2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
42 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
44 static struct vfsmount *shm_mnt;
48 * This virtual memory filesystem is heavily based on the ramfs. It
49 * extends ramfs by the ability to use swap and honor resource limits
50 * which makes it a completely usable filesystem.
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/blkdev.h>
64 #include <linux/pagevec.h>
65 #include <linux/percpu_counter.h>
66 #include <linux/falloc.h>
67 #include <linux/splice.h>
68 #include <linux/security.h>
69 #include <linux/swapops.h>
70 #include <linux/mempolicy.h>
71 #include <linux/namei.h>
72 #include <linux/ctype.h>
73 #include <linux/migrate.h>
74 #include <linux/highmem.h>
75 #include <linux/seq_file.h>
76 #include <linux/magic.h>
77 #include <linux/syscalls.h>
78 #include <linux/fcntl.h>
79 #include <uapi/linux/memfd.h>
80 #include <linux/userfaultfd_k.h>
81 #include <linux/rmap.h>
82 #include <linux/uuid.h>
84 #include <linux/uaccess.h>
88 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
89 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
91 /* Pretend that each entry is of this size in directory's i_size */
92 #define BOGO_DIRENT_SIZE 20
94 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
95 #define SHORT_SYMLINK_LEN 128
98 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
99 * inode->i_private (with i_mutex making sure that it has only one user at
100 * a time): we would prefer not to enlarge the shmem inode just for that.
102 struct shmem_falloc {
103 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
104 pgoff_t start; /* start of range currently being fallocated */
105 pgoff_t next; /* the next page offset to be fallocated */
106 pgoff_t nr_falloced; /* how many new pages have been fallocated */
107 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
110 struct shmem_options {
111 unsigned long long blocks;
112 unsigned long long inodes;
113 struct mempolicy *mpol;
120 #define SHMEM_SEEN_BLOCKS 1
121 #define SHMEM_SEEN_INODES 2
122 #define SHMEM_SEEN_HUGE 4
123 #define SHMEM_SEEN_INUMS 8
127 static unsigned long shmem_default_max_blocks(void)
129 return totalram_pages() / 2;
132 static unsigned long shmem_default_max_inodes(void)
134 unsigned long nr_pages = totalram_pages();
136 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
140 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
141 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
142 struct shmem_inode_info *info, pgoff_t index);
143 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 vm_fault_t *fault_type);
147 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
148 struct page **pagep, enum sgp_type sgp,
149 gfp_t gfp, struct vm_area_struct *vma,
150 struct vm_fault *vmf, vm_fault_t *fault_type);
152 int shmem_getpage(struct inode *inode, pgoff_t index,
153 struct page **pagep, enum sgp_type sgp)
155 return shmem_getpage_gfp(inode, index, pagep, sgp,
156 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
159 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
161 return sb->s_fs_info;
165 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
166 * for shared memory and for shared anonymous (/dev/zero) mappings
167 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
168 * consistent with the pre-accounting of private mappings ...
170 static inline int shmem_acct_size(unsigned long flags, loff_t size)
172 return (flags & VM_NORESERVE) ?
173 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
176 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
178 if (!(flags & VM_NORESERVE))
179 vm_unacct_memory(VM_ACCT(size));
182 static inline int shmem_reacct_size(unsigned long flags,
183 loff_t oldsize, loff_t newsize)
185 if (!(flags & VM_NORESERVE)) {
186 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
187 return security_vm_enough_memory_mm(current->mm,
188 VM_ACCT(newsize) - VM_ACCT(oldsize));
189 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
190 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
196 * ... whereas tmpfs objects are accounted incrementally as
197 * pages are allocated, in order to allow large sparse files.
198 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
199 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201 static inline int shmem_acct_block(unsigned long flags, long pages)
203 if (!(flags & VM_NORESERVE))
206 return security_vm_enough_memory_mm(current->mm,
207 pages * VM_ACCT(PAGE_SIZE));
210 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
212 if (flags & VM_NORESERVE)
213 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
216 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
218 struct shmem_inode_info *info = SHMEM_I(inode);
219 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
221 if (shmem_acct_block(info->flags, pages))
224 if (sbinfo->max_blocks) {
225 if (percpu_counter_compare(&sbinfo->used_blocks,
226 sbinfo->max_blocks - pages) > 0)
228 percpu_counter_add(&sbinfo->used_blocks, pages);
234 shmem_unacct_blocks(info->flags, pages);
238 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
240 struct shmem_inode_info *info = SHMEM_I(inode);
241 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
243 if (sbinfo->max_blocks)
244 percpu_counter_sub(&sbinfo->used_blocks, pages);
245 shmem_unacct_blocks(info->flags, pages);
248 static const struct super_operations shmem_ops;
249 const struct address_space_operations shmem_aops;
250 static const struct file_operations shmem_file_operations;
251 static const struct inode_operations shmem_inode_operations;
252 static const struct inode_operations shmem_dir_inode_operations;
253 static const struct inode_operations shmem_special_inode_operations;
254 static const struct vm_operations_struct shmem_vm_ops;
255 static struct file_system_type shmem_fs_type;
257 bool vma_is_shmem(struct vm_area_struct *vma)
259 return vma->vm_ops == &shmem_vm_ops;
262 static LIST_HEAD(shmem_swaplist);
263 static DEFINE_MUTEX(shmem_swaplist_mutex);
266 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
267 * produces a novel ino for the newly allocated inode.
269 * It may also be called when making a hard link to permit the space needed by
270 * each dentry. However, in that case, no new inode number is needed since that
271 * internally draws from another pool of inode numbers (currently global
272 * get_next_ino()). This case is indicated by passing NULL as inop.
274 #define SHMEM_INO_BATCH 1024
275 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
277 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
280 if (!(sb->s_flags & SB_KERNMOUNT)) {
281 spin_lock(&sbinfo->stat_lock);
282 if (sbinfo->max_inodes) {
283 if (!sbinfo->free_inodes) {
284 spin_unlock(&sbinfo->stat_lock);
287 sbinfo->free_inodes--;
290 ino = sbinfo->next_ino++;
291 if (unlikely(is_zero_ino(ino)))
292 ino = sbinfo->next_ino++;
293 if (unlikely(!sbinfo->full_inums &&
296 * Emulate get_next_ino uint wraparound for
299 if (IS_ENABLED(CONFIG_64BIT))
300 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
301 __func__, MINOR(sb->s_dev));
302 sbinfo->next_ino = 1;
303 ino = sbinfo->next_ino++;
307 spin_unlock(&sbinfo->stat_lock);
310 * __shmem_file_setup, one of our callers, is lock-free: it
311 * doesn't hold stat_lock in shmem_reserve_inode since
312 * max_inodes is always 0, and is called from potentially
313 * unknown contexts. As such, use a per-cpu batched allocator
314 * which doesn't require the per-sb stat_lock unless we are at
315 * the batch boundary.
317 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
318 * shmem mounts are not exposed to userspace, so we don't need
319 * to worry about things like glibc compatibility.
322 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
324 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
325 spin_lock(&sbinfo->stat_lock);
326 ino = sbinfo->next_ino;
327 sbinfo->next_ino += SHMEM_INO_BATCH;
328 spin_unlock(&sbinfo->stat_lock);
329 if (unlikely(is_zero_ino(ino)))
340 static void shmem_free_inode(struct super_block *sb)
342 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
343 if (sbinfo->max_inodes) {
344 spin_lock(&sbinfo->stat_lock);
345 sbinfo->free_inodes++;
346 spin_unlock(&sbinfo->stat_lock);
351 * shmem_recalc_inode - recalculate the block usage of an inode
352 * @inode: inode to recalc
354 * We have to calculate the free blocks since the mm can drop
355 * undirtied hole pages behind our back.
357 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
358 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
360 * It has to be called with the spinlock held.
362 static void shmem_recalc_inode(struct inode *inode)
364 struct shmem_inode_info *info = SHMEM_I(inode);
367 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
369 info->alloced -= freed;
370 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
371 shmem_inode_unacct_blocks(inode, freed);
375 bool shmem_charge(struct inode *inode, long pages)
377 struct shmem_inode_info *info = SHMEM_I(inode);
380 if (!shmem_inode_acct_block(inode, pages))
383 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
384 inode->i_mapping->nrpages += pages;
386 spin_lock_irqsave(&info->lock, flags);
387 info->alloced += pages;
388 inode->i_blocks += pages * BLOCKS_PER_PAGE;
389 shmem_recalc_inode(inode);
390 spin_unlock_irqrestore(&info->lock, flags);
395 void shmem_uncharge(struct inode *inode, long pages)
397 struct shmem_inode_info *info = SHMEM_I(inode);
400 /* nrpages adjustment done by __delete_from_page_cache() or caller */
402 spin_lock_irqsave(&info->lock, flags);
403 info->alloced -= pages;
404 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
405 shmem_recalc_inode(inode);
406 spin_unlock_irqrestore(&info->lock, flags);
408 shmem_inode_unacct_blocks(inode, pages);
412 * Replace item expected in xarray by a new item, while holding xa_lock.
414 static int shmem_replace_entry(struct address_space *mapping,
415 pgoff_t index, void *expected, void *replacement)
417 XA_STATE(xas, &mapping->i_pages, index);
420 VM_BUG_ON(!expected);
421 VM_BUG_ON(!replacement);
422 item = xas_load(&xas);
423 if (item != expected)
425 xas_store(&xas, replacement);
430 * Sometimes, before we decide whether to proceed or to fail, we must check
431 * that an entry was not already brought back from swap by a racing thread.
433 * Checking page is not enough: by the time a SwapCache page is locked, it
434 * might be reused, and again be SwapCache, using the same swap as before.
436 static bool shmem_confirm_swap(struct address_space *mapping,
437 pgoff_t index, swp_entry_t swap)
439 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
443 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
446 * disables huge pages for the mount;
448 * enables huge pages for the mount;
449 * SHMEM_HUGE_WITHIN_SIZE:
450 * only allocate huge pages if the page will be fully within i_size,
451 * also respect fadvise()/madvise() hints;
453 * only allocate huge pages if requested with fadvise()/madvise();
456 #define SHMEM_HUGE_NEVER 0
457 #define SHMEM_HUGE_ALWAYS 1
458 #define SHMEM_HUGE_WITHIN_SIZE 2
459 #define SHMEM_HUGE_ADVISE 3
463 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
466 * disables huge on shm_mnt and all mounts, for emergency use;
468 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
471 #define SHMEM_HUGE_DENY (-1)
472 #define SHMEM_HUGE_FORCE (-2)
474 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
475 /* ifdef here to avoid bloating shmem.o when not necessary */
477 static int shmem_huge __read_mostly;
479 #if defined(CONFIG_SYSFS)
480 static int shmem_parse_huge(const char *str)
482 if (!strcmp(str, "never"))
483 return SHMEM_HUGE_NEVER;
484 if (!strcmp(str, "always"))
485 return SHMEM_HUGE_ALWAYS;
486 if (!strcmp(str, "within_size"))
487 return SHMEM_HUGE_WITHIN_SIZE;
488 if (!strcmp(str, "advise"))
489 return SHMEM_HUGE_ADVISE;
490 if (!strcmp(str, "deny"))
491 return SHMEM_HUGE_DENY;
492 if (!strcmp(str, "force"))
493 return SHMEM_HUGE_FORCE;
498 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
499 static const char *shmem_format_huge(int huge)
502 case SHMEM_HUGE_NEVER:
504 case SHMEM_HUGE_ALWAYS:
506 case SHMEM_HUGE_WITHIN_SIZE:
507 return "within_size";
508 case SHMEM_HUGE_ADVISE:
510 case SHMEM_HUGE_DENY:
512 case SHMEM_HUGE_FORCE:
521 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
522 struct shrink_control *sc, unsigned long nr_to_split)
524 LIST_HEAD(list), *pos, *next;
525 LIST_HEAD(to_remove);
527 struct shmem_inode_info *info;
529 unsigned long batch = sc ? sc->nr_to_scan : 128;
530 int removed = 0, split = 0;
532 if (list_empty(&sbinfo->shrinklist))
535 spin_lock(&sbinfo->shrinklist_lock);
536 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
537 info = list_entry(pos, struct shmem_inode_info, shrinklist);
540 inode = igrab(&info->vfs_inode);
542 /* inode is about to be evicted */
544 list_del_init(&info->shrinklist);
549 /* Check if there's anything to gain */
550 if (round_up(inode->i_size, PAGE_SIZE) ==
551 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
552 list_move(&info->shrinklist, &to_remove);
557 list_move(&info->shrinklist, &list);
562 spin_unlock(&sbinfo->shrinklist_lock);
564 list_for_each_safe(pos, next, &to_remove) {
565 info = list_entry(pos, struct shmem_inode_info, shrinklist);
566 inode = &info->vfs_inode;
567 list_del_init(&info->shrinklist);
571 list_for_each_safe(pos, next, &list) {
574 info = list_entry(pos, struct shmem_inode_info, shrinklist);
575 inode = &info->vfs_inode;
577 if (nr_to_split && split >= nr_to_split)
580 page = find_get_page(inode->i_mapping,
581 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
585 /* No huge page at the end of the file: nothing to split */
586 if (!PageTransHuge(page)) {
592 * Leave the inode on the list if we failed to lock
593 * the page at this time.
