path: root/mm/shmem.c (plain)
blob: 21521551b6b5a86fe750ca21a0e386f6b24d67d2

1	/*
2	* Resizable virtual memory filesystem for Linux.
3	*
4	* Copyright (C) 2000 Linus Torvalds.
5	* 2000 Transmeta Corp.
6	* 2000-2001 Christoph Rohland
7	* 2000-2001 SAP AG
8	* 2002 Red Hat Inc.
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
13	*
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>
17	*
18	* tiny-shmem:
19	* Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20	*
21	* This file is released under the GPL.
22	*/
23
24	#include <linux/fs.h>
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>
31	#include <linux/mm.h>
32	#include <linux/export.h>
33	#include <linux/swap.h>
34	#include <linux/uio.h>
35	#include <linux/khugepaged.h>
36
37	static struct vfsmount *shm_mnt;
38
39	#ifdef CONFIG_SHMEM
40	/*
41	* This virtual memory filesystem is heavily based on the ramfs. It
42	* extends ramfs by the ability to use swap and honor resource limits
43	* which makes it a completely usable filesystem.
44	*/
45
46	#include <linux/xattr.h>
47	#include <linux/exportfs.h>
48	#include <linux/posix_acl.h>
49	#include <linux/posix_acl_xattr.h>
50	#include <linux/mman.h>
51	#include <linux/string.h>
52	#include <linux/slab.h>
53	#include <linux/backing-dev.h>
54	#include <linux/shmem_fs.h>
55	#include <linux/writeback.h>
56	#include <linux/blkdev.h>
57	#include <linux/pagevec.h>
58	#include <linux/percpu_counter.h>
59	#include <linux/falloc.h>
60	#include <linux/splice.h>
61	#include <linux/security.h>
62	#include <linux/swapops.h>
63	#include <linux/mempolicy.h>
64	#include <linux/namei.h>
65	#include <linux/ctype.h>
66	#include <linux/migrate.h>
67	#include <linux/highmem.h>
68	#include <linux/seq_file.h>
69	#include <linux/magic.h>
70	#include <linux/syscalls.h>
71	#include <linux/fcntl.h>
72	#include <uapi/linux/memfd.h>
73
74	#include <asm/uaccess.h>
75	#include <asm/pgtable.h>
76
77	#include "internal.h"
78
79	#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
80	#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
81
82	/* Pretend that each entry is of this size in directory's i_size */
83	#define BOGO_DIRENT_SIZE 20
84
85	/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
86	#define SHORT_SYMLINK_LEN 128
87
88	/*
89	* shmem_fallocate communicates with shmem_fault or shmem_writepage via
90	* inode->i_private (with i_mutex making sure that it has only one user at
91	* a time): we would prefer not to enlarge the shmem inode just for that.
92	*/
93	struct shmem_falloc {
94	wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
95	pgoff_t start; /* start of range currently being fallocated */
96	pgoff_t next; /* the next page offset to be fallocated */
97	pgoff_t nr_falloced; /* how many new pages have been fallocated */
98	pgoff_t nr_unswapped; /* how often writepage refused to swap out */
99	};
100
101	#ifdef CONFIG_TMPFS
102	static unsigned long shmem_default_max_blocks(void)
103	{
104	return totalram_pages / 2;
105	}
106
107	static unsigned long shmem_default_max_inodes(void)
108	{
109	return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
110	}
111	#endif
112
113	static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
114	static int shmem_replace_page(struct page **pagep, gfp_t gfp,
115	struct shmem_inode_info *info, pgoff_t index);
116	static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
117	struct page **pagep, enum sgp_type sgp,
118	gfp_t gfp, struct mm_struct *fault_mm, int *fault_type);
119
120	int shmem_getpage(struct inode *inode, pgoff_t index,
121	struct page **pagep, enum sgp_type sgp)
122	{
123	return shmem_getpage_gfp(inode, index, pagep, sgp,
124	mapping_gfp_mask(inode->i_mapping), NULL, NULL);
125	}
126
127	static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
128	{
129	return sb->s_fs_info;
130	}
131
132	/*
133	* shmem_file_setup pre-accounts the whole fixed size of a VM object,
134	* for shared memory and for shared anonymous (/dev/zero) mappings
135	* (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
136	* consistent with the pre-accounting of private mappings ...
137	*/
138	static inline int shmem_acct_size(unsigned long flags, loff_t size)
139	{
140	return (flags & VM_NORESERVE) ?
141	0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
142	}
143
144	static inline void shmem_unacct_size(unsigned long flags, loff_t size)
145	{
146	if (!(flags & VM_NORESERVE))
147	vm_unacct_memory(VM_ACCT(size));
148	}
149
150	static inline int shmem_reacct_size(unsigned long flags,
151	loff_t oldsize, loff_t newsize)
152	{
153	if (!(flags & VM_NORESERVE)) {
154	if (VM_ACCT(newsize) > VM_ACCT(oldsize))
155	return security_vm_enough_memory_mm(current->mm,
156	VM_ACCT(newsize) - VM_ACCT(oldsize));
157	else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
158	vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
159	}
160	return 0;
161	}
162
163	/*
164	* ... whereas tmpfs objects are accounted incrementally as
165	* pages are allocated, in order to allow large sparse files.
166	* shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167	* so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
168	*/
169	static inline int shmem_acct_block(unsigned long flags, long pages)
170	{
171	if (!(flags & VM_NORESERVE))
172	return 0;
173
174	return security_vm_enough_memory_mm(current->mm,
175	pages * VM_ACCT(PAGE_SIZE));
176	}
177
178	static inline void shmem_unacct_blocks(unsigned long flags, long pages)
179	{
180	if (flags & VM_NORESERVE)
181	vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
182	}
183
184	static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
185	{
186	struct shmem_inode_info *info = SHMEM_I(inode);
187	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
188
189	if (shmem_acct_block(info->flags, pages))
190	return false;
191
192	if (sbinfo->max_blocks) {
193	if (percpu_counter_compare(&sbinfo->used_blocks,
194	sbinfo->max_blocks - pages) > 0)
195	goto unacct;
196	percpu_counter_add(&sbinfo->used_blocks, pages);
197	}
198
199	return true;
200
201	unacct:
202	shmem_unacct_blocks(info->flags, pages);
203	return false;
204	}
205
206	static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
207	{
208	struct shmem_inode_info *info = SHMEM_I(inode);
209	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
210
211	if (sbinfo->max_blocks)
212	percpu_counter_sub(&sbinfo->used_blocks, pages);
213	shmem_unacct_blocks(info->flags, pages);
214	}
215
216	static const struct super_operations shmem_ops;
217	static const struct address_space_operations shmem_aops;
218	static const struct file_operations shmem_file_operations;
219	static const struct inode_operations shmem_inode_operations;
220	static const struct inode_operations shmem_dir_inode_operations;
221	static const struct inode_operations shmem_special_inode_operations;
222	static const struct vm_operations_struct shmem_vm_ops;
223	static struct file_system_type shmem_fs_type;
224
225	static LIST_HEAD(shmem_swaplist);
226	static DEFINE_MUTEX(shmem_swaplist_mutex);
227
228	static int shmem_reserve_inode(struct super_block *sb)
229	{
230	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
231	if (sbinfo->max_inodes) {
232	spin_lock(&sbinfo->stat_lock);
233	if (!sbinfo->free_inodes) {
234	spin_unlock(&sbinfo->stat_lock);
235	return -ENOSPC;
236	}
237	sbinfo->free_inodes--;
238	spin_unlock(&sbinfo->stat_lock);
239	}
240	return 0;
241	}
242
243	static void shmem_free_inode(struct super_block *sb)
244	{
245	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246	if (sbinfo->max_inodes) {
247	spin_lock(&sbinfo->stat_lock);
248	sbinfo->free_inodes++;
249	spin_unlock(&sbinfo->stat_lock);
250	}
251	}
252
253	/**
254	* shmem_recalc_inode - recalculate the block usage of an inode
255	* @inode: inode to recalc
256	*
257	* We have to calculate the free blocks since the mm can drop
258	* undirtied hole pages behind our back.
259	*
260	* But normally info->alloced == inode->i_mapping->nrpages + info->swapped
261	* So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
262	*
263	* It has to be called with the spinlock held.
264	*/
265	static void shmem_recalc_inode(struct inode *inode)
266	{
267	struct shmem_inode_info *info = SHMEM_I(inode);
268	long freed;
269
270	freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
271	if (freed > 0) {
272	info->alloced -= freed;
273	inode->i_blocks -= freed * BLOCKS_PER_PAGE;
274	shmem_inode_unacct_blocks(inode, freed);
275	}
276	}
277
278	bool shmem_charge(struct inode *inode, long pages)
279	{
280	struct shmem_inode_info *info = SHMEM_I(inode);
281	unsigned long flags;
282
283	if (!shmem_inode_acct_block(inode, pages))
284	return false;
285
286	/* nrpages adjustment first, then shmem_recalc_inode() when balanced */
287	inode->i_mapping->nrpages += pages;
288
289	spin_lock_irqsave(&info->lock, flags);
290	info->alloced += pages;
291	inode->i_blocks += pages * BLOCKS_PER_PAGE;
292	shmem_recalc_inode(inode);
293	spin_unlock_irqrestore(&info->lock, flags);
294
295	return true;
296	}
297
298	void shmem_uncharge(struct inode *inode, long pages)
299	{
300	struct shmem_inode_info *info = SHMEM_I(inode);
301	unsigned long flags;
302
303	/* nrpages adjustment done by __delete_from_page_cache() or caller */
304
305	spin_lock_irqsave(&info->lock, flags);
306	info->alloced -= pages;
307	inode->i_blocks -= pages * BLOCKS_PER_PAGE;
308	shmem_recalc_inode(inode);
309	spin_unlock_irqrestore(&info->lock, flags);
310
311	shmem_inode_unacct_blocks(inode, pages);
312	}
313
314	/*
315	* Replace item expected in radix tree by a new item, while holding tree lock.
316	*/
317	static int shmem_radix_tree_replace(struct address_space *mapping,
318	pgoff_t index, void *expected, void *replacement)
319	{
320	void **pslot;
321	void *item;
322
323	VM_BUG_ON(!expected);
324	VM_BUG_ON(!replacement);
325	pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
326	if (!pslot)
327	return -ENOENT;
328	item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
329	if (item != expected)
330	return -ENOENT;
331	radix_tree_replace_slot(pslot, replacement);
332	return 0;
333	}
334
335	/*
336	* Sometimes, before we decide whether to proceed or to fail, we must check
337	* that an entry was not already brought back from swap by a racing thread.
338	*
339	* Checking page is not enough: by the time a SwapCache page is locked, it
340	* might be reused, and again be SwapCache, using the same swap as before.
341	*/
342	static bool shmem_confirm_swap(struct address_space *mapping,
343	pgoff_t index, swp_entry_t swap)
344	{
345	void *item;
346
347	rcu_read_lock();
348	item = radix_tree_lookup(&mapping->page_tree, index);
349	rcu_read_unlock();
350	return item == swp_to_radix_entry(swap);
351	}
352
353	/*
354	* Definitions for "huge tmpfs": tmpfs mounted with the huge= option
355	*
356	* SHMEM_HUGE_NEVER:
357	* disables huge pages for the mount;
358	* SHMEM_HUGE_ALWAYS:
359	* enables huge pages for the mount;
360	* SHMEM_HUGE_WITHIN_SIZE:
361	* only allocate huge pages if the page will be fully within i_size,
362	* also respect fadvise()/madvise() hints;
363	* SHMEM_HUGE_ADVISE:
364	* only allocate huge pages if requested with fadvise()/madvise();
365	*/
366
367	#define SHMEM_HUGE_NEVER 0
368	#define SHMEM_HUGE_ALWAYS 1
369	#define SHMEM_HUGE_WITHIN_SIZE 2
370	#define SHMEM_HUGE_ADVISE 3
371
372	/*
373	* Special values.
374	* Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
375	*
376	* SHMEM_HUGE_DENY:
377	* disables huge on shm_mnt and all mounts, for emergency use;
378	* SHMEM_HUGE_FORCE:
379	* enables huge on shm_mnt and all mounts, w/o needing option, for testing;
380	*
381	*/
382	#define SHMEM_HUGE_DENY (-1)
383	#define SHMEM_HUGE_FORCE (-2)
384
385	#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
386	/* ifdef here to avoid bloating shmem.o when not necessary */
387
388	int shmem_huge __read_mostly;
389
390	#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
391	static int shmem_parse_huge(const char *str)
392	{
393	if (!strcmp(str, "never"))
394	return SHMEM_HUGE_NEVER;
395	if (!strcmp(str, "always"))
396	return SHMEM_HUGE_ALWAYS;
397	if (!strcmp(str, "within_size"))
398	return SHMEM_HUGE_WITHIN_SIZE;
399	if (!strcmp(str, "advise"))
400	return SHMEM_HUGE_ADVISE;
401	if (!strcmp(str, "deny"))
402	return SHMEM_HUGE_DENY;
403	if (!strcmp(str, "force"))
404	return SHMEM_HUGE_FORCE;
405	return -EINVAL;
406	}
407
408	static const char *shmem_format_huge(int huge)
409	{
410	switch (huge) {
411	case SHMEM_HUGE_NEVER:
412	return "never";
413	case SHMEM_HUGE_ALWAYS:
414	return "always";
415	case SHMEM_HUGE_WITHIN_SIZE:
416	return "within_size";
417	case SHMEM_HUGE_ADVISE:
418	return "advise";
419	case SHMEM_HUGE_DENY:
420	return "deny";
421	case SHMEM_HUGE_FORCE:
422	return "force";
423	default:
424	VM_BUG_ON(1);
425	return "bad_val";
426	}
427	}
428	#endif
429
430	static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
431	struct shrink_control *sc, unsigned long nr_to_split)
432	{
433	LIST_HEAD(list), *pos, *next;
434	LIST_HEAD(to_remove);
435	struct inode *inode;
436	struct shmem_inode_info *info;
437	struct page *page;
438	unsigned long batch = sc ? sc->nr_to_scan : 128;
439	int removed = 0, split = 0;
440
441	if (list_empty(&sbinfo->shrinklist))
442	return SHRINK_STOP;
443
444	spin_lock(&sbinfo->shrinklist_lock);
445	list_for_each_safe(pos, next, &sbinfo->shrinklist) {
446	info = list_entry(pos, struct shmem_inode_info, shrinklist);
447
448	/* pin the inode */
449	inode = igrab(&info->vfs_inode);
450
451	/* inode is about to be evicted */
452	if (!inode) {
453	list_del_init(&info->shrinklist);
454	removed++;
455	goto next;
456	}
457
458	/* Check if there's anything to gain */
459	if (round_up(inode->i_size, PAGE_SIZE) ==
460	round_up(inode->i_size, HPAGE_PMD_SIZE)) {
461	list_move(&info->shrinklist, &to_remove);
462	removed++;
463	goto next;
464	}
465
466	list_move(&info->shrinklist, &list);
467	next:
468	if (!--batch)
469	break;
470	}
471	spin_unlock(&sbinfo->shrinklist_lock);
472
473	list_for_each_safe(pos, next, &to_remove) {
474	info = list_entry(pos, struct shmem_inode_info, shrinklist);
475	inode = &info->vfs_inode;
476	list_del_init(&info->shrinklist);
477	iput(inode);
478	}
479
480	list_for_each_safe(pos, next, &list) {
481	int ret;
482
483	info = list_entry(pos, struct shmem_inode_info, shrinklist);
484	inode = &info->vfs_inode;
485
486	if (nr_to_split && split >= nr_to_split)
487	goto leave;
488
489	page = find_get_page(inode->i_mapping,
490	(inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
491	if (!page)
492	goto drop;
493
494	/* No huge page at the end of the file: nothing to split */
495	if (!PageTransHuge(page)) {
496	put_page(page);
497	goto drop;
498	}
499
500	/*
501	* Leave the inode on the list if we failed to lock
502	* the page at this time.
503	*
504	* Waiting for the lock may lead to deadlock in the
505	* reclaim path.
