xref: /linux/mm/shmem.c (revision d6296cb65320be16dbf20f2fd584ddc25f3437cd)
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/fileattr.h>
32 #include <linux/mm.h>
33 #include <linux/random.h>
34 #include <linux/sched/signal.h>
35 #include <linux/export.h>
36 #include <linux/swap.h>
37 #include <linux/uio.h>
38 #include <linux/hugetlb.h>
39 #include <linux/fs_parser.h>
40 #include <linux/swapfile.h>
41 #include <linux/iversion.h>
42 #include "swap.h"
43 
44 static struct vfsmount *shm_mnt;
45 
46 #ifdef CONFIG_SHMEM
47 /*
48  * This virtual memory filesystem is heavily based on the ramfs. It
49  * extends ramfs by the ability to use swap and honor resource limits
50  * which makes it a completely usable filesystem.
51  */
52 
53 #include <linux/xattr.h>
54 #include <linux/exportfs.h>
55 #include <linux/posix_acl.h>
56 #include <linux/posix_acl_xattr.h>
57 #include <linux/mman.h>
58 #include <linux/string.h>
59 #include <linux/slab.h>
60 #include <linux/backing-dev.h>
61 #include <linux/shmem_fs.h>
62 #include <linux/writeback.h>
63 #include <linux/pagevec.h>
64 #include <linux/percpu_counter.h>
65 #include <linux/falloc.h>
66 #include <linux/splice.h>
67 #include <linux/security.h>
68 #include <linux/swapops.h>
69 #include <linux/mempolicy.h>
70 #include <linux/namei.h>
71 #include <linux/ctype.h>
72 #include <linux/migrate.h>
73 #include <linux/highmem.h>
74 #include <linux/seq_file.h>
75 #include <linux/magic.h>
76 #include <linux/syscalls.h>
77 #include <linux/fcntl.h>
78 #include <uapi/linux/memfd.h>
79 #include <linux/userfaultfd_k.h>
80 #include <linux/rmap.h>
81 #include <linux/uuid.h>
82 
83 #include <linux/uaccess.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_rwsem making sure that it has only one user at
99  * a time): we would prefer not to enlarge the shmem inode just for that.
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 struct shmem_options {
110 	unsigned long long blocks;
111 	unsigned long long inodes;
112 	struct mempolicy *mpol;
113 	kuid_t uid;
114 	kgid_t gid;
115 	umode_t mode;
116 	bool full_inums;
117 	int huge;
118 	int seen;
119 #define SHMEM_SEEN_BLOCKS 1
120 #define SHMEM_SEEN_INODES 2
121 #define SHMEM_SEEN_HUGE 4
122 #define SHMEM_SEEN_INUMS 8
123 };
124 
125 #ifdef CONFIG_TMPFS
126 static unsigned long shmem_default_max_blocks(void)
127 {
128 	return totalram_pages() / 2;
129 }
130 
131 static unsigned long shmem_default_max_inodes(void)
132 {
133 	unsigned long nr_pages = totalram_pages();
134 
135 	return min(nr_pages - totalhigh_pages(), nr_pages / 2);
136 }
137 #endif
138 
139 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
140 			     struct folio **foliop, enum sgp_type sgp,
141 			     gfp_t gfp, struct vm_area_struct *vma,
142 			     vm_fault_t *fault_type);
143 
144 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
145 {
146 	return sb->s_fs_info;
147 }
148 
149 /*
150  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
151  * for shared memory and for shared anonymous (/dev/zero) mappings
152  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
153  * consistent with the pre-accounting of private mappings ...
154  */
155 static inline int shmem_acct_size(unsigned long flags, loff_t size)
156 {
157 	return (flags & VM_NORESERVE) ?
158 		0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
159 }
160 
161 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
162 {
163 	if (!(flags & VM_NORESERVE))
164 		vm_unacct_memory(VM_ACCT(size));
165 }
166 
167 static inline int shmem_reacct_size(unsigned long flags,
168 		loff_t oldsize, loff_t newsize)
169 {
170 	if (!(flags & VM_NORESERVE)) {
171 		if (VM_ACCT(newsize) > VM_ACCT(oldsize))
172 			return security_vm_enough_memory_mm(current->mm,
173 					VM_ACCT(newsize) - VM_ACCT(oldsize));
174 		else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
175 			vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
176 	}
177 	return 0;
178 }
179 
180 /*
181  * ... whereas tmpfs objects are accounted incrementally as
182  * pages are allocated, in order to allow large sparse files.
183  * shmem_get_folio reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
184  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
185  */
186 static inline int shmem_acct_block(unsigned long flags, long pages)
187 {
188 	if (!(flags & VM_NORESERVE))
189 		return 0;
190 
191 	return security_vm_enough_memory_mm(current->mm,
192 			pages * VM_ACCT(PAGE_SIZE));
193 }
194 
195 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
196 {
197 	if (flags & VM_NORESERVE)
198 		vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
199 }
200 
201 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
202 {
203 	struct shmem_inode_info *info = SHMEM_I(inode);
204 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
205 
206 	if (shmem_acct_block(info->flags, pages))
207 		return false;
208 
209 	if (sbinfo->max_blocks) {
210 		if (percpu_counter_compare(&sbinfo->used_blocks,
211 					   sbinfo->max_blocks - pages) > 0)
212 			goto unacct;
213 		percpu_counter_add(&sbinfo->used_blocks, pages);
214 	}
215 
216 	return true;
217 
218 unacct:
219 	shmem_unacct_blocks(info->flags, pages);
220 	return false;
221 }
222 
223 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
224 {
225 	struct shmem_inode_info *info = SHMEM_I(inode);
226 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
227 
228 	if (sbinfo->max_blocks)
229 		percpu_counter_sub(&sbinfo->used_blocks, pages);
230 	shmem_unacct_blocks(info->flags, pages);
231 }
232 
233 static const struct super_operations shmem_ops;
234 const struct address_space_operations shmem_aops;
235 static const struct file_operations shmem_file_operations;
236 static const struct inode_operations shmem_inode_operations;
237 static const struct inode_operations shmem_dir_inode_operations;
238 static const struct inode_operations shmem_special_inode_operations;
239 static const struct vm_operations_struct shmem_vm_ops;
240 static const struct vm_operations_struct shmem_anon_vm_ops;
241 static struct file_system_type shmem_fs_type;
242 
243 bool vma_is_anon_shmem(struct vm_area_struct *vma)
244 {
245 	return vma->vm_ops == &shmem_anon_vm_ops;
246 }
247 
248 bool vma_is_shmem(struct vm_area_struct *vma)
249 {
250 	return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
251 }
252 
253 static LIST_HEAD(shmem_swaplist);
254 static DEFINE_MUTEX(shmem_swaplist_mutex);
255 
256 /*
257  * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
258  * produces a novel ino for the newly allocated inode.
259  *
260  * It may also be called when making a hard link to permit the space needed by
261  * each dentry. However, in that case, no new inode number is needed since that
262  * internally draws from another pool of inode numbers (currently global
263  * get_next_ino()). This case is indicated by passing NULL as inop.
264  */
265 #define SHMEM_INO_BATCH 1024
266 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
267 {
268 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
269 	ino_t ino;
270 
271 	if (!(sb->s_flags & SB_KERNMOUNT)) {
272 		raw_spin_lock(&sbinfo->stat_lock);
273 		if (sbinfo->max_inodes) {
274 			if (!sbinfo->free_inodes) {
275 				raw_spin_unlock(&sbinfo->stat_lock);
276 				return -ENOSPC;
277 			}
278 			sbinfo->free_inodes--;
279 		}
280 		if (inop) {
281 			ino = sbinfo->next_ino++;
282 			if (unlikely(is_zero_ino(ino)))
283 				ino = sbinfo->next_ino++;
284 			if (unlikely(!sbinfo->full_inums &&
285 				     ino > UINT_MAX)) {
286 				/*
287 				 * Emulate get_next_ino uint wraparound for
288 				 * compatibility
289 				 */
290 				if (IS_ENABLED(CONFIG_64BIT))
291 					pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
292 						__func__, MINOR(sb->s_dev));
293 				sbinfo->next_ino = 1;
294 				ino = sbinfo->next_ino++;
295 			}
296 			*inop = ino;
297 		}
298 		raw_spin_unlock(&sbinfo->stat_lock);
299 	} else if (inop) {
300 		/*
301 		 * __shmem_file_setup, one of our callers, is lock-free: it
302 		 * doesn't hold stat_lock in shmem_reserve_inode since
303 		 * max_inodes is always 0, and is called from potentially
304 		 * unknown contexts. As such, use a per-cpu batched allocator
305 		 * which doesn't require the per-sb stat_lock unless we are at
306 		 * the batch boundary.
307 		 *
308 		 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
309 		 * shmem mounts are not exposed to userspace, so we don't need
310 		 * to worry about things like glibc compatibility.
311 		 */
312 		ino_t *next_ino;
313 
314 		next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
315 		ino = *next_ino;
316 		if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
317 			raw_spin_lock(&sbinfo->stat_lock);
318 			ino = sbinfo->next_ino;
319 			sbinfo->next_ino += SHMEM_INO_BATCH;
320 			raw_spin_unlock(&sbinfo->stat_lock);
321 			if (unlikely(is_zero_ino(ino)))
322 				ino++;
323 		}
324 		*inop = ino;
325 		*next_ino = ++ino;
326 		put_cpu();
327 	}
328 
329 	return 0;
330 }
331 
332 static void shmem_free_inode(struct super_block *sb)
333 {
334 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
335 	if (sbinfo->max_inodes) {
336 		raw_spin_lock(&sbinfo->stat_lock);
337 		sbinfo->free_inodes++;
338 		raw_spin_unlock(&sbinfo->stat_lock);
339 	}
340 }
341 
342 /**
343  * shmem_recalc_inode - recalculate the block usage of an inode
344  * @inode: inode to recalc
345  *
346  * We have to calculate the free blocks since the mm can drop
347  * undirtied hole pages behind our back.
348  *
349  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
350  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
351  *
352  * It has to be called with the spinlock held.
353  */
354 static void shmem_recalc_inode(struct inode *inode)
355 {
356 	struct shmem_inode_info *info = SHMEM_I(inode);
357 	long freed;
358 
359 	freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
360 	if (freed > 0) {
361 		info->alloced -= freed;
362 		inode->i_blocks -= freed * BLOCKS_PER_PAGE;
363 		shmem_inode_unacct_blocks(inode, freed);
364 	}
365 }
366 
367 bool shmem_charge(struct inode *inode, long pages)
368 {
369 	struct shmem_inode_info *info = SHMEM_I(inode);
370 	unsigned long flags;
371 
372 	if (!shmem_inode_acct_block(inode, pages))
373 		return false;
374 
375 	/* nrpages adjustment first, then shmem_recalc_inode() when balanced */
376 	inode->i_mapping->nrpages += pages;
377 
378 	spin_lock_irqsave(&info->lock, flags);
379 	info->alloced += pages;
380 	inode->i_blocks += pages * BLOCKS_PER_PAGE;
381 	shmem_recalc_inode(inode);
382 	spin_unlock_irqrestore(&info->lock, flags);
383 
384 	return true;
385 }
386 
387 void shmem_uncharge(struct inode *inode, long pages)
388 {
389 	struct shmem_inode_info *info = SHMEM_I(inode);
390 	unsigned long flags;
391 
392 	/* nrpages adjustment done by __filemap_remove_folio() or caller */
393 
394 	spin_lock_irqsave(&info->lock, flags);
395 	info->alloced -= pages;
396 	inode->i_blocks -= pages * BLOCKS_PER_PAGE;
397 	shmem_recalc_inode(inode);
398 	spin_unlock_irqrestore(&info->lock, flags);
399 
400 	shmem_inode_unacct_blocks(inode, pages);
401 }
402 
403 /*
404  * Replace item expected in xarray by a new item, while holding xa_lock.
405  */
406 static int shmem_replace_entry(struct address_space *mapping,
407 			pgoff_t index, void *expected, void *replacement)
408 {
409 	XA_STATE(xas, &mapping->i_pages, index);
410 	void *item;
411 
412 	VM_BUG_ON(!expected);
413 	VM_BUG_ON(!replacement);
414 	item = xas_load(&xas);
415 	if (item != expected)
416 		return -ENOENT;
417 	xas_store(&xas, replacement);
418 	return 0;
419 }
420 
421 /*
422  * Sometimes, before we decide whether to proceed or to fail, we must check
423  * that an entry was not already brought back from swap by a racing thread.
424  *
425  * Checking page is not enough: by the time a SwapCache page is locked, it
426  * might be reused, and again be SwapCache, using the same swap as before.
427  */
428 static bool shmem_confirm_swap(struct address_space *mapping,
429 			       pgoff_t index, swp_entry_t swap)
430 {
431 	return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
432 }
433 
434 /*
435  * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
436  *
437  * SHMEM_HUGE_NEVER:
438  *	disables huge pages for the mount;
439  * SHMEM_HUGE_ALWAYS:
440  *	enables huge pages for the mount;
441  * SHMEM_HUGE_WITHIN_SIZE:
442  *	only allocate huge pages if the page will be fully within i_size,
443  *	also respect fadvise()/madvise() hints;
444  * SHMEM_HUGE_ADVISE:
445  *	only allocate huge pages if requested with fadvise()/madvise();
446  */
447 
448 #define SHMEM_HUGE_NEVER	0
449 #define SHMEM_HUGE_ALWAYS	1
450 #define SHMEM_HUGE_WITHIN_SIZE	2
451 #define SHMEM_HUGE_ADVISE	3
452 
453 /*
454  * Special values.
455  * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
456  *
457  * SHMEM_HUGE_DENY:
458  *	disables huge on shm_mnt and all mounts, for emergency use;
459  * SHMEM_HUGE_FORCE:
460  *	enables huge on shm_mnt and all mounts, w/o needing option, for testing;
461  *
462  */
463 #define SHMEM_HUGE_DENY		(-1)
464 #define SHMEM_HUGE_FORCE	(-2)
465 
466 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
467 /* ifdef here to avoid bloating shmem.o when not necessary */
468 
469 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
470 
471 bool shmem_is_huge(struct vm_area_struct *vma, struct inode *inode,
472 		   pgoff_t index, bool shmem_huge_force)
473 {
474 	loff_t i_size;
475 
476 	if (!S_ISREG(inode->i_mode))
477 		return false;
478 	if (vma && ((vma->vm_flags & VM_NOHUGEPAGE) ||
479 	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
480 		return false;
481 	if (shmem_huge == SHMEM_HUGE_DENY)
482 		return false;
483 	if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
484 		return true;
485 
486 	switch (SHMEM_SB(inode->i_sb)->huge) {
487 	case SHMEM_HUGE_ALWAYS:
488 		return true;
489 	case SHMEM_HUGE_WITHIN_SIZE:
490 		index = round_up(index + 1, HPAGE_PMD_NR);
491 		i_size = round_up(i_size_read(inode), PAGE_SIZE);
492 		if (i_size >> PAGE_SHIFT >= index)
493 			return true;
494 		fallthrough;
495 	case SHMEM_HUGE_ADVISE:
496 		if (vma && (vma->vm_flags & VM_HUGEPAGE))
497 			return true;
498 		fallthrough;
499 	default:
500 		return false;
501 	}
502 }
503 
504 #if defined(CONFIG_SYSFS)
505 static int shmem_parse_huge(const char *str)
506 {
507 	if (!strcmp(str, "never"))
508 		return SHMEM_HUGE_NEVER;
509 	if (!strcmp(str, "always"))
510 		return SHMEM_HUGE_ALWAYS;
511 	if (!strcmp(str, "within_size"))
512 		return SHMEM_HUGE_WITHIN_SIZE;
513 	if (!strcmp(str, "advise"))
514 		return SHMEM_HUGE_ADVISE;
515 	if (!strcmp(str, "deny"))
516 		return SHMEM_HUGE_DENY;
517 	if (!strcmp(str, "force"))
518 		return SHMEM_HUGE_FORCE;
519 	return -EINVAL;
520 }
521 #endif
522 
523 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
524 static const char *shmem_format_huge(int huge)
525 {
526 	switch (huge) {
527 	case SHMEM_HUGE_NEVER:
528 		return "never";
529 	case SHMEM_HUGE_ALWAYS:
530 		return "always";
531 	case SHMEM_HUGE_WITHIN_SIZE:
532 		return "within_size";
533 	case SHMEM_HUGE_ADVISE:
534 		return "advise";
535 	case SHMEM_HUGE_DENY:
536 		return "deny";
537 	case SHMEM_HUGE_FORCE:
538 		return "force";
539 	default:
540 		VM_BUG_ON(1);
541 		return "bad_val";
542 	}
543 }
544 #endif
545 
546 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
547 		struct shrink_control *sc, unsigned long nr_to_split)
548 {
549 	LIST_HEAD(list), *pos, *next;
550 	LIST_HEAD(to_remove);
551 	struct inode *inode;
552 	struct shmem_inode_info *info;
553 	struct folio *folio;
554 	unsigned long batch = sc ? sc->nr_to_scan : 128;
555 	int split = 0;
556 
557 	if (list_empty(&sbinfo->shrinklist))
558 		return SHRINK_STOP;
559 
560 	spin_lock(&sbinfo->shrinklist_lock);
561 	list_for_each_safe(pos, next, &sbinfo->shrinklist) {
562 		info = list_entry(pos, struct shmem_inode_info, shrinklist);
563 
564 		/* pin the inode */
565 		inode = igrab(&info->vfs_inode);
566 
567 		/* inode is about to be evicted */
568 		if (!inode) {
569 			list_del_init(&info->shrinklist);
570 			goto next;
571 		}
572 
573 		/* Check if there's anything to gain */
574 		if (round_up(inode->i_size, PAGE_SIZE) ==
575 				round_up(inode->i_size, HPAGE_PMD_SIZE)) {
576 			list_move(&info->shrinklist, &to_remove);
577 			goto next;
578 		}
579 
580 		list_move(&info->shrinklist, &list);
581 next:
582 		sbinfo->shrinklist_len--;
583 		if (!--batch)
584 			break;
585 	}
586 	spin_unlock(&sbinfo->shrinklist_lock);
587 
588 	list_for_each_safe(pos, next, &to_remove) {
589 		info = list_entry(pos, struct shmem_inode_info, shrinklist);
590 		inode = &info->vfs_inode;
591 		list_del_init(&info->shrinklist);
592 		iput(inode);
593 	}
594 
595 	list_for_each_safe(pos, next, &list) {
596 		int ret;
597 		pgoff_t index;
598 
599 		info = list_entry(pos, struct shmem_inode_info, shrinklist);
600 		inode = &info->vfs_inode;
601 
602 		if (nr_to_split && split >= nr_to_split)
603 			goto move_back;
604 
605 		index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT;
606 		folio = filemap_get_folio(inode->i_mapping, index);
607 		if (!folio)
608 			goto drop;
609 
610 		/* No huge page at the end of the file: nothing to split */
611 		if (!folio_test_large(folio)) {
612 			folio_put(folio);
613 			goto drop;
614 		}
615 
616 		/*
617 		 * Move the inode on the list back to shrinklist if we failed
618 		 * to lock the page at this time.
