2 * Resizable virtual memory filesystem for Linux.
4 * Copyright (C) 2000 Linus Torvalds.
6 * 2000-2001 Christoph Rohland
9 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
21 * This file is released under the GPL.
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
32 #include <linux/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 #include <linux/frontswap.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
43 static struct vfsmount *shm_mnt;
47 * This virtual memory filesystem is heavily based on the ramfs. It
48 * extends ramfs by the ability to use swap and honor resource limits
49 * which makes it a completely usable filesystem.
52 #include <linux/xattr.h>
53 #include <linux/exportfs.h>
54 #include <linux/posix_acl.h>
55 #include <linux/posix_acl_xattr.h>
56 #include <linux/mman.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/backing-dev.h>
60 #include <linux/shmem_fs.h>
61 #include <linux/writeback.h>
62 #include <linux/blkdev.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/userfaultfd_k.h>
80 #include <linux/rmap.h>
81 #include <linux/uuid.h>
83 #include <linux/uaccess.h>
87 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
88 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
93 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
97 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98 * inode->i_private (with i_rwsem making sure that it has only one user at
99 * a time): we would prefer not to enlarge the shmem inode just for that.
101 struct shmem_falloc {
102 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
103 pgoff_t start; /* start of range currently being fallocated */
104 pgoff_t next; /* the next page offset to be fallocated */
105 pgoff_t nr_falloced; /* how many new pages have been fallocated */
106 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
109 struct shmem_options {
110 unsigned long long blocks;
111 unsigned long long inodes;
112 struct mempolicy *mpol;
119 #define SHMEM_SEEN_BLOCKS 1
120 #define SHMEM_SEEN_INODES 2
121 #define SHMEM_SEEN_HUGE 4
122 #define SHMEM_SEEN_INUMS 8
126 static unsigned long shmem_default_max_blocks(void)
128 return totalram_pages() / 2;
131 static unsigned long shmem_default_max_inodes(void)
133 unsigned long nr_pages = totalram_pages();
135 return min(nr_pages - totalhigh_pages(), nr_pages / 2);
139 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
140 struct page **pagep, enum sgp_type sgp,
141 gfp_t gfp, struct vm_area_struct *vma,
142 vm_fault_t *fault_type);
143 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
144 struct page **pagep, enum sgp_type sgp,
145 gfp_t gfp, struct vm_area_struct *vma,
146 struct vm_fault *vmf, vm_fault_t *fault_type);
148 int shmem_getpage(struct inode *inode, pgoff_t index,
149 struct page **pagep, enum sgp_type sgp)
151 return shmem_getpage_gfp(inode, index, pagep, sgp,
152 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
155 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
157 return sb->s_fs_info;
161 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
162 * for shared memory and for shared anonymous (/dev/zero) mappings
163 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
164 * consistent with the pre-accounting of private mappings ...
166 static inline int shmem_acct_size(unsigned long flags, loff_t size)
168 return (flags & VM_NORESERVE) ?
169 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
172 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
174 if (!(flags & VM_NORESERVE))
175 vm_unacct_memory(VM_ACCT(size));
178 static inline int shmem_reacct_size(unsigned long flags,
179 loff_t oldsize, loff_t newsize)
181 if (!(flags & VM_NORESERVE)) {
182 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
183 return security_vm_enough_memory_mm(current->mm,
184 VM_ACCT(newsize) - VM_ACCT(oldsize));
185 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
186 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
192 * ... whereas tmpfs objects are accounted incrementally as
193 * pages are allocated, in order to allow large sparse files.
194 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
195 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
197 static inline int shmem_acct_block(unsigned long flags, long pages)
199 if (!(flags & VM_NORESERVE))
202 return security_vm_enough_memory_mm(current->mm,
203 pages * VM_ACCT(PAGE_SIZE));
206 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
208 if (flags & VM_NORESERVE)
209 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
212 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
214 struct shmem_inode_info *info = SHMEM_I(inode);
215 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
217 if (shmem_acct_block(info->flags, pages))
220 if (sbinfo->max_blocks) {
221 if (percpu_counter_compare(&sbinfo->used_blocks,
222 sbinfo->max_blocks - pages) > 0)
224 percpu_counter_add(&sbinfo->used_blocks, pages);
230 shmem_unacct_blocks(info->flags, pages);
234 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
236 struct shmem_inode_info *info = SHMEM_I(inode);
237 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
239 if (sbinfo->max_blocks)
240 percpu_counter_sub(&sbinfo->used_blocks, pages);
241 shmem_unacct_blocks(info->flags, pages);
244 static const struct super_operations shmem_ops;
245 const struct address_space_operations shmem_aops;
246 static const struct file_operations shmem_file_operations;
247 static const struct inode_operations shmem_inode_operations;
248 static const struct inode_operations shmem_dir_inode_operations;
249 static const struct inode_operations shmem_special_inode_operations;
250 static const struct vm_operations_struct shmem_vm_ops;
251 static struct file_system_type shmem_fs_type;
253 bool vma_is_shmem(struct vm_area_struct *vma)
255 return vma->vm_ops == &shmem_vm_ops;
258 static LIST_HEAD(shmem_swaplist);
259 static DEFINE_MUTEX(shmem_swaplist_mutex);
262 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
263 * produces a novel ino for the newly allocated inode.
265 * It may also be called when making a hard link to permit the space needed by
266 * each dentry. However, in that case, no new inode number is needed since that
267 * internally draws from another pool of inode numbers (currently global
268 * get_next_ino()). This case is indicated by passing NULL as inop.
270 #define SHMEM_INO_BATCH 1024
271 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
273 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
276 if (!(sb->s_flags & SB_KERNMOUNT)) {
277 raw_spin_lock(&sbinfo->stat_lock);
278 if (sbinfo->max_inodes) {
279 if (!sbinfo->free_inodes) {
280 raw_spin_unlock(&sbinfo->stat_lock);
283 sbinfo->free_inodes--;
286 ino = sbinfo->next_ino++;
287 if (unlikely(is_zero_ino(ino)))
288 ino = sbinfo->next_ino++;
289 if (unlikely(!sbinfo->full_inums &&
292 * Emulate get_next_ino uint wraparound for
295 if (IS_ENABLED(CONFIG_64BIT))
296 pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
297 __func__, MINOR(sb->s_dev));
298 sbinfo->next_ino = 1;
299 ino = sbinfo->next_ino++;
303 raw_spin_unlock(&sbinfo->stat_lock);
306 * __shmem_file_setup, one of our callers, is lock-free: it
307 * doesn't hold stat_lock in shmem_reserve_inode since
308 * max_inodes is always 0, and is called from potentially
309 * unknown contexts. As such, use a per-cpu batched allocator
310 * which doesn't require the per-sb stat_lock unless we are at
311 * the batch boundary.
313 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
314 * shmem mounts are not exposed to userspace, so we don't need
315 * to worry about things like glibc compatibility.
319 next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
321 if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
322 raw_spin_lock(&sbinfo->stat_lock);
323 ino = sbinfo->next_ino;
324 sbinfo->next_ino += SHMEM_INO_BATCH;
325 raw_spin_unlock(&sbinfo->stat_lock);
326 if (unlikely(is_zero_ino(ino)))
337 static void shmem_free_inode(struct super_block *sb)
339 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
340 if (sbinfo->max_inodes) {
341 raw_spin_lock(&sbinfo->stat_lock);
342 sbinfo->free_inodes++;
343 raw_spin_unlock(&sbinfo->stat_lock);
348 * shmem_recalc_inode - recalculate the block usage of an inode
349 * @inode: inode to recalc
351 * We have to calculate the free blocks since the mm can drop
352 * undirtied hole pages behind our back.
354 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
355 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
357 * It has to be called with the spinlock held.
359 static void shmem_recalc_inode(struct inode *inode)
361 struct shmem_inode_info *info = SHMEM_I(inode);
364 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
366 info->alloced -= freed;
367 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
368 shmem_inode_unacct_blocks(inode, freed);
372 bool shmem_charge(struct inode *inode, long pages)
374 struct shmem_inode_info *info = SHMEM_I(inode);
377 if (!shmem_inode_acct_block(inode, pages))
380 /* nrpages adjustment first, then shmem_recalc_inode() when balanced */
381 inode->i_mapping->nrpages += pages;
383 spin_lock_irqsave(&info->lock, flags);
384 info->alloced += pages;
385 inode->i_blocks += pages * BLOCKS_PER_PAGE;
386 shmem_recalc_inode(inode);
387 spin_unlock_irqrestore(&info->lock, flags);
392 void shmem_uncharge(struct inode *inode, long pages)
394 struct shmem_inode_info *info = SHMEM_I(inode);
397 /* nrpages adjustment done by __delete_from_page_cache() or caller */
399 spin_lock_irqsave(&info->lock, flags);
400 info->alloced -= pages;
401 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
402 shmem_recalc_inode(inode);
403 spin_unlock_irqrestore(&info->lock, flags);
405 shmem_inode_unacct_blocks(inode, pages);
409 * Replace item expected in xarray by a new item, while holding xa_lock.
411 static int shmem_replace_entry(struct address_space *mapping,
412 pgoff_t index, void *expected, void *replacement)
414 XA_STATE(xas, &mapping->i_pages, index);
417 VM_BUG_ON(!expected);
418 VM_BUG_ON(!replacement);
419 item = xas_load(&xas);
420 if (item != expected)
422 xas_store(&xas, replacement);
427 * Sometimes, before we decide whether to proceed or to fail, we must check
428 * that an entry was not already brought back from swap by a racing thread.
430 * Checking page is not enough: by the time a SwapCache page is locked, it
431 * might be reused, and again be SwapCache, using the same swap as before.
433 static bool shmem_confirm_swap(struct address_space *mapping,
434 pgoff_t index, swp_entry_t swap)
436 return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
440 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
443 * disables huge pages for the mount;
445 * enables huge pages for the mount;
446 * SHMEM_HUGE_WITHIN_SIZE:
447 * only allocate huge pages if the page will be fully within i_size,
448 * also respect fadvise()/madvise() hints;
450 * only allocate huge pages if requested with fadvise()/madvise();
453 #define SHMEM_HUGE_NEVER 0
454 #define SHMEM_HUGE_ALWAYS 1
455 #define SHMEM_HUGE_WITHIN_SIZE 2
456 #define SHMEM_HUGE_ADVISE 3
460 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
463 * disables huge on shm_mnt and all mounts, for emergency use;
465 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
468 #define SHMEM_HUGE_DENY (-1)
469 #define SHMEM_HUGE_FORCE (-2)
471 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
472 /* ifdef here to avoid bloating shmem.o when not necessary */
474 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
476 bool shmem_is_huge(struct vm_area_struct *vma,
477 struct inode *inode, pgoff_t index)
481 if (shmem_huge == SHMEM_HUGE_DENY)
483 if (vma && ((vma->vm_flags & VM_NOHUGEPAGE) ||
484 test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
486 if (shmem_huge == SHMEM_HUGE_FORCE)
489 switch (SHMEM_SB(inode->i_sb)->huge) {
490 case SHMEM_HUGE_ALWAYS:
492 case SHMEM_HUGE_WITHIN_SIZE:
493 index = round_up(index, HPAGE_PMD_NR);
494 i_size = round_up(i_size_read(inode), PAGE_SIZE);
495 if (i_size >= HPAGE_PMD_SIZE && (i_size >> PAGE_SHIFT) >= index)
498 case SHMEM_HUGE_ADVISE:
499 if (vma && (vma->vm_flags & VM_HUGEPAGE))
507 #if defined(CONFIG_SYSFS)
508 static int shmem_parse_huge(const char *str)
510 if (!strcmp(str, "never"))
511 return SHMEM_HUGE_NEVER;
512 if (!strcmp(str, "always"))
513 return SHMEM_HUGE_ALWAYS;
514 if (!strcmp(str, "within_size"))
515 return SHMEM_HUGE_WITHIN_SIZE;
516 if (!strcmp(str, "advise"))
517 return SHMEM_HUGE_ADVISE;
518 if (!strcmp(str, "deny"))
519 return SHMEM_HUGE_DENY;
520 if (!strcmp(str, "force"))
521 return SHMEM_HUGE_FORCE;
526 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
527 static const char *shmem_format_huge(int huge)
530 case SHMEM_HUGE_NEVER:
532 case SHMEM_HUGE_ALWAYS:
534 case SHMEM_HUGE_WITHIN_SIZE:
535 return "within_size";
536 case SHMEM_HUGE_ADVISE:
538 case SHMEM_HUGE_DENY:
540 case SHMEM_HUGE_FORCE:
549 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
550 struct shrink_control *sc, unsigned long nr_to_split)
552 LIST_HEAD(list), *pos, *next;
553 LIST_HEAD(to_remove);
555 struct shmem_inode_info *info;
557 unsigned long batch = sc ? sc->nr_to_scan : 128;
558 int removed = 0, split = 0;
560 if (list_empty(&sbinfo->shrinklist))
563 spin_lock(&sbinfo->shrinklist_lock);
564 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
565 info = list_entry(pos, struct shmem_inode_info, shrinklist);
568 inode = igrab(&info->vfs_inode);
570 /* inode is about to be evicted */
572 list_del_init(&info->shrinklist);
577 /* Check if there's anything to gain */
578 if (round_up(inode->i_size, PAGE_SIZE) ==
579 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
580 list_move(&info->shrinklist, &to_remove);
585 list_move(&info->shrinklist, &list);
590 spin_unlock(&sbinfo->shrinklist_lock);
592 list_for_each_safe(pos, next, &to_remove) {
593 info = list_entry(pos, struct shmem_inode_info, shrinklist);
594 inode = &info->vfs_inode;
595 list_del_init(&info->shrinklist);
599 list_for_each_safe(pos, next, &list) {
602 info = list_entry(pos, struct shmem_inode_info, shrinklist);
603 inode = &info->vfs_inode;
605 if (nr_to_split && split >= nr_to_split)
608 page = find_get_page(inode->i_mapping,
609 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
613 /* No huge page at the end of the file: nothing to split */
614 if (!PageTransHuge(page)) {
620 * Leave the inode on the list if we failed to lock
621 * the page at this time.
