xref: /openbmc/linux/mm/shmem.c (revision ff4a7481c3898ffc3cc271d6aca431d190c37247)
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/random.h>
33 #include <linux/sched/signal.h>
34 #include <linux/export.h>
35 #include <linux/swap.h>
36 #include <linux/uio.h>
37 #include <linux/khugepaged.h>
38 #include <linux/hugetlb.h>
39 
40 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
41 
42 static struct vfsmount *shm_mnt;
43 
44 #ifdef CONFIG_SHMEM
45 /*
46  * This virtual memory filesystem is heavily based on the ramfs. It
47  * extends ramfs by the ability to use swap and honor resource limits
48  * which makes it a completely usable filesystem.
49  */
50 
51 #include <linux/xattr.h>
52 #include <linux/exportfs.h>
53 #include <linux/posix_acl.h>
54 #include <linux/posix_acl_xattr.h>
55 #include <linux/mman.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/backing-dev.h>
59 #include <linux/shmem_fs.h>
60 #include <linux/writeback.h>
61 #include <linux/blkdev.h>
62 #include <linux/pagevec.h>
63 #include <linux/percpu_counter.h>
64 #include <linux/falloc.h>
65 #include <linux/splice.h>
66 #include <linux/security.h>
67 #include <linux/swapops.h>
68 #include <linux/mempolicy.h>
69 #include <linux/namei.h>
70 #include <linux/ctype.h>
71 #include <linux/migrate.h>
72 #include <linux/highmem.h>
73 #include <linux/seq_file.h>
74 #include <linux/magic.h>
75 #include <linux/syscalls.h>
76 #include <linux/fcntl.h>
77 #include <uapi/linux/memfd.h>
78 #include <linux/userfaultfd_k.h>
79 #include <linux/rmap.h>
80 #include <linux/uuid.h>
81 
82 #include <linux/uaccess.h>
83 #include <asm/pgtable.h>
84 
85 #include "internal.h"
86 
87 #define BLOCKS_PER_PAGE  (PAGE_SIZE/512)
88 #define VM_ACCT(size)    (PAGE_ALIGN(size) >> PAGE_SHIFT)
89 
90 /* Pretend that each entry is of this size in directory's i_size */
91 #define BOGO_DIRENT_SIZE 20
92 
93 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
94 #define SHORT_SYMLINK_LEN 128
95 
96 /*
97  * shmem_fallocate communicates with shmem_fault or shmem_writepage via
98  * inode->i_private (with i_mutex making sure that it has only one user at
99  * a time): we would prefer not to enlarge the shmem inode just for that.
100  */
101 struct shmem_falloc {
102 	wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
103 	pgoff_t start;		/* start of range currently being fallocated */
104 	pgoff_t next;		/* the next page offset to be fallocated */
105 	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
106 	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
107 };
108 
109 #ifdef CONFIG_TMPFS
110 static unsigned long shmem_default_max_blocks(void)
111 {
112 	return totalram_pages / 2;
113 }
114 
115 static unsigned long shmem_default_max_inodes(void)
116 {
117 	return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
118 }
119 #endif
120 
121 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
122 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
123 				struct shmem_inode_info *info, pgoff_t index);
124 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
125 		struct page **pagep, enum sgp_type sgp,
126 		gfp_t gfp, struct vm_area_struct *vma,
127 		struct vm_fault *vmf, vm_fault_t *fault_type);
128 
129 int shmem_getpage(struct inode *inode, pgoff_t index,
130 		struct page **pagep, enum sgp_type sgp)
131 {
132 	return shmem_getpage_gfp(inode, index, pagep, sgp,
133 		mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
134 }
135 
136 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
137 {
138 	return sb->s_fs_info;
139 }
140 
141 /*
142  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
143  * for shared memory and for shared anonymous (/dev/zero) mappings
144  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
145  * consistent with the pre-accounting of private mappings ...
146  */
147 static inline int shmem_acct_size(unsigned long flags, loff_t size)
148 {
149 	return (flags & VM_NORESERVE) ?
150 		0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
151 }
152 
153 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
154 {
155 	if (!(flags & VM_NORESERVE))
156 		vm_unacct_memory(VM_ACCT(size));
157 }
158 
159 static inline int shmem_reacct_size(unsigned long flags,
160 		loff_t oldsize, loff_t newsize)
161 {
162 	if (!(flags & VM_NORESERVE)) {
163 		if (VM_ACCT(newsize) > VM_ACCT(oldsize))
164 			return security_vm_enough_memory_mm(current->mm,
165 					VM_ACCT(newsize) - VM_ACCT(oldsize));
166 		else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
167 			vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
168 	}
169 	return 0;
170 }
171 
172 /*
173  * ... whereas tmpfs objects are accounted incrementally as
174  * pages are allocated, in order to allow large sparse files.
175  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
176  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
177  */
178 static inline int shmem_acct_block(unsigned long flags, long pages)
179 {
180 	if (!(flags & VM_NORESERVE))
181 		return 0;
182 
183 	return security_vm_enough_memory_mm(current->mm,
184 			pages * VM_ACCT(PAGE_SIZE));
185 }
186 
187 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
188 {
189 	if (flags & VM_NORESERVE)
190 		vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
191 }
192 
193 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
194 {
195 	struct shmem_inode_info *info = SHMEM_I(inode);
196 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
197 
198 	if (shmem_acct_block(info->flags, pages))
199 		return false;
200 
201 	if (sbinfo->max_blocks) {
202 		if (percpu_counter_compare(&sbinfo->used_blocks,
203 					   sbinfo->max_blocks - pages) > 0)
204 			goto unacct;
205 		percpu_counter_add(&sbinfo->used_blocks, pages);
206 	}
207 
208 	return true;
209 
210 unacct:
211 	shmem_unacct_blocks(info->flags, pages);
212 	return false;
213 }
214 
215 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
216 {
217 	struct shmem_inode_info *info = SHMEM_I(inode);
218 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
219 
220 	if (sbinfo->max_blocks)
221 		percpu_counter_sub(&sbinfo->used_blocks, pages);
222 	shmem_unacct_blocks(info->flags, pages);
223 }
224 
225 static const struct super_operations shmem_ops;
226 static const struct address_space_operations shmem_aops;
227 static const struct file_operations shmem_file_operations;
228 static const struct inode_operations shmem_inode_operations;
229 static const struct inode_operations shmem_dir_inode_operations;
230 static const struct inode_operations shmem_special_inode_operations;
231 static const struct vm_operations_struct shmem_vm_ops;
232 static struct file_system_type shmem_fs_type;
233 
234 bool vma_is_shmem(struct vm_area_struct *vma)
235 {
236 	return vma->vm_ops == &shmem_vm_ops;
237 }
238 
239 static LIST_HEAD(shmem_swaplist);
240 static DEFINE_MUTEX(shmem_swaplist_mutex);
241 
242 static int shmem_reserve_inode(struct super_block *sb)
243 {
244 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
245 	if (sbinfo->max_inodes) {
246 		spin_lock(&sbinfo->stat_lock);
247 		if (!sbinfo->free_inodes) {
248 			spin_unlock(&sbinfo->stat_lock);
249 			return -ENOSPC;
250 		}
251 		sbinfo->free_inodes--;
252 		spin_unlock(&sbinfo->stat_lock);
253 	}
254 	return 0;
255 }
256 
257 static void shmem_free_inode(struct super_block *sb)
258 {
259 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
260 	if (sbinfo->max_inodes) {
261 		spin_lock(&sbinfo->stat_lock);
262 		sbinfo->free_inodes++;
263 		spin_unlock(&sbinfo->stat_lock);
264 	}
265 }
266 
267 /**
268  * shmem_recalc_inode - recalculate the block usage of an inode
269  * @inode: inode to recalc
270  *
271  * We have to calculate the free blocks since the mm can drop
272  * undirtied hole pages behind our back.
273  *
274  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
275  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
276  *
277  * It has to be called with the spinlock held.
278  */
279 static void shmem_recalc_inode(struct inode *inode)
280 {
281 	struct shmem_inode_info *info = SHMEM_I(inode);
282 	long freed;
283 
284 	freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
285 	if (freed > 0) {
286 		info->alloced -= freed;
287 		inode->i_blocks -= freed * BLOCKS_PER_PAGE;
288 		shmem_inode_unacct_blocks(inode, freed);
289 	}
290 }
291 
292 bool shmem_charge(struct inode *inode, long pages)
293 {
294 	struct shmem_inode_info *info = SHMEM_I(inode);
295 	unsigned long flags;
296 
297 	if (!shmem_inode_acct_block(inode, pages))
298 		return false;
299 
300 	/* nrpages adjustment first, then shmem_recalc_inode() when balanced */
301 	inode->i_mapping->nrpages += pages;
302 
303 	spin_lock_irqsave(&info->lock, flags);
304 	info->alloced += pages;
305 	inode->i_blocks += pages * BLOCKS_PER_PAGE;
306 	shmem_recalc_inode(inode);
307 	spin_unlock_irqrestore(&info->lock, flags);
308 
309 	return true;
310 }
311 
312 void shmem_uncharge(struct inode *inode, long pages)
313 {
314 	struct shmem_inode_info *info = SHMEM_I(inode);
315 	unsigned long flags;
316 
317 	/* nrpages adjustment done by __delete_from_page_cache() or caller */
318 
319 	spin_lock_irqsave(&info->lock, flags);
320 	info->alloced -= pages;
321 	inode->i_blocks -= pages * BLOCKS_PER_PAGE;
322 	shmem_recalc_inode(inode);
323 	spin_unlock_irqrestore(&info->lock, flags);
324 
325 	shmem_inode_unacct_blocks(inode, pages);
326 }
327 
328 /*
329  * Replace item expected in xarray by a new item, while holding xa_lock.
330  */
331 static int shmem_replace_entry(struct address_space *mapping,
332 			pgoff_t index, void *expected, void *replacement)
333 {
334 	XA_STATE(xas, &mapping->i_pages, index);
335 	void *item;
336 
337 	VM_BUG_ON(!expected);
338 	VM_BUG_ON(!replacement);
339 	item = xas_load(&xas);
340 	if (item != expected)
341 		return -ENOENT;
342 	xas_store(&xas, replacement);
343 	return 0;
344 }
345 
346 /*
347  * Sometimes, before we decide whether to proceed or to fail, we must check
348  * that an entry was not already brought back from swap by a racing thread.
349  *
350  * Checking page is not enough: by the time a SwapCache page is locked, it
351  * might be reused, and again be SwapCache, using the same swap as before.
352  */
353 static bool shmem_confirm_swap(struct address_space *mapping,
354 			       pgoff_t index, swp_entry_t swap)
355 {
356 	return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
357 }
358 
359 /*
360  * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
361  *
362  * SHMEM_HUGE_NEVER:
363  *	disables huge pages for the mount;
364  * SHMEM_HUGE_ALWAYS:
365  *	enables huge pages for the mount;
366  * SHMEM_HUGE_WITHIN_SIZE:
367  *	only allocate huge pages if the page will be fully within i_size,
368  *	also respect fadvise()/madvise() hints;
369  * SHMEM_HUGE_ADVISE:
370  *	only allocate huge pages if requested with fadvise()/madvise();
371  */
372 
373 #define SHMEM_HUGE_NEVER	0
374 #define SHMEM_HUGE_ALWAYS	1
375 #define SHMEM_HUGE_WITHIN_SIZE	2
376 #define SHMEM_HUGE_ADVISE	3
377 
378 /*
379  * Special values.
380  * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
381  *
382  * SHMEM_HUGE_DENY:
383  *	disables huge on shm_mnt and all mounts, for emergency use;
384  * SHMEM_HUGE_FORCE:
385  *	enables huge on shm_mnt and all mounts, w/o needing option, for testing;
386  *
387  */
388 #define SHMEM_HUGE_DENY		(-1)
389 #define SHMEM_HUGE_FORCE	(-2)
390 
391 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
392 /* ifdef here to avoid bloating shmem.o when not necessary */
393 
394 static int shmem_huge __read_mostly;
395 
396 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
397 static int shmem_parse_huge(const char *str)
398 {
399 	if (!strcmp(str, "never"))
400 		return SHMEM_HUGE_NEVER;
401 	if (!strcmp(str, "always"))
402 		return SHMEM_HUGE_ALWAYS;
403 	if (!strcmp(str, "within_size"))
404 		return SHMEM_HUGE_WITHIN_SIZE;
405 	if (!strcmp(str, "advise"))
406 		return SHMEM_HUGE_ADVISE;
407 	if (!strcmp(str, "deny"))
408 		return SHMEM_HUGE_DENY;
409 	if (!strcmp(str, "force"))
410 		return SHMEM_HUGE_FORCE;
411 	return -EINVAL;
412 }
413 
414 static const char *shmem_format_huge(int huge)
415 {
416 	switch (huge) {
417 	case SHMEM_HUGE_NEVER:
418 		return "never";
419 	case SHMEM_HUGE_ALWAYS:
420 		return "always";
421 	case SHMEM_HUGE_WITHIN_SIZE:
422 		return "within_size";
423 	case SHMEM_HUGE_ADVISE:
424 		return "advise";
425 	case SHMEM_HUGE_DENY:
426 		return "deny";
427 	case SHMEM_HUGE_FORCE:
428 		return "force";
429 	default:
430 		VM_BUG_ON(1);
431 		return "bad_val";
432 	}
433 }
434 #endif
435 
436 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
437 		struct shrink_control *sc, unsigned long nr_to_split)
438 {
439 	LIST_HEAD(list), *pos, *next;
440 	LIST_HEAD(to_remove);
441 	struct inode *inode;
442 	struct shmem_inode_info *info;
443 	struct page *page;
444 	unsigned long batch = sc ? sc->nr_to_scan : 128;
445 	int removed = 0, split = 0;
446 
447 	if (list_empty(&sbinfo->shrinklist))
448 		return SHRINK_STOP;
449 
450 	spin_lock(&sbinfo->shrinklist_lock);
451 	list_for_each_safe(pos, next, &sbinfo->shrinklist) {
452 		info = list_entry(pos, struct shmem_inode_info, shrinklist);
453 
454 		/* pin the inode */
455 		inode = igrab(&info->vfs_inode);
456 
457 		/* inode is about to be evicted */
458 		if (!inode) {
459 			list_del_init(&info->shrinklist);
460 			removed++;
461 			goto next;
462 		}
463 
464 		/* Check if there's anything to gain */
465 		if (round_up(inode->i_size, PAGE_SIZE) ==
466 				round_up(inode->i_size, HPAGE_PMD_SIZE)) {
467 			list_move(&info->shrinklist, &to_remove);
468 			removed++;
469 			goto next;
470 		}
471 
472 		list_move(&info->shrinklist, &list);
473 next:
474 		if (!--batch)
475 			break;
476 	}
477 	spin_unlock(&sbinfo->shrinklist_lock);
478 
479 	list_for_each_safe(pos, next, &to_remove) {
480 		info = list_entry(pos, struct shmem_inode_info, shrinklist);
481 		inode = &info->vfs_inode;
482 		list_del_init(&info->shrinklist);
483 		iput(inode);
484 	}
485 
486 	list_for_each_safe(pos, next, &list) {
487 		int ret;
488 
489 		info = list_entry(pos, struct shmem_inode_info, shrinklist);
490 		inode = &info->vfs_inode;
491 
492 		if (nr_to_split && split >= nr_to_split)
493 			goto leave;
494 
495 		page = find_get_page(inode->i_mapping,
496 				(inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
497 		if (!page)
498 			goto drop;
499 
500 		/* No huge page at the end of the file: nothing to split */
501 		if (!PageTransHuge(page)) {
502 			put_page(page);
503 			goto drop;
504 		}
505 
506 		/*
507 		 * Leave the inode on the list if we failed to lock
508 		 * the page at this time.
509 		 *
510 		 * Waiting for the lock may lead to deadlock in the
511 		 * reclaim path.
512 		 */
513 		if (!trylock_page(page)) {
514 			put_page(page);
515 			goto leave;
516 		}
517 
518 		ret = split_huge_page(page);
519 		unlock_page(page);
520 		put_page(page);
521 
522 		/* If split failed leave the inode on the list */
523 		if (ret)
524 			goto leave;
525 
526 		split++;
527 drop:
528 		list_del_init(&info->shrinklist);
529 		removed++;
530 leave:
531 		iput(inode);
532 	}
533 
534 	spin_lock(&sbinfo->shrinklist_lock);
535 	list_splice_tail(&list, &sbinfo->shrinklist);
536 	sbinfo->shrinklist_len -= removed;
537 	spin_unlock(&sbinfo->shrinklist_lock);
538 
539 	return split;
540 }
541 
542 static long shmem_unused_huge_scan(struct super_block *sb,
543 		struct shrink_control *sc)
544 {
545 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
546 
547 	if (!READ_ONCE(sbinfo->shrinklist_len))
548 		return SHRINK_STOP;
549 
550 	return shmem_unused_huge_shrink(sbinfo, sc, 0);
551 }
552 
553 static long shmem_unused_huge_count(struct super_block *sb,
554 		struct shrink_control *sc)
555 {
556 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
557 	return READ_ONCE(sbinfo->shrinklist_len);
558 }
559 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
560 
561 #define shmem_huge SHMEM_HUGE_DENY
562 
563 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
564 		struct shrink_control *sc, unsigned long nr_to_split)
565 {
566 	return 0;
567 }
568 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
569 
570 static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
571 {
572 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
573 	    (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
574 	    shmem_huge != SHMEM_HUGE_DENY)
575 		return true;
576 	return false;
577 }
578 
579 /*
580  * Like add_to_page_cache_locked, but error if expected item has gone.