506	*/
507	if (!trylock_page(page)) {
508	put_page(page);
509	goto leave;
510	}
511
512	ret = split_huge_page(page);
513	unlock_page(page);
514	put_page(page);
515
516	/* If split failed leave the inode on the list */
517	if (ret)
518	goto leave;
519
520	split++;
521	drop:
522	list_del_init(&info->shrinklist);
523	removed++;
524	leave:
525	iput(inode);
526	}
527
528	spin_lock(&sbinfo->shrinklist_lock);
529	list_splice_tail(&list, &sbinfo->shrinklist);
530	sbinfo->shrinklist_len -= removed;
531	spin_unlock(&sbinfo->shrinklist_lock);
532
533	return split;
534	}
535
536	static long shmem_unused_huge_scan(struct super_block *sb,
537	struct shrink_control *sc)
538	{
539	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
540
541	if (!READ_ONCE(sbinfo->shrinklist_len))
542	return SHRINK_STOP;
543
544	return shmem_unused_huge_shrink(sbinfo, sc, 0);
545	}
546
547	static long shmem_unused_huge_count(struct super_block *sb,
548	struct shrink_control *sc)
549	{
550	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
551	return READ_ONCE(sbinfo->shrinklist_len);
552	}
553	#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
554
555	#define shmem_huge SHMEM_HUGE_DENY
556
557	static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
558	struct shrink_control *sc, unsigned long nr_to_split)
559	{
560	return 0;
561	}
562	#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
563
564	/*
565	* Like add_to_page_cache_locked, but error if expected item has gone.
566	*/
567	static int shmem_add_to_page_cache(struct page *page,
568	struct address_space *mapping,
569	pgoff_t index, void *expected)
570	{
571	int error, nr = hpage_nr_pages(page);
572
573	VM_BUG_ON_PAGE(PageTail(page), page);
574	VM_BUG_ON_PAGE(index != round_down(index, nr), page);
575	VM_BUG_ON_PAGE(!PageLocked(page), page);
576	VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
577	VM_BUG_ON(expected && PageTransHuge(page));
578
579	page_ref_add(page, nr);
580	page->mapping = mapping;
581	page->index = index;
582
583	spin_lock_irq(&mapping->tree_lock);
584	if (PageTransHuge(page)) {
585	void __rcu **results;
586	pgoff_t idx;
587	int i;
588
589	error = 0;
590	if (radix_tree_gang_lookup_slot(&mapping->page_tree,
591	&results, &idx, index, 1) &&
592	idx < index + HPAGE_PMD_NR) {
593	error = -EEXIST;
594	}
595
596	if (!error) {
597	for (i = 0; i < HPAGE_PMD_NR; i++) {
598	error = radix_tree_insert(&mapping->page_tree,
599	index + i, page + i);
600	VM_BUG_ON(error);
601	}
602	count_vm_event(THP_FILE_ALLOC);
603	}
604	} else if (!expected) {
605	error = radix_tree_insert(&mapping->page_tree, index, page);
606	} else {
607	error = shmem_radix_tree_replace(mapping, index, expected,
608	page);
609	}
610
611	if (!error) {
612	mapping->nrpages += nr;
613	if (PageTransHuge(page))
614	__inc_node_page_state(page, NR_SHMEM_THPS);
615	__mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
616	__mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
617	spin_unlock_irq(&mapping->tree_lock);
618	} else {
619	page->mapping = NULL;
620	spin_unlock_irq(&mapping->tree_lock);
621	page_ref_sub(page, nr);
622	}
623	return error;
624	}
625
626	/*
627	* Like delete_from_page_cache, but substitutes swap for page.
628	*/
629	static void shmem_delete_from_page_cache(struct page *page, void *radswap)
630	{
631	struct address_space *mapping = page->mapping;
632	int error;
633
634	VM_BUG_ON_PAGE(PageCompound(page), page);
635
636	spin_lock_irq(&mapping->tree_lock);
637	error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
638	page->mapping = NULL;
639	mapping->nrpages--;
640	__dec_node_page_state(page, NR_FILE_PAGES);
641	__dec_node_page_state(page, NR_SHMEM);
642	spin_unlock_irq(&mapping->tree_lock);
643	put_page(page);
644	BUG_ON(error);
645	}
646
647	/*
648	* Remove swap entry from radix tree, free the swap and its page cache.
649	*/
650	static int shmem_free_swap(struct address_space *mapping,
651	pgoff_t index, void *radswap)
652	{
653	void *old;
654
655	spin_lock_irq(&mapping->tree_lock);
656	old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
657	spin_unlock_irq(&mapping->tree_lock);
658	if (old != radswap)
659	return -ENOENT;
660	free_swap_and_cache(radix_to_swp_entry(radswap));
661	return 0;
662	}
663
664	/*
665	* Determine (in bytes) how many of the shmem object's pages mapped by the
666	* given offsets are swapped out.
667	*
668	* This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
669	* as long as the inode doesn't go away and racy results are not a problem.
670	*/
671	unsigned long shmem_partial_swap_usage(struct address_space *mapping,
672	pgoff_t start, pgoff_t end)
673	{
674	struct radix_tree_iter iter;
675	void **slot;
676	struct page *page;
677	unsigned long swapped = 0;
678
679	rcu_read_lock();
680
681	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
682	if (iter.index >= end)
683	break;
684
685	page = radix_tree_deref_slot(slot);
686
687	if (radix_tree_deref_retry(page)) {
688	slot = radix_tree_iter_retry(&iter);
689	continue;
690	}
691
692	if (radix_tree_exceptional_entry(page))
693	swapped++;
694
695	if (need_resched()) {
696	cond_resched_rcu();
697	slot = radix_tree_iter_next(&iter);
698	}
699	}
700
701	rcu_read_unlock();
702
703	return swapped << PAGE_SHIFT;
704	}
705
706	/*
707	* Determine (in bytes) how many of the shmem object's pages mapped by the
708	* given vma is swapped out.
709	*
710	* This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
711	* as long as the inode doesn't go away and racy results are not a problem.
712	*/
713	unsigned long shmem_swap_usage(struct vm_area_struct *vma)
714	{
715	struct inode *inode = file_inode(vma->vm_file);
716	struct shmem_inode_info *info = SHMEM_I(inode);
717	struct address_space *mapping = inode->i_mapping;
718	unsigned long swapped;
719
720	/* Be careful as we don't hold info->lock */
721	swapped = READ_ONCE(info->swapped);
722
723	/*
724	* The easier cases are when the shmem object has nothing in swap, or
725	* the vma maps it whole. Then we can simply use the stats that we
726	* already track.
727	*/
728	if (!swapped)
729	return 0;
730
731	if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
732	return swapped << PAGE_SHIFT;
733
734	/* Here comes the more involved part */
735	return shmem_partial_swap_usage(mapping,
736	linear_page_index(vma, vma->vm_start),
737	linear_page_index(vma, vma->vm_end));
738	}
739
740	/*
741	* SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
742	*/
743	void shmem_unlock_mapping(struct address_space *mapping)
744	{
745	struct pagevec pvec;
746	pgoff_t indices[PAGEVEC_SIZE];
747	pgoff_t index = 0;
748
749	pagevec_init(&pvec, 0);
750	/*
751	* Minor point, but we might as well stop if someone else SHM_LOCKs it.
752	*/
753	while (!mapping_unevictable(mapping)) {
754	/*
755	* Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
756	* has finished, if it hits a row of PAGEVEC_SIZE swap entries.
757	*/
758	pvec.nr = find_get_entries(mapping, index,
759	PAGEVEC_SIZE, pvec.pages, indices);
760	if (!pvec.nr)
761	break;
762	index = indices[pvec.nr - 1] + 1;
763	pagevec_remove_exceptionals(&pvec);
764	check_move_unevictable_pages(pvec.pages, pvec.nr);
765	pagevec_release(&pvec);
766	cond_resched();
767	}
768	}
769
770	/*
771	* Remove range of pages and swap entries from radix tree, and free them.
772	* If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
773	*/
774	static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
775	bool unfalloc)
776	{
777	struct address_space *mapping = inode->i_mapping;
778	struct shmem_inode_info *info = SHMEM_I(inode);
779	pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
780	pgoff_t end = (lend + 1) >> PAGE_SHIFT;
781	unsigned int partial_start = lstart & (PAGE_SIZE - 1);
782	unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
783	struct pagevec pvec;
784	pgoff_t indices[PAGEVEC_SIZE];
785	long nr_swaps_freed = 0;
786	pgoff_t index;
787	int i;
788
789	if (lend == -1)
790	end = -1; /* unsigned, so actually very big */
791
792	pagevec_init(&pvec, 0);
793	index = start;
794	while (index < end) {
795	pvec.nr = find_get_entries(mapping, index,
796	min(end - index, (pgoff_t)PAGEVEC_SIZE),
797	pvec.pages, indices);
798	if (!pvec.nr)
799	break;
800	for (i = 0; i < pagevec_count(&pvec); i++) {
801	struct page *page = pvec.pages[i];
802
803	index = indices[i];
804	if (index >= end)
805	break;
806
807	if (radix_tree_exceptional_entry(page)) {
808	if (unfalloc)
809	continue;
810	nr_swaps_freed += !shmem_free_swap(mapping,
811	index, page);
812	continue;
813	}
814
815	VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
816
817	if (!trylock_page(page))
818	continue;
819
820	if (PageTransTail(page)) {
821	/* Middle of THP: zero out the page */
822	clear_highpage(page);
823	unlock_page(page);
824	continue;
825	} else if (PageTransHuge(page)) {
826	if (index == round_down(end, HPAGE_PMD_NR)) {
827	/*
828	* Range ends in the middle of THP:
829	* zero out the page
830	*/
831	clear_highpage(page);
832	unlock_page(page);
833	continue;
834	}
835	index += HPAGE_PMD_NR - 1;
836	i += HPAGE_PMD_NR - 1;
837	}
838
839	if (!unfalloc || !PageUptodate(page)) {
840	VM_BUG_ON_PAGE(PageTail(page), page);
841	if (page_mapping(page) == mapping) {
842	VM_BUG_ON_PAGE(PageWriteback(page), page);
843	truncate_inode_page(mapping, page);
844	}
845	}
846	unlock_page(page);
847	}
848	pagevec_remove_exceptionals(&pvec);
849	pagevec_release(&pvec);
850	cond_resched();
851	index++;
852	}
853
854	if (partial_start) {
855	struct page *page = NULL;
856	shmem_getpage(inode, start - 1, &page, SGP_READ);
857	if (page) {
858	unsigned int top = PAGE_SIZE;
859	if (start > end) {
860	top = partial_end;
861	partial_end = 0;
862	}
863	zero_user_segment(page, partial_start, top);
864	set_page_dirty(page);
865	unlock_page(page);
866	put_page(page);
867	}
868	}
869	if (partial_end) {
870	struct page *page = NULL;
871	shmem_getpage(inode, end, &page, SGP_READ);
872	if (page) {
873	zero_user_segment(page, 0, partial_end);
874	set_page_dirty(page);
875	unlock_page(page);
876	put_page(page);
877	}
878	}
879	if (start >= end)
880	return;
881
882	index = start;
883	while (index < end) {
884	cond_resched();
885
886	pvec.nr = find_get_entries(mapping, index,
887	min(end - index, (pgoff_t)PAGEVEC_SIZE),
888	pvec.pages, indices);
889	if (!pvec.nr) {
890	/* If all gone or hole-punch or unfalloc, we're done */
891	if (index == start || end != -1)
892	break;
893	/* But if truncating, restart to make sure all gone */
894	index = start;
895	continue;
896	}
897	for (i = 0; i < pagevec_count(&pvec); i++) {
898	struct page *page = pvec.pages[i];
899
900	index = indices[i];
901	if (index >= end)
902	break;
903
904	if (radix_tree_exceptional_entry(page)) {
905	if (unfalloc)
906	continue;
907	if (shmem_free_swap(mapping, index, page)) {
908	/* Swap was replaced by page: retry */
909	index--;
910	break;
911	}
912	nr_swaps_freed++;
913	continue;
914	}
915
916	lock_page(page);
917
918	if (PageTransTail(page)) {
919	/* Middle of THP: zero out the page */
920	clear_highpage(page);
921	unlock_page(page);
922	/*
923	* Partial thp truncate due 'start' in middle
924	* of THP: don't need to look on these pages
925	* again on !pvec.nr restart.
926	*/
927	if (index != round_down(end, HPAGE_PMD_NR))
928	start++;
929	continue;
930	} else if (PageTransHuge(page)) {
931	if (index == round_down(end, HPAGE_PMD_NR)) {
932	/*
933	* Range ends in the middle of THP:
934	* zero out the page
935	*/
936	clear_highpage(page);
937	unlock_page(page);
938	continue;
939	}
940	index += HPAGE_PMD_NR - 1;
941	i += HPAGE_PMD_NR - 1;
942	}
943
944	if (!unfalloc || !PageUptodate(page)) {
945	VM_BUG_ON_PAGE(PageTail(page), page);
946	if (page_mapping(page) == mapping) {
947	VM_BUG_ON_PAGE(PageWriteback(page), page);
948	truncate_inode_page(mapping, page);
949	} else {
950	/* Page was replaced by swap: retry */
951	unlock_page(page);
952	index--;
953	break;
954	}
955	}
956	unlock_page(page);
957	}
958	pagevec_remove_exceptionals(&pvec);
959	pagevec_release(&pvec);
960	index++;
961	}
962
963	spin_lock_irq(&info->lock);
964	info->swapped -= nr_swaps_freed;
965	shmem_recalc_inode(inode);
966	spin_unlock_irq(&info->lock);
967	}
968
969	void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
970	{
971	shmem_undo_range(inode, lstart, lend, false);
972	inode->i_ctime = inode->i_mtime = current_time(inode);
973	}
974	EXPORT_SYMBOL_GPL(shmem_truncate_range);
975
976	static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
977	struct kstat *stat)
978	{
979	struct inode *inode = dentry->d_inode;
980	struct shmem_inode_info *info = SHMEM_I(inode);
981
982	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
983	spin_lock_irq(&info->lock);
984	shmem_recalc_inode(inode);
985	spin_unlock_irq(&info->lock);
986	}
987	generic_fillattr(inode, stat);
988	return 0;
989	}
990
991	static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
992	{
993	struct inode *inode = d_inode(dentry);
994	struct shmem_inode_info *info = SHMEM_I(inode);
995	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
996	int error;
997
998	error = setattr_prepare(dentry, attr);
999	if (error)
1000	return error;
1001
1002	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1003	loff_t oldsize = inode->i_size;
1004	loff_t newsize = attr->ia_size;
1005
1006	/* protected by i_mutex */
1007	if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1008	(newsize > oldsize && (info->seals & F_SEAL_GROW)))
1009	return -EPERM;
1010
1011	if (newsize != oldsize) {
1012	error = shmem_reacct_size(SHMEM_I(inode)->flags,
1013	oldsize, newsize);
1014	if (error)
1015	return error;
1016	i_size_write(inode, newsize);
1017	inode->i_ctime = inode->i_mtime = current_time(inode);
1018	}
1019	if (newsize <= oldsize) {
1020	loff_t holebegin = round_up(newsize, PAGE_SIZE);
1021	if (oldsize > holebegin)
1022	unmap_mapping_range(inode->i_mapping,
1023	holebegin, 0, 1);
1024	if (info->alloced)
1025	shmem_truncate_range(inode,
1026	newsize, (loff_t)-1);
1027	/* unmap again to remove racily COWed private pages */
1028	if (oldsize > holebegin)
1029	unmap_mapping_range(inode->i_mapping,
1030	holebegin, 0, 1);
1031
1032	/*
1033	* Part of the huge page can be beyond i_size: subject
1034	* to shrink under memory pressure.
1035	*/
1036	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1037	spin_lock(&sbinfo->shrinklist_lock);
1038	/*
1039	* _careful to defend against unlocked access to
1040	* ->shrink_list in shmem_unused_huge_shrink()
1041	*/
1042	if (list_empty_careful(&info->shrinklist)) {
1043	list_add_tail(&info->shrinklist,
1044	&sbinfo->shrinklist);
1045	sbinfo->shrinklist_len++;
1046	}
1047	spin_unlock(&sbinfo->shrinklist_lock);
1048	}
1049	}
1050	}
1051
1052	setattr_copy(inode, attr);
1053	if (attr->ia_valid & ATTR_MODE)
1054	error = posix_acl_chmod(inode, inode->i_mode);
1055	return error;
1056	}
1057
1058	static void shmem_evict_inode(struct inode *inode)
1059	{
1060	struct shmem_inode_info *info = SHMEM_I(inode);
1061	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1062
1063	if (inode->i_mapping->a_ops == &shmem_aops) {
1064	shmem_unacct_size(info->flags, inode->i_size);
1065	inode->i_size = 0;
1066	shmem_truncate_range(inode, 0, (loff_t)-1);
1067	if (!list_empty(&info->shrinklist)) {
1068	spin_lock(&sbinfo->shrinklist_lock);
1069	if (!list_empty(&info->shrinklist)) {
1070	list_del_init(&info->shrinklist);
1071	sbinfo->shrinklist_len--;
1072	}
1073	spin_unlock(&sbinfo->shrinklist_lock);
1074	}
1075	if (!list_empty(&info->swaplist)) {
1076	mutex_lock(&shmem_swaplist_mutex);
1077	list_del_init(&info->swaplist);
1078	mutex_unlock(&shmem_swaplist_mutex);
1079	}
1080	}
1081
1082	simple_xattrs_free(&info->xattrs);
1083	WARN_ON(inode->i_blocks);
1084	shmem_free_inode(inode->i_sb);
1085	clear_inode(inode);
1086	}
1087
1088	/*
1089	* If swap found in inode, free it and move page from swapcache to filecache.