623 * Waiting for the lock may lead to deadlock in the
626 if (!trylock_page(page)) {
631 ret = split_huge_page(page);
635 /* If split failed leave the inode on the list */
641 list_del_init(&info->shrinklist);
647 spin_lock(&sbinfo->shrinklist_lock);
648 list_splice_tail(&list, &sbinfo->shrinklist);
649 sbinfo->shrinklist_len -= removed;
650 spin_unlock(&sbinfo->shrinklist_lock);
655 static long shmem_unused_huge_scan(struct super_block *sb,
656 struct shrink_control *sc)
658 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
660 if (!READ_ONCE(sbinfo->shrinklist_len))
663 return shmem_unused_huge_shrink(sbinfo, sc, 0);
666 static long shmem_unused_huge_count(struct super_block *sb,
667 struct shrink_control *sc)
669 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
670 return READ_ONCE(sbinfo->shrinklist_len);
672 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
674 #define shmem_huge SHMEM_HUGE_DENY
676 bool shmem_is_huge(struct vm_area_struct *vma,
677 struct inode *inode, pgoff_t index)
682 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
683 struct shrink_control *sc, unsigned long nr_to_split)
687 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
690 * Like add_to_page_cache_locked, but error if expected item has gone.
692 static int shmem_add_to_page_cache(struct page *page,
693 struct address_space *mapping,
694 pgoff_t index, void *expected, gfp_t gfp,
695 struct mm_struct *charge_mm)
697 XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
699 unsigned long nr = compound_nr(page);
702 VM_BUG_ON_PAGE(PageTail(page), page);
703 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
704 VM_BUG_ON_PAGE(!PageLocked(page), page);
705 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
706 VM_BUG_ON(expected && PageTransHuge(page));
708 page_ref_add(page, nr);
709 page->mapping = mapping;
712 if (!PageSwapCache(page)) {
713 error = mem_cgroup_charge(page, charge_mm, gfp);
715 if (PageTransHuge(page)) {
716 count_vm_event(THP_FILE_FALLBACK);
717 count_vm_event(THP_FILE_FALLBACK_CHARGE);
722 cgroup_throttle_swaprate(page, gfp);
727 entry = xas_find_conflict(&xas);
728 if (entry != expected)
729 xas_set_err(&xas, -EEXIST);
730 xas_create_range(&xas);
734 xas_store(&xas, page);
739 if (PageTransHuge(page)) {
740 count_vm_event(THP_FILE_ALLOC);
741 __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr);
743 mapping->nrpages += nr;
744 __mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
745 __mod_lruvec_page_state(page, NR_SHMEM, nr);
747 xas_unlock_irq(&xas);
748 } while (xas_nomem(&xas, gfp));
750 if (xas_error(&xas)) {
751 error = xas_error(&xas);
757 page->mapping = NULL;
758 page_ref_sub(page, nr);
763 * Like delete_from_page_cache, but substitutes swap for page.
765 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
767 struct address_space *mapping = page->mapping;
770 VM_BUG_ON_PAGE(PageCompound(page), page);
772 xa_lock_irq(&mapping->i_pages);
773 error = shmem_replace_entry(mapping, page->index, page, radswap);
774 page->mapping = NULL;
776 __dec_lruvec_page_state(page, NR_FILE_PAGES);
777 __dec_lruvec_page_state(page, NR_SHMEM);
778 xa_unlock_irq(&mapping->i_pages);
784 * Remove swap entry from page cache, free the swap and its page cache.
786 static int shmem_free_swap(struct address_space *mapping,
787 pgoff_t index, void *radswap)
791 old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
794 free_swap_and_cache(radix_to_swp_entry(radswap));
799 * Determine (in bytes) how many of the shmem object's pages mapped by the
800 * given offsets are swapped out.
802 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
803 * as long as the inode doesn't go away and racy results are not a problem.
805 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
806 pgoff_t start, pgoff_t end)
808 XA_STATE(xas, &mapping->i_pages, start);
810 unsigned long swapped = 0;
813 xas_for_each(&xas, page, end - 1) {
814 if (xas_retry(&xas, page))
816 if (xa_is_value(page))
819 if (need_resched()) {
827 return swapped << PAGE_SHIFT;
831 * Determine (in bytes) how many of the shmem object's pages mapped by the
832 * given vma is swapped out.
834 * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
835 * as long as the inode doesn't go away and racy results are not a problem.
837 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
839 struct inode *inode = file_inode(vma->vm_file);
840 struct shmem_inode_info *info = SHMEM_I(inode);
841 struct address_space *mapping = inode->i_mapping;
842 unsigned long swapped;
844 /* Be careful as we don't hold info->lock */
845 swapped = READ_ONCE(info->swapped);
848 * The easier cases are when the shmem object has nothing in swap, or
849 * the vma maps it whole. Then we can simply use the stats that we
855 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
856 return swapped << PAGE_SHIFT;
858 /* Here comes the more involved part */
859 return shmem_partial_swap_usage(mapping,
860 linear_page_index(vma, vma->vm_start),
861 linear_page_index(vma, vma->vm_end));
865 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
867 void shmem_unlock_mapping(struct address_space *mapping)
874 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
876 while (!mapping_unevictable(mapping)) {
877 if (!pagevec_lookup(&pvec, mapping, &index))
879 check_move_unevictable_pages(&pvec);
880 pagevec_release(&pvec);
886 * Check whether a hole-punch or truncation needs to split a huge page,
887 * returning true if no split was required, or the split has been successful.
889 * Eviction (or truncation to 0 size) should never need to split a huge page;
890 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
891 * head, and then succeeded to trylock on tail.
893 * A split can only succeed when there are no additional references on the
894 * huge page: so the split below relies upon find_get_entries() having stopped
895 * when it found a subpage of the huge page, without getting further references.
897 static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
899 if (!PageTransCompound(page))
902 /* Just proceed to delete a huge page wholly within the range punched */
903 if (PageHead(page) &&
904 page->index >= start && page->index + HPAGE_PMD_NR <= end)
907 /* Try to split huge page, so we can truly punch the hole or truncate */
908 return split_huge_page(page) >= 0;
912 * Remove range of pages and swap entries from page cache, and free them.
913 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
915 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
918 struct address_space *mapping = inode->i_mapping;
919 struct shmem_inode_info *info = SHMEM_I(inode);
920 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
921 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
922 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
923 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
925 pgoff_t indices[PAGEVEC_SIZE];
926 long nr_swaps_freed = 0;
931 end = -1; /* unsigned, so actually very big */
933 if (info->fallocend > start && info->fallocend <= end && !unfalloc)
934 info->fallocend = start;
938 while (index < end && find_lock_entries(mapping, index, end - 1,
940 for (i = 0; i < pagevec_count(&pvec); i++) {
941 struct page *page = pvec.pages[i];
945 if (xa_is_value(page)) {
948 nr_swaps_freed += !shmem_free_swap(mapping,
952 index += thp_nr_pages(page) - 1;
954 if (!unfalloc || !PageUptodate(page))
955 truncate_inode_page(mapping, page);
958 pagevec_remove_exceptionals(&pvec);
959 pagevec_release(&pvec);
965 struct page *page = NULL;
966 shmem_getpage(inode, start - 1, &page, SGP_READ);
968 unsigned int top = PAGE_SIZE;
973 zero_user_segment(page, partial_start, top);
974 set_page_dirty(page);
980 struct page *page = NULL;
981 shmem_getpage(inode, end, &page, SGP_READ);
983 zero_user_segment(page, 0, partial_end);
984 set_page_dirty(page);
993 while (index < end) {
996 if (!find_get_entries(mapping, index, end - 1, &pvec,
998 /* If all gone or hole-punch or unfalloc, we're done */
999 if (index == start || end != -1)
1001 /* But if truncating, restart to make sure all gone */
1005 for (i = 0; i < pagevec_count(&pvec); i++) {
1006 struct page *page = pvec.pages[i];
1009 if (xa_is_value(page)) {
1012 if (shmem_free_swap(mapping, index, page)) {
1013 /* Swap was replaced by page: retry */
1023 if (!unfalloc || !PageUptodate(page)) {
1024 if (page_mapping(page) != mapping) {
1025 /* Page was replaced by swap: retry */
1030 VM_BUG_ON_PAGE(PageWriteback(page), page);
1031 if (shmem_punch_compound(page, start, end))
1032 truncate_inode_page(mapping, page);
1033 else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1034 /* Wipe the page and don't get stuck */
1035 clear_highpage(page);
1036 flush_dcache_page(page);
1037 set_page_dirty(page);
1039 round_up(start, HPAGE_PMD_NR))
1045 pagevec_remove_exceptionals(&pvec);
1046 pagevec_release(&pvec);
1050 spin_lock_irq(&info->lock);
1051 info->swapped -= nr_swaps_freed;
1052 shmem_recalc_inode(inode);
1053 spin_unlock_irq(&info->lock);
1056 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1058 shmem_undo_range(inode, lstart, lend, false);
1059 inode->i_ctime = inode->i_mtime = current_time(inode);
1061 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1063 static int shmem_getattr(struct user_namespace *mnt_userns,
1064 const struct path *path, struct kstat *stat,
1065 u32 request_mask, unsigned int query_flags)
1067 struct inode *inode = path->dentry->d_inode;
1068 struct shmem_inode_info *info = SHMEM_I(inode);
1070 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1071 spin_lock_irq(&info->lock);
1072 shmem_recalc_inode(inode);
1073 spin_unlock_irq(&info->lock);
1075 generic_fillattr(&init_user_ns, inode, stat);
1077 if (shmem_is_huge(NULL, inode, 0))
1078 stat->blksize = HPAGE_PMD_SIZE;
1083 static int shmem_setattr(struct user_namespace *mnt_userns,
1084 struct dentry *dentry, struct iattr *attr)
1086 struct inode *inode = d_inode(dentry);
1087 struct shmem_inode_info *info = SHMEM_I(inode);
1090 error = setattr_prepare(&init_user_ns, dentry, attr);
1094 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1095 loff_t oldsize = inode->i_size;
1096 loff_t newsize = attr->ia_size;
1098 /* protected by i_rwsem */
1099 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1100 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1103 if (newsize != oldsize) {
1104 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1108 i_size_write(inode, newsize);
1109 inode->i_ctime = inode->i_mtime = current_time(inode);
1111 if (newsize <= oldsize) {
1112 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1113 if (oldsize > holebegin)
1114 unmap_mapping_range(inode->i_mapping,
1117 shmem_truncate_range(inode,
1118 newsize, (loff_t)-1);
1119 /* unmap again to remove racily COWed private pages */
1120 if (oldsize > holebegin)
1121 unmap_mapping_range(inode->i_mapping,
1126 setattr_copy(&init_user_ns, inode, attr);
1127 if (attr->ia_valid & ATTR_MODE)
1128 error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
1132 static void shmem_evict_inode(struct inode *inode)
1134 struct shmem_inode_info *info = SHMEM_I(inode);
1135 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1137 if (shmem_mapping(inode->i_mapping)) {
1138 shmem_unacct_size(info->flags, inode->i_size);
1140 shmem_truncate_range(inode, 0, (loff_t)-1);
1141 if (!list_empty(&info->shrinklist)) {
1142 spin_lock(&sbinfo->shrinklist_lock);
1143 if (!list_empty(&info->shrinklist)) {
1144 list_del_init(&info->shrinklist);
1145 sbinfo->shrinklist_len--;
1147 spin_unlock(&sbinfo->shrinklist_lock);
1149 while (!list_empty(&info->swaplist)) {
1150 /* Wait while shmem_unuse() is scanning this inode... */
1151 wait_var_event(&info->stop_eviction,
1152 !atomic_read(&info->stop_eviction));
1153 mutex_lock(&shmem_swaplist_mutex);
1154 /* ...but beware of the race if we peeked too early */
1155 if (!atomic_read(&info->stop_eviction))
1156 list_del_init(&info->swaplist);
1157 mutex_unlock(&shmem_swaplist_mutex);
1161 simple_xattrs_free(&info->xattrs);
1162 WARN_ON(inode->i_blocks);
1163 shmem_free_inode(inode->i_sb);
1167 static int shmem_find_swap_entries(struct address_space *mapping,
1168 pgoff_t start, unsigned int nr_entries,
1169 struct page **entries, pgoff_t *indices,
1170 unsigned int type, bool frontswap)
1172 XA_STATE(xas, &mapping->i_pages, start);
1175 unsigned int ret = 0;
1181 xas_for_each(&xas, page, ULONG_MAX) {
1182 if (xas_retry(&xas, page))
1185 if (!xa_is_value(page))
1188 entry = radix_to_swp_entry(page);
1189 if (swp_type(entry) != type)
1192 !frontswap_test(swap_info[type], swp_offset(entry)))
1195 indices[ret] = xas.xa_index;
1196 entries[ret] = page;
1198 if (need_resched()) {
1202 if (++ret == nr_entries)
1211 * Move the swapped pages for an inode to page cache. Returns the count
1212 * of pages swapped in, or the error in case of failure.
