Version:  2.0.40 2.2.26 2.4.37 3.11 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

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;
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 = inode_change_ok(inode, 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;
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         SetPageUptodate(newpage);
1487         set_page_private(newpage, swap_index);
1488         SetPageSwapCache(newpage);
1489 
1490         /*
1491          * Our caller will very soon move newpage out of swapcache, but it's
1492          * a nice clean interface for us to replace oldpage by newpage there.
1493          */
1494         spin_lock_irq(&swap_mapping->tree_lock);
1495         error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1496                                                                    newpage);
1497         if (!error) {
1498                 __inc_node_page_state(newpage, NR_FILE_PAGES);
1499                 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1500         }
1501         spin_unlock_irq(&swap_mapping->tree_lock);
1502 
1503         if (unlikely(error)) {
1504                 /*
1505                  * Is this possible?  I think not, now that our callers check
1506                  * both PageSwapCache and page_private after getting page lock;
1507                  * but be defensive.  Reverse old to newpage for clear and free.
1508                  */
1509                 oldpage = newpage;
1510         } else {
1511                 mem_cgroup_migrate(oldpage, newpage);
1512                 lru_cache_add_anon(newpage);
1513                 *pagep = newpage;
1514         }
1515 
1516         ClearPageSwapCache(oldpage);
1517         set_page_private(oldpage, 0);
1518 
1519         unlock_page(oldpage);
1520         put_page(oldpage);
1521         put_page(oldpage);
1522         return error;
1523 }
1524 
1525 /*
1526  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1527  *
1528  * If we allocate a new one we do not mark it dirty. That's up to the
1529  * vm. If we swap it in we mark it dirty since we also free the swap
1530  * entry since a page cannot live in both the swap and page cache.
1531  *
1532  * fault_mm and fault_type are only supplied by shmem_fault:
1533  * otherwise they are NULL.
1534  */
1535 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1536         struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1537         struct mm_struct *fault_mm, int *fault_type)
1538 {
1539         struct address_space *mapping = inode->i_mapping;
1540         struct shmem_inode_info *info;
1541         struct shmem_sb_info *sbinfo;
1542         struct mm_struct *charge_mm;
1543         struct mem_cgroup *memcg;
1544         struct page *page;
1545         swp_entry_t swap;
1546         enum sgp_type sgp_huge = sgp;
1547         pgoff_t hindex = index;
1548         int error;
1549         int once = 0;
1550         int alloced = 0;
1551 
1552         if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1553                 return -EFBIG;
1554         if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1555                 sgp = SGP_CACHE;
1556 repeat:
1557         swap.val = 0;
1558         page = find_lock_entry(mapping, index);
1559         if (radix_tree_exceptional_entry(page)) {
1560                 swap = radix_to_swp_entry(page);
1561                 page = NULL;
1562         }
1563 
1564         if (sgp <= SGP_CACHE &&
1565             ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1566                 error = -EINVAL;
1567                 goto unlock;
1568         }
1569 
1570         if (page && sgp == SGP_WRITE)
1571                 mark_page_accessed(page);
1572 
1573         /* fallocated page? */
1574         if (page && !PageUptodate(page)) {
1575                 if (sgp != SGP_READ)
1576                         goto clear;
1577                 unlock_page(page);
1578                 put_page(page);
1579                 page = NULL;
1580         }
1581         if (page || (sgp == SGP_READ && !swap.val)) {
1582                 *pagep = page;
1583                 return 0;
1584         }
1585 
1586         /*
1587          * Fast cache lookup did not find it:
1588          * bring it back from swap or allocate.
1589          */
1590         info = SHMEM_I(inode);
1591         sbinfo = SHMEM_SB(inode->i_sb);
1592         charge_mm = fault_mm ? : current->mm;
1593 
1594         if (swap.val) {
1595                 /* Look it up and read it in.. */
1596                 page = lookup_swap_cache(swap);
1597                 if (!page) {
1598                         /* Or update major stats only when swapin succeeds?? */
1599                         if (fault_type) {
1600                                 *fault_type |= VM_FAULT_MAJOR;
1601                                 count_vm_event(PGMAJFAULT);
1602                                 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1603                         }
1604                         /* Here we actually start the io */
1605                         page = shmem_swapin(swap, gfp, info, index);
1606                         if (!page) {
1607                                 error = -ENOMEM;
1608                                 goto failed;
1609                         }
1610                 }
1611 
1612                 /* We have to do this with page locked to prevent races */
1613                 lock_page(page);
1614                 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1615                     !shmem_confirm_swap(mapping, index, swap)) {
1616                         error = -EEXIST;        /* try again */
1617                         goto unlock;
1618                 }
1619                 if (!PageUptodate(page)) {
1620                         error = -EIO;
1621                         goto failed;
1622                 }
1623                 wait_on_page_writeback(page);
1624 
1625                 if (shmem_should_replace_page(page, gfp)) {
1626                         error = shmem_replace_page(&page, gfp, info, index);
1627                         if (error)
1628                                 goto failed;
1629                 }
1630 
1631                 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1632                                 false);
1633                 if (!error) {
1634                         error = shmem_add_to_page_cache(page, mapping, index,
1635                                                 swp_to_radix_entry(swap));
1636                         /*
1637                          * We already confirmed swap under page lock, and make
1638                          * no memory allocation here, so usually no possibility
1639                          * of error; but free_swap_and_cache() only trylocks a
1640                          * page, so it is just possible that the entry has been
1641                          * truncated or holepunched since swap was confirmed.
1642                          * shmem_undo_range() will have done some of the
1643                          * unaccounting, now delete_from_swap_cache() will do
1644                          * the rest.
1645                          * Reset swap.val? No, leave it so "failed" goes back to
1646                          * "repeat": reading a hole and writing should succeed.
1647                          */
1648                         if (error) {
1649                                 mem_cgroup_cancel_charge(page, memcg, false);
1650                                 delete_from_swap_cache(page);
1651                         }
1652                 }
1653                 if (error)
1654                         goto failed;
1655 
1656                 mem_cgroup_commit_charge(page, memcg, true, false);
1657 
1658                 spin_lock_irq(&info->lock);
1659                 info->swapped--;
1660                 shmem_recalc_inode(inode);
1661                 spin_unlock_irq(&info->lock);
1662 
1663                 if (sgp == SGP_WRITE)
1664                         mark_page_accessed(page);
1665 
1666                 delete_from_swap_cache(page);
1667                 set_page_dirty(page);
1668                 swap_free(swap);
1669 
1670         } else {
1671                 /* shmem_symlink() */
1672                 if (mapping->a_ops != &shmem_aops)
1673                         goto alloc_nohuge;
1674                 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1675                         goto alloc_nohuge;
1676                 if (shmem_huge == SHMEM_HUGE_FORCE)
1677                         goto alloc_huge;
1678                 switch (sbinfo->huge) {
1679                         loff_t i_size;
1680                         pgoff_t off;
1681                 case SHMEM_HUGE_NEVER:
1682                         goto alloc_nohuge;
1683                 case SHMEM_HUGE_WITHIN_SIZE:
1684                         off = round_up(index, HPAGE_PMD_NR);
1685                         i_size = round_up(i_size_read(inode), PAGE_SIZE);
1686                         if (i_size >= HPAGE_PMD_SIZE &&
1687                                         i_size >> PAGE_SHIFT >= off)
1688                                 goto alloc_huge;
1689                         /* fallthrough */
1690                 case SHMEM_HUGE_ADVISE:
1691                         if (sgp_huge == SGP_HUGE)
1692                                 goto alloc_huge;
1693                         /* TODO: implement fadvise() hints */
1694                         goto alloc_nohuge;
1695                 }
1696 
1697 alloc_huge:
1698                 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1699                                 index, true);
1700                 if (IS_ERR(page)) {
1701 alloc_nohuge:           page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1702                                         index, false);
1703                 }
1704                 if (IS_ERR(page)) {
1705                         int retry = 5;
1706                         error = PTR_ERR(page);
1707                         page = NULL;
1708                         if (error != -ENOSPC)
1709                                 goto failed;
1710                         /*
1711                          * Try to reclaim some spece by splitting a huge page
1712                          * beyond i_size on the filesystem.
1713                          */
1714                         while (retry--) {
1715                                 int ret;
1716                                 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1717                                 if (ret == SHRINK_STOP)
1718                                         break;
1719                                 if (ret)
1720                                         goto alloc_nohuge;
1721                         }
1722                         goto failed;
1723                 }
1724 
1725                 if (PageTransHuge(page))
1726                         hindex = round_down(index, HPAGE_PMD_NR);
1727                 else
1728                         hindex = index;
1729 
1730                 if (sgp == SGP_WRITE)
1731                         __SetPageReferenced(page);
1732 
1733                 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1734                                 PageTransHuge(page));
1735                 if (error)
1736                         goto unacct;
1737                 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1738                                 compound_order(page));
1739                 if (!error) {
1740                         error = shmem_add_to_page_cache(page, mapping, hindex,
1741                                                         NULL);
1742                         radix_tree_preload_end();
1743                 }
1744                 if (error) {
1745                         mem_cgroup_cancel_charge(page, memcg,
1746                                         PageTransHuge(page));
1747                         goto unacct;
1748                 }
1749                 mem_cgroup_commit_charge(page, memcg, false,
1750                                 PageTransHuge(page));
1751                 lru_cache_add_anon(page);
1752 
1753                 spin_lock_irq(&info->lock);
1754                 info->alloced += 1 << compound_order(page);
1755                 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1756                 shmem_recalc_inode(inode);
1757                 spin_unlock_irq(&info->lock);
1758                 alloced = true;
1759 
1760                 if (PageTransHuge(page) &&
1761                                 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1762                                 hindex + HPAGE_PMD_NR - 1) {
1763                         /*
1764                          * Part of the huge page is beyond i_size: subject
1765                          * to shrink under memory pressure.
1766                          */
1767                         spin_lock(&sbinfo->shrinklist_lock);
1768                         if (list_empty(&info->shrinklist)) {
1769                                 list_add_tail(&info->shrinklist,
1770                                                 &sbinfo->shrinklist);
1771                                 sbinfo->shrinklist_len++;
1772                         }
1773                         spin_unlock(&sbinfo->shrinklist_lock);
1774                 }
1775 
1776                 /*
1777                  * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1778                  */
1779                 if (sgp == SGP_FALLOC)
1780                         sgp = SGP_WRITE;
1781 clear:
1782                 /*
1783                  * Let SGP_WRITE caller clear ends if write does not fill page;
1784                  * but SGP_FALLOC on a page fallocated earlier must initialize
1785                  * it now, lest undo on failure cancel our earlier guarantee.
1786                  */
1787                 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1788                         struct page *head = compound_head(page);
1789                         int i;
1790 
1791                         for (i = 0; i < (1 << compound_order(head)); i++) {
1792                                 clear_highpage(head + i);
1793                                 flush_dcache_page(head + i);
1794                         }
1795                         SetPageUptodate(head);
1796                 }
1797         }
1798 
1799         /* Perhaps the file has been truncated since we checked */
1800         if (sgp <= SGP_CACHE &&
1801             ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1802                 if (alloced) {
1803                         ClearPageDirty(page);
1804                         delete_from_page_cache(page);
1805                         spin_lock_irq(&info->lock);
1806                         shmem_recalc_inode(inode);
1807                         spin_unlock_irq(&info->lock);
1808                 }
1809                 error = -EINVAL;
1810                 goto unlock;
1811         }
1812         *pagep = page + index - hindex;
1813         return 0;
1814 
1815         /*
1816          * Error recovery.