619 		 *
620 		 * Waiting for the lock may lead to deadlock in the
621 		 * reclaim path.
622 		 */
623 		if (!folio_trylock(folio)) {
624 			folio_put(folio);
625 			goto move_back;
626 		}
627 
628 		ret = split_folio(folio);
629 		folio_unlock(folio);
630 		folio_put(folio);
631 
632 		/* If split failed move the inode on the list back to shrinklist */
633 		if (ret)
634 			goto move_back;
635 
636 		split++;
637 drop:
638 		list_del_init(&info->shrinklist);
639 		goto put;
640 move_back:
641 		/*
642 		 * Make sure the inode is either on the global list or deleted
643 		 * from any local list before iput() since it could be deleted
644 		 * in another thread once we put the inode (then the local list
645 		 * is corrupted).
646 		 */
647 		spin_lock(&sbinfo->shrinklist_lock);
648 		list_move(&info->shrinklist, &sbinfo->shrinklist);
649 		sbinfo->shrinklist_len++;
650 		spin_unlock(&sbinfo->shrinklist_lock);
651 put:
652 		iput(inode);
653 	}
654 
655 	return split;
656 }
657 
658 static long shmem_unused_huge_scan(struct super_block *sb,
659 		struct shrink_control *sc)
660 {
661 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
662 
663 	if (!READ_ONCE(sbinfo->shrinklist_len))
664 		return SHRINK_STOP;
665 
666 	return shmem_unused_huge_shrink(sbinfo, sc, 0);
667 }
668 
669 static long shmem_unused_huge_count(struct super_block *sb,
670 		struct shrink_control *sc)
671 {
672 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
673 	return READ_ONCE(sbinfo->shrinklist_len);
674 }
675 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
676 
677 #define shmem_huge SHMEM_HUGE_DENY
678 
679 bool shmem_is_huge(struct vm_area_struct *vma, struct inode *inode,
680 		   pgoff_t index, bool shmem_huge_force)
681 {
682 	return false;
683 }
684 
685 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
686 		struct shrink_control *sc, unsigned long nr_to_split)
687 {
688 	return 0;
689 }
690 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
691 
692 /*
693  * Like filemap_add_folio, but error if expected item has gone.
694  */
695 static int shmem_add_to_page_cache(struct folio *folio,
696 				   struct address_space *mapping,
697 				   pgoff_t index, void *expected, gfp_t gfp,
698 				   struct mm_struct *charge_mm)
699 {
700 	XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
701 	long nr = folio_nr_pages(folio);
702 	int error;
703 
704 	VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
705 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
706 	VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
707 	VM_BUG_ON(expected && folio_test_large(folio));
708 
709 	folio_ref_add(folio, nr);
710 	folio->mapping = mapping;
711 	folio->index = index;
712 
713 	if (!folio_test_swapcache(folio)) {
714 		error = mem_cgroup_charge(folio, charge_mm, gfp);
715 		if (error) {
716 			if (folio_test_pmd_mappable(folio)) {
717 				count_vm_event(THP_FILE_FALLBACK);
718 				count_vm_event(THP_FILE_FALLBACK_CHARGE);
719 			}
720 			goto error;
721 		}
722 	}
723 	folio_throttle_swaprate(folio, gfp);
724 
725 	do {
726 		xas_lock_irq(&xas);
727 		if (expected != xas_find_conflict(&xas)) {
728 			xas_set_err(&xas, -EEXIST);
729 			goto unlock;
730 		}
731 		if (expected && xas_find_conflict(&xas)) {
732 			xas_set_err(&xas, -EEXIST);
733 			goto unlock;
734 		}
735 		xas_store(&xas, folio);
736 		if (xas_error(&xas))
737 			goto unlock;
738 		if (folio_test_pmd_mappable(folio)) {
739 			count_vm_event(THP_FILE_ALLOC);
740 			__lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
741 		}
742 		mapping->nrpages += nr;
743 		__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
744 		__lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
745 unlock:
746 		xas_unlock_irq(&xas);
747 	} while (xas_nomem(&xas, gfp));
748 
749 	if (xas_error(&xas)) {
750 		error = xas_error(&xas);
751 		goto error;
752 	}
753 
754 	return 0;
755 error:
756 	folio->mapping = NULL;
757 	folio_ref_sub(folio, nr);
758 	return error;
759 }
760 
761 /*
762  * Like delete_from_page_cache, but substitutes swap for @folio.
763  */
764 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
765 {
766 	struct address_space *mapping = folio->mapping;
767 	long nr = folio_nr_pages(folio);
768 	int error;
769 
770 	xa_lock_irq(&mapping->i_pages);
771 	error = shmem_replace_entry(mapping, folio->index, folio, radswap);
772 	folio->mapping = NULL;
773 	mapping->nrpages -= nr;
774 	__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
775 	__lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
776 	xa_unlock_irq(&mapping->i_pages);
777 	folio_put(folio);
778 	BUG_ON(error);
779 }
780 
781 /*
782  * Remove swap entry from page cache, free the swap and its page cache.
783  */
784 static int shmem_free_swap(struct address_space *mapping,
785 			   pgoff_t index, void *radswap)
786 {
787 	void *old;
788 
789 	old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
790 	if (old != radswap)
791 		return -ENOENT;
792 	free_swap_and_cache(radix_to_swp_entry(radswap));
793 	return 0;
794 }
795 
796 /*
797  * Determine (in bytes) how many of the shmem object's pages mapped by the
798  * given offsets are swapped out.
799  *
800  * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
801  * as long as the inode doesn't go away and racy results are not a problem.
802  */
803 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
804 						pgoff_t start, pgoff_t end)
805 {
806 	XA_STATE(xas, &mapping->i_pages, start);
807 	struct page *page;
808 	unsigned long swapped = 0;
809 
810 	rcu_read_lock();
811 	xas_for_each(&xas, page, end - 1) {
812 		if (xas_retry(&xas, page))
813 			continue;
814 		if (xa_is_value(page))
815 			swapped++;
816 
817 		if (need_resched()) {
818 			xas_pause(&xas);
819 			cond_resched_rcu();
820 		}
821 	}
822 
823 	rcu_read_unlock();
824 
825 	return swapped << PAGE_SHIFT;
826 }
827 
828 /*
829  * Determine (in bytes) how many of the shmem object's pages mapped by the
830  * given vma is swapped out.
831  *
832  * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
833  * as long as the inode doesn't go away and racy results are not a problem.
834  */
835 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
836 {
837 	struct inode *inode = file_inode(vma->vm_file);
838 	struct shmem_inode_info *info = SHMEM_I(inode);
839 	struct address_space *mapping = inode->i_mapping;
840 	unsigned long swapped;
841 
842 	/* Be careful as we don't hold info->lock */
843 	swapped = READ_ONCE(info->swapped);
844 
845 	/*
846 	 * The easier cases are when the shmem object has nothing in swap, or
847 	 * the vma maps it whole. Then we can simply use the stats that we
848 	 * already track.
849 	 */
850 	if (!swapped)
851 		return 0;
852 
853 	if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
854 		return swapped << PAGE_SHIFT;
855 
856 	/* Here comes the more involved part */
857 	return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
858 					vma->vm_pgoff + vma_pages(vma));
859 }
860 
861 /*
862  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
863  */
864 void shmem_unlock_mapping(struct address_space *mapping)
865 {
866 	struct folio_batch fbatch;
867 	pgoff_t index = 0;
868 
869 	folio_batch_init(&fbatch);
870 	/*
871 	 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
872 	 */
873 	while (!mapping_unevictable(mapping) &&
874 	       filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
875 		check_move_unevictable_folios(&fbatch);
876 		folio_batch_release(&fbatch);
877 		cond_resched();
878 	}
879 }
880 
881 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
882 {
883 	struct folio *folio;
884 
885 	/*
886 	 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
887 	 * beyond i_size, and reports fallocated pages as holes.
888 	 */
889 	folio = __filemap_get_folio(inode->i_mapping, index,
890 					FGP_ENTRY | FGP_LOCK, 0);
891 	if (!xa_is_value(folio))
892 		return folio;
893 	/*
894 	 * But read a page back from swap if any of it is within i_size
895 	 * (although in some cases this is just a waste of time).
896 	 */
897 	folio = NULL;
898 	shmem_get_folio(inode, index, &folio, SGP_READ);
899 	return folio;
900 }
901 
902 /*
903  * Remove range of pages and swap entries from page cache, and free them.
904  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
905  */
906 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
907 								 bool unfalloc)
908 {
909 	struct address_space *mapping = inode->i_mapping;
910 	struct shmem_inode_info *info = SHMEM_I(inode);
911 	pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
912 	pgoff_t end = (lend + 1) >> PAGE_SHIFT;
913 	struct folio_batch fbatch;
914 	pgoff_t indices[PAGEVEC_SIZE];
915 	struct folio *folio;
916 	bool same_folio;
917 	long nr_swaps_freed = 0;
918 	pgoff_t index;
919 	int i;
920 
921 	if (lend == -1)
922 		end = -1;	/* unsigned, so actually very big */
923 
924 	if (info->fallocend > start && info->fallocend <= end && !unfalloc)
925 		info->fallocend = start;
926 
927 	folio_batch_init(&fbatch);
928 	index = start;
929 	while (index < end && find_lock_entries(mapping, &index, end - 1,
930 			&fbatch, indices)) {
931 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
932 			folio = fbatch.folios[i];
933 
934 			if (xa_is_value(folio)) {
935 				if (unfalloc)
936 					continue;
937 				nr_swaps_freed += !shmem_free_swap(mapping,
938 							indices[i], folio);
939 				continue;
940 			}
941 
942 			if (!unfalloc || !folio_test_uptodate(folio))
943 				truncate_inode_folio(mapping, folio);
944 			folio_unlock(folio);
945 		}
946 		folio_batch_remove_exceptionals(&fbatch);
947 		folio_batch_release(&fbatch);
948 		cond_resched();
949 	}
950 
951 	/*
952 	 * When undoing a failed fallocate, we want none of the partial folio
953 	 * zeroing and splitting below, but shall want to truncate the whole
954 	 * folio when !uptodate indicates that it was added by this fallocate,
955 	 * even when [lstart, lend] covers only a part of the folio.
956 	 */
957 	if (unfalloc)
958 		goto whole_folios;
959 
960 	same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
961 	folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
962 	if (folio) {
963 		same_folio = lend < folio_pos(folio) + folio_size(folio);
964 		folio_mark_dirty(folio);
965 		if (!truncate_inode_partial_folio(folio, lstart, lend)) {
966 			start = folio->index + folio_nr_pages(folio);
967 			if (same_folio)
968 				end = folio->index;
969 		}
970 		folio_unlock(folio);
971 		folio_put(folio);
972 		folio = NULL;
973 	}
974 
975 	if (!same_folio)
976 		folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
977 	if (folio) {
978 		folio_mark_dirty(folio);
979 		if (!truncate_inode_partial_folio(folio, lstart, lend))
980 			end = folio->index;
981 		folio_unlock(folio);
982 		folio_put(folio);
983 	}
984 
985 whole_folios:
986 
987 	index = start;
988 	while (index < end) {
989 		cond_resched();
990 
991 		if (!find_get_entries(mapping, &index, end - 1, &fbatch,
992 				indices)) {
993 			/* If all gone or hole-punch or unfalloc, we're done */
994 			if (index == start || end != -1)
995 				break;
996 			/* But if truncating, restart to make sure all gone */
997 			index = start;
998 			continue;
999 		}
1000 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
1001 			folio = fbatch.folios[i];
1002 
1003 			if (xa_is_value(folio)) {
1004 				if (unfalloc)
1005 					continue;
1006 				if (shmem_free_swap(mapping, indices[i], folio)) {
1007 					/* Swap was replaced by page: retry */
1008 					index = indices[i];
1009 					break;
1010 				}
1011 				nr_swaps_freed++;
1012 				continue;
1013 			}
1014 
1015 			folio_lock(folio);
1016 
1017 			if (!unfalloc || !folio_test_uptodate(folio)) {
1018 				if (folio_mapping(folio) != mapping) {
1019 					/* Page was replaced by swap: retry */
1020 					folio_unlock(folio);
1021 					index = indices[i];
1022 					break;
1023 				}
1024 				VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1025 						folio);
1026 				truncate_inode_folio(mapping, folio);
1027 			}
1028 			folio_unlock(folio);
1029 		}
1030 		folio_batch_remove_exceptionals(&fbatch);
1031 		folio_batch_release(&fbatch);
1032 	}
1033 
1034 	spin_lock_irq(&info->lock);
1035 	info->swapped -= nr_swaps_freed;
1036 	shmem_recalc_inode(inode);
1037 	spin_unlock_irq(&info->lock);
1038 }
1039 
1040 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1041 {
1042 	shmem_undo_range(inode, lstart, lend, false);
1043 	inode->i_ctime = inode->i_mtime = current_time(inode);
1044 	inode_inc_iversion(inode);
1045 }
1046 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1047 
1048 static int shmem_getattr(struct mnt_idmap *idmap,
1049 			 const struct path *path, struct kstat *stat,
1050 			 u32 request_mask, unsigned int query_flags)
1051 {
1052 	struct inode *inode = path->dentry->d_inode;
1053 	struct shmem_inode_info *info = SHMEM_I(inode);
1054 
1055 	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1056 		spin_lock_irq(&info->lock);
1057 		shmem_recalc_inode(inode);
1058 		spin_unlock_irq(&info->lock);
1059 	}
1060 	if (info->fsflags & FS_APPEND_FL)
1061 		stat->attributes |= STATX_ATTR_APPEND;
1062 	if (info->fsflags & FS_IMMUTABLE_FL)
1063 		stat->attributes |= STATX_ATTR_IMMUTABLE;
1064 	if (info->fsflags & FS_NODUMP_FL)
1065 		stat->attributes |= STATX_ATTR_NODUMP;
1066 	stat->attributes_mask |= (STATX_ATTR_APPEND |
1067 			STATX_ATTR_IMMUTABLE |
1068 			STATX_ATTR_NODUMP);
1069 	generic_fillattr(idmap, inode, stat);
1070 
1071 	if (shmem_is_huge(NULL, inode, 0, false))
1072 		stat->blksize = HPAGE_PMD_SIZE;
1073 
1074 	if (request_mask & STATX_BTIME) {
1075 		stat->result_mask |= STATX_BTIME;
1076 		stat->btime.tv_sec = info->i_crtime.tv_sec;
1077 		stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1078 	}
1079 
1080 	return 0;
1081 }
1082 
1083 static int shmem_setattr(struct mnt_idmap *idmap,
1084 			 struct dentry *dentry, struct iattr *attr)
1085 {
1086 	struct inode *inode = d_inode(dentry);
1087 	struct shmem_inode_info *info = SHMEM_I(inode);
1088 	int error;
1089 	bool update_mtime = false;
1090 	bool update_ctime = true;
1091 
1092 	error = setattr_prepare(idmap, dentry, attr);
1093 	if (error)
1094 		return error;
1095 
1096 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1097 		loff_t oldsize = inode->i_size;
1098 		loff_t newsize = attr->ia_size;
1099 
1100 		/* protected by i_rwsem */
1101 		if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1102 		    (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1103 			return -EPERM;
1104 
1105 		if (newsize != oldsize) {
1106 			error = shmem_reacct_size(SHMEM_I(inode)->flags,
1107 					oldsize, newsize);
1108 			if (error)
1109 				return error;
1110 			i_size_write(inode, newsize);
1111 			update_mtime = true;
1112 		} else {
1113 			update_ctime = false;
1114 		}
1115 		if (newsize <= oldsize) {
1116 			loff_t holebegin = round_up(newsize, PAGE_SIZE);
1117 			if (oldsize > holebegin)
1118 				unmap_mapping_range(inode->i_mapping,
1119 							holebegin, 0, 1);
1120 			if (info->alloced)
1121 				shmem_truncate_range(inode,
1122 							newsize, (loff_t)-1);
1123 			/* unmap again to remove racily COWed private pages */
1124 			if (oldsize > holebegin)
1125 				unmap_mapping_range(inode->i_mapping,
1126 							holebegin, 0, 1);
1127 		}
1128 	}
1129 
1130 	setattr_copy(idmap, inode, attr);
1131 	if (attr->ia_valid & ATTR_MODE)
1132 		error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1133 	if (!error && update_ctime) {
1134 		inode->i_ctime = current_time(inode);
1135 		if (update_mtime)
1136 			inode->i_mtime = inode->i_ctime;
1137 		inode_inc_iversion(inode);
1138 	}
1139 	return error;
1140 }
1141 
1142 static void shmem_evict_inode(struct inode *inode)
1143 {
1144 	struct shmem_inode_info *info = SHMEM_I(inode);
1145 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1146 
1147 	if (shmem_mapping(inode->i_mapping)) {
1148 		shmem_unacct_size(info->flags, inode->i_size);
1149 		inode->i_size = 0;
1150 		mapping_set_exiting(inode->i_mapping);
1151 		shmem_truncate_range(inode, 0, (loff_t)-1);
1152 		if (!list_empty(&info->shrinklist)) {
1153 			spin_lock(&sbinfo->shrinklist_lock);
1154 			if (!list_empty(&info->shrinklist)) {
1155 				list_del_init(&info->shrinklist);
1156 				sbinfo->shrinklist_len--;
1157 			}
1158 			spin_unlock(&sbinfo->shrinklist_lock);
1159 		}
1160 		while (!list_empty(&info->swaplist)) {
1161 			/* Wait while shmem_unuse() is scanning this inode... */
1162 			wait_var_event(&info->stop_eviction,
1163 				       !atomic_read(&info->stop_eviction));
1164 			mutex_lock(&shmem_swaplist_mutex);
1165 			/* ...but beware of the race if we peeked too early */
1166 			if (!atomic_read(&info->stop_eviction))
1167 				list_del_init(&info->swaplist);
1168 			mutex_unlock(&shmem_swaplist_mutex);
1169 		}
1170 	}
1171 
1172 	simple_xattrs_free(&info->xattrs);
1173 	WARN_ON(inode->i_blocks);
1174 	shmem_free_inode(inode->i_sb);
1175 	clear_inode(inode);
1176 }
1177 
1178 static int shmem_find_swap_entries(struct address_space *mapping,
1179 				   pgoff_t start, struct folio_batch *fbatch,
1180 				   pgoff_t *indices, unsigned int type)
1181 {
1182 	XA_STATE(xas, &mapping->i_pages, start);
1183 	struct folio *folio;
1184 	swp_entry_t entry;
1185 
1186 	rcu_read_lock();
1187 	xas_for_each(&xas, folio, ULONG_MAX) {
1188 		if (xas_retry(&xas, folio))
1189 			continue;
1190 
1191 		if (!xa_is_value(folio))
1192 			continue;
1193 
1194 		entry = radix_to_swp_entry(folio);
1195 		/*
1196 		 * swapin error entries can be found in the mapping. But they're
1197 		 * deliberately ignored here as we've done everything we can do.