1214 static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1220 struct address_space *mapping = inode->i_mapping;
1222 for (i = 0; i < pvec.nr; i++) {
1223 struct page *page = pvec.pages[i];
1225 if (!xa_is_value(page))
1227 error = shmem_swapin_page(inode, indices[i],
1229 mapping_gfp_mask(mapping),
1236 if (error == -ENOMEM)
1240 return error ? error : ret;
1244 * If swap found in inode, free it and move page from swapcache to filecache.
1246 static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1247 bool frontswap, unsigned long *fs_pages_to_unuse)
1249 struct address_space *mapping = inode->i_mapping;
1251 struct pagevec pvec;
1252 pgoff_t indices[PAGEVEC_SIZE];
1253 bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1256 pagevec_init(&pvec);
1258 unsigned int nr_entries = PAGEVEC_SIZE;
1260 if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1261 nr_entries = *fs_pages_to_unuse;
1263 pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1264 pvec.pages, indices,
1271 ret = shmem_unuse_swap_entries(inode, pvec, indices);
1275 if (frontswap_partial) {
1276 *fs_pages_to_unuse -= ret;
1277 if (*fs_pages_to_unuse == 0) {
1278 ret = FRONTSWAP_PAGES_UNUSED;
1283 start = indices[pvec.nr - 1];
1290 * Read all the shared memory data that resides in the swap
1291 * device 'type' back into memory, so the swap device can be
1294 int shmem_unuse(unsigned int type, bool frontswap,
1295 unsigned long *fs_pages_to_unuse)
1297 struct shmem_inode_info *info, *next;
1300 if (list_empty(&shmem_swaplist))
1303 mutex_lock(&shmem_swaplist_mutex);
1304 list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1305 if (!info->swapped) {
1306 list_del_init(&info->swaplist);
1310 * Drop the swaplist mutex while searching the inode for swap;
1311 * but before doing so, make sure shmem_evict_inode() will not
1312 * remove placeholder inode from swaplist, nor let it be freed
1313 * (igrab() would protect from unlink, but not from unmount).
1315 atomic_inc(&info->stop_eviction);
1316 mutex_unlock(&shmem_swaplist_mutex);
1318 error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1322 mutex_lock(&shmem_swaplist_mutex);
1323 next = list_next_entry(info, swaplist);
1325 list_del_init(&info->swaplist);
1326 if (atomic_dec_and_test(&info->stop_eviction))
1327 wake_up_var(&info->stop_eviction);
1331 mutex_unlock(&shmem_swaplist_mutex);
1337 * Move the page from the page cache to the swap cache.
1339 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1341 struct shmem_inode_info *info;
1342 struct address_space *mapping;
1343 struct inode *inode;
1348 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1349 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1350 * and its shmem_writeback() needs them to be split when swapping.
1352 if (PageTransCompound(page)) {
1353 /* Ensure the subpages are still dirty */
1355 if (split_huge_page(page) < 0)
1357 ClearPageDirty(page);
1360 BUG_ON(!PageLocked(page));
1361 mapping = page->mapping;
1362 index = page->index;
1363 inode = mapping->host;
1364 info = SHMEM_I(inode);
1365 if (info->flags & VM_LOCKED)
1367 if (!total_swap_pages)
1371 * Our capabilities prevent regular writeback or sync from ever calling
1372 * shmem_writepage; but a stacking filesystem might use ->writepage of
1373 * its underlying filesystem, in which case tmpfs should write out to
1374 * swap only in response to memory pressure, and not for the writeback
1377 if (!wbc->for_reclaim) {
1378 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1383 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1384 * value into swapfile.c, the only way we can correctly account for a
1385 * fallocated page arriving here is now to initialize it and write it.
1387 * That's okay for a page already fallocated earlier, but if we have
1388 * not yet completed the fallocation, then (a) we want to keep track
1389 * of this page in case we have to undo it, and (b) it may not be a
1390 * good idea to continue anyway, once we're pushing into swap. So
1391 * reactivate the page, and let shmem_fallocate() quit when too many.
1393 if (!PageUptodate(page)) {
1394 if (inode->i_private) {
1395 struct shmem_falloc *shmem_falloc;
1396 spin_lock(&inode->i_lock);
1397 shmem_falloc = inode->i_private;
1399 !shmem_falloc->waitq &&
1400 index >= shmem_falloc->start &&
1401 index < shmem_falloc->next)
1402 shmem_falloc->nr_unswapped++;
1404 shmem_falloc = NULL;
1405 spin_unlock(&inode->i_lock);
1409 clear_highpage(page);
1410 flush_dcache_page(page);
1411 SetPageUptodate(page);
1414 swap = get_swap_page(page);
1419 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1420 * if it's not already there. Do it now before the page is
1421 * moved to swap cache, when its pagelock no longer protects
1422 * the inode from eviction. But don't unlock the mutex until
1423 * we've incremented swapped, because shmem_unuse_inode() will
1424 * prune a !swapped inode from the swaplist under this mutex.
1426 mutex_lock(&shmem_swaplist_mutex);
1427 if (list_empty(&info->swaplist))
1428 list_add(&info->swaplist, &shmem_swaplist);
1430 if (add_to_swap_cache(page, swap,
1431 __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1433 spin_lock_irq(&info->lock);
1434 shmem_recalc_inode(inode);
1436 spin_unlock_irq(&info->lock);
1438 swap_shmem_alloc(swap);
1439 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1441 mutex_unlock(&shmem_swaplist_mutex);
1442 BUG_ON(page_mapped(page));
1443 swap_writepage(page, wbc);
1447 mutex_unlock(&shmem_swaplist_mutex);
1448 put_swap_page(page, swap);
1450 set_page_dirty(page);
1451 if (wbc->for_reclaim)
1452 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1457 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1458 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1462 if (!mpol || mpol->mode == MPOL_DEFAULT)
1463 return; /* show nothing */
1465 mpol_to_str(buffer, sizeof(buffer), mpol);
1467 seq_printf(seq, ",mpol=%s", buffer);
1470 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1472 struct mempolicy *mpol = NULL;
1474 raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1475 mpol = sbinfo->mpol;
1477 raw_spin_unlock(&sbinfo->stat_lock);
1481 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1482 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1485 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1489 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1491 #define vm_policy vm_private_data
1494 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1495 struct shmem_inode_info *info, pgoff_t index)
1497 /* Create a pseudo vma that just contains the policy */
1498 vma_init(vma, NULL);
1499 /* Bias interleave by inode number to distribute better across nodes */
1500 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1501 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1504 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1506 /* Drop reference taken by mpol_shared_policy_lookup() */
1507 mpol_cond_put(vma->vm_policy);
1510 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1511 struct shmem_inode_info *info, pgoff_t index)
1513 struct vm_area_struct pvma;
1515 struct vm_fault vmf = {
1519 shmem_pseudo_vma_init(&pvma, info, index);
1520 page = swap_cluster_readahead(swap, gfp, &vmf);
1521 shmem_pseudo_vma_destroy(&pvma);
1527 * Make sure huge_gfp is always more limited than limit_gfp.
1528 * Some of the flags set permissions, while others set limitations.
1530 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1532 gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1533 gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1534 gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1535 gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1537 /* Allow allocations only from the originally specified zones. */
1538 result |= zoneflags;
1541 * Minimize the result gfp by taking the union with the deny flags,
1542 * and the intersection of the allow flags.
1544 result |= (limit_gfp & denyflags);
1545 result |= (huge_gfp & limit_gfp) & allowflags;
1550 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1551 struct shmem_inode_info *info, pgoff_t index)
1553 struct vm_area_struct pvma;
1554 struct address_space *mapping = info->vfs_inode.i_mapping;
1558 hindex = round_down(index, HPAGE_PMD_NR);
1559 if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1563 shmem_pseudo_vma_init(&pvma, info, hindex);
1564 page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(),
1566 shmem_pseudo_vma_destroy(&pvma);
1568 prep_transhuge_page(page);
1570 count_vm_event(THP_FILE_FALLBACK);
1574 static struct page *shmem_alloc_page(gfp_t gfp,
1575 struct shmem_inode_info *info, pgoff_t index)
1577 struct vm_area_struct pvma;
1580 shmem_pseudo_vma_init(&pvma, info, index);
1581 page = alloc_page_vma(gfp, &pvma, 0);
1582 shmem_pseudo_vma_destroy(&pvma);
1587 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1588 struct inode *inode,
1589 pgoff_t index, bool huge)
1591 struct shmem_inode_info *info = SHMEM_I(inode);
1596 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1598 nr = huge ? HPAGE_PMD_NR : 1;
1600 if (!shmem_inode_acct_block(inode, nr))
1604 page = shmem_alloc_hugepage(gfp, info, index);
1606 page = shmem_alloc_page(gfp, info, index);
1608 __SetPageLocked(page);
1609 __SetPageSwapBacked(page);
1614 shmem_inode_unacct_blocks(inode, nr);
1616 return ERR_PTR(err);
1620 * When a page is moved from swapcache to shmem filecache (either by the
1621 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1622 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1623 * ignorance of the mapping it belongs to. If that mapping has special
1624 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1625 * we may need to copy to a suitable page before moving to filecache.
