~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/shmem.c

Version: ~ [ linux-5.16 ] ~ [ linux-5.15.13 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.90 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.170 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.224 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.261 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.296 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.298 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Resizable virtual memory filesystem for Linux.
  3  *
  4  * Copyright (C) 2000 Linus Torvalds.
  5  *               2000 Transmeta Corp.
  6  *               2000-2001 Christoph Rohland
  7  *               2000-2001 SAP AG
  8  *               2002 Red Hat Inc.
  9  * Copyright (C) 2002-2011 Hugh Dickins.
 10  * Copyright (C) 2011 Google Inc.
 11  * Copyright (C) 2002-2005 VERITAS Software Corporation.
 12  * Copyright (C) 2004 Andi Kleen, SuSE Labs
 13  *
 14  * Extended attribute support for tmpfs:
 15  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
 16  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
 17  *
 18  * tiny-shmem:
 19  * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
 20  *
 21  * This file is released under the GPL.
 22  */
 23 
 24 #include <linux/fs.h>
 25 #include <linux/init.h>
 26 #include <linux/vfs.h>
 27 #include <linux/mount.h>
 28 #include <linux/ramfs.h>
 29 #include <linux/pagemap.h>
 30 #include <linux/file.h>
 31 #include <linux/mm.h>
 32 #include <linux/sched/signal.h>
 33 #include <linux/export.h>
 34 #include <linux/swap.h>
 35 #include <linux/uio.h>
 36 #include <linux/khugepaged.h>
 37 #include <linux/hugetlb.h>
 38 
 39 #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
 40 
 41 static struct vfsmount *shm_mnt;
 42 
 43 #ifdef CONFIG_SHMEM
 44 /*
 45  * This virtual memory filesystem is heavily based on the ramfs. It
 46  * extends ramfs by the ability to use swap and honor resource limits
 47  * which makes it a completely usable filesystem.
 48  */
 49 
 50 #include <linux/xattr.h>
 51 #include <linux/exportfs.h>
 52 #include <linux/posix_acl.h>
 53 #include <linux/posix_acl_xattr.h>
 54 #include <linux/mman.h>
 55 #include <linux/string.h>
 56 #include <linux/slab.h>
 57 #include <linux/backing-dev.h>
 58 #include <linux/shmem_fs.h>
 59 #include <linux/writeback.h>
 60 #include <linux/blkdev.h>
 61 #include <linux/pagevec.h>
 62 #include <linux/percpu_counter.h>
 63 #include <linux/falloc.h>
 64 #include <linux/splice.h>
 65 #include <linux/security.h>
 66 #include <linux/swapops.h>
 67 #include <linux/mempolicy.h>
 68 #include <linux/namei.h>
 69 #include <linux/ctype.h>
 70 #include <linux/migrate.h>
 71 #include <linux/highmem.h>
 72 #include <linux/seq_file.h>
 73 #include <linux/magic.h>
 74 #include <linux/syscalls.h>
 75 #include <linux/fcntl.h>
 76 #include <uapi/linux/memfd.h>
 77 #include <linux/userfaultfd_k.h>
 78 #include <linux/rmap.h>
 79 #include <linux/uuid.h>
 80 
 81 #include <linux/uaccess.h>
 82 #include <asm/pgtable.h>
 83 
 84 #include "internal.h"
 85 
 86 #define BLOCKS_PER_PAGE  (PAGE_SIZE/512)
 87 #define VM_ACCT(size)    (PAGE_ALIGN(size) >> PAGE_SHIFT)
 88 
 89 /* Pretend that each entry is of this size in directory's i_size */
 90 #define BOGO_DIRENT_SIZE 20
 91 
 92 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
 93 #define SHORT_SYMLINK_LEN 128
 94 
 95 /*
 96  * shmem_fallocate communicates with shmem_fault or shmem_writepage via
 97  * inode->i_private (with i_mutex making sure that it has only one user at
 98  * a time): we would prefer not to enlarge the shmem inode just for that.
 99  */
100 struct shmem_falloc {
101         wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
102         pgoff_t start;          /* start of range currently being fallocated */
103         pgoff_t next;           /* the next page offset to be fallocated */
104         pgoff_t nr_falloced;    /* how many new pages have been fallocated */
105         pgoff_t nr_unswapped;   /* how often writepage refused to swap out */
106 };
107 
108 #ifdef CONFIG_TMPFS
109 static unsigned long shmem_default_max_blocks(void)
110 {
111         return totalram_pages / 2;
112 }
113 
114 static unsigned long shmem_default_max_inodes(void)
115 {
116         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
117 }
118 #endif
119 
120 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122                                 struct shmem_inode_info *info, pgoff_t index);
123 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
124                 struct page **pagep, enum sgp_type sgp,
125                 gfp_t gfp, struct vm_area_struct *vma,
126                 struct vm_fault *vmf, int *fault_type);
127 
128 int shmem_getpage(struct inode *inode, pgoff_t index,
129                 struct page **pagep, enum sgp_type sgp)
130 {
131         return shmem_getpage_gfp(inode, index, pagep, sgp,
132                 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
133 }
134 
135 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
136 {
137         return sb->s_fs_info;
138 }
139 
140 /*
141  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
142  * for shared memory and for shared anonymous (/dev/zero) mappings
143  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
144  * consistent with the pre-accounting of private mappings ...
145  */
146 static inline int shmem_acct_size(unsigned long flags, loff_t size)
147 {
148         return (flags & VM_NORESERVE) ?
149                 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
150 }
151 
152 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
153 {
154         if (!(flags & VM_NORESERVE))
155                 vm_unacct_memory(VM_ACCT(size));
156 }
157 
158 static inline int shmem_reacct_size(unsigned long flags,
159                 loff_t oldsize, loff_t newsize)
160 {
161         if (!(flags & VM_NORESERVE)) {
162                 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
163                         return security_vm_enough_memory_mm(current->mm,
164                                         VM_ACCT(newsize) - VM_ACCT(oldsize));
165                 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
166                         vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
167         }
168         return 0;
169 }
170 
171 /*
172  * ... whereas tmpfs objects are accounted incrementally as
173  * pages are allocated, in order to allow large sparse files.
174  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
175  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
176  */
177 static inline int shmem_acct_block(unsigned long flags, long pages)
178 {
179         if (!(flags & VM_NORESERVE))
180                 return 0;
181 
182         return security_vm_enough_memory_mm(current->mm,
183                         pages * VM_ACCT(PAGE_SIZE));
184 }
185 
186 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
187 {
188         if (flags & VM_NORESERVE)
189                 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
190 }
191 
192 static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
193 {
194         struct shmem_inode_info *info = SHMEM_I(inode);
195         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
196 
197         if (shmem_acct_block(info->flags, pages))
198                 return false;
199 
200         if (sbinfo->max_blocks) {
201                 if (percpu_counter_compare(&sbinfo->used_blocks,
202                                            sbinfo->max_blocks - pages) > 0)
203                         goto unacct;
204                 percpu_counter_add(&sbinfo->used_blocks, pages);
205         }
206 
207         return true;
208 
209 unacct:
210         shmem_unacct_blocks(info->flags, pages);
211         return false;
212 }
213 
214 static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
215 {
216         struct shmem_inode_info *info = SHMEM_I(inode);
217         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
218 
219         if (sbinfo->max_blocks)
220                 percpu_counter_sub(&sbinfo->used_blocks, pages);
221         shmem_unacct_blocks(info->flags, pages);
222 }
223 
224 static const struct super_operations shmem_ops;
225 static const struct address_space_operations shmem_aops;
226 static const struct file_operations shmem_file_operations;
227 static const struct inode_operations shmem_inode_operations;
228 static const struct inode_operations shmem_dir_inode_operations;
229 static const struct inode_operations shmem_special_inode_operations;
230 static const struct vm_operations_struct shmem_vm_ops;
231 static struct file_system_type shmem_fs_type;
232 
233 bool vma_is_shmem(struct vm_area_struct *vma)
234 {
235         return vma->vm_ops == &shmem_vm_ops;
236 }
237 
238 static LIST_HEAD(shmem_swaplist);
239 static DEFINE_MUTEX(shmem_swaplist_mutex);
240 
241 static int shmem_reserve_inode(struct super_block *sb)
242 {
243         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
244         if (sbinfo->max_inodes) {
245                 spin_lock(&sbinfo->stat_lock);
246                 if (!sbinfo->free_inodes) {
247                         spin_unlock(&sbinfo->stat_lock);
248                         return -ENOSPC;
249                 }
250                 sbinfo->free_inodes--;
251                 spin_unlock(&sbinfo->stat_lock);
252         }
253         return 0;
254 }
255 
256 static void shmem_free_inode(struct super_block *sb)
257 {
258         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
259         if (sbinfo->max_inodes) {
260                 spin_lock(&sbinfo->stat_lock);
261                 sbinfo->free_inodes++;
262                 spin_unlock(&sbinfo->stat_lock);
263         }
264 }
265 
266 /**
267  * shmem_recalc_inode - recalculate the block usage of an inode
268  * @inode: inode to recalc
269  *
270  * We have to calculate the free blocks since the mm can drop
271  * undirtied hole pages behind our back.
272  *
273  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
274  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
275  *
276  * It has to be called with the spinlock held.
277  */
278 static void shmem_recalc_inode(struct inode *inode)
279 {
280         struct shmem_inode_info *info = SHMEM_I(inode);
281         long freed;
282 
283         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
284         if (freed > 0) {
285                 info->alloced -= freed;
286                 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
287                 shmem_inode_unacct_blocks(inode, freed);
288         }
289 }
290 
291 bool shmem_charge(struct inode *inode, long pages)
292 {
293         struct shmem_inode_info *info = SHMEM_I(inode);
294         unsigned long flags;
295 
296         if (!shmem_inode_acct_block(inode, pages))
297                 return false;
298 
299         spin_lock_irqsave(&info->lock, flags);
300         info->alloced += pages;
301         inode->i_blocks += pages * BLOCKS_PER_PAGE;
302         shmem_recalc_inode(inode);
303         spin_unlock_irqrestore(&info->lock, flags);
304         inode->i_mapping->nrpages += pages;
305 
306         return true;
307 }
308 
309 void shmem_uncharge(struct inode *inode, long pages)
310 {
311         struct shmem_inode_info *info = SHMEM_I(inode);
312         unsigned long flags;
313 
314         spin_lock_irqsave(&info->lock, flags);
315         info->alloced -= pages;
316         inode->i_blocks -= pages * BLOCKS_PER_PAGE;
317         shmem_recalc_inode(inode);
318         spin_unlock_irqrestore(&info->lock, flags);
319 
320         shmem_inode_unacct_blocks(inode, pages);
321 }
322 
323 /*
324  * Replace item expected in radix tree by a new item, while holding tree lock.
325  */
326 static int shmem_radix_tree_replace(struct address_space *mapping,
327                         pgoff_t index, void *expected, void *replacement)
328 {
329         struct radix_tree_node *node;
330         void **pslot;
331         void *item;
332 
333         VM_BUG_ON(!expected);
334         VM_BUG_ON(!replacement);
335         item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot);
336         if (!item)
337                 return -ENOENT;
338         if (item != expected)
339                 return -ENOENT;
340         __radix_tree_replace(&mapping->page_tree, node, pslot,
341                              replacement, NULL);
342         return 0;
343 }
344 
345 /*
346  * Sometimes, before we decide whether to proceed or to fail, we must check
347  * that an entry was not already brought back from swap by a racing thread.
348  *
349  * Checking page is not enough: by the time a SwapCache page is locked, it
350  * might be reused, and again be SwapCache, using the same swap as before.
351  */
352 static bool shmem_confirm_swap(struct address_space *mapping,
353                                pgoff_t index, swp_entry_t swap)
354 {
355         void *item;
356 
357         rcu_read_lock();
358         item = radix_tree_lookup(&mapping->page_tree, index);
359         rcu_read_unlock();
360         return item == swp_to_radix_entry(swap);
361 }
362 
363 /*
364  * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
365  *
366  * SHMEM_HUGE_NEVER:
367  *      disables huge pages for the mount;
368  * SHMEM_HUGE_ALWAYS:
369  *      enables huge pages for the mount;
370  * SHMEM_HUGE_WITHIN_SIZE:
371  *      only allocate huge pages if the page will be fully within i_size,
372  *      also respect fadvise()/madvise() hints;
373  * SHMEM_HUGE_ADVISE:
374  *      only allocate huge pages if requested with fadvise()/madvise();
375  */
376 
377 #define SHMEM_HUGE_NEVER        0
378 #define SHMEM_HUGE_ALWAYS       1
379 #define SHMEM_HUGE_WITHIN_SIZE  2
380 #define SHMEM_HUGE_ADVISE       3
381 
382 /*
383  * Special values.
384  * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
385  *
386  * SHMEM_HUGE_DENY:
387  *      disables huge on shm_mnt and all mounts, for emergency use;
388  * SHMEM_HUGE_FORCE:
389  *      enables huge on shm_mnt and all mounts, w/o needing option, for testing;
390  *
391  */
392 #define SHMEM_HUGE_DENY         (-1)
393 #define SHMEM_HUGE_FORCE        (-2)
394 
395 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
396 /* ifdef here to avoid bloating shmem.o when not necessary */
397 
398 int shmem_huge __read_mostly;
399 
400 #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
401 static int shmem_parse_huge(const char *str)
402 {
403         if (!strcmp(str, "never"))
404                 return SHMEM_HUGE_NEVER;
405         if (!strcmp(str, "always"))
406                 return SHMEM_HUGE_ALWAYS;
407         if (!strcmp(str, "within_size"))
408                 return SHMEM_HUGE_WITHIN_SIZE;
409         if (!strcmp(str, "advise"))
410                 return SHMEM_HUGE_ADVISE;
411         if (!strcmp(str, "deny"))
412                 return SHMEM_HUGE_DENY;
413         if (!strcmp(str, "force"))
414                 return SHMEM_HUGE_FORCE;
415         return -EINVAL;
416 }
417 
418 static const char *shmem_format_huge(int huge)
419 {
420         switch (huge) {
421         case SHMEM_HUGE_NEVER:
422                 return "never";
423         case SHMEM_HUGE_ALWAYS:
424                 return "always";
425         case SHMEM_HUGE_WITHIN_SIZE:
426                 return "within_size";
427         case SHMEM_HUGE_ADVISE:
428                 return "advise";
429         case SHMEM_HUGE_DENY:
430                 return "deny";
431         case SHMEM_HUGE_FORCE:
432                 return "force";
433         default:
434                 VM_BUG_ON(1);
435                 return "bad_val";
436         }
437 }
438 #endif
439 
440 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
441                 struct shrink_control *sc, unsigned long nr_to_split)
442 {
443         LIST_HEAD(list), *pos, *next;
444         LIST_HEAD(to_remove);
445         struct inode *inode;
446         struct shmem_inode_info *info;
447         struct page *page;
448         unsigned long batch = sc ? sc->nr_to_scan : 128;
449         int removed = 0, split = 0;
450 
451         if (list_empty(&sbinfo->shrinklist))
452                 return SHRINK_STOP;
453 
454         spin_lock(&sbinfo->shrinklist_lock);
455         list_for_each_safe(pos, next, &sbinfo->shrinklist) {
456                 info = list_entry(pos, struct shmem_inode_info, shrinklist);
457 
458                 /* pin the inode */
459                 inode = igrab(&info->vfs_inode);
460 
461                 /* inode is about to be evicted */
462                 if (!inode) {
463                         list_del_init(&info->shrinklist);
464                         removed++;
465                         goto next;
466                 }
467 
468                 /* Check if there's anything to gain */
469                 if (round_up(inode->i_size, PAGE_SIZE) ==
470                                 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
471                         list_move(&info->shrinklist, &to_remove);
472                         removed++;
473                         goto next;
474                 }
475 
476                 list_move(&info->shrinklist, &list);
477 next:
478                 if (!--batch)
479                         break;
480         }
481         spin_unlock(&sbinfo->shrinklist_lock);
482 
483         list_for_each_safe(pos, next, &to_remove) {
484                 info = list_entry(pos, struct shmem_inode_info, shrinklist);
485                 inode = &info->vfs_inode;
486                 list_del_init(&info->shrinklist);
487                 iput(inode);
488         }
489 
490         list_for_each_safe(pos, next, &list) {
491                 int ret;
492 
493                 info = list_entry(pos, struct shmem_inode_info, shrinklist);
494                 inode = &info->vfs_inode;
495 
496                 if (nr_to_split && split >= nr_to_split)
497                         goto leave;
498 
499                 page = find_get_page(inode->i_mapping,
500                                 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
501                 if (!page)
502                         goto drop;
503 
504                 /* No huge page at the end of the file: nothing to split */
505                 if (!PageTransHuge(page)) {
506                         put_page(page);
507                         goto drop;
508                 }
509 
510                 /*
511                  * Leave the inode on the list if we failed to lock
512                  * the page at this time.
513                  *
514                  * Waiting for the lock may lead to deadlock in the
515                  * reclaim path.
516                  */
517                 if (!trylock_page(page)) {
518                         put_page(page);
519                         goto leave;
520                 }
521 
522                 ret = split_huge_page(page);
523                 unlock_page(page);
524                 put_page(page);
525 
526                 /* If split failed leave the inode on the list */
527                 if (ret)
528                         goto leave;
529 
530                 split++;
531 drop:
532                 list_del_init(&info->shrinklist);
533                 removed++;
534 leave:
535                 iput(inode);
536         }
537 
538         spin_lock(&sbinfo->shrinklist_lock);
539         list_splice_tail(&list, &sbinfo->shrinklist);
540         sbinfo->shrinklist_len -= removed;
541         spin_unlock(&sbinfo->shrinklist_lock);
542 
543         return split;
544 }
545 
546 static long shmem_unused_huge_scan(struct super_block *sb,
547                 struct shrink_control *sc)
548 {
549         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
550 
551         if (!READ_ONCE(sbinfo->shrinklist_len))
552                 return SHRINK_STOP;
553 
554         return shmem_unused_huge_shrink(sbinfo, sc, 0);
555 }
556 
557 static long shmem_unused_huge_count(struct super_block *sb,
558                 struct shrink_control *sc)
559 {
560         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
561         return READ_ONCE(sbinfo->shrinklist_len);
562 }
563 #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
564 
565 #define shmem_huge SHMEM_HUGE_DENY
566 
567 static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
568                 struct shrink_control *sc, unsigned long nr_to_split)
569 {
570         return 0;
571 }
572 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
573 
574 /*
575  * Like add_to_page_cache_locked, but error if expected item has gone.