1198 		 */
1199 		if (swp_type(entry) != type)
1200 			continue;
1201 
1202 		indices[folio_batch_count(fbatch)] = xas.xa_index;
1203 		if (!folio_batch_add(fbatch, folio))
1204 			break;
1205 
1206 		if (need_resched()) {
1207 			xas_pause(&xas);
1208 			cond_resched_rcu();
1209 		}
1210 	}
1211 	rcu_read_unlock();
1212 
1213 	return xas.xa_index;
1214 }
1215 
1216 /*
1217  * Move the swapped pages for an inode to page cache. Returns the count
1218  * of pages swapped in, or the error in case of failure.
1219  */
1220 static int shmem_unuse_swap_entries(struct inode *inode,
1221 		struct folio_batch *fbatch, pgoff_t *indices)
1222 {
1223 	int i = 0;
1224 	int ret = 0;
1225 	int error = 0;
1226 	struct address_space *mapping = inode->i_mapping;
1227 
1228 	for (i = 0; i < folio_batch_count(fbatch); i++) {
1229 		struct folio *folio = fbatch->folios[i];
1230 
1231 		if (!xa_is_value(folio))
1232 			continue;
1233 		error = shmem_swapin_folio(inode, indices[i],
1234 					  &folio, SGP_CACHE,
1235 					  mapping_gfp_mask(mapping),
1236 					  NULL, NULL);
1237 		if (error == 0) {
1238 			folio_unlock(folio);
1239 			folio_put(folio);
1240 			ret++;
1241 		}
1242 		if (error == -ENOMEM)
1243 			break;
1244 		error = 0;
1245 	}
1246 	return error ? error : ret;
1247 }
1248 
1249 /*
1250  * If swap found in inode, free it and move page from swapcache to filecache.
1251  */
1252 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1253 {
1254 	struct address_space *mapping = inode->i_mapping;
1255 	pgoff_t start = 0;
1256 	struct folio_batch fbatch;
1257 	pgoff_t indices[PAGEVEC_SIZE];
1258 	int ret = 0;
1259 
1260 	do {
1261 		folio_batch_init(&fbatch);
1262 		shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1263 		if (folio_batch_count(&fbatch) == 0) {
1264 			ret = 0;
1265 			break;
1266 		}
1267 
1268 		ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1269 		if (ret < 0)
1270 			break;
1271 
1272 		start = indices[folio_batch_count(&fbatch) - 1];
1273 	} while (true);
1274 
1275 	return ret;
1276 }
1277 
1278 /*
1279  * Read all the shared memory data that resides in the swap
1280  * device 'type' back into memory, so the swap device can be
1281  * unused.
1282  */
1283 int shmem_unuse(unsigned int type)
1284 {
1285 	struct shmem_inode_info *info, *next;
1286 	int error = 0;
1287 
1288 	if (list_empty(&shmem_swaplist))
1289 		return 0;
1290 
1291 	mutex_lock(&shmem_swaplist_mutex);
1292 	list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1293 		if (!info->swapped) {
1294 			list_del_init(&info->swaplist);
1295 			continue;
1296 		}
1297 		/*
1298 		 * Drop the swaplist mutex while searching the inode for swap;
1299 		 * but before doing so, make sure shmem_evict_inode() will not
1300 		 * remove placeholder inode from swaplist, nor let it be freed
1301 		 * (igrab() would protect from unlink, but not from unmount).
1302 		 */
1303 		atomic_inc(&info->stop_eviction);
1304 		mutex_unlock(&shmem_swaplist_mutex);
1305 
1306 		error = shmem_unuse_inode(&info->vfs_inode, type);
1307 		cond_resched();
1308 
1309 		mutex_lock(&shmem_swaplist_mutex);
1310 		next = list_next_entry(info, swaplist);
1311 		if (!info->swapped)
1312 			list_del_init(&info->swaplist);
1313 		if (atomic_dec_and_test(&info->stop_eviction))
1314 			wake_up_var(&info->stop_eviction);
1315 		if (error)
1316 			break;
1317 	}
1318 	mutex_unlock(&shmem_swaplist_mutex);
1319 
1320 	return error;
1321 }
1322 
1323 /*
1324  * Move the page from the page cache to the swap cache.
1325  */
1326 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1327 {
1328 	struct folio *folio = page_folio(page);
1329 	struct shmem_inode_info *info;
1330 	struct address_space *mapping;
1331 	struct inode *inode;
1332 	swp_entry_t swap;
1333 	pgoff_t index;
1334 
1335 	/*
1336 	 * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or
1337 	 * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages,
1338 	 * and its shmem_writeback() needs them to be split when swapping.
1339 	 */
1340 	if (folio_test_large(folio)) {
1341 		/* Ensure the subpages are still dirty */
1342 		folio_test_set_dirty(folio);
1343 		if (split_huge_page(page) < 0)
1344 			goto redirty;
1345 		folio = page_folio(page);
1346 		folio_clear_dirty(folio);
1347 	}
1348 
1349 	BUG_ON(!folio_test_locked(folio));
1350 	mapping = folio->mapping;
1351 	index = folio->index;
1352 	inode = mapping->host;
1353 	info = SHMEM_I(inode);
1354 	if (info->flags & VM_LOCKED)
1355 		goto redirty;
1356 	if (!total_swap_pages)
1357 		goto redirty;
1358 
1359 	/*
1360 	 * Our capabilities prevent regular writeback or sync from ever calling
1361 	 * shmem_writepage; but a stacking filesystem might use ->writepage of
1362 	 * its underlying filesystem, in which case tmpfs should write out to
1363 	 * swap only in response to memory pressure, and not for the writeback
1364 	 * threads or sync.
1365 	 */
1366 	if (!wbc->for_reclaim) {
1367 		WARN_ON_ONCE(1);	/* Still happens? Tell us about it! */
1368 		goto redirty;
1369 	}
1370 
1371 	/*
1372 	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1373 	 * value into swapfile.c, the only way we can correctly account for a
1374 	 * fallocated folio arriving here is now to initialize it and write it.
1375 	 *
1376 	 * That's okay for a folio already fallocated earlier, but if we have
1377 	 * not yet completed the fallocation, then (a) we want to keep track
1378 	 * of this folio in case we have to undo it, and (b) it may not be a
1379 	 * good idea to continue anyway, once we're pushing into swap.  So
1380 	 * reactivate the folio, and let shmem_fallocate() quit when too many.
1381 	 */
1382 	if (!folio_test_uptodate(folio)) {
1383 		if (inode->i_private) {
1384 			struct shmem_falloc *shmem_falloc;
1385 			spin_lock(&inode->i_lock);
1386 			shmem_falloc = inode->i_private;
1387 			if (shmem_falloc &&
1388 			    !shmem_falloc->waitq &&
1389 			    index >= shmem_falloc->start &&
1390 			    index < shmem_falloc->next)
1391 				shmem_falloc->nr_unswapped++;
1392 			else
1393 				shmem_falloc = NULL;
1394 			spin_unlock(&inode->i_lock);
1395 			if (shmem_falloc)
1396 				goto redirty;
1397 		}
1398 		folio_zero_range(folio, 0, folio_size(folio));
1399 		flush_dcache_folio(folio);
1400 		folio_mark_uptodate(folio);
1401 	}
1402 
1403 	swap = folio_alloc_swap(folio);
1404 	if (!swap.val)
1405 		goto redirty;
1406 
1407 	/*
1408 	 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1409 	 * if it's not already there.  Do it now before the folio is
1410 	 * moved to swap cache, when its pagelock no longer protects
1411 	 * the inode from eviction.  But don't unlock the mutex until
1412 	 * we've incremented swapped, because shmem_unuse_inode() will
1413 	 * prune a !swapped inode from the swaplist under this mutex.
1414 	 */
1415 	mutex_lock(&shmem_swaplist_mutex);
1416 	if (list_empty(&info->swaplist))
1417 		list_add(&info->swaplist, &shmem_swaplist);
1418 
1419 	if (add_to_swap_cache(folio, swap,
1420 			__GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1421 			NULL) == 0) {
1422 		spin_lock_irq(&info->lock);
1423 		shmem_recalc_inode(inode);
1424 		info->swapped++;
1425 		spin_unlock_irq(&info->lock);
1426 
1427 		swap_shmem_alloc(swap);
1428 		shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1429 
1430 		mutex_unlock(&shmem_swaplist_mutex);
1431 		BUG_ON(folio_mapped(folio));
1432 		swap_writepage(&folio->page, wbc);
1433 		return 0;
1434 	}
1435 
1436 	mutex_unlock(&shmem_swaplist_mutex);
1437 	put_swap_folio(folio, swap);
1438 redirty:
1439 	folio_mark_dirty(folio);
1440 	if (wbc->for_reclaim)
1441 		return AOP_WRITEPAGE_ACTIVATE;	/* Return with folio locked */
1442 	folio_unlock(folio);
1443 	return 0;
1444 }
1445 
1446 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1447 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1448 {
1449 	char buffer[64];
1450 
1451 	if (!mpol || mpol->mode == MPOL_DEFAULT)
1452 		return;		/* show nothing */
1453 
1454 	mpol_to_str(buffer, sizeof(buffer), mpol);
1455 
1456 	seq_printf(seq, ",mpol=%s", buffer);
1457 }
1458 
1459 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1460 {
1461 	struct mempolicy *mpol = NULL;
1462 	if (sbinfo->mpol) {
1463 		raw_spin_lock(&sbinfo->stat_lock);	/* prevent replace/use races */
1464 		mpol = sbinfo->mpol;
1465 		mpol_get(mpol);
1466 		raw_spin_unlock(&sbinfo->stat_lock);
1467 	}
1468 	return mpol;
1469 }
1470 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1471 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1472 {
1473 }
1474 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1475 {
1476 	return NULL;
1477 }
1478 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1479 #ifndef CONFIG_NUMA
1480 #define vm_policy vm_private_data
1481 #endif
1482 
1483 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1484 		struct shmem_inode_info *info, pgoff_t index)
1485 {
1486 	/* Create a pseudo vma that just contains the policy */
1487 	vma_init(vma, NULL);
1488 	/* Bias interleave by inode number to distribute better across nodes */
1489 	vma->vm_pgoff = index + info->vfs_inode.i_ino;
1490 	vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1491 }
1492 
1493 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1494 {
1495 	/* Drop reference taken by mpol_shared_policy_lookup() */
1496 	mpol_cond_put(vma->vm_policy);
1497 }
1498 
1499 static struct folio *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1500 			struct shmem_inode_info *info, pgoff_t index)
1501 {
1502 	struct vm_area_struct pvma;
1503 	struct page *page;
1504 	struct vm_fault vmf = {
1505 		.vma = &pvma,
1506 	};
1507 
1508 	shmem_pseudo_vma_init(&pvma, info, index);
1509 	page = swap_cluster_readahead(swap, gfp, &vmf);
1510 	shmem_pseudo_vma_destroy(&pvma);
1511 
1512 	if (!page)
1513 		return NULL;
1514 	return page_folio(page);
1515 }
1516 
1517 /*
1518  * Make sure huge_gfp is always more limited than limit_gfp.
1519  * Some of the flags set permissions, while others set limitations.
1520  */
1521 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1522 {
1523 	gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1524 	gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1525 	gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1526 	gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1527 
1528 	/* Allow allocations only from the originally specified zones. */
1529 	result |= zoneflags;
1530 
1531 	/*
1532 	 * Minimize the result gfp by taking the union with the deny flags,
1533 	 * and the intersection of the allow flags.
1534 	 */
1535 	result |= (limit_gfp & denyflags);
1536 	result |= (huge_gfp & limit_gfp) & allowflags;
1537 
1538 	return result;
1539 }
1540 
1541 static struct folio *shmem_alloc_hugefolio(gfp_t gfp,
1542 		struct shmem_inode_info *info, pgoff_t index)
1543 {
1544 	struct vm_area_struct pvma;
1545 	struct address_space *mapping = info->vfs_inode.i_mapping;
1546 	pgoff_t hindex;
1547 	struct folio *folio;
1548 
1549 	hindex = round_down(index, HPAGE_PMD_NR);
1550 	if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1551 								XA_PRESENT))
1552 		return NULL;
1553 
1554 	shmem_pseudo_vma_init(&pvma, info, hindex);
1555 	folio = vma_alloc_folio(gfp, HPAGE_PMD_ORDER, &pvma, 0, true);
1556 	shmem_pseudo_vma_destroy(&pvma);
1557 	if (!folio)
1558 		count_vm_event(THP_FILE_FALLBACK);
1559 	return folio;
1560 }
1561 
1562 static struct folio *shmem_alloc_folio(gfp_t gfp,
1563 			struct shmem_inode_info *info, pgoff_t index)
1564 {
1565 	struct vm_area_struct pvma;
1566 	struct folio *folio;
1567 
1568 	shmem_pseudo_vma_init(&pvma, info, index);
1569 	folio = vma_alloc_folio(gfp, 0, &pvma, 0, false);
1570 	shmem_pseudo_vma_destroy(&pvma);
1571 
1572 	return folio;
1573 }
1574 
1575 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
1576 		pgoff_t index, bool huge)
1577 {
1578 	struct shmem_inode_info *info = SHMEM_I(inode);
1579 	struct folio *folio;
1580 	int nr;
1581 	int err = -ENOSPC;
1582 
1583 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1584 		huge = false;
1585 	nr = huge ? HPAGE_PMD_NR : 1;
1586 
1587 	if (!shmem_inode_acct_block(inode, nr))
1588 		goto failed;
1589 
1590 	if (huge)
1591 		folio = shmem_alloc_hugefolio(gfp, info, index);
1592 	else
1593 		folio = shmem_alloc_folio(gfp, info, index);
1594 	if (folio) {
1595 		__folio_set_locked(folio);
1596 		__folio_set_swapbacked(folio);
1597 		return folio;
1598 	}
1599 
1600 	err = -ENOMEM;
1601 	shmem_inode_unacct_blocks(inode, nr);
1602 failed:
1603 	return ERR_PTR(err);
1604 }
1605 
1606 /*
1607  * When a page is moved from swapcache to shmem filecache (either by the
1608  * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1609  * shmem_unuse_inode()), it may have been read in earlier from swap, in
1610  * ignorance of the mapping it belongs to.  If that mapping has special
1611  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1612  * we may need to copy to a suitable page before moving to filecache.
1613  *
1614  * In a future release, this may well be extended to respect cpuset and
1615  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1616  * but for now it is a simple matter of zone.
1617  */
1618 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1619 {
1620 	return folio_zonenum(folio) > gfp_zone(gfp);
1621 }
1622 
1623 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1624 				struct shmem_inode_info *info, pgoff_t index)
1625 {
1626 	struct folio *old, *new;
1627 	struct address_space *swap_mapping;
1628 	swp_entry_t entry;
1629 	pgoff_t swap_index;
1630 	int error;
1631 
1632 	old = *foliop;
1633 	entry = folio_swap_entry(old);
1634 	swap_index = swp_offset(entry);
1635 	swap_mapping = swap_address_space(entry);
1636 
1637 	/*
1638 	 * We have arrived here because our zones are constrained, so don't
1639 	 * limit chance of success by further cpuset and node constraints.
1640 	 */
1641 	gfp &= ~GFP_CONSTRAINT_MASK;
1642 	VM_BUG_ON_FOLIO(folio_test_large(old), old);
1643 	new = shmem_alloc_folio(gfp, info, index);
1644 	if (!new)
1645 		return -ENOMEM;
1646 
1647 	folio_get(new);
1648 	folio_copy(new, old);
1649 	flush_dcache_folio(new);
1650 
1651 	__folio_set_locked(new);
1652 	__folio_set_swapbacked(new);
1653 	folio_mark_uptodate(new);
1654 	folio_set_swap_entry(new, entry);
1655 	folio_set_swapcache(new);
1656 
1657 	/*
1658 	 * Our caller will very soon move newpage out of swapcache, but it's
1659 	 * a nice clean interface for us to replace oldpage by newpage there.
1660 	 */
1661 	xa_lock_irq(&swap_mapping->i_pages);
1662 	error = shmem_replace_entry(swap_mapping, swap_index, old, new);
1663 	if (!error) {
1664 		mem_cgroup_migrate(old, new);
1665 		__lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
1666 		__lruvec_stat_mod_folio(new, NR_SHMEM, 1);
1667 		__lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
1668 		__lruvec_stat_mod_folio(old, NR_SHMEM, -1);
1669 	}
1670 	xa_unlock_irq(&swap_mapping->i_pages);
1671 
1672 	if (unlikely(error)) {
1673 		/*
1674 		 * Is this possible?  I think not, now that our callers check
1675 		 * both PageSwapCache and page_private after getting page lock;
1676 		 * but be defensive.  Reverse old to newpage for clear and free.