1090	*/
1091	static int shmem_unuse_inode(struct shmem_inode_info *info,
1092	swp_entry_t swap, struct page **pagep)
1093	{
1094	struct address_space *mapping = info->vfs_inode.i_mapping;
1095	void *radswap;
1096	pgoff_t index;
1097	gfp_t gfp;
1098	int error = 0;
1099
1100	radswap = swp_to_radix_entry(swap);
1101	index = radix_tree_locate_item(&mapping->page_tree, radswap);
1102	if (index == -1)
1103	return -EAGAIN; /* tell shmem_unuse we found nothing */
1104
1105	/*
1106	* Move _head_ to start search for next from here.
1107	* But be careful: shmem_evict_inode checks list_empty without taking
1108	* mutex, and there's an instant in list_move_tail when info->swaplist
1109	* would appear empty, if it were the only one on shmem_swaplist.
1110	*/
1111	if (shmem_swaplist.next != &info->swaplist)
1112	list_move_tail(&shmem_swaplist, &info->swaplist);
1113
1114	gfp = mapping_gfp_mask(mapping);
1115	if (shmem_should_replace_page(*pagep, gfp)) {
1116	mutex_unlock(&shmem_swaplist_mutex);
1117	error = shmem_replace_page(pagep, gfp, info, index);
1118	mutex_lock(&shmem_swaplist_mutex);
1119	/*
1120	* We needed to drop mutex to make that restrictive page
1121	* allocation, but the inode might have been freed while we
1122	* dropped it: although a racing shmem_evict_inode() cannot
1123	* complete without emptying the radix_tree, our page lock
1124	* on this swapcache page is not enough to prevent that -
1125	* free_swap_and_cache() of our swap entry will only
1126	* trylock_page(), removing swap from radix_tree whatever.
1127	*
1128	* We must not proceed to shmem_add_to_page_cache() if the
1129	* inode has been freed, but of course we cannot rely on
1130	* inode or mapping or info to check that. However, we can
1131	* safely check if our swap entry is still in use (and here
1132	* it can't have got reused for another page): if it's still
1133	* in use, then the inode cannot have been freed yet, and we
1134	* can safely proceed (if it's no longer in use, that tells
1135	* nothing about the inode, but we don't need to unuse swap).
1136	*/
1137	if (!page_swapcount(*pagep))
1138	error = -ENOENT;
1139	}
1140
1141	/*
1142	* We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1143	* but also to hold up shmem_evict_inode(): so inode cannot be freed
1144	* beneath us (pagelock doesn't help until the page is in pagecache).
1145	*/
1146	if (!error)
1147	error = shmem_add_to_page_cache(*pagep, mapping, index,
1148	radswap);
1149	if (error != -ENOMEM) {
1150	/*
1151	* Truncation and eviction use free_swap_and_cache(), which
1152	* only does trylock page: if we raced, best clean up here.
1153	*/
1154	delete_from_swap_cache(*pagep);
1155	set_page_dirty(*pagep);
1156	if (!error) {
1157	spin_lock_irq(&info->lock);
1158	info->swapped--;
1159	spin_unlock_irq(&info->lock);
1160	swap_free(swap);
1161	}
1162	}
1163	return error;
1164	}
1165
1166	/*
1167	* Search through swapped inodes to find and replace swap by page.
1168	*/
1169	int shmem_unuse(swp_entry_t swap, struct page *page)
1170	{
1171	struct list_head *this, *next;
1172	struct shmem_inode_info *info;
1173	struct mem_cgroup *memcg;
1174	int error = 0;
1175
1176	/*
1177	* There's a faint possibility that swap page was replaced before
1178	* caller locked it: caller will come back later with the right page.
1179	*/
1180	if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1181	goto out;
1182
1183	/*
1184	* Charge page using GFP_KERNEL while we can wait, before taking
1185	* the shmem_swaplist_mutex which might hold up shmem_writepage().
1186	* Charged back to the user (not to caller) when swap account is used.
1187	*/
1188	error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1189	false);
1190	if (error)
1191	goto out;
1192	/* No radix_tree_preload: swap entry keeps a place for page in tree */
1193	error = -EAGAIN;
1194
1195	mutex_lock(&shmem_swaplist_mutex);
1196	list_for_each_safe(this, next, &shmem_swaplist) {
1197	info = list_entry(this, struct shmem_inode_info, swaplist);
1198	if (info->swapped)
1199	error = shmem_unuse_inode(info, swap, &page);
1200	else
1201	list_del_init(&info->swaplist);
1202	cond_resched();
1203	if (error != -EAGAIN)
1204	break;
1205	/* found nothing in this: move on to search the next */
1206	}
1207	mutex_unlock(&shmem_swaplist_mutex);
1208
1209	if (error) {
1210	if (error != -ENOMEM)
1211	error = 0;
1212	mem_cgroup_cancel_charge(page, memcg, false);
1213	} else
1214	mem_cgroup_commit_charge(page, memcg, true, false);
1215	out:
1216	unlock_page(page);
1217	put_page(page);
1218	return error;
1219	}
1220
1221	/*
1222	* Move the page from the page cache to the swap cache.
1223	*/
1224	static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1225	{
1226	struct shmem_inode_info *info;
1227	struct address_space *mapping;
1228	struct inode *inode;
1229	swp_entry_t swap;
1230	pgoff_t index;
1231
1232	VM_BUG_ON_PAGE(PageCompound(page), page);
1233	BUG_ON(!PageLocked(page));
1234	mapping = page->mapping;
1235	index = page->index;
1236	inode = mapping->host;
1237	info = SHMEM_I(inode);
1238	if (info->flags & VM_LOCKED)
1239	goto redirty;
1240	if (!total_swap_pages)
1241	goto redirty;
1242
1243	/*
1244	* Our capabilities prevent regular writeback or sync from ever calling
1245	* shmem_writepage; but a stacking filesystem might use ->writepage of
1246	* its underlying filesystem, in which case tmpfs should write out to
1247	* swap only in response to memory pressure, and not for the writeback
1248	* threads or sync.
1249	*/
1250	if (!wbc->for_reclaim) {
1251	WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1252	goto redirty;
1253	}
1254
1255	/*
1256	* This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1257	* value into swapfile.c, the only way we can correctly account for a
1258	* fallocated page arriving here is now to initialize it and write it.
1259	*
1260	* That's okay for a page already fallocated earlier, but if we have
1261	* not yet completed the fallocation, then (a) we want to keep track
1262	* of this page in case we have to undo it, and (b) it may not be a
1263	* good idea to continue anyway, once we're pushing into swap. So
1264	* reactivate the page, and let shmem_fallocate() quit when too many.
1265	*/
1266	if (!PageUptodate(page)) {
1267	if (inode->i_private) {
1268	struct shmem_falloc *shmem_falloc;
1269	spin_lock(&inode->i_lock);
1270	shmem_falloc = inode->i_private;
1271	if (shmem_falloc &&
1272	!shmem_falloc->waitq &&
1273	index >= shmem_falloc->start &&
1274	index < shmem_falloc->next)
1275	shmem_falloc->nr_unswapped++;
1276	else
1277	shmem_falloc = NULL;
1278	spin_unlock(&inode->i_lock);
1279	if (shmem_falloc)
1280	goto redirty;
1281	}
1282	clear_highpage(page);
1283	flush_dcache_page(page);
1284	SetPageUptodate(page);
1285	}
1286
1287	swap = get_swap_page();
1288	if (!swap.val)
1289	goto redirty;
1290
1291	if (mem_cgroup_try_charge_swap(page, swap))
1292	goto free_swap;
1293
1294	/*
1295	* Add inode to shmem_unuse()'s list of swapped-out inodes,
1296	* if it's not already there. Do it now before the page is
1297	* moved to swap cache, when its pagelock no longer protects
1298	* the inode from eviction. But don't unlock the mutex until
1299	* we've incremented swapped, because shmem_unuse_inode() will
1300	* prune a !swapped inode from the swaplist under this mutex.
1301	*/
1302	mutex_lock(&shmem_swaplist_mutex);
1303	if (list_empty(&info->swaplist))
1304	list_add_tail(&info->swaplist, &shmem_swaplist);
1305
1306	if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1307	spin_lock_irq(&info->lock);
1308	shmem_recalc_inode(inode);
1309	info->swapped++;
1310	spin_unlock_irq(&info->lock);
1311
1312	swap_shmem_alloc(swap);
1313	shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1314
1315	mutex_unlock(&shmem_swaplist_mutex);
1316	BUG_ON(page_mapped(page));
1317	swap_writepage(page, wbc);
1318	return 0;
1319	}
1320
1321	mutex_unlock(&shmem_swaplist_mutex);
1322	free_swap:
1323	swapcache_free(swap);
1324	redirty:
1325	set_page_dirty(page);
1326	if (wbc->for_reclaim)
1327	return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1328	unlock_page(page);
1329	return 0;
1330	}
1331
1332	#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1333	static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1334	{
1335	char buffer[64];
1336
1337	if (!mpol || mpol->mode == MPOL_DEFAULT)
1338	return; /* show nothing */
1339
1340	mpol_to_str(buffer, sizeof(buffer), mpol);
1341
1342	seq_printf(seq, ",mpol=%s", buffer);
1343	}
1344
1345	static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1346	{
1347	struct mempolicy *mpol = NULL;
1348	if (sbinfo->mpol) {
1349	spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1350	mpol = sbinfo->mpol;
1351	mpol_get(mpol);
1352	spin_unlock(&sbinfo->stat_lock);
1353	}
1354	return mpol;
1355	}
1356	#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1357	static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1358	{
1359	}
1360	static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1361	{
1362	return NULL;
1363	}
1364	#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1365	#ifndef CONFIG_NUMA
1366	#define vm_policy vm_private_data
1367	#endif
1368
1369	static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1370	struct shmem_inode_info *info, pgoff_t index)
1371	{
1372	/* Create a pseudo vma that just contains the policy */
1373	vma->vm_start = 0;
1374	/* Bias interleave by inode number to distribute better across nodes */
1375	vma->vm_pgoff = index + info->vfs_inode.i_ino;
1376	vma->vm_ops = NULL;
1377	vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1378	}
1379
1380	static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1381	{
1382	/* Drop reference taken by mpol_shared_policy_lookup() */
1383	mpol_cond_put(vma->vm_policy);
1384	}
1385
1386	static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1387	struct shmem_inode_info *info, pgoff_t index)
1388	{
1389	struct vm_area_struct pvma;
1390	struct page *page;
1391
1392	shmem_pseudo_vma_init(&pvma, info, index);
1393	#ifdef CONFIG_AMLOGIC_CMA
1394	page = swapin_readahead(swap, gfp | __GFP_BDEV, &pvma, 0);
1395	#else
1396	page = swapin_readahead(swap, gfp, &pvma, 0);
1397	#endif
1398	shmem_pseudo_vma_destroy(&pvma);
1399
1400	return page;
1401	}
1402
1403	static struct page *shmem_alloc_hugepage(gfp_t gfp,
1404	struct shmem_inode_info *info, pgoff_t index)
1405	{
1406	struct vm_area_struct pvma;
1407	struct inode *inode = &info->vfs_inode;
1408	struct address_space *mapping = inode->i_mapping;
1409	pgoff_t idx, hindex;
1410	void __rcu **results;
1411	struct page *page;
1412
1413	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1414	return NULL;
1415
1416	hindex = round_down(index, HPAGE_PMD_NR);
1417	rcu_read_lock();
1418	if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1419	hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1420	rcu_read_unlock();
1421	return NULL;
1422	}
1423	rcu_read_unlock();
1424
1425	shmem_pseudo_vma_init(&pvma, info, hindex);
1426	page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1427	HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1428	shmem_pseudo_vma_destroy(&pvma);
1429	if (page)
1430	prep_transhuge_page(page);
1431	return page;
1432	}
1433
1434	static struct page *shmem_alloc_page(gfp_t gfp,
1435	struct shmem_inode_info *info, pgoff_t index)
1436	{
1437	struct vm_area_struct pvma;
1438	struct page *page;
1439
1440	shmem_pseudo_vma_init(&pvma, info, index);
1441	#ifdef CONFIG_AMLOGIC_CMA
1442	page = alloc_page_vma(gfp | __GFP_BDEV, &pvma, 0);
1443	#else
1444	page = alloc_page_vma(gfp, &pvma, 0);
1445	#endif
1446	shmem_pseudo_vma_destroy(&pvma);
1447
1448	return page;
1449	}
1450
1451	static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1452	struct inode *inode,
1453	pgoff_t index, bool huge)
1454	{
1455	struct shmem_inode_info *info = SHMEM_I(inode);
1456	struct page *page;
1457	int nr;
1458	int err = -ENOSPC;
1459
1460	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1461	huge = false;
1462	nr = huge ? HPAGE_PMD_NR : 1;
1463
1464	if (!shmem_inode_acct_block(inode, nr))
1465	goto failed;
1466
1467	if (huge)
1468	page = shmem_alloc_hugepage(gfp, info, index);
1469	else
1470	page = shmem_alloc_page(gfp, info, index);
1471	if (page) {
1472	__SetPageLocked(page);
1473	__SetPageSwapBacked(page);
1474	return page;
1475	}
1476
1477	err = -ENOMEM;
1478	shmem_inode_unacct_blocks(inode, nr);
1479	failed:
1480	return ERR_PTR(err);
1481	}
1482
1483	/*
1484	* When a page is moved from swapcache to shmem filecache (either by the
1485	* usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1486	* shmem_unuse_inode()), it may have been read in earlier from swap, in
1487	* ignorance of the mapping it belongs to. If that mapping has special
1488	* constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1489	* we may need to copy to a suitable page before moving to filecache.
1490	*
1491	* In a future release, this may well be extended to respect cpuset and
1492	* NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1493	* but for now it is a simple matter of zone.
1494	*/
1495	static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1496	{
1497	return page_zonenum(page) > gfp_zone(gfp);
1498	}
1499
1500	static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1501	struct shmem_inode_info *info, pgoff_t index)
1502	{
1503	struct page *oldpage, *newpage;
1504	struct address_space *swap_mapping;
1505	swp_entry_t entry;
1506	pgoff_t swap_index;
1507	int error;
1508
1509	oldpage = *pagep;
1510	entry.val = page_private(oldpage);
1511	swap_index = swp_offset(entry);
1512	swap_mapping = page_mapping(oldpage);
1513
1514	/*
1515	* We have arrived here because our zones are constrained, so don't
1516	* limit chance of success by further cpuset and node constraints.
1517	*/
1518	gfp &= ~GFP_CONSTRAINT_MASK;
1519	newpage = shmem_alloc_page(gfp, info, index);
1520	if (!newpage)
1521	return -ENOMEM;
1522
1523	get_page(newpage);
1524	copy_highpage(newpage, oldpage);
1525	flush_dcache_page(newpage);
1526
1527	__SetPageLocked(newpage);
1528	__SetPageSwapBacked(newpage);
1529	SetPageUptodate(newpage);
1530	set_page_private(newpage, entry.val);
1531	SetPageSwapCache(newpage);
1532
1533	/*
1534	* Our caller will very soon move newpage out of swapcache, but it's
1535	* a nice clean interface for us to replace oldpage by newpage there.
1536	*/
1537	spin_lock_irq(&swap_mapping->tree_lock);
1538	error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1539	newpage);
1540	if (!error) {
1541	__inc_node_page_state(newpage, NR_FILE_PAGES);
1542	__dec_node_page_state(oldpage, NR_FILE_PAGES);
1543	}
1544	spin_unlock_irq(&swap_mapping->tree_lock);
1545
1546	if (unlikely(error)) {
1547	/*
1548	* Is this possible? I think not, now that our callers check
1549	* both PageSwapCache and page_private after getting page lock;
1550	* but be defensive. Reverse old to newpage for clear and free.