595 * Waiting for the lock may lead to deadlock in the
598 if (!trylock_page(page)) {
603 ret = split_huge_page(page);
607 /* If split failed leave the inode on the list */
613 list_del_init(&info->shrinklist);
619 spin_lock(&sbinfo->shrinklist_lock);
620 list_splice_tail(&list, &sbinfo->shrinklist);
621 sbinfo->shrinklist_len -= removed;
622 spin_unlock(&sbinfo->shrinklist_lock);
627 static long shmem_unused_huge_scan(struct super_block *sb,
628 struct shrink_control *sc)
630 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
632 if (!READ_ONCE(sbinfo->shrinklist_len))
635 return shmem_unused_huge_shrink(sbinfo, sc, 0);
638 static long shmem_unused_huge_count(struct super_block *sb,
639 struct shrink_control *sc)
641 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
642 return READ_ONCE(sbinfo->shrinklist_len);
644 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
646 #define shmem_huge SHMEM_HUGE_DENY
648 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
649 struct shrink_control *sc, unsigned long nr_to_split)
653 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
655 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
657 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
658 (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
659 shmem_huge != SHMEM_HUGE_DENY)
665 * Like add_to_page_cache_locked, but error if expected item has gone.
667 static int shmem_add_to_page_cache(struct page *page,
668 struct address_space *mapping,
669 pgoff_t index, void *expected, gfp_t gfp,
670 struct mm_struct *charge_mm)
672 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
674 unsigned long nr = compound_nr(page);
677 VM_BUG_ON_PAGE(PageTail(page), page);
678 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
679 VM_BUG_ON_PAGE(!PageLocked(page), page);
680 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
681 VM_BUG_ON(expected && PageTransHuge(page));
683 page_ref_add(page, nr);
684 page->mapping = mapping;
687 if (!PageSwapCache(page)) {
688 error = mem_cgroup_charge(page, charge_mm, gfp);
690 if (PageTransHuge(page)) {
691 count_vm_event(THP_FILE_FALLBACK);
692 count_vm_event(THP_FILE_FALLBACK_CHARGE);
697 cgroup_throttle_swaprate(page, gfp);
702 entry = xas_find_conflict(&xas);
703 if (entry != expected)
704 xas_set_err(&xas, -EEXIST);
705 xas_create_range(&xas);
709 xas_store(&xas, page);
714 if (PageTransHuge(page)) {
715 count_vm_event(THP_FILE_ALLOC);
716 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
718 mapping->nrpages += nr;
719 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
720 __mod_lruvec_page_state(page, NR_SHMEM, nr);
722 xas_unlock_irq(&xas);
723 } while (xas_nomem(&xas, gfp));
725 if (xas_error(&xas)) {
726 error = xas_error(&xas);
732 page->mapping = NULL;
733 page_ref_sub(page, nr);
738 * Like delete_from_page_cache, but substitutes swap for page.
740 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
742 struct address_space *mapping = page->mapping;
745 VM_BUG_ON_PAGE(PageCompound(page), page);
747 xa_lock_irq(&mapping->i_pages);
748 error = shmem_replace_entry(mapping, page->index, page, radswap);
749 page->mapping = NULL;
751 __dec_lruvec_page_state(page, NR_FILE_PAGES);
752 __dec_lruvec_page_state(page, NR_SHMEM);
753 xa_unlock_irq(&mapping->i_pages);
759 * Remove swap entry from page cache, free the swap and its page cache.
761 static int shmem_free_swap(struct address_space *mapping,
762 pgoff_t index, void *radswap)
766 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
769 free_swap_and_cache(radix_to_swp_entry(radswap));
774 * Determine (in bytes) how many of the shmem object's pages mapped by the
775 * given offsets are swapped out.
777 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
778 * as long as the inode doesn't go away and racy results are not a problem.
780 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
781 pgoff_t start, pgoff_t end)
783 XA_STATE(xas, &mapping->i_pages, start);
785 unsigned long swapped = 0;
788 xas_for_each(&xas, page, end - 1) {
789 if (xas_retry(&xas, page))
791 if (xa_is_value(page))
794 if (need_resched()) {
802 return swapped << PAGE_SHIFT;
806 * Determine (in bytes) how many of the shmem object's pages mapped by the
807 * given vma is swapped out.
809 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
810 * as long as the inode doesn't go away and racy results are not a problem.
812 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
814 struct inode *inode = file_inode(vma->vm_file);
815 struct shmem_inode_info *info = SHMEM_I(inode);
816 struct address_space *mapping = inode->i_mapping;
817 unsigned long swapped;
819 /* Be careful as we don't hold info->lock */
820 swapped = READ_ONCE(info->swapped);
823 * The easier cases are when the shmem object has nothing in swap, or
824 * the vma maps it whole. Then we can simply use the stats that we
830 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
831 return swapped << PAGE_SHIFT;
833 /* Here comes the more involved part */
834 return shmem_partial_swap_usage(mapping,
835 linear_page_index(vma, vma->vm_start),
836 linear_page_index(vma, vma->vm_end));
840 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
842 void shmem_unlock_mapping(struct address_space *mapping)
849 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
851 while (!mapping_unevictable(mapping)) {
852 if (!pagevec_lookup(&pvec, mapping, &index))
854 check_move_unevictable_pages(&pvec);
855 pagevec_release(&pvec);
861 * Check whether a hole-punch or truncation needs to split a huge page,
862 * returning true if no split was required, or the split has been successful.
864 * Eviction (or truncation to 0 size) should never need to split a huge page;
865 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
866 * head, and then succeeded to trylock on tail.
868 * A split can only succeed when there are no additional references on the
869 * huge page: so the split below relies upon find_get_entries() having stopped
870 * when it found a subpage of the huge page, without getting further references.
872 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
874 if (!PageTransCompound(page))
877 /* Just proceed to delete a huge page wholly within the range punched */
878 if (PageHead(page) &&
879 page->index >= start && page->index + HPAGE_PMD_NR <= end)
882 /* Try to split huge page, so we can truly punch the hole or truncate */
883 return split_huge_page(page) >= 0;
887 * Remove range of pages and swap entries from page cache, and free them.
888 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
890 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
893 struct address_space *mapping = inode->i_mapping;
894 struct shmem_inode_info *info = SHMEM_I(inode);
895 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
896 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
897 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
898 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
900 pgoff_t indices[PAGEVEC_SIZE];
901 long nr_swaps_freed = 0;
906 end = -1; /* unsigned, so actually very big */
910 while (index < end && find_lock_entries(mapping, index, end - 1,
912 for (i = 0; i < pagevec_count(&pvec); i++) {
913 struct page *page = pvec.pages[i];
917 if (xa_is_value(page)) {
920 nr_swaps_freed += !shmem_free_swap(mapping,
924 index += thp_nr_pages(page) - 1;
926 if (!unfalloc || !PageUptodate(page))
927 truncate_inode_page(mapping, page);
930 pagevec_remove_exceptionals(&pvec);
931 pagevec_release(&pvec);
937 struct page *page = NULL;
938 shmem_getpage(inode, start - 1, &page, SGP_READ);
940 unsigned int top = PAGE_SIZE;
945 zero_user_segment(page, partial_start, top);
946 set_page_dirty(page);
952 struct page *page = NULL;
953 shmem_getpage(inode, end, &page, SGP_READ);
955 zero_user_segment(page, 0, partial_end);
956 set_page_dirty(page);
965 while (index < end) {
968 pvec.nr = find_get_entries(mapping, index, end - 1,
969 PAGEVEC_SIZE, pvec.pages, indices);
971 /* If all gone or hole-punch or unfalloc, we're done */
972 if (index == start || end != -1)
974 /* But if truncating, restart to make sure all gone */
978 for (i = 0; i < pagevec_count(&pvec); i++) {
979 struct page *page = pvec.pages[i];
982 if (xa_is_value(page)) {
985 if (shmem_free_swap(mapping, index, page)) {
986 /* Swap was replaced by page: retry */
996 if (!unfalloc || !PageUptodate(page)) {
997 if (page_mapping(page) != mapping) {
998 /* Page was replaced by swap: retry */
1003 VM_BUG_ON_PAGE(PageWriteback(page), page);
1004 if (shmem_punch_compound(page, start, end))
1005 truncate_inode_page(mapping, page);
1006 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1007 /* Wipe the page and don't get stuck */
1008 clear_highpage(page);
1009 flush_dcache_page(page);
1010 set_page_dirty(page);
1012 round_up(start, HPAGE_PMD_NR))
1018 pagevec_remove_exceptionals(&pvec);
1019 pagevec_release(&pvec);
1023 spin_lock_irq(&info->lock);
1024 info->swapped -= nr_swaps_freed;
1025 shmem_recalc_inode(inode);
1026 spin_unlock_irq(&info->lock);
1029 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1031 shmem_undo_range(inode, lstart, lend, false);
1032 inode->i_ctime = inode->i_mtime = current_time(inode);
1034 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1036 static int shmem_getattr(struct user_namespace *mnt_userns,
1037 const struct path *path, struct kstat *stat,
1038 u32 request_mask, unsigned int query_flags)
1040 struct inode *inode = path->dentry->d_inode;
1041 struct shmem_inode_info *info = SHMEM_I(inode);
1042 struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1044 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1045 spin_lock_irq(&info->lock);
1046 shmem_recalc_inode(inode);
1047 spin_unlock_irq(&info->lock);
1049 generic_fillattr(&init_user_ns, inode, stat);
1051 if (is_huge_enabled(sb_info))
1052 stat->blksize = HPAGE_PMD_SIZE;
1057 static int shmem_setattr(struct user_namespace *mnt_userns,
1058 struct dentry *dentry, struct iattr *attr)
1060 struct inode *inode = d_inode(dentry);
1061 struct shmem_inode_info *info = SHMEM_I(inode);
1062 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1065 error = setattr_prepare(&init_user_ns, dentry, attr);
1069 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1070 loff_t oldsize = inode->i_size;
1071 loff_t newsize = attr->ia_size;
1073 /* protected by i_mutex */
1074 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1075 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1078 if (newsize != oldsize) {
1079 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1083 i_size_write(inode, newsize);
1084 inode->i_ctime = inode->i_mtime = current_time(inode);
1086 if (newsize <= oldsize) {
1087 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1088 if (oldsize > holebegin)
1089 unmap_mapping_range(inode->i_mapping,
1092 shmem_truncate_range(inode,
1093 newsize, (loff_t)-1);
1094 /* unmap again to remove racily COWed private pages */
1095 if (oldsize > holebegin)
1096 unmap_mapping_range(inode->i_mapping,
1100 * Part of the huge page can be beyond i_size: subject
1101 * to shrink under memory pressure.