576  */
577 static int shmem_add_to_page_cache(struct page *page,
578                                    struct address_space *mapping,
579                                    pgoff_t index, void *expected)
580 {
581         int error, nr = hpage_nr_pages(page);
582 
583         VM_BUG_ON_PAGE(PageTail(page), page);
584         VM_BUG_ON_PAGE(index != round_down(index, nr), page);
585         VM_BUG_ON_PAGE(!PageLocked(page), page);
586         VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
587         VM_BUG_ON(expected && PageTransHuge(page));
588 
589         page_ref_add(page, nr);
590         page->mapping = mapping;
591         page->index = index;
592 
593         spin_lock_irq(&mapping->tree_lock);
594         if (PageTransHuge(page)) {
595                 void __rcu **results;
596                 pgoff_t idx;
597                 int i;
598 
599                 error = 0;
600                 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
601                                         &results, &idx, index, 1) &&
602                                 idx < index + HPAGE_PMD_NR) {
603                         error = -EEXIST;
604                 }
605 
606                 if (!error) {
607                         for (i = 0; i < HPAGE_PMD_NR; i++) {
608                                 error = radix_tree_insert(&mapping->page_tree,
609                                                 index + i, page + i);
610                                 VM_BUG_ON(error);
611                         }
612                         count_vm_event(THP_FILE_ALLOC);
613                 }
614         } else if (!expected) {
615                 error = radix_tree_insert(&mapping->page_tree, index, page);
616         } else {
617                 error = shmem_radix_tree_replace(mapping, index, expected,
618                                                                  page);
619         }
620 
621         if (!error) {
622                 mapping->nrpages += nr;
623                 if (PageTransHuge(page))
624                         __inc_node_page_state(page, NR_SHMEM_THPS);
625                 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
626                 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
627                 spin_unlock_irq(&mapping->tree_lock);
628         } else {
629                 page->mapping = NULL;
630                 spin_unlock_irq(&mapping->tree_lock);
631                 page_ref_sub(page, nr);
632         }
633         return error;
634 }
635 
636 /*
637  * Like delete_from_page_cache, but substitutes swap for page.
638  */
639 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
640 {
641         struct address_space *mapping = page->mapping;
642         int error;
643 
644         VM_BUG_ON_PAGE(PageCompound(page), page);
645 
646         spin_lock_irq(&mapping->tree_lock);
647         error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
648         page->mapping = NULL;
649         mapping->nrpages--;
650         __dec_node_page_state(page, NR_FILE_PAGES);
651         __dec_node_page_state(page, NR_SHMEM);
652         spin_unlock_irq(&mapping->tree_lock);
653         put_page(page);
654         BUG_ON(error);
655 }
656 
657 /*
658  * Remove swap entry from radix tree, free the swap and its page cache.
659  */
660 static int shmem_free_swap(struct address_space *mapping,
661                            pgoff_t index, void *radswap)
662 {
663         void *old;
664 
665         spin_lock_irq(&mapping->tree_lock);
666         old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
667         spin_unlock_irq(&mapping->tree_lock);
668         if (old != radswap)
669                 return -ENOENT;
670         free_swap_and_cache(radix_to_swp_entry(radswap));
671         return 0;
672 }
673 
674 /*
675  * Determine (in bytes) how many of the shmem object's pages mapped by the
676  * given offsets are swapped out.
677  *
678  * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
679  * as long as the inode doesn't go away and racy results are not a problem.
680  */
681 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
682                                                 pgoff_t start, pgoff_t end)
683 {
684         struct radix_tree_iter iter;
685         void **slot;
686         struct page *page;
687         unsigned long swapped = 0;
688 
689         rcu_read_lock();
690 
691         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
692                 if (iter.index >= end)
693                         break;
694 
695                 page = radix_tree_deref_slot(slot);
696 
697                 if (radix_tree_deref_retry(page)) {
698                         slot = radix_tree_iter_retry(&iter);
699                         continue;
700                 }
701 
702                 if (radix_tree_exceptional_entry(page))
703                         swapped++;
704 
705                 if (need_resched()) {
706                         slot = radix_tree_iter_resume(slot, &iter);
707                         cond_resched_rcu();
708                 }
709         }
710 
711         rcu_read_unlock();
712 
713         return swapped << PAGE_SHIFT;
714 }
715 
716 /*
717  * Determine (in bytes) how many of the shmem object's pages mapped by the
718  * given vma is swapped out.
719  *
720  * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
721  * as long as the inode doesn't go away and racy results are not a problem.
722  */
723 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
724 {
725         struct inode *inode = file_inode(vma->vm_file);
726         struct shmem_inode_info *info = SHMEM_I(inode);
727         struct address_space *mapping = inode->i_mapping;
728         unsigned long swapped;
729 
730         /* Be careful as we don't hold info->lock */
731         swapped = READ_ONCE(info->swapped);
732 
733         /*
734          * The easier cases are when the shmem object has nothing in swap, or
735          * the vma maps it whole. Then we can simply use the stats that we
736          * already track.
737          */
738         if (!swapped)
739                 return 0;
740 
741         if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
742                 return swapped << PAGE_SHIFT;
743 
744         /* Here comes the more involved part */
745         return shmem_partial_swap_usage(mapping,
746                         linear_page_index(vma, vma->vm_start),
747                         linear_page_index(vma, vma->vm_end));
748 }
749 
750 /*
751  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
752  */
753 void shmem_unlock_mapping(struct address_space *mapping)
754 {
755         struct pagevec pvec;
756         pgoff_t indices[PAGEVEC_SIZE];
757         pgoff_t index = 0;
758 
759         pagevec_init(&pvec);
760         /*
761          * Minor point, but we might as well stop if someone else SHM_LOCKs it.
762          */
763         while (!mapping_unevictable(mapping)) {
764                 /*
765                  * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
766                  * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
767                  */
768                 pvec.nr = find_get_entries(mapping, index,
769                                            PAGEVEC_SIZE, pvec.pages, indices);
770                 if (!pvec.nr)
771                         break;
772                 index = indices[pvec.nr - 1] + 1;
773                 pagevec_remove_exceptionals(&pvec);
774                 check_move_unevictable_pages(pvec.pages, pvec.nr);
775                 pagevec_release(&pvec);
776                 cond_resched();
777         }
778 }
779 
780 /*
781  * Remove range of pages and swap entries from radix tree, and free them.
782  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
783  */
784 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
785                                                                  bool unfalloc)
786 {
787         struct address_space *mapping = inode->i_mapping;
788         struct shmem_inode_info *info = SHMEM_I(inode);
789         pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
790         pgoff_t end = (lend + 1) >> PAGE_SHIFT;
791         unsigned int partial_start = lstart & (PAGE_SIZE - 1);
792         unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
793         struct pagevec pvec;
794         pgoff_t indices[PAGEVEC_SIZE];
795         long nr_swaps_freed = 0;
796         pgoff_t index;
797         int i;
798 
799         if (lend == -1)
800                 end = -1;       /* unsigned, so actually very big */
801 
802         pagevec_init(&pvec);
803         index = start;
804         while (index < end) {
805                 pvec.nr = find_get_entries(mapping, index,
806                         min(end - index, (pgoff_t)PAGEVEC_SIZE),
807                         pvec.pages, indices);
808                 if (!pvec.nr)
809                         break;
810                 for (i = 0; i < pagevec_count(&pvec); i++) {
811                         struct page *page = pvec.pages[i];
812 
813                         index = indices[i];
814                         if (index >= end)
815                                 break;
816 
817                         if (radix_tree_exceptional_entry(page)) {
818                                 if (unfalloc)
819                                         continue;
820                                 nr_swaps_freed += !shmem_free_swap(mapping,
821                                                                 index, page);
822                                 continue;
823                         }
824 
825                         VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
826 
827                         if (!trylock_page(page))
828                                 continue;
829 
830                         if (PageTransTail(page)) {
831                                 /* Middle of THP: zero out the page */
832                                 clear_highpage(page);
833                                 unlock_page(page);
834                                 continue;
835                         } else if (PageTransHuge(page)) {
836                                 if (index == round_down(end, HPAGE_PMD_NR)) {
837                                         /*
838                                          * Range ends in the middle of THP:
839                                          * zero out the page
840                                          */
841                                         clear_highpage(page);
842                                         unlock_page(page);
843                                         continue;
844                                 }
845                                 index += HPAGE_PMD_NR - 1;
846                                 i += HPAGE_PMD_NR - 1;
847                         }
848 
849                         if (!unfalloc || !PageUptodate(page)) {
850                                 VM_BUG_ON_PAGE(PageTail(page), page);
851                                 if (page_mapping(page) == mapping) {
852                                         VM_BUG_ON_PAGE(PageWriteback(page), page);
853                                         truncate_inode_page(mapping, page);
854                                 }
855                         }
856                         unlock_page(page);
857                 }
858                 pagevec_remove_exceptionals(&pvec);
859                 pagevec_release(&pvec);
860                 cond_resched();
861                 index++;
862         }
863 
864         if (partial_start) {
865                 struct page *page = NULL;
866                 shmem_getpage(inode, start - 1, &page, SGP_READ);
867                 if (page) {
868                         unsigned int top = PAGE_SIZE;
869                         if (start > end) {
870                                 top = partial_end;
871                                 partial_end = 0;
872                         }
873                         zero_user_segment(page, partial_start, top);
874                         set_page_dirty(page);
875                         unlock_page(page);
876                         put_page(page);
877                 }
878         }
879         if (partial_end) {
880                 struct page *page = NULL;
881                 shmem_getpage(inode, end, &page, SGP_READ);
882                 if (page) {
883                         zero_user_segment(page, 0, partial_end);
884                         set_page_dirty(page);
885                         unlock_page(page);
886                         put_page(page);
887                 }
888         }
889         if (start >= end)
890                 return;
891 
892         index = start;
893         while (index < end) {
894                 cond_resched();
895 
896                 pvec.nr = find_get_entries(mapping, index,
897                                 min(end - index, (pgoff_t)PAGEVEC_SIZE),
898                                 pvec.pages, indices);
899                 if (!pvec.nr) {
900                         /* If all gone or hole-punch or unfalloc, we're done */
901                         if (index == start || end != -1)
902                                 break;
903                         /* But if truncating, restart to make sure all gone */
904                         index = start;
905                         continue;
906                 }
907                 for (i = 0; i < pagevec_count(&pvec); i++) {
908                         struct page *page = pvec.pages[i];
909 
910                         index = indices[i];
911                         if (index >= end)
912                                 break;
913 
914                         if (radix_tree_exceptional_entry(page)) {
915                                 if (unfalloc)
916                                         continue;
917                                 if (shmem_free_swap(mapping, index, page)) {
918                                         /* Swap was replaced by page: retry */
919                                         index--;
920                                         break;
921                                 }
922                                 nr_swaps_freed++;
923                                 continue;
924                         }
925 
926                         lock_page(page);
927 
928                         if (PageTransTail(page)) {
929                                 /* Middle of THP: zero out the page */
930                                 clear_highpage(page);
931                                 unlock_page(page);
932                                 /*
933                                  * Partial thp truncate due 'start' in middle
934                                  * of THP: don't need to look on these pages
935                                  * again on !pvec.nr restart.
936                                  */
937                                 if (index != round_down(end, HPAGE_PMD_NR))
938                                         start++;
939                                 continue;
940                         } else if (PageTransHuge(page)) {
941                                 if (index == round_down(end, HPAGE_PMD_NR)) {
942                                         /*
943                                          * Range ends in the middle of THP:
944                                          * zero out the page
945                                          */
946                                         clear_highpage(page);
947                                         unlock_page(page);
948                                         continue;
949                                 }
950                                 index += HPAGE_PMD_NR - 1;
951                                 i += HPAGE_PMD_NR - 1;
952                         }
953 
954                         if (!unfalloc || !PageUptodate(page)) {
955                                 VM_BUG_ON_PAGE(PageTail(page), page);
956                                 if (page_mapping(page) == mapping) {
957                                         VM_BUG_ON_PAGE(PageWriteback(page), page);
958                                         truncate_inode_page(mapping, page);
959                                 } else {
960                                         /* Page was replaced by swap: retry */
961                                         unlock_page(page);
962                                         index--;
963                                         break;
964                                 }
965                         }
966                         unlock_page(page);
967                 }
968                 pagevec_remove_exceptionals(&pvec);
969                 pagevec_release(&pvec);
970                 index++;
971         }
972 
973         spin_lock_irq(&info->lock);
974         info->swapped -= nr_swaps_freed;
975         shmem_recalc_inode(inode);
976         spin_unlock_irq(&info->lock);
977 }
978 
979 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
980 {
981         shmem_undo_range(inode, lstart, lend, false);
982         inode->i_ctime = inode->i_mtime = current_time(inode);
983 }
984 EXPORT_SYMBOL_GPL(shmem_truncate_range);
985 
986 static int shmem_getattr(const struct path *path, struct kstat *stat,
987                          u32 request_mask, unsigned int query_flags)
988 {
989         struct inode *inode = path->dentry->d_inode;
990         struct shmem_inode_info *info = SHMEM_I(inode);
991 
992         if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
993                 spin_lock_irq(&info->lock);
994                 shmem_recalc_inode(inode);
995                 spin_unlock_irq(&info->lock);
996         }
997         generic_fillattr(inode, stat);
998         return 0;
999 }
1000 
1001 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
1002 {
1003         struct inode *inode = d_inode(dentry);
1004         struct shmem_inode_info *info = SHMEM_I(inode);
1005         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1006         int error;
1007 
1008         error = setattr_prepare(dentry, attr);
1009         if (error)
1010                 return error;
1011 
1012         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1013                 loff_t oldsize = inode->i_size;
1014                 loff_t newsize = attr->ia_size;
1015 
1016                 /* protected by i_mutex */
1017                 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1018                     (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1019                         return -EPERM;
1020 
1021                 if (newsize != oldsize) {
1022                         error = shmem_reacct_size(SHMEM_I(inode)->flags,
1023                                         oldsize, newsize);
1024                         if (error)
1025                                 return error;
1026                         i_size_write(inode, newsize);
1027                         inode->i_ctime = inode->i_mtime = current_time(inode);
1028                 }
1029                 if (newsize <= oldsize) {
1030                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
1031                         if (oldsize > holebegin)
1032                                 unmap_mapping_range(inode->i_mapping,
1033                                                         holebegin, 0, 1);
1034                         if (info->alloced)
1035                                 shmem_truncate_range(inode,
1036                                                         newsize, (loff_t)-1);
1037                         /* unmap again to remove racily COWed private pages */
1038                         if (oldsize > holebegin)
1039                                 unmap_mapping_range(inode->i_mapping,
1040                                                         holebegin, 0, 1);
1041 
1042                         /*
1043                          * Part of the huge page can be beyond i_size: subject
1044                          * to shrink under memory pressure.
1045                          */
1046                         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1047                                 spin_lock(&sbinfo->shrinklist_lock);
1048                                 /*
1049                                  * _careful to defend against unlocked access to
1050                                  * ->shrink_list in shmem_unused_huge_shrink()
1051                                  */
1052                                 if (list_empty_careful(&info->shrinklist)) {
1053                                         list_add_tail(&info->shrinklist,
1054                                                         &sbinfo->shrinklist);
1055                                         sbinfo->shrinklist_len++;
1056                                 }
1057                                 spin_unlock(&sbinfo->shrinklist_lock);
1058                         }
1059                 }
1060         }
1061 
1062         setattr_copy(inode, attr);
1063         if (attr->ia_valid & ATTR_MODE)
1064                 error = posix_acl_chmod(inode, inode->i_mode);
1065         return error;
1066 }
1067 
1068 static void shmem_evict_inode(struct inode *inode)
1069 {
1070         struct shmem_inode_info *info = SHMEM_I(inode);
1071         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1072 
1073         if (inode->i_mapping->a_ops == &shmem_aops) {
1074                 shmem_unacct_size(info->flags, inode->i_size);
1075                 inode->i_size = 0;
1076                 shmem_truncate_range(inode, 0, (loff_t)-1);
1077                 if (!list_empty(&info->shrinklist)) {
1078                         spin_lock(&sbinfo->shrinklist_lock);
1079                         if (!list_empty(&info->shrinklist)) {
1080                                 list_del_init(&info->shrinklist);
1081                                 sbinfo->shrinklist_len--;
1082                         }
1083                         spin_unlock(&sbinfo->shrinklist_lock);
1084                 }
1085                 if (!list_empty(&info->swaplist)) {
1086                         mutex_lock(&shmem_swaplist_mutex);
1087                         list_del_init(&info->swaplist);
1088                         mutex_unlock(&shmem_swaplist_mutex);
1089                 }
1090         }
1091 
1092         simple_xattrs_free(&info->xattrs);
1093         WARN_ON(inode->i_blocks);
1094         shmem_free_inode(inode->i_sb);
1095         clear_inode(inode);
1096 }
1097 
1098 static unsigned long find_swap_entry(struct radix_tree_root *root, void *item)
1099 {
1100         struct radix_tree_iter iter;
1101         void **slot;
1102         unsigned long found = -1;
1103         unsigned int checked = 0;
1104 
1105         rcu_read_lock();
1106         radix_tree_for_each_slot(slot, root, &iter, 0) {
1107                 if (*slot == item) {
1108                         found = iter.index;
1109                         break;
1110                 }
1111                 checked++;
1112                 if ((checked % 4096) != 0)
1113                         continue;
1114                 slot = radix_tree_iter_resume(slot, &iter);
1115                 cond_resched_rcu();
1116         }
1117 
1118         rcu_read_unlock();
1119         return found;
1120 }
1121 
1122 /*
1123  * If swap found in inode, free it and move page from swapcache to filecache.
