * for "/dev/zero", to create a shared anonymous object.
2181 if (IS_ERR(shm_mnt))
2183 sb = shm_mnt->mnt_sb;
2185 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2189 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2190 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2192 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2195 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2196 if (inflated_len > TASK_SIZE)
2198 if (inflated_len < len)
2201 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2202 if (IS_ERR_VALUE(inflated_addr))
2204 if (inflated_addr & ~PAGE_MASK)
2207 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2208 inflated_addr += offset - inflated_offset;
2209 if (inflated_offset > offset)
2210 inflated_addr += HPAGE_PMD_SIZE;
2212 if (inflated_addr > TASK_SIZE - len)
2214 return inflated_addr;
2218 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2220 struct inode *inode = file_inode(vma->vm_file);
2221 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2224 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2227 struct inode *inode = file_inode(vma->vm_file);
2230 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2231 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2235 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2237 struct inode *inode = file_inode(file);
2238 struct shmem_inode_info *info = SHMEM_I(inode);
2239 int retval = -ENOMEM;
2242 * What serializes the accesses to info->flags?
2243 * ipc_lock_object() when called from shmctl_do_lock(),
2244 * no serialization needed when called from shm_destroy().
2246 if (lock && !(info->flags & VM_LOCKED)) {
2247 if (!user_shm_lock(inode->i_size, ucounts))
2249 info->flags |= VM_LOCKED;
2250 mapping_set_unevictable(file->f_mapping);
2252 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2253 user_shm_unlock(inode->i_size, ucounts);
2254 info->flags &= ~VM_LOCKED;
2255 mapping_clear_unevictable(file->f_mapping);
2263 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2265 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2268 ret = seal_check_future_write(info->seals, vma);
2272 /* arm64 - allow memory tagging on RAM-based files */
2273 vma->vm_flags |= VM_MTE_ALLOWED;
2275 file_accessed(file);
2276 vma->vm_ops = &shmem_vm_ops;
2277 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2278 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2279 (vma->vm_end & HPAGE_PMD_MASK)) {
2280 khugepaged_enter(vma, vma->vm_flags);
2285 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2286 umode_t mode, dev_t dev, unsigned long flags)
2288 struct inode *inode;
2289 struct shmem_inode_info *info;
2290 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2293 if (shmem_reserve_inode(sb, &ino))
2296 inode = new_inode(sb);
2299 inode_init_owner(&init_user_ns, inode, dir, mode);
2300 inode->i_blocks = 0;
2301 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2302 inode->i_generation = prandom_u32();
2303 info = SHMEM_I(inode);
2304 memset(info, 0, (char *)inode - (char *)info);
2305 spin_lock_init(&info->lock);
2306 atomic_set(&info->stop_eviction, 0);
2307 info->seals = F_SEAL_SEAL;
2308 info->flags = flags & VM_NORESERVE;
2309 INIT_LIST_HEAD(&info->shrinklist);
2310 INIT_LIST_HEAD(&info->swaplist);
2311 simple_xattrs_init(&info->xattrs);
2312 cache_no_acl(inode);
2314 switch (mode & S_IFMT) {
2316 inode->i_op = &shmem_special_inode_operations;
2317 init_special_inode(inode, mode, dev);
2320 inode->i_mapping->a_ops = &shmem_aops;
2321 inode->i_op = &shmem_inode_operations;
2322 inode->i_fop = &shmem_file_operations;
2323 mpol_shared_policy_init(&info->policy,
2324 shmem_get_sbmpol(sbinfo));
2328 /* Some things misbehave if size == 0 on a directory */
2329 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2330 inode->i_op = &shmem_dir_inode_operations;
2331 inode->i_fop = &simple_dir_operations;
2335 * Must not load anything in the rbtree,
2336 * mpol_free_shared_policy will not be called.
2338 mpol_shared_policy_init(&info->policy, NULL);
2342 lockdep_annotate_inode_mutex_key(inode);
2344 shmem_free_inode(sb);
2348 #ifdef CONFIG_USERFAULTFD
2349 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2351 struct vm_area_struct *dst_vma,
2352 unsigned long dst_addr,
2353 unsigned long src_addr,
2355 struct page **pagep)
2357 struct inode *inode = file_inode(dst_vma->vm_file);
2358 struct shmem_inode_info *info = SHMEM_I(inode);
2359 struct address_space *mapping = inode->i_mapping;
2360 gfp_t gfp = mapping_gfp_mask(mapping);
2361 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2367 if (!shmem_inode_acct_block(inode, 1)) {
2369 * We may have got a page, returned -ENOENT triggering a retry,
2370 * and now we find ourselves with -ENOMEM. Release the page, to
2371 * avoid a BUG_ON in our caller.