581  */
582 static int shmem_add_to_page_cache(struct page *page,
583 				   struct address_space *mapping,
584 				   pgoff_t index, void *expected, gfp_t gfp)
585 {
586 	XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
587 	unsigned long i = 0;
588 	unsigned long nr = 1UL << compound_order(page);
589 
590 	VM_BUG_ON_PAGE(PageTail(page), page);
591 	VM_BUG_ON_PAGE(index != round_down(index, nr), page);
592 	VM_BUG_ON_PAGE(!PageLocked(page), page);
593 	VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
594 	VM_BUG_ON(expected && PageTransHuge(page));
595 
596 	page_ref_add(page, nr);
597 	page->mapping = mapping;
598 	page->index = index;
599 
600 	do {
601 		void *entry;
602 		xas_lock_irq(&xas);
603 		entry = xas_find_conflict(&xas);
604 		if (entry != expected)
605 			xas_set_err(&xas, -EEXIST);
606 		xas_create_range(&xas);
607 		if (xas_error(&xas))
608 			goto unlock;
609 next:
610 		xas_store(&xas, page + i);
611 		if (++i < nr) {
612 			xas_next(&xas);
613 			goto next;
614 		}
615 		if (PageTransHuge(page)) {
616 			count_vm_event(THP_FILE_ALLOC);
617 			__inc_node_page_state(page, NR_SHMEM_THPS);
618 		}
619 		mapping->nrpages += nr;
620 		__mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
621 		__mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
622 unlock:
623 		xas_unlock_irq(&xas);
624 	} while (xas_nomem(&xas, gfp));
625 
626 	if (xas_error(&xas)) {
627 		page->mapping = NULL;
628 		page_ref_sub(page, nr);
629 		return xas_error(&xas);
630 	}
631 
632 	return 0;
633 }
634 
635 /*
636  * Like delete_from_page_cache, but substitutes swap for page.
637  */
638 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
639 {
640 	struct address_space *mapping = page->mapping;
641 	int error;
642 
643 	VM_BUG_ON_PAGE(PageCompound(page), page);
644 
645 	xa_lock_irq(&mapping->i_pages);
646 	error = shmem_replace_entry(mapping, page->index, page, radswap);
647 	page->mapping = NULL;
648 	mapping->nrpages--;
649 	__dec_node_page_state(page, NR_FILE_PAGES);
650 	__dec_node_page_state(page, NR_SHMEM);
651 	xa_unlock_irq(&mapping->i_pages);
652 	put_page(page);
653 	BUG_ON(error);
654 }
655 
656 /*
657  * Remove swap entry from page cache, free the swap and its page cache.
658  */
659 static int shmem_free_swap(struct address_space *mapping,
660 			   pgoff_t index, void *radswap)
661 {
662 	void *old;
663 
664 	old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
665 	if (old != radswap)
666 		return -ENOENT;
667 	free_swap_and_cache(radix_to_swp_entry(radswap));
668 	return 0;
669 }
670 
671 /*
672  * Determine (in bytes) how many of the shmem object's pages mapped by the
673  * given offsets are swapped out.
674  *
675  * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
676  * as long as the inode doesn't go away and racy results are not a problem.
677  */
678 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
679 						pgoff_t start, pgoff_t end)
680 {
681 	XA_STATE(xas, &mapping->i_pages, start);
682 	struct page *page;
683 	unsigned long swapped = 0;
684 
685 	rcu_read_lock();
686 	xas_for_each(&xas, page, end - 1) {
687 		if (xas_retry(&xas, page))
688 			continue;
689 		if (xa_is_value(page))
690 			swapped++;
691 
692 		if (need_resched()) {
693 			xas_pause(&xas);
694 			cond_resched_rcu();
695 		}
696 	}
697 
698 	rcu_read_unlock();
699 
700 	return swapped << PAGE_SHIFT;
701 }
702 
703 /*
704  * Determine (in bytes) how many of the shmem object's pages mapped by the
705  * given vma is swapped out.
706  *
707  * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
708  * as long as the inode doesn't go away and racy results are not a problem.
709  */
710 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
711 {
712 	struct inode *inode = file_inode(vma->vm_file);
713 	struct shmem_inode_info *info = SHMEM_I(inode);
714 	struct address_space *mapping = inode->i_mapping;
715 	unsigned long swapped;
716 
717 	/* Be careful as we don't hold info->lock */
718 	swapped = READ_ONCE(info->swapped);
719 
720 	/*
721 	 * The easier cases are when the shmem object has nothing in swap, or
722 	 * the vma maps it whole. Then we can simply use the stats that we
723 	 * already track.
724 	 */
725 	if (!swapped)
726 		return 0;
727 
728 	if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
729 		return swapped << PAGE_SHIFT;
730 
731 	/* Here comes the more involved part */
732 	return shmem_partial_swap_usage(mapping,
733 			linear_page_index(vma, vma->vm_start),
734 			linear_page_index(vma, vma->vm_end));
735 }
736 
737 /*
738  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
739  */
740 void shmem_unlock_mapping(struct address_space *mapping)
741 {
742 	struct pagevec pvec;
743 	pgoff_t indices[PAGEVEC_SIZE];
744 	pgoff_t index = 0;
745 
746 	pagevec_init(&pvec);
747 	/*
748 	 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
749 	 */
750 	while (!mapping_unevictable(mapping)) {
751 		/*
752 		 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
753 		 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
754 		 */
755 		pvec.nr = find_get_entries(mapping, index,
756 					   PAGEVEC_SIZE, pvec.pages, indices);
757 		if (!pvec.nr)
758 			break;
759 		index = indices[pvec.nr - 1] + 1;
760 		pagevec_remove_exceptionals(&pvec);
761 		check_move_unevictable_pages(&pvec);
762 		pagevec_release(&pvec);
763 		cond_resched();
764 	}
765 }
766 
767 /*
768  * Remove range of pages and swap entries from page cache, and free them.
769  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
770  */
771 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
772 								 bool unfalloc)
773 {
774 	struct address_space *mapping = inode->i_mapping;
775 	struct shmem_inode_info *info = SHMEM_I(inode);
776 	pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
777 	pgoff_t end = (lend + 1) >> PAGE_SHIFT;
778 	unsigned int partial_start = lstart & (PAGE_SIZE - 1);
779 	unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
780 	struct pagevec pvec;
781 	pgoff_t indices[PAGEVEC_SIZE];
782 	long nr_swaps_freed = 0;
783 	pgoff_t index;
784 	int i;
785 
786 	if (lend == -1)
787 		end = -1;	/* unsigned, so actually very big */
788 
789 	pagevec_init(&pvec);
790 	index = start;
791 	while (index < end) {
792 		pvec.nr = find_get_entries(mapping, index,
793 			min(end - index, (pgoff_t)PAGEVEC_SIZE),
794 			pvec.pages, indices);
795 		if (!pvec.nr)
796 			break;
797 		for (i = 0; i < pagevec_count(&pvec); i++) {
798 			struct page *page = pvec.pages[i];
799 
800 			index = indices[i];
801 			if (index >= end)
802 				break;
803 
804 			if (xa_is_value(page)) {
805 				if (unfalloc)
806 					continue;
807 				nr_swaps_freed += !shmem_free_swap(mapping,
808 								index, page);
809 				continue;
810 			}
811 
812 			VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
813 
814 			if (!trylock_page(page))
815 				continue;
816 
817 			if (PageTransTail(page)) {
818 				/* Middle of THP: zero out the page */
819 				clear_highpage(page);
820 				unlock_page(page);
821 				continue;
822 			} else if (PageTransHuge(page)) {
823 				if (index == round_down(end, HPAGE_PMD_NR)) {
824 					/*
825 					 * Range ends in the middle of THP:
826 					 * zero out the page
827 					 */
828 					clear_highpage(page);
829 					unlock_page(page);
830 					continue;
831 				}
832 				index += HPAGE_PMD_NR - 1;
833 				i += HPAGE_PMD_NR - 1;
834 			}
835 
836 			if (!unfalloc || !PageUptodate(page)) {
837 				VM_BUG_ON_PAGE(PageTail(page), page);
838 				if (page_mapping(page) == mapping) {
839 					VM_BUG_ON_PAGE(PageWriteback(page), page);
840 					truncate_inode_page(mapping, page);
841 				}
842 			}
843 			unlock_page(page);
844 		}
845 		pagevec_remove_exceptionals(&pvec);
846 		pagevec_release(&pvec);
847 		cond_resched();
848 		index++;
849 	}
850 
851 	if (partial_start) {
852 		struct page *page = NULL;
853 		shmem_getpage(inode, start - 1, &page, SGP_READ);
854 		if (page) {
855 			unsigned int top = PAGE_SIZE;
856 			if (start > end) {
857 				top = partial_end;
858 				partial_end = 0;
859 			}
860 			zero_user_segment(page, partial_start, top);
861 			set_page_dirty(page);
862 			unlock_page(page);
863 			put_page(page);
864 		}
865 	}
866 	if (partial_end) {
867 		struct page *page = NULL;
868 		shmem_getpage(inode, end, &page, SGP_READ);
869 		if (page) {
870 			zero_user_segment(page, 0, partial_end);
871 			set_page_dirty(page);
872 			unlock_page(page);
873 			put_page(page);
874 		}
875 	}
876 	if (start >= end)
877 		return;
878 
879 	index = start;
880 	while (index < end) {
881 		cond_resched();
882 
883 		pvec.nr = find_get_entries(mapping, index,
884 				min(end - index, (pgoff_t)PAGEVEC_SIZE),
885 				pvec.pages, indices);
886 		if (!pvec.nr) {
887 			/* If all gone or hole-punch or unfalloc, we're done */
888 			if (index == start || end != -1)
889 				break;
890 			/* But if truncating, restart to make sure all gone */
891 			index = start;
892 			continue;
893 		}
894 		for (i = 0; i < pagevec_count(&pvec); i++) {
895 			struct page *page = pvec.pages[i];
896 
897 			index = indices[i];
898 			if (index >= end)
899 				break;
900 
901 			if (xa_is_value(page)) {
902 				if (unfalloc)
903 					continue;
904 				if (shmem_free_swap(mapping, index, page)) {
905 					/* Swap was replaced by page: retry */
906 					index--;
907 					break;
908 				}
909 				nr_swaps_freed++;
910 				continue;
911 			}
912 
913 			lock_page(page);
914 
915 			if (PageTransTail(page)) {
916 				/* Middle of THP: zero out the page */
917 				clear_highpage(page);
918 				unlock_page(page);
919 				/*
920 				 * Partial thp truncate due 'start' in middle
921 				 * of THP: don't need to look on these pages
922 				 * again on !pvec.nr restart.
923 				 */
924 				if (index != round_down(end, HPAGE_PMD_NR))
925 					start++;
926 				continue;
927 			} else if (PageTransHuge(page)) {
928 				if (index == round_down(end, HPAGE_PMD_NR)) {
929 					/*
930 					 * Range ends in the middle of THP:
931 					 * zero out the page
932 					 */
933 					clear_highpage(page);
934 					unlock_page(page);
935 					continue;
936 				}
937 				index += HPAGE_PMD_NR - 1;
938 				i += HPAGE_PMD_NR - 1;
939 			}
940 
941 			if (!unfalloc || !PageUptodate(page)) {
942 				VM_BUG_ON_PAGE(PageTail(page), page);
943 				if (page_mapping(page) == mapping) {
944 					VM_BUG_ON_PAGE(PageWriteback(page), page);
945 					truncate_inode_page(mapping, page);
946 				} else {
947 					/* Page was replaced by swap: retry */
948 					unlock_page(page);
949 					index--;
950 					break;
951 				}
952 			}
953 			unlock_page(page);
954 		}
955 		pagevec_remove_exceptionals(&pvec);
956 		pagevec_release(&pvec);
957 		index++;
958 	}
959 
960 	spin_lock_irq(&info->lock);
961 	info->swapped -= nr_swaps_freed;
962 	shmem_recalc_inode(inode);
963 	spin_unlock_irq(&info->lock);
964 }
965 
966 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
967 {
968 	shmem_undo_range(inode, lstart, lend, false);
969 	inode->i_ctime = inode->i_mtime = current_time(inode);
970 }
971 EXPORT_SYMBOL_GPL(shmem_truncate_range);
972 
973 static int shmem_getattr(const struct path *path, struct kstat *stat,
974 			 u32 request_mask, unsigned int query_flags)
975 {
976 	struct inode *inode = path->dentry->d_inode;
977 	struct shmem_inode_info *info = SHMEM_I(inode);
978 	struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
979 
980 	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
981 		spin_lock_irq(&info->lock);
982 		shmem_recalc_inode(inode);
983 		spin_unlock_irq(&info->lock);
984 	}
985 	generic_fillattr(inode, stat);
986 
987 	if (is_huge_enabled(sb_info))
988 		stat->blksize = HPAGE_PMD_SIZE;
989 
990 	return 0;
991 }
992 
993 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
994 {
995 	struct inode *inode = d_inode(dentry);
996 	struct shmem_inode_info *info = SHMEM_I(inode);
997 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
998 	int error;
999 
1000 	error = setattr_prepare(dentry, attr);
1001 	if (error)
1002 		return error;
1003 
1004 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1005 		loff_t oldsize = inode->i_size;
1006 		loff_t newsize = attr->ia_size;
1007 
1008 		/* protected by i_mutex */
1009 		if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1010 		    (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1011 			return -EPERM;
1012 
1013 		if (newsize != oldsize) {
1014 			error = shmem_reacct_size(SHMEM_I(inode)->flags,
1015 					oldsize, newsize);
1016 			if (error)
1017 				return error;
1018 			i_size_write(inode, newsize);
1019 			inode->i_ctime = inode->i_mtime = current_time(inode);
1020 		}
1021 		if (newsize <= oldsize) {
1022 			loff_t holebegin = round_up(newsize, PAGE_SIZE);
1023 			if (oldsize > holebegin)
1024 				unmap_mapping_range(inode->i_mapping,
1025 							holebegin, 0, 1);
1026 			if (info->alloced)
1027 				shmem_truncate_range(inode,
1028 							newsize, (loff_t)-1);
1029 			/* unmap again to remove racily COWed private pages */
1030 			if (oldsize > holebegin)
1031 				unmap_mapping_range(inode->i_mapping,
1032 							holebegin, 0, 1);
1033 
1034 			/*
1035 			 * Part of the huge page can be beyond i_size: subject
1036 			 * to shrink under memory pressure.
1037 			 */
1038 			if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1039 				spin_lock(&sbinfo->shrinklist_lock);
1040 				/*
1041 				 * _careful to defend against unlocked access to
1042 				 * ->shrink_list in shmem_unused_huge_shrink()
1043 				 */
1044 				if (list_empty_careful(&info->shrinklist)) {
1045 					list_add_tail(&info->shrinklist,
1046 							&sbinfo->shrinklist);
1047 					sbinfo->shrinklist_len++;
1048 				}
1049 				spin_unlock(&sbinfo->shrinklist_lock);
1050 			}
1051 		}
1052 	}
1053 
1054 	setattr_copy(inode, attr);
1055 	if (attr->ia_valid & ATTR_MODE)
1056 		error = posix_acl_chmod(inode, inode->i_mode);
1057 	return error;
1058 }
1059 
1060 static void shmem_evict_inode(struct inode *inode)
1061 {
1062 	struct shmem_inode_info *info = SHMEM_I(inode);
1063 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1064 
1065 	if (inode->i_mapping->a_ops == &shmem_aops) {
1066 		shmem_unacct_size(info->flags, inode->i_size);
1067 		inode->i_size = 0;
1068 		shmem_truncate_range(inode, 0, (loff_t)-1);
1069 		if (!list_empty(&info->shrinklist)) {
1070 			spin_lock(&sbinfo->shrinklist_lock);
1071 			if (!list_empty(&info->shrinklist)) {
1072 				list_del_init(&info->shrinklist);
1073 				sbinfo->shrinklist_len--;
1074 			}
1075 			spin_unlock(&sbinfo->shrinklist_lock);
1076 		}
1077 		if (!list_empty(&info->swaplist)) {
1078 			mutex_lock(&shmem_swaplist_mutex);
1079 			list_del_init(&info->swaplist);
1080 			mutex_unlock(&shmem_swaplist_mutex);
1081 		}
1082 	}
1083 
1084 	simple_xattrs_free(&info->xattrs);
1085 	WARN_ON(inode->i_blocks);
1086 	shmem_free_inode(inode->i_sb);
1087 	clear_inode(inode);
1088 }
1089 
1090 static unsigned long find_swap_entry(struct xarray *xa, void *item)
1091 {
1092 	XA_STATE(xas, xa, 0);
1093 	unsigned int checked = 0;
1094 	void *entry;
1095 
1096 	rcu_read_lock();
1097 	xas_for_each(&xas, entry, ULONG_MAX) {
1098 		if (xas_retry(&xas, entry))
1099 			continue;
1100 		if (entry == item)
1101 			break;
1102 		checked++;
1103 		if ((checked % XA_CHECK_SCHED) != 0)
1104 			continue;
1105 		xas_pause(&xas);
1106 		cond_resched_rcu();
1107 	}
1108 	rcu_read_unlock();
1109 
1110 	return entry ? xas.xa_index : -1;
1111 }
1112 
1113 /*
1114  * If swap found in inode, free it and move page from swapcache to filecache.
