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

TOMOYO Linux Cross Reference
Linux/mm/shmem.c

Version: ~ [ linux-5.10-rc1 ] ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Resizable virtual memory filesystem for Linux.
  3  *
  4  * Copyright (C) 2000 Linus Torvalds.
  5  *               2000 Transmeta Corp.
  6  *               2000-2001 Christoph Rohland
  7  *               2000-2001 SAP AG
  8  *               2002 Red Hat Inc.
  9  * Copyright (C) 2002-2011 Hugh Dickins.
 10  * Copyright (C) 2011 Google Inc.
 11  * Copyright (C) 2002-2005 VERITAS Software Corporation.
 12  * Copyright (C) 2004 Andi Kleen, SuSE Labs
 13  *
 14  * Extended attribute support for tmpfs:
 15  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
 16  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
 17  *
 18  * tiny-shmem:
 19  * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
 20  *
 21  * This file is released under the GPL.
 22  */
 23 
 24 #include <linux/fs.h>
 25 #include <linux/init.h>
 26 #include <linux/vfs.h>
 27 #include <linux/mount.h>
 28 #include <linux/ramfs.h>
 29 #include <linux/pagemap.h>
 30 #include <linux/file.h>
 31 #include <linux/mm.h>
 32 #include <linux/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         __SetPageLocked(newpage);
1487         __SetPageSwapBacked(newpage);
1488         SetPageUptodate(newpage);
1489         set_page_private(newpage, swap_index);
1490         SetPageSwapCache(newpage);
1491 
1492         /*
1493          * Our caller will very soon move newpage out of swapcache, but it's
1494          * a nice clean interface for us to replace oldpage by newpage there.
1495          */
1496         spin_lock_irq(&swap_mapping->tree_lock);
1497         error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1498                                                                    newpage);
1499         if (!error) {
1500                 __inc_node_page_state(newpage, NR_FILE_PAGES);
1501                 __dec_node_page_state(oldpage, NR_FILE_PAGES);
1502         }
1503         spin_unlock_irq(&swap_mapping->tree_lock);
1504 
1505         if (unlikely(error)) {
1506                 /*
1507                  * Is this possible?  I think not, now that our callers check
1508                  * both PageSwapCache and page_private after getting page lock;
1509                  * but be defensive.  Reverse old to newpage for clear and free.
1510                  */
1511                 oldpage = newpage;
1512         } else {
1513                 mem_cgroup_migrate(oldpage, newpage);
1514                 lru_cache_add_anon(newpage);
1515                 *pagep = newpage;
1516         }
1517 
1518         ClearPageSwapCache(oldpage);
1519         set_page_private(oldpage, 0);
1520 
1521         unlock_page(oldpage);
1522         put_page(oldpage);
1523         put_page(oldpage);
1524         return error;
1525 }
1526 
1527 /*
1528  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1529  *
1530  * If we allocate a new one we do not mark it dirty. That's up to the
1531  * vm. If we swap it in we mark it dirty since we also free the swap
1532  * entry since a page cannot live in both the swap and page cache.
1533  *
1534  * fault_mm and fault_type are only supplied by shmem_fault:
1535  * otherwise they are NULL.
1536  */
1537 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1538         struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1539         struct mm_struct *fault_mm, int *fault_type)
1540 {
1541         struct address_space *mapping = inode->i_mapping;
1542         struct shmem_inode_info *info;
1543         struct shmem_sb_info *sbinfo;
1544         struct mm_struct *charge_mm;
1545         struct mem_cgroup *memcg;
1546         struct page *page;
1547         swp_entry_t swap;
1548         enum sgp_type sgp_huge = sgp;
1549         pgoff_t hindex = index;
1550         int error;
1551         int once = 0;
1552         int alloced = 0;
1553 
1554         if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1555                 return -EFBIG;
1556         if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1557                 sgp = SGP_CACHE;
1558 repeat:
1559         swap.val = 0;
1560         page = find_lock_entry(mapping, index);
1561         if (radix_tree_exceptional_entry(page)) {
1562                 swap = radix_to_swp_entry(page);
1563                 page = NULL;
1564         }
1565 
1566         if (sgp <= SGP_CACHE &&
1567             ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1568                 error = -EINVAL;
1569                 goto unlock;
1570         }
1571 
1572         if (page && sgp == SGP_WRITE)
1573                 mark_page_accessed(page);
1574 
1575         /* fallocated page? */
1576         if (page && !PageUptodate(page)) {
1577                 if (sgp != SGP_READ)
1578                         goto clear;
1579                 unlock_page(page);
1580                 put_page(page);
1581                 page = NULL;
1582         }
1583         if (page || (sgp == SGP_READ && !swap.val)) {
1584                 *pagep = page;
1585                 return 0;
1586         }
1587 
1588         /*
1589          * Fast cache lookup did not find it:
1590          * bring it back from swap or allocate.
1591          */
1592         info = SHMEM_I(inode);
1593         sbinfo = SHMEM_SB(inode->i_sb);
1594         charge_mm = fault_mm ? : current->mm;
1595 
1596         if (swap.val) {
1597                 /* Look it up and read it in.. */
1598                 page = lookup_swap_cache(swap);
1599                 if (!page) {
1600                         /* Or update major stats only when swapin succeeds?? */
1601                         if (fault_type) {
1602                                 *fault_type |= VM_FAULT_MAJOR;
1603                                 count_vm_event(PGMAJFAULT);
1604                                 mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT);
1605                         }
1606                         /* Here we actually start the io */
1607                         page = shmem_swapin(swap, gfp, info, index);
1608                         if (!page) {
1609                                 error = -ENOMEM;
1610                                 goto failed;
1611                         }
1612                 }
1613 
1614                 /* We have to do this with page locked to prevent races */
1615                 lock_page(page);
1616                 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1617                     !shmem_confirm_swap(mapping, index, swap)) {
1618                         error = -EEXIST;        /* try again */
1619                         goto unlock;
1620                 }
1621                 if (!PageUptodate(page)) {
1622                         error = -EIO;
1623                         goto failed;
1624                 }
1625                 wait_on_page_writeback(page);
1626 
1627                 if (shmem_should_replace_page(page, gfp)) {
1628                         error = shmem_replace_page(&page, gfp, info, index);
1629                         if (error)
1630                                 goto failed;
1631                 }
1632 
1633                 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1634                                 false);
1635                 if (!error) {
1636                         error = shmem_add_to_page_cache(page, mapping, index,
1637                                                 swp_to_radix_entry(swap));
1638                         /*
1639                          * We already confirmed swap under page lock, and make
1640                          * no memory allocation here, so usually no possibility
1641                          * of error; but free_swap_and_cache() only trylocks a
1642                          * page, so it is just possible that the entry has been
1643                          * truncated or holepunched since swap was confirmed.
1644                          * shmem_undo_range() will have done some of the
1645                          * unaccounting, now delete_from_swap_cache() will do
1646                          * the rest.
1647                          * Reset swap.val? No, leave it so "failed" goes back to
1648                          * "repeat": reading a hole and writing should succeed.
1649                          */
1650                         if (error) {
1651                                 mem_cgroup_cancel_charge(page, memcg, false);
1652                                 delete_from_swap_cache(page);
1653                         }
1654                 }
1655                 if (error)
1656                         goto failed;
1657 
1658                 mem_cgroup_commit_charge(page, memcg, true, false);
1659 
1660                 spin_lock_irq(&info->lock);
1661                 info->swapped--;
1662                 shmem_recalc_inode(inode);
1663                 spin_unlock_irq(&info->lock);
1664 
1665                 if (sgp == SGP_WRITE)
1666                         mark_page_accessed(page);
1667 
1668                 delete_from_swap_cache(page);
1669                 set_page_dirty(page);
1670                 swap_free(swap);
1671 
1672         } else {
1673                 /* shmem_symlink() */
1674                 if (mapping->a_ops != &shmem_aops)
1675                         goto alloc_nohuge;
1676                 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1677                         goto alloc_nohuge;
1678                 if (shmem_huge == SHMEM_HUGE_FORCE)
1679                         goto alloc_huge;
1680                 switch (sbinfo->huge) {
1681                         loff_t i_size;
1682                         pgoff_t off;
1683                 case SHMEM_HUGE_NEVER:
1684                         goto alloc_nohuge;
1685                 case SHMEM_HUGE_WITHIN_SIZE:
1686                         off = round_up(index, HPAGE_PMD_NR);
1687                         i_size = round_up(i_size_read(inode), PAGE_SIZE);
1688                         if (i_size >= HPAGE_PMD_SIZE &&
1689                                         i_size >> PAGE_SHIFT >= off)
1690                                 goto alloc_huge;
1691                         /* fallthrough */
1692                 case SHMEM_HUGE_ADVISE:
1693                         if (sgp_huge == SGP_HUGE)
1694                                 goto alloc_huge;
1695                         /* TODO: implement fadvise() hints */
1696                         goto alloc_nohuge;
1697                 }
1698 
1699 alloc_huge:
1700                 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1701                                 index, true);
1702                 if (IS_ERR(page)) {
1703 alloc_nohuge:           page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1704                                         index, false);
1705                 }
1706                 if (IS_ERR(page)) {
1707                         int retry = 5;
1708                         error = PTR_ERR(page);
1709                         page = NULL;
1710                         if (error != -ENOSPC)
1711                                 goto failed;
1712                         /*
1713                          * Try to reclaim some spece by splitting a huge page
1714                          * beyond i_size on the filesystem.
1715                          */
1716                         while (retry--) {
1717                                 int ret;
1718                                 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1719                                 if (ret == SHRINK_STOP)
1720                                         break;
1721                                 if (ret)
1722                                         goto alloc_nohuge;
1723                         }
1724                         goto failed;
1725                 }
1726 
1727                 if (PageTransHuge(page))
1728                         hindex = round_down(index, HPAGE_PMD_NR);
1729                 else
1730                         hindex = index;
1731 
1732                 if (sgp == SGP_WRITE)
1733                         __SetPageReferenced(page);
1734 
1735                 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
1736                                 PageTransHuge(page));
1737                 if (error)
1738                         goto unacct;
1739                 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1740                                 compound_order(page));
1741                 if (!error) {
1742                         error = shmem_add_to_page_cache(page, mapping, hindex,
1743                                                         NULL);
1744                         radix_tree_preload_end();
1745                 }
1746                 if (error) {
1747                         mem_cgroup_cancel_charge(page, memcg,
1748                                         PageTransHuge(page));
1749                         goto unacct;
1750                 }
1751                 mem_cgroup_commit_charge(page, memcg, false,
1752                                 PageTransHuge(page));
1753                 lru_cache_add_anon(page);
1754 
1755                 spin_lock_irq(&info->lock);
1756                 info->alloced += 1 << compound_order(page);
1757                 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1758                 shmem_recalc_inode(inode);
1759                 spin_unlock_irq(&info->lock);
1760                 alloced = true;
1761 
1762                 if (PageTransHuge(page) &&
1763                                 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1764                                 hindex + HPAGE_PMD_NR - 1) {
1765                         /*
1766                          * Part of the huge page is beyond i_size: subject
1767                          * to shrink under memory pressure.
1768                          */
1769                         spin_lock(&sbinfo->shrinklist_lock);
1770                         if (list_empty(&info->shrinklist)) {
1771                                 list_add_tail(&info->shrinklist,
1772                                                 &sbinfo->shrinklist);
1773                                 sbinfo->shrinklist_len++;
1774                         }
1775                         spin_unlock(&sbinfo->shrinklist_lock);
1776                 }
1777 
1778                 /*
1779                  * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1780                  */
1781                 if (sgp == SGP_FALLOC)
1782                         sgp = SGP_WRITE;
1783 clear:
1784                 /*
1785                  * Let SGP_WRITE caller clear ends if write does not fill page;
1786                  * but SGP_FALLOC on a page fallocated earlier must initialize
1787                  * it now, lest undo on failure cancel our earlier guarantee.
