xref: /linux/mm/shmem.c (revision f4db95b68ae68ebaf91d35cc0487ac1cbd04261e)
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/shmem_fs.h>
37 #include <linux/swap.h>
38 #include <linux/uio.h>
39 #include <linux/hugetlb.h>
40 #include <linux/fs_parser.h>
41 #include <linux/swapfile.h>
42 #include <linux/iversion.h>
43 #include "swap.h"
44 
45 static struct vfsmount *shm_mnt __ro_after_init;
46 
47 #ifdef CONFIG_SHMEM
48 /*
49  * This virtual memory filesystem is heavily based on the ramfs. It
50  * extends ramfs by the ability to use swap and honor resource limits
51  * which makes it a completely usable filesystem.
52  */
53 
54 #include <linux/xattr.h>
55 #include <linux/exportfs.h>
56 #include <linux/posix_acl.h>
57 #include <linux/posix_acl_xattr.h>
58 #include <linux/mman.h>
59 #include <linux/string.h>
60 #include <linux/slab.h>
61 #include <linux/backing-dev.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/rmap.h>
80 #include <linux/uuid.h>
81 #include <linux/quotaops.h>
82 #include <linux/rcupdate_wait.h>
83 
84 #include <linux/uaccess.h>
85 
86 #include "internal.h"
87 
88 #define BLOCKS_PER_PAGE  (PAGE_SIZE/512)
89 #define VM_ACCT(size)    (PAGE_ALIGN(size) >> PAGE_SHIFT)
90 
91 /* Pretend that each entry is of this size in directory's i_size */
92 #define BOGO_DIRENT_SIZE 20
93 
94 /* Pretend that one inode + its dentry occupy this much memory */
95 #define BOGO_INODE_SIZE 1024
96 
97 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
98 #define SHORT_SYMLINK_LEN 128
99 
100 /*
101  * shmem_fallocate communicates with shmem_fault or shmem_writepage via
102  * inode->i_private (with i_rwsem making sure that it has only one user at
103  * a time): we would prefer not to enlarge the shmem inode just for that.
104  */
105 struct shmem_falloc {
106 	wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
107 	pgoff_t start;		/* start of range currently being fallocated */
108 	pgoff_t next;		/* the next page offset to be fallocated */
109 	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
110 	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
111 };
112 
113 struct shmem_options {
114 	unsigned long long blocks;
115 	unsigned long long inodes;
116 	struct mempolicy *mpol;
117 	kuid_t uid;
118 	kgid_t gid;
119 	umode_t mode;
120 	bool full_inums;
121 	int huge;
122 	int seen;
123 	bool noswap;
124 	unsigned short quota_types;
125 	struct shmem_quota_limits qlimits;
126 #define SHMEM_SEEN_BLOCKS 1
127 #define SHMEM_SEEN_INODES 2
128 #define SHMEM_SEEN_HUGE 4
129 #define SHMEM_SEEN_INUMS 8
130 #define SHMEM_SEEN_NOSWAP 16
131 #define SHMEM_SEEN_QUOTA 32
132 };
133 
134 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
135 static unsigned long huge_shmem_orders_always __read_mostly;
136 static unsigned long huge_shmem_orders_madvise __read_mostly;
137 static unsigned long huge_shmem_orders_inherit __read_mostly;
138 static unsigned long huge_shmem_orders_within_size __read_mostly;
139 #endif
140 
141 #ifdef CONFIG_TMPFS
142 static unsigned long shmem_default_max_blocks(void)
143 {
144 	return totalram_pages() / 2;
145 }
146 
147 static unsigned long shmem_default_max_inodes(void)
148 {
149 	unsigned long nr_pages = totalram_pages();
150 
151 	return min3(nr_pages - totalhigh_pages(), nr_pages / 2,
152 			ULONG_MAX / BOGO_INODE_SIZE);
153 }
154 #endif
155 
156 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
157 			struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
158 			struct vm_area_struct *vma, vm_fault_t *fault_type);
159 
160 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
161 {
162 	return sb->s_fs_info;
163 }
164 
165 /*
166  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
167  * for shared memory and for shared anonymous (/dev/zero) mappings
168  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
169  * consistent with the pre-accounting of private mappings ...
170  */
171 static inline int shmem_acct_size(unsigned long flags, loff_t size)
172 {
173 	return (flags & VM_NORESERVE) ?
174 		0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
175 }
176 
177 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
178 {
179 	if (!(flags & VM_NORESERVE))
180 		vm_unacct_memory(VM_ACCT(size));
181 }
182 
183 static inline int shmem_reacct_size(unsigned long flags,
184 		loff_t oldsize, loff_t newsize)
185 {
186 	if (!(flags & VM_NORESERVE)) {
187 		if (VM_ACCT(newsize) > VM_ACCT(oldsize))
188 			return security_vm_enough_memory_mm(current->mm,
189 					VM_ACCT(newsize) - VM_ACCT(oldsize));
190 		else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
191 			vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
192 	}
193 	return 0;
194 }
195 
196 /*
197  * ... whereas tmpfs objects are accounted incrementally as
198  * pages are allocated, in order to allow large sparse files.
199  * shmem_get_folio reports shmem_acct_blocks failure as -ENOSPC not -ENOMEM,
200  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
201  */
202 static inline int shmem_acct_blocks(unsigned long flags, long pages)
203 {
204 	if (!(flags & VM_NORESERVE))
205 		return 0;
206 
207 	return security_vm_enough_memory_mm(current->mm,
208 			pages * VM_ACCT(PAGE_SIZE));
209 }
210 
211 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
212 {
213 	if (flags & VM_NORESERVE)
214 		vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
215 }
216 
217 static int shmem_inode_acct_blocks(struct inode *inode, long pages)
218 {
219 	struct shmem_inode_info *info = SHMEM_I(inode);
220 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
221 	int err = -ENOSPC;
222 
223 	if (shmem_acct_blocks(info->flags, pages))
224 		return err;
225 
226 	might_sleep();	/* when quotas */
227 	if (sbinfo->max_blocks) {
228 		if (!percpu_counter_limited_add(&sbinfo->used_blocks,
229 						sbinfo->max_blocks, pages))
230 			goto unacct;
231 
232 		err = dquot_alloc_block_nodirty(inode, pages);
233 		if (err) {
234 			percpu_counter_sub(&sbinfo->used_blocks, pages);
235 			goto unacct;
236 		}
237 	} else {
238 		err = dquot_alloc_block_nodirty(inode, pages);
239 		if (err)
240 			goto unacct;
241 	}
242 
243 	return 0;
244 
245 unacct:
246 	shmem_unacct_blocks(info->flags, pages);
247 	return err;
248 }
249 
250 static void shmem_inode_unacct_blocks(struct inode *inode, long pages)
251 {
252 	struct shmem_inode_info *info = SHMEM_I(inode);
253 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
254 
255 	might_sleep();	/* when quotas */
256 	dquot_free_block_nodirty(inode, pages);
257 
258 	if (sbinfo->max_blocks)
259 		percpu_counter_sub(&sbinfo->used_blocks, pages);
260 	shmem_unacct_blocks(info->flags, pages);
261 }
262 
263 static const struct super_operations shmem_ops;
264 static const struct address_space_operations shmem_aops;
265 static const struct file_operations shmem_file_operations;
266 static const struct inode_operations shmem_inode_operations;
267 static const struct inode_operations shmem_dir_inode_operations;
268 static const struct inode_operations shmem_special_inode_operations;
269 static const struct vm_operations_struct shmem_vm_ops;
270 static const struct vm_operations_struct shmem_anon_vm_ops;
271 static struct file_system_type shmem_fs_type;
272 
273 bool shmem_mapping(struct address_space *mapping)
274 {
275 	return mapping->a_ops == &shmem_aops;
276 }
277 EXPORT_SYMBOL_GPL(shmem_mapping);
278 
279 bool vma_is_anon_shmem(struct vm_area_struct *vma)
280 {
281 	return vma->vm_ops == &shmem_anon_vm_ops;
282 }
283 
284 bool vma_is_shmem(struct vm_area_struct *vma)
285 {
286 	return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops;
287 }
288 
289 static LIST_HEAD(shmem_swaplist);
290 static DEFINE_MUTEX(shmem_swaplist_mutex);
291 
292 #ifdef CONFIG_TMPFS_QUOTA
293 
294 static int shmem_enable_quotas(struct super_block *sb,
295 			       unsigned short quota_types)
296 {
297 	int type, err = 0;
298 
299 	sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY;
300 	for (type = 0; type < SHMEM_MAXQUOTAS; type++) {
301 		if (!(quota_types & (1 << type)))
302 			continue;
303 		err = dquot_load_quota_sb(sb, type, QFMT_SHMEM,
304 					  DQUOT_USAGE_ENABLED |
305 					  DQUOT_LIMITS_ENABLED);
306 		if (err)
307 			goto out_err;
308 	}
309 	return 0;
310 
311 out_err:
312 	pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n",
313 		type, err);
314 	for (type--; type >= 0; type--)
315 		dquot_quota_off(sb, type);
316 	return err;
317 }
318 
319 static void shmem_disable_quotas(struct super_block *sb)
320 {
321 	int type;
322 
323 	for (type = 0; type < SHMEM_MAXQUOTAS; type++)
324 		dquot_quota_off(sb, type);
325 }
326 
327 static struct dquot __rcu **shmem_get_dquots(struct inode *inode)
328 {
329 	return SHMEM_I(inode)->i_dquot;
330 }
331 #endif /* CONFIG_TMPFS_QUOTA */
332 
333 /*
334  * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
335  * produces a novel ino for the newly allocated inode.
336  *
337  * It may also be called when making a hard link to permit the space needed by
338  * each dentry. However, in that case, no new inode number is needed since that
339  * internally draws from another pool of inode numbers (currently global
340  * get_next_ino()). This case is indicated by passing NULL as inop.
341  */
342 #define SHMEM_INO_BATCH 1024
343 static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
344 {
345 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
346 	ino_t ino;
347 
348 	if (!(sb->s_flags & SB_KERNMOUNT)) {
349 		raw_spin_lock(&sbinfo->stat_lock);
350 		if (sbinfo->max_inodes) {
351 			if (sbinfo->free_ispace < BOGO_INODE_SIZE) {
352 				raw_spin_unlock(&sbinfo->stat_lock);
353 				return -ENOSPC;
354 			}
355 			sbinfo->free_ispace -= BOGO_INODE_SIZE;
356 		}
357 		if (inop) {
358 			ino = sbinfo->next_ino++;
359 			if (unlikely(is_zero_ino(ino)))
360 				ino = sbinfo->next_ino++;
361 			if (unlikely(!sbinfo->full_inums &&
362 				     ino > UINT_MAX)) {
363 				/*
364 				 * Emulate get_next_ino uint wraparound for
365 				 * compatibility
366 				 */
367 				if (IS_ENABLED(CONFIG_64BIT))
368 					pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
369 						__func__, MINOR(sb->s_dev));
370 				sbinfo->next_ino = 1;
371 				ino = sbinfo->next_ino++;
372 			}
373 			*inop = ino;
374 		}
375 		raw_spin_unlock(&sbinfo->stat_lock);
376 	} else if (inop) {
377 		/*
378 		 * __shmem_file_setup, one of our callers, is lock-free: it
379 		 * doesn't hold stat_lock in shmem_reserve_inode since
380 		 * max_inodes is always 0, and is called from potentially
381 		 * unknown contexts. As such, use a per-cpu batched allocator
382 		 * which doesn't require the per-sb stat_lock unless we are at
383 		 * the batch boundary.
384 		 *
385 		 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
386 		 * shmem mounts are not exposed to userspace, so we don't need
387 		 * to worry about things like glibc compatibility.
388 		 */
389 		ino_t *next_ino;
390 
391 		next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
392 		ino = *next_ino;
393 		if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
394 			raw_spin_lock(&sbinfo->stat_lock);
395 			ino = sbinfo->next_ino;
396 			sbinfo->next_ino += SHMEM_INO_BATCH;
397 			raw_spin_unlock(&sbinfo->stat_lock);
398 			if (unlikely(is_zero_ino(ino)))
399 				ino++;
400 		}
401 		*inop = ino;
402 		*next_ino = ++ino;
403 		put_cpu();
404 	}
405 
406 	return 0;
407 }
408 
409 static void shmem_free_inode(struct super_block *sb, size_t freed_ispace)
410 {
411 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
412 	if (sbinfo->max_inodes) {
413 		raw_spin_lock(&sbinfo->stat_lock);
414 		sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace;
415 		raw_spin_unlock(&sbinfo->stat_lock);
416 	}
417 }
418 
419 /**
420  * shmem_recalc_inode - recalculate the block usage of an inode
421  * @inode: inode to recalc
422  * @alloced: the change in number of pages allocated to inode
423  * @swapped: the change in number of pages swapped from inode
424  *
425  * We have to calculate the free blocks since the mm can drop
426  * undirtied hole pages behind our back.
427  *
428  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
429  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
430  */
431 static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped)
432 {
433 	struct shmem_inode_info *info = SHMEM_I(inode);
434 	long freed;
435 
436 	spin_lock(&info->lock);
437 	info->alloced += alloced;
438 	info->swapped += swapped;
439 	freed = info->alloced - info->swapped -
440 		READ_ONCE(inode->i_mapping->nrpages);
441 	/*
442 	 * Special case: whereas normally shmem_recalc_inode() is called
443 	 * after i_mapping->nrpages has already been adjusted (up or down),
444 	 * shmem_writepage() has to raise swapped before nrpages is lowered -
445 	 * to stop a racing shmem_recalc_inode() from thinking that a page has
446 	 * been freed.  Compensate here, to avoid the need for a followup call.
447 	 */
448 	if (swapped > 0)
449 		freed += swapped;
450 	if (freed > 0)
451 		info->alloced -= freed;
452 	spin_unlock(&info->lock);
453 
454 	/* The quota case may block */
455 	if (freed > 0)
456 		shmem_inode_unacct_blocks(inode, freed);
457 }
458 
459 bool shmem_charge(struct inode *inode, long pages)
460 {
461 	struct address_space *mapping = inode->i_mapping;
462 
463 	if (shmem_inode_acct_blocks(inode, pages))
464 		return false;
465 
466 	/* nrpages adjustment first, then shmem_recalc_inode() when balanced */
467 	xa_lock_irq(&mapping->i_pages);
468 	mapping->nrpages += pages;
469 	xa_unlock_irq(&mapping->i_pages);
470 
471 	shmem_recalc_inode(inode, pages, 0);
472 	return true;
473 }
474 
475 void shmem_uncharge(struct inode *inode, long pages)
476 {
477 	/* pages argument is currently unused: keep it to help debugging */
478 	/* nrpages adjustment done by __filemap_remove_folio() or caller */
479 
480 	shmem_recalc_inode(inode, 0, 0);
481 }
482 
483 /*
484  * Replace item expected in xarray by a new item, while holding xa_lock.
485  */
486 static int shmem_replace_entry(struct address_space *mapping,
487 			pgoff_t index, void *expected, void *replacement)
488 {
489 	XA_STATE(xas, &mapping->i_pages, index);
490 	void *item;
491 
492 	VM_BUG_ON(!expected);
493 	VM_BUG_ON(!replacement);
494 	item = xas_load(&xas);
495 	if (item != expected)
496 		return -ENOENT;
497 	xas_store(&xas, replacement);
498 	return 0;
499 }
500 
501 /*
502  * Sometimes, before we decide whether to proceed or to fail, we must check
503  * that an entry was not already brought back from swap by a racing thread.
504  *
505  * Checking folio is not enough: by the time a swapcache folio is locked, it
506  * might be reused, and again be swapcache, using the same swap as before.
507  */
508 static bool shmem_confirm_swap(struct address_space *mapping,
509 			       pgoff_t index, swp_entry_t swap)
510 {
511 	return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
512 }
513 
514 /*
515  * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
516  *
517  * SHMEM_HUGE_NEVER:
518  *	disables huge pages for the mount;
519  * SHMEM_HUGE_ALWAYS:
520  *	enables huge pages for the mount;
521  * SHMEM_HUGE_WITHIN_SIZE:
522  *	only allocate huge pages if the page will be fully within i_size,
523  *	also respect fadvise()/madvise() hints;
524  * SHMEM_HUGE_ADVISE:
525  *	only allocate huge pages if requested with fadvise()/madvise();
526  */
527 
528 #define SHMEM_HUGE_NEVER	0
529 #define SHMEM_HUGE_ALWAYS	1
530 #define SHMEM_HUGE_WITHIN_SIZE	2
531 #define SHMEM_HUGE_ADVISE	3
532 
533 /*
534  * Special values.
535  * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
536  *
537  * SHMEM_HUGE_DENY:
538  *	disables huge on shm_mnt and all mounts, for emergency use;
539  * SHMEM_HUGE_FORCE:
540  *	enables huge on shm_mnt and all mounts, w/o needing option, for testing;
541  *
542  */
543 #define SHMEM_HUGE_DENY		(-1)
544 #define SHMEM_HUGE_FORCE	(-2)
545 
546 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
547 /* ifdef here to avoid bloating shmem.o when not necessary */
548 
549 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
550 
551 static bool __shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
552 					loff_t write_end, bool shmem_huge_force,
553 					struct vm_area_struct *vma,
554 					unsigned long vm_flags)
555 {
556 	struct mm_struct *mm = vma ? vma->vm_mm : NULL;
557 	loff_t i_size;
558 
559 	if (!S_ISREG(inode->i_mode))
560 		return false;
561 	if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags)))
562 		return false;
563 	if (shmem_huge == SHMEM_HUGE_DENY)
564 		return false;
565 	if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE)
566 		return true;
567 
568 	switch (SHMEM_SB(inode->i_sb)->huge) {
569 	case SHMEM_HUGE_ALWAYS:
570 		return true;
571 	case SHMEM_HUGE_WITHIN_SIZE:
572 		index = round_up(index + 1, HPAGE_PMD_NR);
573 		i_size = max(write_end, i_size_read(inode));
574 		i_size = round_up(i_size, PAGE_SIZE);
575 		if (i_size >> PAGE_SHIFT >= index)
576 			return true;
577 		fallthrough;
578 	case SHMEM_HUGE_ADVISE:
579 		if (mm && (vm_flags & VM_HUGEPAGE))
580 			return true;
581 		fallthrough;
582 	default:
583 		return false;
584 	}
585 }
586 
587 static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
588 		   loff_t write_end, bool shmem_huge_force,
589 		   struct vm_area_struct *vma, unsigned long vm_flags)
590 {
591 	if (HPAGE_PMD_ORDER > MAX_PAGECACHE_ORDER)
592 		return false;
593 
594 	return __shmem_huge_global_enabled(inode, index, write_end,
595 					   shmem_huge_force, vma, vm_flags);
596 }
597 
598 #if defined(CONFIG_SYSFS)
599 static int shmem_parse_huge(const char *str)
600 {
601 	if (!strcmp(str, "never"))
602 		return SHMEM_HUGE_NEVER;
603 	if (!strcmp(str, "always"))
604 		return SHMEM_HUGE_ALWAYS;
605 	if (!strcmp(str, "within_size"))
606 		return SHMEM_HUGE_WITHIN_SIZE;
607 	if (!strcmp(str, "advise"))
608 		return SHMEM_HUGE_ADVISE;
609 	if (!strcmp(str, "deny"))
610 		return SHMEM_HUGE_DENY;
611 	if (!strcmp(str, "force"))
612 		return SHMEM_HUGE_FORCE;
613 	return -EINVAL;
614 }
615 #endif
616 
617 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
618 static const char *shmem_format_huge(int huge)
619 {
620 	switch (huge) {
621 	case SHMEM_HUGE_NEVER:
622 		return "never";
623 	case SHMEM_HUGE_ALWAYS:
624 		return "always";
625 	case SHMEM_HUGE_WITHIN_SIZE:
626 		return "within_size";
627 	case SHMEM_HUGE_ADVISE:
628 		return "advise";
629 	case SHMEM_HUGE_DENY:
630 		return "deny";
631 	case SHMEM_HUGE_FORCE:
632 		return "force";
633 	default:
634 		VM_BUG_ON(1);
635 		return "bad_val";
636 	}
637 }
638 #endif
639 
640 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
641 		struct shrink_control *sc, unsigned long nr_to_free)
642 {
643 	LIST_HEAD(list), *pos, *next;
644 	struct inode *inode;
645 	struct shmem_inode_info *info;
646 	struct folio *folio;
647 	unsigned long batch = sc ? sc->nr_to_scan : 128;
648 	unsigned long split = 0, freed = 0;
649 
650 	if (list_empty(&sbinfo->shrinklist))
651 		return SHRINK_STOP;
652 
653 	spin_lock(&sbinfo->shrinklist_lock);
654 	list_for_each_safe(pos, next, &sbinfo->shrinklist) {
655 		info = list_entry(pos, struct shmem_inode_info, shrinklist);
656 
657 		/* pin the inode */
658 		inode = igrab(&info->vfs_inode);
659 
660 		/* inode is about to be evicted */
661 		if (!inode) {
662 			list_del_init(&info->shrinklist);
663 			goto next;
664 		}
665 
666 		list_move(&info->shrinklist, &list);
667 next:
668 		sbinfo->shrinklist_len--;
669 		if (!--batch)
670 			break;
671 	}
672 	spin_unlock(&sbinfo->shrinklist_lock);
673 
674 	list_for_each_safe(pos, next, &list) {
675 		pgoff_t next, end;
676 		loff_t i_size;
677 		int ret;
678 
679 		info = list_entry(pos, struct shmem_inode_info, shrinklist);
680 		inode = &info->vfs_inode;
681 
682 		if (nr_to_free && freed >= nr_to_free)
683 			goto move_back;
684 
685 		i_size = i_size_read(inode);
686 		folio = filemap_get_entry(inode->i_mapping, i_size / PAGE_SIZE);
687 		if (!folio || xa_is_value(folio))
688 			goto drop;
689 
690 		/* No large folio at the end of the file: nothing to split */
691 		if (!folio_test_large(folio)) {
692 			folio_put(folio);
693 			goto drop;
694 		}
695 
696 		/* Check if there is anything to gain from splitting */
697 		next = folio_next_index(folio);
698 		end = shmem_fallocend(inode, DIV_ROUND_UP(i_size, PAGE_SIZE));
699 		if (end <= folio->index || end >= next) {
700 			folio_put(folio);
701 			goto drop;
702 		}
703 
704 		/*
705 		 * Move the inode on the list back to shrinklist if we failed
706 		 * to lock the page at this time.
707 		 *
708 		 * Waiting for the lock may lead to deadlock in the
709 		 * reclaim path.
710 		 */
711 		if (!folio_trylock(folio)) {
712 			folio_put(folio);
713 			goto move_back;
714 		}
715 
716 		ret = split_folio(folio);
717 		folio_unlock(folio);
718 		folio_put(folio);
719 
720 		/* If split failed move the inode on the list back to shrinklist */
721 		if (ret)
722 			goto move_back;
723 
724 		freed += next - end;
725 		split++;
726 drop:
727 		list_del_init(&info->shrinklist);
728 		goto put;
729 move_back:
730 		/*
731 		 * Make sure the inode is either on the global list or deleted
732 		 * from any local list before iput() since it could be deleted
733 		 * in another thread once we put the inode (then the local list
734 		 * is corrupted).