1103 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1104 spin_lock(&sbinfo->shrinklist_lock);
1106 * _careful to defend against unlocked access to
1107 * ->shrink_list in shmem_unused_huge_shrink()
1109 if (list_empty_careful(&info->shrinklist)) {
1110 list_add_tail(&info->shrinklist,
1111 &sbinfo->shrinklist);
1112 sbinfo->shrinklist_len++;
1114 spin_unlock(&sbinfo->shrinklist_lock);
1119 setattr_copy(&init_user_ns, inode, attr);
1120 if (attr->ia_valid & ATTR_MODE)
1121 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1125 static void shmem_evict_inode(struct inode *inode)
1127 struct shmem_inode_info *info = SHMEM_I(inode);
1128 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1130 if (shmem_mapping(inode->i_mapping)) {
1131 shmem_unacct_size(info->flags, inode->i_size);
1133 shmem_truncate_range(inode, 0, (loff_t)-1);
1134 if (!list_empty(&info->shrinklist)) {
1135 spin_lock(&sbinfo->shrinklist_lock);
1136 if (!list_empty(&info->shrinklist)) {
1137 list_del_init(&info->shrinklist);
1138 sbinfo->shrinklist_len--;
1140 spin_unlock(&sbinfo->shrinklist_lock);
1142 while (!list_empty(&info->swaplist)) {
1143 /* Wait while shmem_unuse() is scanning this inode... */
1144 wait_var_event(&info->stop_eviction,
1145 !atomic_read(&info->stop_eviction));
1146 mutex_lock(&shmem_swaplist_mutex);
1147 /* ...but beware of the race if we peeked too early */
1148 if (!atomic_read(&info->stop_eviction))
1149 list_del_init(&info->swaplist);
1150 mutex_unlock(&shmem_swaplist_mutex);
1154 simple_xattrs_free(&info->xattrs);
1155 WARN_ON(inode->i_blocks);
1156 shmem_free_inode(inode->i_sb);
1160 extern struct swap_info_struct *swap_info[];
1162 static int shmem_find_swap_entries(struct address_space *mapping,
1163 pgoff_t start, unsigned int nr_entries,
1164 struct page **entries, pgoff_t *indices,
1165 unsigned int type, bool frontswap)
1167 XA_STATE(xas, &mapping->i_pages, start);
1170 unsigned int ret = 0;
1176 xas_for_each(&xas, page, ULONG_MAX) {
1177 if (xas_retry(&xas, page))
1180 if (!xa_is_value(page))
1183 entry = radix_to_swp_entry(page);
1184 if (swp_type(entry) != type)
1187 !frontswap_test(swap_info[type], swp_offset(entry)))
1190 indices[ret] = xas.xa_index;
1191 entries[ret] = page;
1193 if (need_resched()) {
1197 if (++ret == nr_entries)
1206 * Move the swapped pages for an inode to page cache. Returns the count
1207 * of pages swapped in, or the error in case of failure.
1209 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1215 struct address_space *mapping = inode->i_mapping;
1217 for (i = 0; i < pvec.nr; i++) {
1218 struct page *page = pvec.pages[i];
1220 if (!xa_is_value(page))
1222 error = shmem_swapin_page(inode, indices[i],
1224 mapping_gfp_mask(mapping),
1231 if (error == -ENOMEM)
1235 return error ? error : ret;
1239 * If swap found in inode, free it and move page from swapcache to filecache.
1241 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1242 bool frontswap, unsigned long *fs_pages_to_unuse)
1244 struct address_space *mapping = inode->i_mapping;
1246 struct pagevec pvec;
1247 pgoff_t indices[PAGEVEC_SIZE];
1248 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1251 pagevec_init(&pvec);
1253 unsigned int nr_entries = PAGEVEC_SIZE;
1255 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1256 nr_entries = *fs_pages_to_unuse;
1258 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1259 pvec.pages, indices,
1266 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1270 if (frontswap_partial) {
1271 *fs_pages_to_unuse -= ret;
1272 if (*fs_pages_to_unuse == 0) {
1273 ret = FRONTSWAP_PAGES_UNUSED;
1278 start = indices[pvec.nr - 1];
1285 * Read all the shared memory data that resides in the swap
1286 * device 'type' back into memory, so the swap device can be
1289 int shmem_unuse(unsigned int type, bool frontswap,
1290 unsigned long *fs_pages_to_unuse)
1292 struct shmem_inode_info *info, *next;
1295 if (list_empty(&shmem_swaplist))
1298 mutex_lock(&shmem_swaplist_mutex);
1299 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1300 if (!info->swapped) {
1301 list_del_init(&info->swaplist);
1305 * Drop the swaplist mutex while searching the inode for swap;
1306 * but before doing so, make sure shmem_evict_inode() will not
1307 * remove placeholder inode from swaplist, nor let it be freed
1308 * (igrab() would protect from unlink, but not from unmount).
1310 atomic_inc(&info->stop_eviction);
1311 mutex_unlock(&shmem_swaplist_mutex);
1313 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1317 mutex_lock(&shmem_swaplist_mutex);
1318 next = list_next_entry(info, swaplist);
1320 list_del_init(&info->swaplist);
1321 if (atomic_dec_and_test(&info->stop_eviction))
1322 wake_up_var(&info->stop_eviction);
1326 mutex_unlock(&shmem_swaplist_mutex);
1332 * Move the page from the page cache to the swap cache.
1334 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1336 struct shmem_inode_info *info;
1337 struct address_space *mapping;
1338 struct inode *inode;
1342 VM_BUG_ON_PAGE(PageCompound(page), page);
1343 BUG_ON(!PageLocked(page));
1344 mapping = page->mapping;
1345 index = page->index;
1346 inode = mapping->host;
1347 info = SHMEM_I(inode);
1348 if (info->flags & VM_LOCKED)
1350 if (!total_swap_pages)
1354 * Our capabilities prevent regular writeback or sync from ever calling
1355 * shmem_writepage; but a stacking filesystem might use ->writepage of
1356 * its underlying filesystem, in which case tmpfs should write out to
1357 * swap only in response to memory pressure, and not for the writeback
1360 if (!wbc->for_reclaim) {
1361 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1366 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1367 * value into swapfile.c, the only way we can correctly account for a
1368 * fallocated page arriving here is now to initialize it and write it.
1370 * That's okay for a page already fallocated earlier, but if we have
1371 * not yet completed the fallocation, then (a) we want to keep track
1372 * of this page in case we have to undo it, and (b) it may not be a
1373 * good idea to continue anyway, once we're pushing into swap. So
1374 * reactivate the page, and let shmem_fallocate() quit when too many.
1376 if (!PageUptodate(page)) {
1377 if (inode->i_private) {
1378 struct shmem_falloc *shmem_falloc;
1379 spin_lock(&inode->i_lock);
1380 shmem_falloc = inode->i_private;
1382 !shmem_falloc->waitq &&
1383 index >= shmem_falloc->start &&
1384 index < shmem_falloc->next)
1385 shmem_falloc->nr_unswapped++;
1387 shmem_falloc = NULL;
1388 spin_unlock(&inode->i_lock);
1392 clear_highpage(page);
1393 flush_dcache_page(page);
1394 SetPageUptodate(page);
1397 swap = get_swap_page(page);
1402 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1403 * if it's not already there. Do it now before the page is
1404 * moved to swap cache, when its pagelock no longer protects
1405 * the inode from eviction. But don't unlock the mutex until
1406 * we've incremented swapped, because shmem_unuse_inode() will
1407 * prune a !swapped inode from the swaplist under this mutex.
1409 mutex_lock(&shmem_swaplist_mutex);
1410 if (list_empty(&info->swaplist))
1411 list_add(&info->swaplist, &shmem_swaplist);
1413 if (add_to_swap_cache(page, swap,
1414 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1416 spin_lock_irq(&info->lock);
1417 shmem_recalc_inode(inode);
1419 spin_unlock_irq(&info->lock);
1421 swap_shmem_alloc(swap);
1422 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1424 mutex_unlock(&shmem_swaplist_mutex);
1425 BUG_ON(page_mapped(page));
1426 swap_writepage(page, wbc);
1430 mutex_unlock(&shmem_swaplist_mutex);
1431 put_swap_page(page, swap);
1433 set_page_dirty(page);
1434 if (wbc->for_reclaim)
1435 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1440 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1441 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1445 if (!mpol || mpol->mode == MPOL_DEFAULT)
1446 return; /* show nothing */
1448 mpol_to_str(buffer, sizeof(buffer), mpol);
1450 seq_printf(seq, ",mpol=%s", buffer);
1453 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1455 struct mempolicy *mpol = NULL;
1457 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1458 mpol = sbinfo->mpol;
1460 spin_unlock(&sbinfo->stat_lock);
1464 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1465 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1468 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1472 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1474 #define vm_policy vm_private_data
1477 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1478 struct shmem_inode_info *info, pgoff_t index)
1480 /* Create a pseudo vma that just contains the policy */
1481 vma_init(vma, NULL);
1482 /* Bias interleave by inode number to distribute better across nodes */
1483 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1484 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1487 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1489 /* Drop reference taken by mpol_shared_policy_lookup() */
1490 mpol_cond_put(vma->vm_policy);
1493 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1494 struct shmem_inode_info *info, pgoff_t index)
1496 struct vm_area_struct pvma;
1498 struct vm_fault vmf = {
1502 shmem_pseudo_vma_init(&pvma, info, index);
1503 page = swap_cluster_readahead(swap, gfp, &vmf);
1504 shmem_pseudo_vma_destroy(&pvma);
1509 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1510 struct shmem_inode_info *info, pgoff_t index)
1512 struct vm_area_struct pvma;
1513 struct address_space *mapping = info->vfs_inode.i_mapping;
1517 hindex = round_down(index, HPAGE_PMD_NR);
1518 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1522 shmem_pseudo_vma_init(&pvma, info, hindex);
1523 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1524 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1525 shmem_pseudo_vma_destroy(&pvma);
1527 prep_transhuge_page(page);
1529 count_vm_event(THP_FILE_FALLBACK);
1533 static struct page *shmem_alloc_page(gfp_t gfp,
1534 struct shmem_inode_info *info, pgoff_t index)
1536 struct vm_area_struct pvma;
1539 shmem_pseudo_vma_init(&pvma, info, index);
1540 page = alloc_page_vma(gfp, &pvma, 0);
1541 shmem_pseudo_vma_destroy(&pvma);
1546 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1547 struct inode *inode,
1548 pgoff_t index, bool huge)
1550 struct shmem_inode_info *info = SHMEM_I(inode);
1555 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1557 nr = huge ? HPAGE_PMD_NR : 1;
1559 if (!shmem_inode_acct_block(inode, nr))
1563 page = shmem_alloc_hugepage(gfp, info, index);
1565 page = shmem_alloc_page(gfp, info, index);
1567 __SetPageLocked(page);
1568 __SetPageSwapBacked(page);
1573 shmem_inode_unacct_blocks(inode, nr);
1575 return ERR_PTR(err);
1579 * When a page is moved from swapcache to shmem filecache (either by the
1580 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1581 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1582 * ignorance of the mapping it belongs to. If that mapping has special
1583 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1584 * we may need to copy to a suitable page before moving to filecache.
1586 * In a future release, this may well be extended to respect cpuset and
1587 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1588 * but for now it is a simple matter of zone.
1590 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1592 return page_zonenum(page) > gfp_zone(gfp);
1595 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1596 struct shmem_inode_info *info, pgoff_t index)
1598 struct page *oldpage, *newpage;
1599 struct address_space *swap_mapping;
1605 entry.val = page_private(oldpage);
1606 swap_index = swp_offset(entry);
1607 swap_mapping = page_mapping(oldpage);
1610 * We have arrived here because our zones are constrained, so don't
1611 * limit chance of success by further cpuset and node constraints.