1788                  */
1789                 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1790                         struct page *head = compound_head(page);
1791                         int i;
1792 
1793                         for (i = 0; i < (1 << compound_order(head)); i++) {
1794                                 clear_highpage(head + i);
1795                                 flush_dcache_page(head + i);
1796                         }
1797                         SetPageUptodate(head);
1798                 }
1799         }
1800 
1801         /* Perhaps the file has been truncated since we checked */
1802         if (sgp <= SGP_CACHE &&
1803             ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1804                 if (alloced) {
1805                         ClearPageDirty(page);
1806                         delete_from_page_cache(page);
1807                         spin_lock_irq(&info->lock);
1808                         shmem_recalc_inode(inode);
1809                         spin_unlock_irq(&info->lock);
1810                 }
1811                 error = -EINVAL;
1812                 goto unlock;
1813         }
1814         *pagep = page + index - hindex;
1815         return 0;
1816 
1817         /*
1818          * Error recovery.
1819          */
1820 unacct:
1821         if (sbinfo->max_blocks)
1822                 percpu_counter_sub(&sbinfo->used_blocks,
1823                                 1 << compound_order(page));
1824         shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1825 
1826         if (PageTransHuge(page)) {
1827                 unlock_page(page);
1828                 put_page(page);
1829                 goto alloc_nohuge;
1830         }
1831 failed:
1832         if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1833                 error = -EEXIST;
1834 unlock:
1835         if (page) {
1836                 unlock_page(page);
1837                 put_page(page);
1838         }
1839         if (error == -ENOSPC && !once++) {
1840                 info = SHMEM_I(inode);
1841                 spin_lock_irq(&info->lock);
1842                 shmem_recalc_inode(inode);
1843                 spin_unlock_irq(&info->lock);
1844                 goto repeat;
1845         }
1846         if (error == -EEXIST)   /* from above or from radix_tree_insert */
1847                 goto repeat;
1848         return error;
1849 }
1850 
1851 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1852 {
1853         struct inode *inode = file_inode(vma->vm_file);
1854         gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1855         enum sgp_type sgp;
1856         int error;
1857         int ret = VM_FAULT_LOCKED;
1858 
1859         /*
1860          * Trinity finds that probing a hole which tmpfs is punching can
1861          * prevent the hole-punch from ever completing: which in turn
1862          * locks writers out with its hold on i_mutex.  So refrain from
1863          * faulting pages into the hole while it's being punched.  Although
1864          * shmem_undo_range() does remove the additions, it may be unable to
1865          * keep up, as each new page needs its own unmap_mapping_range() call,
1866          * and the i_mmap tree grows ever slower to scan if new vmas are added.
1867          *
1868          * It does not matter if we sometimes reach this check just before the
1869          * hole-punch begins, so that one fault then races with the punch:
1870          * we just need to make racing faults a rare case.
1871          *
1872          * The implementation below would be much simpler if we just used a
1873          * standard mutex or completion: but we cannot take i_mutex in fault,
1874          * and bloating every shmem inode for this unlikely case would be sad.
1875          */
1876         if (unlikely(inode->i_private)) {
1877                 struct shmem_falloc *shmem_falloc;
1878 
1879                 spin_lock(&inode->i_lock);
1880                 shmem_falloc = inode->i_private;
1881                 if (shmem_falloc &&
1882                     shmem_falloc->waitq &&
1883                     vmf->pgoff >= shmem_falloc->start &&
1884                     vmf->pgoff < shmem_falloc->next) {
1885                         wait_queue_head_t *shmem_falloc_waitq;
1886                         DEFINE_WAIT(shmem_fault_wait);
1887 
1888                         ret = VM_FAULT_NOPAGE;
1889                         if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1890                            !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1891                                 /* It's polite to up mmap_sem if we can */
1892                                 up_read(&vma->vm_mm->mmap_sem);
1893                                 ret = VM_FAULT_RETRY;
1894                         }
1895 
1896                         shmem_falloc_waitq = shmem_falloc->waitq;
1897                         prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1898                                         TASK_UNINTERRUPTIBLE);
1899                         spin_unlock(&inode->i_lock);
1900                         schedule();
1901 
1902                         /*
1903                          * shmem_falloc_waitq points into the shmem_fallocate()
1904                          * stack of the hole-punching task: shmem_falloc_waitq
1905                          * is usually invalid by the time we reach here, but
1906                          * finish_wait() does not dereference it in that case;
1907                          * though i_lock needed lest racing with wake_up_all().
1908                          */
1909                         spin_lock(&inode->i_lock);
1910                         finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1911                         spin_unlock(&inode->i_lock);
1912                         return ret;
1913                 }
1914                 spin_unlock(&inode->i_lock);
1915         }
1916 
1917         sgp = SGP_CACHE;
1918         if (vma->vm_flags & VM_HUGEPAGE)
1919                 sgp = SGP_HUGE;
1920         else if (vma->vm_flags & VM_NOHUGEPAGE)
1921                 sgp = SGP_NOHUGE;
1922 
1923         error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
1924                                   gfp, vma->vm_mm, &ret);
1925         if (error)
1926                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1927         return ret;
1928 }
1929 
1930 unsigned long shmem_get_unmapped_area(struct file *file,
1931                                       unsigned long uaddr, unsigned long len,
1932                                       unsigned long pgoff, unsigned long flags)
1933 {
1934         unsigned long (*get_area)(struct file *,
1935                 unsigned long, unsigned long, unsigned long, unsigned long);
1936         unsigned long addr;
1937         unsigned long offset;
1938         unsigned long inflated_len;
1939         unsigned long inflated_addr;
1940         unsigned long inflated_offset;
1941 
1942         if (len > TASK_SIZE)
1943                 return -ENOMEM;
1944 
1945         get_area = current->mm->get_unmapped_area;
1946         addr = get_area(file, uaddr, len, pgoff, flags);
1947 
1948         if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1949                 return addr;
1950         if (IS_ERR_VALUE(addr))
1951                 return addr;
1952         if (addr & ~PAGE_MASK)
1953                 return addr;
1954         if (addr > TASK_SIZE - len)
1955                 return addr;
1956 
1957         if (shmem_huge == SHMEM_HUGE_DENY)
1958                 return addr;
1959         if (len < HPAGE_PMD_SIZE)
1960                 return addr;
1961         if (flags & MAP_FIXED)
1962                 return addr;
1963         /*
1964          * Our priority is to support MAP_SHARED mapped hugely;
1965          * and support MAP_PRIVATE mapped hugely too, until it is COWed.
1966          * But if caller specified an address hint, respect that as before.
1967          */
1968         if (uaddr)
1969                 return addr;
1970 
1971         if (shmem_huge != SHMEM_HUGE_FORCE) {
1972                 struct super_block *sb;
1973 
1974                 if (file) {
1975                         VM_BUG_ON(file->f_op != &shmem_file_operations);
1976                         sb = file_inode(file)->i_sb;
1977                 } else {
1978                         /*
1979                          * Called directly from mm/mmap.c, or drivers/char/mem.c
1980                          * for "/dev/zero", to create a shared anonymous object.
1981                          */
1982                         if (IS_ERR(shm_mnt))
1983                                 return addr;
1984                         sb = shm_mnt->mnt_sb;
1985                 }
1986                 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
1987                         return addr;
1988         }
1989 
1990         offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
1991         if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
1992                 return addr;
1993         if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
1994                 return addr;
1995 
1996         inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
1997         if (inflated_len > TASK_SIZE)
1998                 return addr;
1999         if (inflated_len < len)
2000                 return addr;
2001 
2002         inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2003         if (IS_ERR_VALUE(inflated_addr))
2004                 return addr;
2005         if (inflated_addr & ~PAGE_MASK)
2006                 return addr;
2007 
2008         inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2009         inflated_addr += offset - inflated_offset;
2010         if (inflated_offset > offset)
2011                 inflated_addr += HPAGE_PMD_SIZE;
2012 
2013         if (inflated_addr > TASK_SIZE - len)
2014                 return addr;
2015         return inflated_addr;
2016 }
2017 
2018 #ifdef CONFIG_NUMA
2019 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2020 {
2021         struct inode *inode = file_inode(vma->vm_file);
2022         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2023 }
2024 
2025 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2026                                           unsigned long addr)
2027 {
2028         struct inode *inode = file_inode(vma->vm_file);
2029         pgoff_t index;
2030 
2031         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2032         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2033 }
2034 #endif
2035 
2036 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2037 {
2038         struct inode *inode = file_inode(file);
2039         struct shmem_inode_info *info = SHMEM_I(inode);
2040         int retval = -ENOMEM;
2041 
2042         spin_lock_irq(&info->lock);
2043         if (lock && !(info->flags & VM_LOCKED)) {
2044                 if (!user_shm_lock(inode->i_size, user))
2045                         goto out_nomem;
2046                 info->flags |= VM_LOCKED;
2047                 mapping_set_unevictable(file->f_mapping);
2048         }
2049         if (!lock && (info->flags & VM_LOCKED) && user) {
2050                 user_shm_unlock(inode->i_size, user);
2051                 info->flags &= ~VM_LOCKED;
2052                 mapping_clear_unevictable(file->f_mapping);
2053         }
2054         retval = 0;
2055 
2056 out_nomem:
2057         spin_unlock_irq(&info->lock);
2058         return retval;
2059 }
2060 
2061 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2062 {
2063         file_accessed(file);
2064         vma->vm_ops = &shmem_vm_ops;
2065         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2066                         ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2067                         (vma->vm_end & HPAGE_PMD_MASK)) {
2068                 khugepaged_enter(vma, vma->vm_flags);
2069         }
2070         return 0;
2071 }
2072 
2073 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2074                                      umode_t mode, dev_t dev, unsigned long flags)
2075 {
2076         struct inode *inode;
2077         struct shmem_inode_info *info;
2078         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2079 
2080         if (shmem_reserve_inode(sb))
2081                 return NULL;
2082 
2083         inode = new_inode(sb);
2084         if (inode) {
2085                 inode->i_ino = get_next_ino();
2086                 inode_init_owner(inode, dir, mode);
2087                 inode->i_blocks = 0;
2088                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2089                 inode->i_generation = get_seconds();
2090                 info = SHMEM_I(inode);
2091                 memset(info, 0, (char *)inode - (char *)info);
2092                 spin_lock_init(&info->lock);
2093                 info->seals = F_SEAL_SEAL;
2094                 info->flags = flags & VM_NORESERVE;
2095                 INIT_LIST_HEAD(&info->shrinklist);
2096                 INIT_LIST_HEAD(&info->swaplist);
2097                 simple_xattrs_init(&info->xattrs);
2098                 cache_no_acl(inode);
2099 
2100                 switch (mode & S_IFMT) {
2101                 default:
2102                         inode->i_op = &shmem_special_inode_operations;
2103                         init_special_inode(inode, mode, dev);
2104                         break;
2105                 case S_IFREG:
2106                         inode->i_mapping->a_ops = &shmem_aops;
2107                         inode->i_op = &shmem_inode_operations;
2108                         inode->i_fop = &shmem_file_operations;
2109                         mpol_shared_policy_init(&info->policy,
2110                                                  shmem_get_sbmpol(sbinfo));
2111                         break;
2112                 case S_IFDIR:
2113                         inc_nlink(inode);
2114                         /* Some things misbehave if size == 0 on a directory */
2115                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
2116                         inode->i_op = &shmem_dir_inode_operations;
2117                         inode->i_fop = &simple_dir_operations;
2118                         break;
2119                 case S_IFLNK:
2120                         /*
2121                          * Must not load anything in the rbtree,
2122                          * mpol_free_shared_policy will not be called.