1677 		 */
1678 		old = new;
1679 	} else {
1680 		folio_add_lru(new);
1681 		*foliop = new;
1682 	}
1683 
1684 	folio_clear_swapcache(old);
1685 	old->private = NULL;
1686 
1687 	folio_unlock(old);
1688 	folio_put_refs(old, 2);
1689 	return error;
1690 }
1691 
1692 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1693 					 struct folio *folio, swp_entry_t swap)
1694 {
1695 	struct address_space *mapping = inode->i_mapping;
1696 	struct shmem_inode_info *info = SHMEM_I(inode);
1697 	swp_entry_t swapin_error;
1698 	void *old;
1699 
1700 	swapin_error = make_swapin_error_entry();
1701 	old = xa_cmpxchg_irq(&mapping->i_pages, index,
1702 			     swp_to_radix_entry(swap),
1703 			     swp_to_radix_entry(swapin_error), 0);
1704 	if (old != swp_to_radix_entry(swap))
1705 		return;
1706 
1707 	folio_wait_writeback(folio);
1708 	delete_from_swap_cache(folio);
1709 	spin_lock_irq(&info->lock);
1710 	/*
1711 	 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks won't
1712 	 * be 0 when inode is released and thus trigger WARN_ON(inode->i_blocks) in
1713 	 * shmem_evict_inode.
1714 	 */
1715 	info->alloced--;
1716 	info->swapped--;
1717 	shmem_recalc_inode(inode);
1718 	spin_unlock_irq(&info->lock);
1719 	swap_free(swap);
1720 }
1721 
1722 /*
1723  * Swap in the folio pointed to by *foliop.
1724  * Caller has to make sure that *foliop contains a valid swapped folio.
1725  * Returns 0 and the folio in foliop if success. On failure, returns the
1726  * error code and NULL in *foliop.
1727  */
1728 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
1729 			     struct folio **foliop, enum sgp_type sgp,
1730 			     gfp_t gfp, struct vm_area_struct *vma,
1731 			     vm_fault_t *fault_type)
1732 {
1733 	struct address_space *mapping = inode->i_mapping;
1734 	struct shmem_inode_info *info = SHMEM_I(inode);
1735 	struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL;
1736 	struct folio *folio = NULL;
1737 	swp_entry_t swap;
1738 	int error;
1739 
1740 	VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
1741 	swap = radix_to_swp_entry(*foliop);
1742 	*foliop = NULL;
1743 
1744 	if (is_swapin_error_entry(swap))
1745 		return -EIO;
1746 
1747 	/* Look it up and read it in.. */
1748 	folio = swap_cache_get_folio(swap, NULL, 0);
1749 	if (!folio) {
1750 		/* Or update major stats only when swapin succeeds?? */
1751 		if (fault_type) {
1752 			*fault_type |= VM_FAULT_MAJOR;
1753 			count_vm_event(PGMAJFAULT);
1754 			count_memcg_event_mm(charge_mm, PGMAJFAULT);
1755 		}
1756 		/* Here we actually start the io */
1757 		folio = shmem_swapin(swap, gfp, info, index);
1758 		if (!folio) {
1759 			error = -ENOMEM;
1760 			goto failed;
1761 		}
1762 	}
1763 
1764 	/* We have to do this with folio locked to prevent races */
1765 	folio_lock(folio);
1766 	if (!folio_test_swapcache(folio) ||
1767 	    folio_swap_entry(folio).val != swap.val ||
1768 	    !shmem_confirm_swap(mapping, index, swap)) {
1769 		error = -EEXIST;
1770 		goto unlock;
1771 	}
1772 	if (!folio_test_uptodate(folio)) {
1773 		error = -EIO;
1774 		goto failed;
1775 	}
1776 	folio_wait_writeback(folio);
1777 
1778 	/*
1779 	 * Some architectures may have to restore extra metadata to the
1780 	 * folio after reading from swap.
1781 	 */
1782 	arch_swap_restore(swap, folio);
1783 
1784 	if (shmem_should_replace_folio(folio, gfp)) {
1785 		error = shmem_replace_folio(&folio, gfp, info, index);
1786 		if (error)
1787 			goto failed;
1788 	}
1789 
1790 	error = shmem_add_to_page_cache(folio, mapping, index,
1791 					swp_to_radix_entry(swap), gfp,
1792 					charge_mm);
1793 	if (error)
1794 		goto failed;
1795 
1796 	spin_lock_irq(&info->lock);
1797 	info->swapped--;
1798 	shmem_recalc_inode(inode);
1799 	spin_unlock_irq(&info->lock);
1800 
1801 	if (sgp == SGP_WRITE)
1802 		folio_mark_accessed(folio);
1803 
1804 	delete_from_swap_cache(folio);
1805 	folio_mark_dirty(folio);
1806 	swap_free(swap);
1807 
1808 	*foliop = folio;
1809 	return 0;
1810 failed:
1811 	if (!shmem_confirm_swap(mapping, index, swap))
1812 		error = -EEXIST;
1813 	if (error == -EIO)
1814 		shmem_set_folio_swapin_error(inode, index, folio, swap);
1815 unlock:
1816 	if (folio) {
1817 		folio_unlock(folio);
1818 		folio_put(folio);
1819 	}
1820 
1821 	return error;
1822 }
1823 
1824 /*
1825  * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
1826  *
1827  * If we allocate a new one we do not mark it dirty. That's up to the
1828  * vm. If we swap it in we mark it dirty since we also free the swap
1829  * entry since a page cannot live in both the swap and page cache.
1830  *
1831  * vma, vmf, and fault_type are only supplied by shmem_fault:
1832  * otherwise they are NULL.
1833  */
1834 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
1835 		struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
1836 		struct vm_area_struct *vma, struct vm_fault *vmf,
1837 		vm_fault_t *fault_type)
1838 {
1839 	struct address_space *mapping = inode->i_mapping;
1840 	struct shmem_inode_info *info = SHMEM_I(inode);
1841 	struct shmem_sb_info *sbinfo;
1842 	struct mm_struct *charge_mm;
1843 	struct folio *folio;
1844 	pgoff_t hindex;
1845 	gfp_t huge_gfp;
1846 	int error;
1847 	int once = 0;
1848 	int alloced = 0;
1849 
1850 	if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1851 		return -EFBIG;
1852 repeat:
1853 	if (sgp <= SGP_CACHE &&
1854 	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1855 		return -EINVAL;
1856 	}
1857 
1858 	sbinfo = SHMEM_SB(inode->i_sb);
1859 	charge_mm = vma ? vma->vm_mm : NULL;
1860 
1861 	folio = __filemap_get_folio(mapping, index, FGP_ENTRY | FGP_LOCK, 0);
1862 	if (folio && vma && userfaultfd_minor(vma)) {
1863 		if (!xa_is_value(folio)) {
1864 			folio_unlock(folio);
1865 			folio_put(folio);
1866 		}
1867 		*fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
1868 		return 0;
1869 	}
1870 
1871 	if (xa_is_value(folio)) {
1872 		error = shmem_swapin_folio(inode, index, &folio,
1873 					  sgp, gfp, vma, fault_type);
1874 		if (error == -EEXIST)
1875 			goto repeat;
1876 
1877 		*foliop = folio;
1878 		return error;
1879 	}
1880 
1881 	if (folio) {
1882 		if (sgp == SGP_WRITE)
1883 			folio_mark_accessed(folio);
1884 		if (folio_test_uptodate(folio))
1885 			goto out;
1886 		/* fallocated folio */
1887 		if (sgp != SGP_READ)
1888 			goto clear;
1889 		folio_unlock(folio);
1890 		folio_put(folio);
1891 	}
1892 
1893 	/*
1894 	 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
1895 	 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
1896 	 */
1897 	*foliop = NULL;
1898 	if (sgp == SGP_READ)
1899 		return 0;
1900 	if (sgp == SGP_NOALLOC)
1901 		return -ENOENT;
1902 
1903 	/*
1904 	 * Fast cache lookup and swap lookup did not find it: allocate.
1905 	 */
1906 
1907 	if (vma && userfaultfd_missing(vma)) {
1908 		*fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1909 		return 0;
1910 	}
1911 
1912 	if (!shmem_is_huge(vma, inode, index, false))
1913 		goto alloc_nohuge;
1914 
1915 	huge_gfp = vma_thp_gfp_mask(vma);
1916 	huge_gfp = limit_gfp_mask(huge_gfp, gfp);
1917 	folio = shmem_alloc_and_acct_folio(huge_gfp, inode, index, true);
1918 	if (IS_ERR(folio)) {
1919 alloc_nohuge:
1920 		folio = shmem_alloc_and_acct_folio(gfp, inode, index, false);
1921 	}
1922 	if (IS_ERR(folio)) {
1923 		int retry = 5;
1924 
1925 		error = PTR_ERR(folio);
1926 		folio = NULL;
1927 		if (error != -ENOSPC)
1928 			goto unlock;
1929 		/*
1930 		 * Try to reclaim some space by splitting a large folio
1931 		 * beyond i_size on the filesystem.
1932 		 */
1933 		while (retry--) {
1934 			int ret;
1935 
1936 			ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1937 			if (ret == SHRINK_STOP)
1938 				break;
1939 			if (ret)
1940 				goto alloc_nohuge;
1941 		}
1942 		goto unlock;
1943 	}
1944 
1945 	hindex = round_down(index, folio_nr_pages(folio));
1946 
1947 	if (sgp == SGP_WRITE)
1948 		__folio_set_referenced(folio);
1949 
1950 	error = shmem_add_to_page_cache(folio, mapping, hindex,
1951 					NULL, gfp & GFP_RECLAIM_MASK,
1952 					charge_mm);
1953 	if (error)
1954 		goto unacct;
1955 	folio_add_lru(folio);
1956 
1957 	spin_lock_irq(&info->lock);
1958 	info->alloced += folio_nr_pages(folio);
1959 	inode->i_blocks += (blkcnt_t)BLOCKS_PER_PAGE << folio_order(folio);
1960 	shmem_recalc_inode(inode);
1961 	spin_unlock_irq(&info->lock);
1962 	alloced = true;
1963 
1964 	if (folio_test_pmd_mappable(folio) &&
1965 	    DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1966 					folio_next_index(folio) - 1) {
1967 		/*
1968 		 * Part of the large folio is beyond i_size: subject
1969 		 * to shrink under memory pressure.
1970 		 */
1971 		spin_lock(&sbinfo->shrinklist_lock);
1972 		/*
1973 		 * _careful to defend against unlocked access to
1974 		 * ->shrink_list in shmem_unused_huge_shrink()
1975 		 */
1976 		if (list_empty_careful(&info->shrinklist)) {
1977 			list_add_tail(&info->shrinklist,
1978 				      &sbinfo->shrinklist);
1979 			sbinfo->shrinklist_len++;
1980 		}
1981 		spin_unlock(&sbinfo->shrinklist_lock);
1982 	}
1983 
1984 	/*
1985 	 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
1986 	 */
1987 	if (sgp == SGP_FALLOC)
1988 		sgp = SGP_WRITE;
1989 clear:
1990 	/*
1991 	 * Let SGP_WRITE caller clear ends if write does not fill folio;
1992 	 * but SGP_FALLOC on a folio fallocated earlier must initialize
1993 	 * it now, lest undo on failure cancel our earlier guarantee.
1994 	 */
1995 	if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
1996 		long i, n = folio_nr_pages(folio);
1997 
1998 		for (i = 0; i < n; i++)
1999 			clear_highpage(folio_page(folio, i));
2000 		flush_dcache_folio(folio);
2001 		folio_mark_uptodate(folio);
2002 	}
2003 
2004 	/* Perhaps the file has been truncated since we checked */
2005 	if (sgp <= SGP_CACHE &&
2006 	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2007 		if (alloced) {
2008 			folio_clear_dirty(folio);
2009 			filemap_remove_folio(folio);
2010 			spin_lock_irq(&info->lock);
2011 			shmem_recalc_inode(inode);
2012 			spin_unlock_irq(&info->lock);
2013 		}
2014 		error = -EINVAL;
2015 		goto unlock;
2016 	}
2017 out:
2018 	*foliop = folio;
2019 	return 0;
2020 
2021 	/*
2022 	 * Error recovery.
2023 	 */
2024 unacct:
2025 	shmem_inode_unacct_blocks(inode, folio_nr_pages(folio));
2026 
2027 	if (folio_test_large(folio)) {
2028 		folio_unlock(folio);
2029 		folio_put(folio);
2030 		goto alloc_nohuge;
2031 	}
2032 unlock:
2033 	if (folio) {
2034 		folio_unlock(folio);
2035 		folio_put(folio);
2036 	}
2037 	if (error == -ENOSPC && !once++) {
2038 		spin_lock_irq(&info->lock);
2039 		shmem_recalc_inode(inode);
2040 		spin_unlock_irq(&info->lock);
2041 		goto repeat;
2042 	}
2043 	if (error == -EEXIST)
2044 		goto repeat;
2045 	return error;
2046 }
2047 
2048 int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
2049 		enum sgp_type sgp)
2050 {
2051 	return shmem_get_folio_gfp(inode, index, foliop, sgp,
2052 			mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
2053 }
2054 
2055 /*
2056  * This is like autoremove_wake_function, but it removes the wait queue
2057  * entry unconditionally - even if something else had already woken the
2058  * target.
2059  */
2060 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2061 {
2062 	int ret = default_wake_function(wait, mode, sync, key);
2063 	list_del_init(&wait->entry);
2064 	return ret;
2065 }
2066 
2067 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2068 {
2069 	struct vm_area_struct *vma = vmf->vma;
2070 	struct inode *inode = file_inode(vma->vm_file);
2071 	gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2072 	struct folio *folio = NULL;
2073 	int err;
2074 	vm_fault_t ret = VM_FAULT_LOCKED;
2075 
2076 	/*
2077 	 * Trinity finds that probing a hole which tmpfs is punching can
2078 	 * prevent the hole-punch from ever completing: which in turn
2079 	 * locks writers out with its hold on i_rwsem.  So refrain from
2080 	 * faulting pages into the hole while it's being punched.  Although
2081 	 * shmem_undo_range() does remove the additions, it may be unable to
2082 	 * keep up, as each new page needs its own unmap_mapping_range() call,
2083 	 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2084 	 *
2085 	 * It does not matter if we sometimes reach this check just before the
2086 	 * hole-punch begins, so that one fault then races with the punch:
2087 	 * we just need to make racing faults a rare case.
2088 	 *
2089 	 * The implementation below would be much simpler if we just used a
2090 	 * standard mutex or completion: but we cannot take i_rwsem in fault,
2091 	 * and bloating every shmem inode for this unlikely case would be sad.
2092 	 */
2093 	if (unlikely(inode->i_private)) {
2094 		struct shmem_falloc *shmem_falloc;
2095 
2096 		spin_lock(&inode->i_lock);
2097 		shmem_falloc = inode->i_private;
2098 		if (shmem_falloc &&
2099 		    shmem_falloc->waitq &&
2100 		    vmf->pgoff >= shmem_falloc->start &&
2101 		    vmf->pgoff < shmem_falloc->next) {
2102 			struct file *fpin;
2103 			wait_queue_head_t *shmem_falloc_waitq;
2104 			DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2105 
2106 			ret = VM_FAULT_NOPAGE;
2107 			fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2108 			if (fpin)
2109 				ret = VM_FAULT_RETRY;
2110 
2111 			shmem_falloc_waitq = shmem_falloc->waitq;
2112 			prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2113 					TASK_UNINTERRUPTIBLE);
2114 			spin_unlock(&inode->i_lock);
2115 			schedule();
2116 
2117 			/*
2118 			 * shmem_falloc_waitq points into the shmem_fallocate()
2119 			 * stack of the hole-punching task: shmem_falloc_waitq
2120 			 * is usually invalid by the time we reach here, but
2121 			 * finish_wait() does not dereference it in that case;
2122 			 * though i_lock needed lest racing with wake_up_all().
2123 			 */
2124 			spin_lock(&inode->i_lock);
2125 			finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2126 			spin_unlock(&inode->i_lock);
2127 
2128 			if (fpin)
2129 				fput(fpin);
2130 			return ret;
2131 		}
2132 		spin_unlock(&inode->i_lock);
2133 	}
2134 
2135 	err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
2136 				  gfp, vma, vmf, &ret);
2137 	if (err)
2138 		return vmf_error(err);
2139 	if (folio)
2140 		vmf->page = folio_file_page(folio, vmf->pgoff);
2141 	return ret;
2142 }
2143 
2144 unsigned long shmem_get_unmapped_area(struct file *file,
2145 				      unsigned long uaddr, unsigned long len,
2146 				      unsigned long pgoff, unsigned long flags)
2147 {
2148 	unsigned long (*get_area)(struct file *,
2149 		unsigned long, unsigned long, unsigned long, unsigned long);
2150 	unsigned long addr;
2151 	unsigned long offset;
2152 	unsigned long inflated_len;
2153 	unsigned long inflated_addr;
2154 	unsigned long inflated_offset;
2155 
2156 	if (len > TASK_SIZE)
2157 		return -ENOMEM;
2158 
2159 	get_area = current->mm->get_unmapped_area;
2160 	addr = get_area(file, uaddr, len, pgoff, flags);
2161 
2162 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2163 		return addr;
2164 	if (IS_ERR_VALUE(addr))
2165 		return addr;
2166 	if (addr & ~PAGE_MASK)
2167 		return addr;
2168 	if (addr > TASK_SIZE - len)
2169 		return addr;
2170 
2171 	if (shmem_huge == SHMEM_HUGE_DENY)
2172 		return addr;
2173 	if (len < HPAGE_PMD_SIZE)
2174 		return addr;
2175 	if (flags & MAP_FIXED)
2176 		return addr;
2177 	/*
2178 	 * Our priority is to support MAP_SHARED mapped hugely;
2179 	 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2180 	 * But if caller specified an address hint and we allocated area there
2181 	 * successfully, respect that as before.
2182 	 */
2183 	if (uaddr == addr)
2184 		return addr;
2185 
2186 	if (shmem_huge != SHMEM_HUGE_FORCE) {
2187 		struct super_block *sb;
2188 
2189 		if (file) {
2190 			VM_BUG_ON(file->f_op != &shmem_file_operations);
2191 			sb = file_inode(file)->i_sb;
2192 		} else {
2193 			/*
2194 			 * Called directly from mm/mmap.c, or drivers/char/mem.c
2195 			 * for "/dev/zero", to create a shared anonymous object.