1817          */
1818 unacct:
1819         if (sbinfo->max_blocks)
1820                 percpu_counter_sub(&sbinfo->used_blocks,
1821                                 1 << compound_order(page));
1822         shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1823 
1824         if (PageTransHuge(page)) {
1825                 unlock_page(page);
1826                 put_page(page);
1827                 goto alloc_nohuge;
1828         }
1829 failed:
1830         if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1831                 error = -EEXIST;
1832 unlock:
1833         if (page) {
1834                 unlock_page(page);
1835                 put_page(page);
1836         }
1837         if (error == -ENOSPC && !once++) {
1838                 info = SHMEM_I(inode);
1839                 spin_lock_irq(&info->lock);
1840                 shmem_recalc_inode(inode);
1841                 spin_unlock_irq(&info->lock);
1842                 goto repeat;
1843         }
1844         if (error == -EEXIST)   /* from above or from radix_tree_insert */
1845                 goto repeat;
1846         return error;
1847 }
1848 
1849 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1850 {
1851         struct inode *inode = file_inode(vma->vm_file);
1852         gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1853         enum sgp_type sgp;
1854         int error;
1855         int ret = VM_FAULT_LOCKED;
1856 
1857         /*
1858          * Trinity finds that probing a hole which tmpfs is punching can
1859          * prevent the hole-punch from ever completing: which in turn
1860          * locks writers out with its hold on i_mutex.  So refrain from
1861          * faulting pages into the hole while it's being punched.  Although
1862          * shmem_undo_range() does remove the additions, it may be unable to
1863          * keep up, as each new page needs its own unmap_mapping_range() call,
1864          * and the i_mmap tree grows ever slower to scan if new vmas are added.
1865          *
1866          * It does not matter if we sometimes reach this check just before the
1867          * hole-punch begins, so that one fault then races with the punch:
1868          * we just need to make racing faults a rare case.
1869          *
1870          * The implementation below would be much simpler if we just used a
1871          * standard mutex or completion: but we cannot take i_mutex in fault,
1872          * and bloating every shmem inode for this unlikely case would be sad.
1873          */
1874         if (unlikely(inode->i_private)) {
1875                 struct shmem_falloc *shmem_falloc;
1876 
1877                 spin_lock(&inode->i_lock);
1878                 shmem_falloc = inode->i_private;
1879                 if (shmem_falloc &&
1880                     shmem_falloc->waitq &&
1881                     vmf->pgoff >= shmem_falloc->start &&
1882                     vmf->pgoff < shmem_falloc->next) {
1883                         wait_queue_head_t *shmem_falloc_waitq;
1884                         DEFINE_WAIT(shmem_fault_wait);
1885 
1886                         ret = VM_FAULT_NOPAGE;
1887                         if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1888                            !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1889                                 /* It's polite to up mmap_sem if we can */
1890                                 up_read(&vma->vm_mm->mmap_sem);
1891                                 ret = VM_FAULT_RETRY;
1892                         }
1893 
1894                         shmem_falloc_waitq = shmem_falloc->waitq;
1895                         prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1896                                         TASK_UNINTERRUPTIBLE);
1897                         spin_unlock(&inode->i_lock);
1898                         schedule();
1899 
1900                         /*
1901                          * shmem_falloc_waitq points into the shmem_fallocate()
1902                          * stack of the hole-punching task: shmem_falloc_waitq
1903                          * is usually invalid by the time we reach here, but
1904                          * finish_wait() does not dereference it in that case;
1905                          * though i_lock needed lest racing with wake_up_all().
1906                          */
1907                         spin_lock(&inode->i_lock);
1908                         finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1909                         spin_unlock(&inode->i_lock);
1910                         return ret;
1911                 }
1912                 spin_unlock(&inode->i_lock);
1913         }
1914 
1915         sgp = SGP_CACHE;
1916         if (vma->vm_flags & VM_HUGEPAGE)
1917                 sgp = SGP_HUGE;
1918         else if (vma->vm_flags & VM_NOHUGEPAGE)
1919                 sgp = SGP_NOHUGE;
1920 
1921         error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1922                                   gfp, vma->vm_mm, &ret);
1923         if (error)
1924                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1925         return ret;
1926 }
1927 
1928 unsigned long shmem_get_unmapped_area(struct file *file,
1929                                       unsigned long uaddr, unsigned long len,
1930                                       unsigned long pgoff, unsigned long flags)
1931 {
1932         unsigned long (*get_area)(struct file *,
1933                 unsigned long, unsigned long, unsigned long, unsigned long);
1934         unsigned long addr;
1935         unsigned long offset;
1936         unsigned long inflated_len;
1937         unsigned long inflated_addr;
1938         unsigned long inflated_offset;
1939 
1940         if (len > TASK_SIZE)
1941                 return -ENOMEM;
1942 
1943         get_area = current->mm->get_unmapped_area;
1944         addr = get_area(file, uaddr, len, pgoff, flags);
1945 
1946         if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1947                 return addr;
1948         if (IS_ERR_VALUE(addr))
1949                 return addr;
1950         if (addr & ~PAGE_MASK)
1951                 return addr;
1952         if (addr > TASK_SIZE - len)
1953                 return addr;
1954 
1955         if (shmem_huge == SHMEM_HUGE_DENY)
1956                 return addr;
1957         if (len < HPAGE_PMD_SIZE)
1958                 return addr;
1959         if (flags & MAP_FIXED)
1960                 return addr;
1961         /*
1962          * Our priority is to support MAP_SHARED mapped hugely;
1963          * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1964          * But if caller specified an address hint, respect that as before.
1965          */
1966         if (uaddr)
1967                 return addr;
1968 
1969         if (shmem_huge != SHMEM_HUGE_FORCE) {
1970                 struct super_block *sb;
1971 
1972                 if (file) {
1973                         VM_BUG_ON(file->f_op != &shmem_file_operations);
1974                         sb = file_inode(file)->i_sb;
1975                 } else {
1976                         /*
1977                          * Called directly from mm/mmap.c, or drivers/char/mem.c
1978                          * for "/dev/zero", to create a shared anonymous object.
1979                          */
1980                         if (IS_ERR(shm_mnt))
1981                                 return addr;
1982                         sb = shm_mnt->mnt_sb;
1983                 }
1984                 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
1985                         return addr;
1986         }
1987 
1988         offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
1989         if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
1990                 return addr;
1991         if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
1992                 return addr;
1993 
1994         inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
1995         if (inflated_len > TASK_SIZE)
1996                 return addr;
1997         if (inflated_len < len)
1998                 return addr;
1999 
2000         inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2001         if (IS_ERR_VALUE(inflated_addr))
2002                 return addr;
2003         if (inflated_addr & ~PAGE_MASK)
2004                 return addr;
2005 
2006         inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2007         inflated_addr += offset - inflated_offset;
2008         if (inflated_offset > offset)
2009                 inflated_addr += HPAGE_PMD_SIZE;
2010 
2011         if (inflated_addr > TASK_SIZE - len)
2012                 return addr;
2013         return inflated_addr;
2014 }
2015 
2016 #ifdef CONFIG_NUMA
2017 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2018 {
2019         struct inode *inode = file_inode(vma->vm_file);
2020         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2021 }
2022 
2023 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2024                                           unsigned long addr)
2025 {
2026         struct inode *inode = file_inode(vma->vm_file);
2027         pgoff_t index;
2028 
2029         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2030         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2031 }
2032 #endif
2033 
2034 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2035 {
2036         struct inode *inode = file_inode(file);
2037         struct shmem_inode_info *info = SHMEM_I(inode);
2038         int retval = -ENOMEM;
2039 
2040         spin_lock_irq(&info->lock);
2041         if (lock && !(info->flags & VM_LOCKED)) {
2042                 if (!user_shm_lock(inode->i_size, user))
2043                         goto out_nomem;
2044                 info->flags |= VM_LOCKED;
2045                 mapping_set_unevictable(file->f_mapping);
2046         }
2047         if (!lock && (info->flags & VM_LOCKED) && user) {
2048                 user_shm_unlock(inode->i_size, user);
2049                 info->flags &= ~VM_LOCKED;
2050                 mapping_clear_unevictable(file->f_mapping);
2051         }
2052         retval = 0;
2053 
2054 out_nomem:
2055         spin_unlock_irq(&info->lock);
2056         return retval;
2057 }
2058 
2059 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2060 {
2061         file_accessed(file);
2062         vma->vm_ops = &shmem_vm_ops;
2063         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2064                         ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2065                         (vma->vm_end & HPAGE_PMD_MASK)) {
2066                 khugepaged_enter(vma, vma->vm_flags);
2067         }
2068         return 0;
2069 }
2070 
2071 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2072                                      umode_t mode, dev_t dev, unsigned long flags)
2073 {
2074         struct inode *inode;
2075         struct shmem_inode_info *info;
2076         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2077 
2078         if (shmem_reserve_inode(sb))
2079                 return NULL;
2080 
2081         inode = new_inode(sb);
2082         if (inode) {
2083                 inode->i_ino = get_next_ino();
2084                 inode_init_owner(inode, dir, mode);
2085                 inode->i_blocks = 0;
2086                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2087                 inode->i_generation = get_seconds();
2088                 info = SHMEM_I(inode);
2089                 memset(info, 0, (char *)inode - (char *)info);
2090                 spin_lock_init(&info->lock);
2091                 info->seals = F_SEAL_SEAL;
2092                 info->flags = flags & VM_NORESERVE;
2093                 INIT_LIST_HEAD(&info->shrinklist);
2094                 INIT_LIST_HEAD(&info->swaplist);
2095                 simple_xattrs_init(&info->xattrs);
2096                 cache_no_acl(inode);
2097 
2098                 switch (mode & S_IFMT) {
2099                 default:
2100                         inode->i_op = &shmem_special_inode_operations;
2101                         init_special_inode(inode, mode, dev);
2102                         break;
2103                 case S_IFREG:
2104                         inode->i_mapping->a_ops = &shmem_aops;
2105                         inode->i_op = &shmem_inode_operations;
2106                         inode->i_fop = &shmem_file_operations;
2107                         mpol_shared_policy_init(&info->policy,
2108                                                  shmem_get_sbmpol(sbinfo));
2109                         break;
2110                 case S_IFDIR:
2111                         inc_nlink(inode);
2112                         /* Some things misbehave if size == 0 on a directory */
2113                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
2114                         inode->i_op = &shmem_dir_inode_operations;
2115                         inode->i_fop = &simple_dir_operations;
2116                         break;
2117                 case S_IFLNK:
2118                         /*
2119                          * Must not load anything in the rbtree,
2120                          * mpol_free_shared_policy will not be called.
