2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
43 static struct vfsmount *shm_mnt;
47 * This virtual memory filesystem is heavily based on the ramfs. It
48 * extends ramfs by the ability to use swap and honor resource limits
49 * which makes it a completely usable filesystem.
52 #include <linux/xattr.h>
53 #include <linux/exportfs.h>
54 #include <linux/posix_acl.h>
55 #include <linux/posix_acl_xattr.h>
56 #include <linux/mman.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/backing-dev.h>
60 #include <linux/shmem_fs.h>
61 #include <linux/writeback.h>
62 #include <linux/blkdev.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/userfaultfd_k.h>
80 #include <linux/rmap.h>
81 #include <linux/uuid.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_mutex making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc {
102 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
103 pgoff_t start; /* start of range currently being fallocated */
104 pgoff_t next; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
109 struct shmem_options {
110 unsigned long long blocks;
111 unsigned long long inodes;
112 struct mempolicy *mpol;
119 #define SHMEM_SEEN_BLOCKS 1
120 #define SHMEM_SEEN_INODES 2
121 #define SHMEM_SEEN_HUGE 4
122 #define SHMEM_SEEN_INUMS 8
126 static unsigned long shmem_default_max_blocks(void)
128 return totalram_pages() / 2;
131 static unsigned long shmem_default_max_inodes(void)
133 unsigned long nr_pages = totalram_pages();
135 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
139 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
140 struct page **pagep, enum sgp_type sgp,
141 gfp_t gfp, struct vm_area_struct *vma,
142 vm_fault_t *fault_type);
143 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 struct vm_fault *vmf, vm_fault_t *fault_type);
148 int shmem_getpage(struct inode *inode, pgoff_t index,
149 struct page **pagep, enum sgp_type sgp)
151 return shmem_getpage_gfp(inode, index, pagep, sgp,
152 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
155 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
157 return sb->s_fs_info;
161 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
162 * for shared memory and for shared anonymous (/dev/zero) mappings
163 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
164 * consistent with the pre-accounting of private mappings ...
166 static inline int shmem_acct_size(unsigned long flags, loff_t size)
168 return (flags & VM_NORESERVE) ?
169 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
172 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
174 if (!(flags & VM_NORESERVE))
175 vm_unacct_memory(VM_ACCT(size));
178 static inline int shmem_reacct_size(unsigned long flags,
179 loff_t oldsize, loff_t newsize)
181 if (!(flags & VM_NORESERVE)) {
182 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
183 return security_vm_enough_memory_mm(current->mm,
184 VM_ACCT(newsize) - VM_ACCT(oldsize));
185 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
186 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
192 * ... whereas tmpfs objects are accounted incrementally as
193 * pages are allocated, in order to allow large sparse files.
194 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
195 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
197 static inline int shmem_acct_block(unsigned long flags, long pages)
199 if (!(flags & VM_NORESERVE))
202 return security_vm_enough_memory_mm(current->mm,
203 pages * VM_ACCT(PAGE_SIZE));
206 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
208 if (flags & VM_NORESERVE)
209 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
212 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
214 struct shmem_inode_info *info = SHMEM_I(inode);
215 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
217 if (shmem_acct_block(info->flags, pages))
220 if (sbinfo->max_blocks) {
221 if (percpu_counter_compare(&sbinfo->used_blocks,
222 sbinfo->max_blocks - pages) > 0)
224 percpu_counter_add(&sbinfo->used_blocks, pages);
230 shmem_unacct_blocks(info->flags, pages);
234 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
236 struct shmem_inode_info *info = SHMEM_I(inode);
237 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
239 if (sbinfo->max_blocks)
240 percpu_counter_sub(&sbinfo->used_blocks, pages);
241 shmem_unacct_blocks(info->flags, pages);
244 static const struct super_operations shmem_ops;
245 const struct address_space_operations shmem_aops;
246 static const struct file_operations shmem_file_operations;
247 static const struct inode_operations shmem_inode_operations;
248 static const struct inode_operations shmem_dir_inode_operations;
249 static const struct inode_operations shmem_special_inode_operations;
250 static const struct vm_operations_struct shmem_vm_ops;
251 static struct file_system_type shmem_fs_type;
253 bool vma_is_shmem(struct vm_area_struct *vma)
255 return vma->vm_ops == &shmem_vm_ops;
258 static LIST_HEAD(shmem_swaplist);
259 static DEFINE_MUTEX(shmem_swaplist_mutex);
262 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
263 * produces a novel ino for the newly allocated inode.
265 * It may also be called when making a hard link to permit the space needed by
266 * each dentry. However, in that case, no new inode number is needed since that
267 * internally draws from another pool of inode numbers (currently global
268 * get_next_ino()). This case is indicated by passing NULL as inop.
270 #define SHMEM_INO_BATCH 1024
271 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
273 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
276 if (!(sb->s_flags & SB_KERNMOUNT)) {
277 raw_spin_lock(&sbinfo->stat_lock);
278 if (sbinfo->max_inodes) {
279 if (!sbinfo->free_inodes) {
280 raw_spin_unlock(&sbinfo->stat_lock);
283 sbinfo->free_inodes--;
286 ino = sbinfo->next_ino++;
287 if (unlikely(is_zero_ino(ino)))
288 ino = sbinfo->next_ino++;
289 if (unlikely(!sbinfo->full_inums &&
292 * Emulate get_next_ino uint wraparound for
295 if (IS_ENABLED(CONFIG_64BIT))
296 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
297 __func__, MINOR(sb->s_dev));
298 sbinfo->next_ino = 1;
299 ino = sbinfo->next_ino++;
303 raw_spin_unlock(&sbinfo->stat_lock);
306 * __shmem_file_setup, one of our callers, is lock-free: it
307 * doesn't hold stat_lock in shmem_reserve_inode since
308 * max_inodes is always 0, and is called from potentially
309 * unknown contexts. As such, use a per-cpu batched allocator
310 * which doesn't require the per-sb stat_lock unless we are at
311 * the batch boundary.
313 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
314 * shmem mounts are not exposed to userspace, so we don't need
315 * to worry about things like glibc compatibility.
319 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
321 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
322 raw_spin_lock(&sbinfo->stat_lock);
323 ino = sbinfo->next_ino;
324 sbinfo->next_ino += SHMEM_INO_BATCH;
325 raw_spin_unlock(&sbinfo->stat_lock);
326 if (unlikely(is_zero_ino(ino)))
337 static void shmem_free_inode(struct super_block *sb)
339 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
340 if (sbinfo->max_inodes) {
341 raw_spin_lock(&sbinfo->stat_lock);
342 sbinfo->free_inodes++;
343 raw_spin_unlock(&sbinfo->stat_lock);
348 * shmem_recalc_inode - recalculate the block usage of an inode
349 * @inode: inode to recalc
351 * We have to calculate the free blocks since the mm can drop
352 * undirtied hole pages behind our back.
354 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
355 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
357 * It has to be called with the spinlock held.
359 static void shmem_recalc_inode(struct inode *inode)
361 struct shmem_inode_info *info = SHMEM_I(inode);
364 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
366 info->alloced -= freed;
367 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
368 shmem_inode_unacct_blocks(inode, freed);
372 bool shmem_charge(struct inode *inode, long pages)
374 struct shmem_inode_info *info = SHMEM_I(inode);
377 if (!shmem_inode_acct_block(inode, pages))
380 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
381 inode->i_mapping->nrpages += pages;
383 spin_lock_irqsave(&info->lock, flags);
384 info->alloced += pages;
385 inode->i_blocks += pages * BLOCKS_PER_PAGE;
386 shmem_recalc_inode(inode);
387 spin_unlock_irqrestore(&info->lock, flags);
392 void shmem_uncharge(struct inode *inode, long pages)
394 struct shmem_inode_info *info = SHMEM_I(inode);
397 /* nrpages adjustment done by __delete_from_page_cache() or caller */
399 spin_lock_irqsave(&info->lock, flags);
400 info->alloced -= pages;
401 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
402 shmem_recalc_inode(inode);
403 spin_unlock_irqrestore(&info->lock, flags);
405 shmem_inode_unacct_blocks(inode, pages);
409 * Replace item expected in xarray by a new item, while holding xa_lock.
411 static int shmem_replace_entry(struct address_space *mapping,
412 pgoff_t index, void *expected, void *replacement)
414 XA_STATE(xas, &mapping->i_pages, index);
417 VM_BUG_ON(!expected);
418 VM_BUG_ON(!replacement);
419 item = xas_load(&xas);
420 if (item != expected)
422 xas_store(&xas, replacement);
427 * Sometimes, before we decide whether to proceed or to fail, we must check
428 * that an entry was not already brought back from swap by a racing thread.
430 * Checking page is not enough: by the time a SwapCache page is locked, it
431 * might be reused, and again be SwapCache, using the same swap as before.
433 static bool shmem_confirm_swap(struct address_space *mapping,
434 pgoff_t index, swp_entry_t swap)
436 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
440 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
443 * disables huge pages for the mount;
445 * enables huge pages for the mount;
446 * SHMEM_HUGE_WITHIN_SIZE:
447 * only allocate huge pages if the page will be fully within i_size,
448 * also respect fadvise()/madvise() hints;
450 * only allocate huge pages if requested with fadvise()/madvise();
453 #define SHMEM_HUGE_NEVER 0
454 #define SHMEM_HUGE_ALWAYS 1
455 #define SHMEM_HUGE_WITHIN_SIZE 2
456 #define SHMEM_HUGE_ADVISE 3
460 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
463 * disables huge on shm_mnt and all mounts, for emergency use;
465 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
468 #define SHMEM_HUGE_DENY (-1)
469 #define SHMEM_HUGE_FORCE (-2)
471 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
472 /* ifdef here to avoid bloating shmem.o when not necessary */
474 static int shmem_huge __read_mostly;
476 bool shmem_huge_enabled(struct vm_area_struct *vma)
478 struct inode *inode = file_inode(vma->vm_file);
479 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
483 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
484 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
486 if (shmem_huge == SHMEM_HUGE_FORCE)
488 if (shmem_huge == SHMEM_HUGE_DENY)
490 switch (sbinfo->huge) {
491 case SHMEM_HUGE_NEVER:
493 case SHMEM_HUGE_ALWAYS:
495 case SHMEM_HUGE_WITHIN_SIZE:
496 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
497 i_size = round_up(i_size_read(inode), PAGE_SIZE);
498 if (i_size >= HPAGE_PMD_SIZE &&
499 i_size >> PAGE_SHIFT >= off)
502 case SHMEM_HUGE_ADVISE:
503 /* TODO: implement fadvise() hints */
504 return (vma->vm_flags & VM_HUGEPAGE);
511 #if defined(CONFIG_SYSFS)
512 static int shmem_parse_huge(const char *str)
514 if (!strcmp(str, "never"))
515 return SHMEM_HUGE_NEVER;
516 if (!strcmp(str, "always"))
517 return SHMEM_HUGE_ALWAYS;
518 if (!strcmp(str, "within_size"))
519 return SHMEM_HUGE_WITHIN_SIZE;
520 if (!strcmp(str, "advise"))
521 return SHMEM_HUGE_ADVISE;
522 if (!strcmp(str, "deny"))
523 return SHMEM_HUGE_DENY;
524 if (!strcmp(str, "force"))
525 return SHMEM_HUGE_FORCE;
530 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
531 static const char *shmem_format_huge(int huge)
534 case SHMEM_HUGE_NEVER:
536 case SHMEM_HUGE_ALWAYS:
538 case SHMEM_HUGE_WITHIN_SIZE:
539 return "within_size";
540 case SHMEM_HUGE_ADVISE:
542 case SHMEM_HUGE_DENY:
544 case SHMEM_HUGE_FORCE:
553 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
554 struct shrink_control *sc, unsigned long nr_to_split)
556 LIST_HEAD(list), *pos, *next;
557 LIST_HEAD(to_remove);
559 struct shmem_inode_info *info;
561 unsigned long batch = sc ? sc->nr_to_scan : 128;
562 int removed = 0, split = 0;
564 if (list_empty(&sbinfo->shrinklist))
567 spin_lock(&sbinfo->shrinklist_lock);
568 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
569 info = list_entry(pos, struct shmem_inode_info, shrinklist);
572 inode = igrab(&info->vfs_inode);
574 /* inode is about to be evicted */
576 list_del_init(&info->shrinklist);
581 /* Check if there's anything to gain */
582 if (round_up(inode->i_size, PAGE_SIZE) ==
583 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
584 list_move(&info->shrinklist, &to_remove);
589 list_move(&info->shrinklist, &list);
594 spin_unlock(&sbinfo->shrinklist_lock);
596 list_for_each_safe(pos, next, &to_remove) {
597 info = list_entry(pos, struct shmem_inode_info, shrinklist);
598 inode = &info->vfs_inode;
599 list_del_init(&info->shrinklist);
603 list_for_each_safe(pos, next, &list) {
606 info = list_entry(pos, struct shmem_inode_info, shrinklist);
607 inode = &info->vfs_inode;
609 if (nr_to_split && split >= nr_to_split)
612 page = find_get_page(inode->i_mapping,
613 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
617 /* No huge page at the end of the file: nothing to split */
618 if (!PageTransHuge(page)) {
624 * Leave the inode on the list if we failed to lock
625 * the page at this time.