2373 if (unlikely(*pagep)) {
2382 page = shmem_alloc_page(gfp, info, pgoff);
2384 goto out_unacct_blocks;
2386 if (!zeropage) { /* COPY */
2387 page_kaddr = kmap_atomic(page);
2388 ret = copy_from_user(page_kaddr,
2389 (const void __user *)src_addr,
2391 kunmap_atomic(page_kaddr);
2393 /* fallback to copy_from_user outside mmap_lock */
2394 if (unlikely(ret)) {
2397 /* don't free the page */
2398 goto out_unacct_blocks;
2400 } else { /* ZEROPAGE */
2401 clear_highpage(page);
2408 VM_BUG_ON(PageLocked(page));
2409 VM_BUG_ON(PageSwapBacked(page));
2410 __SetPageLocked(page);
2411 __SetPageSwapBacked(page);
2412 __SetPageUptodate(page);
2415 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2416 if (unlikely(pgoff >= max_off))
2419 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2420 gfp & GFP_RECLAIM_MASK, dst_mm);
2424 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2427 goto out_delete_from_cache;
2429 spin_lock_irq(&info->lock);
2431 inode->i_blocks += BLOCKS_PER_PAGE;
2432 shmem_recalc_inode(inode);
2433 spin_unlock_irq(&info->lock);
2438 out_delete_from_cache:
2439 delete_from_page_cache(page);
2444 shmem_inode_unacct_blocks(inode, 1);
2447 #endif /* CONFIG_USERFAULTFD */
2450 static const struct inode_operations shmem_symlink_inode_operations;
2451 static const struct inode_operations shmem_short_symlink_operations;
2453 #ifdef CONFIG_TMPFS_XATTR
2454 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2456 #define shmem_initxattrs NULL
2460 shmem_write_begin(struct file *file, struct address_space *mapping,
2461 loff_t pos, unsigned len, unsigned flags,
2462 struct page **pagep, void **fsdata)
2464 struct inode *inode = mapping->host;
2465 struct shmem_inode_info *info = SHMEM_I(inode);
2466 pgoff_t index = pos >> PAGE_SHIFT;
2468 /* i_mutex is held by caller */
2469 if (unlikely(info->seals & (F_SEAL_GROW |
2470 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2471 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2473 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2477 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2481 shmem_write_end(struct file *file, struct address_space *mapping,
2482 loff_t pos, unsigned len, unsigned copied,
2483 struct page *page, void *fsdata)
2485 struct inode *inode = mapping->host;
2487 if (pos + copied > inode->i_size)
2488 i_size_write(inode, pos + copied);
2490 if (!PageUptodate(page)) {
2491 struct page *head = compound_head(page);
2492 if (PageTransCompound(page)) {
2495 for (i = 0; i < HPAGE_PMD_NR; i++) {
2496 if (head + i == page)
2498 clear_highpage(head + i);
2499 flush_dcache_page(head + i);
2502 if (copied < PAGE_SIZE) {
2503 unsigned from = pos & (PAGE_SIZE - 1);
2504 zero_user_segments(page, 0, from,
2505 from + copied, PAGE_SIZE);
2507 SetPageUptodate(head);
2509 set_page_dirty(page);
2516 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2518 struct file *file = iocb->ki_filp;
2519 struct inode *inode = file_inode(file);
2520 struct address_space *mapping = inode->i_mapping;
2522 unsigned long offset;
2523 enum sgp_type sgp = SGP_READ;
2526 loff_t *ppos = &iocb->ki_pos;
2529 * Might this read be for a stacking filesystem? Then when reading
2530 * holes of a sparse file, we actually need to allocate those pages,
2531 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2533 if (!iter_is_iovec(to))
2536 index = *ppos >> PAGE_SHIFT;
2537 offset = *ppos & ~PAGE_MASK;
2540 struct page *page = NULL;
2542 unsigned long nr, ret;
2543 loff_t i_size = i_size_read(inode);
2545 end_index = i_size >> PAGE_SHIFT;
2546 if (index > end_index)
2548 if (index == end_index) {
2549 nr = i_size & ~PAGE_MASK;
2554 error = shmem_getpage(inode, index, &page, sgp);
2556 if (error == -EINVAL)
2561 if (sgp == SGP_CACHE)
2562 set_page_dirty(page);
2567 * We must evaluate after, since reads (unlike writes)
2568 * are called without i_mutex protection against truncate
2571 i_size = i_size_read(inode);
2572 end_index = i_size >> PAGE_SHIFT;
2573 if (index == end_index) {
2574 nr = i_size & ~PAGE_MASK;
2585 * If users can be writing to this page using arbitrary
2586 * virtual addresses, take care about potential aliasing
2587 * before reading the page on the kernel side.
2589 if (mapping_writably_mapped(mapping))
2590 flush_dcache_page(page);
2592 * Mark the page accessed if we read the beginning.
2595 mark_page_accessed(page);
2597 page = ZERO_PAGE(0);
2602 * Ok, we have the page, and it's up-to-date, so
2603 * now we can copy it to user space...
2605 ret = copy_page_to_iter(page, offset, nr, to);
2608 index += offset >> PAGE_SHIFT;
2609 offset &= ~PAGE_MASK;
2612 if (!iov_iter_count(to))
2621 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2622 file_accessed(file);
2623 return retval ? retval : error;
2626 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2628 struct address_space *mapping = file->f_mapping;
2629 struct inode *inode = mapping->host;
2631 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2632 return generic_file_llseek_size(file, offset, whence,
2633 MAX_LFS_FILESIZE, i_size_read(inode));
2638 /* We're holding i_mutex so we can access i_size directly */
2639 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2641 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2642 inode_unlock(inode);
2646 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2649 struct inode *inode = file_inode(file);
2650 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2651 struct shmem_inode_info *info = SHMEM_I(inode);
2652 struct shmem_falloc shmem_falloc;
2653 pgoff_t start, index, end;
2656 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2661 if (mode & FALLOC_FL_PUNCH_HOLE) {
2662 struct address_space *mapping = file->f_mapping;
2663 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2664 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2665 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2667 /* protected by i_mutex */
2668 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2673 shmem_falloc.waitq = &shmem_falloc_waitq;
2674 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2675 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2676 spin_lock(&inode->i_lock);
2677 inode->i_private = &shmem_falloc;
2678 spin_unlock(&inode->i_lock);
2680 if ((u64)unmap_end > (u64)unmap_start)
2681 unmap_mapping_range(mapping, unmap_start,
2682 1 + unmap_end - unmap_start, 0);
2683 shmem_truncate_range(inode, offset, offset + len - 1);
2684 /* No need to unmap again: hole-punching leaves COWed pages */
2686 spin_lock(&inode->i_lock);
2687 inode->i_private = NULL;
2688 wake_up_all(&shmem_falloc_waitq);
2689 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2690 spin_unlock(&inode->i_lock);
2695 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2696 error = inode_newsize_ok(inode, offset + len);
2700 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2705 start = offset >> PAGE_SHIFT;
2706 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2707 /* Try to avoid a swapstorm if len is impossible to satisfy */
2708 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2713 shmem_falloc.waitq = NULL;
2714 shmem_falloc.start = start;
2715 shmem_falloc.next = start;
2716 shmem_falloc.nr_falloced = 0;
2717 shmem_falloc.nr_unswapped = 0;
2718 spin_lock(&inode->i_lock);
2719 inode->i_private = &shmem_falloc;
2720 spin_unlock(&inode->i_lock);
2722 for (index = start; index < end; index++) {
2726 * Good, the fallocate(2) manpage permits EINTR: we may have
2727 * been interrupted because we are using up too much memory.