2123                          */
2124                         mpol_shared_policy_init(&info->policy, NULL);
2125                         break;
2126                 }
2127         } else
2128                 shmem_free_inode(sb);
2129         return inode;
2130 }
2131 
2132 bool shmem_mapping(struct address_space *mapping)
2133 {
2134         if (!mapping->host)
2135                 return false;
2136 
2137         return mapping->host->i_sb->s_op == &shmem_ops;
2138 }
2139 
2140 #ifdef CONFIG_TMPFS
2141 static const struct inode_operations shmem_symlink_inode_operations;
2142 static const struct inode_operations shmem_short_symlink_operations;
2143 
2144 #ifdef CONFIG_TMPFS_XATTR
2145 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2146 #else
2147 #define shmem_initxattrs NULL
2148 #endif
2149 
2150 static int
2151 shmem_write_begin(struct file *file, struct address_space *mapping,
2152                         loff_t pos, unsigned len, unsigned flags,
2153                         struct page **pagep, void **fsdata)
2154 {
2155         struct inode *inode = mapping->host;
2156         struct shmem_inode_info *info = SHMEM_I(inode);
2157         pgoff_t index = pos >> PAGE_SHIFT;
2158 
2159         /* i_mutex is held by caller */
2160         if (unlikely(info->seals)) {
2161                 if (info->seals & F_SEAL_WRITE)
2162                         return -EPERM;
2163                 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2164                         return -EPERM;
2165         }
2166 
2167         return shmem_getpage(inode, index, pagep, SGP_WRITE);
2168 }
2169 
2170 static int
2171 shmem_write_end(struct file *file, struct address_space *mapping,
2172                         loff_t pos, unsigned len, unsigned copied,
2173                         struct page *page, void *fsdata)
2174 {
2175         struct inode *inode = mapping->host;
2176 
2177         if (pos + copied > inode->i_size)
2178                 i_size_write(inode, pos + copied);
2179 
2180         if (!PageUptodate(page)) {
2181                 struct page *head = compound_head(page);
2182                 if (PageTransCompound(page)) {
2183                         int i;
2184 
2185                         for (i = 0; i < HPAGE_PMD_NR; i++) {
2186                                 if (head + i == page)
2187                                         continue;
2188                                 clear_highpage(head + i);
2189                                 flush_dcache_page(head + i);
2190                         }
2191                 }
2192                 if (copied < PAGE_SIZE) {
2193                         unsigned from = pos & (PAGE_SIZE - 1);
2194                         zero_user_segments(page, 0, from,
2195                                         from + copied, PAGE_SIZE);
2196                 }
2197                 SetPageUptodate(head);
2198         }
2199         set_page_dirty(page);
2200         unlock_page(page);
2201         put_page(page);
2202 
2203         return copied;
2204 }
2205 
2206 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2207 {
2208         struct file *file = iocb->ki_filp;
2209         struct inode *inode = file_inode(file);
2210         struct address_space *mapping = inode->i_mapping;
2211         pgoff_t index;
2212         unsigned long offset;
2213         enum sgp_type sgp = SGP_READ;
2214         int error = 0;
2215         ssize_t retval = 0;
2216         loff_t *ppos = &iocb->ki_pos;
2217 
2218         /*
2219          * Might this read be for a stacking filesystem?  Then when reading
2220          * holes of a sparse file, we actually need to allocate those pages,
2221          * and even mark them dirty, so it cannot exceed the max_blocks limit.
2222          */
2223         if (!iter_is_iovec(to))
2224                 sgp = SGP_CACHE;
2225 
2226         index = *ppos >> PAGE_SHIFT;
2227         offset = *ppos & ~PAGE_MASK;
2228 
2229         for (;;) {
2230                 struct page *page = NULL;
2231                 pgoff_t end_index;
2232                 unsigned long nr, ret;
2233                 loff_t i_size = i_size_read(inode);
2234 
2235                 end_index = i_size >> PAGE_SHIFT;
2236                 if (index > end_index)
2237                         break;
2238                 if (index == end_index) {
2239                         nr = i_size & ~PAGE_MASK;
2240                         if (nr <= offset)
2241                                 break;
2242                 }
2243 
2244                 error = shmem_getpage(inode, index, &page, sgp);
2245                 if (error) {
2246                         if (error == -EINVAL)
2247                                 error = 0;
2248                         break;
2249                 }
2250                 if (page) {
2251                         if (sgp == SGP_CACHE)
2252                                 set_page_dirty(page);
2253                         unlock_page(page);
2254                 }
2255 
2256                 /*
2257                  * We must evaluate after, since reads (unlike writes)
2258                  * are called without i_mutex protection against truncate
2259                  */
2260                 nr = PAGE_SIZE;
2261                 i_size = i_size_read(inode);
2262                 end_index = i_size >> PAGE_SHIFT;
2263                 if (index == end_index) {
2264                         nr = i_size & ~PAGE_MASK;
2265                         if (nr <= offset) {
2266                                 if (page)
2267                                         put_page(page);
2268                                 break;
2269                         }
2270                 }
2271                 nr -= offset;
2272 
2273                 if (page) {
2274                         /*
2275                          * If users can be writing to this page using arbitrary
2276                          * virtual addresses, take care about potential aliasing
2277                          * before reading the page on the kernel side.
2278                          */
2279                         if (mapping_writably_mapped(mapping))
2280                                 flush_dcache_page(page);
2281                         /*
2282                          * Mark the page accessed if we read the beginning.
2283                          */
2284                         if (!offset)
2285                                 mark_page_accessed(page);
2286                 } else {
2287                         page = ZERO_PAGE(0);
2288                         get_page(page);
2289                 }
2290 
2291                 /*
2292                  * Ok, we have the page, and it's up-to-date, so
2293                  * now we can copy it to user space...
2294                  */
2295                 ret = copy_page_to_iter(page, offset, nr, to);
2296                 retval += ret;
2297                 offset += ret;
2298                 index += offset >> PAGE_SHIFT;
2299                 offset &= ~PAGE_MASK;
2300 
2301                 put_page(page);
2302                 if (!iov_iter_count(to))
2303                         break;
2304                 if (ret < nr) {
2305                         error = -EFAULT;
2306                         break;
2307                 }
2308                 cond_resched();
2309         }
2310 
2311         *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2312         file_accessed(file);
2313         return retval ? retval : error;
2314 }
2315 
2316 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
2317                                 struct pipe_inode_info *pipe, size_t len,
2318                                 unsigned int flags)
2319 {
2320         struct address_space *mapping = in->f_mapping;
2321         struct inode *inode = mapping->host;
2322         unsigned int loff, nr_pages, req_pages;
2323         struct page *pages[PIPE_DEF_BUFFERS];
2324         struct partial_page partial[PIPE_DEF_BUFFERS];
2325         struct page *page;
2326         pgoff_t index, end_index;
2327         loff_t isize, left;
2328         int error, page_nr;
2329         struct splice_pipe_desc spd = {
2330                 .pages = pages,
2331                 .partial = partial,
2332                 .nr_pages_max = PIPE_DEF_BUFFERS,
2333                 .flags = flags,
2334                 .ops = &page_cache_pipe_buf_ops,
2335                 .spd_release = spd_release_page,
2336         };
2337 
2338         isize = i_size_read(inode);
2339         if (unlikely(*ppos >= isize))
2340                 return 0;
2341 
2342         left = isize - *ppos;
2343         if (unlikely(left < len))
2344                 len = left;
2345 
2346         if (splice_grow_spd(pipe, &spd))
2347                 return -ENOMEM;
2348 
2349         index = *ppos >> PAGE_SHIFT;
2350         loff = *ppos & ~PAGE_MASK;
2351         req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT;
2352         nr_pages = min(req_pages, spd.nr_pages_max);
2353 
2354         spd.nr_pages = find_get_pages_contig(mapping, index,
2355                                                 nr_pages, spd.pages);
2356         index += spd.nr_pages;
2357         error = 0;
2358 
2359         while (spd.nr_pages < nr_pages) {
2360                 error = shmem_getpage(inode, index, &page, SGP_CACHE);
2361                 if (error)
2362                         break;
2363                 unlock_page(page);
2364                 spd.pages[spd.nr_pages++] = page;
2365                 index++;
2366         }
2367 
2368         index = *ppos >> PAGE_SHIFT;
2369         nr_pages = spd.nr_pages;
2370         spd.nr_pages = 0;
2371 
2372         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
2373                 unsigned int this_len;
2374 
2375                 if (!len)
2376                         break;
2377 
2378                 this_len = min_t(unsigned long, len, PAGE_SIZE - loff);
2379                 page = spd.pages[page_nr];
2380 
2381                 if (!PageUptodate(page) || page->mapping != mapping) {
2382                         error = shmem_getpage(inode, index, &page, SGP_CACHE);
2383                         if (error)
2384                                 break;
2385                         unlock_page(page);
2386                         put_page(spd.pages[page_nr]);
2387                         spd.pages[page_nr] = page;
2388                 }
2389 
2390                 isize = i_size_read(inode);
2391                 end_index = (isize - 1) >> PAGE_SHIFT;
2392                 if (unlikely(!isize || index > end_index))
2393                         break;
2394 
2395                 if (end_index == index) {
2396                         unsigned int plen;
2397 
2398                         plen = ((isize - 1) & ~PAGE_MASK) + 1;
2399                         if (plen <= loff)
2400                                 break;
2401 
2402                         this_len = min(this_len, plen - loff);
2403                         len = this_len;
2404                 }
2405 
2406                 spd.partial[page_nr].offset = loff;
2407                 spd.partial[page_nr].len = this_len;
2408                 len -= this_len;
2409                 loff = 0;
2410                 spd.nr_pages++;
2411                 index++;
2412         }
2413 
2414         while (page_nr < nr_pages)
2415                 put_page(spd.pages[page_nr++]);
2416 
2417         if (spd.nr_pages)
2418                 error = splice_to_pipe(pipe, &spd);
2419 
2420         splice_shrink_spd(&spd);
2421 
2422         if (error > 0) {
2423                 *ppos += error;
2424                 file_accessed(in);
2425         }
2426         return error;
2427 }
2428 
2429 /*
2430  * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2431  */
2432 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2433                                     pgoff_t index, pgoff_t end, int whence)
2434 {
2435         struct page *page;
2436         struct pagevec pvec;
2437         pgoff_t indices[PAGEVEC_SIZE];
2438         bool done = false;
2439         int i;
2440 
2441         pagevec_init(&pvec, 0);
2442         pvec.nr = 1;            /* start small: we may be there already */
2443         while (!done) {
2444                 pvec.nr = find_get_entries(mapping, index,
2445                                         pvec.nr, pvec.pages, indices);
2446                 if (!pvec.nr) {
2447                         if (whence == SEEK_DATA)
2448                                 index = end;
2449                         break;
2450                 }
2451                 for (i = 0; i < pvec.nr; i++, index++) {
2452                         if (index < indices[i]) {
2453                                 if (whence == SEEK_HOLE) {
2454                                         done = true;
2455                                         break;
2456                                 }
2457                                 index = indices[i];
2458                         }
2459                         page = pvec.pages[i];
2460                         if (page && !radix_tree_exceptional_entry(page)) {
2461                                 if (!PageUptodate(page))
2462                                         page = NULL;
2463                         }
2464                         if (index >= end ||
2465                             (page && whence == SEEK_DATA) ||
2466                             (!page && whence == SEEK_HOLE)) {
2467                                 done = true;
2468                                 break;
2469                         }
2470                 }
2471                 pagevec_remove_exceptionals(&pvec);
2472                 pagevec_release(&pvec);
2473                 pvec.nr = PAGEVEC_SIZE;
2474                 cond_resched();
2475         }
2476         return index;
2477 }
2478 
2479 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2480 {
2481         struct address_space *mapping = file->f_mapping;
2482         struct inode *inode = mapping->host;
2483         pgoff_t start, end;
2484         loff_t new_offset;
2485 
2486         if (whence != SEEK_DATA && whence != SEEK_HOLE)
2487                 return generic_file_llseek_size(file, offset, whence,
2488                                         MAX_LFS_FILESIZE, i_size_read(inode));
2489         inode_lock(inode);
2490         /* We're holding i_mutex so we can access i_size directly */
2491 
2492         if (offset < 0)
2493                 offset = -EINVAL;
2494         else if (offset >= inode->i_size)
2495                 offset = -ENXIO;
2496         else {
2497                 start = offset >> PAGE_SHIFT;
2498                 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2499                 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2500                 new_offset <<= PAGE_SHIFT;
2501                 if (new_offset > offset) {
2502                         if (new_offset < inode->i_size)
2503                                 offset = new_offset;
2504                         else if (whence == SEEK_DATA)
2505                                 offset = -ENXIO;
2506                         else
2507                                 offset = inode->i_size;
2508                 }
2509         }
2510 
2511         if (offset >= 0)
2512                 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2513         inode_unlock(inode);
2514         return offset;
2515 }
2516 
2517 /*
2518  * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2519  * so reuse a tag which we firmly believe is never set or cleared on shmem.