627 * Waiting for the lock may lead to deadlock in the
630 if (!trylock_page(page)) {
635 ret = split_huge_page(page);
639 /* If split failed leave the inode on the list */
645 list_del_init(&info->shrinklist);
651 spin_lock(&sbinfo->shrinklist_lock);
652 list_splice_tail(&list, &sbinfo->shrinklist);
653 sbinfo->shrinklist_len -= removed;
654 spin_unlock(&sbinfo->shrinklist_lock);
659 static long shmem_unused_huge_scan(struct super_block *sb,
660 struct shrink_control *sc)
662 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
664 if (!READ_ONCE(sbinfo->shrinklist_len))
667 return shmem_unused_huge_shrink(sbinfo, sc, 0);
670 static long shmem_unused_huge_count(struct super_block *sb,
671 struct shrink_control *sc)
673 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
674 return READ_ONCE(sbinfo->shrinklist_len);
676 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
678 #define shmem_huge SHMEM_HUGE_DENY
680 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
681 struct shrink_control *sc, unsigned long nr_to_split)
685 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
687 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
689 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
690 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
691 shmem_huge != SHMEM_HUGE_DENY)
697 * Like add_to_page_cache_locked, but error if expected item has gone.
699 static int shmem_add_to_page_cache(struct page *page,
700 struct address_space *mapping,
701 pgoff_t index, void *expected, gfp_t gfp,
702 struct mm_struct *charge_mm)
704 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
706 unsigned long nr = compound_nr(page);
709 VM_BUG_ON_PAGE(PageTail(page), page);
710 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
711 VM_BUG_ON_PAGE(!PageLocked(page), page);
712 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
713 VM_BUG_ON(expected && PageTransHuge(page));
715 page_ref_add(page, nr);
716 page->mapping = mapping;
719 if (!PageSwapCache(page)) {
720 error = mem_cgroup_charge(page, charge_mm, gfp);
722 if (PageTransHuge(page)) {
723 count_vm_event(THP_FILE_FALLBACK);
724 count_vm_event(THP_FILE_FALLBACK_CHARGE);
729 cgroup_throttle_swaprate(page, gfp);
734 entry = xas_find_conflict(&xas);
735 if (entry != expected)
736 xas_set_err(&xas, -EEXIST);
737 xas_create_range(&xas);
741 xas_store(&xas, page);
746 if (PageTransHuge(page)) {
747 count_vm_event(THP_FILE_ALLOC);
748 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
750 mapping->nrpages += nr;
751 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
752 __mod_lruvec_page_state(page, NR_SHMEM, nr);
754 xas_unlock_irq(&xas);
755 } while (xas_nomem(&xas, gfp));
757 if (xas_error(&xas)) {
758 error = xas_error(&xas);
764 page->mapping = NULL;
765 page_ref_sub(page, nr);
770 * Like delete_from_page_cache, but substitutes swap for page.
772 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
774 struct address_space *mapping = page->mapping;
777 VM_BUG_ON_PAGE(PageCompound(page), page);
779 xa_lock_irq(&mapping->i_pages);
780 error = shmem_replace_entry(mapping, page->index, page, radswap);
781 page->mapping = NULL;
783 __dec_lruvec_page_state(page, NR_FILE_PAGES);
784 __dec_lruvec_page_state(page, NR_SHMEM);
785 xa_unlock_irq(&mapping->i_pages);
791 * Remove swap entry from page cache, free the swap and its page cache.
793 static int shmem_free_swap(struct address_space *mapping,
794 pgoff_t index, void *radswap)
798 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
801 free_swap_and_cache(radix_to_swp_entry(radswap));
806 * Determine (in bytes) how many of the shmem object's pages mapped by the
807 * given offsets are swapped out.
809 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
810 * as long as the inode doesn't go away and racy results are not a problem.
812 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
813 pgoff_t start, pgoff_t end)
815 XA_STATE(xas, &mapping->i_pages, start);
817 unsigned long swapped = 0;
820 xas_for_each(&xas, page, end - 1) {
821 if (xas_retry(&xas, page))
823 if (xa_is_value(page))
826 if (need_resched()) {
834 return swapped << PAGE_SHIFT;
838 * Determine (in bytes) how many of the shmem object's pages mapped by the
839 * given vma is swapped out.
841 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
842 * as long as the inode doesn't go away and racy results are not a problem.
844 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
846 struct inode *inode = file_inode(vma->vm_file);
847 struct shmem_inode_info *info = SHMEM_I(inode);
848 struct address_space *mapping = inode->i_mapping;
849 unsigned long swapped;
851 /* Be careful as we don't hold info->lock */
852 swapped = READ_ONCE(info->swapped);
855 * The easier cases are when the shmem object has nothing in swap, or
856 * the vma maps it whole. Then we can simply use the stats that we
862 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
863 return swapped << PAGE_SHIFT;
865 /* Here comes the more involved part */
866 return shmem_partial_swap_usage(mapping,
867 linear_page_index(vma, vma->vm_start),
868 linear_page_index(vma, vma->vm_end));
872 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
874 void shmem_unlock_mapping(struct address_space *mapping)
881 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
883 while (!mapping_unevictable(mapping)) {
884 if (!pagevec_lookup(&pvec, mapping, &index))
886 check_move_unevictable_pages(&pvec);
887 pagevec_release(&pvec);
893 * Check whether a hole-punch or truncation needs to split a huge page,
894 * returning true if no split was required, or the split has been successful.
896 * Eviction (or truncation to 0 size) should never need to split a huge page;
897 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
898 * head, and then succeeded to trylock on tail.
900 * A split can only succeed when there are no additional references on the
901 * huge page: so the split below relies upon find_get_entries() having stopped
902 * when it found a subpage of the huge page, without getting further references.
904 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
906 if (!PageTransCompound(page))
909 /* Just proceed to delete a huge page wholly within the range punched */
910 if (PageHead(page) &&
911 page->index >= start && page->index + HPAGE_PMD_NR <= end)
914 /* Try to split huge page, so we can truly punch the hole or truncate */
915 return split_huge_page(page) >= 0;
919 * Remove range of pages and swap entries from page cache, and free them.
920 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
922 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
925 struct address_space *mapping = inode->i_mapping;
926 struct shmem_inode_info *info = SHMEM_I(inode);
927 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
928 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
929 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
930 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
932 pgoff_t indices[PAGEVEC_SIZE];
933 long nr_swaps_freed = 0;
938 end = -1; /* unsigned, so actually very big */
940 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
941 info->fallocend = start;
945 while (index < end && find_lock_entries(mapping, index, end - 1,
947 for (i = 0; i < pagevec_count(&pvec); i++) {
948 struct page *page = pvec.pages[i];
952 if (xa_is_value(page)) {
955 nr_swaps_freed += !shmem_free_swap(mapping,
959 index += thp_nr_pages(page) - 1;
961 if (!unfalloc || !PageUptodate(page))
962 truncate_inode_page(mapping, page);
965 pagevec_remove_exceptionals(&pvec);
966 pagevec_release(&pvec);
972 struct page *page = NULL;
973 shmem_getpage(inode, start - 1, &page, SGP_READ);
975 unsigned int top = PAGE_SIZE;
980 zero_user_segment(page, partial_start, top);
981 set_page_dirty(page);
987 struct page *page = NULL;
988 shmem_getpage(inode, end, &page, SGP_READ);
990 zero_user_segment(page, 0, partial_end);
991 set_page_dirty(page);
1000 while (index < end) {
1003 if (!find_get_entries(mapping, index, end - 1, &pvec,
1005 /* If all gone or hole-punch or unfalloc, we're done */
1006 if (index == start || end != -1)
1008 /* But if truncating, restart to make sure all gone */
1012 for (i = 0; i < pagevec_count(&pvec); i++) {
1013 struct page *page = pvec.pages[i];
1016 if (xa_is_value(page)) {
1019 if (shmem_free_swap(mapping, index, page)) {
1020 /* Swap was replaced by page: retry */
1030 if (!unfalloc || !PageUptodate(page)) {
1031 if (page_mapping(page) != mapping) {
1032 /* Page was replaced by swap: retry */
1037 VM_BUG_ON_PAGE(PageWriteback(page), page);
1038 if (shmem_punch_compound(page, start, end))
1039 truncate_inode_page(mapping, page);
1040 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1041 /* Wipe the page and don't get stuck */
1042 clear_highpage(page);
1043 flush_dcache_page(page);
1044 set_page_dirty(page);
1046 round_up(start, HPAGE_PMD_NR))
1052 pagevec_remove_exceptionals(&pvec);
1053 pagevec_release(&pvec);
1057 spin_lock_irq(&info->lock);
1058 info->swapped -= nr_swaps_freed;
1059 shmem_recalc_inode(inode);
1060 spin_unlock_irq(&info->lock);
1063 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1065 shmem_undo_range(inode, lstart, lend, false);
1066 inode->i_ctime = inode->i_mtime = current_time(inode);
1068 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1070 static int shmem_getattr(struct user_namespace *mnt_userns,
1071 const struct path *path, struct kstat *stat,
1072 u32 request_mask, unsigned int query_flags)
1074 struct inode *inode = path->dentry->d_inode;
1075 struct shmem_inode_info *info = SHMEM_I(inode);
1076 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1078 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1079 spin_lock_irq(&info->lock);
1080 shmem_recalc_inode(inode);
1081 spin_unlock_irq(&info->lock);
1083 generic_fillattr(&init_user_ns, inode, stat);
1085 if (is_huge_enabled(sb_info))
1086 stat->blksize = HPAGE_PMD_SIZE;
1091 static int shmem_setattr(struct user_namespace *mnt_userns,
1092 struct dentry *dentry, struct iattr *attr)
1094 struct inode *inode = d_inode(dentry);
1095 struct shmem_inode_info *info = SHMEM_I(inode);
1098 error = setattr_prepare(&init_user_ns, dentry, attr);
1102 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1103 loff_t oldsize = inode->i_size;
1104 loff_t newsize = attr->ia_size;
1106 /* protected by i_mutex */
1107 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1108 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1111 if (newsize != oldsize) {
1112 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1116 i_size_write(inode, newsize);
1117 inode->i_ctime = inode->i_mtime = current_time(inode);
1119 if (newsize <= oldsize) {
1120 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1121 if (oldsize > holebegin)
1122 unmap_mapping_range(inode->i_mapping,
1125 shmem_truncate_range(inode,
1126 newsize, (loff_t)-1);
1127 /* unmap again to remove racily COWed private pages */
1128 if (oldsize > holebegin)
1129 unmap_mapping_range(inode->i_mapping,
1134 setattr_copy(&init_user_ns, inode, attr);
1135 if (attr->ia_valid & ATTR_MODE)
1136 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1140 static void shmem_evict_inode(struct inode *inode)
1142 struct shmem_inode_info *info = SHMEM_I(inode);
1143 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1145 if (shmem_mapping(inode->i_mapping)) {
1146 shmem_unacct_size(info->flags, inode->i_size);
1148 shmem_truncate_range(inode, 0, (loff_t)-1);
1149 if (!list_empty(&info->shrinklist)) {
1150 spin_lock(&sbinfo->shrinklist_lock);
1151 if (!list_empty(&info->shrinklist)) {
1152 list_del_init(&info->shrinklist);
1153 sbinfo->shrinklist_len--;
1155 spin_unlock(&sbinfo->shrinklist_lock);
1157 while (!list_empty(&info->swaplist)) {
1158 /* Wait while shmem_unuse() is scanning this inode... */
1159 wait_var_event(&info->stop_eviction,
1160 !atomic_read(&info->stop_eviction));
1161 mutex_lock(&shmem_swaplist_mutex);
1162 /* ...but beware of the race if we peeked too early */
1163 if (!atomic_read(&info->stop_eviction))
1164 list_del_init(&info->swaplist);
1165 mutex_unlock(&shmem_swaplist_mutex);
1169 simple_xattrs_free(&info->xattrs);
1170 WARN_ON(inode->i_blocks);
1171 shmem_free_inode(inode->i_sb);
1175 static int shmem_find_swap_entries(struct address_space *mapping,
1176 pgoff_t start, unsigned int nr_entries,
1177 struct page **entries, pgoff_t *indices,
1178 unsigned int type, bool frontswap)
1180 XA_STATE(xas, &mapping->i_pages, start);
1183 unsigned int ret = 0;
1189 xas_for_each(&xas, page, ULONG_MAX) {
1190 if (xas_retry(&xas, page))
1193 if (!xa_is_value(page))
1196 entry = radix_to_swp_entry(page);
1197 if (swp_type(entry) != type)
1200 !frontswap_test(swap_info[type], swp_offset(entry)))
1203 indices[ret] = xas.xa_index;
1204 entries[ret] = page;
1206 if (need_resched()) {
1210 if (++ret == nr_entries)
1219 * Move the swapped pages for an inode to page cache. Returns the count
1220 * of pages swapped in, or the error in case of failure.