1124  */
1125 static int shmem_unuse_inode(struct shmem_inode_info *info,
1126                              swp_entry_t swap, struct page **pagep)
1127 {
1128         struct address_space *mapping = info->vfs_inode.i_mapping;
1129         void *radswap;
1130         pgoff_t index;
1131         gfp_t gfp;
1132         int error = 0;
1133 
1134         radswap = swp_to_radix_entry(swap);
1135         index = find_swap_entry(&mapping->page_tree, radswap);
1136         if (index == -1)
1137                 return -EAGAIN; /* tell shmem_unuse we found nothing */
1138 
1139         /*
1140          * Move _head_ to start search for next from here.
1141          * But be careful: shmem_evict_inode checks list_empty without taking
1142          * mutex, and there's an instant in list_move_tail when info->swaplist
1143          * would appear empty, if it were the only one on shmem_swaplist.
1144          */
1145         if (shmem_swaplist.next != &info->swaplist)
1146                 list_move_tail(&shmem_swaplist, &info->swaplist);
1147 
1148         gfp = mapping_gfp_mask(mapping);
1149         if (shmem_should_replace_page(*pagep, gfp)) {
1150                 mutex_unlock(&shmem_swaplist_mutex);
1151                 error = shmem_replace_page(pagep, gfp, info, index);
1152                 mutex_lock(&shmem_swaplist_mutex);
1153                 /*
1154                  * We needed to drop mutex to make that restrictive page
1155                  * allocation, but the inode might have been freed while we
1156                  * dropped it: although a racing shmem_evict_inode() cannot
1157                  * complete without emptying the radix_tree, our page lock
1158                  * on this swapcache page is not enough to prevent that -
1159                  * free_swap_and_cache() of our swap entry will only
1160                  * trylock_page(), removing swap from radix_tree whatever.
1161                  *
1162                  * We must not proceed to shmem_add_to_page_cache() if the
1163                  * inode has been freed, but of course we cannot rely on
1164                  * inode or mapping or info to check that.  However, we can
1165                  * safely check if our swap entry is still in use (and here
1166                  * it can't have got reused for another page): if it's still
1167                  * in use, then the inode cannot have been freed yet, and we
1168                  * can safely proceed (if it's no longer in use, that tells
1169                  * nothing about the inode, but we don't need to unuse swap).
1170                  */
1171                 if (!page_swapcount(*pagep))
1172                         error = -ENOENT;
1173         }
1174 
1175         /*
1176          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1177          * but also to hold up shmem_evict_inode(): so inode cannot be freed
1178          * beneath us (pagelock doesn't help until the page is in pagecache).
1179          */
1180         if (!error)
1181                 error = shmem_add_to_page_cache(*pagep, mapping, index,
1182                                                 radswap);
1183         if (error != -ENOMEM) {
1184                 /*
1185                  * Truncation and eviction use free_swap_and_cache(), which
1186                  * only does trylock page: if we raced, best clean up here.
1187                  */
1188                 delete_from_swap_cache(*pagep);
1189                 set_page_dirty(*pagep);
1190                 if (!error) {
1191                         spin_lock_irq(&info->lock);
1192                         info->swapped--;
1193                         spin_unlock_irq(&info->lock);
1194                         swap_free(swap);
1195                 }
1196         }
1197         return error;
1198 }
1199 
1200 /*
1201  * Search through swapped inodes to find and replace swap by page.
1202  */
1203 int shmem_unuse(swp_entry_t swap, struct page *page)
1204 {
1205         struct list_head *this, *next;
1206         struct shmem_inode_info *info;
1207         struct mem_cgroup *memcg;
1208         int error = 0;
1209 
1210         /*
1211          * There's a faint possibility that swap page was replaced before
1212          * caller locked it: caller will come back later with the right page.
1213          */
1214         if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
1215                 goto out;
1216 
1217         /*
1218          * Charge page using GFP_KERNEL while we can wait, before taking
1219          * the shmem_swaplist_mutex which might hold up shmem_writepage().
1220          * Charged back to the user (not to caller) when swap account is used.
1221          */
1222         error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1223                         false);
1224         if (error)
1225                 goto out;
1226         /* No radix_tree_preload: swap entry keeps a place for page in tree */
1227         error = -EAGAIN;
1228 
1229         mutex_lock(&shmem_swaplist_mutex);
1230         list_for_each_safe(this, next, &shmem_swaplist) {
1231                 info = list_entry(this, struct shmem_inode_info, swaplist);
1232                 if (info->swapped)
1233                         error = shmem_unuse_inode(info, swap, &page);
1234                 else
1235                         list_del_init(&info->swaplist);
1236                 cond_resched();
1237                 if (error != -EAGAIN)
1238                         break;
1239                 /* found nothing in this: move on to search the next */
1240         }
1241         mutex_unlock(&shmem_swaplist_mutex);
1242 
1243         if (error) {
1244                 if (error != -ENOMEM)
1245                         error = 0;
1246                 mem_cgroup_cancel_charge(page, memcg, false);
1247         } else
1248                 mem_cgroup_commit_charge(page, memcg, true, false);
1249 out:
1250         unlock_page(page);
1251         put_page(page);
1252         return error;
1253 }
1254 
1255 /*
1256  * Move the page from the page cache to the swap cache.
1257  */
1258 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1259 {
1260         struct shmem_inode_info *info;
1261         struct address_space *mapping;
1262         struct inode *inode;
1263         swp_entry_t swap;
1264         pgoff_t index;
1265 
1266         VM_BUG_ON_PAGE(PageCompound(page), page);
1267         BUG_ON(!PageLocked(page));
1268         mapping = page->mapping;
1269         index = page->index;
1270         inode = mapping->host;
1271         info = SHMEM_I(inode);
1272         if (info->flags & VM_LOCKED)
1273                 goto redirty;
1274         if (!total_swap_pages)
1275                 goto redirty;
1276 
1277         /*
1278          * Our capabilities prevent regular writeback or sync from ever calling
1279          * shmem_writepage; but a stacking filesystem might use ->writepage of
1280          * its underlying filesystem, in which case tmpfs should write out to
1281          * swap only in response to memory pressure, and not for the writeback
1282          * threads or sync.
1283          */
1284         if (!wbc->for_reclaim) {
1285                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
1286                 goto redirty;
1287         }
1288 
1289         /*
1290          * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1291          * value into swapfile.c, the only way we can correctly account for a
1292          * fallocated page arriving here is now to initialize it and write it.
1293          *
1294          * That's okay for a page already fallocated earlier, but if we have
1295          * not yet completed the fallocation, then (a) we want to keep track
1296          * of this page in case we have to undo it, and (b) it may not be a
1297          * good idea to continue anyway, once we're pushing into swap.  So
1298          * reactivate the page, and let shmem_fallocate() quit when too many.
1299          */
1300         if (!PageUptodate(page)) {
1301                 if (inode->i_private) {
1302                         struct shmem_falloc *shmem_falloc;
1303                         spin_lock(&inode->i_lock);
1304                         shmem_falloc = inode->i_private;
1305                         if (shmem_falloc &&
1306                             !shmem_falloc->waitq &&
1307                             index >= shmem_falloc->start &&
1308                             index < shmem_falloc->next)
1309                                 shmem_falloc->nr_unswapped++;
1310                         else
1311                                 shmem_falloc = NULL;
1312                         spin_unlock(&inode->i_lock);
1313                         if (shmem_falloc)
1314                                 goto redirty;
1315                 }
1316                 clear_highpage(page);
1317                 flush_dcache_page(page);
1318                 SetPageUptodate(page);
1319         }
1320 
1321         swap = get_swap_page(page);
1322         if (!swap.val)
1323                 goto redirty;
1324 
1325         if (mem_cgroup_try_charge_swap(page, swap))
1326                 goto free_swap;
1327 
1328         /*
1329          * Add inode to shmem_unuse()'s list of swapped-out inodes,
1330          * if it's not already there.  Do it now before the page is
1331          * moved to swap cache, when its pagelock no longer protects
1332          * the inode from eviction.  But don't unlock the mutex until
1333          * we've incremented swapped, because shmem_unuse_inode() will
1334          * prune a !swapped inode from the swaplist under this mutex.
1335          */
1336         mutex_lock(&shmem_swaplist_mutex);
1337         if (list_empty(&info->swaplist))
1338                 list_add_tail(&info->swaplist, &shmem_swaplist);
1339 
1340         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1341                 spin_lock_irq(&info->lock);
1342                 shmem_recalc_inode(inode);
1343                 info->swapped++;
1344                 spin_unlock_irq(&info->lock);
1345 
1346                 swap_shmem_alloc(swap);
1347                 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1348 
1349                 mutex_unlock(&shmem_swaplist_mutex);
1350                 BUG_ON(page_mapped(page));
1351                 swap_writepage(page, wbc);
1352                 return 0;
1353         }
1354 
1355         mutex_unlock(&shmem_swaplist_mutex);
1356 free_swap:
1357         put_swap_page(page, swap);
1358 redirty:
1359         set_page_dirty(page);
1360         if (wbc->for_reclaim)
1361                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
1362         unlock_page(page);
1363         return 0;
1364 }
1365 
1366 #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1367 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1368 {
1369         char buffer[64];
1370 
1371         if (!mpol || mpol->mode == MPOL_DEFAULT)
1372                 return;         /* show nothing */
1373 
1374         mpol_to_str(buffer, sizeof(buffer), mpol);
1375 
1376         seq_printf(seq, ",mpol=%s", buffer);
1377 }
1378 
1379 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1380 {
1381         struct mempolicy *mpol = NULL;
1382         if (sbinfo->mpol) {
1383                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
1384                 mpol = sbinfo->mpol;
1385                 mpol_get(mpol);
1386                 spin_unlock(&sbinfo->stat_lock);
1387         }
1388         return mpol;
1389 }
1390 #else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1391 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1392 {
1393 }
1394 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1395 {
1396         return NULL;
1397 }
1398 #endif /* CONFIG_NUMA && CONFIG_TMPFS */
1399 #ifndef CONFIG_NUMA
1400 #define vm_policy vm_private_data
1401 #endif
1402 
1403 static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1404                 struct shmem_inode_info *info, pgoff_t index)
1405 {
1406         /* Create a pseudo vma that just contains the policy */
1407         vma->vm_start = 0;
1408         /* Bias interleave by inode number to distribute better across nodes */
1409         vma->vm_pgoff = index + info->vfs_inode.i_ino;
1410         vma->vm_ops = NULL;
1411         vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1412 }
1413 
1414 static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1415 {
1416         /* Drop reference taken by mpol_shared_policy_lookup() */
1417         mpol_cond_put(vma->vm_policy);
1418 }
1419 
1420 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1421                         struct shmem_inode_info *info, pgoff_t index)
1422 {
1423         struct vm_area_struct pvma;
1424         struct page *page;
1425 
1426         shmem_pseudo_vma_init(&pvma, info, index);
1427         page = swapin_readahead(swap, gfp, &pvma, 0);
1428         shmem_pseudo_vma_destroy(&pvma);
1429 
1430         return page;
1431 }
1432 
1433 static struct page *shmem_alloc_hugepage(gfp_t gfp,
1434                 struct shmem_inode_info *info, pgoff_t index)
1435 {
1436         struct vm_area_struct pvma;
1437         struct inode *inode = &info->vfs_inode;
1438         struct address_space *mapping = inode->i_mapping;
1439         pgoff_t idx, hindex;
1440         void __rcu **results;
1441         struct page *page;
1442 
1443         if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1444                 return NULL;
1445 
1446         hindex = round_down(index, HPAGE_PMD_NR);
1447         rcu_read_lock();
1448         if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1449                                 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1450                 rcu_read_unlock();
1451                 return NULL;
1452         }
1453         rcu_read_unlock();
1454 
1455         shmem_pseudo_vma_init(&pvma, info, hindex);
1456         page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1457                         HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1458         shmem_pseudo_vma_destroy(&pvma);
1459         if (page)
1460                 prep_transhuge_page(page);
1461         return page;
1462 }
1463 
1464 static struct page *shmem_alloc_page(gfp_t gfp,
1465                         struct shmem_inode_info *info, pgoff_t index)
1466 {
1467         struct vm_area_struct pvma;
1468         struct page *page;
1469 
1470         shmem_pseudo_vma_init(&pvma, info, index);
1471         page = alloc_page_vma(gfp, &pvma, 0);
1472         shmem_pseudo_vma_destroy(&pvma);
1473 
1474         return page;
1475 }
1476 
1477 static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1478                 struct inode *inode,
1479                 pgoff_t index, bool huge)
1480 {
1481         struct shmem_inode_info *info = SHMEM_I(inode);
1482         struct page *page;
1483         int nr;
1484         int err = -ENOSPC;
1485 
1486         if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1487                 huge = false;
1488         nr = huge ? HPAGE_PMD_NR : 1;
1489 
1490         if (!shmem_inode_acct_block(inode, nr))
1491                 goto failed;
1492 
1493         if (huge)
1494                 page = shmem_alloc_hugepage(gfp, info, index);
1495         else
1496                 page = shmem_alloc_page(gfp, info, index);
1497         if (page) {
1498                 __SetPageLocked(page);
1499                 __SetPageSwapBacked(page);
1500                 return page;
1501         }
1502 
1503         err = -ENOMEM;
1504         shmem_inode_unacct_blocks(inode, nr);
1505 failed:
1506         return ERR_PTR(err);
1507 }
1508 
1509 /*
1510  * When a page is moved from swapcache to shmem filecache (either by the
1511  * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1512  * shmem_unuse_inode()), it may have been read in earlier from swap, in
1513  * ignorance of the mapping it belongs to.  If that mapping has special
1514  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1515  * we may need to copy to a suitable page before moving to filecache.
1516  *
1517  * In a future release, this may well be extended to respect cpuset and
1518  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1519  * but for now it is a simple matter of zone.
1520  */
1521 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1522 {
1523         return page_zonenum(page) > gfp_zone(gfp);
1524 }
1525 
1526 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1527                                 struct shmem_inode_info *info, pgoff_t index)
1528 {
1529         struct page *oldpage, *newpage;
1530         struct address_space *swap_mapping;
1531         pgoff_t swap_index;
1532         int error;
1533 
1534         oldpage = *pagep;
1535         swap_index = page_private(oldpage);
1536         swap_mapping = page_mapping(oldpage);
1537 
1538         /*
1539          * We have arrived here because our zones are constrained, so don't
1540          * limit chance of success by further cpuset and node constraints.
1541          */
1542         gfp &= ~GFP_CONSTRAINT_MASK;
1543         newpage = shmem_alloc_page(gfp, info, index);
1544         if (!newpage)
1545                 return -ENOMEM;
1546 
1547         get_page(newpage);
1548         copy_highpage(newpage, oldpage);
1549         flush_dcache_page(newpage);
1550 
1551         __SetPageLocked(newpage);
1552         __SetPageSwapBacked(newpage);
1553         SetPageUptodate(newpage);
1554         set_page_private(newpage, swap_index);
1555         SetPageSwapCache(newpage);
1556 
1557         /*
1558          * Our caller will very soon move newpage out of swapcache, but it's
1559          * a nice clean interface for us to replace oldpage by newpage there.
1560          */
1561         spin_lock_irq(&swap_mapping->tree_lock);
1562         error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1563                                                                    newpage);
1564         if (!error) {
1565                 __inc_node_page_state(newpage, NR_FILE_PAGES);
1566                 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1567         }
1568         spin_unlock_irq(&swap_mapping->tree_lock);
1569 
1570         if (unlikely(error)) {
1571                 /*
1572                  * Is this possible?  I think not, now that our callers check
1573                  * both PageSwapCache and page_private after getting page lock;
1574                  * but be defensive.  Reverse old to newpage for clear and free.
1575                  */
1576                 oldpage = newpage;
1577         } else {
1578                 mem_cgroup_migrate(oldpage, newpage);
1579                 lru_cache_add_anon(newpage);
1580                 *pagep = newpage;
1581         }
1582 
1583         ClearPageSwapCache(oldpage);
1584         set_page_private(oldpage, 0);
1585 
1586         unlock_page(oldpage);
1587         put_page(oldpage);
1588         put_page(oldpage);
1589         return error;
1590 }
1591 
1592 /*
1593  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1594  *
1595  * If we allocate a new one we do not mark it dirty. That's up to the
1596  * vm. If we swap it in we mark it dirty since we also free the swap
1597  * entry since a page cannot live in both the swap and page cache.
1598  *
1599  * fault_mm and fault_type are only supplied by shmem_fault:
1600  * otherwise they are NULL.
1601  */
1602 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1603         struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1604         struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type)
1605 {
1606         struct address_space *mapping = inode->i_mapping;
1607         struct shmem_inode_info *info = SHMEM_I(inode);
1608         struct shmem_sb_info *sbinfo;
1609         struct mm_struct *charge_mm;
1610         struct mem_cgroup *memcg;
1611         struct page *page;
1612         swp_entry_t swap;
1613         enum sgp_type sgp_huge = sgp;
1614         pgoff_t hindex = index;
1615         int error;
1616         int once = 0;
1617         int alloced = 0;
1618 
1619         if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1620                 return -EFBIG;
1621         if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1622                 sgp = SGP_CACHE;
1623 repeat:
1624         swap.val = 0;
1625         page = find_lock_entry(mapping, index);
1626         if (radix_tree_exceptional_entry(page)) {
1627                 swap = radix_to_swp_entry(page);
1628                 page = NULL;
1629         }
1630 
1631         if (sgp <= SGP_CACHE &&
1632             ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1633                 error = -EINVAL;
1634                 goto unlock;
1635         }
1636 
1637         if (page && sgp == SGP_WRITE)
1638                 mark_page_accessed(page);
1639 
1640         /* fallocated page? */
1641         if (page && !PageUptodate(page)) {
1642                 if (sgp != SGP_READ)
1643                         goto clear;
1644                 unlock_page(page);
1645                 put_page(page);
1646                 page = NULL;
1647         }
1648         if (page || (sgp == SGP_READ && !swap.val)) {
1649                 *pagep = page;
1650                 return 0;
1651         }
1652 
1653         /*
1654          * Fast cache lookup did not find it:
1655          * bring it back from swap or allocate.