735 		 */
736 		spin_lock(&sbinfo->shrinklist_lock);
737 		list_move(&info->shrinklist, &sbinfo->shrinklist);
738 		sbinfo->shrinklist_len++;
739 		spin_unlock(&sbinfo->shrinklist_lock);
740 put:
741 		iput(inode);
742 	}
743 
744 	return split;
745 }
746 
747 static long shmem_unused_huge_scan(struct super_block *sb,
748 		struct shrink_control *sc)
749 {
750 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
751 
752 	if (!READ_ONCE(sbinfo->shrinklist_len))
753 		return SHRINK_STOP;
754 
755 	return shmem_unused_huge_shrink(sbinfo, sc, 0);
756 }
757 
758 static long shmem_unused_huge_count(struct super_block *sb,
759 		struct shrink_control *sc)
760 {
761 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
762 	return READ_ONCE(sbinfo->shrinklist_len);
763 }
764 #else /* !CONFIG_TRANSPARENT_HUGEPAGE */
765 
766 #define shmem_huge SHMEM_HUGE_DENY
767 
768 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
769 		struct shrink_control *sc, unsigned long nr_to_free)
770 {
771 	return 0;
772 }
773 
774 static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
775 		loff_t write_end, bool shmem_huge_force,
776 		struct vm_area_struct *vma, unsigned long vm_flags)
777 {
778 	return false;
779 }
780 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
781 
782 /*
783  * Somewhat like filemap_add_folio, but error if expected item has gone.
784  */
785 static int shmem_add_to_page_cache(struct folio *folio,
786 				   struct address_space *mapping,
787 				   pgoff_t index, void *expected, gfp_t gfp)
788 {
789 	XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio));
790 	long nr = folio_nr_pages(folio);
791 
792 	VM_BUG_ON_FOLIO(index != round_down(index, nr), folio);
793 	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
794 	VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
795 
796 	folio_ref_add(folio, nr);
797 	folio->mapping = mapping;
798 	folio->index = index;
799 
800 	gfp &= GFP_RECLAIM_MASK;
801 	folio_throttle_swaprate(folio, gfp);
802 
803 	do {
804 		xas_lock_irq(&xas);
805 		if (expected != xas_find_conflict(&xas)) {
806 			xas_set_err(&xas, -EEXIST);
807 			goto unlock;
808 		}
809 		if (expected && xas_find_conflict(&xas)) {
810 			xas_set_err(&xas, -EEXIST);
811 			goto unlock;
812 		}
813 		xas_store(&xas, folio);
814 		if (xas_error(&xas))
815 			goto unlock;
816 		if (folio_test_pmd_mappable(folio))
817 			__lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr);
818 		__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
819 		__lruvec_stat_mod_folio(folio, NR_SHMEM, nr);
820 		mapping->nrpages += nr;
821 unlock:
822 		xas_unlock_irq(&xas);
823 	} while (xas_nomem(&xas, gfp));
824 
825 	if (xas_error(&xas)) {
826 		folio->mapping = NULL;
827 		folio_ref_sub(folio, nr);
828 		return xas_error(&xas);
829 	}
830 
831 	return 0;
832 }
833 
834 /*
835  * Somewhat like filemap_remove_folio, but substitutes swap for @folio.
836  */
837 static void shmem_delete_from_page_cache(struct folio *folio, void *radswap)
838 {
839 	struct address_space *mapping = folio->mapping;
840 	long nr = folio_nr_pages(folio);
841 	int error;
842 
843 	xa_lock_irq(&mapping->i_pages);
844 	error = shmem_replace_entry(mapping, folio->index, folio, radswap);
845 	folio->mapping = NULL;
846 	mapping->nrpages -= nr;
847 	__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
848 	__lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
849 	xa_unlock_irq(&mapping->i_pages);
850 	folio_put_refs(folio, nr);
851 	BUG_ON(error);
852 }
853 
854 /*
855  * Remove swap entry from page cache, free the swap and its page cache. Returns
856  * the number of pages being freed. 0 means entry not found in XArray (0 pages
857  * being freed).
858  */
859 static long shmem_free_swap(struct address_space *mapping,
860 			    pgoff_t index, void *radswap)
861 {
862 	int order = xa_get_order(&mapping->i_pages, index);
863 	void *old;
864 
865 	old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
866 	if (old != radswap)
867 		return 0;
868 	free_swap_and_cache_nr(radix_to_swp_entry(radswap), 1 << order);
869 
870 	return 1 << order;
871 }
872 
873 /*
874  * Determine (in bytes) how many of the shmem object's pages mapped by the
875  * given offsets are swapped out.
876  *
877  * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
878  * as long as the inode doesn't go away and racy results are not a problem.
879  */
880 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
881 						pgoff_t start, pgoff_t end)
882 {
883 	XA_STATE(xas, &mapping->i_pages, start);
884 	struct page *page;
885 	unsigned long swapped = 0;
886 	unsigned long max = end - 1;
887 
888 	rcu_read_lock();
889 	xas_for_each(&xas, page, max) {
890 		if (xas_retry(&xas, page))
891 			continue;
892 		if (xa_is_value(page))
893 			swapped += 1 << xas_get_order(&xas);
894 		if (xas.xa_index == max)
895 			break;
896 		if (need_resched()) {
897 			xas_pause(&xas);
898 			cond_resched_rcu();
899 		}
900 	}
901 	rcu_read_unlock();
902 
903 	return swapped << PAGE_SHIFT;
904 }
905 
906 /*
907  * Determine (in bytes) how many of the shmem object's pages mapped by the
908  * given vma is swapped out.
909  *
910  * This is safe to call without i_rwsem or the i_pages lock thanks to RCU,
911  * as long as the inode doesn't go away and racy results are not a problem.
912  */
913 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
914 {
915 	struct inode *inode = file_inode(vma->vm_file);
916 	struct shmem_inode_info *info = SHMEM_I(inode);
917 	struct address_space *mapping = inode->i_mapping;
918 	unsigned long swapped;
919 
920 	/* Be careful as we don't hold info->lock */
921 	swapped = READ_ONCE(info->swapped);
922 
923 	/*
924 	 * The easier cases are when the shmem object has nothing in swap, or
925 	 * the vma maps it whole. Then we can simply use the stats that we
926 	 * already track.
927 	 */
928 	if (!swapped)
929 		return 0;
930 
931 	if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
932 		return swapped << PAGE_SHIFT;
933 
934 	/* Here comes the more involved part */
935 	return shmem_partial_swap_usage(mapping, vma->vm_pgoff,
936 					vma->vm_pgoff + vma_pages(vma));
937 }
938 
939 /*
940  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
941  */
942 void shmem_unlock_mapping(struct address_space *mapping)
943 {
944 	struct folio_batch fbatch;
945 	pgoff_t index = 0;
946 
947 	folio_batch_init(&fbatch);
948 	/*
949 	 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
950 	 */
951 	while (!mapping_unevictable(mapping) &&
952 	       filemap_get_folios(mapping, &index, ~0UL, &fbatch)) {
953 		check_move_unevictable_folios(&fbatch);
954 		folio_batch_release(&fbatch);
955 		cond_resched();
956 	}
957 }
958 
959 static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
960 {
961 	struct folio *folio;
962 
963 	/*
964 	 * At first avoid shmem_get_folio(,,,SGP_READ): that fails
965 	 * beyond i_size, and reports fallocated folios as holes.
966 	 */
967 	folio = filemap_get_entry(inode->i_mapping, index);
968 	if (!folio)
969 		return folio;
970 	if (!xa_is_value(folio)) {
971 		folio_lock(folio);
972 		if (folio->mapping == inode->i_mapping)
973 			return folio;
974 		/* The folio has been swapped out */
975 		folio_unlock(folio);
976 		folio_put(folio);
977 	}
978 	/*
979 	 * But read a folio back from swap if any of it is within i_size
980 	 * (although in some cases this is just a waste of time).
981 	 */
982 	folio = NULL;
983 	shmem_get_folio(inode, index, 0, &folio, SGP_READ);
984 	return folio;
985 }
986 
987 /*
988  * Remove range of pages and swap entries from page cache, and free them.
989  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
990  */
991 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
992 								 bool unfalloc)
993 {
994 	struct address_space *mapping = inode->i_mapping;
995 	struct shmem_inode_info *info = SHMEM_I(inode);
996 	pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
997 	pgoff_t end = (lend + 1) >> PAGE_SHIFT;
998 	struct folio_batch fbatch;
999 	pgoff_t indices[PAGEVEC_SIZE];
1000 	struct folio *folio;
1001 	bool same_folio;
1002 	long nr_swaps_freed = 0;
1003 	pgoff_t index;
1004 	int i;
1005 
1006 	if (lend == -1)
1007 		end = -1;	/* unsigned, so actually very big */
1008 
1009 	if (info->fallocend > start && info->fallocend <= end && !unfalloc)
1010 		info->fallocend = start;
1011 
1012 	folio_batch_init(&fbatch);
1013 	index = start;
1014 	while (index < end && find_lock_entries(mapping, &index, end - 1,
1015 			&fbatch, indices)) {
1016 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
1017 			folio = fbatch.folios[i];
1018 
1019 			if (xa_is_value(folio)) {
1020 				if (unfalloc)
1021 					continue;
1022 				nr_swaps_freed += shmem_free_swap(mapping,
1023 							indices[i], folio);
1024 				continue;
1025 			}
1026 
1027 			if (!unfalloc || !folio_test_uptodate(folio))
1028 				truncate_inode_folio(mapping, folio);
1029 			folio_unlock(folio);
1030 		}
1031 		folio_batch_remove_exceptionals(&fbatch);
1032 		folio_batch_release(&fbatch);
1033 		cond_resched();
1034 	}
1035 
1036 	/*
1037 	 * When undoing a failed fallocate, we want none of the partial folio
1038 	 * zeroing and splitting below, but shall want to truncate the whole
1039 	 * folio when !uptodate indicates that it was added by this fallocate,
1040 	 * even when [lstart, lend] covers only a part of the folio.
1041 	 */
1042 	if (unfalloc)
1043 		goto whole_folios;
1044 
1045 	same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
1046 	folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT);
1047 	if (folio) {
1048 		same_folio = lend < folio_pos(folio) + folio_size(folio);
1049 		folio_mark_dirty(folio);
1050 		if (!truncate_inode_partial_folio(folio, lstart, lend)) {
1051 			start = folio_next_index(folio);
1052 			if (same_folio)
1053 				end = folio->index;
1054 		}
1055 		folio_unlock(folio);
1056 		folio_put(folio);
1057 		folio = NULL;
1058 	}
1059 
1060 	if (!same_folio)
1061 		folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT);
1062 	if (folio) {
1063 		folio_mark_dirty(folio);
1064 		if (!truncate_inode_partial_folio(folio, lstart, lend))
1065 			end = folio->index;
1066 		folio_unlock(folio);
1067 		folio_put(folio);
1068 	}
1069 
1070 whole_folios:
1071 
1072 	index = start;
1073 	while (index < end) {
1074 		cond_resched();
1075 
1076 		if (!find_get_entries(mapping, &index, end - 1, &fbatch,
1077 				indices)) {
1078 			/* If all gone or hole-punch or unfalloc, we're done */
1079 			if (index == start || end != -1)
1080 				break;
1081 			/* But if truncating, restart to make sure all gone */
1082 			index = start;
1083 			continue;
1084 		}
1085 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
1086 			folio = fbatch.folios[i];
1087 
1088 			if (xa_is_value(folio)) {
1089 				long swaps_freed;
1090 
1091 				if (unfalloc)
1092 					continue;
1093 				swaps_freed = shmem_free_swap(mapping, indices[i], folio);
1094 				if (!swaps_freed) {
1095 					/* Swap was replaced by page: retry */
1096 					index = indices[i];
1097 					break;
1098 				}
1099 				nr_swaps_freed += swaps_freed;
1100 				continue;
1101 			}
1102 
1103 			folio_lock(folio);
1104 
1105 			if (!unfalloc || !folio_test_uptodate(folio)) {
1106 				if (folio_mapping(folio) != mapping) {
1107 					/* Page was replaced by swap: retry */
1108 					folio_unlock(folio);
1109 					index = indices[i];
1110 					break;
1111 				}
1112 				VM_BUG_ON_FOLIO(folio_test_writeback(folio),
1113 						folio);
1114 
1115 				if (!folio_test_large(folio)) {
1116 					truncate_inode_folio(mapping, folio);
1117 				} else if (truncate_inode_partial_folio(folio, lstart, lend)) {
1118 					/*
1119 					 * If we split a page, reset the loop so
1120 					 * that we pick up the new sub pages.
1121 					 * Otherwise the THP was entirely
1122 					 * dropped or the target range was
1123 					 * zeroed, so just continue the loop as
1124 					 * is.
1125 					 */
1126 					if (!folio_test_large(folio)) {
1127 						folio_unlock(folio);
1128 						index = start;
1129 						break;
1130 					}
1131 				}
1132 			}
1133 			folio_unlock(folio);
1134 		}
1135 		folio_batch_remove_exceptionals(&fbatch);
1136 		folio_batch_release(&fbatch);
1137 	}
1138 
1139 	shmem_recalc_inode(inode, 0, -nr_swaps_freed);
1140 }
1141 
1142 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1143 {
1144 	shmem_undo_range(inode, lstart, lend, false);
1145 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1146 	inode_inc_iversion(inode);
1147 }
1148 EXPORT_SYMBOL_GPL(shmem_truncate_range);
1149 
1150 static int shmem_getattr(struct mnt_idmap *idmap,
1151 			 const struct path *path, struct kstat *stat,
1152 			 u32 request_mask, unsigned int query_flags)
1153 {
1154 	struct inode *inode = path->dentry->d_inode;
1155 	struct shmem_inode_info *info = SHMEM_I(inode);
1156 
1157 	if (info->alloced - info->swapped != inode->i_mapping->nrpages)
1158 		shmem_recalc_inode(inode, 0, 0);
1159 
1160 	if (info->fsflags & FS_APPEND_FL)
1161 		stat->attributes |= STATX_ATTR_APPEND;
1162 	if (info->fsflags & FS_IMMUTABLE_FL)
1163 		stat->attributes |= STATX_ATTR_IMMUTABLE;
1164 	if (info->fsflags & FS_NODUMP_FL)
1165 		stat->attributes |= STATX_ATTR_NODUMP;
1166 	stat->attributes_mask |= (STATX_ATTR_APPEND |
1167 			STATX_ATTR_IMMUTABLE |
1168 			STATX_ATTR_NODUMP);
1169 	inode_lock_shared(inode);
1170 	generic_fillattr(idmap, request_mask, inode, stat);
1171 	inode_unlock_shared(inode);
1172 
1173 	if (shmem_huge_global_enabled(inode, 0, 0, false, NULL, 0))
1174 		stat->blksize = HPAGE_PMD_SIZE;
1175 
1176 	if (request_mask & STATX_BTIME) {
1177 		stat->result_mask |= STATX_BTIME;
1178 		stat->btime.tv_sec = info->i_crtime.tv_sec;
1179 		stat->btime.tv_nsec = info->i_crtime.tv_nsec;
1180 	}
1181 
1182 	return 0;
1183 }
1184 
1185 static int shmem_setattr(struct mnt_idmap *idmap,
1186 			 struct dentry *dentry, struct iattr *attr)
1187 {
1188 	struct inode *inode = d_inode(dentry);
1189 	struct shmem_inode_info *info = SHMEM_I(inode);
1190 	int error;
1191 	bool update_mtime = false;
1192 	bool update_ctime = true;
1193 
1194 	error = setattr_prepare(idmap, dentry, attr);
1195 	if (error)
1196 		return error;
1197 
1198 	if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) {
1199 		if ((inode->i_mode ^ attr->ia_mode) & 0111) {
1200 			return -EPERM;
1201 		}
1202 	}
1203 
1204 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1205 		loff_t oldsize = inode->i_size;
1206 		loff_t newsize = attr->ia_size;
1207 
1208 		/* protected by i_rwsem */
1209 		if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1210 		    (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1211 			return -EPERM;
1212 
1213 		if (newsize != oldsize) {
1214 			error = shmem_reacct_size(SHMEM_I(inode)->flags,
1215 					oldsize, newsize);
1216 			if (error)
1217 				return error;
1218 			i_size_write(inode, newsize);
1219 			update_mtime = true;
1220 		} else {
1221 			update_ctime = false;
1222 		}
1223 		if (newsize <= oldsize) {
1224 			loff_t holebegin = round_up(newsize, PAGE_SIZE);
1225 			if (oldsize > holebegin)
1226 				unmap_mapping_range(inode->i_mapping,
1227 							holebegin, 0, 1);
1228 			if (info->alloced)
1229 				shmem_truncate_range(inode,
1230 							newsize, (loff_t)-1);
1231 			/* unmap again to remove racily COWed private pages */
1232 			if (oldsize > holebegin)
1233 				unmap_mapping_range(inode->i_mapping,
1234 							holebegin, 0, 1);
1235 		}
1236 	}
1237 
1238 	if (is_quota_modification(idmap, inode, attr)) {
1239 		error = dquot_initialize(inode);
1240 		if (error)
1241 			return error;
1242 	}
1243 
1244 	/* Transfer quota accounting */
1245 	if (i_uid_needs_update(idmap, attr, inode) ||
1246 	    i_gid_needs_update(idmap, attr, inode)) {
1247 		error = dquot_transfer(idmap, inode, attr);
1248 		if (error)
1249 			return error;
1250 	}
1251 
1252 	setattr_copy(idmap, inode, attr);
1253 	if (attr->ia_valid & ATTR_MODE)
1254 		error = posix_acl_chmod(idmap, dentry, inode->i_mode);
1255 	if (!error && update_ctime) {
1256 		inode_set_ctime_current(inode);
1257 		if (update_mtime)
1258 			inode_set_mtime_to_ts(inode, inode_get_ctime(inode));
1259 		inode_inc_iversion(inode);
1260 	}
1261 	return error;
1262 }
1263 
1264 static void shmem_evict_inode(struct inode *inode)
1265 {
1266 	struct shmem_inode_info *info = SHMEM_I(inode);
1267 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1268 	size_t freed = 0;
1269 
1270 	if (shmem_mapping(inode->i_mapping)) {
1271 		shmem_unacct_size(info->flags, inode->i_size);
1272 		inode->i_size = 0;
1273 		mapping_set_exiting(inode->i_mapping);
1274 		shmem_truncate_range(inode, 0, (loff_t)-1);
1275 		if (!list_empty(&info->shrinklist)) {
1276 			spin_lock(&sbinfo->shrinklist_lock);
1277 			if (!list_empty(&info->shrinklist)) {
1278 				list_del_init(&info->shrinklist);
1279 				sbinfo->shrinklist_len--;
1280 			}
1281 			spin_unlock(&sbinfo->shrinklist_lock);
1282 		}
1283 		while (!list_empty(&info->swaplist)) {
1284 			/* Wait while shmem_unuse() is scanning this inode... */
1285 			wait_var_event(&info->stop_eviction,
1286 				       !atomic_read(&info->stop_eviction));
1287 			mutex_lock(&shmem_swaplist_mutex);
1288 			/* ...but beware of the race if we peeked too early */
1289 			if (!atomic_read(&info->stop_eviction))
1290 				list_del_init(&info->swaplist);
1291 			mutex_unlock(&shmem_swaplist_mutex);
1292 		}
1293 	}
1294 
1295 	simple_xattrs_free(&info->xattrs, sbinfo->max_inodes ? &freed : NULL);
1296 	shmem_free_inode(inode->i_sb, freed);
1297 	WARN_ON(inode->i_blocks);
1298 	clear_inode(inode);
1299 #ifdef CONFIG_TMPFS_QUOTA
1300 	dquot_free_inode(inode);
1301 	dquot_drop(inode);
1302 #endif
1303 }
1304 
1305 static int shmem_find_swap_entries(struct address_space *mapping,
1306 				   pgoff_t start, struct folio_batch *fbatch,
1307 				   pgoff_t *indices, unsigned int type)
1308 {
1309 	XA_STATE(xas, &mapping->i_pages, start);
1310 	struct folio *folio;
1311 	swp_entry_t entry;
1312 
1313 	rcu_read_lock();
1314 	xas_for_each(&xas, folio, ULONG_MAX) {
1315 		if (xas_retry(&xas, folio))
1316 			continue;
1317 
1318 		if (!xa_is_value(folio))
1319 			continue;
1320 
1321 		entry = radix_to_swp_entry(folio);
1322 		/*
1323 		 * swapin error entries can be found in the mapping. But they're
1324 		 * deliberately ignored here as we've done everything we can do.
1325 		 */
1326 		if (swp_type(entry) != type)
1327 			continue;
1328 
1329 		indices[folio_batch_count(fbatch)] = xas.xa_index;
1330 		if (!folio_batch_add(fbatch, folio))
1331 			break;
1332 
1333 		if (need_resched()) {
1334 			xas_pause(&xas);
1335 			cond_resched_rcu();
1336 		}
1337 	}
1338 	rcu_read_unlock();
1339 
1340 	return xas.xa_index;
1341 }
1342 
1343 /*
1344  * Move the swapped pages for an inode to page cache. Returns the count
1345  * of pages swapped in, or the error in case of failure.
1346  */
1347 static int shmem_unuse_swap_entries(struct inode *inode,
1348 		struct folio_batch *fbatch, pgoff_t *indices)
1349 {
1350 	int i = 0;
1351 	int ret = 0;
1352 	int error = 0;
1353 	struct address_space *mapping = inode->i_mapping;
1354 
1355 	for (i = 0; i < folio_batch_count(fbatch); i++) {
1356 		struct folio *folio = fbatch->folios[i];
1357 
1358 		if (!xa_is_value(folio))
1359 			continue;
1360 		error = shmem_swapin_folio(inode, indices[i], &folio, SGP_CACHE,
1361 					mapping_gfp_mask(mapping), NULL, NULL);
1362 		if (error == 0) {
1363 			folio_unlock(folio);
1364 			folio_put(folio);
1365 			ret++;
1366 		}
1367 		if (error == -ENOMEM)
1368 			break;
1369 		error = 0;
1370 	}
1371 	return error ? error : ret;
1372 }
1373 
1374 /*
1375  * If swap found in inode, free it and move page from swapcache to filecache.
1376  */
1377 static int shmem_unuse_inode(struct inode *inode, unsigned int type)
1378 {
1379 	struct address_space *mapping = inode->i_mapping;
1380 	pgoff_t start = 0;
1381 	struct folio_batch fbatch;
1382 	pgoff_t indices[PAGEVEC_SIZE];
1383 	int ret = 0;
1384 
1385 	do {
1386 		folio_batch_init(&fbatch);
1387 		shmem_find_swap_entries(mapping, start, &fbatch, indices, type);
1388 		if (folio_batch_count(&fbatch) == 0) {
1389 			ret = 0;
1390 			break;
1391 		}
1392 
1393 		ret = shmem_unuse_swap_entries(inode, &fbatch, indices);
1394 		if (ret < 0)
1395 			break;
1396 
1397 		start = indices[folio_batch_count(&fbatch) - 1];
1398 	} while (true);
1399 
1400 	return ret;
1401 }
1402 
1403 /*
1404  * Read all the shared memory data that resides in the swap
1405  * device 'type' back into memory, so the swap device can be
1406  * unused.
1407  */
1408 int shmem_unuse(unsigned int type)
1409 {
1410 	struct shmem_inode_info *info, *next;
1411 	int error = 0;
1412 
1413 	if (list_empty(&shmem_swaplist))
1414 		return 0;
1415 
1416 	mutex_lock(&shmem_swaplist_mutex);
1417 	list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1418 		if (!info->swapped) {
1419 			list_del_init(&info->swaplist);
1420 			continue;
1421 		}
1422 		/*
1423 		 * Drop the swaplist mutex while searching the inode for swap;
1424 		 * but before doing so, make sure shmem_evict_inode() will not
1425 		 * remove placeholder inode from swaplist, nor let it be freed
1426 		 * (igrab() would protect from unlink, but not from unmount).
1427 		 */
1428 		atomic_inc(&info->stop_eviction);
1429 		mutex_unlock(&shmem_swaplist_mutex);
1430 
1431 		error = shmem_unuse_inode(&info->vfs_inode, type);
1432 		cond_resched();
1433 
1434 		mutex_lock(&shmem_swaplist_mutex);
1435 		next = list_next_entry(info, swaplist);
1436 		if (!info->swapped)
1437 			list_del_init(&info->swaplist);
1438 		if (atomic_dec_and_test(&info->stop_eviction))
1439 			wake_up_var(&info->stop_eviction);
1440 		if (error)
1441 			break;
1442 	}
1443 	mutex_unlock(&shmem_swaplist_mutex);
1444 
1445 	return error;
1446 }
1447 
1448 /*
1449  * Move the page from the page cache to the swap cache.