2196 			 */
2197 			if (IS_ERR(shm_mnt))
2198 				return addr;
2199 			sb = shm_mnt->mnt_sb;
2200 		}
2201 		if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2202 			return addr;
2203 	}
2204 
2205 	offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2206 	if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2207 		return addr;
2208 	if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2209 		return addr;
2210 
2211 	inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2212 	if (inflated_len > TASK_SIZE)
2213 		return addr;
2214 	if (inflated_len < len)
2215 		return addr;
2216 
2217 	inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2218 	if (IS_ERR_VALUE(inflated_addr))
2219 		return addr;
2220 	if (inflated_addr & ~PAGE_MASK)
2221 		return addr;
2222 
2223 	inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2224 	inflated_addr += offset - inflated_offset;
2225 	if (inflated_offset > offset)
2226 		inflated_addr += HPAGE_PMD_SIZE;
2227 
2228 	if (inflated_addr > TASK_SIZE - len)
2229 		return addr;
2230 	return inflated_addr;
2231 }
2232 
2233 #ifdef CONFIG_NUMA
2234 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2235 {
2236 	struct inode *inode = file_inode(vma->vm_file);
2237 	return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2238 }
2239 
2240 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2241 					  unsigned long addr)
2242 {
2243 	struct inode *inode = file_inode(vma->vm_file);
2244 	pgoff_t index;
2245 
2246 	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2247 	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2248 }
2249 #endif
2250 
2251 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2252 {
2253 	struct inode *inode = file_inode(file);
2254 	struct shmem_inode_info *info = SHMEM_I(inode);
2255 	int retval = -ENOMEM;
2256 
2257 	/*
2258 	 * What serializes the accesses to info->flags?
2259 	 * ipc_lock_object() when called from shmctl_do_lock(),
2260 	 * no serialization needed when called from shm_destroy().
2261 	 */
2262 	if (lock && !(info->flags & VM_LOCKED)) {
2263 		if (!user_shm_lock(inode->i_size, ucounts))
2264 			goto out_nomem;
2265 		info->flags |= VM_LOCKED;
2266 		mapping_set_unevictable(file->f_mapping);
2267 	}
2268 	if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2269 		user_shm_unlock(inode->i_size, ucounts);
2270 		info->flags &= ~VM_LOCKED;
2271 		mapping_clear_unevictable(file->f_mapping);
2272 	}
2273 	retval = 0;
2274 
2275 out_nomem:
2276 	return retval;
2277 }
2278 
2279 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2280 {
2281 	struct inode *inode = file_inode(file);
2282 	struct shmem_inode_info *info = SHMEM_I(inode);
2283 	int ret;
2284 
2285 	ret = seal_check_future_write(info->seals, vma);
2286 	if (ret)
2287 		return ret;
2288 
2289 	/* arm64 - allow memory tagging on RAM-based files */
2290 	vma->vm_flags |= VM_MTE_ALLOWED;
2291 
2292 	file_accessed(file);
2293 	/* This is anonymous shared memory if it is unlinked at the time of mmap */
2294 	if (inode->i_nlink)
2295 		vma->vm_ops = &shmem_vm_ops;
2296 	else
2297 		vma->vm_ops = &shmem_anon_vm_ops;
2298 	return 0;
2299 }
2300 
2301 #ifdef CONFIG_TMPFS_XATTR
2302 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2303 
2304 /*
2305  * chattr's fsflags are unrelated to extended attributes,
2306  * but tmpfs has chosen to enable them under the same config option.
2307  */
2308 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2309 {
2310 	unsigned int i_flags = 0;
2311 
2312 	if (fsflags & FS_NOATIME_FL)
2313 		i_flags |= S_NOATIME;
2314 	if (fsflags & FS_APPEND_FL)
2315 		i_flags |= S_APPEND;
2316 	if (fsflags & FS_IMMUTABLE_FL)
2317 		i_flags |= S_IMMUTABLE;
2318 	/*
2319 	 * But FS_NODUMP_FL does not require any action in i_flags.
2320 	 */
2321 	inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2322 }
2323 #else
2324 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2325 {
2326 }
2327 #define shmem_initxattrs NULL
2328 #endif
2329 
2330 static struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block *sb,
2331 				     struct inode *dir, umode_t mode, dev_t dev,
2332 				     unsigned long flags)
2333 {
2334 	struct inode *inode;
2335 	struct shmem_inode_info *info;
2336 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2337 	ino_t ino;
2338 
2339 	if (shmem_reserve_inode(sb, &ino))
2340 		return NULL;
2341 
2342 	inode = new_inode(sb);
2343 	if (inode) {
2344 		inode->i_ino = ino;
2345 		inode_init_owner(idmap, inode, dir, mode);
2346 		inode->i_blocks = 0;
2347 		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2348 		inode->i_generation = get_random_u32();
2349 		info = SHMEM_I(inode);
2350 		memset(info, 0, (char *)inode - (char *)info);
2351 		spin_lock_init(&info->lock);
2352 		atomic_set(&info->stop_eviction, 0);
2353 		info->seals = F_SEAL_SEAL;
2354 		info->flags = flags & VM_NORESERVE;
2355 		info->i_crtime = inode->i_mtime;
2356 		info->fsflags = (dir == NULL) ? 0 :
2357 			SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2358 		if (info->fsflags)
2359 			shmem_set_inode_flags(inode, info->fsflags);
2360 		INIT_LIST_HEAD(&info->shrinklist);
2361 		INIT_LIST_HEAD(&info->swaplist);
2362 		simple_xattrs_init(&info->xattrs);
2363 		cache_no_acl(inode);
2364 		mapping_set_large_folios(inode->i_mapping);
2365 
2366 		switch (mode & S_IFMT) {
2367 		default:
2368 			inode->i_op = &shmem_special_inode_operations;
2369 			init_special_inode(inode, mode, dev);
2370 			break;
2371 		case S_IFREG:
2372 			inode->i_mapping->a_ops = &shmem_aops;
2373 			inode->i_op = &shmem_inode_operations;
2374 			inode->i_fop = &shmem_file_operations;
2375 			mpol_shared_policy_init(&info->policy,
2376 						 shmem_get_sbmpol(sbinfo));
2377 			break;
2378 		case S_IFDIR:
2379 			inc_nlink(inode);
2380 			/* Some things misbehave if size == 0 on a directory */
2381 			inode->i_size = 2 * BOGO_DIRENT_SIZE;
2382 			inode->i_op = &shmem_dir_inode_operations;
2383 			inode->i_fop = &simple_dir_operations;
2384 			break;
2385 		case S_IFLNK:
2386 			/*
2387 			 * Must not load anything in the rbtree,
2388 			 * mpol_free_shared_policy will not be called.
2389 			 */
2390 			mpol_shared_policy_init(&info->policy, NULL);
2391 			break;
2392 		}
2393 
2394 		lockdep_annotate_inode_mutex_key(inode);
2395 	} else
2396 		shmem_free_inode(sb);
2397 	return inode;
2398 }
2399 
2400 #ifdef CONFIG_USERFAULTFD
2401 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2402 			   pmd_t *dst_pmd,
2403 			   struct vm_area_struct *dst_vma,
2404 			   unsigned long dst_addr,
2405 			   unsigned long src_addr,
2406 			   bool zeropage, bool wp_copy,
2407 			   struct page **pagep)
2408 {
2409 	struct inode *inode = file_inode(dst_vma->vm_file);
2410 	struct shmem_inode_info *info = SHMEM_I(inode);
2411 	struct address_space *mapping = inode->i_mapping;
2412 	gfp_t gfp = mapping_gfp_mask(mapping);
2413 	pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2414 	void *page_kaddr;
2415 	struct folio *folio;
2416 	int ret;
2417 	pgoff_t max_off;
2418 
2419 	if (!shmem_inode_acct_block(inode, 1)) {
2420 		/*
2421 		 * We may have got a page, returned -ENOENT triggering a retry,
2422 		 * and now we find ourselves with -ENOMEM. Release the page, to
2423 		 * avoid a BUG_ON in our caller.
2424 		 */
2425 		if (unlikely(*pagep)) {
2426 			put_page(*pagep);
2427 			*pagep = NULL;
2428 		}
2429 		return -ENOMEM;
2430 	}
2431 
2432 	if (!*pagep) {
2433 		ret = -ENOMEM;
2434 		folio = shmem_alloc_folio(gfp, info, pgoff);
2435 		if (!folio)
2436 			goto out_unacct_blocks;
2437 
2438 		if (!zeropage) {	/* COPY */
2439 			page_kaddr = kmap_local_folio(folio, 0);
2440 			/*
2441 			 * The read mmap_lock is held here.  Despite the
2442 			 * mmap_lock being read recursive a deadlock is still
2443 			 * possible if a writer has taken a lock.  For example:
2444 			 *
2445 			 * process A thread 1 takes read lock on own mmap_lock
2446 			 * process A thread 2 calls mmap, blocks taking write lock
2447 			 * process B thread 1 takes page fault, read lock on own mmap lock
2448 			 * process B thread 2 calls mmap, blocks taking write lock
2449 			 * process A thread 1 blocks taking read lock on process B
2450 			 * process B thread 1 blocks taking read lock on process A
2451 			 *
2452 			 * Disable page faults to prevent potential deadlock
2453 			 * and retry the copy outside the mmap_lock.
2454 			 */
2455 			pagefault_disable();
2456 			ret = copy_from_user(page_kaddr,
2457 					     (const void __user *)src_addr,
2458 					     PAGE_SIZE);
2459 			pagefault_enable();
2460 			kunmap_local(page_kaddr);
2461 
2462 			/* fallback to copy_from_user outside mmap_lock */
2463 			if (unlikely(ret)) {
2464 				*pagep = &folio->page;
2465 				ret = -ENOENT;
2466 				/* don't free the page */
2467 				goto out_unacct_blocks;
2468 			}
2469 
2470 			flush_dcache_folio(folio);
2471 		} else {		/* ZEROPAGE */
2472 			clear_user_highpage(&folio->page, dst_addr);
2473 		}
2474 	} else {
2475 		folio = page_folio(*pagep);
2476 		VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2477 		*pagep = NULL;
2478 	}
2479 
2480 	VM_BUG_ON(folio_test_locked(folio));
2481 	VM_BUG_ON(folio_test_swapbacked(folio));
2482 	__folio_set_locked(folio);
2483 	__folio_set_swapbacked(folio);
2484 	__folio_mark_uptodate(folio);
2485 
2486 	ret = -EFAULT;
2487 	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2488 	if (unlikely(pgoff >= max_off))
2489 		goto out_release;
2490 
2491 	ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
2492 				      gfp & GFP_RECLAIM_MASK, dst_mm);
2493 	if (ret)
2494 		goto out_release;
2495 
2496 	ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
2497 				       &folio->page, true, wp_copy);
2498 	if (ret)
2499 		goto out_delete_from_cache;
2500 
2501 	spin_lock_irq(&info->lock);
2502 	info->alloced++;
2503 	inode->i_blocks += BLOCKS_PER_PAGE;
2504 	shmem_recalc_inode(inode);
2505 	spin_unlock_irq(&info->lock);
2506 
2507 	folio_unlock(folio);
2508 	return 0;
2509 out_delete_from_cache:
2510 	filemap_remove_folio(folio);
2511 out_release:
2512 	folio_unlock(folio);
2513 	folio_put(folio);
2514 out_unacct_blocks:
2515 	shmem_inode_unacct_blocks(inode, 1);
2516 	return ret;
2517 }
2518 #endif /* CONFIG_USERFAULTFD */
2519 
2520 #ifdef CONFIG_TMPFS
2521 static const struct inode_operations shmem_symlink_inode_operations;
2522 static const struct inode_operations shmem_short_symlink_operations;
2523 
2524 static int
2525 shmem_write_begin(struct file *file, struct address_space *mapping,
2526 			loff_t pos, unsigned len,
2527 			struct page **pagep, void **fsdata)
2528 {
2529 	struct inode *inode = mapping->host;
2530 	struct shmem_inode_info *info = SHMEM_I(inode);
2531 	pgoff_t index = pos >> PAGE_SHIFT;
2532 	struct folio *folio;
2533 	int ret = 0;
2534 
2535 	/* i_rwsem is held by caller */
2536 	if (unlikely(info->seals & (F_SEAL_GROW |
2537 				   F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2538 		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2539 			return -EPERM;
2540 		if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2541 			return -EPERM;
2542 	}
2543 
2544 	ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
2545 
2546 	if (ret)
2547 		return ret;
2548 
2549 	*pagep = folio_file_page(folio, index);
2550 	if (PageHWPoison(*pagep)) {
2551 		folio_unlock(folio);
2552 		folio_put(folio);
2553 		*pagep = NULL;
2554 		return -EIO;
2555 	}
2556 
2557 	return 0;
2558 }
2559 
2560 static int
2561 shmem_write_end(struct file *file, struct address_space *mapping,
2562 			loff_t pos, unsigned len, unsigned copied,
2563 			struct page *page, void *fsdata)
2564 {
2565 	struct inode *inode = mapping->host;
2566 
2567 	if (pos + copied > inode->i_size)
2568 		i_size_write(inode, pos + copied);
2569 
2570 	if (!PageUptodate(page)) {
2571 		struct page *head = compound_head(page);
2572 		if (PageTransCompound(page)) {
2573 			int i;
2574 
2575 			for (i = 0; i < HPAGE_PMD_NR; i++) {
2576 				if (head + i == page)
2577 					continue;
2578 				clear_highpage(head + i);
2579 				flush_dcache_page(head + i);
2580 			}
2581 		}
2582 		if (copied < PAGE_SIZE) {
2583 			unsigned from = pos & (PAGE_SIZE - 1);
2584 			zero_user_segments(page, 0, from,
2585 					from + copied, PAGE_SIZE);
2586 		}
2587 		SetPageUptodate(head);
2588 	}
2589 	set_page_dirty(page);
2590 	unlock_page(page);
2591 	put_page(page);
2592 
2593 	return copied;
2594 }
2595 
2596 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2597 {
2598 	struct file *file = iocb->ki_filp;
2599 	struct inode *inode = file_inode(file);
2600 	struct address_space *mapping = inode->i_mapping;
2601 	pgoff_t index;
2602 	unsigned long offset;
2603 	int error = 0;
2604 	ssize_t retval = 0;
2605 	loff_t *ppos = &iocb->ki_pos;
2606 
2607 	index = *ppos >> PAGE_SHIFT;
2608 	offset = *ppos & ~PAGE_MASK;
2609 
2610 	for (;;) {
2611 		struct folio *folio = NULL;
2612 		struct page *page = NULL;
2613 		pgoff_t end_index;
2614 		unsigned long nr, ret;
2615 		loff_t i_size = i_size_read(inode);
2616 
2617 		end_index = i_size >> PAGE_SHIFT;
2618 		if (index > end_index)
2619 			break;
2620 		if (index == end_index) {
2621 			nr = i_size & ~PAGE_MASK;
2622 			if (nr <= offset)
2623 				break;
2624 		}
2625 
2626 		error = shmem_get_folio(inode, index, &folio, SGP_READ);
2627 		if (error) {
2628 			if (error == -EINVAL)
2629 				error = 0;
2630 			break;
2631 		}
2632 		if (folio) {
2633 			folio_unlock(folio);
2634 
2635 			page = folio_file_page(folio, index);
2636 			if (PageHWPoison(page)) {
2637 				folio_put(folio);
2638 				error = -EIO;
2639 				break;
2640 			}
2641 		}
2642 
2643 		/*
2644 		 * We must evaluate after, since reads (unlike writes)
2645 		 * are called without i_rwsem protection against truncate
2646 		 */
2647 		nr = PAGE_SIZE;
2648 		i_size = i_size_read(inode);
2649 		end_index = i_size >> PAGE_SHIFT;
2650 		if (index == end_index) {
2651 			nr = i_size & ~PAGE_MASK;
2652 			if (nr <= offset) {
2653 				if (folio)
2654 					folio_put(folio);
2655 				break;
2656 			}
2657 		}
2658 		nr -= offset;
2659 
2660 		if (folio) {
2661 			/*
2662 			 * If users can be writing to this page using arbitrary
2663 			 * virtual addresses, take care about potential aliasing
2664 			 * before reading the page on the kernel side.
2665 			 */
2666 			if (mapping_writably_mapped(mapping))
2667 				flush_dcache_page(page);
2668 			/*
2669 			 * Mark the page accessed if we read the beginning.
2670 			 */
2671 			if (!offset)
2672 				folio_mark_accessed(folio);
2673 			/*
2674 			 * Ok, we have the page, and it's up-to-date, so
2675 			 * now we can copy it to user space...
2676 			 */
2677 			ret = copy_page_to_iter(page, offset, nr, to);
2678 			folio_put(folio);
2679 
2680 		} else if (user_backed_iter(to)) {
2681 			/*
2682 			 * Copy to user tends to be so well optimized, but
2683 			 * clear_user() not so much, that it is noticeably
2684 			 * faster to copy the zero page instead of clearing.
2685 			 */
2686 			ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
2687 		} else {
2688 			/*
2689 			 * But submitting the same page twice in a row to
2690 			 * splice() - or others? - can result in confusion:
2691 			 * so don't attempt that optimization on pipes etc.
