Version:  2.0.40 2.2.26 2.4.37 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5

Linux/mm/shmem.c

  1 /*
  2  * Resizable virtual memory filesystem for Linux.
  3  *
  4  * Copyright (C) 2000 Linus Torvalds.
  5  *               2000 Transmeta Corp.
  6  *               2000-2001 Christoph Rohland
  7  *               2000-2001 SAP AG
  8  *               2002 Red Hat Inc.
  9  * Copyright (C) 2002-2011 Hugh Dickins.
 10  * Copyright (C) 2011 Google Inc.
 11  * Copyright (C) 2002-2005 VERITAS Software Corporation.
 12  * Copyright (C) 2004 Andi Kleen, SuSE Labs
 13  *
 14  * Extended attribute support for tmpfs:
 15  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
 16  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
 17  *
 18  * tiny-shmem:
 19  * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
 20  *
 21  * This file is released under the GPL.
 22  */
 23 
 24 #include <linux/fs.h>
 25 #include <linux/init.h>
 26 #include <linux/vfs.h>
 27 #include <linux/mount.h>
 28 #include <linux/ramfs.h>
 29 #include <linux/pagemap.h>
 30 #include <linux/file.h>
 31 #include <linux/mm.h>
 32 #include <linux/export.h>
 33 #include <linux/swap.h>
 34 #include <linux/uio.h>
 35 
 36 static struct vfsmount *shm_mnt;
 37 
 38 #ifdef CONFIG_SHMEM
 39 /*
 40  * This virtual memory filesystem is heavily based on the ramfs. It
 41  * extends ramfs by the ability to use swap and honor resource limits
 42  * which makes it a completely usable filesystem.
 43  */
 44 
 45 #include <linux/xattr.h>
 46 #include <linux/exportfs.h>
 47 #include <linux/posix_acl.h>
 48 #include <linux/posix_acl_xattr.h>
 49 #include <linux/mman.h>
 50 #include <linux/string.h>
 51 #include <linux/slab.h>
 52 #include <linux/backing-dev.h>
 53 #include <linux/shmem_fs.h>
 54 #include <linux/writeback.h>
 55 #include <linux/blkdev.h>
 56 #include <linux/pagevec.h>
 57 #include <linux/percpu_counter.h>
 58 #include <linux/falloc.h>
 59 #include <linux/splice.h>
 60 #include <linux/security.h>
 61 #include <linux/swapops.h>
 62 #include <linux/mempolicy.h>
 63 #include <linux/namei.h>
 64 #include <linux/ctype.h>
 65 #include <linux/migrate.h>
 66 #include <linux/highmem.h>
 67 #include <linux/seq_file.h>
 68 #include <linux/magic.h>
 69 #include <linux/syscalls.h>
 70 #include <linux/fcntl.h>
 71 #include <uapi/linux/memfd.h>
 72 
 73 #include <asm/uaccess.h>
 74 #include <asm/pgtable.h>
 75 
 76 #include "internal.h"
 77 
 78 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
 79 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
 80 
 81 /* Pretend that each entry is of this size in directory's i_size */
 82 #define BOGO_DIRENT_SIZE 20
 83 
 84 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
 85 #define SHORT_SYMLINK_LEN 128
 86 
 87 /*
 88  * shmem_fallocate communicates with shmem_fault or shmem_writepage via
 89  * inode->i_private (with i_mutex making sure that it has only one user at
 90  * a time): we would prefer not to enlarge the shmem inode just for that.
 91  */
 92 struct shmem_falloc {
 93         wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
 94         pgoff_t start;          /* start of range currently being fallocated */
 95         pgoff_t next;           /* the next page offset to be fallocated */
 96         pgoff_t nr_falloced;    /* how many new pages have been fallocated */
 97         pgoff_t nr_unswapped;   /* how often writepage refused to swap out */
 98 };
 99 
100 /* Flag allocation requirements to shmem_getpage */
101 enum sgp_type {
102         SGP_READ,       /* don't exceed i_size, don't allocate page */
103         SGP_CACHE,      /* don't exceed i_size, may allocate page */
104         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
105         SGP_WRITE,      /* may exceed i_size, may allocate !Uptodate page */
106         SGP_FALLOC,     /* like SGP_WRITE, but make existing page Uptodate */
107 };
108 
109 #ifdef CONFIG_TMPFS
110 static unsigned long shmem_default_max_blocks(void)
111 {
112         return totalram_pages / 2;
113 }
114 
115 static unsigned long shmem_default_max_inodes(void)
116 {
117         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
118 }
119 #endif
120 
121 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
122 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
123                                 struct shmem_inode_info *info, pgoff_t index);
124 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
125         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
126 
127 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
128         struct page **pagep, enum sgp_type sgp, int *fault_type)
129 {
130         return shmem_getpage_gfp(inode, index, pagep, sgp,
131                         mapping_gfp_mask(inode->i_mapping), fault_type);
132 }
133 
134 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
135 {
136         return sb->s_fs_info;
137 }
138 
139 /*
140  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
141  * for shared memory and for shared anonymous (/dev/zero) mappings
142  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
143  * consistent with the pre-accounting of private mappings ...
144  */
145 static inline int shmem_acct_size(unsigned long flags, loff_t size)
146 {
147         return (flags & VM_NORESERVE) ?
148                 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
149 }
150 
151 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
152 {
153         if (!(flags & VM_NORESERVE))
154                 vm_unacct_memory(VM_ACCT(size));
155 }
156 
157 static inline int shmem_reacct_size(unsigned long flags,
158                 loff_t oldsize, loff_t newsize)
159 {
160         if (!(flags & VM_NORESERVE)) {
161                 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
162                         return security_vm_enough_memory_mm(current->mm,
163                                         VM_ACCT(newsize) - VM_ACCT(oldsize));
164                 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
165                         vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
166         }
167         return 0;
168 }
169 
170 /*
171  * ... whereas tmpfs objects are accounted incrementally as
172  * pages are allocated, in order to allow huge sparse files.
173  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
174  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
175  */
176 static inline int shmem_acct_block(unsigned long flags)
177 {
178         return (flags & VM_NORESERVE) ?
179                 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
180 }
181 
182 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
183 {
184         if (flags & VM_NORESERVE)
185                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
186 }
187 
188 static const struct super_operations shmem_ops;
189 static const struct address_space_operations shmem_aops;
190 static const struct file_operations shmem_file_operations;
191 static const struct inode_operations shmem_inode_operations;
192 static const struct inode_operations shmem_dir_inode_operations;
193 static const struct inode_operations shmem_special_inode_operations;
194 static const struct vm_operations_struct shmem_vm_ops;
195 
196 static LIST_HEAD(shmem_swaplist);
197 static DEFINE_MUTEX(shmem_swaplist_mutex);
198 
199 static int shmem_reserve_inode(struct super_block *sb)
200 {
201         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
202         if (sbinfo->max_inodes) {
203                 spin_lock(&sbinfo->stat_lock);
204                 if (!sbinfo->free_inodes) {
205                         spin_unlock(&sbinfo->stat_lock);
206                         return -ENOSPC;
207                 }
208                 sbinfo->free_inodes--;
209                 spin_unlock(&sbinfo->stat_lock);
210         }
211         return 0;
212 }
213 
214 static void shmem_free_inode(struct super_block *sb)
215 {
216         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
217         if (sbinfo->max_inodes) {
218                 spin_lock(&sbinfo->stat_lock);
219                 sbinfo->free_inodes++;
220                 spin_unlock(&sbinfo->stat_lock);
221         }
222 }
223 
224 /**
225  * shmem_recalc_inode - recalculate the block usage of an inode
226  * @inode: inode to recalc
227  *
228  * We have to calculate the free blocks since the mm can drop
229  * undirtied hole pages behind our back.
230  *
231  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
232  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
233  *
234  * It has to be called with the spinlock held.
235  */
236 static void shmem_recalc_inode(struct inode *inode)
237 {
238         struct shmem_inode_info *info = SHMEM_I(inode);
239         long freed;
240 
241         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
242         if (freed > 0) {
243                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
244                 if (sbinfo->max_blocks)
245                         percpu_counter_add(&sbinfo->used_blocks, -freed);
246                 info->alloced -= freed;
247                 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
248                 shmem_unacct_blocks(info->flags, freed);
249         }
250 }
251 
252 /*
253  * Replace item expected in radix tree by a new item, while holding tree lock.
254  */
255 static int shmem_radix_tree_replace(struct address_space *mapping,
256                         pgoff_t index, void *expected, void *replacement)
257 {
258         void **pslot;
259         void *item;
260 
261         VM_BUG_ON(!expected);
262         VM_BUG_ON(!replacement);
263         pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
264         if (!pslot)
265                 return -ENOENT;
266         item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
267         if (item != expected)
268                 return -ENOENT;
269         radix_tree_replace_slot(pslot, replacement);
270         return 0;
271 }
272 
273 /*
274  * Sometimes, before we decide whether to proceed or to fail, we must check
275  * that an entry was not already brought back from swap by a racing thread.
276  *
277  * Checking page is not enough: by the time a SwapCache page is locked, it
278  * might be reused, and again be SwapCache, using the same swap as before.
279  */
280 static bool shmem_confirm_swap(struct address_space *mapping,
281                                pgoff_t index, swp_entry_t swap)
282 {
283         void *item;
284 
285         rcu_read_lock();
286         item = radix_tree_lookup(&mapping->page_tree, index);
287         rcu_read_unlock();
288         return item == swp_to_radix_entry(swap);
289 }
290 
291 /*
292  * Like add_to_page_cache_locked, but error if expected item has gone.
293  */
294 static int shmem_add_to_page_cache(struct page *page,
295                                    struct address_space *mapping,
296                                    pgoff_t index, void *expected)
297 {
298         int error;
299 
300         VM_BUG_ON_PAGE(!PageLocked(page), page);
301         VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
302 
303         page_cache_get(page);
304         page->mapping = mapping;
305         page->index = index;
306 
307         spin_lock_irq(&mapping->tree_lock);
308         if (!expected)
309                 error = radix_tree_insert(&mapping->page_tree, index, page);
310         else
311                 error = shmem_radix_tree_replace(mapping, index, expected,
312                                                                  page);
313         if (!error) {
314                 mapping->nrpages++;
315                 __inc_zone_page_state(page, NR_FILE_PAGES);
316                 __inc_zone_page_state(page, NR_SHMEM);
317                 spin_unlock_irq(&mapping->tree_lock);
318         } else {
319                 page->mapping = NULL;
320                 spin_unlock_irq(&mapping->tree_lock);
321                 page_cache_release(page);
322         }
323         return error;
324 }
325 
326 /*
327  * Like delete_from_page_cache, but substitutes swap for page.