1450  */
1451 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1452 {
1453 	struct folio *folio = page_folio(page);
1454 	struct address_space *mapping = folio->mapping;
1455 	struct inode *inode = mapping->host;
1456 	struct shmem_inode_info *info = SHMEM_I(inode);
1457 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1458 	swp_entry_t swap;
1459 	pgoff_t index;
1460 	int nr_pages;
1461 	bool split = false;
1462 
1463 	/*
1464 	 * Our capabilities prevent regular writeback or sync from ever calling
1465 	 * shmem_writepage; but a stacking filesystem might use ->writepage of
1466 	 * its underlying filesystem, in which case tmpfs should write out to
1467 	 * swap only in response to memory pressure, and not for the writeback
1468 	 * threads or sync.
1469 	 */
1470 	if (WARN_ON_ONCE(!wbc->for_reclaim))
1471 		goto redirty;
1472 
1473 	if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap))
1474 		goto redirty;
1475 
1476 	if (!total_swap_pages)
1477 		goto redirty;
1478 
1479 	/*
1480 	 * If CONFIG_THP_SWAP is not enabled, the large folio should be
1481 	 * split when swapping.
1482 	 *
1483 	 * And shrinkage of pages beyond i_size does not split swap, so
1484 	 * swapout of a large folio crossing i_size needs to split too
1485 	 * (unless fallocate has been used to preallocate beyond EOF).
1486 	 */
1487 	if (folio_test_large(folio)) {
1488 		index = shmem_fallocend(inode,
1489 			DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE));
1490 		if ((index > folio->index && index < folio_next_index(folio)) ||
1491 		    !IS_ENABLED(CONFIG_THP_SWAP))
1492 			split = true;
1493 	}
1494 
1495 	if (split) {
1496 try_split:
1497 		/* Ensure the subpages are still dirty */
1498 		folio_test_set_dirty(folio);
1499 		if (split_huge_page_to_list_to_order(page, wbc->list, 0))
1500 			goto redirty;
1501 		folio = page_folio(page);
1502 		folio_clear_dirty(folio);
1503 	}
1504 
1505 	index = folio->index;
1506 	nr_pages = folio_nr_pages(folio);
1507 
1508 	/*
1509 	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1510 	 * value into swapfile.c, the only way we can correctly account for a
1511 	 * fallocated folio arriving here is now to initialize it and write it.
1512 	 *
1513 	 * That's okay for a folio already fallocated earlier, but if we have
1514 	 * not yet completed the fallocation, then (a) we want to keep track
1515 	 * of this folio in case we have to undo it, and (b) it may not be a
1516 	 * good idea to continue anyway, once we're pushing into swap.  So
1517 	 * reactivate the folio, and let shmem_fallocate() quit when too many.
1518 	 */
1519 	if (!folio_test_uptodate(folio)) {
1520 		if (inode->i_private) {
1521 			struct shmem_falloc *shmem_falloc;
1522 			spin_lock(&inode->i_lock);
1523 			shmem_falloc = inode->i_private;
1524 			if (shmem_falloc &&
1525 			    !shmem_falloc->waitq &&
1526 			    index >= shmem_falloc->start &&
1527 			    index < shmem_falloc->next)
1528 				shmem_falloc->nr_unswapped++;
1529 			else
1530 				shmem_falloc = NULL;
1531 			spin_unlock(&inode->i_lock);
1532 			if (shmem_falloc)
1533 				goto redirty;
1534 		}
1535 		folio_zero_range(folio, 0, folio_size(folio));
1536 		flush_dcache_folio(folio);
1537 		folio_mark_uptodate(folio);
1538 	}
1539 
1540 	swap = folio_alloc_swap(folio);
1541 	if (!swap.val) {
1542 		if (nr_pages > 1)
1543 			goto try_split;
1544 
1545 		goto redirty;
1546 	}
1547 
1548 	/*
1549 	 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1550 	 * if it's not already there.  Do it now before the folio is
1551 	 * moved to swap cache, when its pagelock no longer protects
1552 	 * the inode from eviction.  But don't unlock the mutex until
1553 	 * we've incremented swapped, because shmem_unuse_inode() will
1554 	 * prune a !swapped inode from the swaplist under this mutex.
1555 	 */
1556 	mutex_lock(&shmem_swaplist_mutex);
1557 	if (list_empty(&info->swaplist))
1558 		list_add(&info->swaplist, &shmem_swaplist);
1559 
1560 	if (add_to_swap_cache(folio, swap,
1561 			__GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1562 			NULL) == 0) {
1563 		shmem_recalc_inode(inode, 0, nr_pages);
1564 		swap_shmem_alloc(swap, nr_pages);
1565 		shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap));
1566 
1567 		mutex_unlock(&shmem_swaplist_mutex);
1568 		BUG_ON(folio_mapped(folio));
1569 		return swap_writepage(&folio->page, wbc);
1570 	}
1571 
1572 	mutex_unlock(&shmem_swaplist_mutex);
1573 	put_swap_folio(folio, swap);
1574 redirty:
1575 	folio_mark_dirty(folio);
1576 	if (wbc->for_reclaim)
1577 		return AOP_WRITEPAGE_ACTIVATE;	/* Return with folio locked */
1578 	folio_unlock(folio);
1579 	return 0;
1580 }
1581 
1582 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1583 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1584 {
1585 	char buffer[64];
1586 
1587 	if (!mpol || mpol->mode == MPOL_DEFAULT)
1588 		return;		/* show nothing */
1589 
1590 	mpol_to_str(buffer, sizeof(buffer), mpol);
1591 
1592 	seq_printf(seq, ",mpol=%s", buffer);
1593 }
1594 
1595 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1596 {
1597 	struct mempolicy *mpol = NULL;
1598 	if (sbinfo->mpol) {
1599 		raw_spin_lock(&sbinfo->stat_lock);	/* prevent replace/use races */
1600 		mpol = sbinfo->mpol;
1601 		mpol_get(mpol);
1602 		raw_spin_unlock(&sbinfo->stat_lock);
1603 	}
1604 	return mpol;
1605 }
1606 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1607 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1608 {
1609 }
1610 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1611 {
1612 	return NULL;
1613 }
1614 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1615 
1616 static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
1617 			pgoff_t index, unsigned int order, pgoff_t *ilx);
1618 
1619 static struct folio *shmem_swapin_cluster(swp_entry_t swap, gfp_t gfp,
1620 			struct shmem_inode_info *info, pgoff_t index)
1621 {
1622 	struct mempolicy *mpol;
1623 	pgoff_t ilx;
1624 	struct folio *folio;
1625 
1626 	mpol = shmem_get_pgoff_policy(info, index, 0, &ilx);
1627 	folio = swap_cluster_readahead(swap, gfp, mpol, ilx);
1628 	mpol_cond_put(mpol);
1629 
1630 	return folio;
1631 }
1632 
1633 /*
1634  * Make sure huge_gfp is always more limited than limit_gfp.
1635  * Some of the flags set permissions, while others set limitations.
1636  */
1637 static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
1638 {
1639 	gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM;
1640 	gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY;
1641 	gfp_t zoneflags = limit_gfp & GFP_ZONEMASK;
1642 	gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK);
1643 
1644 	/* Allow allocations only from the originally specified zones. */
1645 	result |= zoneflags;
1646 
1647 	/*
1648 	 * Minimize the result gfp by taking the union with the deny flags,
1649 	 * and the intersection of the allow flags.
1650 	 */
1651 	result |= (limit_gfp & denyflags);
1652 	result |= (huge_gfp & limit_gfp) & allowflags;
1653 
1654 	return result;
1655 }
1656 
1657 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1658 unsigned long shmem_allowable_huge_orders(struct inode *inode,
1659 				struct vm_area_struct *vma, pgoff_t index,
1660 				loff_t write_end, bool shmem_huge_force)
1661 {
1662 	unsigned long mask = READ_ONCE(huge_shmem_orders_always);
1663 	unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
1664 	unsigned long vm_flags = vma ? vma->vm_flags : 0;
1665 	bool global_huge;
1666 	loff_t i_size;
1667 	int order;
1668 
1669 	if (thp_disabled_by_hw() || (vma && vma_thp_disabled(vma, vm_flags)))
1670 		return 0;
1671 
1672 	global_huge = shmem_huge_global_enabled(inode, index, write_end,
1673 					shmem_huge_force, vma, vm_flags);
1674 	if (!vma || !vma_is_anon_shmem(vma)) {
1675 		/*
1676 		 * For tmpfs, we now only support PMD sized THP if huge page
1677 		 * is enabled, otherwise fallback to order 0.
1678 		 */
1679 		return global_huge ? BIT(HPAGE_PMD_ORDER) : 0;
1680 	}
1681 
1682 	/*
1683 	 * Following the 'deny' semantics of the top level, force the huge
1684 	 * option off from all mounts.
1685 	 */
1686 	if (shmem_huge == SHMEM_HUGE_DENY)
1687 		return 0;
1688 
1689 	/*
1690 	 * Only allow inherit orders if the top-level value is 'force', which
1691 	 * means non-PMD sized THP can not override 'huge' mount option now.
1692 	 */
1693 	if (shmem_huge == SHMEM_HUGE_FORCE)
1694 		return READ_ONCE(huge_shmem_orders_inherit);
1695 
1696 	/* Allow mTHP that will be fully within i_size. */
1697 	order = highest_order(within_size_orders);
1698 	while (within_size_orders) {
1699 		index = round_up(index + 1, order);
1700 		i_size = round_up(i_size_read(inode), PAGE_SIZE);
1701 		if (i_size >> PAGE_SHIFT >= index) {
1702 			mask |= within_size_orders;
1703 			break;
1704 		}
1705 
1706 		order = next_order(&within_size_orders, order);
1707 	}
1708 
1709 	if (vm_flags & VM_HUGEPAGE)
1710 		mask |= READ_ONCE(huge_shmem_orders_madvise);
1711 
1712 	if (global_huge)
1713 		mask |= READ_ONCE(huge_shmem_orders_inherit);
1714 
1715 	return THP_ORDERS_ALL_FILE_DEFAULT & mask;
1716 }
1717 
1718 static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
1719 					   struct address_space *mapping, pgoff_t index,
1720 					   unsigned long orders)
1721 {
1722 	struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
1723 	pgoff_t aligned_index;
1724 	unsigned long pages;
1725 	int order;
1726 
1727 	if (vma) {
1728 		orders = thp_vma_suitable_orders(vma, vmf->address, orders);
1729 		if (!orders)
1730 			return 0;
1731 	}
1732 
1733 	/* Find the highest order that can add into the page cache */
1734 	order = highest_order(orders);
1735 	while (orders) {
1736 		pages = 1UL << order;
1737 		aligned_index = round_down(index, pages);
1738 		/*
1739 		 * Check for conflict before waiting on a huge allocation.
1740 		 * Conflict might be that a huge page has just been allocated
1741 		 * and added to page cache by a racing thread, or that there
1742 		 * is already at least one small page in the huge extent.
1743 		 * Be careful to retry when appropriate, but not forever!
1744 		 * Elsewhere -EEXIST would be the right code, but not here.
1745 		 */
1746 		if (!xa_find(&mapping->i_pages, &aligned_index,
1747 			     aligned_index + pages - 1, XA_PRESENT))
1748 			break;
1749 		order = next_order(&orders, order);
1750 	}
1751 
1752 	return orders;
1753 }
1754 #else
1755 static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault *vmf,
1756 					   struct address_space *mapping, pgoff_t index,
1757 					   unsigned long orders)
1758 {
1759 	return 0;
1760 }
1761 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1762 
1763 static struct folio *shmem_alloc_folio(gfp_t gfp, int order,
1764 		struct shmem_inode_info *info, pgoff_t index)
1765 {
1766 	struct mempolicy *mpol;
1767 	pgoff_t ilx;
1768 	struct folio *folio;
1769 
1770 	mpol = shmem_get_pgoff_policy(info, index, order, &ilx);
1771 	folio = folio_alloc_mpol(gfp, order, mpol, ilx, numa_node_id());
1772 	mpol_cond_put(mpol);
1773 
1774 	return folio;
1775 }
1776 
1777 static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
1778 		gfp_t gfp, struct inode *inode, pgoff_t index,
1779 		struct mm_struct *fault_mm, unsigned long orders)
1780 {
1781 	struct address_space *mapping = inode->i_mapping;
1782 	struct shmem_inode_info *info = SHMEM_I(inode);
1783 	unsigned long suitable_orders = 0;
1784 	struct folio *folio = NULL;
1785 	long pages;
1786 	int error, order;
1787 
1788 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1789 		orders = 0;
1790 
1791 	if (orders > 0) {
1792 		suitable_orders = shmem_suitable_orders(inode, vmf,
1793 							mapping, index, orders);
1794 
1795 		order = highest_order(suitable_orders);
1796 		while (suitable_orders) {
1797 			pages = 1UL << order;
1798 			index = round_down(index, pages);
1799 			folio = shmem_alloc_folio(gfp, order, info, index);
1800 			if (folio)
1801 				goto allocated;
1802 
1803 			if (pages == HPAGE_PMD_NR)
1804 				count_vm_event(THP_FILE_FALLBACK);
1805 			count_mthp_stat(order, MTHP_STAT_SHMEM_FALLBACK);
1806 			order = next_order(&suitable_orders, order);
1807 		}
1808 	} else {
1809 		pages = 1;
1810 		folio = shmem_alloc_folio(gfp, 0, info, index);
1811 	}
1812 	if (!folio)
1813 		return ERR_PTR(-ENOMEM);
1814 
1815 allocated:
1816 	__folio_set_locked(folio);
1817 	__folio_set_swapbacked(folio);
1818 
1819 	gfp &= GFP_RECLAIM_MASK;
1820 	error = mem_cgroup_charge(folio, fault_mm, gfp);
1821 	if (error) {
1822 		if (xa_find(&mapping->i_pages, &index,
1823 				index + pages - 1, XA_PRESENT)) {
1824 			error = -EEXIST;
1825 		} else if (pages > 1) {
1826 			if (pages == HPAGE_PMD_NR) {
1827 				count_vm_event(THP_FILE_FALLBACK);
1828 				count_vm_event(THP_FILE_FALLBACK_CHARGE);
1829 			}
1830 			count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK);
1831 			count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK_CHARGE);
1832 		}
1833 		goto unlock;
1834 	}
1835 
1836 	error = shmem_add_to_page_cache(folio, mapping, index, NULL, gfp);
1837 	if (error)
1838 		goto unlock;
1839 
1840 	error = shmem_inode_acct_blocks(inode, pages);
1841 	if (error) {
1842 		struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1843 		long freed;
1844 		/*
1845 		 * Try to reclaim some space by splitting a few
1846 		 * large folios beyond i_size on the filesystem.
1847 		 */
1848 		shmem_unused_huge_shrink(sbinfo, NULL, pages);
1849 		/*
1850 		 * And do a shmem_recalc_inode() to account for freed pages:
1851 		 * except our folio is there in cache, so not quite balanced.
1852 		 */
1853 		spin_lock(&info->lock);
1854 		freed = pages + info->alloced - info->swapped -
1855 			READ_ONCE(mapping->nrpages);
1856 		if (freed > 0)
1857 			info->alloced -= freed;
1858 		spin_unlock(&info->lock);
1859 		if (freed > 0)
1860 			shmem_inode_unacct_blocks(inode, freed);
1861 		error = shmem_inode_acct_blocks(inode, pages);
1862 		if (error) {
1863 			filemap_remove_folio(folio);
1864 			goto unlock;
1865 		}
1866 	}
1867 
1868 	shmem_recalc_inode(inode, pages, 0);
1869 	folio_add_lru(folio);
1870 	return folio;
1871 
1872 unlock:
1873 	folio_unlock(folio);
1874 	folio_put(folio);
1875 	return ERR_PTR(error);
1876 }
1877 
1878 /*
1879  * When a page is moved from swapcache to shmem filecache (either by the
1880  * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of
1881  * shmem_unuse_inode()), it may have been read in earlier from swap, in
1882  * ignorance of the mapping it belongs to.  If that mapping has special
1883  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1884  * we may need to copy to a suitable page before moving to filecache.
1885  *
1886  * In a future release, this may well be extended to respect cpuset and
1887  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1888  * but for now it is a simple matter of zone.
1889  */
1890 static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
1891 {
1892 	return folio_zonenum(folio) > gfp_zone(gfp);
1893 }
1894 
1895 static int shmem_replace_folio(struct folio **foliop, gfp_t gfp,
1896 				struct shmem_inode_info *info, pgoff_t index,
1897 				struct vm_area_struct *vma)
1898 {
1899 	struct folio *new, *old = *foliop;
1900 	swp_entry_t entry = old->swap;
1901 	struct address_space *swap_mapping = swap_address_space(entry);
1902 	pgoff_t swap_index = swap_cache_index(entry);
1903 	XA_STATE(xas, &swap_mapping->i_pages, swap_index);
1904 	int nr_pages = folio_nr_pages(old);
1905 	int error = 0, i;
1906 
1907 	/*
1908 	 * We have arrived here because our zones are constrained, so don't
1909 	 * limit chance of success by further cpuset and node constraints.
1910 	 */
1911 	gfp &= ~GFP_CONSTRAINT_MASK;
1912 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1913 	if (nr_pages > 1) {
1914 		gfp_t huge_gfp = vma_thp_gfp_mask(vma);
1915 
1916 		gfp = limit_gfp_mask(huge_gfp, gfp);
1917 	}
1918 #endif
1919 
1920 	new = shmem_alloc_folio(gfp, folio_order(old), info, index);
1921 	if (!new)
1922 		return -ENOMEM;
1923 
1924 	folio_ref_add(new, nr_pages);
1925 	folio_copy(new, old);
1926 	flush_dcache_folio(new);
1927 
1928 	__folio_set_locked(new);
1929 	__folio_set_swapbacked(new);
1930 	folio_mark_uptodate(new);
1931 	new->swap = entry;
1932 	folio_set_swapcache(new);
1933 
1934 	/* Swap cache still stores N entries instead of a high-order entry */
1935 	xa_lock_irq(&swap_mapping->i_pages);
1936 	for (i = 0; i < nr_pages; i++) {
1937 		void *item = xas_load(&xas);
1938 
1939 		if (item != old) {
1940 			error = -ENOENT;
1941 			break;
1942 		}
1943 
1944 		xas_store(&xas, new);
1945 		xas_next(&xas);
1946 	}
1947 	if (!error) {
1948 		mem_cgroup_replace_folio(old, new);
1949 		__lruvec_stat_mod_folio(new, NR_FILE_PAGES, nr_pages);
1950 		__lruvec_stat_mod_folio(new, NR_SHMEM, nr_pages);
1951 		__lruvec_stat_mod_folio(old, NR_FILE_PAGES, -nr_pages);
1952 		__lruvec_stat_mod_folio(old, NR_SHMEM, -nr_pages);
1953 	}
1954 	xa_unlock_irq(&swap_mapping->i_pages);
1955 
1956 	if (unlikely(error)) {
1957 		/*
1958 		 * Is this possible?  I think not, now that our callers
1959 		 * check both the swapcache flag and folio->private
1960 		 * after getting the folio lock; but be defensive.
1961 		 * Reverse old to newpage for clear and free.
1962 		 */
1963 		old = new;
1964 	} else {
1965 		folio_add_lru(new);
1966 		*foliop = new;
1967 	}
1968 
1969 	folio_clear_swapcache(old);
1970 	old->private = NULL;
1971 
1972 	folio_unlock(old);
1973 	/*
1974 	 * The old folio are removed from swap cache, drop the 'nr_pages'
1975 	 * reference, as well as one temporary reference getting from swap
1976 	 * cache.
1977 	 */
1978 	folio_put_refs(old, nr_pages + 1);
1979 	return error;
1980 }
1981 
1982 static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index,
1983 					 struct folio *folio, swp_entry_t swap)
1984 {
1985 	struct address_space *mapping = inode->i_mapping;
1986 	swp_entry_t swapin_error;
1987 	void *old;
1988 	int nr_pages;
1989 
1990 	swapin_error = make_poisoned_swp_entry();
1991 	old = xa_cmpxchg_irq(&mapping->i_pages, index,
1992 			     swp_to_radix_entry(swap),
1993 			     swp_to_radix_entry(swapin_error), 0);
1994 	if (old != swp_to_radix_entry(swap))
1995 		return;
1996 
1997 	nr_pages = folio_nr_pages(folio);
1998 	folio_wait_writeback(folio);
1999 	delete_from_swap_cache(folio);
2000 	/*
2001 	 * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks
2002 	 * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks)
2003 	 * in shmem_evict_inode().
2004 	 */
2005 	shmem_recalc_inode(inode, -nr_pages, -nr_pages);
2006 	swap_free_nr(swap, nr_pages);
2007 }
2008 
2009 static int shmem_split_large_entry(struct inode *inode, pgoff_t index,
2010 				   swp_entry_t swap, gfp_t gfp)
2011 {
2012 	struct address_space *mapping = inode->i_mapping;
2013 	XA_STATE_ORDER(xas, &mapping->i_pages, index, 0);
2014 	void *alloced_shadow = NULL;
2015 	int alloced_order = 0, i;
2016 
2017 	/* Convert user data gfp flags to xarray node gfp flags */
2018 	gfp &= GFP_RECLAIM_MASK;
2019 
2020 	for (;;) {
2021 		int order = -1, split_order = 0;
2022 		void *old = NULL;
2023 
2024 		xas_lock_irq(&xas);
2025 		old = xas_load(&xas);
2026 		if (!xa_is_value(old) || swp_to_radix_entry(swap) != old) {
2027 			xas_set_err(&xas, -EEXIST);
2028 			goto unlock;
2029 		}
2030 
2031 		order = xas_get_order(&xas);
2032 
2033 		/* Swap entry may have changed before we re-acquire the lock */
2034 		if (alloced_order &&
2035 		    (old != alloced_shadow || order != alloced_order)) {
2036 			xas_destroy(&xas);
2037 			alloced_order = 0;
2038 		}
2039 
2040 		/* Try to split large swap entry in pagecache */
2041 		if (order > 0) {
2042 			if (!alloced_order) {
2043 				split_order = order;
2044 				goto unlock;
2045 			}
2046 			xas_split(&xas, old, order);
2047 
2048 			/*
2049 			 * Re-set the swap entry after splitting, and the swap
2050 			 * offset of the original large entry must be continuous.
2051 			 */
2052 			for (i = 0; i < 1 << order; i++) {
2053 				pgoff_t aligned_index = round_down(index, 1 << order);
2054 				swp_entry_t tmp;
2055 
2056 				tmp = swp_entry(swp_type(swap), swp_offset(swap) + i);
2057 				__xa_store(&mapping->i_pages, aligned_index + i,
2058 					   swp_to_radix_entry(tmp), 0);
2059 			}
2060 		}
2061 
2062 unlock:
2063 		xas_unlock_irq(&xas);
2064 
2065 		/* split needed, alloc here and retry. */
2066 		if (split_order) {
2067 			xas_split_alloc(&xas, old, split_order, gfp);
2068 			if (xas_error(&xas))
2069 				goto error;
2070 			alloced_shadow = old;
2071 			alloced_order = split_order;
2072 			xas_reset(&xas);
2073 			continue;
2074 		}
2075 
2076 		if (!xas_nomem(&xas, gfp))
2077 			break;
2078 	}
2079 
2080 error:
2081 	if (xas_error(&xas))
2082 		return xas_error(&xas);
2083 
2084 	return alloced_order;
2085 }
2086 
2087 /*
2088  * Swap in the folio pointed to by *foliop.
2089  * Caller has to make sure that *foliop contains a valid swapped folio.
2090  * Returns 0 and the folio in foliop if success. On failure, returns the
2091  * error code and NULL in *foliop.