1115  */
1116 static int shmem_unuse_inode(struct shmem_inode_info *info,
1117 			     swp_entry_t swap, struct page **pagep)
1118 {
1119 	struct address_space *mapping = info->vfs_inode.i_mapping;
1120 	void *radswap;
1121 	pgoff_t index;
1122 	gfp_t gfp;
1123 	int error = 0;
1124 
1125 	radswap = swp_to_radix_entry(swap);
1126 	index = find_swap_entry(&mapping->i_pages, radswap);
1127 	if (index == -1)
1128 		return -EAGAIN;	/* tell shmem_unuse we found nothing */
1129 
1130 	/*
1131 	 * Move _head_ to start search for next from here.
1132 	 * But be careful: shmem_evict_inode checks list_empty without taking
1133 	 * mutex, and there's an instant in list_move_tail when info->swaplist
1134 	 * would appear empty, if it were the only one on shmem_swaplist.
1135 	 */
1136 	if (shmem_swaplist.next != &info->swaplist)
1137 		list_move_tail(&shmem_swaplist, &info->swaplist);
1138 
1139 	gfp = mapping_gfp_mask(mapping);
1140 	if (shmem_should_replace_page(*pagep, gfp)) {
1141 		mutex_unlock(&shmem_swaplist_mutex);
1142 		error = shmem_replace_page(pagep, gfp, info, index);
1143 		mutex_lock(&shmem_swaplist_mutex);
1144 		/*
1145 		 * We needed to drop mutex to make that restrictive page
1146 		 * allocation, but the inode might have been freed while we
1147 		 * dropped it: although a racing shmem_evict_inode() cannot
1148 		 * complete without emptying the page cache, our page lock
1149 		 * on this swapcache page is not enough to prevent that -
1150 		 * free_swap_and_cache() of our swap entry will only
1151 		 * trylock_page(), removing swap from page cache whatever.
1152 		 *
1153 		 * We must not proceed to shmem_add_to_page_cache() if the
1154 		 * inode has been freed, but of course we cannot rely on
1155 		 * inode or mapping or info to check that.  However, we can
1156 		 * safely check if our swap entry is still in use (and here
1157 		 * it can't have got reused for another page): if it's still
1158 		 * in use, then the inode cannot have been freed yet, and we
1159 		 * can safely proceed (if it's no longer in use, that tells
1160 		 * nothing about the inode, but we don't need to unuse swap).
1161 		 */
1162 		if (!page_swapcount(*pagep))
1163 			error = -ENOENT;
1164 	}
1165 
1166 	/*
1167 	 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1168 	 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1169 	 * beneath us (pagelock doesn't help until the page is in pagecache).
1170 	 */
1171 	if (!error)
1172 		error = shmem_add_to_page_cache(*pagep, mapping, index,
1173 						radswap, gfp);
1174 	if (error != -ENOMEM) {
1175 		/*
1176 		 * Truncation and eviction use free_swap_and_cache(), which
1177 		 * only does trylock page: if we raced, best clean up here.
1178 		 */
1179 		delete_from_swap_cache(*pagep);
1180 		set_page_dirty(*pagep);
1181 		if (!error) {
1182 			spin_lock_irq(&info->lock);
1183 			info->swapped--;
1184 			spin_unlock_irq(&info->lock);
1185 			swap_free(swap);
1186 		}
1187 	}
1188 	return error;
1189 }
1190 
1191 /*
1192  * Search through swapped inodes to find and replace swap by page.
1193  */
1194 int shmem_unuse(swp_entry_t swap, struct page *page)
1195 {
1196 	struct list_head *this, *next;
1197 	struct shmem_inode_info *info;
1198 	struct mem_cgroup *memcg;
1199 	int error = 0;
1200 
1201 	/*
1202 	 * There's a faint possibility that swap page was replaced before
1203 	 * caller locked it: caller will come back later with the right page.
1204 	 */
1205 	if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1206 		goto out;
1207 
1208 	/*
1209 	 * Charge page using GFP_KERNEL while we can wait, before taking
1210 	 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1211 	 * Charged back to the user (not to caller) when swap account is used.
1212 	 */
1213 	error = mem_cgroup_try_charge_delay(page, current->mm, GFP_KERNEL,
1214 					    &memcg, false);
1215 	if (error)
1216 		goto out;
1217 	/* No memory allocation: swap entry occupies the slot for the page */
1218 	error = -EAGAIN;
1219 
1220 	mutex_lock(&shmem_swaplist_mutex);
1221 	list_for_each_safe(this, next, &shmem_swaplist) {
1222 		info = list_entry(this, struct shmem_inode_info, swaplist);
1223 		if (info->swapped)
1224 			error = shmem_unuse_inode(info, swap, &page);
1225 		else
1226 			list_del_init(&info->swaplist);
1227 		cond_resched();
1228 		if (error != -EAGAIN)
1229 			break;
1230 		/* found nothing in this: move on to search the next */
1231 	}
1232 	mutex_unlock(&shmem_swaplist_mutex);
1233 
1234 	if (error) {
1235 		if (error != -ENOMEM)
1236 			error = 0;
1237 		mem_cgroup_cancel_charge(page, memcg, false);
1238 	} else
1239 		mem_cgroup_commit_charge(page, memcg, true, false);
1240 out:
1241 	unlock_page(page);
1242 	put_page(page);
1243 	return error;
1244 }
1245 
1246 /*
1247  * Move the page from the page cache to the swap cache.
1248  */
1249 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1250 {
1251 	struct shmem_inode_info *info;
1252 	struct address_space *mapping;
1253 	struct inode *inode;
1254 	swp_entry_t swap;
1255 	pgoff_t index;
1256 
1257 	VM_BUG_ON_PAGE(PageCompound(page), page);
1258 	BUG_ON(!PageLocked(page));
1259 	mapping = page->mapping;
1260 	index = page->index;
1261 	inode = mapping->host;
1262 	info = SHMEM_I(inode);
1263 	if (info->flags & VM_LOCKED)
1264 		goto redirty;
1265 	if (!total_swap_pages)
1266 		goto redirty;
1267 
1268 	/*
1269 	 * Our capabilities prevent regular writeback or sync from ever calling
1270 	 * shmem_writepage; but a stacking filesystem might use ->writepage of
1271 	 * its underlying filesystem, in which case tmpfs should write out to
1272 	 * swap only in response to memory pressure, and not for the writeback
1273 	 * threads or sync.
1274 	 */
1275 	if (!wbc->for_reclaim) {
1276 		WARN_ON_ONCE(1);	/* Still happens? Tell us about it! */
1277 		goto redirty;
1278 	}
1279 
1280 	/*
1281 	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1282 	 * value into swapfile.c, the only way we can correctly account for a
1283 	 * fallocated page arriving here is now to initialize it and write it.
1284 	 *
1285 	 * That's okay for a page already fallocated earlier, but if we have
1286 	 * not yet completed the fallocation, then (a) we want to keep track
1287 	 * of this page in case we have to undo it, and (b) it may not be a
1288 	 * good idea to continue anyway, once we're pushing into swap.  So
1289 	 * reactivate the page, and let shmem_fallocate() quit when too many.
1290 	 */
1291 	if (!PageUptodate(page)) {
1292 		if (inode->i_private) {
1293 			struct shmem_falloc *shmem_falloc;
1294 			spin_lock(&inode->i_lock);
1295 			shmem_falloc = inode->i_private;
1296 			if (shmem_falloc &&
1297 			    !shmem_falloc->waitq &&
1298 			    index >= shmem_falloc->start &&
1299 			    index < shmem_falloc->next)
1300 				shmem_falloc->nr_unswapped++;
1301 			else
1302 				shmem_falloc = NULL;
1303 			spin_unlock(&inode->i_lock);
1304 			if (shmem_falloc)
1305 				goto redirty;
1306 		}
1307 		clear_highpage(page);
1308 		flush_dcache_page(page);
1309 		SetPageUptodate(page);
1310 	}
1311 
1312 	swap = get_swap_page(page);
1313 	if (!swap.val)
1314 		goto redirty;
1315 
1316 	/*
1317 	 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1318 	 * if it's not already there.  Do it now before the page is
1319 	 * moved to swap cache, when its pagelock no longer protects
1320 	 * the inode from eviction.  But don't unlock the mutex until
1321 	 * we've incremented swapped, because shmem_unuse_inode() will
1322 	 * prune a !swapped inode from the swaplist under this mutex.
1323 	 */
1324 	mutex_lock(&shmem_swaplist_mutex);
1325 	if (list_empty(&info->swaplist))
1326 		list_add_tail(&info->swaplist, &shmem_swaplist);
1327 
1328 	if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1329 		spin_lock_irq(&info->lock);
1330 		shmem_recalc_inode(inode);
1331 		info->swapped++;
1332 		spin_unlock_irq(&info->lock);
1333 
1334 		swap_shmem_alloc(swap);
1335 		shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1336 
1337 		mutex_unlock(&shmem_swaplist_mutex);
1338 		BUG_ON(page_mapped(page));
1339 		swap_writepage(page, wbc);
1340 		return 0;
1341 	}
1342 
1343 	mutex_unlock(&shmem_swaplist_mutex);
1344 	put_swap_page(page, swap);
1345 redirty:
1346 	set_page_dirty(page);
1347 	if (wbc->for_reclaim)
1348 		return AOP_WRITEPAGE_ACTIVATE;	/* Return with page locked */
1349 	unlock_page(page);
1350 	return 0;
1351 }
1352 
1353 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1354 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1355 {
1356 	char buffer[64];
1357 
1358 	if (!mpol || mpol->mode == MPOL_DEFAULT)
1359 		return;		/* show nothing */
1360 
1361 	mpol_to_str(buffer, sizeof(buffer), mpol);
1362 
1363 	seq_printf(seq, ",mpol=%s", buffer);
1364 }
1365 
1366 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1367 {
1368 	struct mempolicy *mpol = NULL;
1369 	if (sbinfo->mpol) {
1370 		spin_lock(&sbinfo->stat_lock);	/* prevent replace/use races */
1371 		mpol = sbinfo->mpol;
1372 		mpol_get(mpol);
1373 		spin_unlock(&sbinfo->stat_lock);
1374 	}
1375 	return mpol;
1376 }
1377 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1378 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1379 {
1380 }
1381 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1382 {
1383 	return NULL;
1384 }
1385 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1386 #ifndef CONFIG_NUMA
1387 #define vm_policy vm_private_data
1388 #endif
1389 
1390 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1391 		struct shmem_inode_info *info, pgoff_t index)
1392 {
1393 	/* Create a pseudo vma that just contains the policy */
1394 	vma_init(vma, NULL);
1395 	/* Bias interleave by inode number to distribute better across nodes */
1396 	vma->vm_pgoff = index + info->vfs_inode.i_ino;
1397 	vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1398 }
1399 
1400 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1401 {
1402 	/* Drop reference taken by mpol_shared_policy_lookup() */
1403 	mpol_cond_put(vma->vm_policy);
1404 }
1405 
1406 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1407 			struct shmem_inode_info *info, pgoff_t index)
1408 {
1409 	struct vm_area_struct pvma;
1410 	struct page *page;
1411 	struct vm_fault vmf;
1412 
1413 	shmem_pseudo_vma_init(&pvma, info, index);
1414 	vmf.vma = &pvma;
1415 	vmf.address = 0;
1416 	page = swap_cluster_readahead(swap, gfp, &vmf);
1417 	shmem_pseudo_vma_destroy(&pvma);
1418 
1419 	return page;
1420 }
1421 
1422 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1423 		struct shmem_inode_info *info, pgoff_t index)
1424 {
1425 	struct vm_area_struct pvma;
1426 	struct address_space *mapping = info->vfs_inode.i_mapping;
1427 	pgoff_t hindex;
1428 	struct page *page;
1429 
1430 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1431 		return NULL;
1432 
1433 	hindex = round_down(index, HPAGE_PMD_NR);
1434 	if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1435 								XA_PRESENT))
1436 		return NULL;
1437 
1438 	shmem_pseudo_vma_init(&pvma, info, hindex);
1439 	page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1440 			HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1441 	shmem_pseudo_vma_destroy(&pvma);
1442 	if (page)
1443 		prep_transhuge_page(page);
1444 	return page;
1445 }
1446 
1447 static struct page *shmem_alloc_page(gfp_t gfp,
1448 			struct shmem_inode_info *info, pgoff_t index)
1449 {
1450 	struct vm_area_struct pvma;
1451 	struct page *page;
1452 
1453 	shmem_pseudo_vma_init(&pvma, info, index);
1454 	page = alloc_page_vma(gfp, &pvma, 0);
1455 	shmem_pseudo_vma_destroy(&pvma);
1456 
1457 	return page;
1458 }
1459 
1460 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1461 		struct inode *inode,
1462 		pgoff_t index, bool huge)
1463 {
1464 	struct shmem_inode_info *info = SHMEM_I(inode);
1465 	struct page *page;
1466 	int nr;
1467 	int err = -ENOSPC;
1468 
1469 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1470 		huge = false;
1471 	nr = huge ? HPAGE_PMD_NR : 1;
1472 
1473 	if (!shmem_inode_acct_block(inode, nr))
1474 		goto failed;
1475 
1476 	if (huge)
1477 		page = shmem_alloc_hugepage(gfp, info, index);
1478 	else
1479 		page = shmem_alloc_page(gfp, info, index);
1480 	if (page) {
1481 		__SetPageLocked(page);
1482 		__SetPageSwapBacked(page);
1483 		return page;
1484 	}
1485 
1486 	err = -ENOMEM;
1487 	shmem_inode_unacct_blocks(inode, nr);
1488 failed:
1489 	return ERR_PTR(err);
1490 }
1491 
1492 /*
1493  * When a page is moved from swapcache to shmem filecache (either by the
1494  * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1495  * shmem_unuse_inode()), it may have been read in earlier from swap, in
1496  * ignorance of the mapping it belongs to.  If that mapping has special
1497  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1498  * we may need to copy to a suitable page before moving to filecache.
1499  *
1500  * In a future release, this may well be extended to respect cpuset and
1501  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1502  * but for now it is a simple matter of zone.
1503  */
1504 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1505 {
1506 	return page_zonenum(page) > gfp_zone(gfp);
1507 }
1508 
1509 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1510 				struct shmem_inode_info *info, pgoff_t index)
1511 {
1512 	struct page *oldpage, *newpage;
1513 	struct address_space *swap_mapping;
1514 	swp_entry_t entry;
1515 	pgoff_t swap_index;
1516 	int error;
1517 
1518 	oldpage = *pagep;
1519 	entry.val = page_private(oldpage);
1520 	swap_index = swp_offset(entry);
1521 	swap_mapping = page_mapping(oldpage);
1522 
1523 	/*
1524 	 * We have arrived here because our zones are constrained, so don't
1525 	 * limit chance of success by further cpuset and node constraints.
1526 	 */
1527 	gfp &= ~GFP_CONSTRAINT_MASK;
1528 	newpage = shmem_alloc_page(gfp, info, index);
1529 	if (!newpage)
1530 		return -ENOMEM;
1531 
1532 	get_page(newpage);
1533 	copy_highpage(newpage, oldpage);
1534 	flush_dcache_page(newpage);
1535 
1536 	__SetPageLocked(newpage);
1537 	__SetPageSwapBacked(newpage);
1538 	SetPageUptodate(newpage);
1539 	set_page_private(newpage, entry.val);
1540 	SetPageSwapCache(newpage);
1541 
1542 	/*
1543 	 * Our caller will very soon move newpage out of swapcache, but it's
1544 	 * a nice clean interface for us to replace oldpage by newpage there.