1613 gfp &= ~GFP_CONSTRAINT_MASK;
1614 newpage = shmem_alloc_page(gfp, info, index);
1619 copy_highpage(newpage, oldpage);
1620 flush_dcache_page(newpage);
1622 __SetPageLocked(newpage);
1623 __SetPageSwapBacked(newpage);
1624 SetPageUptodate(newpage);
1625 set_page_private(newpage, entry.val);
1626 SetPageSwapCache(newpage);
1629 * Our caller will very soon move newpage out of swapcache, but it's
1630 * a nice clean interface for us to replace oldpage by newpage there.
1632 xa_lock_irq(&swap_mapping->i_pages);
1633 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1635 mem_cgroup_migrate(oldpage, newpage);
1636 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1637 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1639 xa_unlock_irq(&swap_mapping->i_pages);
1641 if (unlikely(error)) {
1643 * Is this possible? I think not, now that our callers check
1644 * both PageSwapCache and page_private after getting page lock;
1645 * but be defensive. Reverse old to newpage for clear and free.
1649 lru_cache_add(newpage);
1653 ClearPageSwapCache(oldpage);
1654 set_page_private(oldpage, 0);
1656 unlock_page(oldpage);
1663 * Swap in the page pointed to by *pagep.
1664 * Caller has to make sure that *pagep contains a valid swapped page.
1665 * Returns 0 and the page in pagep if success. On failure, returns the
1666 * error code and NULL in *pagep.
1668 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1669 struct page **pagep, enum sgp_type sgp,
1670 gfp_t gfp, struct vm_area_struct *vma,
1671 vm_fault_t *fault_type)
1673 struct address_space *mapping = inode->i_mapping;
1674 struct shmem_inode_info *info = SHMEM_I(inode);
1675 struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
1680 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1681 swap = radix_to_swp_entry(*pagep);
1684 /* Look it up and read it in.. */
1685 page = lookup_swap_cache(swap, NULL, 0);
1687 /* Or update major stats only when swapin succeeds?? */
1689 *fault_type |= VM_FAULT_MAJOR;
1690 count_vm_event(PGMAJFAULT);
1691 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1693 /* Here we actually start the io */
1694 page = shmem_swapin(swap, gfp, info, index);
1701 /* We have to do this with page locked to prevent races */
1703 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1704 !shmem_confirm_swap(mapping, index, swap)) {
1708 if (!PageUptodate(page)) {
1712 wait_on_page_writeback(page);
1715 * Some architectures may have to restore extra metadata to the
1716 * physical page after reading from swap.
1718 arch_swap_restore(swap, page);
1720 if (shmem_should_replace_page(page, gfp)) {
1721 error = shmem_replace_page(&page, gfp, info, index);
1726 error = shmem_add_to_page_cache(page, mapping, index,
1727 swp_to_radix_entry(swap), gfp,
1732 spin_lock_irq(&info->lock);
1734 shmem_recalc_inode(inode);
1735 spin_unlock_irq(&info->lock);
1737 if (sgp == SGP_WRITE)
1738 mark_page_accessed(page);
1740 delete_from_swap_cache(page);
1741 set_page_dirty(page);
1747 if (!shmem_confirm_swap(mapping, index, swap))
1759 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1761 * If we allocate a new one we do not mark it dirty. That's up to the
1762 * vm. If we swap it in we mark it dirty since we also free the swap
1763 * entry since a page cannot live in both the swap and page cache.
1765 * vmf and fault_type are only supplied by shmem_fault:
1766 * otherwise they are NULL.
1768 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1769 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1770 struct vm_area_struct *vma, struct vm_fault *vmf,
1771 vm_fault_t *fault_type)
1773 struct address_space *mapping = inode->i_mapping;
1774 struct shmem_inode_info *info = SHMEM_I(inode);
1775 struct shmem_sb_info *sbinfo;
1776 struct mm_struct *charge_mm;
1778 enum sgp_type sgp_huge = sgp;
1779 pgoff_t hindex = index;
1784 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1786 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1789 if (sgp <= SGP_CACHE &&
1790 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1794 sbinfo = SHMEM_SB(inode->i_sb);
1795 charge_mm = vma ? vma->vm_mm : current->mm;
1797 page = pagecache_get_page(mapping, index,
1798 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1799 if (xa_is_value(page)) {
1800 error = shmem_swapin_page(inode, index, &page,
1801 sgp, gfp, vma, fault_type);
1802 if (error == -EEXIST)
1810 hindex = page->index;
1811 if (page && sgp == SGP_WRITE)
1812 mark_page_accessed(page);
1814 /* fallocated page? */
1815 if (page && !PageUptodate(page)) {
1816 if (sgp != SGP_READ)
1823 if (page || sgp == SGP_READ)
1827 * Fast cache lookup did not find it:
1828 * bring it back from swap or allocate.
1831 if (vma && userfaultfd_missing(vma)) {
1832 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1836 /* shmem_symlink() */
1837 if (!shmem_mapping(mapping))
1839 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1841 if (shmem_huge == SHMEM_HUGE_FORCE)
1843 switch (sbinfo->huge) {
1844 case SHMEM_HUGE_NEVER:
1846 case SHMEM_HUGE_WITHIN_SIZE: {
1850 off = round_up(index, HPAGE_PMD_NR);
1851 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1852 if (i_size >= HPAGE_PMD_SIZE &&
1853 i_size >> PAGE_SHIFT >= off)
1858 case SHMEM_HUGE_ADVISE:
1859 if (sgp_huge == SGP_HUGE)
1861 /* TODO: implement fadvise() hints */
1866 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1869 page = shmem_alloc_and_acct_page(gfp, inode,
1875 error = PTR_ERR(page);
1877 if (error != -ENOSPC)
1880 * Try to reclaim some space by splitting a huge page
1881 * beyond i_size on the filesystem.
1886 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1887 if (ret == SHRINK_STOP)
1895 if (PageTransHuge(page))
1896 hindex = round_down(index, HPAGE_PMD_NR);
1900 if (sgp == SGP_WRITE)
1901 __SetPageReferenced(page);
1903 error = shmem_add_to_page_cache(page, mapping, hindex,
1904 NULL, gfp & GFP_RECLAIM_MASK,
1908 lru_cache_add(page);
1910 spin_lock_irq(&info->lock);
1911 info->alloced += compound_nr(page);
1912 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1913 shmem_recalc_inode(inode);
1914 spin_unlock_irq(&info->lock);
1917 if (PageTransHuge(page) &&
1918 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1919 hindex + HPAGE_PMD_NR - 1) {
1921 * Part of the huge page is beyond i_size: subject
1922 * to shrink under memory pressure.
1924 spin_lock(&sbinfo->shrinklist_lock);
1926 * _careful to defend against unlocked access to
1927 * ->shrink_list in shmem_unused_huge_shrink()
1929 if (list_empty_careful(&info->shrinklist)) {
1930 list_add_tail(&info->shrinklist,
1931 &sbinfo->shrinklist);
1932 sbinfo->shrinklist_len++;
1934 spin_unlock(&sbinfo->shrinklist_lock);
1938 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1940 if (sgp == SGP_FALLOC)
1944 * Let SGP_WRITE caller clear ends if write does not fill page;
1945 * but SGP_FALLOC on a page fallocated earlier must initialize
1946 * it now, lest undo on failure cancel our earlier guarantee.
1948 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1951 for (i = 0; i < compound_nr(page); i++) {
1952 clear_highpage(page + i);
1953 flush_dcache_page(page + i);
1955 SetPageUptodate(page);
1958 /* Perhaps the file has been truncated since we checked */
1959 if (sgp <= SGP_CACHE &&
1960 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1962 ClearPageDirty(page);
1963 delete_from_page_cache(page);
1964 spin_lock_irq(&info->lock);
1965 shmem_recalc_inode(inode);
1966 spin_unlock_irq(&info->lock);
1972 *pagep = page + index - hindex;
1979 shmem_inode_unacct_blocks(inode, compound_nr(page));
1981 if (PageTransHuge(page)) {
1991 if (error == -ENOSPC && !once++) {
1992 spin_lock_irq(&info->lock);
1993 shmem_recalc_inode(inode);
1994 spin_unlock_irq(&info->lock);
1997 if (error == -EEXIST)
2003 * This is like autoremove_wake_function, but it removes the wait queue
2004 * entry unconditionally - even if something else had already woken the
2007 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2009 int ret = default_wake_function(wait, mode, sync, key);
2010 list_del_init(&wait->entry);
2014 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2016 struct vm_area_struct *vma = vmf->vma;
2017 struct inode *inode = file_inode(vma->vm_file);
2018 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2021 vm_fault_t ret = VM_FAULT_LOCKED;
2024 * Trinity finds that probing a hole which tmpfs is punching can
2025 * prevent the hole-punch from ever completing: which in turn
2026 * locks writers out with its hold on i_mutex. So refrain from
2027 * faulting pages into the hole while it's being punched. Although
2028 * shmem_undo_range() does remove the additions, it may be unable to
2029 * keep up, as each new page needs its own unmap_mapping_range() call,
2030 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2032 * It does not matter if we sometimes reach this check just before the
2033 * hole-punch begins, so that one fault then races with the punch:
2034 * we just need to make racing faults a rare case.
2036 * The implementation below would be much simpler if we just used a
2037 * standard mutex or completion: but we cannot take i_mutex in fault,
2038 * and bloating every shmem inode for this unlikely case would be sad.
2040 if (unlikely(inode->i_private)) {
2041 struct shmem_falloc *shmem_falloc;
2043 spin_lock(&inode->i_lock);
2044 shmem_falloc = inode->i_private;
2046 shmem_falloc->waitq &&
2047 vmf->pgoff >= shmem_falloc->start &&
2048 vmf->pgoff < shmem_falloc->next) {
2050 wait_queue_head_t *shmem_falloc_waitq;
2051 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2053 ret = VM_FAULT_NOPAGE;
2054 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2056 ret = VM_FAULT_RETRY;
2058 shmem_falloc_waitq = shmem_falloc->waitq;
2059 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2060 TASK_UNINTERRUPTIBLE);
2061 spin_unlock(&inode->i_lock);
2065 * shmem_falloc_waitq points into the shmem_fallocate()
2066 * stack of the hole-punching task: shmem_falloc_waitq
2067 * is usually invalid by the time we reach here, but
2068 * finish_wait() does not dereference it in that case;
2069 * though i_lock needed lest racing with wake_up_all().
2071 spin_lock(&inode->i_lock);
2072 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2073 spin_unlock(&inode->i_lock);
2079 spin_unlock(&inode->i_lock);
2084 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2085 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2087 else if (vma->vm_flags & VM_HUGEPAGE)
2090 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2091 gfp, vma, vmf, &ret);
2093 return vmf_error(err);
2097 unsigned long shmem_get_unmapped_area(struct file *file,
2098 unsigned long uaddr, unsigned long len,
2099 unsigned long pgoff, unsigned long flags)
2101 unsigned long (*get_area)(struct file *,
2102 unsigned long, unsigned long, unsigned long, unsigned long);
2104 unsigned long offset;
2105 unsigned long inflated_len;
2106 unsigned long inflated_addr;
2107 unsigned long inflated_offset;
2109 if (len > TASK_SIZE)
2112 get_area = current->mm->get_unmapped_area;
2113 addr = get_area(file, uaddr, len, pgoff, flags);
2115 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2117 if (IS_ERR_VALUE(addr))
2119 if (addr & ~PAGE_MASK)
2121 if (addr > TASK_SIZE - len)
2124 if (shmem_huge == SHMEM_HUGE_DENY)
2126 if (len < HPAGE_PMD_SIZE)
2128 if (flags & MAP_FIXED)
2131 * Our priority is to support MAP_SHARED mapped hugely;
2132 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2133 * But if caller specified an address hint and we allocated area there
2134 * successfully, respect that as before.
2139 if (shmem_huge != SHMEM_HUGE_FORCE) {
2140 struct super_block *sb;
2143 VM_BUG_ON(file->f_op != &shmem_file_operations);
2144 sb = file_inode(file)->i_sb;
2147 * Called directly from mm/mmap.c, or drivers/char/mem.c
2148 * for "/dev/zero", to create a shared anonymous object.