2729 if (signal_pending(current))
2731 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2734 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2736 /* Remove the !PageUptodate pages we added */
2737 if (index > start) {
2738 shmem_undo_range(inode,
2739 (loff_t)start << PAGE_SHIFT,
2740 ((loff_t)index << PAGE_SHIFT) - 1, true);
2746 * Inform shmem_writepage() how far we have reached.
2747 * No need for lock or barrier: we have the page lock.
2749 shmem_falloc.next++;
2750 if (!PageUptodate(page))
2751 shmem_falloc.nr_falloced++;
2754 * If !PageUptodate, leave it that way so that freeable pages
2755 * can be recognized if we need to rollback on error later.
2756 * But set_page_dirty so that memory pressure will swap rather
2757 * than free the pages we are allocating (and SGP_CACHE pages
2758 * might still be clean: we now need to mark those dirty too).
2760 set_page_dirty(page);
2766 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2767 i_size_write(inode, offset + len);
2768 inode->i_ctime = current_time(inode);
2770 spin_lock(&inode->i_lock);
2771 inode->i_private = NULL;
2772 spin_unlock(&inode->i_lock);
2774 inode_unlock(inode);
2778 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2780 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2782 buf->f_type = TMPFS_MAGIC;
2783 buf->f_bsize = PAGE_SIZE;
2784 buf->f_namelen = NAME_MAX;
2785 if (sbinfo->max_blocks) {
2786 buf->f_blocks = sbinfo->max_blocks;
2788 buf->f_bfree = sbinfo->max_blocks -
2789 percpu_counter_sum(&sbinfo->used_blocks);
2791 if (sbinfo->max_inodes) {
2792 buf->f_files = sbinfo->max_inodes;
2793 buf->f_ffree = sbinfo->free_inodes;
2795 /* else leave those fields 0 like simple_statfs */
2797 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2803 * File creation. Allocate an inode, and we're done..
2806 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2807 struct dentry *dentry, umode_t mode, dev_t dev)
2809 struct inode *inode;
2810 int error = -ENOSPC;
2812 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2814 error = simple_acl_create(dir, inode);
2817 error = security_inode_init_security(inode, dir,
2819 shmem_initxattrs, NULL);
2820 if (error && error != -EOPNOTSUPP)
2824 dir->i_size += BOGO_DIRENT_SIZE;
2825 dir->i_ctime = dir->i_mtime = current_time(dir);
2826 d_instantiate(dentry, inode);
2827 dget(dentry); /* Extra count - pin the dentry in core */
2836 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2837 struct dentry *dentry, umode_t mode)
2839 struct inode *inode;
2840 int error = -ENOSPC;
2842 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2844 error = security_inode_init_security(inode, dir,
2846 shmem_initxattrs, NULL);
2847 if (error && error != -EOPNOTSUPP)
2849 error = simple_acl_create(dir, inode);
2852 d_tmpfile(dentry, inode);
2860 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2861 struct dentry *dentry, umode_t mode)
2865 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2866 mode | S_IFDIR, 0)))
2872 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2873 struct dentry *dentry, umode_t mode, bool excl)
2875 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2881 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2883 struct inode *inode = d_inode(old_dentry);
2887 * No ordinary (disk based) filesystem counts links as inodes;
2888 * but each new link needs a new dentry, pinning lowmem, and
2889 * tmpfs dentries cannot be pruned until they are unlinked.
2890 * But if an O_TMPFILE file is linked into the tmpfs, the
2891 * first link must skip that, to get the accounting right.
2893 if (inode->i_nlink) {
2894 ret = shmem_reserve_inode(inode->i_sb, NULL);
2899 dir->i_size += BOGO_DIRENT_SIZE;
2900 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2902 ihold(inode); /* New dentry reference */
2903 dget(dentry); /* Extra pinning count for the created dentry */
2904 d_instantiate(dentry, inode);
2909 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2911 struct inode *inode = d_inode(dentry);
2913 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2914 shmem_free_inode(inode->i_sb);
2916 dir->i_size -= BOGO_DIRENT_SIZE;
2917 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2919 dput(dentry); /* Undo the count from "create" - this does all the work */
2923 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2925 if (!simple_empty(dentry))
2928 drop_nlink(d_inode(dentry));
2930 return shmem_unlink(dir, dentry);
2933 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2935 bool old_is_dir = d_is_dir(old_dentry);
2936 bool new_is_dir = d_is_dir(new_dentry);
2938 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2940 drop_nlink(old_dir);
2943 drop_nlink(new_dir);
2947 old_dir->i_ctime = old_dir->i_mtime =
2948 new_dir->i_ctime = new_dir->i_mtime =
2949 d_inode(old_dentry)->i_ctime =
2950 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2955 static int shmem_whiteout(struct user_namespace *mnt_userns,
2956 struct inode *old_dir, struct dentry *old_dentry)
2958 struct dentry *whiteout;
2961 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2965 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2966 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2972 * Cheat and hash the whiteout while the old dentry is still in
2973 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2975 * d_lookup() will consistently find one of them at this point,
2976 * not sure which one, but that isn't even important.