1222 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1228 struct address_space *mapping = inode->i_mapping;
1230 for (i = 0; i < pvec.nr; i++) {
1231 struct page *page = pvec.pages[i];
1233 if (!xa_is_value(page))
1235 error = shmem_swapin_page(inode, indices[i],
1237 mapping_gfp_mask(mapping),
1244 if (error == -ENOMEM)
1248 return error ? error : ret;
1252 * If swap found in inode, free it and move page from swapcache to filecache.
1254 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1255 bool frontswap, unsigned long *fs_pages_to_unuse)
1257 struct address_space *mapping = inode->i_mapping;
1259 struct pagevec pvec;
1260 pgoff_t indices[PAGEVEC_SIZE];
1261 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1264 pagevec_init(&pvec);
1266 unsigned int nr_entries = PAGEVEC_SIZE;
1268 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1269 nr_entries = *fs_pages_to_unuse;
1271 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1272 pvec.pages, indices,
1279 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1283 if (frontswap_partial) {
1284 *fs_pages_to_unuse -= ret;
1285 if (*fs_pages_to_unuse == 0) {
1286 ret = FRONTSWAP_PAGES_UNUSED;
1291 start = indices[pvec.nr - 1];
1298 * Read all the shared memory data that resides in the swap
1299 * device 'type' back into memory, so the swap device can be
1302 int shmem_unuse(unsigned int type, bool frontswap,
1303 unsigned long *fs_pages_to_unuse)
1305 struct shmem_inode_info *info, *next;
1308 if (list_empty(&shmem_swaplist))
1311 mutex_lock(&shmem_swaplist_mutex);
1312 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1313 if (!info->swapped) {
1314 list_del_init(&info->swaplist);
1318 * Drop the swaplist mutex while searching the inode for swap;
1319 * but before doing so, make sure shmem_evict_inode() will not
1320 * remove placeholder inode from swaplist, nor let it be freed
1321 * (igrab() would protect from unlink, but not from unmount).
1323 atomic_inc(&info->stop_eviction);
1324 mutex_unlock(&shmem_swaplist_mutex);
1326 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1330 mutex_lock(&shmem_swaplist_mutex);
1331 next = list_next_entry(info, swaplist);
1333 list_del_init(&info->swaplist);
1334 if (atomic_dec_and_test(&info->stop_eviction))
1335 wake_up_var(&info->stop_eviction);
1339 mutex_unlock(&shmem_swaplist_mutex);
1345 * Move the page from the page cache to the swap cache.
1347 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1349 struct shmem_inode_info *info;
1350 struct address_space *mapping;
1351 struct inode *inode;
1355 VM_BUG_ON_PAGE(PageCompound(page), page);
1356 BUG_ON(!PageLocked(page));
1357 mapping = page->mapping;
1358 index = page->index;
1359 inode = mapping->host;
1360 info = SHMEM_I(inode);
1361 if (info->flags & VM_LOCKED)
1363 if (!total_swap_pages)
1367 * Our capabilities prevent regular writeback or sync from ever calling
1368 * shmem_writepage; but a stacking filesystem might use ->writepage of
1369 * its underlying filesystem, in which case tmpfs should write out to
1370 * swap only in response to memory pressure, and not for the writeback
1373 if (!wbc->for_reclaim) {
1374 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1379 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1380 * value into swapfile.c, the only way we can correctly account for a
1381 * fallocated page arriving here is now to initialize it and write it.
1383 * That's okay for a page already fallocated earlier, but if we have
1384 * not yet completed the fallocation, then (a) we want to keep track
1385 * of this page in case we have to undo it, and (b) it may not be a
1386 * good idea to continue anyway, once we're pushing into swap. So
1387 * reactivate the page, and let shmem_fallocate() quit when too many.
1389 if (!PageUptodate(page)) {
1390 if (inode->i_private) {
1391 struct shmem_falloc *shmem_falloc;
1392 spin_lock(&inode->i_lock);
1393 shmem_falloc = inode->i_private;
1395 !shmem_falloc->waitq &&
1396 index >= shmem_falloc->start &&
1397 index < shmem_falloc->next)
1398 shmem_falloc->nr_unswapped++;
1400 shmem_falloc = NULL;
1401 spin_unlock(&inode->i_lock);
1405 clear_highpage(page);
1406 flush_dcache_page(page);
1407 SetPageUptodate(page);
1410 swap = get_swap_page(page);
1415 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1416 * if it's not already there. Do it now before the page is
1417 * moved to swap cache, when its pagelock no longer protects
1418 * the inode from eviction. But don't unlock the mutex until
1419 * we've incremented swapped, because shmem_unuse_inode() will
1420 * prune a !swapped inode from the swaplist under this mutex.
1422 mutex_lock(&shmem_swaplist_mutex);
1423 if (list_empty(&info->swaplist))
1424 list_add(&info->swaplist, &shmem_swaplist);
1426 if (add_to_swap_cache(page, swap,
1427 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1429 spin_lock_irq(&info->lock);
1430 shmem_recalc_inode(inode);
1432 spin_unlock_irq(&info->lock);
1434 swap_shmem_alloc(swap);
1435 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1437 mutex_unlock(&shmem_swaplist_mutex);
1438 BUG_ON(page_mapped(page));
1439 swap_writepage(page, wbc);
1443 mutex_unlock(&shmem_swaplist_mutex);
1444 put_swap_page(page, swap);
1446 set_page_dirty(page);
1447 if (wbc->for_reclaim)
1448 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1453 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1454 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1458 if (!mpol || mpol->mode == MPOL_DEFAULT)
1459 return; /* show nothing */
1461 mpol_to_str(buffer, sizeof(buffer), mpol);
1463 seq_printf(seq, ",mpol=%s", buffer);
1466 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1468 struct mempolicy *mpol = NULL;
1470 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1471 mpol = sbinfo->mpol;
1473 raw_spin_unlock(&sbinfo->stat_lock);
1477 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1478 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1481 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1485 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1487 #define vm_policy vm_private_data
1490 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1491 struct shmem_inode_info *info, pgoff_t index)
1493 /* Create a pseudo vma that just contains the policy */
1494 vma_init(vma, NULL);
1495 /* Bias interleave by inode number to distribute better across nodes */
1496 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1497 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1500 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1502 /* Drop reference taken by mpol_shared_policy_lookup() */
1503 mpol_cond_put(vma->vm_policy);
1506 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1507 struct shmem_inode_info *info, pgoff_t index)
1509 struct vm_area_struct pvma;
1511 struct vm_fault vmf = {
1515 shmem_pseudo_vma_init(&pvma, info, index);
1516 page = swap_cluster_readahead(swap, gfp, &vmf);
1517 shmem_pseudo_vma_destroy(&pvma);
1523 * Make sure huge_gfp is always more limited than limit_gfp.
1524 * Some of the flags set permissions, while others set limitations.
1526 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1528 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1529 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1530 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1531 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1533 /* Allow allocations only from the originally specified zones. */
1534 result |= zoneflags;
1537 * Minimize the result gfp by taking the union with the deny flags,
1538 * and the intersection of the allow flags.
1540 result |= (limit_gfp & denyflags);
1541 result |= (huge_gfp & limit_gfp) & allowflags;
1546 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1547 struct shmem_inode_info *info, pgoff_t index)
1549 struct vm_area_struct pvma;
1550 struct address_space *mapping = info->vfs_inode.i_mapping;
1554 hindex = round_down(index, HPAGE_PMD_NR);
1555 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1559 shmem_pseudo_vma_init(&pvma, info, hindex);
1560 page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(),
1562 shmem_pseudo_vma_destroy(&pvma);
1564 prep_transhuge_page(page);
1566 count_vm_event(THP_FILE_FALLBACK);
1570 static struct page *shmem_alloc_page(gfp_t gfp,
1571 struct shmem_inode_info *info, pgoff_t index)
1573 struct vm_area_struct pvma;
1576 shmem_pseudo_vma_init(&pvma, info, index);
1577 page = alloc_page_vma(gfp, &pvma, 0);
1578 shmem_pseudo_vma_destroy(&pvma);
1583 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1584 struct inode *inode,
1585 pgoff_t index, bool huge)
1587 struct shmem_inode_info *info = SHMEM_I(inode);
1592 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1594 nr = huge ? HPAGE_PMD_NR : 1;
1596 if (!shmem_inode_acct_block(inode, nr))
1600 page = shmem_alloc_hugepage(gfp, info, index);
1602 page = shmem_alloc_page(gfp, info, index);
1604 __SetPageLocked(page);
1605 __SetPageSwapBacked(page);
1610 shmem_inode_unacct_blocks(inode, nr);
1612 return ERR_PTR(err);
1616 * When a page is moved from swapcache to shmem filecache (either by the
1617 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1618 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1619 * ignorance of the mapping it belongs to. If that mapping has special
1620 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1621 * we may need to copy to a suitable page before moving to filecache.
1623 * In a future release, this may well be extended to respect cpuset and
1624 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1625 * but for now it is a simple matter of zone.