1545 	 */
1546 	xa_lock_irq(&swap_mapping->i_pages);
1547 	error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1548 	if (!error) {
1549 		__inc_node_page_state(newpage, NR_FILE_PAGES);
1550 		__dec_node_page_state(oldpage, NR_FILE_PAGES);
1551 	}
1552 	xa_unlock_irq(&swap_mapping->i_pages);
1553 
1554 	if (unlikely(error)) {
1555 		/*
1556 		 * Is this possible?  I think not, now that our callers check
1557 		 * both PageSwapCache and page_private after getting page lock;
1558 		 * but be defensive.  Reverse old to newpage for clear and free.
1559 		 */
1560 		oldpage = newpage;
1561 	} else {
1562 		mem_cgroup_migrate(oldpage, newpage);
1563 		lru_cache_add_anon(newpage);
1564 		*pagep = newpage;
1565 	}
1566 
1567 	ClearPageSwapCache(oldpage);
1568 	set_page_private(oldpage, 0);
1569 
1570 	unlock_page(oldpage);
1571 	put_page(oldpage);
1572 	put_page(oldpage);
1573 	return error;
1574 }
1575 
1576 /*
1577  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1578  *
1579  * If we allocate a new one we do not mark it dirty. That's up to the
1580  * vm. If we swap it in we mark it dirty since we also free the swap
1581  * entry since a page cannot live in both the swap and page cache.
1582  *
1583  * fault_mm and fault_type are only supplied by shmem_fault:
1584  * otherwise they are NULL.
1585  */
1586 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1587 	struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1588 	struct vm_area_struct *vma, struct vm_fault *vmf,
1589 			vm_fault_t *fault_type)
1590 {
1591 	struct address_space *mapping = inode->i_mapping;
1592 	struct shmem_inode_info *info = SHMEM_I(inode);
1593 	struct shmem_sb_info *sbinfo;
1594 	struct mm_struct *charge_mm;
1595 	struct mem_cgroup *memcg;
1596 	struct page *page;
1597 	swp_entry_t swap;
1598 	enum sgp_type sgp_huge = sgp;
1599 	pgoff_t hindex = index;
1600 	int error;
1601 	int once = 0;
1602 	int alloced = 0;
1603 
1604 	if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1605 		return -EFBIG;
1606 	if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1607 		sgp = SGP_CACHE;
1608 repeat:
1609 	swap.val = 0;
1610 	page = find_lock_entry(mapping, index);
1611 	if (xa_is_value(page)) {
1612 		swap = radix_to_swp_entry(page);
1613 		page = NULL;
1614 	}
1615 
1616 	if (sgp <= SGP_CACHE &&
1617 	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1618 		error = -EINVAL;
1619 		goto unlock;
1620 	}
1621 
1622 	if (page && sgp == SGP_WRITE)
1623 		mark_page_accessed(page);
1624 
1625 	/* fallocated page? */
1626 	if (page && !PageUptodate(page)) {
1627 		if (sgp != SGP_READ)
1628 			goto clear;
1629 		unlock_page(page);
1630 		put_page(page);
1631 		page = NULL;
1632 	}
1633 	if (page || (sgp == SGP_READ && !swap.val)) {
1634 		*pagep = page;
1635 		return 0;
1636 	}
1637 
1638 	/*
1639 	 * Fast cache lookup did not find it:
1640 	 * bring it back from swap or allocate.
1641 	 */
1642 	sbinfo = SHMEM_SB(inode->i_sb);
1643 	charge_mm = vma ? vma->vm_mm : current->mm;
1644 
1645 	if (swap.val) {
1646 		/* Look it up and read it in.. */
1647 		page = lookup_swap_cache(swap, NULL, 0);
1648 		if (!page) {
1649 			/* Or update major stats only when swapin succeeds?? */
1650 			if (fault_type) {
1651 				*fault_type |= VM_FAULT_MAJOR;
1652 				count_vm_event(PGMAJFAULT);
1653 				count_memcg_event_mm(charge_mm, PGMAJFAULT);
1654 			}
1655 			/* Here we actually start the io */
1656 			page = shmem_swapin(swap, gfp, info, index);
1657 			if (!page) {
1658 				error = -ENOMEM;
1659 				goto failed;
1660 			}
1661 		}
1662 
1663 		/* We have to do this with page locked to prevent races */
1664 		lock_page(page);
1665 		if (!PageSwapCache(page) || page_private(page) != swap.val ||
1666 		    !shmem_confirm_swap(mapping, index, swap)) {
1667 			error = -EEXIST;	/* try again */
1668 			goto unlock;
1669 		}
1670 		if (!PageUptodate(page)) {
1671 			error = -EIO;
1672 			goto failed;
1673 		}
1674 		wait_on_page_writeback(page);
1675 
1676 		if (shmem_should_replace_page(page, gfp)) {
1677 			error = shmem_replace_page(&page, gfp, info, index);
1678 			if (error)
1679 				goto failed;
1680 		}
1681 
1682 		error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1683 				false);
1684 		if (!error) {
1685 			error = shmem_add_to_page_cache(page, mapping, index,
1686 						swp_to_radix_entry(swap), gfp);
1687 			/*
1688 			 * We already confirmed swap under page lock, and make
1689 			 * no memory allocation here, so usually no possibility
1690 			 * of error; but free_swap_and_cache() only trylocks a
1691 			 * page, so it is just possible that the entry has been
1692 			 * truncated or holepunched since swap was confirmed.
1693 			 * shmem_undo_range() will have done some of the
1694 			 * unaccounting, now delete_from_swap_cache() will do
1695 			 * the rest.
1696 			 * Reset swap.val? No, leave it so "failed" goes back to
1697 			 * "repeat": reading a hole and writing should succeed.
1698 			 */
1699 			if (error) {
1700 				mem_cgroup_cancel_charge(page, memcg, false);
1701 				delete_from_swap_cache(page);
1702 			}
1703 		}
1704 		if (error)
1705 			goto failed;
1706 
1707 		mem_cgroup_commit_charge(page, memcg, true, false);
1708 
1709 		spin_lock_irq(&info->lock);
1710 		info->swapped--;
1711 		shmem_recalc_inode(inode);
1712 		spin_unlock_irq(&info->lock);
1713 
1714 		if (sgp == SGP_WRITE)
1715 			mark_page_accessed(page);
1716 
1717 		delete_from_swap_cache(page);
1718 		set_page_dirty(page);
1719 		swap_free(swap);
1720 
1721 	} else {
1722 		if (vma && userfaultfd_missing(vma)) {
1723 			*fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1724 			return 0;
1725 		}
1726 
1727 		/* shmem_symlink() */
1728 		if (mapping->a_ops != &shmem_aops)
1729 			goto alloc_nohuge;
1730 		if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1731 			goto alloc_nohuge;
1732 		if (shmem_huge == SHMEM_HUGE_FORCE)
1733 			goto alloc_huge;
1734 		switch (sbinfo->huge) {
1735 			loff_t i_size;
1736 			pgoff_t off;
1737 		case SHMEM_HUGE_NEVER:
1738 			goto alloc_nohuge;
1739 		case SHMEM_HUGE_WITHIN_SIZE:
1740 			off = round_up(index, HPAGE_PMD_NR);
1741 			i_size = round_up(i_size_read(inode), PAGE_SIZE);
1742 			if (i_size >= HPAGE_PMD_SIZE &&
1743 					i_size >> PAGE_SHIFT >= off)
1744 				goto alloc_huge;
1745 			/* fallthrough */
1746 		case SHMEM_HUGE_ADVISE:
1747 			if (sgp_huge == SGP_HUGE)
1748 				goto alloc_huge;
1749 			/* TODO: implement fadvise() hints */
1750 			goto alloc_nohuge;
1751 		}
1752 
1753 alloc_huge:
1754 		page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1755 		if (IS_ERR(page)) {
1756 alloc_nohuge:		page = shmem_alloc_and_acct_page(gfp, inode,
1757 					index, false);
1758 		}
1759 		if (IS_ERR(page)) {
1760 			int retry = 5;
1761 			error = PTR_ERR(page);
1762 			page = NULL;
1763 			if (error != -ENOSPC)
1764 				goto failed;
1765 			/*
1766 			 * Try to reclaim some spece by splitting a huge page
1767 			 * beyond i_size on the filesystem.
1768 			 */
1769 			while (retry--) {
1770 				int ret;
1771 				ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1772 				if (ret == SHRINK_STOP)
1773 					break;
1774 				if (ret)
1775 					goto alloc_nohuge;
1776 			}
1777 			goto failed;
1778 		}
1779 
1780 		if (PageTransHuge(page))
1781 			hindex = round_down(index, HPAGE_PMD_NR);
1782 		else
1783 			hindex = index;
1784 
1785 		if (sgp == SGP_WRITE)
1786 			__SetPageReferenced(page);
1787 
1788 		error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1789 				PageTransHuge(page));
1790 		if (error)
1791 			goto unacct;
1792 		error = shmem_add_to_page_cache(page, mapping, hindex,
1793 						NULL, gfp & GFP_RECLAIM_MASK);
1794 		if (error) {
1795 			mem_cgroup_cancel_charge(page, memcg,
1796 					PageTransHuge(page));
1797 			goto unacct;
1798 		}
1799 		mem_cgroup_commit_charge(page, memcg, false,
1800 				PageTransHuge(page));
1801 		lru_cache_add_anon(page);
1802 
1803 		spin_lock_irq(&info->lock);
1804 		info->alloced += 1 << compound_order(page);
1805 		inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1806 		shmem_recalc_inode(inode);
1807 		spin_unlock_irq(&info->lock);
1808 		alloced = true;
1809 
1810 		if (PageTransHuge(page) &&
1811 				DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1812 				hindex + HPAGE_PMD_NR - 1) {
1813 			/*
1814 			 * Part of the huge page is beyond i_size: subject
1815 			 * to shrink under memory pressure.
1816 			 */
1817 			spin_lock(&sbinfo->shrinklist_lock);
1818 			/*
1819 			 * _careful to defend against unlocked access to
1820 			 * ->shrink_list in shmem_unused_huge_shrink()
1821 			 */
1822 			if (list_empty_careful(&info->shrinklist)) {
1823 				list_add_tail(&info->shrinklist,
1824 						&sbinfo->shrinklist);
1825 				sbinfo->shrinklist_len++;
1826 			}
1827 			spin_unlock(&sbinfo->shrinklist_lock);
1828 		}
1829 
1830 		/*
1831 		 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1832 		 */
1833 		if (sgp == SGP_FALLOC)
1834 			sgp = SGP_WRITE;
1835 clear:
1836 		/*
1837 		 * Let SGP_WRITE caller clear ends if write does not fill page;
1838 		 * but SGP_FALLOC on a page fallocated earlier must initialize
1839 		 * it now, lest undo on failure cancel our earlier guarantee.
1840 		 */
1841 		if (sgp != SGP_WRITE && !PageUptodate(page)) {
1842 			struct page *head = compound_head(page);
1843 			int i;
1844 
1845 			for (i = 0; i < (1 << compound_order(head)); i++) {
1846 				clear_highpage(head + i);
1847 				flush_dcache_page(head + i);
1848 			}
1849 			SetPageUptodate(head);
1850 		}
1851 	}
1852 
1853 	/* Perhaps the file has been truncated since we checked */
1854 	if (sgp <= SGP_CACHE &&
1855 	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1856 		if (alloced) {
1857 			ClearPageDirty(page);
1858 			delete_from_page_cache(page);
1859 			spin_lock_irq(&info->lock);
1860 			shmem_recalc_inode(inode);
1861 			spin_unlock_irq(&info->lock);
1862 		}
1863 		error = -EINVAL;
1864 		goto unlock;
1865 	}
1866 	*pagep = page + index - hindex;
1867 	return 0;
1868 
1869 	/*
1870 	 * Error recovery.
1871 	 */
1872 unacct:
1873 	shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
1874 
1875 	if (PageTransHuge(page)) {
1876 		unlock_page(page);
1877 		put_page(page);
1878 		goto alloc_nohuge;
1879 	}
1880 failed:
1881 	if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1882 		error = -EEXIST;
1883 unlock:
1884 	if (page) {
1885 		unlock_page(page);
1886 		put_page(page);
1887 	}
1888 	if (error == -ENOSPC && !once++) {
1889 		spin_lock_irq(&info->lock);
1890 		shmem_recalc_inode(inode);
1891 		spin_unlock_irq(&info->lock);
1892 		goto repeat;
1893 	}
1894 	if (error == -EEXIST)
1895 		goto repeat;
1896 	return error;
1897 }
1898 
1899 /*
1900  * This is like autoremove_wake_function, but it removes the wait queue
1901  * entry unconditionally - even if something else had already woken the
1902  * target.
1903  */
1904 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1905 {
1906 	int ret = default_wake_function(wait, mode, sync, key);
1907 	list_del_init(&wait->entry);
1908 	return ret;
1909 }
1910 
1911 static vm_fault_t shmem_fault(struct vm_fault *vmf)
1912 {
1913 	struct vm_area_struct *vma = vmf->vma;
1914 	struct inode *inode = file_inode(vma->vm_file);
1915 	gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1916 	enum sgp_type sgp;
1917 	int err;
1918 	vm_fault_t ret = VM_FAULT_LOCKED;
1919 
1920 	/*
1921 	 * Trinity finds that probing a hole which tmpfs is punching can
1922 	 * prevent the hole-punch from ever completing: which in turn
1923 	 * locks writers out with its hold on i_mutex.  So refrain from
1924 	 * faulting pages into the hole while it's being punched.  Although
1925 	 * shmem_undo_range() does remove the additions, it may be unable to
1926 	 * keep up, as each new page needs its own unmap_mapping_range() call,
1927 	 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1928 	 *
1929 	 * It does not matter if we sometimes reach this check just before the
1930 	 * hole-punch begins, so that one fault then races with the punch:
1931 	 * we just need to make racing faults a rare case.
1932 	 *
1933 	 * The implementation below would be much simpler if we just used a
1934 	 * standard mutex or completion: but we cannot take i_mutex in fault,
1935 	 * and bloating every shmem inode for this unlikely case would be sad.
1936 	 */
1937 	if (unlikely(inode->i_private)) {
1938 		struct shmem_falloc *shmem_falloc;
1939 
1940 		spin_lock(&inode->i_lock);
1941 		shmem_falloc = inode->i_private;
1942 		if (shmem_falloc &&
1943 		    shmem_falloc->waitq &&
1944 		    vmf->pgoff >= shmem_falloc->start &&
1945 		    vmf->pgoff < shmem_falloc->next) {
1946 			wait_queue_head_t *shmem_falloc_waitq;
1947 			DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1948 
1949 			ret = VM_FAULT_NOPAGE;
1950 			if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1951 			   !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1952 				/* It's polite to up mmap_sem if we can */
1953 				up_read(&vma->vm_mm->mmap_sem);
1954 				ret = VM_FAULT_RETRY;
1955 			}
1956 
1957 			shmem_falloc_waitq = shmem_falloc->waitq;
1958 			prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1959 					TASK_UNINTERRUPTIBLE);
1960 			spin_unlock(&inode->i_lock);
1961 			schedule();
1962 
1963 			/*
1964 			 * shmem_falloc_waitq points into the shmem_fallocate()
1965 			 * stack of the hole-punching task: shmem_falloc_waitq
1966 			 * is usually invalid by the time we reach here, but
1967 			 * finish_wait() does not dereference it in that case;
1968 			 * though i_lock needed lest racing with wake_up_all().
1969 			 */
1970 			spin_lock(&inode->i_lock);
1971 			finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1972 			spin_unlock(&inode->i_lock);
1973 			return ret;
1974 		}
1975 		spin_unlock(&inode->i_lock);
1976 	}
1977 
1978 	sgp = SGP_CACHE;
1979 
1980 	if ((vma->vm_flags & VM_NOHUGEPAGE) ||
1981 	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
1982 		sgp = SGP_NOHUGE;
1983 	else if (vma->vm_flags & VM_HUGEPAGE)
1984 		sgp = SGP_HUGE;
1985 
1986 	err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1987 				  gfp, vma, vmf, &ret);
1988 	if (err)
1989 		return vmf_error(err);
1990 	return ret;
1991 }
1992 
1993 unsigned long shmem_get_unmapped_area(struct file *file,
1994 				      unsigned long uaddr, unsigned long len,
1995 				      unsigned long pgoff, unsigned long flags)
1996 {
1997 	unsigned long (*get_area)(struct file *,
1998 		unsigned long, unsigned long, unsigned long, unsigned long);
1999 	unsigned long addr;
2000 	unsigned long offset;
2001 	unsigned long inflated_len;
2002 	unsigned long inflated_addr;
2003 	unsigned long inflated_offset;
2004 
2005 	if (len > TASK_SIZE)
2006 		return -ENOMEM;
2007 
2008 	get_area = current->mm->get_unmapped_area;
2009 	addr = get_area(file, uaddr, len, pgoff, flags);
2010 
2011 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2012 		return addr;
2013 	if (IS_ERR_VALUE(addr))
2014 		return addr;
2015 	if (addr & ~PAGE_MASK)
2016 		return addr;
2017 	if (addr > TASK_SIZE - len)
2018 		return addr;
2019 
2020 	if (shmem_huge == SHMEM_HUGE_DENY)
2021 		return addr;
2022 	if (len < HPAGE_PMD_SIZE)
2023 		return addr;
2024 	if (flags & MAP_FIXED)
2025 		return addr;
2026 	/*
2027 	 * Our priority is to support MAP_SHARED mapped hugely;
2028 	 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2029 	 * But if caller specified an address hint, respect that as before.