2983 * The VFS layer already does all the dentry stuff for rename,
2984 * we just have to decrement the usage count for the target if
2985 * it exists so that the VFS layer correctly free's it when it
2988 static int shmem_rename2(struct user_namespace *mnt_userns,
2989 struct inode *old_dir, struct dentry *old_dentry,
2990 struct inode *new_dir, struct dentry *new_dentry,
2993 struct inode *inode = d_inode(old_dentry);
2994 int they_are_dirs = S_ISDIR(inode->i_mode);
2996 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2999 if (flags & RENAME_EXCHANGE)
3000 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3002 if (!simple_empty(new_dentry))
3005 if (flags & RENAME_WHITEOUT) {
3008 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3013 if (d_really_is_positive(new_dentry)) {
3014 (void) shmem_unlink(new_dir, new_dentry);
3015 if (they_are_dirs) {
3016 drop_nlink(d_inode(new_dentry));
3017 drop_nlink(old_dir);
3019 } else if (they_are_dirs) {
3020 drop_nlink(old_dir);
3024 old_dir->i_size -= BOGO_DIRENT_SIZE;
3025 new_dir->i_size += BOGO_DIRENT_SIZE;
3026 old_dir->i_ctime = old_dir->i_mtime =
3027 new_dir->i_ctime = new_dir->i_mtime =
3028 inode->i_ctime = current_time(old_dir);
3032 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3033 struct dentry *dentry, const char *symname)
3037 struct inode *inode;
3040 len = strlen(symname) + 1;
3041 if (len > PAGE_SIZE)
3042 return -ENAMETOOLONG;
3044 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3049 error = security_inode_init_security(inode, dir, &dentry->d_name,
3050 shmem_initxattrs, NULL);
3051 if (error && error != -EOPNOTSUPP) {
3056 inode->i_size = len-1;
3057 if (len <= SHORT_SYMLINK_LEN) {
3058 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3059 if (!inode->i_link) {
3063 inode->i_op = &shmem_short_symlink_operations;
3065 inode_nohighmem(inode);
3066 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3071 inode->i_mapping->a_ops = &shmem_aops;
3072 inode->i_op = &shmem_symlink_inode_operations;
3073 memcpy(page_address(page), symname, len);
3074 SetPageUptodate(page);
3075 set_page_dirty(page);
3079 dir->i_size += BOGO_DIRENT_SIZE;
3080 dir->i_ctime = dir->i_mtime = current_time(dir);
3081 d_instantiate(dentry, inode);
3086 static void shmem_put_link(void *arg)
3088 mark_page_accessed(arg);
3092 static const char *shmem_get_link(struct dentry *dentry,
3093 struct inode *inode,
3094 struct delayed_call *done)
3096 struct page *page = NULL;
3099 page = find_get_page(inode->i_mapping, 0);
3101 return ERR_PTR(-ECHILD);
3102 if (!PageUptodate(page)) {
3104 return ERR_PTR(-ECHILD);
3107 error = shmem_getpage(inode, 0, &page, SGP_READ);
3109 return ERR_PTR(error);
3112 set_delayed_call(done, shmem_put_link, page);
3113 return page_address(page);
3116 #ifdef CONFIG_TMPFS_XATTR
3118 * Superblocks without xattr inode operations may get some security.* xattr
3119 * support from the LSM "for free". As soon as we have any other xattrs
3120 * like ACLs, we also need to implement the security.* handlers at
3121 * filesystem level, though.
3125 * Callback for security_inode_init_security() for acquiring xattrs.
3127 static int shmem_initxattrs(struct inode *inode,
3128 const struct xattr *xattr_array,
3131 struct shmem_inode_info *info = SHMEM_I(inode);
3132 const struct xattr *xattr;
3133 struct simple_xattr *new_xattr;
3136 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3137 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3141 len = strlen(xattr->name) + 1;
3142 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3144 if (!new_xattr->name) {
3149 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3150 XATTR_SECURITY_PREFIX_LEN);
3151 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3154 simple_xattr_list_add(&info->xattrs, new_xattr);
3160 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3161 struct dentry *unused, struct inode *inode,
3162 const char *name, void *buffer, size_t size)
3164 struct shmem_inode_info *info = SHMEM_I(inode);
3166 name = xattr_full_name(handler, name);
3167 return simple_xattr_get(&info->xattrs, name, buffer, size);
3170 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3171 struct user_namespace *mnt_userns,
3172 struct dentry *unused, struct inode *inode,
3173 const char *name, const void *value,
3174 size_t size, int flags)
3176 struct shmem_inode_info *info = SHMEM_I(inode);
3178 name = xattr_full_name(handler, name);
3179 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3182 static const struct xattr_handler shmem_security_xattr_handler = {
3183 .prefix = XATTR_SECURITY_PREFIX,
3184 .get = shmem_xattr_handler_get,
3185 .set = shmem_xattr_handler_set,
3188 static const struct xattr_handler shmem_trusted_xattr_handler = {
3189 .prefix = XATTR_TRUSTED_PREFIX,
3190 .get = shmem_xattr_handler_get,
3191 .set = shmem_xattr_handler_set,
3194 static const struct xattr_handler *shmem_xattr_handlers[] = {
3195 #ifdef CONFIG_TMPFS_POSIX_ACL
3196 &posix_acl_access_xattr_handler,
3197 &posix_acl_default_xattr_handler,
3199 &shmem_security_xattr_handler,
3200 &shmem_trusted_xattr_handler,
3204 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3206 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3207 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3209 #endif /* CONFIG_TMPFS_XATTR */
3211 static const struct inode_operations shmem_short_symlink_operations = {
3212 .get_link = simple_get_link,
3213 #ifdef CONFIG_TMPFS_XATTR
3214 .listxattr = shmem_listxattr,
3218 static const struct inode_operations shmem_symlink_inode_operations = {
3219 .get_link = shmem_get_link,
3220 #ifdef CONFIG_TMPFS_XATTR
3221 .listxattr = shmem_listxattr,
3225 static struct dentry *shmem_get_parent(struct dentry *child)
3227 return ERR_PTR(-ESTALE);
3230 static int shmem_match(struct inode *ino, void *vfh)
3234 inum = (inum << 32) | fh[1];