2520  */
2521 #define SHMEM_TAG_PINNED        PAGECACHE_TAG_TOWRITE
2522 #define LAST_SCAN               4       /* about 150ms max */
2523 
2524 static void shmem_tag_pins(struct address_space *mapping)
2525 {
2526         struct radix_tree_iter iter;
2527         void **slot;
2528         pgoff_t start;
2529         struct page *page;
2530 
2531         lru_add_drain();
2532         start = 0;
2533         rcu_read_lock();
2534 
2535         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2536                 page = radix_tree_deref_slot(slot);
2537                 if (!page || radix_tree_exception(page)) {
2538                         if (radix_tree_deref_retry(page)) {
2539                                 slot = radix_tree_iter_retry(&iter);
2540                                 continue;
2541                         }
2542                 } else if (page_count(page) - page_mapcount(page) > 1) {
2543                         spin_lock_irq(&mapping->tree_lock);
2544                         radix_tree_tag_set(&mapping->page_tree, iter.index,
2545                                            SHMEM_TAG_PINNED);
2546                         spin_unlock_irq(&mapping->tree_lock);
2547                 }
2548 
2549                 if (need_resched()) {
2550                         cond_resched_rcu();
2551                         slot = radix_tree_iter_next(&iter);
2552                 }
2553         }
2554         rcu_read_unlock();
2555 }
2556 
2557 /*
2558  * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2559  * via get_user_pages(), drivers might have some pending I/O without any active
2560  * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2561  * and see whether it has an elevated ref-count. If so, we tag them and wait for
2562  * them to be dropped.
2563  * The caller must guarantee that no new user will acquire writable references
2564  * to those pages to avoid races.
2565  */
2566 static int shmem_wait_for_pins(struct address_space *mapping)
2567 {
2568         struct radix_tree_iter iter;
2569         void **slot;
2570         pgoff_t start;
2571         struct page *page;
2572         int error, scan;
2573 
2574         shmem_tag_pins(mapping);
2575 
2576         error = 0;
2577         for (scan = 0; scan <= LAST_SCAN; scan++) {
2578                 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2579                         break;
2580 
2581                 if (!scan)
2582                         lru_add_drain_all();
2583                 else if (schedule_timeout_killable((HZ << scan) / 200))
2584                         scan = LAST_SCAN;
2585 
2586                 start = 0;
2587                 rcu_read_lock();
2588                 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2589                                            start, SHMEM_TAG_PINNED) {
2590 
2591                         page = radix_tree_deref_slot(slot);
2592                         if (radix_tree_exception(page)) {
2593                                 if (radix_tree_deref_retry(page)) {
2594                                         slot = radix_tree_iter_retry(&iter);
2595                                         continue;
2596                                 }
2597 
2598                                 page = NULL;
2599                         }
2600 
2601                         if (page &&
2602                             page_count(page) - page_mapcount(page) != 1) {
2603                                 if (scan < LAST_SCAN)
2604                                         goto continue_resched;
2605 
2606                                 /*
2607                                  * On the last scan, we clean up all those tags
2608                                  * we inserted; but make a note that we still
2609                                  * found pages pinned.
2610                                  */
2611                                 error = -EBUSY;
2612                         }
2613 
2614                         spin_lock_irq(&mapping->tree_lock);
2615                         radix_tree_tag_clear(&mapping->page_tree,
2616                                              iter.index, SHMEM_TAG_PINNED);
2617                         spin_unlock_irq(&mapping->tree_lock);
2618 continue_resched:
2619                         if (need_resched()) {
2620                                 cond_resched_rcu();
2621                                 slot = radix_tree_iter_next(&iter);
2622                         }
2623                 }
2624                 rcu_read_unlock();
2625         }
2626 
2627         return error;
2628 }
2629 
2630 #define F_ALL_SEALS (F_SEAL_SEAL | \
2631                      F_SEAL_SHRINK | \
2632                      F_SEAL_GROW | \
2633                      F_SEAL_WRITE)
2634 
2635 int shmem_add_seals(struct file *file, unsigned int seals)
2636 {
2637         struct inode *inode = file_inode(file);
2638         struct shmem_inode_info *info = SHMEM_I(inode);
2639         int error;
2640 
2641         /*
2642          * SEALING
2643          * Sealing allows multiple parties to share a shmem-file but restrict
2644          * access to a specific subset of file operations. Seals can only be
2645          * added, but never removed. This way, mutually untrusted parties can
2646          * share common memory regions with a well-defined policy. A malicious
2647          * peer can thus never perform unwanted operations on a shared object.
2648          *
2649          * Seals are only supported on special shmem-files and always affect
2650          * the whole underlying inode. Once a seal is set, it may prevent some
2651          * kinds of access to the file. Currently, the following seals are
2652          * defined:
2653          *   SEAL_SEAL: Prevent further seals from being set on this file
2654          *   SEAL_SHRINK: Prevent the file from shrinking
2655          *   SEAL_GROW: Prevent the file from growing
2656          *   SEAL_WRITE: Prevent write access to the file
2657          *
2658          * As we don't require any trust relationship between two parties, we
2659          * must prevent seals from being removed. Therefore, sealing a file
2660          * only adds a given set of seals to the file, it never touches
2661          * existing seals. Furthermore, the "setting seals"-operation can be
2662          * sealed itself, which basically prevents any further seal from being
2663          * added.
2664          *
2665          * Semantics of sealing are only defined on volatile files. Only
2666          * anonymous shmem files support sealing. More importantly, seals are
2667          * never written to disk. Therefore, there's no plan to support it on
2668          * other file types.
2669          */
2670 
2671         if (file->f_op != &shmem_file_operations)
2672                 return -EINVAL;
2673         if (!(file->f_mode & FMODE_WRITE))
2674                 return -EPERM;
2675         if (seals & ~(unsigned int)F_ALL_SEALS)
2676                 return -EINVAL;
2677 
2678         inode_lock(inode);
2679 
2680         if (info->seals & F_SEAL_SEAL) {
2681                 error = -EPERM;
2682                 goto unlock;
2683         }
2684 
2685         if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2686                 error = mapping_deny_writable(file->f_mapping);
2687                 if (error)
2688                         goto unlock;
2689 
2690                 error = shmem_wait_for_pins(file->f_mapping);
2691                 if (error) {
2692                         mapping_allow_writable(file->f_mapping);
2693                         goto unlock;
2694                 }
2695         }
2696 
2697         info->seals |= seals;
2698         error = 0;
2699 
2700 unlock:
2701         inode_unlock(inode);
2702         return error;
2703 }
2704 EXPORT_SYMBOL_GPL(shmem_add_seals);
2705 
2706 int shmem_get_seals(struct file *file)
2707 {
2708         if (file->f_op != &shmem_file_operations)
2709                 return -EINVAL;
2710 
2711         return SHMEM_I(file_inode(file))->seals;
2712 }
2713 EXPORT_SYMBOL_GPL(shmem_get_seals);
2714 
2715 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2716 {
2717         long error;
2718 
2719         switch (cmd) {
2720         case F_ADD_SEALS:
2721                 /* disallow upper 32bit */
2722                 if (arg > UINT_MAX)
2723                         return -EINVAL;
2724 
2725                 error = shmem_add_seals(file, arg);
2726                 break;
2727         case F_GET_SEALS:
2728                 error = shmem_get_seals(file);
2729                 break;
2730         default:
2731                 error = -EINVAL;
2732                 break;
2733         }
2734 
2735         return error;
2736 }
2737 
2738 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2739                                                          loff_t len)
2740 {
2741         struct inode *inode = file_inode(file);
2742         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2743         struct shmem_inode_info *info = SHMEM_I(inode);
2744         struct shmem_falloc shmem_falloc;
2745         pgoff_t start, index, end;
2746         int error;
2747 
2748         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2749                 return -EOPNOTSUPP;
2750 
2751         inode_lock(inode);
2752 
2753         if (mode & FALLOC_FL_PUNCH_HOLE) {
2754                 struct address_space *mapping = file->f_mapping;
2755                 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2756                 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2757                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2758 
2759                 /* protected by i_mutex */
2760                 if (info->seals & F_SEAL_WRITE) {
2761                         error = -EPERM;
2762                         goto out;
2763                 }
2764 
2765                 shmem_falloc.waitq = &shmem_falloc_waitq;
2766                 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2767                 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2768                 spin_lock(&inode->i_lock);
2769                 inode->i_private = &shmem_falloc;
2770                 spin_unlock(&inode->i_lock);
2771 
2772                 if ((u64)unmap_end > (u64)unmap_start)
2773                         unmap_mapping_range(mapping, unmap_start,
2774                                             1 + unmap_end - unmap_start, 0);
2775                 shmem_truncate_range(inode, offset, offset + len - 1);
2776                 /* No need to unmap again: hole-punching leaves COWed pages */
2777 
2778                 spin_lock(&inode->i_lock);
2779                 inode->i_private = NULL;
2780                 wake_up_all(&shmem_falloc_waitq);
2781                 spin_unlock(&inode->i_lock);
2782                 error = 0;
2783                 goto out;
2784         }
2785 
2786         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2787         error = inode_newsize_ok(inode, offset + len);
2788         if (error)
2789                 goto out;
2790 
2791         if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2792                 error = -EPERM;
2793                 goto out;
2794         }
2795 
2796         start = offset >> PAGE_SHIFT;
2797         end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2798         /* Try to avoid a swapstorm if len is impossible to satisfy */
2799         if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2800                 error = -ENOSPC;
2801                 goto out;
2802         }
2803 
2804         shmem_falloc.waitq = NULL;
2805         shmem_falloc.start = start;
2806         shmem_falloc.next  = start;
2807         shmem_falloc.nr_falloced = 0;
2808         shmem_falloc.nr_unswapped = 0;
2809         spin_lock(&inode->i_lock);
2810         inode->i_private = &shmem_falloc;
2811         spin_unlock(&inode->i_lock);
2812 
2813         for (index = start; index < end; index++) {
2814                 struct page *page;
2815 
2816                 /*
2817                  * Good, the fallocate(2) manpage permits EINTR: we may have
2818                  * been interrupted because we are using up too much memory.