2030 	 */
2031 	if (uaddr)
2032 		return addr;
2033 
2034 	if (shmem_huge != SHMEM_HUGE_FORCE) {
2035 		struct super_block *sb;
2036 
2037 		if (file) {
2038 			VM_BUG_ON(file->f_op != &shmem_file_operations);
2039 			sb = file_inode(file)->i_sb;
2040 		} else {
2041 			/*
2042 			 * Called directly from mm/mmap.c, or drivers/char/mem.c
2043 			 * for "/dev/zero", to create a shared anonymous object.
2044 			 */
2045 			if (IS_ERR(shm_mnt))
2046 				return addr;
2047 			sb = shm_mnt->mnt_sb;
2048 		}
2049 		if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2050 			return addr;
2051 	}
2052 
2053 	offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2054 	if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2055 		return addr;
2056 	if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2057 		return addr;
2058 
2059 	inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2060 	if (inflated_len > TASK_SIZE)
2061 		return addr;
2062 	if (inflated_len < len)
2063 		return addr;
2064 
2065 	inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2066 	if (IS_ERR_VALUE(inflated_addr))
2067 		return addr;
2068 	if (inflated_addr & ~PAGE_MASK)
2069 		return addr;
2070 
2071 	inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2072 	inflated_addr += offset - inflated_offset;
2073 	if (inflated_offset > offset)
2074 		inflated_addr += HPAGE_PMD_SIZE;
2075 
2076 	if (inflated_addr > TASK_SIZE - len)
2077 		return addr;
2078 	return inflated_addr;
2079 }
2080 
2081 #ifdef CONFIG_NUMA
2082 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2083 {
2084 	struct inode *inode = file_inode(vma->vm_file);
2085 	return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2086 }
2087 
2088 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2089 					  unsigned long addr)
2090 {
2091 	struct inode *inode = file_inode(vma->vm_file);
2092 	pgoff_t index;
2093 
2094 	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2095 	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2096 }
2097 #endif
2098 
2099 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2100 {
2101 	struct inode *inode = file_inode(file);
2102 	struct shmem_inode_info *info = SHMEM_I(inode);
2103 	int retval = -ENOMEM;
2104 
2105 	spin_lock_irq(&info->lock);
2106 	if (lock && !(info->flags & VM_LOCKED)) {
2107 		if (!user_shm_lock(inode->i_size, user))
2108 			goto out_nomem;
2109 		info->flags |= VM_LOCKED;
2110 		mapping_set_unevictable(file->f_mapping);
2111 	}
2112 	if (!lock && (info->flags & VM_LOCKED) && user) {
2113 		user_shm_unlock(inode->i_size, user);
2114 		info->flags &= ~VM_LOCKED;
2115 		mapping_clear_unevictable(file->f_mapping);
2116 	}
2117 	retval = 0;
2118 
2119 out_nomem:
2120 	spin_unlock_irq(&info->lock);
2121 	return retval;
2122 }
2123 
2124 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2125 {
2126 	file_accessed(file);
2127 	vma->vm_ops = &shmem_vm_ops;
2128 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2129 			((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2130 			(vma->vm_end & HPAGE_PMD_MASK)) {
2131 		khugepaged_enter(vma, vma->vm_flags);
2132 	}
2133 	return 0;
2134 }
2135 
2136 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2137 				     umode_t mode, dev_t dev, unsigned long flags)
2138 {
2139 	struct inode *inode;
2140 	struct shmem_inode_info *info;
2141 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2142 
2143 	if (shmem_reserve_inode(sb))
2144 		return NULL;
2145 
2146 	inode = new_inode(sb);
2147 	if (inode) {
2148 		inode->i_ino = get_next_ino();
2149 		inode_init_owner(inode, dir, mode);
2150 		inode->i_blocks = 0;
2151 		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2152 		inode->i_generation = prandom_u32();
2153 		info = SHMEM_I(inode);
2154 		memset(info, 0, (char *)inode - (char *)info);
2155 		spin_lock_init(&info->lock);
2156 		info->seals = F_SEAL_SEAL;
2157 		info->flags = flags & VM_NORESERVE;
2158 		INIT_LIST_HEAD(&info->shrinklist);
2159 		INIT_LIST_HEAD(&info->swaplist);
2160 		simple_xattrs_init(&info->xattrs);
2161 		cache_no_acl(inode);
2162 
2163 		switch (mode & S_IFMT) {
2164 		default:
2165 			inode->i_op = &shmem_special_inode_operations;
2166 			init_special_inode(inode, mode, dev);
2167 			break;
2168 		case S_IFREG:
2169 			inode->i_mapping->a_ops = &shmem_aops;
2170 			inode->i_op = &shmem_inode_operations;
2171 			inode->i_fop = &shmem_file_operations;
2172 			mpol_shared_policy_init(&info->policy,
2173 						 shmem_get_sbmpol(sbinfo));
2174 			break;
2175 		case S_IFDIR:
2176 			inc_nlink(inode);
2177 			/* Some things misbehave if size == 0 on a directory */
2178 			inode->i_size = 2 * BOGO_DIRENT_SIZE;
2179 			inode->i_op = &shmem_dir_inode_operations;
2180 			inode->i_fop = &simple_dir_operations;
2181 			break;
2182 		case S_IFLNK:
2183 			/*
2184 			 * Must not load anything in the rbtree,
2185 			 * mpol_free_shared_policy will not be called.
2186 			 */
2187 			mpol_shared_policy_init(&info->policy, NULL);
2188 			break;
2189 		}
2190 
2191 		lockdep_annotate_inode_mutex_key(inode);
2192 	} else
2193 		shmem_free_inode(sb);
2194 	return inode;
2195 }
2196 
2197 bool shmem_mapping(struct address_space *mapping)
2198 {
2199 	return mapping->a_ops == &shmem_aops;
2200 }
2201 
2202 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2203 				  pmd_t *dst_pmd,
2204 				  struct vm_area_struct *dst_vma,
2205 				  unsigned long dst_addr,
2206 				  unsigned long src_addr,
2207 				  bool zeropage,
2208 				  struct page **pagep)
2209 {
2210 	struct inode *inode = file_inode(dst_vma->vm_file);
2211 	struct shmem_inode_info *info = SHMEM_I(inode);
2212 	struct address_space *mapping = inode->i_mapping;
2213 	gfp_t gfp = mapping_gfp_mask(mapping);
2214 	pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2215 	struct mem_cgroup *memcg;
2216 	spinlock_t *ptl;
2217 	void *page_kaddr;
2218 	struct page *page;
2219 	pte_t _dst_pte, *dst_pte;
2220 	int ret;
2221 	pgoff_t offset, max_off;
2222 
2223 	ret = -ENOMEM;
2224 	if (!shmem_inode_acct_block(inode, 1))
2225 		goto out;
2226 
2227 	if (!*pagep) {
2228 		page = shmem_alloc_page(gfp, info, pgoff);
2229 		if (!page)
2230 			goto out_unacct_blocks;
2231 
2232 		if (!zeropage) {	/* mcopy_atomic */
2233 			page_kaddr = kmap_atomic(page);
2234 			ret = copy_from_user(page_kaddr,
2235 					     (const void __user *)src_addr,
2236 					     PAGE_SIZE);
2237 			kunmap_atomic(page_kaddr);
2238 
2239 			/* fallback to copy_from_user outside mmap_sem */
2240 			if (unlikely(ret)) {
2241 				*pagep = page;
2242 				shmem_inode_unacct_blocks(inode, 1);
2243 				/* don't free the page */
2244 				return -ENOENT;
2245 			}
2246 		} else {		/* mfill_zeropage_atomic */
2247 			clear_highpage(page);
2248 		}
2249 	} else {
2250 		page = *pagep;
2251 		*pagep = NULL;
2252 	}
2253 
2254 	VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2255 	__SetPageLocked(page);
2256 	__SetPageSwapBacked(page);
2257 	__SetPageUptodate(page);
2258 
2259 	ret = -EFAULT;
2260 	offset = linear_page_index(dst_vma, dst_addr);
2261 	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2262 	if (unlikely(offset >= max_off))
2263 		goto out_release;
2264 
2265 	ret = mem_cgroup_try_charge_delay(page, dst_mm, gfp, &memcg, false);
2266 	if (ret)
2267 		goto out_release;
2268 
2269 	ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2270 						gfp & GFP_RECLAIM_MASK);
2271 	if (ret)
2272 		goto out_release_uncharge;
2273 
2274 	mem_cgroup_commit_charge(page, memcg, false, false);
2275 
2276 	_dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2277 	if (dst_vma->vm_flags & VM_WRITE)
2278 		_dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2279 	else {
2280 		/*
2281 		 * We don't set the pte dirty if the vma has no
2282 		 * VM_WRITE permission, so mark the page dirty or it
2283 		 * could be freed from under us. We could do it
2284 		 * unconditionally before unlock_page(), but doing it
2285 		 * only if VM_WRITE is not set is faster.
2286 		 */
2287 		set_page_dirty(page);
2288 	}
2289 
2290 	dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2291 
2292 	ret = -EFAULT;
2293 	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2294 	if (unlikely(offset >= max_off))
2295 		goto out_release_uncharge_unlock;
2296 
2297 	ret = -EEXIST;
2298 	if (!pte_none(*dst_pte))
2299 		goto out_release_uncharge_unlock;
2300 
2301 	lru_cache_add_anon(page);
2302 
2303 	spin_lock(&info->lock);
2304 	info->alloced++;
2305 	inode->i_blocks += BLOCKS_PER_PAGE;
2306 	shmem_recalc_inode(inode);
2307 	spin_unlock(&info->lock);
2308 
2309 	inc_mm_counter(dst_mm, mm_counter_file(page));
2310 	page_add_file_rmap(page, false);
2311 	set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2312 
2313 	/* No need to invalidate - it was non-present before */
2314 	update_mmu_cache(dst_vma, dst_addr, dst_pte);
2315 	pte_unmap_unlock(dst_pte, ptl);
2316 	unlock_page(page);
2317 	ret = 0;
2318 out:
2319 	return ret;
2320 out_release_uncharge_unlock:
2321 	pte_unmap_unlock(dst_pte, ptl);
2322 	ClearPageDirty(page);
2323 	delete_from_page_cache(page);
2324 out_release_uncharge:
2325 	mem_cgroup_cancel_charge(page, memcg, false);
2326 out_release:
2327 	unlock_page(page);
2328 	put_page(page);
2329 out_unacct_blocks:
2330 	shmem_inode_unacct_blocks(inode, 1);
2331 	goto out;
2332 }
2333 
2334 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2335 			   pmd_t *dst_pmd,
2336 			   struct vm_area_struct *dst_vma,
2337 			   unsigned long dst_addr,
2338 			   unsigned long src_addr,
2339 			   struct page **pagep)
2340 {
2341 	return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2342 				      dst_addr, src_addr, false, pagep);
2343 }
2344 
2345 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2346 			     pmd_t *dst_pmd,
2347 			     struct vm_area_struct *dst_vma,
2348 			     unsigned long dst_addr)
2349 {
2350 	struct page *page = NULL;
2351 
2352 	return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2353 				      dst_addr, 0, true, &page);
2354 }
2355 
2356 #ifdef CONFIG_TMPFS
2357 static const struct inode_operations shmem_symlink_inode_operations;
2358 static const struct inode_operations shmem_short_symlink_operations;
2359 
2360 #ifdef CONFIG_TMPFS_XATTR
2361 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2362 #else
2363 #define shmem_initxattrs NULL
2364 #endif
2365 
2366 static int
2367 shmem_write_begin(struct file *file, struct address_space *mapping,
2368 			loff_t pos, unsigned len, unsigned flags,
2369 			struct page **pagep, void **fsdata)
2370 {
2371 	struct inode *inode = mapping->host;
2372 	struct shmem_inode_info *info = SHMEM_I(inode);
2373 	pgoff_t index = pos >> PAGE_SHIFT;
2374 
2375 	/* i_mutex is held by caller */
2376 	if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) {
2377 		if (info->seals & F_SEAL_WRITE)
2378 			return -EPERM;
2379 		if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2380 			return -EPERM;
2381 	}
2382 
2383 	return shmem_getpage(inode, index, pagep, SGP_WRITE);
2384 }
2385 
2386 static int
2387 shmem_write_end(struct file *file, struct address_space *mapping,
2388 			loff_t pos, unsigned len, unsigned copied,
2389 			struct page *page, void *fsdata)
2390 {
2391 	struct inode *inode = mapping->host;
2392 
2393 	if (pos + copied > inode->i_size)
2394 		i_size_write(inode, pos + copied);
2395 
2396 	if (!PageUptodate(page)) {
2397 		struct page *head = compound_head(page);
2398 		if (PageTransCompound(page)) {
2399 			int i;
2400 
2401 			for (i = 0; i < HPAGE_PMD_NR; i++) {
2402 				if (head + i == page)
2403 					continue;
2404 				clear_highpage(head + i);
2405 				flush_dcache_page(head + i);
2406 			}
2407 		}
2408 		if (copied < PAGE_SIZE) {
2409 			unsigned from = pos & (PAGE_SIZE - 1);
2410 			zero_user_segments(page, 0, from,
2411 					from + copied, PAGE_SIZE);
2412 		}
2413 		SetPageUptodate(head);
2414 	}
2415 	set_page_dirty(page);
2416 	unlock_page(page);
2417 	put_page(page);
2418 
2419 	return copied;
2420 }
2421 
2422 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2423 {
2424 	struct file *file = iocb->ki_filp;
2425 	struct inode *inode = file_inode(file);
2426 	struct address_space *mapping = inode->i_mapping;
2427 	pgoff_t index;
2428 	unsigned long offset;
2429 	enum sgp_type sgp = SGP_READ;
2430 	int error = 0;
2431 	ssize_t retval = 0;
2432 	loff_t *ppos = &iocb->ki_pos;
2433 
2434 	/*
2435 	 * Might this read be for a stacking filesystem?  Then when reading
2436 	 * holes of a sparse file, we actually need to allocate those pages,
2437 	 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2438 	 */
2439 	if (!iter_is_iovec(to))
2440 		sgp = SGP_CACHE;
2441 
2442 	index = *ppos >> PAGE_SHIFT;
2443 	offset = *ppos & ~PAGE_MASK;
2444 
2445 	for (;;) {
2446 		struct page *page = NULL;
2447 		pgoff_t end_index;
2448 		unsigned long nr, ret;
2449 		loff_t i_size = i_size_read(inode);
2450 
2451 		end_index = i_size >> PAGE_SHIFT;
2452 		if (index > end_index)
2453 			break;
2454 		if (index == end_index) {
2455 			nr = i_size & ~PAGE_MASK;
2456 			if (nr <= offset)
2457 				break;
2458 		}
2459 
2460 		error = shmem_getpage(inode, index, &page, sgp);
2461 		if (error) {
2462 			if (error == -EINVAL)
2463 				error = 0;
2464 			break;
2465 		}
2466 		if (page) {
2467 			if (sgp == SGP_CACHE)
2468 				set_page_dirty(page);
2469 			unlock_page(page);
2470 		}
2471 
2472 		/*
2473 		 * We must evaluate after, since reads (unlike writes)
2474 		 * are called without i_mutex protection against truncate
2475 		 */
2476 		nr = PAGE_SIZE;
2477 		i_size = i_size_read(inode);
2478 		end_index = i_size >> PAGE_SHIFT;
2479 		if (index == end_index) {
2480 			nr = i_size & ~PAGE_MASK;
2481 			if (nr <= offset) {
2482 				if (page)
2483 					put_page(page);
2484 				break;
2485 			}
2486 		}
2487 		nr -= offset;
2488 
2489 		if (page) {
2490 			/*
2491 			 * If users can be writing to this page using arbitrary
2492 			 * virtual addresses, take care about potential aliasing
2493 			 * before reading the page on the kernel side.
2494 			 */
2495 			if (mapping_writably_mapped(mapping))
2496 				flush_dcache_page(page);
2497 			/*
2498 			 * Mark the page accessed if we read the beginning.
2499 			 */
2500 			if (!offset)
2501 				mark_page_accessed(page);
2502 		} else {
2503 			page = ZERO_PAGE(0);
2504 			get_page(page);
2505 		}
2506 
2507 		/*
2508 		 * Ok, we have the page, and it's up-to-date, so
2509 		 * now we can copy it to user space...