2150 if (IS_ERR(shm_mnt))
2152 sb = shm_mnt->mnt_sb;
2154 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2158 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2159 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2161 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2164 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2165 if (inflated_len > TASK_SIZE)
2167 if (inflated_len < len)
2170 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2171 if (IS_ERR_VALUE(inflated_addr))
2173 if (inflated_addr & ~PAGE_MASK)
2176 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2177 inflated_addr += offset - inflated_offset;
2178 if (inflated_offset > offset)
2179 inflated_addr += HPAGE_PMD_SIZE;
2181 if (inflated_addr > TASK_SIZE - len)
2183 return inflated_addr;
2187 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2189 struct inode *inode = file_inode(vma->vm_file);
2190 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2193 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2196 struct inode *inode = file_inode(vma->vm_file);
2199 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2200 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2204 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2206 struct inode *inode = file_inode(file);
2207 struct shmem_inode_info *info = SHMEM_I(inode);
2208 int retval = -ENOMEM;
2211 * What serializes the accesses to info->flags?
2212 * ipc_lock_object() when called from shmctl_do_lock(),
2213 * no serialization needed when called from shm_destroy().
2215 if (lock && !(info->flags & VM_LOCKED)) {
2216 if (!user_shm_lock(inode->i_size, user))
2218 info->flags |= VM_LOCKED;
2219 mapping_set_unevictable(file->f_mapping);
2221 if (!lock && (info->flags & VM_LOCKED) && user) {
2222 user_shm_unlock(inode->i_size, user);
2223 info->flags &= ~VM_LOCKED;
2224 mapping_clear_unevictable(file->f_mapping);
2232 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2234 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2236 if (info->seals & F_SEAL_FUTURE_WRITE) {
2238 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2239 * "future write" seal active.
2241 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2245 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2246 * MAP_SHARED and read-only, take care to not allow mprotect to
2247 * revert protections on such mappings. Do this only for shared
2248 * mappings. For private mappings, don't need to mask
2249 * VM_MAYWRITE as we still want them to be COW-writable.
2251 if (vma->vm_flags & VM_SHARED)
2252 vma->vm_flags &= ~(VM_MAYWRITE);
2255 /* arm64 - allow memory tagging on RAM-based files */
2256 vma->vm_flags |= VM_MTE_ALLOWED;
2258 file_accessed(file);
2259 vma->vm_ops = &shmem_vm_ops;
2260 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2261 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2262 (vma->vm_end & HPAGE_PMD_MASK)) {
2263 khugepaged_enter(vma, vma->vm_flags);
2268 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2269 umode_t mode, dev_t dev, unsigned long flags)
2271 struct inode *inode;
2272 struct shmem_inode_info *info;
2273 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2276 if (shmem_reserve_inode(sb, &ino))
2279 inode = new_inode(sb);
2282 inode_init_owner(&init_user_ns, inode, dir, mode);
2283 inode->i_blocks = 0;
2284 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2285 inode->i_generation = prandom_u32();
2286 info = SHMEM_I(inode);
2287 memset(info, 0, (char *)inode - (char *)info);
2288 spin_lock_init(&info->lock);
2289 atomic_set(&info->stop_eviction, 0);
2290 info->seals = F_SEAL_SEAL;
2291 info->flags = flags & VM_NORESERVE;
2292 INIT_LIST_HEAD(&info->shrinklist);
2293 INIT_LIST_HEAD(&info->swaplist);
2294 simple_xattrs_init(&info->xattrs);
2295 cache_no_acl(inode);
2297 switch (mode & S_IFMT) {
2299 inode->i_op = &shmem_special_inode_operations;
2300 init_special_inode(inode, mode, dev);
2303 inode->i_mapping->a_ops = &shmem_aops;
2304 inode->i_op = &shmem_inode_operations;
2305 inode->i_fop = &shmem_file_operations;
2306 mpol_shared_policy_init(&info->policy,
2307 shmem_get_sbmpol(sbinfo));
2311 /* Some things misbehave if size == 0 on a directory */
2312 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2313 inode->i_op = &shmem_dir_inode_operations;
2314 inode->i_fop = &simple_dir_operations;
2318 * Must not load anything in the rbtree,
2319 * mpol_free_shared_policy will not be called.
2321 mpol_shared_policy_init(&info->policy, NULL);
2325 lockdep_annotate_inode_mutex_key(inode);
2327 shmem_free_inode(sb);
2331 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2333 struct vm_area_struct *dst_vma,
2334 unsigned long dst_addr,
2335 unsigned long src_addr,
2337 struct page **pagep)
2339 struct inode *inode = file_inode(dst_vma->vm_file);
2340 struct shmem_inode_info *info = SHMEM_I(inode);
2341 struct address_space *mapping = inode->i_mapping;
2342 gfp_t gfp = mapping_gfp_mask(mapping);
2343 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2347 pte_t _dst_pte, *dst_pte;
2349 pgoff_t offset, max_off;
2352 if (!shmem_inode_acct_block(inode, 1))
2356 page = shmem_alloc_page(gfp, info, pgoff);
2358 goto out_unacct_blocks;
2360 if (!zeropage) { /* mcopy_atomic */
2361 page_kaddr = kmap_atomic(page);
2362 ret = copy_from_user(page_kaddr,
2363 (const void __user *)src_addr,
2365 kunmap_atomic(page_kaddr);
2367 /* fallback to copy_from_user outside mmap_lock */
2368 if (unlikely(ret)) {
2370 shmem_inode_unacct_blocks(inode, 1);
2371 /* don't free the page */
2374 } else { /* mfill_zeropage_atomic */
2375 clear_highpage(page);
2382 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2383 __SetPageLocked(page);
2384 __SetPageSwapBacked(page);
2385 __SetPageUptodate(page);
2388 offset = linear_page_index(dst_vma, dst_addr);
2389 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2390 if (unlikely(offset >= max_off))
2393 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2394 gfp & GFP_RECLAIM_MASK, dst_mm);
2398 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2399 if (dst_vma->vm_flags & VM_WRITE)
2400 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2403 * We don't set the pte dirty if the vma has no
2404 * VM_WRITE permission, so mark the page dirty or it
2405 * could be freed from under us. We could do it
2406 * unconditionally before unlock_page(), but doing it
2407 * only if VM_WRITE is not set is faster.
2409 set_page_dirty(page);
2412 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2415 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2416 if (unlikely(offset >= max_off))
2417 goto out_release_unlock;
2420 if (!pte_none(*dst_pte))
2421 goto out_release_unlock;
2423 lru_cache_add(page);
2425 spin_lock_irq(&info->lock);
2427 inode->i_blocks += BLOCKS_PER_PAGE;
2428 shmem_recalc_inode(inode);
2429 spin_unlock_irq(&info->lock);
2431 inc_mm_counter(dst_mm, mm_counter_file(page));
2432 page_add_file_rmap(page, false);
2433 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2435 /* No need to invalidate - it was non-present before */
2436 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2437 pte_unmap_unlock(dst_pte, ptl);
2443 pte_unmap_unlock(dst_pte, ptl);
2444 ClearPageDirty(page);
2445 delete_from_page_cache(page);
2450 shmem_inode_unacct_blocks(inode, 1);
2454 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2456 struct vm_area_struct *dst_vma,
2457 unsigned long dst_addr,
2458 unsigned long src_addr,
2459 struct page **pagep)
2461 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2462 dst_addr, src_addr, false, pagep);
2465 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2467 struct vm_area_struct *dst_vma,
2468 unsigned long dst_addr)
2470 struct page *page = NULL;
2472 return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2473 dst_addr, 0, true, &page);
2477 static const struct inode_operations shmem_symlink_inode_operations;
2478 static const struct inode_operations shmem_short_symlink_operations;
2480 #ifdef CONFIG_TMPFS_XATTR
2481 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2483 #define shmem_initxattrs NULL
2487 shmem_write_begin(struct file *file, struct address_space *mapping,
2488 loff_t pos, unsigned len, unsigned flags,
2489 struct page **pagep, void **fsdata)
2491 struct inode *inode = mapping->host;
2492 struct shmem_inode_info *info = SHMEM_I(inode);
2493 pgoff_t index = pos >> PAGE_SHIFT;
2495 /* i_mutex is held by caller */
2496 if (unlikely(info->seals & (F_SEAL_GROW |
2497 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2498 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2500 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2504 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2508 shmem_write_end(struct file *file, struct address_space *mapping,
2509 loff_t pos, unsigned len, unsigned copied,
2510 struct page *page, void *fsdata)
2512 struct inode *inode = mapping->host;
2514 if (pos + copied > inode->i_size)
2515 i_size_write(inode, pos + copied);
2517 if (!PageUptodate(page)) {
2518 struct page *head = compound_head(page);
2519 if (PageTransCompound(page)) {
2522 for (i = 0; i < HPAGE_PMD_NR; i++) {
2523 if (head + i == page)
2525 clear_highpage(head + i);
2526 flush_dcache_page(head + i);
2529 if (copied < PAGE_SIZE) {
2530 unsigned from = pos & (PAGE_SIZE - 1);
2531 zero_user_segments(page, 0, from,
2532 from + copied, PAGE_SIZE);
2534 SetPageUptodate(head);
2536 set_page_dirty(page);
2543 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2545 struct file *file = iocb->ki_filp;
2546 struct inode *inode = file_inode(file);
2547 struct address_space *mapping = inode->i_mapping;
2549 unsigned long offset;
2550 enum sgp_type sgp = SGP_READ;
2553 loff_t *ppos = &iocb->ki_pos;
2556 * Might this read be for a stacking filesystem? Then when reading
2557 * holes of a sparse file, we actually need to allocate those pages,
2558 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2560 if (!iter_is_iovec(to))
2563 index = *ppos >> PAGE_SHIFT;
2564 offset = *ppos & ~PAGE_MASK;
2567 struct page *page = NULL;
2569 unsigned long nr, ret;
2570 loff_t i_size = i_size_read(inode);
2572 end_index = i_size >> PAGE_SHIFT;
2573 if (index > end_index)
2575 if (index == end_index) {
2576 nr = i_size & ~PAGE_MASK;
2581 error = shmem_getpage(inode, index, &page, sgp);
2583 if (error == -EINVAL)
2588 if (sgp == SGP_CACHE)
2589 set_page_dirty(page);
2594 * We must evaluate after, since reads (unlike writes)
2595 * are called without i_mutex protection against truncate
2598 i_size = i_size_read(inode);
2599 end_index = i_size >> PAGE_SHIFT;
2600 if (index == end_index) {
2601 nr = i_size & ~PAGE_MASK;
2612 * If users can be writing to this page using arbitrary
2613 * virtual addresses, take care about potential aliasing
2614 * before reading the page on the kernel side.
2616 if (mapping_writably_mapped(mapping))
2617 flush_dcache_page(page);
2619 * Mark the page accessed if we read the beginning.
2622 mark_page_accessed(page);
2624 page = ZERO_PAGE(0);
2629 * Ok, we have the page, and it's up-to-date, so
2630 * now we can copy it to user space...