328  */
329 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
330 {
331         struct address_space *mapping = page->mapping;
332         int error;
333 
334         spin_lock_irq(&mapping->tree_lock);
335         error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
336         page->mapping = NULL;
337         mapping->nrpages--;
338         __dec_zone_page_state(page, NR_FILE_PAGES);
339         __dec_zone_page_state(page, NR_SHMEM);
340         spin_unlock_irq(&mapping->tree_lock);
341         page_cache_release(page);
342         BUG_ON(error);
343 }
344 
345 /*
346  * Remove swap entry from radix tree, free the swap and its page cache.
347  */
348 static int shmem_free_swap(struct address_space *mapping,
349                            pgoff_t index, void *radswap)
350 {
351         void *old;
352 
353         spin_lock_irq(&mapping->tree_lock);
354         old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
355         spin_unlock_irq(&mapping->tree_lock);
356         if (old != radswap)
357                 return -ENOENT;
358         free_swap_and_cache(radix_to_swp_entry(radswap));
359         return 0;
360 }
361 
362 /*
363  * Determine (in bytes) how many of the shmem object's pages mapped by the
364  * given offsets are swapped out.
365  *
366  * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
367  * as long as the inode doesn't go away and racy results are not a problem.
368  */
369 unsigned long shmem_partial_swap_usage(struct address_space *mapping,
370                                                 pgoff_t start, pgoff_t end)
371 {
372         struct radix_tree_iter iter;
373         void **slot;
374         struct page *page;
375         unsigned long swapped = 0;
376 
377         rcu_read_lock();
378 
379 restart:
380         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
381                 if (iter.index >= end)
382                         break;
383 
384                 page = radix_tree_deref_slot(slot);
385 
386                 /*
387                  * This should only be possible to happen at index 0, so we
388                  * don't need to reset the counter, nor do we risk infinite
389                  * restarts.
390                  */
391                 if (radix_tree_deref_retry(page))
392                         goto restart;
393 
394                 if (radix_tree_exceptional_entry(page))
395                         swapped++;
396 
397                 if (need_resched()) {
398                         cond_resched_rcu();
399                         start = iter.index + 1;
400                         goto restart;
401                 }
402         }
403 
404         rcu_read_unlock();
405 
406         return swapped << PAGE_SHIFT;
407 }
408 
409 /*
410  * Determine (in bytes) how many of the shmem object's pages mapped by the
411  * given vma is swapped out.
412  *
413  * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
414  * as long as the inode doesn't go away and racy results are not a problem.
415  */
416 unsigned long shmem_swap_usage(struct vm_area_struct *vma)
417 {
418         struct inode *inode = file_inode(vma->vm_file);
419         struct shmem_inode_info *info = SHMEM_I(inode);
420         struct address_space *mapping = inode->i_mapping;
421         unsigned long swapped;
422 
423         /* Be careful as we don't hold info->lock */
424         swapped = READ_ONCE(info->swapped);
425 
426         /*
427          * The easier cases are when the shmem object has nothing in swap, or
428          * the vma maps it whole. Then we can simply use the stats that we
429          * already track.
430          */
431         if (!swapped)
432                 return 0;
433 
434         if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
435                 return swapped << PAGE_SHIFT;
436 
437         /* Here comes the more involved part */
438         return shmem_partial_swap_usage(mapping,
439                         linear_page_index(vma, vma->vm_start),
440                         linear_page_index(vma, vma->vm_end));
441 }
442 
443 /*
444  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
445  */
446 void shmem_unlock_mapping(struct address_space *mapping)
447 {
448         struct pagevec pvec;
449         pgoff_t indices[PAGEVEC_SIZE];
450         pgoff_t index = 0;
451 
452         pagevec_init(&pvec, 0);
453         /*
454          * Minor point, but we might as well stop if someone else SHM_LOCKs it.
455          */
456         while (!mapping_unevictable(mapping)) {
457                 /*
458                  * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
459                  * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
460                  */
461                 pvec.nr = find_get_entries(mapping, index,
462                                            PAGEVEC_SIZE, pvec.pages, indices);
463                 if (!pvec.nr)
464                         break;
465                 index = indices[pvec.nr - 1] + 1;
466                 pagevec_remove_exceptionals(&pvec);
467                 check_move_unevictable_pages(pvec.pages, pvec.nr);
468                 pagevec_release(&pvec);
469                 cond_resched();
470         }
471 }
472 
473 /*
474  * Remove range of pages and swap entries from radix tree, and free them.
475  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
476  */
477 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
478                                                                  bool unfalloc)
479 {
480         struct address_space *mapping = inode->i_mapping;
481         struct shmem_inode_info *info = SHMEM_I(inode);
482         pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
483         pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
484         unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
485         unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
486         struct pagevec pvec;
487         pgoff_t indices[PAGEVEC_SIZE];
488         long nr_swaps_freed = 0;
489         pgoff_t index;
490         int i;
491 
492         if (lend == -1)
493                 end = -1;       /* unsigned, so actually very big */
494 
495         pagevec_init(&pvec, 0);
496         index = start;
497         while (index < end) {
498                 pvec.nr = find_get_entries(mapping, index,
499                         min(end - index, (pgoff_t)PAGEVEC_SIZE),
500                         pvec.pages, indices);
501                 if (!pvec.nr)
502                         break;
503                 for (i = 0; i < pagevec_count(&pvec); i++) {
504                         struct page *page = pvec.pages[i];
505 
506                         index = indices[i];
507                         if (index >= end)
508                                 break;
509 
510                         if (radix_tree_exceptional_entry(page)) {
511                                 if (unfalloc)
512                                         continue;
513                                 nr_swaps_freed += !shmem_free_swap(mapping,
514                                                                 index, page);
515                                 continue;
516                         }
517 
518                         if (!trylock_page(page))
519                                 continue;
520                         if (!unfalloc || !PageUptodate(page)) {
521                                 if (page->mapping == mapping) {
522                                         VM_BUG_ON_PAGE(PageWriteback(page), page);
523                                         truncate_inode_page(mapping, page);
524                                 }
525                         }
526                         unlock_page(page);
527                 }
528                 pagevec_remove_exceptionals(&pvec);
529                 pagevec_release(&pvec);
530                 cond_resched();
531                 index++;
532         }
533 
534         if (partial_start) {
535                 struct page *page = NULL;
536                 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
537                 if (page) {
538                         unsigned int top = PAGE_CACHE_SIZE;
539                         if (start > end) {
540                                 top = partial_end;
541                                 partial_end = 0;
542                         }
543                         zero_user_segment(page, partial_start, top);
544                         set_page_dirty(page);
545                         unlock_page(page);
546                         page_cache_release(page);
547                 }
548         }
549         if (partial_end) {
550                 struct page *page = NULL;
551                 shmem_getpage(inode, end, &page, SGP_READ, NULL);
552                 if (page) {
553                         zero_user_segment(page, 0, partial_end);
554                         set_page_dirty(page);
555                         unlock_page(page);
556                         page_cache_release(page);
557                 }
558         }
559         if (start >= end)
560                 return;
561 
562         index = start;
563         while (index < end) {
564                 cond_resched();
565 
566                 pvec.nr = find_get_entries(mapping, index,
567                                 min(end - index, (pgoff_t)PAGEVEC_SIZE),
568                                 pvec.pages, indices);
569                 if (!pvec.nr) {
570                         /* If all gone or hole-punch or unfalloc, we're done */
571                         if (index == start || end != -1)
572                                 break;
573                         /* But if truncating, restart to make sure all gone */
574                         index = start;
575                         continue;
576                 }
577                 for (i = 0; i < pagevec_count(&pvec); i++) {
578                         struct page *page = pvec.pages[i];
579 
580                         index = indices[i];
581                         if (index >= end)
582                                 break;
583 
584                         if (radix_tree_exceptional_entry(page)) {
585                                 if (unfalloc)
586                                         continue;
587                                 if (shmem_free_swap(mapping, index, page)) {
588                                         /* Swap was replaced by page: retry */
589                                         index--;
590                                         break;
591                                 }
592                                 nr_swaps_freed++;
593                                 continue;
594                         }
595 
596                         lock_page(page);
597                         if (!unfalloc || !PageUptodate(page)) {
598                                 if (page->mapping == mapping) {
599                                         VM_BUG_ON_PAGE(PageWriteback(page), page);
600                                         truncate_inode_page(mapping, page);
601                                 } else {
602                                         /* Page was replaced by swap: retry */
603                                         unlock_page(page);
604                                         index--;
605                                         break;
606                                 }
607                         }
608                         unlock_page(page);
609                 }
610                 pagevec_remove_exceptionals(&pvec);
611                 pagevec_release(&pvec);
612                 index++;
613         }
614 
615         spin_lock(&info->lock);
616         info->swapped -= nr_swaps_freed;
617         shmem_recalc_inode(inode);
618         spin_unlock(&info->lock);
619 }
620 
621 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
622 {
623         shmem_undo_range(inode, lstart, lend, false);
624         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
625 }
626 EXPORT_SYMBOL_GPL(shmem_truncate_range);
627 
628 static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry,
629                          struct kstat *stat)
630 {
631         struct inode *inode = dentry->d_inode;
632         struct shmem_inode_info *info = SHMEM_I(inode);
633 
634         if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
635                 spin_lock(&info->lock);
636                 shmem_recalc_inode(inode);
637                 spin_unlock(&info->lock);
638         }
639         generic_fillattr(inode, stat);
640         return 0;
641 }
642 
643 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
644 {
645         struct inode *inode = d_inode(dentry);
646         struct shmem_inode_info *info = SHMEM_I(inode);
647         int error;
648 
649         error = inode_change_ok(inode, attr);
650         if (error)
651                 return error;
652 
653         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
654                 loff_t oldsize = inode->i_size;
655                 loff_t newsize = attr->ia_size;
656 
657                 /* protected by i_mutex */
658                 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
659                     (newsize > oldsize && (info->seals & F_SEAL_GROW)))
660                         return -EPERM;
661 
662                 if (newsize != oldsize) {
663                         error = shmem_reacct_size(SHMEM_I(inode)->flags,
664                                         oldsize, newsize);
665                         if (error)
666                                 return error;
667                         i_size_write(inode, newsize);
668                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
669                 }
670                 if (newsize <= oldsize) {
671                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
672                         if (oldsize > holebegin)
673                                 unmap_mapping_range(inode->i_mapping,
674                                                         holebegin, 0, 1);
675                         if (info->alloced)
676                                 shmem_truncate_range(inode,
677                                                         newsize, (loff_t)-1);
678                         /* unmap again to remove racily COWed private pages */
679                         if (oldsize > holebegin)
680                                 unmap_mapping_range(inode->i_mapping,
681                                                         holebegin, 0, 1);
682                 }
683         }
684 
685         setattr_copy(inode, attr);
686         if (attr->ia_valid & ATTR_MODE)
687                 error = posix_acl_chmod(inode, inode->i_mode);
688         return error;
689 }
690 
691 static void shmem_evict_inode(struct inode *inode)
692 {
693         struct shmem_inode_info *info = SHMEM_I(inode);
694 
695         if (inode->i_mapping->a_ops == &shmem_aops) {
696                 shmem_unacct_size(info->flags, inode->i_size);
697                 inode->i_size = 0;
698                 shmem_truncate_range(inode, 0, (loff_t)-1);
699                 if (!list_empty(&info->swaplist)) {
700                         mutex_lock(&shmem_swaplist_mutex);
701                         list_del_init(&info->swaplist);
702                         mutex_unlock(&shmem_swaplist_mutex);
703                 }
704         }
705 
706         simple_xattrs_free(&info->xattrs);
707         WARN_ON(inode->i_blocks);
708         shmem_free_inode(inode->i_sb);
709         clear_inode(inode);
710 }
711 
712 /*
713  * If swap found in inode, free it and move page from swapcache to filecache.