2510 		 */
2511 		ret = copy_page_to_iter(page, offset, nr, to);
2512 		retval += ret;
2513 		offset += ret;
2514 		index += offset >> PAGE_SHIFT;
2515 		offset &= ~PAGE_MASK;
2516 
2517 		put_page(page);
2518 		if (!iov_iter_count(to))
2519 			break;
2520 		if (ret < nr) {
2521 			error = -EFAULT;
2522 			break;
2523 		}
2524 		cond_resched();
2525 	}
2526 
2527 	*ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2528 	file_accessed(file);
2529 	return retval ? retval : error;
2530 }
2531 
2532 /*
2533  * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2534  */
2535 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2536 				    pgoff_t index, pgoff_t end, int whence)
2537 {
2538 	struct page *page;
2539 	struct pagevec pvec;
2540 	pgoff_t indices[PAGEVEC_SIZE];
2541 	bool done = false;
2542 	int i;
2543 
2544 	pagevec_init(&pvec);
2545 	pvec.nr = 1;		/* start small: we may be there already */
2546 	while (!done) {
2547 		pvec.nr = find_get_entries(mapping, index,
2548 					pvec.nr, pvec.pages, indices);
2549 		if (!pvec.nr) {
2550 			if (whence == SEEK_DATA)
2551 				index = end;
2552 			break;
2553 		}
2554 		for (i = 0; i < pvec.nr; i++, index++) {
2555 			if (index < indices[i]) {
2556 				if (whence == SEEK_HOLE) {
2557 					done = true;
2558 					break;
2559 				}
2560 				index = indices[i];
2561 			}
2562 			page = pvec.pages[i];
2563 			if (page && !xa_is_value(page)) {
2564 				if (!PageUptodate(page))
2565 					page = NULL;
2566 			}
2567 			if (index >= end ||
2568 			    (page && whence == SEEK_DATA) ||
2569 			    (!page && whence == SEEK_HOLE)) {
2570 				done = true;
2571 				break;
2572 			}
2573 		}
2574 		pagevec_remove_exceptionals(&pvec);
2575 		pagevec_release(&pvec);
2576 		pvec.nr = PAGEVEC_SIZE;
2577 		cond_resched();
2578 	}
2579 	return index;
2580 }
2581 
2582 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2583 {
2584 	struct address_space *mapping = file->f_mapping;
2585 	struct inode *inode = mapping->host;
2586 	pgoff_t start, end;
2587 	loff_t new_offset;
2588 
2589 	if (whence != SEEK_DATA && whence != SEEK_HOLE)
2590 		return generic_file_llseek_size(file, offset, whence,
2591 					MAX_LFS_FILESIZE, i_size_read(inode));
2592 	inode_lock(inode);
2593 	/* We're holding i_mutex so we can access i_size directly */
2594 
2595 	if (offset < 0 || offset >= inode->i_size)
2596 		offset = -ENXIO;
2597 	else {
2598 		start = offset >> PAGE_SHIFT;
2599 		end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2600 		new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2601 		new_offset <<= PAGE_SHIFT;
2602 		if (new_offset > offset) {
2603 			if (new_offset < inode->i_size)
2604 				offset = new_offset;
2605 			else if (whence == SEEK_DATA)
2606 				offset = -ENXIO;
2607 			else
2608 				offset = inode->i_size;
2609 		}
2610 	}
2611 
2612 	if (offset >= 0)
2613 		offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2614 	inode_unlock(inode);
2615 	return offset;
2616 }
2617 
2618 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2619 							 loff_t len)
2620 {
2621 	struct inode *inode = file_inode(file);
2622 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2623 	struct shmem_inode_info *info = SHMEM_I(inode);
2624 	struct shmem_falloc shmem_falloc;
2625 	pgoff_t start, index, end;
2626 	int error;
2627 
2628 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2629 		return -EOPNOTSUPP;
2630 
2631 	inode_lock(inode);
2632 
2633 	if (mode & FALLOC_FL_PUNCH_HOLE) {
2634 		struct address_space *mapping = file->f_mapping;
2635 		loff_t unmap_start = round_up(offset, PAGE_SIZE);
2636 		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2637 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2638 
2639 		/* protected by i_mutex */
2640 		if (info->seals & F_SEAL_WRITE) {
2641 			error = -EPERM;
2642 			goto out;
2643 		}
2644 
2645 		shmem_falloc.waitq = &shmem_falloc_waitq;
2646 		shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2647 		shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2648 		spin_lock(&inode->i_lock);
2649 		inode->i_private = &shmem_falloc;
2650 		spin_unlock(&inode->i_lock);
2651 
2652 		if ((u64)unmap_end > (u64)unmap_start)
2653 			unmap_mapping_range(mapping, unmap_start,
2654 					    1 + unmap_end - unmap_start, 0);
2655 		shmem_truncate_range(inode, offset, offset + len - 1);
2656 		/* No need to unmap again: hole-punching leaves COWed pages */
2657 
2658 		spin_lock(&inode->i_lock);
2659 		inode->i_private = NULL;
2660 		wake_up_all(&shmem_falloc_waitq);
2661 		WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2662 		spin_unlock(&inode->i_lock);
2663 		error = 0;
2664 		goto out;
2665 	}
2666 
2667 	/* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2668 	error = inode_newsize_ok(inode, offset + len);
2669 	if (error)
2670 		goto out;
2671 
2672 	if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2673 		error = -EPERM;
2674 		goto out;
2675 	}
2676 
2677 	start = offset >> PAGE_SHIFT;
2678 	end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2679 	/* Try to avoid a swapstorm if len is impossible to satisfy */
2680 	if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2681 		error = -ENOSPC;
2682 		goto out;
2683 	}
2684 
2685 	shmem_falloc.waitq = NULL;
2686 	shmem_falloc.start = start;
2687 	shmem_falloc.next  = start;
2688 	shmem_falloc.nr_falloced = 0;
2689 	shmem_falloc.nr_unswapped = 0;
2690 	spin_lock(&inode->i_lock);
2691 	inode->i_private = &shmem_falloc;
2692 	spin_unlock(&inode->i_lock);
2693 
2694 	for (index = start; index < end; index++) {
2695 		struct page *page;
2696 
2697 		/*
2698 		 * Good, the fallocate(2) manpage permits EINTR: we may have
2699 		 * been interrupted because we are using up too much memory.
2700 		 */
2701 		if (signal_pending(current))
2702 			error = -EINTR;
2703 		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2704 			error = -ENOMEM;
2705 		else
2706 			error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2707 		if (error) {
2708 			/* Remove the !PageUptodate pages we added */
2709 			if (index > start) {
2710 				shmem_undo_range(inode,
2711 				    (loff_t)start << PAGE_SHIFT,
2712 				    ((loff_t)index << PAGE_SHIFT) - 1, true);
2713 			}
2714 			goto undone;
2715 		}
2716 
2717 		/*
2718 		 * Inform shmem_writepage() how far we have reached.
2719 		 * No need for lock or barrier: we have the page lock.
2720 		 */
2721 		shmem_falloc.next++;
2722 		if (!PageUptodate(page))
2723 			shmem_falloc.nr_falloced++;
2724 
2725 		/*
2726 		 * If !PageUptodate, leave it that way so that freeable pages
2727 		 * can be recognized if we need to rollback on error later.
2728 		 * But set_page_dirty so that memory pressure will swap rather
2729 		 * than free the pages we are allocating (and SGP_CACHE pages
2730 		 * might still be clean: we now need to mark those dirty too).
2731 		 */
2732 		set_page_dirty(page);
2733 		unlock_page(page);
2734 		put_page(page);
2735 		cond_resched();
2736 	}
2737 
2738 	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2739 		i_size_write(inode, offset + len);
2740 	inode->i_ctime = current_time(inode);
2741 undone:
2742 	spin_lock(&inode->i_lock);
2743 	inode->i_private = NULL;
2744 	spin_unlock(&inode->i_lock);
2745 out:
2746 	inode_unlock(inode);
2747 	return error;
2748 }
2749 
2750 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2751 {
2752 	struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2753 
2754 	buf->f_type = TMPFS_MAGIC;
2755 	buf->f_bsize = PAGE_SIZE;
2756 	buf->f_namelen = NAME_MAX;
2757 	if (sbinfo->max_blocks) {
2758 		buf->f_blocks = sbinfo->max_blocks;
2759 		buf->f_bavail =
2760 		buf->f_bfree  = sbinfo->max_blocks -
2761 				percpu_counter_sum(&sbinfo->used_blocks);
2762 	}
2763 	if (sbinfo->max_inodes) {
2764 		buf->f_files = sbinfo->max_inodes;
2765 		buf->f_ffree = sbinfo->free_inodes;
2766 	}
2767 	/* else leave those fields 0 like simple_statfs */
2768 	return 0;
2769 }
2770 
2771 /*
2772  * File creation. Allocate an inode, and we're done..
2773  */
2774 static int
2775 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2776 {
2777 	struct inode *inode;
2778 	int error = -ENOSPC;
2779 
2780 	inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2781 	if (inode) {
2782 		error = simple_acl_create(dir, inode);
2783 		if (error)
2784 			goto out_iput;
2785 		error = security_inode_init_security(inode, dir,
2786 						     &dentry->d_name,
2787 						     shmem_initxattrs, NULL);
2788 		if (error && error != -EOPNOTSUPP)
2789 			goto out_iput;
2790 
2791 		error = 0;
2792 		dir->i_size += BOGO_DIRENT_SIZE;
2793 		dir->i_ctime = dir->i_mtime = current_time(dir);
2794 		d_instantiate(dentry, inode);
2795 		dget(dentry); /* Extra count - pin the dentry in core */
2796 	}
2797 	return error;
2798 out_iput:
2799 	iput(inode);
2800 	return error;
2801 }
2802 
2803 static int
2804 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2805 {
2806 	struct inode *inode;
2807 	int error = -ENOSPC;
2808 
2809 	inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2810 	if (inode) {
2811 		error = security_inode_init_security(inode, dir,
2812 						     NULL,
2813 						     shmem_initxattrs, NULL);
2814 		if (error && error != -EOPNOTSUPP)
2815 			goto out_iput;
2816 		error = simple_acl_create(dir, inode);
2817 		if (error)
2818 			goto out_iput;
2819 		d_tmpfile(dentry, inode);
2820 	}
2821 	return error;
2822 out_iput:
2823 	iput(inode);
2824 	return error;
2825 }
2826 
2827 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2828 {
2829 	int error;
2830 
2831 	if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2832 		return error;
2833 	inc_nlink(dir);
2834 	return 0;
2835 }
2836 
2837 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2838 		bool excl)
2839 {
2840 	return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2841 }
2842 
2843 /*
2844  * Link a file..
2845  */
2846 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2847 {
2848 	struct inode *inode = d_inode(old_dentry);
2849 	int ret;
2850 
2851 	/*
2852 	 * No ordinary (disk based) filesystem counts links as inodes;
2853 	 * but each new link needs a new dentry, pinning lowmem, and
2854 	 * tmpfs dentries cannot be pruned until they are unlinked.
2855 	 */
2856 	ret = shmem_reserve_inode(inode->i_sb);
2857 	if (ret)
2858 		goto out;
2859 
2860 	dir->i_size += BOGO_DIRENT_SIZE;
2861 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2862 	inc_nlink(inode);
2863 	ihold(inode);	/* New dentry reference */
2864 	dget(dentry);		/* Extra pinning count for the created dentry */
2865 	d_instantiate(dentry, inode);
2866 out:
2867 	return ret;
2868 }
2869 
2870 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2871 {
2872 	struct inode *inode = d_inode(dentry);
2873 
2874 	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2875 		shmem_free_inode(inode->i_sb);
2876 
2877 	dir->i_size -= BOGO_DIRENT_SIZE;
2878 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2879 	drop_nlink(inode);
2880 	dput(dentry);	/* Undo the count from "create" - this does all the work */
2881 	return 0;
2882 }
2883 
2884 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2885 {
2886 	if (!simple_empty(dentry))
2887 		return -ENOTEMPTY;
2888 
2889 	drop_nlink(d_inode(dentry));
2890 	drop_nlink(dir);
2891 	return shmem_unlink(dir, dentry);
2892 }
2893 
2894 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2895 {
2896 	bool old_is_dir = d_is_dir(old_dentry);
2897 	bool new_is_dir = d_is_dir(new_dentry);
2898 
2899 	if (old_dir != new_dir && old_is_dir != new_is_dir) {
2900 		if (old_is_dir) {
2901 			drop_nlink(old_dir);
2902 			inc_nlink(new_dir);
2903 		} else {
2904 			drop_nlink(new_dir);
2905 			inc_nlink(old_dir);
2906 		}
2907 	}
2908 	old_dir->i_ctime = old_dir->i_mtime =
2909 	new_dir->i_ctime = new_dir->i_mtime =
2910 	d_inode(old_dentry)->i_ctime =
2911 	d_inode(new_dentry)->i_ctime = current_time(old_dir);
2912 
2913 	return 0;
2914 }
2915 
2916 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2917 {
2918 	struct dentry *whiteout;
2919 	int error;
2920 
2921 	whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2922 	if (!whiteout)
2923 		return -ENOMEM;
2924 
2925 	error = shmem_mknod(old_dir, whiteout,
2926 			    S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2927 	dput(whiteout);
2928 	if (error)
2929 		return error;
2930 
2931 	/*
2932 	 * Cheat and hash the whiteout while the old dentry is still in
2933 	 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2934 	 *
2935 	 * d_lookup() will consistently find one of them at this point,
2936 	 * not sure which one, but that isn't even important.
2937 	 */
2938 	d_rehash(whiteout);
2939 	return 0;
2940 }
2941 
2942 /*
2943  * The VFS layer already does all the dentry stuff for rename,
2944  * we just have to decrement the usage count for the target if
2945  * it exists so that the VFS layer correctly free's it when it
2946  * gets overwritten.
2947  */
2948 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
2949 {
2950 	struct inode *inode = d_inode(old_dentry);
2951 	int they_are_dirs = S_ISDIR(inode->i_mode);
2952 
2953 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2954 		return -EINVAL;
2955 
2956 	if (flags & RENAME_EXCHANGE)
2957 		return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2958 
2959 	if (!simple_empty(new_dentry))
2960 		return -ENOTEMPTY;
2961 
2962 	if (flags & RENAME_WHITEOUT) {
2963 		int error;
2964 
2965 		error = shmem_whiteout(old_dir, old_dentry);
2966 		if (error)
2967 			return error;
2968 	}
2969 
2970 	if (d_really_is_positive(new_dentry)) {
2971 		(void) shmem_unlink(new_dir, new_dentry);
2972 		if (they_are_dirs) {
2973 			drop_nlink(d_inode(new_dentry));
2974 			drop_nlink(old_dir);
2975 		}
2976 	} else if (they_are_dirs) {
2977 		drop_nlink(old_dir);
2978 		inc_nlink(new_dir);
2979 	}
2980 
2981 	old_dir->i_size -= BOGO_DIRENT_SIZE;
2982 	new_dir->i_size += BOGO_DIRENT_SIZE;
2983 	old_dir->i_ctime = old_dir->i_mtime =
2984 	new_dir->i_ctime = new_dir->i_mtime =
2985 	inode->i_ctime = current_time(old_dir);
2986 	return 0;
2987 }
2988 
2989 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2990 {
2991 	int error;
2992 	int len;
2993 	struct inode *inode;
2994 	struct page *page;
2995 
2996 	len = strlen(symname) + 1;
2997 	if (len > PAGE_SIZE)
2998 		return -ENAMETOOLONG;
2999 
3000 	inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3001 				VM_NORESERVE);
3002 	if (!inode)
3003 		return -ENOSPC;
3004 
3005 	error = security_inode_init_security(inode, dir, &dentry->d_name,
3006 					     shmem_initxattrs, NULL);
3007 	if (error) {
3008 		if (error != -EOPNOTSUPP) {
3009 			iput(inode);
3010 			return error;
3011 		}
3012 		error = 0;
3013 	}
3014 
3015 	inode->i_size = len-1;
3016 	if (len <= SHORT_SYMLINK_LEN) {
3017 		inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3018 		if (!inode->i_link) {
3019 			iput(inode);
3020 			return -ENOMEM;
3021 		}
3022 		inode->i_op = &shmem_short_symlink_operations;
3023 	} else {
3024 		inode_nohighmem(inode);
3025 		error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3026 		if (error) {
3027 			iput(inode);
3028 			return error;
3029 		}
3030 		inode->i_mapping->a_ops = &shmem_aops;
3031 		inode->i_op = &shmem_symlink_inode_operations;
3032 		memcpy(page_address(page), symname, len);
3033 		SetPageUptodate(page);
3034 		set_page_dirty(page);
3035 		unlock_page(page);
3036 		put_page(page);
3037 	}
3038 	dir->i_size += BOGO_DIRENT_SIZE;
3039 	dir->i_ctime = dir->i_mtime = current_time(dir);
3040 	d_instantiate(dentry, inode);
3041 	dget(dentry);
3042 	return 0;
3043 }
3044 
3045 static void shmem_put_link(void *arg)
3046 {
3047 	mark_page_accessed(arg);
3048 	put_page(arg);
3049 }
3050 
3051 static const char *shmem_get_link(struct dentry *dentry,
3052 				  struct inode *inode,
3053 				  struct delayed_call *done)
3054 {
3055 	struct page *page = NULL;
3056 	int error;
3057 	if (!dentry) {
3058 		page = find_get_page(inode->i_mapping, 0);
3059 		if (!page)
3060 			return ERR_PTR(-ECHILD);
3061 		if (!PageUptodate(page)) {
3062 			put_page(page);
3063 			return ERR_PTR(-ECHILD);
3064 		}
3065 	} else {
3066 		error = shmem_getpage(inode, 0, &page, SGP_READ);
3067 		if (error)
3068 			return ERR_PTR(error);
3069 		unlock_page(page);
3070 	}
3071 	set_delayed_call(done, shmem_put_link, page);
3072 	return page_address(page);
3073 }
3074 
3075 #ifdef CONFIG_TMPFS_XATTR
3076 /*
3077  * Superblocks without xattr inode operations may get some security.* xattr
3078  * support from the LSM "for free". As soon as we have any other xattrs
3079  * like ACLs, we also need to implement the security.* handlers at
3080  * filesystem level, though.