1627 * In a future release, this may well be extended to respect cpuset and
1628 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1629 * but for now it is a simple matter of zone.
1631 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1633 return page_zonenum(page) > gfp_zone(gfp);
1636 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1637 struct shmem_inode_info *info, pgoff_t index)
1639 struct page *oldpage, *newpage;
1640 struct address_space *swap_mapping;
1646 entry.val = page_private(oldpage);
1647 swap_index = swp_offset(entry);
1648 swap_mapping = page_mapping(oldpage);
1651 * We have arrived here because our zones are constrained, so don't
1652 * limit chance of success by further cpuset and node constraints.
1654 gfp &= ~GFP_CONSTRAINT_MASK;
1655 newpage = shmem_alloc_page(gfp, info, index);
1660 copy_highpage(newpage, oldpage);
1661 flush_dcache_page(newpage);
1663 __SetPageLocked(newpage);
1664 __SetPageSwapBacked(newpage);
1665 SetPageUptodate(newpage);
1666 set_page_private(newpage, entry.val);
1667 SetPageSwapCache(newpage);
1670 * Our caller will very soon move newpage out of swapcache, but it's
1671 * a nice clean interface for us to replace oldpage by newpage there.
1673 xa_lock_irq(&swap_mapping->i_pages);
1674 error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1676 mem_cgroup_migrate(oldpage, newpage);
1677 __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1678 __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1680 xa_unlock_irq(&swap_mapping->i_pages);
1682 if (unlikely(error)) {
1684 * Is this possible? I think not, now that our callers check
1685 * both PageSwapCache and page_private after getting page lock;
1686 * but be defensive. Reverse old to newpage for clear and free.
1690 lru_cache_add(newpage);
1694 ClearPageSwapCache(oldpage);
1695 set_page_private(oldpage, 0);
1697 unlock_page(oldpage);
1704 * Swap in the page pointed to by *pagep.
1705 * Caller has to make sure that *pagep contains a valid swapped page.
1706 * Returns 0 and the page in pagep if success. On failure, returns the
1707 * error code and NULL in *pagep.
1709 static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1710 struct page **pagep, enum sgp_type sgp,
1711 gfp_t gfp, struct vm_area_struct *vma,
1712 vm_fault_t *fault_type)
1714 struct address_space *mapping = inode->i_mapping;
1715 struct shmem_inode_info *info = SHMEM_I(inode);
1716 struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1721 VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1722 swap = radix_to_swp_entry(*pagep);
1725 /* Look it up and read it in.. */
1726 page = lookup_swap_cache(swap, NULL, 0);
1728 /* Or update major stats only when swapin succeeds?? */
1730 *fault_type |= VM_FAULT_MAJOR;
1731 count_vm_event(PGMAJFAULT);
1732 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1734 /* Here we actually start the io */
1735 page = shmem_swapin(swap, gfp, info, index);
1742 /* We have to do this with page locked to prevent races */
1744 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1745 !shmem_confirm_swap(mapping, index, swap)) {
1749 if (!PageUptodate(page)) {
1753 wait_on_page_writeback(page);
1756 * Some architectures may have to restore extra metadata to the
1757 * physical page after reading from swap.
1759 arch_swap_restore(swap, page);
1761 if (shmem_should_replace_page(page, gfp)) {
1762 error = shmem_replace_page(&page, gfp, info, index);
1767 error = shmem_add_to_page_cache(page, mapping, index,
1768 swp_to_radix_entry(swap), gfp,
1773 spin_lock_irq(&info->lock);
1775 shmem_recalc_inode(inode);
1776 spin_unlock_irq(&info->lock);
1778 if (sgp == SGP_WRITE)
1779 mark_page_accessed(page);
1781 delete_from_swap_cache(page);
1782 set_page_dirty(page);
1788 if (!shmem_confirm_swap(mapping, index, swap))
1800 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1802 * If we allocate a new one we do not mark it dirty. That's up to the
1803 * vm. If we swap it in we mark it dirty since we also free the swap
1804 * entry since a page cannot live in both the swap and page cache.
1806 * vma, vmf, and fault_type are only supplied by shmem_fault:
1807 * otherwise they are NULL.
1809 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1810 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1811 struct vm_area_struct *vma, struct vm_fault *vmf,
1812 vm_fault_t *fault_type)
1814 struct address_space *mapping = inode->i_mapping;
1815 struct shmem_inode_info *info = SHMEM_I(inode);
1816 struct shmem_sb_info *sbinfo;
1817 struct mm_struct *charge_mm;
1819 pgoff_t hindex = index;
1825 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1828 if (sgp <= SGP_CACHE &&
1829 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1833 sbinfo = SHMEM_SB(inode->i_sb);
1834 charge_mm = vma ? vma->vm_mm : NULL;
1836 page = pagecache_get_page(mapping, index,
1837 FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0);
1839 if (page && vma && userfaultfd_minor(vma)) {
1840 if (!xa_is_value(page)) {
1844 *fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1848 if (xa_is_value(page)) {
1849 error = shmem_swapin_page(inode, index, &page,
1850 sgp, gfp, vma, fault_type);
1851 if (error == -EEXIST)
1859 hindex = page->index;
1860 if (sgp == SGP_WRITE)
1861 mark_page_accessed(page);
1862 if (PageUptodate(page))
1864 /* fallocated page */
1865 if (sgp != SGP_READ)
1872 * SGP_READ: succeed on hole, with NULL page, letting caller zero.
1873 * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail.
1876 if (sgp == SGP_READ)
1878 if (sgp == SGP_NOALLOC)
1882 * Fast cache lookup and swap lookup did not find it: allocate.
1885 if (vma && userfaultfd_missing(vma)) {
1886 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1890 /* Never use a huge page for shmem_symlink() */
1891 if (S_ISLNK(inode->i_mode))
1893 if (!shmem_is_huge(vma, inode, index))
1896 huge_gfp = vma_thp_gfp_mask(vma);
1897 huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1898 page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true);
1901 page = shmem_alloc_and_acct_page(gfp, inode,
1907 error = PTR_ERR(page);
1909 if (error != -ENOSPC)
1912 * Try to reclaim some space by splitting a huge page
1913 * beyond i_size on the filesystem.
1918 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1919 if (ret == SHRINK_STOP)
1927 if (PageTransHuge(page))
1928 hindex = round_down(index, HPAGE_PMD_NR);
1932 if (sgp == SGP_WRITE)
1933 __SetPageReferenced(page);
1935 error = shmem_add_to_page_cache(page, mapping, hindex,
1936 NULL, gfp & GFP_RECLAIM_MASK,
1940 lru_cache_add(page);
1942 spin_lock_irq(&info->lock);
1943 info->alloced += compound_nr(page);
1944 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1945 shmem_recalc_inode(inode);
1946 spin_unlock_irq(&info->lock);
1949 if (PageTransHuge(page) &&
1950 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1951 hindex + HPAGE_PMD_NR - 1) {
1953 * Part of the huge page is beyond i_size: subject
1954 * to shrink under memory pressure.
1956 spin_lock(&sbinfo->shrinklist_lock);
1958 * _careful to defend against unlocked access to
1959 * ->shrink_list in shmem_unused_huge_shrink()
1961 if (list_empty_careful(&info->shrinklist)) {
1962 list_add_tail(&info->shrinklist,
1963 &sbinfo->shrinklist);
1964 sbinfo->shrinklist_len++;
1966 spin_unlock(&sbinfo->shrinklist_lock);
1970 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1972 if (sgp == SGP_FALLOC)
1976 * Let SGP_WRITE caller clear ends if write does not fill page;
1977 * but SGP_FALLOC on a page fallocated earlier must initialize
1978 * it now, lest undo on failure cancel our earlier guarantee.
1980 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1983 for (i = 0; i < compound_nr(page); i++) {
1984 clear_highpage(page + i);
1985 flush_dcache_page(page + i);
1987 SetPageUptodate(page);
1990 /* Perhaps the file has been truncated since we checked */
1991 if (sgp <= SGP_CACHE &&
1992 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1994 ClearPageDirty(page);
1995 delete_from_page_cache(page);
1996 spin_lock_irq(&info->lock);
1997 shmem_recalc_inode(inode);
1998 spin_unlock_irq(&info->lock);
2004 *pagep = page + index - hindex;
2011 shmem_inode_unacct_blocks(inode, compound_nr(page));
2013 if (PageTransHuge(page)) {
2023 if (error == -ENOSPC && !once++) {
2024 spin_lock_irq(&info->lock);
2025 shmem_recalc_inode(inode);
2026 spin_unlock_irq(&info->lock);
2029 if (error == -EEXIST)
2035 * This is like autoremove_wake_function, but it removes the wait queue
2036 * entry unconditionally - even if something else had already woken the
2039 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2041 int ret = default_wake_function(wait, mode, sync, key);
2042 list_del_init(&wait->entry);
2046 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2048 struct vm_area_struct *vma = vmf->vma;
2049 struct inode *inode = file_inode(vma->vm_file);
2050 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2052 vm_fault_t ret = VM_FAULT_LOCKED;
2055 * Trinity finds that probing a hole which tmpfs is punching can
2056 * prevent the hole-punch from ever completing: which in turn
2057 * locks writers out with its hold on i_rwsem. So refrain from
2058 * faulting pages into the hole while it's being punched. Although
2059 * shmem_undo_range() does remove the additions, it may be unable to
2060 * keep up, as each new page needs its own unmap_mapping_range() call,
2061 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2063 * It does not matter if we sometimes reach this check just before the
2064 * hole-punch begins, so that one fault then races with the punch:
2065 * we just need to make racing faults a rare case.
2067 * The implementation below would be much simpler if we just used a
2068 * standard mutex or completion: but we cannot take i_rwsem in fault,
2069 * and bloating every shmem inode for this unlikely case would be sad.
2071 if (unlikely(inode->i_private)) {
2072 struct shmem_falloc *shmem_falloc;
2074 spin_lock(&inode->i_lock);
2075 shmem_falloc = inode->i_private;
2077 shmem_falloc->waitq &&
2078 vmf->pgoff >= shmem_falloc->start &&
2079 vmf->pgoff < shmem_falloc->next) {
2081 wait_queue_head_t *shmem_falloc_waitq;
2082 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2084 ret = VM_FAULT_NOPAGE;
2085 fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2087 ret = VM_FAULT_RETRY;
2089 shmem_falloc_waitq = shmem_falloc->waitq;
2090 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2091 TASK_UNINTERRUPTIBLE);
2092 spin_unlock(&inode->i_lock);
2096 * shmem_falloc_waitq points into the shmem_fallocate()
2097 * stack of the hole-punching task: shmem_falloc_waitq
2098 * is usually invalid by the time we reach here, but
2099 * finish_wait() does not dereference it in that case;
2100 * though i_lock needed lest racing with wake_up_all().
2102 spin_lock(&inode->i_lock);
2103 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2104 spin_unlock(&inode->i_lock);
2110 spin_unlock(&inode->i_lock);
2113 err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE,
2114 gfp, vma, vmf, &ret);
2116 return vmf_error(err);
2120 unsigned long shmem_get_unmapped_area(struct file *file,
2121 unsigned long uaddr, unsigned long len,
2122 unsigned long pgoff, unsigned long flags)
2124 unsigned long (*get_area)(struct file *,
2125 unsigned long, unsigned long, unsigned long, unsigned long);
2127 unsigned long offset;
2128 unsigned long inflated_len;
2129 unsigned long inflated_addr;
2130 unsigned long inflated_offset;
2132 if (len > TASK_SIZE)
2135 get_area = current->mm->get_unmapped_area;
2136 addr = get_area(file, uaddr, len, pgoff, flags);
2138 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2140 if (IS_ERR_VALUE(addr))
2142 if (addr & ~PAGE_MASK)
2144 if (addr > TASK_SIZE - len)
2147 if (shmem_huge == SHMEM_HUGE_DENY)
2149 if (len < HPAGE_PMD_SIZE)
2151 if (flags & MAP_FIXED)
2154 * Our priority is to support MAP_SHARED mapped hugely;
2155 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2156 * But if caller specified an address hint and we allocated area there
2157 * successfully, respect that as before.