2121                          */
2122                         mpol_shared_policy_init(&info->policy, NULL);
2123                         break;
2124                 }
2125         } else
2126                 shmem_free_inode(sb);
2127         return inode;
2128 }
2129 
2130 bool shmem_mapping(struct address_space *mapping)
2131 {
2132         if (!mapping->host)
2133                 return false;
2134 
2135         return mapping->host->i_sb->s_op == &shmem_ops;
2136 }
2137 
2138 #ifdef CONFIG_TMPFS
2139 static const struct inode_operations shmem_symlink_inode_operations;
2140 static const struct inode_operations shmem_short_symlink_operations;
2141 
2142 #ifdef CONFIG_TMPFS_XATTR
2143 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2144 #else
2145 #define shmem_initxattrs NULL
2146 #endif
2147 
2148 static int
2149 shmem_write_begin(struct file *file, struct address_space *mapping,
2150                         loff_t pos, unsigned len, unsigned flags,
2151                         struct page **pagep, void **fsdata)
2152 {
2153         struct inode *inode = mapping->host;
2154         struct shmem_inode_info *info = SHMEM_I(inode);
2155         pgoff_t index = pos >> PAGE_SHIFT;
2156 
2157         /* i_mutex is held by caller */
2158         if (unlikely(info->seals)) {
2159                 if (info->seals & F_SEAL_WRITE)
2160                         return -EPERM;
2161                 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2162                         return -EPERM;
2163         }
2164 
2165         return shmem_getpage(inode, index, pagep, SGP_WRITE);
2166 }
2167 
2168 static int
2169 shmem_write_end(struct file *file, struct address_space *mapping,
2170                         loff_t pos, unsigned len, unsigned copied,
2171                         struct page *page, void *fsdata)
2172 {
2173         struct inode *inode = mapping->host;
2174 
2175         if (pos + copied > inode->i_size)
2176                 i_size_write(inode, pos + copied);
2177 
2178         if (!PageUptodate(page)) {
2179                 struct page *head = compound_head(page);
2180                 if (PageTransCompound(page)) {
2181                         int i;
2182 
2183                         for (i = 0; i < HPAGE_PMD_NR; i++) {
2184                                 if (head + i == page)
2185                                         continue;
2186                                 clear_highpage(head + i);
2187                                 flush_dcache_page(head + i);
2188                         }
2189                 }
2190                 if (copied < PAGE_SIZE) {
2191                         unsigned from = pos & (PAGE_SIZE - 1);
2192                         zero_user_segments(page, 0, from,
2193                                         from + copied, PAGE_SIZE);
2194                 }
2195                 SetPageUptodate(head);
2196         }
2197         set_page_dirty(page);
2198         unlock_page(page);
2199         put_page(page);
2200 
2201         return copied;
2202 }
2203 
2204 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2205 {
2206         struct file *file = iocb->ki_filp;
2207         struct inode *inode = file_inode(file);
2208         struct address_space *mapping = inode->i_mapping;
2209         pgoff_t index;
2210         unsigned long offset;
2211         enum sgp_type sgp = SGP_READ;
2212         int error = 0;
2213         ssize_t retval = 0;
2214         loff_t *ppos = &iocb->ki_pos;
2215 
2216         /*
2217          * Might this read be for a stacking filesystem?  Then when reading
2218          * holes of a sparse file, we actually need to allocate those pages,
2219          * and even mark them dirty, so it cannot exceed the max_blocks limit.
2220          */
2221         if (!iter_is_iovec(to))
2222                 sgp = SGP_CACHE;
2223 
2224         index = *ppos >> PAGE_SHIFT;
2225         offset = *ppos & ~PAGE_MASK;
2226 
2227         for (;;) {
2228                 struct page *page = NULL;
2229                 pgoff_t end_index;
2230                 unsigned long nr, ret;
2231                 loff_t i_size = i_size_read(inode);
2232 
2233                 end_index = i_size >> PAGE_SHIFT;
2234                 if (index > end_index)
2235                         break;
2236                 if (index == end_index) {
2237                         nr = i_size & ~PAGE_MASK;
2238                         if (nr <= offset)
2239                                 break;
2240                 }
2241 
2242                 error = shmem_getpage(inode, index, &page, sgp);
2243                 if (error) {
2244                         if (error == -EINVAL)
2245                                 error = 0;
2246                         break;
2247                 }
2248                 if (page) {
2249                         if (sgp == SGP_CACHE)
2250                                 set_page_dirty(page);
2251                         unlock_page(page);
2252                 }
2253 
2254                 /*
2255                  * We must evaluate after, since reads (unlike writes)
2256                  * are called without i_mutex protection against truncate
2257                  */
2258                 nr = PAGE_SIZE;
2259                 i_size = i_size_read(inode);
2260                 end_index = i_size >> PAGE_SHIFT;
2261                 if (index == end_index) {
2262                         nr = i_size & ~PAGE_MASK;
2263                         if (nr <= offset) {
2264                                 if (page)
2265                                         put_page(page);
2266                                 break;
2267                         }
2268                 }
2269                 nr -= offset;
2270 
2271                 if (page) {
2272                         /*
2273                          * If users can be writing to this page using arbitrary
2274                          * virtual addresses, take care about potential aliasing
2275                          * before reading the page on the kernel side.
2276                          */
2277                         if (mapping_writably_mapped(mapping))
2278                                 flush_dcache_page(page);
2279                         /*
2280                          * Mark the page accessed if we read the beginning.
2281                          */
2282                         if (!offset)
2283                                 mark_page_accessed(page);
2284                 } else {
2285                         page = ZERO_PAGE(0);
2286                         get_page(page);
2287                 }
2288 
2289                 /*
2290                  * Ok, we have the page, and it's up-to-date, so
2291                  * now we can copy it to user space...
2292                  */
2293                 ret = copy_page_to_iter(page, offset, nr, to);
2294                 retval += ret;
2295                 offset += ret;
2296                 index += offset >> PAGE_SHIFT;
2297                 offset &= ~PAGE_MASK;
2298 
2299                 put_page(page);
2300                 if (!iov_iter_count(to))
2301                         break;
2302                 if (ret < nr) {
2303                         error = -EFAULT;
2304                         break;
2305                 }
2306                 cond_resched();
2307         }
2308 
2309         *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2310         file_accessed(file);
2311         return retval ? retval : error;
2312 }
2313 
2314 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2315                                 struct pipe_inode_info *pipe, size_t len,
2316                                 unsigned int flags)
2317 {
2318         struct address_space *mapping = in->f_mapping;
2319         struct inode *inode = mapping->host;
2320         unsigned int loff, nr_pages, req_pages;
2321         struct page *pages[PIPE_DEF_BUFFERS];
2322         struct partial_page partial[PIPE_DEF_BUFFERS];
2323         struct page *page;
2324         pgoff_t index, end_index;
2325         loff_t isize, left;
2326         int error, page_nr;
2327         struct splice_pipe_desc spd = {
2328                 .pages = pages,
2329                 .partial = partial,
2330                 .nr_pages_max = PIPE_DEF_BUFFERS,
2331                 .flags = flags,
2332                 .ops = &page_cache_pipe_buf_ops,
2333                 .spd_release = spd_release_page,
2334         };
2335 
2336         isize = i_size_read(inode);
2337         if (unlikely(*ppos >= isize))
2338                 return 0;
2339 
2340         left = isize - *ppos;
2341         if (unlikely(left < len))
2342                 len = left;
2343 
2344         if (splice_grow_spd(pipe, &spd))
2345                 return -ENOMEM;
2346 
2347         index = *ppos >> PAGE_SHIFT;
2348         loff = *ppos & ~PAGE_MASK;
2349         req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT;
2350         nr_pages = min(req_pages, spd.nr_pages_max);
2351 
2352         spd.nr_pages = find_get_pages_contig(mapping, index,
2353                                                 nr_pages, spd.pages);
2354         index += spd.nr_pages;
2355         error = 0;
2356 
2357         while (spd.nr_pages < nr_pages) {
2358                 error = shmem_getpage(inode, index, &page, SGP_CACHE);
2359                 if (error)
2360                         break;
2361                 unlock_page(page);
2362                 spd.pages[spd.nr_pages++] = page;
2363                 index++;
2364         }
2365 
2366         index = *ppos >> PAGE_SHIFT;
2367         nr_pages = spd.nr_pages;
2368         spd.nr_pages = 0;
2369 
2370         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
2371                 unsigned int this_len;
2372 
2373                 if (!len)
2374                         break;
2375 
2376                 this_len = min_t(unsigned long, len, PAGE_SIZE - loff);
2377                 page = spd.pages[page_nr];
2378 
2379                 if (!PageUptodate(page) || page->mapping != mapping) {
2380                         error = shmem_getpage(inode, index, &page, SGP_CACHE);
2381                         if (error)
2382                                 break;
2383                         unlock_page(page);
2384                         put_page(spd.pages[page_nr]);
2385                         spd.pages[page_nr] = page;
2386                 }
2387 
2388                 isize = i_size_read(inode);
2389                 end_index = (isize - 1) >> PAGE_SHIFT;
2390                 if (unlikely(!isize || index > end_index))
2391                         break;
2392 
2393                 if (end_index == index) {
2394                         unsigned int plen;
2395 
2396                         plen = ((isize - 1) & ~PAGE_MASK) + 1;
2397                         if (plen <= loff)
2398                                 break;
2399 
2400                         this_len = min(this_len, plen - loff);
2401                         len = this_len;
2402                 }
2403 
2404                 spd.partial[page_nr].offset = loff;
2405                 spd.partial[page_nr].len = this_len;
2406                 len -= this_len;
2407                 loff = 0;
2408                 spd.nr_pages++;
2409                 index++;
2410         }
2411 
2412         while (page_nr < nr_pages)
2413                 put_page(spd.pages[page_nr++]);
2414 
2415         if (spd.nr_pages)
2416                 error = splice_to_pipe(pipe, &spd);
2417 
2418         splice_shrink_spd(&spd);
2419 
2420         if (error > 0) {
2421                 *ppos += error;
2422                 file_accessed(in);
2423         }
2424         return error;
2425 }
2426 
2427 /*
2428  * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2429  */
2430 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2431                                     pgoff_t index, pgoff_t end, int whence)
2432 {
2433         struct page *page;
2434         struct pagevec pvec;
2435         pgoff_t indices[PAGEVEC_SIZE];
2436         bool done = false;
2437         int i;
2438 
2439         pagevec_init(&pvec, 0);
2440         pvec.nr = 1;            /* start small: we may be there already */
2441         while (!done) {
2442                 pvec.nr = find_get_entries(mapping, index,
2443                                         pvec.nr, pvec.pages, indices);
2444                 if (!pvec.nr) {
2445                         if (whence == SEEK_DATA)
2446                                 index = end;
2447                         break;
2448                 }
2449                 for (i = 0; i < pvec.nr; i++, index++) {
2450                         if (index < indices[i]) {
2451                                 if (whence == SEEK_HOLE) {
2452                                         done = true;
2453                                         break;
2454                                 }
2455                                 index = indices[i];
2456                         }
2457                         page = pvec.pages[i];
2458                         if (page && !radix_tree_exceptional_entry(page)) {
2459                                 if (!PageUptodate(page))
2460                                         page = NULL;
2461                         }
2462                         if (index >= end ||
2463                             (page && whence == SEEK_DATA) ||
2464                             (!page && whence == SEEK_HOLE)) {
2465                                 done = true;
2466                                 break;
2467                         }
2468                 }
2469                 pagevec_remove_exceptionals(&pvec);
2470                 pagevec_release(&pvec);
2471                 pvec.nr = PAGEVEC_SIZE;
2472                 cond_resched();
2473         }
2474         return index;
2475 }
2476 
2477 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2478 {
2479         struct address_space *mapping = file->f_mapping;
2480         struct inode *inode = mapping->host;
2481         pgoff_t start, end;
2482         loff_t new_offset;
2483 
2484         if (whence != SEEK_DATA && whence != SEEK_HOLE)
2485                 return generic_file_llseek_size(file, offset, whence,
2486                                         MAX_LFS_FILESIZE, i_size_read(inode));
2487         inode_lock(inode);
2488         /* We're holding i_mutex so we can access i_size directly */
2489 
2490         if (offset < 0)
2491                 offset = -EINVAL;
2492         else if (offset >= inode->i_size)
2493                 offset = -ENXIO;
2494         else {
2495                 start = offset >> PAGE_SHIFT;
2496                 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2497                 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2498                 new_offset <<= PAGE_SHIFT;
2499                 if (new_offset > offset) {
2500                         if (new_offset < inode->i_size)
2501                                 offset = new_offset;
2502                         else if (whence == SEEK_DATA)
2503                                 offset = -ENXIO;
2504                         else
2505                                 offset = inode->i_size;
2506                 }
2507         }
2508 
2509         if (offset >= 0)
2510                 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2511         inode_unlock(inode);
2512         return offset;
2513 }
2514 
2515 /*
2516  * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2517  * so reuse a tag which we firmly believe is never set or cleared on shmem.