2819                  */
2820                 if (signal_pending(current))
2821                         error = -EINTR;
2822                 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2823                         error = -ENOMEM;
2824                 else
2825                         error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2826                 if (error) {
2827                         /* Remove the !PageUptodate pages we added */
2828                         if (index > start) {
2829                                 shmem_undo_range(inode,
2830                                     (loff_t)start << PAGE_SHIFT,
2831                                     ((loff_t)index << PAGE_SHIFT) - 1, true);
2832                         }
2833                         goto undone;
2834                 }
2835 
2836                 /*
2837                  * Inform shmem_writepage() how far we have reached.
2838                  * No need for lock or barrier: we have the page lock.
2839                  */
2840                 shmem_falloc.next++;
2841                 if (!PageUptodate(page))
2842                         shmem_falloc.nr_falloced++;
2843 
2844                 /*
2845                  * If !PageUptodate, leave it that way so that freeable pages
2846                  * can be recognized if we need to rollback on error later.
2847                  * But set_page_dirty so that memory pressure will swap rather
2848                  * than free the pages we are allocating (and SGP_CACHE pages
2849                  * might still be clean: we now need to mark those dirty too).
2850                  */
2851                 set_page_dirty(page);
2852                 unlock_page(page);
2853                 put_page(page);
2854                 cond_resched();
2855         }
2856 
2857         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2858                 i_size_write(inode, offset + len);
2859         inode->i_ctime = CURRENT_TIME;
2860 undone:
2861         spin_lock(&inode->i_lock);
2862         inode->i_private = NULL;
2863         spin_unlock(&inode->i_lock);
2864 out:
2865         inode_unlock(inode);
2866         return error;
2867 }
2868 
2869 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2870 {
2871         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2872 
2873         buf->f_type = TMPFS_MAGIC;
2874         buf->f_bsize = PAGE_SIZE;
2875         buf->f_namelen = NAME_MAX;
2876         if (sbinfo->max_blocks) {
2877                 buf->f_blocks = sbinfo->max_blocks;
2878                 buf->f_bavail =
2879                 buf->f_bfree  = sbinfo->max_blocks -
2880                                 percpu_counter_sum(&sbinfo->used_blocks);
2881         }
2882         if (sbinfo->max_inodes) {
2883                 buf->f_files = sbinfo->max_inodes;
2884                 buf->f_ffree = sbinfo->free_inodes;
2885         }
2886         /* else leave those fields 0 like simple_statfs */
2887         return 0;
2888 }
2889 
2890 /*
2891  * File creation. Allocate an inode, and we're done..
2892  */
2893 static int
2894 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2895 {
2896         struct inode *inode;
2897         int error = -ENOSPC;
2898 
2899         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2900         if (inode) {
2901                 error = simple_acl_create(dir, inode);
2902                 if (error)
2903                         goto out_iput;
2904                 error = security_inode_init_security(inode, dir,
2905                                                      &dentry->d_name,
2906                                                      shmem_initxattrs, NULL);
2907                 if (error && error != -EOPNOTSUPP)
2908                         goto out_iput;
2909 
2910                 error = 0;
2911                 dir->i_size += BOGO_DIRENT_SIZE;
2912                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2913                 d_instantiate(dentry, inode);
2914                 dget(dentry); /* Extra count - pin the dentry in core */
2915         }
2916         return error;
2917 out_iput:
2918         iput(inode);
2919         return error;
2920 }
2921 
2922 static int
2923 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2924 {
2925         struct inode *inode;
2926         int error = -ENOSPC;
2927 
2928         inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2929         if (inode) {
2930                 error = security_inode_init_security(inode, dir,
2931                                                      NULL,
2932                                                      shmem_initxattrs, NULL);
2933                 if (error && error != -EOPNOTSUPP)
2934                         goto out_iput;
2935                 error = simple_acl_create(dir, inode);
2936                 if (error)
2937                         goto out_iput;
2938                 d_tmpfile(dentry, inode);
2939         }
2940         return error;
2941 out_iput:
2942         iput(inode);
2943         return error;
2944 }
2945 
2946 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2947 {
2948         int error;
2949 
2950         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2951                 return error;
2952         inc_nlink(dir);
2953         return 0;
2954 }
2955 
2956 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2957                 bool excl)
2958 {
2959         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2960 }
2961 
2962 /*
2963  * Link a file..
2964  */
2965 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2966 {
2967         struct inode *inode = d_inode(old_dentry);
2968         int ret;
2969 
2970         /*
2971          * No ordinary (disk based) filesystem counts links as inodes;
2972          * but each new link needs a new dentry, pinning lowmem, and
2973          * tmpfs dentries cannot be pruned until they are unlinked.
2974          */
2975         ret = shmem_reserve_inode(inode->i_sb);
2976         if (ret)
2977                 goto out;
2978 
2979         dir->i_size += BOGO_DIRENT_SIZE;
2980         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2981         inc_nlink(inode);
2982         ihold(inode);   /* New dentry reference */
2983         dget(dentry);           /* Extra pinning count for the created dentry */
2984         d_instantiate(dentry, inode);
2985 out:
2986         return ret;
2987 }
2988 
2989 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2990 {
2991         struct inode *inode = d_inode(dentry);
2992 
2993         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2994                 shmem_free_inode(inode->i_sb);
2995 
2996         dir->i_size -= BOGO_DIRENT_SIZE;
2997         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2998         drop_nlink(inode);
2999         dput(dentry);   /* Undo the count from "create" - this does all the work */
3000         return 0;
3001 }
3002 
3003 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3004 {
3005         if (!simple_empty(dentry))
3006                 return -ENOTEMPTY;
3007 
3008         drop_nlink(d_inode(dentry));
3009         drop_nlink(dir);
3010         return shmem_unlink(dir, dentry);
3011 }
3012 
3013 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3014 {
3015         bool old_is_dir = d_is_dir(old_dentry);
3016         bool new_is_dir = d_is_dir(new_dentry);
3017 
3018         if (old_dir != new_dir && old_is_dir != new_is_dir) {
3019                 if (old_is_dir) {
3020                         drop_nlink(old_dir);
3021                         inc_nlink(new_dir);
3022                 } else {
3023                         drop_nlink(new_dir);
3024                         inc_nlink(old_dir);
3025                 }
3026         }
3027         old_dir->i_ctime = old_dir->i_mtime =
3028         new_dir->i_ctime = new_dir->i_mtime =
3029         d_inode(old_dentry)->i_ctime =
3030         d_inode(new_dentry)->i_ctime = CURRENT_TIME;
3031 
3032         return 0;
3033 }
3034 
3035 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3036 {
3037         struct dentry *whiteout;
3038         int error;
3039 
3040         whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3041         if (!whiteout)
3042                 return -ENOMEM;
3043 
3044         error = shmem_mknod(old_dir, whiteout,
3045                             S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3046         dput(whiteout);
3047         if (error)
3048                 return error;
3049 
3050         /*
3051          * Cheat and hash the whiteout while the old dentry is still in
3052          * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3053          *
3054          * d_lookup() will consistently find one of them at this point,
3055          * not sure which one, but that isn't even important.
3056          */
3057         d_rehash(whiteout);
3058         return 0;
3059 }
3060 
3061 /*
3062  * The VFS layer already does all the dentry stuff for rename,
3063  * we just have to decrement the usage count for the target if
3064  * it exists so that the VFS layer correctly free's it when it
3065  * gets overwritten.
3066  */
3067 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3068 {
3069         struct inode *inode = d_inode(old_dentry);
3070         int they_are_dirs = S_ISDIR(inode->i_mode);
3071 
3072         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3073                 return -EINVAL;
3074 
3075         if (flags & RENAME_EXCHANGE)
3076                 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3077 
3078         if (!simple_empty(new_dentry))
3079                 return -ENOTEMPTY;
3080 
3081         if (flags & RENAME_WHITEOUT) {
3082                 int error;
3083 
3084                 error = shmem_whiteout(old_dir, old_dentry);
3085                 if (error)
3086                         return error;
3087         }
3088 
3089         if (d_really_is_positive(new_dentry)) {
3090                 (void) shmem_unlink(new_dir, new_dentry);
3091                 if (they_are_dirs) {
3092                         drop_nlink(d_inode(new_dentry));
3093                         drop_nlink(old_dir);
3094                 }
3095         } else if (they_are_dirs) {
3096                 drop_nlink(old_dir);
3097                 inc_nlink(new_dir);
3098         }
3099 
3100         old_dir->i_size -= BOGO_DIRENT_SIZE;
3101         new_dir->i_size += BOGO_DIRENT_SIZE;
3102         old_dir->i_ctime = old_dir->i_mtime =
3103         new_dir->i_ctime = new_dir->i_mtime =
3104         inode->i_ctime = CURRENT_TIME;
3105         return 0;
3106 }
3107 
3108 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3109 {
3110         int error;
3111         int len;
3112         struct inode *inode;
3113         struct page *page;
3114         struct shmem_inode_info *info;
3115 
3116         len = strlen(symname) + 1;
3117         if (len > PAGE_SIZE)
3118                 return -ENAMETOOLONG;
3119 
3120         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
3121         if (!inode)
3122                 return -ENOSPC;
3123 
3124         error = security_inode_init_security(inode, dir, &dentry->d_name,
3125                                              shmem_initxattrs, NULL);
3126         if (error) {
3127                 if (error != -EOPNOTSUPP) {
3128                         iput(inode);
3129                         return error;
3130                 }
3131                 error = 0;
3132         }
3133 
3134         info = SHMEM_I(inode);
3135         inode->i_size = len-1;
3136         if (len <= SHORT_SYMLINK_LEN) {
3137                 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3138                 if (!inode->i_link) {
3139                         iput(inode);
3140                         return -ENOMEM;
3141                 }
3142                 inode->i_op = &shmem_short_symlink_operations;
3143         } else {
3144                 inode_nohighmem(inode);
3145                 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3146                 if (error) {
3147                         iput(inode);
3148                         return error;
3149                 }
3150                 inode->i_mapping->a_ops = &shmem_aops;
3151                 inode->i_op = &shmem_symlink_inode_operations;
3152                 memcpy(page_address(page), symname, len);
3153                 SetPageUptodate(page);
3154                 set_page_dirty(page);
3155                 unlock_page(page);
3156                 put_page(page);
3157         }
3158         dir->i_size += BOGO_DIRENT_SIZE;
3159         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
3160         d_instantiate(dentry, inode);
3161         dget(dentry);
3162         return 0;
3163 }
3164 
3165 static void shmem_put_link(void *arg)
3166 {
3167         mark_page_accessed(arg);
3168         put_page(arg);
3169 }
3170 
3171 static const char *shmem_get_link(struct dentry *dentry,
3172                                   struct inode *inode,
3173                                   struct delayed_call *done)
3174 {
3175         struct page *page = NULL;
3176         int error;
3177         if (!dentry) {
3178                 page = find_get_page(inode->i_mapping, 0);
3179                 if (!page)
3180                         return ERR_PTR(-ECHILD);
3181                 if (!PageUptodate(page)) {
3182                         put_page(page);
3183                         return ERR_PTR(-ECHILD);
3184                 }
3185         } else {
3186                 error = shmem_getpage(inode, 0, &page, SGP_READ);
3187                 if (error)
3188                         return ERR_PTR(error);
3189                 unlock_page(page);
3190         }
3191         set_delayed_call(done, shmem_put_link, page);
3192         return page_address(page);
3193 }
3194 
3195 #ifdef CONFIG_TMPFS_XATTR
3196 /*
3197  * Superblocks without xattr inode operations may get some security.* xattr
3198  * support from the LSM "for free". As soon as we have any other xattrs
3199  * like ACLs, we also need to implement the security.* handlers at
3200  * filesystem level, though.