1656          */
1657         sbinfo = SHMEM_SB(inode->i_sb);
1658         charge_mm = vma ? vma->vm_mm : current->mm;
1659 
1660         if (swap.val) {
1661                 /* Look it up and read it in.. */
1662                 page = lookup_swap_cache(swap, NULL, 0);
1663                 if (!page) {
1664                         /* Or update major stats only when swapin succeeds?? */
1665                         if (fault_type) {
1666                                 *fault_type |= VM_FAULT_MAJOR;
1667                                 count_vm_event(PGMAJFAULT);
1668                                 count_memcg_event_mm(charge_mm, PGMAJFAULT);
1669                         }
1670                         /* Here we actually start the io */
1671                         page = shmem_swapin(swap, gfp, info, index);
1672                         if (!page) {
1673                                 error = -ENOMEM;
1674                                 goto failed;
1675                         }
1676                 }
1677 
1678                 /* We have to do this with page locked to prevent races */
1679                 lock_page(page);
1680                 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1681                     !shmem_confirm_swap(mapping, index, swap)) {
1682                         error = -EEXIST;        /* try again */
1683                         goto unlock;
1684                 }
1685                 if (!PageUptodate(page)) {
1686                         error = -EIO;
1687                         goto failed;
1688                 }
1689                 wait_on_page_writeback(page);
1690 
1691                 if (shmem_should_replace_page(page, gfp)) {
1692                         error = shmem_replace_page(&page, gfp, info, index);
1693                         if (error)
1694                                 goto failed;
1695                 }
1696 
1697                 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1698                                 false);
1699                 if (!error) {
1700                         error = shmem_add_to_page_cache(page, mapping, index,
1701                                                 swp_to_radix_entry(swap));
1702                         /*
1703                          * We already confirmed swap under page lock, and make
1704                          * no memory allocation here, so usually no possibility
1705                          * of error; but free_swap_and_cache() only trylocks a
1706                          * page, so it is just possible that the entry has been
1707                          * truncated or holepunched since swap was confirmed.
1708                          * shmem_undo_range() will have done some of the
1709                          * unaccounting, now delete_from_swap_cache() will do
1710                          * the rest.
1711                          * Reset swap.val? No, leave it so "failed" goes back to
1712                          * "repeat": reading a hole and writing should succeed.
1713                          */
1714                         if (error) {
1715                                 mem_cgroup_cancel_charge(page, memcg, false);
1716                                 delete_from_swap_cache(page);
1717                         }
1718                 }
1719                 if (error)
1720                         goto failed;
1721 
1722                 mem_cgroup_commit_charge(page, memcg, true, false);
1723 
1724                 spin_lock_irq(&info->lock);
1725                 info->swapped--;
1726                 shmem_recalc_inode(inode);
1727                 spin_unlock_irq(&info->lock);
1728 
1729                 if (sgp == SGP_WRITE)
1730                         mark_page_accessed(page);
1731 
1732                 delete_from_swap_cache(page);
1733                 set_page_dirty(page);
1734                 swap_free(swap);
1735 
1736         } else {
1737                 if (vma && userfaultfd_missing(vma)) {
1738                         *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1739                         return 0;
1740                 }
1741 
1742                 /* shmem_symlink() */
1743                 if (mapping->a_ops != &shmem_aops)
1744                         goto alloc_nohuge;
1745                 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1746                         goto alloc_nohuge;
1747                 if (shmem_huge == SHMEM_HUGE_FORCE)
1748                         goto alloc_huge;
1749                 switch (sbinfo->huge) {
1750                         loff_t i_size;
1751                         pgoff_t off;
1752                 case SHMEM_HUGE_NEVER:
1753                         goto alloc_nohuge;
1754                 case SHMEM_HUGE_WITHIN_SIZE:
1755                         off = round_up(index, HPAGE_PMD_NR);
1756                         i_size = round_up(i_size_read(inode), PAGE_SIZE);
1757                         if (i_size >= HPAGE_PMD_SIZE &&
1758                                         i_size >> PAGE_SHIFT >= off)
1759                                 goto alloc_huge;
1760                         /* fallthrough */
1761                 case SHMEM_HUGE_ADVISE:
1762                         if (sgp_huge == SGP_HUGE)
1763                                 goto alloc_huge;
1764                         /* TODO: implement fadvise() hints */
1765                         goto alloc_nohuge;
1766                 }
1767 
1768 alloc_huge:
1769                 page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1770                 if (IS_ERR(page)) {
1771 alloc_nohuge:           page = shmem_alloc_and_acct_page(gfp, inode,
1772                                         index, false);
1773                 }
1774                 if (IS_ERR(page)) {
1775                         int retry = 5;
1776                         error = PTR_ERR(page);
1777                         page = NULL;
1778                         if (error != -ENOSPC)
1779                                 goto failed;
1780                         /*
1781                          * Try to reclaim some spece by splitting a huge page
1782                          * beyond i_size on the filesystem.
1783                          */
1784                         while (retry--) {
1785                                 int ret;
1786                                 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1787                                 if (ret == SHRINK_STOP)
1788                                         break;
1789                                 if (ret)
1790                                         goto alloc_nohuge;
1791                         }
1792                         goto failed;
1793                 }
1794 
1795                 if (PageTransHuge(page))
1796                         hindex = round_down(index, HPAGE_PMD_NR);
1797                 else
1798                         hindex = index;
1799 
1800                 if (sgp == SGP_WRITE)
1801                         __SetPageReferenced(page);
1802 
1803                 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1804                                 PageTransHuge(page));
1805                 if (error)
1806                         goto unacct;
1807                 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1808                                 compound_order(page));
1809                 if (!error) {
1810                         error = shmem_add_to_page_cache(page, mapping, hindex,
1811                                                         NULL);
1812                         radix_tree_preload_end();
1813                 }
1814                 if (error) {
1815                         mem_cgroup_cancel_charge(page, memcg,
1816                                         PageTransHuge(page));
1817                         goto unacct;
1818                 }
1819                 mem_cgroup_commit_charge(page, memcg, false,
1820                                 PageTransHuge(page));
1821                 lru_cache_add_anon(page);
1822 
1823                 spin_lock_irq(&info->lock);
1824                 info->alloced += 1 << compound_order(page);
1825                 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1826                 shmem_recalc_inode(inode);
1827                 spin_unlock_irq(&info->lock);
1828                 alloced = true;
1829 
1830                 if (PageTransHuge(page) &&
1831                                 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1832                                 hindex + HPAGE_PMD_NR - 1) {
1833                         /*
1834                          * Part of the huge page is beyond i_size: subject
1835                          * to shrink under memory pressure.
1836                          */
1837                         spin_lock(&sbinfo->shrinklist_lock);
1838                         /*
1839                          * _careful to defend against unlocked access to
1840                          * ->shrink_list in shmem_unused_huge_shrink()
1841                          */
1842                         if (list_empty_careful(&info->shrinklist)) {
1843                                 list_add_tail(&info->shrinklist,
1844                                                 &sbinfo->shrinklist);
1845                                 sbinfo->shrinklist_len++;
1846                         }
1847                         spin_unlock(&sbinfo->shrinklist_lock);
1848                 }
1849 
1850                 /*
1851                  * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1852                  */
1853                 if (sgp == SGP_FALLOC)
1854                         sgp = SGP_WRITE;
1855 clear:
1856                 /*
1857                  * Let SGP_WRITE caller clear ends if write does not fill page;
1858                  * but SGP_FALLOC on a page fallocated earlier must initialize
1859                  * it now, lest undo on failure cancel our earlier guarantee.
1860                  */
1861                 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1862                         struct page *head = compound_head(page);
1863                         int i;
1864 
1865                         for (i = 0; i < (1 << compound_order(head)); i++) {
1866                                 clear_highpage(head + i);
1867                                 flush_dcache_page(head + i);
1868                         }
1869                         SetPageUptodate(head);
1870                 }
1871         }
1872 
1873         /* Perhaps the file has been truncated since we checked */
1874         if (sgp <= SGP_CACHE &&
1875             ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1876                 if (alloced) {
1877                         ClearPageDirty(page);
1878                         delete_from_page_cache(page);
1879                         spin_lock_irq(&info->lock);
1880                         shmem_recalc_inode(inode);
1881                         spin_unlock_irq(&info->lock);
1882                 }
1883                 error = -EINVAL;
1884                 goto unlock;
1885         }
1886         *pagep = page + index - hindex;
1887         return 0;
1888 
1889         /*
1890          * Error recovery.
1891          */
1892 unacct:
1893         shmem_inode_unacct_blocks(inode, 1 << compound_order(page));
1894 
1895         if (PageTransHuge(page)) {
1896                 unlock_page(page);
1897                 put_page(page);
1898                 goto alloc_nohuge;
1899         }
1900 failed:
1901         if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1902                 error = -EEXIST;
1903 unlock:
1904         if (page) {
1905                 unlock_page(page);
1906                 put_page(page);
1907         }
1908         if (error == -ENOSPC && !once++) {
1909                 spin_lock_irq(&info->lock);
1910                 shmem_recalc_inode(inode);
1911                 spin_unlock_irq(&info->lock);
1912                 goto repeat;
1913         }
1914         if (error == -EEXIST)   /* from above or from radix_tree_insert */
1915                 goto repeat;
1916         return error;
1917 }
1918 
1919 /*
1920  * This is like autoremove_wake_function, but it removes the wait queue
1921  * entry unconditionally - even if something else had already woken the
1922  * target.
1923  */
1924 static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1925 {
1926         int ret = default_wake_function(wait, mode, sync, key);
1927         list_del_init(&wait->entry);
1928         return ret;
1929 }
1930 
1931 static int shmem_fault(struct vm_fault *vmf)
1932 {
1933         struct vm_area_struct *vma = vmf->vma;
1934         struct inode *inode = file_inode(vma->vm_file);
1935         gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1936         enum sgp_type sgp;
1937         int error;
1938         int ret = VM_FAULT_LOCKED;
1939 
1940         /*
1941          * Trinity finds that probing a hole which tmpfs is punching can
1942          * prevent the hole-punch from ever completing: which in turn
1943          * locks writers out with its hold on i_mutex.  So refrain from
1944          * faulting pages into the hole while it's being punched.  Although
1945          * shmem_undo_range() does remove the additions, it may be unable to
1946          * keep up, as each new page needs its own unmap_mapping_range() call,
1947          * and the i_mmap tree grows ever slower to scan if new vmas are added.
1948          *
1949          * It does not matter if we sometimes reach this check just before the
1950          * hole-punch begins, so that one fault then races with the punch:
1951          * we just need to make racing faults a rare case.
1952          *
1953          * The implementation below would be much simpler if we just used a
1954          * standard mutex or completion: but we cannot take i_mutex in fault,
1955          * and bloating every shmem inode for this unlikely case would be sad.
1956          */
1957         if (unlikely(inode->i_private)) {
1958                 struct shmem_falloc *shmem_falloc;
1959 
1960                 spin_lock(&inode->i_lock);
1961                 shmem_falloc = inode->i_private;
1962                 if (shmem_falloc &&
1963                     shmem_falloc->waitq &&
1964                     vmf->pgoff >= shmem_falloc->start &&
1965                     vmf->pgoff < shmem_falloc->next) {
1966                         wait_queue_head_t *shmem_falloc_waitq;
1967                         DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
1968 
1969                         ret = VM_FAULT_NOPAGE;
1970                         if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1971                            !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1972                                 /* It's polite to up mmap_sem if we can */
1973                                 up_read(&vma->vm_mm->mmap_sem);
1974                                 ret = VM_FAULT_RETRY;
1975                         }
1976 
1977                         shmem_falloc_waitq = shmem_falloc->waitq;
1978                         prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1979                                         TASK_UNINTERRUPTIBLE);
1980                         spin_unlock(&inode->i_lock);
1981                         schedule();
1982 
1983                         /*
1984                          * shmem_falloc_waitq points into the shmem_fallocate()
1985                          * stack of the hole-punching task: shmem_falloc_waitq
1986                          * is usually invalid by the time we reach here, but
1987                          * finish_wait() does not dereference it in that case;
1988                          * though i_lock needed lest racing with wake_up_all().
1989                          */
1990                         spin_lock(&inode->i_lock);
1991                         finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1992                         spin_unlock(&inode->i_lock);
1993                         return ret;
1994                 }
1995                 spin_unlock(&inode->i_lock);
1996         }
1997 
1998         sgp = SGP_CACHE;
1999 
2000         if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2001             test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2002                 sgp = SGP_NOHUGE;
2003         else if (vma->vm_flags & VM_HUGEPAGE)
2004                 sgp = SGP_HUGE;
2005 
2006         error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2007                                   gfp, vma, vmf, &ret);
2008         if (error)
2009                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
2010         return ret;
2011 }
2012 
2013 unsigned long shmem_get_unmapped_area(struct file *file,
2014                                       unsigned long uaddr, unsigned long len,
2015                                       unsigned long pgoff, unsigned long flags)
2016 {
2017         unsigned long (*get_area)(struct file *,
2018                 unsigned long, unsigned long, unsigned long, unsigned long);
2019         unsigned long addr;
2020         unsigned long offset;
2021         unsigned long inflated_len;
2022         unsigned long inflated_addr;
2023         unsigned long inflated_offset;
2024 
2025         if (len > TASK_SIZE)
2026                 return -ENOMEM;
2027 
2028         get_area = current->mm->get_unmapped_area;
2029         addr = get_area(file, uaddr, len, pgoff, flags);
2030 
2031         if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2032                 return addr;
2033         if (IS_ERR_VALUE(addr))
2034                 return addr;
2035         if (addr & ~PAGE_MASK)
2036                 return addr;
2037         if (addr > TASK_SIZE - len)
2038                 return addr;
2039 
2040         if (shmem_huge == SHMEM_HUGE_DENY)
2041                 return addr;
2042         if (len < HPAGE_PMD_SIZE)
2043                 return addr;
2044         if (flags & MAP_FIXED)
2045                 return addr;
2046         /*
2047          * Our priority is to support MAP_SHARED mapped hugely;
2048          * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2049          * But if caller specified an address hint, respect that as before.
2050          */
2051         if (uaddr)
2052                 return addr;
2053 
2054         if (shmem_huge != SHMEM_HUGE_FORCE) {
2055                 struct super_block *sb;
2056 
2057                 if (file) {
2058                         VM_BUG_ON(file->f_op != &shmem_file_operations);
2059                         sb = file_inode(file)->i_sb;
2060                 } else {
2061                         /*
2062                          * Called directly from mm/mmap.c, or drivers/char/mem.c
2063                          * for "/dev/zero", to create a shared anonymous object.
2064                          */
2065                         if (IS_ERR(shm_mnt))
2066                                 return addr;
2067                         sb = shm_mnt->mnt_sb;
2068                 }
2069                 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2070                         return addr;
2071         }
2072 
2073         offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2074         if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2075                 return addr;
2076         if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2077                 return addr;
2078 
2079         inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2080         if (inflated_len > TASK_SIZE)
2081                 return addr;
2082         if (inflated_len < len)
2083                 return addr;
2084 
2085         inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2086         if (IS_ERR_VALUE(inflated_addr))
2087                 return addr;
2088         if (inflated_addr & ~PAGE_MASK)
2089                 return addr;
2090 
2091         inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2092         inflated_addr += offset - inflated_offset;
2093         if (inflated_offset > offset)
2094                 inflated_addr += HPAGE_PMD_SIZE;
2095 
2096         if (inflated_addr > TASK_SIZE - len)
2097                 return addr;
2098         return inflated_addr;
2099 }
2100 
2101 #ifdef CONFIG_NUMA
2102 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2103 {
2104         struct inode *inode = file_inode(vma->vm_file);
2105         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2106 }
2107 
2108 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2109                                           unsigned long addr)
2110 {
2111         struct inode *inode = file_inode(vma->vm_file);
2112         pgoff_t index;
2113 
2114         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2115         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2116 }
2117 #endif
2118 
2119 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2120 {
2121         struct inode *inode = file_inode(file);
2122         struct shmem_inode_info *info = SHMEM_I(inode);
2123         int retval = -ENOMEM;
2124 
2125         spin_lock_irq(&info->lock);
2126         if (lock && !(info->flags & VM_LOCKED)) {
2127                 if (!user_shm_lock(inode->i_size, user))
2128                         goto out_nomem;
2129                 info->flags |= VM_LOCKED;
2130                 mapping_set_unevictable(file->f_mapping);
2131         }
2132         if (!lock && (info->flags & VM_LOCKED) && user) {
2133                 user_shm_unlock(inode->i_size, user);
2134                 info->flags &= ~VM_LOCKED;
2135                 mapping_clear_unevictable(file->f_mapping);
2136         }
2137         retval = 0;
2138 
2139 out_nomem:
2140         spin_unlock_irq(&info->lock);
2141         return retval;
2142 }
2143 
2144 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2145 {
2146         file_accessed(file);
2147         vma->vm_ops = &shmem_vm_ops;
2148         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2149                         ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2150                         (vma->vm_end & HPAGE_PMD_MASK)) {
2151                 khugepaged_enter(vma, vma->vm_flags);
2152         }
2153         return 0;
2154 }
2155 
2156 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2157                                      umode_t mode, dev_t dev, unsigned long flags)
2158 {
2159         struct inode *inode;
2160         struct shmem_inode_info *info;
2161         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2162 
2163         if (shmem_reserve_inode(sb))
2164                 return NULL;
2165 
2166         inode = new_inode(sb);
2167         if (inode) {
2168                 inode->i_ino = get_next_ino();
2169                 inode_init_owner(inode, dir, mode);
2170                 inode->i_blocks = 0;
2171                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2172                 inode->i_generation = get_seconds();
2173                 info = SHMEM_I(inode);
2174                 memset(info, 0, (char *)inode - (char *)info);
2175                 spin_lock_init(&info->lock);
2176                 info->seals = F_SEAL_SEAL;
2177                 info->flags = flags & VM_NORESERVE;
2178                 INIT_LIST_HEAD(&info->shrinklist);
2179                 INIT_LIST_HEAD(&info->swaplist);
2180                 simple_xattrs_init(&info->xattrs);
2181                 cache_no_acl(inode);
2182 
2183                 switch (mode & S_IFMT) {
2184                 default:
2185                         inode->i_op = &shmem_special_inode_operations;
2186                         init_special_inode(inode, mode, dev);
2187                         break;
2188                 case S_IFREG:
2189                         inode->i_mapping->a_ops = &shmem_aops;
2190                         inode->i_op = &shmem_inode_operations;
2191                         inode->i_fop = &shmem_file_operations;
2192                         mpol_shared_policy_init(&info->policy,
2193                                                  shmem_get_sbmpol(sbinfo));
2194                         break;
2195                 case S_IFDIR:
2196                         inc_nlink(inode);
2197                         /* Some things misbehave if size == 0 on a directory */
2198                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
2199                         inode->i_op = &shmem_dir_inode_operations;
2200                         inode->i_fop = &simple_dir_operations;
2201                         break;
2202                 case S_IFLNK:
2203                         /*
2204                          * Must not load anything in the rbtree,
2205                          * mpol_free_shared_policy will not be called.