3081  */
3082 
3083 /*
3084  * Callback for security_inode_init_security() for acquiring xattrs.
3085  */
3086 static int shmem_initxattrs(struct inode *inode,
3087 			    const struct xattr *xattr_array,
3088 			    void *fs_info)
3089 {
3090 	struct shmem_inode_info *info = SHMEM_I(inode);
3091 	const struct xattr *xattr;
3092 	struct simple_xattr *new_xattr;
3093 	size_t len;
3094 
3095 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3096 		new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3097 		if (!new_xattr)
3098 			return -ENOMEM;
3099 
3100 		len = strlen(xattr->name) + 1;
3101 		new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3102 					  GFP_KERNEL);
3103 		if (!new_xattr->name) {
3104 			kfree(new_xattr);
3105 			return -ENOMEM;
3106 		}
3107 
3108 		memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3109 		       XATTR_SECURITY_PREFIX_LEN);
3110 		memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3111 		       xattr->name, len);
3112 
3113 		simple_xattr_list_add(&info->xattrs, new_xattr);
3114 	}
3115 
3116 	return 0;
3117 }
3118 
3119 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3120 				   struct dentry *unused, struct inode *inode,
3121 				   const char *name, void *buffer, size_t size)
3122 {
3123 	struct shmem_inode_info *info = SHMEM_I(inode);
3124 
3125 	name = xattr_full_name(handler, name);
3126 	return simple_xattr_get(&info->xattrs, name, buffer, size);
3127 }
3128 
3129 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3130 				   struct dentry *unused, struct inode *inode,
3131 				   const char *name, const void *value,
3132 				   size_t size, int flags)
3133 {
3134 	struct shmem_inode_info *info = SHMEM_I(inode);
3135 
3136 	name = xattr_full_name(handler, name);
3137 	return simple_xattr_set(&info->xattrs, name, value, size, flags);
3138 }
3139 
3140 static const struct xattr_handler shmem_security_xattr_handler = {
3141 	.prefix = XATTR_SECURITY_PREFIX,
3142 	.get = shmem_xattr_handler_get,
3143 	.set = shmem_xattr_handler_set,
3144 };
3145 
3146 static const struct xattr_handler shmem_trusted_xattr_handler = {
3147 	.prefix = XATTR_TRUSTED_PREFIX,
3148 	.get = shmem_xattr_handler_get,
3149 	.set = shmem_xattr_handler_set,
3150 };
3151 
3152 static const struct xattr_handler *shmem_xattr_handlers[] = {
3153 #ifdef CONFIG_TMPFS_POSIX_ACL
3154 	&posix_acl_access_xattr_handler,
3155 	&posix_acl_default_xattr_handler,
3156 #endif
3157 	&shmem_security_xattr_handler,
3158 	&shmem_trusted_xattr_handler,
3159 	NULL
3160 };
3161 
3162 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3163 {
3164 	struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3165 	return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3166 }
3167 #endif /* CONFIG_TMPFS_XATTR */
3168 
3169 static const struct inode_operations shmem_short_symlink_operations = {
3170 	.get_link	= simple_get_link,
3171 #ifdef CONFIG_TMPFS_XATTR
3172 	.listxattr	= shmem_listxattr,
3173 #endif
3174 };
3175 
3176 static const struct inode_operations shmem_symlink_inode_operations = {
3177 	.get_link	= shmem_get_link,
3178 #ifdef CONFIG_TMPFS_XATTR
3179 	.listxattr	= shmem_listxattr,
3180 #endif
3181 };
3182 
3183 static struct dentry *shmem_get_parent(struct dentry *child)
3184 {
3185 	return ERR_PTR(-ESTALE);
3186 }
3187 
3188 static int shmem_match(struct inode *ino, void *vfh)
3189 {
3190 	__u32 *fh = vfh;
3191 	__u64 inum = fh[2];
3192 	inum = (inum << 32) | fh[1];
3193 	return ino->i_ino == inum && fh[0] == ino->i_generation;
3194 }
3195 
3196 /* Find any alias of inode, but prefer a hashed alias */
3197 static struct dentry *shmem_find_alias(struct inode *inode)
3198 {
3199 	struct dentry *alias = d_find_alias(inode);
3200 
3201 	return alias ?: d_find_any_alias(inode);
3202 }
3203 
3204 
3205 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3206 		struct fid *fid, int fh_len, int fh_type)
3207 {
3208 	struct inode *inode;
3209 	struct dentry *dentry = NULL;
3210 	u64 inum;
3211 
3212 	if (fh_len < 3)
3213 		return NULL;
3214 
3215 	inum = fid->raw[2];
3216 	inum = (inum << 32) | fid->raw[1];
3217 
3218 	inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3219 			shmem_match, fid->raw);
3220 	if (inode) {
3221 		dentry = shmem_find_alias(inode);
3222 		iput(inode);
3223 	}
3224 
3225 	return dentry;
3226 }
3227 
3228 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3229 				struct inode *parent)
3230 {
3231 	if (*len < 3) {
3232 		*len = 3;
3233 		return FILEID_INVALID;
3234 	}
3235 
3236 	if (inode_unhashed(inode)) {
3237 		/* Unfortunately insert_inode_hash is not idempotent,
3238 		 * so as we hash inodes here rather than at creation
3239 		 * time, we need a lock to ensure we only try
3240 		 * to do it once
3241 		 */
3242 		static DEFINE_SPINLOCK(lock);
3243 		spin_lock(&lock);
3244 		if (inode_unhashed(inode))
3245 			__insert_inode_hash(inode,
3246 					    inode->i_ino + inode->i_generation);
3247 		spin_unlock(&lock);
3248 	}
3249 
3250 	fh[0] = inode->i_generation;
3251 	fh[1] = inode->i_ino;
3252 	fh[2] = ((__u64)inode->i_ino) >> 32;
3253 
3254 	*len = 3;
3255 	return 1;
3256 }
3257 
3258 static const struct export_operations shmem_export_ops = {
3259 	.get_parent     = shmem_get_parent,
3260 	.encode_fh      = shmem_encode_fh,
3261 	.fh_to_dentry	= shmem_fh_to_dentry,
3262 };
3263 
3264 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3265 			       bool remount)
3266 {
3267 	char *this_char, *value, *rest;
3268 	struct mempolicy *mpol = NULL;
3269 	uid_t uid;
3270 	gid_t gid;
3271 
3272 	while (options != NULL) {
3273 		this_char = options;
3274 		for (;;) {
3275 			/*
3276 			 * NUL-terminate this option: unfortunately,
3277 			 * mount options form a comma-separated list,
3278 			 * but mpol's nodelist may also contain commas.
3279 			 */
3280 			options = strchr(options, ',');
3281 			if (options == NULL)
3282 				break;
3283 			options++;
3284 			if (!isdigit(*options)) {
3285 				options[-1] = '\0';
3286 				break;
3287 			}
3288 		}
3289 		if (!*this_char)
3290 			continue;
3291 		if ((value = strchr(this_char,'=')) != NULL) {
3292 			*value++ = 0;
3293 		} else {
3294 			pr_err("tmpfs: No value for mount option '%s'\n",
3295 			       this_char);
3296 			goto error;
3297 		}
3298 
3299 		if (!strcmp(this_char,"size")) {
3300 			unsigned long long size;
3301 			size = memparse(value,&rest);
3302 			if (*rest == '%') {
3303 				size <<= PAGE_SHIFT;
3304 				size *= totalram_pages;
3305 				do_div(size, 100);
3306 				rest++;
3307 			}
3308 			if (*rest)
3309 				goto bad_val;
3310 			sbinfo->max_blocks =
3311 				DIV_ROUND_UP(size, PAGE_SIZE);
3312 		} else if (!strcmp(this_char,"nr_blocks")) {
3313 			sbinfo->max_blocks = memparse(value, &rest);
3314 			if (*rest)
3315 				goto bad_val;
3316 		} else if (!strcmp(this_char,"nr_inodes")) {
3317 			sbinfo->max_inodes = memparse(value, &rest);
3318 			if (*rest)
3319 				goto bad_val;
3320 		} else if (!strcmp(this_char,"mode")) {
3321 			if (remount)
3322 				continue;
3323 			sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3324 			if (*rest)
3325 				goto bad_val;
3326 		} else if (!strcmp(this_char,"uid")) {
3327 			if (remount)
3328 				continue;
3329 			uid = simple_strtoul(value, &rest, 0);
3330 			if (*rest)
3331 				goto bad_val;
3332 			sbinfo->uid = make_kuid(current_user_ns(), uid);
3333 			if (!uid_valid(sbinfo->uid))
3334 				goto bad_val;
3335 		} else if (!strcmp(this_char,"gid")) {
3336 			if (remount)
3337 				continue;
3338 			gid = simple_strtoul(value, &rest, 0);
3339 			if (*rest)
3340 				goto bad_val;
3341 			sbinfo->gid = make_kgid(current_user_ns(), gid);
3342 			if (!gid_valid(sbinfo->gid))
3343 				goto bad_val;
3344 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3345 		} else if (!strcmp(this_char, "huge")) {
3346 			int huge;
3347 			huge = shmem_parse_huge(value);
3348 			if (huge < 0)
3349 				goto bad_val;
3350 			if (!has_transparent_hugepage() &&
3351 					huge != SHMEM_HUGE_NEVER)
3352 				goto bad_val;
3353 			sbinfo->huge = huge;
3354 #endif
3355 #ifdef CONFIG_NUMA
3356 		} else if (!strcmp(this_char,"mpol")) {
3357 			mpol_put(mpol);
3358 			mpol = NULL;
3359 			if (mpol_parse_str(value, &mpol))
3360 				goto bad_val;
3361 #endif
3362 		} else {
3363 			pr_err("tmpfs: Bad mount option %s\n", this_char);
3364 			goto error;
3365 		}
3366 	}
3367 	sbinfo->mpol = mpol;
3368 	return 0;
3369 
3370 bad_val:
3371 	pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3372 	       value, this_char);
3373 error:
3374 	mpol_put(mpol);
3375 	return 1;
3376 
3377 }
3378 
3379 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3380 {
3381 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3382 	struct shmem_sb_info config = *sbinfo;
3383 	unsigned long inodes;
3384 	int error = -EINVAL;
3385 
3386 	config.mpol = NULL;
3387 	if (shmem_parse_options(data, &config, true))
3388 		return error;
3389 
3390 	spin_lock(&sbinfo->stat_lock);
3391 	inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3392 	if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3393 		goto out;
3394 	if (config.max_inodes < inodes)
3395 		goto out;
3396 	/*
3397 	 * Those tests disallow limited->unlimited while any are in use;
3398 	 * but we must separately disallow unlimited->limited, because
3399 	 * in that case we have no record of how much is already in use.
3400 	 */
3401 	if (config.max_blocks && !sbinfo->max_blocks)
3402 		goto out;
3403 	if (config.max_inodes && !sbinfo->max_inodes)
3404 		goto out;
3405 
3406 	error = 0;
3407 	sbinfo->huge = config.huge;
3408 	sbinfo->max_blocks  = config.max_blocks;
3409 	sbinfo->max_inodes  = config.max_inodes;
3410 	sbinfo->free_inodes = config.max_inodes - inodes;
3411 
3412 	/*
3413 	 * Preserve previous mempolicy unless mpol remount option was specified.
3414 	 */
3415 	if (config.mpol) {
3416 		mpol_put(sbinfo->mpol);
3417 		sbinfo->mpol = config.mpol;	/* transfers initial ref */
3418 	}
3419 out:
3420 	spin_unlock(&sbinfo->stat_lock);
3421 	return error;
3422 }
3423 
3424 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3425 {
3426 	struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3427 
3428 	if (sbinfo->max_blocks != shmem_default_max_blocks())
3429 		seq_printf(seq, ",size=%luk",
3430 			sbinfo->max_blocks << (PAGE_SHIFT - 10));
3431 	if (sbinfo->max_inodes != shmem_default_max_inodes())
3432 		seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3433 	if (sbinfo->mode != (0777 | S_ISVTX))
3434 		seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3435 	if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3436 		seq_printf(seq, ",uid=%u",
3437 				from_kuid_munged(&init_user_ns, sbinfo->uid));
3438 	if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3439 		seq_printf(seq, ",gid=%u",
3440 				from_kgid_munged(&init_user_ns, sbinfo->gid));
3441 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3442 	/* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3443 	if (sbinfo->huge)
3444 		seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3445 #endif
3446 	shmem_show_mpol(seq, sbinfo->mpol);
3447 	return 0;
3448 }
3449 
3450 #endif /* CONFIG_TMPFS */
3451 
3452 static void shmem_put_super(struct super_block *sb)
3453 {
3454 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3455 
3456 	percpu_counter_destroy(&sbinfo->used_blocks);
3457 	mpol_put(sbinfo->mpol);
3458 	kfree(sbinfo);
3459 	sb->s_fs_info = NULL;
3460 }
3461 
3462 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3463 {
3464 	struct inode *inode;
3465 	struct shmem_sb_info *sbinfo;
3466 	int err = -ENOMEM;
3467 
3468 	/* Round up to L1_CACHE_BYTES to resist false sharing */
3469 	sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3470 				L1_CACHE_BYTES), GFP_KERNEL);
3471 	if (!sbinfo)
3472 		return -ENOMEM;
3473 
3474 	sbinfo->mode = 0777 | S_ISVTX;
3475 	sbinfo->uid = current_fsuid();
3476 	sbinfo->gid = current_fsgid();
3477 	sb->s_fs_info = sbinfo;
3478 
3479 #ifdef CONFIG_TMPFS
3480 	/*
3481 	 * Per default we only allow half of the physical ram per
3482 	 * tmpfs instance, limiting inodes to one per page of lowmem;
3483 	 * but the internal instance is left unlimited.