714  */
715 static int shmem_unuse_inode(struct shmem_inode_info *info,
716                              swp_entry_t swap, struct page **pagep)
717 {
718         struct address_space *mapping = info->vfs_inode.i_mapping;
719         void *radswap;
720         pgoff_t index;
721         gfp_t gfp;
722         int error = 0;
723 
724         radswap = swp_to_radix_entry(swap);
725         index = radix_tree_locate_item(&mapping->page_tree, radswap);
726         if (index == -1)
727                 return -EAGAIN; /* tell shmem_unuse we found nothing */
728 
729         /*
730          * Move _head_ to start search for next from here.
731          * But be careful: shmem_evict_inode checks list_empty without taking
732          * mutex, and there's an instant in list_move_tail when info->swaplist
733          * would appear empty, if it were the only one on shmem_swaplist.
734          */
735         if (shmem_swaplist.next != &info->swaplist)
736                 list_move_tail(&shmem_swaplist, &info->swaplist);
737 
738         gfp = mapping_gfp_mask(mapping);
739         if (shmem_should_replace_page(*pagep, gfp)) {
740                 mutex_unlock(&shmem_swaplist_mutex);
741                 error = shmem_replace_page(pagep, gfp, info, index);
742                 mutex_lock(&shmem_swaplist_mutex);
743                 /*
744                  * We needed to drop mutex to make that restrictive page
745                  * allocation, but the inode might have been freed while we
746                  * dropped it: although a racing shmem_evict_inode() cannot
747                  * complete without emptying the radix_tree, our page lock
748                  * on this swapcache page is not enough to prevent that -
749                  * free_swap_and_cache() of our swap entry will only
750                  * trylock_page(), removing swap from radix_tree whatever.
751                  *
752                  * We must not proceed to shmem_add_to_page_cache() if the
753                  * inode has been freed, but of course we cannot rely on
754                  * inode or mapping or info to check that.  However, we can
755                  * safely check if our swap entry is still in use (and here
756                  * it can't have got reused for another page): if it's still
757                  * in use, then the inode cannot have been freed yet, and we
758                  * can safely proceed (if it's no longer in use, that tells
759                  * nothing about the inode, but we don't need to unuse swap).
760                  */
761                 if (!page_swapcount(*pagep))
762                         error = -ENOENT;
763         }
764 
765         /*
766          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
767          * but also to hold up shmem_evict_inode(): so inode cannot be freed
768          * beneath us (pagelock doesn't help until the page is in pagecache).
769          */
770         if (!error)
771                 error = shmem_add_to_page_cache(*pagep, mapping, index,
772                                                 radswap);
773         if (error != -ENOMEM) {
774                 /*
775                  * Truncation and eviction use free_swap_and_cache(), which
776                  * only does trylock page: if we raced, best clean up here.
777                  */
778                 delete_from_swap_cache(*pagep);
779                 set_page_dirty(*pagep);
780                 if (!error) {
781                         spin_lock(&info->lock);
782                         info->swapped--;
783                         spin_unlock(&info->lock);
784                         swap_free(swap);
785                 }
786         }
787         return error;
788 }
789 
790 /*
791  * Search through swapped inodes to find and replace swap by page.
792  */
793 int shmem_unuse(swp_entry_t swap, struct page *page)
794 {
795         struct list_head *this, *next;
796         struct shmem_inode_info *info;
797         struct mem_cgroup *memcg;
798         int error = 0;
799 
800         /*
801          * There's a faint possibility that swap page was replaced before
802          * caller locked it: caller will come back later with the right page.
803          */
804         if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
805                 goto out;
806 
807         /*
808          * Charge page using GFP_KERNEL while we can wait, before taking
809          * the shmem_swaplist_mutex which might hold up shmem_writepage().
810          * Charged back to the user (not to caller) when swap account is used.
811          */
812         error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
813                         false);
814         if (error)
815                 goto out;
816         /* No radix_tree_preload: swap entry keeps a place for page in tree */
817         error = -EAGAIN;
818 
819         mutex_lock(&shmem_swaplist_mutex);
820         list_for_each_safe(this, next, &shmem_swaplist) {
821                 info = list_entry(this, struct shmem_inode_info, swaplist);
822                 if (info->swapped)
823                         error = shmem_unuse_inode(info, swap, &page);
824                 else
825                         list_del_init(&info->swaplist);
826                 cond_resched();
827                 if (error != -EAGAIN)
828                         break;
829                 /* found nothing in this: move on to search the next */
830         }
831         mutex_unlock(&shmem_swaplist_mutex);
832 
833         if (error) {
834                 if (error != -ENOMEM)
835                         error = 0;
836                 mem_cgroup_cancel_charge(page, memcg, false);
837         } else
838                 mem_cgroup_commit_charge(page, memcg, true, false);
839 out:
840         unlock_page(page);
841         page_cache_release(page);
842         return error;
843 }
844 
845 /*
846  * Move the page from the page cache to the swap cache.
847  */
848 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
849 {
850         struct shmem_inode_info *info;
851         struct address_space *mapping;
852         struct inode *inode;
853         swp_entry_t swap;
854         pgoff_t index;
855 
856         BUG_ON(!PageLocked(page));
857         mapping = page->mapping;
858         index = page->index;
859         inode = mapping->host;
860         info = SHMEM_I(inode);
861         if (info->flags & VM_LOCKED)
862                 goto redirty;
863         if (!total_swap_pages)
864                 goto redirty;
865 
866         /*
867          * Our capabilities prevent regular writeback or sync from ever calling
868          * shmem_writepage; but a stacking filesystem might use ->writepage of
869          * its underlying filesystem, in which case tmpfs should write out to
870          * swap only in response to memory pressure, and not for the writeback
871          * threads or sync.
872          */
873         if (!wbc->for_reclaim) {
874                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
875                 goto redirty;
876         }
877 
878         /*
879          * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
880          * value into swapfile.c, the only way we can correctly account for a
881          * fallocated page arriving here is now to initialize it and write it.
882          *
883          * That's okay for a page already fallocated earlier, but if we have
884          * not yet completed the fallocation, then (a) we want to keep track
885          * of this page in case we have to undo it, and (b) it may not be a
886          * good idea to continue anyway, once we're pushing into swap.  So
887          * reactivate the page, and let shmem_fallocate() quit when too many.
888          */
889         if (!PageUptodate(page)) {
890                 if (inode->i_private) {
891                         struct shmem_falloc *shmem_falloc;
892                         spin_lock(&inode->i_lock);
893                         shmem_falloc = inode->i_private;
894                         if (shmem_falloc &&
895                             !shmem_falloc->waitq &&
896                             index >= shmem_falloc->start &&
897                             index < shmem_falloc->next)
898                                 shmem_falloc->nr_unswapped++;
899                         else
900                                 shmem_falloc = NULL;
901                         spin_unlock(&inode->i_lock);
902                         if (shmem_falloc)
903                                 goto redirty;
904                 }
905                 clear_highpage(page);
906                 flush_dcache_page(page);
907                 SetPageUptodate(page);
908         }
909 
910         swap = get_swap_page();
911         if (!swap.val)
912                 goto redirty;
913 
914         if (mem_cgroup_try_charge_swap(page, swap))
915                 goto free_swap;
916 
917         /*
918          * Add inode to shmem_unuse()'s list of swapped-out inodes,
919          * if it's not already there.  Do it now before the page is
920          * moved to swap cache, when its pagelock no longer protects
921          * the inode from eviction.  But don't unlock the mutex until
922          * we've incremented swapped, because shmem_unuse_inode() will
923          * prune a !swapped inode from the swaplist under this mutex.
924          */
925         mutex_lock(&shmem_swaplist_mutex);
926         if (list_empty(&info->swaplist))
927                 list_add_tail(&info->swaplist, &shmem_swaplist);
928 
929         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
930                 spin_lock(&info->lock);
931                 shmem_recalc_inode(inode);
932                 info->swapped++;
933                 spin_unlock(&info->lock);
934 
935                 swap_shmem_alloc(swap);
936                 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
937 
938                 mutex_unlock(&shmem_swaplist_mutex);
939                 BUG_ON(page_mapped(page));
940                 swap_writepage(page, wbc);
941                 return 0;
942         }
943 
944         mutex_unlock(&shmem_swaplist_mutex);
945 free_swap:
946         swapcache_free(swap);
947 redirty:
948         set_page_dirty(page);
949         if (wbc->for_reclaim)
950                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
951         unlock_page(page);
952         return 0;
953 }
954 
955 #ifdef CONFIG_NUMA
956 #ifdef CONFIG_TMPFS
957 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
958 {
959         char buffer[64];
960 
961         if (!mpol || mpol->mode == MPOL_DEFAULT)
962                 return;         /* show nothing */
963 
964         mpol_to_str(buffer, sizeof(buffer), mpol);
965 
966         seq_printf(seq, ",mpol=%s", buffer);
967 }
968 
969 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
970 {
971         struct mempolicy *mpol = NULL;
972         if (sbinfo->mpol) {
973                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
974                 mpol = sbinfo->mpol;
975                 mpol_get(mpol);
976                 spin_unlock(&sbinfo->stat_lock);
977         }
978         return mpol;
979 }
980 #endif /* CONFIG_TMPFS */
981 
982 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
983                         struct shmem_inode_info *info, pgoff_t index)
984 {
985         struct vm_area_struct pvma;
986         struct page *page;
987 
988         /* Create a pseudo vma that just contains the policy */
989         pvma.vm_start = 0;
990         /* Bias interleave by inode number to distribute better across nodes */
991         pvma.vm_pgoff = index + info->vfs_inode.i_ino;
992         pvma.vm_ops = NULL;
993         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
994 
995         page = swapin_readahead(swap, gfp, &pvma, 0);
996 
997         /* Drop reference taken by mpol_shared_policy_lookup() */
998         mpol_cond_put(pvma.vm_policy);
999 
1000         return page;
1001 }
1002 
1003 static struct page *shmem_alloc_page(gfp_t gfp,
1004                         struct shmem_inode_info *info, pgoff_t index)
1005 {
1006         struct vm_area_struct pvma;
1007         struct page *page;
1008 
1009         /* Create a pseudo vma that just contains the policy */
1010         pvma.vm_start = 0;
1011         /* Bias interleave by inode number to distribute better across nodes */
1012         pvma.vm_pgoff = index + info->vfs_inode.i_ino;
1013         pvma.vm_ops = NULL;
1014         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1015 
1016         page = alloc_page_vma(gfp, &pvma, 0);
1017 
1018         /* Drop reference taken by mpol_shared_policy_lookup() */
1019         mpol_cond_put(pvma.vm_policy);
1020 
1021         return page;
1022 }
1023 #else /* !CONFIG_NUMA */
1024 #ifdef CONFIG_TMPFS
1025 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1026 {
1027 }
1028 #endif /* CONFIG_TMPFS */
1029 
1030 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1031                         struct shmem_inode_info *info, pgoff_t index)
1032 {
1033         return swapin_readahead(swap, gfp, NULL, 0);
1034 }
1035 
1036 static inline struct page *shmem_alloc_page(gfp_t gfp,
1037                         struct shmem_inode_info *info, pgoff_t index)
1038 {
1039         return alloc_page(gfp);
1040 }
1041 #endif /* CONFIG_NUMA */
1042 
1043 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1044 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1045 {
1046         return NULL;
1047 }
1048 #endif
1049 
1050 /*
1051  * When a page is moved from swapcache to shmem filecache (either by the
1052  * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1053  * shmem_unuse_inode()), it may have been read in earlier from swap, in
1054  * ignorance of the mapping it belongs to.  If that mapping has special
1055  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1056  * we may need to copy to a suitable page before moving to filecache.