2632 ret = copy_page_to_iter(page, offset, nr, to);
2635 index += offset >> PAGE_SHIFT;
2636 offset &= ~PAGE_MASK;
2639 if (!iov_iter_count(to))
2648 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2649 file_accessed(file);
2650 return retval ? retval : error;
2653 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2655 struct address_space *mapping = file->f_mapping;
2656 struct inode *inode = mapping->host;
2658 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2659 return generic_file_llseek_size(file, offset, whence,
2660 MAX_LFS_FILESIZE, i_size_read(inode));
2665 /* We're holding i_mutex so we can access i_size directly */
2666 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2668 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2669 inode_unlock(inode);
2673 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2676 struct inode *inode = file_inode(file);
2677 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2678 struct shmem_inode_info *info = SHMEM_I(inode);
2679 struct shmem_falloc shmem_falloc;
2680 pgoff_t start, index, end;
2683 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2688 if (mode & FALLOC_FL_PUNCH_HOLE) {
2689 struct address_space *mapping = file->f_mapping;
2690 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2691 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2692 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2694 /* protected by i_mutex */
2695 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2700 shmem_falloc.waitq = &shmem_falloc_waitq;
2701 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2702 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2703 spin_lock(&inode->i_lock);
2704 inode->i_private = &shmem_falloc;
2705 spin_unlock(&inode->i_lock);
2707 if ((u64)unmap_end > (u64)unmap_start)
2708 unmap_mapping_range(mapping, unmap_start,
2709 1 + unmap_end - unmap_start, 0);
2710 shmem_truncate_range(inode, offset, offset + len - 1);
2711 /* No need to unmap again: hole-punching leaves COWed pages */
2713 spin_lock(&inode->i_lock);
2714 inode->i_private = NULL;
2715 wake_up_all(&shmem_falloc_waitq);
2716 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2717 spin_unlock(&inode->i_lock);
2722 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2723 error = inode_newsize_ok(inode, offset + len);
2727 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2732 start = offset >> PAGE_SHIFT;
2733 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2734 /* Try to avoid a swapstorm if len is impossible to satisfy */
2735 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2740 shmem_falloc.waitq = NULL;
2741 shmem_falloc.start = start;
2742 shmem_falloc.next = start;
2743 shmem_falloc.nr_falloced = 0;
2744 shmem_falloc.nr_unswapped = 0;
2745 spin_lock(&inode->i_lock);
2746 inode->i_private = &shmem_falloc;
2747 spin_unlock(&inode->i_lock);
2749 for (index = start; index < end; index++) {
2753 * Good, the fallocate(2) manpage permits EINTR: we may have
2754 * been interrupted because we are using up too much memory.
2756 if (signal_pending(current))
2758 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2761 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2763 /* Remove the !PageUptodate pages we added */
2764 if (index > start) {
2765 shmem_undo_range(inode,
2766 (loff_t)start << PAGE_SHIFT,
2767 ((loff_t)index << PAGE_SHIFT) - 1, true);
2773 * Inform shmem_writepage() how far we have reached.
2774 * No need for lock or barrier: we have the page lock.
2776 shmem_falloc.next++;
2777 if (!PageUptodate(page))
2778 shmem_falloc.nr_falloced++;
2781 * If !PageUptodate, leave it that way so that freeable pages
2782 * can be recognized if we need to rollback on error later.
2783 * But set_page_dirty so that memory pressure will swap rather
2784 * than free the pages we are allocating (and SGP_CACHE pages
2785 * might still be clean: we now need to mark those dirty too).
2787 set_page_dirty(page);
2793 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2794 i_size_write(inode, offset + len);
2795 inode->i_ctime = current_time(inode);
2797 spin_lock(&inode->i_lock);
2798 inode->i_private = NULL;
2799 spin_unlock(&inode->i_lock);
2801 inode_unlock(inode);
2805 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2807 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2809 buf->f_type = TMPFS_MAGIC;
2810 buf->f_bsize = PAGE_SIZE;
2811 buf->f_namelen = NAME_MAX;
2812 if (sbinfo->max_blocks) {
2813 buf->f_blocks = sbinfo->max_blocks;
2815 buf->f_bfree = sbinfo->max_blocks -
2816 percpu_counter_sum(&sbinfo->used_blocks);
2818 if (sbinfo->max_inodes) {
2819 buf->f_files = sbinfo->max_inodes;
2820 buf->f_ffree = sbinfo->free_inodes;
2822 /* else leave those fields 0 like simple_statfs */
2827 * File creation. Allocate an inode, and we're done..
2830 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2831 struct dentry *dentry, umode_t mode, dev_t dev)
2833 struct inode *inode;
2834 int error = -ENOSPC;
2836 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2838 error = simple_acl_create(dir, inode);
2841 error = security_inode_init_security(inode, dir,
2843 shmem_initxattrs, NULL);
2844 if (error && error != -EOPNOTSUPP)
2848 dir->i_size += BOGO_DIRENT_SIZE;
2849 dir->i_ctime = dir->i_mtime = current_time(dir);
2850 d_instantiate(dentry, inode);
2851 dget(dentry); /* Extra count - pin the dentry in core */
2860 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2861 struct dentry *dentry, umode_t mode)
2863 struct inode *inode;
2864 int error = -ENOSPC;
2866 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2868 error = security_inode_init_security(inode, dir,
2870 shmem_initxattrs, NULL);
2871 if (error && error != -EOPNOTSUPP)
2873 error = simple_acl_create(dir, inode);
2876 d_tmpfile(dentry, inode);
2884 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2885 struct dentry *dentry, umode_t mode)
2889 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2890 mode | S_IFDIR, 0)))
2896 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2897 struct dentry *dentry, umode_t mode, bool excl)
2899 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2905 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2907 struct inode *inode = d_inode(old_dentry);
2911 * No ordinary (disk based) filesystem counts links as inodes;
2912 * but each new link needs a new dentry, pinning lowmem, and
2913 * tmpfs dentries cannot be pruned until they are unlinked.
2914 * But if an O_TMPFILE file is linked into the tmpfs, the
2915 * first link must skip that, to get the accounting right.
2917 if (inode->i_nlink) {
2918 ret = shmem_reserve_inode(inode->i_sb, NULL);
2923 dir->i_size += BOGO_DIRENT_SIZE;
2924 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2926 ihold(inode); /* New dentry reference */
2927 dget(dentry); /* Extra pinning count for the created dentry */
2928 d_instantiate(dentry, inode);
2933 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2935 struct inode *inode = d_inode(dentry);
2937 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2938 shmem_free_inode(inode->i_sb);
2940 dir->i_size -= BOGO_DIRENT_SIZE;
2941 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2943 dput(dentry); /* Undo the count from "create" - this does all the work */
2947 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2949 if (!simple_empty(dentry))
2952 drop_nlink(d_inode(dentry));
2954 return shmem_unlink(dir, dentry);
2957 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2959 bool old_is_dir = d_is_dir(old_dentry);
2960 bool new_is_dir = d_is_dir(new_dentry);
2962 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2964 drop_nlink(old_dir);
2967 drop_nlink(new_dir);
2971 old_dir->i_ctime = old_dir->i_mtime =
2972 new_dir->i_ctime = new_dir->i_mtime =
2973 d_inode(old_dentry)->i_ctime =
2974 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2979 static int shmem_whiteout(struct user_namespace *mnt_userns,
2980 struct inode *old_dir, struct dentry *old_dentry)
2982 struct dentry *whiteout;
2985 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2989 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2990 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2996 * Cheat and hash the whiteout while the old dentry is still in
2997 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2999 * d_lookup() will consistently find one of them at this point,
3000 * not sure which one, but that isn't even important.
3007 * The VFS layer already does all the dentry stuff for rename,
3008 * we just have to decrement the usage count for the target if
3009 * it exists so that the VFS layer correctly free's it when it
3012 static int shmem_rename2(struct user_namespace *mnt_userns,
3013 struct inode *old_dir, struct dentry *old_dentry,
3014 struct inode *new_dir, struct dentry *new_dentry,
3017 struct inode *inode = d_inode(old_dentry);
3018 int they_are_dirs = S_ISDIR(inode->i_mode);
3020 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3023 if (flags & RENAME_EXCHANGE)
3024 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3026 if (!simple_empty(new_dentry))
3029 if (flags & RENAME_WHITEOUT) {
3032 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3037 if (d_really_is_positive(new_dentry)) {
3038 (void) shmem_unlink(new_dir, new_dentry);
3039 if (they_are_dirs) {
3040 drop_nlink(d_inode(new_dentry));
3041 drop_nlink(old_dir);
3043 } else if (they_are_dirs) {
3044 drop_nlink(old_dir);
3048 old_dir->i_size -= BOGO_DIRENT_SIZE;
3049 new_dir->i_size += BOGO_DIRENT_SIZE;
3050 old_dir->i_ctime = old_dir->i_mtime =
3051 new_dir->i_ctime = new_dir->i_mtime =
3052 inode->i_ctime = current_time(old_dir);
3056 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3057 struct dentry *dentry, const char *symname)
3061 struct inode *inode;
3064 len = strlen(symname) + 1;
3065 if (len > PAGE_SIZE)
3066 return -ENAMETOOLONG;
3068 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3073 error = security_inode_init_security(inode, dir, &dentry->d_name,
3074 shmem_initxattrs, NULL);
3075 if (error && error != -EOPNOTSUPP) {
3080 inode->i_size = len-1;
3081 if (len <= SHORT_SYMLINK_LEN) {
3082 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3083 if (!inode->i_link) {
3087 inode->i_op = &shmem_short_symlink_operations;
3089 inode_nohighmem(inode);
3090 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3095 inode->i_mapping->a_ops = &shmem_aops;
3096 inode->i_op = &shmem_symlink_inode_operations;
3097 memcpy(page_address(page), symname, len);
3098 SetPageUptodate(page);
3099 set_page_dirty(page);
3103 dir->i_size += BOGO_DIRENT_SIZE;
3104 dir->i_ctime = dir->i_mtime = current_time(dir);
3105 d_instantiate(dentry, inode);
3110 static void shmem_put_link(void *arg)
3112 mark_page_accessed(arg);
3116 static const char *shmem_get_link(struct dentry *dentry,
3117 struct inode *inode,
3118 struct delayed_call *done)
3120 struct page *page = NULL;
3123 page = find_get_page(inode->i_mapping, 0);
3125 return ERR_PTR(-ECHILD);
3126 if (!PageUptodate(page)) {
3128 return ERR_PTR(-ECHILD);
3131 error = shmem_getpage(inode, 0, &page, SGP_READ);
3133 return ERR_PTR(error);
3136 set_delayed_call(done, shmem_put_link, page);
3137 return page_address(page);
3140 #ifdef CONFIG_TMPFS_XATTR
3142 * Superblocks without xattr inode operations may get some security.* xattr
3143 * support from the LSM "for free". As soon as we have any other xattrs
3144 * like ACLs, we also need to implement the security.* handlers at
3145 * filesystem level, though.
3149 * Callback for security_inode_init_security() for acquiring xattrs.