3201  */
3202 
3203 /*
3204  * Callback for security_inode_init_security() for acquiring xattrs.
3205  */
3206 static int shmem_initxattrs(struct inode *inode,
3207                             const struct xattr *xattr_array,
3208                             void *fs_info)
3209 {
3210         struct shmem_inode_info *info = SHMEM_I(inode);
3211         const struct xattr *xattr;
3212         struct simple_xattr *new_xattr;
3213         size_t len;
3214 
3215         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3216                 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3217                 if (!new_xattr)
3218                         return -ENOMEM;
3219 
3220                 len = strlen(xattr->name) + 1;
3221                 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3222                                           GFP_KERNEL);
3223                 if (!new_xattr->name) {
3224                         kfree(new_xattr);
3225                         return -ENOMEM;
3226                 }
3227 
3228                 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3229                        XATTR_SECURITY_PREFIX_LEN);
3230                 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3231                        xattr->name, len);
3232 
3233                 simple_xattr_list_add(&info->xattrs, new_xattr);
3234         }
3235 
3236         return 0;
3237 }
3238 
3239 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3240                                    struct dentry *unused, struct inode *inode,
3241                                    const char *name, void *buffer, size_t size)
3242 {
3243         struct shmem_inode_info *info = SHMEM_I(inode);
3244 
3245         name = xattr_full_name(handler, name);
3246         return simple_xattr_get(&info->xattrs, name, buffer, size);
3247 }
3248 
3249 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3250                                    struct dentry *unused, struct inode *inode,
3251                                    const char *name, const void *value,
3252                                    size_t size, int flags)
3253 {
3254         struct shmem_inode_info *info = SHMEM_I(inode);
3255 
3256         name = xattr_full_name(handler, name);
3257         return simple_xattr_set(&info->xattrs, name, value, size, flags);
3258 }
3259 
3260 static const struct xattr_handler shmem_security_xattr_handler = {
3261         .prefix = XATTR_SECURITY_PREFIX,
3262         .get = shmem_xattr_handler_get,
3263         .set = shmem_xattr_handler_set,
3264 };
3265 
3266 static const struct xattr_handler shmem_trusted_xattr_handler = {
3267         .prefix = XATTR_TRUSTED_PREFIX,
3268         .get = shmem_xattr_handler_get,
3269         .set = shmem_xattr_handler_set,
3270 };
3271 
3272 static const struct xattr_handler *shmem_xattr_handlers[] = {
3273 #ifdef CONFIG_TMPFS_POSIX_ACL
3274         &posix_acl_access_xattr_handler,
3275         &posix_acl_default_xattr_handler,
3276 #endif
3277         &shmem_security_xattr_handler,
3278         &shmem_trusted_xattr_handler,
3279         NULL
3280 };
3281 
3282 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3283 {
3284         struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3285         return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3286 }
3287 #endif /* CONFIG_TMPFS_XATTR */
3288 
3289 static const struct inode_operations shmem_short_symlink_operations = {
3290         .readlink       = generic_readlink,
3291         .get_link       = simple_get_link,
3292 #ifdef CONFIG_TMPFS_XATTR
3293         .setxattr       = generic_setxattr,
3294         .getxattr       = generic_getxattr,
3295         .listxattr      = shmem_listxattr,
3296         .removexattr    = generic_removexattr,
3297 #endif
3298 };
3299 
3300 static const struct inode_operations shmem_symlink_inode_operations = {
3301         .readlink       = generic_readlink,
3302         .get_link       = shmem_get_link,
3303 #ifdef CONFIG_TMPFS_XATTR
3304         .setxattr       = generic_setxattr,
3305         .getxattr       = generic_getxattr,
3306         .listxattr      = shmem_listxattr,
3307         .removexattr    = generic_removexattr,
3308 #endif
3309 };
3310 
3311 static struct dentry *shmem_get_parent(struct dentry *child)
3312 {
3313         return ERR_PTR(-ESTALE);
3314 }
3315 
3316 static int shmem_match(struct inode *ino, void *vfh)
3317 {
3318         __u32 *fh = vfh;
3319         __u64 inum = fh[2];
3320         inum = (inum << 32) | fh[1];
3321         return ino->i_ino == inum && fh[0] == ino->i_generation;
3322 }
3323 
3324 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3325                 struct fid *fid, int fh_len, int fh_type)
3326 {
3327         struct inode *inode;
3328         struct dentry *dentry = NULL;
3329         u64 inum;
3330 
3331         if (fh_len < 3)
3332                 return NULL;
3333 
3334         inum = fid->raw[2];
3335         inum = (inum << 32) | fid->raw[1];
3336 
3337         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3338                         shmem_match, fid->raw);
3339         if (inode) {
3340                 dentry = d_find_alias(inode);
3341                 iput(inode);
3342         }
3343 
3344         return dentry;
3345 }
3346 
3347 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3348                                 struct inode *parent)
3349 {
3350         if (*len < 3) {
3351                 *len = 3;
3352                 return FILEID_INVALID;
3353         }
3354 
3355         if (inode_unhashed(inode)) {
3356                 /* Unfortunately insert_inode_hash is not idempotent,
3357                  * so as we hash inodes here rather than at creation
3358                  * time, we need a lock to ensure we only try
3359                  * to do it once
3360                  */
3361                 static DEFINE_SPINLOCK(lock);
3362                 spin_lock(&lock);
3363                 if (inode_unhashed(inode))
3364                         __insert_inode_hash(inode,
3365                                             inode->i_ino + inode->i_generation);
3366                 spin_unlock(&lock);
3367         }
3368 
3369         fh[0] = inode->i_generation;
3370         fh[1] = inode->i_ino;
3371         fh[2] = ((__u64)inode->i_ino) >> 32;
3372 
3373         *len = 3;
3374         return 1;
3375 }
3376 
3377 static const struct export_operations shmem_export_ops = {
3378         .get_parent     = shmem_get_parent,
3379         .encode_fh      = shmem_encode_fh,
3380         .fh_to_dentry   = shmem_fh_to_dentry,
3381 };
3382 
3383 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3384                                bool remount)
3385 {
3386         char *this_char, *value, *rest;
3387         struct mempolicy *mpol = NULL;
3388         uid_t uid;
3389         gid_t gid;
3390 
3391         while (options != NULL) {
3392                 this_char = options;
3393                 for (;;) {
3394                         /*
3395                          * NUL-terminate this option: unfortunately,
3396                          * mount options form a comma-separated list,
3397                          * but mpol's nodelist may also contain commas.
3398                          */
3399                         options = strchr(options, ',');
3400                         if (options == NULL)
3401                                 break;
3402                         options++;
3403                         if (!isdigit(*options)) {
3404                                 options[-1] = '\0';
3405                                 break;
3406                         }
3407                 }
3408                 if (!*this_char)
3409                         continue;
3410                 if ((value = strchr(this_char,'=')) != NULL) {
3411                         *value++ = 0;
3412                 } else {
3413                         pr_err("tmpfs: No value for mount option '%s'\n",
3414                                this_char);
3415                         goto error;
3416                 }
3417 
3418                 if (!strcmp(this_char,"size")) {
3419                         unsigned long long size;
3420                         size = memparse(value,&rest);
3421                         if (*rest == '%') {
3422                                 size <<= PAGE_SHIFT;
3423                                 size *= totalram_pages;
3424                                 do_div(size, 100);
3425                                 rest++;
3426                         }
3427                         if (*rest)
3428                                 goto bad_val;
3429                         sbinfo->max_blocks =
3430                                 DIV_ROUND_UP(size, PAGE_SIZE);
3431                 } else if (!strcmp(this_char,"nr_blocks")) {
3432                         sbinfo->max_blocks = memparse(value, &rest);
3433                         if (*rest)
3434                                 goto bad_val;
3435                 } else if (!strcmp(this_char,"nr_inodes")) {
3436                         sbinfo->max_inodes = memparse(value, &rest);
3437                         if (*rest)
3438                                 goto bad_val;
3439                 } else if (!strcmp(this_char,"mode")) {
3440                         if (remount)
3441                                 continue;
3442                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
3443                         if (*rest)
3444                                 goto bad_val;
3445                 } else if (!strcmp(this_char,"uid")) {
3446                         if (remount)
3447                                 continue;
3448                         uid = simple_strtoul(value, &rest, 0);
3449                         if (*rest)
3450                                 goto bad_val;
3451                         sbinfo->uid = make_kuid(current_user_ns(), uid);
3452                         if (!uid_valid(sbinfo->uid))
3453                                 goto bad_val;
3454                 } else if (!strcmp(this_char,"gid")) {
3455                         if (remount)
3456                                 continue;
3457                         gid = simple_strtoul(value, &rest, 0);
3458                         if (*rest)
3459                                 goto bad_val;
3460                         sbinfo->gid = make_kgid(current_user_ns(), gid);
3461                         if (!gid_valid(sbinfo->gid))
3462                                 goto bad_val;
3463 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3464                 } else if (!strcmp(this_char, "huge")) {
3465                         int huge;
3466                         huge = shmem_parse_huge(value);
3467                         if (huge < 0)
3468                                 goto bad_val;
3469                         if (!has_transparent_hugepage() &&
3470                                         huge != SHMEM_HUGE_NEVER)
3471                                 goto bad_val;
3472                         sbinfo->huge = huge;
3473 #endif
3474 #ifdef CONFIG_NUMA
3475                 } else if (!strcmp(this_char,"mpol")) {
3476                         mpol_put(mpol);
3477                         mpol = NULL;
3478                         if (mpol_parse_str(value, &mpol))
3479                                 goto bad_val;
3480 #endif
3481                 } else {
3482                         pr_err("tmpfs: Bad mount option %s\n", this_char);
3483                         goto error;
3484                 }
3485         }
3486         sbinfo->mpol = mpol;
3487         return 0;
3488 
3489 bad_val:
3490         pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
3491                value, this_char);
3492 error:
3493         mpol_put(mpol);
3494         return 1;
3495 
3496 }
3497 
3498 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3499 {
3500         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3501         struct shmem_sb_info config = *sbinfo;
3502         unsigned long inodes;
3503         int error = -EINVAL;
3504 
3505         config.mpol = NULL;
3506         if (shmem_parse_options(data, &config, true))
3507                 return error;
3508 
3509         spin_lock(&sbinfo->stat_lock);
3510         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3511         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
3512                 goto out;
3513         if (config.max_inodes < inodes)
3514                 goto out;
3515         /*
3516          * Those tests disallow limited->unlimited while any are in use;
3517          * but we must separately disallow unlimited->limited, because
3518          * in that case we have no record of how much is already in use.
3519          */
3520         if (config.max_blocks && !sbinfo->max_blocks)
3521                 goto out;
3522         if (config.max_inodes && !sbinfo->max_inodes)
3523                 goto out;
3524 
3525         error = 0;
3526         sbinfo->huge = config.huge;
3527         sbinfo->max_blocks  = config.max_blocks;
3528         sbinfo->max_inodes  = config.max_inodes;
3529         sbinfo->free_inodes = config.max_inodes - inodes;
3530 
3531         /*
3532          * Preserve previous mempolicy unless mpol remount option was specified.