1057  *
1058  * In a future release, this may well be extended to respect cpuset and
1059  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1060  * but for now it is a simple matter of zone.
1061  */
1062 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1063 {
1064         return page_zonenum(page) > gfp_zone(gfp);
1065 }
1066 
1067 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1068                                 struct shmem_inode_info *info, pgoff_t index)
1069 {
1070         struct page *oldpage, *newpage;
1071         struct address_space *swap_mapping;
1072         pgoff_t swap_index;
1073         int error;
1074 
1075         oldpage = *pagep;
1076         swap_index = page_private(oldpage);
1077         swap_mapping = page_mapping(oldpage);
1078 
1079         /*
1080          * We have arrived here because our zones are constrained, so don't
1081          * limit chance of success by further cpuset and node constraints.
1082          */
1083         gfp &= ~GFP_CONSTRAINT_MASK;
1084         newpage = shmem_alloc_page(gfp, info, index);
1085         if (!newpage)
1086                 return -ENOMEM;
1087 
1088         page_cache_get(newpage);
1089         copy_highpage(newpage, oldpage);
1090         flush_dcache_page(newpage);
1091 
1092         __SetPageLocked(newpage);
1093         SetPageUptodate(newpage);
1094         SetPageSwapBacked(newpage);
1095         set_page_private(newpage, swap_index);
1096         SetPageSwapCache(newpage);
1097 
1098         /*
1099          * Our caller will very soon move newpage out of swapcache, but it's
1100          * a nice clean interface for us to replace oldpage by newpage there.
1101          */
1102         spin_lock_irq(&swap_mapping->tree_lock);
1103         error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1104                                                                    newpage);
1105         if (!error) {
1106                 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1107                 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1108         }
1109         spin_unlock_irq(&swap_mapping->tree_lock);
1110 
1111         if (unlikely(error)) {
1112                 /*
1113                  * Is this possible?  I think not, now that our callers check
1114                  * both PageSwapCache and page_private after getting page lock;
1115                  * but be defensive.  Reverse old to newpage for clear and free.
1116                  */
1117                 oldpage = newpage;
1118         } else {
1119                 mem_cgroup_replace_page(oldpage, newpage);
1120                 lru_cache_add_anon(newpage);
1121                 *pagep = newpage;
1122         }
1123 
1124         ClearPageSwapCache(oldpage);
1125         set_page_private(oldpage, 0);
1126 
1127         unlock_page(oldpage);
1128         page_cache_release(oldpage);
1129         page_cache_release(oldpage);
1130         return error;
1131 }
1132 
1133 /*
1134  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1135  *
1136  * If we allocate a new one we do not mark it dirty. That's up to the
1137  * vm. If we swap it in we mark it dirty since we also free the swap
1138  * entry since a page cannot live in both the swap and page cache
1139  */
1140 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1141         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1142 {
1143         struct address_space *mapping = inode->i_mapping;
1144         struct shmem_inode_info *info;
1145         struct shmem_sb_info *sbinfo;
1146         struct mem_cgroup *memcg;
1147         struct page *page;
1148         swp_entry_t swap;
1149         int error;
1150         int once = 0;
1151         int alloced = 0;
1152 
1153         if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1154                 return -EFBIG;
1155 repeat:
1156         swap.val = 0;
1157         page = find_lock_entry(mapping, index);
1158         if (radix_tree_exceptional_entry(page)) {
1159                 swap = radix_to_swp_entry(page);
1160                 page = NULL;
1161         }
1162 
1163         if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1164             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1165                 error = -EINVAL;
1166                 goto unlock;
1167         }
1168 
1169         if (page && sgp == SGP_WRITE)
1170                 mark_page_accessed(page);
1171 
1172         /* fallocated page? */
1173         if (page && !PageUptodate(page)) {
1174                 if (sgp != SGP_READ)
1175                         goto clear;
1176                 unlock_page(page);
1177                 page_cache_release(page);
1178                 page = NULL;
1179         }
1180         if (page || (sgp == SGP_READ && !swap.val)) {
1181                 *pagep = page;
1182                 return 0;
1183         }
1184 
1185         /*
1186          * Fast cache lookup did not find it:
1187          * bring it back from swap or allocate.
1188          */
1189         info = SHMEM_I(inode);
1190         sbinfo = SHMEM_SB(inode->i_sb);
1191 
1192         if (swap.val) {
1193                 /* Look it up and read it in.. */
1194                 page = lookup_swap_cache(swap);
1195                 if (!page) {
1196                         /* here we actually do the io */
1197                         if (fault_type)
1198                                 *fault_type |= VM_FAULT_MAJOR;
1199                         page = shmem_swapin(swap, gfp, info, index);
1200                         if (!page) {
1201                                 error = -ENOMEM;
1202                                 goto failed;
1203                         }
1204                 }
1205 
1206                 /* We have to do this with page locked to prevent races */
1207                 lock_page(page);
1208                 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1209                     !shmem_confirm_swap(mapping, index, swap)) {
1210                         error = -EEXIST;        /* try again */
1211                         goto unlock;
1212                 }
1213                 if (!PageUptodate(page)) {
1214                         error = -EIO;
1215                         goto failed;
1216                 }
1217                 wait_on_page_writeback(page);
1218 
1219                 if (shmem_should_replace_page(page, gfp)) {
1220                         error = shmem_replace_page(&page, gfp, info, index);
1221                         if (error)
1222                                 goto failed;
1223                 }
1224 
1225                 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
1226                                 false);
1227                 if (!error) {
1228                         error = shmem_add_to_page_cache(page, mapping, index,
1229                                                 swp_to_radix_entry(swap));
1230                         /*
1231                          * We already confirmed swap under page lock, and make
1232                          * no memory allocation here, so usually no possibility
1233                          * of error; but free_swap_and_cache() only trylocks a
1234                          * page, so it is just possible that the entry has been
1235                          * truncated or holepunched since swap was confirmed.
1236                          * shmem_undo_range() will have done some of the
1237                          * unaccounting, now delete_from_swap_cache() will do
1238                          * the rest.
1239                          * Reset swap.val? No, leave it so "failed" goes back to
1240                          * "repeat": reading a hole and writing should succeed.
1241                          */
1242                         if (error) {
1243                                 mem_cgroup_cancel_charge(page, memcg, false);
1244                                 delete_from_swap_cache(page);
1245                         }
1246                 }
1247                 if (error)
1248                         goto failed;
1249 
1250                 mem_cgroup_commit_charge(page, memcg, true, false);
1251 
1252                 spin_lock(&info->lock);
1253                 info->swapped--;
1254                 shmem_recalc_inode(inode);
1255                 spin_unlock(&info->lock);
1256 
1257                 if (sgp == SGP_WRITE)
1258                         mark_page_accessed(page);
1259 
1260                 delete_from_swap_cache(page);
1261                 set_page_dirty(page);
1262                 swap_free(swap);
1263 
1264         } else {
1265                 if (shmem_acct_block(info->flags)) {
1266                         error = -ENOSPC;
1267                         goto failed;
1268                 }
1269                 if (sbinfo->max_blocks) {
1270                         if (percpu_counter_compare(&sbinfo->used_blocks,
1271                                                 sbinfo->max_blocks) >= 0) {
1272                                 error = -ENOSPC;
1273                                 goto unacct;
1274                         }
1275                         percpu_counter_inc(&sbinfo->used_blocks);
1276                 }
1277 
1278                 page = shmem_alloc_page(gfp, info, index);
1279                 if (!page) {
1280                         error = -ENOMEM;
1281                         goto decused;
1282                 }
1283 
1284                 __SetPageSwapBacked(page);
1285                 __SetPageLocked(page);
1286                 if (sgp == SGP_WRITE)
1287                         __SetPageReferenced(page);
1288 
1289                 error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg,
1290                                 false);
1291                 if (error)
1292                         goto decused;
1293                 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1294                 if (!error) {
1295                         error = shmem_add_to_page_cache(page, mapping, index,
1296                                                         NULL);
1297                         radix_tree_preload_end();
1298                 }
1299                 if (error) {
1300                         mem_cgroup_cancel_charge(page, memcg, false);
1301                         goto decused;
1302                 }
1303                 mem_cgroup_commit_charge(page, memcg, false, false);
1304                 lru_cache_add_anon(page);
1305 
1306                 spin_lock(&info->lock);
1307                 info->alloced++;
1308                 inode->i_blocks += BLOCKS_PER_PAGE;
1309                 shmem_recalc_inode(inode);
1310                 spin_unlock(&info->lock);
1311                 alloced = true;
1312 
1313                 /*
1314                  * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1315                  */
1316                 if (sgp == SGP_FALLOC)
1317                         sgp = SGP_WRITE;
1318 clear:
1319                 /*
1320                  * Let SGP_WRITE caller clear ends if write does not fill page;
1321                  * but SGP_FALLOC on a page fallocated earlier must initialize
1322                  * it now, lest undo on failure cancel our earlier guarantee.
1323                  */
1324                 if (sgp != SGP_WRITE) {
1325                         clear_highpage(page);
1326                         flush_dcache_page(page);
1327                         SetPageUptodate(page);
1328                 }
1329                 if (sgp == SGP_DIRTY)
1330                         set_page_dirty(page);
1331         }
1332 
1333         /* Perhaps the file has been truncated since we checked */
1334         if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1335             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1336                 if (alloced) {
1337                         ClearPageDirty(page);
1338                         delete_from_page_cache(page);
1339                         spin_lock(&info->lock);
1340                         shmem_recalc_inode(inode);
1341                         spin_unlock(&info->lock);
1342                 }
1343                 error = -EINVAL;
1344                 goto unlock;
1345         }
1346         *pagep = page;
1347         return 0;
1348 
1349         /*
1350          * Error recovery.