1551	*/
1552	oldpage = newpage;
1553	} else {
1554	mem_cgroup_migrate(oldpage, newpage);
1555	lru_cache_add_anon(newpage);
1556	*pagep = newpage;
1557	}
1558
1559	ClearPageSwapCache(oldpage);
1560	set_page_private(oldpage, 0);
1561
1562	unlock_page(oldpage);
1563	put_page(oldpage);
1564	put_page(oldpage);
1565	return error;
1566	}
1567
1568	/*
1569	* shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1570	*
1571	* If we allocate a new one we do not mark it dirty. That's up to the
1572	* vm. If we swap it in we mark it dirty since we also free the swap
1573	* entry since a page cannot live in both the swap and page cache.
1574	*
1575	* fault_mm and fault_type are only supplied by shmem_fault:
1576	* otherwise they are NULL.
1577	*/
1578	static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1579	struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1580	struct mm_struct *fault_mm, int *fault_type)
1581	{
1582	struct address_space *mapping = inode->i_mapping;
1583	struct shmem_inode_info *info = SHMEM_I(inode);
1584	struct shmem_sb_info *sbinfo;
1585	struct mm_struct *charge_mm;
1586	struct mem_cgroup *memcg;
1587	struct page *page;
1588	swp_entry_t swap;
1589	enum sgp_type sgp_huge = sgp;
1590	pgoff_t hindex = index;
1591	int error;
1592	int once = 0;
1593	int alloced = 0;
1594
1595	if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1596	return -EFBIG;
1597	if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1598	sgp = SGP_CACHE;
1599	repeat:
1600	swap.val = 0;
1601	page = find_lock_entry(mapping, index);
1602	if (radix_tree_exceptional_entry(page)) {
1603	swap = radix_to_swp_entry(page);
1604	page = NULL;
1605	}
1606
1607	if (sgp <= SGP_CACHE &&
1608	((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1609	error = -EINVAL;
1610	goto unlock;
1611	}
1612
1613	if (page && sgp == SGP_WRITE)
1614	mark_page_accessed(page);
1615
1616	/* fallocated page? */
1617	if (page && !PageUptodate(page)) {
1618	if (sgp != SGP_READ)
1619	goto clear;
1620	unlock_page(page);
1621	put_page(page);
1622	page = NULL;
1623	}
1624	if (page || (sgp == SGP_READ && !swap.val)) {
1625	*pagep = page;
1626	return 0;
1627	}
1628
1629	/*
1630	* Fast cache lookup did not find it:
1631	* bring it back from swap or allocate.
1632	*/
1633	sbinfo = SHMEM_SB(inode->i_sb);
1634	charge_mm = fault_mm ? : current->mm;
1635
1636	if (swap.val) {
1637	/* Look it up and read it in.. */
1638	page = lookup_swap_cache(swap);
1639	if (!page) {
1640	/* Or update major stats only when swapin succeeds?? */
1641	if (fault_type) {
1642	*fault_type |= VM_FAULT_MAJOR;
1643	count_vm_event(PGMAJFAULT);
1644	mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1645	}
1646	/* Here we actually start the io */
1647	page = shmem_swapin(swap, gfp, info, index);
1648	if (!page) {
1649	error = -ENOMEM;
1650	goto failed;
1651	}
1652	}
1653
1654	/* We have to do this with page locked to prevent races */
1655	lock_page(page);
1656	if (!PageSwapCache(page) || page_private(page) != swap.val ||
1657	!shmem_confirm_swap(mapping, index, swap)) {
1658	error = -EEXIST; /* try again */
1659	goto unlock;
1660	}
1661	if (!PageUptodate(page)) {
1662	error = -EIO;
1663	goto failed;
1664	}
1665	wait_on_page_writeback(page);
1666
1667	if (shmem_should_replace_page(page, gfp)) {
1668	error = shmem_replace_page(&page, gfp, info, index);
1669	if (error)
1670	goto failed;
1671	}
1672
1673	error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1674	false);
1675	if (!error) {
1676	error = shmem_add_to_page_cache(page, mapping, index,
1677	swp_to_radix_entry(swap));
1678	/*
1679	* We already confirmed swap under page lock, and make
1680	* no memory allocation here, so usually no possibility
1681	* of error; but free_swap_and_cache() only trylocks a
1682	* page, so it is just possible that the entry has been
1683	* truncated or holepunched since swap was confirmed.
1684	* shmem_undo_range() will have done some of the
1685	* unaccounting, now delete_from_swap_cache() will do
1686	* the rest.
1687	* Reset swap.val? No, leave it so "failed" goes back to
1688	* "repeat": reading a hole and writing should succeed.
1689	*/
1690	if (error) {
1691	mem_cgroup_cancel_charge(page, memcg, false);
1692	delete_from_swap_cache(page);
1693	}
1694	}
1695	if (error)
1696	goto failed;
1697
1698	mem_cgroup_commit_charge(page, memcg, true, false);
1699
1700	spin_lock_irq(&info->lock);
1701	info->swapped--;
1702	shmem_recalc_inode(inode);
1703	spin_unlock_irq(&info->lock);
1704
1705	if (sgp == SGP_WRITE)
1706	mark_page_accessed(page);
1707
1708	delete_from_swap_cache(page);
1709	set_page_dirty(page);
1710	swap_free(swap);
1711
1712	} else {
1713	/* shmem_symlink() */
1714	if (mapping->a_ops != &shmem_aops)
1715	goto alloc_nohuge;
1716	if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1717	goto alloc_nohuge;
1718	if (shmem_huge == SHMEM_HUGE_FORCE)
1719	goto alloc_huge;
1720	switch (sbinfo->huge) {
1721	loff_t i_size;
1722	pgoff_t off;
1723	case SHMEM_HUGE_NEVER:
1724	goto alloc_nohuge;
1725	case SHMEM_HUGE_WITHIN_SIZE:
1726	off = round_up(index, HPAGE_PMD_NR);
1727	i_size = round_up(i_size_read(inode), PAGE_SIZE);
1728	if (i_size >= HPAGE_PMD_SIZE &&
1729	i_size >> PAGE_SHIFT >= off)
1730	goto alloc_huge;
1731	/* fallthrough */
1732	case SHMEM_HUGE_ADVISE:
1733	if (sgp_huge == SGP_HUGE)
1734	goto alloc_huge;
1735	/* TODO: implement fadvise() hints */
1736	goto alloc_nohuge;
1737	}
1738
1739	alloc_huge:
1740	page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1741	if (IS_ERR(page)) {
1742	alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode,
1743	index, false);
1744	}
1745	if (IS_ERR(page)) {
1746	int retry = 5;
1747	error = PTR_ERR(page);
1748	page = NULL;
1749	if (error != -ENOSPC)
1750	goto failed;
1751	/*
1752	* Try to reclaim some spece by splitting a huge page
1753	* beyond i_size on the filesystem.
1754	*/
1755	while (retry--) {
1756	int ret;
1757	ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1758	if (ret == SHRINK_STOP)
1759	break;
1760	if (ret)
1761	goto alloc_nohuge;
1762	}
1763	goto failed;
1764	}
1765
1766	if (PageTransHuge(page))
1767	hindex = round_down(index, HPAGE_PMD_NR);
1768	else
1769	hindex = index;
1770
1771	if (sgp == SGP_WRITE)
1772	__SetPageReferenced(page);
1773
1774	error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1775	PageTransHuge(page));
1776	if (error)
1777	goto unacct;
1778	error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1779	compound_order(page));
1780	if (!error) {
1781	error = shmem_add_to_page_cache(page, mapping, hindex,
1782	NULL);
1783	radix_tree_preload_end();
1784	}
1785	if (error) {
1786	mem_cgroup_cancel_charge(page, memcg,
1787	PageTransHuge(page));
1788	goto unacct;
1789	}
1790	mem_cgroup_commit_charge(page, memcg, false,
1791	PageTransHuge(page));
1792	lru_cache_add_anon(page);
1793
1794	spin_lock_irq(&info->lock);
1795	info->alloced += 1 << compound_order(page);
1796	inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1797	shmem_recalc_inode(inode);
1798	spin_unlock_irq(&info->lock);
1799	alloced = true;
1800
1801	if (PageTransHuge(page) &&
1802	DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1803	hindex + HPAGE_PMD_NR - 1) {
1804	/*
1805	* Part of the huge page is beyond i_size: subject
1806	* to shrink under memory pressure.
1807	*/
1808	spin_lock(&sbinfo->shrinklist_lock);
1809	/*
1810	* _careful to defend against unlocked access to
1811	* ->shrink_list in shmem_unused_huge_shrink()
1812	*/
1813	if (list_empty_careful(&info->shrinklist)) {
1814	list_add_tail(&info->shrinklist,
1815	&sbinfo->shrinklist);
1816	sbinfo->shrinklist_len++;
1817	}
1818	spin_unlock(&sbinfo->shrinklist_lock);
1819	}
1820
1821	/*
1822	* Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1823	*/
1824	if (sgp == SGP_FALLOC)
1825	sgp = SGP_WRITE;
1826	clear:
1827	/*
1828	* Let SGP_WRITE caller clear ends if write does not fill page;
1829	* but SGP_FALLOC on a page fallocated earlier must initialize
1830	* it now, lest undo on failure cancel our earlier guarantee.
1831	*/
1832	if (sgp != SGP_WRITE && !PageUptodate(page)) {
1833	struct page *head = compound_head(page);
1834	int i;
1835
1836	for (i = 0; i < (1 << compound_order(head)); i++) {
1837	clear_highpage(head + i);
1838	flush_dcache_page(head + i);
1839	}
1840	SetPageUptodate(head);
1841	}
1842	}
1843
1844	/* Perhaps the file has been truncated since we checked */
1845	if (sgp <= SGP_CACHE &&
1846	((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1847	if (alloced) {
1848	ClearPageDirty(page);
1849	delete_from_page_cache(page);
1850	spin_lock_irq(&info->lock);
1851	shmem_recalc_inode(inode);
1852	spin_unlock_irq(&info->lock);
1853	}
1854	error = -EINVAL;
1855	goto unlock;
1856	}
1857	*pagep = page + index - hindex;
1858	return 0;
1859
1860	/*
1861	* Error recovery.
1862	*/
1863	unacct:
1864	shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
1865
1866	if (PageTransHuge(page)) {
1867	unlock_page(page);
1868	put_page(page);
1869	goto alloc_nohuge;
1870	}
1871	failed:
1872	if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1873	error = -EEXIST;
1874	unlock:
1875	if (page) {
1876	unlock_page(page);
1877	put_page(page);
1878	}
1879	if (error == -ENOSPC && !once++) {
1880	spin_lock_irq(&info->lock);
1881	shmem_recalc_inode(inode);
1882	spin_unlock_irq(&info->lock);
1883	goto repeat;
1884	}
1885	if (error == -EEXIST) /* from above or from radix_tree_insert */
1886	goto repeat;
1887	return error;
1888	}
1889
1890	/*
1891	* This is like autoremove_wake_function, but it removes the wait queue
1892	* entry unconditionally - even if something else had already woken the
1893	* target.
1894	*/
1895	static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
1896	{
1897	int ret = default_wake_function(wait, mode, sync, key);
1898	list_del_init(&wait->task_list);
1899	return ret;
1900	}
1901
1902	static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1903	{
1904	struct inode *inode = file_inode(vma->vm_file);
1905	gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1906	enum sgp_type sgp;
1907	int error;
1908	int ret = VM_FAULT_LOCKED;
1909
1910	/*
1911	* Trinity finds that probing a hole which tmpfs is punching can
1912	* prevent the hole-punch from ever completing: which in turn
1913	* locks writers out with its hold on i_mutex. So refrain from
1914	* faulting pages into the hole while it's being punched. Although
1915	* shmem_undo_range() does remove the additions, it may be unable to
1916	* keep up, as each new page needs its own unmap_mapping_range() call,
1917	* and the i_mmap tree grows ever slower to scan if new vmas are added.
1918	*
1919	* It does not matter if we sometimes reach this check just before the
1920	* hole-punch begins, so that one fault then races with the punch:
1921	* we just need to make racing faults a rare case.
1922	*
1923	* The implementation below would be much simpler if we just used a
1924	* standard mutex or completion: but we cannot take i_mutex in fault,
1925	* and bloating every shmem inode for this unlikely case would be sad.
1926	*/
1927	if (unlikely(inode->i_private)) {
1928	struct shmem_falloc *shmem_falloc;
1929
1930	spin_lock(&inode->i_lock);
1931	shmem_falloc = inode->i_private;
1932	if (shmem_falloc &&
1933	shmem_falloc->waitq &&
1934	vmf->pgoff >= shmem_falloc->start &&
1935	vmf->pgoff < shmem_falloc->next) {
1936	wait_queue_head_t *shmem_falloc_waitq;
1937	DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1938
1939	ret = VM_FAULT_NOPAGE;
1940	if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1941	!(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1942	/* It's polite to up mmap_sem if we can */
1943	up_read(&vma->vm_mm->mmap_sem);
1944	ret = VM_FAULT_RETRY;
1945	}
1946
1947	shmem_falloc_waitq = shmem_falloc->waitq;
1948	prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1949	TASK_UNINTERRUPTIBLE);
1950	spin_unlock(&inode->i_lock);
1951	schedule();
1952
1953	/*
1954	* shmem_falloc_waitq points into the shmem_fallocate()
1955	* stack of the hole-punching task: shmem_falloc_waitq
1956	* is usually invalid by the time we reach here, but
1957	* finish_wait() does not dereference it in that case;
1958	* though i_lock needed lest racing with wake_up_all().
1959	*/
1960	spin_lock(&inode->i_lock);
1961	finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1962	spin_unlock(&inode->i_lock);
1963	return ret;
1964	}
1965	spin_unlock(&inode->i_lock);
1966	}
1967
1968	sgp = SGP_CACHE;
1969	if (vma->vm_flags & VM_HUGEPAGE)
1970	sgp = SGP_HUGE;
1971	else if (vma->vm_flags & VM_NOHUGEPAGE)
1972	sgp = SGP_NOHUGE;
1973
1974	error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1975	gfp, vma->vm_mm, &ret);
1976	if (error)
1977	return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1978	return ret;
1979	}
1980
1981	unsigned long shmem_get_unmapped_area(struct file *file,
1982	unsigned long uaddr, unsigned long len,
1983	unsigned long pgoff, unsigned long flags)
1984	{
1985	unsigned long (*get_area)(struct file *,
1986	unsigned long, unsigned long, unsigned long, unsigned long);
1987	unsigned long addr;
1988	unsigned long offset;
1989	unsigned long inflated_len;
1990	unsigned long inflated_addr;
1991	unsigned long inflated_offset;
1992
1993	if (len > TASK_SIZE)
1994	return -ENOMEM;
1995
1996	get_area = current->mm->get_unmapped_area;
1997	addr = get_area(file, uaddr, len, pgoff, flags);
1998
1999	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2000	return addr;
2001	if (IS_ERR_VALUE(addr))
2002	return addr;
2003	if (addr & ~PAGE_MASK)
2004	return addr;
2005	if (addr > TASK_SIZE - len)
2006	return addr;
2007
2008	if (shmem_huge == SHMEM_HUGE_DENY)
2009	return addr;
2010	if (len < HPAGE_PMD_SIZE)
2011	return addr;
2012	if (flags & MAP_FIXED)
2013	return addr;
2014	/*
2015	* Our priority is to support MAP_SHARED mapped hugely;
2016	* and support MAP_PRIVATE mapped hugely too, until it is COWed.
2017	* But if caller specified an address hint, respect that as before.
2018	*/
2019	if (uaddr)
2020	return addr;
2021
2022	if (shmem_huge != SHMEM_HUGE_FORCE) {
2023	struct super_block *sb;
2024
2025	if (file) {
2026	VM_BUG_ON(file->f_op != &shmem_file_operations);
2027	sb = file_inode(file)->i_sb;
2028	} else {
2029	/*
2030	* Called directly from mm/mmap.c, or drivers/char/mem.c
2031	* for "/dev/zero", to create a shared anonymous object.