3533          */
3534         if (config.mpol) {
3535                 mpol_put(sbinfo->mpol);
3536                 sbinfo->mpol = config.mpol;     /* transfers initial ref */
3537         }
3538 out:
3539         spin_unlock(&sbinfo->stat_lock);
3540         return error;
3541 }
3542 
3543 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3544 {
3545         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3546 
3547         if (sbinfo->max_blocks != shmem_default_max_blocks())
3548                 seq_printf(seq, ",size=%luk",
3549                         sbinfo->max_blocks << (PAGE_SHIFT - 10));
3550         if (sbinfo->max_inodes != shmem_default_max_inodes())
3551                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3552         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
3553                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3554         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3555                 seq_printf(seq, ",uid=%u",
3556                                 from_kuid_munged(&init_user_ns, sbinfo->uid));
3557         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3558                 seq_printf(seq, ",gid=%u",
3559                                 from_kgid_munged(&init_user_ns, sbinfo->gid));
3560 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3561         /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3562         if (sbinfo->huge)
3563                 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3564 #endif
3565         shmem_show_mpol(seq, sbinfo->mpol);
3566         return 0;
3567 }
3568 
3569 #define MFD_NAME_PREFIX "memfd:"
3570 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3571 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3572 
3573 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3574 
3575 SYSCALL_DEFINE2(memfd_create,
3576                 const char __user *, uname,
3577                 unsigned int, flags)
3578 {
3579         struct shmem_inode_info *info;
3580         struct file *file;
3581         int fd, error;
3582         char *name;
3583         long len;
3584 
3585         if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3586                 return -EINVAL;
3587 
3588         /* length includes terminating zero */
3589         len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3590         if (len <= 0)
3591                 return -EFAULT;
3592         if (len > MFD_NAME_MAX_LEN + 1)
3593                 return -EINVAL;
3594 
3595         name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3596         if (!name)
3597                 return -ENOMEM;
3598 
3599         strcpy(name, MFD_NAME_PREFIX);
3600         if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3601                 error = -EFAULT;
3602                 goto err_name;
3603         }
3604 
3605         /* terminating-zero may have changed after strnlen_user() returned */
3606         if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3607                 error = -EFAULT;
3608                 goto err_name;
3609         }
3610 
3611         fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3612         if (fd < 0) {
3613                 error = fd;
3614                 goto err_name;
3615         }
3616 
3617         file = shmem_file_setup(name, 0, VM_NORESERVE);
3618         if (IS_ERR(file)) {
3619                 error = PTR_ERR(file);
3620                 goto err_fd;
3621         }
3622         info = SHMEM_I(file_inode(file));
3623         file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3624         file->f_flags |= O_RDWR | O_LARGEFILE;
3625         if (flags & MFD_ALLOW_SEALING)
3626                 info->seals &= ~F_SEAL_SEAL;
3627 
3628         fd_install(fd, file);
3629         kfree(name);
3630         return fd;
3631 
3632 err_fd:
3633         put_unused_fd(fd);
3634 err_name:
3635         kfree(name);
3636         return error;
3637 }
3638 
3639 #endif /* CONFIG_TMPFS */
3640 
3641 static void shmem_put_super(struct super_block *sb)
3642 {
3643         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3644 
3645         percpu_counter_destroy(&sbinfo->used_blocks);
3646         mpol_put(sbinfo->mpol);
3647         kfree(sbinfo);
3648         sb->s_fs_info = NULL;
3649 }
3650 
3651 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3652 {
3653         struct inode *inode;
3654         struct shmem_sb_info *sbinfo;
3655         int err = -ENOMEM;
3656 
3657         /* Round up to L1_CACHE_BYTES to resist false sharing */
3658         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3659                                 L1_CACHE_BYTES), GFP_KERNEL);
3660         if (!sbinfo)
3661                 return -ENOMEM;
3662 
3663         sbinfo->mode = S_IRWXUGO | S_ISVTX;
3664         sbinfo->uid = current_fsuid();
3665         sbinfo->gid = current_fsgid();
3666         sb->s_fs_info = sbinfo;
3667 
3668 #ifdef CONFIG_TMPFS
3669         /*
3670          * Per default we only allow half of the physical ram per
3671          * tmpfs instance, limiting inodes to one per page of lowmem;
3672          * but the internal instance is left unlimited.
3673          */
3674         if (!(sb->s_flags & MS_KERNMOUNT)) {
3675                 sbinfo->max_blocks = shmem_default_max_blocks();
3676                 sbinfo->max_inodes = shmem_default_max_inodes();
3677                 if (shmem_parse_options(data, sbinfo, false)) {
3678                         err = -EINVAL;
3679                         goto failed;
3680                 }
3681         } else {
3682                 sb->s_flags |= MS_NOUSER;
3683         }
3684         sb->s_export_op = &shmem_export_ops;
3685         sb->s_flags |= MS_NOSEC;
3686 #else
3687         sb->s_flags |= MS_NOUSER;
3688 #endif
3689 
3690         spin_lock_init(&sbinfo->stat_lock);
3691         if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3692                 goto failed;
3693         sbinfo->free_inodes = sbinfo->max_inodes;
3694         spin_lock_init(&sbinfo->shrinklist_lock);
3695         INIT_LIST_HEAD(&sbinfo->shrinklist);
3696 
3697         sb->s_maxbytes = MAX_LFS_FILESIZE;
3698         sb->s_blocksize = PAGE_SIZE;
3699         sb->s_blocksize_bits = PAGE_SHIFT;
3700         sb->s_magic = TMPFS_MAGIC;
3701         sb->s_op = &shmem_ops;
3702         sb->s_time_gran = 1;
3703 #ifdef CONFIG_TMPFS_XATTR
3704         sb->s_xattr = shmem_xattr_handlers;
3705 #endif
3706 #ifdef CONFIG_TMPFS_POSIX_ACL
3707         sb->s_flags |= MS_POSIXACL;
3708 #endif
3709 
3710         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3711         if (!inode)
3712                 goto failed;
3713         inode->i_uid = sbinfo->uid;
3714         inode->i_gid = sbinfo->gid;
3715         sb->s_root = d_make_root(inode);
3716         if (!sb->s_root)
3717                 goto failed;
3718         return 0;
3719 
3720 failed:
3721         shmem_put_super(sb);
3722         return err;
3723 }
3724 
3725 static struct kmem_cache *shmem_inode_cachep;
3726 
3727 static struct inode *shmem_alloc_inode(struct super_block *sb)
3728 {
3729         struct shmem_inode_info *info;
3730         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3731         if (!info)
3732                 return NULL;
3733         return &info->vfs_inode;
3734 }
3735 
3736 static void shmem_destroy_callback(struct rcu_head *head)
3737 {
3738         struct inode *inode = container_of(head, struct inode, i_rcu);
3739         if (S_ISLNK(inode->i_mode))
3740                 kfree(inode->i_link);
3741         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3742 }
3743 
3744 static void shmem_destroy_inode(struct inode *inode)
3745 {
3746         if (S_ISREG(inode->i_mode))
3747                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3748         call_rcu(&inode->i_rcu, shmem_destroy_callback);
3749 }
3750 
3751 static void shmem_init_inode(void *foo)
3752 {
3753         struct shmem_inode_info *info = foo;
3754         inode_init_once(&info->vfs_inode);
3755 }
3756 
3757 static int shmem_init_inodecache(void)
3758 {
3759         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3760                                 sizeof(struct shmem_inode_info),
3761                                 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3762         return 0;
3763 }
3764 
3765 static void shmem_destroy_inodecache(void)
3766 {
3767         kmem_cache_destroy(shmem_inode_cachep);
3768 }
3769 
3770 static const struct address_space_operations shmem_aops = {
3771         .writepage      = shmem_writepage,
3772         .set_page_dirty = __set_page_dirty_no_writeback,
3773 #ifdef CONFIG_TMPFS
3774         .write_begin    = shmem_write_begin,
3775         .write_end      = shmem_write_end,
3776 #endif
3777 #ifdef CONFIG_MIGRATION
3778         .migratepage    = migrate_page,
3779 #endif
3780         .error_remove_page = generic_error_remove_page,
3781 };
3782 
3783 static const struct file_operations shmem_file_operations = {
3784         .mmap           = shmem_mmap,
3785         .get_unmapped_area = shmem_get_unmapped_area,
3786 #ifdef CONFIG_TMPFS
3787         .llseek         = shmem_file_llseek,
3788         .read_iter      = shmem_file_read_iter,
3789         .write_iter     = generic_file_write_iter,
3790         .fsync          = noop_fsync,
3791         .splice_read    = shmem_file_splice_read,
3792         .splice_write   = iter_file_splice_write,
3793         .fallocate      = shmem_fallocate,
3794 #endif
3795 };
3796 
3797 static const struct inode_operations shmem_inode_operations = {
3798         .getattr        = shmem_getattr,
3799         .setattr        = shmem_setattr,
3800 #ifdef CONFIG_TMPFS_XATTR
3801         .setxattr       = generic_setxattr,
3802         .getxattr       = generic_getxattr,
3803         .listxattr      = shmem_listxattr,
3804         .removexattr    = generic_removexattr,
3805         .set_acl        = simple_set_acl,
3806 #endif
3807 };
3808 
3809 static const struct inode_operations shmem_dir_inode_operations = {
3810 #ifdef CONFIG_TMPFS
3811         .create         = shmem_create,
3812         .lookup         = simple_lookup,
3813         .link           = shmem_link,
3814         .unlink         = shmem_unlink,
3815         .symlink        = shmem_symlink,
3816         .mkdir          = shmem_mkdir,
3817         .rmdir          = shmem_rmdir,
3818         .mknod          = shmem_mknod,
3819         .rename2        = shmem_rename2,
3820         .tmpfile        = shmem_tmpfile,
3821 #endif
3822 #ifdef CONFIG_TMPFS_XATTR
3823         .setxattr       = generic_setxattr,
3824         .getxattr       = generic_getxattr,
3825         .listxattr      = shmem_listxattr,
3826         .removexattr    = generic_removexattr,
3827 #endif
3828 #ifdef CONFIG_TMPFS_POSIX_ACL
3829         .setattr        = shmem_setattr,
3830         .set_acl        = simple_set_acl,
3831 #endif
3832 };
3833 
3834 static const struct inode_operations shmem_special_inode_operations = {
3835 #ifdef CONFIG_TMPFS_XATTR
3836         .setxattr       = generic_setxattr,
3837         .getxattr       = generic_getxattr,
3838         .listxattr      = shmem_listxattr,
3839         .removexattr    = generic_removexattr,
3840 #endif
3841 #ifdef CONFIG_TMPFS_POSIX_ACL
3842         .setattr        = shmem_setattr,
3843         .set_acl        = simple_set_acl,
3844 #endif
3845 };
3846 
3847 static const struct super_operations shmem_ops = {
3848         .alloc_inode    = shmem_alloc_inode,
3849         .destroy_inode  = shmem_destroy_inode,
3850 #ifdef CONFIG_TMPFS
3851         .statfs         = shmem_statfs,
3852         .remount_fs     = shmem_remount_fs,
3853         .show_options   = shmem_show_options,
3854 #endif
3855         .evict_inode    = shmem_evict_inode,
3856         .drop_inode     = generic_delete_inode,
3857         .put_super      = shmem_put_super,
3858 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3859         .nr_cached_objects      = shmem_unused_huge_count,
3860         .free_cached_objects    = shmem_unused_huge_scan,
3861 #endif
3862 };
3863 
3864 static const struct vm_operations_struct shmem_vm_ops = {
3865         .fault          = shmem_fault,
3866         .map_pages      = filemap_map_pages,
3867 #ifdef CONFIG_NUMA
3868         .set_policy     = shmem_set_policy,
3869         .get_policy     = shmem_get_policy,
3870 #endif
3871 };
3872 
3873 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3874         int flags, const char *dev_name, void *data)
3875 {
3876         return mount_nodev(fs_type, flags, data, shmem_fill_super);
3877 }
3878 
3879 static struct file_system_type shmem_fs_type = {
3880         .owner          = THIS_MODULE,
3881         .name           = "tmpfs",
3882         .mount          = shmem_mount,
3883         .kill_sb        = kill_litter_super,
3884         .fs_flags       = FS_USERNS_MOUNT,
3885 };
3886 
3887 int __init shmem_init(void)
3888 {
3889         int error;