2518  */
2519 #define SHMEM_TAG_PINNED        PAGECACHE_TAG_TOWRITE
2520 #define LAST_SCAN               4       /* about 150ms max */
2521 
2522 static void shmem_tag_pins(struct address_space *mapping)
2523 {
2524         struct radix_tree_iter iter;
2525         void **slot;
2526         pgoff_t start;
2527         struct page *page;
2528 
2529         lru_add_drain();
2530         start = 0;
2531         rcu_read_lock();
2532 
2533         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2534                 page = radix_tree_deref_slot(slot);
2535                 if (!page || radix_tree_exception(page)) {
2536                         if (radix_tree_deref_retry(page)) {
2537                                 slot = radix_tree_iter_retry(&iter);
2538                                 continue;
2539                         }
2540                 } else if (page_count(page) - page_mapcount(page) > 1) {
2541                         spin_lock_irq(&mapping->tree_lock);
2542                         radix_tree_tag_set(&mapping->page_tree, iter.index,
2543                                            SHMEM_TAG_PINNED);
2544                         spin_unlock_irq(&mapping->tree_lock);
2545                 }
2546 
2547                 if (need_resched()) {
2548                         cond_resched_rcu();
2549                         slot = radix_tree_iter_next(&iter);
2550                 }
2551         }
2552         rcu_read_unlock();
2553 }
2554 
2555 /*
2556  * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2557  * via get_user_pages(), drivers might have some pending I/O without any active
2558  * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2559  * and see whether it has an elevated ref-count. If so, we tag them and wait for
2560  * them to be dropped.
2561  * The caller must guarantee that no new user will acquire writable references
2562  * to those pages to avoid races.
2563  */
2564 static int shmem_wait_for_pins(struct address_space *mapping)
2565 {
2566         struct radix_tree_iter iter;
2567         void **slot;
2568         pgoff_t start;
2569         struct page *page;
2570         int error, scan;
2571 
2572         shmem_tag_pins(mapping);
2573 
2574         error = 0;
2575         for (scan = 0; scan <= LAST_SCAN; scan++) {
2576                 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2577                         break;
2578 
2579                 if (!scan)
2580                         lru_add_drain_all();
2581                 else if (schedule_timeout_killable((HZ << scan) / 200))
2582                         scan = LAST_SCAN;
2583 
2584                 start = 0;
2585                 rcu_read_lock();
2586                 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2587                                            start, SHMEM_TAG_PINNED) {
2588 
2589                         page = radix_tree_deref_slot(slot);
2590                         if (radix_tree_exception(page)) {
2591                                 if (radix_tree_deref_retry(page)) {
2592                                         slot = radix_tree_iter_retry(&iter);
2593                                         continue;
2594                                 }
2595 
2596                                 page = NULL;
2597                         }
2598 
2599                         if (page &&
2600                             page_count(page) - page_mapcount(page) != 1) {
2601                                 if (scan < LAST_SCAN)
2602                                         goto continue_resched;
2603 
2604                                 /*
2605                                  * On the last scan, we clean up all those tags
2606                                  * we inserted; but make a note that we still
2607                                  * found pages pinned.
2608                                  */
2609                                 error = -EBUSY;
2610                         }
2611 
2612                         spin_lock_irq(&mapping->tree_lock);
2613                         radix_tree_tag_clear(&mapping->page_tree,
2614                                              iter.index, SHMEM_TAG_PINNED);
2615                         spin_unlock_irq(&mapping->tree_lock);
2616 continue_resched:
2617                         if (need_resched()) {
2618                                 cond_resched_rcu();
2619                                 slot = radix_tree_iter_next(&iter);
2620                         }
2621                 }
2622                 rcu_read_unlock();
2623         }
2624 
2625         return error;
2626 }
2627 
2628 #define F_ALL_SEALS (F_SEAL_SEAL | \
2629                      F_SEAL_SHRINK | \
2630                      F_SEAL_GROW | \
2631                      F_SEAL_WRITE)
2632 
2633 int shmem_add_seals(struct file *file, unsigned int seals)
2634 {
2635         struct inode *inode = file_inode(file);
2636         struct shmem_inode_info *info = SHMEM_I(inode);
2637         int error;
2638 
2639         /*
2640          * SEALING
2641          * Sealing allows multiple parties to share a shmem-file but restrict
2642          * access to a specific subset of file operations. Seals can only be
2643          * added, but never removed. This way, mutually untrusted parties can
2644          * share common memory regions with a well-defined policy. A malicious
2645          * peer can thus never perform unwanted operations on a shared object.
2646          *
2647          * Seals are only supported on special shmem-files and always affect
2648          * the whole underlying inode. Once a seal is set, it may prevent some
2649          * kinds of access to the file. Currently, the following seals are
2650          * defined:
2651          *   SEAL_SEAL: Prevent further seals from being set on this file
2652          *   SEAL_SHRINK: Prevent the file from shrinking
2653          *   SEAL_GROW: Prevent the file from growing
2654          *   SEAL_WRITE: Prevent write access to the file
2655          *
2656          * As we don't require any trust relationship between two parties, we
2657          * must prevent seals from being removed. Therefore, sealing a file
2658          * only adds a given set of seals to the file, it never touches
2659          * existing seals. Furthermore, the "setting seals"-operation can be
2660          * sealed itself, which basically prevents any further seal from being
2661          * added.
2662          *
2663          * Semantics of sealing are only defined on volatile files. Only
2664          * anonymous shmem files support sealing. More importantly, seals are
2665          * never written to disk. Therefore, there's no plan to support it on
2666          * other file types.
2667          */
2668 
2669         if (file->f_op != &shmem_file_operations)
2670                 return -EINVAL;
2671         if (!(file->f_mode & FMODE_WRITE))
2672                 return -EPERM;
2673         if (seals & ~(unsigned int)F_ALL_SEALS)
2674                 return -EINVAL;
2675 
2676         inode_lock(inode);
2677 
2678         if (info->seals & F_SEAL_SEAL) {
2679                 error = -EPERM;
2680                 goto unlock;
2681         }
2682 
2683         if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2684                 error = mapping_deny_writable(file->f_mapping);
2685                 if (error)
2686                         goto unlock;
2687 
2688                 error = shmem_wait_for_pins(file->f_mapping);
2689                 if (error) {
2690                         mapping_allow_writable(file->f_mapping);
2691                         goto unlock;
2692                 }
2693         }
2694 
2695         info->seals |= seals;
2696         error = 0;
2697 
2698 unlock:
2699         inode_unlock(inode);
2700         return error;
2701 }
2702 EXPORT_SYMBOL_GPL(shmem_add_seals);
2703 
2704 int shmem_get_seals(struct file *file)
2705 {
2706         if (file->f_op != &shmem_file_operations)
2707                 return -EINVAL;
2708 
2709         return SHMEM_I(file_inode(file))->seals;
2710 }
2711 EXPORT_SYMBOL_GPL(shmem_get_seals);
2712 
2713 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2714 {
2715         long error;
2716 
2717         switch (cmd) {
2718         case F_ADD_SEALS:
2719                 /* disallow upper 32bit */
2720                 if (arg > UINT_MAX)
2721                         return -EINVAL;
2722 
2723                 error = shmem_add_seals(file, arg);
2724                 break;
2725         case F_GET_SEALS:
2726                 error = shmem_get_seals(file);
2727                 break;
2728         default:
2729                 error = -EINVAL;
2730                 break;
2731         }
2732 
2733         return error;
2734 }
2735 
2736 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2737                                                          loff_t len)
2738 {
2739         struct inode *inode = file_inode(file);
2740         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2741         struct shmem_inode_info *info = SHMEM_I(inode);
2742         struct shmem_falloc shmem_falloc;
2743         pgoff_t start, index, end;
2744         int error;
2745 
2746         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2747                 return -EOPNOTSUPP;
2748 
2749         inode_lock(inode);
2750 
2751         if (mode & FALLOC_FL_PUNCH_HOLE) {
2752                 struct address_space *mapping = file->f_mapping;
2753                 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2754                 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2755                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2756 
2757                 /* protected by i_mutex */
2758                 if (info->seals & F_SEAL_WRITE) {
2759                         error = -EPERM;
2760                         goto out;
2761                 }
2762 
2763                 shmem_falloc.waitq = &shmem_falloc_waitq;
2764                 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2765                 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2766                 spin_lock(&inode->i_lock);
2767                 inode->i_private = &shmem_falloc;
2768                 spin_unlock(&inode->i_lock);
2769 
2770                 if ((u64)unmap_end > (u64)unmap_start)
2771                         unmap_mapping_range(mapping, unmap_start,
2772                                             1 + unmap_end - unmap_start, 0);
2773                 shmem_truncate_range(inode, offset, offset + len - 1);
2774                 /* No need to unmap again: hole-punching leaves COWed pages */
2775 
2776                 spin_lock(&inode->i_lock);
2777                 inode->i_private = NULL;
2778                 wake_up_all(&shmem_falloc_waitq);
2779                 spin_unlock(&inode->i_lock);
2780                 error = 0;
2781                 goto out;
2782         }
2783 
2784         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2785         error = inode_newsize_ok(inode, offset + len);
2786         if (error)
2787                 goto out;
2788 
2789         if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2790                 error = -EPERM;
2791                 goto out;
2792         }
2793 
2794         start = offset >> PAGE_SHIFT;
2795         end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2796         /* Try to avoid a swapstorm if len is impossible to satisfy */
2797         if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2798                 error = -ENOSPC;
2799                 goto out;
2800         }
2801 
2802         shmem_falloc.waitq = NULL;
2803         shmem_falloc.start = start;
2804         shmem_falloc.next  = start;
2805         shmem_falloc.nr_falloced = 0;
2806         shmem_falloc.nr_unswapped = 0;
2807         spin_lock(&inode->i_lock);
2808         inode->i_private = &shmem_falloc;
2809         spin_unlock(&inode->i_lock);
2810 
2811         for (index = start; index < end; index++) {
2812                 struct page *page;
2813 
2814                 /*
2815                  * Good, the fallocate(2) manpage permits EINTR: we may have
2816                  * been interrupted because we are using up too much memory.
2817                  */
2818                 if (signal_pending(current))
2819                         error = -EINTR;
2820                 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2821                         error = -ENOMEM;
2822                 else
2823                         error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2824                 if (error) {
2825                         /* Remove the !PageUptodate pages we added */
2826                         if (index > start) {
2827                                 shmem_undo_range(inode,
2828                                     (loff_t)start << PAGE_SHIFT,
2829                                     ((loff_t)index << PAGE_SHIFT) - 1, true);
2830                         }
2831                         goto undone;
2832                 }
2833 
2834                 /*
2835                  * Inform shmem_writepage() how far we have reached.