2692 			 */
2693 			ret = iov_iter_zero(nr, to);
2694 		}
2695 
2696 		retval += ret;
2697 		offset += ret;
2698 		index += offset >> PAGE_SHIFT;
2699 		offset &= ~PAGE_MASK;
2700 
2701 		if (!iov_iter_count(to))
2702 			break;
2703 		if (ret < nr) {
2704 			error = -EFAULT;
2705 			break;
2706 		}
2707 		cond_resched();
2708 	}
2709 
2710 	*ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2711 	file_accessed(file);
2712 	return retval ? retval : error;
2713 }
2714 
2715 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2716 {
2717 	struct address_space *mapping = file->f_mapping;
2718 	struct inode *inode = mapping->host;
2719 
2720 	if (whence != SEEK_DATA && whence != SEEK_HOLE)
2721 		return generic_file_llseek_size(file, offset, whence,
2722 					MAX_LFS_FILESIZE, i_size_read(inode));
2723 	if (offset < 0)
2724 		return -ENXIO;
2725 
2726 	inode_lock(inode);
2727 	/* We're holding i_rwsem so we can access i_size directly */
2728 	offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
2729 	if (offset >= 0)
2730 		offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2731 	inode_unlock(inode);
2732 	return offset;
2733 }
2734 
2735 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2736 							 loff_t len)
2737 {
2738 	struct inode *inode = file_inode(file);
2739 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2740 	struct shmem_inode_info *info = SHMEM_I(inode);
2741 	struct shmem_falloc shmem_falloc;
2742 	pgoff_t start, index, end, undo_fallocend;
2743 	int error;
2744 
2745 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2746 		return -EOPNOTSUPP;
2747 
2748 	inode_lock(inode);
2749 
2750 	if (mode & FALLOC_FL_PUNCH_HOLE) {
2751 		struct address_space *mapping = file->f_mapping;
2752 		loff_t unmap_start = round_up(offset, PAGE_SIZE);
2753 		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2754 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2755 
2756 		/* protected by i_rwsem */
2757 		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2758 			error = -EPERM;
2759 			goto out;
2760 		}
2761 
2762 		shmem_falloc.waitq = &shmem_falloc_waitq;
2763 		shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2764 		shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2765 		spin_lock(&inode->i_lock);
2766 		inode->i_private = &shmem_falloc;
2767 		spin_unlock(&inode->i_lock);
2768 
2769 		if ((u64)unmap_end > (u64)unmap_start)
2770 			unmap_mapping_range(mapping, unmap_start,
2771 					    1 + unmap_end - unmap_start, 0);
2772 		shmem_truncate_range(inode, offset, offset + len - 1);
2773 		/* No need to unmap again: hole-punching leaves COWed pages */
2774 
2775 		spin_lock(&inode->i_lock);
2776 		inode->i_private = NULL;
2777 		wake_up_all(&shmem_falloc_waitq);
2778 		WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2779 		spin_unlock(&inode->i_lock);
2780 		error = 0;
2781 		goto out;
2782 	}
2783 
2784 	/* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2785 	error = inode_newsize_ok(inode, offset + len);
2786 	if (error)
2787 		goto out;
2788 
2789 	if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2790 		error = -EPERM;
2791 		goto out;
2792 	}
2793 
2794 	start = offset >> PAGE_SHIFT;
2795 	end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2796 	/* Try to avoid a swapstorm if len is impossible to satisfy */
2797 	if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2798 		error = -ENOSPC;
2799 		goto out;
2800 	}
2801 
2802 	shmem_falloc.waitq = NULL;
2803 	shmem_falloc.start = start;
2804 	shmem_falloc.next  = start;
2805 	shmem_falloc.nr_falloced = 0;
2806 	shmem_falloc.nr_unswapped = 0;
2807 	spin_lock(&inode->i_lock);
2808 	inode->i_private = &shmem_falloc;
2809 	spin_unlock(&inode->i_lock);
2810 
2811 	/*
2812 	 * info->fallocend is only relevant when huge pages might be
2813 	 * involved: to prevent split_huge_page() freeing fallocated
2814 	 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
2815 	 */
2816 	undo_fallocend = info->fallocend;
2817 	if (info->fallocend < end)
2818 		info->fallocend = end;
2819 
2820 	for (index = start; index < end; ) {
2821 		struct folio *folio;
2822 
2823 		/*
2824 		 * Good, the fallocate(2) manpage permits EINTR: we may have
2825 		 * been interrupted because we are using up too much memory.
2826 		 */
2827 		if (signal_pending(current))
2828 			error = -EINTR;
2829 		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2830 			error = -ENOMEM;
2831 		else
2832 			error = shmem_get_folio(inode, index, &folio,
2833 						SGP_FALLOC);
2834 		if (error) {
2835 			info->fallocend = undo_fallocend;
2836 			/* Remove the !uptodate folios we added */
2837 			if (index > start) {
2838 				shmem_undo_range(inode,
2839 				    (loff_t)start << PAGE_SHIFT,
2840 				    ((loff_t)index << PAGE_SHIFT) - 1, true);
2841 			}
2842 			goto undone;
2843 		}
2844 
2845 		/*
2846 		 * Here is a more important optimization than it appears:
2847 		 * a second SGP_FALLOC on the same large folio will clear it,
2848 		 * making it uptodate and un-undoable if we fail later.
2849 		 */
2850 		index = folio_next_index(folio);
2851 		/* Beware 32-bit wraparound */
2852 		if (!index)
2853 			index--;
2854 
2855 		/*
2856 		 * Inform shmem_writepage() how far we have reached.
2857 		 * No need for lock or barrier: we have the page lock.
2858 		 */
2859 		if (!folio_test_uptodate(folio))
2860 			shmem_falloc.nr_falloced += index - shmem_falloc.next;
2861 		shmem_falloc.next = index;
2862 
2863 		/*
2864 		 * If !uptodate, leave it that way so that freeable folios
2865 		 * can be recognized if we need to rollback on error later.
2866 		 * But mark it dirty so that memory pressure will swap rather
2867 		 * than free the folios we are allocating (and SGP_CACHE folios
2868 		 * might still be clean: we now need to mark those dirty too).
2869 		 */
2870 		folio_mark_dirty(folio);
2871 		folio_unlock(folio);
2872 		folio_put(folio);
2873 		cond_resched();
2874 	}
2875 
2876 	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2877 		i_size_write(inode, offset + len);
2878 undone:
2879 	spin_lock(&inode->i_lock);
2880 	inode->i_private = NULL;
2881 	spin_unlock(&inode->i_lock);
2882 out:
2883 	if (!error)
2884 		file_modified(file);
2885 	inode_unlock(inode);
2886 	return error;
2887 }
2888 
2889 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2890 {
2891 	struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2892 
2893 	buf->f_type = TMPFS_MAGIC;
2894 	buf->f_bsize = PAGE_SIZE;
2895 	buf->f_namelen = NAME_MAX;
2896 	if (sbinfo->max_blocks) {
2897 		buf->f_blocks = sbinfo->max_blocks;
2898 		buf->f_bavail =
2899 		buf->f_bfree  = sbinfo->max_blocks -
2900 				percpu_counter_sum(&sbinfo->used_blocks);
2901 	}
2902 	if (sbinfo->max_inodes) {
2903 		buf->f_files = sbinfo->max_inodes;
2904 		buf->f_ffree = sbinfo->free_inodes;
2905 	}
2906 	/* else leave those fields 0 like simple_statfs */
2907 
2908 	buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
2909 
2910 	return 0;
2911 }
2912 
2913 /*
2914  * File creation. Allocate an inode, and we're done..
2915  */
2916 static int
2917 shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
2918 	    struct dentry *dentry, umode_t mode, dev_t dev)
2919 {
2920 	struct inode *inode;
2921 	int error = -ENOSPC;
2922 
2923 	inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
2924 	if (inode) {
2925 		error = simple_acl_create(dir, inode);
2926 		if (error)
2927 			goto out_iput;
2928 		error = security_inode_init_security(inode, dir,
2929 						     &dentry->d_name,
2930 						     shmem_initxattrs, NULL);
2931 		if (error && error != -EOPNOTSUPP)
2932 			goto out_iput;
2933 
2934 		error = 0;
2935 		dir->i_size += BOGO_DIRENT_SIZE;
2936 		dir->i_ctime = dir->i_mtime = current_time(dir);
2937 		inode_inc_iversion(dir);
2938 		d_instantiate(dentry, inode);
2939 		dget(dentry); /* Extra count - pin the dentry in core */
2940 	}
2941 	return error;
2942 out_iput:
2943 	iput(inode);
2944 	return error;
2945 }
2946 
2947 static int
2948 shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2949 	      struct file *file, umode_t mode)
2950 {
2951 	struct inode *inode;
2952 	int error = -ENOSPC;
2953 
2954 	inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
2955 	if (inode) {
2956 		error = security_inode_init_security(inode, dir,
2957 						     NULL,
2958 						     shmem_initxattrs, NULL);
2959 		if (error && error != -EOPNOTSUPP)
2960 			goto out_iput;
2961 		error = simple_acl_create(dir, inode);
2962 		if (error)
2963 			goto out_iput;
2964 		d_tmpfile(file, inode);
2965 	}
2966 	return finish_open_simple(file, error);
2967 out_iput:
2968 	iput(inode);
2969 	return error;
2970 }
2971 
2972 static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2973 		       struct dentry *dentry, umode_t mode)
2974 {
2975 	int error;
2976 
2977 	error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
2978 	if (error)
2979 		return error;
2980 	inc_nlink(dir);
2981 	return 0;
2982 }
2983 
2984 static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
2985 			struct dentry *dentry, umode_t mode, bool excl)
2986 {
2987 	return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
2988 }
2989 
2990 /*
2991  * Link a file..
2992  */
2993 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2994 {
2995 	struct inode *inode = d_inode(old_dentry);
2996 	int ret = 0;
2997 
2998 	/*
2999 	 * No ordinary (disk based) filesystem counts links as inodes;
3000 	 * but each new link needs a new dentry, pinning lowmem, and
3001 	 * tmpfs dentries cannot be pruned until they are unlinked.
3002 	 * But if an O_TMPFILE file is linked into the tmpfs, the
3003 	 * first link must skip that, to get the accounting right.
3004 	 */
3005 	if (inode->i_nlink) {
3006 		ret = shmem_reserve_inode(inode->i_sb, NULL);
3007 		if (ret)
3008 			goto out;
3009 	}
3010 
3011 	dir->i_size += BOGO_DIRENT_SIZE;
3012 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3013 	inode_inc_iversion(dir);
3014 	inc_nlink(inode);
3015 	ihold(inode);	/* New dentry reference */
3016 	dget(dentry);		/* Extra pinning count for the created dentry */
3017 	d_instantiate(dentry, inode);
3018 out:
3019 	return ret;
3020 }
3021 
3022 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3023 {
3024 	struct inode *inode = d_inode(dentry);
3025 
3026 	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3027 		shmem_free_inode(inode->i_sb);
3028 
3029 	dir->i_size -= BOGO_DIRENT_SIZE;
3030 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3031 	inode_inc_iversion(dir);
3032 	drop_nlink(inode);
3033 	dput(dentry);	/* Undo the count from "create" - this does all the work */
3034 	return 0;
3035 }
3036 
3037 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3038 {
3039 	if (!simple_empty(dentry))
3040 		return -ENOTEMPTY;
3041 
3042 	drop_nlink(d_inode(dentry));
3043 	drop_nlink(dir);
3044 	return shmem_unlink(dir, dentry);
3045 }
3046 
3047 static int shmem_whiteout(struct mnt_idmap *idmap,
3048 			  struct inode *old_dir, struct dentry *old_dentry)
3049 {
3050 	struct dentry *whiteout;
3051 	int error;
3052 
3053 	whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3054 	if (!whiteout)
3055 		return -ENOMEM;
3056 
3057 	error = shmem_mknod(idmap, old_dir, whiteout,
3058 			    S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3059 	dput(whiteout);
3060 	if (error)
3061 		return error;
3062 
3063 	/*
3064 	 * Cheat and hash the whiteout while the old dentry is still in
3065 	 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3066 	 *
3067 	 * d_lookup() will consistently find one of them at this point,
3068 	 * not sure which one, but that isn't even important.
3069 	 */
3070 	d_rehash(whiteout);
3071 	return 0;
3072 }
3073 
3074 /*
3075  * The VFS layer already does all the dentry stuff for rename,
3076  * we just have to decrement the usage count for the target if
3077  * it exists so that the VFS layer correctly free's it when it
3078  * gets overwritten.
3079  */
3080 static int shmem_rename2(struct mnt_idmap *idmap,
3081 			 struct inode *old_dir, struct dentry *old_dentry,
3082 			 struct inode *new_dir, struct dentry *new_dentry,
3083 			 unsigned int flags)
3084 {
3085 	struct inode *inode = d_inode(old_dentry);
3086 	int they_are_dirs = S_ISDIR(inode->i_mode);
3087 
3088 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3089 		return -EINVAL;
3090 
3091 	if (flags & RENAME_EXCHANGE)
3092 		return simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
3093 
3094 	if (!simple_empty(new_dentry))
3095 		return -ENOTEMPTY;
3096 
3097 	if (flags & RENAME_WHITEOUT) {
3098 		int error;
3099 
3100 		error = shmem_whiteout(idmap, old_dir, old_dentry);
3101 		if (error)
3102 			return error;
3103 	}
3104 
3105 	if (d_really_is_positive(new_dentry)) {
3106 		(void) shmem_unlink(new_dir, new_dentry);
3107 		if (they_are_dirs) {
3108 			drop_nlink(d_inode(new_dentry));
3109 			drop_nlink(old_dir);
3110 		}
3111 	} else if (they_are_dirs) {
3112 		drop_nlink(old_dir);
3113 		inc_nlink(new_dir);
3114 	}
3115 
3116 	old_dir->i_size -= BOGO_DIRENT_SIZE;
3117 	new_dir->i_size += BOGO_DIRENT_SIZE;
3118 	old_dir->i_ctime = old_dir->i_mtime =
3119 	new_dir->i_ctime = new_dir->i_mtime =
3120 	inode->i_ctime = current_time(old_dir);
3121 	inode_inc_iversion(old_dir);
3122 	inode_inc_iversion(new_dir);
3123 	return 0;
3124 }
3125 
3126 static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3127 			 struct dentry *dentry, const char *symname)
3128 {
3129 	int error;
3130 	int len;
3131 	struct inode *inode;
3132 	struct folio *folio;
3133 
3134 	len = strlen(symname) + 1;
3135 	if (len > PAGE_SIZE)
3136 		return -ENAMETOOLONG;
3137 
3138 	inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3139 				VM_NORESERVE);
3140 	if (!inode)
3141 		return -ENOSPC;
3142 
3143 	error = security_inode_init_security(inode, dir, &dentry->d_name,
3144 					     shmem_initxattrs, NULL);
3145 	if (error && error != -EOPNOTSUPP) {
3146 		iput(inode);
3147 		return error;
3148 	}
3149 
3150 	inode->i_size = len-1;
3151 	if (len <= SHORT_SYMLINK_LEN) {
3152 		inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3153 		if (!inode->i_link) {
3154 			iput(inode);
3155 			return -ENOMEM;
3156 		}
3157 		inode->i_op = &shmem_short_symlink_operations;
3158 	} else {
3159 		inode_nohighmem(inode);
3160 		error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
3161 		if (error) {
3162 			iput(inode);
3163 			return error;
3164 		}
3165 		inode->i_mapping->a_ops = &shmem_aops;
3166 		inode->i_op = &shmem_symlink_inode_operations;
3167 		memcpy(folio_address(folio), symname, len);
3168 		folio_mark_uptodate(folio);
3169 		folio_mark_dirty(folio);
3170 		folio_unlock(folio);
3171 		folio_put(folio);
3172 	}
3173 	dir->i_size += BOGO_DIRENT_SIZE;
3174 	dir->i_ctime = dir->i_mtime = current_time(dir);
3175 	inode_inc_iversion(dir);
3176 	d_instantiate(dentry, inode);
3177 	dget(dentry);
3178 	return 0;
3179 }
3180 
3181 static void shmem_put_link(void *arg)
3182 {
3183 	folio_mark_accessed(arg);
3184 	folio_put(arg);
3185 }
3186 
3187 static const char *shmem_get_link(struct dentry *dentry,
3188 				  struct inode *inode,
3189 				  struct delayed_call *done)
3190 {
3191 	struct folio *folio = NULL;
3192 	int error;
3193 
3194 	if (!dentry) {
3195 		folio = filemap_get_folio(inode->i_mapping, 0);
3196 		if (!folio)
3197 			return ERR_PTR(-ECHILD);
3198 		if (PageHWPoison(folio_page(folio, 0)) ||
3199 		    !folio_test_uptodate(folio)) {
3200 			folio_put(folio);
3201 			return ERR_PTR(-ECHILD);
3202 		}
3203 	} else {
3204 		error = shmem_get_folio(inode, 0, &folio, SGP_READ);
3205 		if (error)
3206 			return ERR_PTR(error);
3207 		if (!folio)
3208 			return ERR_PTR(-ECHILD);
3209 		if (PageHWPoison(folio_page(folio, 0))) {
3210 			folio_unlock(folio);
3211 			folio_put(folio);
3212 			return ERR_PTR(-ECHILD);
3213 		}
3214 		folio_unlock(folio);
3215 	}
3216 	set_delayed_call(done, shmem_put_link, folio);
3217 	return folio_address(folio);
3218 }
3219 
3220 #ifdef CONFIG_TMPFS_XATTR
3221 
3222 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3223 {
3224 	struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3225 
3226 	fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3227 
3228 	return 0;
3229 }
3230 
3231 static int shmem_fileattr_set(struct mnt_idmap *idmap,
3232 			      struct dentry *dentry, struct fileattr *fa)
3233 {
3234 	struct inode *inode = d_inode(dentry);
3235 	struct shmem_inode_info *info = SHMEM_I(inode);
3236 
3237 	if (fileattr_has_fsx(fa))
3238 		return -EOPNOTSUPP;
3239 	if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3240 		return -EOPNOTSUPP;
3241 
3242 	info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3243 		(fa->flags & SHMEM_FL_USER_MODIFIABLE);
3244 
3245 	shmem_set_inode_flags(inode, info->fsflags);
3246 	inode->i_ctime = current_time(inode);
3247 	inode_inc_iversion(inode);
3248 	return 0;
3249 }
3250 
3251 /*
3252  * Superblocks without xattr inode operations may get some security.* xattr
3253  * support from the LSM "for free". As soon as we have any other xattrs
3254  * like ACLs, we also need to implement the security.* handlers at
3255  * filesystem level, though.
3256  */
3257 
3258 /*
3259  * Callback for security_inode_init_security() for acquiring xattrs.
