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/export.h>
33 #include <linux/swap.h>
34 #include <linux/uio.h>
35 #include <linux/khugepaged.h>
37 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
39 static struct vfsmount *shm_mnt;
43 * This virtual memory filesystem is heavily based on the ramfs. It
44 * extends ramfs by the ability to use swap and honor resource limits
45 * which makes it a completely usable filesystem.
48 #include <linux/xattr.h>
49 #include <linux/exportfs.h>
50 #include <linux/posix_acl.h>
51 #include <linux/posix_acl_xattr.h>
52 #include <linux/mman.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/backing-dev.h>
56 #include <linux/shmem_fs.h>
57 #include <linux/writeback.h>
58 #include <linux/blkdev.h>
59 #include <linux/pagevec.h>
60 #include <linux/percpu_counter.h>
61 #include <linux/falloc.h>
62 #include <linux/splice.h>
63 #include <linux/security.h>
64 #include <linux/swapops.h>
65 #include <linux/mempolicy.h>
66 #include <linux/namei.h>
67 #include <linux/ctype.h>
68 #include <linux/migrate.h>
69 #include <linux/highmem.h>
70 #include <linux/seq_file.h>
71 #include <linux/magic.h>
72 #include <linux/syscalls.h>
73 #include <linux/fcntl.h>
74 #include <uapi/linux/memfd.h>
75 #include <linux/userfaultfd_k.h>
76 #include <linux/rmap.h>
78 #include <linux/uaccess.h>
79 #include <asm/pgtable.h>
83 #define BLOCKS_PER_PAGE (PAGE_SIZE/512)
84 #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
86 /* Pretend that each entry is of this size in directory's i_size */
87 #define BOGO_DIRENT_SIZE 20
89 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
90 #define SHORT_SYMLINK_LEN 128
93 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
94 * inode->i_private (with i_mutex making sure that it has only one user at
95 * a time): we would prefer not to enlarge the shmem inode just for that.
98 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
99 pgoff_t start; /* start of range currently being fallocated */
100 pgoff_t next; /* the next page offset to be fallocated */
101 pgoff_t nr_falloced; /* how many new pages have been fallocated */
102 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
106 static unsigned long shmem_default_max_blocks(void)
108 return totalram_pages / 2;
111 static unsigned long shmem_default_max_inodes(void)
113 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
117 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
118 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
119 struct shmem_inode_info *info, pgoff_t index);
120 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
121 struct page **pagep, enum sgp_type sgp,
122 gfp_t gfp, struct vm_area_struct *vma,
123 struct vm_fault *vmf, int *fault_type);
125 int shmem_getpage(struct inode *inode, pgoff_t index,
126 struct page **pagep, enum sgp_type sgp)
128 return shmem_getpage_gfp(inode, index, pagep, sgp,
129 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
132 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
134 return sb->s_fs_info;
138 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
139 * for shared memory and for shared anonymous (/dev/zero) mappings
140 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
141 * consistent with the pre-accounting of private mappings ...
143 static inline int shmem_acct_size(unsigned long flags, loff_t size)
145 return (flags & VM_NORESERVE) ?
146 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
149 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
151 if (!(flags & VM_NORESERVE))
152 vm_unacct_memory(VM_ACCT(size));
155 static inline int shmem_reacct_size(unsigned long flags,
156 loff_t oldsize, loff_t newsize)
158 if (!(flags & VM_NORESERVE)) {
159 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
160 return security_vm_enough_memory_mm(current->mm,
161 VM_ACCT(newsize) - VM_ACCT(oldsize));
162 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
163 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
169 * ... whereas tmpfs objects are accounted incrementally as
170 * pages are allocated, in order to allow large sparse files.
171 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
172 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
174 static inline int shmem_acct_block(unsigned long flags, long pages)
176 if (!(flags & VM_NORESERVE))
179 return security_vm_enough_memory_mm(current->mm,
180 pages * VM_ACCT(PAGE_SIZE));
183 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
185 if (flags & VM_NORESERVE)
186 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
189 static const struct super_operations shmem_ops;
190 static const struct address_space_operations shmem_aops;
191 static const struct file_operations shmem_file_operations;
192 static const struct inode_operations shmem_inode_operations;
193 static const struct inode_operations shmem_dir_inode_operations;
194 static const struct inode_operations shmem_special_inode_operations;
195 static const struct vm_operations_struct shmem_vm_ops;
196 static struct file_system_type shmem_fs_type;
198 bool vma_is_shmem(struct vm_area_struct *vma)
200 return vma->vm_ops == &shmem_vm_ops;
203 static LIST_HEAD(shmem_swaplist);
204 static DEFINE_MUTEX(shmem_swaplist_mutex);
206 static int shmem_reserve_inode(struct super_block *sb)
208 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
209 if (sbinfo->max_inodes) {
210 spin_lock(&sbinfo->stat_lock);
211 if (!sbinfo->free_inodes) {
212 spin_unlock(&sbinfo->stat_lock);
215 sbinfo->free_inodes--;
216 spin_unlock(&sbinfo->stat_lock);
221 static void shmem_free_inode(struct super_block *sb)
223 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
224 if (sbinfo->max_inodes) {
225 spin_lock(&sbinfo->stat_lock);
226 sbinfo->free_inodes++;
227 spin_unlock(&sbinfo->stat_lock);
232 * shmem_recalc_inode - recalculate the block usage of an inode
233 * @inode: inode to recalc
235 * We have to calculate the free blocks since the mm can drop
236 * undirtied hole pages behind our back.
238 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
239 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
241 * It has to be called with the spinlock held.
243 static void shmem_recalc_inode(struct inode *inode)
245 struct shmem_inode_info *info = SHMEM_I(inode);
248 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
250 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
251 if (sbinfo->max_blocks)
252 percpu_counter_add(&sbinfo->used_blocks, -freed);
253 info->alloced -= freed;
254 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
255 shmem_unacct_blocks(info->flags, freed);
259 bool shmem_charge(struct inode *inode, long pages)
261 struct shmem_inode_info *info = SHMEM_I(inode);
262 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
265 if (shmem_acct_block(info->flags, pages))
267 spin_lock_irqsave(&info->lock, flags);
268 info->alloced += pages;
269 inode->i_blocks += pages * BLOCKS_PER_PAGE;
270 shmem_recalc_inode(inode);
271 spin_unlock_irqrestore(&info->lock, flags);
272 inode->i_mapping->nrpages += pages;
274 if (!sbinfo->max_blocks)
276 if (percpu_counter_compare(&sbinfo->used_blocks,
277 sbinfo->max_blocks - pages) > 0) {
278 inode->i_mapping->nrpages -= pages;
279 spin_lock_irqsave(&info->lock, flags);
280 info->alloced -= pages;
281 shmem_recalc_inode(inode);
282 spin_unlock_irqrestore(&info->lock, flags);
283 shmem_unacct_blocks(info->flags, pages);
286 percpu_counter_add(&sbinfo->used_blocks, pages);
290 void shmem_uncharge(struct inode *inode, long pages)
292 struct shmem_inode_info *info = SHMEM_I(inode);
293 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
296 spin_lock_irqsave(&info->lock, flags);
297 info->alloced -= pages;
298 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
299 shmem_recalc_inode(inode);
300 spin_unlock_irqrestore(&info->lock, flags);
302 if (sbinfo->max_blocks)
303 percpu_counter_sub(&sbinfo->used_blocks, pages);
304 shmem_unacct_blocks(info->flags, pages);
308 * Replace item expected in radix tree by a new item, while holding tree lock.
310 static int shmem_radix_tree_replace(struct address_space *mapping,
311 pgoff_t index, void *expected, void *replacement)
313 struct radix_tree_node *node;
317 VM_BUG_ON(!expected);
318 VM_BUG_ON(!replacement);
319 item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot);
322 if (item != expected)
324 __radix_tree_replace(&mapping->page_tree, node, pslot,
325 replacement, NULL, NULL);
330 * Sometimes, before we decide whether to proceed or to fail, we must check
331 * that an entry was not already brought back from swap by a racing thread.
333 * Checking page is not enough: by the time a SwapCache page is locked, it
334 * might be reused, and again be SwapCache, using the same swap as before.
336 static bool shmem_confirm_swap(struct address_space *mapping,
337 pgoff_t index, swp_entry_t swap)
342 item = radix_tree_lookup(&mapping->page_tree, index);
344 return item == swp_to_radix_entry(swap);
348 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
351 * disables huge pages for the mount;
353 * enables huge pages for the mount;
354 * SHMEM_HUGE_WITHIN_SIZE:
355 * only allocate huge pages if the page will be fully within i_size,
356 * also respect fadvise()/madvise() hints;
358 * only allocate huge pages if requested with fadvise()/madvise();
361 #define SHMEM_HUGE_NEVER 0
362 #define SHMEM_HUGE_ALWAYS 1
363 #define SHMEM_HUGE_WITHIN_SIZE 2
364 #define SHMEM_HUGE_ADVISE 3
368 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
371 * disables huge on shm_mnt and all mounts, for emergency use;
373 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
376 #define SHMEM_HUGE_DENY (-1)
377 #define SHMEM_HUGE_FORCE (-2)
379 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
380 /* ifdef here to avoid bloating shmem.o when not necessary */
382 int shmem_huge __read_mostly;
384 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
385 static int shmem_parse_huge(const char *str)
387 if (!strcmp(str, "never"))
388 return SHMEM_HUGE_NEVER;
389 if (!strcmp(str, "always"))
390 return SHMEM_HUGE_ALWAYS;
391 if (!strcmp(str, "within_size"))
392 return SHMEM_HUGE_WITHIN_SIZE;
393 if (!strcmp(str, "advise"))
394 return SHMEM_HUGE_ADVISE;
395 if (!strcmp(str, "deny"))
396 return SHMEM_HUGE_DENY;
397 if (!strcmp(str, "force"))
398 return SHMEM_HUGE_FORCE;
402 static const char *shmem_format_huge(int huge)
405 case SHMEM_HUGE_NEVER:
407 case SHMEM_HUGE_ALWAYS:
409 case SHMEM_HUGE_WITHIN_SIZE:
410 return "within_size";
411 case SHMEM_HUGE_ADVISE:
413 case SHMEM_HUGE_DENY:
415 case SHMEM_HUGE_FORCE:
424 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
425 struct shrink_control *sc, unsigned long nr_to_split)
427 LIST_HEAD(list), *pos, *next;
428 LIST_HEAD(to_remove);
430 struct shmem_inode_info *info;
432 unsigned long batch = sc ? sc->nr_to_scan : 128;
433 int removed = 0, split = 0;
435 if (list_empty(&sbinfo->shrinklist))
438 spin_lock(&sbinfo->shrinklist_lock);
439 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
440 info = list_entry(pos, struct shmem_inode_info, shrinklist);
443 inode = igrab(&info->vfs_inode);
445 /* inode is about to be evicted */
447 list_del_init(&info->shrinklist);
452 /* Check if there's anything to gain */
453 if (round_up(inode->i_size, PAGE_SIZE) ==
454 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
455 list_move(&info->shrinklist, &to_remove);
460 list_move(&info->shrinklist, &list);
465 spin_unlock(&sbinfo->shrinklist_lock);
467 list_for_each_safe(pos, next, &to_remove) {
468 info = list_entry(pos, struct shmem_inode_info, shrinklist);
469 inode = &info->vfs_inode;
470 list_del_init(&info->shrinklist);
474 list_for_each_safe(pos, next, &list) {
477 info = list_entry(pos, struct shmem_inode_info, shrinklist);
478 inode = &info->vfs_inode;
480 if (nr_to_split && split >= nr_to_split) {
485 page = find_lock_page(inode->i_mapping,
486 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
490 if (!PageTransHuge(page)) {
496 ret = split_huge_page(page);
501 /* split failed: leave it on the list */
508 list_del_init(&info->shrinklist);
513 spin_lock(&sbinfo->shrinklist_lock);
514 list_splice_tail(&list, &sbinfo->shrinklist);
515 sbinfo->shrinklist_len -= removed;
516 spin_unlock(&sbinfo->shrinklist_lock);
521 static long shmem_unused_huge_scan(struct super_block *sb,
522 struct shrink_control *sc)
524 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
526 if (!READ_ONCE(sbinfo->shrinklist_len))
529 return shmem_unused_huge_shrink(sbinfo, sc, 0);
532 static long shmem_unused_huge_count(struct super_block *sb,
533 struct shrink_control *sc)
535 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
536 return READ_ONCE(sbinfo->shrinklist_len);
538 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
540 #define shmem_huge SHMEM_HUGE_DENY
542 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
543 struct shrink_control *sc, unsigned long nr_to_split)
547 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
550 * Like add_to_page_cache_locked, but error if expected item has gone.