2032	*/
2033	if (IS_ERR(shm_mnt))
2034	return addr;
2035	sb = shm_mnt->mnt_sb;
2036	}
2037	if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2038	return addr;
2039	}
2040
2041	offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2042	if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2043	return addr;
2044	if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2045	return addr;
2046
2047	inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2048	if (inflated_len > TASK_SIZE)
2049	return addr;
2050	if (inflated_len < len)
2051	return addr;
2052
2053	inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2054	if (IS_ERR_VALUE(inflated_addr))
2055	return addr;
2056	if (inflated_addr & ~PAGE_MASK)
2057	return addr;
2058
2059	inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2060	inflated_addr += offset - inflated_offset;
2061	if (inflated_offset > offset)
2062	inflated_addr += HPAGE_PMD_SIZE;
2063
2064	if (inflated_addr > TASK_SIZE - len)
2065	return addr;
2066	return inflated_addr;
2067	}
2068
2069	#ifdef CONFIG_NUMA
2070	static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2071	{
2072	struct inode *inode = file_inode(vma->vm_file);
2073	return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2074	}
2075
2076	static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2077	unsigned long addr)
2078	{
2079	struct inode *inode = file_inode(vma->vm_file);
2080	pgoff_t index;
2081
2082	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2083	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2084	}
2085	#endif
2086
2087	int shmem_lock(struct file *file, int lock, struct user_struct *user)
2088	{
2089	struct inode *inode = file_inode(file);
2090	struct shmem_inode_info *info = SHMEM_I(inode);
2091	int retval = -ENOMEM;
2092
2093	spin_lock_irq(&info->lock);
2094	if (lock && !(info->flags & VM_LOCKED)) {
2095	if (!user_shm_lock(inode->i_size, user))
2096	goto out_nomem;
2097	info->flags |= VM_LOCKED;
2098	mapping_set_unevictable(file->f_mapping);
2099	}
2100	if (!lock && (info->flags & VM_LOCKED) && user) {
2101	user_shm_unlock(inode->i_size, user);
2102	info->flags &= ~VM_LOCKED;
2103	mapping_clear_unevictable(file->f_mapping);
2104	}
2105	retval = 0;
2106
2107	out_nomem:
2108	spin_unlock_irq(&info->lock);
2109	return retval;
2110	}
2111
2112	static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2113	{
2114	file_accessed(file);
2115	vma->vm_ops = &shmem_vm_ops;
2116	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2117	((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2118	(vma->vm_end & HPAGE_PMD_MASK)) {
2119	khugepaged_enter(vma, vma->vm_flags);
2120	}
2121	return 0;
2122	}
2123
2124	static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2125	umode_t mode, dev_t dev, unsigned long flags)
2126	{
2127	struct inode *inode;
2128	struct shmem_inode_info *info;
2129	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2130
2131	if (shmem_reserve_inode(sb))
2132	return NULL;
2133
2134	inode = new_inode(sb);
2135	if (inode) {
2136	inode->i_ino = get_next_ino();
2137	inode_init_owner(inode, dir, mode);
2138	inode->i_blocks = 0;
2139	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2140	inode->i_generation = get_seconds();
2141	info = SHMEM_I(inode);
2142	memset(info, 0, (char *)inode - (char *)info);
2143	spin_lock_init(&info->lock);
2144	info->seals = F_SEAL_SEAL;
2145	info->flags = flags & VM_NORESERVE;
2146	INIT_LIST_HEAD(&info->shrinklist);
2147	INIT_LIST_HEAD(&info->swaplist);
2148	simple_xattrs_init(&info->xattrs);
2149	cache_no_acl(inode);
2150
2151	switch (mode & S_IFMT) {
2152	default:
2153	inode->i_op = &shmem_special_inode_operations;
2154	init_special_inode(inode, mode, dev);
2155	break;
2156	case S_IFREG:
2157	inode->i_mapping->a_ops = &shmem_aops;
2158	inode->i_op = &shmem_inode_operations;
2159	inode->i_fop = &shmem_file_operations;
2160	mpol_shared_policy_init(&info->policy,
2161	shmem_get_sbmpol(sbinfo));
2162	break;
2163	case S_IFDIR:
2164	inc_nlink(inode);
2165	/* Some things misbehave if size == 0 on a directory */
2166	inode->i_size = 2 * BOGO_DIRENT_SIZE;
2167	inode->i_op = &shmem_dir_inode_operations;
2168	inode->i_fop = &simple_dir_operations;
2169	break;
2170	case S_IFLNK:
2171	/*
2172	* Must not load anything in the rbtree,
2173	* mpol_free_shared_policy will not be called.
2174	*/
2175	mpol_shared_policy_init(&info->policy, NULL);
2176	break;
2177	}
2178
2179	lockdep_annotate_inode_mutex_key(inode);
2180	} else
2181	shmem_free_inode(sb);
2182	return inode;
2183	}
2184
2185	bool shmem_mapping(struct address_space *mapping)
2186	{
2187	if (!mapping->host)
2188	return false;
2189
2190	return mapping->host->i_sb->s_op == &shmem_ops;
2191	}
2192
2193	#ifdef CONFIG_TMPFS
2194	static const struct inode_operations shmem_symlink_inode_operations;
2195	static const struct inode_operations shmem_short_symlink_operations;
2196
2197	#ifdef CONFIG_TMPFS_XATTR
2198	static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2199	#else
2200	#define shmem_initxattrs NULL
2201	#endif
2202
2203	static int
2204	shmem_write_begin(struct file *file, struct address_space *mapping,
2205	loff_t pos, unsigned len, unsigned flags,
2206	struct page **pagep, void **fsdata)
2207	{
2208	struct inode *inode = mapping->host;
2209	struct shmem_inode_info *info = SHMEM_I(inode);
2210	pgoff_t index = pos >> PAGE_SHIFT;
2211
2212	/* i_mutex is held by caller */
2213	if (unlikely(info->seals)) {
2214	if (info->seals & F_SEAL_WRITE)
2215	return -EPERM;
2216	if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2217	return -EPERM;
2218	}
2219
2220	return shmem_getpage(inode, index, pagep, SGP_WRITE);
2221	}
2222
2223	static int
2224	shmem_write_end(struct file *file, struct address_space *mapping,
2225	loff_t pos, unsigned len, unsigned copied,
2226	struct page *page, void *fsdata)
2227	{
2228	struct inode *inode = mapping->host;
2229
2230	if (pos + copied > inode->i_size)
2231	i_size_write(inode, pos + copied);
2232
2233	if (!PageUptodate(page)) {
2234	struct page *head = compound_head(page);
2235	if (PageTransCompound(page)) {
2236	int i;
2237
2238	for (i = 0; i < HPAGE_PMD_NR; i++) {
2239	if (head + i == page)
2240	continue;
2241	clear_highpage(head + i);
2242	flush_dcache_page(head + i);
2243	}
2244	}
2245	if (copied < PAGE_SIZE) {
2246	unsigned from = pos & (PAGE_SIZE - 1);
2247	zero_user_segments(page, 0, from,
2248	from + copied, PAGE_SIZE);
2249	}
2250	SetPageUptodate(head);
2251	}
2252	set_page_dirty(page);
2253	unlock_page(page);
2254	put_page(page);
2255
2256	return copied;
2257	}
2258
2259	static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2260	{
2261	struct file *file = iocb->ki_filp;
2262	struct inode *inode = file_inode(file);
2263	struct address_space *mapping = inode->i_mapping;
2264	pgoff_t index;
2265	unsigned long offset;
2266	enum sgp_type sgp = SGP_READ;
2267	int error = 0;
2268	ssize_t retval = 0;
2269	loff_t *ppos = &iocb->ki_pos;
2270
2271	/*
2272	* Might this read be for a stacking filesystem? Then when reading
2273	* holes of a sparse file, we actually need to allocate those pages,
2274	* and even mark them dirty, so it cannot exceed the max_blocks limit.
2275	*/
2276	if (!iter_is_iovec(to))
2277	sgp = SGP_CACHE;
2278
2279	index = *ppos >> PAGE_SHIFT;
2280	offset = *ppos & ~PAGE_MASK;
2281
2282	for (;;) {
2283	struct page *page = NULL;
2284	pgoff_t end_index;
2285	unsigned long nr, ret;
2286	loff_t i_size = i_size_read(inode);
2287
2288	end_index = i_size >> PAGE_SHIFT;
2289	if (index > end_index)
2290	break;
2291	if (index == end_index) {
2292	nr = i_size & ~PAGE_MASK;
2293	if (nr <= offset)
2294	break;
2295	}
2296
2297	error = shmem_getpage(inode, index, &page, sgp);
2298	if (error) {
2299	if (error == -EINVAL)
2300	error = 0;
2301	break;
2302	}
2303	if (page) {
2304	if (sgp == SGP_CACHE)
2305	set_page_dirty(page);
2306	unlock_page(page);
2307	}
2308
2309	/*
2310	* We must evaluate after, since reads (unlike writes)
2311	* are called without i_mutex protection against truncate
2312	*/
2313	nr = PAGE_SIZE;
2314	i_size = i_size_read(inode);
2315	end_index = i_size >> PAGE_SHIFT;
2316	if (index == end_index) {
2317	nr = i_size & ~PAGE_MASK;
2318	if (nr <= offset) {
2319	if (page)
2320	put_page(page);
2321	break;
2322	}
2323	}
2324	nr -= offset;
2325
2326	if (page) {
2327	/*
2328	* If users can be writing to this page using arbitrary
2329	* virtual addresses, take care about potential aliasing
2330	* before reading the page on the kernel side.
2331	*/
2332	if (mapping_writably_mapped(mapping))
2333	flush_dcache_page(page);
2334	/*
2335	* Mark the page accessed if we read the beginning.
2336	*/
2337	if (!offset)
2338	mark_page_accessed(page);
2339	} else {
2340	page = ZERO_PAGE(0);
2341	get_page(page);
2342	}
2343
2344	/*
2345	* Ok, we have the page, and it's up-to-date, so
2346	* now we can copy it to user space...
2347	*/
2348	ret = copy_page_to_iter(page, offset, nr, to);
2349	retval += ret;
2350	offset += ret;
2351	index += offset >> PAGE_SHIFT;
2352	offset &= ~PAGE_MASK;
2353
2354	put_page(page);
2355	if (!iov_iter_count(to))
2356	break;
2357	if (ret < nr) {
2358	error = -EFAULT;
2359	break;
2360	}
2361	cond_resched();
2362	}
2363
2364	*ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2365	file_accessed(file);
2366	return retval ? retval : error;
2367	}
2368
2369	/*
2370	* llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2371	*/
2372	static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2373	pgoff_t index, pgoff_t end, int whence)
2374	{
2375	struct page *page;
2376	struct pagevec pvec;
2377	pgoff_t indices[PAGEVEC_SIZE];
2378	bool done = false;
2379	int i;
2380
2381	pagevec_init(&pvec, 0);
2382	pvec.nr = 1; /* start small: we may be there already */
2383	while (!done) {
2384	pvec.nr = find_get_entries(mapping, index,
2385	pvec.nr, pvec.pages, indices);
2386	if (!pvec.nr) {
2387	if (whence == SEEK_DATA)
2388	index = end;
2389	break;
2390	}
2391	for (i = 0; i < pvec.nr; i++, index++) {
2392	if (index < indices[i]) {
2393	if (whence == SEEK_HOLE) {
2394	done = true;
2395	break;
2396	}
2397	index = indices[i];
2398	}
2399	page = pvec.pages[i];
2400	if (page && !radix_tree_exceptional_entry(page)) {
2401	if (!PageUptodate(page))
2402	page = NULL;
2403	}
2404	if (index >= end ||
2405	(page && whence == SEEK_DATA) ||
2406	(!page && whence == SEEK_HOLE)) {
2407	done = true;
2408	break;
2409	}
2410	}
2411	pagevec_remove_exceptionals(&pvec);
2412	pagevec_release(&pvec);
2413	pvec.nr = PAGEVEC_SIZE;
2414	cond_resched();
2415	}
2416	return index;
2417	}
2418
2419	static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2420	{
2421	struct address_space *mapping = file->f_mapping;
2422	struct inode *inode = mapping->host;
2423	pgoff_t start, end;
2424	loff_t new_offset;
2425
2426	if (whence != SEEK_DATA && whence != SEEK_HOLE)
2427	return generic_file_llseek_size(file, offset, whence,
2428	MAX_LFS_FILESIZE, i_size_read(inode));
2429	inode_lock(inode);
2430	/* We're holding i_mutex so we can access i_size directly */
2431
2432	if (offset < 0 || offset >= inode->i_size)
2433	offset = -ENXIO;
2434	else {
2435	start = offset >> PAGE_SHIFT;
2436	end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2437	new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2438	new_offset <<= PAGE_SHIFT;
2439	if (new_offset > offset) {
2440	if (new_offset < inode->i_size)
2441	offset = new_offset;
2442	else if (whence == SEEK_DATA)
2443	offset = -ENXIO;
2444	else
2445	offset = inode->i_size;
2446	}
2447	}
2448
2449	if (offset >= 0)
2450	offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2451	inode_unlock(inode);
2452	return offset;
2453	}
2454
2455	/*
2456	* We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2457	* so reuse a tag which we firmly believe is never set or cleared on shmem.
2458	*/
2459	#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2460	#define LAST_SCAN 4 /* about 150ms max */
2461
2462	static void shmem_tag_pins(struct address_space *mapping)
2463	{
2464	struct radix_tree_iter iter;
2465	void **slot;
2466	pgoff_t start;
2467	struct page *page;
2468
2469	lru_add_drain();
2470	start = 0;
2471	rcu_read_lock();
2472
2473	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2474	page = radix_tree_deref_slot(slot);
2475	if (!page || radix_tree_exception(page)) {
2476	if (radix_tree_deref_retry(page)) {
2477	slot = radix_tree_iter_retry(&iter);
2478	continue;
2479	}
2480	} else if (page_count(page) - page_mapcount(page) > 1) {
2481	spin_lock_irq(&mapping->tree_lock);
2482	radix_tree_tag_set(&mapping->page_tree, iter.index,
2483	SHMEM_TAG_PINNED);
2484	spin_unlock_irq(&mapping->tree_lock);
2485	}
2486
2487	if (need_resched()) {
2488	cond_resched_rcu();
2489	slot = radix_tree_iter_next(&iter);
2490	}
2491	}
2492	rcu_read_unlock();
2493	}
2494
2495	/*
2496	* Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2497	* via get_user_pages(), drivers might have some pending I/O without any active
2498	* user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2499	* and see whether it has an elevated ref-count. If so, we tag them and wait for
2500	* them to be dropped.
2501	* The caller must guarantee that no new user will acquire writable references
2502	* to those pages to avoid races.
2503	*/
2504	static int shmem_wait_for_pins(struct address_space *mapping)
2505	{
2506	struct radix_tree_iter iter;
2507	void **slot;
2508	pgoff_t start;
2509	struct page *page;
2510	int error, scan;
2511
2512	shmem_tag_pins(mapping);
2513
2514	error = 0;
2515	for (scan = 0; scan <= LAST_SCAN; scan++) {
2516	if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2517	break;
2518
2519	if (!scan)
2520	lru_add_drain_all();
2521	else if (schedule_timeout_killable((HZ << scan) / 200))
2522	scan = LAST_SCAN;
2523
2524	start = 0;
2525	rcu_read_lock();
2526	radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2527	start, SHMEM_TAG_PINNED) {
2528
2529	page = radix_tree_deref_slot(slot);
2530	if (radix_tree_exception(page)) {
2531	if (radix_tree_deref_retry(page)) {
2532	slot = radix_tree_iter_retry(&iter);
2533	continue;
2534	}
2535
2536	page = NULL;
2537	}
2538
2539	if (page &&
2540	page_count(page) - page_mapcount(page) != 1) {
2541	if (scan < LAST_SCAN)
2542	goto continue_resched;
2543
2544	/*
2545	* On the last scan, we clean up all those tags
2546	* we inserted; but make a note that we still
2547	* found pages pinned.
2548	*/
2549	error = -EBUSY;
2550	}
2551
2552	spin_lock_irq(&mapping->tree_lock);
2553	radix_tree_tag_clear(&mapping->page_tree,
2554	iter.index, SHMEM_TAG_PINNED);
2555	spin_unlock_irq(&mapping->tree_lock);
2556	continue_resched:
2557	if (need_resched()) {
2558	cond_resched_rcu();
2559	slot = radix_tree_iter_next(&iter);
2560	}
2561	}
2562	rcu_read_unlock();
2563	}
2564
2565	return error;
2566	}
2567
2568	#define F_ALL_SEALS (F_SEAL_SEAL | \
2569	F_SEAL_SHRINK | \
2570	F_SEAL_GROW | \
2571	F_SEAL_WRITE)
2572
2573	int shmem_add_seals(struct file *file, unsigned int seals)
2574	{
2575	struct inode *inode = file_inode(file);
2576	struct shmem_inode_info *info = SHMEM_I(inode);
2577	int error;
2578
2579	/*
2580	* SEALING
2581	* Sealing allows multiple parties to share a shmem-file but restrict
2582	* access to a specific subset of file operations. Seals can only be
2583	* added, but never removed. This way, mutually untrusted parties can
2584	* share common memory regions with a well-defined policy. A malicious
2585	* peer can thus never perform unwanted operations on a shared object.