1627 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1629 return page_zonenum(page) > gfp_zone(gfp);
1632 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1633 struct shmem_inode_info *info, pgoff_t index)
1635 struct page *oldpage, *newpage;
1636 struct address_space *swap_mapping;
1642 entry.val = page_private(oldpage);
1643 swap_index = swp_offset(entry);
1644 swap_mapping = page_mapping(oldpage);
1647 * We have arrived here because our zones are constrained, so don't
1648 * limit chance of success by further cpuset and node constraints.
1650 gfp &= ~GFP_CONSTRAINT_MASK;
1651 newpage = shmem_alloc_page(gfp, info, index);
1656 copy_highpage(newpage, oldpage);
1657 flush_dcache_page(newpage);
1659 __SetPageLocked(newpage);
1660 __SetPageSwapBacked(newpage);
1661 SetPageUptodate(newpage);
1662 set_page_private(newpage, entry.val);
1663 SetPageSwapCache(newpage);
1666 * Our caller will very soon move newpage out of swapcache, but it's
1667 * a nice clean interface for us to replace oldpage by newpage there.
1669 xa_lock_irq(&swap_mapping->i_pages);
1670 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1672 mem_cgroup_migrate(oldpage, newpage);
1673 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1674 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1676 xa_unlock_irq(&swap_mapping->i_pages);
1678 if (unlikely(error)) {
1680 * Is this possible? I think not, now that our callers check
1681 * both PageSwapCache and page_private after getting page lock;
1682 * but be defensive. Reverse old to newpage for clear and free.
1686 lru_cache_add(newpage);
1690 ClearPageSwapCache(oldpage);
1691 set_page_private(oldpage, 0);
1693 unlock_page(oldpage);
1700 * Swap in the page pointed to by *pagep.
1701 * Caller has to make sure that *pagep contains a valid swapped page.
1702 * Returns 0 and the page in pagep if success. On failure, returns the
1703 * error code and NULL in *pagep.
1705 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1706 struct page **pagep, enum sgp_type sgp,
1707 gfp_t gfp, struct vm_area_struct *vma,
1708 vm_fault_t *fault_type)
1710 struct address_space *mapping = inode->i_mapping;
1711 struct shmem_inode_info *info = SHMEM_I(inode);
1712 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1717 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1718 swap = radix_to_swp_entry(*pagep);
1721 /* Look it up and read it in.. */
1722 page = lookup_swap_cache(swap, NULL, 0);
1724 /* Or update major stats only when swapin succeeds?? */
1726 *fault_type |= VM_FAULT_MAJOR;
1727 count_vm_event(PGMAJFAULT);
1728 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1730 /* Here we actually start the io */
1731 page = shmem_swapin(swap, gfp, info, index);
1738 /* We have to do this with page locked to prevent races */
1740 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1741 !shmem_confirm_swap(mapping, index, swap)) {
1745 if (!PageUptodate(page)) {
1749 wait_on_page_writeback(page);
1752 * Some architectures may have to restore extra metadata to the
1753 * physical page after reading from swap.
1755 arch_swap_restore(swap, page);
1757 if (shmem_should_replace_page(page, gfp)) {
1758 error = shmem_replace_page(&page, gfp, info, index);
1763 error = shmem_add_to_page_cache(page, mapping, index,
1764 swp_to_radix_entry(swap), gfp,
1769 spin_lock_irq(&info->lock);
1771 shmem_recalc_inode(inode);
1772 spin_unlock_irq(&info->lock);
1774 if (sgp == SGP_WRITE)
1775 mark_page_accessed(page);
1777 delete_from_swap_cache(page);
1778 set_page_dirty(page);
1784 if (!shmem_confirm_swap(mapping, index, swap))
1796 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1798 * If we allocate a new one we do not mark it dirty. That's up to the
1799 * vm. If we swap it in we mark it dirty since we also free the swap
1800 * entry since a page cannot live in both the swap and page cache.
1802 * vma, vmf, and fault_type are only supplied by shmem_fault:
1803 * otherwise they are NULL.
1805 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1806 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1807 struct vm_area_struct *vma, struct vm_fault *vmf,
1808 vm_fault_t *fault_type)
1810 struct address_space *mapping = inode->i_mapping;
1811 struct shmem_inode_info *info = SHMEM_I(inode);
1812 struct shmem_sb_info *sbinfo;
1813 struct mm_struct *charge_mm;
1815 enum sgp_type sgp_huge = sgp;
1816 pgoff_t hindex = index;
1822 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1824 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1827 if (sgp <= SGP_CACHE &&
1828 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1832 sbinfo = SHMEM_SB(inode->i_sb);
1833 charge_mm = vma ? vma->vm_mm : NULL;
1835 page = pagecache_get_page(mapping, index,
1836 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1838 if (page && vma && userfaultfd_minor(vma)) {
1839 if (!xa_is_value(page)) {
1843 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1847 if (xa_is_value(page)) {
1848 error = shmem_swapin_page(inode, index, &page,
1849 sgp, gfp, vma, fault_type);
1850 if (error == -EEXIST)
1858 hindex = page->index;
1859 if (sgp == SGP_WRITE)
1860 mark_page_accessed(page);
1861 if (PageUptodate(page))
1863 /* fallocated page */
1864 if (sgp != SGP_READ)
1871 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1872 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1875 if (sgp == SGP_READ)
1877 if (sgp == SGP_NOALLOC)
1881 * Fast cache lookup and swap lookup did not find it: allocate.
1884 if (vma && userfaultfd_missing(vma)) {
1885 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1889 /* shmem_symlink() */
1890 if (!shmem_mapping(mapping))
1892 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1894 if (shmem_huge == SHMEM_HUGE_FORCE)
1896 switch (sbinfo->huge) {
1897 case SHMEM_HUGE_NEVER:
1899 case SHMEM_HUGE_WITHIN_SIZE: {
1903 off = round_up(index, HPAGE_PMD_NR);
1904 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1905 if (i_size >= HPAGE_PMD_SIZE &&
1906 i_size >> PAGE_SHIFT >= off)
1911 case SHMEM_HUGE_ADVISE:
1912 if (sgp_huge == SGP_HUGE)
1914 /* TODO: implement fadvise() hints */
1919 huge_gfp = vma_thp_gfp_mask(vma);
1920 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1921 page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true);
1924 page = shmem_alloc_and_acct_page(gfp, inode,
1930 error = PTR_ERR(page);
1932 if (error != -ENOSPC)
1935 * Try to reclaim some space by splitting a huge page
1936 * beyond i_size on the filesystem.
1941 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1942 if (ret == SHRINK_STOP)
1950 if (PageTransHuge(page))
1951 hindex = round_down(index, HPAGE_PMD_NR);
1955 if (sgp == SGP_WRITE)
1956 __SetPageReferenced(page);
1958 error = shmem_add_to_page_cache(page, mapping, hindex,
1959 NULL, gfp & GFP_RECLAIM_MASK,
1963 lru_cache_add(page);
1965 spin_lock_irq(&info->lock);
1966 info->alloced += compound_nr(page);
1967 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1968 shmem_recalc_inode(inode);
1969 spin_unlock_irq(&info->lock);
1972 if (PageTransHuge(page) &&
1973 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1974 hindex + HPAGE_PMD_NR - 1) {
1976 * Part of the huge page is beyond i_size: subject
1977 * to shrink under memory pressure.
1979 spin_lock(&sbinfo->shrinklist_lock);
1981 * _careful to defend against unlocked access to
1982 * ->shrink_list in shmem_unused_huge_shrink()
1984 if (list_empty_careful(&info->shrinklist)) {
1985 list_add_tail(&info->shrinklist,
1986 &sbinfo->shrinklist);
1987 sbinfo->shrinklist_len++;
1989 spin_unlock(&sbinfo->shrinklist_lock);
1993 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1995 if (sgp == SGP_FALLOC)
1999 * Let SGP_WRITE caller clear ends if write does not fill page;
2000 * but SGP_FALLOC on a page fallocated earlier must initialize
2001 * it now, lest undo on failure cancel our earlier guarantee.
2003 if (sgp != SGP_WRITE && !PageUptodate(page)) {
2006 for (i = 0; i < compound_nr(page); i++) {
2007 clear_highpage(page + i);
2008 flush_dcache_page(page + i);
2010 SetPageUptodate(page);
2013 /* Perhaps the file has been truncated since we checked */
2014 if (sgp <= SGP_CACHE &&
2015 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2017 ClearPageDirty(page);
2018 delete_from_page_cache(page);
2019 spin_lock_irq(&info->lock);
2020 shmem_recalc_inode(inode);
2021 spin_unlock_irq(&info->lock);
2027 *pagep = page + index - hindex;
2034 shmem_inode_unacct_blocks(inode, compound_nr(page));
2036 if (PageTransHuge(page)) {
2046 if (error == -ENOSPC && !once++) {
2047 spin_lock_irq(&info->lock);
2048 shmem_recalc_inode(inode);
2049 spin_unlock_irq(&info->lock);
2052 if (error == -EEXIST)
2058 * This is like autoremove_wake_function, but it removes the wait queue
2059 * entry unconditionally - even if something else had already woken the
2062 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2064 int ret = default_wake_function(wait, mode, sync, key);
2065 list_del_init(&wait->entry);
2069 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2071 struct vm_area_struct *vma = vmf->vma;
2072 struct inode *inode = file_inode(vma->vm_file);
2073 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2076 vm_fault_t ret = VM_FAULT_LOCKED;
2079 * Trinity finds that probing a hole which tmpfs is punching can
2080 * prevent the hole-punch from ever completing: which in turn
2081 * locks writers out with its hold on i_mutex. So refrain from
2082 * faulting pages into the hole while it's being punched. Although
2083 * shmem_undo_range() does remove the additions, it may be unable to
2084 * keep up, as each new page needs its own unmap_mapping_range() call,
2085 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2087 * It does not matter if we sometimes reach this check just before the
2088 * hole-punch begins, so that one fault then races with the punch:
2089 * we just need to make racing faults a rare case.
2091 * The implementation below would be much simpler if we just used a
2092 * standard mutex or completion: but we cannot take i_mutex in fault,
2093 * and bloating every shmem inode for this unlikely case would be sad.
2095 if (unlikely(inode->i_private)) {
2096 struct shmem_falloc *shmem_falloc;
2098 spin_lock(&inode->i_lock);
2099 shmem_falloc = inode->i_private;
2101 shmem_falloc->waitq &&
2102 vmf->pgoff >= shmem_falloc->start &&
2103 vmf->pgoff < shmem_falloc->next) {
2105 wait_queue_head_t *shmem_falloc_waitq;
2106 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2108 ret = VM_FAULT_NOPAGE;
2109 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2111 ret = VM_FAULT_RETRY;
2113 shmem_falloc_waitq = shmem_falloc->waitq;
2114 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2115 TASK_UNINTERRUPTIBLE);
2116 spin_unlock(&inode->i_lock);
2120 * shmem_falloc_waitq points into the shmem_fallocate()
2121 * stack of the hole-punching task: shmem_falloc_waitq
2122 * is usually invalid by the time we reach here, but
2123 * finish_wait() does not dereference it in that case;
2124 * though i_lock needed lest racing with wake_up_all().