1351          */
1352 decused:
1353         if (sbinfo->max_blocks)
1354                 percpu_counter_add(&sbinfo->used_blocks, -1);
1355 unacct:
1356         shmem_unacct_blocks(info->flags, 1);
1357 failed:
1358         if (swap.val && !shmem_confirm_swap(mapping, index, swap))
1359                 error = -EEXIST;
1360 unlock:
1361         if (page) {
1362                 unlock_page(page);
1363                 page_cache_release(page);
1364         }
1365         if (error == -ENOSPC && !once++) {
1366                 info = SHMEM_I(inode);
1367                 spin_lock(&info->lock);
1368                 shmem_recalc_inode(inode);
1369                 spin_unlock(&info->lock);
1370                 goto repeat;
1371         }
1372         if (error == -EEXIST)   /* from above or from radix_tree_insert */
1373                 goto repeat;
1374         return error;
1375 }
1376 
1377 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1378 {
1379         struct inode *inode = file_inode(vma->vm_file);
1380         int error;
1381         int ret = VM_FAULT_LOCKED;
1382 
1383         /*
1384          * Trinity finds that probing a hole which tmpfs is punching can
1385          * prevent the hole-punch from ever completing: which in turn
1386          * locks writers out with its hold on i_mutex.  So refrain from
1387          * faulting pages into the hole while it's being punched.  Although
1388          * shmem_undo_range() does remove the additions, it may be unable to
1389          * keep up, as each new page needs its own unmap_mapping_range() call,
1390          * and the i_mmap tree grows ever slower to scan if new vmas are added.
1391          *
1392          * It does not matter if we sometimes reach this check just before the
1393          * hole-punch begins, so that one fault then races with the punch:
1394          * we just need to make racing faults a rare case.
1395          *
1396          * The implementation below would be much simpler if we just used a
1397          * standard mutex or completion: but we cannot take i_mutex in fault,
1398          * and bloating every shmem inode for this unlikely case would be sad.
1399          */
1400         if (unlikely(inode->i_private)) {
1401                 struct shmem_falloc *shmem_falloc;
1402 
1403                 spin_lock(&inode->i_lock);
1404                 shmem_falloc = inode->i_private;
1405                 if (shmem_falloc &&
1406                     shmem_falloc->waitq &&
1407                     vmf->pgoff >= shmem_falloc->start &&
1408                     vmf->pgoff < shmem_falloc->next) {
1409                         wait_queue_head_t *shmem_falloc_waitq;
1410                         DEFINE_WAIT(shmem_fault_wait);
1411 
1412                         ret = VM_FAULT_NOPAGE;
1413                         if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1414                            !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1415                                 /* It's polite to up mmap_sem if we can */
1416                                 up_read(&vma->vm_mm->mmap_sem);
1417                                 ret = VM_FAULT_RETRY;
1418                         }
1419 
1420                         shmem_falloc_waitq = shmem_falloc->waitq;
1421                         prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1422                                         TASK_UNINTERRUPTIBLE);
1423                         spin_unlock(&inode->i_lock);
1424                         schedule();
1425 
1426                         /*
1427                          * shmem_falloc_waitq points into the shmem_fallocate()
1428                          * stack of the hole-punching task: shmem_falloc_waitq
1429                          * is usually invalid by the time we reach here, but
1430                          * finish_wait() does not dereference it in that case;
1431                          * though i_lock needed lest racing with wake_up_all().
1432                          */
1433                         spin_lock(&inode->i_lock);
1434                         finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1435                         spin_unlock(&inode->i_lock);
1436                         return ret;
1437                 }
1438                 spin_unlock(&inode->i_lock);
1439         }
1440 
1441         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1442         if (error)
1443                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1444 
1445         if (ret & VM_FAULT_MAJOR) {
1446                 count_vm_event(PGMAJFAULT);
1447                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1448         }
1449         return ret;
1450 }
1451 
1452 #ifdef CONFIG_NUMA
1453 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1454 {
1455         struct inode *inode = file_inode(vma->vm_file);
1456         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1457 }
1458 
1459 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1460                                           unsigned long addr)
1461 {
1462         struct inode *inode = file_inode(vma->vm_file);
1463         pgoff_t index;
1464 
1465         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1466         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1467 }
1468 #endif
1469 
1470 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1471 {
1472         struct inode *inode = file_inode(file);
1473         struct shmem_inode_info *info = SHMEM_I(inode);
1474         int retval = -ENOMEM;
1475 
1476         spin_lock(&info->lock);
1477         if (lock && !(info->flags & VM_LOCKED)) {
1478                 if (!user_shm_lock(inode->i_size, user))
1479                         goto out_nomem;
1480                 info->flags |= VM_LOCKED;
1481                 mapping_set_unevictable(file->f_mapping);
1482         }
1483         if (!lock && (info->flags & VM_LOCKED) && user) {
1484                 user_shm_unlock(inode->i_size, user);
1485                 info->flags &= ~VM_LOCKED;
1486                 mapping_clear_unevictable(file->f_mapping);
1487         }
1488         retval = 0;
1489 
1490 out_nomem:
1491         spin_unlock(&info->lock);
1492         return retval;
1493 }
1494 
1495 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1496 {
1497         file_accessed(file);
1498         vma->vm_ops = &shmem_vm_ops;
1499         return 0;
1500 }
1501 
1502 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1503                                      umode_t mode, dev_t dev, unsigned long flags)
1504 {
1505         struct inode *inode;
1506         struct shmem_inode_info *info;
1507         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1508 
1509         if (shmem_reserve_inode(sb))
1510                 return NULL;
1511 
1512         inode = new_inode(sb);
1513         if (inode) {
1514                 inode->i_ino = get_next_ino();
1515                 inode_init_owner(inode, dir, mode);
1516                 inode->i_blocks = 0;
1517                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1518                 inode->i_generation = get_seconds();
1519                 info = SHMEM_I(inode);
1520                 memset(info, 0, (char *)inode - (char *)info);
1521                 spin_lock_init(&info->lock);
1522                 info->seals = F_SEAL_SEAL;
1523                 info->flags = flags & VM_NORESERVE;
1524                 INIT_LIST_HEAD(&info->swaplist);
1525                 simple_xattrs_init(&info->xattrs);
1526                 cache_no_acl(inode);
1527 
1528                 switch (mode & S_IFMT) {
1529                 default:
1530                         inode->i_op = &shmem_special_inode_operations;
1531                         init_special_inode(inode, mode, dev);
1532                         break;
1533                 case S_IFREG:
1534                         inode->i_mapping->a_ops = &shmem_aops;
1535                         inode->i_op = &shmem_inode_operations;
1536                         inode->i_fop = &shmem_file_operations;
1537                         mpol_shared_policy_init(&info->policy,
1538                                                  shmem_get_sbmpol(sbinfo));
1539                         break;
1540                 case S_IFDIR:
1541                         inc_nlink(inode);
1542                         /* Some things misbehave if size == 0 on a directory */
1543                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1544                         inode->i_op = &shmem_dir_inode_operations;
1545                         inode->i_fop = &simple_dir_operations;
1546                         break;
1547                 case S_IFLNK:
1548                         /*
1549                          * Must not load anything in the rbtree,
1550                          * mpol_free_shared_policy will not be called.
1551                          */
1552                         mpol_shared_policy_init(&info->policy, NULL);
1553                         break;
1554                 }
1555         } else
1556                 shmem_free_inode(sb);
1557         return inode;
1558 }
1559 
1560 bool shmem_mapping(struct address_space *mapping)
1561 {
1562         if (!mapping->host)
1563                 return false;
1564 
1565         return mapping->host->i_sb->s_op == &shmem_ops;
1566 }
1567 
1568 #ifdef CONFIG_TMPFS
1569 static const struct inode_operations shmem_symlink_inode_operations;
1570 static const struct inode_operations shmem_short_symlink_operations;
1571 
1572 #ifdef CONFIG_TMPFS_XATTR
1573 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1574 #else
1575 #define shmem_initxattrs NULL
1576 #endif
1577 
1578 static int
1579 shmem_write_begin(struct file *file, struct address_space *mapping,
1580                         loff_t pos, unsigned len, unsigned flags,
1581                         struct page **pagep, void **fsdata)
1582 {
1583         struct inode *inode = mapping->host;
1584         struct shmem_inode_info *info = SHMEM_I(inode);
1585         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1586 
1587         /* i_mutex is held by caller */
1588         if (unlikely(info->seals)) {
1589                 if (info->seals & F_SEAL_WRITE)
1590                         return -EPERM;
1591                 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
1592                         return -EPERM;
1593         }
1594 
1595         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1596 }
1597 
1598 static int
1599 shmem_write_end(struct file *file, struct address_space *mapping,
1600                         loff_t pos, unsigned len, unsigned copied,
1601                         struct page *page, void *fsdata)
1602 {
1603         struct inode *inode = mapping->host;
1604 
1605         if (pos + copied > inode->i_size)
1606                 i_size_write(inode, pos + copied);
1607 
1608         if (!PageUptodate(page)) {
1609                 if (copied < PAGE_CACHE_SIZE) {
1610                         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1611                         zero_user_segments(page, 0, from,
1612                                         from + copied, PAGE_CACHE_SIZE);
1613                 }
1614                 SetPageUptodate(page);
1615         }
1616         set_page_dirty(page);
1617         unlock_page(page);
1618         page_cache_release(page);
1619 
1620         return copied;
1621 }
1622 
1623 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1624 {
1625         struct file *file = iocb->ki_filp;
1626         struct inode *inode = file_inode(file);
1627         struct address_space *mapping = inode->i_mapping;
1628         pgoff_t index;
1629         unsigned long offset;
1630         enum sgp_type sgp = SGP_READ;
1631         int error = 0;
1632         ssize_t retval = 0;
1633         loff_t *ppos = &iocb->ki_pos;
1634 
1635         /*
1636          * Might this read be for a stacking filesystem?  Then when reading
1637          * holes of a sparse file, we actually need to allocate those pages,
1638          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1639          */
1640         if (!iter_is_iovec(to))
1641                 sgp = SGP_DIRTY;
1642 
1643         index = *ppos >> PAGE_CACHE_SHIFT;
1644         offset = *ppos & ~PAGE_CACHE_MASK;
1645 
1646         for (;;) {
1647                 struct page *page = NULL;
1648                 pgoff_t end_index;
1649                 unsigned long nr, ret;
1650                 loff_t i_size = i_size_read(inode);
1651 
1652                 end_index = i_size >> PAGE_CACHE_SHIFT;
1653                 if (index > end_index)
1654                         break;
1655                 if (index == end_index) {
1656                         nr = i_size & ~PAGE_CACHE_MASK;
1657                         if (nr <= offset)
1658                                 break;
1659                 }
1660 
1661                 error = shmem_getpage(inode, index, &page, sgp, NULL);
1662                 if (error) {
1663                         if (error == -EINVAL)
1664                                 error = 0;
1665                         break;
1666                 }
1667                 if (page)
1668                         unlock_page(page);
1669 
1670                 /*
1671                  * We must evaluate after, since reads (unlike writes)
1672                  * are called without i_mutex protection against truncate
1673                  */
1674                 nr = PAGE_CACHE_SIZE;
1675                 i_size = i_size_read(inode);
1676                 end_index = i_size >> PAGE_CACHE_SHIFT;
1677                 if (index == end_index) {
1678                         nr = i_size & ~PAGE_CACHE_MASK;
1679                         if (nr <= offset) {
1680                                 if (page)
1681                                         page_cache_release(page);
1682                                 break;
1683                         }
1684                 }
1685                 nr -= offset;
1686 
1687                 if (page) {
1688                         /*
1689                          * If users can be writing to this page using arbitrary
1690                          * virtual addresses, take care about potential aliasing
1691                          * before reading the page on the kernel side.