2162 if (shmem_huge != SHMEM_HUGE_FORCE) {
2163 struct super_block *sb;
2166 VM_BUG_ON(file->f_op != &shmem_file_operations);
2167 sb = file_inode(file)->i_sb;
2170 * Called directly from mm/mmap.c, or drivers/char/mem.c
2171 * for "/dev/zero", to create a shared anonymous object.
2173 if (IS_ERR(shm_mnt))
2175 sb = shm_mnt->mnt_sb;
2177 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2181 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2182 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2184 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2187 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2188 if (inflated_len > TASK_SIZE)
2190 if (inflated_len < len)
2193 inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2194 if (IS_ERR_VALUE(inflated_addr))
2196 if (inflated_addr & ~PAGE_MASK)
2199 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2200 inflated_addr += offset - inflated_offset;
2201 if (inflated_offset > offset)
2202 inflated_addr += HPAGE_PMD_SIZE;
2204 if (inflated_addr > TASK_SIZE - len)
2206 return inflated_addr;
2210 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2212 struct inode *inode = file_inode(vma->vm_file);
2213 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2216 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2219 struct inode *inode = file_inode(vma->vm_file);
2222 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2223 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2227 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2229 struct inode *inode = file_inode(file);
2230 struct shmem_inode_info *info = SHMEM_I(inode);
2231 int retval = -ENOMEM;
2234 * What serializes the accesses to info->flags?
2235 * ipc_lock_object() when called from shmctl_do_lock(),
2236 * no serialization needed when called from shm_destroy().
2238 if (lock && !(info->flags & VM_LOCKED)) {
2239 if (!user_shm_lock(inode->i_size, ucounts))
2241 info->flags |= VM_LOCKED;
2242 mapping_set_unevictable(file->f_mapping);
2244 if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2245 user_shm_unlock(inode->i_size, ucounts);
2246 info->flags &= ~VM_LOCKED;
2247 mapping_clear_unevictable(file->f_mapping);
2255 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2257 struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2260 ret = seal_check_future_write(info->seals, vma);
2264 /* arm64 - allow memory tagging on RAM-based files */
2265 vma->vm_flags |= VM_MTE_ALLOWED;
2267 file_accessed(file);
2268 vma->vm_ops = &shmem_vm_ops;
2269 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2270 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2271 (vma->vm_end & HPAGE_PMD_MASK)) {
2272 khugepaged_enter(vma, vma->vm_flags);
2277 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2278 umode_t mode, dev_t dev, unsigned long flags)
2280 struct inode *inode;
2281 struct shmem_inode_info *info;
2282 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2285 if (shmem_reserve_inode(sb, &ino))
2288 inode = new_inode(sb);
2291 inode_init_owner(&init_user_ns, inode, dir, mode);
2292 inode->i_blocks = 0;
2293 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2294 inode->i_generation = prandom_u32();
2295 info = SHMEM_I(inode);
2296 memset(info, 0, (char *)inode - (char *)info);
2297 spin_lock_init(&info->lock);
2298 atomic_set(&info->stop_eviction, 0);
2299 info->seals = F_SEAL_SEAL;
2300 info->flags = flags & VM_NORESERVE;
2301 INIT_LIST_HEAD(&info->shrinklist);
2302 INIT_LIST_HEAD(&info->swaplist);
2303 simple_xattrs_init(&info->xattrs);
2304 cache_no_acl(inode);
2306 switch (mode & S_IFMT) {
2308 inode->i_op = &shmem_special_inode_operations;
2309 init_special_inode(inode, mode, dev);
2312 inode->i_mapping->a_ops = &shmem_aops;
2313 inode->i_op = &shmem_inode_operations;
2314 inode->i_fop = &shmem_file_operations;
2315 mpol_shared_policy_init(&info->policy,
2316 shmem_get_sbmpol(sbinfo));
2320 /* Some things misbehave if size == 0 on a directory */
2321 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2322 inode->i_op = &shmem_dir_inode_operations;
2323 inode->i_fop = &simple_dir_operations;
2327 * Must not load anything in the rbtree,
2328 * mpol_free_shared_policy will not be called.
2330 mpol_shared_policy_init(&info->policy, NULL);
2334 lockdep_annotate_inode_mutex_key(inode);
2336 shmem_free_inode(sb);
2340 #ifdef CONFIG_USERFAULTFD
2341 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2343 struct vm_area_struct *dst_vma,
2344 unsigned long dst_addr,
2345 unsigned long src_addr,
2347 struct page **pagep)
2349 struct inode *inode = file_inode(dst_vma->vm_file);
2350 struct shmem_inode_info *info = SHMEM_I(inode);
2351 struct address_space *mapping = inode->i_mapping;
2352 gfp_t gfp = mapping_gfp_mask(mapping);
2353 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2359 if (!shmem_inode_acct_block(inode, 1)) {
2361 * We may have got a page, returned -ENOENT triggering a retry,
2362 * and now we find ourselves with -ENOMEM. Release the page, to
2363 * avoid a BUG_ON in our caller.
2365 if (unlikely(*pagep)) {
2374 page = shmem_alloc_page(gfp, info, pgoff);
2376 goto out_unacct_blocks;
2378 if (!zeropage) { /* COPY */
2379 page_kaddr = kmap_atomic(page);
2380 ret = copy_from_user(page_kaddr,
2381 (const void __user *)src_addr,
2383 kunmap_atomic(page_kaddr);
2385 /* fallback to copy_from_user outside mmap_lock */
2386 if (unlikely(ret)) {
2389 /* don't free the page */
2390 goto out_unacct_blocks;
2392 } else { /* ZEROPAGE */
2393 clear_highpage(page);
2400 VM_BUG_ON(PageLocked(page));
2401 VM_BUG_ON(PageSwapBacked(page));
2402 __SetPageLocked(page);
2403 __SetPageSwapBacked(page);
2404 __SetPageUptodate(page);
2407 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2408 if (unlikely(pgoff >= max_off))
2411 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2412 gfp & GFP_RECLAIM_MASK, dst_mm);
2416 ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2419 goto out_delete_from_cache;
2421 spin_lock_irq(&info->lock);
2423 inode->i_blocks += BLOCKS_PER_PAGE;
2424 shmem_recalc_inode(inode);
2425 spin_unlock_irq(&info->lock);
2430 out_delete_from_cache:
2431 delete_from_page_cache(page);
2436 shmem_inode_unacct_blocks(inode, 1);
2439 #endif /* CONFIG_USERFAULTFD */
2442 static const struct inode_operations shmem_symlink_inode_operations;
2443 static const struct inode_operations shmem_short_symlink_operations;
2445 #ifdef CONFIG_TMPFS_XATTR
2446 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2448 #define shmem_initxattrs NULL
2452 shmem_write_begin(struct file *file, struct address_space *mapping,
2453 loff_t pos, unsigned len, unsigned flags,
2454 struct page **pagep, void **fsdata)
2456 struct inode *inode = mapping->host;
2457 struct shmem_inode_info *info = SHMEM_I(inode);
2458 pgoff_t index = pos >> PAGE_SHIFT;
2460 /* i_rwsem is held by caller */
2461 if (unlikely(info->seals & (F_SEAL_GROW |
2462 F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2463 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2465 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2469 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2473 shmem_write_end(struct file *file, struct address_space *mapping,
2474 loff_t pos, unsigned len, unsigned copied,
2475 struct page *page, void *fsdata)
2477 struct inode *inode = mapping->host;
2479 if (pos + copied > inode->i_size)
2480 i_size_write(inode, pos + copied);
2482 if (!PageUptodate(page)) {
2483 struct page *head = compound_head(page);
2484 if (PageTransCompound(page)) {
2487 for (i = 0; i < HPAGE_PMD_NR; i++) {
2488 if (head + i == page)
2490 clear_highpage(head + i);
2491 flush_dcache_page(head + i);
2494 if (copied < PAGE_SIZE) {
2495 unsigned from = pos & (PAGE_SIZE - 1);
2496 zero_user_segments(page, 0, from,
2497 from + copied, PAGE_SIZE);
2499 SetPageUptodate(head);
2501 set_page_dirty(page);
2508 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2510 struct file *file = iocb->ki_filp;
2511 struct inode *inode = file_inode(file);
2512 struct address_space *mapping = inode->i_mapping;
2514 unsigned long offset;
2515 enum sgp_type sgp = SGP_READ;
2518 loff_t *ppos = &iocb->ki_pos;
2521 * Might this read be for a stacking filesystem? Then when reading
2522 * holes of a sparse file, we actually need to allocate those pages,
2523 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2525 if (!iter_is_iovec(to))
2528 index = *ppos >> PAGE_SHIFT;
2529 offset = *ppos & ~PAGE_MASK;
2532 struct page *page = NULL;
2534 unsigned long nr, ret;
2535 loff_t i_size = i_size_read(inode);
2537 end_index = i_size >> PAGE_SHIFT;
2538 if (index > end_index)
2540 if (index == end_index) {
2541 nr = i_size & ~PAGE_MASK;
2546 error = shmem_getpage(inode, index, &page, sgp);
2548 if (error == -EINVAL)
2553 if (sgp == SGP_CACHE)
2554 set_page_dirty(page);
2559 * We must evaluate after, since reads (unlike writes)
2560 * are called without i_rwsem protection against truncate
2563 i_size = i_size_read(inode);
2564 end_index = i_size >> PAGE_SHIFT;
2565 if (index == end_index) {
2566 nr = i_size & ~PAGE_MASK;
2577 * If users can be writing to this page using arbitrary
2578 * virtual addresses, take care about potential aliasing
2579 * before reading the page on the kernel side.
2581 if (mapping_writably_mapped(mapping))
2582 flush_dcache_page(page);
2584 * Mark the page accessed if we read the beginning.
2587 mark_page_accessed(page);
2589 page = ZERO_PAGE(0);
2594 * Ok, we have the page, and it's up-to-date, so
2595 * now we can copy it to user space...
2597 ret = copy_page_to_iter(page, offset, nr, to);
2600 index += offset >> PAGE_SHIFT;
2601 offset &= ~PAGE_MASK;
2604 if (!iov_iter_count(to))
2613 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2614 file_accessed(file);
2615 return retval ? retval : error;
2618 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2620 struct address_space *mapping = file->f_mapping;
2621 struct inode *inode = mapping->host;
2623 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2624 return generic_file_llseek_size(file, offset, whence,
2625 MAX_LFS_FILESIZE, i_size_read(inode));
2630 /* We're holding i_rwsem so we can access i_size directly */
2631 offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2633 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2634 inode_unlock(inode);
2638 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2641 struct inode *inode = file_inode(file);
2642 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2643 struct shmem_inode_info *info = SHMEM_I(inode);
2644 struct shmem_falloc shmem_falloc;
2645 pgoff_t start, index, end, undo_fallocend;
2648 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2653 if (mode & FALLOC_FL_PUNCH_HOLE) {
2654 struct address_space *mapping = file->f_mapping;
2655 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2656 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2657 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2659 /* protected by i_rwsem */
2660 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2665 shmem_falloc.waitq = &shmem_falloc_waitq;
2666 shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2667 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2668 spin_lock(&inode->i_lock);
2669 inode->i_private = &shmem_falloc;
2670 spin_unlock(&inode->i_lock);
2672 if ((u64)unmap_end > (u64)unmap_start)
2673 unmap_mapping_range(mapping, unmap_start,
2674 1 + unmap_end - unmap_start, 0);
2675 shmem_truncate_range(inode, offset, offset + len - 1);
2676 /* No need to unmap again: hole-punching leaves COWed pages */
2678 spin_lock(&inode->i_lock);
2679 inode->i_private = NULL;
2680 wake_up_all(&shmem_falloc_waitq);
2681 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2682 spin_unlock(&inode->i_lock);
2687 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2688 error = inode_newsize_ok(inode, offset + len);
2692 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2697 start = offset >> PAGE_SHIFT;
2698 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2699 /* Try to avoid a swapstorm if len is impossible to satisfy */
2700 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2705 shmem_falloc.waitq = NULL;
2706 shmem_falloc.start = start;
2707 shmem_falloc.next = start;
2708 shmem_falloc.nr_falloced = 0;
2709 shmem_falloc.nr_unswapped = 0;
2710 spin_lock(&inode->i_lock);
2711 inode->i_private = &shmem_falloc;
2712 spin_unlock(&inode->i_lock);
2715 * info->fallocend is only relevant when huge pages might be
2716 * involved: to prevent split_huge_page() freeing fallocated
2717 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2719 undo_fallocend = info->fallocend;
2720 if (info->fallocend < end)
2721 info->fallocend = end;
2723 for (index = start; index < end; ) {
2727 * Good, the fallocate(2) manpage permits EINTR: we may have
2728 * been interrupted because we are using up too much memory.