2092  */
2093 static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
2094 			     struct folio **foliop, enum sgp_type sgp,
2095 			     gfp_t gfp, struct vm_area_struct *vma,
2096 			     vm_fault_t *fault_type)
2097 {
2098 	struct address_space *mapping = inode->i_mapping;
2099 	struct mm_struct *fault_mm = vma ? vma->vm_mm : NULL;
2100 	struct shmem_inode_info *info = SHMEM_I(inode);
2101 	struct swap_info_struct *si;
2102 	struct folio *folio = NULL;
2103 	swp_entry_t swap;
2104 	int error, nr_pages;
2105 
2106 	VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
2107 	swap = radix_to_swp_entry(*foliop);
2108 	*foliop = NULL;
2109 
2110 	if (is_poisoned_swp_entry(swap))
2111 		return -EIO;
2112 
2113 	si = get_swap_device(swap);
2114 	if (!si) {
2115 		if (!shmem_confirm_swap(mapping, index, swap))
2116 			return -EEXIST;
2117 		else
2118 			return -EINVAL;
2119 	}
2120 
2121 	/* Look it up and read it in.. */
2122 	folio = swap_cache_get_folio(swap, NULL, 0);
2123 	if (!folio) {
2124 		int split_order;
2125 
2126 		/* Or update major stats only when swapin succeeds?? */
2127 		if (fault_type) {
2128 			*fault_type |= VM_FAULT_MAJOR;
2129 			count_vm_event(PGMAJFAULT);
2130 			count_memcg_event_mm(fault_mm, PGMAJFAULT);
2131 		}
2132 
2133 		/*
2134 		 * Now swap device can only swap in order 0 folio, then we
2135 		 * should split the large swap entry stored in the pagecache
2136 		 * if necessary.
2137 		 */
2138 		split_order = shmem_split_large_entry(inode, index, swap, gfp);
2139 		if (split_order < 0) {
2140 			error = split_order;
2141 			goto failed;
2142 		}
2143 
2144 		/*
2145 		 * If the large swap entry has already been split, it is
2146 		 * necessary to recalculate the new swap entry based on
2147 		 * the old order alignment.
2148 		 */
2149 		if (split_order > 0) {
2150 			pgoff_t offset = index - round_down(index, 1 << split_order);
2151 
2152 			swap = swp_entry(swp_type(swap), swp_offset(swap) + offset);
2153 		}
2154 
2155 		/* Here we actually start the io */
2156 		folio = shmem_swapin_cluster(swap, gfp, info, index);
2157 		if (!folio) {
2158 			error = -ENOMEM;
2159 			goto failed;
2160 		}
2161 	}
2162 
2163 	/* We have to do this with folio locked to prevent races */
2164 	folio_lock(folio);
2165 	if (!folio_test_swapcache(folio) ||
2166 	    folio->swap.val != swap.val ||
2167 	    !shmem_confirm_swap(mapping, index, swap)) {
2168 		error = -EEXIST;
2169 		goto unlock;
2170 	}
2171 	if (!folio_test_uptodate(folio)) {
2172 		error = -EIO;
2173 		goto failed;
2174 	}
2175 	folio_wait_writeback(folio);
2176 	nr_pages = folio_nr_pages(folio);
2177 
2178 	/*
2179 	 * Some architectures may have to restore extra metadata to the
2180 	 * folio after reading from swap.
2181 	 */
2182 	arch_swap_restore(folio_swap(swap, folio), folio);
2183 
2184 	if (shmem_should_replace_folio(folio, gfp)) {
2185 		error = shmem_replace_folio(&folio, gfp, info, index, vma);
2186 		if (error)
2187 			goto failed;
2188 	}
2189 
2190 	error = shmem_add_to_page_cache(folio, mapping,
2191 					round_down(index, nr_pages),
2192 					swp_to_radix_entry(swap), gfp);
2193 	if (error)
2194 		goto failed;
2195 
2196 	shmem_recalc_inode(inode, 0, -nr_pages);
2197 
2198 	if (sgp == SGP_WRITE)
2199 		folio_mark_accessed(folio);
2200 
2201 	delete_from_swap_cache(folio);
2202 	folio_mark_dirty(folio);
2203 	swap_free_nr(swap, nr_pages);
2204 	put_swap_device(si);
2205 
2206 	*foliop = folio;
2207 	return 0;
2208 failed:
2209 	if (!shmem_confirm_swap(mapping, index, swap))
2210 		error = -EEXIST;
2211 	if (error == -EIO)
2212 		shmem_set_folio_swapin_error(inode, index, folio, swap);
2213 unlock:
2214 	if (folio) {
2215 		folio_unlock(folio);
2216 		folio_put(folio);
2217 	}
2218 	put_swap_device(si);
2219 
2220 	return error;
2221 }
2222 
2223 /*
2224  * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate
2225  *
2226  * If we allocate a new one we do not mark it dirty. That's up to the
2227  * vm. If we swap it in we mark it dirty since we also free the swap
2228  * entry since a page cannot live in both the swap and page cache.
2229  *
2230  * vmf and fault_type are only supplied by shmem_fault: otherwise they are NULL.
2231  */
2232 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
2233 		loff_t write_end, struct folio **foliop, enum sgp_type sgp,
2234 		gfp_t gfp, struct vm_fault *vmf, vm_fault_t *fault_type)
2235 {
2236 	struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
2237 	struct mm_struct *fault_mm;
2238 	struct folio *folio;
2239 	int error;
2240 	bool alloced;
2241 	unsigned long orders = 0;
2242 
2243 	if (WARN_ON_ONCE(!shmem_mapping(inode->i_mapping)))
2244 		return -EINVAL;
2245 
2246 	if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
2247 		return -EFBIG;
2248 repeat:
2249 	if (sgp <= SGP_CACHE &&
2250 	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode))
2251 		return -EINVAL;
2252 
2253 	alloced = false;
2254 	fault_mm = vma ? vma->vm_mm : NULL;
2255 
2256 	folio = filemap_get_entry(inode->i_mapping, index);
2257 	if (folio && vma && userfaultfd_minor(vma)) {
2258 		if (!xa_is_value(folio))
2259 			folio_put(folio);
2260 		*fault_type = handle_userfault(vmf, VM_UFFD_MINOR);
2261 		return 0;
2262 	}
2263 
2264 	if (xa_is_value(folio)) {
2265 		error = shmem_swapin_folio(inode, index, &folio,
2266 					   sgp, gfp, vma, fault_type);
2267 		if (error == -EEXIST)
2268 			goto repeat;
2269 
2270 		*foliop = folio;
2271 		return error;
2272 	}
2273 
2274 	if (folio) {
2275 		folio_lock(folio);
2276 
2277 		/* Has the folio been truncated or swapped out? */
2278 		if (unlikely(folio->mapping != inode->i_mapping)) {
2279 			folio_unlock(folio);
2280 			folio_put(folio);
2281 			goto repeat;
2282 		}
2283 		if (sgp == SGP_WRITE)
2284 			folio_mark_accessed(folio);
2285 		if (folio_test_uptodate(folio))
2286 			goto out;
2287 		/* fallocated folio */
2288 		if (sgp != SGP_READ)
2289 			goto clear;
2290 		folio_unlock(folio);
2291 		folio_put(folio);
2292 	}
2293 
2294 	/*
2295 	 * SGP_READ: succeed on hole, with NULL folio, letting caller zero.
2296 	 * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail.
2297 	 */
2298 	*foliop = NULL;
2299 	if (sgp == SGP_READ)
2300 		return 0;
2301 	if (sgp == SGP_NOALLOC)
2302 		return -ENOENT;
2303 
2304 	/*
2305 	 * Fast cache lookup and swap lookup did not find it: allocate.
2306 	 */
2307 
2308 	if (vma && userfaultfd_missing(vma)) {
2309 		*fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
2310 		return 0;
2311 	}
2312 
2313 	/* Find hugepage orders that are allowed for anonymous shmem and tmpfs. */
2314 	orders = shmem_allowable_huge_orders(inode, vma, index, write_end, false);
2315 	if (orders > 0) {
2316 		gfp_t huge_gfp;
2317 
2318 		huge_gfp = vma_thp_gfp_mask(vma);
2319 		huge_gfp = limit_gfp_mask(huge_gfp, gfp);
2320 		folio = shmem_alloc_and_add_folio(vmf, huge_gfp,
2321 				inode, index, fault_mm, orders);
2322 		if (!IS_ERR(folio)) {
2323 			if (folio_test_pmd_mappable(folio))
2324 				count_vm_event(THP_FILE_ALLOC);
2325 			count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_ALLOC);
2326 			goto alloced;
2327 		}
2328 		if (PTR_ERR(folio) == -EEXIST)
2329 			goto repeat;
2330 	}
2331 
2332 	folio = shmem_alloc_and_add_folio(vmf, gfp, inode, index, fault_mm, 0);
2333 	if (IS_ERR(folio)) {
2334 		error = PTR_ERR(folio);
2335 		if (error == -EEXIST)
2336 			goto repeat;
2337 		folio = NULL;
2338 		goto unlock;
2339 	}
2340 
2341 alloced:
2342 	alloced = true;
2343 	if (folio_test_large(folio) &&
2344 	    DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
2345 					folio_next_index(folio)) {
2346 		struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2347 		struct shmem_inode_info *info = SHMEM_I(inode);
2348 		/*
2349 		 * Part of the large folio is beyond i_size: subject
2350 		 * to shrink under memory pressure.
2351 		 */
2352 		spin_lock(&sbinfo->shrinklist_lock);
2353 		/*
2354 		 * _careful to defend against unlocked access to
2355 		 * ->shrink_list in shmem_unused_huge_shrink()
2356 		 */
2357 		if (list_empty_careful(&info->shrinklist)) {
2358 			list_add_tail(&info->shrinklist,
2359 				      &sbinfo->shrinklist);
2360 			sbinfo->shrinklist_len++;
2361 		}
2362 		spin_unlock(&sbinfo->shrinklist_lock);
2363 	}
2364 
2365 	if (sgp == SGP_WRITE)
2366 		folio_set_referenced(folio);
2367 	/*
2368 	 * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio.
2369 	 */
2370 	if (sgp == SGP_FALLOC)
2371 		sgp = SGP_WRITE;
2372 clear:
2373 	/*
2374 	 * Let SGP_WRITE caller clear ends if write does not fill folio;
2375 	 * but SGP_FALLOC on a folio fallocated earlier must initialize
2376 	 * it now, lest undo on failure cancel our earlier guarantee.
2377 	 */
2378 	if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) {
2379 		long i, n = folio_nr_pages(folio);
2380 
2381 		for (i = 0; i < n; i++)
2382 			clear_highpage(folio_page(folio, i));
2383 		flush_dcache_folio(folio);
2384 		folio_mark_uptodate(folio);
2385 	}
2386 
2387 	/* Perhaps the file has been truncated since we checked */
2388 	if (sgp <= SGP_CACHE &&
2389 	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
2390 		error = -EINVAL;
2391 		goto unlock;
2392 	}
2393 out:
2394 	*foliop = folio;
2395 	return 0;
2396 
2397 	/*
2398 	 * Error recovery.
2399 	 */
2400 unlock:
2401 	if (alloced)
2402 		filemap_remove_folio(folio);
2403 	shmem_recalc_inode(inode, 0, 0);
2404 	if (folio) {
2405 		folio_unlock(folio);
2406 		folio_put(folio);
2407 	}
2408 	return error;
2409 }
2410 
2411 /**
2412  * shmem_get_folio - find, and lock a shmem folio.
2413  * @inode:	inode to search
2414  * @index:	the page index.
2415  * @write_end:	end of a write, could extend inode size
2416  * @foliop:	pointer to the folio if found
2417  * @sgp:	SGP_* flags to control behavior
2418  *
2419  * Looks up the page cache entry at @inode & @index.  If a folio is
2420  * present, it is returned locked with an increased refcount.
2421  *
2422  * If the caller modifies data in the folio, it must call folio_mark_dirty()
2423  * before unlocking the folio to ensure that the folio is not reclaimed.
2424  * There is no need to reserve space before calling folio_mark_dirty().
2425  *
2426  * When no folio is found, the behavior depends on @sgp:
2427  *  - for SGP_READ, *@foliop is %NULL and 0 is returned
2428  *  - for SGP_NOALLOC, *@foliop is %NULL and -ENOENT is returned
2429  *  - for all other flags a new folio is allocated, inserted into the
2430  *    page cache and returned locked in @foliop.
2431  *
2432  * Context: May sleep.
2433  * Return: 0 if successful, else a negative error code.
2434  */
2435 int shmem_get_folio(struct inode *inode, pgoff_t index, loff_t write_end,
2436 		    struct folio **foliop, enum sgp_type sgp)
2437 {
2438 	return shmem_get_folio_gfp(inode, index, write_end, foliop, sgp,
2439 			mapping_gfp_mask(inode->i_mapping), NULL, NULL);
2440 }
2441 EXPORT_SYMBOL_GPL(shmem_get_folio);
2442 
2443 /*
2444  * This is like autoremove_wake_function, but it removes the wait queue
2445  * entry unconditionally - even if something else had already woken the
2446  * target.
2447  */
2448 static int synchronous_wake_function(wait_queue_entry_t *wait,
2449 			unsigned int mode, int sync, void *key)
2450 {
2451 	int ret = default_wake_function(wait, mode, sync, key);
2452 	list_del_init(&wait->entry);
2453 	return ret;
2454 }
2455 
2456 /*
2457  * Trinity finds that probing a hole which tmpfs is punching can
2458  * prevent the hole-punch from ever completing: which in turn
2459  * locks writers out with its hold on i_rwsem.  So refrain from
2460  * faulting pages into the hole while it's being punched.  Although
2461  * shmem_undo_range() does remove the additions, it may be unable to
2462  * keep up, as each new page needs its own unmap_mapping_range() call,
2463  * and the i_mmap tree grows ever slower to scan if new vmas are added.
2464  *
2465  * It does not matter if we sometimes reach this check just before the
2466  * hole-punch begins, so that one fault then races with the punch:
2467  * we just need to make racing faults a rare case.
2468  *
2469  * The implementation below would be much simpler if we just used a
2470  * standard mutex or completion: but we cannot take i_rwsem in fault,
2471  * and bloating every shmem inode for this unlikely case would be sad.
2472  */
2473 static vm_fault_t shmem_falloc_wait(struct vm_fault *vmf, struct inode *inode)
2474 {
2475 	struct shmem_falloc *shmem_falloc;
2476 	struct file *fpin = NULL;
2477 	vm_fault_t ret = 0;
2478 
2479 	spin_lock(&inode->i_lock);
2480 	shmem_falloc = inode->i_private;
2481 	if (shmem_falloc &&
2482 	    shmem_falloc->waitq &&
2483 	    vmf->pgoff >= shmem_falloc->start &&
2484 	    vmf->pgoff < shmem_falloc->next) {
2485 		wait_queue_head_t *shmem_falloc_waitq;
2486 		DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2487 
2488 		ret = VM_FAULT_NOPAGE;
2489 		fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2490 		shmem_falloc_waitq = shmem_falloc->waitq;
2491 		prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2492 				TASK_UNINTERRUPTIBLE);
2493 		spin_unlock(&inode->i_lock);
2494 		schedule();
2495 
2496 		/*
2497 		 * shmem_falloc_waitq points into the shmem_fallocate()
2498 		 * stack of the hole-punching task: shmem_falloc_waitq
2499 		 * is usually invalid by the time we reach here, but
2500 		 * finish_wait() does not dereference it in that case;
2501 		 * though i_lock needed lest racing with wake_up_all().
2502 		 */
2503 		spin_lock(&inode->i_lock);
2504 		finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2505 	}
2506 	spin_unlock(&inode->i_lock);
2507 	if (fpin) {
2508 		fput(fpin);
2509 		ret = VM_FAULT_RETRY;
2510 	}
2511 	return ret;
2512 }
2513 
2514 static vm_fault_t shmem_fault(struct vm_fault *vmf)
2515 {
2516 	struct inode *inode = file_inode(vmf->vma->vm_file);
2517 	gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2518 	struct folio *folio = NULL;
2519 	vm_fault_t ret = 0;
2520 	int err;
2521 
2522 	/*
2523 	 * Trinity finds that probing a hole which tmpfs is punching can
2524 	 * prevent the hole-punch from ever completing: noted in i_private.
2525 	 */
2526 	if (unlikely(inode->i_private)) {
2527 		ret = shmem_falloc_wait(vmf, inode);
2528 		if (ret)
2529 			return ret;
2530 	}
2531 
2532 	WARN_ON_ONCE(vmf->page != NULL);
2533 	err = shmem_get_folio_gfp(inode, vmf->pgoff, 0, &folio, SGP_CACHE,
2534 				  gfp, vmf, &ret);
2535 	if (err)
2536 		return vmf_error(err);
2537 	if (folio) {
2538 		vmf->page = folio_file_page(folio, vmf->pgoff);
2539 		ret |= VM_FAULT_LOCKED;
2540 	}
2541 	return ret;
2542 }
2543 
2544 unsigned long shmem_get_unmapped_area(struct file *file,
2545 				      unsigned long uaddr, unsigned long len,
2546 				      unsigned long pgoff, unsigned long flags)
2547 {
2548 	unsigned long addr;
2549 	unsigned long offset;
2550 	unsigned long inflated_len;
2551 	unsigned long inflated_addr;
2552 	unsigned long inflated_offset;
2553 	unsigned long hpage_size;
2554 
2555 	if (len > TASK_SIZE)
2556 		return -ENOMEM;
2557 
2558 	addr = mm_get_unmapped_area(current->mm, file, uaddr, len, pgoff,
2559 				    flags);
2560 
2561 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2562 		return addr;
2563 	if (IS_ERR_VALUE(addr))
2564 		return addr;
2565 	if (addr & ~PAGE_MASK)
2566 		return addr;
2567 	if (addr > TASK_SIZE - len)
2568 		return addr;
2569 
2570 	if (shmem_huge == SHMEM_HUGE_DENY)
2571 		return addr;
2572 	if (flags & MAP_FIXED)
2573 		return addr;
2574 	/*
2575 	 * Our priority is to support MAP_SHARED mapped hugely;
2576 	 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2577 	 * But if caller specified an address hint and we allocated area there
2578 	 * successfully, respect that as before.
2579 	 */
2580 	if (uaddr == addr)
2581 		return addr;
2582 
2583 	hpage_size = HPAGE_PMD_SIZE;
2584 	if (shmem_huge != SHMEM_HUGE_FORCE) {
2585 		struct super_block *sb;
2586 		unsigned long __maybe_unused hpage_orders;
2587 		int order = 0;
2588 
2589 		if (file) {
2590 			VM_BUG_ON(file->f_op != &shmem_file_operations);
2591 			sb = file_inode(file)->i_sb;
2592 		} else {
2593 			/*
2594 			 * Called directly from mm/mmap.c, or drivers/char/mem.c
2595 			 * for "/dev/zero", to create a shared anonymous object.
2596 			 */
2597 			if (IS_ERR(shm_mnt))
2598 				return addr;
2599 			sb = shm_mnt->mnt_sb;
2600 
2601 			/*
2602 			 * Find the highest mTHP order used for anonymous shmem to
2603 			 * provide a suitable alignment address.
2604 			 */
2605 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2606 			hpage_orders = READ_ONCE(huge_shmem_orders_always);
2607 			hpage_orders |= READ_ONCE(huge_shmem_orders_within_size);
2608 			hpage_orders |= READ_ONCE(huge_shmem_orders_madvise);
2609 			if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER)
2610 				hpage_orders |= READ_ONCE(huge_shmem_orders_inherit);
2611 
2612 			if (hpage_orders > 0) {
2613 				order = highest_order(hpage_orders);
2614 				hpage_size = PAGE_SIZE << order;
2615 			}
2616 #endif
2617 		}
2618 		if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER && !order)
2619 			return addr;
2620 	}
2621 
2622 	if (len < hpage_size)
2623 		return addr;
2624 
2625 	offset = (pgoff << PAGE_SHIFT) & (hpage_size - 1);
2626 	if (offset && offset + len < 2 * hpage_size)
2627 		return addr;
2628 	if ((addr & (hpage_size - 1)) == offset)
2629 		return addr;
2630 
2631 	inflated_len = len + hpage_size - PAGE_SIZE;
2632 	if (inflated_len > TASK_SIZE)
2633 		return addr;
2634 	if (inflated_len < len)
2635 		return addr;
2636 
2637 	inflated_addr = mm_get_unmapped_area(current->mm, NULL, uaddr,
2638 					     inflated_len, 0, flags);
2639 	if (IS_ERR_VALUE(inflated_addr))
2640 		return addr;
2641 	if (inflated_addr & ~PAGE_MASK)
2642 		return addr;
2643 
2644 	inflated_offset = inflated_addr & (hpage_size - 1);
2645 	inflated_addr += offset - inflated_offset;
2646 	if (inflated_offset > offset)
2647 		inflated_addr += hpage_size;
2648 
2649 	if (inflated_addr > TASK_SIZE - len)
2650 		return addr;
2651 	return inflated_addr;
2652 }
2653 
2654 #ifdef CONFIG_NUMA
2655 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2656 {
2657 	struct inode *inode = file_inode(vma->vm_file);
2658 	return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2659 }
2660 
2661 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2662 					  unsigned long addr, pgoff_t *ilx)
2663 {
2664 	struct inode *inode = file_inode(vma->vm_file);
2665 	pgoff_t index;
2666 
2667 	/*
2668 	 * Bias interleave by inode number to distribute better across nodes;
2669 	 * but this interface is independent of which page order is used, so
2670 	 * supplies only that bias, letting caller apply the offset (adjusted
2671 	 * by page order, as in shmem_get_pgoff_policy() and get_vma_policy()).
2672 	 */
2673 	*ilx = inode->i_ino;
2674 	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2675 	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2676 }
2677 
2678 static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
2679 			pgoff_t index, unsigned int order, pgoff_t *ilx)
2680 {
2681 	struct mempolicy *mpol;
2682 
2683 	/* Bias interleave by inode number to distribute better across nodes */
2684 	*ilx = info->vfs_inode.i_ino + (index >> order);
2685 
2686 	mpol = mpol_shared_policy_lookup(&info->policy, index);
2687 	return mpol ? mpol : get_task_policy(current);
2688 }
2689 #else
2690 static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info,
2691 			pgoff_t index, unsigned int order, pgoff_t *ilx)
2692 {
2693 	*ilx = 0;
2694 	return NULL;
2695 }
2696 #endif /* CONFIG_NUMA */
2697 
2698 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
2699 {
2700 	struct inode *inode = file_inode(file);
2701 	struct shmem_inode_info *info = SHMEM_I(inode);
2702 	int retval = -ENOMEM;
2703 
2704 	/*
2705 	 * What serializes the accesses to info->flags?
2706 	 * ipc_lock_object() when called from shmctl_do_lock(),
2707 	 * no serialization needed when called from shm_destroy().
2708 	 */
2709 	if (lock && !(info->flags & VM_LOCKED)) {
2710 		if (!user_shm_lock(inode->i_size, ucounts))
2711 			goto out_nomem;
2712 		info->flags |= VM_LOCKED;
2713 		mapping_set_unevictable(file->f_mapping);
2714 	}
2715 	if (!lock && (info->flags & VM_LOCKED) && ucounts) {
2716 		user_shm_unlock(inode->i_size, ucounts);
2717 		info->flags &= ~VM_LOCKED;
2718 		mapping_clear_unevictable(file->f_mapping);
2719 	}
2720 	retval = 0;
2721 
2722 out_nomem:
2723 	return retval;
2724 }
2725 
2726 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2727 {
2728 	struct inode *inode = file_inode(file);
2729 	struct shmem_inode_info *info = SHMEM_I(inode);
2730 	int ret;
2731 
2732 	ret = seal_check_write(info->seals, vma);
2733 	if (ret)
2734 		return ret;
2735 
2736 	/* arm64 - allow memory tagging on RAM-based files */
2737 	vm_flags_set(vma, VM_MTE_ALLOWED);
2738 
2739 	file_accessed(file);
2740 	/* This is anonymous shared memory if it is unlinked at the time of mmap */
2741 	if (inode->i_nlink)
2742 		vma->vm_ops = &shmem_vm_ops;
2743 	else
2744 		vma->vm_ops = &shmem_anon_vm_ops;
2745 	return 0;
2746 }
2747 
2748 static int shmem_file_open(struct inode *inode, struct file *file)
2749 {
2750 	file->f_mode |= FMODE_CAN_ODIRECT;
2751 	return generic_file_open(inode, file);
2752 }
2753 
2754 #ifdef CONFIG_TMPFS_XATTR
2755 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2756 
2757 /*
2758  * chattr's fsflags are unrelated to extended attributes,
2759  * but tmpfs has chosen to enable them under the same config option.