552 static int shmem_add_to_page_cache(struct page *page,
553 struct address_space *mapping,
554 pgoff_t index, void *expected)
556 int error, nr = hpage_nr_pages(page);
558 VM_BUG_ON_PAGE(PageTail(page), page);
559 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
560 VM_BUG_ON_PAGE(!PageLocked(page), page);
561 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
562 VM_BUG_ON(expected && PageTransHuge(page));
564 page_ref_add(page, nr);
565 page->mapping = mapping;
568 spin_lock_irq(&mapping->tree_lock);
569 if (PageTransHuge(page)) {
570 void __rcu **results;
575 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
576 &results, &idx, index, 1) &&
577 idx < index + HPAGE_PMD_NR) {
582 for (i = 0; i < HPAGE_PMD_NR; i++) {
583 error = radix_tree_insert(&mapping->page_tree,
584 index + i, page + i);
587 count_vm_event(THP_FILE_ALLOC);
589 } else if (!expected) {
590 error = radix_tree_insert(&mapping->page_tree, index, page);
592 error = shmem_radix_tree_replace(mapping, index, expected,
597 mapping->nrpages += nr;
598 if (PageTransHuge(page))
599 __inc_node_page_state(page, NR_SHMEM_THPS);
600 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
601 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
602 spin_unlock_irq(&mapping->tree_lock);
604 page->mapping = NULL;
605 spin_unlock_irq(&mapping->tree_lock);
606 page_ref_sub(page, nr);
612 * Like delete_from_page_cache, but substitutes swap for page.
614 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
616 struct address_space *mapping = page->mapping;
619 VM_BUG_ON_PAGE(PageCompound(page), page);
621 spin_lock_irq(&mapping->tree_lock);
622 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
623 page->mapping = NULL;
625 __dec_node_page_state(page, NR_FILE_PAGES);
626 __dec_node_page_state(page, NR_SHMEM);
627 spin_unlock_irq(&mapping->tree_lock);
633 * Remove swap entry from radix tree, free the swap and its page cache.
635 static int shmem_free_swap(struct address_space *mapping,
636 pgoff_t index, void *radswap)
640 spin_lock_irq(&mapping->tree_lock);
641 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
642 spin_unlock_irq(&mapping->tree_lock);
645 free_swap_and_cache(radix_to_swp_entry(radswap));
650 * Determine (in bytes) how many of the shmem object's pages mapped by the
651 * given offsets are swapped out.
653 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
654 * as long as the inode doesn't go away and racy results are not a problem.
656 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
657 pgoff_t start, pgoff_t end)
659 struct radix_tree_iter iter;
662 unsigned long swapped = 0;
666 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
667 if (iter.index >= end)
670 page = radix_tree_deref_slot(slot);
672 if (radix_tree_deref_retry(page)) {
673 slot = radix_tree_iter_retry(&iter);
677 if (radix_tree_exceptional_entry(page))
680 if (need_resched()) {
681 slot = radix_tree_iter_resume(slot, &iter);
688 return swapped << PAGE_SHIFT;
692 * Determine (in bytes) how many of the shmem object's pages mapped by the
693 * given vma is swapped out.
695 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
696 * as long as the inode doesn't go away and racy results are not a problem.
698 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
700 struct inode *inode = file_inode(vma->vm_file);
701 struct shmem_inode_info *info = SHMEM_I(inode);
702 struct address_space *mapping = inode->i_mapping;
703 unsigned long swapped;
705 /* Be careful as we don't hold info->lock */
706 swapped = READ_ONCE(info->swapped);
709 * The easier cases are when the shmem object has nothing in swap, or
710 * the vma maps it whole. Then we can simply use the stats that we
716 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
717 return swapped << PAGE_SHIFT;
719 /* Here comes the more involved part */
720 return shmem_partial_swap_usage(mapping,
721 linear_page_index(vma, vma->vm_start),
722 linear_page_index(vma, vma->vm_end));
726 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
728 void shmem_unlock_mapping(struct address_space *mapping)
731 pgoff_t indices[PAGEVEC_SIZE];
734 pagevec_init(&pvec, 0);
736 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
738 while (!mapping_unevictable(mapping)) {
740 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
741 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
743 pvec.nr = find_get_entries(mapping, index,
744 PAGEVEC_SIZE, pvec.pages, indices);
747 index = indices[pvec.nr - 1] + 1;
748 pagevec_remove_exceptionals(&pvec);
749 check_move_unevictable_pages(pvec.pages, pvec.nr);
750 pagevec_release(&pvec);
756 * Remove range of pages and swap entries from radix tree, and free them.
757 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
759 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
762 struct address_space *mapping = inode->i_mapping;
763 struct shmem_inode_info *info = SHMEM_I(inode);
764 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
765 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
766 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
767 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
769 pgoff_t indices[PAGEVEC_SIZE];
770 long nr_swaps_freed = 0;
775 end = -1; /* unsigned, so actually very big */
777 pagevec_init(&pvec, 0);
779 while (index < end) {
780 pvec.nr = find_get_entries(mapping, index,
781 min(end - index, (pgoff_t)PAGEVEC_SIZE),
782 pvec.pages, indices);
785 for (i = 0; i < pagevec_count(&pvec); i++) {
786 struct page *page = pvec.pages[i];
792 if (radix_tree_exceptional_entry(page)) {
795 nr_swaps_freed += !shmem_free_swap(mapping,
800 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
802 if (!trylock_page(page))
805 if (PageTransTail(page)) {
806 /* Middle of THP: zero out the page */
807 clear_highpage(page);
810 } else if (PageTransHuge(page)) {
811 if (index == round_down(end, HPAGE_PMD_NR)) {
813 * Range ends in the middle of THP:
816 clear_highpage(page);
820 index += HPAGE_PMD_NR - 1;
821 i += HPAGE_PMD_NR - 1;
824 if (!unfalloc || !PageUptodate(page)) {
825 VM_BUG_ON_PAGE(PageTail(page), page);
826 if (page_mapping(page) == mapping) {
827 VM_BUG_ON_PAGE(PageWriteback(page), page);
828 truncate_inode_page(mapping, page);
833 pagevec_remove_exceptionals(&pvec);
834 pagevec_release(&pvec);
840 struct page *page = NULL;
841 shmem_getpage(inode, start - 1, &page, SGP_READ);
843 unsigned int top = PAGE_SIZE;
848 zero_user_segment(page, partial_start, top);
849 set_page_dirty(page);
855 struct page *page = NULL;
856 shmem_getpage(inode, end, &page, SGP_READ);
858 zero_user_segment(page, 0, partial_end);
859 set_page_dirty(page);
868 while (index < end) {
871 pvec.nr = find_get_entries(mapping, index,
872 min(end - index, (pgoff_t)PAGEVEC_SIZE),
873 pvec.pages, indices);
875 /* If all gone or hole-punch or unfalloc, we're done */
876 if (index == start || end != -1)
878 /* But if truncating, restart to make sure all gone */
882 for (i = 0; i < pagevec_count(&pvec); i++) {
883 struct page *page = pvec.pages[i];
889 if (radix_tree_exceptional_entry(page)) {
892 if (shmem_free_swap(mapping, index, page)) {
893 /* Swap was replaced by page: retry */
903 if (PageTransTail(page)) {
904 /* Middle of THP: zero out the page */
905 clear_highpage(page);
908 * Partial thp truncate due 'start' in middle
909 * of THP: don't need to look on these pages
910 * again on !pvec.nr restart.
912 if (index != round_down(end, HPAGE_PMD_NR))
915 } else if (PageTransHuge(page)) {
916 if (index == round_down(end, HPAGE_PMD_NR)) {
918 * Range ends in the middle of THP:
921 clear_highpage(page);
925 index += HPAGE_PMD_NR - 1;
926 i += HPAGE_PMD_NR - 1;
929 if (!unfalloc || !PageUptodate(page)) {
930 VM_BUG_ON_PAGE(PageTail(page), page);
931 if (page_mapping(page) == mapping) {
932 VM_BUG_ON_PAGE(PageWriteback(page), page);
933 truncate_inode_page(mapping, page);
935 /* Page was replaced by swap: retry */
943 pagevec_remove_exceptionals(&pvec);
944 pagevec_release(&pvec);
948 spin_lock_irq(&info->lock);
949 info->swapped -= nr_swaps_freed;
950 shmem_recalc_inode(inode);
951 spin_unlock_irq(&info->lock);
954 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
956 shmem_undo_range(inode, lstart, lend, false);
957 inode->i_ctime = inode->i_mtime = current_time(inode);
959 EXPORT_SYMBOL_GPL(shmem_truncate_range);
961 static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
964 struct inode *inode = dentry->d_inode;
965 struct shmem_inode_info *info = SHMEM_I(inode);
967 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
968 spin_lock_irq(&info->lock);
969 shmem_recalc_inode(inode);
970 spin_unlock_irq(&info->lock);
972 generic_fillattr(inode, stat);
976 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
978 struct inode *inode = d_inode(dentry);
979 struct shmem_inode_info *info = SHMEM_I(inode);
980 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
983 error = setattr_prepare(dentry, attr);
987 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
988 loff_t oldsize = inode->i_size;
989 loff_t newsize = attr->ia_size;
991 /* protected by i_mutex */
992 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
993 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
996 if (newsize != oldsize) {
997 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1001 i_size_write(inode, newsize);
1002 inode->i_ctime = inode->i_mtime = current_time(inode);
1004 if (newsize <= oldsize) {
1005 loff_t holebegin = round_up(newsize, PAGE_SIZE);
1006 if (oldsize > holebegin)
1007 unmap_mapping_range(inode->i_mapping,
1010 shmem_truncate_range(inode,
1011 newsize, (loff_t)-1);
1012 /* unmap again to remove racily COWed private pages */
1013 if (oldsize > holebegin)
1014 unmap_mapping_range(inode->i_mapping,
1018 * Part of the huge page can be beyond i_size: subject
1019 * to shrink under memory pressure.
1021 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1022 spin_lock(&sbinfo->shrinklist_lock);
1023 if (list_empty(&info->shrinklist)) {
1024 list_add_tail(&info->shrinklist,
1025 &sbinfo->shrinklist);
1026 sbinfo->shrinklist_len++;
1028 spin_unlock(&sbinfo->shrinklist_lock);
1033 setattr_copy(inode, attr);
1034 if (attr->ia_valid & ATTR_MODE)
1035 error = posix_acl_chmod(inode, inode->i_mode);
1039 static void shmem_evict_inode(struct inode *inode)
1041 struct shmem_inode_info *info = SHMEM_I(inode);
1042 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1044 if (inode->i_mapping->a_ops == &shmem_aops) {
1045 shmem_unacct_size(info->flags, inode->i_size);
1047 shmem_truncate_range(inode, 0, (loff_t)-1);
1048 if (!list_empty(&info->shrinklist)) {
1049 spin_lock(&sbinfo->shrinklist_lock);
1050 if (!list_empty(&info->shrinklist)) {
1051 list_del_init(&info->shrinklist);
1052 sbinfo->shrinklist_len--;
1054 spin_unlock(&sbinfo->shrinklist_lock);
1056 if (!list_empty(&info->swaplist)) {
1057 mutex_lock(&shmem_swaplist_mutex);
1058 list_del_init(&info->swaplist);
1059 mutex_unlock(&shmem_swaplist_mutex);
1063 simple_xattrs_free(&info->xattrs);
1064 WARN_ON(inode->i_blocks);
1065 shmem_free_inode(inode->i_sb);
1069 static unsigned long find_swap_entry(struct radix_tree_root *root, void *item)
1071 struct radix_tree_iter iter;
1073 unsigned long found = -1;
1074 unsigned int checked = 0;
1077 radix_tree_for_each_slot(slot, root, &iter, 0) {
1078 if (*slot == item) {
1083 if ((checked % 4096) != 0)
1085 slot = radix_tree_iter_resume(slot, &iter);
1094 * If swap found in inode, free it and move page from swapcache to filecache.