1692                          */
1693                         if (mapping_writably_mapped(mapping))
1694                                 flush_dcache_page(page);
1695                         /*
1696                          * Mark the page accessed if we read the beginning.
1697                          */
1698                         if (!offset)
1699                                 mark_page_accessed(page);
1700                 } else {
1701                         page = ZERO_PAGE(0);
1702                         page_cache_get(page);
1703                 }
1704 
1705                 /*
1706                  * Ok, we have the page, and it's up-to-date, so
1707                  * now we can copy it to user space...
1708                  */
1709                 ret = copy_page_to_iter(page, offset, nr, to);
1710                 retval += ret;
1711                 offset += ret;
1712                 index += offset >> PAGE_CACHE_SHIFT;
1713                 offset &= ~PAGE_CACHE_MASK;
1714 
1715                 page_cache_release(page);
1716                 if (!iov_iter_count(to))
1717                         break;
1718                 if (ret < nr) {
1719                         error = -EFAULT;
1720                         break;
1721                 }
1722                 cond_resched();
1723         }
1724 
1725         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1726         file_accessed(file);
1727         return retval ? retval : error;
1728 }
1729 
1730 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1731                                 struct pipe_inode_info *pipe, size_t len,
1732                                 unsigned int flags)
1733 {
1734         struct address_space *mapping = in->f_mapping;
1735         struct inode *inode = mapping->host;
1736         unsigned int loff, nr_pages, req_pages;
1737         struct page *pages[PIPE_DEF_BUFFERS];
1738         struct partial_page partial[PIPE_DEF_BUFFERS];
1739         struct page *page;
1740         pgoff_t index, end_index;
1741         loff_t isize, left;
1742         int error, page_nr;
1743         struct splice_pipe_desc spd = {
1744                 .pages = pages,
1745                 .partial = partial,
1746                 .nr_pages_max = PIPE_DEF_BUFFERS,
1747                 .flags = flags,
1748                 .ops = &page_cache_pipe_buf_ops,
1749                 .spd_release = spd_release_page,
1750         };
1751 
1752         isize = i_size_read(inode);
1753         if (unlikely(*ppos >= isize))
1754                 return 0;
1755 
1756         left = isize - *ppos;
1757         if (unlikely(left < len))
1758                 len = left;
1759 
1760         if (splice_grow_spd(pipe, &spd))
1761                 return -ENOMEM;
1762 
1763         index = *ppos >> PAGE_CACHE_SHIFT;
1764         loff = *ppos & ~PAGE_CACHE_MASK;
1765         req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1766         nr_pages = min(req_pages, spd.nr_pages_max);
1767 
1768         spd.nr_pages = find_get_pages_contig(mapping, index,
1769                                                 nr_pages, spd.pages);
1770         index += spd.nr_pages;
1771         error = 0;
1772 
1773         while (spd.nr_pages < nr_pages) {
1774                 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1775                 if (error)
1776                         break;
1777                 unlock_page(page);
1778                 spd.pages[spd.nr_pages++] = page;
1779                 index++;
1780         }
1781 
1782         index = *ppos >> PAGE_CACHE_SHIFT;
1783         nr_pages = spd.nr_pages;
1784         spd.nr_pages = 0;
1785 
1786         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1787                 unsigned int this_len;
1788 
1789                 if (!len)
1790                         break;
1791 
1792                 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1793                 page = spd.pages[page_nr];
1794 
1795                 if (!PageUptodate(page) || page->mapping != mapping) {
1796                         error = shmem_getpage(inode, index, &page,
1797                                                         SGP_CACHE, NULL);
1798                         if (error)
1799                                 break;
1800                         unlock_page(page);
1801                         page_cache_release(spd.pages[page_nr]);
1802                         spd.pages[page_nr] = page;
1803                 }
1804 
1805                 isize = i_size_read(inode);
1806                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1807                 if (unlikely(!isize || index > end_index))
1808                         break;
1809 
1810                 if (end_index == index) {
1811                         unsigned int plen;
1812 
1813                         plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1814                         if (plen <= loff)
1815                                 break;
1816 
1817                         this_len = min(this_len, plen - loff);
1818                         len = this_len;
1819                 }
1820 
1821                 spd.partial[page_nr].offset = loff;
1822                 spd.partial[page_nr].len = this_len;
1823                 len -= this_len;
1824                 loff = 0;
1825                 spd.nr_pages++;
1826                 index++;
1827         }
1828 
1829         while (page_nr < nr_pages)
1830                 page_cache_release(spd.pages[page_nr++]);
1831 
1832         if (spd.nr_pages)
1833                 error = splice_to_pipe(pipe, &spd);
1834 
1835         splice_shrink_spd(&spd);
1836 
1837         if (error > 0) {
1838                 *ppos += error;
1839                 file_accessed(in);
1840         }
1841         return error;
1842 }
1843 
1844 /*
1845  * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1846  */
1847 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1848                                     pgoff_t index, pgoff_t end, int whence)
1849 {
1850         struct page *page;
1851         struct pagevec pvec;
1852         pgoff_t indices[PAGEVEC_SIZE];
1853         bool done = false;
1854         int i;
1855 
1856         pagevec_init(&pvec, 0);
1857         pvec.nr = 1;            /* start small: we may be there already */
1858         while (!done) {
1859                 pvec.nr = find_get_entries(mapping, index,
1860                                         pvec.nr, pvec.pages, indices);
1861                 if (!pvec.nr) {
1862                         if (whence == SEEK_DATA)
1863                                 index = end;
1864                         break;
1865                 }
1866                 for (i = 0; i < pvec.nr; i++, index++) {
1867                         if (index < indices[i]) {
1868                                 if (whence == SEEK_HOLE) {
1869                                         done = true;
1870                                         break;
1871                                 }
1872                                 index = indices[i];
1873                         }
1874                         page = pvec.pages[i];
1875                         if (page && !radix_tree_exceptional_entry(page)) {
1876                                 if (!PageUptodate(page))
1877                                         page = NULL;
1878                         }
1879                         if (index >= end ||
1880                             (page && whence == SEEK_DATA) ||
1881                             (!page && whence == SEEK_HOLE)) {
1882                                 done = true;
1883                                 break;
1884                         }
1885                 }
1886                 pagevec_remove_exceptionals(&pvec);
1887                 pagevec_release(&pvec);
1888                 pvec.nr = PAGEVEC_SIZE;
1889                 cond_resched();
1890         }
1891         return index;
1892 }
1893 
1894 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1895 {
1896         struct address_space *mapping = file->f_mapping;
1897         struct inode *inode = mapping->host;
1898         pgoff_t start, end;
1899         loff_t new_offset;
1900 
1901         if (whence != SEEK_DATA && whence != SEEK_HOLE)
1902                 return generic_file_llseek_size(file, offset, whence,
1903                                         MAX_LFS_FILESIZE, i_size_read(inode));
1904         inode_lock(inode);
1905         /* We're holding i_mutex so we can access i_size directly */
1906 
1907         if (offset < 0)
1908                 offset = -EINVAL;
1909         else if (offset >= inode->i_size)
1910                 offset = -ENXIO;
1911         else {
1912                 start = offset >> PAGE_CACHE_SHIFT;
1913                 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1914                 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1915                 new_offset <<= PAGE_CACHE_SHIFT;
1916                 if (new_offset > offset) {
1917                         if (new_offset < inode->i_size)
1918                                 offset = new_offset;
1919                         else if (whence == SEEK_DATA)
1920                                 offset = -ENXIO;
1921                         else
1922                                 offset = inode->i_size;
1923                 }
1924         }
1925 
1926         if (offset >= 0)
1927                 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1928         inode_unlock(inode);
1929         return offset;
1930 }
1931 
1932 /*
1933  * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
1934  * so reuse a tag which we firmly believe is never set or cleared on shmem.
1935  */
1936 #define SHMEM_TAG_PINNED        PAGECACHE_TAG_TOWRITE
1937 #define LAST_SCAN               4       /* about 150ms max */
1938 
1939 static void shmem_tag_pins(struct address_space *mapping)
1940 {
1941         struct radix_tree_iter iter;
1942         void **slot;
1943         pgoff_t start;
1944         struct page *page;
1945 
1946         lru_add_drain();
1947         start = 0;
1948         rcu_read_lock();
1949 
1950 restart:
1951         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1952                 page = radix_tree_deref_slot(slot);
1953                 if (!page || radix_tree_exception(page)) {
1954                         if (radix_tree_deref_retry(page))
1955                                 goto restart;
1956                 } else if (page_count(page) - page_mapcount(page) > 1) {
1957                         spin_lock_irq(&mapping->tree_lock);
1958                         radix_tree_tag_set(&mapping->page_tree, iter.index,
1959                                            SHMEM_TAG_PINNED);
1960                         spin_unlock_irq(&mapping->tree_lock);
1961                 }
1962 
1963                 if (need_resched()) {
1964                         cond_resched_rcu();
1965                         start = iter.index + 1;
1966                         goto restart;
1967                 }
1968         }
1969         rcu_read_unlock();
1970 }
1971 
1972 /*
1973  * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
1974  * via get_user_pages(), drivers might have some pending I/O without any active
1975  * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
1976  * and see whether it has an elevated ref-count. If so, we tag them and wait for
1977  * them to be dropped.
1978  * The caller must guarantee that no new user will acquire writable references
1979  * to those pages to avoid races.