2126 spin_lock(&inode->i_lock);
2127 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2128 spin_unlock(&inode->i_lock);
2134 spin_unlock(&inode->i_lock);
2139 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2140 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2142 else if (vma->vm_flags & VM_HUGEPAGE)
2145 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2146 gfp, vma, vmf, &ret);
2148 return vmf_error(err);
2152 unsigned long shmem_get_unmapped_area(struct file *file,
2153 unsigned long uaddr, unsigned long len,
2154 unsigned long pgoff, unsigned long flags)
2156 unsigned long (*get_area)(struct file *,
2157 unsigned long, unsigned long, unsigned long, unsigned long);
2159 unsigned long offset;
2160 unsigned long inflated_len;
2161 unsigned long inflated_addr;
2162 unsigned long inflated_offset;
2164 if (len > TASK_SIZE)
2167 get_area = current->mm->get_unmapped_area;
2168 addr = get_area(file, uaddr, len, pgoff, flags);
2170 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2172 if (IS_ERR_VALUE(addr))
2174 if (addr & ~PAGE_MASK)
2176 if (addr > TASK_SIZE - len)
2179 if (shmem_huge == SHMEM_HUGE_DENY)
2181 if (len < HPAGE_PMD_SIZE)
2183 if (flags & MAP_FIXED)
2186 * Our priority is to support MAP_SHARED mapped hugely;
2187 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2188 * But if caller specified an address hint and we allocated area there
2189 * successfully, respect that as before.
2194 if (shmem_huge != SHMEM_HUGE_FORCE) {
2195 struct super_block *sb;
2198 VM_BUG_ON(file->f_op != &shmem_file_operations);
2199 sb = file_inode(file)->i_sb;
2202 * Called directly from mm/mmap.c, or drivers/char/mem.c
2203 * for "/dev/zero", to create a shared anonymous object.
2205 if (IS_ERR(shm_mnt))
2207 sb = shm_mnt->mnt_sb;
2209 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2213 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2214 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2216 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2219 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2220 if (inflated_len > TASK_SIZE)
2222 if (inflated_len < len)
2225 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2226 if (IS_ERR_VALUE(inflated_addr))
2228 if (inflated_addr & ~PAGE_MASK)
2231 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2232 inflated_addr += offset - inflated_offset;
2233 if (inflated_offset > offset)
2234 inflated_addr += HPAGE_PMD_SIZE;
2236 if (inflated_addr > TASK_SIZE - len)
2238 return inflated_addr;
2242 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2244 struct inode *inode = file_inode(vma->vm_file);
2245 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2248 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2251 struct inode *inode = file_inode(vma->vm_file);
2254 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2255 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2259 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2261 struct inode *inode = file_inode(file);
2262 struct shmem_inode_info *info = SHMEM_I(inode);
2263 int retval = -ENOMEM;
2266 * What serializes the accesses to info->flags?
2267 * ipc_lock_object() when called from shmctl_do_lock(),
2268 * no serialization needed when called from shm_destroy().
2270 if (lock && !(info->flags & VM_LOCKED)) {
2271 if (!user_shm_lock(inode->i_size, ucounts))
2273 info->flags |= VM_LOCKED;
2274 mapping_set_unevictable(file->f_mapping);
2276 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2277 user_shm_unlock(inode->i_size, ucounts);
2278 info->flags &= ~VM_LOCKED;
2279 mapping_clear_unevictable(file->f_mapping);
2287 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2289 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2292 ret = seal_check_future_write(info->seals, vma);
2296 /* arm64 - allow memory tagging on RAM-based files */
2297 vma->vm_flags |= VM_MTE_ALLOWED;
2299 file_accessed(file);
2300 vma->vm_ops = &shmem_vm_ops;
2301 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2302 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2303 (vma->vm_end & HPAGE_PMD_MASK)) {
2304 khugepaged_enter(vma, vma->vm_flags);
2309 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2310 umode_t mode, dev_t dev, unsigned long flags)
2312 struct inode *inode;
2313 struct shmem_inode_info *info;
2314 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2317 if (shmem_reserve_inode(sb, &ino))
2320 inode = new_inode(sb);
2323 inode_init_owner(&init_user_ns, inode, dir, mode);
2324 inode->i_blocks = 0;
2325 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2326 inode->i_generation = prandom_u32();
2327 info = SHMEM_I(inode);
2328 memset(info, 0, (char *)inode - (char *)info);
2329 spin_lock_init(&info->lock);
2330 atomic_set(&info->stop_eviction, 0);
2331 info->seals = F_SEAL_SEAL;
2332 info->flags = flags & VM_NORESERVE;
2333 INIT_LIST_HEAD(&info->shrinklist);
2334 INIT_LIST_HEAD(&info->swaplist);
2335 simple_xattrs_init(&info->xattrs);
2336 cache_no_acl(inode);
2338 switch (mode & S_IFMT) {
2340 inode->i_op = &shmem_special_inode_operations;
2341 init_special_inode(inode, mode, dev);
2344 inode->i_mapping->a_ops = &shmem_aops;
2345 inode->i_op = &shmem_inode_operations;
2346 inode->i_fop = &shmem_file_operations;
2347 mpol_shared_policy_init(&info->policy,
2348 shmem_get_sbmpol(sbinfo));
2352 /* Some things misbehave if size == 0 on a directory */
2353 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2354 inode->i_op = &shmem_dir_inode_operations;
2355 inode->i_fop = &simple_dir_operations;
2359 * Must not load anything in the rbtree,
2360 * mpol_free_shared_policy will not be called.
2362 mpol_shared_policy_init(&info->policy, NULL);
2366 lockdep_annotate_inode_mutex_key(inode);
2368 shmem_free_inode(sb);
2372 #ifdef CONFIG_USERFAULTFD
2373 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2375 struct vm_area_struct *dst_vma,
2376 unsigned long dst_addr,
2377 unsigned long src_addr,
2379 struct page **pagep)
2381 struct inode *inode = file_inode(dst_vma->vm_file);
2382 struct shmem_inode_info *info = SHMEM_I(inode);
2383 struct address_space *mapping = inode->i_mapping;
2384 gfp_t gfp = mapping_gfp_mask(mapping);
2385 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2391 if (!shmem_inode_acct_block(inode, 1)) {
2393 * We may have got a page, returned -ENOENT triggering a retry,
2394 * and now we find ourselves with -ENOMEM. Release the page, to
2395 * avoid a BUG_ON in our caller.
2397 if (unlikely(*pagep)) {
2406 page = shmem_alloc_page(gfp, info, pgoff);
2408 goto out_unacct_blocks;
2410 if (!zeropage) { /* COPY */
2411 page_kaddr = kmap_atomic(page);
2412 ret = copy_from_user(page_kaddr,
2413 (const void __user *)src_addr,
2415 kunmap_atomic(page_kaddr);
2417 /* fallback to copy_from_user outside mmap_lock */
2418 if (unlikely(ret)) {
2421 /* don't free the page */
2422 goto out_unacct_blocks;
2424 } else { /* ZEROPAGE */
2425 clear_highpage(page);
2432 VM_BUG_ON(PageLocked(page));
2433 VM_BUG_ON(PageSwapBacked(page));
2434 __SetPageLocked(page);
2435 __SetPageSwapBacked(page);
2436 __SetPageUptodate(page);
2439 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2440 if (unlikely(pgoff >= max_off))
2443 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2444 gfp & GFP_RECLAIM_MASK, dst_mm);
2448 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2451 goto out_delete_from_cache;
2453 spin_lock_irq(&info->lock);
2455 inode->i_blocks += BLOCKS_PER_PAGE;
2456 shmem_recalc_inode(inode);
2457 spin_unlock_irq(&info->lock);
2462 out_delete_from_cache:
2463 delete_from_page_cache(page);
2468 shmem_inode_unacct_blocks(inode, 1);
2471 #endif /* CONFIG_USERFAULTFD */
2474 static const struct inode_operations shmem_symlink_inode_operations;
2475 static const struct inode_operations shmem_short_symlink_operations;
2477 #ifdef CONFIG_TMPFS_XATTR
2478 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2480 #define shmem_initxattrs NULL
2484 shmem_write_begin(struct file *file, struct address_space *mapping,
2485 loff_t pos, unsigned len, unsigned flags,
2486 struct page **pagep, void **fsdata)
2488 struct inode *inode = mapping->host;
2489 struct shmem_inode_info *info = SHMEM_I(inode);
2490 pgoff_t index = pos >> PAGE_SHIFT;
2492 /* i_mutex is held by caller */
2493 if (unlikely(info->seals & (F_SEAL_GROW |
2494 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2495 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2497 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2501 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2505 shmem_write_end(struct file *file, struct address_space *mapping,
2506 loff_t pos, unsigned len, unsigned copied,
2507 struct page *page, void *fsdata)
2509 struct inode *inode = mapping->host;
2511 if (pos + copied > inode->i_size)
2512 i_size_write(inode, pos + copied);
2514 if (!PageUptodate(page)) {
2515 struct page *head = compound_head(page);
2516 if (PageTransCompound(page)) {
2519 for (i = 0; i < HPAGE_PMD_NR; i++) {
2520 if (head + i == page)
2522 clear_highpage(head + i);
2523 flush_dcache_page(head + i);
2526 if (copied < PAGE_SIZE) {
2527 unsigned from = pos & (PAGE_SIZE - 1);
2528 zero_user_segments(page, 0, from,
2529 from + copied, PAGE_SIZE);
2531 SetPageUptodate(head);
2533 set_page_dirty(page);
2540 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2542 struct file *file = iocb->ki_filp;
2543 struct inode *inode = file_inode(file);
2544 struct address_space *mapping = inode->i_mapping;
2546 unsigned long offset;
2547 enum sgp_type sgp = SGP_READ;
2550 loff_t *ppos = &iocb->ki_pos;
2553 * Might this read be for a stacking filesystem? Then when reading
2554 * holes of a sparse file, we actually need to allocate those pages,
2555 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2557 if (!iter_is_iovec(to))
2560 index = *ppos >> PAGE_SHIFT;
2561 offset = *ppos & ~PAGE_MASK;
2564 struct page *page = NULL;
2566 unsigned long nr, ret;
2567 loff_t i_size = i_size_read(inode);
2569 end_index = i_size >> PAGE_SHIFT;
2570 if (index > end_index)
2572 if (index == end_index) {
2573 nr = i_size & ~PAGE_MASK;
2578 error = shmem_getpage(inode, index, &page, sgp);
2580 if (error == -EINVAL)
2585 if (sgp == SGP_CACHE)
2586 set_page_dirty(page);
2591 * We must evaluate after, since reads (unlike writes)
2592 * are called without i_mutex protection against truncate
2595 i_size = i_size_read(inode);
2596 end_index = i_size >> PAGE_SHIFT;
2597 if (index == end_index) {
2598 nr = i_size & ~PAGE_MASK;
2609 * If users can be writing to this page using arbitrary
2610 * virtual addresses, take care about potential aliasing
2611 * before reading the page on the kernel side.
2613 if (mapping_writably_mapped(mapping))
2614 flush_dcache_page(page);
2616 * Mark the page accessed if we read the beginning.
2619 mark_page_accessed(page);
2621 page = ZERO_PAGE(0);
2626 * Ok, we have the page, and it's up-to-date, so
2627 * now we can copy it to user space...