1096 static int shmem_unuse_inode(struct shmem_inode_info *info,
1097 swp_entry_t swap, struct page **pagep)
1099 struct address_space *mapping = info->vfs_inode.i_mapping;
1105 radswap = swp_to_radix_entry(swap);
1106 index = find_swap_entry(&mapping->page_tree, radswap);
1108 return -EAGAIN; /* tell shmem_unuse we found nothing */
1111 * Move _head_ to start search for next from here.
1112 * But be careful: shmem_evict_inode checks list_empty without taking
1113 * mutex, and there's an instant in list_move_tail when info->swaplist
1114 * would appear empty, if it were the only one on shmem_swaplist.
1116 if (shmem_swaplist.next != &info->swaplist)
1117 list_move_tail(&shmem_swaplist, &info->swaplist);
1119 gfp = mapping_gfp_mask(mapping);
1120 if (shmem_should_replace_page(*pagep, gfp)) {
1121 mutex_unlock(&shmem_swaplist_mutex);
1122 error = shmem_replace_page(pagep, gfp, info, index);
1123 mutex_lock(&shmem_swaplist_mutex);
1125 * We needed to drop mutex to make that restrictive page
1126 * allocation, but the inode might have been freed while we
1127 * dropped it: although a racing shmem_evict_inode() cannot
1128 * complete without emptying the radix_tree, our page lock
1129 * on this swapcache page is not enough to prevent that -
1130 * free_swap_and_cache() of our swap entry will only
1131 * trylock_page(), removing swap from radix_tree whatever.
1133 * We must not proceed to shmem_add_to_page_cache() if the
1134 * inode has been freed, but of course we cannot rely on
1135 * inode or mapping or info to check that. However, we can
1136 * safely check if our swap entry is still in use (and here
1137 * it can't have got reused for another page): if it's still
1138 * in use, then the inode cannot have been freed yet, and we
1139 * can safely proceed (if it's no longer in use, that tells
1140 * nothing about the inode, but we don't need to unuse swap).
1142 if (!page_swapcount(*pagep))
1147 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1148 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1149 * beneath us (pagelock doesn't help until the page is in pagecache).
1152 error = shmem_add_to_page_cache(*pagep, mapping, index,
1154 if (error != -ENOMEM) {
1156 * Truncation and eviction use free_swap_and_cache(), which
1157 * only does trylock page: if we raced, best clean up here.
1159 delete_from_swap_cache(*pagep);
1160 set_page_dirty(*pagep);
1162 spin_lock_irq(&info->lock);
1164 spin_unlock_irq(&info->lock);
1172 * Search through swapped inodes to find and replace swap by page.
1174 int shmem_unuse(swp_entry_t swap, struct page *page)
1176 struct list_head *this, *next;
1177 struct shmem_inode_info *info;
1178 struct mem_cgroup *memcg;
1182 * There's a faint possibility that swap page was replaced before
1183 * caller locked it: caller will come back later with the right page.
1185 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1189 * Charge page using GFP_KERNEL while we can wait, before taking
1190 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1191 * Charged back to the user (not to caller) when swap account is used.
1193 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1197 /* No radix_tree_preload: swap entry keeps a place for page in tree */
1200 mutex_lock(&shmem_swaplist_mutex);
1201 list_for_each_safe(this, next, &shmem_swaplist) {
1202 info = list_entry(this, struct shmem_inode_info, swaplist);
1204 error = shmem_unuse_inode(info, swap, &page);
1206 list_del_init(&info->swaplist);
1208 if (error != -EAGAIN)
1210 /* found nothing in this: move on to search the next */
1212 mutex_unlock(&shmem_swaplist_mutex);
1215 if (error != -ENOMEM)
1217 mem_cgroup_cancel_charge(page, memcg, false);
1219 mem_cgroup_commit_charge(page, memcg, true, false);
1227 * Move the page from the page cache to the swap cache.
1229 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1231 struct shmem_inode_info *info;
1232 struct address_space *mapping;
1233 struct inode *inode;
1237 VM_BUG_ON_PAGE(PageCompound(page), page);
1238 BUG_ON(!PageLocked(page));
1239 mapping = page->mapping;
1240 index = page->index;
1241 inode = mapping->host;
1242 info = SHMEM_I(inode);
1243 if (info->flags & VM_LOCKED)
1245 if (!total_swap_pages)
1249 * Our capabilities prevent regular writeback or sync from ever calling
1250 * shmem_writepage; but a stacking filesystem might use ->writepage of
1251 * its underlying filesystem, in which case tmpfs should write out to
1252 * swap only in response to memory pressure, and not for the writeback
1255 if (!wbc->for_reclaim) {
1256 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1261 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1262 * value into swapfile.c, the only way we can correctly account for a
1263 * fallocated page arriving here is now to initialize it and write it.
1265 * That's okay for a page already fallocated earlier, but if we have
1266 * not yet completed the fallocation, then (a) we want to keep track
1267 * of this page in case we have to undo it, and (b) it may not be a
1268 * good idea to continue anyway, once we're pushing into swap. So
1269 * reactivate the page, and let shmem_fallocate() quit when too many.
1271 if (!PageUptodate(page)) {
1272 if (inode->i_private) {
1273 struct shmem_falloc *shmem_falloc;
1274 spin_lock(&inode->i_lock);
1275 shmem_falloc = inode->i_private;
1277 !shmem_falloc->waitq &&
1278 index >= shmem_falloc->start &&
1279 index < shmem_falloc->next)
1280 shmem_falloc->nr_unswapped++;
1282 shmem_falloc = NULL;
1283 spin_unlock(&inode->i_lock);
1287 clear_highpage(page);
1288 flush_dcache_page(page);
1289 SetPageUptodate(page);
1292 swap = get_swap_page();
1296 if (mem_cgroup_try_charge_swap(page, swap))
1300 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1301 * if it's not already there. Do it now before the page is
1302 * moved to swap cache, when its pagelock no longer protects
1303 * the inode from eviction. But don't unlock the mutex until
1304 * we've incremented swapped, because shmem_unuse_inode() will
1305 * prune a !swapped inode from the swaplist under this mutex.
1307 mutex_lock(&shmem_swaplist_mutex);
1308 if (list_empty(&info->swaplist))
1309 list_add_tail(&info->swaplist, &shmem_swaplist);
1311 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1312 spin_lock_irq(&info->lock);
1313 shmem_recalc_inode(inode);
1315 spin_unlock_irq(&info->lock);
1317 swap_shmem_alloc(swap);
1318 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1320 mutex_unlock(&shmem_swaplist_mutex);
1321 BUG_ON(page_mapped(page));
1322 swap_writepage(page, wbc);
1326 mutex_unlock(&shmem_swaplist_mutex);
1328 swapcache_free(swap);
1330 set_page_dirty(page);
1331 if (wbc->for_reclaim)
1332 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1337 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1338 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1342 if (!mpol || mpol->mode == MPOL_DEFAULT)
1343 return; /* show nothing */
1345 mpol_to_str(buffer, sizeof(buffer), mpol);
1347 seq_printf(seq, ",mpol=%s", buffer);
1350 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1352 struct mempolicy *mpol = NULL;
1354 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1355 mpol = sbinfo->mpol;
1357 spin_unlock(&sbinfo->stat_lock);
1361 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1362 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1365 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1369 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1371 #define vm_policy vm_private_data
1374 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1375 struct shmem_inode_info *info, pgoff_t index)
1377 /* Create a pseudo vma that just contains the policy */
1379 /* Bias interleave by inode number to distribute better across nodes */
1380 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1382 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1385 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1387 /* Drop reference taken by mpol_shared_policy_lookup() */
1388 mpol_cond_put(vma->vm_policy);
1391 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1392 struct shmem_inode_info *info, pgoff_t index)
1394 struct vm_area_struct pvma;
1397 shmem_pseudo_vma_init(&pvma, info, index);
1398 page = swapin_readahead(swap, gfp, &pvma, 0);
1399 shmem_pseudo_vma_destroy(&pvma);
1404 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1405 struct shmem_inode_info *info, pgoff_t index)
1407 struct vm_area_struct pvma;
1408 struct inode *inode = &info->vfs_inode;
1409 struct address_space *mapping = inode->i_mapping;
1410 pgoff_t idx, hindex;
1411 void __rcu **results;
1414 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1417 hindex = round_down(index, HPAGE_PMD_NR);
1419 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1420 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1426 shmem_pseudo_vma_init(&pvma, info, hindex);
1427 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1428 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1429 shmem_pseudo_vma_destroy(&pvma);
1431 prep_transhuge_page(page);
1435 static struct page *shmem_alloc_page(gfp_t gfp,
1436 struct shmem_inode_info *info, pgoff_t index)
1438 struct vm_area_struct pvma;
1441 shmem_pseudo_vma_init(&pvma, info, index);
1442 page = alloc_page_vma(gfp, &pvma, 0);
1443 shmem_pseudo_vma_destroy(&pvma);
1448 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1449 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1450 pgoff_t index, bool huge)
1456 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1458 nr = huge ? HPAGE_PMD_NR : 1;
1460 if (shmem_acct_block(info->flags, nr))
1462 if (sbinfo->max_blocks) {
1463 if (percpu_counter_compare(&sbinfo->used_blocks,
1464 sbinfo->max_blocks - nr) > 0)
1466 percpu_counter_add(&sbinfo->used_blocks, nr);
1470 page = shmem_alloc_hugepage(gfp, info, index);
1472 page = shmem_alloc_page(gfp, info, index);
1474 __SetPageLocked(page);
1475 __SetPageSwapBacked(page);
1480 if (sbinfo->max_blocks)
1481 percpu_counter_add(&sbinfo->used_blocks, -nr);
1483 shmem_unacct_blocks(info->flags, nr);
1485 return ERR_PTR(err);
1489 * When a page is moved from swapcache to shmem filecache (either by the
1490 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1491 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1492 * ignorance of the mapping it belongs to. If that mapping has special
1493 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1494 * we may need to copy to a suitable page before moving to filecache.
1496 * In a future release, this may well be extended to respect cpuset and
1497 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1498 * but for now it is a simple matter of zone.
1500 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1502 return page_zonenum(page) > gfp_zone(gfp);
1505 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1506 struct shmem_inode_info *info, pgoff_t index)
1508 struct page *oldpage, *newpage;
1509 struct address_space *swap_mapping;
1514 swap_index = page_private(oldpage);
1515 swap_mapping = page_mapping(oldpage);
1518 * We have arrived here because our zones are constrained, so don't
1519 * limit chance of success by further cpuset and node constraints.
1521 gfp &= ~GFP_CONSTRAINT_MASK;
1522 newpage = shmem_alloc_page(gfp, info, index);
1527 copy_highpage(newpage, oldpage);
1528 flush_dcache_page(newpage);
1530 __SetPageLocked(newpage);
1531 __SetPageSwapBacked(newpage);
1532 SetPageUptodate(newpage);
1533 set_page_private(newpage, swap_index);
1534 SetPageSwapCache(newpage);
1537 * Our caller will very soon move newpage out of swapcache, but it's
1538 * a nice clean interface for us to replace oldpage by newpage there.
1540 spin_lock_irq(&swap_mapping->tree_lock);
1541 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1544 __inc_node_page_state(newpage, NR_FILE_PAGES);
1545 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1547 spin_unlock_irq(&swap_mapping->tree_lock);
1549 if (unlikely(error)) {
1551 * Is this possible? I think not, now that our callers check
1552 * both PageSwapCache and page_private after getting page lock;
1553 * but be defensive. Reverse old to newpage for clear and free.