2206                          */
2207                         mpol_shared_policy_init(&info->policy, NULL);
2208                         break;
2209                 }
2210         } else
2211                 shmem_free_inode(sb);
2212         return inode;
2213 }
2214 
2215 bool shmem_mapping(struct address_space *mapping)
2216 {
2217         return mapping->a_ops == &shmem_aops;
2218 }
2219 
2220 static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2221                                   pmd_t *dst_pmd,
2222                                   struct vm_area_struct *dst_vma,
2223                                   unsigned long dst_addr,
2224                                   unsigned long src_addr,
2225                                   bool zeropage,
2226                                   struct page **pagep)
2227 {
2228         struct inode *inode = file_inode(dst_vma->vm_file);
2229         struct shmem_inode_info *info = SHMEM_I(inode);
2230         struct address_space *mapping = inode->i_mapping;
2231         gfp_t gfp = mapping_gfp_mask(mapping);
2232         pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2233         struct mem_cgroup *memcg;
2234         spinlock_t *ptl;
2235         void *page_kaddr;
2236         struct page *page;
2237         pte_t _dst_pte, *dst_pte;
2238         int ret;
2239 
2240         ret = -ENOMEM;
2241         if (!shmem_inode_acct_block(inode, 1))
2242                 goto out;
2243 
2244         if (!*pagep) {
2245                 page = shmem_alloc_page(gfp, info, pgoff);
2246                 if (!page)
2247                         goto out_unacct_blocks;
2248 
2249                 if (!zeropage) {        /* mcopy_atomic */
2250                         page_kaddr = kmap_atomic(page);
2251                         ret = copy_from_user(page_kaddr,
2252                                              (const void __user *)src_addr,
2253                                              PAGE_SIZE);
2254                         kunmap_atomic(page_kaddr);
2255 
2256                         /* fallback to copy_from_user outside mmap_sem */
2257                         if (unlikely(ret)) {
2258                                 *pagep = page;
2259                                 shmem_inode_unacct_blocks(inode, 1);
2260                                 /* don't free the page */
2261                                 return -EFAULT;
2262                         }
2263                 } else {                /* mfill_zeropage_atomic */
2264                         clear_highpage(page);
2265                 }
2266         } else {
2267                 page = *pagep;
2268                 *pagep = NULL;
2269         }
2270 
2271         VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2272         __SetPageLocked(page);
2273         __SetPageSwapBacked(page);
2274         __SetPageUptodate(page);
2275 
2276         ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false);
2277         if (ret)
2278                 goto out_release;
2279 
2280         ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
2281         if (!ret) {
2282                 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL);
2283                 radix_tree_preload_end();
2284         }
2285         if (ret)
2286                 goto out_release_uncharge;
2287 
2288         mem_cgroup_commit_charge(page, memcg, false, false);
2289 
2290         _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2291         if (dst_vma->vm_flags & VM_WRITE)
2292                 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2293 
2294         ret = -EEXIST;
2295         dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2296         if (!pte_none(*dst_pte))
2297                 goto out_release_uncharge_unlock;
2298 
2299         lru_cache_add_anon(page);
2300 
2301         spin_lock(&info->lock);
2302         info->alloced++;
2303         inode->i_blocks += BLOCKS_PER_PAGE;
2304         shmem_recalc_inode(inode);
2305         spin_unlock(&info->lock);
2306 
2307         inc_mm_counter(dst_mm, mm_counter_file(page));
2308         page_add_file_rmap(page, false);
2309         set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2310 
2311         /* No need to invalidate - it was non-present before */
2312         update_mmu_cache(dst_vma, dst_addr, dst_pte);
2313         unlock_page(page);
2314         pte_unmap_unlock(dst_pte, ptl);
2315         ret = 0;
2316 out:
2317         return ret;
2318 out_release_uncharge_unlock:
2319         pte_unmap_unlock(dst_pte, ptl);
2320 out_release_uncharge:
2321         mem_cgroup_cancel_charge(page, memcg, false);
2322 out_release:
2323         unlock_page(page);
2324         put_page(page);
2325 out_unacct_blocks:
2326         shmem_inode_unacct_blocks(inode, 1);
2327         goto out;
2328 }
2329 
2330 int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2331                            pmd_t *dst_pmd,
2332                            struct vm_area_struct *dst_vma,
2333                            unsigned long dst_addr,
2334                            unsigned long src_addr,
2335                            struct page **pagep)
2336 {
2337         return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2338                                       dst_addr, src_addr, false, pagep);
2339 }
2340 
2341 int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2342                              pmd_t *dst_pmd,
2343                              struct vm_area_struct *dst_vma,
2344                              unsigned long dst_addr)
2345 {
2346         struct page *page = NULL;
2347 
2348         return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2349                                       dst_addr, 0, true, &page);
2350 }
2351 
2352 #ifdef CONFIG_TMPFS
2353 static const struct inode_operations shmem_symlink_inode_operations;
2354 static const struct inode_operations shmem_short_symlink_operations;
2355 
2356 #ifdef CONFIG_TMPFS_XATTR
2357 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2358 #else
2359 #define shmem_initxattrs NULL
2360 #endif
2361 
2362 static int
2363 shmem_write_begin(struct file *file, struct address_space *mapping,
2364                         loff_t pos, unsigned len, unsigned flags,
2365                         struct page **pagep, void **fsdata)
2366 {
2367         struct inode *inode = mapping->host;
2368         struct shmem_inode_info *info = SHMEM_I(inode);
2369         pgoff_t index = pos >> PAGE_SHIFT;
2370 
2371         /* i_mutex is held by caller */
2372         if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) {
2373                 if (info->seals & F_SEAL_WRITE)
2374                         return -EPERM;
2375                 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2376                         return -EPERM;
2377         }
2378 
2379         return shmem_getpage(inode, index, pagep, SGP_WRITE);
2380 }
2381 
2382 static int
2383 shmem_write_end(struct file *file, struct address_space *mapping,
2384                         loff_t pos, unsigned len, unsigned copied,
2385                         struct page *page, void *fsdata)
2386 {
2387         struct inode *inode = mapping->host;
2388 
2389         if (pos + copied > inode->i_size)
2390                 i_size_write(inode, pos + copied);
2391 
2392         if (!PageUptodate(page)) {
2393                 struct page *head = compound_head(page);
2394                 if (PageTransCompound(page)) {
2395                         int i;
2396 
2397                         for (i = 0; i < HPAGE_PMD_NR; i++) {
2398                                 if (head + i == page)
2399                                         continue;
2400                                 clear_highpage(head + i);
2401                                 flush_dcache_page(head + i);
2402                         }
2403                 }
2404                 if (copied < PAGE_SIZE) {
2405                         unsigned from = pos & (PAGE_SIZE - 1);
2406                         zero_user_segments(page, 0, from,
2407                                         from + copied, PAGE_SIZE);
2408                 }
2409                 SetPageUptodate(head);
2410         }
2411         set_page_dirty(page);
2412         unlock_page(page);
2413         put_page(page);
2414 
2415         return copied;
2416 }
2417 
2418 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2419 {
2420         struct file *file = iocb->ki_filp;
2421         struct inode *inode = file_inode(file);
2422         struct address_space *mapping = inode->i_mapping;
2423         pgoff_t index;
2424         unsigned long offset;
2425         enum sgp_type sgp = SGP_READ;
2426         int error = 0;
2427         ssize_t retval = 0;
2428         loff_t *ppos = &iocb->ki_pos;
2429 
2430         /*
2431          * Might this read be for a stacking filesystem?  Then when reading
2432          * holes of a sparse file, we actually need to allocate those pages,
2433          * and even mark them dirty, so it cannot exceed the max_blocks limit.
2434          */
2435         if (!iter_is_iovec(to))
2436                 sgp = SGP_CACHE;
2437 
2438         index = *ppos >> PAGE_SHIFT;
2439         offset = *ppos & ~PAGE_MASK;
2440 
2441         for (;;) {
2442                 struct page *page = NULL;
2443                 pgoff_t end_index;
2444                 unsigned long nr, ret;
2445                 loff_t i_size = i_size_read(inode);
2446 
2447                 end_index = i_size >> PAGE_SHIFT;
2448                 if (index > end_index)
2449                         break;
2450                 if (index == end_index) {
2451                         nr = i_size & ~PAGE_MASK;
2452                         if (nr <= offset)
2453                                 break;
2454                 }
2455 
2456                 error = shmem_getpage(inode, index, &page, sgp);
2457                 if (error) {
2458                         if (error == -EINVAL)
2459                                 error = 0;
2460                         break;
2461                 }
2462                 if (page) {
2463                         if (sgp == SGP_CACHE)
2464                                 set_page_dirty(page);
2465                         unlock_page(page);
2466                 }
2467 
2468                 /*
2469                  * We must evaluate after, since reads (unlike writes)
2470                  * are called without i_mutex protection against truncate
2471                  */
2472                 nr = PAGE_SIZE;
2473                 i_size = i_size_read(inode);
2474                 end_index = i_size >> PAGE_SHIFT;
2475                 if (index == end_index) {
2476                         nr = i_size & ~PAGE_MASK;
2477                         if (nr <= offset) {
2478                                 if (page)
2479                                         put_page(page);
2480                                 break;
2481                         }
2482                 }
2483                 nr -= offset;
2484 
2485                 if (page) {
2486                         /*
2487                          * If users can be writing to this page using arbitrary
2488                          * virtual addresses, take care about potential aliasing
2489                          * before reading the page on the kernel side.
2490                          */
2491                         if (mapping_writably_mapped(mapping))
2492                                 flush_dcache_page(page);
2493                         /*
2494                          * Mark the page accessed if we read the beginning.
2495                          */
2496                         if (!offset)
2497                                 mark_page_accessed(page);
2498                 } else {
2499                         page = ZERO_PAGE(0);
2500                         get_page(page);
2501                 }
2502 
2503                 /*
2504                  * Ok, we have the page, and it's up-to-date, so
2505                  * now we can copy it to user space...
2506                  */
2507                 ret = copy_page_to_iter(page, offset, nr, to);
2508                 retval += ret;
2509                 offset += ret;
2510                 index += offset >> PAGE_SHIFT;
2511                 offset &= ~PAGE_MASK;
2512 
2513                 put_page(page);
2514                 if (!iov_iter_count(to))
2515                         break;
2516                 if (ret < nr) {
2517                         error = -EFAULT;
2518                         break;
2519                 }
2520                 cond_resched();
2521         }
2522 
2523         *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2524         file_accessed(file);
2525         return retval ? retval : error;
2526 }
2527 
2528 /*
2529  * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2530  */
2531 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2532                                     pgoff_t index, pgoff_t end, int whence)
2533 {
2534         struct page *page;
2535         struct pagevec pvec;
2536         pgoff_t indices[PAGEVEC_SIZE];
2537         bool done = false;
2538         int i;
2539 
2540         pagevec_init(&pvec);
2541         pvec.nr = 1;            /* start small: we may be there already */
2542         while (!done) {
2543                 pvec.nr = find_get_entries(mapping, index,
2544                                         pvec.nr, pvec.pages, indices);
2545                 if (!pvec.nr) {
2546                         if (whence == SEEK_DATA)
2547                                 index = end;
2548                         break;
2549                 }
2550                 for (i = 0; i < pvec.nr; i++, index++) {
2551                         if (index < indices[i]) {
2552                                 if (whence == SEEK_HOLE) {
2553                                         done = true;
2554                                         break;
2555                                 }
2556                                 index = indices[i];
2557                         }
2558                         page = pvec.pages[i];
2559                         if (page && !radix_tree_exceptional_entry(page)) {
2560                                 if (!PageUptodate(page))
2561                                         page = NULL;
2562                         }
2563                         if (index >= end ||
2564                             (page && whence == SEEK_DATA) ||
2565                             (!page && whence == SEEK_HOLE)) {
2566                                 done = true;
2567                                 break;
2568                         }
2569                 }
2570                 pagevec_remove_exceptionals(&pvec);
2571                 pagevec_release(&pvec);
2572                 pvec.nr = PAGEVEC_SIZE;
2573                 cond_resched();
2574         }
2575         return index;
2576 }
2577 
2578 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2579 {
2580         struct address_space *mapping = file->f_mapping;
2581         struct inode *inode = mapping->host;
2582         pgoff_t start, end;
2583         loff_t new_offset;
2584 
2585         if (whence != SEEK_DATA && whence != SEEK_HOLE)
2586                 return generic_file_llseek_size(file, offset, whence,
2587                                         MAX_LFS_FILESIZE, i_size_read(inode));
2588         inode_lock(inode);
2589         /* We're holding i_mutex so we can access i_size directly */
2590 
2591         if (offset < 0)
2592                 offset = -EINVAL;
2593         else if (offset >= inode->i_size)
2594                 offset = -ENXIO;
2595         else {
2596                 start = offset >> PAGE_SHIFT;
2597                 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2598                 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2599                 new_offset <<= PAGE_SHIFT;
2600                 if (new_offset > offset) {
2601                         if (new_offset < inode->i_size)
2602                                 offset = new_offset;
2603                         else if (whence == SEEK_DATA)
2604                                 offset = -ENXIO;
2605                         else
2606                                 offset = inode->i_size;
2607                 }
2608         }
2609 
2610         if (offset >= 0)
2611                 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2612         inode_unlock(inode);
2613         return offset;
2614 }
2615 
2616 /*
2617  * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2618  * so reuse a tag which we firmly believe is never set or cleared on shmem.
2619  */
2620 #define SHMEM_TAG_PINNED        PAGECACHE_TAG_TOWRITE
2621 #define LAST_SCAN               4       /* about 150ms max */
2622 
2623 static void shmem_tag_pins(struct address_space *mapping)
2624 {
2625         struct radix_tree_iter iter;
2626         void **slot;
2627         pgoff_t start;
2628         struct page *page;
2629 
2630         lru_add_drain();
2631         start = 0;
2632         rcu_read_lock();
2633 
2634         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2635                 page = radix_tree_deref_slot(slot);
2636                 if (!page || radix_tree_exception(page)) {
2637                         if (radix_tree_deref_retry(page)) {
2638                                 slot = radix_tree_iter_retry(&iter);
2639                                 continue;
2640                         }
2641                 } else if (page_count(page) - page_mapcount(page) > 1) {
2642                         spin_lock_irq(&mapping->tree_lock);
2643                         radix_tree_tag_set(&mapping->page_tree, iter.index,
2644                                            SHMEM_TAG_PINNED);
2645                         spin_unlock_irq(&mapping->tree_lock);
2646                 }
2647 
2648                 if (need_resched()) {
2649                         slot = radix_tree_iter_resume(slot, &iter);
2650                         cond_resched_rcu();
2651                 }
2652         }
2653         rcu_read_unlock();
2654 }
2655 
2656 /*
2657  * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2658  * via get_user_pages(), drivers might have some pending I/O without any active
2659  * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2660  * and see whether it has an elevated ref-count. If so, we tag them and wait for
2661  * them to be dropped.
2662  * The caller must guarantee that no new user will acquire writable references
2663  * to those pages to avoid races.
2664  */
2665 static int shmem_wait_for_pins(struct address_space *mapping)
2666 {
2667         struct radix_tree_iter iter;
2668         void **slot;
2669         pgoff_t start;
2670         struct page *page;
2671         int error, scan;
2672 
2673         shmem_tag_pins(mapping);
2674 
2675         error = 0;
2676         for (scan = 0; scan <= LAST_SCAN; scan++) {
2677                 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2678                         break;
2679 
2680                 if (!scan)
2681                         lru_add_drain_all();
2682                 else if (schedule_timeout_killable((HZ << scan) / 200))
2683                         scan = LAST_SCAN;
2684 
2685                 start = 0;
2686                 rcu_read_lock();
2687                 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2688                                            start, SHMEM_TAG_PINNED) {
2689 
2690                         page = radix_tree_deref_slot(slot);
2691                         if (radix_tree_exception(page)) {
2692                                 if (radix_tree_deref_retry(page)) {
2693                                         slot = radix_tree_iter_retry(&iter);
2694                                         continue;
2695                                 }
2696 
2697                                 page = NULL;
2698                         }
2699 
2700                         if (page &&
2701                             page_count(page) - page_mapcount(page) != 1) {
2702                                 if (scan < LAST_SCAN)
2703                                         goto continue_resched;
2704 
2705                                 /*
2706                                  * On the last scan, we clean up all those tags
2707                                  * we inserted; but make a note that we still
2708                                  * found pages pinned.