2760  */
2761 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2762 {
2763 	unsigned int i_flags = 0;
2764 
2765 	if (fsflags & FS_NOATIME_FL)
2766 		i_flags |= S_NOATIME;
2767 	if (fsflags & FS_APPEND_FL)
2768 		i_flags |= S_APPEND;
2769 	if (fsflags & FS_IMMUTABLE_FL)
2770 		i_flags |= S_IMMUTABLE;
2771 	/*
2772 	 * But FS_NODUMP_FL does not require any action in i_flags.
2773 	 */
2774 	inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE);
2775 }
2776 #else
2777 static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags)
2778 {
2779 }
2780 #define shmem_initxattrs NULL
2781 #endif
2782 
2783 static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode)
2784 {
2785 	return &SHMEM_I(inode)->dir_offsets;
2786 }
2787 
2788 static struct inode *__shmem_get_inode(struct mnt_idmap *idmap,
2789 					     struct super_block *sb,
2790 					     struct inode *dir, umode_t mode,
2791 					     dev_t dev, unsigned long flags)
2792 {
2793 	struct inode *inode;
2794 	struct shmem_inode_info *info;
2795 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2796 	ino_t ino;
2797 	int err;
2798 
2799 	err = shmem_reserve_inode(sb, &ino);
2800 	if (err)
2801 		return ERR_PTR(err);
2802 
2803 	inode = new_inode(sb);
2804 	if (!inode) {
2805 		shmem_free_inode(sb, 0);
2806 		return ERR_PTR(-ENOSPC);
2807 	}
2808 
2809 	inode->i_ino = ino;
2810 	inode_init_owner(idmap, inode, dir, mode);
2811 	inode->i_blocks = 0;
2812 	simple_inode_init_ts(inode);
2813 	inode->i_generation = get_random_u32();
2814 	info = SHMEM_I(inode);
2815 	memset(info, 0, (char *)inode - (char *)info);
2816 	spin_lock_init(&info->lock);
2817 	atomic_set(&info->stop_eviction, 0);
2818 	info->seals = F_SEAL_SEAL;
2819 	info->flags = flags & VM_NORESERVE;
2820 	info->i_crtime = inode_get_mtime(inode);
2821 	info->fsflags = (dir == NULL) ? 0 :
2822 		SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED;
2823 	if (info->fsflags)
2824 		shmem_set_inode_flags(inode, info->fsflags);
2825 	INIT_LIST_HEAD(&info->shrinklist);
2826 	INIT_LIST_HEAD(&info->swaplist);
2827 	simple_xattrs_init(&info->xattrs);
2828 	cache_no_acl(inode);
2829 	if (sbinfo->noswap)
2830 		mapping_set_unevictable(inode->i_mapping);
2831 	mapping_set_large_folios(inode->i_mapping);
2832 
2833 	switch (mode & S_IFMT) {
2834 	default:
2835 		inode->i_op = &shmem_special_inode_operations;
2836 		init_special_inode(inode, mode, dev);
2837 		break;
2838 	case S_IFREG:
2839 		inode->i_mapping->a_ops = &shmem_aops;
2840 		inode->i_op = &shmem_inode_operations;
2841 		inode->i_fop = &shmem_file_operations;
2842 		mpol_shared_policy_init(&info->policy,
2843 					 shmem_get_sbmpol(sbinfo));
2844 		break;
2845 	case S_IFDIR:
2846 		inc_nlink(inode);
2847 		/* Some things misbehave if size == 0 on a directory */
2848 		inode->i_size = 2 * BOGO_DIRENT_SIZE;
2849 		inode->i_op = &shmem_dir_inode_operations;
2850 		inode->i_fop = &simple_offset_dir_operations;
2851 		simple_offset_init(shmem_get_offset_ctx(inode));
2852 		break;
2853 	case S_IFLNK:
2854 		/*
2855 		 * Must not load anything in the rbtree,
2856 		 * mpol_free_shared_policy will not be called.
2857 		 */
2858 		mpol_shared_policy_init(&info->policy, NULL);
2859 		break;
2860 	}
2861 
2862 	lockdep_annotate_inode_mutex_key(inode);
2863 	return inode;
2864 }
2865 
2866 #ifdef CONFIG_TMPFS_QUOTA
2867 static struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2868 				     struct super_block *sb, struct inode *dir,
2869 				     umode_t mode, dev_t dev, unsigned long flags)
2870 {
2871 	int err;
2872 	struct inode *inode;
2873 
2874 	inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2875 	if (IS_ERR(inode))
2876 		return inode;
2877 
2878 	err = dquot_initialize(inode);
2879 	if (err)
2880 		goto errout;
2881 
2882 	err = dquot_alloc_inode(inode);
2883 	if (err) {
2884 		dquot_drop(inode);
2885 		goto errout;
2886 	}
2887 	return inode;
2888 
2889 errout:
2890 	inode->i_flags |= S_NOQUOTA;
2891 	iput(inode);
2892 	return ERR_PTR(err);
2893 }
2894 #else
2895 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
2896 				     struct super_block *sb, struct inode *dir,
2897 				     umode_t mode, dev_t dev, unsigned long flags)
2898 {
2899 	return __shmem_get_inode(idmap, sb, dir, mode, dev, flags);
2900 }
2901 #endif /* CONFIG_TMPFS_QUOTA */
2902 
2903 #ifdef CONFIG_USERFAULTFD
2904 int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
2905 			   struct vm_area_struct *dst_vma,
2906 			   unsigned long dst_addr,
2907 			   unsigned long src_addr,
2908 			   uffd_flags_t flags,
2909 			   struct folio **foliop)
2910 {
2911 	struct inode *inode = file_inode(dst_vma->vm_file);
2912 	struct shmem_inode_info *info = SHMEM_I(inode);
2913 	struct address_space *mapping = inode->i_mapping;
2914 	gfp_t gfp = mapping_gfp_mask(mapping);
2915 	pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2916 	void *page_kaddr;
2917 	struct folio *folio;
2918 	int ret;
2919 	pgoff_t max_off;
2920 
2921 	if (shmem_inode_acct_blocks(inode, 1)) {
2922 		/*
2923 		 * We may have got a page, returned -ENOENT triggering a retry,
2924 		 * and now we find ourselves with -ENOMEM. Release the page, to
2925 		 * avoid a BUG_ON in our caller.
2926 		 */
2927 		if (unlikely(*foliop)) {
2928 			folio_put(*foliop);
2929 			*foliop = NULL;
2930 		}
2931 		return -ENOMEM;
2932 	}
2933 
2934 	if (!*foliop) {
2935 		ret = -ENOMEM;
2936 		folio = shmem_alloc_folio(gfp, 0, info, pgoff);
2937 		if (!folio)
2938 			goto out_unacct_blocks;
2939 
2940 		if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) {
2941 			page_kaddr = kmap_local_folio(folio, 0);
2942 			/*
2943 			 * The read mmap_lock is held here.  Despite the
2944 			 * mmap_lock being read recursive a deadlock is still
2945 			 * possible if a writer has taken a lock.  For example:
2946 			 *
2947 			 * process A thread 1 takes read lock on own mmap_lock
2948 			 * process A thread 2 calls mmap, blocks taking write lock
2949 			 * process B thread 1 takes page fault, read lock on own mmap lock
2950 			 * process B thread 2 calls mmap, blocks taking write lock
2951 			 * process A thread 1 blocks taking read lock on process B
2952 			 * process B thread 1 blocks taking read lock on process A
2953 			 *
2954 			 * Disable page faults to prevent potential deadlock
2955 			 * and retry the copy outside the mmap_lock.
2956 			 */
2957 			pagefault_disable();
2958 			ret = copy_from_user(page_kaddr,
2959 					     (const void __user *)src_addr,
2960 					     PAGE_SIZE);
2961 			pagefault_enable();
2962 			kunmap_local(page_kaddr);
2963 
2964 			/* fallback to copy_from_user outside mmap_lock */
2965 			if (unlikely(ret)) {
2966 				*foliop = folio;
2967 				ret = -ENOENT;
2968 				/* don't free the page */
2969 				goto out_unacct_blocks;
2970 			}
2971 
2972 			flush_dcache_folio(folio);
2973 		} else {		/* ZEROPAGE */
2974 			clear_user_highpage(&folio->page, dst_addr);
2975 		}
2976 	} else {
2977 		folio = *foliop;
2978 		VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
2979 		*foliop = NULL;
2980 	}
2981 
2982 	VM_BUG_ON(folio_test_locked(folio));
2983 	VM_BUG_ON(folio_test_swapbacked(folio));
2984 	__folio_set_locked(folio);
2985 	__folio_set_swapbacked(folio);
2986 	__folio_mark_uptodate(folio);
2987 
2988 	ret = -EFAULT;
2989 	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2990 	if (unlikely(pgoff >= max_off))
2991 		goto out_release;
2992 
2993 	ret = mem_cgroup_charge(folio, dst_vma->vm_mm, gfp);
2994 	if (ret)
2995 		goto out_release;
2996 	ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL, gfp);
2997 	if (ret)
2998 		goto out_release;
2999 
3000 	ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
3001 				       &folio->page, true, flags);
3002 	if (ret)
3003 		goto out_delete_from_cache;
3004 
3005 	shmem_recalc_inode(inode, 1, 0);
3006 	folio_unlock(folio);
3007 	return 0;
3008 out_delete_from_cache:
3009 	filemap_remove_folio(folio);
3010 out_release:
3011 	folio_unlock(folio);
3012 	folio_put(folio);
3013 out_unacct_blocks:
3014 	shmem_inode_unacct_blocks(inode, 1);
3015 	return ret;
3016 }
3017 #endif /* CONFIG_USERFAULTFD */
3018 
3019 #ifdef CONFIG_TMPFS
3020 static const struct inode_operations shmem_symlink_inode_operations;
3021 static const struct inode_operations shmem_short_symlink_operations;
3022 
3023 static int
3024 shmem_write_begin(struct file *file, struct address_space *mapping,
3025 			loff_t pos, unsigned len,
3026 			struct folio **foliop, void **fsdata)
3027 {
3028 	struct inode *inode = mapping->host;
3029 	struct shmem_inode_info *info = SHMEM_I(inode);
3030 	pgoff_t index = pos >> PAGE_SHIFT;
3031 	struct folio *folio;
3032 	int ret = 0;
3033 
3034 	/* i_rwsem is held by caller */
3035 	if (unlikely(info->seals & (F_SEAL_GROW |
3036 				   F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
3037 		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
3038 			return -EPERM;
3039 		if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
3040 			return -EPERM;
3041 	}
3042 
3043 	ret = shmem_get_folio(inode, index, pos + len, &folio, SGP_WRITE);
3044 	if (ret)
3045 		return ret;
3046 
3047 	if (folio_test_hwpoison(folio) ||
3048 	    (folio_test_large(folio) && folio_test_has_hwpoisoned(folio))) {
3049 		folio_unlock(folio);
3050 		folio_put(folio);
3051 		return -EIO;
3052 	}
3053 
3054 	*foliop = folio;
3055 	return 0;
3056 }
3057 
3058 static int
3059 shmem_write_end(struct file *file, struct address_space *mapping,
3060 			loff_t pos, unsigned len, unsigned copied,
3061 			struct folio *folio, void *fsdata)
3062 {
3063 	struct inode *inode = mapping->host;
3064 
3065 	if (pos + copied > inode->i_size)
3066 		i_size_write(inode, pos + copied);
3067 
3068 	if (!folio_test_uptodate(folio)) {
3069 		if (copied < folio_size(folio)) {
3070 			size_t from = offset_in_folio(folio, pos);
3071 			folio_zero_segments(folio, 0, from,
3072 					from + copied, folio_size(folio));
3073 		}
3074 		folio_mark_uptodate(folio);
3075 	}
3076 	folio_mark_dirty(folio);
3077 	folio_unlock(folio);
3078 	folio_put(folio);
3079 
3080 	return copied;
3081 }
3082 
3083 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3084 {
3085 	struct file *file = iocb->ki_filp;
3086 	struct inode *inode = file_inode(file);
3087 	struct address_space *mapping = inode->i_mapping;
3088 	pgoff_t index;
3089 	unsigned long offset;
3090 	int error = 0;
3091 	ssize_t retval = 0;
3092 	loff_t *ppos = &iocb->ki_pos;
3093 
3094 	index = *ppos >> PAGE_SHIFT;
3095 	offset = *ppos & ~PAGE_MASK;
3096 
3097 	for (;;) {
3098 		struct folio *folio = NULL;
3099 		struct page *page = NULL;
3100 		pgoff_t end_index;
3101 		unsigned long nr, ret;
3102 		loff_t i_size = i_size_read(inode);
3103 
3104 		end_index = i_size >> PAGE_SHIFT;
3105 		if (index > end_index)
3106 			break;
3107 		if (index == end_index) {
3108 			nr = i_size & ~PAGE_MASK;
3109 			if (nr <= offset)
3110 				break;
3111 		}
3112 
3113 		error = shmem_get_folio(inode, index, 0, &folio, SGP_READ);
3114 		if (error) {
3115 			if (error == -EINVAL)
3116 				error = 0;
3117 			break;
3118 		}
3119 		if (folio) {
3120 			folio_unlock(folio);
3121 
3122 			page = folio_file_page(folio, index);
3123 			if (PageHWPoison(page)) {
3124 				folio_put(folio);
3125 				error = -EIO;
3126 				break;
3127 			}
3128 		}
3129 
3130 		/*
3131 		 * We must evaluate after, since reads (unlike writes)
3132 		 * are called without i_rwsem protection against truncate
3133 		 */
3134 		nr = PAGE_SIZE;
3135 		i_size = i_size_read(inode);
3136 		end_index = i_size >> PAGE_SHIFT;
3137 		if (index == end_index) {
3138 			nr = i_size & ~PAGE_MASK;
3139 			if (nr <= offset) {
3140 				if (folio)
3141 					folio_put(folio);
3142 				break;
3143 			}
3144 		}
3145 		nr -= offset;
3146 
3147 		if (folio) {
3148 			/*
3149 			 * If users can be writing to this page using arbitrary
3150 			 * virtual addresses, take care about potential aliasing
3151 			 * before reading the page on the kernel side.
3152 			 */
3153 			if (mapping_writably_mapped(mapping))
3154 				flush_dcache_page(page);
3155 			/*
3156 			 * Mark the page accessed if we read the beginning.
3157 			 */
3158 			if (!offset)
3159 				folio_mark_accessed(folio);
3160 			/*
3161 			 * Ok, we have the page, and it's up-to-date, so
3162 			 * now we can copy it to user space...
3163 			 */
3164 			ret = copy_page_to_iter(page, offset, nr, to);
3165 			folio_put(folio);
3166 
3167 		} else if (user_backed_iter(to)) {
3168 			/*
3169 			 * Copy to user tends to be so well optimized, but
3170 			 * clear_user() not so much, that it is noticeably
3171 			 * faster to copy the zero page instead of clearing.
3172 			 */
3173 			ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to);
3174 		} else {
3175 			/*
3176 			 * But submitting the same page twice in a row to
3177 			 * splice() - or others? - can result in confusion:
3178 			 * so don't attempt that optimization on pipes etc.
3179 			 */
3180 			ret = iov_iter_zero(nr, to);
3181 		}
3182 
3183 		retval += ret;
3184 		offset += ret;
3185 		index += offset >> PAGE_SHIFT;
3186 		offset &= ~PAGE_MASK;
3187 
3188 		if (!iov_iter_count(to))
3189 			break;
3190 		if (ret < nr) {
3191 			error = -EFAULT;
3192 			break;
3193 		}
3194 		cond_resched();
3195 	}
3196 
3197 	*ppos = ((loff_t) index << PAGE_SHIFT) + offset;
3198 	file_accessed(file);
3199 	return retval ? retval : error;
3200 }
3201 
3202 static ssize_t shmem_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3203 {
3204 	struct file *file = iocb->ki_filp;
3205 	struct inode *inode = file->f_mapping->host;
3206 	ssize_t ret;
3207 
3208 	inode_lock(inode);
3209 	ret = generic_write_checks(iocb, from);
3210 	if (ret <= 0)
3211 		goto unlock;
3212 	ret = file_remove_privs(file);
3213 	if (ret)
3214 		goto unlock;
3215 	ret = file_update_time(file);
3216 	if (ret)
3217 		goto unlock;
3218 	ret = generic_perform_write(iocb, from);
3219 unlock:
3220 	inode_unlock(inode);
3221 	return ret;
3222 }
3223 
3224 static bool zero_pipe_buf_get(struct pipe_inode_info *pipe,
3225 			      struct pipe_buffer *buf)
3226 {
3227 	return true;
3228 }
3229 
3230 static void zero_pipe_buf_release(struct pipe_inode_info *pipe,
3231 				  struct pipe_buffer *buf)
3232 {
3233 }
3234 
3235 static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe,
3236 				    struct pipe_buffer *buf)
3237 {
3238 	return false;
3239 }
3240 
3241 static const struct pipe_buf_operations zero_pipe_buf_ops = {
3242 	.release	= zero_pipe_buf_release,
3243 	.try_steal	= zero_pipe_buf_try_steal,
3244 	.get		= zero_pipe_buf_get,
3245 };
3246 
3247 static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe,
3248 					loff_t fpos, size_t size)
3249 {
3250 	size_t offset = fpos & ~PAGE_MASK;
3251 
3252 	size = min_t(size_t, size, PAGE_SIZE - offset);
3253 
3254 	if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
3255 		struct pipe_buffer *buf = pipe_head_buf(pipe);
3256 
3257 		*buf = (struct pipe_buffer) {
3258 			.ops	= &zero_pipe_buf_ops,
3259 			.page	= ZERO_PAGE(0),
3260 			.offset	= offset,
3261 			.len	= size,
3262 		};
3263 		pipe->head++;
3264 	}
3265 
3266 	return size;
3267 }
3268 
3269 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
3270 				      struct pipe_inode_info *pipe,
3271 				      size_t len, unsigned int flags)
3272 {
3273 	struct inode *inode = file_inode(in);
3274 	struct address_space *mapping = inode->i_mapping;
3275 	struct folio *folio = NULL;
3276 	size_t total_spliced = 0, used, npages, n, part;
3277 	loff_t isize;
3278 	int error = 0;
3279 
3280 	/* Work out how much data we can actually add into the pipe */
3281 	used = pipe_occupancy(pipe->head, pipe->tail);
3282 	npages = max_t(ssize_t, pipe->max_usage - used, 0);
3283 	len = min_t(size_t, len, npages * PAGE_SIZE);
3284 
3285 	do {
3286 		if (*ppos >= i_size_read(inode))
3287 			break;
3288 
3289 		error = shmem_get_folio(inode, *ppos / PAGE_SIZE, 0, &folio,
3290 					SGP_READ);
3291 		if (error) {
3292 			if (error == -EINVAL)
3293 				error = 0;
3294 			break;
3295 		}
3296 		if (folio) {
3297 			folio_unlock(folio);
3298 
3299 			if (folio_test_hwpoison(folio) ||
3300 			    (folio_test_large(folio) &&
3301 			     folio_test_has_hwpoisoned(folio))) {
3302 				error = -EIO;
3303 				break;
3304 			}
3305 		}
3306 
3307 		/*
3308 		 * i_size must be checked after we know the pages are Uptodate.
3309 		 *
3310 		 * Checking i_size after the check allows us to calculate
3311 		 * the correct value for "nr", which means the zero-filled
3312 		 * part of the page is not copied back to userspace (unless
3313 		 * another truncate extends the file - this is desired though).
3314 		 */
3315 		isize = i_size_read(inode);
3316 		if (unlikely(*ppos >= isize))
3317 			break;
3318 		part = min_t(loff_t, isize - *ppos, len);
3319 
3320 		if (folio) {
3321 			/*
3322 			 * If users can be writing to this page using arbitrary
3323 			 * virtual addresses, take care about potential aliasing
3324 			 * before reading the page on the kernel side.
3325 			 */
3326 			if (mapping_writably_mapped(mapping))
3327 				flush_dcache_folio(folio);
3328 			folio_mark_accessed(folio);
3329 			/*
3330 			 * Ok, we have the page, and it's up-to-date, so we can
3331 			 * now splice it into the pipe.
3332 			 */
3333 			n = splice_folio_into_pipe(pipe, folio, *ppos, part);
3334 			folio_put(folio);
3335 			folio = NULL;
3336 		} else {
3337 			n = splice_zeropage_into_pipe(pipe, *ppos, part);
3338 		}
3339 
3340 		if (!n)
3341 			break;
3342 		len -= n;
3343 		total_spliced += n;
3344 		*ppos += n;
3345 		in->f_ra.prev_pos = *ppos;
3346 		if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
3347 			break;
3348 
3349 		cond_resched();
3350 	} while (len);
3351 
3352 	if (folio)
3353 		folio_put(folio);
3354 
3355 	file_accessed(in);
3356 	return total_spliced ? total_spliced : error;
3357 }
3358 
3359 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
3360 {
3361 	struct address_space *mapping = file->f_mapping;
3362 	struct inode *inode = mapping->host;
3363 
3364 	if (whence != SEEK_DATA && whence != SEEK_HOLE)
3365 		return generic_file_llseek_size(file, offset, whence,
3366 					MAX_LFS_FILESIZE, i_size_read(inode));
3367 	if (offset < 0)
3368 		return -ENXIO;
3369 
3370 	inode_lock(inode);
3371 	/* We're holding i_rwsem so we can access i_size directly */
3372 	offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence);
3373 	if (offset >= 0)
3374 		offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
3375 	inode_unlock(inode);
3376 	return offset;
3377 }
3378 
3379 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
3380 							 loff_t len)
3381 {
3382 	struct inode *inode = file_inode(file);
3383 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3384 	struct shmem_inode_info *info = SHMEM_I(inode);
3385 	struct shmem_falloc shmem_falloc;
3386 	pgoff_t start, index, end, undo_fallocend;
3387 	int error;
3388 
3389 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3390 		return -EOPNOTSUPP;
3391 
3392 	inode_lock(inode);
3393 
3394 	if (mode & FALLOC_FL_PUNCH_HOLE) {
3395 		struct address_space *mapping = file->f_mapping;
3396 		loff_t unmap_start = round_up(offset, PAGE_SIZE);
3397 		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
3398 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
3399 
3400 		/* protected by i_rwsem */
3401 		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
3402 			error = -EPERM;
3403 			goto out;
3404 		}
3405 
3406 		shmem_falloc.waitq = &shmem_falloc_waitq;
3407 		shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
3408 		shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
3409 		spin_lock(&inode->i_lock);
3410 		inode->i_private = &shmem_falloc;
3411 		spin_unlock(&inode->i_lock);
3412 
3413 		if ((u64)unmap_end > (u64)unmap_start)
3414 			unmap_mapping_range(mapping, unmap_start,
3415 					    1 + unmap_end - unmap_start, 0);
3416 		shmem_truncate_range(inode, offset, offset + len - 1);
3417 		/* No need to unmap again: hole-punching leaves COWed pages */
3418 
3419 		spin_lock(&inode->i_lock);
3420 		inode->i_private = NULL;
3421 		wake_up_all(&shmem_falloc_waitq);
3422 		WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
3423 		spin_unlock(&inode->i_lock);
3424 		error = 0;
3425 		goto out;
3426 	}
3427 
3428 	/* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
3429 	error = inode_newsize_ok(inode, offset + len);
3430 	if (error)
3431 		goto out;
3432 
3433 	if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
3434 		error = -EPERM;
3435 		goto out;
3436 	}
3437 
3438 	start = offset >> PAGE_SHIFT;
3439 	end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
3440 	/* Try to avoid a swapstorm if len is impossible to satisfy */
3441 	if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
3442 		error = -ENOSPC;
3443 		goto out;
3444 	}
3445 
3446 	shmem_falloc.waitq = NULL;
3447 	shmem_falloc.start = start;
3448 	shmem_falloc.next  = start;
3449 	shmem_falloc.nr_falloced = 0;
3450 	shmem_falloc.nr_unswapped = 0;
3451 	spin_lock(&inode->i_lock);
3452 	inode->i_private = &shmem_falloc;
3453 	spin_unlock(&inode->i_lock);
3454 
3455 	/*
3456 	 * info->fallocend is only relevant when huge pages might be
3457 	 * involved: to prevent split_huge_page() freeing fallocated
3458 	 * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size.