3151 static int shmem_initxattrs(struct inode *inode,
3152 const struct xattr *xattr_array,
3155 struct shmem_inode_info *info = SHMEM_I(inode);
3156 const struct xattr *xattr;
3157 struct simple_xattr *new_xattr;
3160 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3161 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3165 len = strlen(xattr->name) + 1;
3166 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3168 if (!new_xattr->name) {
3173 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3174 XATTR_SECURITY_PREFIX_LEN);
3175 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3178 simple_xattr_list_add(&info->xattrs, new_xattr);
3184 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3185 struct dentry *unused, struct inode *inode,
3186 const char *name, void *buffer, size_t size)
3188 struct shmem_inode_info *info = SHMEM_I(inode);
3190 name = xattr_full_name(handler, name);
3191 return simple_xattr_get(&info->xattrs, name, buffer, size);
3194 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3195 struct user_namespace *mnt_userns,
3196 struct dentry *unused, struct inode *inode,
3197 const char *name, const void *value,
3198 size_t size, int flags)
3200 struct shmem_inode_info *info = SHMEM_I(inode);
3202 name = xattr_full_name(handler, name);
3203 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3206 static const struct xattr_handler shmem_security_xattr_handler = {
3207 .prefix = XATTR_SECURITY_PREFIX,
3208 .get = shmem_xattr_handler_get,
3209 .set = shmem_xattr_handler_set,
3212 static const struct xattr_handler shmem_trusted_xattr_handler = {
3213 .prefix = XATTR_TRUSTED_PREFIX,
3214 .get = shmem_xattr_handler_get,
3215 .set = shmem_xattr_handler_set,
3218 static const struct xattr_handler *shmem_xattr_handlers[] = {
3219 #ifdef CONFIG_TMPFS_POSIX_ACL
3220 &posix_acl_access_xattr_handler,
3221 &posix_acl_default_xattr_handler,
3223 &shmem_security_xattr_handler,
3224 &shmem_trusted_xattr_handler,
3228 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3230 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3231 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3233 #endif /* CONFIG_TMPFS_XATTR */
3235 static const struct inode_operations shmem_short_symlink_operations = {
3236 .get_link = simple_get_link,
3237 #ifdef CONFIG_TMPFS_XATTR
3238 .listxattr = shmem_listxattr,
3242 static const struct inode_operations shmem_symlink_inode_operations = {
3243 .get_link = shmem_get_link,
3244 #ifdef CONFIG_TMPFS_XATTR
3245 .listxattr = shmem_listxattr,
3249 static struct dentry *shmem_get_parent(struct dentry *child)
3251 return ERR_PTR(-ESTALE);
3254 static int shmem_match(struct inode *ino, void *vfh)
3258 inum = (inum << 32) | fh[1];
3259 return ino->i_ino == inum && fh[0] == ino->i_generation;
3262 /* Find any alias of inode, but prefer a hashed alias */
3263 static struct dentry *shmem_find_alias(struct inode *inode)
3265 struct dentry *alias = d_find_alias(inode);
3267 return alias ?: d_find_any_alias(inode);
3271 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3272 struct fid *fid, int fh_len, int fh_type)
3274 struct inode *inode;
3275 struct dentry *dentry = NULL;
3282 inum = (inum << 32) | fid->raw[1];
3284 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3285 shmem_match, fid->raw);
3287 dentry = shmem_find_alias(inode);
3294 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3295 struct inode *parent)
3299 return FILEID_INVALID;
3302 if (inode_unhashed(inode)) {
3303 /* Unfortunately insert_inode_hash is not idempotent,
3304 * so as we hash inodes here rather than at creation
3305 * time, we need a lock to ensure we only try
3308 static DEFINE_SPINLOCK(lock);
3310 if (inode_unhashed(inode))
3311 __insert_inode_hash(inode,
3312 inode->i_ino + inode->i_generation);
3316 fh[0] = inode->i_generation;
3317 fh[1] = inode->i_ino;
3318 fh[2] = ((__u64)inode->i_ino) >> 32;
3324 static const struct export_operations shmem_export_ops = {
3325 .get_parent = shmem_get_parent,
3326 .encode_fh = shmem_encode_fh,
3327 .fh_to_dentry = shmem_fh_to_dentry,
3343 static const struct constant_table shmem_param_enums_huge[] = {
3344 {"never", SHMEM_HUGE_NEVER },
3345 {"always", SHMEM_HUGE_ALWAYS },
3346 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3347 {"advise", SHMEM_HUGE_ADVISE },
3351 const struct fs_parameter_spec shmem_fs_parameters[] = {
3352 fsparam_u32 ("gid", Opt_gid),
3353 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3354 fsparam_u32oct("mode", Opt_mode),
3355 fsparam_string("mpol", Opt_mpol),
3356 fsparam_string("nr_blocks", Opt_nr_blocks),
3357 fsparam_string("nr_inodes", Opt_nr_inodes),
3358 fsparam_string("size", Opt_size),
3359 fsparam_u32 ("uid", Opt_uid),
3360 fsparam_flag ("inode32", Opt_inode32),
3361 fsparam_flag ("inode64", Opt_inode64),
3365 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3367 struct shmem_options *ctx = fc->fs_private;
3368 struct fs_parse_result result;
3369 unsigned long long size;
3373 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3379 size = memparse(param->string, &rest);
3381 size <<= PAGE_SHIFT;
3382 size *= totalram_pages();
3388 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3389 ctx->seen |= SHMEM_SEEN_BLOCKS;
3392 ctx->blocks = memparse(param->string, &rest);
3395 ctx->seen |= SHMEM_SEEN_BLOCKS;
3398 ctx->inodes = memparse(param->string, &rest);
3401 ctx->seen |= SHMEM_SEEN_INODES;
3404 ctx->mode = result.uint_32 & 07777;
3407 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3408 if (!uid_valid(ctx->uid))
3412 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3413 if (!gid_valid(ctx->gid))
3417 ctx->huge = result.uint_32;
3418 if (ctx->huge != SHMEM_HUGE_NEVER &&
3419 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3420 has_transparent_hugepage()))
3421 goto unsupported_parameter;
3422 ctx->seen |= SHMEM_SEEN_HUGE;
3425 if (IS_ENABLED(CONFIG_NUMA)) {
3426 mpol_put(ctx->mpol);
3428 if (mpol_parse_str(param->string, &ctx->mpol))
3432 goto unsupported_parameter;
3434 ctx->full_inums = false;
3435 ctx->seen |= SHMEM_SEEN_INUMS;
3438 if (sizeof(ino_t) < 8) {
3440 "Cannot use inode64 with <64bit inums in kernel\n");
3442 ctx->full_inums = true;
3443 ctx->seen |= SHMEM_SEEN_INUMS;
3448 unsupported_parameter:
3449 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3451 return invalfc(fc, "Bad value for '%s'", param->key);
3454 static int shmem_parse_options(struct fs_context *fc, void *data)
3456 char *options = data;
3459 int err = security_sb_eat_lsm_opts(options, &fc->security);
3464 while (options != NULL) {
3465 char *this_char = options;
3468 * NUL-terminate this option: unfortunately,
3469 * mount options form a comma-separated list,
3470 * but mpol's nodelist may also contain commas.
3472 options = strchr(options, ',');
3473 if (options == NULL)
3476 if (!isdigit(*options)) {
3482 char *value = strchr(this_char,'=');
3488 len = strlen(value);
3490 err = vfs_parse_fs_string(fc, this_char, value, len);
3499 * Reconfigure a shmem filesystem.
3501 * Note that we disallow change from limited->unlimited blocks/inodes while any
3502 * are in use; but we must separately disallow unlimited->limited, because in
3503 * that case we have no record of how much is already in use.
3505 static int shmem_reconfigure(struct fs_context *fc)
3507 struct shmem_options *ctx = fc->fs_private;
3508 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3509 unsigned long inodes;
3512 spin_lock(&sbinfo->stat_lock);
3513 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3514 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3515 if (!sbinfo->max_blocks) {
3516 err = "Cannot retroactively limit size";
3519 if (percpu_counter_compare(&sbinfo->used_blocks,
3521 err = "Too small a size for current use";
3525 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3526 if (!sbinfo->max_inodes) {
3527 err = "Cannot retroactively limit inodes";
3530 if (ctx->inodes < inodes) {
3531 err = "Too few inodes for current use";
3536 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3537 sbinfo->next_ino > UINT_MAX) {
3538 err = "Current inum too high to switch to 32-bit inums";
3542 if (ctx->seen & SHMEM_SEEN_HUGE)
3543 sbinfo->huge = ctx->huge;
3544 if (ctx->seen & SHMEM_SEEN_INUMS)
3545 sbinfo->full_inums = ctx->full_inums;
3546 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3547 sbinfo->max_blocks = ctx->blocks;
3548 if (ctx->seen & SHMEM_SEEN_INODES) {
3549 sbinfo->max_inodes = ctx->inodes;
3550 sbinfo->free_inodes = ctx->inodes - inodes;
3554 * Preserve previous mempolicy unless mpol remount option was specified.
3557 mpol_put(sbinfo->mpol);
3558 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3561 spin_unlock(&sbinfo->stat_lock);
3564 spin_unlock(&sbinfo->stat_lock);
3565 return invalfc(fc, "%s", err);
3568 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3570 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3572 if (sbinfo->max_blocks != shmem_default_max_blocks())
3573 seq_printf(seq, ",size=%luk",
3574 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3575 if (sbinfo->max_inodes != shmem_default_max_inodes())
3576 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3577 if (sbinfo->mode != (0777 | S_ISVTX))
3578 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3579 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3580 seq_printf(seq, ",uid=%u",
3581 from_kuid_munged(&init_user_ns, sbinfo->uid));
3582 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3583 seq_printf(seq, ",gid=%u",
3584 from_kgid_munged(&init_user_ns, sbinfo->gid));
3587 * Showing inode{64,32} might be useful even if it's the system default,
3588 * since then people don't have to resort to checking both here and
3589 * /proc/config.gz to confirm 64-bit inums were successfully applied
3590 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3592 * We hide it when inode64 isn't the default and we are using 32-bit
3593 * inodes, since that probably just means the feature isn't even under
3598 * +-----------------+-----------------+
3599 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3600 * +------------------+-----------------+-----------------+
3601 * | full_inums=true | show | show |
3602 * | full_inums=false | show | hide |
3603 * +------------------+-----------------+-----------------+
3606 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3607 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3608 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3609 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3611 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3613 shmem_show_mpol(seq, sbinfo->mpol);
3617 #endif /* CONFIG_TMPFS */
3619 static void shmem_put_super(struct super_block *sb)
3621 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3623 free_percpu(sbinfo->ino_batch);
3624 percpu_counter_destroy(&sbinfo->used_blocks);
3625 mpol_put(sbinfo->mpol);
3627 sb->s_fs_info = NULL;
3630 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3632 struct shmem_options *ctx = fc->fs_private;
3633 struct inode *inode;
3634 struct shmem_sb_info *sbinfo;
3637 /* Round up to L1_CACHE_BYTES to resist false sharing */
3638 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3639 L1_CACHE_BYTES), GFP_KERNEL);
3643 sb->s_fs_info = sbinfo;
3647 * Per default we only allow half of the physical ram per
3648 * tmpfs instance, limiting inodes to one per page of lowmem;
3649 * but the internal instance is left unlimited.
