Version:  2.0.40 2.2.26 2.4.37 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9

Linux/mm/shmem.c

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

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us