2629 ret = copy_page_to_iter(page, offset, nr, to);
2632 index += offset >> PAGE_SHIFT;
2633 offset &= ~PAGE_MASK;
2636 if (!iov_iter_count(to))
2645 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2646 file_accessed(file);
2647 return retval ? retval : error;
2650 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2652 struct address_space *mapping = file->f_mapping;
2653 struct inode *inode = mapping->host;
2655 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2656 return generic_file_llseek_size(file, offset, whence,
2657 MAX_LFS_FILESIZE, i_size_read(inode));
2662 /* We're holding i_mutex so we can access i_size directly */
2663 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2665 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2666 inode_unlock(inode);
2670 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2673 struct inode *inode = file_inode(file);
2674 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2675 struct shmem_inode_info *info = SHMEM_I(inode);
2676 struct shmem_falloc shmem_falloc;
2677 pgoff_t start, index, end, undo_fallocend;
2680 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2685 if (mode & FALLOC_FL_PUNCH_HOLE) {
2686 struct address_space *mapping = file->f_mapping;
2687 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2688 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2689 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2691 /* protected by i_mutex */
2692 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2697 shmem_falloc.waitq = &shmem_falloc_waitq;
2698 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2699 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2700 spin_lock(&inode->i_lock);
2701 inode->i_private = &shmem_falloc;
2702 spin_unlock(&inode->i_lock);
2704 if ((u64)unmap_end > (u64)unmap_start)
2705 unmap_mapping_range(mapping, unmap_start,
2706 1 + unmap_end - unmap_start, 0);
2707 shmem_truncate_range(inode, offset, offset + len - 1);
2708 /* No need to unmap again: hole-punching leaves COWed pages */
2710 spin_lock(&inode->i_lock);
2711 inode->i_private = NULL;
2712 wake_up_all(&shmem_falloc_waitq);
2713 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2714 spin_unlock(&inode->i_lock);
2719 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2720 error = inode_newsize_ok(inode, offset + len);
2724 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2729 start = offset >> PAGE_SHIFT;
2730 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2731 /* Try to avoid a swapstorm if len is impossible to satisfy */
2732 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2737 shmem_falloc.waitq = NULL;
2738 shmem_falloc.start = start;
2739 shmem_falloc.next = start;
2740 shmem_falloc.nr_falloced = 0;
2741 shmem_falloc.nr_unswapped = 0;
2742 spin_lock(&inode->i_lock);
2743 inode->i_private = &shmem_falloc;
2744 spin_unlock(&inode->i_lock);
2747 * info->fallocend is only relevant when huge pages might be
2748 * involved: to prevent split_huge_page() freeing fallocated
2749 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2751 undo_fallocend = info->fallocend;
2752 if (info->fallocend < end)
2753 info->fallocend = end;
2755 for (index = start; index < end; ) {
2759 * Good, the fallocate(2) manpage permits EINTR: we may have
2760 * been interrupted because we are using up too much memory.
2762 if (signal_pending(current))
2764 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2767 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2769 info->fallocend = undo_fallocend;
2770 /* Remove the !PageUptodate pages we added */
2771 if (index > start) {
2772 shmem_undo_range(inode,
2773 (loff_t)start << PAGE_SHIFT,
2774 ((loff_t)index << PAGE_SHIFT) - 1, true);
2781 * Here is a more important optimization than it appears:
2782 * a second SGP_FALLOC on the same huge page will clear it,
2783 * making it PageUptodate and un-undoable if we fail later.
2785 if (PageTransCompound(page)) {
2786 index = round_up(index, HPAGE_PMD_NR);
2787 /* Beware 32-bit wraparound */
2793 * Inform shmem_writepage() how far we have reached.
2794 * No need for lock or barrier: we have the page lock.
2796 if (!PageUptodate(page))
2797 shmem_falloc.nr_falloced += index - shmem_falloc.next;
2798 shmem_falloc.next = index;
2801 * If !PageUptodate, leave it that way so that freeable pages
2802 * can be recognized if we need to rollback on error later.
2803 * But set_page_dirty so that memory pressure will swap rather
2804 * than free the pages we are allocating (and SGP_CACHE pages
2805 * might still be clean: we now need to mark those dirty too).
2807 set_page_dirty(page);
2813 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2814 i_size_write(inode, offset + len);
2815 inode->i_ctime = current_time(inode);
2817 spin_lock(&inode->i_lock);
2818 inode->i_private = NULL;
2819 spin_unlock(&inode->i_lock);
2821 inode_unlock(inode);
2825 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2827 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2829 buf->f_type = TMPFS_MAGIC;
2830 buf->f_bsize = PAGE_SIZE;
2831 buf->f_namelen = NAME_MAX;
2832 if (sbinfo->max_blocks) {
2833 buf->f_blocks = sbinfo->max_blocks;
2835 buf->f_bfree = sbinfo->max_blocks -
2836 percpu_counter_sum(&sbinfo->used_blocks);
2838 if (sbinfo->max_inodes) {
2839 buf->f_files = sbinfo->max_inodes;
2840 buf->f_ffree = sbinfo->free_inodes;
2842 /* else leave those fields 0 like simple_statfs */
2844 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2850 * File creation. Allocate an inode, and we're done..
2853 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2854 struct dentry *dentry, umode_t mode, dev_t dev)
2856 struct inode *inode;
2857 int error = -ENOSPC;
2859 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2861 error = simple_acl_create(dir, inode);
2864 error = security_inode_init_security(inode, dir,
2866 shmem_initxattrs, NULL);
2867 if (error && error != -EOPNOTSUPP)
2871 dir->i_size += BOGO_DIRENT_SIZE;
2872 dir->i_ctime = dir->i_mtime = current_time(dir);
2873 d_instantiate(dentry, inode);
2874 dget(dentry); /* Extra count - pin the dentry in core */
2883 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2884 struct dentry *dentry, umode_t mode)
2886 struct inode *inode;
2887 int error = -ENOSPC;
2889 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2891 error = security_inode_init_security(inode, dir,
2893 shmem_initxattrs, NULL);
2894 if (error && error != -EOPNOTSUPP)
2896 error = simple_acl_create(dir, inode);
2899 d_tmpfile(dentry, inode);
2907 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2908 struct dentry *dentry, umode_t mode)
2912 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2913 mode | S_IFDIR, 0)))
2919 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2920 struct dentry *dentry, umode_t mode, bool excl)
2922 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2928 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2930 struct inode *inode = d_inode(old_dentry);
2934 * No ordinary (disk based) filesystem counts links as inodes;
2935 * but each new link needs a new dentry, pinning lowmem, and
2936 * tmpfs dentries cannot be pruned until they are unlinked.
2937 * But if an O_TMPFILE file is linked into the tmpfs, the
2938 * first link must skip that, to get the accounting right.
2940 if (inode->i_nlink) {
2941 ret = shmem_reserve_inode(inode->i_sb, NULL);
2946 dir->i_size += BOGO_DIRENT_SIZE;
2947 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2949 ihold(inode); /* New dentry reference */
2950 dget(dentry); /* Extra pinning count for the created dentry */
2951 d_instantiate(dentry, inode);
2956 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2958 struct inode *inode = d_inode(dentry);
2960 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2961 shmem_free_inode(inode->i_sb);
2963 dir->i_size -= BOGO_DIRENT_SIZE;
2964 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2966 dput(dentry); /* Undo the count from "create" - this does all the work */
2970 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2972 if (!simple_empty(dentry))
2975 drop_nlink(d_inode(dentry));
2977 return shmem_unlink(dir, dentry);
2980 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2982 bool old_is_dir = d_is_dir(old_dentry);
2983 bool new_is_dir = d_is_dir(new_dentry);
2985 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2987 drop_nlink(old_dir);
2990 drop_nlink(new_dir);
2994 old_dir->i_ctime = old_dir->i_mtime =
2995 new_dir->i_ctime = new_dir->i_mtime =
2996 d_inode(old_dentry)->i_ctime =
2997 d_inode(new_dentry)->i_ctime = current_time(old_dir);
3002 static int shmem_whiteout(struct user_namespace *mnt_userns,
3003 struct inode *old_dir, struct dentry *old_dentry)
3005 struct dentry *whiteout;
3008 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3012 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
3013 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3019 * Cheat and hash the whiteout while the old dentry is still in
3020 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3022 * d_lookup() will consistently find one of them at this point,
3023 * not sure which one, but that isn't even important.
3030 * The VFS layer already does all the dentry stuff for rename,
3031 * we just have to decrement the usage count for the target if
3032 * it exists so that the VFS layer correctly free's it when it
3035 static int shmem_rename2(struct user_namespace *mnt_userns,
3036 struct inode *old_dir, struct dentry *old_dentry,
3037 struct inode *new_dir, struct dentry *new_dentry,
3040 struct inode *inode = d_inode(old_dentry);
3041 int they_are_dirs = S_ISDIR(inode->i_mode);
3043 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3046 if (flags & RENAME_EXCHANGE)
3047 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3049 if (!simple_empty(new_dentry))
3052 if (flags & RENAME_WHITEOUT) {
3055 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3060 if (d_really_is_positive(new_dentry)) {
3061 (void) shmem_unlink(new_dir, new_dentry);
3062 if (they_are_dirs) {
3063 drop_nlink(d_inode(new_dentry));
3064 drop_nlink(old_dir);
3066 } else if (they_are_dirs) {
3067 drop_nlink(old_dir);
3071 old_dir->i_size -= BOGO_DIRENT_SIZE;
3072 new_dir->i_size += BOGO_DIRENT_SIZE;
3073 old_dir->i_ctime = old_dir->i_mtime =
3074 new_dir->i_ctime = new_dir->i_mtime =
3075 inode->i_ctime = current_time(old_dir);
3079 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3080 struct dentry *dentry, const char *symname)
3084 struct inode *inode;
3087 len = strlen(symname) + 1;
3088 if (len > PAGE_SIZE)
3089 return -ENAMETOOLONG;
3091 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3096 error = security_inode_init_security(inode, dir, &dentry->d_name,
3097 shmem_initxattrs, NULL);
3098 if (error && error != -EOPNOTSUPP) {
3103 inode->i_size = len-1;
3104 if (len <= SHORT_SYMLINK_LEN) {
3105 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3106 if (!inode->i_link) {
3110 inode->i_op = &shmem_short_symlink_operations;
3112 inode_nohighmem(inode);
3113 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3118 inode->i_mapping->a_ops = &shmem_aops;
3119 inode->i_op = &shmem_symlink_inode_operations;
3120 memcpy(page_address(page), symname, len);
3121 SetPageUptodate(page);
3122 set_page_dirty(page);
3126 dir->i_size += BOGO_DIRENT_SIZE;
3127 dir->i_ctime = dir->i_mtime = current_time(dir);
3128 d_instantiate(dentry, inode);
3133 static void shmem_put_link(void *arg)
3135 mark_page_accessed(arg);
3139 static const char *shmem_get_link(struct dentry *dentry,
3140 struct inode *inode,
3141 struct delayed_call *done)
3143 struct page *page = NULL;
3146 page = find_get_page(inode->i_mapping, 0);
3148 return ERR_PTR(-ECHILD);
3149 if (!PageUptodate(page)) {
3151 return ERR_PTR(-ECHILD);
3154 error = shmem_getpage(inode, 0, &page, SGP_READ);
3156 return ERR_PTR(error);
3159 set_delayed_call(done, shmem_put_link, page);
3160 return page_address(page);
3163 #ifdef CONFIG_TMPFS_XATTR
3165 * Superblocks without xattr inode operations may get some security.* xattr
3166 * support from the LSM "for free". As soon as we have any other xattrs
3167 * like ACLs, we also need to implement the security.* handlers at
3168 * filesystem level, though.
3172 * Callback for security_inode_init_security() for acquiring xattrs.