2730 if (signal_pending(current))
2732 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2735 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2737 info->fallocend = undo_fallocend;
2738 /* Remove the !PageUptodate pages we added */
2739 if (index > start) {
2740 shmem_undo_range(inode,
2741 (loff_t)start << PAGE_SHIFT,
2742 ((loff_t)index << PAGE_SHIFT) - 1, true);
2749 * Here is a more important optimization than it appears:
2750 * a second SGP_FALLOC on the same huge page will clear it,
2751 * making it PageUptodate and un-undoable if we fail later.
2753 if (PageTransCompound(page)) {
2754 index = round_up(index, HPAGE_PMD_NR);
2755 /* Beware 32-bit wraparound */
2761 * Inform shmem_writepage() how far we have reached.
2762 * No need for lock or barrier: we have the page lock.
2764 if (!PageUptodate(page))
2765 shmem_falloc.nr_falloced += index - shmem_falloc.next;
2766 shmem_falloc.next = index;
2769 * If !PageUptodate, leave it that way so that freeable pages
2770 * can be recognized if we need to rollback on error later.
2771 * But set_page_dirty so that memory pressure will swap rather
2772 * than free the pages we are allocating (and SGP_CACHE pages
2773 * might still be clean: we now need to mark those dirty too).
2775 set_page_dirty(page);
2781 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2782 i_size_write(inode, offset + len);
2783 inode->i_ctime = current_time(inode);
2785 spin_lock(&inode->i_lock);
2786 inode->i_private = NULL;
2787 spin_unlock(&inode->i_lock);
2789 inode_unlock(inode);
2793 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2795 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2797 buf->f_type = TMPFS_MAGIC;
2798 buf->f_bsize = PAGE_SIZE;
2799 buf->f_namelen = NAME_MAX;
2800 if (sbinfo->max_blocks) {
2801 buf->f_blocks = sbinfo->max_blocks;
2803 buf->f_bfree = sbinfo->max_blocks -
2804 percpu_counter_sum(&sbinfo->used_blocks);
2806 if (sbinfo->max_inodes) {
2807 buf->f_files = sbinfo->max_inodes;
2808 buf->f_ffree = sbinfo->free_inodes;
2810 /* else leave those fields 0 like simple_statfs */
2812 buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2818 * File creation. Allocate an inode, and we're done..
2821 shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2822 struct dentry *dentry, umode_t mode, dev_t dev)
2824 struct inode *inode;
2825 int error = -ENOSPC;
2827 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2829 error = simple_acl_create(dir, inode);
2832 error = security_inode_init_security(inode, dir,
2834 shmem_initxattrs, NULL);
2835 if (error && error != -EOPNOTSUPP)
2839 dir->i_size += BOGO_DIRENT_SIZE;
2840 dir->i_ctime = dir->i_mtime = current_time(dir);
2841 d_instantiate(dentry, inode);
2842 dget(dentry); /* Extra count - pin the dentry in core */
2851 shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2852 struct dentry *dentry, umode_t mode)
2854 struct inode *inode;
2855 int error = -ENOSPC;
2857 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2859 error = security_inode_init_security(inode, dir,
2861 shmem_initxattrs, NULL);
2862 if (error && error != -EOPNOTSUPP)
2864 error = simple_acl_create(dir, inode);
2867 d_tmpfile(dentry, inode);
2875 static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2876 struct dentry *dentry, umode_t mode)
2880 if ((error = shmem_mknod(&init_user_ns, dir, dentry,
2881 mode | S_IFDIR, 0)))
2887 static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir,
2888 struct dentry *dentry, umode_t mode, bool excl)
2890 return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
2896 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2898 struct inode *inode = d_inode(old_dentry);
2902 * No ordinary (disk based) filesystem counts links as inodes;
2903 * but each new link needs a new dentry, pinning lowmem, and
2904 * tmpfs dentries cannot be pruned until they are unlinked.
2905 * But if an O_TMPFILE file is linked into the tmpfs, the
2906 * first link must skip that, to get the accounting right.
2908 if (inode->i_nlink) {
2909 ret = shmem_reserve_inode(inode->i_sb, NULL);
2914 dir->i_size += BOGO_DIRENT_SIZE;
2915 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2917 ihold(inode); /* New dentry reference */
2918 dget(dentry); /* Extra pinning count for the created dentry */
2919 d_instantiate(dentry, inode);
2924 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2926 struct inode *inode = d_inode(dentry);
2928 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2929 shmem_free_inode(inode->i_sb);
2931 dir->i_size -= BOGO_DIRENT_SIZE;
2932 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2934 dput(dentry); /* Undo the count from "create" - this does all the work */
2938 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2940 if (!simple_empty(dentry))
2943 drop_nlink(d_inode(dentry));
2945 return shmem_unlink(dir, dentry);
2948 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2950 bool old_is_dir = d_is_dir(old_dentry);
2951 bool new_is_dir = d_is_dir(new_dentry);
2953 if (old_dir != new_dir && old_is_dir != new_is_dir) {
2955 drop_nlink(old_dir);
2958 drop_nlink(new_dir);
2962 old_dir->i_ctime = old_dir->i_mtime =
2963 new_dir->i_ctime = new_dir->i_mtime =
2964 d_inode(old_dentry)->i_ctime =
2965 d_inode(new_dentry)->i_ctime = current_time(old_dir);
2970 static int shmem_whiteout(struct user_namespace *mnt_userns,
2971 struct inode *old_dir, struct dentry *old_dentry)
2973 struct dentry *whiteout;
2976 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2980 error = shmem_mknod(&init_user_ns, old_dir, whiteout,
2981 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2987 * Cheat and hash the whiteout while the old dentry is still in
2988 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2990 * d_lookup() will consistently find one of them at this point,
2991 * not sure which one, but that isn't even important.
2998 * The VFS layer already does all the dentry stuff for rename,
2999 * we just have to decrement the usage count for the target if
3000 * it exists so that the VFS layer correctly free's it when it
3003 static int shmem_rename2(struct user_namespace *mnt_userns,
3004 struct inode *old_dir, struct dentry *old_dentry,
3005 struct inode *new_dir, struct dentry *new_dentry,
3008 struct inode *inode = d_inode(old_dentry);
3009 int they_are_dirs = S_ISDIR(inode->i_mode);
3011 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3014 if (flags & RENAME_EXCHANGE)
3015 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3017 if (!simple_empty(new_dentry))
3020 if (flags & RENAME_WHITEOUT) {
3023 error = shmem_whiteout(&init_user_ns, old_dir, old_dentry);
3028 if (d_really_is_positive(new_dentry)) {
3029 (void) shmem_unlink(new_dir, new_dentry);
3030 if (they_are_dirs) {
3031 drop_nlink(d_inode(new_dentry));
3032 drop_nlink(old_dir);
3034 } else if (they_are_dirs) {
3035 drop_nlink(old_dir);
3039 old_dir->i_size -= BOGO_DIRENT_SIZE;
3040 new_dir->i_size += BOGO_DIRENT_SIZE;
3041 old_dir->i_ctime = old_dir->i_mtime =
3042 new_dir->i_ctime = new_dir->i_mtime =
3043 inode->i_ctime = current_time(old_dir);
3047 static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3048 struct dentry *dentry, const char *symname)
3052 struct inode *inode;
3055 len = strlen(symname) + 1;
3056 if (len > PAGE_SIZE)
3057 return -ENAMETOOLONG;
3059 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3064 error = security_inode_init_security(inode, dir, &dentry->d_name,
3065 shmem_initxattrs, NULL);
3066 if (error && error != -EOPNOTSUPP) {
3071 inode->i_size = len-1;
3072 if (len <= SHORT_SYMLINK_LEN) {
3073 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3074 if (!inode->i_link) {
3078 inode->i_op = &shmem_short_symlink_operations;
3080 inode_nohighmem(inode);
3081 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3086 inode->i_mapping->a_ops = &shmem_aops;
3087 inode->i_op = &shmem_symlink_inode_operations;
3088 memcpy(page_address(page), symname, len);
3089 SetPageUptodate(page);
3090 set_page_dirty(page);
3094 dir->i_size += BOGO_DIRENT_SIZE;
3095 dir->i_ctime = dir->i_mtime = current_time(dir);
3096 d_instantiate(dentry, inode);
3101 static void shmem_put_link(void *arg)
3103 mark_page_accessed(arg);
3107 static const char *shmem_get_link(struct dentry *dentry,
3108 struct inode *inode,
3109 struct delayed_call *done)
3111 struct page *page = NULL;
3114 page = find_get_page(inode->i_mapping, 0);
3116 return ERR_PTR(-ECHILD);
3117 if (!PageUptodate(page)) {
3119 return ERR_PTR(-ECHILD);
3122 error = shmem_getpage(inode, 0, &page, SGP_READ);
3124 return ERR_PTR(error);
3127 set_delayed_call(done, shmem_put_link, page);
3128 return page_address(page);
3131 #ifdef CONFIG_TMPFS_XATTR
3133 * Superblocks without xattr inode operations may get some security.* xattr
3134 * support from the LSM "for free". As soon as we have any other xattrs
3135 * like ACLs, we also need to implement the security.* handlers at
3136 * filesystem level, though.
3140 * Callback for security_inode_init_security() for acquiring xattrs.