3890 
3891         /* If rootfs called this, don't re-init */
3892         if (shmem_inode_cachep)
3893                 return 0;
3894 
3895         error = shmem_init_inodecache();
3896         if (error)
3897                 goto out3;
3898 
3899         error = register_filesystem(&shmem_fs_type);
3900         if (error) {
3901                 pr_err("Could not register tmpfs\n");
3902                 goto out2;
3903         }
3904 
3905         shm_mnt = kern_mount(&shmem_fs_type);
3906         if (IS_ERR(shm_mnt)) {
3907                 error = PTR_ERR(shm_mnt);
3908                 pr_err("Could not kern_mount tmpfs\n");
3909                 goto out1;
3910         }
3911 
3912 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3913         if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3914                 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3915         else
3916                 shmem_huge = 0; /* just in case it was patched */
3917 #endif
3918         return 0;
3919 
3920 out1:
3921         unregister_filesystem(&shmem_fs_type);
3922 out2:
3923         shmem_destroy_inodecache();
3924 out3:
3925         shm_mnt = ERR_PTR(error);
3926         return error;
3927 }
3928 
3929 #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3930 static ssize_t shmem_enabled_show(struct kobject *kobj,
3931                 struct kobj_attribute *attr, char *buf)
3932 {
3933         int values[] = {
3934                 SHMEM_HUGE_ALWAYS,
3935                 SHMEM_HUGE_WITHIN_SIZE,
3936                 SHMEM_HUGE_ADVISE,
3937                 SHMEM_HUGE_NEVER,
3938                 SHMEM_HUGE_DENY,
3939                 SHMEM_HUGE_FORCE,
3940         };
3941         int i, count;
3942 
3943         for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3944                 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3945 
3946                 count += sprintf(buf + count, fmt,
3947                                 shmem_format_huge(values[i]));
3948         }
3949         buf[count - 1] = '\n';
3950         return count;
3951 }
3952 
3953 static ssize_t shmem_enabled_store(struct kobject *kobj,
3954                 struct kobj_attribute *attr, const char *buf, size_t count)
3955 {
3956         char tmp[16];
3957         int huge;
3958 
3959         if (count + 1 > sizeof(tmp))
3960                 return -EINVAL;
3961         memcpy(tmp, buf, count);
3962         tmp[count] = '\0';
3963         if (count && tmp[count - 1] == '\n')
3964                 tmp[count - 1] = '\0';
3965 
3966         huge = shmem_parse_huge(tmp);
3967         if (huge == -EINVAL)
3968                 return -EINVAL;
3969         if (!has_transparent_hugepage() &&
3970                         huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3971                 return -EINVAL;
3972 
3973         shmem_huge = huge;
3974         if (shmem_huge < SHMEM_HUGE_DENY)
3975                 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3976         return count;
3977 }
3978 
3979 struct kobj_attribute shmem_enabled_attr =
3980         __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3981 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3982 
3983 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3984 bool shmem_huge_enabled(struct vm_area_struct *vma)
3985 {
3986         struct inode *inode = file_inode(vma->vm_file);
3987         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3988         loff_t i_size;
3989         pgoff_t off;
3990 
3991         if (shmem_huge == SHMEM_HUGE_FORCE)
3992                 return true;
3993         if (shmem_huge == SHMEM_HUGE_DENY)
3994                 return false;
3995         switch (sbinfo->huge) {
3996                 case SHMEM_HUGE_NEVER:
3997                         return false;
3998                 case SHMEM_HUGE_ALWAYS:
3999                         return true;
4000                 case SHMEM_HUGE_WITHIN_SIZE:
4001                         off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4002                         i_size = round_up(i_size_read(inode), PAGE_SIZE);
4003                         if (i_size >= HPAGE_PMD_SIZE &&
4004                                         i_size >> PAGE_SHIFT >= off)
4005                                 return true;
4006                 case SHMEM_HUGE_ADVISE:
4007                         /* TODO: implement fadvise() hints */
4008                         return (vma->vm_flags & VM_HUGEPAGE);
4009                 default:
4010                         VM_BUG_ON(1);
4011                         return false;
4012         }
4013 }
4014 #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4015 
4016 #else /* !CONFIG_SHMEM */
4017 
4018 /*
4019  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4020  *
4021  * This is intended for small system where the benefits of the full
4022  * shmem code (swap-backed and resource-limited) are outweighed by
4023  * their complexity. On systems without swap this code should be
4024  * effectively equivalent, but much lighter weight.
4025  */
4026 
4027 static struct file_system_type shmem_fs_type = {
4028         .name           = "tmpfs",
4029         .mount          = ramfs_mount,
4030         .kill_sb        = kill_litter_super,
4031         .fs_flags       = FS_USERNS_MOUNT,
4032 };
4033 
4034 int __init shmem_init(void)
4035 {
4036         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4037 
4038         shm_mnt = kern_mount(&shmem_fs_type);
4039         BUG_ON(IS_ERR(shm_mnt));
4040 
4041         return 0;
4042 }
4043 
4044 int shmem_unuse(swp_entry_t swap, struct page *page)
4045 {
4046         return 0;
4047 }
4048 
4049 int shmem_lock(struct file *file, int lock, struct user_struct *user)
4050 {
4051         return 0;
4052 }
4053 
4054 void shmem_unlock_mapping(struct address_space *mapping)
4055 {
4056 }
4057 
4058 #ifdef CONFIG_MMU
4059 unsigned long shmem_get_unmapped_area(struct file *file,
4060                                       unsigned long addr, unsigned long len,
4061                                       unsigned long pgoff, unsigned long flags)
4062 {
4063         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4064 }
4065 #endif
4066 
4067 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4068 {
4069         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4070 }
4071 EXPORT_SYMBOL_GPL(shmem_truncate_range);
4072 
4073 #define shmem_vm_ops                            generic_file_vm_ops
4074 #define shmem_file_operations                   ramfs_file_operations
4075 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
4076 #define shmem_acct_size(flags, size)            0
4077 #define shmem_unacct_size(flags, size)          do {} while (0)
4078 
4079 #endif /* CONFIG_SHMEM */
4080 
4081 /* common code */
4082 
4083 static struct dentry_operations anon_ops = {
4084         .d_dname = simple_dname
4085 };
4086 
4087 static struct file *__shmem_file_setup(const char *name, loff_t size,
4088                                        unsigned long flags, unsigned int i_flags)
4089 {
4090         struct file *res;
4091         struct inode *inode;
4092         struct path path;
4093         struct super_block *sb;
4094         struct qstr this;
4095 
4096         if (IS_ERR(shm_mnt))
4097                 return ERR_CAST(shm_mnt);
4098 
4099         if (size < 0 || size > MAX_LFS_FILESIZE)
4100                 return ERR_PTR(-EINVAL);
4101 
4102         if (shmem_acct_size(flags, size))
4103                 return ERR_PTR(-ENOMEM);
4104 
4105         res = ERR_PTR(-ENOMEM);
4106         this.name = name;
4107         this.len = strlen(name);
4108         this.hash = 0; /* will go */
4109         sb = shm_mnt->mnt_sb;
4110         path.mnt = mntget(shm_mnt);
4111         path.dentry = d_alloc_pseudo(sb, &this);
4112         if (!path.dentry)
4113                 goto put_memory;
4114         d_set_d_op(path.dentry, &anon_ops);
4115 
4116         res = ERR_PTR(-ENOSPC);
4117         inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
4118         if (!inode)
4119                 goto put_memory;
4120 
4121         inode->i_flags |= i_flags;
4122         d_instantiate(path.dentry, inode);
4123         inode->i_size = size;
4124         clear_nlink(inode);     /* It is unlinked */
4125         res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4126         if (IS_ERR(res))
4127                 goto put_path;
4128 
4129         res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
4130                   &shmem_file_operations);
4131         if (IS_ERR(res))
4132                 goto put_path;
4133 
4134         return res;
4135 
4136 put_memory:
4137         shmem_unacct_size(flags, size);
4138 put_path:
4139         path_put(&path);
4140         return res;
4141 }
4142 
4143 /**
4144  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4145  *      kernel internal.  There will be NO LSM permission checks against the
4146  *      underlying inode.  So users of this interface must do LSM checks at a
4147  *      higher layer.  The users are the big_key and shm implementations.  LSM
4148  *      checks are provided at the key or shm level rather than the inode.
4149  * @name: name for dentry (to be seen in /proc/<pid>/maps
4150  * @size: size to be set for the file
4151  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4152  */
4153 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4154 {
4155         return __shmem_file_setup(name, size, flags, S_PRIVATE);
4156 }
4157 
4158 /**
4159  * shmem_file_setup - get an unlinked file living in tmpfs
4160  * @name: name for dentry (to be seen in /proc/<pid>/maps
4161  * @size: size to be set for the file
4162  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4163  */
4164 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4165 {
4166         return __shmem_file_setup(name, size, flags, 0);
4167 }
4168 EXPORT_SYMBOL_GPL(shmem_file_setup);
4169 
4170 /**
4171  * shmem_zero_setup - setup a shared anonymous mapping
4172  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4173  */
4174 int shmem_zero_setup(struct vm_area_struct *vma)
4175 {
4176         struct file *file;
4177         loff_t size = vma->vm_end - vma->vm_start;
4178 
4179         /*
4180          * Cloning a new file under mmap_sem leads to a lock ordering conflict
4181          * between XFS directory reading and selinux: since this file is only
4182          * accessible to the user through its mapping, use S_PRIVATE flag to
4183          * bypass file security, in the same way as shmem_kernel_file_setup().
4184          */
4185         file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
4186         if (IS_ERR(file))
4187                 return PTR_ERR(file);
4188 
4189         if (vma->vm_file)
4190                 fput(vma->vm_file);
4191         vma->vm_file = file;
4192         vma->vm_ops = &shmem_vm_ops;
4193 
4194         if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4195                         ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4196                         (vma->vm_end & HPAGE_PMD_MASK)) {
4197                 khugepaged_enter(vma, vma->vm_flags);
4198         }
4199 
4200         return 0;
4201 }
4202 
4203 /**
4204  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4205  * @mapping:    the page's address_space
4206  * @index:      the page index
4207  * @gfp:        the page allocator flags to use if allocating
4208  *
4209  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4210  * with any new page allocations done using the specified allocation flags.
4211  * But read_cache_page_gfp() uses the ->readpage() method: which does not
4212  * suit tmpfs, since it may have pages in swapcache, and needs to find those
4213  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4214  *
4215  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4216  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4217  */
4218 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4219                                          pgoff_t index, gfp_t gfp)
4220 {
4221 #ifdef CONFIG_SHMEM
4222         struct inode *inode = mapping->host;
4223         struct page *page;
4224         int error;
4225 
4226         BUG_ON(mapping->a_ops != &shmem_aops);
4227         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4228                                   gfp, NULL, NULL);
4229         if (error)
4230                 page = ERR_PTR(error);
4231         else
4232                 unlock_page(page);
4233         return page;
4234 #else
4235         /*
4236          * The tiny !SHMEM case uses ramfs without swap
4237          */
4238         return read_cache_page_gfp(mapping, index, gfp);
4239 #endif
4240 }
4241 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
4242 

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

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

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

osdn.jp