3459 	 */
3460 	undo_fallocend = info->fallocend;
3461 	if (info->fallocend < end)
3462 		info->fallocend = end;
3463 
3464 	for (index = start; index < end; ) {
3465 		struct folio *folio;
3466 
3467 		/*
3468 		 * Check for fatal signal so that we abort early in OOM
3469 		 * situations. We don't want to abort in case of non-fatal
3470 		 * signals as large fallocate can take noticeable time and
3471 		 * e.g. periodic timers may result in fallocate constantly
3472 		 * restarting.
3473 		 */
3474 		if (fatal_signal_pending(current))
3475 			error = -EINTR;
3476 		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
3477 			error = -ENOMEM;
3478 		else
3479 			error = shmem_get_folio(inode, index, offset + len,
3480 						&folio, SGP_FALLOC);
3481 		if (error) {
3482 			info->fallocend = undo_fallocend;
3483 			/* Remove the !uptodate folios we added */
3484 			if (index > start) {
3485 				shmem_undo_range(inode,
3486 				    (loff_t)start << PAGE_SHIFT,
3487 				    ((loff_t)index << PAGE_SHIFT) - 1, true);
3488 			}
3489 			goto undone;
3490 		}
3491 
3492 		/*
3493 		 * Here is a more important optimization than it appears:
3494 		 * a second SGP_FALLOC on the same large folio will clear it,
3495 		 * making it uptodate and un-undoable if we fail later.
3496 		 */
3497 		index = folio_next_index(folio);
3498 		/* Beware 32-bit wraparound */
3499 		if (!index)
3500 			index--;
3501 
3502 		/*
3503 		 * Inform shmem_writepage() how far we have reached.
3504 		 * No need for lock or barrier: we have the page lock.
3505 		 */
3506 		if (!folio_test_uptodate(folio))
3507 			shmem_falloc.nr_falloced += index - shmem_falloc.next;
3508 		shmem_falloc.next = index;
3509 
3510 		/*
3511 		 * If !uptodate, leave it that way so that freeable folios
3512 		 * can be recognized if we need to rollback on error later.
3513 		 * But mark it dirty so that memory pressure will swap rather
3514 		 * than free the folios we are allocating (and SGP_CACHE folios
3515 		 * might still be clean: we now need to mark those dirty too).
3516 		 */
3517 		folio_mark_dirty(folio);
3518 		folio_unlock(folio);
3519 		folio_put(folio);
3520 		cond_resched();
3521 	}
3522 
3523 	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
3524 		i_size_write(inode, offset + len);
3525 undone:
3526 	spin_lock(&inode->i_lock);
3527 	inode->i_private = NULL;
3528 	spin_unlock(&inode->i_lock);
3529 out:
3530 	if (!error)
3531 		file_modified(file);
3532 	inode_unlock(inode);
3533 	return error;
3534 }
3535 
3536 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
3537 {
3538 	struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
3539 
3540 	buf->f_type = TMPFS_MAGIC;
3541 	buf->f_bsize = PAGE_SIZE;
3542 	buf->f_namelen = NAME_MAX;
3543 	if (sbinfo->max_blocks) {
3544 		buf->f_blocks = sbinfo->max_blocks;
3545 		buf->f_bavail =
3546 		buf->f_bfree  = sbinfo->max_blocks -
3547 				percpu_counter_sum(&sbinfo->used_blocks);
3548 	}
3549 	if (sbinfo->max_inodes) {
3550 		buf->f_files = sbinfo->max_inodes;
3551 		buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE;
3552 	}
3553 	/* else leave those fields 0 like simple_statfs */
3554 
3555 	buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b);
3556 
3557 	return 0;
3558 }
3559 
3560 /*
3561  * File creation. Allocate an inode, and we're done..
3562  */
3563 static int
3564 shmem_mknod(struct mnt_idmap *idmap, struct inode *dir,
3565 	    struct dentry *dentry, umode_t mode, dev_t dev)
3566 {
3567 	struct inode *inode;
3568 	int error;
3569 
3570 	inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE);
3571 	if (IS_ERR(inode))
3572 		return PTR_ERR(inode);
3573 
3574 	error = simple_acl_create(dir, inode);
3575 	if (error)
3576 		goto out_iput;
3577 	error = security_inode_init_security(inode, dir, &dentry->d_name,
3578 					     shmem_initxattrs, NULL);
3579 	if (error && error != -EOPNOTSUPP)
3580 		goto out_iput;
3581 
3582 	error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3583 	if (error)
3584 		goto out_iput;
3585 
3586 	dir->i_size += BOGO_DIRENT_SIZE;
3587 	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3588 	inode_inc_iversion(dir);
3589 	d_instantiate(dentry, inode);
3590 	dget(dentry); /* Extra count - pin the dentry in core */
3591 	return error;
3592 
3593 out_iput:
3594 	iput(inode);
3595 	return error;
3596 }
3597 
3598 static int
3599 shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
3600 	      struct file *file, umode_t mode)
3601 {
3602 	struct inode *inode;
3603 	int error;
3604 
3605 	inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE);
3606 	if (IS_ERR(inode)) {
3607 		error = PTR_ERR(inode);
3608 		goto err_out;
3609 	}
3610 	error = security_inode_init_security(inode, dir, NULL,
3611 					     shmem_initxattrs, NULL);
3612 	if (error && error != -EOPNOTSUPP)
3613 		goto out_iput;
3614 	error = simple_acl_create(dir, inode);
3615 	if (error)
3616 		goto out_iput;
3617 	d_tmpfile(file, inode);
3618 
3619 err_out:
3620 	return finish_open_simple(file, error);
3621 out_iput:
3622 	iput(inode);
3623 	return error;
3624 }
3625 
3626 static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3627 		       struct dentry *dentry, umode_t mode)
3628 {
3629 	int error;
3630 
3631 	error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0);
3632 	if (error)
3633 		return error;
3634 	inc_nlink(dir);
3635 	return 0;
3636 }
3637 
3638 static int shmem_create(struct mnt_idmap *idmap, struct inode *dir,
3639 			struct dentry *dentry, umode_t mode, bool excl)
3640 {
3641 	return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
3642 }
3643 
3644 /*
3645  * Link a file..
3646  */
3647 static int shmem_link(struct dentry *old_dentry, struct inode *dir,
3648 		      struct dentry *dentry)
3649 {
3650 	struct inode *inode = d_inode(old_dentry);
3651 	int ret = 0;
3652 
3653 	/*
3654 	 * No ordinary (disk based) filesystem counts links as inodes;
3655 	 * but each new link needs a new dentry, pinning lowmem, and
3656 	 * tmpfs dentries cannot be pruned until they are unlinked.
3657 	 * But if an O_TMPFILE file is linked into the tmpfs, the
3658 	 * first link must skip that, to get the accounting right.
3659 	 */
3660 	if (inode->i_nlink) {
3661 		ret = shmem_reserve_inode(inode->i_sb, NULL);
3662 		if (ret)
3663 			goto out;
3664 	}
3665 
3666 	ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3667 	if (ret) {
3668 		if (inode->i_nlink)
3669 			shmem_free_inode(inode->i_sb, 0);
3670 		goto out;
3671 	}
3672 
3673 	dir->i_size += BOGO_DIRENT_SIZE;
3674 	inode_set_mtime_to_ts(dir,
3675 			      inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
3676 	inode_inc_iversion(dir);
3677 	inc_nlink(inode);
3678 	ihold(inode);	/* New dentry reference */
3679 	dget(dentry);	/* Extra pinning count for the created dentry */
3680 	d_instantiate(dentry, inode);
3681 out:
3682 	return ret;
3683 }
3684 
3685 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3686 {
3687 	struct inode *inode = d_inode(dentry);
3688 
3689 	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3690 		shmem_free_inode(inode->i_sb, 0);
3691 
3692 	simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3693 
3694 	dir->i_size -= BOGO_DIRENT_SIZE;
3695 	inode_set_mtime_to_ts(dir,
3696 			      inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode)));
3697 	inode_inc_iversion(dir);
3698 	drop_nlink(inode);
3699 	dput(dentry);	/* Undo the count from "create" - does all the work */
3700 	return 0;
3701 }
3702 
3703 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3704 {
3705 	if (!simple_offset_empty(dentry))
3706 		return -ENOTEMPTY;
3707 
3708 	drop_nlink(d_inode(dentry));
3709 	drop_nlink(dir);
3710 	return shmem_unlink(dir, dentry);
3711 }
3712 
3713 static int shmem_whiteout(struct mnt_idmap *idmap,
3714 			  struct inode *old_dir, struct dentry *old_dentry)
3715 {
3716 	struct dentry *whiteout;
3717 	int error;
3718 
3719 	whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3720 	if (!whiteout)
3721 		return -ENOMEM;
3722 
3723 	error = shmem_mknod(idmap, old_dir, whiteout,
3724 			    S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3725 	dput(whiteout);
3726 	if (error)
3727 		return error;
3728 
3729 	/*
3730 	 * Cheat and hash the whiteout while the old dentry is still in
3731 	 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3732 	 *
3733 	 * d_lookup() will consistently find one of them at this point,
3734 	 * not sure which one, but that isn't even important.
3735 	 */
3736 	d_rehash(whiteout);
3737 	return 0;
3738 }
3739 
3740 /*
3741  * The VFS layer already does all the dentry stuff for rename,
3742  * we just have to decrement the usage count for the target if
3743  * it exists so that the VFS layer correctly free's it when it
3744  * gets overwritten.
3745  */
3746 static int shmem_rename2(struct mnt_idmap *idmap,
3747 			 struct inode *old_dir, struct dentry *old_dentry,
3748 			 struct inode *new_dir, struct dentry *new_dentry,
3749 			 unsigned int flags)
3750 {
3751 	struct inode *inode = d_inode(old_dentry);
3752 	int they_are_dirs = S_ISDIR(inode->i_mode);
3753 	int error;
3754 
3755 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3756 		return -EINVAL;
3757 
3758 	if (flags & RENAME_EXCHANGE)
3759 		return simple_offset_rename_exchange(old_dir, old_dentry,
3760 						     new_dir, new_dentry);
3761 
3762 	if (!simple_offset_empty(new_dentry))
3763 		return -ENOTEMPTY;
3764 
3765 	if (flags & RENAME_WHITEOUT) {
3766 		error = shmem_whiteout(idmap, old_dir, old_dentry);
3767 		if (error)
3768 			return error;
3769 	}
3770 
3771 	error = simple_offset_rename(old_dir, old_dentry, new_dir, new_dentry);
3772 	if (error)
3773 		return error;
3774 
3775 	if (d_really_is_positive(new_dentry)) {
3776 		(void) shmem_unlink(new_dir, new_dentry);
3777 		if (they_are_dirs) {
3778 			drop_nlink(d_inode(new_dentry));
3779 			drop_nlink(old_dir);
3780 		}
3781 	} else if (they_are_dirs) {
3782 		drop_nlink(old_dir);
3783 		inc_nlink(new_dir);
3784 	}
3785 
3786 	old_dir->i_size -= BOGO_DIRENT_SIZE;
3787 	new_dir->i_size += BOGO_DIRENT_SIZE;
3788 	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
3789 	inode_inc_iversion(old_dir);
3790 	inode_inc_iversion(new_dir);
3791 	return 0;
3792 }
3793 
3794 static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
3795 			 struct dentry *dentry, const char *symname)
3796 {
3797 	int error;
3798 	int len;
3799 	struct inode *inode;
3800 	struct folio *folio;
3801 
3802 	len = strlen(symname) + 1;
3803 	if (len > PAGE_SIZE)
3804 		return -ENAMETOOLONG;
3805 
3806 	inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0,
3807 				VM_NORESERVE);
3808 	if (IS_ERR(inode))
3809 		return PTR_ERR(inode);
3810 
3811 	error = security_inode_init_security(inode, dir, &dentry->d_name,
3812 					     shmem_initxattrs, NULL);
3813 	if (error && error != -EOPNOTSUPP)
3814 		goto out_iput;
3815 
3816 	error = simple_offset_add(shmem_get_offset_ctx(dir), dentry);
3817 	if (error)
3818 		goto out_iput;
3819 
3820 	inode->i_size = len-1;
3821 	if (len <= SHORT_SYMLINK_LEN) {
3822 		inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3823 		if (!inode->i_link) {
3824 			error = -ENOMEM;
3825 			goto out_remove_offset;
3826 		}
3827 		inode->i_op = &shmem_short_symlink_operations;
3828 	} else {
3829 		inode_nohighmem(inode);
3830 		inode->i_mapping->a_ops = &shmem_aops;
3831 		error = shmem_get_folio(inode, 0, 0, &folio, SGP_WRITE);
3832 		if (error)
3833 			goto out_remove_offset;
3834 		inode->i_op = &shmem_symlink_inode_operations;
3835 		memcpy(folio_address(folio), symname, len);
3836 		folio_mark_uptodate(folio);
3837 		folio_mark_dirty(folio);
3838 		folio_unlock(folio);
3839 		folio_put(folio);
3840 	}
3841 	dir->i_size += BOGO_DIRENT_SIZE;
3842 	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3843 	inode_inc_iversion(dir);
3844 	d_instantiate(dentry, inode);
3845 	dget(dentry);
3846 	return 0;
3847 
3848 out_remove_offset:
3849 	simple_offset_remove(shmem_get_offset_ctx(dir), dentry);
3850 out_iput:
3851 	iput(inode);
3852 	return error;
3853 }
3854 
3855 static void shmem_put_link(void *arg)
3856 {
3857 	folio_mark_accessed(arg);
3858 	folio_put(arg);
3859 }
3860 
3861 static const char *shmem_get_link(struct dentry *dentry, struct inode *inode,
3862 				  struct delayed_call *done)
3863 {
3864 	struct folio *folio = NULL;
3865 	int error;
3866 
3867 	if (!dentry) {
3868 		folio = filemap_get_folio(inode->i_mapping, 0);
3869 		if (IS_ERR(folio))
3870 			return ERR_PTR(-ECHILD);
3871 		if (PageHWPoison(folio_page(folio, 0)) ||
3872 		    !folio_test_uptodate(folio)) {
3873 			folio_put(folio);
3874 			return ERR_PTR(-ECHILD);
3875 		}
3876 	} else {
3877 		error = shmem_get_folio(inode, 0, 0, &folio, SGP_READ);
3878 		if (error)
3879 			return ERR_PTR(error);
3880 		if (!folio)
3881 			return ERR_PTR(-ECHILD);
3882 		if (PageHWPoison(folio_page(folio, 0))) {
3883 			folio_unlock(folio);
3884 			folio_put(folio);
3885 			return ERR_PTR(-ECHILD);
3886 		}
3887 		folio_unlock(folio);
3888 	}
3889 	set_delayed_call(done, shmem_put_link, folio);
3890 	return folio_address(folio);
3891 }
3892 
3893 #ifdef CONFIG_TMPFS_XATTR
3894 
3895 static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa)
3896 {
3897 	struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3898 
3899 	fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE);
3900 
3901 	return 0;
3902 }
3903 
3904 static int shmem_fileattr_set(struct mnt_idmap *idmap,
3905 			      struct dentry *dentry, struct fileattr *fa)
3906 {
3907 	struct inode *inode = d_inode(dentry);
3908 	struct shmem_inode_info *info = SHMEM_I(inode);
3909 
3910 	if (fileattr_has_fsx(fa))
3911 		return -EOPNOTSUPP;
3912 	if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE)
3913 		return -EOPNOTSUPP;
3914 
3915 	info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) |
3916 		(fa->flags & SHMEM_FL_USER_MODIFIABLE);
3917 
3918 	shmem_set_inode_flags(inode, info->fsflags);
3919 	inode_set_ctime_current(inode);
3920 	inode_inc_iversion(inode);
3921 	return 0;
3922 }
3923 
3924 /*
3925  * Superblocks without xattr inode operations may get some security.* xattr
3926  * support from the LSM "for free". As soon as we have any other xattrs
3927  * like ACLs, we also need to implement the security.* handlers at
3928  * filesystem level, though.
3929  */
3930 
3931 /*
3932  * Callback for security_inode_init_security() for acquiring xattrs.
3933  */
3934 static int shmem_initxattrs(struct inode *inode,
3935 			    const struct xattr *xattr_array, void *fs_info)
3936 {
3937 	struct shmem_inode_info *info = SHMEM_I(inode);
3938 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3939 	const struct xattr *xattr;
3940 	struct simple_xattr *new_xattr;
3941 	size_t ispace = 0;
3942 	size_t len;
3943 
3944 	if (sbinfo->max_inodes) {
3945 		for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3946 			ispace += simple_xattr_space(xattr->name,
3947 				xattr->value_len + XATTR_SECURITY_PREFIX_LEN);
3948 		}
3949 		if (ispace) {
3950 			raw_spin_lock(&sbinfo->stat_lock);
3951 			if (sbinfo->free_ispace < ispace)
3952 				ispace = 0;
3953 			else
3954 				sbinfo->free_ispace -= ispace;
3955 			raw_spin_unlock(&sbinfo->stat_lock);
3956 			if (!ispace)
3957 				return -ENOSPC;
3958 		}
3959 	}
3960 
3961 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3962 		new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3963 		if (!new_xattr)
3964 			break;
3965 
3966 		len = strlen(xattr->name) + 1;
3967 		new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3968 					  GFP_KERNEL_ACCOUNT);
3969 		if (!new_xattr->name) {
3970 			kvfree(new_xattr);
3971 			break;
3972 		}
3973 
3974 		memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3975 		       XATTR_SECURITY_PREFIX_LEN);
3976 		memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3977 		       xattr->name, len);
3978 
3979 		simple_xattr_add(&info->xattrs, new_xattr);
3980 	}
3981 
3982 	if (xattr->name != NULL) {
3983 		if (ispace) {
3984 			raw_spin_lock(&sbinfo->stat_lock);
3985 			sbinfo->free_ispace += ispace;
3986 			raw_spin_unlock(&sbinfo->stat_lock);
3987 		}
3988 		simple_xattrs_free(&info->xattrs, NULL);
3989 		return -ENOMEM;
3990 	}
3991 
3992 	return 0;
3993 }
3994 
3995 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3996 				   struct dentry *unused, struct inode *inode,
3997 				   const char *name, void *buffer, size_t size)
3998 {
3999 	struct shmem_inode_info *info = SHMEM_I(inode);
4000 
4001 	name = xattr_full_name(handler, name);
4002 	return simple_xattr_get(&info->xattrs, name, buffer, size);
4003 }
4004 
4005 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
4006 				   struct mnt_idmap *idmap,
4007 				   struct dentry *unused, struct inode *inode,
4008 				   const char *name, const void *value,
4009 				   size_t size, int flags)
4010 {
4011 	struct shmem_inode_info *info = SHMEM_I(inode);
4012 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4013 	struct simple_xattr *old_xattr;
4014 	size_t ispace = 0;
4015 
4016 	name = xattr_full_name(handler, name);
4017 	if (value && sbinfo->max_inodes) {
4018 		ispace = simple_xattr_space(name, size);
4019 		raw_spin_lock(&sbinfo->stat_lock);
4020 		if (sbinfo->free_ispace < ispace)
4021 			ispace = 0;
4022 		else
4023 			sbinfo->free_ispace -= ispace;
4024 		raw_spin_unlock(&sbinfo->stat_lock);
4025 		if (!ispace)
4026 			return -ENOSPC;
4027 	}
4028 
4029 	old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags);
4030 	if (!IS_ERR(old_xattr)) {
4031 		ispace = 0;
4032 		if (old_xattr && sbinfo->max_inodes)
4033 			ispace = simple_xattr_space(old_xattr->name,
4034 						    old_xattr->size);
4035 		simple_xattr_free(old_xattr);
4036 		old_xattr = NULL;
4037 		inode_set_ctime_current(inode);
4038 		inode_inc_iversion(inode);
4039 	}
4040 	if (ispace) {
4041 		raw_spin_lock(&sbinfo->stat_lock);
4042 		sbinfo->free_ispace += ispace;
4043 		raw_spin_unlock(&sbinfo->stat_lock);
4044 	}
4045 	return PTR_ERR(old_xattr);
4046 }
4047 
4048 static const struct xattr_handler shmem_security_xattr_handler = {
4049 	.prefix = XATTR_SECURITY_PREFIX,
4050 	.get = shmem_xattr_handler_get,
4051 	.set = shmem_xattr_handler_set,
4052 };
4053 
4054 static const struct xattr_handler shmem_trusted_xattr_handler = {
4055 	.prefix = XATTR_TRUSTED_PREFIX,
4056 	.get = shmem_xattr_handler_get,
4057 	.set = shmem_xattr_handler_set,
4058 };
4059 
4060 static const struct xattr_handler shmem_user_xattr_handler = {
4061 	.prefix = XATTR_USER_PREFIX,
4062 	.get = shmem_xattr_handler_get,
4063 	.set = shmem_xattr_handler_set,
4064 };
4065 
4066 static const struct xattr_handler * const shmem_xattr_handlers[] = {
4067 	&shmem_security_xattr_handler,
4068 	&shmem_trusted_xattr_handler,
4069 	&shmem_user_xattr_handler,
4070 	NULL
4071 };
4072 
4073 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
4074 {
4075 	struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
4076 	return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
4077 }
4078 #endif /* CONFIG_TMPFS_XATTR */
4079 
4080 static const struct inode_operations shmem_short_symlink_operations = {
4081 	.getattr	= shmem_getattr,
4082 	.setattr	= shmem_setattr,
4083 	.get_link	= simple_get_link,
4084 #ifdef CONFIG_TMPFS_XATTR
4085 	.listxattr	= shmem_listxattr,
4086 #endif
4087 };
4088 
4089 static const struct inode_operations shmem_symlink_inode_operations = {
4090 	.getattr	= shmem_getattr,
4091 	.setattr	= shmem_setattr,
4092 	.get_link	= shmem_get_link,
4093 #ifdef CONFIG_TMPFS_XATTR
4094 	.listxattr	= shmem_listxattr,
4095 #endif
4096 };
4097 
4098 static struct dentry *shmem_get_parent(struct dentry *child)
4099 {
4100 	return ERR_PTR(-ESTALE);
4101 }
4102 
4103 static int shmem_match(struct inode *ino, void *vfh)
4104 {
4105 	__u32 *fh = vfh;
4106 	__u64 inum = fh[2];
4107 	inum = (inum << 32) | fh[1];
4108 	return ino->i_ino == inum && fh[0] == ino->i_generation;
4109 }
4110 
4111 /* Find any alias of inode, but prefer a hashed alias */
4112 static struct dentry *shmem_find_alias(struct inode *inode)
4113 {
4114 	struct dentry *alias = d_find_alias(inode);
4115 
4116 	return alias ?: d_find_any_alias(inode);
4117 }
4118 
4119 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
4120 		struct fid *fid, int fh_len, int fh_type)
4121 {
4122 	struct inode *inode;
4123 	struct dentry *dentry = NULL;
4124 	u64 inum;
4125 
4126 	if (fh_len < 3)
4127 		return NULL;
4128 
4129 	inum = fid->raw[2];
4130 	inum = (inum << 32) | fid->raw[1];
4131 
4132 	inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
4133 			shmem_match, fid->raw);
4134 	if (inode) {
4135 		dentry = shmem_find_alias(inode);
4136 		iput(inode);
4137 	}
4138 
4139 	return dentry;
4140 }
4141 
4142 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
4143 				struct inode *parent)
4144 {
4145 	if (*len < 3) {
4146 		*len = 3;
4147 		return FILEID_INVALID;
4148 	}
4149 
4150 	if (inode_unhashed(inode)) {
4151 		/* Unfortunately insert_inode_hash is not idempotent,
4152 		 * so as we hash inodes here rather than at creation
4153 		 * time, we need a lock to ensure we only try
4154 		 * to do it once
4155 		 */
4156 		static DEFINE_SPINLOCK(lock);
4157 		spin_lock(&lock);
4158 		if (inode_unhashed(inode))
4159 			__insert_inode_hash(inode,
4160 					    inode->i_ino + inode->i_generation);
4161 		spin_unlock(&lock);
4162 	}
4163 
4164 	fh[0] = inode->i_generation;
4165 	fh[1] = inode->i_ino;
4166 	fh[2] = ((__u64)inode->i_ino) >> 32;
4167 
4168 	*len = 3;
4169 	return 1;
4170 }
4171 
4172 static const struct export_operations shmem_export_ops = {
4173 	.get_parent     = shmem_get_parent,
4174 	.encode_fh      = shmem_encode_fh,
4175 	.fh_to_dentry	= shmem_fh_to_dentry,
4176 };
4177 
4178 enum shmem_param {
4179 	Opt_gid,
4180 	Opt_huge,
4181 	Opt_mode,
4182 	Opt_mpol,
4183 	Opt_nr_blocks,
4184 	Opt_nr_inodes,
4185 	Opt_size,
4186 	Opt_uid,
4187 	Opt_inode32,
4188 	Opt_inode64,
4189 	Opt_noswap,
4190 	Opt_quota,
4191 	Opt_usrquota,
4192 	Opt_grpquota,
4193 	Opt_usrquota_block_hardlimit,
4194 	Opt_usrquota_inode_hardlimit,
4195 	Opt_grpquota_block_hardlimit,
4196 	Opt_grpquota_inode_hardlimit,
4197 };
4198 
4199 static const struct constant_table shmem_param_enums_huge[] = {
4200 	{"never",	SHMEM_HUGE_NEVER },
4201 	{"always",	SHMEM_HUGE_ALWAYS },
4202 	{"within_size",	SHMEM_HUGE_WITHIN_SIZE },
4203 	{"advise",	SHMEM_HUGE_ADVISE },
4204 	{}
4205 };
4206 
4207 const struct fs_parameter_spec shmem_fs_parameters[] = {
4208 	fsparam_gid   ("gid",		Opt_gid),
4209 	fsparam_enum  ("huge",		Opt_huge,  shmem_param_enums_huge),
4210 	fsparam_u32oct("mode",		Opt_mode),
4211 	fsparam_string("mpol",		Opt_mpol),
4212 	fsparam_string("nr_blocks",	Opt_nr_blocks),
4213 	fsparam_string("nr_inodes",	Opt_nr_inodes),
4214 	fsparam_string("size",		Opt_size),
4215 	fsparam_uid   ("uid",		Opt_uid),
4216 	fsparam_flag  ("inode32",	Opt_inode32),
4217 	fsparam_flag  ("inode64",	Opt_inode64),
4218 	fsparam_flag  ("noswap",	Opt_noswap),
4219 #ifdef CONFIG_TMPFS_QUOTA
4220 	fsparam_flag  ("quota",		Opt_quota),
4221 	fsparam_flag  ("usrquota",	Opt_usrquota),
4222 	fsparam_flag  ("grpquota",	Opt_grpquota),
4223 	fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit),
4224 	fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit),
4225 	fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit),
4226 	fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit),
4227 #endif
4228 	{}
4229 };
4230 
4231 static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
4232 {
4233 	struct shmem_options *ctx = fc->fs_private;
4234 	struct fs_parse_result result;
4235 	unsigned long long size;
4236 	char *rest;
4237 	int opt;
4238 	kuid_t kuid;
4239 	kgid_t kgid;
4240 
4241 	opt = fs_parse(fc, shmem_fs_parameters, param, &result);
4242 	if (opt < 0)
4243 		return opt;
4244 
4245 	switch (opt) {
4246 	case Opt_size:
4247 		size = memparse(param->string, &rest);
4248 		if (*rest == '%') {
4249 			size <<= PAGE_SHIFT;
4250 			size *= totalram_pages();
4251 			do_div(size, 100);
4252 			rest++;
4253 		}
4254 		if (*rest)
4255 			goto bad_value;
4256 		ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
4257 		ctx->seen |= SHMEM_SEEN_BLOCKS;
4258 		break;
4259 	case Opt_nr_blocks:
4260 		ctx->blocks = memparse(param->string, &rest);
4261 		if (*rest || ctx->blocks > LONG_MAX)
4262 			goto bad_value;
4263 		ctx->seen |= SHMEM_SEEN_BLOCKS;
4264 		break;
4265 	case Opt_nr_inodes:
4266 		ctx->inodes = memparse(param->string, &rest);
4267 		if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE)
4268 			goto bad_value;
4269 		ctx->seen |= SHMEM_SEEN_INODES;
4270 		break;
4271 	case Opt_mode:
4272 		ctx->mode = result.uint_32 & 07777;
4273 		break;
4274 	case Opt_uid:
4275 		kuid = result.uid;
4276 
4277 		/*
4278 		 * The requested uid must be representable in the
4279 		 * filesystem's idmapping.