3235 return ino->i_ino == inum && fh[0] == ino->i_generation;
3238 /* Find any alias of inode, but prefer a hashed alias */
3239 static struct dentry *shmem_find_alias(struct inode *inode)
3241 struct dentry *alias = d_find_alias(inode);
3243 return alias ?: d_find_any_alias(inode);
3247 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3248 struct fid *fid, int fh_len, int fh_type)
3250 struct inode *inode;
3251 struct dentry *dentry = NULL;
3258 inum = (inum << 32) | fid->raw[1];
3260 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3261 shmem_match, fid->raw);
3263 dentry = shmem_find_alias(inode);
3270 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3271 struct inode *parent)
3275 return FILEID_INVALID;
3278 if (inode_unhashed(inode)) {
3279 /* Unfortunately insert_inode_hash is not idempotent,
3280 * so as we hash inodes here rather than at creation
3281 * time, we need a lock to ensure we only try
3284 static DEFINE_SPINLOCK(lock);
3286 if (inode_unhashed(inode))
3287 __insert_inode_hash(inode,
3288 inode->i_ino + inode->i_generation);
3292 fh[0] = inode->i_generation;
3293 fh[1] = inode->i_ino;
3294 fh[2] = ((__u64)inode->i_ino) >> 32;
3300 static const struct export_operations shmem_export_ops = {
3301 .get_parent = shmem_get_parent,
3302 .encode_fh = shmem_encode_fh,
3303 .fh_to_dentry = shmem_fh_to_dentry,
3319 static const struct constant_table shmem_param_enums_huge[] = {
3320 {"never", SHMEM_HUGE_NEVER },
3321 {"always", SHMEM_HUGE_ALWAYS },
3322 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3323 {"advise", SHMEM_HUGE_ADVISE },
3327 const struct fs_parameter_spec shmem_fs_parameters[] = {
3328 fsparam_u32 ("gid", Opt_gid),
3329 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3330 fsparam_u32oct("mode", Opt_mode),
3331 fsparam_string("mpol", Opt_mpol),
3332 fsparam_string("nr_blocks", Opt_nr_blocks),
3333 fsparam_string("nr_inodes", Opt_nr_inodes),
3334 fsparam_string("size", Opt_size),
3335 fsparam_u32 ("uid", Opt_uid),
3336 fsparam_flag ("inode32", Opt_inode32),
3337 fsparam_flag ("inode64", Opt_inode64),
3341 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3343 struct shmem_options *ctx = fc->fs_private;
3344 struct fs_parse_result result;
3345 unsigned long long size;
3349 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3355 size = memparse(param->string, &rest);
3357 size <<= PAGE_SHIFT;
3358 size *= totalram_pages();
3364 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3365 ctx->seen |= SHMEM_SEEN_BLOCKS;
3368 ctx->blocks = memparse(param->string, &rest);
3371 ctx->seen |= SHMEM_SEEN_BLOCKS;
3374 ctx->inodes = memparse(param->string, &rest);
3377 ctx->seen |= SHMEM_SEEN_INODES;
3380 ctx->mode = result.uint_32 & 07777;
3383 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3384 if (!uid_valid(ctx->uid))
3388 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3389 if (!gid_valid(ctx->gid))
3393 ctx->huge = result.uint_32;
3394 if (ctx->huge != SHMEM_HUGE_NEVER &&
3395 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3396 has_transparent_hugepage()))
3397 goto unsupported_parameter;
3398 ctx->seen |= SHMEM_SEEN_HUGE;
3401 if (IS_ENABLED(CONFIG_NUMA)) {
3402 mpol_put(ctx->mpol);
3404 if (mpol_parse_str(param->string, &ctx->mpol))
3408 goto unsupported_parameter;
3410 ctx->full_inums = false;
3411 ctx->seen |= SHMEM_SEEN_INUMS;
3414 if (sizeof(ino_t) < 8) {
3416 "Cannot use inode64 with <64bit inums in kernel\n");
3418 ctx->full_inums = true;
3419 ctx->seen |= SHMEM_SEEN_INUMS;
3424 unsupported_parameter:
3425 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3427 return invalfc(fc, "Bad value for '%s'", param->key);
3430 static int shmem_parse_options(struct fs_context *fc, void *data)
3432 char *options = data;
3435 int err = security_sb_eat_lsm_opts(options, &fc->security);
3440 while (options != NULL) {
3441 char *this_char = options;
3444 * NUL-terminate this option: unfortunately,
3445 * mount options form a comma-separated list,
3446 * but mpol's nodelist may also contain commas.
3448 options = strchr(options, ',');
3449 if (options == NULL)
3452 if (!isdigit(*options)) {
3458 char *value = strchr(this_char, '=');
3464 len = strlen(value);
3466 err = vfs_parse_fs_string(fc, this_char, value, len);
3475 * Reconfigure a shmem filesystem.
3477 * Note that we disallow change from limited->unlimited blocks/inodes while any
3478 * are in use; but we must separately disallow unlimited->limited, because in
3479 * that case we have no record of how much is already in use.
3481 static int shmem_reconfigure(struct fs_context *fc)
3483 struct shmem_options *ctx = fc->fs_private;
3484 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3485 unsigned long inodes;
3486 struct mempolicy *mpol = NULL;
3489 raw_spin_lock(&sbinfo->stat_lock);
3490 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3491 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3492 if (!sbinfo->max_blocks) {
3493 err = "Cannot retroactively limit size";
3496 if (percpu_counter_compare(&sbinfo->used_blocks,
3498 err = "Too small a size for current use";
3502 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3503 if (!sbinfo->max_inodes) {
3504 err = "Cannot retroactively limit inodes";
3507 if (ctx->inodes < inodes) {
3508 err = "Too few inodes for current use";
3513 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3514 sbinfo->next_ino > UINT_MAX) {
3515 err = "Current inum too high to switch to 32-bit inums";
3519 if (ctx->seen & SHMEM_SEEN_HUGE)
3520 sbinfo->huge = ctx->huge;
3521 if (ctx->seen & SHMEM_SEEN_INUMS)
3522 sbinfo->full_inums = ctx->full_inums;
3523 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3524 sbinfo->max_blocks = ctx->blocks;
3525 if (ctx->seen & SHMEM_SEEN_INODES) {
3526 sbinfo->max_inodes = ctx->inodes;
3527 sbinfo->free_inodes = ctx->inodes - inodes;
3531 * Preserve previous mempolicy unless mpol remount option was specified.