2836                  * No need for lock or barrier: we have the page lock.
2837                  */
2838                 shmem_falloc.next++;
2839                 if (!PageUptodate(page))
2840                         shmem_falloc.nr_falloced++;
2841 
2842                 /*
2843                  * If !PageUptodate, leave it that way so that freeable pages
2844                  * can be recognized if we need to rollback on error later.
2845                  * But set_page_dirty so that memory pressure will swap rather
2846                  * than free the pages we are allocating (and SGP_CACHE pages
2847                  * might still be clean: we now need to mark those dirty too).
2848                  */
2849                 set_page_dirty(page);
2850                 unlock_page(page);
2851                 put_page(page);
2852                 cond_resched();
2853         }
2854 
2855         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2856                 i_size_write(inode, offset + len);
2857         inode->i_ctime = CURRENT_TIME;
2858 undone:
2859         spin_lock(&inode->i_lock);
2860         inode->i_private = NULL;
2861         spin_unlock(&inode->i_lock);
2862 out:
2863         inode_unlock(inode);
2864         return error;
2865 }
2866 
2867 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2868 {
2869         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2870 
2871         buf->f_type = TMPFS_MAGIC;
2872         buf->f_bsize = PAGE_SIZE;
2873         buf->f_namelen = NAME_MAX;
2874         if (sbinfo->max_blocks) {
2875                 buf->f_blocks = sbinfo->max_blocks;
2876                 buf->f_bavail =
2877                 buf->f_bfree  = sbinfo->max_blocks -
2878                                 percpu_counter_sum(&sbinfo->used_blocks);
2879         }
2880         if (sbinfo->max_inodes) {
2881                 buf->f_files = sbinfo->max_inodes;
2882                 buf->f_ffree = sbinfo->free_inodes;
2883         }
2884         /* else leave those fields 0 like simple_statfs */
2885         return 0;
2886 }
2887 
2888 /*
2889  * File creation. Allocate an inode, and we're done..
2890  */
2891 static int
2892 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2893 {
2894         struct inode *inode;
2895         int error = -ENOSPC;
2896 
2897         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2898         if (inode) {
2899                 error = simple_acl_create(dir, inode);
2900                 if (error)
2901                         goto out_iput;
2902                 error = security_inode_init_security(inode, dir,
2903                                                      &dentry->d_name,
2904                                                      shmem_initxattrs, NULL);
2905                 if (error && error != -EOPNOTSUPP)
2906                         goto out_iput;
2907 
2908                 error = 0;
2909                 dir->i_size += BOGO_DIRENT_SIZE;
2910                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2911                 d_instantiate(dentry, inode);
2912                 dget(dentry); /* Extra count - pin the dentry in core */
2913         }
2914         return error;
2915 out_iput:
2916         iput(inode);
2917         return error;
2918 }
2919 
2920 static int
2921 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2922 {
2923         struct inode *inode;
2924         int error = -ENOSPC;
2925 
2926         inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2927         if (inode) {
2928                 error = security_inode_init_security(inode, dir,
2929                                                      NULL,
2930                                                      shmem_initxattrs, NULL);
2931                 if (error && error != -EOPNOTSUPP)
2932                         goto out_iput;
2933                 error = simple_acl_create(dir, inode);
2934                 if (error)
2935                         goto out_iput;
2936                 d_tmpfile(dentry, inode);
2937         }
2938         return error;
2939 out_iput:
2940         iput(inode);
2941         return error;
2942 }
2943 
2944 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2945 {
2946         int error;
2947 
2948         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2949                 return error;
2950         inc_nlink(dir);
2951         return 0;
2952 }
2953 
2954 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2955                 bool excl)
2956 {
2957         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2958 }
2959 
2960 /*
2961  * Link a file..
2962  */
2963 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2964 {
2965         struct inode *inode = d_inode(old_dentry);
2966         int ret;
2967 
2968         /*
2969          * No ordinary (disk based) filesystem counts links as inodes;
2970          * but each new link needs a new dentry, pinning lowmem, and
2971          * tmpfs dentries cannot be pruned until they are unlinked.
2972          */
2973         ret = shmem_reserve_inode(inode->i_sb);
2974         if (ret)
2975                 goto out;
2976 
2977         dir->i_size += BOGO_DIRENT_SIZE;
2978         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2979         inc_nlink(inode);
2980         ihold(inode);   /* New dentry reference */
2981         dget(dentry);           /* Extra pinning count for the created dentry */
2982         d_instantiate(dentry, inode);
2983 out:
2984         return ret;
2985 }
2986 
2987 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2988 {
2989         struct inode *inode = d_inode(dentry);
2990 
2991         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2992                 shmem_free_inode(inode->i_sb);
2993 
2994         dir->i_size -= BOGO_DIRENT_SIZE;
2995         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2996         drop_nlink(inode);
2997         dput(dentry);   /* Undo the count from "create" - this does all the work */
2998         return 0;
2999 }
3000 
3001 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3002 {
3003         if (!simple_empty(dentry))
3004                 return -ENOTEMPTY;
3005 
3006         drop_nlink(d_inode(dentry));
3007         drop_nlink(dir);
3008         return shmem_unlink(dir, dentry);
3009 }
3010 
3011 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3012 {
3013         bool old_is_dir = d_is_dir(old_dentry);
3014         bool new_is_dir = d_is_dir(new_dentry);
3015 
3016         if (old_dir != new_dir && old_is_dir != new_is_dir) {
3017                 if (old_is_dir) {
3018                         drop_nlink(old_dir);
3019                         inc_nlink(new_dir);
3020                 } else {
3021                         drop_nlink(new_dir);
3022                         inc_nlink(old_dir);
3023                 }
3024         }
3025         old_dir->i_ctime = old_dir->i_mtime =
3026         new_dir->i_ctime = new_dir->i_mtime =
3027         d_inode(old_dentry)->i_ctime =
3028         d_inode(new_dentry)->i_ctime = CURRENT_TIME;
3029 
3030         return 0;
3031 }
3032 
3033 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3034 {
3035         struct dentry *whiteout;
3036         int error;
3037 
3038         whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3039         if (!whiteout)
3040                 return -ENOMEM;
3041 
3042         error = shmem_mknod(old_dir, whiteout,
3043                             S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3044         dput(whiteout);
3045         if (error)
3046                 return error;
3047 
3048         /*
3049          * Cheat and hash the whiteout while the old dentry is still in
3050          * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3051          *
3052          * d_lookup() will consistently find one of them at this point,
3053          * not sure which one, but that isn't even important.
3054          */
3055         d_rehash(whiteout);
3056         return 0;
3057 }
3058 
3059 /*
3060  * The VFS layer already does all the dentry stuff for rename,
3061  * we just have to decrement the usage count for the target if
3062  * it exists so that the VFS layer correctly free's it when it
3063  * gets overwritten.
3064  */
3065 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3066 {
3067         struct inode *inode = d_inode(old_dentry);
3068         int they_are_dirs = S_ISDIR(inode->i_mode);
3069 
3070         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3071                 return -EINVAL;
3072 
3073         if (flags & RENAME_EXCHANGE)
3074                 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3075 
3076         if (!simple_empty(new_dentry))
3077                 return -ENOTEMPTY;
3078 
3079         if (flags & RENAME_WHITEOUT) {
3080                 int error;
3081 
3082                 error = shmem_whiteout(old_dir, old_dentry);
3083                 if (error)
3084                         return error;
3085         }
3086 
3087         if (d_really_is_positive(new_dentry)) {
3088                 (void) shmem_unlink(new_dir, new_dentry);
3089                 if (they_are_dirs) {
3090                         drop_nlink(d_inode(new_dentry));
3091                         drop_nlink(old_dir);
3092                 }
3093         } else if (they_are_dirs) {
3094                 drop_nlink(old_dir);
3095                 inc_nlink(new_dir);
3096         }
3097 
3098         old_dir->i_size -= BOGO_DIRENT_SIZE;
3099         new_dir->i_size += BOGO_DIRENT_SIZE;
3100         old_dir->i_ctime = old_dir->i_mtime =
3101         new_dir->i_ctime = new_dir->i_mtime =
3102         inode->i_ctime = CURRENT_TIME;
3103         return 0;
3104 }
3105 
3106 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3107 {
3108         int error;
3109         int len;
3110         struct inode *inode;
3111         struct page *page;
3112         struct shmem_inode_info *info;
3113 
3114         len = strlen(symname) + 1;
3115         if (len > PAGE_SIZE)
3116                 return -ENAMETOOLONG;
3117 
3118         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3119         if (!inode)
3120                 return -ENOSPC;
3121 
3122         error = security_inode_init_security(inode, dir, &dentry->d_name,
3123                                              shmem_initxattrs, NULL);
3124         if (error) {
3125                 if (error != -EOPNOTSUPP) {
3126                         iput(inode);
3127                         return error;
3128                 }
3129                 error = 0;
3130         }
3131 
3132         info = SHMEM_I(inode);
3133         inode->i_size = len-1;
3134         if (len <= SHORT_SYMLINK_LEN) {
3135                 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3136                 if (!inode->i_link) {
3137                         iput(inode);
3138                         return -ENOMEM;
3139                 }
3140                 inode->i_op = &shmem_short_symlink_operations;
3141         } else {
3142                 inode_nohighmem(inode);
3143                 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3144                 if (error) {
3145                         iput(inode);
3146                         return error;
3147                 }
3148                 inode->i_mapping->a_ops = &shmem_aops;
3149                 inode->i_op = &shmem_symlink_inode_operations;
3150                 memcpy(page_address(page), symname, len);
3151                 SetPageUptodate(page);
3152                 set_page_dirty(page);
3153                 unlock_page(page);
3154                 put_page(page);
3155         }
3156         dir->i_size += BOGO_DIRENT_SIZE;
3157         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
3158         d_instantiate(dentry, inode);
3159         dget(dentry);
3160         return 0;
3161 }
3162 
3163 static void shmem_put_link(void *arg)
3164 {
3165         mark_page_accessed(arg);
3166         put_page(arg);
3167 }
3168 
3169 static const char *shmem_get_link(struct dentry *dentry,
3170                                   struct inode *inode,
3171                                   struct delayed_call *done)
3172 {
3173         struct page *page = NULL;
3174         int error;
3175         if (!dentry) {
3176                 page = find_get_page(inode->i_mapping, 0);
3177                 if (!page)
3178                         return ERR_PTR(-ECHILD);
3179                 if (!PageUptodate(page)) {
3180                         put_page(page);
3181                         return ERR_PTR(-ECHILD);
3182                 }
3183         } else {
3184                 error = shmem_getpage(inode, 0, &page, SGP_READ);
3185                 if (error)
3186                         return ERR_PTR(error);
3187                 unlock_page(page);
3188         }
3189         set_delayed_call(done, shmem_put_link, page);
3190         return page_address(page);
3191 }
3192 
3193 #ifdef CONFIG_TMPFS_XATTR
3194 /*
3195  * Superblocks without xattr inode operations may get some security.* xattr
3196  * support from the LSM "for free". As soon as we have any other xattrs
3197  * like ACLs, we also need to implement the security.* handlers at
3198  * filesystem level, though.
3199  */
3200 
3201 /*
3202  * Callback for security_inode_init_security() for acquiring xattrs.