4280 		 */
4281 		if (!kuid_has_mapping(fc->user_ns, kuid))
4282 			goto bad_value;
4283 
4284 		ctx->uid = kuid;
4285 		break;
4286 	case Opt_gid:
4287 		kgid = result.gid;
4288 
4289 		/*
4290 		 * The requested gid must be representable in the
4291 		 * filesystem's idmapping.
4292 		 */
4293 		if (!kgid_has_mapping(fc->user_ns, kgid))
4294 			goto bad_value;
4295 
4296 		ctx->gid = kgid;
4297 		break;
4298 	case Opt_huge:
4299 		ctx->huge = result.uint_32;
4300 		if (ctx->huge != SHMEM_HUGE_NEVER &&
4301 		    !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4302 		      has_transparent_hugepage()))
4303 			goto unsupported_parameter;
4304 		ctx->seen |= SHMEM_SEEN_HUGE;
4305 		break;
4306 	case Opt_mpol:
4307 		if (IS_ENABLED(CONFIG_NUMA)) {
4308 			mpol_put(ctx->mpol);
4309 			ctx->mpol = NULL;
4310 			if (mpol_parse_str(param->string, &ctx->mpol))
4311 				goto bad_value;
4312 			break;
4313 		}
4314 		goto unsupported_parameter;
4315 	case Opt_inode32:
4316 		ctx->full_inums = false;
4317 		ctx->seen |= SHMEM_SEEN_INUMS;
4318 		break;
4319 	case Opt_inode64:
4320 		if (sizeof(ino_t) < 8) {
4321 			return invalfc(fc,
4322 				       "Cannot use inode64 with <64bit inums in kernel\n");
4323 		}
4324 		ctx->full_inums = true;
4325 		ctx->seen |= SHMEM_SEEN_INUMS;
4326 		break;
4327 	case Opt_noswap:
4328 		if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) {
4329 			return invalfc(fc,
4330 				       "Turning off swap in unprivileged tmpfs mounts unsupported");
4331 		}
4332 		ctx->noswap = true;
4333 		ctx->seen |= SHMEM_SEEN_NOSWAP;
4334 		break;
4335 	case Opt_quota:
4336 		if (fc->user_ns != &init_user_ns)
4337 			return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4338 		ctx->seen |= SHMEM_SEEN_QUOTA;
4339 		ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP);
4340 		break;
4341 	case Opt_usrquota:
4342 		if (fc->user_ns != &init_user_ns)
4343 			return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4344 		ctx->seen |= SHMEM_SEEN_QUOTA;
4345 		ctx->quota_types |= QTYPE_MASK_USR;
4346 		break;
4347 	case Opt_grpquota:
4348 		if (fc->user_ns != &init_user_ns)
4349 			return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported");
4350 		ctx->seen |= SHMEM_SEEN_QUOTA;
4351 		ctx->quota_types |= QTYPE_MASK_GRP;
4352 		break;
4353 	case Opt_usrquota_block_hardlimit:
4354 		size = memparse(param->string, &rest);
4355 		if (*rest || !size)
4356 			goto bad_value;
4357 		if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4358 			return invalfc(fc,
4359 				       "User quota block hardlimit too large.");
4360 		ctx->qlimits.usrquota_bhardlimit = size;
4361 		break;
4362 	case Opt_grpquota_block_hardlimit:
4363 		size = memparse(param->string, &rest);
4364 		if (*rest || !size)
4365 			goto bad_value;
4366 		if (size > SHMEM_QUOTA_MAX_SPC_LIMIT)
4367 			return invalfc(fc,
4368 				       "Group quota block hardlimit too large.");
4369 		ctx->qlimits.grpquota_bhardlimit = size;
4370 		break;
4371 	case Opt_usrquota_inode_hardlimit:
4372 		size = memparse(param->string, &rest);
4373 		if (*rest || !size)
4374 			goto bad_value;
4375 		if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4376 			return invalfc(fc,
4377 				       "User quota inode hardlimit too large.");
4378 		ctx->qlimits.usrquota_ihardlimit = size;
4379 		break;
4380 	case Opt_grpquota_inode_hardlimit:
4381 		size = memparse(param->string, &rest);
4382 		if (*rest || !size)
4383 			goto bad_value;
4384 		if (size > SHMEM_QUOTA_MAX_INO_LIMIT)
4385 			return invalfc(fc,
4386 				       "Group quota inode hardlimit too large.");
4387 		ctx->qlimits.grpquota_ihardlimit = size;
4388 		break;
4389 	}
4390 	return 0;
4391 
4392 unsupported_parameter:
4393 	return invalfc(fc, "Unsupported parameter '%s'", param->key);
4394 bad_value:
4395 	return invalfc(fc, "Bad value for '%s'", param->key);
4396 }
4397 
4398 static int shmem_parse_options(struct fs_context *fc, void *data)
4399 {
4400 	char *options = data;
4401 
4402 	if (options) {
4403 		int err = security_sb_eat_lsm_opts(options, &fc->security);
4404 		if (err)
4405 			return err;
4406 	}
4407 
4408 	while (options != NULL) {
4409 		char *this_char = options;
4410 		for (;;) {
4411 			/*
4412 			 * NUL-terminate this option: unfortunately,
4413 			 * mount options form a comma-separated list,
4414 			 * but mpol's nodelist may also contain commas.
4415 			 */
4416 			options = strchr(options, ',');
4417 			if (options == NULL)
4418 				break;
4419 			options++;
4420 			if (!isdigit(*options)) {
4421 				options[-1] = '\0';
4422 				break;
4423 			}
4424 		}
4425 		if (*this_char) {
4426 			char *value = strchr(this_char, '=');
4427 			size_t len = 0;
4428 			int err;
4429 
4430 			if (value) {
4431 				*value++ = '\0';
4432 				len = strlen(value);
4433 			}
4434 			err = vfs_parse_fs_string(fc, this_char, value, len);
4435 			if (err < 0)
4436 				return err;
4437 		}
4438 	}
4439 	return 0;
4440 }
4441 
4442 /*
4443  * Reconfigure a shmem filesystem.
4444  */
4445 static int shmem_reconfigure(struct fs_context *fc)
4446 {
4447 	struct shmem_options *ctx = fc->fs_private;
4448 	struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
4449 	unsigned long used_isp;
4450 	struct mempolicy *mpol = NULL;
4451 	const char *err;
4452 
4453 	raw_spin_lock(&sbinfo->stat_lock);
4454 	used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace;
4455 
4456 	if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
4457 		if (!sbinfo->max_blocks) {
4458 			err = "Cannot retroactively limit size";
4459 			goto out;
4460 		}
4461 		if (percpu_counter_compare(&sbinfo->used_blocks,
4462 					   ctx->blocks) > 0) {
4463 			err = "Too small a size for current use";
4464 			goto out;
4465 		}
4466 	}
4467 	if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
4468 		if (!sbinfo->max_inodes) {
4469 			err = "Cannot retroactively limit inodes";
4470 			goto out;
4471 		}
4472 		if (ctx->inodes * BOGO_INODE_SIZE < used_isp) {
4473 			err = "Too few inodes for current use";
4474 			goto out;
4475 		}
4476 	}
4477 
4478 	if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
4479 	    sbinfo->next_ino > UINT_MAX) {
4480 		err = "Current inum too high to switch to 32-bit inums";
4481 		goto out;
4482 	}
4483 	if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) {
4484 		err = "Cannot disable swap on remount";
4485 		goto out;
4486 	}
4487 	if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) {
4488 		err = "Cannot enable swap on remount if it was disabled on first mount";
4489 		goto out;
4490 	}
4491 
4492 	if (ctx->seen & SHMEM_SEEN_QUOTA &&
4493 	    !sb_any_quota_loaded(fc->root->d_sb)) {
4494 		err = "Cannot enable quota on remount";
4495 		goto out;
4496 	}
4497 
4498 #ifdef CONFIG_TMPFS_QUOTA
4499 #define CHANGED_LIMIT(name)						\
4500 	(ctx->qlimits.name## hardlimit &&				\
4501 	(ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit))
4502 
4503 	if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) ||
4504 	    CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) {
4505 		err = "Cannot change global quota limit on remount";
4506 		goto out;
4507 	}
4508 #endif /* CONFIG_TMPFS_QUOTA */
4509 
4510 	if (ctx->seen & SHMEM_SEEN_HUGE)
4511 		sbinfo->huge = ctx->huge;
4512 	if (ctx->seen & SHMEM_SEEN_INUMS)
4513 		sbinfo->full_inums = ctx->full_inums;
4514 	if (ctx->seen & SHMEM_SEEN_BLOCKS)
4515 		sbinfo->max_blocks  = ctx->blocks;
4516 	if (ctx->seen & SHMEM_SEEN_INODES) {
4517 		sbinfo->max_inodes  = ctx->inodes;
4518 		sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp;
4519 	}
4520 
4521 	/*
4522 	 * Preserve previous mempolicy unless mpol remount option was specified.
4523 	 */
4524 	if (ctx->mpol) {
4525 		mpol = sbinfo->mpol;
4526 		sbinfo->mpol = ctx->mpol;	/* transfers initial ref */
4527 		ctx->mpol = NULL;
4528 	}
4529 
4530 	if (ctx->noswap)
4531 		sbinfo->noswap = true;
4532 
4533 	raw_spin_unlock(&sbinfo->stat_lock);
4534 	mpol_put(mpol);
4535 	return 0;
4536 out:
4537 	raw_spin_unlock(&sbinfo->stat_lock);
4538 	return invalfc(fc, "%s", err);
4539 }
4540 
4541 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
4542 {
4543 	struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
4544 	struct mempolicy *mpol;
4545 
4546 	if (sbinfo->max_blocks != shmem_default_max_blocks())
4547 		seq_printf(seq, ",size=%luk", K(sbinfo->max_blocks));
4548 	if (sbinfo->max_inodes != shmem_default_max_inodes())
4549 		seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
4550 	if (sbinfo->mode != (0777 | S_ISVTX))
4551 		seq_printf(seq, ",mode=%03ho", sbinfo->mode);
4552 	if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
4553 		seq_printf(seq, ",uid=%u",
4554 				from_kuid_munged(&init_user_ns, sbinfo->uid));
4555 	if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
4556 		seq_printf(seq, ",gid=%u",
4557 				from_kgid_munged(&init_user_ns, sbinfo->gid));
4558 
4559 	/*
4560 	 * Showing inode{64,32} might be useful even if it's the system default,
4561 	 * since then people don't have to resort to checking both here and
4562 	 * /proc/config.gz to confirm 64-bit inums were successfully applied
4563 	 * (which may not even exist if IKCONFIG_PROC isn't enabled).
4564 	 *
4565 	 * We hide it when inode64 isn't the default and we are using 32-bit
4566 	 * inodes, since that probably just means the feature isn't even under
4567 	 * consideration.
4568 	 *
4569 	 * As such:
4570 	 *
4571 	 *                     +-----------------+-----------------+
4572 	 *                     | TMPFS_INODE64=y | TMPFS_INODE64=n |
4573 	 *  +------------------+-----------------+-----------------+
4574 	 *  | full_inums=true  | show            | show            |
4575 	 *  | full_inums=false | show            | hide            |
4576 	 *  +------------------+-----------------+-----------------+
4577 	 *
4578 	 */
4579 	if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
4580 		seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
4581 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4582 	/* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
4583 	if (sbinfo->huge)
4584 		seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
4585 #endif
4586 	mpol = shmem_get_sbmpol(sbinfo);
4587 	shmem_show_mpol(seq, mpol);
4588 	mpol_put(mpol);
4589 	if (sbinfo->noswap)
4590 		seq_printf(seq, ",noswap");
4591 #ifdef CONFIG_TMPFS_QUOTA
4592 	if (sb_has_quota_active(root->d_sb, USRQUOTA))
4593 		seq_printf(seq, ",usrquota");
4594 	if (sb_has_quota_active(root->d_sb, GRPQUOTA))
4595 		seq_printf(seq, ",grpquota");
4596 	if (sbinfo->qlimits.usrquota_bhardlimit)
4597 		seq_printf(seq, ",usrquota_block_hardlimit=%lld",
4598 			   sbinfo->qlimits.usrquota_bhardlimit);
4599 	if (sbinfo->qlimits.grpquota_bhardlimit)
4600 		seq_printf(seq, ",grpquota_block_hardlimit=%lld",
4601 			   sbinfo->qlimits.grpquota_bhardlimit);
4602 	if (sbinfo->qlimits.usrquota_ihardlimit)
4603 		seq_printf(seq, ",usrquota_inode_hardlimit=%lld",
4604 			   sbinfo->qlimits.usrquota_ihardlimit);
4605 	if (sbinfo->qlimits.grpquota_ihardlimit)
4606 		seq_printf(seq, ",grpquota_inode_hardlimit=%lld",
4607 			   sbinfo->qlimits.grpquota_ihardlimit);
4608 #endif
4609 	return 0;
4610 }
4611 
4612 #endif /* CONFIG_TMPFS */
4613 
4614 static void shmem_put_super(struct super_block *sb)
4615 {
4616 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
4617 
4618 #ifdef CONFIG_TMPFS_QUOTA
4619 	shmem_disable_quotas(sb);
4620 #endif
4621 	free_percpu(sbinfo->ino_batch);
4622 	percpu_counter_destroy(&sbinfo->used_blocks);
4623 	mpol_put(sbinfo->mpol);
4624 	kfree(sbinfo);
4625 	sb->s_fs_info = NULL;
4626 }
4627 
4628 static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
4629 {
4630 	struct shmem_options *ctx = fc->fs_private;
4631 	struct inode *inode;
4632 	struct shmem_sb_info *sbinfo;
4633 	int error = -ENOMEM;
4634 
4635 	/* Round up to L1_CACHE_BYTES to resist false sharing */
4636 	sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
4637 				L1_CACHE_BYTES), GFP_KERNEL);
4638 	if (!sbinfo)
4639 		return error;
4640 
4641 	sb->s_fs_info = sbinfo;
4642 
4643 #ifdef CONFIG_TMPFS
4644 	/*
4645 	 * Per default we only allow half of the physical ram per
4646 	 * tmpfs instance, limiting inodes to one per page of lowmem;
4647 	 * but the internal instance is left unlimited.