3174 static int shmem_initxattrs(struct inode *inode,
3175 const struct xattr *xattr_array,
3178 struct shmem_inode_info *info = SHMEM_I(inode);
3179 const struct xattr *xattr;
3180 struct simple_xattr *new_xattr;
3183 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3184 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3188 len = strlen(xattr->name) + 1;
3189 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3191 if (!new_xattr->name) {
3196 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3197 XATTR_SECURITY_PREFIX_LEN);
3198 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3201 simple_xattr_list_add(&info->xattrs, new_xattr);
3207 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3208 struct dentry *unused, struct inode *inode,
3209 const char *name, void *buffer, size_t size)
3211 struct shmem_inode_info *info = SHMEM_I(inode);
3213 name = xattr_full_name(handler, name);
3214 return simple_xattr_get(&info->xattrs, name, buffer, size);
3217 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3218 struct user_namespace *mnt_userns,
3219 struct dentry *unused, struct inode *inode,
3220 const char *name, const void *value,
3221 size_t size, int flags)
3223 struct shmem_inode_info *info = SHMEM_I(inode);
3225 name = xattr_full_name(handler, name);
3226 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3229 static const struct xattr_handler shmem_security_xattr_handler = {
3230 .prefix = XATTR_SECURITY_PREFIX,
3231 .get = shmem_xattr_handler_get,
3232 .set = shmem_xattr_handler_set,
3235 static const struct xattr_handler shmem_trusted_xattr_handler = {
3236 .prefix = XATTR_TRUSTED_PREFIX,
3237 .get = shmem_xattr_handler_get,
3238 .set = shmem_xattr_handler_set,
3241 static const struct xattr_handler *shmem_xattr_handlers[] = {
3242 #ifdef CONFIG_TMPFS_POSIX_ACL
3243 &posix_acl_access_xattr_handler,
3244 &posix_acl_default_xattr_handler,
3246 &shmem_security_xattr_handler,
3247 &shmem_trusted_xattr_handler,
3251 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3253 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3254 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3256 #endif /* CONFIG_TMPFS_XATTR */
3258 static const struct inode_operations shmem_short_symlink_operations = {
3259 .get_link = simple_get_link,
3260 #ifdef CONFIG_TMPFS_XATTR
3261 .listxattr = shmem_listxattr,
3265 static const struct inode_operations shmem_symlink_inode_operations = {
3266 .get_link = shmem_get_link,
3267 #ifdef CONFIG_TMPFS_XATTR
3268 .listxattr = shmem_listxattr,
3272 static struct dentry *shmem_get_parent(struct dentry *child)
3274 return ERR_PTR(-ESTALE);
3277 static int shmem_match(struct inode *ino, void *vfh)
3281 inum = (inum << 32) | fh[1];
3282 return ino->i_ino == inum && fh[0] == ino->i_generation;
3285 /* Find any alias of inode, but prefer a hashed alias */
3286 static struct dentry *shmem_find_alias(struct inode *inode)
3288 struct dentry *alias = d_find_alias(inode);
3290 return alias ?: d_find_any_alias(inode);
3294 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3295 struct fid *fid, int fh_len, int fh_type)
3297 struct inode *inode;
3298 struct dentry *dentry = NULL;
3305 inum = (inum << 32) | fid->raw[1];
3307 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3308 shmem_match, fid->raw);
3310 dentry = shmem_find_alias(inode);
3317 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3318 struct inode *parent)
3322 return FILEID_INVALID;
3325 if (inode_unhashed(inode)) {
3326 /* Unfortunately insert_inode_hash is not idempotent,
3327 * so as we hash inodes here rather than at creation
3328 * time, we need a lock to ensure we only try
3331 static DEFINE_SPINLOCK(lock);
3333 if (inode_unhashed(inode))
3334 __insert_inode_hash(inode,
3335 inode->i_ino + inode->i_generation);
3339 fh[0] = inode->i_generation;
3340 fh[1] = inode->i_ino;
3341 fh[2] = ((__u64)inode->i_ino) >> 32;
3347 static const struct export_operations shmem_export_ops = {
3348 .get_parent = shmem_get_parent,
3349 .encode_fh = shmem_encode_fh,
3350 .fh_to_dentry = shmem_fh_to_dentry,
3366 static const struct constant_table shmem_param_enums_huge[] = {
3367 {"never", SHMEM_HUGE_NEVER },
3368 {"always", SHMEM_HUGE_ALWAYS },
3369 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3370 {"advise", SHMEM_HUGE_ADVISE },
3374 const struct fs_parameter_spec shmem_fs_parameters[] = {
3375 fsparam_u32 ("gid", Opt_gid),
3376 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3377 fsparam_u32oct("mode", Opt_mode),
3378 fsparam_string("mpol", Opt_mpol),
3379 fsparam_string("nr_blocks", Opt_nr_blocks),
3380 fsparam_string("nr_inodes", Opt_nr_inodes),
3381 fsparam_string("size", Opt_size),
3382 fsparam_u32 ("uid", Opt_uid),
3383 fsparam_flag ("inode32", Opt_inode32),
3384 fsparam_flag ("inode64", Opt_inode64),
3388 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3390 struct shmem_options *ctx = fc->fs_private;
3391 struct fs_parse_result result;
3392 unsigned long long size;
3396 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3402 size = memparse(param->string, &rest);
3404 size <<= PAGE_SHIFT;
3405 size *= totalram_pages();
3411 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3412 ctx->seen |= SHMEM_SEEN_BLOCKS;
3415 ctx->blocks = memparse(param->string, &rest);
3418 ctx->seen |= SHMEM_SEEN_BLOCKS;
3421 ctx->inodes = memparse(param->string, &rest);
3424 ctx->seen |= SHMEM_SEEN_INODES;
3427 ctx->mode = result.uint_32 & 07777;
3430 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3431 if (!uid_valid(ctx->uid))
3435 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3436 if (!gid_valid(ctx->gid))
3440 ctx->huge = result.uint_32;
3441 if (ctx->huge != SHMEM_HUGE_NEVER &&
3442 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3443 has_transparent_hugepage()))
3444 goto unsupported_parameter;
3445 ctx->seen |= SHMEM_SEEN_HUGE;
3448 if (IS_ENABLED(CONFIG_NUMA)) {
3449 mpol_put(ctx->mpol);
3451 if (mpol_parse_str(param->string, &ctx->mpol))
3455 goto unsupported_parameter;
3457 ctx->full_inums = false;
3458 ctx->seen |= SHMEM_SEEN_INUMS;
3461 if (sizeof(ino_t) < 8) {
3463 "Cannot use inode64 with <64bit inums in kernel\n");
3465 ctx->full_inums = true;
3466 ctx->seen |= SHMEM_SEEN_INUMS;
3471 unsupported_parameter:
3472 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3474 return invalfc(fc, "Bad value for '%s'", param->key);
3477 static int shmem_parse_options(struct fs_context *fc, void *data)
3479 char *options = data;
3482 int err = security_sb_eat_lsm_opts(options, &fc->security);
3487 while (options != NULL) {
3488 char *this_char = options;
3491 * NUL-terminate this option: unfortunately,
3492 * mount options form a comma-separated list,
3493 * but mpol's nodelist may also contain commas.
3495 options = strchr(options, ',');
3496 if (options == NULL)
3499 if (!isdigit(*options)) {
3505 char *value = strchr(this_char, '=');
3511 len = strlen(value);
3513 err = vfs_parse_fs_string(fc, this_char, value, len);
3522 * Reconfigure a shmem filesystem.
3524 * Note that we disallow change from limited->unlimited blocks/inodes while any
3525 * are in use; but we must separately disallow unlimited->limited, because in
3526 * that case we have no record of how much is already in use.
3528 static int shmem_reconfigure(struct fs_context *fc)
3530 struct shmem_options *ctx = fc->fs_private;
3531 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3532 unsigned long inodes;
3533 struct mempolicy *mpol = NULL;
3536 raw_spin_lock(&sbinfo->stat_lock);
3537 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3538 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3539 if (!sbinfo->max_blocks) {
3540 err = "Cannot retroactively limit size";
3543 if (percpu_counter_compare(&sbinfo->used_blocks,
3545 err = "Too small a size for current use";
3549 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3550 if (!sbinfo->max_inodes) {
3551 err = "Cannot retroactively limit inodes";
3554 if (ctx->inodes < inodes) {
3555 err = "Too few inodes for current use";
3560 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3561 sbinfo->next_ino > UINT_MAX) {
3562 err = "Current inum too high to switch to 32-bit inums";
3566 if (ctx->seen & SHMEM_SEEN_HUGE)
3567 sbinfo->huge = ctx->huge;
3568 if (ctx->seen & SHMEM_SEEN_INUMS)
3569 sbinfo->full_inums = ctx->full_inums;
3570 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3571 sbinfo->max_blocks = ctx->blocks;
3572 if (ctx->seen & SHMEM_SEEN_INODES) {
3573 sbinfo->max_inodes = ctx->inodes;
3574 sbinfo->free_inodes = ctx->inodes - inodes;
3578 * Preserve previous mempolicy unless mpol remount option was specified.