3203  */
3204 static int shmem_initxattrs(struct inode *inode,
3205                             const struct xattr *xattr_array,
3206                             void *fs_info)
3207 {
3208         struct shmem_inode_info *info = SHMEM_I(inode);
3209         const struct xattr *xattr;
3210         struct simple_xattr *new_xattr;
3211         size_t len;
3212 
3213         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3214                 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3215                 if (!new_xattr)
3216                         return -ENOMEM;
3217 
3218                 len = strlen(xattr->name) + 1;
3219                 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3220                                           GFP_KERNEL);
3221                 if (!new_xattr->name) {
3222                         kfree(new_xattr);
3223                         return -ENOMEM;
3224                 }
3225 
3226                 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3227                        XATTR_SECURITY_PREFIX_LEN);
3228                 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3229                        xattr->name, len);
3230 
3231                 simple_xattr_list_add(&info->xattrs, new_xattr);
3232         }
3233 
3234         return 0;
3235 }
3236 
3237 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3238                                    struct dentry *unused, struct inode *inode,
3239                                    const char *name, void *buffer, size_t size)
3240 {
3241         struct shmem_inode_info *info = SHMEM_I(inode);
3242 
3243         name = xattr_full_name(handler, name);
3244         return simple_xattr_get(&info->xattrs, name, buffer, size);
3245 }
3246 
3247 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3248                                    struct dentry *unused, struct inode *inode,
3249                                    const char *name, const void *value,
3250                                    size_t size, int flags)
3251 {
3252         struct shmem_inode_info *info = SHMEM_I(inode);
3253 
3254         name = xattr_full_name(handler, name);
3255         return simple_xattr_set(&info->xattrs, name, value, size, flags);
3256 }
3257 
3258 static const struct xattr_handler shmem_security_xattr_handler = {
3259         .prefix = XATTR_SECURITY_PREFIX,
3260         .get = shmem_xattr_handler_get,
3261         .set = shmem_xattr_handler_set,
3262 };
3263 
3264 static const struct xattr_handler shmem_trusted_xattr_handler = {
3265         .prefix = XATTR_TRUSTED_PREFIX,
3266         .get = shmem_xattr_handler_get,
3267         .set = shmem_xattr_handler_set,
3268 };
3269 
3270 static const struct xattr_handler *shmem_xattr_handlers[] = {
3271 #ifdef CONFIG_TMPFS_POSIX_ACL
3272         &posix_acl_access_xattr_handler,
3273         &posix_acl_default_xattr_handler,
3274 #endif
3275         &shmem_security_xattr_handler,
3276         &shmem_trusted_xattr_handler,
3277         NULL
3278 };
3279 
3280 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3281 {
3282         struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3283         return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3284 }
3285 #endif /* CONFIG_TMPFS_XATTR */
3286 
3287 static const struct inode_operations shmem_short_symlink_operations = {
3288         .readlink       = generic_readlink,
3289         .get_link       = simple_get_link,
3290 #ifdef CONFIG_TMPFS_XATTR
3291         .setxattr       = generic_setxattr,
3292         .getxattr       = generic_getxattr,
3293         .listxattr      = shmem_listxattr,
3294         .removexattr    = generic_removexattr,
3295 #endif
3296 };
3297 
3298 static const struct inode_operations shmem_symlink_inode_operations = {
3299         .readlink       = generic_readlink,
3300         .get_link       = shmem_get_link,
3301 #ifdef CONFIG_TMPFS_XATTR
3302         .setxattr       = generic_setxattr,
3303         .getxattr       = generic_getxattr,
3304         .listxattr      = shmem_listxattr,
3305         .removexattr    = generic_removexattr,
3306 #endif
3307 };
3308 
3309 static struct dentry *shmem_get_parent(struct dentry *child)
3310 {
3311         return ERR_PTR(-ESTALE);
3312 }
3313 
3314 static int shmem_match(struct inode *ino, void *vfh)
3315 {
3316         __u32 *fh = vfh;
3317         __u64 inum = fh[2];
3318         inum = (inum << 32) | fh[1];
3319         return ino->i_ino == inum && fh[0] == ino->i_generation;
3320 }
3321 
3322 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3323                 struct fid *fid, int fh_len, int fh_type)
3324 {
3325         struct inode *inode;
3326         struct dentry *dentry = NULL;
3327         u64 inum;
3328 
3329         if (fh_len < 3)
3330                 return NULL;
3331 
3332         inum = fid->raw[2];
3333         inum = (inum << 32) | fid->raw[1];
3334 
3335         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3336                         shmem_match, fid->raw);
3337         if (inode) {
3338                 dentry = d_find_alias(inode);
3339                 iput(inode);
3340         }
3341 
3342         return dentry;
3343 }
3344 
3345 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3346                                 struct inode *parent)
3347 {
3348         if (*len < 3) {
3349                 *len = 3;
3350                 return FILEID_INVALID;
3351         }
3352 
3353         if (inode_unhashed(inode)) {
3354                 /* Unfortunately insert_inode_hash is not idempotent,
3355                  * so as we hash inodes here rather than at creation
3356                  * time, we need a lock to ensure we only try
3357                  * to do it once
3358                  */
3359                 static DEFINE_SPINLOCK(lock);
3360                 spin_lock(&lock);
3361                 if (inode_unhashed(inode))
3362                         __insert_inode_hash(inode,
3363                                             inode->i_ino + inode->i_generation);
3364                 spin_unlock(&lock);
3365         }
3366 
3367         fh[0] = inode->i_generation;
3368         fh[1] = inode->i_ino;
3369         fh[2] = ((__u64)inode->i_ino) >> 32;
3370 
3371         *len = 3;
3372         return 1;
3373 }
3374 
3375 static const struct export_operations shmem_export_ops = {
3376         .get_parent     = shmem_get_parent,
3377         .encode_fh      = shmem_encode_fh,
3378         .fh_to_dentry   = shmem_fh_to_dentry,
3379 };
3380 
3381 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3382                                bool remount)
3383 {
3384         char *this_char, *value, *rest;
3385         struct mempolicy *mpol = NULL;
3386         uid_t uid;
3387         gid_t gid;
3388 
3389         while (options != NULL) {
3390                 this_char = options;
3391                 for (;;) {
3392                         /*
3393                          * NUL-terminate this option: unfortunately,
3394                          * mount options form a comma-separated list,
3395                          * but mpol's nodelist may also contain commas.
3396                          */
3397                         options = strchr(options, ',');
3398                         if (options == NULL)
3399                                 break;
3400                         options++;
3401                         if (!isdigit(*options)) {
3402                                 options[-1] = '\0';
3403                                 break;
3404                         }
3405                 }
3406                 if (!*this_char)
3407                         continue;
3408                 if ((value = strchr(this_char,'=')) != NULL) {
3409                         *value++ = 0;
3410                 } else {
3411                         pr_err("tmpfs: No value for mount option '%s'\n",
3412                                this_char);
3413                         goto error;
3414                 }
3415 
3416                 if (!strcmp(this_char,"size")) {
3417                         unsigned long long size;
3418                         size = memparse(value,&rest);
3419                         if (*rest == '%') {
3420                                 size <<= PAGE_SHIFT;
3421                                 size *= totalram_pages;
3422                                 do_div(size, 100);
3423                                 rest++;
3424                         }
3425                         if (*rest)
3426                                 goto bad_val;
3427                         sbinfo->max_blocks =
3428                                 DIV_ROUND_UP(size, PAGE_SIZE);
3429                 } else if (!strcmp(this_char,"nr_blocks")) {
3430                         sbinfo->max_blocks = memparse(value, &rest);
3431                         if (*rest)
3432                                 goto bad_val;
3433                 } else if (!strcmp(this_char,"nr_inodes")) {
3434                         sbinfo->max_inodes = memparse(value, &rest);
3435                         if (*rest)
3436                                 goto bad_val;
3437                 } else if (!strcmp(this_char,"mode")) {
3438                         if (remount)
3439                                 continue;
3440                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3441                         if (*rest)
3442                                 goto bad_val;
3443                 } else if (!strcmp(this_char,"uid")) {
3444                         if (remount)
3445                                 continue;
3446                         uid = simple_strtoul(value, &rest, 0);
3447                         if (*rest)
3448                                 goto bad_val;
3449                         sbinfo->uid = make_kuid(current_user_ns(), uid);
3450                         if (!uid_valid(sbinfo->uid))
3451                                 goto bad_val;
3452                 } else if (!strcmp(this_char,"gid")) {
3453                         if (remount)
3454                                 continue;
3455                         gid = simple_strtoul(value, &rest, 0);
3456                         if (*rest)
3457                                 goto bad_val;
3458                         sbinfo->gid = make_kgid(current_user_ns(), gid);
3459                         if (!gid_valid(sbinfo->gid))
3460                                 goto bad_val;
3461 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3462                 } else if (!strcmp(this_char, "huge")) {
3463                         int huge;
3464                         huge = shmem_parse_huge(value);
3465                         if (huge < 0)
3466                                 goto bad_val;
3467                         if (!has_transparent_hugepage() &&
3468                                         huge != SHMEM_HUGE_NEVER)
3469                                 goto bad_val;
3470                         sbinfo->huge = huge;
3471 #endif
3472 #ifdef CONFIG_NUMA
3473                 } else if (!strcmp(this_char,"mpol")) {
3474                         mpol_put(mpol);
3475                         mpol = NULL;
3476                         if (mpol_parse_str(value, &mpol))
3477                                 goto bad_val;
3478 #endif
3479                 } else {
3480                         pr_err("tmpfs: Bad mount option %s\n", this_char);
3481                         goto error;
3482                 }
3483         }
3484         sbinfo->mpol = mpol;
3485         return 0;
3486 
3487 bad_val:
3488         pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3489                value, this_char);
3490 error:
3491         mpol_put(mpol);
3492         return 1;
3493 
3494 }
3495 
3496 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3497 {
3498         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3499         struct shmem_sb_info config = *sbinfo;
3500         unsigned long inodes;
3501         int error = -EINVAL;
3502 
3503         config.mpol = NULL;
3504         if (shmem_parse_options(data, &config, true))
3505                 return error;
3506 
3507         spin_lock(&sbinfo->stat_lock);
3508         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3509         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3510                 goto out;
3511         if (config.max_inodes < inodes)
3512                 goto out;
3513         /*
3514          * Those tests disallow limited->unlimited while any are in use;
3515          * but we must separately disallow unlimited->limited, because
3516          * in that case we have no record of how much is already in use.
3517          */
3518         if (config.max_blocks && !sbinfo->max_blocks)
3519                 goto out;
3520         if (config.max_inodes && !sbinfo->max_inodes)
3521                 goto out;
3522 
3523         error = 0;
3524         sbinfo->huge = config.huge;
3525         sbinfo->max_blocks  = config.max_blocks;
3526         sbinfo->max_inodes  = config.max_inodes;
3527         sbinfo->free_inodes = config.max_inodes - inodes;
3528 
3529         /*
3530          * Preserve previous mempolicy unless mpol remount option was specified.