1557 mem_cgroup_migrate(oldpage, newpage);
1558 lru_cache_add_anon(newpage);
1562 ClearPageSwapCache(oldpage);
1563 set_page_private(oldpage, 0);
1565 unlock_page(oldpage);
1572 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1574 * If we allocate a new one we do not mark it dirty. That's up to the
1575 * vm. If we swap it in we mark it dirty since we also free the swap
1576 * entry since a page cannot live in both the swap and page cache.
1578 * fault_mm and fault_type are only supplied by shmem_fault:
1579 * otherwise they are NULL.
1581 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1582 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1583 struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type)
1585 struct address_space *mapping = inode->i_mapping;
1586 struct shmem_inode_info *info = SHMEM_I(inode);
1587 struct shmem_sb_info *sbinfo;
1588 struct mm_struct *charge_mm;
1589 struct mem_cgroup *memcg;
1592 enum sgp_type sgp_huge = sgp;
1593 pgoff_t hindex = index;
1598 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1600 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1604 page = find_lock_entry(mapping, index);
1605 if (radix_tree_exceptional_entry(page)) {
1606 swap = radix_to_swp_entry(page);
1610 if (sgp <= SGP_CACHE &&
1611 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1616 if (page && sgp == SGP_WRITE)
1617 mark_page_accessed(page);
1619 /* fallocated page? */
1620 if (page && !PageUptodate(page)) {
1621 if (sgp != SGP_READ)
1627 if (page || (sgp == SGP_READ && !swap.val)) {
1633 * Fast cache lookup did not find it:
1634 * bring it back from swap or allocate.
1636 sbinfo = SHMEM_SB(inode->i_sb);
1637 charge_mm = vma ? vma->vm_mm : current->mm;
1640 /* Look it up and read it in.. */
1641 page = lookup_swap_cache(swap);
1643 /* Or update major stats only when swapin succeeds?? */
1645 *fault_type |= VM_FAULT_MAJOR;
1646 count_vm_event(PGMAJFAULT);
1647 mem_cgroup_count_vm_event(charge_mm,
1650 /* Here we actually start the io */
1651 page = shmem_swapin(swap, gfp, info, index);
1658 /* We have to do this with page locked to prevent races */
1660 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1661 !shmem_confirm_swap(mapping, index, swap)) {
1662 error = -EEXIST; /* try again */
1665 if (!PageUptodate(page)) {
1669 wait_on_page_writeback(page);
1671 if (shmem_should_replace_page(page, gfp)) {
1672 error = shmem_replace_page(&page, gfp, info, index);
1677 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1680 error = shmem_add_to_page_cache(page, mapping, index,
1681 swp_to_radix_entry(swap));
1683 * We already confirmed swap under page lock, and make
1684 * no memory allocation here, so usually no possibility
1685 * of error; but free_swap_and_cache() only trylocks a
1686 * page, so it is just possible that the entry has been
1687 * truncated or holepunched since swap was confirmed.
1688 * shmem_undo_range() will have done some of the
1689 * unaccounting, now delete_from_swap_cache() will do
1691 * Reset swap.val? No, leave it so "failed" goes back to
1692 * "repeat": reading a hole and writing should succeed.
1695 mem_cgroup_cancel_charge(page, memcg, false);
1696 delete_from_swap_cache(page);
1702 mem_cgroup_commit_charge(page, memcg, true, false);
1704 spin_lock_irq(&info->lock);
1706 shmem_recalc_inode(inode);
1707 spin_unlock_irq(&info->lock);
1709 if (sgp == SGP_WRITE)
1710 mark_page_accessed(page);
1712 delete_from_swap_cache(page);
1713 set_page_dirty(page);
1717 if (vma && userfaultfd_missing(vma)) {
1718 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1722 /* shmem_symlink() */
1723 if (mapping->a_ops != &shmem_aops)
1725 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1727 if (shmem_huge == SHMEM_HUGE_FORCE)
1729 switch (sbinfo->huge) {
1732 case SHMEM_HUGE_NEVER:
1734 case SHMEM_HUGE_WITHIN_SIZE:
1735 off = round_up(index, HPAGE_PMD_NR);
1736 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1737 if (i_size >= HPAGE_PMD_SIZE &&
1738 i_size >> PAGE_SHIFT >= off)
1741 case SHMEM_HUGE_ADVISE:
1742 if (sgp_huge == SGP_HUGE)
1744 /* TODO: implement fadvise() hints */
1749 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1752 alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1757 error = PTR_ERR(page);
1759 if (error != -ENOSPC)
1762 * Try to reclaim some spece by splitting a huge page
1763 * beyond i_size on the filesystem.
1767 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1768 if (ret == SHRINK_STOP)
1776 if (PageTransHuge(page))
1777 hindex = round_down(index, HPAGE_PMD_NR);
1781 if (sgp == SGP_WRITE)
1782 __SetPageReferenced(page);
1784 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1785 PageTransHuge(page));
1788 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1789 compound_order(page));
1791 error = shmem_add_to_page_cache(page, mapping, hindex,
1793 radix_tree_preload_end();
1796 mem_cgroup_cancel_charge(page, memcg,
1797 PageTransHuge(page));
1800 mem_cgroup_commit_charge(page, memcg, false,
1801 PageTransHuge(page));
1802 lru_cache_add_anon(page);
1804 spin_lock_irq(&info->lock);
1805 info->alloced += 1 << compound_order(page);
1806 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1807 shmem_recalc_inode(inode);
1808 spin_unlock_irq(&info->lock);
1811 if (PageTransHuge(page) &&
1812 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1813 hindex + HPAGE_PMD_NR - 1) {
1815 * Part of the huge page is beyond i_size: subject
1816 * to shrink under memory pressure.
1818 spin_lock(&sbinfo->shrinklist_lock);
1819 if (list_empty(&info->shrinklist)) {
1820 list_add_tail(&info->shrinklist,
1821 &sbinfo->shrinklist);
1822 sbinfo->shrinklist_len++;
1824 spin_unlock(&sbinfo->shrinklist_lock);
1828 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1830 if (sgp == SGP_FALLOC)
1834 * Let SGP_WRITE caller clear ends if write does not fill page;
1835 * but SGP_FALLOC on a page fallocated earlier must initialize
1836 * it now, lest undo on failure cancel our earlier guarantee.
1838 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1839 struct page *head = compound_head(page);
1842 for (i = 0; i < (1 << compound_order(head)); i++) {
1843 clear_highpage(head + i);
1844 flush_dcache_page(head + i);
1846 SetPageUptodate(head);
1850 /* Perhaps the file has been truncated since we checked */
1851 if (sgp <= SGP_CACHE &&
1852 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1854 ClearPageDirty(page);
1855 delete_from_page_cache(page);
1856 spin_lock_irq(&info->lock);
1857 shmem_recalc_inode(inode);
1858 spin_unlock_irq(&info->lock);
1863 *pagep = page + index - hindex;
1870 if (sbinfo->max_blocks)
1871 percpu_counter_sub(&sbinfo->used_blocks,
1872 1 << compound_order(page));
1873 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1875 if (PageTransHuge(page)) {
1881 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1888 if (error == -ENOSPC && !once++) {
1889 spin_lock_irq(&info->lock);
1890 shmem_recalc_inode(inode);
1891 spin_unlock_irq(&info->lock);
1894 if (error == -EEXIST) /* from above or from radix_tree_insert */
1900 * This is like autoremove_wake_function, but it removes the wait queue
1901 * entry unconditionally - even if something else had already woken the
1904 static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1906 int ret = default_wake_function(wait, mode, sync, key);
1907 list_del_init(&wait->task_list);
1911 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1913 struct inode *inode = file_inode(vma->vm_file);
1914 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1917 int ret = VM_FAULT_LOCKED;
1920 * Trinity finds that probing a hole which tmpfs is punching can
1921 * prevent the hole-punch from ever completing: which in turn
1922 * locks writers out with its hold on i_mutex. So refrain from
1923 * faulting pages into the hole while it's being punched. Although
1924 * shmem_undo_range() does remove the additions, it may be unable to
1925 * keep up, as each new page needs its own unmap_mapping_range() call,
1926 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1928 * It does not matter if we sometimes reach this check just before the
1929 * hole-punch begins, so that one fault then races with the punch:
1930 * we just need to make racing faults a rare case.
1932 * The implementation below would be much simpler if we just used a
1933 * standard mutex or completion: but we cannot take i_mutex in fault,
1934 * and bloating every shmem inode for this unlikely case would be sad.
1936 if (unlikely(inode->i_private)) {
1937 struct shmem_falloc *shmem_falloc;
1939 spin_lock(&inode->i_lock);
1940 shmem_falloc = inode->i_private;
1942 shmem_falloc->waitq &&
1943 vmf->pgoff >= shmem_falloc->start &&
1944 vmf->pgoff < shmem_falloc->next) {
1945 wait_queue_head_t *shmem_falloc_waitq;
1946 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1948 ret = VM_FAULT_NOPAGE;
1949 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1950 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1951 /* It's polite to up mmap_sem if we can */
1952 up_read(&vma->vm_mm->mmap_sem);
1953 ret = VM_FAULT_RETRY;
1956 shmem_falloc_waitq = shmem_falloc->waitq;
1957 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1958 TASK_UNINTERRUPTIBLE);
1959 spin_unlock(&inode->i_lock);
1963 * shmem_falloc_waitq points into the shmem_fallocate()
1964 * stack of the hole-punching task: shmem_falloc_waitq
1965 * is usually invalid by the time we reach here, but
1966 * finish_wait() does not dereference it in that case;
1967 * though i_lock needed lest racing with wake_up_all().
1969 spin_lock(&inode->i_lock);
1970 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1971 spin_unlock(&inode->i_lock);
1974 spin_unlock(&inode->i_lock);
1978 if (vma->vm_flags & VM_HUGEPAGE)
1980 else if (vma->vm_flags & VM_NOHUGEPAGE)
1983 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1984 gfp, vma, vmf, &ret);
1986 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1990 unsigned long shmem_get_unmapped_area(struct file *file,
1991 unsigned long uaddr, unsigned long len,
1992 unsigned long pgoff, unsigned long flags)
1994 unsigned long (*get_area)(struct file *,
1995 unsigned long, unsigned long, unsigned long, unsigned long);
1997 unsigned long offset;
1998 unsigned long inflated_len;
1999 unsigned long inflated_addr;
2000 unsigned long inflated_offset;
2002 if (len > TASK_SIZE)
2005 get_area = current->mm->get_unmapped_area;
2006 addr = get_area(file, uaddr, len, pgoff, flags);
2008 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2010 if (IS_ERR_VALUE(addr))
2012 if (addr & ~PAGE_MASK)
2014 if (addr > TASK_SIZE - len)
2017 if (shmem_huge == SHMEM_HUGE_DENY)
2019 if (len < HPAGE_PMD_SIZE)
2021 if (flags & MAP_FIXED)
2024 * Our priority is to support MAP_SHARED mapped hugely;
2025 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2026 * But if caller specified an address hint, respect that as before.
2031 if (shmem_huge != SHMEM_HUGE_FORCE) {
2032 struct super_block *sb;
2035 VM_BUG_ON(file->f_op != &shmem_file_operations);
2036 sb = file_inode(file)->i_sb;
2039 * Called directly from mm/mmap.c, or drivers/char/mem.c
2040 * for "/dev/zero", to create a shared anonymous object.