2586	*
2587	* Seals are only supported on special shmem-files and always affect
2588	* the whole underlying inode. Once a seal is set, it may prevent some
2589	* kinds of access to the file. Currently, the following seals are
2590	* defined:
2591	* SEAL_SEAL: Prevent further seals from being set on this file
2592	* SEAL_SHRINK: Prevent the file from shrinking
2593	* SEAL_GROW: Prevent the file from growing
2594	* SEAL_WRITE: Prevent write access to the file
2595	*
2596	* As we don't require any trust relationship between two parties, we
2597	* must prevent seals from being removed. Therefore, sealing a file
2598	* only adds a given set of seals to the file, it never touches
2599	* existing seals. Furthermore, the "setting seals"-operation can be
2600	* sealed itself, which basically prevents any further seal from being
2601	* added.
2602	*
2603	* Semantics of sealing are only defined on volatile files. Only
2604	* anonymous shmem files support sealing. More importantly, seals are
2605	* never written to disk. Therefore, there's no plan to support it on
2606	* other file types.
2607	*/
2608
2609	if (file->f_op != &shmem_file_operations)
2610	return -EINVAL;
2611	if (!(file->f_mode & FMODE_WRITE))
2612	return -EPERM;
2613	if (seals & ~(unsigned int)F_ALL_SEALS)
2614	return -EINVAL;
2615
2616	inode_lock(inode);
2617
2618	if (info->seals & F_SEAL_SEAL) {
2619	error = -EPERM;
2620	goto unlock;
2621	}
2622
2623	if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2624	error = mapping_deny_writable(file->f_mapping);
2625	if (error)
2626	goto unlock;
2627
2628	error = shmem_wait_for_pins(file->f_mapping);
2629	if (error) {
2630	mapping_allow_writable(file->f_mapping);
2631	goto unlock;
2632	}
2633	}
2634
2635	info->seals |= seals;
2636	error = 0;
2637
2638	unlock:
2639	inode_unlock(inode);
2640	return error;
2641	}
2642	EXPORT_SYMBOL_GPL(shmem_add_seals);
2643
2644	int shmem_get_seals(struct file *file)
2645	{
2646	if (file->f_op != &shmem_file_operations)
2647	return -EINVAL;
2648
2649	return SHMEM_I(file_inode(file))->seals;
2650	}
2651	EXPORT_SYMBOL_GPL(shmem_get_seals);
2652
2653	long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2654	{
2655	long error;
2656
2657	switch (cmd) {
2658	case F_ADD_SEALS:
2659	/* disallow upper 32bit */
2660	if (arg > UINT_MAX)
2661	return -EINVAL;
2662
2663	error = shmem_add_seals(file, arg);
2664	break;
2665	case F_GET_SEALS:
2666	error = shmem_get_seals(file);
2667	break;
2668	default:
2669	error = -EINVAL;
2670	break;
2671	}
2672
2673	return error;
2674	}
2675
2676	static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2677	loff_t len)
2678	{
2679	struct inode *inode = file_inode(file);
2680	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2681	struct shmem_inode_info *info = SHMEM_I(inode);
2682	struct shmem_falloc shmem_falloc;
2683	pgoff_t start, index, end;
2684	int error;
2685
2686	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2687	return -EOPNOTSUPP;
2688
2689	inode_lock(inode);
2690
2691	if (mode & FALLOC_FL_PUNCH_HOLE) {
2692	struct address_space *mapping = file->f_mapping;
2693	loff_t unmap_start = round_up(offset, PAGE_SIZE);
2694	loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2695	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2696
2697	/* protected by i_mutex */
2698	if (info->seals & F_SEAL_WRITE) {
2699	error = -EPERM;
2700	goto out;
2701	}
2702
2703	shmem_falloc.waitq = &shmem_falloc_waitq;
2704	shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2705	shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2706	spin_lock(&inode->i_lock);
2707	inode->i_private = &shmem_falloc;
2708	spin_unlock(&inode->i_lock);
2709
2710	if ((u64)unmap_end > (u64)unmap_start)
2711	unmap_mapping_range(mapping, unmap_start,
2712	1 + unmap_end - unmap_start, 0);
2713	shmem_truncate_range(inode, offset, offset + len - 1);
2714	/* No need to unmap again: hole-punching leaves COWed pages */
2715
2716	spin_lock(&inode->i_lock);
2717	inode->i_private = NULL;
2718	wake_up_all(&shmem_falloc_waitq);
2719	WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list));
2720	spin_unlock(&inode->i_lock);
2721	error = 0;
2722	goto out;
2723	}
2724
2725	/* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2726	error = inode_newsize_ok(inode, offset + len);
2727	if (error)
2728	goto out;
2729
2730	if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2731	error = -EPERM;
2732	goto out;
2733	}
2734
2735	start = offset >> PAGE_SHIFT;
2736	end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2737	/* Try to avoid a swapstorm if len is impossible to satisfy */
2738	if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2739	error = -ENOSPC;
2740	goto out;
2741	}
2742
2743	shmem_falloc.waitq = NULL;
2744	shmem_falloc.start = start;
2745	shmem_falloc.next = start;
2746	shmem_falloc.nr_falloced = 0;
2747	shmem_falloc.nr_unswapped = 0;
2748	spin_lock(&inode->i_lock);
2749	inode->i_private = &shmem_falloc;
2750	spin_unlock(&inode->i_lock);
2751
2752	for (index = start; index < end; index++) {
2753	struct page *page;
2754
2755	/*
2756	* Good, the fallocate(2) manpage permits EINTR: we may have
2757	* been interrupted because we are using up too much memory.
2758	*/
2759	if (signal_pending(current))
2760	error = -EINTR;
2761	else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2762	error = -ENOMEM;
2763	else
2764	error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2765	if (error) {
2766	/* Remove the !PageUptodate pages we added */
2767	if (index > start) {
2768	shmem_undo_range(inode,
2769	(loff_t)start << PAGE_SHIFT,
2770	((loff_t)index << PAGE_SHIFT) - 1, true);
2771	}
2772	goto undone;
2773	}
2774
2775	/*
2776	* Inform shmem_writepage() how far we have reached.
2777	* No need for lock or barrier: we have the page lock.
2778	*/
2779	shmem_falloc.next++;
2780	if (!PageUptodate(page))
2781	shmem_falloc.nr_falloced++;
2782
2783	/*
2784	* If !PageUptodate, leave it that way so that freeable pages
2785	* can be recognized if we need to rollback on error later.
2786	* But set_page_dirty so that memory pressure will swap rather
2787	* than free the pages we are allocating (and SGP_CACHE pages
2788	* might still be clean: we now need to mark those dirty too).
2789	*/
2790	set_page_dirty(page);
2791	unlock_page(page);
2792	put_page(page);
2793	cond_resched();
2794	}
2795
2796	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2797	i_size_write(inode, offset + len);
2798	inode->i_ctime = current_time(inode);
2799	undone:
2800	spin_lock(&inode->i_lock);
2801	inode->i_private = NULL;
2802	spin_unlock(&inode->i_lock);
2803	out:
2804	inode_unlock(inode);
2805	return error;
2806	}
2807
2808	static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2809	{
2810	struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2811
2812	buf->f_type = TMPFS_MAGIC;
2813	buf->f_bsize = PAGE_SIZE;
2814	buf->f_namelen = NAME_MAX;
2815	if (sbinfo->max_blocks) {
2816	buf->f_blocks = sbinfo->max_blocks;
2817	buf->f_bavail =
2818	buf->f_bfree = sbinfo->max_blocks -
2819	percpu_counter_sum(&sbinfo->used_blocks);
2820	}
2821	if (sbinfo->max_inodes) {
2822	buf->f_files = sbinfo->max_inodes;
2823	buf->f_ffree = sbinfo->free_inodes;
2824	}
2825	/* else leave those fields 0 like simple_statfs */
2826	return 0;
2827	}
2828
2829	/*
2830	* File creation. Allocate an inode, and we're done..
2831	*/
2832	static int
2833	shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2834	{
2835	struct inode *inode;
2836	int error = -ENOSPC;
2837
2838	inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2839	if (inode) {
2840	error = simple_acl_create(dir, inode);
2841	if (error)
2842	goto out_iput;
2843	error = security_inode_init_security(inode, dir,
2844	&dentry->d_name,
2845	shmem_initxattrs, NULL);
2846	if (error && error != -EOPNOTSUPP)
2847	goto out_iput;
2848
2849	error = 0;
2850	dir->i_size += BOGO_DIRENT_SIZE;
2851	dir->i_ctime = dir->i_mtime = current_time(dir);
2852	d_instantiate(dentry, inode);
2853	dget(dentry); /* Extra count - pin the dentry in core */
2854	}
2855	return error;
2856	out_iput:
2857	iput(inode);
2858	return error;
2859	}
2860
2861	static int
2862	shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2863	{
2864	struct inode *inode;
2865	int error = -ENOSPC;
2866
2867	inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2868	if (inode) {
2869	error = security_inode_init_security(inode, dir,
2870	NULL,
2871	shmem_initxattrs, NULL);
2872	if (error && error != -EOPNOTSUPP)
2873	goto out_iput;
2874	error = simple_acl_create(dir, inode);
2875	if (error)
2876	goto out_iput;
2877	d_tmpfile(dentry, inode);
2878	}
2879	return error;
2880	out_iput:
2881	iput(inode);
2882	return error;
2883	}
2884
2885	static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2886	{
2887	int error;
2888
2889	if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2890	return error;
2891	inc_nlink(dir);
2892	return 0;
2893	}
2894
2895	static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2896	bool excl)
2897	{
2898	return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2899	}
2900
2901	/*
2902	* Link a file..
2903	*/
2904	static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2905	{
2906	struct inode *inode = d_inode(old_dentry);
2907	int ret = 0;
2908
2909	/*
2910	* No ordinary (disk based) filesystem counts links as inodes;
2911	* but each new link needs a new dentry, pinning lowmem, and
2912	* tmpfs dentries cannot be pruned until they are unlinked.
2913	* But if an O_TMPFILE file is linked into the tmpfs, the
2914	* first link must skip that, to get the accounting right.
2915	*/
2916	if (inode->i_nlink) {
2917	ret = shmem_reserve_inode(inode->i_sb);
2918	if (ret)
2919	goto out;
2920	}
2921
2922	dir->i_size += BOGO_DIRENT_SIZE;
2923	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2924	inc_nlink(inode);
2925	ihold(inode); /* New dentry reference */
2926	dget(dentry); /* Extra pinning count for the created dentry */
2927	d_instantiate(dentry, inode);
2928	out:
2929	return ret;
2930	}
2931
2932	static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2933	{
2934	struct inode *inode = d_inode(dentry);
2935
2936	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2937	shmem_free_inode(inode->i_sb);
2938
2939	dir->i_size -= BOGO_DIRENT_SIZE;
2940	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2941	drop_nlink(inode);
2942	dput(dentry); /* Undo the count from "create" - this does all the work */
2943	return 0;
2944	}
2945
2946	static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2947	{
2948	if (!simple_empty(dentry))
2949	return -ENOTEMPTY;
2950
2951	drop_nlink(d_inode(dentry));
2952	drop_nlink(dir);
2953	return shmem_unlink(dir, dentry);
2954	}
2955
2956	static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2957	{
2958	bool old_is_dir = d_is_dir(old_dentry);
2959	bool new_is_dir = d_is_dir(new_dentry);
2960
2961	if (old_dir != new_dir && old_is_dir != new_is_dir) {
2962	if (old_is_dir) {
2963	drop_nlink(old_dir);
2964	inc_nlink(new_dir);
2965	} else {
2966	drop_nlink(new_dir);
2967	inc_nlink(old_dir);
2968	}
2969	}
2970	old_dir->i_ctime = old_dir->i_mtime =
2971	new_dir->i_ctime = new_dir->i_mtime =
2972	d_inode(old_dentry)->i_ctime =
2973	d_inode(new_dentry)->i_ctime = current_time(old_dir);
2974
2975	return 0;
2976	}
2977
2978	static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2979	{
2980	struct dentry *whiteout;
2981	int error;
2982
2983	whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2984	if (!whiteout)
2985	return -ENOMEM;
2986
2987	error = shmem_mknod(old_dir, whiteout,
2988	S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2989	dput(whiteout);
2990	if (error)
2991	return error;
2992
2993	/*
2994	* Cheat and hash the whiteout while the old dentry is still in
2995	* place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2996	*
2997	* d_lookup() will consistently find one of them at this point,
2998	* not sure which one, but that isn't even important.
2999	*/
3000	d_rehash(whiteout);
3001	return 0;
3002	}
3003
3004	/*
3005	* The VFS layer already does all the dentry stuff for rename,
3006	* we just have to decrement the usage count for the target if
3007	* it exists so that the VFS layer correctly free's it when it
3008	* gets overwritten.
3009	*/
3010	static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3011	{
3012	struct inode *inode = d_inode(old_dentry);
3013	int they_are_dirs = S_ISDIR(inode->i_mode);
3014
3015	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3016	return -EINVAL;
3017
3018	if (flags & RENAME_EXCHANGE)
3019	return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3020
3021	if (!simple_empty(new_dentry))
3022	return -ENOTEMPTY;
3023
3024	if (flags & RENAME_WHITEOUT) {
3025	int error;
3026
3027	error = shmem_whiteout(old_dir, old_dentry);
3028	if (error)
3029	return error;
3030	}
3031
3032	if (d_really_is_positive(new_dentry)) {
3033	(void) shmem_unlink(new_dir, new_dentry);
3034	if (they_are_dirs) {
3035	drop_nlink(d_inode(new_dentry));
3036	drop_nlink(old_dir);
3037	}
3038	} else if (they_are_dirs) {
3039	drop_nlink(old_dir);
3040	inc_nlink(new_dir);
3041	}
3042
3043	old_dir->i_size -= BOGO_DIRENT_SIZE;
3044	new_dir->i_size += BOGO_DIRENT_SIZE;
3045	old_dir->i_ctime = old_dir->i_mtime =
3046	new_dir->i_ctime = new_dir->i_mtime =
3047	inode->i_ctime = current_time(old_dir);
3048	return 0;
3049	}
3050
3051	static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3052	{
3053	int error;
3054	int len;
3055	struct inode *inode;
3056	struct page *page;
3057	struct shmem_inode_info *info;
3058
3059	len = strlen(symname) + 1;
3060	if (len > PAGE_SIZE)
3061	return -ENAMETOOLONG;
3062
3063	inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3064	if (!inode)
3065	return -ENOSPC;
3066
3067	error = security_inode_init_security(inode, dir, &dentry->d_name,
3068	shmem_initxattrs, NULL);
3069	if (error) {
3070	if (error != -EOPNOTSUPP) {
3071	iput(inode);
3072	return error;
3073	}
3074	error = 0;
3075	}
3076
3077	info = SHMEM_I(inode);
3078	inode->i_size = len-1;
3079	if (len <= SHORT_SYMLINK_LEN) {
3080	inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3081	if (!inode->i_link) {
3082	iput(inode);
3083	return -ENOMEM;
3084	}
3085	inode->i_op = &shmem_short_symlink_operations;
3086	} else {
3087	inode_nohighmem(inode);
3088	error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3089	if (error) {
3090	iput(inode);
3091	return error;
3092	}
3093	inode->i_mapping->a_ops = &shmem_aops;
3094	inode->i_op = &shmem_symlink_inode_operations;
3095	memcpy(page_address(page), symname, len);
3096	SetPageUptodate(page);
3097	set_page_dirty(page);
3098	unlock_page(page);
3099	put_page(page);
3100	}
3101	dir->i_size += BOGO_DIRENT_SIZE;
3102	dir->i_ctime = dir->i_mtime = current_time(dir);
3103	d_instantiate(dentry, inode);
3104	dget(dentry);
3105	return 0;
3106	}
3107
3108	static void shmem_put_link(void *arg)
3109	{
3110	mark_page_accessed(arg);
3111	put_page(arg);
3112	}
3113
3114	static const char *shmem_get_link(struct dentry *dentry,
3115	struct inode *inode,
3116	struct delayed_call *done)
3117	{
3118	struct page *page = NULL;
3119	int error;
3120	if (!dentry) {
3121	page = find_get_page(inode->i_mapping, 0);
3122	if (!page)
3123	return ERR_PTR(-ECHILD);
3124	if (!PageUptodate(page)) {
3125	put_page(page);
3126	return ERR_PTR(-ECHILD);
3127	}
3128	} else {
3129	error = shmem_getpage(inode, 0, &page, SGP_READ);
3130	if (error)
3131	return ERR_PTR(error);
3132	unlock_page(page);
3133	}
3134	set_delayed_call(done, shmem_put_link, page);
3135	return page_address(page);
3136	}
3137
3138	#ifdef CONFIG_TMPFS_XATTR
3139	/*
3140	* Superblocks without xattr inode operations may get some security.* xattr
3141	* support from the LSM "for free". As soon as we have any other xattrs
3142	* like ACLs, we also need to implement the security.* handlers at
3143	* filesystem level, though.