3531          */
3532         if (config.mpol) {
3533                 mpol_put(sbinfo->mpol);
3534                 sbinfo->mpol = config.mpol;     /* transfers initial ref */
3535         }
3536 out:
3537         spin_unlock(&sbinfo->stat_lock);
3538         return error;
3539 }
3540 
3541 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3542 {
3543         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3544 
3545         if (sbinfo->max_blocks != shmem_default_max_blocks())
3546                 seq_printf(seq, ",size=%luk",
3547                         sbinfo->max_blocks << (PAGE_SHIFT - 10));
3548         if (sbinfo->max_inodes != shmem_default_max_inodes())
3549                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3550         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3551                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3552         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3553                 seq_printf(seq, ",uid=%u",
3554                                 from_kuid_munged(&init_user_ns, sbinfo->uid));
3555         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3556                 seq_printf(seq, ",gid=%u",
3557                                 from_kgid_munged(&init_user_ns, sbinfo->gid));
3558 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3559         /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3560         if (sbinfo->huge)
3561                 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3562 #endif
3563         shmem_show_mpol(seq, sbinfo->mpol);
3564         return 0;
3565 }
3566 
3567 #define MFD_NAME_PREFIX "memfd:"
3568 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3569 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3570 
3571 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3572 
3573 SYSCALL_DEFINE2(memfd_create,
3574                 const char __user *, uname,
3575                 unsigned int, flags)
3576 {
3577         struct shmem_inode_info *info;
3578         struct file *file;
3579         int fd, error;
3580         char *name;
3581         long len;
3582 
3583         if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3584                 return -EINVAL;
3585 
3586         /* length includes terminating zero */
3587         len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3588         if (len <= 0)
3589                 return -EFAULT;
3590         if (len > MFD_NAME_MAX_LEN + 1)
3591                 return -EINVAL;
3592 
3593         name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3594         if (!name)
3595                 return -ENOMEM;
3596 
3597         strcpy(name, MFD_NAME_PREFIX);
3598         if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3599                 error = -EFAULT;
3600                 goto err_name;
3601         }
3602 
3603         /* terminating-zero may have changed after strnlen_user() returned */
3604         if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3605                 error = -EFAULT;
3606                 goto err_name;
3607         }
3608 
3609         fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3610         if (fd < 0) {
3611                 error = fd;
3612                 goto err_name;
3613         }
3614 
3615         file = shmem_file_setup(name, 0, VM_NORESERVE);
3616         if (IS_ERR(file)) {
3617                 error = PTR_ERR(file);
3618                 goto err_fd;
3619         }
3620         info = SHMEM_I(file_inode(file));
3621         file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3622         file->f_flags |= O_RDWR | O_LARGEFILE;
3623         if (flags & MFD_ALLOW_SEALING)
3624                 info->seals &= ~F_SEAL_SEAL;
3625 
3626         fd_install(fd, file);
3627         kfree(name);
3628         return fd;
3629 
3630 err_fd:
3631         put_unused_fd(fd);
3632 err_name:
3633         kfree(name);
3634         return error;
3635 }
3636 
3637 #endif /* CONFIG_TMPFS */
3638 
3639 static void shmem_put_super(struct super_block *sb)
3640 {
3641         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3642 
3643         percpu_counter_destroy(&sbinfo->used_blocks);
3644         mpol_put(sbinfo->mpol);
3645         kfree(sbinfo);
3646         sb->s_fs_info = NULL;
3647 }
3648 
3649 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3650 {
3651         struct inode *inode;
3652         struct shmem_sb_info *sbinfo;
3653         int err = -ENOMEM;
3654 
3655         /* Round up to L1_CACHE_BYTES to resist false sharing */
3656         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3657                                 L1_CACHE_BYTES), GFP_KERNEL);
3658         if (!sbinfo)
3659                 return -ENOMEM;
3660 
3661         sbinfo->mode = S_IRWXUGO | S_ISVTX;
3662         sbinfo->uid = current_fsuid();
3663         sbinfo->gid = current_fsgid();
3664         sb->s_fs_info = sbinfo;
3665 
3666 #ifdef CONFIG_TMPFS
3667         /*
3668          * Per default we only allow half of the physical ram per
3669          * tmpfs instance, limiting inodes to one per page of lowmem;
3670          * but the internal instance is left unlimited.
3671          */
3672         if (!(sb->s_flags & MS_KERNMOUNT)) {
3673                 sbinfo->max_blocks = shmem_default_max_blocks();
3674                 sbinfo->max_inodes = shmem_default_max_inodes();
3675                 if (shmem_parse_options(data, sbinfo, false)) {
3676                         err = -EINVAL;
3677                         goto failed;
3678                 }
3679         } else {
3680                 sb->s_flags |= MS_NOUSER;
3681         }
3682         sb->s_export_op = &shmem_export_ops;
3683         sb->s_flags |= MS_NOSEC;
3684 #else
3685         sb->s_flags |= MS_NOUSER;
3686 #endif
3687 
3688         spin_lock_init(&sbinfo->stat_lock);
3689         if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3690                 goto failed;
3691         sbinfo->free_inodes = sbinfo->max_inodes;
3692         spin_lock_init(&sbinfo->shrinklist_lock);
3693         INIT_LIST_HEAD(&sbinfo->shrinklist);
3694 
3695         sb->s_maxbytes = MAX_LFS_FILESIZE;
3696         sb->s_blocksize = PAGE_SIZE;
3697         sb->s_blocksize_bits = PAGE_SHIFT;
3698         sb->s_magic = TMPFS_MAGIC;
3699         sb->s_op = &shmem_ops;
3700         sb->s_time_gran = 1;
3701 #ifdef CONFIG_TMPFS_XATTR
3702         sb->s_xattr = shmem_xattr_handlers;
3703 #endif
3704 #ifdef CONFIG_TMPFS_POSIX_ACL
3705         sb->s_flags |= MS_POSIXACL;
3706 #endif
3707 
3708         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3709         if (!inode)
3710                 goto failed;
3711         inode->i_uid = sbinfo->uid;
3712         inode->i_gid = sbinfo->gid;
3713         sb->s_root = d_make_root(inode);
3714         if (!sb->s_root)
3715                 goto failed;
3716         return 0;
3717 
3718 failed:
3719         shmem_put_super(sb);
3720         return err;
3721 }
3722 
3723 static struct kmem_cache *shmem_inode_cachep;
3724 
3725 static struct inode *shmem_alloc_inode(struct super_block *sb)
3726 {
3727         struct shmem_inode_info *info;
3728         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3729         if (!info)
3730                 return NULL;
3731         return &info->vfs_inode;
3732 }
3733 
3734 static void shmem_destroy_callback(struct rcu_head *head)
3735 {
3736         struct inode *inode = container_of(head, struct inode, i_rcu);
3737         if (S_ISLNK(inode->i_mode))
3738                 kfree(inode->i_link);
3739         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3740 }
3741 
3742 static void shmem_destroy_inode(struct inode *inode)
3743 {
3744         if (S_ISREG(inode->i_mode))
3745                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3746         call_rcu(&inode->i_rcu, shmem_destroy_callback);
3747 }
3748 
3749 static void shmem_init_inode(void *foo)
3750 {
3751         struct shmem_inode_info *info = foo;
3752         inode_init_once(&info->vfs_inode);
3753 }
3754 
3755 static int shmem_init_inodecache(void)
3756 {
3757         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3758                                 sizeof(struct shmem_inode_info),
3759                                 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3760         return 0;
3761 }
3762 
3763 static void shmem_destroy_inodecache(void)
3764 {
3765         kmem_cache_destroy(shmem_inode_cachep);
3766 }
3767 
3768 static const struct address_space_operations shmem_aops = {
3769         .writepage      = shmem_writepage,
3770         .set_page_dirty = __set_page_dirty_no_writeback,
3771 #ifdef CONFIG_TMPFS
3772         .write_begin    = shmem_write_begin,
3773         .write_end      = shmem_write_end,
3774 #endif
3775 #ifdef CONFIG_MIGRATION
3776         .migratepage    = migrate_page,
3777 #endif
3778         .error_remove_page = generic_error_remove_page,
3779 };
3780 
3781 static const struct file_operations shmem_file_operations = {
3782         .mmap           = shmem_mmap,
3783         .get_unmapped_area = shmem_get_unmapped_area,
3784 #ifdef CONFIG_TMPFS
3785         .llseek         = shmem_file_llseek,
3786         .read_iter      = shmem_file_read_iter,
3787         .write_iter     = generic_file_write_iter,
3788         .fsync          = noop_fsync,
3789         .splice_read    = shmem_file_splice_read,
3790         .splice_write   = iter_file_splice_write,
3791         .fallocate      = shmem_fallocate,
3792 #endif
3793 };
3794 
3795 static const struct inode_operations shmem_inode_operations = {
3796         .getattr        = shmem_getattr,
3797         .setattr        = shmem_setattr,
3798 #ifdef CONFIG_TMPFS_XATTR
3799         .setxattr       = generic_setxattr,
3800         .getxattr       = generic_getxattr,
3801         .listxattr      = shmem_listxattr,
3802         .removexattr    = generic_removexattr,
3803         .set_acl        = simple_set_acl,
3804 #endif
3805 };
3806 
3807 static const struct inode_operations shmem_dir_inode_operations = {
3808 #ifdef CONFIG_TMPFS
3809         .create         = shmem_create,
3810         .lookup         = simple_lookup,
3811         .link           = shmem_link,
3812         .unlink         = shmem_unlink,
3813         .symlink        = shmem_symlink,
3814         .mkdir          = shmem_mkdir,
3815         .rmdir          = shmem_rmdir,
3816         .mknod          = shmem_mknod,
3817         .rename2        = shmem_rename2,
3818         .tmpfile        = shmem_tmpfile,
3819 #endif
3820 #ifdef CONFIG_TMPFS_XATTR
3821         .setxattr       = generic_setxattr,
3822         .getxattr       = generic_getxattr,
3823         .listxattr      = shmem_listxattr,
3824         .removexattr    = generic_removexattr,
3825 #endif
3826 #ifdef CONFIG_TMPFS_POSIX_ACL
3827         .setattr        = shmem_setattr,
3828         .set_acl        = simple_set_acl,
3829 #endif
3830 };
3831 
3832 static const struct inode_operations shmem_special_inode_operations = {
3833 #ifdef CONFIG_TMPFS_XATTR
3834         .setxattr       = generic_setxattr,
3835         .getxattr       = generic_getxattr,
3836         .listxattr      = shmem_listxattr,
3837         .removexattr    = generic_removexattr,
3838 #endif
3839 #ifdef CONFIG_TMPFS_POSIX_ACL
3840         .setattr        = shmem_setattr,
3841         .set_acl        = simple_set_acl,
3842 #endif
3843 };
3844 
3845 static const struct super_operations shmem_ops = {
3846         .alloc_inode    = shmem_alloc_inode,
3847         .destroy_inode  = shmem_destroy_inode,
3848 #ifdef CONFIG_TMPFS
3849         .statfs         = shmem_statfs,
3850         .remount_fs     = shmem_remount_fs,
3851         .show_options   = shmem_show_options,
3852 #endif
3853         .evict_inode    = shmem_evict_inode,
3854         .drop_inode     = generic_delete_inode,
3855         .put_super      = shmem_put_super,
3856 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3857         .nr_cached_objects      = shmem_unused_huge_count,
3858         .free_cached_objects    = shmem_unused_huge_scan,
3859 #endif
3860 };
3861 
3862 static const struct vm_operations_struct shmem_vm_ops = {
3863         .fault          = shmem_fault,
3864         .map_pages      = filemap_map_pages,
3865 #ifdef CONFIG_NUMA
3866         .set_policy     = shmem_set_policy,
3867         .get_policy     = shmem_get_policy,
3868 #endif
3869 };
3870 
3871 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3872         int flags, const char *dev_name, void *data)
3873 {
3874         return mount_nodev(fs_type, flags, data, shmem_fill_super);
3875 }
3876 
3877 static struct file_system_type shmem_fs_type = {
3878         .owner          = THIS_MODULE,
3879         .name           = "tmpfs",
3880         .mount          = shmem_mount,
3881         .kill_sb        = kill_litter_super,
3882         .fs_flags       = FS_USERNS_MOUNT,