3534 mpol = sbinfo->mpol;
3535 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3538 raw_spin_unlock(&sbinfo->stat_lock);
3542 raw_spin_unlock(&sbinfo->stat_lock);
3543 return invalfc(fc, "%s", err);
3546 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3548 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3550 if (sbinfo->max_blocks != shmem_default_max_blocks())
3551 seq_printf(seq, ",size=%luk",
3552 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3553 if (sbinfo->max_inodes != shmem_default_max_inodes())
3554 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3555 if (sbinfo->mode != (0777 | S_ISVTX))
3556 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3557 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3558 seq_printf(seq, ",uid=%u",
3559 from_kuid_munged(&init_user_ns, sbinfo->uid));
3560 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3561 seq_printf(seq, ",gid=%u",
3562 from_kgid_munged(&init_user_ns, sbinfo->gid));
3565 * Showing inode{64,32} might be useful even if it's the system default,
3566 * since then people don't have to resort to checking both here and
3567 * /proc/config.gz to confirm 64-bit inums were successfully applied
3568 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3570 * We hide it when inode64 isn't the default and we are using 32-bit
3571 * inodes, since that probably just means the feature isn't even under
3576 * +-----------------+-----------------+
3577 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3578 * +------------------+-----------------+-----------------+
3579 * | full_inums=true | show | show |
3580 * | full_inums=false | show | hide |
3581 * +------------------+-----------------+-----------------+
3584 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3585 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3586 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3587 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3589 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3591 shmem_show_mpol(seq, sbinfo->mpol);
3595 #endif /* CONFIG_TMPFS */
3597 static void shmem_put_super(struct super_block *sb)
3599 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3601 free_percpu(sbinfo->ino_batch);
3602 percpu_counter_destroy(&sbinfo->used_blocks);
3603 mpol_put(sbinfo->mpol);
3605 sb->s_fs_info = NULL;
3608 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3610 struct shmem_options *ctx = fc->fs_private;
3611 struct inode *inode;
3612 struct shmem_sb_info *sbinfo;
3614 /* Round up to L1_CACHE_BYTES to resist false sharing */
3615 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3616 L1_CACHE_BYTES), GFP_KERNEL);
3620 sb->s_fs_info = sbinfo;
3624 * Per default we only allow half of the physical ram per
3625 * tmpfs instance, limiting inodes to one per page of lowmem;
3626 * but the internal instance is left unlimited.
3628 if (!(sb->s_flags & SB_KERNMOUNT)) {
3629 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3630 ctx->blocks = shmem_default_max_blocks();
3631 if (!(ctx->seen & SHMEM_SEEN_INODES))
3632 ctx->inodes = shmem_default_max_inodes();
3633 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3634 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3636 sb->s_flags |= SB_NOUSER;
3638 sb->s_export_op = &shmem_export_ops;
3639 sb->s_flags |= SB_NOSEC;
3641 sb->s_flags |= SB_NOUSER;
3643 sbinfo->max_blocks = ctx->blocks;
3644 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3645 if (sb->s_flags & SB_KERNMOUNT) {
3646 sbinfo->ino_batch = alloc_percpu(ino_t);
3647 if (!sbinfo->ino_batch)
3650 sbinfo->uid = ctx->uid;
3651 sbinfo->gid = ctx->gid;
3652 sbinfo->full_inums = ctx->full_inums;
3653 sbinfo->mode = ctx->mode;
3654 sbinfo->huge = ctx->huge;
3655 sbinfo->mpol = ctx->mpol;
3658 raw_spin_lock_init(&sbinfo->stat_lock);
3659 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3661 spin_lock_init(&sbinfo->shrinklist_lock);
3662 INIT_LIST_HEAD(&sbinfo->shrinklist);
3664 sb->s_maxbytes = MAX_LFS_FILESIZE;
3665 sb->s_blocksize = PAGE_SIZE;
3666 sb->s_blocksize_bits = PAGE_SHIFT;
3667 sb->s_magic = TMPFS_MAGIC;
3668 sb->s_op = &shmem_ops;
3669 sb->s_time_gran = 1;
3670 #ifdef CONFIG_TMPFS_XATTR
3671 sb->s_xattr = shmem_xattr_handlers;
3673 #ifdef CONFIG_TMPFS_POSIX_ACL
3674 sb->s_flags |= SB_POSIXACL;
3676 uuid_gen(&sb->s_uuid);
3678 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3681 inode->i_uid = sbinfo->uid;
3682 inode->i_gid = sbinfo->gid;
3683 sb->s_root = d_make_root(inode);
3689 shmem_put_super(sb);
3693 static int shmem_get_tree(struct fs_context *fc)
3695 return get_tree_nodev(fc, shmem_fill_super);
3698 static void shmem_free_fc(struct fs_context *fc)
3700 struct shmem_options *ctx = fc->fs_private;
3703 mpol_put(ctx->mpol);
3708 static const struct fs_context_operations shmem_fs_context_ops = {
3709 .free = shmem_free_fc,
3710 .get_tree = shmem_get_tree,
3712 .parse_monolithic = shmem_parse_options,
3713 .parse_param = shmem_parse_one,
3714 .reconfigure = shmem_reconfigure,
3718 static struct kmem_cache *shmem_inode_cachep;
3720 static struct inode *shmem_alloc_inode(struct super_block *sb)
3722 struct shmem_inode_info *info;
3723 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3726 return &info->vfs_inode;
3729 static void shmem_free_in_core_inode(struct inode *inode)
3731 if (S_ISLNK(inode->i_mode))
3732 kfree(inode->i_link);
3733 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3736 static void shmem_destroy_inode(struct inode *inode)
3738 if (S_ISREG(inode->i_mode))
3739 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3742 static void shmem_init_inode(void *foo)
3744 struct shmem_inode_info *info = foo;
3745 inode_init_once(&info->vfs_inode);
3748 static void shmem_init_inodecache(void)
3750 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3751 sizeof(struct shmem_inode_info),
3752 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3755 static void shmem_destroy_inodecache(void)
3757 kmem_cache_destroy(shmem_inode_cachep);