3144	*/
3145
3146	/*
3147	* Callback for security_inode_init_security() for acquiring xattrs.
3148	*/
3149	static int shmem_initxattrs(struct inode *inode,
3150	const struct xattr *xattr_array,
3151	void *fs_info)
3152	{
3153	struct shmem_inode_info *info = SHMEM_I(inode);
3154	const struct xattr *xattr;
3155	struct simple_xattr *new_xattr;
3156	size_t len;
3157
3158	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3159	new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3160	if (!new_xattr)
3161	return -ENOMEM;
3162
3163	len = strlen(xattr->name) + 1;
3164	new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3165	GFP_KERNEL);
3166	if (!new_xattr->name) {
3167	kfree(new_xattr);
3168	return -ENOMEM;
3169	}
3170
3171	memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3172	XATTR_SECURITY_PREFIX_LEN);
3173	memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3174	xattr->name, len);
3175
3176	simple_xattr_list_add(&info->xattrs, new_xattr);
3177	}
3178
3179	return 0;
3180	}
3181
3182	static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3183	struct dentry *unused, struct inode *inode,
3184	const char *name, void *buffer, size_t size)
3185	{
3186	struct shmem_inode_info *info = SHMEM_I(inode);
3187
3188	name = xattr_full_name(handler, name);
3189	return simple_xattr_get(&info->xattrs, name, buffer, size);
3190	}
3191
3192	static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3193	struct dentry *unused, struct inode *inode,
3194	const char *name, const void *value,
3195	size_t size, int flags)
3196	{
3197	struct shmem_inode_info *info = SHMEM_I(inode);
3198
3199	name = xattr_full_name(handler, name);
3200	return simple_xattr_set(&info->xattrs, name, value, size, flags);
3201	}
3202
3203	static const struct xattr_handler shmem_security_xattr_handler = {
3204	.prefix = XATTR_SECURITY_PREFIX,
3205	.get = shmem_xattr_handler_get,
3206	.set = shmem_xattr_handler_set,
3207	};
3208
3209	static const struct xattr_handler shmem_trusted_xattr_handler = {
3210	.prefix = XATTR_TRUSTED_PREFIX,
3211	.get = shmem_xattr_handler_get,
3212	.set = shmem_xattr_handler_set,
3213	};
3214
3215	static const struct xattr_handler *shmem_xattr_handlers[] = {
3216	#ifdef CONFIG_TMPFS_POSIX_ACL
3217	&posix_acl_access_xattr_handler,
3218	&posix_acl_default_xattr_handler,
3219	#endif
3220	&shmem_security_xattr_handler,
3221	&shmem_trusted_xattr_handler,
3222	NULL
3223	};
3224
3225	static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3226	{
3227	struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3228	return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3229	}
3230	#endif /* CONFIG_TMPFS_XATTR */
3231
3232	static const struct inode_operations shmem_short_symlink_operations = {
3233	.readlink = generic_readlink,
3234	.get_link = simple_get_link,
3235	#ifdef CONFIG_TMPFS_XATTR
3236	.listxattr = shmem_listxattr,
3237	#endif
3238	};
3239
3240	static const struct inode_operations shmem_symlink_inode_operations = {
3241	.readlink = generic_readlink,
3242	.get_link = shmem_get_link,
3243	#ifdef CONFIG_TMPFS_XATTR
3244	.listxattr = shmem_listxattr,
3245	#endif
3246	};
3247
3248	static struct dentry *shmem_get_parent(struct dentry *child)
3249	{
3250	return ERR_PTR(-ESTALE);
3251	}
3252
3253	static int shmem_match(struct inode *ino, void *vfh)
3254	{
3255	__u32 *fh = vfh;
3256	__u64 inum = fh[2];
3257	inum = (inum << 32) | fh[1];
3258	return ino->i_ino == inum && fh[0] == ino->i_generation;
3259	}
3260
3261	static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3262	struct fid *fid, int fh_len, int fh_type)
3263	{
3264	struct inode *inode;
3265	struct dentry *dentry = NULL;
3266	u64 inum;
3267
3268	if (fh_len < 3)
3269	return NULL;
3270
3271	inum = fid->raw[2];
3272	inum = (inum << 32) | fid->raw[1];
3273
3274	inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3275	shmem_match, fid->raw);
3276	if (inode) {
3277	dentry = d_find_alias(inode);
3278	iput(inode);
3279	}
3280
3281	return dentry;
3282	}
3283
3284	static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3285	struct inode *parent)
3286	{
3287	if (*len < 3) {
3288	*len = 3;
3289	return FILEID_INVALID;
3290	}
3291
3292	if (inode_unhashed(inode)) {
3293	/* Unfortunately insert_inode_hash is not idempotent,
3294	* so as we hash inodes here rather than at creation
3295	* time, we need a lock to ensure we only try
3296	* to do it once
3297	*/
3298	static DEFINE_SPINLOCK(lock);
3299	spin_lock(&lock);
3300	if (inode_unhashed(inode))
3301	__insert_inode_hash(inode,
3302	inode->i_ino + inode->i_generation);
3303	spin_unlock(&lock);
3304	}
3305
3306	fh[0] = inode->i_generation;
3307	fh[1] = inode->i_ino;
3308	fh[2] = ((__u64)inode->i_ino) >> 32;
3309
3310	*len = 3;
3311	return 1;
3312	}
3313
3314	static const struct export_operations shmem_export_ops = {
3315	.get_parent = shmem_get_parent,
3316	.encode_fh = shmem_encode_fh,
3317	.fh_to_dentry = shmem_fh_to_dentry,
3318	};
3319
3320	static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3321	bool remount)
3322	{
3323	char *this_char, *value, *rest;
3324	struct mempolicy *mpol = NULL;
3325	uid_t uid;
3326	gid_t gid;
3327
3328	while (options != NULL) {
3329	this_char = options;
3330	for (;;) {
3331	/*
3332	* NUL-terminate this option: unfortunately,
3333	* mount options form a comma-separated list,
3334	* but mpol's nodelist may also contain commas.
3335	*/
3336	options = strchr(options, ',');
3337	if (options == NULL)
3338	break;
3339	options++;
3340	if (!isdigit(*options)) {
3341	options[-1] = '\0';
3342	break;
3343	}
3344	}
3345	if (!*this_char)
3346	continue;
3347	if ((value = strchr(this_char,'=')) != NULL) {
3348	*value++ = 0;
3349	} else {
3350	pr_err("tmpfs: No value for mount option '%s'\n",
3351	this_char);
3352	goto error;
3353	}
3354
3355	if (!strcmp(this_char,"size")) {
3356	unsigned long long size;
3357	size = memparse(value,&rest);
3358	if (*rest == '%') {
3359	size <<= PAGE_SHIFT;
3360	size *= totalram_pages;
3361	do_div(size, 100);
3362	rest++;
3363	}
3364	if (*rest)
3365	goto bad_val;
3366	sbinfo->max_blocks =
3367	DIV_ROUND_UP(size, PAGE_SIZE);
3368	} else if (!strcmp(this_char,"nr_blocks")) {
3369	sbinfo->max_blocks = memparse(value, &rest);
3370	if (*rest)
3371	goto bad_val;
3372	} else if (!strcmp(this_char,"nr_inodes")) {
3373	sbinfo->max_inodes = memparse(value, &rest);
3374	if (*rest)
3375	goto bad_val;
3376	} else if (!strcmp(this_char,"mode")) {
3377	if (remount)
3378	continue;
3379	sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3380	if (*rest)
3381	goto bad_val;
3382	} else if (!strcmp(this_char,"uid")) {
3383	if (remount)
3384	continue;
3385	uid = simple_strtoul(value, &rest, 0);
3386	if (*rest)
3387	goto bad_val;
3388	sbinfo->uid = make_kuid(current_user_ns(), uid);
3389	if (!uid_valid(sbinfo->uid))
3390	goto bad_val;
3391	} else if (!strcmp(this_char,"gid")) {
3392	if (remount)
3393	continue;
3394	gid = simple_strtoul(value, &rest, 0);
3395	if (*rest)
3396	goto bad_val;
3397	sbinfo->gid = make_kgid(current_user_ns(), gid);
3398	if (!gid_valid(sbinfo->gid))
3399	goto bad_val;
3400	#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3401	} else if (!strcmp(this_char, "huge")) {
3402	int huge;
3403	huge = shmem_parse_huge(value);
3404	if (huge < 0)
3405	goto bad_val;
3406	if (!has_transparent_hugepage() &&
3407	huge != SHMEM_HUGE_NEVER)
3408	goto bad_val;
3409	sbinfo->huge = huge;
3410	#endif
3411	#ifdef CONFIG_NUMA
3412	} else if (!strcmp(this_char,"mpol")) {
3413	mpol_put(mpol);
3414	mpol = NULL;
3415	if (mpol_parse_str(value, &mpol))
3416	goto bad_val;
3417	#endif
3418	} else {
3419	pr_err("tmpfs: Bad mount option %s\n", this_char);
3420	goto error;
3421	}
3422	}
3423	sbinfo->mpol = mpol;
3424	return 0;
3425
3426	bad_val:
3427	pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3428	value, this_char);
3429	error:
3430	mpol_put(mpol);
3431	return 1;
3432
3433	}
3434
3435	static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3436	{
3437	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3438	struct shmem_sb_info config = *sbinfo;
3439	unsigned long inodes;
3440	int error = -EINVAL;
3441
3442	config.mpol = NULL;
3443	if (shmem_parse_options(data, &config, true))
3444	return error;
3445
3446	spin_lock(&sbinfo->stat_lock);
3447	inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3448	if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3449	goto out;
3450	if (config.max_inodes < inodes)
3451	goto out;
3452	/*
3453	* Those tests disallow limited->unlimited while any are in use;
3454	* but we must separately disallow unlimited->limited, because
3455	* in that case we have no record of how much is already in use.
3456	*/
3457	if (config.max_blocks && !sbinfo->max_blocks)
3458	goto out;
3459	if (config.max_inodes && !sbinfo->max_inodes)
3460	goto out;
3461
3462	error = 0;
3463	sbinfo->huge = config.huge;
3464	sbinfo->max_blocks = config.max_blocks;
3465	sbinfo->max_inodes = config.max_inodes;
3466	sbinfo->free_inodes = config.max_inodes - inodes;
3467
3468	/*
3469	* Preserve previous mempolicy unless mpol remount option was specified.
3470	*/
3471	if (config.mpol) {
3472	mpol_put(sbinfo->mpol);
3473	sbinfo->mpol = config.mpol; /* transfers initial ref */
3474	}
3475	out:
3476	spin_unlock(&sbinfo->stat_lock);
3477	return error;
3478	}
3479
3480	static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3481	{
3482	struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3483
3484	if (sbinfo->max_blocks != shmem_default_max_blocks())
3485	seq_printf(seq, ",size=%luk",
3486	sbinfo->max_blocks << (PAGE_SHIFT - 10));
3487	if (sbinfo->max_inodes != shmem_default_max_inodes())
3488	seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3489	if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3490	seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3491	if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3492	seq_printf(seq, ",uid=%u",
3493	from_kuid_munged(&init_user_ns, sbinfo->uid));
3494	if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3495	seq_printf(seq, ",gid=%u",
3496	from_kgid_munged(&init_user_ns, sbinfo->gid));
3497	#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3498	/* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3499	if (sbinfo->huge)
3500	seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3501	#endif
3502	shmem_show_mpol(seq, sbinfo->mpol);
3503	return 0;
3504	}
3505
3506	#define MFD_NAME_PREFIX "memfd:"
3507	#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3508	#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3509
3510	#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3511
3512	SYSCALL_DEFINE2(memfd_create,
3513	const char __user *, uname,
3514	unsigned int, flags)
3515	{
3516	struct shmem_inode_info *info;
3517	struct file *file;
3518	int fd, error;
3519	char *name;
3520	long len;
3521
3522	if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3523	return -EINVAL;
3524
3525	/* length includes terminating zero */
3526	len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3527	if (len <= 0)
3528	return -EFAULT;
3529	if (len > MFD_NAME_MAX_LEN + 1)
3530	return -EINVAL;
3531
3532	name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3533	if (!name)
3534	return -ENOMEM;
3535
3536	strcpy(name, MFD_NAME_PREFIX);
3537	if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3538	error = -EFAULT;
3539	goto err_name;
3540	}
3541
3542	/* terminating-zero may have changed after strnlen_user() returned */
3543	if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3544	error = -EFAULT;
3545	goto err_name;
3546	}
3547
3548	fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3549	if (fd < 0) {
3550	error = fd;
3551	goto err_name;
3552	}
3553
3554	file = shmem_file_setup(name, 0, VM_NORESERVE);
3555	if (IS_ERR(file)) {
3556	error = PTR_ERR(file);
3557	goto err_fd;
3558	}
3559	info = SHMEM_I(file_inode(file));
3560	file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3561	file->f_flags |= O_RDWR | O_LARGEFILE;
3562	if (flags & MFD_ALLOW_SEALING)
3563	info->seals &= ~F_SEAL_SEAL;
3564
3565	fd_install(fd, file);
3566	kfree(name);
3567	return fd;
3568
3569	err_fd:
3570	put_unused_fd(fd);
3571	err_name:
3572	kfree(name);
3573	return error;
3574	}
3575
3576	#endif /* CONFIG_TMPFS */
3577
3578	static void shmem_put_super(struct super_block *sb)
3579	{
3580	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3581
3582	percpu_counter_destroy(&sbinfo->used_blocks);
3583	mpol_put(sbinfo->mpol);
3584	kfree(sbinfo);
3585	sb->s_fs_info = NULL;
3586	}
3587
3588	int shmem_fill_super(struct super_block *sb, void *data, int silent)
3589	{
3590	struct inode *inode;
3591	struct shmem_sb_info *sbinfo;
3592	int err = -ENOMEM;
3593
3594	/* Round up to L1_CACHE_BYTES to resist false sharing */
3595	sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3596	L1_CACHE_BYTES), GFP_KERNEL);
3597	if (!sbinfo)
3598	return -ENOMEM;
3599
3600	sbinfo->mode = S_IRWXUGO | S_ISVTX;
3601	sbinfo->uid = current_fsuid();
3602	sbinfo->gid = current_fsgid();
3603	sb->s_fs_info = sbinfo;
3604
3605	#ifdef CONFIG_TMPFS
3606	/*
3607	* Per default we only allow half of the physical ram per
3608	* tmpfs instance, limiting inodes to one per page of lowmem;
3609	* but the internal instance is left unlimited.