2042 if (IS_ERR(shm_mnt))
2044 sb = shm_mnt->mnt_sb;
2046 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2050 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2051 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2053 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2056 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2057 if (inflated_len > TASK_SIZE)
2059 if (inflated_len < len)
2062 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2063 if (IS_ERR_VALUE(inflated_addr))
2065 if (inflated_addr & ~PAGE_MASK)
2068 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2069 inflated_addr += offset - inflated_offset;
2070 if (inflated_offset > offset)
2071 inflated_addr += HPAGE_PMD_SIZE;
2073 if (inflated_addr > TASK_SIZE - len)
2075 return inflated_addr;
2079 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2081 struct inode *inode = file_inode(vma->vm_file);
2082 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2085 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2088 struct inode *inode = file_inode(vma->vm_file);
2091 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2092 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2096 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2098 struct inode *inode = file_inode(file);
2099 struct shmem_inode_info *info = SHMEM_I(inode);
2100 int retval = -ENOMEM;
2102 spin_lock_irq(&info->lock);
2103 if (lock && !(info->flags & VM_LOCKED)) {
2104 if (!user_shm_lock(inode->i_size, user))
2106 info->flags |= VM_LOCKED;
2107 mapping_set_unevictable(file->f_mapping);
2109 if (!lock && (info->flags & VM_LOCKED) && user) {
2110 user_shm_unlock(inode->i_size, user);
2111 info->flags &= ~VM_LOCKED;
2112 mapping_clear_unevictable(file->f_mapping);
2117 spin_unlock_irq(&info->lock);
2121 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2123 file_accessed(file);
2124 vma->vm_ops = &shmem_vm_ops;
2125 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2126 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2127 (vma->vm_end & HPAGE_PMD_MASK)) {
2128 khugepaged_enter(vma, vma->vm_flags);
2133 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2134 umode_t mode, dev_t dev, unsigned long flags)
2136 struct inode *inode;
2137 struct shmem_inode_info *info;
2138 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2140 if (shmem_reserve_inode(sb))
2143 inode = new_inode(sb);
2145 inode->i_ino = get_next_ino();
2146 inode_init_owner(inode, dir, mode);
2147 inode->i_blocks = 0;
2148 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2149 inode->i_generation = get_seconds();
2150 info = SHMEM_I(inode);
2151 memset(info, 0, (char *)inode - (char *)info);
2152 spin_lock_init(&info->lock);
2153 info->seals = F_SEAL_SEAL;
2154 info->flags = flags & VM_NORESERVE;
2155 INIT_LIST_HEAD(&info->shrinklist);
2156 INIT_LIST_HEAD(&info->swaplist);
2157 simple_xattrs_init(&info->xattrs);
2158 cache_no_acl(inode);
2160 switch (mode & S_IFMT) {
2162 inode->i_op = &shmem_special_inode_operations;
2163 init_special_inode(inode, mode, dev);
2166 inode->i_mapping->a_ops = &shmem_aops;
2167 inode->i_op = &shmem_inode_operations;
2168 inode->i_fop = &shmem_file_operations;
2169 mpol_shared_policy_init(&info->policy,
2170 shmem_get_sbmpol(sbinfo));
2174 /* Some things misbehave if size == 0 on a directory */
2175 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2176 inode->i_op = &shmem_dir_inode_operations;
2177 inode->i_fop = &simple_dir_operations;
2181 * Must not load anything in the rbtree,
2182 * mpol_free_shared_policy will not be called.
2184 mpol_shared_policy_init(&info->policy, NULL);
2188 shmem_free_inode(sb);
2192 bool shmem_mapping(struct address_space *mapping)
2194 return mapping->a_ops == &shmem_aops;
2197 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2199 struct vm_area_struct *dst_vma,
2200 unsigned long dst_addr,
2201 unsigned long src_addr,
2202 struct page **pagep)
2204 struct inode *inode = file_inode(dst_vma->vm_file);
2205 struct shmem_inode_info *info = SHMEM_I(inode);
2206 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2207 struct address_space *mapping = inode->i_mapping;
2208 gfp_t gfp = mapping_gfp_mask(mapping);
2209 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2210 struct mem_cgroup *memcg;
2214 pte_t _dst_pte, *dst_pte;
2219 if (shmem_acct_block(info->flags, 1))
2221 if (sbinfo->max_blocks) {
2222 if (percpu_counter_compare(&sbinfo->used_blocks,
2223 sbinfo->max_blocks) >= 0)
2224 goto out_unacct_blocks;
2225 percpu_counter_inc(&sbinfo->used_blocks);
2228 page = shmem_alloc_page(gfp, info, pgoff);
2230 goto out_dec_used_blocks;
2232 page_kaddr = kmap_atomic(page);
2233 ret = copy_from_user(page_kaddr, (const void __user *)src_addr,
2235 kunmap_atomic(page_kaddr);
2237 /* fallback to copy_from_user outside mmap_sem */
2238 if (unlikely(ret)) {
2240 /* don't free the page */
2248 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2249 __SetPageLocked(page);
2250 __SetPageSwapBacked(page);
2251 __SetPageUptodate(page);
2253 ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false);
2257 ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
2259 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL);
2260 radix_tree_preload_end();
2263 goto out_release_uncharge;
2265 mem_cgroup_commit_charge(page, memcg, false, false);
2267 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2268 if (dst_vma->vm_flags & VM_WRITE)
2269 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2272 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2273 if (!pte_none(*dst_pte))
2274 goto out_release_uncharge_unlock;
2276 lru_cache_add_anon(page);
2278 spin_lock(&info->lock);
2280 inode->i_blocks += BLOCKS_PER_PAGE;
2281 shmem_recalc_inode(inode);
2282 spin_unlock(&info->lock);
2284 inc_mm_counter(dst_mm, mm_counter_file(page));
2285 page_add_file_rmap(page, false);
2286 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2288 /* No need to invalidate - it was non-present before */
2289 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2291 pte_unmap_unlock(dst_pte, ptl);
2295 out_release_uncharge_unlock:
2296 pte_unmap_unlock(dst_pte, ptl);
2297 out_release_uncharge:
2298 mem_cgroup_cancel_charge(page, memcg, false);
2302 out_dec_used_blocks:
2303 if (sbinfo->max_blocks)
2304 percpu_counter_add(&sbinfo->used_blocks, -1);
2306 shmem_unacct_blocks(info->flags, 1);
2311 static const struct inode_operations shmem_symlink_inode_operations;
2312 static const struct inode_operations shmem_short_symlink_operations;
2314 #ifdef CONFIG_TMPFS_XATTR
2315 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2317 #define shmem_initxattrs NULL
2321 shmem_write_begin(struct file *file, struct address_space *mapping,
2322 loff_t pos, unsigned len, unsigned flags,
2323 struct page **pagep, void **fsdata)
2325 struct inode *inode = mapping->host;
2326 struct shmem_inode_info *info = SHMEM_I(inode);
2327 pgoff_t index = pos >> PAGE_SHIFT;
2329 /* i_mutex is held by caller */
2330 if (unlikely(info->seals)) {
2331 if (info->seals & F_SEAL_WRITE)
2333 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2337 return shmem_getpage(inode, index, pagep, SGP_WRITE);
2341 shmem_write_end(struct file *file, struct address_space *mapping,
2342 loff_t pos, unsigned len, unsigned copied,
2343 struct page *page, void *fsdata)
2345 struct inode *inode = mapping->host;
2347 if (pos + copied > inode->i_size)
2348 i_size_write(inode, pos + copied);
2350 if (!PageUptodate(page)) {
2351 struct page *head = compound_head(page);
2352 if (PageTransCompound(page)) {
2355 for (i = 0; i < HPAGE_PMD_NR; i++) {
2356 if (head + i == page)
2358 clear_highpage(head + i);
2359 flush_dcache_page(head + i);
2362 if (copied < PAGE_SIZE) {
2363 unsigned from = pos & (PAGE_SIZE - 1);
2364 zero_user_segments(page, 0, from,
2365 from + copied, PAGE_SIZE);
2367 SetPageUptodate(head);
2369 set_page_dirty(page);
2376 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2378 struct file *file = iocb->ki_filp;
2379 struct inode *inode = file_inode(file);
2380 struct address_space *mapping = inode->i_mapping;
2382 unsigned long offset;
2383 enum sgp_type sgp = SGP_READ;
2386 loff_t *ppos = &iocb->ki_pos;
2389 * Might this read be for a stacking filesystem? Then when reading
2390 * holes of a sparse file, we actually need to allocate those pages,
2391 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2393 if (!iter_is_iovec(to))
2396 index = *ppos >> PAGE_SHIFT;
2397 offset = *ppos & ~PAGE_MASK;
2400 struct page *page = NULL;
2402 unsigned long nr, ret;
2403 loff_t i_size = i_size_read(inode);
2405 end_index = i_size >> PAGE_SHIFT;
2406 if (index > end_index)
2408 if (index == end_index) {
2409 nr = i_size & ~PAGE_MASK;
2414 error = shmem_getpage(inode, index, &page, sgp);
2416 if (error == -EINVAL)
2421 if (sgp == SGP_CACHE)
2422 set_page_dirty(page);
2427 * We must evaluate after, since reads (unlike writes)
2428 * are called without i_mutex protection against truncate
2431 i_size = i_size_read(inode);
2432 end_index = i_size >> PAGE_SHIFT;
2433 if (index == end_index) {
2434 nr = i_size & ~PAGE_MASK;
2445 * If users can be writing to this page using arbitrary
2446 * virtual addresses, take care about potential aliasing
2447 * before reading the page on the kernel side.
2449 if (mapping_writably_mapped(mapping))
2450 flush_dcache_page(page);
2452 * Mark the page accessed if we read the beginning.
2455 mark_page_accessed(page);
2457 page = ZERO_PAGE(0);
2462 * Ok, we have the page, and it's up-to-date, so
2463 * now we can copy it to user space...
2465 ret = copy_page_to_iter(page, offset, nr, to);
2468 index += offset >> PAGE_SHIFT;
2469 offset &= ~PAGE_MASK;
2472 if (!iov_iter_count(to))
2481 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2482 file_accessed(file);
2483 return retval ? retval : error;
2487 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2489 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2490 pgoff_t index, pgoff_t end, int whence)
2493 struct pagevec pvec;
2494 pgoff_t indices[PAGEVEC_SIZE];
2498 pagevec_init(&pvec, 0);
2499 pvec.nr = 1; /* start small: we may be there already */
2501 pvec.nr = find_get_entries(mapping, index,
2502 pvec.nr, pvec.pages, indices);
2504 if (whence == SEEK_DATA)
2508 for (i = 0; i < pvec.nr; i++, index++) {
2509 if (index < indices[i]) {
2510 if (whence == SEEK_HOLE) {
2516 page = pvec.pages[i];
2517 if (page && !radix_tree_exceptional_entry(page)) {
2518 if (!PageUptodate(page))
2522 (page && whence == SEEK_DATA) ||
2523 (!page && whence == SEEK_HOLE)) {
2528 pagevec_remove_exceptionals(&pvec);
2529 pagevec_release(&pvec);
2530 pvec.nr = PAGEVEC_SIZE;
2536 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2538 struct address_space *mapping = file->f_mapping;
2539 struct inode *inode = mapping->host;
2543 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2544 return generic_file_llseek_size(file, offset, whence,
2545 MAX_LFS_FILESIZE, i_size_read(inode));
2547 /* We're holding i_mutex so we can access i_size directly */
2551 else if (offset >= inode->i_size)
2554 start = offset >> PAGE_SHIFT;
2555 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2556 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2557 new_offset <<= PAGE_SHIFT;
2558 if (new_offset > offset) {
2559 if (new_offset < inode->i_size)
2560 offset = new_offset;
2561 else if (whence == SEEK_DATA)
2564 offset = inode->i_size;
2569 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2570 inode_unlock(inode);
2575 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2576 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2578 #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2579 #define LAST_SCAN 4 /* about 150ms max */
2581 static void shmem_tag_pins(struct address_space *mapping)
2583 struct radix_tree_iter iter;
2592 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2593 page = radix_tree_deref_slot(slot);
2594 if (!page || radix_tree_exception(page)) {
2595 if (radix_tree_deref_retry(page)) {
2596 slot = radix_tree_iter_retry(&iter);
2599 } else if (page_count(page) - page_mapcount(page) > 1) {
2600 spin_lock_irq(&mapping->tree_lock);
2601 radix_tree_tag_set(&mapping->page_tree, iter.index,
2603 spin_unlock_irq(&mapping->tree_lock);
2606 if (need_resched()) {
2607 slot = radix_tree_iter_resume(slot, &iter);
2615 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2616 * via get_user_pages(), drivers might have some pending I/O without any active
2617 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2618 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2619 * them to be dropped.
2620 * The caller must guarantee that no new user will acquire writable references
2621 * to those pages to avoid races.