2709                                  */
2710                                 error = -EBUSY;
2711                         }
2712 
2713                         spin_lock_irq(&mapping->tree_lock);
2714                         radix_tree_tag_clear(&mapping->page_tree,
2715                                              iter.index, SHMEM_TAG_PINNED);
2716                         spin_unlock_irq(&mapping->tree_lock);
2717 continue_resched:
2718                         if (need_resched()) {
2719                                 slot = radix_tree_iter_resume(slot, &iter);
2720                                 cond_resched_rcu();
2721                         }
2722                 }
2723                 rcu_read_unlock();
2724         }
2725 
2726         return error;
2727 }
2728 
2729 static unsigned int *memfd_file_seals_ptr(struct file *file)
2730 {
2731         if (file->f_op == &shmem_file_operations)
2732                 return &SHMEM_I(file_inode(file))->seals;
2733 
2734 #ifdef CONFIG_HUGETLBFS
2735         if (file->f_op == &hugetlbfs_file_operations)
2736                 return &HUGETLBFS_I(file_inode(file))->seals;
2737 #endif
2738 
2739         return NULL;
2740 }
2741 
2742 #define F_ALL_SEALS (F_SEAL_SEAL | \
2743                      F_SEAL_SHRINK | \
2744                      F_SEAL_GROW | \
2745                      F_SEAL_WRITE)
2746 
2747 static int memfd_add_seals(struct file *file, unsigned int seals)
2748 {
2749         struct inode *inode = file_inode(file);
2750         unsigned int *file_seals;
2751         int error;
2752 
2753         /*
2754          * SEALING
2755          * Sealing allows multiple parties to share a shmem-file but restrict
2756          * access to a specific subset of file operations. Seals can only be
2757          * added, but never removed. This way, mutually untrusted parties can
2758          * share common memory regions with a well-defined policy. A malicious
2759          * peer can thus never perform unwanted operations on a shared object.
2760          *
2761          * Seals are only supported on special shmem-files and always affect
2762          * the whole underlying inode. Once a seal is set, it may prevent some
2763          * kinds of access to the file. Currently, the following seals are
2764          * defined:
2765          *   SEAL_SEAL: Prevent further seals from being set on this file
2766          *   SEAL_SHRINK: Prevent the file from shrinking
2767          *   SEAL_GROW: Prevent the file from growing
2768          *   SEAL_WRITE: Prevent write access to the file
2769          *
2770          * As we don't require any trust relationship between two parties, we
2771          * must prevent seals from being removed. Therefore, sealing a file
2772          * only adds a given set of seals to the file, it never touches
2773          * existing seals. Furthermore, the "setting seals"-operation can be
2774          * sealed itself, which basically prevents any further seal from being
2775          * added.
2776          *
2777          * Semantics of sealing are only defined on volatile files. Only
2778          * anonymous shmem files support sealing. More importantly, seals are
2779          * never written to disk. Therefore, there's no plan to support it on
2780          * other file types.
2781          */
2782 
2783         if (!(file->f_mode & FMODE_WRITE))
2784                 return -EPERM;
2785         if (seals & ~(unsigned int)F_ALL_SEALS)
2786                 return -EINVAL;
2787 
2788         inode_lock(inode);
2789 
2790         file_seals = memfd_file_seals_ptr(file);
2791         if (!file_seals) {
2792                 error = -EINVAL;
2793                 goto unlock;
2794         }
2795 
2796         if (*file_seals & F_SEAL_SEAL) {
2797                 error = -EPERM;
2798                 goto unlock;
2799         }
2800 
2801         if ((seals & F_SEAL_WRITE) && !(*file_seals & F_SEAL_WRITE)) {
2802                 error = mapping_deny_writable(file->f_mapping);
2803                 if (error)
2804                         goto unlock;
2805 
2806                 error = shmem_wait_for_pins(file->f_mapping);
2807                 if (error) {
2808                         mapping_allow_writable(file->f_mapping);
2809                         goto unlock;
2810                 }
2811         }
2812 
2813         *file_seals |= seals;
2814         error = 0;
2815 
2816 unlock:
2817         inode_unlock(inode);
2818         return error;
2819 }
2820 
2821 static int memfd_get_seals(struct file *file)
2822 {
2823         unsigned int *seals = memfd_file_seals_ptr(file);
2824 
2825         return seals ? *seals : -EINVAL;
2826 }
2827 
2828 long memfd_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2829 {
2830         long error;
2831 
2832         switch (cmd) {
2833         case F_ADD_SEALS:
2834                 /* disallow upper 32bit */
2835                 if (arg > UINT_MAX)
2836                         return -EINVAL;
2837 
2838                 error = memfd_add_seals(file, arg);
2839                 break;
2840         case F_GET_SEALS:
2841                 error = memfd_get_seals(file);
2842                 break;
2843         default:
2844                 error = -EINVAL;
2845                 break;
2846         }
2847 
2848         return error;
2849 }
2850 
2851 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2852                                                          loff_t len)
2853 {
2854         struct inode *inode = file_inode(file);
2855         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2856         struct shmem_inode_info *info = SHMEM_I(inode);
2857         struct shmem_falloc shmem_falloc;
2858         pgoff_t start, index, end;
2859         int error;
2860 
2861         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2862                 return -EOPNOTSUPP;
2863 
2864         inode_lock(inode);
2865 
2866         if (mode & FALLOC_FL_PUNCH_HOLE) {
2867                 struct address_space *mapping = file->f_mapping;
2868                 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2869                 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2870                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2871 
2872                 /* protected by i_mutex */
2873                 if (info->seals & F_SEAL_WRITE) {
2874                         error = -EPERM;
2875                         goto out;
2876                 }
2877 
2878                 shmem_falloc.waitq = &shmem_falloc_waitq;
2879                 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2880                 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2881                 spin_lock(&inode->i_lock);
2882                 inode->i_private = &shmem_falloc;
2883                 spin_unlock(&inode->i_lock);
2884 
2885                 if ((u64)unmap_end > (u64)unmap_start)
2886                         unmap_mapping_range(mapping, unmap_start,
2887                                             1 + unmap_end - unmap_start, 0);
2888                 shmem_truncate_range(inode, offset, offset + len - 1);
2889                 /* No need to unmap again: hole-punching leaves COWed pages */
2890 
2891                 spin_lock(&inode->i_lock);
2892                 inode->i_private = NULL;
2893                 wake_up_all(&shmem_falloc_waitq);
2894                 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2895                 spin_unlock(&inode->i_lock);
2896                 error = 0;
2897                 goto out;
2898         }
2899 
2900         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2901         error = inode_newsize_ok(inode, offset + len);
2902         if (error)
2903                 goto out;
2904 
2905         if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2906                 error = -EPERM;
2907                 goto out;
2908         }
2909 
2910         start = offset >> PAGE_SHIFT;
2911         end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2912         /* Try to avoid a swapstorm if len is impossible to satisfy */
2913         if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2914                 error = -ENOSPC;
2915                 goto out;
2916         }
2917 
2918         shmem_falloc.waitq = NULL;
2919         shmem_falloc.start = start;
2920         shmem_falloc.next  = start;
2921         shmem_falloc.nr_falloced = 0;
2922         shmem_falloc.nr_unswapped = 0;
2923         spin_lock(&inode->i_lock);
2924         inode->i_private = &shmem_falloc;
2925         spin_unlock(&inode->i_lock);
2926 
2927         for (index = start; index < end; index++) {
2928                 struct page *page;
2929 
2930                 /*
2931                  * Good, the fallocate(2) manpage permits EINTR: we may have
2932                  * been interrupted because we are using up too much memory.
2933                  */
2934                 if (signal_pending(current))
2935                         error = -EINTR;
2936                 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2937                         error = -ENOMEM;
2938                 else
2939                         error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2940                 if (error) {
2941                         /* Remove the !PageUptodate pages we added */
2942                         if (index > start) {
2943                                 shmem_undo_range(inode,
2944                                     (loff_t)start << PAGE_SHIFT,
2945                                     ((loff_t)index << PAGE_SHIFT) - 1, true);
2946                         }
2947                         goto undone;
2948                 }
2949 
2950                 /*
2951                  * Inform shmem_writepage() how far we have reached.
2952                  * No need for lock or barrier: we have the page lock.
2953                  */
2954                 shmem_falloc.next++;
2955                 if (!PageUptodate(page))
2956                         shmem_falloc.nr_falloced++;
2957 
2958                 /*
2959                  * If !PageUptodate, leave it that way so that freeable pages
2960                  * can be recognized if we need to rollback on error later.
2961                  * But set_page_dirty so that memory pressure will swap rather
2962                  * than free the pages we are allocating (and SGP_CACHE pages
2963                  * might still be clean: we now need to mark those dirty too).
2964                  */
2965                 set_page_dirty(page);
2966                 unlock_page(page);
2967                 put_page(page);
2968                 cond_resched();
2969         }
2970 
2971         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2972                 i_size_write(inode, offset + len);
2973         inode->i_ctime = current_time(inode);
2974 undone:
2975         spin_lock(&inode->i_lock);
2976         inode->i_private = NULL;
2977         spin_unlock(&inode->i_lock);
2978 out:
2979         inode_unlock(inode);
2980         return error;
2981 }
2982 
2983 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2984 {
2985         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2986 
2987         buf->f_type = TMPFS_MAGIC;
2988         buf->f_bsize = PAGE_SIZE;
2989         buf->f_namelen = NAME_MAX;
2990         if (sbinfo->max_blocks) {
2991                 buf->f_blocks = sbinfo->max_blocks;
2992                 buf->f_bavail =
2993                 buf->f_bfree  = sbinfo->max_blocks -
2994                                 percpu_counter_sum(&sbinfo->used_blocks);
2995         }
2996         if (sbinfo->max_inodes) {
2997                 buf->f_files = sbinfo->max_inodes;
2998                 buf->f_ffree = sbinfo->free_inodes;
2999         }
3000         /* else leave those fields 0 like simple_statfs */
3001         return 0;
3002 }
3003 
3004 /*
3005  * File creation. Allocate an inode, and we're done..
3006  */
3007 static int
3008 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3009 {
3010         struct inode *inode;
3011         int error = -ENOSPC;
3012 
3013         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
3014         if (inode) {
3015                 error = simple_acl_create(dir, inode);
3016                 if (error)
3017                         goto out_iput;
3018                 error = security_inode_init_security(inode, dir,
3019                                                      &dentry->d_name,
3020                                                      shmem_initxattrs, NULL);
3021                 if (error && error != -EOPNOTSUPP)
3022                         goto out_iput;
3023 
3024                 error = 0;
3025                 dir->i_size += BOGO_DIRENT_SIZE;
3026                 dir->i_ctime = dir->i_mtime = current_time(dir);
3027                 d_instantiate(dentry, inode);
3028                 dget(dentry); /* Extra count - pin the dentry in core */
3029         }
3030         return error;
3031 out_iput:
3032         iput(inode);
3033         return error;
3034 }
3035 
3036 static int
3037 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
3038 {
3039         struct inode *inode;
3040         int error = -ENOSPC;
3041 
3042         inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
3043         if (inode) {
3044                 error = security_inode_init_security(inode, dir,
3045                                                      NULL,
3046                                                      shmem_initxattrs, NULL);
3047                 if (error && error != -EOPNOTSUPP)
3048                         goto out_iput;
3049                 error = simple_acl_create(dir, inode);
3050                 if (error)
3051                         goto out_iput;
3052                 d_tmpfile(dentry, inode);
3053         }
3054         return error;
3055 out_iput:
3056         iput(inode);
3057         return error;
3058 }
3059 
3060 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3061 {
3062         int error;
3063 
3064         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
3065                 return error;
3066         inc_nlink(dir);
3067         return 0;
3068 }
3069 
3070 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
3071                 bool excl)
3072 {
3073         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
3074 }
3075 
3076 /*
3077  * Link a file..
3078  */
3079 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3080 {
3081         struct inode *inode = d_inode(old_dentry);
3082         int ret;
3083 
3084         /*
3085          * No ordinary (disk based) filesystem counts links as inodes;
3086          * but each new link needs a new dentry, pinning lowmem, and
3087          * tmpfs dentries cannot be pruned until they are unlinked.
3088          */
3089         ret = shmem_reserve_inode(inode->i_sb);
3090         if (ret)
3091                 goto out;
3092 
3093         dir->i_size += BOGO_DIRENT_SIZE;
3094         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3095         inc_nlink(inode);
3096         ihold(inode);   /* New dentry reference */
3097         dget(dentry);           /* Extra pinning count for the created dentry */
3098         d_instantiate(dentry, inode);
3099 out:
3100         return ret;
3101 }
3102 
3103 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3104 {
3105         struct inode *inode = d_inode(dentry);
3106 
3107         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3108                 shmem_free_inode(inode->i_sb);
3109 
3110         dir->i_size -= BOGO_DIRENT_SIZE;
3111         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3112         drop_nlink(inode);
3113         dput(dentry);   /* Undo the count from "create" - this does all the work */
3114         return 0;
3115 }
3116 
3117 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3118 {
3119         if (!simple_empty(dentry))
3120                 return -ENOTEMPTY;
3121 
3122         drop_nlink(d_inode(dentry));
3123         drop_nlink(dir);
3124         return shmem_unlink(dir, dentry);
3125 }
3126 
3127 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3128 {
3129         bool old_is_dir = d_is_dir(old_dentry);
3130         bool new_is_dir = d_is_dir(new_dentry);
3131 
3132         if (old_dir != new_dir && old_is_dir != new_is_dir) {
3133                 if (old_is_dir) {
3134                         drop_nlink(old_dir);
3135                         inc_nlink(new_dir);
3136                 } else {
3137                         drop_nlink(new_dir);
3138                         inc_nlink(old_dir);
3139                 }
3140         }
3141         old_dir->i_ctime = old_dir->i_mtime =
3142         new_dir->i_ctime = new_dir->i_mtime =
3143         d_inode(old_dentry)->i_ctime =
3144         d_inode(new_dentry)->i_ctime = current_time(old_dir);
3145 
3146         return 0;
3147 }
3148 
3149 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3150 {
3151         struct dentry *whiteout;
3152         int error;
3153 
3154         whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3155         if (!whiteout)
3156                 return -ENOMEM;
3157 
3158         error = shmem_mknod(old_dir, whiteout,
3159                             S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3160         dput(whiteout);
3161         if (error)
3162                 return error;
3163 
3164         /*
3165          * Cheat and hash the whiteout while the old dentry is still in
3166          * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3167          *
3168          * d_lookup() will consistently find one of them at this point,
3169          * not sure which one, but that isn't even important.
3170          */
3171         d_rehash(whiteout);
3172         return 0;
3173 }
3174 
3175 /*
3176  * The VFS layer already does all the dentry stuff for rename,
3177  * we just have to decrement the usage count for the target if
3178  * it exists so that the VFS layer correctly free's it when it
3179  * gets overwritten.
3180  */
3181 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3182 {
3183         struct inode *inode = d_inode(old_dentry);
3184         int they_are_dirs = S_ISDIR(inode->i_mode);
3185 
3186         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3187                 return -EINVAL;
3188 
3189         if (flags & RENAME_EXCHANGE)
3190                 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3191 
3192         if (!simple_empty(new_dentry))
3193                 return -ENOTEMPTY;
3194 
3195         if (flags & RENAME_WHITEOUT) {
3196                 int error;
3197 
3198                 error = shmem_whiteout(old_dir, old_dentry);
3199                 if (error)
3200                         return error;
3201         }
3202 
3203         if (d_really_is_positive(new_dentry)) {
3204                 (void) shmem_unlink(new_dir, new_dentry);
3205                 if (they_are_dirs) {
3206                         drop_nlink(d_inode(new_dentry));
3207                         drop_nlink(old_dir);
3208                 }
3209         } else if (they_are_dirs) {
3210                 drop_nlink(old_dir);
3211                 inc_nlink(new_dir);
3212         }
3213 
3214         old_dir->i_size -= BOGO_DIRENT_SIZE;
3215         new_dir->i_size += BOGO_DIRENT_SIZE;
3216         old_dir->i_ctime = old_dir->i_mtime =
3217         new_dir->i_ctime = new_dir->i_mtime =
3218         inode->i_ctime = current_time(old_dir);
3219         return 0;
3220 }
3221 
3222 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3223 {
3224         int error;
3225         int len;
3226         struct inode *inode;
3227         struct page *page;
3228 
3229         len = strlen(symname) + 1;
3230         if (len > PAGE_SIZE)
3231                 return -ENAMETOOLONG;
3232 
3233         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3234         if (!inode)
3235                 return -ENOSPC;
3236 
3237         error = security_inode_init_security(inode, dir, &dentry->d_name,
3238                                              shmem_initxattrs, NULL);
3239         if (error) {
3240                 if (error != -EOPNOTSUPP) {
3241                         iput(inode);
3242                         return error;
3243                 }
3244                 error = 0;
3245         }
3246 
3247         inode->i_size = len-1;
3248         if (len <= SHORT_SYMLINK_LEN) {
3249                 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3250                 if (!inode->i_link) {
3251                         iput(inode);
3252                         return -ENOMEM;
3253                 }
3254                 inode->i_op = &shmem_short_symlink_operations;
3255         } else {
3256                 inode_nohighmem(inode);
3257                 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3258                 if (error) {
3259                         iput(inode);
3260                         return error;
3261                 }
3262                 inode->i_mapping->a_ops = &shmem_aops;
3263                 inode->i_op = &shmem_symlink_inode_operations;
3264                 memcpy(page_address(page), symname, len);
3265                 SetPageUptodate(page);
3266                 set_page_dirty(page);
3267                 unlock_page(page);
3268                 put_page(page);
3269         }
3270         dir->i_size += BOGO_DIRENT_SIZE;
3271         dir->i_ctime = dir->i_mtime = current_time(dir);
3272         d_instantiate(dentry, inode);
3273         dget(dentry);
3274         return 0;
3275 }
3276 
3277 static void shmem_put_link(void *arg)
3278 {
3279         mark_page_accessed(arg);
3280         put_page(arg);
3281 }
3282 
3283 static const char *shmem_get_link(struct dentry *dentry,
3284                                   struct inode *inode,
3285                                   struct delayed_call *done)
3286 {
3287         struct page *page = NULL;
3288         int error;
3289         if (!dentry) {
3290                 page = find_get_page(inode->i_mapping, 0);
3291                 if (!page)
3292                         return ERR_PTR(-ECHILD);
3293                 if (!PageUptodate(page)) {
3294                         put_page(page);
3295                         return ERR_PTR(-ECHILD);
3296                 }
3297         } else {
3298                 error = shmem_getpage(inode, 0, &page, SGP_READ);
3299                 if (error)
3300                         return ERR_PTR(error);
3301                 unlock_page(page);
3302         }
3303         set_delayed_call(done, shmem_put_link, page);
3304         return page_address(page);
3305 }
3306 
3307 #ifdef CONFIG_TMPFS_XATTR
3308 /*
3309  * Superblocks without xattr inode operations may get some security.* xattr
3310  * support from the LSM "for free". As soon as we have any other xattrs
3311  * like ACLs, we also need to implement the security.* handlers at
3312  * filesystem level, though.