3760 const struct address_space_operations shmem_aops = {
3761 .writepage = shmem_writepage,
3762 .set_page_dirty = __set_page_dirty_no_writeback,
3764 .write_begin = shmem_write_begin,
3765 .write_end = shmem_write_end,
3767 #ifdef CONFIG_MIGRATION
3768 .migratepage = migrate_page,
3770 .error_remove_page = generic_error_remove_page,
3772 EXPORT_SYMBOL(shmem_aops);
3774 static const struct file_operations shmem_file_operations = {
3776 .get_unmapped_area = shmem_get_unmapped_area,
3778 .llseek = shmem_file_llseek,
3779 .read_iter = shmem_file_read_iter,
3780 .write_iter = generic_file_write_iter,
3781 .fsync = noop_fsync,
3782 .splice_read = generic_file_splice_read,
3783 .splice_write = iter_file_splice_write,
3784 .fallocate = shmem_fallocate,
3788 static const struct inode_operations shmem_inode_operations = {
3789 .getattr = shmem_getattr,
3790 .setattr = shmem_setattr,
3791 #ifdef CONFIG_TMPFS_XATTR
3792 .listxattr = shmem_listxattr,
3793 .set_acl = simple_set_acl,
3797 static const struct inode_operations shmem_dir_inode_operations = {
3799 .create = shmem_create,
3800 .lookup = simple_lookup,
3802 .unlink = shmem_unlink,
3803 .symlink = shmem_symlink,
3804 .mkdir = shmem_mkdir,
3805 .rmdir = shmem_rmdir,
3806 .mknod = shmem_mknod,
3807 .rename = shmem_rename2,
3808 .tmpfile = shmem_tmpfile,
3810 #ifdef CONFIG_TMPFS_XATTR
3811 .listxattr = shmem_listxattr,
3813 #ifdef CONFIG_TMPFS_POSIX_ACL
3814 .setattr = shmem_setattr,
3815 .set_acl = simple_set_acl,
3819 static const struct inode_operations shmem_special_inode_operations = {
3820 #ifdef CONFIG_TMPFS_XATTR
3821 .listxattr = shmem_listxattr,
3823 #ifdef CONFIG_TMPFS_POSIX_ACL
3824 .setattr = shmem_setattr,
3825 .set_acl = simple_set_acl,
3829 static const struct super_operations shmem_ops = {
3830 .alloc_inode = shmem_alloc_inode,
3831 .free_inode = shmem_free_in_core_inode,
3832 .destroy_inode = shmem_destroy_inode,
3834 .statfs = shmem_statfs,
3835 .show_options = shmem_show_options,
3837 .evict_inode = shmem_evict_inode,
3838 .drop_inode = generic_delete_inode,
3839 .put_super = shmem_put_super,
3840 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3841 .nr_cached_objects = shmem_unused_huge_count,
3842 .free_cached_objects = shmem_unused_huge_scan,
3846 static const struct vm_operations_struct shmem_vm_ops = {
3847 .fault = shmem_fault,
3848 .map_pages = filemap_map_pages,
3850 .set_policy = shmem_set_policy,
3851 .get_policy = shmem_get_policy,
3855 int shmem_init_fs_context(struct fs_context *fc)
3857 struct shmem_options *ctx;
3859 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3863 ctx->mode = 0777 | S_ISVTX;
3864 ctx->uid = current_fsuid();
3865 ctx->gid = current_fsgid();
3867 fc->fs_private = ctx;
3868 fc->ops = &shmem_fs_context_ops;
3872 static struct file_system_type shmem_fs_type = {
3873 .owner = THIS_MODULE,
3875 .init_fs_context = shmem_init_fs_context,
3877 .parameters = shmem_fs_parameters,
3879 .kill_sb = kill_litter_super,
3880 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
3883 int __init shmem_init(void)
3887 shmem_init_inodecache();
3889 error = register_filesystem(&shmem_fs_type);
3891 pr_err("Could not register tmpfs\n");
3895 shm_mnt = kern_mount(&shmem_fs_type);
3896 if (IS_ERR(shm_mnt)) {
3897 error = PTR_ERR(shm_mnt);
3898 pr_err("Could not kern_mount tmpfs\n");
3902 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3903 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3904 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3906 shmem_huge = 0; /* just in case it was patched */
3911 unregister_filesystem(&shmem_fs_type);
3913 shmem_destroy_inodecache();
3914 shm_mnt = ERR_PTR(error);
3918 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3919 static ssize_t shmem_enabled_show(struct kobject *kobj,
3920 struct kobj_attribute *attr, char *buf)
3922 static const int values[] = {
3924 SHMEM_HUGE_WITHIN_SIZE,
3933 for (i = 0; i < ARRAY_SIZE(values); i++) {
3934 len += sysfs_emit_at(buf, len,
3935 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3937 shmem_format_huge(values[i]));
3940 len += sysfs_emit_at(buf, len, "\n");
3945 static ssize_t shmem_enabled_store(struct kobject *kobj,
3946 struct kobj_attribute *attr, const char *buf, size_t count)
3951 if (count + 1 > sizeof(tmp))
3953 memcpy(tmp, buf, count);
3955 if (count && tmp[count - 1] == '\n')
3956 tmp[count - 1] = '\0';
3958 huge = shmem_parse_huge(tmp);
3959 if (huge == -EINVAL)
3961 if (!has_transparent_hugepage() &&
3962 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3966 if (shmem_huge > SHMEM_HUGE_DENY)
3967 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3971 struct kobj_attribute shmem_enabled_attr =
3972 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3973 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3975 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3976 bool shmem_huge_enabled(struct vm_area_struct *vma)
3978 struct inode *inode = file_inode(vma->vm_file);
3979 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3983 if (!transhuge_vma_enabled(vma, vma->vm_flags))
3985 if (shmem_huge == SHMEM_HUGE_FORCE)
3987 if (shmem_huge == SHMEM_HUGE_DENY)
3989 switch (sbinfo->huge) {
3990 case SHMEM_HUGE_NEVER:
3992 case SHMEM_HUGE_ALWAYS:
3994 case SHMEM_HUGE_WITHIN_SIZE:
3995 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3996 i_size = round_up(i_size_read(inode), PAGE_SIZE);
3997 if (i_size >= HPAGE_PMD_SIZE &&
3998 i_size >> PAGE_SHIFT >= off)
4001 case SHMEM_HUGE_ADVISE:
4002 /* TODO: implement fadvise() hints */
4003 return (vma->vm_flags & VM_HUGEPAGE);
4009 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4011 #else /* !CONFIG_SHMEM */
4014 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4016 * This is intended for small system where the benefits of the full
4017 * shmem code (swap-backed and resource-limited) are outweighed by
4018 * their complexity. On systems without swap this code should be
4019 * effectively equivalent, but much lighter weight.