2623 static int shmem_wait_for_pins(struct address_space *mapping)
2625 struct radix_tree_iter iter;
2631 shmem_tag_pins(mapping);
2634 for (scan = 0; scan <= LAST_SCAN; scan++) {
2635 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2639 lru_add_drain_all();
2640 else if (schedule_timeout_killable((HZ << scan) / 200))
2645 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2646 start, SHMEM_TAG_PINNED) {
2648 page = radix_tree_deref_slot(slot);
2649 if (radix_tree_exception(page)) {
2650 if (radix_tree_deref_retry(page)) {
2651 slot = radix_tree_iter_retry(&iter);
2659 page_count(page) - page_mapcount(page) != 1) {
2660 if (scan < LAST_SCAN)
2661 goto continue_resched;
2664 * On the last scan, we clean up all those tags
2665 * we inserted; but make a note that we still
2666 * found pages pinned.
2671 spin_lock_irq(&mapping->tree_lock);
2672 radix_tree_tag_clear(&mapping->page_tree,
2673 iter.index, SHMEM_TAG_PINNED);
2674 spin_unlock_irq(&mapping->tree_lock);
2676 if (need_resched()) {
2677 slot = radix_tree_iter_resume(slot, &iter);
2687 #define F_ALL_SEALS (F_SEAL_SEAL | \
2692 int shmem_add_seals(struct file *file, unsigned int seals)
2694 struct inode *inode = file_inode(file);
2695 struct shmem_inode_info *info = SHMEM_I(inode);
2700 * Sealing allows multiple parties to share a shmem-file but restrict
2701 * access to a specific subset of file operations. Seals can only be
2702 * added, but never removed. This way, mutually untrusted parties can
2703 * share common memory regions with a well-defined policy. A malicious
2704 * peer can thus never perform unwanted operations on a shared object.
2706 * Seals are only supported on special shmem-files and always affect
2707 * the whole underlying inode. Once a seal is set, it may prevent some
2708 * kinds of access to the file. Currently, the following seals are
2710 * SEAL_SEAL: Prevent further seals from being set on this file
2711 * SEAL_SHRINK: Prevent the file from shrinking
2712 * SEAL_GROW: Prevent the file from growing
2713 * SEAL_WRITE: Prevent write access to the file
2715 * As we don't require any trust relationship between two parties, we
2716 * must prevent seals from being removed. Therefore, sealing a file
2717 * only adds a given set of seals to the file, it never touches
2718 * existing seals. Furthermore, the "setting seals"-operation can be
2719 * sealed itself, which basically prevents any further seal from being
2722 * Semantics of sealing are only defined on volatile files. Only
2723 * anonymous shmem files support sealing. More importantly, seals are
2724 * never written to disk. Therefore, there's no plan to support it on
2728 if (file->f_op != &shmem_file_operations)
2730 if (!(file->f_mode & FMODE_WRITE))
2732 if (seals & ~(unsigned int)F_ALL_SEALS)
2737 if (info->seals & F_SEAL_SEAL) {
2742 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2743 error = mapping_deny_writable(file->f_mapping);
2747 error = shmem_wait_for_pins(file->f_mapping);
2749 mapping_allow_writable(file->f_mapping);
2754 info->seals |= seals;
2758 inode_unlock(inode);
2761 EXPORT_SYMBOL_GPL(shmem_add_seals);
2763 int shmem_get_seals(struct file *file)
2765 if (file->f_op != &shmem_file_operations)
2768 return SHMEM_I(file_inode(file))->seals;
2770 EXPORT_SYMBOL_GPL(shmem_get_seals);
2772 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2778 /* disallow upper 32bit */
2782 error = shmem_add_seals(file, arg);
2785 error = shmem_get_seals(file);
2795 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2798 struct inode *inode = file_inode(file);
2799 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2800 struct shmem_inode_info *info = SHMEM_I(inode);
2801 struct shmem_falloc shmem_falloc;
2802 pgoff_t start, index, end;
2805 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2810 if (mode & FALLOC_FL_PUNCH_HOLE) {
2811 struct address_space *mapping = file->f_mapping;
2812 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2813 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2814 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2816 /* protected by i_mutex */
2817 if (info->seals & F_SEAL_WRITE) {
2822 shmem_falloc.waitq = &shmem_falloc_waitq;
2823 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2824 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2825 spin_lock(&inode->i_lock);
2826 inode->i_private = &shmem_falloc;
2827 spin_unlock(&inode->i_lock);
2829 if ((u64)unmap_end > (u64)unmap_start)
2830 unmap_mapping_range(mapping, unmap_start,
2831 1 + unmap_end - unmap_start, 0);
2832 shmem_truncate_range(inode, offset, offset + len - 1);
2833 /* No need to unmap again: hole-punching leaves COWed pages */
2835 spin_lock(&inode->i_lock);
2836 inode->i_private = NULL;
2837 wake_up_all(&shmem_falloc_waitq);
2838 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
2839 spin_unlock(&inode->i_lock);
2844 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2845 error = inode_newsize_ok(inode, offset + len);
2849 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2854 start = offset >> PAGE_SHIFT;
2855 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2856 /* Try to avoid a swapstorm if len is impossible to satisfy */
2857 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2862 shmem_falloc.waitq = NULL;
2863 shmem_falloc.start = start;
2864 shmem_falloc.next = start;
2865 shmem_falloc.nr_falloced = 0;
2866 shmem_falloc.nr_unswapped = 0;
2867 spin_lock(&inode->i_lock);
2868 inode->i_private = &shmem_falloc;
2869 spin_unlock(&inode->i_lock);
2871 for (index = start; index < end; index++) {
2875 * Good, the fallocate(2) manpage permits EINTR: we may have
2876 * been interrupted because we are using up too much memory.
2878 if (signal_pending(current))
2880 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2883 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2885 /* Remove the !PageUptodate pages we added */
2886 if (index > start) {
2887 shmem_undo_range(inode,
2888 (loff_t)start << PAGE_SHIFT,
2889 ((loff_t)index << PAGE_SHIFT) - 1, true);
2895 * Inform shmem_writepage() how far we have reached.
2896 * No need for lock or barrier: we have the page lock.
2898 shmem_falloc.next++;
2899 if (!PageUptodate(page))
2900 shmem_falloc.nr_falloced++;
2903 * If !PageUptodate, leave it that way so that freeable pages
2904 * can be recognized if we need to rollback on error later.
2905 * But set_page_dirty so that memory pressure will swap rather
2906 * than free the pages we are allocating (and SGP_CACHE pages
2907 * might still be clean: we now need to mark those dirty too).
2909 set_page_dirty(page);
2915 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2916 i_size_write(inode, offset + len);
2917 inode->i_ctime = current_time(inode);
2919 spin_lock(&inode->i_lock);
2920 inode->i_private = NULL;
2921 spin_unlock(&inode->i_lock);
2923 inode_unlock(inode);
2927 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2929 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2931 buf->f_type = TMPFS_MAGIC;
2932 buf->f_bsize = PAGE_SIZE;
2933 buf->f_namelen = NAME_MAX;
2934 if (sbinfo->max_blocks) {
2935 buf->f_blocks = sbinfo->max_blocks;
2937 buf->f_bfree = sbinfo->max_blocks -
2938 percpu_counter_sum(&sbinfo->used_blocks);
2940 if (sbinfo->max_inodes) {
2941 buf->f_files = sbinfo->max_inodes;
2942 buf->f_ffree = sbinfo->free_inodes;
2944 /* else leave those fields 0 like simple_statfs */
2949 * File creation. Allocate an inode, and we're done..
2952 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2954 struct inode *inode;
2955 int error = -ENOSPC;
2957 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2959 error = simple_acl_create(dir, inode);
2962 error = security_inode_init_security(inode, dir,
2964 shmem_initxattrs, NULL);
2965 if (error && error != -EOPNOTSUPP)
2969 dir->i_size += BOGO_DIRENT_SIZE;
2970 dir->i_ctime = dir->i_mtime = current_time(dir);
2971 d_instantiate(dentry, inode);
2972 dget(dentry); /* Extra count - pin the dentry in core */
2981 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2983 struct inode *inode;
2984 int error = -ENOSPC;
2986 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2988 error = security_inode_init_security(inode, dir,
2990 shmem_initxattrs, NULL);
2991 if (error && error != -EOPNOTSUPP)
2993 error = simple_acl_create(dir, inode);
2996 d_tmpfile(dentry, inode);
3004 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3008 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
3014 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
3017 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
3023 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3025 struct inode *inode = d_inode(old_dentry);
3029 * No ordinary (disk based) filesystem counts links as inodes;
3030 * but each new link needs a new dentry, pinning lowmem, and
3031 * tmpfs dentries cannot be pruned until they are unlinked.
3033 ret = shmem_reserve_inode(inode->i_sb);
3037 dir->i_size += BOGO_DIRENT_SIZE;
3038 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3040 ihold(inode); /* New dentry reference */
3041 dget(dentry); /* Extra pinning count for the created dentry */
3042 d_instantiate(dentry, inode);
3047 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3049 struct inode *inode = d_inode(dentry);
3051 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3052 shmem_free_inode(inode->i_sb);
3054 dir->i_size -= BOGO_DIRENT_SIZE;
3055 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3057 dput(dentry); /* Undo the count from "create" - this does all the work */
3061 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3063 if (!simple_empty(dentry))
3066 drop_nlink(d_inode(dentry));
3068 return shmem_unlink(dir, dentry);
3071 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3073 bool old_is_dir = d_is_dir(old_dentry);
3074 bool new_is_dir = d_is_dir(new_dentry);
3076 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3078 drop_nlink(old_dir);
3081 drop_nlink(new_dir);
3085 old_dir->i_ctime = old_dir->i_mtime =
3086 new_dir->i_ctime = new_dir->i_mtime =
3087 d_inode(old_dentry)->i_ctime =
3088 d_inode(new_dentry)->i_ctime = current_time(old_dir);
3093 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3095 struct dentry *whiteout;
3098 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3102 error = shmem_mknod(old_dir, whiteout,
3103 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3109 * Cheat and hash the whiteout while the old dentry is still in
3110 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3112 * d_lookup() will consistently find one of them at this point,
3113 * not sure which one, but that isn't even important.
3120 * The VFS layer already does all the dentry stuff for rename,
3121 * we just have to decrement the usage count for the target if
3122 * it exists so that the VFS layer correctly free's it when it
3125 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3127 struct inode *inode = d_inode(old_dentry);
3128 int they_are_dirs = S_ISDIR(inode->i_mode);
3130 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3133 if (flags & RENAME_EXCHANGE)
3134 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3136 if (!simple_empty(new_dentry))
3139 if (flags & RENAME_WHITEOUT) {
3142 error = shmem_whiteout(old_dir, old_dentry);
3147 if (d_really_is_positive(new_dentry)) {
3148 (void) shmem_unlink(new_dir, new_dentry);
3149 if (they_are_dirs) {
3150 drop_nlink(d_inode(new_dentry));
3151 drop_nlink(old_dir);
3153 } else if (they_are_dirs) {
3154 drop_nlink(old_dir);
3158 old_dir->i_size -= BOGO_DIRENT_SIZE;
3159 new_dir->i_size += BOGO_DIRENT_SIZE;
3160 old_dir->i_ctime = old_dir->i_mtime =
3161 new_dir->i_ctime = new_dir->i_mtime =
3162 inode->i_ctime = current_time(old_dir);
3166 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3170 struct inode *inode;
3172 struct shmem_inode_info *info;
3174 len = strlen(symname) + 1;
3175 if (len > PAGE_SIZE)
3176 return -ENAMETOOLONG;
3178 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3182 error = security_inode_init_security(inode, dir, &dentry->d_name,
3183 shmem_initxattrs, NULL);
3185 if (error != -EOPNOTSUPP) {
3192 info = SHMEM_I(inode);
3193 inode->i_size = len-1;
3194 if (len <= SHORT_SYMLINK_LEN) {
3195 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3196 if (!inode->i_link) {
3200 inode->i_op = &shmem_short_symlink_operations;
3202 inode_nohighmem(inode);
3203 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3208 inode->i_mapping->a_ops = &shmem_aops;
3209 inode->i_op = &shmem_symlink_inode_operations;
3210 memcpy(page_address(page), symname, len);
3211 SetPageUptodate(page);
3212 set_page_dirty(page);
3216 dir->i_size += BOGO_DIRENT_SIZE;
3217 dir->i_ctime = dir->i_mtime = current_time(dir);
3218 d_instantiate(dentry, inode);
3223 static void shmem_put_link(void *arg)
3225 mark_page_accessed(arg);
3229 static const char *shmem_get_link(struct dentry *dentry,
3230 struct inode *inode,
3231 struct delayed_call *done)
3233 struct page *page = NULL;
3236 page = find_get_page(inode->i_mapping, 0);
3238 return ERR_PTR(-ECHILD);
3239 if (!PageUptodate(page)) {
3241 return ERR_PTR(-ECHILD);
3244 error = shmem_getpage(inode, 0, &page, SGP_READ);
3246 return ERR_PTR(error);
3249 set_delayed_call(done, shmem_put_link, page);
3250 return page_address(page);
3253 #ifdef CONFIG_TMPFS_XATTR
3255 * Superblocks without xattr inode operations may get some security.* xattr
3256 * support from the LSM "for free". As soon as we have any other xattrs
3257 * like ACLs, we also need to implement the security.* handlers at
3258 * filesystem level, though.