3610	*/
3611	if (!(sb->s_flags & MS_KERNMOUNT)) {
3612	sbinfo->max_blocks = shmem_default_max_blocks();
3613	sbinfo->max_inodes = shmem_default_max_inodes();
3614	if (shmem_parse_options(data, sbinfo, false)) {
3615	err = -EINVAL;
3616	goto failed;
3617	}
3618	} else {
3619	sb->s_flags |= MS_NOUSER;
3620	}
3621	sb->s_export_op = &shmem_export_ops;
3622	sb->s_flags |= MS_NOSEC;
3623	#else
3624	sb->s_flags |= MS_NOUSER;
3625	#endif
3626
3627	spin_lock_init(&sbinfo->stat_lock);
3628	if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3629	goto failed;
3630	sbinfo->free_inodes = sbinfo->max_inodes;
3631	spin_lock_init(&sbinfo->shrinklist_lock);
3632	INIT_LIST_HEAD(&sbinfo->shrinklist);
3633
3634	sb->s_maxbytes = MAX_LFS_FILESIZE;
3635	sb->s_blocksize = PAGE_SIZE;
3636	sb->s_blocksize_bits = PAGE_SHIFT;
3637	sb->s_magic = TMPFS_MAGIC;
3638	sb->s_op = &shmem_ops;
3639	sb->s_time_gran = 1;
3640	#ifdef CONFIG_TMPFS_XATTR
3641	sb->s_xattr = shmem_xattr_handlers;
3642	#endif
3643	#ifdef CONFIG_TMPFS_POSIX_ACL
3644	sb->s_flags |= MS_POSIXACL;
3645	#endif
3646
3647	inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3648	if (!inode)
3649	goto failed;
3650	inode->i_uid = sbinfo->uid;
3651	inode->i_gid = sbinfo->gid;
3652	sb->s_root = d_make_root(inode);
3653	if (!sb->s_root)
3654	goto failed;
3655	return 0;
3656
3657	failed:
3658	shmem_put_super(sb);
3659	return err;
3660	}
3661
3662	static struct kmem_cache *shmem_inode_cachep;
3663
3664	static struct inode *shmem_alloc_inode(struct super_block *sb)
3665	{
3666	struct shmem_inode_info *info;
3667	info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3668	if (!info)
3669	return NULL;
3670	return &info->vfs_inode;
3671	}
3672
3673	static void shmem_destroy_callback(struct rcu_head *head)
3674	{
3675	struct inode *inode = container_of(head, struct inode, i_rcu);
3676	if (S_ISLNK(inode->i_mode))
3677	kfree(inode->i_link);
3678	kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3679	}
3680
3681	static void shmem_destroy_inode(struct inode *inode)
3682	{
3683	if (S_ISREG(inode->i_mode))
3684	mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3685	call_rcu(&inode->i_rcu, shmem_destroy_callback);
3686	}
3687
3688	static void shmem_init_inode(void *foo)
3689	{
3690	struct shmem_inode_info *info = foo;
3691	inode_init_once(&info->vfs_inode);
3692	}
3693
3694	static int shmem_init_inodecache(void)
3695	{
3696	shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3697	sizeof(struct shmem_inode_info),
3698	0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3699	return 0;
3700	}
3701
3702	static void shmem_destroy_inodecache(void)
3703	{
3704	kmem_cache_destroy(shmem_inode_cachep);
3705	}
3706
3707	static const struct address_space_operations shmem_aops = {
3708	.writepage = shmem_writepage,
3709	.set_page_dirty = __set_page_dirty_no_writeback,
3710	#ifdef CONFIG_TMPFS
3711	.write_begin = shmem_write_begin,
3712	.write_end = shmem_write_end,
3713	#endif
3714	#ifdef CONFIG_MIGRATION
3715	.migratepage = migrate_page,
3716	#endif
3717	.error_remove_page = generic_error_remove_page,
3718	};
3719
3720	static const struct file_operations shmem_file_operations = {
3721	.mmap = shmem_mmap,
3722	.get_unmapped_area = shmem_get_unmapped_area,
3723	#ifdef CONFIG_TMPFS
3724	.llseek = shmem_file_llseek,
3725	.read_iter = shmem_file_read_iter,
3726	.write_iter = generic_file_write_iter,
3727	.fsync = noop_fsync,
3728	.splice_read = generic_file_splice_read,
3729	.splice_write = iter_file_splice_write,
3730	.fallocate = shmem_fallocate,
3731	#endif
3732	};
3733
3734	static const struct inode_operations shmem_inode_operations = {
3735	.getattr = shmem_getattr,
3736	.setattr = shmem_setattr,
3737	#ifdef CONFIG_TMPFS_XATTR
3738	.listxattr = shmem_listxattr,
3739	.set_acl = simple_set_acl,
3740	#endif
3741	};
3742
3743	static const struct inode_operations shmem_dir_inode_operations = {
3744	#ifdef CONFIG_TMPFS
3745	.create = shmem_create,
3746	.lookup = simple_lookup,
3747	.link = shmem_link,
3748	.unlink = shmem_unlink,
3749	.symlink = shmem_symlink,
3750	.mkdir = shmem_mkdir,
3751	.rmdir = shmem_rmdir,
3752	.mknod = shmem_mknod,
3753	.rename = shmem_rename2,
3754	.tmpfile = shmem_tmpfile,
3755	#endif
3756	#ifdef CONFIG_TMPFS_XATTR
3757	.listxattr = shmem_listxattr,
3758	#endif
3759	#ifdef CONFIG_TMPFS_POSIX_ACL
3760	.setattr = shmem_setattr,
3761	.set_acl = simple_set_acl,
3762	#endif
3763	};
3764
3765	static const struct inode_operations shmem_special_inode_operations = {
3766	#ifdef CONFIG_TMPFS_XATTR
3767	.listxattr = shmem_listxattr,
3768	#endif
3769	#ifdef CONFIG_TMPFS_POSIX_ACL
3770	.setattr = shmem_setattr,
3771	.set_acl = simple_set_acl,
3772	#endif
3773	};
3774
3775	static const struct super_operations shmem_ops = {
3776	.alloc_inode = shmem_alloc_inode,
3777	.destroy_inode = shmem_destroy_inode,
3778	#ifdef CONFIG_TMPFS
3779	.statfs = shmem_statfs,
3780	.remount_fs = shmem_remount_fs,
3781	.show_options = shmem_show_options,
3782	#endif
3783	.evict_inode = shmem_evict_inode,
3784	.drop_inode = generic_delete_inode,
3785	.put_super = shmem_put_super,
3786	#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3787	.nr_cached_objects = shmem_unused_huge_count,
3788	.free_cached_objects = shmem_unused_huge_scan,
3789	#endif
3790	};
3791
3792	static const struct vm_operations_struct shmem_vm_ops = {
3793	.fault = shmem_fault,
3794	.map_pages = filemap_map_pages,
3795	#ifdef CONFIG_NUMA
3796	.set_policy = shmem_set_policy,
3797	.get_policy = shmem_get_policy,
3798	#endif
3799	};
3800
3801	static struct dentry *shmem_mount(struct file_system_type *fs_type,
3802	int flags, const char *dev_name, void *data)
3803	{
3804	return mount_nodev(fs_type, flags, data, shmem_fill_super);
3805	}
3806
3807	static struct file_system_type shmem_fs_type = {
3808	.owner = THIS_MODULE,
3809	.name = "tmpfs",
3810	.mount = shmem_mount,
3811	.kill_sb = kill_litter_super,
3812	.fs_flags = FS_USERNS_MOUNT,
3813	};
3814
3815	int __init shmem_init(void)
3816	{
3817	int error;
3818
3819	/* If rootfs called this, don't re-init */
3820	if (shmem_inode_cachep)
3821	return 0;
3822
3823	error = shmem_init_inodecache();
3824	if (error)
3825	goto out3;
3826
3827	error = register_filesystem(&shmem_fs_type);
3828	if (error) {
3829	pr_err("Could not register tmpfs\n");
3830	goto out2;
3831	}
3832
3833	shm_mnt = kern_mount(&shmem_fs_type);
3834	if (IS_ERR(shm_mnt)) {
3835	error = PTR_ERR(shm_mnt);
3836	pr_err("Could not kern_mount tmpfs\n");
3837	goto out1;
3838	}
3839
3840	#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3841	if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3842	SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3843	else
3844	shmem_huge = 0; /* just in case it was patched */
3845	#endif
3846	return 0;
3847
3848	out1:
3849	unregister_filesystem(&shmem_fs_type);
3850	out2:
3851	shmem_destroy_inodecache();
3852	out3:
3853	shm_mnt = ERR_PTR(error);
3854	return error;
3855	}
3856
3857	#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3858	static ssize_t shmem_enabled_show(struct kobject *kobj,
3859	struct kobj_attribute *attr, char *buf)
3860	{
3861	int values[] = {
3862	SHMEM_HUGE_ALWAYS,
3863	SHMEM_HUGE_WITHIN_SIZE,
3864	SHMEM_HUGE_ADVISE,
3865	SHMEM_HUGE_NEVER,
3866	SHMEM_HUGE_DENY,
3867	SHMEM_HUGE_FORCE,
3868	};
3869	int i, count;
3870
3871	for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3872	const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3873
3874	count += sprintf(buf + count, fmt,
3875	shmem_format_huge(values[i]));
3876	}
3877	buf[count - 1] = '\n';
3878	return count;
3879	}
3880
3881	static ssize_t shmem_enabled_store(struct kobject *kobj,
3882	struct kobj_attribute *attr, const char *buf, size_t count)
3883	{
3884	char tmp[16];
3885	int huge;
3886
3887	if (count + 1 > sizeof(tmp))
3888	return -EINVAL;
3889	memcpy(tmp, buf, count);
3890	tmp[count] = '\0';
3891	if (count && tmp[count - 1] == '\n')
3892	tmp[count - 1] = '\0';
3893
3894	huge = shmem_parse_huge(tmp);
3895	if (huge == -EINVAL)
3896	return -EINVAL;
3897	if (!has_transparent_hugepage() &&
3898	huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3899	return -EINVAL;
3900
3901	shmem_huge = huge;
3902	if (shmem_huge > SHMEM_HUGE_DENY)
3903	SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3904	return count;
3905	}
3906
3907	struct kobj_attribute shmem_enabled_attr =
3908	__ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3909	#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3910
3911	#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3912	bool shmem_huge_enabled(struct vm_area_struct *vma)
3913	{
3914	struct inode *inode = file_inode(vma->vm_file);
3915	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3916	loff_t i_size;
3917	pgoff_t off;
3918
3919	if (shmem_huge == SHMEM_HUGE_FORCE)
3920	return true;
3921	if (shmem_huge == SHMEM_HUGE_DENY)
3922	return false;
3923	switch (sbinfo->huge) {
3924	case SHMEM_HUGE_NEVER:
3925	return false;
3926	case SHMEM_HUGE_ALWAYS:
3927	return true;
3928	case SHMEM_HUGE_WITHIN_SIZE:
3929	off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3930	i_size = round_up(i_size_read(inode), PAGE_SIZE);
3931	if (i_size >= HPAGE_PMD_SIZE &&
3932	i_size >> PAGE_SHIFT >= off)
3933	return true;
3934	case SHMEM_HUGE_ADVISE:
3935	/* TODO: implement fadvise() hints */
3936	return (vma->vm_flags & VM_HUGEPAGE);
3937	default:
3938	VM_BUG_ON(1);
3939	return false;
3940	}
3941	}
3942	#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3943
3944	#else /* !CONFIG_SHMEM */
3945
3946	/*
3947	* tiny-shmem: simple shmemfs and tmpfs using ramfs code
3948	*
3949	* This is intended for small system where the benefits of the full
3950	* shmem code (swap-backed and resource-limited) are outweighed by
3951	* their complexity. On systems without swap this code should be
3952	* effectively equivalent, but much lighter weight.
3953	*/
3954
3955	static struct file_system_type shmem_fs_type = {
3956	.name = "tmpfs",
3957	.mount = ramfs_mount,
3958	.kill_sb = kill_litter_super,
3959	.fs_flags = FS_USERNS_MOUNT,
3960	};
3961
3962	int __init shmem_init(void)
3963	{
3964	BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3965
3966	shm_mnt = kern_mount(&shmem_fs_type);
3967	BUG_ON(IS_ERR(shm_mnt));
3968
3969	return 0;
3970	}
3971
3972	int shmem_unuse(swp_entry_t swap, struct page *page)
3973	{
3974	return 0;
3975	}
3976
3977	int shmem_lock(struct file *file, int lock, struct user_struct *user)
3978	{
3979	return 0;
3980	}
3981
3982	void shmem_unlock_mapping(struct address_space *mapping)
3983	{
3984	}
3985
3986	#ifdef CONFIG_MMU
3987	unsigned long shmem_get_unmapped_area(struct file *file,
3988	unsigned long addr, unsigned long len,
3989	unsigned long pgoff, unsigned long flags)
3990	{
3991	return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3992	}
3993	#endif
3994
3995	void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3996	{
3997	truncate_inode_pages_range(inode->i_mapping, lstart, lend);
3998	}
3999	EXPORT_SYMBOL_GPL(shmem_truncate_range);
4000
4001	#define shmem_vm_ops generic_file_vm_ops
4002	#define shmem_file_operations ramfs_file_operations
4003	#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4004	#define shmem_acct_size(flags, size) 0
4005	#define shmem_unacct_size(flags, size) do {} while (0)
4006
4007	#endif /* CONFIG_SHMEM */
4008
4009	/* common code */
4010
4011	static const struct dentry_operations anon_ops = {
4012	.d_dname = simple_dname
4013	};
4014
4015	static struct file *__shmem_file_setup(const char *name, loff_t size,
4016	unsigned long flags, unsigned int i_flags)
4017	{
4018	struct file *res;
4019	struct inode *inode;
4020	struct path path;
4021	struct super_block *sb;
4022	struct qstr this;
4023
4024	if (IS_ERR(shm_mnt))
4025	return ERR_CAST(shm_mnt);
4026
4027	if (size < 0 || size > MAX_LFS_FILESIZE)
4028	return ERR_PTR(-EINVAL);
4029
4030	if (shmem_acct_size(flags, size))
4031	return ERR_PTR(-ENOMEM);
4032
4033	res = ERR_PTR(-ENOMEM);
4034	this.name = name;
4035	this.len = strlen(name);
4036	this.hash = 0; /* will go */
4037	sb = shm_mnt->mnt_sb;
4038	path.mnt = mntget(shm_mnt);
4039	path.dentry = d_alloc_pseudo(sb, &this);
4040	if (!path.dentry)
4041	goto put_memory;
4042	d_set_d_op(path.dentry, &anon_ops);
4043
4044	res = ERR_PTR(-ENOSPC);
4045	inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4046	if (!inode)
4047	goto put_memory;
4048
4049	inode->i_flags |= i_flags;
4050	d_instantiate(path.dentry, inode);
4051	inode->i_size = size;
4052	clear_nlink(inode); /* It is unlinked */
4053	res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4054	if (IS_ERR(res))
4055	goto put_path;
4056
4057	res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4058	&shmem_file_operations);
4059	if (IS_ERR(res))
4060	goto put_path;
4061
4062	return res;
4063
4064	put_memory:
4065	shmem_unacct_size(flags, size);
4066	put_path:
4067	path_put(&path);
4068	return res;
4069	}
4070
4071	/**
4072	* shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4073	* kernel internal. There will be NO LSM permission checks against the
4074	* underlying inode. So users of this interface must do LSM checks at a
4075	* higher layer. The users are the big_key and shm implementations. LSM
4076	* checks are provided at the key or shm level rather than the inode.
4077	* @name: name for dentry (to be seen in /proc/<pid>/maps
4078	* @size: size to be set for the file
4079	* @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4080	*/
4081	struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4082	{
4083	return __shmem_file_setup(name, size, flags, S_PRIVATE);
4084	}
4085
4086	/**
4087	* shmem_file_setup - get an unlinked file living in tmpfs
4088	* @name: name for dentry (to be seen in /proc/<pid>/maps
4089	* @size: size to be set for the file
4090	* @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4091	*/
4092	struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4093	{
4094	return __shmem_file_setup(name, size, flags, 0);
4095	}
4096	EXPORT_SYMBOL_GPL(shmem_file_setup);
4097
4098	void shmem_set_file(struct vm_area_struct *vma, struct file *file)
4099	{
4100	if (vma->vm_file)
4101	fput(vma->vm_file);
4102	vma->vm_file = file;
4103	vma->vm_ops = &shmem_vm_ops;
4104	}
4105
4106	/**
4107	* shmem_zero_setup - setup a shared anonymous mapping
4108	* @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4109	*/
4110	int shmem_zero_setup(struct vm_area_struct *vma)
4111	{
4112	struct file *file;
4113	loff_t size = vma->vm_end - vma->vm_start;
4114
4115	/*
4116	* Cloning a new file under mmap_sem leads to a lock ordering conflict
4117	* between XFS directory reading and selinux: since this file is only
4118	* accessible to the user through its mapping, use S_PRIVATE flag to
4119	* bypass file security, in the same way as shmem_kernel_file_setup().
4120	*/
4121	file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4122	if (IS_ERR(file))
4123	return PTR_ERR(file);
4124
4125	shmem_set_file(vma, file);
4126
4127	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4128	((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4129	(vma->vm_end & HPAGE_PMD_MASK)) {
4130	khugepaged_enter(vma, vma->vm_flags);
4131	}
4132
4133	return 0;
4134	}
4135
4136	/**
4137	* shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4138	* @mapping: the page's address_space
4139	* @index: the page index
4140	* @gfp: the page allocator flags to use if allocating
4141	*
4142	* This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4143	* with any new page allocations done using the specified allocation flags.
4144	* But read_cache_page_gfp() uses the ->readpage() method: which does not
4145	* suit tmpfs, since it may have pages in swapcache, and needs to find those
4146	* for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4147	*
4148	* i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4149	* with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4150	*/
4151	struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4152	pgoff_t index, gfp_t gfp)
4153	{
4154	#ifdef CONFIG_SHMEM
4155	struct inode *inode = mapping->host;
4156	struct page *page;
4157	int error;
4158
4159	BUG_ON(mapping->a_ops != &shmem_aops);
4160	error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4161	gfp, NULL, NULL);
4162	if (error)
4163	page = ERR_PTR(error);
4164	else
4165	unlock_page(page);
4166	return page;
4167	#else
4168	/*
4169	* The tiny !SHMEM case uses ramfs without swap
4170	*/
4171	return read_cache_page_gfp(mapping, index, gfp);
4172	#endif
4173	}
4174	EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
4175