3313  */
3314 
3315 /*
3316  * Callback for security_inode_init_security() for acquiring xattrs.
3317  */
3318 static int shmem_initxattrs(struct inode *inode,
3319                             const struct xattr *xattr_array,
3320                             void *fs_info)
3321 {
3322         struct shmem_inode_info *info = SHMEM_I(inode);
3323         const struct xattr *xattr;
3324         struct simple_xattr *new_xattr;
3325         size_t len;
3326 
3327         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3328                 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3329                 if (!new_xattr)
3330                         return -ENOMEM;
3331 
3332                 len = strlen(xattr->name) + 1;
3333                 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3334                                           GFP_KERNEL);
3335                 if (!new_xattr->name) {
3336                         kfree(new_xattr);
3337                         return -ENOMEM;
3338                 }
3339 
3340                 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3341                        XATTR_SECURITY_PREFIX_LEN);
3342                 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3343                        xattr->name, len);
3344 
3345                 simple_xattr_list_add(&info->xattrs, new_xattr);
3346         }
3347 
3348         return 0;
3349 }
3350 
3351 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3352                                    struct dentry *unused, struct inode *inode,
3353                                    const char *name, void *buffer, size_t size)
3354 {
3355         struct shmem_inode_info *info = SHMEM_I(inode);
3356 
3357         name = xattr_full_name(handler, name);
3358         return simple_xattr_get(&info->xattrs, name, buffer, size);
3359 }
3360 
3361 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3362                                    struct dentry *unused, struct inode *inode,
3363                                    const char *name, const void *value,
3364                                    size_t size, int flags)
3365 {
3366         struct shmem_inode_info *info = SHMEM_I(inode);
3367 
3368         name = xattr_full_name(handler, name);
3369         return simple_xattr_set(&info->xattrs, name, value, size, flags);
3370 }
3371 
3372 static const struct xattr_handler shmem_security_xattr_handler = {
3373         .prefix = XATTR_SECURITY_PREFIX,
3374         .get = shmem_xattr_handler_get,
3375         .set = shmem_xattr_handler_set,
3376 };
3377 
3378 static const struct xattr_handler shmem_trusted_xattr_handler = {
3379         .prefix = XATTR_TRUSTED_PREFIX,
3380         .get = shmem_xattr_handler_get,
3381         .set = shmem_xattr_handler_set,
3382 };
3383 
3384 static const struct xattr_handler *shmem_xattr_handlers[] = {
3385 #ifdef CONFIG_TMPFS_POSIX_ACL
3386         &posix_acl_access_xattr_handler,
3387         &posix_acl_default_xattr_handler,
3388 #endif
3389         &shmem_security_xattr_handler,
3390         &shmem_trusted_xattr_handler,
3391         NULL
3392 };
3393 
3394 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3395 {
3396         struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3397         return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3398 }
3399 #endif /* CONFIG_TMPFS_XATTR */
3400 
3401 static const struct inode_operations shmem_short_symlink_operations = {
3402         .get_link       = simple_get_link,
3403 #ifdef CONFIG_TMPFS_XATTR
3404         .listxattr      = shmem_listxattr,
3405 #endif
3406 };
3407 
3408 static const struct inode_operations shmem_symlink_inode_operations = {
3409         .get_link       = shmem_get_link,
3410 #ifdef CONFIG_TMPFS_XATTR
3411         .listxattr      = shmem_listxattr,
3412 #endif
3413 };
3414 
3415 static struct dentry *shmem_get_parent(struct dentry *child)
3416 {
3417         return ERR_PTR(-ESTALE);
3418 }
3419 
3420 static int shmem_match(struct inode *ino, void *vfh)
3421 {
3422         __u32 *fh = vfh;
3423         __u64 inum = fh[2];
3424         inum = (inum << 32) | fh[1];
3425         return ino->i_ino == inum && fh[0] == ino->i_generation;
3426 }
3427 
3428 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3429                 struct fid *fid, int fh_len, int fh_type)
3430 {
3431         struct inode *inode;
3432         struct dentry *dentry = NULL;
3433         u64 inum;
3434 
3435         if (fh_len < 3)
3436                 return NULL;
3437 
3438         inum = fid->raw[2];
3439         inum = (inum << 32) | fid->raw[1];
3440 
3441         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3442                         shmem_match, fid->raw);
3443         if (inode) {
3444                 dentry = d_find_alias(inode);
3445                 iput(inode);
3446         }
3447 
3448         return dentry;
3449 }
3450 
3451 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3452                                 struct inode *parent)
3453 {
3454         if (*len < 3) {
3455                 *len = 3;
3456                 return FILEID_INVALID;
3457         }
3458 
3459         if (inode_unhashed(inode)) {
3460                 /* Unfortunately insert_inode_hash is not idempotent,
3461                  * so as we hash inodes here rather than at creation
3462                  * time, we need a lock to ensure we only try
3463                  * to do it once
3464                  */
3465                 static DEFINE_SPINLOCK(lock);
3466                 spin_lock(&lock);
3467                 if (inode_unhashed(inode))
3468                         __insert_inode_hash(inode,
3469                                             inode->i_ino + inode->i_generation);
3470                 spin_unlock(&lock);
3471         }
3472 
3473         fh[0] = inode->i_generation;
3474         fh[1] = inode->i_ino;
3475         fh[2] = ((__u64)inode->i_ino) >> 32;
3476 
3477         *len = 3;
3478         return 1;
3479 }
3480 
3481 static const struct export_operations shmem_export_ops = {
3482         .get_parent     = shmem_get_parent,
3483         .encode_fh      = shmem_encode_fh,
3484         .fh_to_dentry   = shmem_fh_to_dentry,
3485 };
3486 
3487 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3488                                bool remount)
3489 {
3490         char *this_char, *value, *rest;
3491         struct mempolicy *mpol = NULL;
3492         uid_t uid;
3493         gid_t gid;
3494 
3495         while (options != NULL) {
3496                 this_char = options;
3497                 for (;;) {
3498                         /*
3499                          * NUL-terminate this option: unfortunately,
3500                          * mount options form a comma-separated list,
3501                          * but mpol's nodelist may also contain commas.
3502                          */
3503                         options = strchr(options, ',');
3504                         if (options == NULL)
3505                                 break;
3506                         options++;
3507                         if (!isdigit(*options)) {
3508                                 options[-1] = '\0';
3509                                 break;
3510                         }
3511                 }
3512                 if (!*this_char)
3513                         continue;
3514                 if ((value = strchr(this_char,'=')) != NULL) {
3515                         *value++ = 0;
3516                 } else {
3517                         pr_err("tmpfs: No value for mount option '%s'\n",
3518                                this_char);
3519                         goto error;
3520                 }
3521 
3522                 if (!strcmp(this_char,"size")) {
3523                         unsigned long long size;
3524                         size = memparse(value,&rest);
3525                         if (*rest == '%') {
3526                                 size <<= PAGE_SHIFT;
3527                                 size *= totalram_pages;
3528                                 do_div(size, 100);
3529                                 rest++;
3530                         }
3531                         if (*rest)
3532                                 goto bad_val;
3533                         sbinfo->max_blocks =
3534                                 DIV_ROUND_UP(size, PAGE_SIZE);
3535                 } else if (!strcmp(this_char,"nr_blocks")) {
3536                         sbinfo->max_blocks = memparse(value, &rest);
3537                         if (*rest)
3538                                 goto bad_val;
3539                 } else if (!strcmp(this_char,"nr_inodes")) {
3540                         sbinfo->max_inodes = memparse(value, &rest);
3541                         if (*rest)
3542                                 goto bad_val;
3543                 } else if (!strcmp(this_char,"mode")) {
3544                         if (remount)
3545                                 continue;
3546                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3547                         if (*rest)
3548                                 goto bad_val;
3549                 } else if (!strcmp(this_char,"uid")) {
3550                         if (remount)
3551                                 continue;
3552                         uid = simple_strtoul(value, &rest, 0);
3553                         if (*rest)
3554                                 goto bad_val;
3555                         sbinfo->uid = make_kuid(current_user_ns(), uid);
3556                         if (!uid_valid(sbinfo->uid))
3557                                 goto bad_val;
3558                 } else if (!strcmp(this_char,"gid")) {
3559                         if (remount)
3560                                 continue;
3561                         gid = simple_strtoul(value, &rest, 0);
3562                         if (*rest)
3563                                 goto bad_val;
3564                         sbinfo->gid = make_kgid(current_user_ns(), gid);
3565                         if (!gid_valid(sbinfo->gid))
3566                                 goto bad_val;
3567 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3568                 } else if (!strcmp(this_char, "huge")) {
3569                         int huge;
3570                         huge = shmem_parse_huge(value);
3571                         if (huge < 0)
3572                                 goto bad_val;
3573                         if (!has_transparent_hugepage() &&
3574                                         huge != SHMEM_HUGE_NEVER)
3575                                 goto bad_val;
3576                         sbinfo->huge = huge;
3577 #endif
3578 #ifdef CONFIG_NUMA
3579                 } else if (!strcmp(this_char,"mpol")) {
3580                         mpol_put(mpol);
3581                         mpol = NULL;
3582                         if (mpol_parse_str(value, &mpol))
3583                                 goto bad_val;
3584 #endif
3585                 } else {
3586                         pr_err("tmpfs: Bad mount option %s\n", this_char);
3587                         goto error;
3588                 }
3589         }
3590         sbinfo->mpol = mpol;
3591         return 0;
3592 
3593 bad_val:
3594         pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3595                value, this_char);
3596 error:
3597         mpol_put(mpol);
3598         return 1;
3599 
3600 }
3601 
3602 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3603 {
3604         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3605         struct shmem_sb_info config = *sbinfo;
3606         unsigned long inodes;
3607         int error = -EINVAL;
3608 
3609         config.mpol = NULL;
3610         if (shmem_parse_options(data, &config, true))
3611                 return error;
3612 
3613         spin_lock(&sbinfo->stat_lock);
3614         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3615         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3616                 goto out;
3617         if (config.max_inodes < inodes)
3618                 goto out;
3619         /*
3620          * Those tests disallow limited->unlimited while any are in use;
3621          * but we must separately disallow unlimited->limited, because
3622          * in that case we have no record of how much is already in use.
3623          */
3624         if (config.max_blocks && !sbinfo->max_blocks)
3625                 goto out;
3626         if (config.max_inodes && !sbinfo->max_inodes)
3627                 goto out;
3628 
3629         error = 0;
3630         sbinfo->huge = config.huge;
3631         sbinfo->max_blocks  = config.max_blocks;
3632         sbinfo->max_inodes  = config.max_inodes;
3633         sbinfo->free_inodes = config.max_inodes - inodes;
3634 
3635         /*
3636          * Preserve previous mempolicy unless mpol remount option was specified.
3637          */
3638         if (config.mpol) {
3639                 mpol_put(sbinfo->mpol);
3640                 sbinfo->mpol = config.mpol;     /* transfers initial ref */
3641         }
3642 out:
3643         spin_unlock(&sbinfo->stat_lock);
3644         return error;
3645 }
3646 
3647 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3648 {
3649         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3650 
3651         if (sbinfo->max_blocks != shmem_default_max_blocks())
3652                 seq_printf(seq, ",size=%luk",
3653                         sbinfo->max_blocks << (PAGE_SHIFT - 10));
3654         if (sbinfo->max_inodes != shmem_default_max_inodes())
3655                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3656         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3657                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3658         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3659                 seq_printf(seq, ",uid=%u",
3660                                 from_kuid_munged(&init_user_ns, sbinfo->uid));
3661         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3662                 seq_printf(seq, ",gid=%u",
3663                                 from_kgid_munged(&init_user_ns, sbinfo->gid));
3664 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3665         /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3666         if (sbinfo->huge)
3667                 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3668 #endif
3669         shmem_show_mpol(seq, sbinfo->mpol);
3670         return 0;
3671 }
3672 
3673 #define MFD_NAME_PREFIX "memfd:"
3674 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3675 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3676 
3677 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB)
3678 
3679 SYSCALL_DEFINE2(memfd_create,
3680                 const char __user *, uname,
3681                 unsigned int, flags)
3682 {
3683         unsigned int *file_seals;
3684         struct file *file;
3685         int fd, error;
3686         char *name;
3687         long len;
3688 
3689         if (!(flags & MFD_HUGETLB)) {
3690                 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3691                         return -EINVAL;
3692         } else {
3693                 /* Allow huge page size encoding in flags. */
3694                 if (flags & ~(unsigned int)(MFD_ALL_FLAGS |
3695                                 (MFD_HUGE_MASK << MFD_HUGE_SHIFT)))
3696                         return -EINVAL;
3697         }
3698 
3699         /* length includes terminating zero */
3700         len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3701         if (len <= 0)
3702                 return -EFAULT;
3703         if (len > MFD_NAME_MAX_LEN + 1)
3704                 return -EINVAL;
3705 
3706         name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL);
3707         if (!name)
3708                 return -ENOMEM;
3709 
3710         strcpy(name, MFD_NAME_PREFIX);
3711         if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3712                 error = -EFAULT;
3713                 goto err_name;
3714         }
3715 
3716         /* terminating-zero may have changed after strnlen_user() returned */
3717         if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3718                 error = -EFAULT;
3719                 goto err_name;
3720         }
3721 
3722         fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3723         if (fd < 0) {
3724                 error = fd;
3725                 goto err_name;
3726         }
3727 
3728         if (flags & MFD_HUGETLB) {
3729                 struct user_struct *user = NULL;
3730 
3731                 file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user,
3732                                         HUGETLB_ANONHUGE_INODE,
3733                                         (flags >> MFD_HUGE_SHIFT) &
3734                                         MFD_HUGE_MASK);
3735         } else
3736                 file = shmem_file_setup(name, 0, VM_NORESERVE);
3737         if (IS_ERR(file)) {
3738                 error = PTR_ERR(file);
3739                 goto err_fd;
3740         }
3741         file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3742         file->f_flags |= O_RDWR | O_LARGEFILE;
3743 
3744         if (flags & MFD_ALLOW_SEALING) {
3745                 file_seals = memfd_file_seals_ptr(file);
3746                 *file_seals &= ~F_SEAL_SEAL;
3747         }
3748 
3749         fd_install(fd, file);
3750         kfree(name);
3751         return fd;
3752 
3753 err_fd:
3754         put_unused_fd(fd);
3755 err_name:
3756         kfree(name);
3757         return error;
3758 }
3759 
3760 #endif /* CONFIG_TMPFS */
3761 
3762 static void shmem_put_super(struct super_block *sb)
3763 {
3764         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3765 
3766         percpu_counter_destroy(&sbinfo->used_blocks);
3767         mpol_put(sbinfo->mpol);
3768         kfree(sbinfo);
3769         sb->s_fs_info = NULL;
3770 }
3771 
3772 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3773 {
3774         struct inode *inode;
3775         struct shmem_sb_info *sbinfo;
3776         int err = -ENOMEM;
3777 
3778         /* Round up to L1_CACHE_BYTES to resist false sharing */
3779         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3780                                 L1_CACHE_BYTES), GFP_KERNEL);
3781         if (!sbinfo)
3782                 return -ENOMEM;
3783 
3784         sbinfo->mode = S_IRWXUGO | S_ISVTX;
3785         sbinfo->uid = current_fsuid();
3786         sbinfo->gid = current_fsgid();
3787         sb->s_fs_info = sbinfo;
3788 
3789 #ifdef CONFIG_TMPFS
3790         /*
3791          * Per default we only allow half of the physical ram per
3792          * tmpfs instance, limiting inodes to one per page of lowmem;
3793          * but the internal instance is left unlimited.