3262 * Callback for security_inode_init_security() for acquiring xattrs.
3264 static int shmem_initxattrs(struct inode *inode,
3265 const struct xattr *xattr_array,
3268 struct shmem_inode_info *info = SHMEM_I(inode);
3269 const struct xattr *xattr;
3270 struct simple_xattr *new_xattr;
3273 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3274 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3278 len = strlen(xattr->name) + 1;
3279 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3281 if (!new_xattr->name) {
3286 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3287 XATTR_SECURITY_PREFIX_LEN);
3288 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3291 simple_xattr_list_add(&info->xattrs, new_xattr);
3297 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3298 struct dentry *unused, struct inode *inode,
3299 const char *name, void *buffer, size_t size)
3301 struct shmem_inode_info *info = SHMEM_I(inode);
3303 name = xattr_full_name(handler, name);
3304 return simple_xattr_get(&info->xattrs, name, buffer, size);
3307 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3308 struct dentry *unused, struct inode *inode,
3309 const char *name, const void *value,
3310 size_t size, int flags)
3312 struct shmem_inode_info *info = SHMEM_I(inode);
3314 name = xattr_full_name(handler, name);
3315 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3318 static const struct xattr_handler shmem_security_xattr_handler = {
3319 .prefix = XATTR_SECURITY_PREFIX,
3320 .get = shmem_xattr_handler_get,
3321 .set = shmem_xattr_handler_set,
3324 static const struct xattr_handler shmem_trusted_xattr_handler = {
3325 .prefix = XATTR_TRUSTED_PREFIX,
3326 .get = shmem_xattr_handler_get,
3327 .set = shmem_xattr_handler_set,
3330 static const struct xattr_handler *shmem_xattr_handlers[] = {
3331 #ifdef CONFIG_TMPFS_POSIX_ACL
3332 &posix_acl_access_xattr_handler,
3333 &posix_acl_default_xattr_handler,
3335 &shmem_security_xattr_handler,
3336 &shmem_trusted_xattr_handler,
3340 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3342 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3343 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3345 #endif /* CONFIG_TMPFS_XATTR */
3347 static const struct inode_operations shmem_short_symlink_operations = {
3348 .get_link = simple_get_link,
3349 #ifdef CONFIG_TMPFS_XATTR
3350 .listxattr = shmem_listxattr,
3354 static const struct inode_operations shmem_symlink_inode_operations = {
3355 .get_link = shmem_get_link,
3356 #ifdef CONFIG_TMPFS_XATTR
3357 .listxattr = shmem_listxattr,
3361 static struct dentry *shmem_get_parent(struct dentry *child)
3363 return ERR_PTR(-ESTALE);
3366 static int shmem_match(struct inode *ino, void *vfh)
3370 inum = (inum << 32) | fh[1];
3371 return ino->i_ino == inum && fh[0] == ino->i_generation;
3374 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3375 struct fid *fid, int fh_len, int fh_type)
3377 struct inode *inode;
3378 struct dentry *dentry = NULL;
3385 inum = (inum << 32) | fid->raw[1];
3387 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3388 shmem_match, fid->raw);
3390 dentry = d_find_alias(inode);
3397 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3398 struct inode *parent)
3402 return FILEID_INVALID;
3405 if (inode_unhashed(inode)) {
3406 /* Unfortunately insert_inode_hash is not idempotent,
3407 * so as we hash inodes here rather than at creation
3408 * time, we need a lock to ensure we only try
3411 static DEFINE_SPINLOCK(lock);
3413 if (inode_unhashed(inode))
3414 __insert_inode_hash(inode,
3415 inode->i_ino + inode->i_generation);
3419 fh[0] = inode->i_generation;
3420 fh[1] = inode->i_ino;
3421 fh[2] = ((__u64)inode->i_ino) >> 32;
3427 static const struct export_operations shmem_export_ops = {
3428 .get_parent = shmem_get_parent,
3429 .encode_fh = shmem_encode_fh,
3430 .fh_to_dentry = shmem_fh_to_dentry,
3433 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3436 char *this_char, *value, *rest;
3437 struct mempolicy *mpol = NULL;
3441 while (options != NULL) {
3442 this_char = options;
3445 * NUL-terminate this option: unfortunately,
3446 * mount options form a comma-separated list,
3447 * but mpol's nodelist may also contain commas.
3449 options = strchr(options, ',');
3450 if (options == NULL)
3453 if (!isdigit(*options)) {
3460 if ((value = strchr(this_char,'=')) != NULL) {
3463 pr_err("tmpfs: No value for mount option '%s'\n",
3468 if (!strcmp(this_char,"size")) {
3469 unsigned long long size;
3470 size = memparse(value,&rest);
3472 size <<= PAGE_SHIFT;
3473 size *= totalram_pages;
3479 sbinfo->max_blocks =
3480 DIV_ROUND_UP(size, PAGE_SIZE);
3481 } else if (!strcmp(this_char,"nr_blocks")) {
3482 sbinfo->max_blocks = memparse(value, &rest);
3485 } else if (!strcmp(this_char,"nr_inodes")) {
3486 sbinfo->max_inodes = memparse(value, &rest);
3489 } else if (!strcmp(this_char,"mode")) {
3492 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3495 } else if (!strcmp(this_char,"uid")) {
3498 uid = simple_strtoul(value, &rest, 0);
3501 sbinfo->uid = make_kuid(current_user_ns(), uid);
3502 if (!uid_valid(sbinfo->uid))
3504 } else if (!strcmp(this_char,"gid")) {
3507 gid = simple_strtoul(value, &rest, 0);
3510 sbinfo->gid = make_kgid(current_user_ns(), gid);
3511 if (!gid_valid(sbinfo->gid))
3513 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3514 } else if (!strcmp(this_char, "huge")) {
3516 huge = shmem_parse_huge(value);
3519 if (!has_transparent_hugepage() &&
3520 huge != SHMEM_HUGE_NEVER)
3522 sbinfo->huge = huge;
3525 } else if (!strcmp(this_char,"mpol")) {
3528 if (mpol_parse_str(value, &mpol))
3532 pr_err("tmpfs: Bad mount option %s\n", this_char);
3536 sbinfo->mpol = mpol;
3540 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3548 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3550 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3551 struct shmem_sb_info config = *sbinfo;
3552 unsigned long inodes;
3553 int error = -EINVAL;
3556 if (shmem_parse_options(data, &config, true))
3559 spin_lock(&sbinfo->stat_lock);
3560 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3561 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3563 if (config.max_inodes < inodes)
3566 * Those tests disallow limited->unlimited while any are in use;
3567 * but we must separately disallow unlimited->limited, because
3568 * in that case we have no record of how much is already in use.
3570 if (config.max_blocks && !sbinfo->max_blocks)
3572 if (config.max_inodes && !sbinfo->max_inodes)
3576 sbinfo->huge = config.huge;
3577 sbinfo->max_blocks = config.max_blocks;
3578 sbinfo->max_inodes = config.max_inodes;
3579 sbinfo->free_inodes = config.max_inodes - inodes;
3582 * Preserve previous mempolicy unless mpol remount option was specified.
3585 mpol_put(sbinfo->mpol);
3586 sbinfo->mpol = config.mpol; /* transfers initial ref */
3589 spin_unlock(&sbinfo->stat_lock);
3593 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3595 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3597 if (sbinfo->max_blocks != shmem_default_max_blocks())
3598 seq_printf(seq, ",size=%luk",
3599 sbinfo->max_blocks << (PAGE_SHIFT - 10));
3600 if (sbinfo->max_inodes != shmem_default_max_inodes())
3601 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3602 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3603 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3604 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3605 seq_printf(seq, ",uid=%u",
3606 from_kuid_munged(&init_user_ns, sbinfo->uid));
3607 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3608 seq_printf(seq, ",gid=%u",
3609 from_kgid_munged(&init_user_ns, sbinfo->gid));
3610 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3611 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3613 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3615 shmem_show_mpol(seq, sbinfo->mpol);
3619 #define MFD_NAME_PREFIX "memfd:"
3620 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3621 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3623 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3625 SYSCALL_DEFINE2(memfd_create,
3626 const char __user *, uname,
3627 unsigned int, flags)
3629 struct shmem_inode_info *info;
3635 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3638 /* length includes terminating zero */
3639 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3642 if (len > MFD_NAME_MAX_LEN + 1)
3645 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3649 strcpy(name, MFD_NAME_PREFIX);
3650 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3655 /* terminating-zero may have changed after strnlen_user() returned */
3656 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3661 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3667 file = shmem_file_setup(name, 0, VM_NORESERVE);
3669 error = PTR_ERR(file);
3672 info = SHMEM_I(file_inode(file));
3673 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3674 file->f_flags |= O_RDWR | O_LARGEFILE;
3675 if (flags & MFD_ALLOW_SEALING)
3676 info->seals &= ~F_SEAL_SEAL;
3678 fd_install(fd, file);
3689 #endif /* CONFIG_TMPFS */
3691 static void shmem_put_super(struct super_block *sb)
3693 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3695 percpu_counter_destroy(&sbinfo->used_blocks);
3696 mpol_put(sbinfo->mpol);
3698 sb->s_fs_info = NULL;
3701 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3703 struct inode *inode;
3704 struct shmem_sb_info *sbinfo;
3707 /* Round up to L1_CACHE_BYTES to resist false sharing */
3708 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3709 L1_CACHE_BYTES), GFP_KERNEL);
3713 sbinfo->mode = S_IRWXUGO | S_ISVTX;
3714 sbinfo->uid = current_fsuid();
3715 sbinfo->gid = current_fsgid();
3716 sb->s_fs_info = sbinfo;
3720 * Per default we only allow half of the physical ram per
3721 * tmpfs instance, limiting inodes to one per page of lowmem;
3722 * but the internal instance is left unlimited.