1980  */
1981 static int shmem_wait_for_pins(struct address_space *mapping)
1982 {
1983         struct radix_tree_iter iter;
1984         void **slot;
1985         pgoff_t start;
1986         struct page *page;
1987         int error, scan;
1988 
1989         shmem_tag_pins(mapping);
1990 
1991         error = 0;
1992         for (scan = 0; scan <= LAST_SCAN; scan++) {
1993                 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
1994                         break;
1995 
1996                 if (!scan)
1997                         lru_add_drain_all();
1998                 else if (schedule_timeout_killable((HZ << scan) / 200))
1999                         scan = LAST_SCAN;
2000 
2001                 start = 0;
2002                 rcu_read_lock();
2003 restart:
2004                 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2005                                            start, SHMEM_TAG_PINNED) {
2006 
2007                         page = radix_tree_deref_slot(slot);
2008                         if (radix_tree_exception(page)) {
2009                                 if (radix_tree_deref_retry(page))
2010                                         goto restart;
2011 
2012                                 page = NULL;
2013                         }
2014 
2015                         if (page &&
2016                             page_count(page) - page_mapcount(page) != 1) {
2017                                 if (scan < LAST_SCAN)
2018                                         goto continue_resched;
2019 
2020                                 /*
2021                                  * On the last scan, we clean up all those tags
2022                                  * we inserted; but make a note that we still
2023                                  * found pages pinned.
2024                                  */
2025                                 error = -EBUSY;
2026                         }
2027 
2028                         spin_lock_irq(&mapping->tree_lock);
2029                         radix_tree_tag_clear(&mapping->page_tree,
2030                                              iter.index, SHMEM_TAG_PINNED);
2031                         spin_unlock_irq(&mapping->tree_lock);
2032 continue_resched:
2033                         if (need_resched()) {
2034                                 cond_resched_rcu();
2035                                 start = iter.index + 1;
2036                                 goto restart;
2037                         }
2038                 }
2039                 rcu_read_unlock();
2040         }
2041 
2042         return error;
2043 }
2044 
2045 #define F_ALL_SEALS (F_SEAL_SEAL | \
2046                      F_SEAL_SHRINK | \
2047                      F_SEAL_GROW | \
2048                      F_SEAL_WRITE)
2049 
2050 int shmem_add_seals(struct file *file, unsigned int seals)
2051 {
2052         struct inode *inode = file_inode(file);
2053         struct shmem_inode_info *info = SHMEM_I(inode);
2054         int error;
2055 
2056         /*
2057          * SEALING
2058          * Sealing allows multiple parties to share a shmem-file but restrict
2059          * access to a specific subset of file operations. Seals can only be
2060          * added, but never removed. This way, mutually untrusted parties can
2061          * share common memory regions with a well-defined policy. A malicious
2062          * peer can thus never perform unwanted operations on a shared object.
2063          *
2064          * Seals are only supported on special shmem-files and always affect
2065          * the whole underlying inode. Once a seal is set, it may prevent some
2066          * kinds of access to the file. Currently, the following seals are
2067          * defined:
2068          *   SEAL_SEAL: Prevent further seals from being set on this file
2069          *   SEAL_SHRINK: Prevent the file from shrinking
2070          *   SEAL_GROW: Prevent the file from growing
2071          *   SEAL_WRITE: Prevent write access to the file
2072          *
2073          * As we don't require any trust relationship between two parties, we
2074          * must prevent seals from being removed. Therefore, sealing a file
2075          * only adds a given set of seals to the file, it never touches
2076          * existing seals. Furthermore, the "setting seals"-operation can be
2077          * sealed itself, which basically prevents any further seal from being
2078          * added.
2079          *
2080          * Semantics of sealing are only defined on volatile files. Only
2081          * anonymous shmem files support sealing. More importantly, seals are
2082          * never written to disk. Therefore, there's no plan to support it on
2083          * other file types.
2084          */
2085 
2086         if (file->f_op != &shmem_file_operations)
2087                 return -EINVAL;
2088         if (!(file->f_mode & FMODE_WRITE))
2089                 return -EPERM;
2090         if (seals & ~(unsigned int)F_ALL_SEALS)
2091                 return -EINVAL;
2092 
2093         inode_lock(inode);
2094 
2095         if (info->seals & F_SEAL_SEAL) {
2096                 error = -EPERM;
2097                 goto unlock;
2098         }
2099 
2100         if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2101                 error = mapping_deny_writable(file->f_mapping);
2102                 if (error)
2103                         goto unlock;
2104 
2105                 error = shmem_wait_for_pins(file->f_mapping);
2106                 if (error) {
2107                         mapping_allow_writable(file->f_mapping);
2108                         goto unlock;
2109                 }
2110         }
2111 
2112         info->seals |= seals;
2113         error = 0;
2114 
2115 unlock:
2116         inode_unlock(inode);
2117         return error;
2118 }
2119 EXPORT_SYMBOL_GPL(shmem_add_seals);
2120 
2121 int shmem_get_seals(struct file *file)
2122 {
2123         if (file->f_op != &shmem_file_operations)
2124                 return -EINVAL;
2125 
2126         return SHMEM_I(file_inode(file))->seals;
2127 }
2128 EXPORT_SYMBOL_GPL(shmem_get_seals);
2129 
2130 long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2131 {
2132         long error;
2133 
2134         switch (cmd) {
2135         case F_ADD_SEALS:
2136                 /* disallow upper 32bit */
2137                 if (arg > UINT_MAX)
2138                         return -EINVAL;
2139 
2140                 error = shmem_add_seals(file, arg);
2141                 break;
2142         case F_GET_SEALS:
2143                 error = shmem_get_seals(file);
2144                 break;
2145         default:
2146                 error = -EINVAL;
2147                 break;
2148         }
2149 
2150         return error;
2151 }
2152 
2153 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2154                                                          loff_t len)
2155 {
2156         struct inode *inode = file_inode(file);
2157         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2158         struct shmem_inode_info *info = SHMEM_I(inode);
2159         struct shmem_falloc shmem_falloc;
2160         pgoff_t start, index, end;
2161         int error;
2162 
2163         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2164                 return -EOPNOTSUPP;
2165 
2166         inode_lock(inode);
2167 
2168         if (mode & FALLOC_FL_PUNCH_HOLE) {
2169                 struct address_space *mapping = file->f_mapping;
2170                 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2171                 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2172                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2173 
2174                 /* protected by i_mutex */
2175                 if (info->seals & F_SEAL_WRITE) {
2176                         error = -EPERM;
2177                         goto out;
2178                 }
2179 
2180                 shmem_falloc.waitq = &shmem_falloc_waitq;
2181                 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2182                 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2183                 spin_lock(&inode->i_lock);
2184                 inode->i_private = &shmem_falloc;
2185                 spin_unlock(&inode->i_lock);
2186 
2187                 if ((u64)unmap_end > (u64)unmap_start)
2188                         unmap_mapping_range(mapping, unmap_start,
2189                                             1 + unmap_end - unmap_start, 0);
2190                 shmem_truncate_range(inode, offset, offset + len - 1);
2191                 /* No need to unmap again: hole-punching leaves COWed pages */
2192 
2193                 spin_lock(&inode->i_lock);
2194                 inode->i_private = NULL;
2195                 wake_up_all(&shmem_falloc_waitq);
2196                 spin_unlock(&inode->i_lock);
2197                 error = 0;
2198                 goto out;
2199         }
2200 
2201         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2202         error = inode_newsize_ok(inode, offset + len);
2203         if (error)
2204                 goto out;
2205 
2206         if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2207                 error = -EPERM;
2208                 goto out;
2209         }
2210 
2211         start = offset >> PAGE_CACHE_SHIFT;
2212         end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2213         /* Try to avoid a swapstorm if len is impossible to satisfy */
2214         if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2215                 error = -ENOSPC;
2216                 goto out;
2217         }
2218 
2219         shmem_falloc.waitq = NULL;
2220         shmem_falloc.start = start;
2221         shmem_falloc.next  = start;
2222         shmem_falloc.nr_falloced = 0;
2223         shmem_falloc.nr_unswapped = 0;
2224         spin_lock(&inode->i_lock);
2225         inode->i_private = &shmem_falloc;
2226         spin_unlock(&inode->i_lock);
2227 
2228         for (index = start; index < end; index++) {
2229                 struct page *page;
2230 
2231                 /*
2232                  * Good, the fallocate(2) manpage permits EINTR: we may have
2233                  * been interrupted because we are using up too much memory.
2234                  */
2235                 if (signal_pending(current))
2236                         error = -EINTR;
2237                 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2238                         error = -ENOMEM;
2239                 else
2240                         error = shmem_getpage(inode, index, &page, SGP_FALLOC,
2241                                                                         NULL);
2242                 if (error) {
2243                         /* Remove the !PageUptodate pages we added */
2244                         shmem_undo_range(inode,
2245                                 (loff_t)start << PAGE_CACHE_SHIFT,
2246                                 (loff_t)index << PAGE_CACHE_SHIFT, true);
2247                         goto undone;
2248                 }
2249 
2250                 /*
2251                  * Inform shmem_writepage() how far we have reached.
2252                  * No need for lock or barrier: we have the page lock.
2253                  */
2254                 shmem_falloc.next++;
2255                 if (!PageUptodate(page))
2256                         shmem_falloc.nr_falloced++;
2257 
2258                 /*
2259                  * If !PageUptodate, leave it that way so that freeable pages
2260                  * can be recognized if we need to rollback on error later.
2261                  * But set_page_dirty so that memory pressure will swap rather
2262                  * than free the pages we are allocating (and SGP_CACHE pages
2263                  * might still be clean: we now need to mark those dirty too).
2264                  */
2265                 set_page_dirty(page);
2266                 unlock_page(page);
2267                 page_cache_release(page);
2268                 cond_resched();
2269         }
2270 
2271         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2272                 i_size_write(inode, offset + len);
2273         inode->i_ctime = CURRENT_TIME;
2274 undone:
2275         spin_lock(&inode->i_lock);
2276         inode->i_private = NULL;
2277         spin_unlock(&inode->i_lock);
2278 out:
2279         inode_unlock(inode);
2280         return error;
2281 }
2282 
2283 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2284 {
2285         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2286 
2287         buf->f_type = TMPFS_MAGIC;
2288         buf->f_bsize = PAGE_CACHE_SIZE;
2289         buf->f_namelen = NAME_MAX;
2290         if (sbinfo->max_blocks) {
2291                 buf->f_blocks = sbinfo->max_blocks;
2292                 buf->f_bavail =
2293                 buf->f_bfree  = sbinfo->max_blocks -
2294                                 percpu_counter_sum(&sbinfo->used_blocks);
2295         }
2296         if (sbinfo->max_inodes) {
2297                 buf->f_files = sbinfo->max_inodes;
2298                 buf->f_ffree = sbinfo->free_inodes;
2299         }
2300         /* else leave those fields 0 like simple_statfs */
2301         return 0;
2302 }
2303 
2304 /*
2305  * File creation. Allocate an inode, and we're done..