3883 };
3884 
3885 int __init shmem_init(void)
3886 {
3887         int error;
3888 
3889         /* If rootfs called this, don't re-init */
3890         if (shmem_inode_cachep)
3891                 return 0;
3892 
3893         error = shmem_init_inodecache();
3894         if (error)
3895                 goto out3;
3896 
3897         error = register_filesystem(&shmem_fs_type);
3898         if (error) {
3899                 pr_err("Could not register tmpfs\n");
3900                 goto out2;
3901         }
3902 
3903         shm_mnt = kern_mount(&shmem_fs_type);
3904         if (IS_ERR(shm_mnt)) {
3905                 error = PTR_ERR(shm_mnt);
3906                 pr_err("Could not kern_mount tmpfs\n");
3907                 goto out1;
3908         }
3909 
3910 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3911         if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3912                 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3913         else
3914                 shmem_huge = 0; /* just in case it was patched */
3915 #endif
3916         return 0;
3917 
3918 out1:
3919         unregister_filesystem(&shmem_fs_type);
3920 out2:
3921         shmem_destroy_inodecache();
3922 out3:
3923         shm_mnt = ERR_PTR(error);
3924         return error;
3925 }
3926 
3927 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3928 static ssize_t shmem_enabled_show(struct kobject *kobj,
3929                 struct kobj_attribute *attr, char *buf)
3930 {
3931         int values[] = {
3932                 SHMEM_HUGE_ALWAYS,
3933                 SHMEM_HUGE_WITHIN_SIZE,
3934                 SHMEM_HUGE_ADVISE,
3935                 SHMEM_HUGE_NEVER,
3936                 SHMEM_HUGE_DENY,
3937                 SHMEM_HUGE_FORCE,
3938         };
3939         int i, count;
3940 
3941         for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3942                 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3943 
3944                 count += sprintf(buf + count, fmt,
3945                                 shmem_format_huge(values[i]));
3946         }
3947         buf[count - 1] = '\n';
3948         return count;
3949 }
3950 
3951 static ssize_t shmem_enabled_store(struct kobject *kobj,
3952                 struct kobj_attribute *attr, const char *buf, size_t count)
3953 {
3954         char tmp[16];
3955         int huge;
3956 
3957         if (count + 1 > sizeof(tmp))
3958                 return -EINVAL;
3959         memcpy(tmp, buf, count);
3960         tmp[count] = '\0';
3961         if (count && tmp[count - 1] == '\n')
3962                 tmp[count - 1] = '\0';
3963 
3964         huge = shmem_parse_huge(tmp);
3965         if (huge == -EINVAL)
3966                 return -EINVAL;
3967         if (!has_transparent_hugepage() &&
3968                         huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3969                 return -EINVAL;
3970 
3971         shmem_huge = huge;
3972         if (shmem_huge < SHMEM_HUGE_DENY)
3973                 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3974         return count;
3975 }
3976 
3977 struct kobj_attribute shmem_enabled_attr =
3978         __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3979 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3980 
3981 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3982 bool shmem_huge_enabled(struct vm_area_struct *vma)
3983 {
3984         struct inode *inode = file_inode(vma->vm_file);
3985         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3986         loff_t i_size;
3987         pgoff_t off;
3988 
3989         if (shmem_huge == SHMEM_HUGE_FORCE)
3990                 return true;
3991         if (shmem_huge == SHMEM_HUGE_DENY)
3992                 return false;
3993         switch (sbinfo->huge) {
3994                 case SHMEM_HUGE_NEVER:
3995                         return false;
3996                 case SHMEM_HUGE_ALWAYS:
3997                         return true;
3998                 case SHMEM_HUGE_WITHIN_SIZE:
3999                         off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4000                         i_size = round_up(i_size_read(inode), PAGE_SIZE);
4001                         if (i_size >= HPAGE_PMD_SIZE &&
4002                                         i_size >> PAGE_SHIFT >= off)
4003                                 return true;
4004                 case SHMEM_HUGE_ADVISE:
4005                         /* TODO: implement fadvise() hints */
4006                         return (vma->vm_flags & VM_HUGEPAGE);
4007                 default:
4008                         VM_BUG_ON(1);
4009                         return false;
4010         }
4011 }
4012 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4013 
4014 #else /* !CONFIG_SHMEM */
4015 
4016 /*
4017  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4018  *
4019  * This is intended for small system where the benefits of the full
4020  * shmem code (swap-backed and resource-limited) are outweighed by
4021  * their complexity. On systems without swap this code should be
4022  * effectively equivalent, but much lighter weight.
4023  */
4024 
4025 static struct file_system_type shmem_fs_type = {
4026         .name           = "tmpfs",
4027         .mount          = ramfs_mount,
4028         .kill_sb        = kill_litter_super,
4029         .fs_flags       = FS_USERNS_MOUNT,
4030 };
4031 
4032 int __init shmem_init(void)
4033 {
4034         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4035 
4036         shm_mnt = kern_mount(&shmem_fs_type);
4037         BUG_ON(IS_ERR(shm_mnt));
4038 
4039         return 0;
4040 }
4041 
4042 int shmem_unuse(swp_entry_t swap, struct page *page)
4043 {
4044         return 0;
4045 }
4046 
4047 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4048 {
4049         return 0;
4050 }
4051 
4052 void shmem_unlock_mapping(struct address_space *mapping)
4053 {
4054 }
4055 
4056 #ifdef CONFIG_MMU
4057 unsigned long shmem_get_unmapped_area(struct file *file,
4058                                       unsigned long addr, unsigned long len,
4059                                       unsigned long pgoff, unsigned long flags)
4060 {
4061         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4062 }
4063 #endif
4064 
4065 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4066 {
4067         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4068 }
4069 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4070 
4071 #define shmem_vm_ops                            generic_file_vm_ops
4072 #define shmem_file_operations                   ramfs_file_operations
4073 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
4074 #define shmem_acct_size(flags, size)            0
4075 #define shmem_unacct_size(flags, size)          do {} while (0)
4076 
4077 #endif /* CONFIG_SHMEM */
4078 
4079 /* common code */
4080 
4081 static struct dentry_operations anon_ops = {
4082         .d_dname = simple_dname
4083 };
4084 
4085 static struct file *__shmem_file_setup(const char *name, loff_t size,
4086                                        unsigned long flags, unsigned int i_flags)
4087 {
4088         struct file *res;
4089         struct inode *inode;
4090         struct path path;
4091         struct super_block *sb;
4092         struct qstr this;
4093 
4094         if (IS_ERR(shm_mnt))
4095                 return ERR_CAST(shm_mnt);
4096 
4097         if (size < 0 || size > MAX_LFS_FILESIZE)
4098                 return ERR_PTR(-EINVAL);
4099 
4100         if (shmem_acct_size(flags, size))
4101                 return ERR_PTR(-ENOMEM);
4102 
4103         res = ERR_PTR(-ENOMEM);
4104         this.name = name;
4105         this.len = strlen(name);
4106         this.hash = 0; /* will go */
4107         sb = shm_mnt->mnt_sb;
4108         path.mnt = mntget(shm_mnt);
4109         path.dentry = d_alloc_pseudo(sb, &this);
4110         if (!path.dentry)
4111                 goto put_memory;
4112         d_set_d_op(path.dentry, &anon_ops);
4113 
4114         res = ERR_PTR(-ENOSPC);
4115         inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4116         if (!inode)
4117                 goto put_memory;
4118 
4119         inode->i_flags |= i_flags;
4120         d_instantiate(path.dentry, inode);
4121         inode->i_size = size;
4122         clear_nlink(inode);     /* It is unlinked */
4123         res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4124         if (IS_ERR(res))
4125                 goto put_path;
4126 
4127         res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4128                   &shmem_file_operations);
4129         if (IS_ERR(res))
4130                 goto put_path;
4131 
4132         return res;
4133 
4134 put_memory:
4135         shmem_unacct_size(flags, size);
4136 put_path:
4137         path_put(&path);
4138         return res;
4139 }
4140 
4141 /**
4142  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4143  *      kernel internal.  There will be NO LSM permission checks against the
4144  *      underlying inode.  So users of this interface must do LSM checks at a
4145  *      higher layer.  The users are the big_key and shm implementations.  LSM
4146  *      checks are provided at the key or shm level rather than the inode.
4147  * @name: name for dentry (to be seen in /proc/<pid>/maps
4148  * @size: size to be set for the file
4149  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4150  */
4151 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4152 {
4153         return __shmem_file_setup(name, size, flags, S_PRIVATE);
4154 }
4155 
4156 /**
4157  * shmem_file_setup - get an unlinked file living in tmpfs
4158  * @name: name for dentry (to be seen in /proc/<pid>/maps
4159  * @size: size to be set for the file
4160  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4161  */
4162 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4163 {
4164         return __shmem_file_setup(name, size, flags, 0);
4165 }
4166 EXPORT_SYMBOL_GPL(shmem_file_setup);
4167 
4168 /**
4169  * shmem_zero_setup - setup a shared anonymous mapping
4170  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4171  */
4172 int shmem_zero_setup(struct vm_area_struct *vma)
4173 {
4174         struct file *file;
4175         loff_t size = vma->vm_end - vma->vm_start;
4176 
4177         /*
4178          * Cloning a new file under mmap_sem leads to a lock ordering conflict
4179          * between XFS directory reading and selinux: since this file is only
4180          * accessible to the user through its mapping, use S_PRIVATE flag to
4181          * bypass file security, in the same way as shmem_kernel_file_setup().
4182          */
4183         file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4184         if (IS_ERR(file))
4185                 return PTR_ERR(file);
4186 
4187         if (vma->vm_file)
4188                 fput(vma->vm_file);
4189         vma->vm_file = file;
4190         vma->vm_ops = &shmem_vm_ops;
4191 
4192         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4193                         ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4194                         (vma->vm_end & HPAGE_PMD_MASK)) {
4195                 khugepaged_enter(vma, vma->vm_flags);
4196         }
4197 
4198         return 0;
4199 }
4200 
4201 /**
4202  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4203  * @mapping:    the page's address_space
4204  * @index:      the page index
4205  * @gfp:        the page allocator flags to use if allocating
4206  *
4207  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4208  * with any new page allocations done using the specified allocation flags.
4209  * But read_cache_page_gfp() uses the ->readpage() method: which does not
4210  * suit tmpfs, since it may have pages in swapcache, and needs to find those
4211  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4212  *
4213  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4214  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4215  */
4216 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4217                                          pgoff_t index, gfp_t gfp)
4218 {
4219 #ifdef CONFIG_SHMEM
4220         struct inode *inode = mapping->host;
4221         struct page *page;
4222         int error;
4223 
4224         BUG_ON(mapping->a_ops != &shmem_aops);
4225         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4226                                   gfp, NULL, NULL);
4227         if (error)
4228                 page = ERR_PTR(error);
4229         else
4230                 unlock_page(page);
4231         return page;
4232 #else
4233         /*
4234          * The tiny !SHMEM case uses ramfs without swap
4235          */
4236         return read_cache_page_gfp(mapping, index, gfp);
4237 #endif
4238 }
4239 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
4240 

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