3651 if (!(sb->s_flags & SB_KERNMOUNT)) {
3652 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3653 ctx->blocks = shmem_default_max_blocks();
3654 if (!(ctx->seen & SHMEM_SEEN_INODES))
3655 ctx->inodes = shmem_default_max_inodes();
3656 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3657 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3659 sb->s_flags |= SB_NOUSER;
3661 sb->s_export_op = &shmem_export_ops;
3662 sb->s_flags |= SB_NOSEC;
3664 sb->s_flags |= SB_NOUSER;
3666 sbinfo->max_blocks = ctx->blocks;
3667 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3668 if (sb->s_flags & SB_KERNMOUNT) {
3669 sbinfo->ino_batch = alloc_percpu(ino_t);
3670 if (!sbinfo->ino_batch)
3673 sbinfo->uid = ctx->uid;
3674 sbinfo->gid = ctx->gid;
3675 sbinfo->full_inums = ctx->full_inums;
3676 sbinfo->mode = ctx->mode;
3677 sbinfo->huge = ctx->huge;
3678 sbinfo->mpol = ctx->mpol;
3681 spin_lock_init(&sbinfo->stat_lock);
3682 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3684 spin_lock_init(&sbinfo->shrinklist_lock);
3685 INIT_LIST_HEAD(&sbinfo->shrinklist);
3687 sb->s_maxbytes = MAX_LFS_FILESIZE;
3688 sb->s_blocksize = PAGE_SIZE;
3689 sb->s_blocksize_bits = PAGE_SHIFT;
3690 sb->s_magic = TMPFS_MAGIC;
3691 sb->s_op = &shmem_ops;
3692 sb->s_time_gran = 1;
3693 #ifdef CONFIG_TMPFS_XATTR
3694 sb->s_xattr = shmem_xattr_handlers;
3696 #ifdef CONFIG_TMPFS_POSIX_ACL
3697 sb->s_flags |= SB_POSIXACL;
3699 uuid_gen(&sb->s_uuid);
3701 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3704 inode->i_uid = sbinfo->uid;
3705 inode->i_gid = sbinfo->gid;
3706 sb->s_root = d_make_root(inode);
3712 shmem_put_super(sb);
3716 static int shmem_get_tree(struct fs_context *fc)
3718 return get_tree_nodev(fc, shmem_fill_super);
3721 static void shmem_free_fc(struct fs_context *fc)
3723 struct shmem_options *ctx = fc->fs_private;
3726 mpol_put(ctx->mpol);
3731 static const struct fs_context_operations shmem_fs_context_ops = {
3732 .free = shmem_free_fc,
3733 .get_tree = shmem_get_tree,
3735 .parse_monolithic = shmem_parse_options,
3736 .parse_param = shmem_parse_one,
3737 .reconfigure = shmem_reconfigure,
3741 static struct kmem_cache *shmem_inode_cachep;
3743 static struct inode *shmem_alloc_inode(struct super_block *sb)
3745 struct shmem_inode_info *info;
3746 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3749 return &info->vfs_inode;
3752 static void shmem_free_in_core_inode(struct inode *inode)
3754 if (S_ISLNK(inode->i_mode))
3755 kfree(inode->i_link);
3756 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3759 static void shmem_destroy_inode(struct inode *inode)
3761 if (S_ISREG(inode->i_mode))
3762 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3765 static void shmem_init_inode(void *foo)
3767 struct shmem_inode_info *info = foo;
3768 inode_init_once(&info->vfs_inode);
3771 static void shmem_init_inodecache(void)
3773 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3774 sizeof(struct shmem_inode_info),
3775 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3778 static void shmem_destroy_inodecache(void)
3780 kmem_cache_destroy(shmem_inode_cachep);
3783 const struct address_space_operations shmem_aops = {
3784 .writepage = shmem_writepage,
3785 .set_page_dirty = __set_page_dirty_no_writeback,
3787 .write_begin = shmem_write_begin,
3788 .write_end = shmem_write_end,
3790 #ifdef CONFIG_MIGRATION
3791 .migratepage = migrate_page,
3793 .error_remove_page = generic_error_remove_page,
3795 EXPORT_SYMBOL(shmem_aops);
3797 static const struct file_operations shmem_file_operations = {
3799 .get_unmapped_area = shmem_get_unmapped_area,
3801 .llseek = shmem_file_llseek,
3802 .read_iter = shmem_file_read_iter,
3803 .write_iter = generic_file_write_iter,
3804 .fsync = noop_fsync,
3805 .splice_read = generic_file_splice_read,
3806 .splice_write = iter_file_splice_write,
3807 .fallocate = shmem_fallocate,
3811 static const struct inode_operations shmem_inode_operations = {
3812 .getattr = shmem_getattr,
3813 .setattr = shmem_setattr,
3814 #ifdef CONFIG_TMPFS_XATTR
3815 .listxattr = shmem_listxattr,
3816 .set_acl = simple_set_acl,
3820 static const struct inode_operations shmem_dir_inode_operations = {
3822 .create = shmem_create,
3823 .lookup = simple_lookup,
3825 .unlink = shmem_unlink,
3826 .symlink = shmem_symlink,
3827 .mkdir = shmem_mkdir,
3828 .rmdir = shmem_rmdir,
3829 .mknod = shmem_mknod,
3830 .rename = shmem_rename2,
3831 .tmpfile = shmem_tmpfile,
3833 #ifdef CONFIG_TMPFS_XATTR
3834 .listxattr = shmem_listxattr,
3836 #ifdef CONFIG_TMPFS_POSIX_ACL
3837 .setattr = shmem_setattr,
3838 .set_acl = simple_set_acl,
3842 static const struct inode_operations shmem_special_inode_operations = {
3843 #ifdef CONFIG_TMPFS_XATTR
3844 .listxattr = shmem_listxattr,
3846 #ifdef CONFIG_TMPFS_POSIX_ACL
3847 .setattr = shmem_setattr,
3848 .set_acl = simple_set_acl,
3852 static const struct super_operations shmem_ops = {
3853 .alloc_inode = shmem_alloc_inode,
3854 .free_inode = shmem_free_in_core_inode,
3855 .destroy_inode = shmem_destroy_inode,
3857 .statfs = shmem_statfs,
3858 .show_options = shmem_show_options,
3860 .evict_inode = shmem_evict_inode,
3861 .drop_inode = generic_delete_inode,
3862 .put_super = shmem_put_super,
3863 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3864 .nr_cached_objects = shmem_unused_huge_count,
3865 .free_cached_objects = shmem_unused_huge_scan,
3869 static const struct vm_operations_struct shmem_vm_ops = {
3870 .fault = shmem_fault,
3871 .map_pages = filemap_map_pages,
3873 .set_policy = shmem_set_policy,
3874 .get_policy = shmem_get_policy,
3878 int shmem_init_fs_context(struct fs_context *fc)
3880 struct shmem_options *ctx;
3882 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3886 ctx->mode = 0777 | S_ISVTX;
3887 ctx->uid = current_fsuid();
3888 ctx->gid = current_fsgid();
3890 fc->fs_private = ctx;
3891 fc->ops = &shmem_fs_context_ops;
3895 static struct file_system_type shmem_fs_type = {
3896 .owner = THIS_MODULE,
3898 .init_fs_context = shmem_init_fs_context,
3900 .parameters = shmem_fs_parameters,
3902 .kill_sb = kill_litter_super,
3903 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
3906 int __init shmem_init(void)
3910 shmem_init_inodecache();
3912 error = register_filesystem(&shmem_fs_type);
3914 pr_err("Could not register tmpfs\n");
3918 shm_mnt = kern_mount(&shmem_fs_type);
3919 if (IS_ERR(shm_mnt)) {
3920 error = PTR_ERR(shm_mnt);
3921 pr_err("Could not kern_mount tmpfs\n");
3925 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3926 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3927 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3929 shmem_huge = 0; /* just in case it was patched */
3934 unregister_filesystem(&shmem_fs_type);
3936 shmem_destroy_inodecache();
3937 shm_mnt = ERR_PTR(error);
3941 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3942 static ssize_t shmem_enabled_show(struct kobject *kobj,
3943 struct kobj_attribute *attr, char *buf)
3945 static const int values[] = {
3947 SHMEM_HUGE_WITHIN_SIZE,
3956 for (i = 0; i < ARRAY_SIZE(values); i++) {
3957 len += sysfs_emit_at(buf, len,
3958 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3960 shmem_format_huge(values[i]));
3963 len += sysfs_emit_at(buf, len, "\n");
3968 static ssize_t shmem_enabled_store(struct kobject *kobj,
3969 struct kobj_attribute *attr, const char *buf, size_t count)
3974 if (count + 1 > sizeof(tmp))
3976 memcpy(tmp, buf, count);
3978 if (count && tmp[count - 1] == '\n')
3979 tmp[count - 1] = '\0';
3981 huge = shmem_parse_huge(tmp);
3982 if (huge == -EINVAL)
3984 if (!has_transparent_hugepage() &&
3985 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3989 if (shmem_huge > SHMEM_HUGE_DENY)
3990 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3994 struct kobj_attribute shmem_enabled_attr =
3995 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3996 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3998 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3999 bool shmem_huge_enabled(struct vm_area_struct *vma)
4001 struct inode *inode = file_inode(vma->vm_file);
4002 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4006 if ((vma->vm_flags & VM_NOHUGEPAGE) ||
4007 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
4009 if (shmem_huge == SHMEM_HUGE_FORCE)
4011 if (shmem_huge == SHMEM_HUGE_DENY)
4013 switch (sbinfo->huge) {
4014 case SHMEM_HUGE_NEVER:
4016 case SHMEM_HUGE_ALWAYS:
4018 case SHMEM_HUGE_WITHIN_SIZE:
4019 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4020 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4021 if (i_size >= HPAGE_PMD_SIZE &&
4022 i_size >> PAGE_SHIFT >= off)
4025 case SHMEM_HUGE_ADVISE:
4026 /* TODO: implement fadvise() hints */
4027 return (vma->vm_flags & VM_HUGEPAGE);
4033 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4035 #else /* !CONFIG_SHMEM */
4038 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4040 * This is intended for small system where the benefits of the full
4041 * shmem code (swap-backed and resource-limited) are outweighed by
4042 * their complexity. On systems without swap this code should be
4043 * effectively equivalent, but much lighter weight.
4046 static struct file_system_type shmem_fs_type = {
4048 .init_fs_context = ramfs_init_fs_context,
4049 .parameters = ramfs_fs_parameters,
4050 .kill_sb = kill_litter_super,
4051 .fs_flags = FS_USERNS_MOUNT,
4054 int __init shmem_init(void)
4056 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4058 shm_mnt = kern_mount(&shmem_fs_type);
4059 BUG_ON(IS_ERR(shm_mnt));
4064 int shmem_unuse(unsigned int type, bool frontswap,
4065 unsigned long *fs_pages_to_unuse)
4070 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4075 void shmem_unlock_mapping(struct address_space *mapping)
4080 unsigned long shmem_get_unmapped_area(struct file *file,
4081 unsigned long addr, unsigned long len,
4082 unsigned long pgoff, unsigned long flags)
4084 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4088 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4090 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4092 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4094 #define shmem_vm_ops generic_file_vm_ops
4095 #define shmem_file_operations ramfs_file_operations
4096 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4097 #define shmem_acct_size(flags, size) 0
4098 #define shmem_unacct_size(flags, size) do {} while (0)
4100 #endif /* CONFIG_SHMEM */
4104 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4105 unsigned long flags, unsigned int i_flags)
4107 struct inode *inode;
4111 return ERR_CAST(mnt);
4113 if (size < 0 || size > MAX_LFS_FILESIZE)
4114 return ERR_PTR(-EINVAL);
4116 if (shmem_acct_size(flags, size))
4117 return ERR_PTR(-ENOMEM);
4119 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4121 if (unlikely(!inode)) {
4122 shmem_unacct_size(flags, size);
4123 return ERR_PTR(-ENOSPC);
4125 inode->i_flags |= i_flags;
4126 inode->i_size = size;
4127 clear_nlink(inode); /* It is unlinked */
4128 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4130 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4131 &shmem_file_operations);
4138 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4139 * kernel internal. There will be NO LSM permission checks against the
4140 * underlying inode. So users of this interface must do LSM checks at a
4141 * higher layer. The users are the big_key and shm implementations. LSM
4142 * checks are provided at the key or shm level rather than the inode.
4143 * @name: name for dentry (to be seen in /proc/<pid>/maps
4144 * @size: size to be set for the file
4145 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4147 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4149 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4153 * shmem_file_setup - get an unlinked file living in tmpfs
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(const char *name, loff_t size, unsigned long flags)
4160 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4162 EXPORT_SYMBOL_GPL(shmem_file_setup);
4165 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4166 * @mnt: the tmpfs mount where the file will be created
4167 * @name: name for dentry (to be seen in /proc/<pid>/maps
4168 * @size: size to be set for the file
4169 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4171 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4172 loff_t size, unsigned long flags)
4174 return __shmem_file_setup(mnt, name, size, flags, 0);
4176 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4179 * shmem_zero_setup - setup a shared anonymous mapping
4180 * @vma: the vma to be mmapped is prepared by do_mmap
4182 int shmem_zero_setup(struct vm_area_struct *vma)
4185 loff_t size = vma->vm_end - vma->vm_start;
4188 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4189 * between XFS directory reading and selinux: since this file is only
4190 * accessible to the user through its mapping, use S_PRIVATE flag to
4191 * bypass file security, in the same way as shmem_kernel_file_setup().
4193 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4195 return PTR_ERR(file);
4199 vma->vm_file = file;
4200 vma->vm_ops = &shmem_vm_ops;
4202 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4203 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4204 (vma->vm_end & HPAGE_PMD_MASK)) {
4205 khugepaged_enter(vma, vma->vm_flags);
4212 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4213 * @mapping: the page's address_space
4214 * @index: the page index
4215 * @gfp: the page allocator flags to use if allocating
4217 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4218 * with any new page allocations done using the specified allocation flags.
4219 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4220 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4221 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4223 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4224 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4226 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4227 pgoff_t index, gfp_t gfp)
4230 struct inode *inode = mapping->host;
4234 BUG_ON(!shmem_mapping(mapping));
4235 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4236 gfp, NULL, NULL, NULL);
4238 page = ERR_PTR(error);
4244 * The tiny !SHMEM case uses ramfs without swap
4246 return read_cache_page_gfp(mapping, index, gfp);
4249 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);