3484 	 */
3485 	if (!(sb->s_flags & SB_KERNMOUNT)) {
3486 		sbinfo->max_blocks = shmem_default_max_blocks();
3487 		sbinfo->max_inodes = shmem_default_max_inodes();
3488 		if (shmem_parse_options(data, sbinfo, false)) {
3489 			err = -EINVAL;
3490 			goto failed;
3491 		}
3492 	} else {
3493 		sb->s_flags |= SB_NOUSER;
3494 	}
3495 	sb->s_export_op = &shmem_export_ops;
3496 	sb->s_flags |= SB_NOSEC;
3497 #else
3498 	sb->s_flags |= SB_NOUSER;
3499 #endif
3500 
3501 	spin_lock_init(&sbinfo->stat_lock);
3502 	if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3503 		goto failed;
3504 	sbinfo->free_inodes = sbinfo->max_inodes;
3505 	spin_lock_init(&sbinfo->shrinklist_lock);
3506 	INIT_LIST_HEAD(&sbinfo->shrinklist);
3507 
3508 	sb->s_maxbytes = MAX_LFS_FILESIZE;
3509 	sb->s_blocksize = PAGE_SIZE;
3510 	sb->s_blocksize_bits = PAGE_SHIFT;
3511 	sb->s_magic = TMPFS_MAGIC;
3512 	sb->s_op = &shmem_ops;
3513 	sb->s_time_gran = 1;
3514 #ifdef CONFIG_TMPFS_XATTR
3515 	sb->s_xattr = shmem_xattr_handlers;
3516 #endif
3517 #ifdef CONFIG_TMPFS_POSIX_ACL
3518 	sb->s_flags |= SB_POSIXACL;
3519 #endif
3520 	uuid_gen(&sb->s_uuid);
3521 
3522 	inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3523 	if (!inode)
3524 		goto failed;
3525 	inode->i_uid = sbinfo->uid;
3526 	inode->i_gid = sbinfo->gid;
3527 	sb->s_root = d_make_root(inode);
3528 	if (!sb->s_root)
3529 		goto failed;
3530 	return 0;
3531 
3532 failed:
3533 	shmem_put_super(sb);
3534 	return err;
3535 }
3536 
3537 static struct kmem_cache *shmem_inode_cachep;
3538 
3539 static struct inode *shmem_alloc_inode(struct super_block *sb)
3540 {
3541 	struct shmem_inode_info *info;
3542 	info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3543 	if (!info)
3544 		return NULL;
3545 	return &info->vfs_inode;
3546 }
3547 
3548 static void shmem_destroy_callback(struct rcu_head *head)
3549 {
3550 	struct inode *inode = container_of(head, struct inode, i_rcu);
3551 	if (S_ISLNK(inode->i_mode))
3552 		kfree(inode->i_link);
3553 	kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3554 }
3555 
3556 static void shmem_destroy_inode(struct inode *inode)
3557 {
3558 	if (S_ISREG(inode->i_mode))
3559 		mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3560 	call_rcu(&inode->i_rcu, shmem_destroy_callback);
3561 }
3562 
3563 static void shmem_init_inode(void *foo)
3564 {
3565 	struct shmem_inode_info *info = foo;
3566 	inode_init_once(&info->vfs_inode);
3567 }
3568 
3569 static void shmem_init_inodecache(void)
3570 {
3571 	shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3572 				sizeof(struct shmem_inode_info),
3573 				0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3574 }
3575 
3576 static void shmem_destroy_inodecache(void)
3577 {
3578 	kmem_cache_destroy(shmem_inode_cachep);
3579 }
3580 
3581 static const struct address_space_operations shmem_aops = {
3582 	.writepage	= shmem_writepage,
3583 	.set_page_dirty	= __set_page_dirty_no_writeback,
3584 #ifdef CONFIG_TMPFS
3585 	.write_begin	= shmem_write_begin,
3586 	.write_end	= shmem_write_end,
3587 #endif
3588 #ifdef CONFIG_MIGRATION
3589 	.migratepage	= migrate_page,
3590 #endif
3591 	.error_remove_page = generic_error_remove_page,
3592 };
3593 
3594 static const struct file_operations shmem_file_operations = {
3595 	.mmap		= shmem_mmap,
3596 	.get_unmapped_area = shmem_get_unmapped_area,
3597 #ifdef CONFIG_TMPFS
3598 	.llseek		= shmem_file_llseek,
3599 	.read_iter	= shmem_file_read_iter,
3600 	.write_iter	= generic_file_write_iter,
3601 	.fsync		= noop_fsync,
3602 	.splice_read	= generic_file_splice_read,
3603 	.splice_write	= iter_file_splice_write,
3604 	.fallocate	= shmem_fallocate,
3605 #endif
3606 };
3607 
3608 static const struct inode_operations shmem_inode_operations = {
3609 	.getattr	= shmem_getattr,
3610 	.setattr	= shmem_setattr,
3611 #ifdef CONFIG_TMPFS_XATTR
3612 	.listxattr	= shmem_listxattr,
3613 	.set_acl	= simple_set_acl,
3614 #endif
3615 };
3616 
3617 static const struct inode_operations shmem_dir_inode_operations = {
3618 #ifdef CONFIG_TMPFS
3619 	.create		= shmem_create,
3620 	.lookup		= simple_lookup,
3621 	.link		= shmem_link,
3622 	.unlink		= shmem_unlink,
3623 	.symlink	= shmem_symlink,
3624 	.mkdir		= shmem_mkdir,
3625 	.rmdir		= shmem_rmdir,
3626 	.mknod		= shmem_mknod,
3627 	.rename		= shmem_rename2,
3628 	.tmpfile	= shmem_tmpfile,
3629 #endif
3630 #ifdef CONFIG_TMPFS_XATTR
3631 	.listxattr	= shmem_listxattr,
3632 #endif
3633 #ifdef CONFIG_TMPFS_POSIX_ACL
3634 	.setattr	= shmem_setattr,
3635 	.set_acl	= simple_set_acl,
3636 #endif
3637 };
3638 
3639 static const struct inode_operations shmem_special_inode_operations = {
3640 #ifdef CONFIG_TMPFS_XATTR
3641 	.listxattr	= shmem_listxattr,
3642 #endif
3643 #ifdef CONFIG_TMPFS_POSIX_ACL
3644 	.setattr	= shmem_setattr,
3645 	.set_acl	= simple_set_acl,
3646 #endif
3647 };
3648 
3649 static const struct super_operations shmem_ops = {
3650 	.alloc_inode	= shmem_alloc_inode,
3651 	.destroy_inode	= shmem_destroy_inode,
3652 #ifdef CONFIG_TMPFS
3653 	.statfs		= shmem_statfs,
3654 	.remount_fs	= shmem_remount_fs,
3655 	.show_options	= shmem_show_options,
3656 #endif
3657 	.evict_inode	= shmem_evict_inode,
3658 	.drop_inode	= generic_delete_inode,
3659 	.put_super	= shmem_put_super,
3660 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3661 	.nr_cached_objects	= shmem_unused_huge_count,
3662 	.free_cached_objects	= shmem_unused_huge_scan,
3663 #endif
3664 };
3665 
3666 static const struct vm_operations_struct shmem_vm_ops = {
3667 	.fault		= shmem_fault,
3668 	.map_pages	= filemap_map_pages,
3669 #ifdef CONFIG_NUMA
3670 	.set_policy     = shmem_set_policy,
3671 	.get_policy     = shmem_get_policy,
3672 #endif
3673 };
3674 
3675 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3676 	int flags, const char *dev_name, void *data)
3677 {
3678 	return mount_nodev(fs_type, flags, data, shmem_fill_super);
3679 }
3680 
3681 static struct file_system_type shmem_fs_type = {
3682 	.owner		= THIS_MODULE,
3683 	.name		= "tmpfs",
3684 	.mount		= shmem_mount,
3685 	.kill_sb	= kill_litter_super,
3686 	.fs_flags	= FS_USERNS_MOUNT,
3687 };
3688 
3689 int __init shmem_init(void)
3690 {
3691 	int error;
3692 
3693 	/* If rootfs called this, don't re-init */
3694 	if (shmem_inode_cachep)
3695 		return 0;
3696 
3697 	shmem_init_inodecache();
3698 
3699 	error = register_filesystem(&shmem_fs_type);
3700 	if (error) {
3701 		pr_err("Could not register tmpfs\n");
3702 		goto out2;
3703 	}
3704 
3705 	shm_mnt = kern_mount(&shmem_fs_type);
3706 	if (IS_ERR(shm_mnt)) {
3707 		error = PTR_ERR(shm_mnt);
3708 		pr_err("Could not kern_mount tmpfs\n");
3709 		goto out1;
3710 	}
3711 
3712 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3713 	if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3714 		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3715 	else
3716 		shmem_huge = 0; /* just in case it was patched */
3717 #endif
3718 	return 0;
3719 
3720 out1:
3721 	unregister_filesystem(&shmem_fs_type);
3722 out2:
3723 	shmem_destroy_inodecache();
3724 	shm_mnt = ERR_PTR(error);
3725 	return error;
3726 }
3727 
3728 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3729 static ssize_t shmem_enabled_show(struct kobject *kobj,
3730 		struct kobj_attribute *attr, char *buf)
3731 {
3732 	int values[] = {
3733 		SHMEM_HUGE_ALWAYS,
3734 		SHMEM_HUGE_WITHIN_SIZE,
3735 		SHMEM_HUGE_ADVISE,
3736 		SHMEM_HUGE_NEVER,
3737 		SHMEM_HUGE_DENY,
3738 		SHMEM_HUGE_FORCE,
3739 	};
3740 	int i, count;
3741 
3742 	for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3743 		const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3744 
3745 		count += sprintf(buf + count, fmt,
3746 				shmem_format_huge(values[i]));
3747 	}
3748 	buf[count - 1] = '\n';
3749 	return count;
3750 }
3751 
3752 static ssize_t shmem_enabled_store(struct kobject *kobj,
3753 		struct kobj_attribute *attr, const char *buf, size_t count)
3754 {
3755 	char tmp[16];
3756 	int huge;
3757 
3758 	if (count + 1 > sizeof(tmp))
3759 		return -EINVAL;
3760 	memcpy(tmp, buf, count);
3761 	tmp[count] = '\0';
3762 	if (count && tmp[count - 1] == '\n')
3763 		tmp[count - 1] = '\0';
3764 
3765 	huge = shmem_parse_huge(tmp);
3766 	if (huge == -EINVAL)
3767 		return -EINVAL;
3768 	if (!has_transparent_hugepage() &&
3769 			huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3770 		return -EINVAL;
3771 
3772 	shmem_huge = huge;
3773 	if (shmem_huge > SHMEM_HUGE_DENY)
3774 		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3775 	return count;
3776 }
3777 
3778 struct kobj_attribute shmem_enabled_attr =
3779 	__ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3780 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3781 
3782 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3783 bool shmem_huge_enabled(struct vm_area_struct *vma)
3784 {
3785 	struct inode *inode = file_inode(vma->vm_file);
3786 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3787 	loff_t i_size;
3788 	pgoff_t off;
3789 
3790 	if (shmem_huge == SHMEM_HUGE_FORCE)
3791 		return true;
3792 	if (shmem_huge == SHMEM_HUGE_DENY)
3793 		return false;
3794 	switch (sbinfo->huge) {
3795 		case SHMEM_HUGE_NEVER:
3796 			return false;
3797 		case SHMEM_HUGE_ALWAYS:
3798 			return true;
3799 		case SHMEM_HUGE_WITHIN_SIZE:
3800 			off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
3801 			i_size = round_up(i_size_read(inode), PAGE_SIZE);
3802 			if (i_size >= HPAGE_PMD_SIZE &&
3803 					i_size >> PAGE_SHIFT >= off)
3804 				return true;
3805 			/* fall through */
3806 		case SHMEM_HUGE_ADVISE:
3807 			/* TODO: implement fadvise() hints */
3808 			return (vma->vm_flags & VM_HUGEPAGE);
3809 		default:
3810 			VM_BUG_ON(1);
3811 			return false;
3812 	}
3813 }
3814 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
3815 
3816 #else /* !CONFIG_SHMEM */
3817 
3818 /*
3819  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3820  *
3821  * This is intended for small system where the benefits of the full
3822  * shmem code (swap-backed and resource-limited) are outweighed by
3823  * their complexity. On systems without swap this code should be
3824  * effectively equivalent, but much lighter weight.
3825  */
3826 
3827 static struct file_system_type shmem_fs_type = {
3828 	.name		= "tmpfs",
3829 	.mount		= ramfs_mount,
3830 	.kill_sb	= kill_litter_super,
3831 	.fs_flags	= FS_USERNS_MOUNT,
3832 };
3833 
3834 int __init shmem_init(void)
3835 {
3836 	BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3837 
3838 	shm_mnt = kern_mount(&shmem_fs_type);
3839 	BUG_ON(IS_ERR(shm_mnt));
3840 
3841 	return 0;
3842 }
3843 
3844 int shmem_unuse(swp_entry_t swap, struct page *page)
3845 {
3846 	return 0;
3847 }
3848 
3849 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3850 {
3851 	return 0;
3852 }
3853 
3854 void shmem_unlock_mapping(struct address_space *mapping)
3855 {
3856 }
3857 
3858 #ifdef CONFIG_MMU
3859 unsigned long shmem_get_unmapped_area(struct file *file,
3860 				      unsigned long addr, unsigned long len,
3861 				      unsigned long pgoff, unsigned long flags)
3862 {
3863 	return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
3864 }
3865 #endif
3866 
3867 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3868 {
3869 	truncate_inode_pages_range(inode->i_mapping, lstart, lend);
3870 }
3871 EXPORT_SYMBOL_GPL(shmem_truncate_range);
3872 
3873 #define shmem_vm_ops				generic_file_vm_ops
3874 #define shmem_file_operations			ramfs_file_operations
3875 #define shmem_get_inode(sb, dir, mode, dev, flags)	ramfs_get_inode(sb, dir, mode, dev)
3876 #define shmem_acct_size(flags, size)		0
3877 #define shmem_unacct_size(flags, size)		do {} while (0)
3878 
3879 #endif /* CONFIG_SHMEM */
3880 
3881 /* common code */
3882 
3883 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
3884 				       unsigned long flags, unsigned int i_flags)
3885 {
3886 	struct inode *inode;
3887 	struct file *res;
3888 
3889 	if (IS_ERR(mnt))
3890 		return ERR_CAST(mnt);
3891 
3892 	if (size < 0 || size > MAX_LFS_FILESIZE)
3893 		return ERR_PTR(-EINVAL);
3894 
3895 	if (shmem_acct_size(flags, size))
3896 		return ERR_PTR(-ENOMEM);
3897 
3898 	inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
3899 				flags);
3900 	if (unlikely(!inode)) {
3901 		shmem_unacct_size(flags, size);
3902 		return ERR_PTR(-ENOSPC);
3903 	}
3904 	inode->i_flags |= i_flags;
3905 	inode->i_size = size;
3906 	clear_nlink(inode);	/* It is unlinked */
3907 	res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3908 	if (!IS_ERR(res))
3909 		res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
3910 				&shmem_file_operations);
3911 	if (IS_ERR(res))
3912 		iput(inode);
3913 	return res;
3914 }
3915 
3916 /**
3917  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3918  * 	kernel internal.  There will be NO LSM permission checks against the
3919  * 	underlying inode.  So users of this interface must do LSM checks at a
3920  *	higher layer.  The users are the big_key and shm implementations.  LSM
3921  *	checks are provided at the key or shm level rather than the inode.
3922  * @name: name for dentry (to be seen in /proc/<pid>/maps
3923  * @size: size to be set for the file
3924  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3925  */
3926 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3927 {
3928 	return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
3929 }
3930 
3931 /**
3932  * shmem_file_setup - get an unlinked file living in tmpfs
3933  * @name: name for dentry (to be seen in /proc/<pid>/maps
3934  * @size: size to be set for the file
3935  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3936  */
3937 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3938 {
3939 	return __shmem_file_setup(shm_mnt, name, size, flags, 0);
3940 }
3941 EXPORT_SYMBOL_GPL(shmem_file_setup);
3942 
3943 /**
3944  * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
3945  * @mnt: the tmpfs mount where the file will be created
3946  * @name: name for dentry (to be seen in /proc/<pid>/maps
3947  * @size: size to be set for the file
3948  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3949  */
3950 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
3951 				       loff_t size, unsigned long flags)
3952 {
3953 	return __shmem_file_setup(mnt, name, size, flags, 0);
3954 }
3955 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
3956 
3957 /**
3958  * shmem_zero_setup - setup a shared anonymous mapping
3959  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3960  */
3961 int shmem_zero_setup(struct vm_area_struct *vma)
3962 {
3963 	struct file *file;
3964 	loff_t size = vma->vm_end - vma->vm_start;
3965 
3966 	/*
3967 	 * Cloning a new file under mmap_sem leads to a lock ordering conflict
3968 	 * between XFS directory reading and selinux: since this file is only
3969 	 * accessible to the user through its mapping, use S_PRIVATE flag to
3970 	 * bypass file security, in the same way as shmem_kernel_file_setup().
3971 	 */
3972 	file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
3973 	if (IS_ERR(file))
3974 		return PTR_ERR(file);
3975 
3976 	if (vma->vm_file)
3977 		fput(vma->vm_file);
3978 	vma->vm_file = file;
3979 	vma->vm_ops = &shmem_vm_ops;
3980 
3981 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
3982 			((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
3983 			(vma->vm_end & HPAGE_PMD_MASK)) {
3984 		khugepaged_enter(vma, vma->vm_flags);
3985 	}
3986 
3987 	return 0;
3988 }
3989 
3990 /**
3991  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3992  * @mapping:	the page's address_space
3993  * @index:	the page index
3994  * @gfp:	the page allocator flags to use if allocating
3995  *
3996  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3997  * with any new page allocations done using the specified allocation flags.
3998  * But read_cache_page_gfp() uses the ->readpage() method: which does not
3999  * suit tmpfs, since it may have pages in swapcache, and needs to find those
4000  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4001  *
4002  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4003  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4004  */
4005 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4006 					 pgoff_t index, gfp_t gfp)
4007 {
4008 #ifdef CONFIG_SHMEM
4009 	struct inode *inode = mapping->host;
4010 	struct page *page;
4011 	int error;
4012 
4013 	BUG_ON(mapping->a_ops != &shmem_aops);
4014 	error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4015 				  gfp, NULL, NULL, NULL);
4016 	if (error)
4017 		page = ERR_PTR(error);
4018 	else
4019 		unlock_page(page);
4020 	return page;
4021 #else
4022 	/*
4023 	 * The tiny !SHMEM case uses ramfs without swap
4024 	 */
4025 	return read_cache_page_gfp(mapping, index, gfp);
4026 #endif
4027 }
4028 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
4029