3260  */
3261 static int shmem_initxattrs(struct inode *inode,
3262 			    const struct xattr *xattr_array,
3263 			    void *fs_info)
3264 {
3265 	struct shmem_inode_info *info = SHMEM_I(inode);
3266 	const struct xattr *xattr;
3267 	struct simple_xattr *new_xattr;
3268 	size_t len;
3269 
3270 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3271 		new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3272 		if (!new_xattr)
3273 			return -ENOMEM;
3274 
3275 		len = strlen(xattr->name) + 1;
3276 		new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3277 					  GFP_KERNEL);
3278 		if (!new_xattr->name) {
3279 			kvfree(new_xattr);
3280 			return -ENOMEM;
3281 		}
3282 
3283 		memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3284 		       XATTR_SECURITY_PREFIX_LEN);
3285 		memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3286 		       xattr->name, len);
3287 
3288 		simple_xattr_add(&info->xattrs, new_xattr);
3289 	}
3290 
3291 	return 0;
3292 }
3293 
3294 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3295 				   struct dentry *unused, struct inode *inode,
3296 				   const char *name, void *buffer, size_t size)
3297 {
3298 	struct shmem_inode_info *info = SHMEM_I(inode);
3299 
3300 	name = xattr_full_name(handler, name);
3301 	return simple_xattr_get(&info->xattrs, name, buffer, size);
3302 }
3303 
3304 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3305 				   struct mnt_idmap *idmap,
3306 				   struct dentry *unused, struct inode *inode,
3307 				   const char *name, const void *value,
3308 				   size_t size, int flags)
3309 {
3310 	struct shmem_inode_info *info = SHMEM_I(inode);
3311 	int err;
3312 
3313 	name = xattr_full_name(handler, name);
3314 	err = simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3315 	if (!err) {
3316 		inode->i_ctime = current_time(inode);
3317 		inode_inc_iversion(inode);
3318 	}
3319 	return err;
3320 }
3321 
3322 static const struct xattr_handler shmem_security_xattr_handler = {
3323 	.prefix = XATTR_SECURITY_PREFIX,
3324 	.get = shmem_xattr_handler_get,
3325 	.set = shmem_xattr_handler_set,
3326 };
3327 
3328 static const struct xattr_handler shmem_trusted_xattr_handler = {
3329 	.prefix = XATTR_TRUSTED_PREFIX,
3330 	.get = shmem_xattr_handler_get,
3331 	.set = shmem_xattr_handler_set,
3332 };
3333 
3334 static const struct xattr_handler *shmem_xattr_handlers[] = {
3335 #ifdef CONFIG_TMPFS_POSIX_ACL
3336 	&posix_acl_access_xattr_handler,
3337 	&posix_acl_default_xattr_handler,
3338 #endif
3339 	&shmem_security_xattr_handler,
3340 	&shmem_trusted_xattr_handler,
3341 	NULL
3342 };
3343 
3344 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3345 {
3346 	struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3347 	return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3348 }
3349 #endif /* CONFIG_TMPFS_XATTR */
3350 
3351 static const struct inode_operations shmem_short_symlink_operations = {
3352 	.getattr	= shmem_getattr,
3353 	.get_link	= simple_get_link,
3354 #ifdef CONFIG_TMPFS_XATTR
3355 	.listxattr	= shmem_listxattr,
3356 #endif
3357 };
3358 
3359 static const struct inode_operations shmem_symlink_inode_operations = {
3360 	.getattr	= shmem_getattr,
3361 	.get_link	= shmem_get_link,
3362 #ifdef CONFIG_TMPFS_XATTR
3363 	.listxattr	= shmem_listxattr,
3364 #endif
3365 };
3366 
3367 static struct dentry *shmem_get_parent(struct dentry *child)
3368 {
3369 	return ERR_PTR(-ESTALE);
3370 }
3371 
3372 static int shmem_match(struct inode *ino, void *vfh)
3373 {
3374 	__u32 *fh = vfh;
3375 	__u64 inum = fh[2];
3376 	inum = (inum << 32) | fh[1];
3377 	return ino->i_ino == inum && fh[0] == ino->i_generation;
3378 }
3379 
3380 /* Find any alias of inode, but prefer a hashed alias */
3381 static struct dentry *shmem_find_alias(struct inode *inode)
3382 {
3383 	struct dentry *alias = d_find_alias(inode);
3384 
3385 	return alias ?: d_find_any_alias(inode);
3386 }
3387 
3388 
3389 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3390 		struct fid *fid, int fh_len, int fh_type)
3391 {
3392 	struct inode *inode;
3393 	struct dentry *dentry = NULL;
3394 	u64 inum;
3395 
3396 	if (fh_len < 3)
3397 		return NULL;
3398 
3399 	inum = fid->raw[2];
3400 	inum = (inum << 32) | fid->raw[1];
3401 
3402 	inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3403 			shmem_match, fid->raw);
3404 	if (inode) {
3405 		dentry = shmem_find_alias(inode);
3406 		iput(inode);
3407 	}
3408 
3409 	return dentry;
3410 }
3411 
3412 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3413 				struct inode *parent)
3414 {
3415 	if (*len < 3) {
3416 		*len = 3;
3417 		return FILEID_INVALID;
3418 	}
3419 
3420 	if (inode_unhashed(inode)) {
3421 		/* Unfortunately insert_inode_hash is not idempotent,
3422 		 * so as we hash inodes here rather than at creation
3423 		 * time, we need a lock to ensure we only try
3424 		 * to do it once
3425 		 */
3426 		static DEFINE_SPINLOCK(lock);
3427 		spin_lock(&lock);
3428 		if (inode_unhashed(inode))
3429 			__insert_inode_hash(inode,
3430 					    inode->i_ino + inode->i_generation);
3431 		spin_unlock(&lock);
3432 	}
3433 
3434 	fh[0] = inode->i_generation;
3435 	fh[1] = inode->i_ino;
3436 	fh[2] = ((__u64)inode->i_ino) >> 32;
3437 
3438 	*len = 3;
3439 	return 1;
3440 }
3441 
3442 static const struct export_operations shmem_export_ops = {
3443 	.get_parent     = shmem_get_parent,
3444 	.encode_fh      = shmem_encode_fh,
3445 	.fh_to_dentry	= shmem_fh_to_dentry,
3446 };
3447 
3448 enum shmem_param {
3449 	Opt_gid,
3450 	Opt_huge,
3451 	Opt_mode,
3452 	Opt_mpol,
3453 	Opt_nr_blocks,
3454 	Opt_nr_inodes,
3455 	Opt_size,
3456 	Opt_uid,
3457 	Opt_inode32,
3458 	Opt_inode64,
3459 };
3460 
3461 static const struct constant_table shmem_param_enums_huge[] = {
3462 	{"never",	SHMEM_HUGE_NEVER },
3463 	{"always",	SHMEM_HUGE_ALWAYS },
3464 	{"within_size",	SHMEM_HUGE_WITHIN_SIZE },
3465 	{"advise",	SHMEM_HUGE_ADVISE },
3466 	{}
3467 };
3468 
3469 const struct fs_parameter_spec shmem_fs_parameters[] = {
3470 	fsparam_u32   ("gid",		Opt_gid),
3471 	fsparam_enum  ("huge",		Opt_huge,  shmem_param_enums_huge),
3472 	fsparam_u32oct("mode",		Opt_mode),
3473 	fsparam_string("mpol",		Opt_mpol),
3474 	fsparam_string("nr_blocks",	Opt_nr_blocks),
3475 	fsparam_string("nr_inodes",	Opt_nr_inodes),
3476 	fsparam_string("size",		Opt_size),
3477 	fsparam_u32   ("uid",		Opt_uid),
3478 	fsparam_flag  ("inode32",	Opt_inode32),
3479 	fsparam_flag  ("inode64",	Opt_inode64),
3480 	{}
3481 };
3482 
3483 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3484 {
3485 	struct shmem_options *ctx = fc->fs_private;
3486 	struct fs_parse_result result;
3487 	unsigned long long size;
3488 	char *rest;
3489 	int opt;
3490 
3491 	opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3492 	if (opt < 0)
3493 		return opt;
3494 
3495 	switch (opt) {
3496 	case Opt_size:
3497 		size = memparse(param->string, &rest);
3498 		if (*rest == '%') {
3499 			size <<= PAGE_SHIFT;
3500 			size *= totalram_pages();
3501 			do_div(size, 100);
3502 			rest++;
3503 		}
3504 		if (*rest)
3505 			goto bad_value;
3506 		ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3507 		ctx->seen |= SHMEM_SEEN_BLOCKS;
3508 		break;
3509 	case Opt_nr_blocks:
3510 		ctx->blocks = memparse(param->string, &rest);
3511 		if (*rest || ctx->blocks > S64_MAX)
3512 			goto bad_value;
3513 		ctx->seen |= SHMEM_SEEN_BLOCKS;
3514 		break;
3515 	case Opt_nr_inodes:
3516 		ctx->inodes = memparse(param->string, &rest);
3517 		if (*rest)
3518 			goto bad_value;
3519 		ctx->seen |= SHMEM_SEEN_INODES;
3520 		break;
3521 	case Opt_mode:
3522 		ctx->mode = result.uint_32 & 07777;
3523 		break;
3524 	case Opt_uid:
3525 		ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3526 		if (!uid_valid(ctx->uid))
3527 			goto bad_value;
3528 		break;
3529 	case Opt_gid:
3530 		ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3531 		if (!gid_valid(ctx->gid))
3532 			goto bad_value;
3533 		break;
3534 	case Opt_huge:
3535 		ctx->huge = result.uint_32;
3536 		if (ctx->huge != SHMEM_HUGE_NEVER &&
3537 		    !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3538 		      has_transparent_hugepage()))
3539 			goto unsupported_parameter;
3540 		ctx->seen |= SHMEM_SEEN_HUGE;
3541 		break;
3542 	case Opt_mpol:
3543 		if (IS_ENABLED(CONFIG_NUMA)) {
3544 			mpol_put(ctx->mpol);
3545 			ctx->mpol = NULL;
3546 			if (mpol_parse_str(param->string, &ctx->mpol))
3547 				goto bad_value;
3548 			break;
3549 		}
3550 		goto unsupported_parameter;
3551 	case Opt_inode32:
3552 		ctx->full_inums = false;
3553 		ctx->seen |= SHMEM_SEEN_INUMS;
3554 		break;
3555 	case Opt_inode64:
3556 		if (sizeof(ino_t) < 8) {
3557 			return invalfc(fc,
3558 				       "Cannot use inode64 with <64bit inums in kernel\n");
3559 		}
3560 		ctx->full_inums = true;
3561 		ctx->seen |= SHMEM_SEEN_INUMS;
3562 		break;
3563 	}
3564 	return 0;
3565 
3566 unsupported_parameter:
3567 	return invalfc(fc, "Unsupported parameter '%s'", param->key);
3568 bad_value:
3569 	return invalfc(fc, "Bad value for '%s'", param->key);
3570 }
3571 
3572 static int shmem_parse_options(struct fs_context *fc, void *data)
3573 {
3574 	char *options = data;
3575 
3576 	if (options) {
3577 		int err = security_sb_eat_lsm_opts(options, &fc->security);
3578 		if (err)
3579 			return err;
3580 	}
3581 
3582 	while (options != NULL) {
3583 		char *this_char = options;
3584 		for (;;) {
3585 			/*
3586 			 * NUL-terminate this option: unfortunately,
3587 			 * mount options form a comma-separated list,
3588 			 * but mpol's nodelist may also contain commas.
3589 			 */
3590 			options = strchr(options, ',');
3591 			if (options == NULL)
3592 				break;
3593 			options++;
3594 			if (!isdigit(*options)) {
3595 				options[-1] = '\0';
3596 				break;
3597 			}
3598 		}
3599 		if (*this_char) {
3600 			char *value = strchr(this_char, '=');
3601 			size_t len = 0;
3602 			int err;
3603 
3604 			if (value) {
3605 				*value++ = '\0';
3606 				len = strlen(value);
3607 			}
3608 			err = vfs_parse_fs_string(fc, this_char, value, len);
3609 			if (err < 0)
3610 				return err;
3611 		}
3612 	}
3613 	return 0;
3614 }
3615 
3616 /*
3617  * Reconfigure a shmem filesystem.
3618  *
3619  * Note that we disallow change from limited->unlimited blocks/inodes while any
3620  * are in use; but we must separately disallow unlimited->limited, because in
3621  * that case we have no record of how much is already in use.
3622  */
3623 static int shmem_reconfigure(struct fs_context *fc)
3624 {
3625 	struct shmem_options *ctx = fc->fs_private;
3626 	struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3627 	unsigned long inodes;
3628 	struct mempolicy *mpol = NULL;
3629 	const char *err;
3630 
3631 	raw_spin_lock(&sbinfo->stat_lock);
3632 	inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3633 
3634 	if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3635 		if (!sbinfo->max_blocks) {
3636 			err = "Cannot retroactively limit size";
3637 			goto out;
3638 		}
3639 		if (percpu_counter_compare(&sbinfo->used_blocks,
3640 					   ctx->blocks) > 0) {
3641 			err = "Too small a size for current use";
3642 			goto out;
3643 		}
3644 	}
3645 	if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3646 		if (!sbinfo->max_inodes) {
3647 			err = "Cannot retroactively limit inodes";
3648 			goto out;
3649 		}
3650 		if (ctx->inodes < inodes) {
3651 			err = "Too few inodes for current use";
3652 			goto out;
3653 		}
3654 	}
3655 
3656 	if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3657 	    sbinfo->next_ino > UINT_MAX) {
3658 		err = "Current inum too high to switch to 32-bit inums";
3659 		goto out;
3660 	}
3661 
3662 	if (ctx->seen & SHMEM_SEEN_HUGE)
3663 		sbinfo->huge = ctx->huge;
3664 	if (ctx->seen & SHMEM_SEEN_INUMS)
3665 		sbinfo->full_inums = ctx->full_inums;
3666 	if (ctx->seen & SHMEM_SEEN_BLOCKS)
3667 		sbinfo->max_blocks  = ctx->blocks;
3668 	if (ctx->seen & SHMEM_SEEN_INODES) {
3669 		sbinfo->max_inodes  = ctx->inodes;
3670 		sbinfo->free_inodes = ctx->inodes - inodes;
3671 	}
3672 
3673 	/*
3674 	 * Preserve previous mempolicy unless mpol remount option was specified.
3675 	 */
3676 	if (ctx->mpol) {
3677 		mpol = sbinfo->mpol;
3678 		sbinfo->mpol = ctx->mpol;	/* transfers initial ref */
3679 		ctx->mpol = NULL;
3680 	}
3681 	raw_spin_unlock(&sbinfo->stat_lock);
3682 	mpol_put(mpol);
3683 	return 0;
3684 out:
3685 	raw_spin_unlock(&sbinfo->stat_lock);
3686 	return invalfc(fc, "%s", err);
3687 }
3688 
3689 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3690 {
3691 	struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3692 
3693 	if (sbinfo->max_blocks != shmem_default_max_blocks())
3694 		seq_printf(seq, ",size=%luk",
3695 			sbinfo->max_blocks << (PAGE_SHIFT - 10));
3696 	if (sbinfo->max_inodes != shmem_default_max_inodes())
3697 		seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3698 	if (sbinfo->mode != (0777 | S_ISVTX))
3699 		seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3700 	if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3701 		seq_printf(seq, ",uid=%u",
3702 				from_kuid_munged(&init_user_ns, sbinfo->uid));
3703 	if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3704 		seq_printf(seq, ",gid=%u",
3705 				from_kgid_munged(&init_user_ns, sbinfo->gid));
3706 
3707 	/*
3708 	 * Showing inode{64,32} might be useful even if it's the system default,
3709 	 * since then people don't have to resort to checking both here and
3710 	 * /proc/config.gz to confirm 64-bit inums were successfully applied
3711 	 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3712 	 *
3713 	 * We hide it when inode64 isn't the default and we are using 32-bit
3714 	 * inodes, since that probably just means the feature isn't even under
3715 	 * consideration.
3716 	 *
3717 	 * As such:
3718 	 *
3719 	 *                     +-----------------+-----------------+
3720 	 *                     | TMPFS_INODE64=y | TMPFS_INODE64=n |
3721 	 *  +------------------+-----------------+-----------------+
3722 	 *  | full_inums=true  | show            | show            |
3723 	 *  | full_inums=false | show            | hide            |
3724 	 *  +------------------+-----------------+-----------------+
3725 	 *
3726 	 */
3727 	if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3728 		seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3729 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3730 	/* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3731 	if (sbinfo->huge)
3732 		seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3733 #endif
3734 	shmem_show_mpol(seq, sbinfo->mpol);
3735 	return 0;
3736 }
3737 
3738 #endif /* CONFIG_TMPFS */
3739 
3740 static void shmem_put_super(struct super_block *sb)
3741 {
3742 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3743 
3744 	free_percpu(sbinfo->ino_batch);
3745 	percpu_counter_destroy(&sbinfo->used_blocks);
3746 	mpol_put(sbinfo->mpol);
3747 	kfree(sbinfo);
3748 	sb->s_fs_info = NULL;
3749 }
3750 
3751 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3752 {
3753 	struct shmem_options *ctx = fc->fs_private;
3754 	struct inode *inode;
3755 	struct shmem_sb_info *sbinfo;
3756 
3757 	/* Round up to L1_CACHE_BYTES to resist false sharing */
3758 	sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3759 				L1_CACHE_BYTES), GFP_KERNEL);
3760 	if (!sbinfo)
3761 		return -ENOMEM;
3762 
3763 	sb->s_fs_info = sbinfo;
3764 
3765 #ifdef CONFIG_TMPFS
3766 	/*
3767 	 * Per default we only allow half of the physical ram per
3768 	 * tmpfs instance, limiting inodes to one per page of lowmem;
3769 	 * but the internal instance is left unlimited.