3724 if (!(sb->s_flags & MS_KERNMOUNT)) {
3725 sbinfo->max_blocks = shmem_default_max_blocks();
3726 sbinfo->max_inodes = shmem_default_max_inodes();
3727 if (shmem_parse_options(data, sbinfo, false)) {
3732 sb->s_flags |= MS_NOUSER;
3734 sb->s_export_op = &shmem_export_ops;
3735 sb->s_flags |= MS_NOSEC;
3737 sb->s_flags |= MS_NOUSER;
3740 spin_lock_init(&sbinfo->stat_lock);
3741 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3743 sbinfo->free_inodes = sbinfo->max_inodes;
3744 spin_lock_init(&sbinfo->shrinklist_lock);
3745 INIT_LIST_HEAD(&sbinfo->shrinklist);
3747 sb->s_maxbytes = MAX_LFS_FILESIZE;
3748 sb->s_blocksize = PAGE_SIZE;
3749 sb->s_blocksize_bits = PAGE_SHIFT;
3750 sb->s_magic = TMPFS_MAGIC;
3751 sb->s_op = &shmem_ops;
3752 sb->s_time_gran = 1;
3753 #ifdef CONFIG_TMPFS_XATTR
3754 sb->s_xattr = shmem_xattr_handlers;
3756 #ifdef CONFIG_TMPFS_POSIX_ACL
3757 sb->s_flags |= MS_POSIXACL;
3760 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3763 inode->i_uid = sbinfo->uid;
3764 inode->i_gid = sbinfo->gid;
3765 sb->s_root = d_make_root(inode);
3771 shmem_put_super(sb);
3775 static struct kmem_cache *shmem_inode_cachep;
3777 static struct inode *shmem_alloc_inode(struct super_block *sb)
3779 struct shmem_inode_info *info;
3780 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3783 return &info->vfs_inode;
3786 static void shmem_destroy_callback(struct rcu_head *head)
3788 struct inode *inode = container_of(head, struct inode, i_rcu);
3789 if (S_ISLNK(inode->i_mode))
3790 kfree(inode->i_link);
3791 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3794 static void shmem_destroy_inode(struct inode *inode)
3796 if (S_ISREG(inode->i_mode))
3797 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3798 call_rcu(&inode->i_rcu, shmem_destroy_callback);
3801 static void shmem_init_inode(void *foo)
3803 struct shmem_inode_info *info = foo;
3804 inode_init_once(&info->vfs_inode);
3807 static int shmem_init_inodecache(void)
3809 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3810 sizeof(struct shmem_inode_info),
3811 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3815 static void shmem_destroy_inodecache(void)
3817 kmem_cache_destroy(shmem_inode_cachep);
3820 static const struct address_space_operations shmem_aops = {
3821 .writepage = shmem_writepage,
3822 .set_page_dirty = __set_page_dirty_no_writeback,
3824 .write_begin = shmem_write_begin,
3825 .write_end = shmem_write_end,
3827 #ifdef CONFIG_MIGRATION
3828 .migratepage = migrate_page,
3830 .error_remove_page = generic_error_remove_page,
3833 static const struct file_operations shmem_file_operations = {
3835 .get_unmapped_area = shmem_get_unmapped_area,
3837 .llseek = shmem_file_llseek,
3838 .read_iter = shmem_file_read_iter,
3839 .write_iter = generic_file_write_iter,
3840 .fsync = noop_fsync,
3841 .splice_read = generic_file_splice_read,
3842 .splice_write = iter_file_splice_write,
3843 .fallocate = shmem_fallocate,
3847 static const struct inode_operations shmem_inode_operations = {
3848 .getattr = shmem_getattr,
3849 .setattr = shmem_setattr,
3850 #ifdef CONFIG_TMPFS_XATTR
3851 .listxattr = shmem_listxattr,
3852 .set_acl = simple_set_acl,
3856 static const struct inode_operations shmem_dir_inode_operations = {
3858 .create = shmem_create,
3859 .lookup = simple_lookup,
3861 .unlink = shmem_unlink,
3862 .symlink = shmem_symlink,
3863 .mkdir = shmem_mkdir,
3864 .rmdir = shmem_rmdir,
3865 .mknod = shmem_mknod,
3866 .rename = shmem_rename2,
3867 .tmpfile = shmem_tmpfile,
3869 #ifdef CONFIG_TMPFS_XATTR
3870 .listxattr = shmem_listxattr,
3872 #ifdef CONFIG_TMPFS_POSIX_ACL
3873 .setattr = shmem_setattr,
3874 .set_acl = simple_set_acl,
3878 static const struct inode_operations shmem_special_inode_operations = {
3879 #ifdef CONFIG_TMPFS_XATTR
3880 .listxattr = shmem_listxattr,
3882 #ifdef CONFIG_TMPFS_POSIX_ACL
3883 .setattr = shmem_setattr,
3884 .set_acl = simple_set_acl,
3888 static const struct super_operations shmem_ops = {
3889 .alloc_inode = shmem_alloc_inode,
3890 .destroy_inode = shmem_destroy_inode,
3892 .statfs = shmem_statfs,
3893 .remount_fs = shmem_remount_fs,
3894 .show_options = shmem_show_options,
3896 .evict_inode = shmem_evict_inode,
3897 .drop_inode = generic_delete_inode,
3898 .put_super = shmem_put_super,
3899 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3900 .nr_cached_objects = shmem_unused_huge_count,
3901 .free_cached_objects = shmem_unused_huge_scan,
3905 static const struct vm_operations_struct shmem_vm_ops = {
3906 .fault = shmem_fault,
3907 .map_pages = filemap_map_pages,
3909 .set_policy = shmem_set_policy,
3910 .get_policy = shmem_get_policy,
3914 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3915 int flags, const char *dev_name, void *data)
3917 return mount_nodev(fs_type, flags, data, shmem_fill_super);
3920 static struct file_system_type shmem_fs_type = {
3921 .owner = THIS_MODULE,
3923 .mount = shmem_mount,
3924 .kill_sb = kill_litter_super,
3925 .fs_flags = FS_USERNS_MOUNT,
3928 int __init shmem_init(void)
3932 /* If rootfs called this, don't re-init */
3933 if (shmem_inode_cachep)
3936 error = shmem_init_inodecache();
3940 error = register_filesystem(&shmem_fs_type);
3942 pr_err("Could not register tmpfs\n");
3946 shm_mnt = kern_mount(&shmem_fs_type);
3947 if (IS_ERR(shm_mnt)) {
3948 error = PTR_ERR(shm_mnt);
3949 pr_err("Could not kern_mount tmpfs\n");
3953 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3954 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3955 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3957 shmem_huge = 0; /* just in case it was patched */
3962 unregister_filesystem(&shmem_fs_type);
3964 shmem_destroy_inodecache();
3966 shm_mnt = ERR_PTR(error);
3970 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3971 static ssize_t shmem_enabled_show(struct kobject *kobj,
3972 struct kobj_attribute *attr, char *buf)
3976 SHMEM_HUGE_WITHIN_SIZE,
3984 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3985 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3987 count += sprintf(buf + count, fmt,
3988 shmem_format_huge(values[i]));
3990 buf[count - 1] = '\n';
3994 static ssize_t shmem_enabled_store(struct kobject *kobj,
3995 struct kobj_attribute *attr, const char *buf, size_t count)
4000 if (count + 1 > sizeof(tmp))
4002 memcpy(tmp, buf, count);
4004 if (count && tmp[count - 1] == '\n')
4005 tmp[count - 1] = '\0';
4007 huge = shmem_parse_huge(tmp);
4008 if (huge == -EINVAL)
4010 if (!has_transparent_hugepage() &&
4011 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4015 if (shmem_huge < SHMEM_HUGE_DENY)
4016 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4020 struct kobj_attribute shmem_enabled_attr =
4021 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4022 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4024 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4025 bool shmem_huge_enabled(struct vm_area_struct *vma)
4027 struct inode *inode = file_inode(vma->vm_file);
4028 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4032 if (shmem_huge == SHMEM_HUGE_FORCE)
4034 if (shmem_huge == SHMEM_HUGE_DENY)
4036 switch (sbinfo->huge) {
4037 case SHMEM_HUGE_NEVER:
4039 case SHMEM_HUGE_ALWAYS:
4041 case SHMEM_HUGE_WITHIN_SIZE:
4042 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4043 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4044 if (i_size >= HPAGE_PMD_SIZE &&
4045 i_size >> PAGE_SHIFT >= off)
4047 case SHMEM_HUGE_ADVISE:
4048 /* TODO: implement fadvise() hints */
4049 return (vma->vm_flags & VM_HUGEPAGE);
4055 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4057 #else /* !CONFIG_SHMEM */
4060 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4062 * This is intended for small system where the benefits of the full
4063 * shmem code (swap-backed and resource-limited) are outweighed by
4064 * their complexity. On systems without swap this code should be
4065 * effectively equivalent, but much lighter weight.
4068 static struct file_system_type shmem_fs_type = {
4070 .mount = ramfs_mount,
4071 .kill_sb = kill_litter_super,
4072 .fs_flags = FS_USERNS_MOUNT,
4075 int __init shmem_init(void)
4077 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4079 shm_mnt = kern_mount(&shmem_fs_type);
4080 BUG_ON(IS_ERR(shm_mnt));
4085 int shmem_unuse(swp_entry_t swap, struct page *page)
4090 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4095 void shmem_unlock_mapping(struct address_space *mapping)
4100 unsigned long shmem_get_unmapped_area(struct file *file,
4101 unsigned long addr, unsigned long len,
4102 unsigned long pgoff, unsigned long flags)
4104 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4108 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4110 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4112 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4114 #define shmem_vm_ops generic_file_vm_ops
4115 #define shmem_file_operations ramfs_file_operations
4116 #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4117 #define shmem_acct_size(flags, size) 0
4118 #define shmem_unacct_size(flags, size) do {} while (0)
4120 #endif /* CONFIG_SHMEM */
4124 static const struct dentry_operations anon_ops = {
4125 .d_dname = simple_dname
4128 static struct file *__shmem_file_setup(const char *name, loff_t size,
4129 unsigned long flags, unsigned int i_flags)
4132 struct inode *inode;
4134 struct super_block *sb;
4137 if (IS_ERR(shm_mnt))
4138 return ERR_CAST(shm_mnt);
4140 if (size < 0 || size > MAX_LFS_FILESIZE)
4141 return ERR_PTR(-EINVAL);
4143 if (shmem_acct_size(flags, size))
4144 return ERR_PTR(-ENOMEM);
4146 res = ERR_PTR(-ENOMEM);
4148 this.len = strlen(name);
4149 this.hash = 0; /* will go */
4150 sb = shm_mnt->mnt_sb;
4151 path.mnt = mntget(shm_mnt);
4152 path.dentry = d_alloc_pseudo(sb, &this);
4155 d_set_d_op(path.dentry, &anon_ops);
4157 res = ERR_PTR(-ENOSPC);
4158 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4162 inode->i_flags |= i_flags;
4163 d_instantiate(path.dentry, inode);
4164 inode->i_size = size;
4165 clear_nlink(inode); /* It is unlinked */
4166 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4170 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4171 &shmem_file_operations);
4178 shmem_unacct_size(flags, size);
4185 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4186 * kernel internal. There will be NO LSM permission checks against the
4187 * underlying inode. So users of this interface must do LSM checks at a
4188 * higher layer. The users are the big_key and shm implementations. LSM
4189 * checks are provided at the key or shm level rather than the inode.
4190 * @name: name for dentry (to be seen in /proc/<pid>/maps
4191 * @size: size to be set for the file
4192 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4194 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4196 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4200 * shmem_file_setup - get an unlinked file living in tmpfs
4201 * @name: name for dentry (to be seen in /proc/<pid>/maps
4202 * @size: size to be set for the file
4203 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4205 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4207 return __shmem_file_setup(name, size, flags, 0);
4209 EXPORT_SYMBOL_GPL(shmem_file_setup);
4212 * shmem_zero_setup - setup a shared anonymous mapping
4213 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4215 int shmem_zero_setup(struct vm_area_struct *vma)
4218 loff_t size = vma->vm_end - vma->vm_start;
4221 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4222 * between XFS directory reading and selinux: since this file is only
4223 * accessible to the user through its mapping, use S_PRIVATE flag to
4224 * bypass file security, in the same way as shmem_kernel_file_setup().
4226 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4228 return PTR_ERR(file);
4232 vma->vm_file = file;
4233 vma->vm_ops = &shmem_vm_ops;
4235 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4236 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4237 (vma->vm_end & HPAGE_PMD_MASK)) {
4238 khugepaged_enter(vma, vma->vm_flags);
4245 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4246 * @mapping: the page's address_space
4247 * @index: the page index
4248 * @gfp: the page allocator flags to use if allocating
4250 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4251 * with any new page allocations done using the specified allocation flags.
4252 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4253 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4254 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4256 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4257 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4259 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4260 pgoff_t index, gfp_t gfp)
4263 struct inode *inode = mapping->host;
4267 BUG_ON(mapping->a_ops != &shmem_aops);
4268 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4269 gfp, NULL, NULL, NULL);
4271 page = ERR_PTR(error);
4277 * The tiny !SHMEM case uses ramfs without swap
4279 return read_cache_page_gfp(mapping, index, gfp);
4282 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);