3581 mpol = sbinfo->mpol;
3582 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3585 raw_spin_unlock(&sbinfo->stat_lock);
3589 raw_spin_unlock(&sbinfo->stat_lock);
3590 return invalfc(fc, "%s", err);
3593 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3595 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3597 if (sbinfo->max_blocks != shmem_default_max_blocks())
3598 seq_printf(seq, ",size=%luk",
3599 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3600 if (sbinfo->max_inodes != shmem_default_max_inodes())
3601 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3602 if (sbinfo->mode != (0777 | S_ISVTX))
3603 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3604 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3605 seq_printf(seq, ",uid=%u",
3606 from_kuid_munged(&init_user_ns, sbinfo->uid));
3607 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3608 seq_printf(seq, ",gid=%u",
3609 from_kgid_munged(&init_user_ns, sbinfo->gid));
3612 * Showing inode{64,32} might be useful even if it's the system default,
3613 * since then people don't have to resort to checking both here and
3614 * /proc/config.gz to confirm 64-bit inums were successfully applied
3615 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3617 * We hide it when inode64 isn't the default and we are using 32-bit
3618 * inodes, since that probably just means the feature isn't even under
3623 * +-----------------+-----------------+
3624 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3625 * +------------------+-----------------+-----------------+
3626 * | full_inums=true | show | show |
3627 * | full_inums=false | show | hide |
3628 * +------------------+-----------------+-----------------+
3631 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3632 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3633 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
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 free_percpu(sbinfo->ino_batch);
3649 percpu_counter_destroy(&sbinfo->used_blocks);
3650 mpol_put(sbinfo->mpol);
3652 sb->s_fs_info = NULL;
3655 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3657 struct shmem_options *ctx = fc->fs_private;
3658 struct inode *inode;
3659 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();
3680 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3681 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3683 sb->s_flags |= SB_NOUSER;
3685 sb->s_export_op = &shmem_export_ops;
3686 sb->s_flags |= SB_NOSEC;
3688 sb->s_flags |= SB_NOUSER;
3690 sbinfo->max_blocks = ctx->blocks;
3691 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3692 if (sb->s_flags & SB_KERNMOUNT) {
3693 sbinfo->ino_batch = alloc_percpu(ino_t);
3694 if (!sbinfo->ino_batch)
3697 sbinfo->uid = ctx->uid;
3698 sbinfo->gid = ctx->gid;
3699 sbinfo->full_inums = ctx->full_inums;
3700 sbinfo->mode = ctx->mode;
3701 sbinfo->huge = ctx->huge;
3702 sbinfo->mpol = ctx->mpol;
3705 raw_spin_lock_init(&sbinfo->stat_lock);
3706 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3708 spin_lock_init(&sbinfo->shrinklist_lock);
3709 INIT_LIST_HEAD(&sbinfo->shrinklist);
3711 sb->s_maxbytes = MAX_LFS_FILESIZE;
3712 sb->s_blocksize = PAGE_SIZE;
3713 sb->s_blocksize_bits = PAGE_SHIFT;
3714 sb->s_magic = TMPFS_MAGIC;
3715 sb->s_op = &shmem_ops;
3716 sb->s_time_gran = 1;
3717 #ifdef CONFIG_TMPFS_XATTR
3718 sb->s_xattr = shmem_xattr_handlers;
3720 #ifdef CONFIG_TMPFS_POSIX_ACL
3721 sb->s_flags |= SB_POSIXACL;
3723 uuid_gen(&sb->s_uuid);
3725 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3728 inode->i_uid = sbinfo->uid;
3729 inode->i_gid = sbinfo->gid;
3730 sb->s_root = d_make_root(inode);
3736 shmem_put_super(sb);
3740 static int shmem_get_tree(struct fs_context *fc)
3742 return get_tree_nodev(fc, shmem_fill_super);
3745 static void shmem_free_fc(struct fs_context *fc)
3747 struct shmem_options *ctx = fc->fs_private;
3750 mpol_put(ctx->mpol);
3755 static const struct fs_context_operations shmem_fs_context_ops = {
3756 .free = shmem_free_fc,
3757 .get_tree = shmem_get_tree,
3759 .parse_monolithic = shmem_parse_options,
3760 .parse_param = shmem_parse_one,
3761 .reconfigure = shmem_reconfigure,
3765 static struct kmem_cache *shmem_inode_cachep;
3767 static struct inode *shmem_alloc_inode(struct super_block *sb)
3769 struct shmem_inode_info *info;
3770 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3773 return &info->vfs_inode;
3776 static void shmem_free_in_core_inode(struct inode *inode)
3778 if (S_ISLNK(inode->i_mode))
3779 kfree(inode->i_link);
3780 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3783 static void shmem_destroy_inode(struct inode *inode)
3785 if (S_ISREG(inode->i_mode))
3786 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3789 static void shmem_init_inode(void *foo)
3791 struct shmem_inode_info *info = foo;
3792 inode_init_once(&info->vfs_inode);
3795 static void shmem_init_inodecache(void)
3797 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3798 sizeof(struct shmem_inode_info),
3799 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3802 static void shmem_destroy_inodecache(void)
3804 kmem_cache_destroy(shmem_inode_cachep);
3807 const struct address_space_operations shmem_aops = {
3808 .writepage = shmem_writepage,
3809 .set_page_dirty = __set_page_dirty_no_writeback,
3811 .write_begin = shmem_write_begin,
3812 .write_end = shmem_write_end,
3814 #ifdef CONFIG_MIGRATION
3815 .migratepage = migrate_page,
3817 .error_remove_page = generic_error_remove_page,
3819 EXPORT_SYMBOL(shmem_aops);
3821 static const struct file_operations shmem_file_operations = {
3823 .get_unmapped_area = shmem_get_unmapped_area,
3825 .llseek = shmem_file_llseek,
3826 .read_iter = shmem_file_read_iter,
3827 .write_iter = generic_file_write_iter,
3828 .fsync = noop_fsync,
3829 .splice_read = generic_file_splice_read,
3830 .splice_write = iter_file_splice_write,
3831 .fallocate = shmem_fallocate,
3835 static const struct inode_operations shmem_inode_operations = {
3836 .getattr = shmem_getattr,
3837 .setattr = shmem_setattr,
3838 #ifdef CONFIG_TMPFS_XATTR
3839 .listxattr = shmem_listxattr,
3840 .set_acl = simple_set_acl,
3844 static const struct inode_operations shmem_dir_inode_operations = {
3846 .create = shmem_create,
3847 .lookup = simple_lookup,
3849 .unlink = shmem_unlink,
3850 .symlink = shmem_symlink,
3851 .mkdir = shmem_mkdir,
3852 .rmdir = shmem_rmdir,
3853 .mknod = shmem_mknod,
3854 .rename = shmem_rename2,
3855 .tmpfile = shmem_tmpfile,
3857 #ifdef CONFIG_TMPFS_XATTR
3858 .listxattr = shmem_listxattr,
3860 #ifdef CONFIG_TMPFS_POSIX_ACL
3861 .setattr = shmem_setattr,
3862 .set_acl = simple_set_acl,
3866 static const struct inode_operations shmem_special_inode_operations = {
3867 #ifdef CONFIG_TMPFS_XATTR
3868 .listxattr = shmem_listxattr,
3870 #ifdef CONFIG_TMPFS_POSIX_ACL
3871 .setattr = shmem_setattr,
3872 .set_acl = simple_set_acl,
3876 static const struct super_operations shmem_ops = {
3877 .alloc_inode = shmem_alloc_inode,
3878 .free_inode = shmem_free_in_core_inode,
3879 .destroy_inode = shmem_destroy_inode,
3881 .statfs = shmem_statfs,
3882 .show_options = shmem_show_options,
3884 .evict_inode = shmem_evict_inode,
3885 .drop_inode = generic_delete_inode,
3886 .put_super = shmem_put_super,
3887 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3888 .nr_cached_objects = shmem_unused_huge_count,
3889 .free_cached_objects = shmem_unused_huge_scan,
3893 static const struct vm_operations_struct shmem_vm_ops = {
3894 .fault = shmem_fault,
3895 .map_pages = filemap_map_pages,
3897 .set_policy = shmem_set_policy,
3898 .get_policy = shmem_get_policy,
3902 int shmem_init_fs_context(struct fs_context *fc)
3904 struct shmem_options *ctx;
3906 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3910 ctx->mode = 0777 | S_ISVTX;
3911 ctx->uid = current_fsuid();
3912 ctx->gid = current_fsgid();
3914 fc->fs_private = ctx;
3915 fc->ops = &shmem_fs_context_ops;
3919 static struct file_system_type shmem_fs_type = {
3920 .owner = THIS_MODULE,
3922 .init_fs_context = shmem_init_fs_context,
3924 .parameters = shmem_fs_parameters,
3926 .kill_sb = kill_litter_super,
3927 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
3930 int __init shmem_init(void)
3934 shmem_init_inodecache();
3936 error = register_filesystem(&shmem_fs_type);
3938 pr_err("Could not register tmpfs\n");
3942 shm_mnt = kern_mount(&shmem_fs_type);
3943 if (IS_ERR(shm_mnt)) {
3944 error = PTR_ERR(shm_mnt);
3945 pr_err("Could not kern_mount tmpfs\n");
3949 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3950 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3951 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3953 shmem_huge = 0; /* just in case it was patched */
3958 unregister_filesystem(&shmem_fs_type);
3960 shmem_destroy_inodecache();
3961 shm_mnt = ERR_PTR(error);
3965 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3966 static ssize_t shmem_enabled_show(struct kobject *kobj,
3967 struct kobj_attribute *attr, char *buf)
3969 static const int values[] = {
3971 SHMEM_HUGE_WITHIN_SIZE,
3980 for (i = 0; i < ARRAY_SIZE(values); i++) {
3981 len += sysfs_emit_at(buf, len,
3982 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3984 shmem_format_huge(values[i]));
3987 len += sysfs_emit_at(buf, len, "\n");
3992 static ssize_t shmem_enabled_store(struct kobject *kobj,
3993 struct kobj_attribute *attr, const char *buf, size_t count)
3998 if (count + 1 > sizeof(tmp))
4000 memcpy(tmp, buf, count);
4002 if (count && tmp[count - 1] == '\n')
4003 tmp[count - 1] = '\0';
4005 huge = shmem_parse_huge(tmp);
4006 if (huge == -EINVAL)
4008 if (!has_transparent_hugepage() &&
4009 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4013 if (shmem_huge > SHMEM_HUGE_DENY)
4014 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4018 struct kobj_attribute shmem_enabled_attr =
4019 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4020 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4022 #else /* !CONFIG_SHMEM */
4025 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4027 * This is intended for small system where the benefits of the full
4028 * shmem code (swap-backed and resource-limited) are outweighed by
4029 * their complexity. On systems without swap this code should be
4030 * effectively equivalent, but much lighter weight.
4033 static struct file_system_type shmem_fs_type = {
4035 .init_fs_context = ramfs_init_fs_context,
4036 .parameters = ramfs_fs_parameters,
4037 .kill_sb = kill_litter_super,
4038 .fs_flags = FS_USERNS_MOUNT,
4041 int __init shmem_init(void)
4043 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4045 shm_mnt = kern_mount(&shmem_fs_type);
4046 BUG_ON(IS_ERR(shm_mnt));
4051 int shmem_unuse(unsigned int type, bool frontswap,
4052 unsigned long *fs_pages_to_unuse)
4057 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4062 void shmem_unlock_mapping(struct address_space *mapping)
4067 unsigned long shmem_get_unmapped_area(struct file *file,
4068 unsigned long addr, unsigned long len,
4069 unsigned long pgoff, unsigned long flags)
4071 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4075 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4077 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4079 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4081 #define shmem_vm_ops generic_file_vm_ops
4082 #define shmem_file_operations ramfs_file_operations
4083 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4084 #define shmem_acct_size(flags, size) 0
4085 #define shmem_unacct_size(flags, size) do {} while (0)
4087 #endif /* CONFIG_SHMEM */
4091 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4092 unsigned long flags, unsigned int i_flags)
4094 struct inode *inode;
4098 return ERR_CAST(mnt);
4100 if (size < 0 || size > MAX_LFS_FILESIZE)
4101 return ERR_PTR(-EINVAL);
4103 if (shmem_acct_size(flags, size))
4104 return ERR_PTR(-ENOMEM);
4106 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4108 if (unlikely(!inode)) {
4109 shmem_unacct_size(flags, size);
4110 return ERR_PTR(-ENOSPC);
4112 inode->i_flags |= i_flags;
4113 inode->i_size = size;
4114 clear_nlink(inode); /* It is unlinked */
4115 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4117 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4118 &shmem_file_operations);
4125 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4126 * kernel internal. There will be NO LSM permission checks against the
4127 * underlying inode. So users of this interface must do LSM checks at a
4128 * higher layer. The users are the big_key and shm implementations. LSM
4129 * checks are provided at the key or shm level rather than the inode.
4130 * @name: name for dentry (to be seen in /proc/<pid>/maps
4131 * @size: size to be set for the file
4132 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4134 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4136 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4140 * shmem_file_setup - get an unlinked file living in tmpfs
4141 * @name: name for dentry (to be seen in /proc/<pid>/maps
4142 * @size: size to be set for the file
4143 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4145 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4147 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4149 EXPORT_SYMBOL_GPL(shmem_file_setup);
4152 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4153 * @mnt: the tmpfs mount where the file will be created
4154 * @name: name for dentry (to be seen in /proc/<pid>/maps
4155 * @size: size to be set for the file
4156 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4158 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4159 loff_t size, unsigned long flags)
4161 return __shmem_file_setup(mnt, name, size, flags, 0);
4163 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4166 * shmem_zero_setup - setup a shared anonymous mapping
4167 * @vma: the vma to be mmapped is prepared by do_mmap
4169 int shmem_zero_setup(struct vm_area_struct *vma)
4172 loff_t size = vma->vm_end - vma->vm_start;
4175 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4176 * between XFS directory reading and selinux: since this file is only
4177 * accessible to the user through its mapping, use S_PRIVATE flag to
4178 * bypass file security, in the same way as shmem_kernel_file_setup().
4180 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4182 return PTR_ERR(file);
4186 vma->vm_file = file;
4187 vma->vm_ops = &shmem_vm_ops;
4189 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4190 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4191 (vma->vm_end & HPAGE_PMD_MASK)) {
4192 khugepaged_enter(vma, vma->vm_flags);
4199 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4200 * @mapping: the page's address_space
4201 * @index: the page index
4202 * @gfp: the page allocator flags to use if allocating
4204 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4205 * with any new page allocations done using the specified allocation flags.
4206 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4207 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4208 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4210 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4211 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4213 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4214 pgoff_t index, gfp_t gfp)
4217 struct inode *inode = mapping->host;
4221 BUG_ON(!shmem_mapping(mapping));
4222 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4223 gfp, NULL, NULL, NULL);
4225 page = ERR_PTR(error);
4231 * The tiny !SHMEM case uses ramfs without swap
4233 return read_cache_page_gfp(mapping, index, gfp);
4236 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);