3770 	 */
3771 	if (!(sb->s_flags & SB_KERNMOUNT)) {
3772 		if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3773 			ctx->blocks = shmem_default_max_blocks();
3774 		if (!(ctx->seen & SHMEM_SEEN_INODES))
3775 			ctx->inodes = shmem_default_max_inodes();
3776 		if (!(ctx->seen & SHMEM_SEEN_INUMS))
3777 			ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3778 	} else {
3779 		sb->s_flags |= SB_NOUSER;
3780 	}
3781 	sb->s_export_op = &shmem_export_ops;
3782 	sb->s_flags |= SB_NOSEC | SB_I_VERSION;
3783 #else
3784 	sb->s_flags |= SB_NOUSER;
3785 #endif
3786 	sbinfo->max_blocks = ctx->blocks;
3787 	sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3788 	if (sb->s_flags & SB_KERNMOUNT) {
3789 		sbinfo->ino_batch = alloc_percpu(ino_t);
3790 		if (!sbinfo->ino_batch)
3791 			goto failed;
3792 	}
3793 	sbinfo->uid = ctx->uid;
3794 	sbinfo->gid = ctx->gid;
3795 	sbinfo->full_inums = ctx->full_inums;
3796 	sbinfo->mode = ctx->mode;
3797 	sbinfo->huge = ctx->huge;
3798 	sbinfo->mpol = ctx->mpol;
3799 	ctx->mpol = NULL;
3800 
3801 	raw_spin_lock_init(&sbinfo->stat_lock);
3802 	if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3803 		goto failed;
3804 	spin_lock_init(&sbinfo->shrinklist_lock);
3805 	INIT_LIST_HEAD(&sbinfo->shrinklist);
3806 
3807 	sb->s_maxbytes = MAX_LFS_FILESIZE;
3808 	sb->s_blocksize = PAGE_SIZE;
3809 	sb->s_blocksize_bits = PAGE_SHIFT;
3810 	sb->s_magic = TMPFS_MAGIC;
3811 	sb->s_op = &shmem_ops;
3812 	sb->s_time_gran = 1;
3813 #ifdef CONFIG_TMPFS_XATTR
3814 	sb->s_xattr = shmem_xattr_handlers;
3815 #endif
3816 #ifdef CONFIG_TMPFS_POSIX_ACL
3817 	sb->s_flags |= SB_POSIXACL;
3818 #endif
3819 	uuid_gen(&sb->s_uuid);
3820 
3821 	inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, S_IFDIR | sbinfo->mode, 0,
3822 				VM_NORESERVE);
3823 	if (!inode)
3824 		goto failed;
3825 	inode->i_uid = sbinfo->uid;
3826 	inode->i_gid = sbinfo->gid;
3827 	sb->s_root = d_make_root(inode);
3828 	if (!sb->s_root)
3829 		goto failed;
3830 	return 0;
3831 
3832 failed:
3833 	shmem_put_super(sb);
3834 	return -ENOMEM;
3835 }
3836 
3837 static int shmem_get_tree(struct fs_context *fc)
3838 {
3839 	return get_tree_nodev(fc, shmem_fill_super);
3840 }
3841 
3842 static void shmem_free_fc(struct fs_context *fc)
3843 {
3844 	struct shmem_options *ctx = fc->fs_private;
3845 
3846 	if (ctx) {
3847 		mpol_put(ctx->mpol);
3848 		kfree(ctx);
3849 	}
3850 }
3851 
3852 static const struct fs_context_operations shmem_fs_context_ops = {
3853 	.free			= shmem_free_fc,
3854 	.get_tree		= shmem_get_tree,
3855 #ifdef CONFIG_TMPFS
3856 	.parse_monolithic	= shmem_parse_options,
3857 	.parse_param		= shmem_parse_one,
3858 	.reconfigure		= shmem_reconfigure,
3859 #endif
3860 };
3861 
3862 static struct kmem_cache *shmem_inode_cachep;
3863 
3864 static struct inode *shmem_alloc_inode(struct super_block *sb)
3865 {
3866 	struct shmem_inode_info *info;
3867 	info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
3868 	if (!info)
3869 		return NULL;
3870 	return &info->vfs_inode;
3871 }
3872 
3873 static void shmem_free_in_core_inode(struct inode *inode)
3874 {
3875 	if (S_ISLNK(inode->i_mode))
3876 		kfree(inode->i_link);
3877 	kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3878 }
3879 
3880 static void shmem_destroy_inode(struct inode *inode)
3881 {
3882 	if (S_ISREG(inode->i_mode))
3883 		mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3884 }
3885 
3886 static void shmem_init_inode(void *foo)
3887 {
3888 	struct shmem_inode_info *info = foo;
3889 	inode_init_once(&info->vfs_inode);
3890 }
3891 
3892 static void shmem_init_inodecache(void)
3893 {
3894 	shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3895 				sizeof(struct shmem_inode_info),
3896 				0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3897 }
3898 
3899 static void shmem_destroy_inodecache(void)
3900 {
3901 	kmem_cache_destroy(shmem_inode_cachep);
3902 }
3903 
3904 /* Keep the page in page cache instead of truncating it */
3905 static int shmem_error_remove_page(struct address_space *mapping,
3906 				   struct page *page)
3907 {
3908 	return 0;
3909 }
3910 
3911 const struct address_space_operations shmem_aops = {
3912 	.writepage	= shmem_writepage,
3913 	.dirty_folio	= noop_dirty_folio,
3914 #ifdef CONFIG_TMPFS
3915 	.write_begin	= shmem_write_begin,
3916 	.write_end	= shmem_write_end,
3917 #endif
3918 #ifdef CONFIG_MIGRATION
3919 	.migrate_folio	= migrate_folio,
3920 #endif
3921 	.error_remove_page = shmem_error_remove_page,
3922 };
3923 EXPORT_SYMBOL(shmem_aops);
3924 
3925 static const struct file_operations shmem_file_operations = {
3926 	.mmap		= shmem_mmap,
3927 	.open		= generic_file_open,
3928 	.get_unmapped_area = shmem_get_unmapped_area,
3929 #ifdef CONFIG_TMPFS
3930 	.llseek		= shmem_file_llseek,
3931 	.read_iter	= shmem_file_read_iter,
3932 	.write_iter	= generic_file_write_iter,
3933 	.fsync		= noop_fsync,
3934 	.splice_read	= generic_file_splice_read,
3935 	.splice_write	= iter_file_splice_write,
3936 	.fallocate	= shmem_fallocate,
3937 #endif
3938 };
3939 
3940 static const struct inode_operations shmem_inode_operations = {
3941 	.getattr	= shmem_getattr,
3942 	.setattr	= shmem_setattr,
3943 #ifdef CONFIG_TMPFS_XATTR
3944 	.listxattr	= shmem_listxattr,
3945 	.set_acl	= simple_set_acl,
3946 	.fileattr_get	= shmem_fileattr_get,
3947 	.fileattr_set	= shmem_fileattr_set,
3948 #endif
3949 };
3950 
3951 static const struct inode_operations shmem_dir_inode_operations = {
3952 #ifdef CONFIG_TMPFS
3953 	.getattr	= shmem_getattr,
3954 	.create		= shmem_create,
3955 	.lookup		= simple_lookup,
3956 	.link		= shmem_link,
3957 	.unlink		= shmem_unlink,
3958 	.symlink	= shmem_symlink,
3959 	.mkdir		= shmem_mkdir,
3960 	.rmdir		= shmem_rmdir,
3961 	.mknod		= shmem_mknod,
3962 	.rename		= shmem_rename2,
3963 	.tmpfile	= shmem_tmpfile,
3964 #endif
3965 #ifdef CONFIG_TMPFS_XATTR
3966 	.listxattr	= shmem_listxattr,
3967 	.fileattr_get	= shmem_fileattr_get,
3968 	.fileattr_set	= shmem_fileattr_set,
3969 #endif
3970 #ifdef CONFIG_TMPFS_POSIX_ACL
3971 	.setattr	= shmem_setattr,
3972 	.set_acl	= simple_set_acl,
3973 #endif
3974 };
3975 
3976 static const struct inode_operations shmem_special_inode_operations = {
3977 	.getattr	= shmem_getattr,
3978 #ifdef CONFIG_TMPFS_XATTR
3979 	.listxattr	= shmem_listxattr,
3980 #endif
3981 #ifdef CONFIG_TMPFS_POSIX_ACL
3982 	.setattr	= shmem_setattr,
3983 	.set_acl	= simple_set_acl,
3984 #endif
3985 };
3986 
3987 static const struct super_operations shmem_ops = {
3988 	.alloc_inode	= shmem_alloc_inode,
3989 	.free_inode	= shmem_free_in_core_inode,
3990 	.destroy_inode	= shmem_destroy_inode,
3991 #ifdef CONFIG_TMPFS
3992 	.statfs		= shmem_statfs,
3993 	.show_options	= shmem_show_options,
3994 #endif
3995 	.evict_inode	= shmem_evict_inode,
3996 	.drop_inode	= generic_delete_inode,
3997 	.put_super	= shmem_put_super,
3998 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3999 	.nr_cached_objects	= shmem_unused_huge_count,
4000 	.free_cached_objects	= shmem_unused_huge_scan,
4001 #endif
4002 };
4003 
4004 static const struct vm_operations_struct shmem_vm_ops = {
4005 	.fault		= shmem_fault,
4006 	.map_pages	= filemap_map_pages,
4007 #ifdef CONFIG_NUMA
4008 	.set_policy     = shmem_set_policy,
4009 	.get_policy     = shmem_get_policy,
4010 #endif
4011 };
4012 
4013 static const struct vm_operations_struct shmem_anon_vm_ops = {
4014 	.fault		= shmem_fault,
4015 	.map_pages	= filemap_map_pages,
4016 #ifdef CONFIG_NUMA
4017 	.set_policy     = shmem_set_policy,
4018 	.get_policy     = shmem_get_policy,
4019 #endif
4020 };
4021 
4022 int shmem_init_fs_context(struct fs_context *fc)
4023 {
4024 	struct shmem_options *ctx;
4025 
4026 	ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4027 	if (!ctx)
4028 		return -ENOMEM;
4029 
4030 	ctx->mode = 0777 | S_ISVTX;
4031 	ctx->uid = current_fsuid();
4032 	ctx->gid = current_fsgid();
4033 
4034 	fc->fs_private = ctx;
4035 	fc->ops = &shmem_fs_context_ops;
4036 	return 0;
4037 }
4038 
4039 static struct file_system_type shmem_fs_type = {
4040 	.owner		= THIS_MODULE,
4041 	.name		= "tmpfs",
4042 	.init_fs_context = shmem_init_fs_context,
4043 #ifdef CONFIG_TMPFS
4044 	.parameters	= shmem_fs_parameters,
4045 #endif
4046 	.kill_sb	= kill_litter_super,
4047 #ifdef CONFIG_SHMEM
4048 	.fs_flags	= FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4049 #else
4050 	.fs_flags	= FS_USERNS_MOUNT,
4051 #endif
4052 };
4053 
4054 void __init shmem_init(void)
4055 {
4056 	int error;
4057 
4058 	shmem_init_inodecache();
4059 
4060 	error = register_filesystem(&shmem_fs_type);
4061 	if (error) {
4062 		pr_err("Could not register tmpfs\n");
4063 		goto out2;
4064 	}
4065 
4066 	shm_mnt = kern_mount(&shmem_fs_type);
4067 	if (IS_ERR(shm_mnt)) {
4068 		error = PTR_ERR(shm_mnt);
4069 		pr_err("Could not kern_mount tmpfs\n");
4070 		goto out1;
4071 	}
4072 
4073 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4074 	if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4075 		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4076 	else
4077 		shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4078 #endif
4079 	return;
4080 
4081 out1:
4082 	unregister_filesystem(&shmem_fs_type);
4083 out2:
4084 	shmem_destroy_inodecache();
4085 	shm_mnt = ERR_PTR(error);
4086 }
4087 
4088 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4089 static ssize_t shmem_enabled_show(struct kobject *kobj,
4090 				  struct kobj_attribute *attr, char *buf)
4091 {
4092 	static const int values[] = {
4093 		SHMEM_HUGE_ALWAYS,
4094 		SHMEM_HUGE_WITHIN_SIZE,
4095 		SHMEM_HUGE_ADVISE,
4096 		SHMEM_HUGE_NEVER,
4097 		SHMEM_HUGE_DENY,
4098 		SHMEM_HUGE_FORCE,
4099 	};
4100 	int len = 0;
4101 	int i;
4102 
4103 	for (i = 0; i < ARRAY_SIZE(values); i++) {
4104 		len += sysfs_emit_at(buf, len,
4105 				     shmem_huge == values[i] ? "%s[%s]" : "%s%s",
4106 				     i ? " " : "",
4107 				     shmem_format_huge(values[i]));
4108 	}
4109 
4110 	len += sysfs_emit_at(buf, len, "\n");
4111 
4112 	return len;
4113 }
4114 
4115 static ssize_t shmem_enabled_store(struct kobject *kobj,
4116 		struct kobj_attribute *attr, const char *buf, size_t count)
4117 {
4118 	char tmp[16];
4119 	int huge;
4120 
4121 	if (count + 1 > sizeof(tmp))
4122 		return -EINVAL;
4123 	memcpy(tmp, buf, count);
4124 	tmp[count] = '\0';
4125 	if (count && tmp[count - 1] == '\n')
4126 		tmp[count - 1] = '\0';
4127 
4128 	huge = shmem_parse_huge(tmp);
4129 	if (huge == -EINVAL)
4130 		return -EINVAL;
4131 	if (!has_transparent_hugepage() &&
4132 			huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4133 		return -EINVAL;
4134 
4135 	shmem_huge = huge;
4136 	if (shmem_huge > SHMEM_HUGE_DENY)
4137 		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4138 	return count;
4139 }
4140 
4141 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
4142 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4143 
4144 #else /* !CONFIG_SHMEM */
4145 
4146 /*
4147  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4148  *
4149  * This is intended for small system where the benefits of the full
4150  * shmem code (swap-backed and resource-limited) are outweighed by
4151  * their complexity. On systems without swap this code should be
4152  * effectively equivalent, but much lighter weight.
4153  */
4154 
4155 static struct file_system_type shmem_fs_type = {
4156 	.name		= "tmpfs",
4157 	.init_fs_context = ramfs_init_fs_context,
4158 	.parameters	= ramfs_fs_parameters,
4159 	.kill_sb	= kill_litter_super,
4160 	.fs_flags	= FS_USERNS_MOUNT,
4161 };
4162 
4163 void __init shmem_init(void)
4164 {
4165 	BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4166 
4167 	shm_mnt = kern_mount(&shmem_fs_type);
4168 	BUG_ON(IS_ERR(shm_mnt));
4169 }
4170 
4171 int shmem_unuse(unsigned int type)
4172 {
4173 	return 0;
4174 }
4175 
4176 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
4177 {
4178 	return 0;
4179 }
4180 
4181 void shmem_unlock_mapping(struct address_space *mapping)
4182 {
4183 }
4184 
4185 #ifdef CONFIG_MMU
4186 unsigned long shmem_get_unmapped_area(struct file *file,
4187 				      unsigned long addr, unsigned long len,
4188 				      unsigned long pgoff, unsigned long flags)
4189 {
4190 	return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4191 }
4192 #endif
4193 
4194 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4195 {
4196 	truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4197 }
4198 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4199 
4200 #define shmem_vm_ops				generic_file_vm_ops
4201 #define shmem_anon_vm_ops			generic_file_vm_ops
4202 #define shmem_file_operations			ramfs_file_operations
4203 #define shmem_get_inode(idmap, sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
4204 #define shmem_acct_size(flags, size)		0
4205 #define shmem_unacct_size(flags, size)		do {} while (0)
4206 
4207 #endif /* CONFIG_SHMEM */
4208 
4209 /* common code */
4210 
4211 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4212 				       unsigned long flags, unsigned int i_flags)
4213 {
4214 	struct inode *inode;
4215 	struct file *res;
4216 
4217 	if (IS_ERR(mnt))
4218 		return ERR_CAST(mnt);
4219 
4220 	if (size < 0 || size > MAX_LFS_FILESIZE)
4221 		return ERR_PTR(-EINVAL);
4222 
4223 	if (shmem_acct_size(flags, size))
4224 		return ERR_PTR(-ENOMEM);
4225 
4226 	if (is_idmapped_mnt(mnt))
4227 		return ERR_PTR(-EINVAL);
4228 
4229 	inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
4230 				S_IFREG | S_IRWXUGO, 0, flags);
4231 	if (unlikely(!inode)) {
4232 		shmem_unacct_size(flags, size);
4233 		return ERR_PTR(-ENOSPC);
4234 	}
4235 	inode->i_flags |= i_flags;
4236 	inode->i_size = size;
4237 	clear_nlink(inode);	/* It is unlinked */
4238 	res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4239 	if (!IS_ERR(res))
4240 		res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4241 				&shmem_file_operations);
4242 	if (IS_ERR(res))
4243 		iput(inode);
4244 	return res;
4245 }
4246 
4247 /**
4248  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4249  * 	kernel internal.  There will be NO LSM permission checks against the
4250  * 	underlying inode.  So users of this interface must do LSM checks at a
4251  *	higher layer.  The users are the big_key and shm implementations.  LSM
4252  *	checks are provided at the key or shm level rather than the inode.
4253  * @name: name for dentry (to be seen in /proc/<pid>/maps
4254  * @size: size to be set for the file
4255  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4256  */
4257 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4258 {
4259 	return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4260 }
4261 
4262 /**
4263  * shmem_file_setup - get an unlinked file living in tmpfs
4264  * @name: name for dentry (to be seen in /proc/<pid>/maps
4265  * @size: size to be set for the file
4266  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4267  */
4268 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4269 {
4270 	return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4271 }
4272 EXPORT_SYMBOL_GPL(shmem_file_setup);
4273 
4274 /**
4275  * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4276  * @mnt: the tmpfs mount where the file will be created
4277  * @name: name for dentry (to be seen in /proc/<pid>/maps
4278  * @size: size to be set for the file
4279  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4280  */
4281 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4282 				       loff_t size, unsigned long flags)
4283 {
4284 	return __shmem_file_setup(mnt, name, size, flags, 0);
4285 }
4286 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4287 
4288 /**
4289  * shmem_zero_setup - setup a shared anonymous mapping
4290  * @vma: the vma to be mmapped is prepared by do_mmap
4291  */
4292 int shmem_zero_setup(struct vm_area_struct *vma)
4293 {
4294 	struct file *file;
4295 	loff_t size = vma->vm_end - vma->vm_start;
4296 
4297 	/*
4298 	 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4299 	 * between XFS directory reading and selinux: since this file is only
4300 	 * accessible to the user through its mapping, use S_PRIVATE flag to
4301 	 * bypass file security, in the same way as shmem_kernel_file_setup().
4302 	 */
4303 	file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4304 	if (IS_ERR(file))
4305 		return PTR_ERR(file);
4306 
4307 	if (vma->vm_file)
4308 		fput(vma->vm_file);
4309 	vma->vm_file = file;
4310 	vma->vm_ops = &shmem_anon_vm_ops;
4311 
4312 	return 0;
4313 }
4314 
4315 /**
4316  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4317  * @mapping:	the page's address_space
4318  * @index:	the page index
4319  * @gfp:	the page allocator flags to use if allocating
4320  *
4321  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4322  * with any new page allocations done using the specified allocation flags.
4323  * But read_cache_page_gfp() uses the ->read_folio() method: which does not
4324  * suit tmpfs, since it may have pages in swapcache, and needs to find those
4325  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4326  *
4327  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4328  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4329  */
4330 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4331 					 pgoff_t index, gfp_t gfp)
4332 {
4333 #ifdef CONFIG_SHMEM
4334 	struct inode *inode = mapping->host;
4335 	struct folio *folio;
4336 	struct page *page;
4337 	int error;
4338 
4339 	BUG_ON(!shmem_mapping(mapping));
4340 	error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
4341 				  gfp, NULL, NULL, NULL);
4342 	if (error)
4343 		return ERR_PTR(error);
4344 
4345 	folio_unlock(folio);
4346 	page = folio_file_page(folio, index);
4347 	if (PageHWPoison(page)) {
4348 		folio_put(folio);
4349 		return ERR_PTR(-EIO);
4350 	}
4351 
4352 	return page;
4353 #else
4354 	/*
4355 	 * The tiny !SHMEM case uses ramfs without swap
4356 	 */
4357 	return read_cache_page_gfp(mapping, index, gfp);
4358 #endif
4359 }
4360 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
4361