3794          */
3795         if (!(sb->s_flags & SB_KERNMOUNT)) {
3796                 sbinfo->max_blocks = shmem_default_max_blocks();
3797                 sbinfo->max_inodes = shmem_default_max_inodes();
3798                 if (shmem_parse_options(data, sbinfo, false)) {
3799                         err = -EINVAL;
3800                         goto failed;
3801                 }
3802         } else {
3803                 sb->s_flags |= SB_NOUSER;
3804         }
3805         sb->s_export_op = &shmem_export_ops;
3806         sb->s_flags |= SB_NOSEC;
3807 #else
3808         sb->s_flags |= SB_NOUSER;
3809 #endif
3810 
3811         spin_lock_init(&sbinfo->stat_lock);
3812         if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3813                 goto failed;
3814         sbinfo->free_inodes = sbinfo->max_inodes;
3815         spin_lock_init(&sbinfo->shrinklist_lock);
3816         INIT_LIST_HEAD(&sbinfo->shrinklist);
3817 
3818         sb->s_maxbytes = MAX_LFS_FILESIZE;
3819         sb->s_blocksize = PAGE_SIZE;
3820         sb->s_blocksize_bits = PAGE_SHIFT;
3821         sb->s_magic = TMPFS_MAGIC;
3822         sb->s_op = &shmem_ops;
3823         sb->s_time_gran = 1;
3824 #ifdef CONFIG_TMPFS_XATTR
3825         sb->s_xattr = shmem_xattr_handlers;
3826 #endif
3827 #ifdef CONFIG_TMPFS_POSIX_ACL
3828         sb->s_flags |= SB_POSIXACL;
3829 #endif
3830         uuid_gen(&sb->s_uuid);
3831 
3832         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3833         if (!inode)
3834                 goto failed;
3835         inode->i_uid = sbinfo->uid;
3836         inode->i_gid = sbinfo->gid;
3837         sb->s_root = d_make_root(inode);
3838         if (!sb->s_root)
3839                 goto failed;
3840         return 0;
3841 
3842 failed:
3843         shmem_put_super(sb);
3844         return err;
3845 }
3846 
3847 static struct kmem_cache *shmem_inode_cachep;
3848 
3849 static struct inode *shmem_alloc_inode(struct super_block *sb)
3850 {
3851         struct shmem_inode_info *info;
3852         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3853         if (!info)
3854                 return NULL;
3855         return &info->vfs_inode;
3856 }
3857 
3858 static void shmem_destroy_callback(struct rcu_head *head)
3859 {
3860         struct inode *inode = container_of(head, struct inode, i_rcu);
3861         if (S_ISLNK(inode->i_mode))
3862                 kfree(inode->i_link);
3863         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3864 }
3865 
3866 static void shmem_destroy_inode(struct inode *inode)
3867 {
3868         if (S_ISREG(inode->i_mode))
3869                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3870         call_rcu(&inode->i_rcu, shmem_destroy_callback);
3871 }
3872 
3873 static void shmem_init_inode(void *foo)
3874 {
3875         struct shmem_inode_info *info = foo;
3876         inode_init_once(&info->vfs_inode);
3877 }
3878 
3879 static void shmem_init_inodecache(void)
3880 {
3881         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3882                                 sizeof(struct shmem_inode_info),
3883                                 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3884 }
3885 
3886 static void shmem_destroy_inodecache(void)
3887 {
3888         kmem_cache_destroy(shmem_inode_cachep);
3889 }
3890 
3891 static const struct address_space_operations shmem_aops = {
3892         .writepage      = shmem_writepage,
3893         .set_page_dirty = __set_page_dirty_no_writeback,
3894 #ifdef CONFIG_TMPFS
3895         .write_begin    = shmem_write_begin,
3896         .write_end      = shmem_write_end,
3897 #endif
3898 #ifdef CONFIG_MIGRATION
3899         .migratepage    = migrate_page,
3900 #endif
3901         .error_remove_page = generic_error_remove_page,
3902 };
3903 
3904 static const struct file_operations shmem_file_operations = {
3905         .mmap           = shmem_mmap,
3906         .get_unmapped_area = shmem_get_unmapped_area,
3907 #ifdef CONFIG_TMPFS
3908         .llseek         = shmem_file_llseek,
3909         .read_iter      = shmem_file_read_iter,
3910         .write_iter     = generic_file_write_iter,
3911         .fsync          = noop_fsync,
3912         .splice_read    = generic_file_splice_read,
3913         .splice_write   = iter_file_splice_write,
3914         .fallocate      = shmem_fallocate,
3915 #endif
3916 };
3917 
3918 static const struct inode_operations shmem_inode_operations = {
3919         .getattr        = shmem_getattr,
3920         .setattr        = shmem_setattr,
3921 #ifdef CONFIG_TMPFS_XATTR
3922         .listxattr      = shmem_listxattr,
3923         .set_acl        = simple_set_acl,
3924 #endif
3925 };
3926 
3927 static const struct inode_operations shmem_dir_inode_operations = {
3928 #ifdef CONFIG_TMPFS
3929         .create         = shmem_create,
3930         .lookup         = simple_lookup,
3931         .link           = shmem_link,
3932         .unlink         = shmem_unlink,
3933         .symlink        = shmem_symlink,
3934         .mkdir          = shmem_mkdir,
3935         .rmdir          = shmem_rmdir,
3936         .mknod          = shmem_mknod,
3937         .rename         = shmem_rename2,
3938         .tmpfile        = shmem_tmpfile,
3939 #endif
3940 #ifdef CONFIG_TMPFS_XATTR
3941         .listxattr      = shmem_listxattr,
3942 #endif
3943 #ifdef CONFIG_TMPFS_POSIX_ACL
3944         .setattr        = shmem_setattr,
3945         .set_acl        = simple_set_acl,
3946 #endif
3947 };
3948 
3949 static const struct inode_operations shmem_special_inode_operations = {
3950 #ifdef CONFIG_TMPFS_XATTR
3951         .listxattr      = shmem_listxattr,
3952 #endif
3953 #ifdef CONFIG_TMPFS_POSIX_ACL
3954         .setattr        = shmem_setattr,
3955         .set_acl        = simple_set_acl,
3956 #endif
3957 };
3958 
3959 static const struct super_operations shmem_ops = {
3960         .alloc_inode    = shmem_alloc_inode,
3961         .destroy_inode  = shmem_destroy_inode,
3962 #ifdef CONFIG_TMPFS
3963         .statfs         = shmem_statfs,
3964         .remount_fs     = shmem_remount_fs,
3965         .show_options   = shmem_show_options,
3966 #endif
3967         .evict_inode    = shmem_evict_inode,
3968         .drop_inode     = generic_delete_inode,
3969         .put_super      = shmem_put_super,
3970 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3971         .nr_cached_objects      = shmem_unused_huge_count,
3972         .free_cached_objects    = shmem_unused_huge_scan,
3973 #endif
3974 };
3975 
3976 static const struct vm_operations_struct shmem_vm_ops = {
3977         .fault          = shmem_fault,
3978         .map_pages      = filemap_map_pages,
3979 #ifdef CONFIG_NUMA
3980         .set_policy     = shmem_set_policy,
3981         .get_policy     = shmem_get_policy,
3982 #endif
3983 };
3984 
3985 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3986         int flags, const char *dev_name, void *data)
3987 {
3988         return mount_nodev(fs_type, flags, data, shmem_fill_super);
3989 }
3990 
3991 static struct file_system_type shmem_fs_type = {
3992         .owner          = THIS_MODULE,
3993         .name           = "tmpfs",
3994         .mount          = shmem_mount,
3995         .kill_sb        = kill_litter_super,
3996         .fs_flags       = FS_USERNS_MOUNT,
3997 };
3998 
3999 int __init shmem_init(void)
4000 {
4001         int error;
4002 
4003         /* If rootfs called this, don't re-init */
4004         if (shmem_inode_cachep)
4005                 return 0;
4006 
4007         shmem_init_inodecache();
4008 
4009         error = register_filesystem(&shmem_fs_type);
4010         if (error) {
4011                 pr_err("Could not register tmpfs\n");
4012                 goto out2;
4013         }
4014 
4015         shm_mnt = kern_mount(&shmem_fs_type);
4016         if (IS_ERR(shm_mnt)) {
4017                 error = PTR_ERR(shm_mnt);
4018                 pr_err("Could not kern_mount tmpfs\n");
4019                 goto out1;
4020         }
4021 
4022 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4023         if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4024                 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4025         else
4026                 shmem_huge = 0; /* just in case it was patched */
4027 #endif
4028         return 0;
4029 
4030 out1:
4031         unregister_filesystem(&shmem_fs_type);
4032 out2:
4033         shmem_destroy_inodecache();
4034         shm_mnt = ERR_PTR(error);
4035         return error;
4036 }
4037 
4038 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
4039 static ssize_t shmem_enabled_show(struct kobject *kobj,
4040                 struct kobj_attribute *attr, char *buf)
4041 {
4042         int values[] = {
4043                 SHMEM_HUGE_ALWAYS,
4044                 SHMEM_HUGE_WITHIN_SIZE,
4045                 SHMEM_HUGE_ADVISE,
4046                 SHMEM_HUGE_NEVER,
4047                 SHMEM_HUGE_DENY,
4048                 SHMEM_HUGE_FORCE,
4049         };
4050         int i, count;
4051 
4052         for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
4053                 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
4054 
4055                 count += sprintf(buf + count, fmt,
4056                                 shmem_format_huge(values[i]));
4057         }
4058         buf[count - 1] = '\n';
4059         return count;
4060 }
4061 
4062 static ssize_t shmem_enabled_store(struct kobject *kobj,
4063                 struct kobj_attribute *attr, const char *buf, size_t count)
4064 {
4065         char tmp[16];
4066         int huge;
4067 
4068         if (count + 1 > sizeof(tmp))
4069                 return -EINVAL;
4070         memcpy(tmp, buf, count);
4071         tmp[count] = '\0';
4072         if (count && tmp[count - 1] == '\n')
4073                 tmp[count - 1] = '\0';
4074 
4075         huge = shmem_parse_huge(tmp);
4076         if (huge == -EINVAL)
4077                 return -EINVAL;
4078         if (!has_transparent_hugepage() &&
4079                         huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4080                 return -EINVAL;
4081 
4082         shmem_huge = huge;
4083         if (shmem_huge > SHMEM_HUGE_DENY)
4084                 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4085         return count;
4086 }
4087 
4088 struct kobj_attribute shmem_enabled_attr =
4089         __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4090 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
4091 
4092 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
4093 bool shmem_huge_enabled(struct vm_area_struct *vma)
4094 {
4095         struct inode *inode = file_inode(vma->vm_file);
4096         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4097         loff_t i_size;
4098         pgoff_t off;
4099 
4100         if (shmem_huge == SHMEM_HUGE_FORCE)
4101                 return true;
4102         if (shmem_huge == SHMEM_HUGE_DENY)
4103                 return false;
4104         switch (sbinfo->huge) {
4105                 case SHMEM_HUGE_NEVER:
4106                         return false;
4107                 case SHMEM_HUGE_ALWAYS:
4108                         return true;
4109                 case SHMEM_HUGE_WITHIN_SIZE:
4110                         off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4111                         i_size = round_up(i_size_read(inode), PAGE_SIZE);
4112                         if (i_size >= HPAGE_PMD_SIZE &&
4113                                         i_size >> PAGE_SHIFT >= off)
4114                                 return true;
4115                         /* fall through */
4116                 case SHMEM_HUGE_ADVISE:
4117                         /* TODO: implement fadvise() hints */
4118                         return (vma->vm_flags & VM_HUGEPAGE);
4119                 default:
4120                         VM_BUG_ON(1);
4121                         return false;
4122         }
4123 }
4124 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4125 
4126 #else /* !CONFIG_SHMEM */
4127 
4128 /*
4129  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4130  *
4131  * This is intended for small system where the benefits of the full
4132  * shmem code (swap-backed and resource-limited) are outweighed by
4133  * their complexity. On systems without swap this code should be
4134  * effectively equivalent, but much lighter weight.
4135  */
4136 
4137 static struct file_system_type shmem_fs_type = {
4138         .name           = "tmpfs",
4139         .mount          = ramfs_mount,
4140         .kill_sb        = kill_litter_super,
4141         .fs_flags       = FS_USERNS_MOUNT,
4142 };
4143 
4144 int __init shmem_init(void)
4145 {
4146         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4147 
4148         shm_mnt = kern_mount(&shmem_fs_type);
4149         BUG_ON(IS_ERR(shm_mnt));
4150 
4151         return 0;
4152 }
4153 
4154 int shmem_unuse(swp_entry_t swap, struct page *page)
4155 {
4156         return 0;
4157 }
4158 
4159 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4160 {
4161         return 0;
4162 }
4163 
4164 void shmem_unlock_mapping(struct address_space *mapping)
4165 {
4166 }
4167 
4168 #ifdef CONFIG_MMU
4169 unsigned long shmem_get_unmapped_area(struct file *file,
4170                                       unsigned long addr, unsigned long len,
4171                                       unsigned long pgoff, unsigned long flags)
4172 {
4173         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4174 }
4175 #endif
4176 
4177 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4178 {
4179         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4180 }
4181 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4182 
4183 #define shmem_vm_ops                            generic_file_vm_ops
4184 #define shmem_file_operations                   ramfs_file_operations
4185 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
4186 #define shmem_acct_size(flags, size)            0
4187 #define shmem_unacct_size(flags, size)          do {} while (0)
4188 
4189 #endif /* CONFIG_SHMEM */
4190 
4191 /* common code */
4192 
4193 static const struct dentry_operations anon_ops = {
4194         .d_dname = simple_dname
4195 };
4196 
4197 static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4198                                        unsigned long flags, unsigned int i_flags)
4199 {
4200         struct file *res;
4201         struct inode *inode;
4202         struct path path;
4203         struct super_block *sb;
4204         struct qstr this;
4205 
4206         if (IS_ERR(mnt))
4207                 return ERR_CAST(mnt);
4208 
4209         if (size < 0 || size > MAX_LFS_FILESIZE)
4210                 return ERR_PTR(-EINVAL);
4211 
4212         if (shmem_acct_size(flags, size))
4213                 return ERR_PTR(-ENOMEM);
4214 
4215         res = ERR_PTR(-ENOMEM);
4216         this.name = name;
4217         this.len = strlen(name);
4218         this.hash = 0; /* will go */
4219         sb = mnt->mnt_sb;
4220         path.mnt = mntget(mnt);
4221         path.dentry = d_alloc_pseudo(sb, &this);
4222         if (!path.dentry)
4223                 goto put_memory;
4224         d_set_d_op(path.dentry, &anon_ops);
4225 
4226         res = ERR_PTR(-ENOSPC);
4227         inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4228         if (!inode)
4229                 goto put_memory;
4230 
4231         inode->i_flags |= i_flags;
4232         d_instantiate(path.dentry, inode);
4233         inode->i_size = size;
4234         clear_nlink(inode);     /* It is unlinked */
4235         res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4236         if (IS_ERR(res))
4237                 goto put_path;
4238 
4239         res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4240                   &shmem_file_operations);
4241         if (IS_ERR(res))
4242                 goto put_path;
4243 
4244         return res;
4245 
4246 put_memory:
4247         shmem_unacct_size(flags, size);
4248 put_path:
4249         path_put(&path);
4250         return res;
4251 }
4252 
4253 /**
4254  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4255  *      kernel internal.  There will be NO LSM permission checks against the
4256  *      underlying inode.  So users of this interface must do LSM checks at a
4257  *      higher layer.  The users are the big_key and shm implementations.  LSM
4258  *      checks are provided at the key or shm level rather than the inode.
4259  * @name: name for dentry (to be seen in /proc/<pid>/maps
4260  * @size: size to be set for the file
4261  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4262  */
4263 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4264 {
4265         return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4266 }
4267 
4268 /**
4269  * shmem_file_setup - get an unlinked file living in tmpfs
4270  * @name: name for dentry (to be seen in /proc/<pid>/maps
4271  * @size: size to be set for the file
4272  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4273  */
4274 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4275 {
4276         return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4277 }
4278 EXPORT_SYMBOL_GPL(shmem_file_setup);
4279 
4280 /**
4281  * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4282  * @mnt: the tmpfs mount where the file will be created
4283  * @name: name for dentry (to be seen in /proc/<pid>/maps
4284  * @size: size to be set for the file
4285  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4286  */
4287 struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4288                                        loff_t size, unsigned long flags)
4289 {
4290         return __shmem_file_setup(mnt, name, size, flags, 0);
4291 }
4292 EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4293 
4294 /**
4295  * shmem_zero_setup - setup a shared anonymous mapping
4296  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4297  */
4298 int shmem_zero_setup(struct vm_area_struct *vma)
4299 {
4300         struct file *file;
4301         loff_t size = vma->vm_end - vma->vm_start;
4302 
4303         /*
4304          * Cloning a new file under mmap_sem leads to a lock ordering conflict
4305          * between XFS directory reading and selinux: since this file is only
4306          * accessible to the user through its mapping, use S_PRIVATE flag to
4307          * bypass file security, in the same way as shmem_kernel_file_setup().
4308          */
4309         file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4310         if (IS_ERR(file))
4311                 return PTR_ERR(file);
4312 
4313         if (vma->vm_file)
4314                 fput(vma->vm_file);
4315         vma->vm_file = file;
4316         vma->vm_ops = &shmem_vm_ops;
4317 
4318         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4319                         ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4320                         (vma->vm_end & HPAGE_PMD_MASK)) {
4321                 khugepaged_enter(vma, vma->vm_flags);
4322         }
4323 
4324         return 0;
4325 }
4326 
4327 /**
4328  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4329  * @mapping:    the page's address_space
4330  * @index:      the page index
4331  * @gfp:        the page allocator flags to use if allocating
4332  *
4333  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4334  * with any new page allocations done using the specified allocation flags.
4335  * But read_cache_page_gfp() uses the ->readpage() method: which does not
4336  * suit tmpfs, since it may have pages in swapcache, and needs to find those
4337  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4338  *
4339  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4340  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4341  */
4342 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4343                                          pgoff_t index, gfp_t gfp)
4344 {
4345 #ifdef CONFIG_SHMEM
4346         struct inode *inode = mapping->host;
4347         struct page *page;
4348         int error;
4349 
4350         BUG_ON(mapping->a_ops != &shmem_aops);
4351         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4352                                   gfp, NULL, NULL, NULL);
4353         if (error)
4354                 page = ERR_PTR(error);
4355         else
4356                 unlock_page(page);
4357         return page;
4358 #else
4359         /*
4360          * The tiny !SHMEM case uses ramfs without swap
4361          */
4362         return read_cache_page_gfp(mapping, index, gfp);
4363 #endif
4364 }
4365 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
4366 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp