2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
42 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
44 static struct vfsmount *shm_mnt;
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/blkdev.h>
64 #include <linux/pagevec.h>
65 #include <linux/percpu_counter.h>
66 #include <linux/falloc.h>
67 #include <linux/splice.h>
68 #include <linux/security.h>
69 #include <linux/swapops.h>
70 #include <linux/mempolicy.h>
71 #include <linux/namei.h>
72 #include <linux/ctype.h>
73 #include <linux/migrate.h>
74 #include <linux/highmem.h>
75 #include <linux/seq_file.h>
76 #include <linux/magic.h>
77 #include <linux/syscalls.h>
78 #include <linux/fcntl.h>
79 #include <uapi/linux/memfd.h>
80 #include <linux/userfaultfd_k.h>
81 #include <linux/rmap.h>
82 #include <linux/uuid.h>
84 #include <linux/uaccess.h>
85 #include <asm/pgtable.h>
89 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
90 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
92 /* Pretend that each entry is of this size in directory's i_size */
93 #define BOGO_DIRENT_SIZE 20
95 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
96 #define SHORT_SYMLINK_LEN 128
99 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
100 * inode->i_private (with i_mutex making sure that it has only one user at
101 * a time): we would prefer not to enlarge the shmem inode just for that.
103 struct shmem_falloc {
104 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
105 pgoff_t start; /* start of range currently being fallocated */
106 pgoff_t next; /* the next page offset to be fallocated */
107 pgoff_t nr_falloced; /* how many new pages have been fallocated */
108 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
111 struct shmem_options {
112 unsigned long long blocks;
113 unsigned long long inodes;
114 struct mempolicy *mpol;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
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 bool shmem_should_replace_page(struct page *page, gfp_t gfp);
140 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
141 struct shmem_inode_info *info, pgoff_t index);
142 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
143 struct page **pagep, enum sgp_type sgp,
144 gfp_t gfp, struct vm_area_struct *vma,
145 vm_fault_t *fault_type);
146 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
147 struct page **pagep, enum sgp_type sgp,
148 gfp_t gfp, struct vm_area_struct *vma,
149 struct vm_fault *vmf, vm_fault_t *fault_type);
151 int shmem_getpage(struct inode *inode, pgoff_t index,
152 struct page **pagep, enum sgp_type sgp)
154 return shmem_getpage_gfp(inode, index, pagep, sgp,
155 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
158 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
160 return sb->s_fs_info;
164 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
165 * for shared memory and for shared anonymous (/dev/zero) mappings
166 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
167 * consistent with the pre-accounting of private mappings ...
169 static inline int shmem_acct_size(unsigned long flags, loff_t size)
171 return (flags & VM_NORESERVE) ?
172 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
175 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
177 if (!(flags & VM_NORESERVE))
178 vm_unacct_memory(VM_ACCT(size));
181 static inline int shmem_reacct_size(unsigned long flags,
182 loff_t oldsize, loff_t newsize)
184 if (!(flags & VM_NORESERVE)) {
185 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
186 return security_vm_enough_memory_mm(current->mm,
187 VM_ACCT(newsize) - VM_ACCT(oldsize));
188 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
189 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
195 * ... whereas tmpfs objects are accounted incrementally as
196 * pages are allocated, in order to allow large sparse files.
197 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
198 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
200 static inline int shmem_acct_block(unsigned long flags, long pages)
202 if (!(flags & VM_NORESERVE))
205 return security_vm_enough_memory_mm(current->mm,
206 pages * VM_ACCT(PAGE_SIZE));
209 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
211 if (flags & VM_NORESERVE)
212 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
215 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
217 struct shmem_inode_info *info = SHMEM_I(inode);
218 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
220 if (shmem_acct_block(info->flags, pages))
223 if (sbinfo->max_blocks) {
224 if (percpu_counter_compare(&sbinfo->used_blocks,
225 sbinfo->max_blocks - pages) > 0)
227 percpu_counter_add(&sbinfo->used_blocks, pages);
233 shmem_unacct_blocks(info->flags, pages);
237 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
239 struct shmem_inode_info *info = SHMEM_I(inode);
240 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
242 if (sbinfo->max_blocks)
243 percpu_counter_sub(&sbinfo->used_blocks, pages);
244 shmem_unacct_blocks(info->flags, pages);
247 static const struct super_operations shmem_ops;
248 static const struct address_space_operations shmem_aops;
249 static const struct file_operations shmem_file_operations;
250 static const struct inode_operations shmem_inode_operations;
251 static const struct inode_operations shmem_dir_inode_operations;
252 static const struct inode_operations shmem_special_inode_operations;
253 static const struct vm_operations_struct shmem_vm_ops;
254 static struct file_system_type shmem_fs_type;
256 bool vma_is_shmem(struct vm_area_struct *vma)
258 return vma->vm_ops == &shmem_vm_ops;
261 static LIST_HEAD(shmem_swaplist);
262 static DEFINE_MUTEX(shmem_swaplist_mutex);
264 static int shmem_reserve_inode(struct super_block *sb)
266 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
267 if (sbinfo->max_inodes) {
268 spin_lock(&sbinfo->stat_lock);
269 if (!sbinfo->free_inodes) {
270 spin_unlock(&sbinfo->stat_lock);
273 sbinfo->free_inodes--;
274 spin_unlock(&sbinfo->stat_lock);
279 static void shmem_free_inode(struct super_block *sb)
281 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
282 if (sbinfo->max_inodes) {
283 spin_lock(&sbinfo->stat_lock);
284 sbinfo->free_inodes++;
285 spin_unlock(&sbinfo->stat_lock);
290 * shmem_recalc_inode - recalculate the block usage of an inode
291 * @inode: inode to recalc
293 * We have to calculate the free blocks since the mm can drop
294 * undirtied hole pages behind our back.
296 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
297 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
299 * It has to be called with the spinlock held.
301 static void shmem_recalc_inode(struct inode *inode)
303 struct shmem_inode_info *info = SHMEM_I(inode);
306 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
308 info->alloced -= freed;
309 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
310 shmem_inode_unacct_blocks(inode, freed);
314 bool shmem_charge(struct inode *inode, long pages)
316 struct shmem_inode_info *info = SHMEM_I(inode);
319 if (!shmem_inode_acct_block(inode, pages))
322 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
323 inode->i_mapping->nrpages += pages;
325 spin_lock_irqsave(&info->lock, flags);
326 info->alloced += pages;
327 inode->i_blocks += pages * BLOCKS_PER_PAGE;
328 shmem_recalc_inode(inode);
329 spin_unlock_irqrestore(&info->lock, flags);
334 void shmem_uncharge(struct inode *inode, long pages)
336 struct shmem_inode_info *info = SHMEM_I(inode);
339 /* nrpages adjustment done by __delete_from_page_cache() or caller */
341 spin_lock_irqsave(&info->lock, flags);
342 info->alloced -= pages;
343 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
344 shmem_recalc_inode(inode);
345 spin_unlock_irqrestore(&info->lock, flags);
347 shmem_inode_unacct_blocks(inode, pages);
351 * Replace item expected in xarray by a new item, while holding xa_lock.
353 static int shmem_replace_entry(struct address_space *mapping,
354 pgoff_t index, void *expected, void *replacement)
356 XA_STATE(xas, &mapping->i_pages, index);
359 VM_BUG_ON(!expected);
360 VM_BUG_ON(!replacement);
361 item = xas_load(&xas);
362 if (item != expected)
364 xas_store(&xas, replacement);
369 * Sometimes, before we decide whether to proceed or to fail, we must check
370 * that an entry was not already brought back from swap by a racing thread.
372 * Checking page is not enough: by the time a SwapCache page is locked, it
373 * might be reused, and again be SwapCache, using the same swap as before.
375 static bool shmem_confirm_swap(struct address_space *mapping,
376 pgoff_t index, swp_entry_t swap)
378 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
382 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
385 * disables huge pages for the mount;
387 * enables huge pages for the mount;
388 * SHMEM_HUGE_WITHIN_SIZE:
389 * only allocate huge pages if the page will be fully within i_size,
390 * also respect fadvise()/madvise() hints;
392 * only allocate huge pages if requested with fadvise()/madvise();
395 #define SHMEM_HUGE_NEVER 0
396 #define SHMEM_HUGE_ALWAYS 1
397 #define SHMEM_HUGE_WITHIN_SIZE 2
398 #define SHMEM_HUGE_ADVISE 3
402 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
405 * disables huge on shm_mnt and all mounts, for emergency use;
407 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
410 #define SHMEM_HUGE_DENY (-1)
411 #define SHMEM_HUGE_FORCE (-2)
413 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
414 /* ifdef here to avoid bloating shmem.o when not necessary */
416 static int shmem_huge __read_mostly;
418 #if defined(CONFIG_SYSFS)
419 static int shmem_parse_huge(const char *str)
421 if (!strcmp(str, "never"))
422 return SHMEM_HUGE_NEVER;
423 if (!strcmp(str, "always"))
424 return SHMEM_HUGE_ALWAYS;
425 if (!strcmp(str, "within_size"))
426 return SHMEM_HUGE_WITHIN_SIZE;
427 if (!strcmp(str, "advise"))
428 return SHMEM_HUGE_ADVISE;
429 if (!strcmp(str, "deny"))
430 return SHMEM_HUGE_DENY;
431 if (!strcmp(str, "force"))
432 return SHMEM_HUGE_FORCE;
437 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
438 static const char *shmem_format_huge(int huge)
441 case SHMEM_HUGE_NEVER:
443 case SHMEM_HUGE_ALWAYS:
445 case SHMEM_HUGE_WITHIN_SIZE:
446 return "within_size";
447 case SHMEM_HUGE_ADVISE:
449 case SHMEM_HUGE_DENY:
451 case SHMEM_HUGE_FORCE:
460 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
461 struct shrink_control *sc, unsigned long nr_to_split)
463 LIST_HEAD(list), *pos, *next;
464 LIST_HEAD(to_remove);
466 struct shmem_inode_info *info;
468 unsigned long batch = sc ? sc->nr_to_scan : 128;
471 if (list_empty(&sbinfo->shrinklist))
474 spin_lock(&sbinfo->shrinklist_lock);
475 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
476 info = list_entry(pos, struct shmem_inode_info, shrinklist);
479 inode = igrab(&info->vfs_inode);
481 /* inode is about to be evicted */
483 list_del_init(&info->shrinklist);
487 /* Check if there's anything to gain */
488 if (round_up(inode->i_size, PAGE_SIZE) ==
489 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
490 list_move(&info->shrinklist, &to_remove);
494 list_move(&info->shrinklist, &list);
496 sbinfo->shrinklist_len--;
500 spin_unlock(&sbinfo->shrinklist_lock);
502 list_for_each_safe(pos, next, &to_remove) {
503 info = list_entry(pos, struct shmem_inode_info, shrinklist);
504 inode = &info->vfs_inode;
505 list_del_init(&info->shrinklist);
509 list_for_each_safe(pos, next, &list) {
512 info = list_entry(pos, struct shmem_inode_info, shrinklist);
513 inode = &info->vfs_inode;
515 if (nr_to_split && split >= nr_to_split)
518 page = find_get_page(inode->i_mapping,
519 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
523 /* No huge page at the end of the file: nothing to split */
524 if (!PageTransHuge(page)) {
530 * Move the inode on the list back to shrinklist if we failed
531 * to lock the page at this time.
533 * Waiting for the lock may lead to deadlock in the
536 if (!trylock_page(page)) {
541 ret = split_huge_page(page);
545 /* If split failed move the inode on the list back to shrinklist */
551 list_del_init(&info->shrinklist);
555 * Make sure the inode is either on the global list or deleted
556 * from any local list before iput() since it could be deleted
557 * in another thread once we put the inode (then the local list
560 spin_lock(&sbinfo->shrinklist_lock);
561 list_move(&info->shrinklist, &sbinfo->shrinklist);
562 sbinfo->shrinklist_len++;
563 spin_unlock(&sbinfo->shrinklist_lock);
571 static long shmem_unused_huge_scan(struct super_block *sb,
572 struct shrink_control *sc)
574 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
576 if (!READ_ONCE(sbinfo->shrinklist_len))
579 return shmem_unused_huge_shrink(sbinfo, sc, 0);
582 static long shmem_unused_huge_count(struct super_block *sb,
583 struct shrink_control *sc)
585 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
586 return READ_ONCE(sbinfo->shrinklist_len);
588 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
590 #define shmem_huge SHMEM_HUGE_DENY
592 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
593 struct shrink_control *sc, unsigned long nr_to_split)
597 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
599 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
601 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
602 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
603 shmem_huge != SHMEM_HUGE_DENY)
609 * Like add_to_page_cache_locked, but error if expected item has gone.
611 static int shmem_add_to_page_cache(struct page *page,
612 struct address_space *mapping,
613 pgoff_t index, void *expected, gfp_t gfp)
615 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
617 unsigned long nr = compound_nr(page);
619 VM_BUG_ON_PAGE(PageTail(page), page);
620 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
621 VM_BUG_ON_PAGE(!PageLocked(page), page);
622 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
623 VM_BUG_ON(expected && PageTransHuge(page));
625 page_ref_add(page, nr);
626 page->mapping = mapping;
632 entry = xas_find_conflict(&xas);
633 if (entry != expected)
634 xas_set_err(&xas, -EEXIST);
635 xas_create_range(&xas);
639 xas_store(&xas, page);
644 if (PageTransHuge(page)) {
645 count_vm_event(THP_FILE_ALLOC);
646 __inc_node_page_state(page, NR_SHMEM_THPS);
648 mapping->nrpages += nr;
649 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
650 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
652 xas_unlock_irq(&xas);
653 } while (xas_nomem(&xas, gfp));
655 if (xas_error(&xas)) {
656 page->mapping = NULL;
657 page_ref_sub(page, nr);
658 return xas_error(&xas);
665 * Like delete_from_page_cache, but substitutes swap for page.
667 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
669 struct address_space *mapping = page->mapping;
672 VM_BUG_ON_PAGE(PageCompound(page), page);
674 xa_lock_irq(&mapping->i_pages);
675 error = shmem_replace_entry(mapping, page->index, page, radswap);
676 page->mapping = NULL;
678 __dec_node_page_state(page, NR_FILE_PAGES);
679 __dec_node_page_state(page, NR_SHMEM);
680 xa_unlock_irq(&mapping->i_pages);
686 * Remove swap entry from page cache, free the swap and its page cache.
688 static int shmem_free_swap(struct address_space *mapping,
689 pgoff_t index, void *radswap)
693 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
696 free_swap_and_cache(radix_to_swp_entry(radswap));
701 * Determine (in bytes) how many of the shmem object's pages mapped by the
702 * given offsets are swapped out.
704 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
705 * as long as the inode doesn't go away and racy results are not a problem.
707 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
708 pgoff_t start, pgoff_t end)
710 XA_STATE(xas, &mapping->i_pages, start);
712 unsigned long swapped = 0;
715 xas_for_each(&xas, page, end - 1) {
716 if (xas_retry(&xas, page))
718 if (xa_is_value(page))
721 if (need_resched()) {
729 return swapped << PAGE_SHIFT;
733 * Determine (in bytes) how many of the shmem object's pages mapped by the
734 * given vma is swapped out.
736 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
737 * as long as the inode doesn't go away and racy results are not a problem.
739 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
741 struct inode *inode = file_inode(vma->vm_file);
742 struct shmem_inode_info *info = SHMEM_I(inode);
743 struct address_space *mapping = inode->i_mapping;
744 unsigned long swapped;
746 /* Be careful as we don't hold info->lock */
747 swapped = READ_ONCE(info->swapped);
750 * The easier cases are when the shmem object has nothing in swap, or
751 * the vma maps it whole. Then we can simply use the stats that we
757 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
758 return swapped << PAGE_SHIFT;
760 /* Here comes the more involved part */
761 return shmem_partial_swap_usage(mapping,
762 linear_page_index(vma, vma->vm_start),
763 linear_page_index(vma, vma->vm_end));
767 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
769 void shmem_unlock_mapping(struct address_space *mapping)
772 pgoff_t indices[PAGEVEC_SIZE];
777 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
779 while (!mapping_unevictable(mapping)) {
781 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
782 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
784 pvec.nr = find_get_entries(mapping, index,
785 PAGEVEC_SIZE, pvec.pages, indices);
788 index = indices[pvec.nr - 1] + 1;
789 pagevec_remove_exceptionals(&pvec);
790 check_move_unevictable_pages(&pvec);
791 pagevec_release(&pvec);
797 * Remove range of pages and swap entries from page cache, and free them.
798 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
800 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
803 struct address_space *mapping = inode->i_mapping;
804 struct shmem_inode_info *info = SHMEM_I(inode);
805 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
806 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
807 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
808 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
810 pgoff_t indices[PAGEVEC_SIZE];
811 long nr_swaps_freed = 0;
816 end = -1; /* unsigned, so actually very big */
820 while (index < end) {
821 pvec.nr = find_get_entries(mapping, index,
822 min(end - index, (pgoff_t)PAGEVEC_SIZE),
823 pvec.pages, indices);
826 for (i = 0; i < pagevec_count(&pvec); i++) {
827 struct page *page = pvec.pages[i];
833 if (xa_is_value(page)) {
836 nr_swaps_freed += !shmem_free_swap(mapping,
841 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
843 if (!trylock_page(page))
846 if (PageTransTail(page)) {
847 /* Middle of THP: zero out the page */
848 clear_highpage(page);
851 } else if (PageTransHuge(page)) {
852 if (index == round_down(end, HPAGE_PMD_NR)) {
854 * Range ends in the middle of THP:
857 clear_highpage(page);
861 index += HPAGE_PMD_NR - 1;
862 i += HPAGE_PMD_NR - 1;
865 if (!unfalloc || !PageUptodate(page)) {
866 VM_BUG_ON_PAGE(PageTail(page), page);
867 if (page_mapping(page) == mapping) {
868 VM_BUG_ON_PAGE(PageWriteback(page), page);
869 truncate_inode_page(mapping, page);
874 pagevec_remove_exceptionals(&pvec);
875 pagevec_release(&pvec);
881 struct page *page = NULL;
882 shmem_getpage(inode, start - 1, &page, SGP_READ);
884 unsigned int top = PAGE_SIZE;
889 zero_user_segment(page, partial_start, top);
890 set_page_dirty(page);
896 struct page *page = NULL;
897 shmem_getpage(inode, end, &page, SGP_READ);
899 zero_user_segment(page, 0, partial_end);
900 set_page_dirty(page);
909 while (index < end) {
912 pvec.nr = find_get_entries(mapping, index,
913 min(end - index, (pgoff_t)PAGEVEC_SIZE),
914 pvec.pages, indices);
916 /* If all gone or hole-punch or unfalloc, we're done */
917 if (index == start || end != -1)
919 /* But if truncating, restart to make sure all gone */
923 for (i = 0; i < pagevec_count(&pvec); i++) {
924 struct page *page = pvec.pages[i];
930 if (xa_is_value(page)) {
933 if (shmem_free_swap(mapping, index, page)) {
934 /* Swap was replaced by page: retry */
944 if (PageTransTail(page)) {
945 /* Middle of THP: zero out the page */
946 clear_highpage(page);
949 * Partial thp truncate due 'start' in middle
950 * of THP: don't need to look on these pages
951 * again on !pvec.nr restart.
953 if (index != round_down(end, HPAGE_PMD_NR))
956 } else if (PageTransHuge(page)) {
957 if (index == round_down(end, HPAGE_PMD_NR)) {
959 * Range ends in the middle of THP:
962 clear_highpage(page);
966 index += HPAGE_PMD_NR - 1;
967 i += HPAGE_PMD_NR - 1;
970 if (!unfalloc || !PageUptodate(page)) {
971 VM_BUG_ON_PAGE(PageTail(page), page);
972 if (page_mapping(page) == mapping) {
973 VM_BUG_ON_PAGE(PageWriteback(page), page);
974 truncate_inode_page(mapping, page);
976 /* Page was replaced by swap: retry */
984 pagevec_remove_exceptionals(&pvec);
985 pagevec_release(&pvec);
989 spin_lock_irq(&info->lock);
990 info->swapped -= nr_swaps_freed;
991 shmem_recalc_inode(inode);
992 spin_unlock_irq(&info->lock);
995 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
997 shmem_undo_range(inode, lstart, lend, false);
998 inode->i_ctime = inode->i_mtime = current_time(inode);
1000 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1002 static int shmem_getattr(const struct path *path, struct kstat *stat,
1003 u32 request_mask, unsigned int query_flags)
1005 struct inode *inode = path->dentry->d_inode;
1006 struct shmem_inode_info *info = SHMEM_I(inode);
1007 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1009 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1010 spin_lock_irq(&info->lock);
1011 shmem_recalc_inode(inode);
1012 spin_unlock_irq(&info->lock);
1014 generic_fillattr(inode, stat);
1016 if (is_huge_enabled(sb_info))
1017 stat->blksize = HPAGE_PMD_SIZE;
1022 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
1024 struct inode *inode = d_inode(dentry);
1025 struct shmem_inode_info *info = SHMEM_I(inode);
1026 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1029 error = setattr_prepare(dentry, attr);
1033 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1034 loff_t oldsize = inode->i_size;
1035 loff_t newsize = attr->ia_size;
1037 /* protected by i_mutex */
1038 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1039 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1042 if (newsize != oldsize) {
1043 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1047 i_size_write(inode, newsize);
1048 inode->i_ctime = inode->i_mtime = current_time(inode);
1050 if (newsize <= oldsize) {
1051 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1052 if (oldsize > holebegin)
1053 unmap_mapping_range(inode->i_mapping,
1056 shmem_truncate_range(inode,
1057 newsize, (loff_t)-1);
1058 /* unmap again to remove racily COWed private pages */
1059 if (oldsize > holebegin)
1060 unmap_mapping_range(inode->i_mapping,
1064 * Part of the huge page can be beyond i_size: subject
1065 * to shrink under memory pressure.
1067 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1068 spin_lock(&sbinfo->shrinklist_lock);
1070 * _careful to defend against unlocked access to
1071 * ->shrink_list in shmem_unused_huge_shrink()
1073 if (list_empty_careful(&info->shrinklist)) {
1074 list_add_tail(&info->shrinklist,
1075 &sbinfo->shrinklist);
1076 sbinfo->shrinklist_len++;
1078 spin_unlock(&sbinfo->shrinklist_lock);
1083 setattr_copy(inode, attr);
1084 if (attr->ia_valid & ATTR_MODE)
1085 error = posix_acl_chmod(inode, inode->i_mode);
1089 static void shmem_evict_inode(struct inode *inode)
1091 struct shmem_inode_info *info = SHMEM_I(inode);
1092 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1094 if (inode->i_mapping->a_ops == &shmem_aops) {
1095 shmem_unacct_size(info->flags, inode->i_size);
1097 shmem_truncate_range(inode, 0, (loff_t)-1);
1098 if (!list_empty(&info->shrinklist)) {
1099 spin_lock(&sbinfo->shrinklist_lock);
1100 if (!list_empty(&info->shrinklist)) {
1101 list_del_init(&info->shrinklist);
1102 sbinfo->shrinklist_len--;
1104 spin_unlock(&sbinfo->shrinklist_lock);
1106 while (!list_empty(&info->swaplist)) {
1107 /* Wait while shmem_unuse() is scanning this inode... */
1108 wait_var_event(&info->stop_eviction,
1109 !atomic_read(&info->stop_eviction));
1110 mutex_lock(&shmem_swaplist_mutex);
1111 /* ...but beware of the race if we peeked too early */
1112 if (!atomic_read(&info->stop_eviction))
1113 list_del_init(&info->swaplist);
1114 mutex_unlock(&shmem_swaplist_mutex);
1118 simple_xattrs_free(&info->xattrs);
1119 WARN_ON(inode->i_blocks);
1120 shmem_free_inode(inode->i_sb);
1124 extern struct swap_info_struct *swap_info[];
1126 static int shmem_find_swap_entries(struct address_space *mapping,
1127 pgoff_t start, unsigned int nr_entries,
1128 struct page **entries, pgoff_t *indices,
1129 unsigned int type, bool frontswap)
1131 XA_STATE(xas, &mapping->i_pages, start);
1134 unsigned int ret = 0;
1140 xas_for_each(&xas, page, ULONG_MAX) {
1141 if (xas_retry(&xas, page))
1144 if (!xa_is_value(page))
1147 entry = radix_to_swp_entry(page);
1148 if (swp_type(entry) != type)
1151 !frontswap_test(swap_info[type], swp_offset(entry)))
1154 indices[ret] = xas.xa_index;
1155 entries[ret] = page;
1157 if (need_resched()) {
1161 if (++ret == nr_entries)
1170 * Move the swapped pages for an inode to page cache. Returns the count
1171 * of pages swapped in, or the error in case of failure.
1173 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1179 struct address_space *mapping = inode->i_mapping;
1181 for (i = 0; i < pvec.nr; i++) {
1182 struct page *page = pvec.pages[i];
1184 if (!xa_is_value(page))
1186 error = shmem_swapin_page(inode, indices[i],
1188 mapping_gfp_mask(mapping),
1195 if (error == -ENOMEM)
1199 return error ? error : ret;
1203 * If swap found in inode, free it and move page from swapcache to filecache.
1205 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1206 bool frontswap, unsigned long *fs_pages_to_unuse)
1208 struct address_space *mapping = inode->i_mapping;
1210 struct pagevec pvec;
1211 pgoff_t indices[PAGEVEC_SIZE];
1212 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1215 pagevec_init(&pvec);
1217 unsigned int nr_entries = PAGEVEC_SIZE;
1219 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1220 nr_entries = *fs_pages_to_unuse;
1222 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1223 pvec.pages, indices,
1230 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1234 if (frontswap_partial) {
1235 *fs_pages_to_unuse -= ret;
1236 if (*fs_pages_to_unuse == 0) {
1237 ret = FRONTSWAP_PAGES_UNUSED;
1242 start = indices[pvec.nr - 1];
1249 * Read all the shared memory data that resides in the swap
1250 * device 'type' back into memory, so the swap device can be
1253 int shmem_unuse(unsigned int type, bool frontswap,
1254 unsigned long *fs_pages_to_unuse)
1256 struct shmem_inode_info *info, *next;
1259 if (list_empty(&shmem_swaplist))
1262 mutex_lock(&shmem_swaplist_mutex);
1263 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1264 if (!info->swapped) {
1265 list_del_init(&info->swaplist);
1269 * Drop the swaplist mutex while searching the inode for swap;
1270 * but before doing so, make sure shmem_evict_inode() will not
1271 * remove placeholder inode from swaplist, nor let it be freed
1272 * (igrab() would protect from unlink, but not from unmount).
1274 atomic_inc(&info->stop_eviction);
1275 mutex_unlock(&shmem_swaplist_mutex);
1277 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1281 mutex_lock(&shmem_swaplist_mutex);
1282 next = list_next_entry(info, swaplist);
1284 list_del_init(&info->swaplist);
1285 if (atomic_dec_and_test(&info->stop_eviction))
1286 wake_up_var(&info->stop_eviction);
1290 mutex_unlock(&shmem_swaplist_mutex);
1296 * Move the page from the page cache to the swap cache.
1298 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1300 struct shmem_inode_info *info;
1301 struct address_space *mapping;
1302 struct inode *inode;
1306 VM_BUG_ON_PAGE(PageCompound(page), page);
1307 BUG_ON(!PageLocked(page));
1308 mapping = page->mapping;
1309 index = page->index;
1310 inode = mapping->host;
1311 info = SHMEM_I(inode);
1312 if (info->flags & VM_LOCKED)
1314 if (!total_swap_pages)
1318 * Our capabilities prevent regular writeback or sync from ever calling
1319 * shmem_writepage; but a stacking filesystem might use ->writepage of
1320 * its underlying filesystem, in which case tmpfs should write out to
1321 * swap only in response to memory pressure, and not for the writeback
1324 if (!wbc->for_reclaim) {
1325 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1330 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1331 * value into swapfile.c, the only way we can correctly account for a
1332 * fallocated page arriving here is now to initialize it and write it.
1334 * That's okay for a page already fallocated earlier, but if we have
1335 * not yet completed the fallocation, then (a) we want to keep track
1336 * of this page in case we have to undo it, and (b) it may not be a
1337 * good idea to continue anyway, once we're pushing into swap. So
1338 * reactivate the page, and let shmem_fallocate() quit when too many.
1340 if (!PageUptodate(page)) {
1341 if (inode->i_private) {
1342 struct shmem_falloc *shmem_falloc;
1343 spin_lock(&inode->i_lock);
1344 shmem_falloc = inode->i_private;
1346 !shmem_falloc->waitq &&
1347 index >= shmem_falloc->start &&
1348 index < shmem_falloc->next)
1349 shmem_falloc->nr_unswapped++;
1351 shmem_falloc = NULL;
1352 spin_unlock(&inode->i_lock);
1356 clear_highpage(page);
1357 flush_dcache_page(page);
1358 SetPageUptodate(page);
1361 swap = get_swap_page(page);
1366 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1367 * if it's not already there. Do it now before the page is
1368 * moved to swap cache, when its pagelock no longer protects
1369 * the inode from eviction. But don't unlock the mutex until
1370 * we've incremented swapped, because shmem_unuse_inode() will
1371 * prune a !swapped inode from the swaplist under this mutex.
1373 mutex_lock(&shmem_swaplist_mutex);
1374 if (list_empty(&info->swaplist))
1375 list_add(&info->swaplist, &shmem_swaplist);
1377 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1378 spin_lock_irq(&info->lock);
1379 shmem_recalc_inode(inode);
1381 spin_unlock_irq(&info->lock);
1383 swap_shmem_alloc(swap);
1384 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1386 mutex_unlock(&shmem_swaplist_mutex);
1387 BUG_ON(page_mapped(page));
1388 swap_writepage(page, wbc);
1392 mutex_unlock(&shmem_swaplist_mutex);
1393 put_swap_page(page, swap);
1395 set_page_dirty(page);
1396 if (wbc->for_reclaim)
1397 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1402 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1403 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1407 if (!mpol || mpol->mode == MPOL_DEFAULT)
1408 return; /* show nothing */
1410 mpol_to_str(buffer, sizeof(buffer), mpol);
1412 seq_printf(seq, ",mpol=%s", buffer);
1415 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1417 struct mempolicy *mpol = NULL;
1419 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1420 mpol = sbinfo->mpol;
1422 spin_unlock(&sbinfo->stat_lock);
1426 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1427 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1430 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1434 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1436 #define vm_policy vm_private_data
1439 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1440 struct shmem_inode_info *info, pgoff_t index)
1442 /* Create a pseudo vma that just contains the policy */
1443 vma_init(vma, NULL);
1444 /* Bias interleave by inode number to distribute better across nodes */
1445 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1446 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1449 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1451 /* Drop reference taken by mpol_shared_policy_lookup() */
1452 mpol_cond_put(vma->vm_policy);
1455 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1456 struct shmem_inode_info *info, pgoff_t index)
1458 struct vm_area_struct pvma;
1460 struct vm_fault vmf;
1462 shmem_pseudo_vma_init(&pvma, info, index);
1465 page = swap_cluster_readahead(swap, gfp, &vmf);
1466 shmem_pseudo_vma_destroy(&pvma);
1471 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1472 struct shmem_inode_info *info, pgoff_t index)
1474 struct vm_area_struct pvma;
1475 struct address_space *mapping = info->vfs_inode.i_mapping;
1479 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1482 hindex = round_down(index, HPAGE_PMD_NR);
1483 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1487 shmem_pseudo_vma_init(&pvma, info, hindex);
1488 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1489 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1490 shmem_pseudo_vma_destroy(&pvma);
1492 prep_transhuge_page(page);
1496 static struct page *shmem_alloc_page(gfp_t gfp,
1497 struct shmem_inode_info *info, pgoff_t index)
1499 struct vm_area_struct pvma;
1502 shmem_pseudo_vma_init(&pvma, info, index);
1503 page = alloc_page_vma(gfp, &pvma, 0);
1504 shmem_pseudo_vma_destroy(&pvma);
1509 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1510 struct inode *inode,
1511 pgoff_t index, bool huge)
1513 struct shmem_inode_info *info = SHMEM_I(inode);
1518 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1520 nr = huge ? HPAGE_PMD_NR : 1;
1522 if (!shmem_inode_acct_block(inode, nr))
1526 page = shmem_alloc_hugepage(gfp, info, index);
1528 page = shmem_alloc_page(gfp, info, index);
1530 __SetPageLocked(page);
1531 __SetPageSwapBacked(page);
1536 shmem_inode_unacct_blocks(inode, nr);
1538 return ERR_PTR(err);
1542 * When a page is moved from swapcache to shmem filecache (either by the
1543 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1544 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1545 * ignorance of the mapping it belongs to. If that mapping has special
1546 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1547 * we may need to copy to a suitable page before moving to filecache.
1549 * In a future release, this may well be extended to respect cpuset and
1550 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1551 * but for now it is a simple matter of zone.
1553 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1555 return page_zonenum(page) > gfp_zone(gfp);
1558 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1559 struct shmem_inode_info *info, pgoff_t index)
1561 struct page *oldpage, *newpage;
1562 struct address_space *swap_mapping;
1568 entry.val = page_private(oldpage);
1569 swap_index = swp_offset(entry);
1570 swap_mapping = page_mapping(oldpage);
1573 * We have arrived here because our zones are constrained, so don't
1574 * limit chance of success by further cpuset and node constraints.
1576 gfp &= ~GFP_CONSTRAINT_MASK;
1577 newpage = shmem_alloc_page(gfp, info, index);
1582 copy_highpage(newpage, oldpage);
1583 flush_dcache_page(newpage);
1585 __SetPageLocked(newpage);
1586 __SetPageSwapBacked(newpage);
1587 SetPageUptodate(newpage);
1588 set_page_private(newpage, entry.val);
1589 SetPageSwapCache(newpage);
1592 * Our caller will very soon move newpage out of swapcache, but it's
1593 * a nice clean interface for us to replace oldpage by newpage there.
1595 xa_lock_irq(&swap_mapping->i_pages);
1596 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1598 __inc_node_page_state(newpage, NR_FILE_PAGES);
1599 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1601 xa_unlock_irq(&swap_mapping->i_pages);
1603 if (unlikely(error)) {
1605 * Is this possible? I think not, now that our callers check
1606 * both PageSwapCache and page_private after getting page lock;
1607 * but be defensive. Reverse old to newpage for clear and free.
1611 mem_cgroup_migrate(oldpage, newpage);
1612 lru_cache_add_anon(newpage);
1616 ClearPageSwapCache(oldpage);
1617 set_page_private(oldpage, 0);
1619 unlock_page(oldpage);
1626 * Swap in the page pointed to by *pagep.
1627 * Caller has to make sure that *pagep contains a valid swapped page.
1628 * Returns 0 and the page in pagep if success. On failure, returns the
1629 * the error code and NULL in *pagep.
1631 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1632 struct page **pagep, enum sgp_type sgp,
1633 gfp_t gfp, struct vm_area_struct *vma,
1634 vm_fault_t *fault_type)
1636 struct address_space *mapping = inode->i_mapping;
1637 struct shmem_inode_info *info = SHMEM_I(inode);
1638 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
1639 struct mem_cgroup *memcg;
1644 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1645 swap = radix_to_swp_entry(*pagep);
1648 /* Look it up and read it in.. */
1649 page = lookup_swap_cache(swap, NULL, 0);
1651 /* Or update major stats only when swapin succeeds?? */
1653 *fault_type |= VM_FAULT_MAJOR;
1654 count_vm_event(PGMAJFAULT);
1655 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1657 /* Here we actually start the io */
1658 page = shmem_swapin(swap, gfp, info, index);
1665 /* We have to do this with page locked to prevent races */
1667 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1668 !shmem_confirm_swap(mapping, index, swap)) {
1672 if (!PageUptodate(page)) {
1676 wait_on_page_writeback(page);
1678 if (shmem_should_replace_page(page, gfp)) {
1679 error = shmem_replace_page(&page, gfp, info, index);
1684 error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1687 error = shmem_add_to_page_cache(page, mapping, index,
1688 swp_to_radix_entry(swap), gfp);
1690 * We already confirmed swap under page lock, and make
1691 * no memory allocation here, so usually no possibility
1692 * of error; but free_swap_and_cache() only trylocks a
1693 * page, so it is just possible that the entry has been
1694 * truncated or holepunched since swap was confirmed.
1695 * shmem_undo_range() will have done some of the
1696 * unaccounting, now delete_from_swap_cache() will do
1700 mem_cgroup_cancel_charge(page, memcg, false);
1701 delete_from_swap_cache(page);
1707 mem_cgroup_commit_charge(page, memcg, true, false);
1709 spin_lock_irq(&info->lock);
1711 shmem_recalc_inode(inode);
1712 spin_unlock_irq(&info->lock);
1714 if (sgp == SGP_WRITE)
1715 mark_page_accessed(page);
1717 delete_from_swap_cache(page);
1718 set_page_dirty(page);
1724 if (!shmem_confirm_swap(mapping, index, swap))
1736 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1738 * If we allocate a new one we do not mark it dirty. That's up to the
1739 * vm. If we swap it in we mark it dirty since we also free the swap
1740 * entry since a page cannot live in both the swap and page cache.
1742 * vmf and fault_type are only supplied by shmem_fault:
1743 * otherwise they are NULL.
1745 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1746 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1747 struct vm_area_struct *vma, struct vm_fault *vmf,
1748 vm_fault_t *fault_type)
1750 struct address_space *mapping = inode->i_mapping;
1751 struct shmem_inode_info *info = SHMEM_I(inode);
1752 struct shmem_sb_info *sbinfo;
1753 struct mm_struct *charge_mm;
1754 struct mem_cgroup *memcg;
1756 enum sgp_type sgp_huge = sgp;
1757 pgoff_t hindex = index;
1762 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1764 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1767 if (sgp <= SGP_CACHE &&
1768 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1772 sbinfo = SHMEM_SB(inode->i_sb);
1773 charge_mm = vma ? vma->vm_mm : current->mm;
1775 page = find_lock_entry(mapping, index);
1776 if (xa_is_value(page)) {
1777 error = shmem_swapin_page(inode, index, &page,
1778 sgp, gfp, vma, fault_type);
1779 if (error == -EEXIST)
1786 if (page && sgp == SGP_WRITE)
1787 mark_page_accessed(page);
1789 /* fallocated page? */
1790 if (page && !PageUptodate(page)) {
1791 if (sgp != SGP_READ)
1797 if (page || sgp == SGP_READ) {
1803 * Fast cache lookup did not find it:
1804 * bring it back from swap or allocate.
1807 if (vma && userfaultfd_missing(vma)) {
1808 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1812 /* shmem_symlink() */
1813 if (mapping->a_ops != &shmem_aops)
1815 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1817 if (shmem_huge == SHMEM_HUGE_FORCE)
1819 switch (sbinfo->huge) {
1822 case SHMEM_HUGE_NEVER:
1824 case SHMEM_HUGE_WITHIN_SIZE:
1825 off = round_up(index, HPAGE_PMD_NR);
1826 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1827 if (i_size >= HPAGE_PMD_SIZE &&
1828 i_size >> PAGE_SHIFT >= off)
1831 case SHMEM_HUGE_ADVISE:
1832 if (sgp_huge == SGP_HUGE)
1834 /* TODO: implement fadvise() hints */
1839 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1842 page = shmem_alloc_and_acct_page(gfp, inode,
1848 error = PTR_ERR(page);
1850 if (error != -ENOSPC)
1853 * Try to reclaim some space by splitting a huge page
1854 * beyond i_size on the filesystem.
1859 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1860 if (ret == SHRINK_STOP)
1868 if (PageTransHuge(page))
1869 hindex = round_down(index, HPAGE_PMD_NR);
1873 if (sgp == SGP_WRITE)
1874 __SetPageReferenced(page);
1876 error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1877 PageTransHuge(page));
1880 error = shmem_add_to_page_cache(page, mapping, hindex,
1881 NULL, gfp & GFP_RECLAIM_MASK);
1883 mem_cgroup_cancel_charge(page, memcg,
1884 PageTransHuge(page));
1887 mem_cgroup_commit_charge(page, memcg, false,
1888 PageTransHuge(page));
1889 lru_cache_add_anon(page);
1891 spin_lock_irq(&info->lock);
1892 info->alloced += compound_nr(page);
1893 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1894 shmem_recalc_inode(inode);
1895 spin_unlock_irq(&info->lock);
1898 if (PageTransHuge(page) &&
1899 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1900 hindex + HPAGE_PMD_NR - 1) {
1902 * Part of the huge page is beyond i_size: subject
1903 * to shrink under memory pressure.
1905 spin_lock(&sbinfo->shrinklist_lock);
1907 * _careful to defend against unlocked access to
1908 * ->shrink_list in shmem_unused_huge_shrink()
1910 if (list_empty_careful(&info->shrinklist)) {
1911 list_add_tail(&info->shrinklist,
1912 &sbinfo->shrinklist);
1913 sbinfo->shrinklist_len++;
1915 spin_unlock(&sbinfo->shrinklist_lock);
1919 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1921 if (sgp == SGP_FALLOC)
1925 * Let SGP_WRITE caller clear ends if write does not fill page;
1926 * but SGP_FALLOC on a page fallocated earlier must initialize
1927 * it now, lest undo on failure cancel our earlier guarantee.
1929 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1930 struct page *head = compound_head(page);
1933 for (i = 0; i < compound_nr(head); i++) {
1934 clear_highpage(head + i);
1935 flush_dcache_page(head + i);
1937 SetPageUptodate(head);
1940 /* Perhaps the file has been truncated since we checked */
1941 if (sgp <= SGP_CACHE &&
1942 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1944 ClearPageDirty(page);
1945 delete_from_page_cache(page);
1946 spin_lock_irq(&info->lock);
1947 shmem_recalc_inode(inode);
1948 spin_unlock_irq(&info->lock);
1953 *pagep = page + index - hindex;
1960 shmem_inode_unacct_blocks(inode, compound_nr(page));
1962 if (PageTransHuge(page)) {
1972 if (error == -ENOSPC && !once++) {
1973 spin_lock_irq(&info->lock);
1974 shmem_recalc_inode(inode);
1975 spin_unlock_irq(&info->lock);
1978 if (error == -EEXIST)
1984 * This is like autoremove_wake_function, but it removes the wait queue
1985 * entry unconditionally - even if something else had already woken the
1988 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1990 int ret = default_wake_function(wait, mode, sync, key);
1991 list_del_init(&wait->entry);
1995 static vm_fault_t shmem_fault(struct vm_fault *vmf)
1997 struct vm_area_struct *vma = vmf->vma;
1998 struct inode *inode = file_inode(vma->vm_file);
1999 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2002 vm_fault_t ret = VM_FAULT_LOCKED;
2005 * Trinity finds that probing a hole which tmpfs is punching can
2006 * prevent the hole-punch from ever completing: which in turn
2007 * locks writers out with its hold on i_mutex. So refrain from
2008 * faulting pages into the hole while it's being punched. Although
2009 * shmem_undo_range() does remove the additions, it may be unable to
2010 * keep up, as each new page needs its own unmap_mapping_range() call,
2011 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2013 * It does not matter if we sometimes reach this check just before the
2014 * hole-punch begins, so that one fault then races with the punch:
2015 * we just need to make racing faults a rare case.
2017 * The implementation below would be much simpler if we just used a
2018 * standard mutex or completion: but we cannot take i_mutex in fault,
2019 * and bloating every shmem inode for this unlikely case would be sad.
2021 if (unlikely(inode->i_private)) {
2022 struct shmem_falloc *shmem_falloc;
2024 spin_lock(&inode->i_lock);
2025 shmem_falloc = inode->i_private;
2027 shmem_falloc->waitq &&
2028 vmf->pgoff >= shmem_falloc->start &&
2029 vmf->pgoff < shmem_falloc->next) {
2031 wait_queue_head_t *shmem_falloc_waitq;
2032 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2034 ret = VM_FAULT_NOPAGE;
2035 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2037 ret = VM_FAULT_RETRY;
2039 shmem_falloc_waitq = shmem_falloc->waitq;
2040 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2041 TASK_UNINTERRUPTIBLE);
2042 spin_unlock(&inode->i_lock);
2046 * shmem_falloc_waitq points into the shmem_fallocate()
2047 * stack of the hole-punching task: shmem_falloc_waitq
2048 * is usually invalid by the time we reach here, but
2049 * finish_wait() does not dereference it in that case;
2050 * though i_lock needed lest racing with wake_up_all().
2052 spin_lock(&inode->i_lock);
2053 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2054 spin_unlock(&inode->i_lock);
2060 spin_unlock(&inode->i_lock);
2065 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2066 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2068 else if (vma->vm_flags & VM_HUGEPAGE)
2071 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2072 gfp, vma, vmf, &ret);
2074 return vmf_error(err);
2078 unsigned long shmem_get_unmapped_area(struct file *file,
2079 unsigned long uaddr, unsigned long len,
2080 unsigned long pgoff, unsigned long flags)
2082 unsigned long (*get_area)(struct file *,
2083 unsigned long, unsigned long, unsigned long, unsigned long);
2085 unsigned long offset;
2086 unsigned long inflated_len;
2087 unsigned long inflated_addr;
2088 unsigned long inflated_offset;
2090 if (len > TASK_SIZE)
2093 get_area = current->mm->get_unmapped_area;
2094 addr = get_area(file, uaddr, len, pgoff, flags);
2096 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2098 if (IS_ERR_VALUE(addr))
2100 if (addr & ~PAGE_MASK)
2102 if (addr > TASK_SIZE - len)
2105 if (shmem_huge == SHMEM_HUGE_DENY)
2107 if (len < HPAGE_PMD_SIZE)
2109 if (flags & MAP_FIXED)
2112 * Our priority is to support MAP_SHARED mapped hugely;
2113 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2114 * But if caller specified an address hint and we allocated area there
2115 * successfully, respect that as before.
2120 if (shmem_huge != SHMEM_HUGE_FORCE) {
2121 struct super_block *sb;
2124 VM_BUG_ON(file->f_op != &shmem_file_operations);
2125 sb = file_inode(file)->i_sb;
2128 * Called directly from mm/mmap.c, or drivers/char/mem.c
2129 * for "/dev/zero", to create a shared anonymous object.
2131 if (IS_ERR(shm_mnt))
2133 sb = shm_mnt->mnt_sb;
2135 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2139 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2140 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2142 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2145 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2146 if (inflated_len > TASK_SIZE)
2148 if (inflated_len < len)
2151 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2152 if (IS_ERR_VALUE(inflated_addr))
2154 if (inflated_addr & ~PAGE_MASK)
2157 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2158 inflated_addr += offset - inflated_offset;
2159 if (inflated_offset > offset)
2160 inflated_addr += HPAGE_PMD_SIZE;
2162 if (inflated_addr > TASK_SIZE - len)
2164 return inflated_addr;
2168 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2170 struct inode *inode = file_inode(vma->vm_file);
2171 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2174 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2177 struct inode *inode = file_inode(vma->vm_file);
2180 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2181 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2185 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2187 struct inode *inode = file_inode(file);
2188 struct shmem_inode_info *info = SHMEM_I(inode);
2189 int retval = -ENOMEM;
2192 * What serializes the accesses to info->flags?
2193 * ipc_lock_object() when called from shmctl_do_lock(),
2194 * no serialization needed when called from shm_destroy().
2196 if (lock && !(info->flags & VM_LOCKED)) {
2197 if (!user_shm_lock(inode->i_size, user))
2199 info->flags |= VM_LOCKED;
2200 mapping_set_unevictable(file->f_mapping);
2202 if (!lock && (info->flags & VM_LOCKED) && user) {
2203 user_shm_unlock(inode->i_size, user);
2204 info->flags &= ~VM_LOCKED;
2205 mapping_clear_unevictable(file->f_mapping);
2213 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2215 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2218 ret = seal_check_future_write(info->seals, vma);
2222 file_accessed(file);
2223 vma->vm_ops = &shmem_vm_ops;
2224 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2225 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2226 (vma->vm_end & HPAGE_PMD_MASK)) {
2227 khugepaged_enter(vma, vma->vm_flags);
2232 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2233 umode_t mode, dev_t dev, unsigned long flags)
2235 struct inode *inode;
2236 struct shmem_inode_info *info;
2237 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2239 if (shmem_reserve_inode(sb))
2242 inode = new_inode(sb);
2244 inode->i_ino = get_next_ino();
2245 inode_init_owner(inode, dir, mode);
2246 inode->i_blocks = 0;
2247 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2248 inode->i_generation = prandom_u32();
2249 info = SHMEM_I(inode);
2250 memset(info, 0, (char *)inode - (char *)info);
2251 spin_lock_init(&info->lock);
2252 atomic_set(&info->stop_eviction, 0);
2253 info->seals = F_SEAL_SEAL;
2254 info->flags = flags & VM_NORESERVE;
2255 INIT_LIST_HEAD(&info->shrinklist);
2256 INIT_LIST_HEAD(&info->swaplist);
2257 simple_xattrs_init(&info->xattrs);
2258 cache_no_acl(inode);
2260 switch (mode & S_IFMT) {
2262 inode->i_op = &shmem_special_inode_operations;
2263 init_special_inode(inode, mode, dev);
2266 inode->i_mapping->a_ops = &shmem_aops;
2267 inode->i_op = &shmem_inode_operations;
2268 inode->i_fop = &shmem_file_operations;
2269 mpol_shared_policy_init(&info->policy,
2270 shmem_get_sbmpol(sbinfo));
2274 /* Some things misbehave if size == 0 on a directory */
2275 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2276 inode->i_op = &shmem_dir_inode_operations;
2277 inode->i_fop = &simple_dir_operations;
2281 * Must not load anything in the rbtree,
2282 * mpol_free_shared_policy will not be called.
2284 mpol_shared_policy_init(&info->policy, NULL);
2288 lockdep_annotate_inode_mutex_key(inode);
2290 shmem_free_inode(sb);
2294 bool shmem_mapping(struct address_space *mapping)
2296 return mapping->a_ops == &shmem_aops;
2299 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2301 struct vm_area_struct *dst_vma,
2302 unsigned long dst_addr,
2303 unsigned long src_addr,
2305 struct page **pagep)
2307 struct inode *inode = file_inode(dst_vma->vm_file);
2308 struct shmem_inode_info *info = SHMEM_I(inode);
2309 struct address_space *mapping = inode->i_mapping;
2310 gfp_t gfp = mapping_gfp_mask(mapping);
2311 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2312 struct mem_cgroup *memcg;
2316 pte_t _dst_pte, *dst_pte;
2318 pgoff_t offset, max_off;
2321 if (!shmem_inode_acct_block(inode, 1)) {
2323 * We may have got a page, returned -ENOENT triggering a retry,
2324 * and now we find ourselves with -ENOMEM. Release the page, to
2325 * avoid a BUG_ON in our caller.
2327 if (unlikely(*pagep)) {
2335 page = shmem_alloc_page(gfp, info, pgoff);
2337 goto out_unacct_blocks;
2339 if (!zeropage) { /* mcopy_atomic */
2340 page_kaddr = kmap_atomic(page);
2341 ret = copy_from_user(page_kaddr,
2342 (const void __user *)src_addr,
2344 kunmap_atomic(page_kaddr);
2346 /* fallback to copy_from_user outside mmap_sem */
2347 if (unlikely(ret)) {
2349 shmem_inode_unacct_blocks(inode, 1);
2350 /* don't free the page */
2353 } else { /* mfill_zeropage_atomic */
2354 clear_highpage(page);
2361 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2362 __SetPageLocked(page);
2363 __SetPageSwapBacked(page);
2364 __SetPageUptodate(page);
2367 offset = linear_page_index(dst_vma, dst_addr);
2368 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2369 if (unlikely(offset >= max_off))
2372 ret = mem_cgroup_try_charge_delay(page, dst_mm, gfp, &memcg, false);
2376 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2377 gfp & GFP_RECLAIM_MASK);
2379 goto out_release_uncharge;
2381 mem_cgroup_commit_charge(page, memcg, false, false);
2383 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2384 if (dst_vma->vm_flags & VM_WRITE)
2385 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2388 * We don't set the pte dirty if the vma has no
2389 * VM_WRITE permission, so mark the page dirty or it
2390 * could be freed from under us. We could do it
2391 * unconditionally before unlock_page(), but doing it
2392 * only if VM_WRITE is not set is faster.
2394 set_page_dirty(page);
2397 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2400 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2401 if (unlikely(offset >= max_off))
2402 goto out_release_uncharge_unlock;
2405 if (!pte_none(*dst_pte))
2406 goto out_release_uncharge_unlock;
2408 lru_cache_add_anon(page);
2410 spin_lock_irq(&info->lock);
2412 inode->i_blocks += BLOCKS_PER_PAGE;
2413 shmem_recalc_inode(inode);
2414 spin_unlock_irq(&info->lock);
2416 inc_mm_counter(dst_mm, mm_counter_file(page));
2417 page_add_file_rmap(page, false);
2418 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2420 /* No need to invalidate - it was non-present before */
2421 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2422 pte_unmap_unlock(dst_pte, ptl);
2427 out_release_uncharge_unlock:
2428 pte_unmap_unlock(dst_pte, ptl);
2429 ClearPageDirty(page);
2430 delete_from_page_cache(page);
2431 out_release_uncharge:
2432 mem_cgroup_cancel_charge(page, memcg, false);
2437 shmem_inode_unacct_blocks(inode, 1);
2441 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2443 struct vm_area_struct *dst_vma,
2444 unsigned long dst_addr,
2445 unsigned long src_addr,
2446 struct page **pagep)
2448 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2449 dst_addr, src_addr, false, pagep);
2452 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2454 struct vm_area_struct *dst_vma,
2455 unsigned long dst_addr)
2457 struct page *page = NULL;
2459 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2460 dst_addr, 0, true, &page);
2464 static const struct inode_operations shmem_symlink_inode_operations;
2465 static const struct inode_operations shmem_short_symlink_operations;
2467 #ifdef CONFIG_TMPFS_XATTR
2468 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2470 #define shmem_initxattrs NULL
2474 shmem_write_begin(struct file *file, struct address_space *mapping,
2475 loff_t pos, unsigned len, unsigned flags,
2476 struct page **pagep, void **fsdata)
2478 struct inode *inode = mapping->host;
2479 struct shmem_inode_info *info = SHMEM_I(inode);
2480 pgoff_t index = pos >> PAGE_SHIFT;
2482 /* i_mutex is held by caller */
2483 if (unlikely(info->seals & (F_SEAL_GROW |
2484 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2485 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2487 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2491 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2495 shmem_write_end(struct file *file, struct address_space *mapping,
2496 loff_t pos, unsigned len, unsigned copied,
2497 struct page *page, void *fsdata)
2499 struct inode *inode = mapping->host;
2501 if (pos + copied > inode->i_size)
2502 i_size_write(inode, pos + copied);
2504 if (!PageUptodate(page)) {
2505 struct page *head = compound_head(page);
2506 if (PageTransCompound(page)) {
2509 for (i = 0; i < HPAGE_PMD_NR; i++) {
2510 if (head + i == page)
2512 clear_highpage(head + i);
2513 flush_dcache_page(head + i);
2516 if (copied < PAGE_SIZE) {
2517 unsigned from = pos & (PAGE_SIZE - 1);
2518 zero_user_segments(page, 0, from,
2519 from + copied, PAGE_SIZE);
2521 SetPageUptodate(head);
2523 set_page_dirty(page);
2530 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2532 struct file *file = iocb->ki_filp;
2533 struct inode *inode = file_inode(file);
2534 struct address_space *mapping = inode->i_mapping;
2536 unsigned long offset;
2537 enum sgp_type sgp = SGP_READ;
2540 loff_t *ppos = &iocb->ki_pos;
2543 * Might this read be for a stacking filesystem? Then when reading
2544 * holes of a sparse file, we actually need to allocate those pages,
2545 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2547 if (!iter_is_iovec(to))
2550 index = *ppos >> PAGE_SHIFT;
2551 offset = *ppos & ~PAGE_MASK;
2554 struct page *page = NULL;
2556 unsigned long nr, ret;
2557 loff_t i_size = i_size_read(inode);
2559 end_index = i_size >> PAGE_SHIFT;
2560 if (index > end_index)
2562 if (index == end_index) {
2563 nr = i_size & ~PAGE_MASK;
2568 error = shmem_getpage(inode, index, &page, sgp);
2570 if (error == -EINVAL)
2575 if (sgp == SGP_CACHE)
2576 set_page_dirty(page);
2581 * We must evaluate after, since reads (unlike writes)
2582 * are called without i_mutex protection against truncate
2585 i_size = i_size_read(inode);
2586 end_index = i_size >> PAGE_SHIFT;
2587 if (index == end_index) {
2588 nr = i_size & ~PAGE_MASK;
2599 * If users can be writing to this page using arbitrary
2600 * virtual addresses, take care about potential aliasing
2601 * before reading the page on the kernel side.
2603 if (mapping_writably_mapped(mapping))
2604 flush_dcache_page(page);
2606 * Mark the page accessed if we read the beginning.
2609 mark_page_accessed(page);
2611 page = ZERO_PAGE(0);
2616 * Ok, we have the page, and it's up-to-date, so
2617 * now we can copy it to user space...
2619 ret = copy_page_to_iter(page, offset, nr, to);
2622 index += offset >> PAGE_SHIFT;
2623 offset &= ~PAGE_MASK;
2626 if (!iov_iter_count(to))
2635 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2636 file_accessed(file);
2637 return retval ? retval : error;
2641 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2643 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2644 pgoff_t index, pgoff_t end, int whence)
2647 struct pagevec pvec;
2648 pgoff_t indices[PAGEVEC_SIZE];
2652 pagevec_init(&pvec);
2653 pvec.nr = 1; /* start small: we may be there already */
2655 pvec.nr = find_get_entries(mapping, index,
2656 pvec.nr, pvec.pages, indices);
2658 if (whence == SEEK_DATA)
2662 for (i = 0; i < pvec.nr; i++, index++) {
2663 if (index < indices[i]) {
2664 if (whence == SEEK_HOLE) {
2670 page = pvec.pages[i];
2671 if (page && !xa_is_value(page)) {
2672 if (!PageUptodate(page))
2676 (page && whence == SEEK_DATA) ||
2677 (!page && whence == SEEK_HOLE)) {
2682 pagevec_remove_exceptionals(&pvec);
2683 pagevec_release(&pvec);
2684 pvec.nr = PAGEVEC_SIZE;
2690 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2692 struct address_space *mapping = file->f_mapping;
2693 struct inode *inode = mapping->host;
2697 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2698 return generic_file_llseek_size(file, offset, whence,
2699 MAX_LFS_FILESIZE, i_size_read(inode));
2701 /* We're holding i_mutex so we can access i_size directly */
2703 if (offset < 0 || offset >= inode->i_size)
2706 start = offset >> PAGE_SHIFT;
2707 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2708 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2709 new_offset <<= PAGE_SHIFT;
2710 if (new_offset > offset) {
2711 if (new_offset < inode->i_size)
2712 offset = new_offset;
2713 else if (whence == SEEK_DATA)
2716 offset = inode->i_size;
2721 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2722 inode_unlock(inode);
2726 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2729 struct inode *inode = file_inode(file);
2730 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2731 struct shmem_inode_info *info = SHMEM_I(inode);
2732 struct shmem_falloc shmem_falloc;
2733 pgoff_t start, index, end;
2736 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2741 if (mode & FALLOC_FL_PUNCH_HOLE) {
2742 struct address_space *mapping = file->f_mapping;
2743 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2744 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2745 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2747 /* protected by i_mutex */
2748 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2753 shmem_falloc.waitq = &shmem_falloc_waitq;
2754 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2755 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2756 spin_lock(&inode->i_lock);
2757 inode->i_private = &shmem_falloc;
2758 spin_unlock(&inode->i_lock);
2760 if ((u64)unmap_end > (u64)unmap_start)
2761 unmap_mapping_range(mapping, unmap_start,
2762 1 + unmap_end - unmap_start, 0);
2763 shmem_truncate_range(inode, offset, offset + len - 1);
2764 /* No need to unmap again: hole-punching leaves COWed pages */
2766 spin_lock(&inode->i_lock);
2767 inode->i_private = NULL;
2768 wake_up_all(&shmem_falloc_waitq);
2769 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2770 spin_unlock(&inode->i_lock);
2775 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2776 error = inode_newsize_ok(inode, offset + len);
2780 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2785 start = offset >> PAGE_SHIFT;
2786 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2787 /* Try to avoid a swapstorm if len is impossible to satisfy */
2788 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2793 shmem_falloc.waitq = NULL;
2794 shmem_falloc.start = start;
2795 shmem_falloc.next = start;
2796 shmem_falloc.nr_falloced = 0;
2797 shmem_falloc.nr_unswapped = 0;
2798 spin_lock(&inode->i_lock);
2799 inode->i_private = &shmem_falloc;
2800 spin_unlock(&inode->i_lock);
2802 for (index = start; index < end; index++) {
2806 * Good, the fallocate(2) manpage permits EINTR: we may have
2807 * been interrupted because we are using up too much memory.
2809 if (signal_pending(current))
2811 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2814 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2816 /* Remove the !PageUptodate pages we added */
2817 if (index > start) {
2818 shmem_undo_range(inode,
2819 (loff_t)start << PAGE_SHIFT,
2820 ((loff_t)index << PAGE_SHIFT) - 1, true);
2826 * Inform shmem_writepage() how far we have reached.
2827 * No need for lock or barrier: we have the page lock.
2829 shmem_falloc.next++;
2830 if (!PageUptodate(page))
2831 shmem_falloc.nr_falloced++;
2834 * If !PageUptodate, leave it that way so that freeable pages
2835 * can be recognized if we need to rollback on error later.
2836 * But set_page_dirty so that memory pressure will swap rather
2837 * than free the pages we are allocating (and SGP_CACHE pages
2838 * might still be clean: we now need to mark those dirty too).
2840 set_page_dirty(page);
2846 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2847 i_size_write(inode, offset + len);
2848 inode->i_ctime = current_time(inode);
2850 spin_lock(&inode->i_lock);
2851 inode->i_private = NULL;
2852 spin_unlock(&inode->i_lock);
2854 inode_unlock(inode);
2858 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2860 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2862 buf->f_type = TMPFS_MAGIC;
2863 buf->f_bsize = PAGE_SIZE;
2864 buf->f_namelen = NAME_MAX;
2865 if (sbinfo->max_blocks) {
2866 buf->f_blocks = sbinfo->max_blocks;
2868 buf->f_bfree = sbinfo->max_blocks -
2869 percpu_counter_sum(&sbinfo->used_blocks);
2871 if (sbinfo->max_inodes) {
2872 buf->f_files = sbinfo->max_inodes;
2873 buf->f_ffree = sbinfo->free_inodes;
2875 /* else leave those fields 0 like simple_statfs */
2880 * File creation. Allocate an inode, and we're done..
2883 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2885 struct inode *inode;
2886 int error = -ENOSPC;
2888 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2890 error = simple_acl_create(dir, inode);
2893 error = security_inode_init_security(inode, dir,
2895 shmem_initxattrs, NULL);
2896 if (error && error != -EOPNOTSUPP)
2900 dir->i_size += BOGO_DIRENT_SIZE;
2901 dir->i_ctime = dir->i_mtime = current_time(dir);
2902 d_instantiate(dentry, inode);
2903 dget(dentry); /* Extra count - pin the dentry in core */
2912 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2914 struct inode *inode;
2915 int error = -ENOSPC;
2917 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2919 error = security_inode_init_security(inode, dir,
2921 shmem_initxattrs, NULL);
2922 if (error && error != -EOPNOTSUPP)
2924 error = simple_acl_create(dir, inode);
2927 d_tmpfile(dentry, inode);
2935 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2939 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2945 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2948 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2954 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2956 struct inode *inode = d_inode(old_dentry);
2960 * No ordinary (disk based) filesystem counts links as inodes;
2961 * but each new link needs a new dentry, pinning lowmem, and
2962 * tmpfs dentries cannot be pruned until they are unlinked.
2963 * But if an O_TMPFILE file is linked into the tmpfs, the
2964 * first link must skip that, to get the accounting right.
2966 if (inode->i_nlink) {
2967 ret = shmem_reserve_inode(inode->i_sb);
2972 dir->i_size += BOGO_DIRENT_SIZE;
2973 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2975 ihold(inode); /* New dentry reference */
2976 dget(dentry); /* Extra pinning count for the created dentry */
2977 d_instantiate(dentry, inode);
2982 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2984 struct inode *inode = d_inode(dentry);
2986 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2987 shmem_free_inode(inode->i_sb);
2989 dir->i_size -= BOGO_DIRENT_SIZE;
2990 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2992 dput(dentry); /* Undo the count from "create" - this does all the work */
2996 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2998 if (!simple_empty(dentry))
3001 drop_nlink(d_inode(dentry));
3003 return shmem_unlink(dir, dentry);
3006 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3008 bool old_is_dir = d_is_dir(old_dentry);
3009 bool new_is_dir = d_is_dir(new_dentry);
3011 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3013 drop_nlink(old_dir);
3016 drop_nlink(new_dir);
3020 old_dir->i_ctime = old_dir->i_mtime =
3021 new_dir->i_ctime = new_dir->i_mtime =
3022 d_inode(old_dentry)->i_ctime =
3023 d_inode(new_dentry)->i_ctime = current_time(old_dir);
3028 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3030 struct dentry *whiteout;
3033 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3037 error = shmem_mknod(old_dir, whiteout,
3038 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3044 * Cheat and hash the whiteout while the old dentry is still in
3045 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3047 * d_lookup() will consistently find one of them at this point,
3048 * not sure which one, but that isn't even important.
3055 * The VFS layer already does all the dentry stuff for rename,
3056 * we just have to decrement the usage count for the target if
3057 * it exists so that the VFS layer correctly free's it when it
3060 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3062 struct inode *inode = d_inode(old_dentry);
3063 int they_are_dirs = S_ISDIR(inode->i_mode);
3065 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3068 if (flags & RENAME_EXCHANGE)
3069 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3071 if (!simple_empty(new_dentry))
3074 if (flags & RENAME_WHITEOUT) {
3077 error = shmem_whiteout(old_dir, old_dentry);
3082 if (d_really_is_positive(new_dentry)) {
3083 (void) shmem_unlink(new_dir, new_dentry);
3084 if (they_are_dirs) {
3085 drop_nlink(d_inode(new_dentry));
3086 drop_nlink(old_dir);
3088 } else if (they_are_dirs) {
3089 drop_nlink(old_dir);
3093 old_dir->i_size -= BOGO_DIRENT_SIZE;
3094 new_dir->i_size += BOGO_DIRENT_SIZE;
3095 old_dir->i_ctime = old_dir->i_mtime =
3096 new_dir->i_ctime = new_dir->i_mtime =
3097 inode->i_ctime = current_time(old_dir);
3101 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3105 struct inode *inode;
3108 len = strlen(symname) + 1;
3109 if (len > PAGE_SIZE)
3110 return -ENAMETOOLONG;
3112 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3117 error = security_inode_init_security(inode, dir, &dentry->d_name,
3118 shmem_initxattrs, NULL);
3120 if (error != -EOPNOTSUPP) {
3127 inode->i_size = len-1;
3128 if (len <= SHORT_SYMLINK_LEN) {
3129 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3130 if (!inode->i_link) {
3134 inode->i_op = &shmem_short_symlink_operations;
3136 inode_nohighmem(inode);
3137 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3142 inode->i_mapping->a_ops = &shmem_aops;
3143 inode->i_op = &shmem_symlink_inode_operations;
3144 memcpy(page_address(page), symname, len);
3145 SetPageUptodate(page);
3146 set_page_dirty(page);
3150 dir->i_size += BOGO_DIRENT_SIZE;
3151 dir->i_ctime = dir->i_mtime = current_time(dir);
3152 d_instantiate(dentry, inode);
3157 static void shmem_put_link(void *arg)
3159 mark_page_accessed(arg);
3163 static const char *shmem_get_link(struct dentry *dentry,
3164 struct inode *inode,
3165 struct delayed_call *done)
3167 struct page *page = NULL;
3170 page = find_get_page(inode->i_mapping, 0);
3172 return ERR_PTR(-ECHILD);
3173 if (!PageUptodate(page)) {
3175 return ERR_PTR(-ECHILD);
3178 error = shmem_getpage(inode, 0, &page, SGP_READ);
3180 return ERR_PTR(error);
3183 set_delayed_call(done, shmem_put_link, page);
3184 return page_address(page);
3187 #ifdef CONFIG_TMPFS_XATTR
3189 * Superblocks without xattr inode operations may get some security.* xattr
3190 * support from the LSM "for free". As soon as we have any other xattrs
3191 * like ACLs, we also need to implement the security.* handlers at
3192 * filesystem level, though.
3196 * Callback for security_inode_init_security() for acquiring xattrs.
3198 static int shmem_initxattrs(struct inode *inode,
3199 const struct xattr *xattr_array,
3202 struct shmem_inode_info *info = SHMEM_I(inode);
3203 const struct xattr *xattr;
3204 struct simple_xattr *new_xattr;
3207 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3208 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3212 len = strlen(xattr->name) + 1;
3213 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3215 if (!new_xattr->name) {
3220 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3221 XATTR_SECURITY_PREFIX_LEN);
3222 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3225 simple_xattr_list_add(&info->xattrs, new_xattr);
3231 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3232 struct dentry *unused, struct inode *inode,
3233 const char *name, void *buffer, size_t size)
3235 struct shmem_inode_info *info = SHMEM_I(inode);
3237 name = xattr_full_name(handler, name);
3238 return simple_xattr_get(&info->xattrs, name, buffer, size);
3241 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3242 struct dentry *unused, struct inode *inode,
3243 const char *name, const void *value,
3244 size_t size, int flags)
3246 struct shmem_inode_info *info = SHMEM_I(inode);
3248 name = xattr_full_name(handler, name);
3249 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3252 static const struct xattr_handler shmem_security_xattr_handler = {
3253 .prefix = XATTR_SECURITY_PREFIX,
3254 .get = shmem_xattr_handler_get,
3255 .set = shmem_xattr_handler_set,
3258 static const struct xattr_handler shmem_trusted_xattr_handler = {
3259 .prefix = XATTR_TRUSTED_PREFIX,
3260 .get = shmem_xattr_handler_get,
3261 .set = shmem_xattr_handler_set,
3264 static const struct xattr_handler *shmem_xattr_handlers[] = {
3265 #ifdef CONFIG_TMPFS_POSIX_ACL
3266 &posix_acl_access_xattr_handler,
3267 &posix_acl_default_xattr_handler,
3269 &shmem_security_xattr_handler,
3270 &shmem_trusted_xattr_handler,
3274 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3276 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3277 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3279 #endif /* CONFIG_TMPFS_XATTR */
3281 static const struct inode_operations shmem_short_symlink_operations = {
3282 .get_link = simple_get_link,
3283 #ifdef CONFIG_TMPFS_XATTR
3284 .listxattr = shmem_listxattr,
3288 static const struct inode_operations shmem_symlink_inode_operations = {
3289 .get_link = shmem_get_link,
3290 #ifdef CONFIG_TMPFS_XATTR
3291 .listxattr = shmem_listxattr,
3295 static struct dentry *shmem_get_parent(struct dentry *child)
3297 return ERR_PTR(-ESTALE);
3300 static int shmem_match(struct inode *ino, void *vfh)
3304 inum = (inum << 32) | fh[1];
3305 return ino->i_ino == inum && fh[0] == ino->i_generation;
3308 /* Find any alias of inode, but prefer a hashed alias */
3309 static struct dentry *shmem_find_alias(struct inode *inode)
3311 struct dentry *alias = d_find_alias(inode);
3313 return alias ?: d_find_any_alias(inode);
3317 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3318 struct fid *fid, int fh_len, int fh_type)
3320 struct inode *inode;
3321 struct dentry *dentry = NULL;
3328 inum = (inum << 32) | fid->raw[1];
3330 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3331 shmem_match, fid->raw);
3333 dentry = shmem_find_alias(inode);
3340 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3341 struct inode *parent)
3345 return FILEID_INVALID;
3348 if (inode_unhashed(inode)) {
3349 /* Unfortunately insert_inode_hash is not idempotent,
3350 * so as we hash inodes here rather than at creation
3351 * time, we need a lock to ensure we only try
3354 static DEFINE_SPINLOCK(lock);
3356 if (inode_unhashed(inode))
3357 __insert_inode_hash(inode,
3358 inode->i_ino + inode->i_generation);
3362 fh[0] = inode->i_generation;
3363 fh[1] = inode->i_ino;
3364 fh[2] = ((__u64)inode->i_ino) >> 32;
3370 static const struct export_operations shmem_export_ops = {
3371 .get_parent = shmem_get_parent,
3372 .encode_fh = shmem_encode_fh,
3373 .fh_to_dentry = shmem_fh_to_dentry,
3387 static const struct fs_parameter_spec shmem_param_specs[] = {
3388 fsparam_u32 ("gid", Opt_gid),
3389 fsparam_enum ("huge", Opt_huge),
3390 fsparam_u32oct("mode", Opt_mode),
3391 fsparam_string("mpol", Opt_mpol),
3392 fsparam_string("nr_blocks", Opt_nr_blocks),
3393 fsparam_string("nr_inodes", Opt_nr_inodes),
3394 fsparam_string("size", Opt_size),
3395 fsparam_u32 ("uid", Opt_uid),
3399 static const struct fs_parameter_enum shmem_param_enums[] = {
3400 { Opt_huge, "never", SHMEM_HUGE_NEVER },
3401 { Opt_huge, "always", SHMEM_HUGE_ALWAYS },
3402 { Opt_huge, "within_size", SHMEM_HUGE_WITHIN_SIZE },
3403 { Opt_huge, "advise", SHMEM_HUGE_ADVISE },
3407 const struct fs_parameter_description shmem_fs_parameters = {
3409 .specs = shmem_param_specs,
3410 .enums = shmem_param_enums,
3413 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3415 struct shmem_options *ctx = fc->fs_private;
3416 struct fs_parse_result result;
3417 unsigned long long size;
3423 opt = fs_parse(fc, &shmem_fs_parameters, param, &result);
3429 size = memparse(param->string, &rest);
3431 size <<= PAGE_SHIFT;
3432 size *= totalram_pages();
3438 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3439 ctx->seen |= SHMEM_SEEN_BLOCKS;
3442 ctx->blocks = memparse(param->string, &rest);
3445 ctx->seen |= SHMEM_SEEN_BLOCKS;
3448 ctx->inodes = memparse(param->string, &rest);
3451 ctx->seen |= SHMEM_SEEN_INODES;
3454 ctx->mode = result.uint_32 & 07777;
3457 kuid = make_kuid(current_user_ns(), result.uint_32);
3458 if (!uid_valid(kuid))
3462 * The requested uid must be representable in the
3463 * filesystem's idmapping.
3465 if (!kuid_has_mapping(fc->user_ns, kuid))
3471 kgid = make_kgid(current_user_ns(), result.uint_32);
3472 if (!gid_valid(kgid))
3476 * The requested gid must be representable in the
3477 * filesystem's idmapping.
3479 if (!kgid_has_mapping(fc->user_ns, kgid))
3485 ctx->huge = result.uint_32;
3486 if (ctx->huge != SHMEM_HUGE_NEVER &&
3487 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
3488 has_transparent_hugepage()))
3489 goto unsupported_parameter;
3490 ctx->seen |= SHMEM_SEEN_HUGE;
3493 if (IS_ENABLED(CONFIG_NUMA)) {
3494 mpol_put(ctx->mpol);
3496 if (mpol_parse_str(param->string, &ctx->mpol))
3500 goto unsupported_parameter;
3504 unsupported_parameter:
3505 return invalf(fc, "tmpfs: Unsupported parameter '%s'", param->key);
3507 return invalf(fc, "tmpfs: Bad value for '%s'", param->key);
3510 static int shmem_parse_options(struct fs_context *fc, void *data)
3512 char *options = data;
3515 int err = security_sb_eat_lsm_opts(options, &fc->security);
3520 while (options != NULL) {
3521 char *this_char = options;
3524 * NUL-terminate this option: unfortunately,
3525 * mount options form a comma-separated list,
3526 * but mpol's nodelist may also contain commas.
3528 options = strchr(options, ',');
3529 if (options == NULL)
3532 if (!isdigit(*options)) {
3538 char *value = strchr(this_char,'=');
3544 len = strlen(value);
3546 err = vfs_parse_fs_string(fc, this_char, value, len);
3555 * Reconfigure a shmem filesystem.
3557 * Note that we disallow change from limited->unlimited blocks/inodes while any
3558 * are in use; but we must separately disallow unlimited->limited, because in
3559 * that case we have no record of how much is already in use.
3561 static int shmem_reconfigure(struct fs_context *fc)
3563 struct shmem_options *ctx = fc->fs_private;
3564 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3565 unsigned long inodes;
3568 spin_lock(&sbinfo->stat_lock);
3569 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3570 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3571 if (!sbinfo->max_blocks) {
3572 err = "Cannot retroactively limit size";
3575 if (percpu_counter_compare(&sbinfo->used_blocks,
3577 err = "Too small a size for current use";
3581 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3582 if (!sbinfo->max_inodes) {
3583 err = "Cannot retroactively limit inodes";
3586 if (ctx->inodes < inodes) {
3587 err = "Too few inodes for current use";
3592 if (ctx->seen & SHMEM_SEEN_HUGE)
3593 sbinfo->huge = ctx->huge;
3594 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3595 sbinfo->max_blocks = ctx->blocks;
3596 if (ctx->seen & SHMEM_SEEN_INODES) {
3597 sbinfo->max_inodes = ctx->inodes;
3598 sbinfo->free_inodes = ctx->inodes - inodes;
3602 * Preserve previous mempolicy unless mpol remount option was specified.
3605 mpol_put(sbinfo->mpol);
3606 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3609 spin_unlock(&sbinfo->stat_lock);
3612 spin_unlock(&sbinfo->stat_lock);
3613 return invalf(fc, "tmpfs: %s", err);
3616 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3618 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3620 if (sbinfo->max_blocks != shmem_default_max_blocks())
3621 seq_printf(seq, ",size=%luk",
3622 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3623 if (sbinfo->max_inodes != shmem_default_max_inodes())
3624 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3625 if (sbinfo->mode != (0777 | S_ISVTX))
3626 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3627 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3628 seq_printf(seq, ",uid=%u",
3629 from_kuid_munged(&init_user_ns, sbinfo->uid));
3630 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3631 seq_printf(seq, ",gid=%u",
3632 from_kgid_munged(&init_user_ns, sbinfo->gid));
3633 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3634 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3636 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3638 shmem_show_mpol(seq, sbinfo->mpol);
3642 #endif /* CONFIG_TMPFS */
3644 static void shmem_put_super(struct super_block *sb)
3646 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3648 percpu_counter_destroy(&sbinfo->used_blocks);
3649 mpol_put(sbinfo->mpol);
3651 sb->s_fs_info = NULL;
3654 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3656 struct shmem_options *ctx = fc->fs_private;
3657 struct inode *inode;
3658 struct shmem_sb_info *sbinfo;
3661 /* Round up to L1_CACHE_BYTES to resist false sharing */
3662 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3663 L1_CACHE_BYTES), GFP_KERNEL);
3667 sb->s_fs_info = sbinfo;
3671 * Per default we only allow half of the physical ram per
3672 * tmpfs instance, limiting inodes to one per page of lowmem;
3673 * but the internal instance is left unlimited.
3675 if (!(sb->s_flags & SB_KERNMOUNT)) {
3676 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3677 ctx->blocks = shmem_default_max_blocks();
3678 if (!(ctx->seen & SHMEM_SEEN_INODES))
3679 ctx->inodes = shmem_default_max_inodes();
3681 sb->s_flags |= SB_NOUSER;
3683 sb->s_export_op = &shmem_export_ops;
3684 sb->s_flags |= SB_NOSEC;
3686 sb->s_flags |= SB_NOUSER;
3688 sbinfo->max_blocks = ctx->blocks;
3689 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3690 sbinfo->uid = ctx->uid;
3691 sbinfo->gid = ctx->gid;
3692 sbinfo->mode = ctx->mode;
3693 sbinfo->huge = ctx->huge;
3694 sbinfo->mpol = ctx->mpol;
3697 spin_lock_init(&sbinfo->stat_lock);
3698 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3700 spin_lock_init(&sbinfo->shrinklist_lock);
3701 INIT_LIST_HEAD(&sbinfo->shrinklist);
3703 sb->s_maxbytes = MAX_LFS_FILESIZE;
3704 sb->s_blocksize = PAGE_SIZE;
3705 sb->s_blocksize_bits = PAGE_SHIFT;
3706 sb->s_magic = TMPFS_MAGIC;
3707 sb->s_op = &shmem_ops;
3708 sb->s_time_gran = 1;
3709 #ifdef CONFIG_TMPFS_XATTR
3710 sb->s_xattr = shmem_xattr_handlers;
3712 #ifdef CONFIG_TMPFS_POSIX_ACL
3713 sb->s_flags |= SB_POSIXACL;
3715 uuid_gen(&sb->s_uuid);
3717 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3720 inode->i_uid = sbinfo->uid;
3721 inode->i_gid = sbinfo->gid;
3722 sb->s_root = d_make_root(inode);
3728 shmem_put_super(sb);
3732 static int shmem_get_tree(struct fs_context *fc)
3734 return get_tree_nodev(fc, shmem_fill_super);
3737 static void shmem_free_fc(struct fs_context *fc)
3739 struct shmem_options *ctx = fc->fs_private;
3742 mpol_put(ctx->mpol);
3747 static const struct fs_context_operations shmem_fs_context_ops = {
3748 .free = shmem_free_fc,
3749 .get_tree = shmem_get_tree,
3751 .parse_monolithic = shmem_parse_options,
3752 .parse_param = shmem_parse_one,
3753 .reconfigure = shmem_reconfigure,
3757 static struct kmem_cache *shmem_inode_cachep;
3759 static struct inode *shmem_alloc_inode(struct super_block *sb)
3761 struct shmem_inode_info *info;
3762 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3765 return &info->vfs_inode;
3768 static void shmem_free_in_core_inode(struct inode *inode)
3770 if (S_ISLNK(inode->i_mode))
3771 kfree(inode->i_link);
3772 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3775 static void shmem_destroy_inode(struct inode *inode)
3777 if (S_ISREG(inode->i_mode))
3778 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3781 static void shmem_init_inode(void *foo)
3783 struct shmem_inode_info *info = foo;
3784 inode_init_once(&info->vfs_inode);
3787 static void shmem_init_inodecache(void)
3789 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3790 sizeof(struct shmem_inode_info),
3791 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3794 static void shmem_destroy_inodecache(void)
3796 kmem_cache_destroy(shmem_inode_cachep);
3799 static const struct address_space_operations shmem_aops = {
3800 .writepage = shmem_writepage,
3801 .set_page_dirty = __set_page_dirty_no_writeback,
3803 .write_begin = shmem_write_begin,
3804 .write_end = shmem_write_end,
3806 #ifdef CONFIG_MIGRATION
3807 .migratepage = migrate_page,
3809 .error_remove_page = generic_error_remove_page,
3812 static const struct file_operations shmem_file_operations = {
3814 .get_unmapped_area = shmem_get_unmapped_area,
3816 .llseek = shmem_file_llseek,
3817 .read_iter = shmem_file_read_iter,
3818 .write_iter = generic_file_write_iter,
3819 .fsync = noop_fsync,
3820 .splice_read = generic_file_splice_read,
3821 .splice_write = iter_file_splice_write,
3822 .fallocate = shmem_fallocate,
3826 static const struct inode_operations shmem_inode_operations = {
3827 .getattr = shmem_getattr,
3828 .setattr = shmem_setattr,
3829 #ifdef CONFIG_TMPFS_XATTR
3830 .listxattr = shmem_listxattr,
3831 .set_acl = simple_set_acl,
3835 static const struct inode_operations shmem_dir_inode_operations = {
3837 .create = shmem_create,
3838 .lookup = simple_lookup,
3840 .unlink = shmem_unlink,
3841 .symlink = shmem_symlink,
3842 .mkdir = shmem_mkdir,
3843 .rmdir = shmem_rmdir,
3844 .mknod = shmem_mknod,
3845 .rename = shmem_rename2,
3846 .tmpfile = shmem_tmpfile,
3848 #ifdef CONFIG_TMPFS_XATTR
3849 .listxattr = shmem_listxattr,
3851 #ifdef CONFIG_TMPFS_POSIX_ACL
3852 .setattr = shmem_setattr,
3853 .set_acl = simple_set_acl,
3857 static const struct inode_operations shmem_special_inode_operations = {
3858 #ifdef CONFIG_TMPFS_XATTR
3859 .listxattr = shmem_listxattr,
3861 #ifdef CONFIG_TMPFS_POSIX_ACL
3862 .setattr = shmem_setattr,
3863 .set_acl = simple_set_acl,
3867 static const struct super_operations shmem_ops = {
3868 .alloc_inode = shmem_alloc_inode,
3869 .free_inode = shmem_free_in_core_inode,
3870 .destroy_inode = shmem_destroy_inode,
3872 .statfs = shmem_statfs,
3873 .show_options = shmem_show_options,
3875 .evict_inode = shmem_evict_inode,
3876 .drop_inode = generic_delete_inode,
3877 .put_super = shmem_put_super,
3878 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3879 .nr_cached_objects = shmem_unused_huge_count,
3880 .free_cached_objects = shmem_unused_huge_scan,
3884 static const struct vm_operations_struct shmem_vm_ops = {
3885 .fault = shmem_fault,
3886 .map_pages = filemap_map_pages,
3888 .set_policy = shmem_set_policy,
3889 .get_policy = shmem_get_policy,
3893 int shmem_init_fs_context(struct fs_context *fc)
3895 struct shmem_options *ctx;
3897 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3901 ctx->mode = 0777 | S_ISVTX;
3902 ctx->uid = current_fsuid();
3903 ctx->gid = current_fsgid();
3905 fc->fs_private = ctx;
3906 fc->ops = &shmem_fs_context_ops;
3910 static struct file_system_type shmem_fs_type = {
3911 .owner = THIS_MODULE,
3913 .init_fs_context = shmem_init_fs_context,
3915 .parameters = &shmem_fs_parameters,
3917 .kill_sb = kill_litter_super,
3918 .fs_flags = FS_USERNS_MOUNT,
3921 int __init shmem_init(void)
3925 shmem_init_inodecache();
3927 error = register_filesystem(&shmem_fs_type);
3929 pr_err("Could not register tmpfs\n");
3933 shm_mnt = kern_mount(&shmem_fs_type);
3934 if (IS_ERR(shm_mnt)) {
3935 error = PTR_ERR(shm_mnt);
3936 pr_err("Could not kern_mount tmpfs\n");
3940 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3941 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3942 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3944 shmem_huge = 0; /* just in case it was patched */
3949 unregister_filesystem(&shmem_fs_type);
3951 shmem_destroy_inodecache();
3952 shm_mnt = ERR_PTR(error);
3956 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3957 static ssize_t shmem_enabled_show(struct kobject *kobj,
3958 struct kobj_attribute *attr, char *buf)
3962 SHMEM_HUGE_WITHIN_SIZE,
3970 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3971 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3973 count += sprintf(buf + count, fmt,
3974 shmem_format_huge(values[i]));
3976 buf[count - 1] = '\n';
3980 static ssize_t shmem_enabled_store(struct kobject *kobj,
3981 struct kobj_attribute *attr, const char *buf, size_t count)
3986 if (count + 1 > sizeof(tmp))
3988 memcpy(tmp, buf, count);
3990 if (count && tmp[count - 1] == '\n')
3991 tmp[count - 1] = '\0';
3993 huge = shmem_parse_huge(tmp);
3994 if (huge == -EINVAL)
3996 if (!has_transparent_hugepage() &&
3997 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4001 if (shmem_huge > SHMEM_HUGE_DENY)
4002 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4006 struct kobj_attribute shmem_enabled_attr =
4007 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4008 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4010 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4011 bool shmem_huge_enabled(struct vm_area_struct *vma)
4013 struct inode *inode = file_inode(vma->vm_file);
4014 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4018 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
4019 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
4021 if (shmem_huge == SHMEM_HUGE_FORCE)
4023 if (shmem_huge == SHMEM_HUGE_DENY)
4025 switch (sbinfo->huge) {
4026 case SHMEM_HUGE_NEVER:
4028 case SHMEM_HUGE_ALWAYS:
4030 case SHMEM_HUGE_WITHIN_SIZE:
4031 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4032 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4033 if (i_size >= HPAGE_PMD_SIZE &&
4034 i_size >> PAGE_SHIFT >= off)
4037 case SHMEM_HUGE_ADVISE:
4038 /* TODO: implement fadvise() hints */
4039 return (vma->vm_flags & VM_HUGEPAGE);
4045 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4047 #else /* !CONFIG_SHMEM */
4050 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4052 * This is intended for small system where the benefits of the full
4053 * shmem code (swap-backed and resource-limited) are outweighed by
4054 * their complexity. On systems without swap this code should be
4055 * effectively equivalent, but much lighter weight.
4058 static struct file_system_type shmem_fs_type = {
4060 .init_fs_context = ramfs_init_fs_context,
4061 .parameters = &ramfs_fs_parameters,
4062 .kill_sb = kill_litter_super,
4063 .fs_flags = FS_USERNS_MOUNT,
4066 int __init shmem_init(void)
4068 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4070 shm_mnt = kern_mount(&shmem_fs_type);
4071 BUG_ON(IS_ERR(shm_mnt));
4076 int shmem_unuse(unsigned int type, bool frontswap,
4077 unsigned long *fs_pages_to_unuse)
4082 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4087 void shmem_unlock_mapping(struct address_space *mapping)
4092 unsigned long shmem_get_unmapped_area(struct file *file,
4093 unsigned long addr, unsigned long len,
4094 unsigned long pgoff, unsigned long flags)
4096 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4100 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4102 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4104 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4106 #define shmem_vm_ops generic_file_vm_ops
4107 #define shmem_file_operations ramfs_file_operations
4108 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4109 #define shmem_acct_size(flags, size) 0
4110 #define shmem_unacct_size(flags, size) do {} while (0)
4112 #endif /* CONFIG_SHMEM */
4116 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4117 unsigned long flags, unsigned int i_flags)
4119 struct inode *inode;
4123 return ERR_CAST(mnt);
4125 if (size < 0 || size > MAX_LFS_FILESIZE)
4126 return ERR_PTR(-EINVAL);
4128 if (shmem_acct_size(flags, size))
4129 return ERR_PTR(-ENOMEM);
4131 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4133 if (unlikely(!inode)) {
4134 shmem_unacct_size(flags, size);
4135 return ERR_PTR(-ENOSPC);
4137 inode->i_flags |= i_flags;
4138 inode->i_size = size;
4139 clear_nlink(inode); /* It is unlinked */
4140 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4142 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4143 &shmem_file_operations);
4150 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4151 * kernel internal. There will be NO LSM permission checks against the
4152 * underlying inode. So users of this interface must do LSM checks at a
4153 * higher layer. The users are the big_key and shm implementations. LSM
4154 * checks are provided at the key or shm level rather than the inode.
4155 * @name: name for dentry (to be seen in /proc/<pid>/maps
4156 * @size: size to be set for the file
4157 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4159 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4161 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4165 * shmem_file_setup - get an unlinked file living in tmpfs
4166 * @name: name for dentry (to be seen in /proc/<pid>/maps
4167 * @size: size to be set for the file
4168 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4170 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4172 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4174 EXPORT_SYMBOL_GPL(shmem_file_setup);
4177 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4178 * @mnt: the tmpfs mount where the file will be created
4179 * @name: name for dentry (to be seen in /proc/<pid>/maps
4180 * @size: size to be set for the file
4181 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4183 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4184 loff_t size, unsigned long flags)
4186 return __shmem_file_setup(mnt, name, size, flags, 0);
4188 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4191 * shmem_zero_setup - setup a shared anonymous mapping
4192 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4194 int shmem_zero_setup(struct vm_area_struct *vma)
4197 loff_t size = vma->vm_end - vma->vm_start;
4200 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4201 * between XFS directory reading and selinux: since this file is only
4202 * accessible to the user through its mapping, use S_PRIVATE flag to
4203 * bypass file security, in the same way as shmem_kernel_file_setup().
4205 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4207 return PTR_ERR(file);
4211 vma->vm_file = file;
4212 vma->vm_ops = &shmem_vm_ops;
4214 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4215 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4216 (vma->vm_end & HPAGE_PMD_MASK)) {
4217 khugepaged_enter(vma, vma->vm_flags);
4224 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4225 * @mapping: the page's address_space
4226 * @index: the page index
4227 * @gfp: the page allocator flags to use if allocating
4229 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4230 * with any new page allocations done using the specified allocation flags.
4231 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4232 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4233 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4235 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4236 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4238 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4239 pgoff_t index, gfp_t gfp)
4242 struct inode *inode = mapping->host;
4246 BUG_ON(mapping->a_ops != &shmem_aops);
4247 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4248 gfp, NULL, NULL, NULL);
4250 page = ERR_PTR(error);
4256 * The tiny !SHMEM case uses ramfs without swap
4258 return read_cache_page_gfp(mapping, index, gfp);
4261 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);