4022 static struct file_system_type shmem_fs_type = {
4024 .init_fs_context = ramfs_init_fs_context,
4025 .parameters = ramfs_fs_parameters,
4026 .kill_sb = kill_litter_super,
4027 .fs_flags = FS_USERNS_MOUNT,
4030 int __init shmem_init(void)
4032 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4034 shm_mnt = kern_mount(&shmem_fs_type);
4035 BUG_ON(IS_ERR(shm_mnt));
4040 int shmem_unuse(unsigned int type, bool frontswap,
4041 unsigned long *fs_pages_to_unuse)
4046 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4051 void shmem_unlock_mapping(struct address_space *mapping)
4056 unsigned long shmem_get_unmapped_area(struct file *file,
4057 unsigned long addr, unsigned long len,
4058 unsigned long pgoff, unsigned long flags)
4060 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4064 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4066 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4068 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4070 #define shmem_vm_ops generic_file_vm_ops
4071 #define shmem_file_operations ramfs_file_operations
4072 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4073 #define shmem_acct_size(flags, size) 0
4074 #define shmem_unacct_size(flags, size) do {} while (0)
4076 #endif /* CONFIG_SHMEM */
4080 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4081 unsigned long flags, unsigned int i_flags)
4083 struct inode *inode;
4087 return ERR_CAST(mnt);
4089 if (size < 0 || size > MAX_LFS_FILESIZE)
4090 return ERR_PTR(-EINVAL);
4092 if (shmem_acct_size(flags, size))
4093 return ERR_PTR(-ENOMEM);
4095 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4097 if (unlikely(!inode)) {
4098 shmem_unacct_size(flags, size);
4099 return ERR_PTR(-ENOSPC);
4101 inode->i_flags |= i_flags;
4102 inode->i_size = size;
4103 clear_nlink(inode); /* It is unlinked */
4104 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4106 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4107 &shmem_file_operations);
4114 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4115 * kernel internal. There will be NO LSM permission checks against the
4116 * underlying inode. So users of this interface must do LSM checks at a
4117 * higher layer. The users are the big_key and shm implementations. LSM
4118 * checks are provided at the key or shm level rather than the inode.
4119 * @name: name for dentry (to be seen in /proc/<pid>/maps
4120 * @size: size to be set for the file
4121 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4123 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4125 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4129 * shmem_file_setup - get an unlinked file living in tmpfs
4130 * @name: name for dentry (to be seen in /proc/<pid>/maps
4131 * @size: size to be set for the file
4132 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4134 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4136 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4138 EXPORT_SYMBOL_GPL(shmem_file_setup);
4141 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4142 * @mnt: the tmpfs mount where the file will be created
4143 * @name: name for dentry (to be seen in /proc/<pid>/maps
4144 * @size: size to be set for the file
4145 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4147 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4148 loff_t size, unsigned long flags)
4150 return __shmem_file_setup(mnt, name, size, flags, 0);
4152 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4155 * shmem_zero_setup - setup a shared anonymous mapping
4156 * @vma: the vma to be mmapped is prepared by do_mmap
4158 int shmem_zero_setup(struct vm_area_struct *vma)
4161 loff_t size = vma->vm_end - vma->vm_start;
4164 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4165 * between XFS directory reading and selinux: since this file is only
4166 * accessible to the user through its mapping, use S_PRIVATE flag to
4167 * bypass file security, in the same way as shmem_kernel_file_setup().
4169 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4171 return PTR_ERR(file);
4175 vma->vm_file = file;
4176 vma->vm_ops = &shmem_vm_ops;
4178 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4179 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4180 (vma->vm_end & HPAGE_PMD_MASK)) {
4181 khugepaged_enter(vma, vma->vm_flags);
4188 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4189 * @mapping: the page's address_space
4190 * @index: the page index
4191 * @gfp: the page allocator flags to use if allocating
4193 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4194 * with any new page allocations done using the specified allocation flags.
4195 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4196 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4197 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4199 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4200 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4202 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4203 pgoff_t index, gfp_t gfp)
4206 struct inode *inode = mapping->host;
4210 BUG_ON(!shmem_mapping(mapping));
4211 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4212 gfp, NULL, NULL, NULL);
4214 page = ERR_PTR(error);
4220 * The tiny !SHMEM case uses ramfs without swap
4222 return read_cache_page_gfp(mapping, index, gfp);
4225 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);