3142 static int shmem_initxattrs(struct inode *inode,
3143 const struct xattr *xattr_array,
3146 struct shmem_inode_info *info = SHMEM_I(inode);
3147 const struct xattr *xattr;
3148 struct simple_xattr *new_xattr;
3151 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3152 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3156 len = strlen(xattr->name) + 1;
3157 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3159 if (!new_xattr->name) {
3164 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3165 XATTR_SECURITY_PREFIX_LEN);
3166 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3169 simple_xattr_list_add(&info->xattrs, new_xattr);
3175 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3176 struct dentry *unused, struct inode *inode,
3177 const char *name, void *buffer, size_t size)
3179 struct shmem_inode_info *info = SHMEM_I(inode);
3181 name = xattr_full_name(handler, name);
3182 return simple_xattr_get(&info->xattrs, name, buffer, size);
3185 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3186 struct user_namespace *mnt_userns,
3187 struct dentry *unused, struct inode *inode,
3188 const char *name, const void *value,
3189 size_t size, int flags)
3191 struct shmem_inode_info *info = SHMEM_I(inode);
3193 name = xattr_full_name(handler, name);
3194 return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3197 static const struct xattr_handler shmem_security_xattr_handler = {
3198 .prefix = XATTR_SECURITY_PREFIX,
3199 .get = shmem_xattr_handler_get,
3200 .set = shmem_xattr_handler_set,
3203 static const struct xattr_handler shmem_trusted_xattr_handler = {
3204 .prefix = XATTR_TRUSTED_PREFIX,
3205 .get = shmem_xattr_handler_get,
3206 .set = shmem_xattr_handler_set,
3209 static const struct xattr_handler *shmem_xattr_handlers[] = {
3210 #ifdef CONFIG_TMPFS_POSIX_ACL
3211 &posix_acl_access_xattr_handler,
3212 &posix_acl_default_xattr_handler,
3214 &shmem_security_xattr_handler,
3215 &shmem_trusted_xattr_handler,
3219 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3221 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3222 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3224 #endif /* CONFIG_TMPFS_XATTR */
3226 static const struct inode_operations shmem_short_symlink_operations = {
3227 .get_link = simple_get_link,
3228 #ifdef CONFIG_TMPFS_XATTR
3229 .listxattr = shmem_listxattr,
3233 static const struct inode_operations shmem_symlink_inode_operations = {
3234 .get_link = shmem_get_link,
3235 #ifdef CONFIG_TMPFS_XATTR
3236 .listxattr = shmem_listxattr,
3240 static struct dentry *shmem_get_parent(struct dentry *child)
3242 return ERR_PTR(-ESTALE);
3245 static int shmem_match(struct inode *ino, void *vfh)
3249 inum = (inum << 32) | fh[1];
3250 return ino->i_ino == inum && fh[0] == ino->i_generation;
3253 /* Find any alias of inode, but prefer a hashed alias */
3254 static struct dentry *shmem_find_alias(struct inode *inode)
3256 struct dentry *alias = d_find_alias(inode);
3258 return alias ?: d_find_any_alias(inode);
3262 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3263 struct fid *fid, int fh_len, int fh_type)
3265 struct inode *inode;
3266 struct dentry *dentry = NULL;
3273 inum = (inum << 32) | fid->raw[1];
3275 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3276 shmem_match, fid->raw);
3278 dentry = shmem_find_alias(inode);
3285 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3286 struct inode *parent)
3290 return FILEID_INVALID;
3293 if (inode_unhashed(inode)) {
3294 /* Unfortunately insert_inode_hash is not idempotent,
3295 * so as we hash inodes here rather than at creation
3296 * time, we need a lock to ensure we only try
3299 static DEFINE_SPINLOCK(lock);
3301 if (inode_unhashed(inode))
3302 __insert_inode_hash(inode,
3303 inode->i_ino + inode->i_generation);
3307 fh[0] = inode->i_generation;
3308 fh[1] = inode->i_ino;
3309 fh[2] = ((__u64)inode->i_ino) >> 32;
3315 static const struct export_operations shmem_export_ops = {
3316 .get_parent = shmem_get_parent,
3317 .encode_fh = shmem_encode_fh,
3318 .fh_to_dentry = shmem_fh_to_dentry,
3334 static const struct constant_table shmem_param_enums_huge[] = {
3335 {"never", SHMEM_HUGE_NEVER },
3336 {"always", SHMEM_HUGE_ALWAYS },
3337 {"within_size", SHMEM_HUGE_WITHIN_SIZE },
3338 {"advise", SHMEM_HUGE_ADVISE },
3342 const struct fs_parameter_spec shmem_fs_parameters[] = {
3343 fsparam_u32 ("gid", Opt_gid),
3344 fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge),
3345 fsparam_u32oct("mode", Opt_mode),
3346 fsparam_string("mpol", Opt_mpol),
3347 fsparam_string("nr_blocks", Opt_nr_blocks),
3348 fsparam_string("nr_inodes", Opt_nr_inodes),
3349 fsparam_string("size", Opt_size),
3350 fsparam_u32 ("uid", Opt_uid),
3351 fsparam_flag ("inode32", Opt_inode32),
3352 fsparam_flag ("inode64", Opt_inode64),
3356 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3358 struct shmem_options *ctx = fc->fs_private;
3359 struct fs_parse_result result;
3360 unsigned long long size;
3364 opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3370 size = memparse(param->string, &rest);
3372 size <<= PAGE_SHIFT;
3373 size *= totalram_pages();
3379 ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3380 ctx->seen |= SHMEM_SEEN_BLOCKS;
3383 ctx->blocks = memparse(param->string, &rest);
3386 ctx->seen |= SHMEM_SEEN_BLOCKS;
3389 ctx->inodes = memparse(param->string, &rest);
3392 ctx->seen |= SHMEM_SEEN_INODES;
3395 ctx->mode = result.uint_32 & 07777;
3398 ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3399 if (!uid_valid(ctx->uid))
3403 ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3404 if (!gid_valid(ctx->gid))
3408 ctx->huge = result.uint_32;
3409 if (ctx->huge != SHMEM_HUGE_NEVER &&
3410 !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3411 has_transparent_hugepage()))
3412 goto unsupported_parameter;
3413 ctx->seen |= SHMEM_SEEN_HUGE;
3416 if (IS_ENABLED(CONFIG_NUMA)) {
3417 mpol_put(ctx->mpol);
3419 if (mpol_parse_str(param->string, &ctx->mpol))
3423 goto unsupported_parameter;
3425 ctx->full_inums = false;
3426 ctx->seen |= SHMEM_SEEN_INUMS;
3429 if (sizeof(ino_t) < 8) {
3431 "Cannot use inode64 with <64bit inums in kernel\n");
3433 ctx->full_inums = true;
3434 ctx->seen |= SHMEM_SEEN_INUMS;
3439 unsupported_parameter:
3440 return invalfc(fc, "Unsupported parameter '%s'", param->key);
3442 return invalfc(fc, "Bad value for '%s'", param->key);
3445 static int shmem_parse_options(struct fs_context *fc, void *data)
3447 char *options = data;
3450 int err = security_sb_eat_lsm_opts(options, &fc->security);
3455 while (options != NULL) {
3456 char *this_char = options;
3459 * NUL-terminate this option: unfortunately,
3460 * mount options form a comma-separated list,
3461 * but mpol's nodelist may also contain commas.
3463 options = strchr(options, ',');
3464 if (options == NULL)
3467 if (!isdigit(*options)) {
3473 char *value = strchr(this_char, '=');
3479 len = strlen(value);
3481 err = vfs_parse_fs_string(fc, this_char, value, len);
3490 * Reconfigure a shmem filesystem.
3492 * Note that we disallow change from limited->unlimited blocks/inodes while any
3493 * are in use; but we must separately disallow unlimited->limited, because in
3494 * that case we have no record of how much is already in use.
3496 static int shmem_reconfigure(struct fs_context *fc)
3498 struct shmem_options *ctx = fc->fs_private;
3499 struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3500 unsigned long inodes;
3501 struct mempolicy *mpol = NULL;
3504 raw_spin_lock(&sbinfo->stat_lock);
3505 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3506 if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3507 if (!sbinfo->max_blocks) {
3508 err = "Cannot retroactively limit size";
3511 if (percpu_counter_compare(&sbinfo->used_blocks,
3513 err = "Too small a size for current use";
3517 if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3518 if (!sbinfo->max_inodes) {
3519 err = "Cannot retroactively limit inodes";
3522 if (ctx->inodes < inodes) {
3523 err = "Too few inodes for current use";
3528 if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3529 sbinfo->next_ino > UINT_MAX) {
3530 err = "Current inum too high to switch to 32-bit inums";
3534 if (ctx->seen & SHMEM_SEEN_HUGE)
3535 sbinfo->huge = ctx->huge;
3536 if (ctx->seen & SHMEM_SEEN_INUMS)
3537 sbinfo->full_inums = ctx->full_inums;
3538 if (ctx->seen & SHMEM_SEEN_BLOCKS)
3539 sbinfo->max_blocks = ctx->blocks;
3540 if (ctx->seen & SHMEM_SEEN_INODES) {
3541 sbinfo->max_inodes = ctx->inodes;
3542 sbinfo->free_inodes = ctx->inodes - inodes;
3546 * Preserve previous mempolicy unless mpol remount option was specified.
3549 mpol = sbinfo->mpol;
3550 sbinfo->mpol = ctx->mpol; /* transfers initial ref */
3553 raw_spin_unlock(&sbinfo->stat_lock);
3557 raw_spin_unlock(&sbinfo->stat_lock);
3558 return invalfc(fc, "%s", err);
3561 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3563 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3565 if (sbinfo->max_blocks != shmem_default_max_blocks())
3566 seq_printf(seq, ",size=%luk",
3567 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3568 if (sbinfo->max_inodes != shmem_default_max_inodes())
3569 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3570 if (sbinfo->mode != (0777 | S_ISVTX))
3571 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3572 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3573 seq_printf(seq, ",uid=%u",
3574 from_kuid_munged(&init_user_ns, sbinfo->uid));
3575 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3576 seq_printf(seq, ",gid=%u",
3577 from_kgid_munged(&init_user_ns, sbinfo->gid));
3580 * Showing inode{64,32} might be useful even if it's the system default,
3581 * since then people don't have to resort to checking both here and
3582 * /proc/config.gz to confirm 64-bit inums were successfully applied
3583 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3585 * We hide it when inode64 isn't the default and we are using 32-bit
3586 * inodes, since that probably just means the feature isn't even under
3591 * +-----------------+-----------------+
3592 * | TMPFS_INODE64=y | TMPFS_INODE64=n |
3593 * +------------------+-----------------+-----------------+
3594 * | full_inums=true | show | show |
3595 * | full_inums=false | show | hide |
3596 * +------------------+-----------------+-----------------+
3599 if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3600 seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3601 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3602 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3604 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3606 shmem_show_mpol(seq, sbinfo->mpol);
3610 #endif /* CONFIG_TMPFS */
3612 static void shmem_put_super(struct super_block *sb)
3614 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3616 free_percpu(sbinfo->ino_batch);
3617 percpu_counter_destroy(&sbinfo->used_blocks);
3618 mpol_put(sbinfo->mpol);
3620 sb->s_fs_info = NULL;
3623 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3625 struct shmem_options *ctx = fc->fs_private;
3626 struct inode *inode;
3627 struct shmem_sb_info *sbinfo;
3629 /* Round up to L1_CACHE_BYTES to resist false sharing */
3630 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3631 L1_CACHE_BYTES), GFP_KERNEL);
3635 sb->s_fs_info = sbinfo;
3639 * Per default we only allow half of the physical ram per
3640 * tmpfs instance, limiting inodes to one per page of lowmem;
3641 * but the internal instance is left unlimited.