4648 	 */
4649 	if (!(sb->s_flags & SB_KERNMOUNT)) {
4650 		if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
4651 			ctx->blocks = shmem_default_max_blocks();
4652 		if (!(ctx->seen & SHMEM_SEEN_INODES))
4653 			ctx->inodes = shmem_default_max_inodes();
4654 		if (!(ctx->seen & SHMEM_SEEN_INUMS))
4655 			ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
4656 		sbinfo->noswap = ctx->noswap;
4657 	} else {
4658 		sb->s_flags |= SB_NOUSER;
4659 	}
4660 	sb->s_export_op = &shmem_export_ops;
4661 	sb->s_flags |= SB_NOSEC | SB_I_VERSION;
4662 #else
4663 	sb->s_flags |= SB_NOUSER;
4664 #endif
4665 	sbinfo->max_blocks = ctx->blocks;
4666 	sbinfo->max_inodes = ctx->inodes;
4667 	sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE;
4668 	if (sb->s_flags & SB_KERNMOUNT) {
4669 		sbinfo->ino_batch = alloc_percpu(ino_t);
4670 		if (!sbinfo->ino_batch)
4671 			goto failed;
4672 	}
4673 	sbinfo->uid = ctx->uid;
4674 	sbinfo->gid = ctx->gid;
4675 	sbinfo->full_inums = ctx->full_inums;
4676 	sbinfo->mode = ctx->mode;
4677 	sbinfo->huge = ctx->huge;
4678 	sbinfo->mpol = ctx->mpol;
4679 	ctx->mpol = NULL;
4680 
4681 	raw_spin_lock_init(&sbinfo->stat_lock);
4682 	if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
4683 		goto failed;
4684 	spin_lock_init(&sbinfo->shrinklist_lock);
4685 	INIT_LIST_HEAD(&sbinfo->shrinklist);
4686 
4687 	sb->s_maxbytes = MAX_LFS_FILESIZE;
4688 	sb->s_blocksize = PAGE_SIZE;
4689 	sb->s_blocksize_bits = PAGE_SHIFT;
4690 	sb->s_magic = TMPFS_MAGIC;
4691 	sb->s_op = &shmem_ops;
4692 	sb->s_time_gran = 1;
4693 #ifdef CONFIG_TMPFS_XATTR
4694 	sb->s_xattr = shmem_xattr_handlers;
4695 #endif
4696 #ifdef CONFIG_TMPFS_POSIX_ACL
4697 	sb->s_flags |= SB_POSIXACL;
4698 #endif
4699 	uuid_t uuid;
4700 	uuid_gen(&uuid);
4701 	super_set_uuid(sb, uuid.b, sizeof(uuid));
4702 
4703 #ifdef CONFIG_TMPFS_QUOTA
4704 	if (ctx->seen & SHMEM_SEEN_QUOTA) {
4705 		sb->dq_op = &shmem_quota_operations;
4706 		sb->s_qcop = &dquot_quotactl_sysfile_ops;
4707 		sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
4708 
4709 		/* Copy the default limits from ctx into sbinfo */
4710 		memcpy(&sbinfo->qlimits, &ctx->qlimits,
4711 		       sizeof(struct shmem_quota_limits));
4712 
4713 		if (shmem_enable_quotas(sb, ctx->quota_types))
4714 			goto failed;
4715 	}
4716 #endif /* CONFIG_TMPFS_QUOTA */
4717 
4718 	inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL,
4719 				S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
4720 	if (IS_ERR(inode)) {
4721 		error = PTR_ERR(inode);
4722 		goto failed;
4723 	}
4724 	inode->i_uid = sbinfo->uid;
4725 	inode->i_gid = sbinfo->gid;
4726 	sb->s_root = d_make_root(inode);
4727 	if (!sb->s_root)
4728 		goto failed;
4729 	return 0;
4730 
4731 failed:
4732 	shmem_put_super(sb);
4733 	return error;
4734 }
4735 
4736 static int shmem_get_tree(struct fs_context *fc)
4737 {
4738 	return get_tree_nodev(fc, shmem_fill_super);
4739 }
4740 
4741 static void shmem_free_fc(struct fs_context *fc)
4742 {
4743 	struct shmem_options *ctx = fc->fs_private;
4744 
4745 	if (ctx) {
4746 		mpol_put(ctx->mpol);
4747 		kfree(ctx);
4748 	}
4749 }
4750 
4751 static const struct fs_context_operations shmem_fs_context_ops = {
4752 	.free			= shmem_free_fc,
4753 	.get_tree		= shmem_get_tree,
4754 #ifdef CONFIG_TMPFS
4755 	.parse_monolithic	= shmem_parse_options,
4756 	.parse_param		= shmem_parse_one,
4757 	.reconfigure		= shmem_reconfigure,
4758 #endif
4759 };
4760 
4761 static struct kmem_cache *shmem_inode_cachep __ro_after_init;
4762 
4763 static struct inode *shmem_alloc_inode(struct super_block *sb)
4764 {
4765 	struct shmem_inode_info *info;
4766 	info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL);
4767 	if (!info)
4768 		return NULL;
4769 	return &info->vfs_inode;
4770 }
4771 
4772 static void shmem_free_in_core_inode(struct inode *inode)
4773 {
4774 	if (S_ISLNK(inode->i_mode))
4775 		kfree(inode->i_link);
4776 	kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
4777 }
4778 
4779 static void shmem_destroy_inode(struct inode *inode)
4780 {
4781 	if (S_ISREG(inode->i_mode))
4782 		mpol_free_shared_policy(&SHMEM_I(inode)->policy);
4783 	if (S_ISDIR(inode->i_mode))
4784 		simple_offset_destroy(shmem_get_offset_ctx(inode));
4785 }
4786 
4787 static void shmem_init_inode(void *foo)
4788 {
4789 	struct shmem_inode_info *info = foo;
4790 	inode_init_once(&info->vfs_inode);
4791 }
4792 
4793 static void __init shmem_init_inodecache(void)
4794 {
4795 	shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
4796 				sizeof(struct shmem_inode_info),
4797 				0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
4798 }
4799 
4800 static void __init shmem_destroy_inodecache(void)
4801 {
4802 	kmem_cache_destroy(shmem_inode_cachep);
4803 }
4804 
4805 /* Keep the page in page cache instead of truncating it */
4806 static int shmem_error_remove_folio(struct address_space *mapping,
4807 				   struct folio *folio)
4808 {
4809 	return 0;
4810 }
4811 
4812 static const struct address_space_operations shmem_aops = {
4813 	.writepage	= shmem_writepage,
4814 	.dirty_folio	= noop_dirty_folio,
4815 #ifdef CONFIG_TMPFS
4816 	.write_begin	= shmem_write_begin,
4817 	.write_end	= shmem_write_end,
4818 #endif
4819 #ifdef CONFIG_MIGRATION
4820 	.migrate_folio	= migrate_folio,
4821 #endif
4822 	.error_remove_folio = shmem_error_remove_folio,
4823 };
4824 
4825 static const struct file_operations shmem_file_operations = {
4826 	.mmap		= shmem_mmap,
4827 	.open		= shmem_file_open,
4828 	.get_unmapped_area = shmem_get_unmapped_area,
4829 #ifdef CONFIG_TMPFS
4830 	.llseek		= shmem_file_llseek,
4831 	.read_iter	= shmem_file_read_iter,
4832 	.write_iter	= shmem_file_write_iter,
4833 	.fsync		= noop_fsync,
4834 	.splice_read	= shmem_file_splice_read,
4835 	.splice_write	= iter_file_splice_write,
4836 	.fallocate	= shmem_fallocate,
4837 #endif
4838 };
4839 
4840 static const struct inode_operations shmem_inode_operations = {
4841 	.getattr	= shmem_getattr,
4842 	.setattr	= shmem_setattr,
4843 #ifdef CONFIG_TMPFS_XATTR
4844 	.listxattr	= shmem_listxattr,
4845 	.set_acl	= simple_set_acl,
4846 	.fileattr_get	= shmem_fileattr_get,
4847 	.fileattr_set	= shmem_fileattr_set,
4848 #endif
4849 };
4850 
4851 static const struct inode_operations shmem_dir_inode_operations = {
4852 #ifdef CONFIG_TMPFS
4853 	.getattr	= shmem_getattr,
4854 	.create		= shmem_create,
4855 	.lookup		= simple_lookup,
4856 	.link		= shmem_link,
4857 	.unlink		= shmem_unlink,
4858 	.symlink	= shmem_symlink,
4859 	.mkdir		= shmem_mkdir,
4860 	.rmdir		= shmem_rmdir,
4861 	.mknod		= shmem_mknod,
4862 	.rename		= shmem_rename2,
4863 	.tmpfile	= shmem_tmpfile,
4864 	.get_offset_ctx	= shmem_get_offset_ctx,
4865 #endif
4866 #ifdef CONFIG_TMPFS_XATTR
4867 	.listxattr	= shmem_listxattr,
4868 	.fileattr_get	= shmem_fileattr_get,
4869 	.fileattr_set	= shmem_fileattr_set,
4870 #endif
4871 #ifdef CONFIG_TMPFS_POSIX_ACL
4872 	.setattr	= shmem_setattr,
4873 	.set_acl	= simple_set_acl,
4874 #endif
4875 };
4876 
4877 static const struct inode_operations shmem_special_inode_operations = {
4878 	.getattr	= shmem_getattr,
4879 #ifdef CONFIG_TMPFS_XATTR
4880 	.listxattr	= shmem_listxattr,
4881 #endif
4882 #ifdef CONFIG_TMPFS_POSIX_ACL
4883 	.setattr	= shmem_setattr,
4884 	.set_acl	= simple_set_acl,
4885 #endif
4886 };
4887 
4888 static const struct super_operations shmem_ops = {
4889 	.alloc_inode	= shmem_alloc_inode,
4890 	.free_inode	= shmem_free_in_core_inode,
4891 	.destroy_inode	= shmem_destroy_inode,
4892 #ifdef CONFIG_TMPFS
4893 	.statfs		= shmem_statfs,
4894 	.show_options	= shmem_show_options,
4895 #endif
4896 #ifdef CONFIG_TMPFS_QUOTA
4897 	.get_dquots	= shmem_get_dquots,
4898 #endif
4899 	.evict_inode	= shmem_evict_inode,
4900 	.drop_inode	= generic_delete_inode,
4901 	.put_super	= shmem_put_super,
4902 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4903 	.nr_cached_objects	= shmem_unused_huge_count,
4904 	.free_cached_objects	= shmem_unused_huge_scan,
4905 #endif
4906 };
4907 
4908 static const struct vm_operations_struct shmem_vm_ops = {
4909 	.fault		= shmem_fault,
4910 	.map_pages	= filemap_map_pages,
4911 #ifdef CONFIG_NUMA
4912 	.set_policy     = shmem_set_policy,
4913 	.get_policy     = shmem_get_policy,
4914 #endif
4915 };
4916 
4917 static const struct vm_operations_struct shmem_anon_vm_ops = {
4918 	.fault		= shmem_fault,
4919 	.map_pages	= filemap_map_pages,
4920 #ifdef CONFIG_NUMA
4921 	.set_policy     = shmem_set_policy,
4922 	.get_policy     = shmem_get_policy,
4923 #endif
4924 };
4925 
4926 int shmem_init_fs_context(struct fs_context *fc)
4927 {
4928 	struct shmem_options *ctx;
4929 
4930 	ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
4931 	if (!ctx)
4932 		return -ENOMEM;
4933 
4934 	ctx->mode = 0777 | S_ISVTX;
4935 	ctx->uid = current_fsuid();
4936 	ctx->gid = current_fsgid();
4937 
4938 	fc->fs_private = ctx;
4939 	fc->ops = &shmem_fs_context_ops;
4940 	return 0;
4941 }
4942 
4943 static struct file_system_type shmem_fs_type = {
4944 	.owner		= THIS_MODULE,
4945 	.name		= "tmpfs",
4946 	.init_fs_context = shmem_init_fs_context,
4947 #ifdef CONFIG_TMPFS
4948 	.parameters	= shmem_fs_parameters,
4949 #endif
4950 	.kill_sb	= kill_litter_super,
4951 	.fs_flags	= FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
4952 };
4953 
4954 void __init shmem_init(void)
4955 {
4956 	int error;
4957 
4958 	shmem_init_inodecache();
4959 
4960 #ifdef CONFIG_TMPFS_QUOTA
4961 	register_quota_format(&shmem_quota_format);
4962 #endif
4963 
4964 	error = register_filesystem(&shmem_fs_type);
4965 	if (error) {
4966 		pr_err("Could not register tmpfs\n");
4967 		goto out2;
4968 	}
4969 
4970 	shm_mnt = kern_mount(&shmem_fs_type);
4971 	if (IS_ERR(shm_mnt)) {
4972 		error = PTR_ERR(shm_mnt);
4973 		pr_err("Could not kern_mount tmpfs\n");
4974 		goto out1;
4975 	}
4976 
4977 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4978 	if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4979 		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4980 	else
4981 		shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */
4982 
4983 	/*
4984 	 * Default to setting PMD-sized THP to inherit the global setting and
4985 	 * disable all other multi-size THPs.
4986 	 */
4987 	huge_shmem_orders_inherit = BIT(HPAGE_PMD_ORDER);
4988 #endif
4989 	return;
4990 
4991 out1:
4992 	unregister_filesystem(&shmem_fs_type);
4993 out2:
4994 #ifdef CONFIG_TMPFS_QUOTA
4995 	unregister_quota_format(&shmem_quota_format);
4996 #endif
4997 	shmem_destroy_inodecache();
4998 	shm_mnt = ERR_PTR(error);
4999 }
5000 
5001 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
5002 static ssize_t shmem_enabled_show(struct kobject *kobj,
5003 				  struct kobj_attribute *attr, char *buf)
5004 {
5005 	static const int values[] = {
5006 		SHMEM_HUGE_ALWAYS,
5007 		SHMEM_HUGE_WITHIN_SIZE,
5008 		SHMEM_HUGE_ADVISE,
5009 		SHMEM_HUGE_NEVER,
5010 		SHMEM_HUGE_DENY,
5011 		SHMEM_HUGE_FORCE,
5012 	};
5013 	int len = 0;
5014 	int i;
5015 
5016 	for (i = 0; i < ARRAY_SIZE(values); i++) {
5017 		len += sysfs_emit_at(buf, len,
5018 				shmem_huge == values[i] ? "%s[%s]" : "%s%s",
5019 				i ? " " : "", shmem_format_huge(values[i]));
5020 	}
5021 	len += sysfs_emit_at(buf, len, "\n");
5022 
5023 	return len;
5024 }
5025 
5026 static ssize_t shmem_enabled_store(struct kobject *kobj,
5027 		struct kobj_attribute *attr, const char *buf, size_t count)
5028 {
5029 	char tmp[16];
5030 	int huge;
5031 
5032 	if (count + 1 > sizeof(tmp))
5033 		return -EINVAL;
5034 	memcpy(tmp, buf, count);
5035 	tmp[count] = '\0';
5036 	if (count && tmp[count - 1] == '\n')
5037 		tmp[count - 1] = '\0';
5038 
5039 	huge = shmem_parse_huge(tmp);
5040 	if (huge == -EINVAL)
5041 		return -EINVAL;
5042 	if (!has_transparent_hugepage() &&
5043 			huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
5044 		return -EINVAL;
5045 
5046 	/* Do not override huge allocation policy with non-PMD sized mTHP */
5047 	if (huge == SHMEM_HUGE_FORCE &&
5048 	    huge_shmem_orders_inherit != BIT(HPAGE_PMD_ORDER))
5049 		return -EINVAL;
5050 
5051 	shmem_huge = huge;
5052 	if (shmem_huge > SHMEM_HUGE_DENY)
5053 		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
5054 	return count;
5055 }
5056 
5057 struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled);
5058 static DEFINE_SPINLOCK(huge_shmem_orders_lock);
5059 
5060 static ssize_t thpsize_shmem_enabled_show(struct kobject *kobj,
5061 					  struct kobj_attribute *attr, char *buf)
5062 {
5063 	int order = to_thpsize(kobj)->order;
5064 	const char *output;
5065 
5066 	if (test_bit(order, &huge_shmem_orders_always))
5067 		output = "[always] inherit within_size advise never";
5068 	else if (test_bit(order, &huge_shmem_orders_inherit))
5069 		output = "always [inherit] within_size advise never";
5070 	else if (test_bit(order, &huge_shmem_orders_within_size))
5071 		output = "always inherit [within_size] advise never";
5072 	else if (test_bit(order, &huge_shmem_orders_madvise))
5073 		output = "always inherit within_size [advise] never";
5074 	else
5075 		output = "always inherit within_size advise [never]";
5076 
5077 	return sysfs_emit(buf, "%s\n", output);
5078 }
5079 
5080 static ssize_t thpsize_shmem_enabled_store(struct kobject *kobj,
5081 					   struct kobj_attribute *attr,
5082 					   const char *buf, size_t count)
5083 {
5084 	int order = to_thpsize(kobj)->order;
5085 	ssize_t ret = count;
5086 
5087 	if (sysfs_streq(buf, "always")) {
5088 		spin_lock(&huge_shmem_orders_lock);
5089 		clear_bit(order, &huge_shmem_orders_inherit);
5090 		clear_bit(order, &huge_shmem_orders_madvise);
5091 		clear_bit(order, &huge_shmem_orders_within_size);
5092 		set_bit(order, &huge_shmem_orders_always);
5093 		spin_unlock(&huge_shmem_orders_lock);
5094 	} else if (sysfs_streq(buf, "inherit")) {
5095 		/* Do not override huge allocation policy with non-PMD sized mTHP */
5096 		if (shmem_huge == SHMEM_HUGE_FORCE &&
5097 		    order != HPAGE_PMD_ORDER)
5098 			return -EINVAL;
5099 
5100 		spin_lock(&huge_shmem_orders_lock);
5101 		clear_bit(order, &huge_shmem_orders_always);
5102 		clear_bit(order, &huge_shmem_orders_madvise);
5103 		clear_bit(order, &huge_shmem_orders_within_size);
5104 		set_bit(order, &huge_shmem_orders_inherit);
5105 		spin_unlock(&huge_shmem_orders_lock);
5106 	} else if (sysfs_streq(buf, "within_size")) {
5107 		spin_lock(&huge_shmem_orders_lock);
5108 		clear_bit(order, &huge_shmem_orders_always);
5109 		clear_bit(order, &huge_shmem_orders_inherit);
5110 		clear_bit(order, &huge_shmem_orders_madvise);
5111 		set_bit(order, &huge_shmem_orders_within_size);
5112 		spin_unlock(&huge_shmem_orders_lock);
5113 	} else if (sysfs_streq(buf, "advise")) {
5114 		spin_lock(&huge_shmem_orders_lock);
5115 		clear_bit(order, &huge_shmem_orders_always);
5116 		clear_bit(order, &huge_shmem_orders_inherit);
5117 		clear_bit(order, &huge_shmem_orders_within_size);
5118 		set_bit(order, &huge_shmem_orders_madvise);
5119 		spin_unlock(&huge_shmem_orders_lock);
5120 	} else if (sysfs_streq(buf, "never")) {
5121 		spin_lock(&huge_shmem_orders_lock);
5122 		clear_bit(order, &huge_shmem_orders_always);
5123 		clear_bit(order, &huge_shmem_orders_inherit);
5124 		clear_bit(order, &huge_shmem_orders_within_size);
5125 		clear_bit(order, &huge_shmem_orders_madvise);
5126 		spin_unlock(&huge_shmem_orders_lock);
5127 	} else {
5128 		ret = -EINVAL;
5129 	}
5130 
5131 	return ret;
5132 }
5133 
5134 struct kobj_attribute thpsize_shmem_enabled_attr =
5135 	__ATTR(shmem_enabled, 0644, thpsize_shmem_enabled_show, thpsize_shmem_enabled_store);
5136 #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
5137 
5138 #else /* !CONFIG_SHMEM */
5139 
5140 /*
5141  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
5142  *
5143  * This is intended for small system where the benefits of the full
5144  * shmem code (swap-backed and resource-limited) are outweighed by
5145  * their complexity. On systems without swap this code should be
5146  * effectively equivalent, but much lighter weight.
5147  */
5148 
5149 static struct file_system_type shmem_fs_type = {
5150 	.name		= "tmpfs",
5151 	.init_fs_context = ramfs_init_fs_context,
5152 	.parameters	= ramfs_fs_parameters,
5153 	.kill_sb	= ramfs_kill_sb,
5154 	.fs_flags	= FS_USERNS_MOUNT,
5155 };
5156 
5157 void __init shmem_init(void)
5158 {
5159 	BUG_ON(register_filesystem(&shmem_fs_type) != 0);
5160 
5161 	shm_mnt = kern_mount(&shmem_fs_type);
5162 	BUG_ON(IS_ERR(shm_mnt));
5163 }
5164 
5165 int shmem_unuse(unsigned int type)
5166 {
5167 	return 0;
5168 }
5169 
5170 int shmem_lock(struct file *file, int lock, struct ucounts *ucounts)
5171 {
5172 	return 0;
5173 }
5174 
5175 void shmem_unlock_mapping(struct address_space *mapping)
5176 {
5177 }
5178 
5179 #ifdef CONFIG_MMU
5180 unsigned long shmem_get_unmapped_area(struct file *file,
5181 				      unsigned long addr, unsigned long len,
5182 				      unsigned long pgoff, unsigned long flags)
5183 {
5184 	return mm_get_unmapped_area(current->mm, file, addr, len, pgoff, flags);
5185 }
5186 #endif
5187 
5188 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
5189 {
5190 	truncate_inode_pages_range(inode->i_mapping, lstart, lend);
5191 }
5192 EXPORT_SYMBOL_GPL(shmem_truncate_range);
5193 
5194 #define shmem_vm_ops				generic_file_vm_ops
5195 #define shmem_anon_vm_ops			generic_file_vm_ops
5196 #define shmem_file_operations			ramfs_file_operations
5197 #define shmem_acct_size(flags, size)		0
5198 #define shmem_unacct_size(flags, size)		do {} while (0)
5199 
5200 static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap,
5201 				struct super_block *sb, struct inode *dir,
5202 				umode_t mode, dev_t dev, unsigned long flags)
5203 {
5204 	struct inode *inode = ramfs_get_inode(sb, dir, mode, dev);
5205 	return inode ? inode : ERR_PTR(-ENOSPC);
5206 }
5207 
5208 #endif /* CONFIG_SHMEM */
5209 
5210 /* common code */
5211 
5212 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name,
5213 			loff_t size, unsigned long flags, unsigned int i_flags)
5214 {
5215 	struct inode *inode;
5216 	struct file *res;
5217 
5218 	if (IS_ERR(mnt))
5219 		return ERR_CAST(mnt);
5220 
5221 	if (size < 0 || size > MAX_LFS_FILESIZE)
5222 		return ERR_PTR(-EINVAL);
5223 
5224 	if (shmem_acct_size(flags, size))
5225 		return ERR_PTR(-ENOMEM);
5226 
5227 	if (is_idmapped_mnt(mnt))
5228 		return ERR_PTR(-EINVAL);
5229 
5230 	inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL,
5231 				S_IFREG | S_IRWXUGO, 0, flags);
5232 	if (IS_ERR(inode)) {
5233 		shmem_unacct_size(flags, size);
5234 		return ERR_CAST(inode);
5235 	}
5236 	inode->i_flags |= i_flags;
5237 	inode->i_size = size;
5238 	clear_nlink(inode);	/* It is unlinked */
5239 	res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
5240 	if (!IS_ERR(res))
5241 		res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
5242 				&shmem_file_operations);
5243 	if (IS_ERR(res))
5244 		iput(inode);
5245 	return res;
5246 }
5247 
5248 /**
5249  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
5250  * 	kernel internal.  There will be NO LSM permission checks against the
5251  * 	underlying inode.  So users of this interface must do LSM checks at a
5252  *	higher layer.  The users are the big_key and shm implementations.  LSM
5253  *	checks are provided at the key or shm level rather than the inode.
5254  * @name: name for dentry (to be seen in /proc/<pid>/maps
5255  * @size: size to be set for the file
5256  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5257  */
5258 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
5259 {
5260 	return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
5261 }
5262 EXPORT_SYMBOL_GPL(shmem_kernel_file_setup);
5263 
5264 /**
5265  * shmem_file_setup - get an unlinked file living in tmpfs
5266  * @name: name for dentry (to be seen in /proc/<pid>/maps
5267  * @size: size to be set for the file
5268  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5269  */
5270 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
5271 {
5272 	return __shmem_file_setup(shm_mnt, name, size, flags, 0);
5273 }
5274 EXPORT_SYMBOL_GPL(shmem_file_setup);
5275 
5276 /**
5277  * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
5278  * @mnt: the tmpfs mount where the file will be created
5279  * @name: name for dentry (to be seen in /proc/<pid>/maps
5280  * @size: size to be set for the file
5281  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
5282  */
5283 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
5284 				       loff_t size, unsigned long flags)
5285 {
5286 	return __shmem_file_setup(mnt, name, size, flags, 0);
5287 }
5288 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
5289 
5290 /**
5291  * shmem_zero_setup - setup a shared anonymous mapping
5292  * @vma: the vma to be mmapped is prepared by do_mmap
5293  */
5294 int shmem_zero_setup(struct vm_area_struct *vma)
5295 {
5296 	struct file *file;
5297 	loff_t size = vma->vm_end - vma->vm_start;
5298 
5299 	/*
5300 	 * Cloning a new file under mmap_lock leads to a lock ordering conflict
5301 	 * between XFS directory reading and selinux: since this file is only
5302 	 * accessible to the user through its mapping, use S_PRIVATE flag to
5303 	 * bypass file security, in the same way as shmem_kernel_file_setup().
5304 	 */
5305 	file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
5306 	if (IS_ERR(file))
5307 		return PTR_ERR(file);
5308 
5309 	if (vma->vm_file)
5310 		fput(vma->vm_file);
5311 	vma->vm_file = file;
5312 	vma->vm_ops = &shmem_anon_vm_ops;
5313 
5314 	return 0;
5315 }
5316 
5317 /**
5318  * shmem_read_folio_gfp - read into page cache, using specified page allocation flags.
5319  * @mapping:	the folio's address_space
5320  * @index:	the folio index
5321  * @gfp:	the page allocator flags to use if allocating
5322  *
5323  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
5324  * with any new page allocations done using the specified allocation flags.
5325  * But read_cache_page_gfp() uses the ->read_folio() method: which does not
5326  * suit tmpfs, since it may have pages in swapcache, and needs to find those
5327  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
5328  *
5329  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
5330  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
5331  */
5332 struct folio *shmem_read_folio_gfp(struct address_space *mapping,
5333 		pgoff_t index, gfp_t gfp)
5334 {
5335 #ifdef CONFIG_SHMEM
5336 	struct inode *inode = mapping->host;
5337 	struct folio *folio;
5338 	int error;
5339 
5340 	error = shmem_get_folio_gfp(inode, index, 0, &folio, SGP_CACHE,
5341 				    gfp, NULL, NULL);
5342 	if (error)
5343 		return ERR_PTR(error);
5344 
5345 	folio_unlock(folio);
5346 	return folio;
5347 #else
5348 	/*
5349 	 * The tiny !SHMEM case uses ramfs without swap
5350 	 */
5351 	return mapping_read_folio_gfp(mapping, index, gfp);
5352 #endif
5353 }
5354 EXPORT_SYMBOL_GPL(shmem_read_folio_gfp);
5355 
5356 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
5357 					 pgoff_t index, gfp_t gfp)
5358 {
5359 	struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp);
5360 	struct page *page;
5361 
5362 	if (IS_ERR(folio))
5363 		return &folio->page;
5364 
5365 	page = folio_file_page(folio, index);
5366 	if (PageHWPoison(page)) {
5367 		folio_put(folio);
5368 		return ERR_PTR(-EIO);
5369 	}
5370 
5371 	return page;
5372 }
5373 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
5374