3643 if (!(sb->s_flags & SB_KERNMOUNT)) {
3644 if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3645 ctx->blocks = shmem_default_max_blocks();
3646 if (!(ctx->seen & SHMEM_SEEN_INODES))
3647 ctx->inodes = shmem_default_max_inodes();
3648 if (!(ctx->seen & SHMEM_SEEN_INUMS))
3649 ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3651 sb->s_flags |= SB_NOUSER;
3653 sb->s_export_op = &shmem_export_ops;
3654 sb->s_flags |= SB_NOSEC;
3656 sb->s_flags |= SB_NOUSER;
3658 sbinfo->max_blocks = ctx->blocks;
3659 sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3660 if (sb->s_flags & SB_KERNMOUNT) {
3661 sbinfo->ino_batch = alloc_percpu(ino_t);
3662 if (!sbinfo->ino_batch)
3665 sbinfo->uid = ctx->uid;
3666 sbinfo->gid = ctx->gid;
3667 sbinfo->full_inums = ctx->full_inums;
3668 sbinfo->mode = ctx->mode;
3669 sbinfo->huge = ctx->huge;
3670 sbinfo->mpol = ctx->mpol;
3673 raw_spin_lock_init(&sbinfo->stat_lock);
3674 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3676 spin_lock_init(&sbinfo->shrinklist_lock);
3677 INIT_LIST_HEAD(&sbinfo->shrinklist);
3679 sb->s_maxbytes = MAX_LFS_FILESIZE;
3680 sb->s_blocksize = PAGE_SIZE;
3681 sb->s_blocksize_bits = PAGE_SHIFT;
3682 sb->s_magic = TMPFS_MAGIC;
3683 sb->s_op = &shmem_ops;
3684 sb->s_time_gran = 1;
3685 #ifdef CONFIG_TMPFS_XATTR
3686 sb->s_xattr = shmem_xattr_handlers;
3688 #ifdef CONFIG_TMPFS_POSIX_ACL
3689 sb->s_flags |= SB_POSIXACL;
3691 uuid_gen(&sb->s_uuid);
3693 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3696 inode->i_uid = sbinfo->uid;
3697 inode->i_gid = sbinfo->gid;
3698 sb->s_root = d_make_root(inode);
3704 shmem_put_super(sb);
3708 static int shmem_get_tree(struct fs_context *fc)
3710 return get_tree_nodev(fc, shmem_fill_super);
3713 static void shmem_free_fc(struct fs_context *fc)
3715 struct shmem_options *ctx = fc->fs_private;
3718 mpol_put(ctx->mpol);
3723 static const struct fs_context_operations shmem_fs_context_ops = {
3724 .free = shmem_free_fc,
3725 .get_tree = shmem_get_tree,
3727 .parse_monolithic = shmem_parse_options,
3728 .parse_param = shmem_parse_one,
3729 .reconfigure = shmem_reconfigure,
3733 static struct kmem_cache *shmem_inode_cachep;
3735 static struct inode *shmem_alloc_inode(struct super_block *sb)
3737 struct shmem_inode_info *info;
3738 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3741 return &info->vfs_inode;
3744 static void shmem_free_in_core_inode(struct inode *inode)
3746 if (S_ISLNK(inode->i_mode))
3747 kfree(inode->i_link);
3748 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3751 static void shmem_destroy_inode(struct inode *inode)
3753 if (S_ISREG(inode->i_mode))
3754 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3757 static void shmem_init_inode(void *foo)
3759 struct shmem_inode_info *info = foo;
3760 inode_init_once(&info->vfs_inode);
3763 static void shmem_init_inodecache(void)
3765 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3766 sizeof(struct shmem_inode_info),
3767 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3770 static void shmem_destroy_inodecache(void)
3772 kmem_cache_destroy(shmem_inode_cachep);
3775 const struct address_space_operations shmem_aops = {
3776 .writepage = shmem_writepage,
3777 .set_page_dirty = __set_page_dirty_no_writeback,
3779 .write_begin = shmem_write_begin,
3780 .write_end = shmem_write_end,
3782 #ifdef CONFIG_MIGRATION
3783 .migratepage = migrate_page,
3785 .error_remove_page = generic_error_remove_page,
3787 EXPORT_SYMBOL(shmem_aops);
3789 static const struct file_operations shmem_file_operations = {
3791 .get_unmapped_area = shmem_get_unmapped_area,
3793 .llseek = shmem_file_llseek,
3794 .read_iter = shmem_file_read_iter,
3795 .write_iter = generic_file_write_iter,
3796 .fsync = noop_fsync,
3797 .splice_read = generic_file_splice_read,
3798 .splice_write = iter_file_splice_write,
3799 .fallocate = shmem_fallocate,
3803 static const struct inode_operations shmem_inode_operations = {
3804 .getattr = shmem_getattr,
3805 .setattr = shmem_setattr,
3806 #ifdef CONFIG_TMPFS_XATTR
3807 .listxattr = shmem_listxattr,
3808 .set_acl = simple_set_acl,
3812 static const struct inode_operations shmem_dir_inode_operations = {
3814 .create = shmem_create,
3815 .lookup = simple_lookup,
3817 .unlink = shmem_unlink,
3818 .symlink = shmem_symlink,
3819 .mkdir = shmem_mkdir,
3820 .rmdir = shmem_rmdir,
3821 .mknod = shmem_mknod,
3822 .rename = shmem_rename2,
3823 .tmpfile = shmem_tmpfile,
3825 #ifdef CONFIG_TMPFS_XATTR
3826 .listxattr = shmem_listxattr,
3828 #ifdef CONFIG_TMPFS_POSIX_ACL
3829 .setattr = shmem_setattr,
3830 .set_acl = simple_set_acl,
3834 static const struct inode_operations shmem_special_inode_operations = {
3835 #ifdef CONFIG_TMPFS_XATTR
3836 .listxattr = shmem_listxattr,
3838 #ifdef CONFIG_TMPFS_POSIX_ACL
3839 .setattr = shmem_setattr,
3840 .set_acl = simple_set_acl,
3844 static const struct super_operations shmem_ops = {
3845 .alloc_inode = shmem_alloc_inode,
3846 .free_inode = shmem_free_in_core_inode,
3847 .destroy_inode = shmem_destroy_inode,
3849 .statfs = shmem_statfs,
3850 .show_options = shmem_show_options,
3852 .evict_inode = shmem_evict_inode,
3853 .drop_inode = generic_delete_inode,
3854 .put_super = shmem_put_super,
3855 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3856 .nr_cached_objects = shmem_unused_huge_count,
3857 .free_cached_objects = shmem_unused_huge_scan,
3861 static const struct vm_operations_struct shmem_vm_ops = {
3862 .fault = shmem_fault,
3863 .map_pages = filemap_map_pages,
3865 .set_policy = shmem_set_policy,
3866 .get_policy = shmem_get_policy,
3870 int shmem_init_fs_context(struct fs_context *fc)
3872 struct shmem_options *ctx;
3874 ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3878 ctx->mode = 0777 | S_ISVTX;
3879 ctx->uid = current_fsuid();
3880 ctx->gid = current_fsgid();
3882 fc->fs_private = ctx;
3883 fc->ops = &shmem_fs_context_ops;
3887 static struct file_system_type shmem_fs_type = {
3888 .owner = THIS_MODULE,
3890 .init_fs_context = shmem_init_fs_context,
3892 .parameters = shmem_fs_parameters,
3894 .kill_sb = kill_litter_super,
3895 .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT,
3898 int __init shmem_init(void)
3902 shmem_init_inodecache();
3904 error = register_filesystem(&shmem_fs_type);
3906 pr_err("Could not register tmpfs\n");
3910 shm_mnt = kern_mount(&shmem_fs_type);
3911 if (IS_ERR(shm_mnt)) {
3912 error = PTR_ERR(shm_mnt);
3913 pr_err("Could not kern_mount tmpfs\n");
3917 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3918 if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3919 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3921 shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
3926 unregister_filesystem(&shmem_fs_type);
3928 shmem_destroy_inodecache();
3929 shm_mnt = ERR_PTR(error);
3933 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
3934 static ssize_t shmem_enabled_show(struct kobject *kobj,
3935 struct kobj_attribute *attr, char *buf)
3937 static const int values[] = {
3939 SHMEM_HUGE_WITHIN_SIZE,
3948 for (i = 0; i < ARRAY_SIZE(values); i++) {
3949 len += sysfs_emit_at(buf, len,
3950 shmem_huge == values[i] ? "%s[%s]" : "%s%s",
3952 shmem_format_huge(values[i]));
3955 len += sysfs_emit_at(buf, len, "\n");
3960 static ssize_t shmem_enabled_store(struct kobject *kobj,
3961 struct kobj_attribute *attr, const char *buf, size_t count)
3966 if (count + 1 > sizeof(tmp))
3968 memcpy(tmp, buf, count);
3970 if (count && tmp[count - 1] == '\n')
3971 tmp[count - 1] = '\0';
3973 huge = shmem_parse_huge(tmp);
3974 if (huge == -EINVAL)
3976 if (!has_transparent_hugepage() &&
3977 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3981 if (shmem_huge > SHMEM_HUGE_DENY)
3982 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3986 struct kobj_attribute shmem_enabled_attr =
3987 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3988 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
3990 #else /* !CONFIG_SHMEM */
3993 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3995 * This is intended for small system where the benefits of the full
3996 * shmem code (swap-backed and resource-limited) are outweighed by
3997 * their complexity. On systems without swap this code should be
3998 * effectively equivalent, but much lighter weight.
4001 static struct file_system_type shmem_fs_type = {
4003 .init_fs_context = ramfs_init_fs_context,
4004 .parameters = ramfs_fs_parameters,
4005 .kill_sb = kill_litter_super,
4006 .fs_flags = FS_USERNS_MOUNT,
4009 int __init shmem_init(void)
4011 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4013 shm_mnt = kern_mount(&shmem_fs_type);
4014 BUG_ON(IS_ERR(shm_mnt));
4019 int shmem_unuse(unsigned int type, bool frontswap,
4020 unsigned long *fs_pages_to_unuse)
4025 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4030 void shmem_unlock_mapping(struct address_space *mapping)
4035 unsigned long shmem_get_unmapped_area(struct file *file,
4036 unsigned long addr, unsigned long len,
4037 unsigned long pgoff, unsigned long flags)
4039 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4043 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4045 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4047 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4049 #define shmem_vm_ops generic_file_vm_ops
4050 #define shmem_file_operations ramfs_file_operations
4051 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4052 #define shmem_acct_size(flags, size) 0
4053 #define shmem_unacct_size(flags, size) do {} while (0)
4055 #endif /* CONFIG_SHMEM */
4059 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4060 unsigned long flags, unsigned int i_flags)
4062 struct inode *inode;
4066 return ERR_CAST(mnt);
4068 if (size < 0 || size > MAX_LFS_FILESIZE)
4069 return ERR_PTR(-EINVAL);
4071 if (shmem_acct_size(flags, size))
4072 return ERR_PTR(-ENOMEM);
4074 inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4076 if (unlikely(!inode)) {
4077 shmem_unacct_size(flags, size);
4078 return ERR_PTR(-ENOSPC);
4080 inode->i_flags |= i_flags;
4081 inode->i_size = size;
4082 clear_nlink(inode); /* It is unlinked */
4083 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4085 res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4086 &shmem_file_operations);
4093 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4094 * kernel internal. There will be NO LSM permission checks against the
4095 * underlying inode. So users of this interface must do LSM checks at a
4096 * higher layer. The users are the big_key and shm implementations. LSM
4097 * checks are provided at the key or shm level rather than the inode.
4098 * @name: name for dentry (to be seen in /proc/<pid>/maps
4099 * @size: size to be set for the file
4100 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4102 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4104 return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4108 * shmem_file_setup - get an unlinked file living in tmpfs
4109 * @name: name for dentry (to be seen in /proc/<pid>/maps
4110 * @size: size to be set for the file
4111 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4113 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4115 return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4117 EXPORT_SYMBOL_GPL(shmem_file_setup);
4120 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4121 * @mnt: the tmpfs mount where the file will be created
4122 * @name: name for dentry (to be seen in /proc/<pid>/maps
4123 * @size: size to be set for the file
4124 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4126 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4127 loff_t size, unsigned long flags)
4129 return __shmem_file_setup(mnt, name, size, flags, 0);
4131 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4134 * shmem_zero_setup - setup a shared anonymous mapping
4135 * @vma: the vma to be mmapped is prepared by do_mmap
4137 int shmem_zero_setup(struct vm_area_struct *vma)
4140 loff_t size = vma->vm_end - vma->vm_start;
4143 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4144 * between XFS directory reading and selinux: since this file is only
4145 * accessible to the user through its mapping, use S_PRIVATE flag to
4146 * bypass file security, in the same way as shmem_kernel_file_setup().
4148 file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4150 return PTR_ERR(file);
4154 vma->vm_file = file;
4155 vma->vm_ops = &shmem_vm_ops;
4157 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4158 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4159 (vma->vm_end & HPAGE_PMD_MASK)) {
4160 khugepaged_enter(vma, vma->vm_flags);
4167 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4168 * @mapping: the page's address_space
4169 * @index: the page index
4170 * @gfp: the page allocator flags to use if allocating
4172 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4173 * with any new page allocations done using the specified allocation flags.
4174 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4175 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4176 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4178 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4179 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4181 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4182 pgoff_t index, gfp_t gfp)
4185 struct inode *inode = mapping->host;
4189 BUG_ON(!shmem_mapping(mapping));
4190 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4191 gfp, NULL, NULL, NULL);
4193 page = ERR_PTR(error);
4199 * The tiny !SHMEM case uses ramfs without swap
4201 return read_cache_page_gfp(mapping, index, gfp);
4204 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);