2306  */
2307 static int
2308 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2309 {
2310         struct inode *inode;
2311         int error = -ENOSPC;
2312 
2313         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2314         if (inode) {
2315                 error = simple_acl_create(dir, inode);
2316                 if (error)
2317                         goto out_iput;
2318                 error = security_inode_init_security(inode, dir,
2319                                                      &dentry->d_name,
2320                                                      shmem_initxattrs, NULL);
2321                 if (error && error != -EOPNOTSUPP)
2322                         goto out_iput;
2323 
2324                 error = 0;
2325                 dir->i_size += BOGO_DIRENT_SIZE;
2326                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2327                 d_instantiate(dentry, inode);
2328                 dget(dentry); /* Extra count - pin the dentry in core */
2329         }
2330         return error;
2331 out_iput:
2332         iput(inode);
2333         return error;
2334 }
2335 
2336 static int
2337 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2338 {
2339         struct inode *inode;
2340         int error = -ENOSPC;
2341 
2342         inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2343         if (inode) {
2344                 error = security_inode_init_security(inode, dir,
2345                                                      NULL,
2346                                                      shmem_initxattrs, NULL);
2347                 if (error && error != -EOPNOTSUPP)
2348                         goto out_iput;
2349                 error = simple_acl_create(dir, inode);
2350                 if (error)
2351                         goto out_iput;
2352                 d_tmpfile(dentry, inode);
2353         }
2354         return error;
2355 out_iput:
2356         iput(inode);
2357         return error;
2358 }
2359 
2360 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2361 {
2362         int error;
2363 
2364         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2365                 return error;
2366         inc_nlink(dir);
2367         return 0;
2368 }
2369 
2370 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2371                 bool excl)
2372 {
2373         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2374 }
2375 
2376 /*
2377  * Link a file..
2378  */
2379 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2380 {
2381         struct inode *inode = d_inode(old_dentry);
2382         int ret;
2383 
2384         /*
2385          * No ordinary (disk based) filesystem counts links as inodes;
2386          * but each new link needs a new dentry, pinning lowmem, and
2387          * tmpfs dentries cannot be pruned until they are unlinked.
2388          */
2389         ret = shmem_reserve_inode(inode->i_sb);
2390         if (ret)
2391                 goto out;
2392 
2393         dir->i_size += BOGO_DIRENT_SIZE;
2394         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2395         inc_nlink(inode);
2396         ihold(inode);   /* New dentry reference */
2397         dget(dentry);           /* Extra pinning count for the created dentry */
2398         d_instantiate(dentry, inode);
2399 out:
2400         return ret;
2401 }
2402 
2403 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2404 {
2405         struct inode *inode = d_inode(dentry);
2406 
2407         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2408                 shmem_free_inode(inode->i_sb);
2409 
2410         dir->i_size -= BOGO_DIRENT_SIZE;
2411         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2412         drop_nlink(inode);
2413         dput(dentry);   /* Undo the count from "create" - this does all the work */
2414         return 0;
2415 }
2416 
2417 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2418 {
2419         if (!simple_empty(dentry))
2420                 return -ENOTEMPTY;
2421 
2422         drop_nlink(d_inode(dentry));
2423         drop_nlink(dir);
2424         return shmem_unlink(dir, dentry);
2425 }
2426 
2427 static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2428 {
2429         bool old_is_dir = d_is_dir(old_dentry);
2430         bool new_is_dir = d_is_dir(new_dentry);
2431 
2432         if (old_dir != new_dir && old_is_dir != new_is_dir) {
2433                 if (old_is_dir) {
2434                         drop_nlink(old_dir);
2435                         inc_nlink(new_dir);
2436                 } else {
2437                         drop_nlink(new_dir);
2438                         inc_nlink(old_dir);
2439                 }
2440         }
2441         old_dir->i_ctime = old_dir->i_mtime =
2442         new_dir->i_ctime = new_dir->i_mtime =
2443         d_inode(old_dentry)->i_ctime =
2444         d_inode(new_dentry)->i_ctime = CURRENT_TIME;
2445 
2446         return 0;
2447 }
2448 
2449 static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
2450 {
2451         struct dentry *whiteout;
2452         int error;
2453 
2454         whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
2455         if (!whiteout)
2456                 return -ENOMEM;
2457 
2458         error = shmem_mknod(old_dir, whiteout,
2459                             S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
2460         dput(whiteout);
2461         if (error)
2462                 return error;
2463 
2464         /*
2465          * Cheat and hash the whiteout while the old dentry is still in
2466          * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
2467          *
2468          * d_lookup() will consistently find one of them at this point,
2469          * not sure which one, but that isn't even important.
2470          */
2471         d_rehash(whiteout);
2472         return 0;
2473 }
2474 
2475 /*
2476  * The VFS layer already does all the dentry stuff for rename,
2477  * we just have to decrement the usage count for the target if
2478  * it exists so that the VFS layer correctly free's it when it
2479  * gets overwritten.
2480  */
2481 static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
2482 {
2483         struct inode *inode = d_inode(old_dentry);
2484         int they_are_dirs = S_ISDIR(inode->i_mode);
2485 
2486         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
2487                 return -EINVAL;
2488 
2489         if (flags & RENAME_EXCHANGE)
2490                 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
2491 
2492         if (!simple_empty(new_dentry))
2493                 return -ENOTEMPTY;
2494 
2495         if (flags & RENAME_WHITEOUT) {
2496                 int error;
2497 
2498                 error = shmem_whiteout(old_dir, old_dentry);
2499                 if (error)
2500                         return error;
2501         }
2502 
2503         if (d_really_is_positive(new_dentry)) {
2504                 (void) shmem_unlink(new_dir, new_dentry);
2505                 if (they_are_dirs) {
2506                         drop_nlink(d_inode(new_dentry));
2507                         drop_nlink(old_dir);
2508                 }
2509         } else if (they_are_dirs) {
2510                 drop_nlink(old_dir);
2511                 inc_nlink(new_dir);
2512         }
2513 
2514         old_dir->i_size -= BOGO_DIRENT_SIZE;
2515         new_dir->i_size += BOGO_DIRENT_SIZE;
2516         old_dir->i_ctime = old_dir->i_mtime =
2517         new_dir->i_ctime = new_dir->i_mtime =
2518         inode->i_ctime = CURRENT_TIME;
2519         return 0;
2520 }
2521 
2522 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2523 {
2524         int error;
2525         int len;
2526         struct inode *inode;
2527         struct page *page;
2528         struct shmem_inode_info *info;
2529 
2530         len = strlen(symname) + 1;
2531         if (len > PAGE_CACHE_SIZE)
2532                 return -ENAMETOOLONG;
2533 
2534         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2535         if (!inode)
2536                 return -ENOSPC;
2537 
2538         error = security_inode_init_security(inode, dir, &dentry->d_name,
2539                                              shmem_initxattrs, NULL);
2540         if (error) {
2541                 if (error != -EOPNOTSUPP) {
2542                         iput(inode);
2543                         return error;
2544                 }
2545                 error = 0;
2546         }
2547 
2548         info = SHMEM_I(inode);
2549         inode->i_size = len-1;
2550         if (len <= SHORT_SYMLINK_LEN) {
2551                 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
2552                 if (!inode->i_link) {
2553                         iput(inode);
2554                         return -ENOMEM;
2555                 }
2556                 inode->i_op = &shmem_short_symlink_operations;
2557         } else {
2558                 inode_nohighmem(inode);
2559                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2560                 if (error) {
2561                         iput(inode);
2562                         return error;
2563                 }
2564                 inode->i_mapping->a_ops = &shmem_aops;
2565                 inode->i_op = &shmem_symlink_inode_operations;
2566                 memcpy(page_address(page), symname, len);
2567                 SetPageUptodate(page);
2568                 set_page_dirty(page);
2569                 unlock_page(page);
2570                 page_cache_release(page);
2571         }
2572         dir->i_size += BOGO_DIRENT_SIZE;
2573         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2574         d_instantiate(dentry, inode);
2575         dget(dentry);
2576         return 0;
2577 }
2578 
2579 static void shmem_put_link(void *arg)
2580 {
2581         mark_page_accessed(arg);
2582         put_page(arg);
2583 }
2584 
2585 static const char *shmem_get_link(struct dentry *dentry,
2586                                   struct inode *inode,
2587                                   struct delayed_call *done)
2588 {
2589         struct page *page = NULL;
2590         int error;
2591         if (!dentry) {
2592                 page = find_get_page(inode->i_mapping, 0);
2593                 if (!page)
2594                         return ERR_PTR(-ECHILD);
2595                 if (!PageUptodate(page)) {
2596                         put_page(page);
2597                         return ERR_PTR(-ECHILD);
2598                 }
2599         } else {
2600                 error = shmem_getpage(inode, 0, &page, SGP_READ, NULL);
2601                 if (error)
2602                         return ERR_PTR(error);
2603                 unlock_page(page);
2604         }
2605         set_delayed_call(done, shmem_put_link, page);
2606         return page_address(page);
2607 }
2608 
2609 #ifdef CONFIG_TMPFS_XATTR
2610 /*
2611  * Superblocks without xattr inode operations may get some security.* xattr
2612  * support from the LSM "for free". As soon as we have any other xattrs
2613  * like ACLs, we also need to implement the security.* handlers at
2614  * filesystem level, though.
2615  */
2616 
2617 /*
2618  * Callback for security_inode_init_security() for acquiring xattrs.
2619  */
2620 static int shmem_initxattrs(struct inode *inode,
2621                             const struct xattr *xattr_array,
2622                             void *fs_info)
2623 {
2624         struct shmem_inode_info *info = SHMEM_I(inode);
2625         const struct xattr *xattr;
2626         struct simple_xattr *new_xattr;
2627         size_t len;
2628 
2629         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2630                 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2631                 if (!new_xattr)
2632                         return -ENOMEM;
2633 
2634                 len = strlen(xattr->name) + 1;
2635                 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2636                                           GFP_KERNEL);
2637                 if (!new_xattr->name) {
2638                         kfree(new_xattr);
2639                         return -ENOMEM;
2640                 }
2641 
2642                 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2643                        XATTR_SECURITY_PREFIX_LEN);
2644                 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2645                        xattr->name, len);
2646 
2647                 simple_xattr_list_add(&info->xattrs, new_xattr);
2648         }
2649 
2650         return 0;
2651 }
2652 
2653 static int shmem_xattr_handler_get(const struct xattr_handler *handler,
2654                                    struct dentry *dentry, const char *name,
2655                                    void *buffer, size_t size)
2656 {
2657         struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2658 
2659         name = xattr_full_name(handler, name);
2660         return simple_xattr_get(&info->xattrs, name, buffer, size);
2661 }
2662 
2663 static int shmem_xattr_handler_set(const struct xattr_handler *handler,
2664                                    struct dentry *dentry, const char *name,
2665                                    const void *value, size_t size, int flags)
2666 {
2667         struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2668 
2669         name = xattr_full_name(handler, name);
2670         return simple_xattr_set(&info->xattrs, name, value, size, flags);
2671 }
2672 
2673 static const struct xattr_handler shmem_security_xattr_handler = {
2674         .prefix = XATTR_SECURITY_PREFIX,
2675         .get = shmem_xattr_handler_get,
2676         .set = shmem_xattr_handler_set,
2677 };
2678 
2679 static const struct xattr_handler shmem_trusted_xattr_handler = {
2680         .prefix = XATTR_TRUSTED_PREFIX,
2681         .get = shmem_xattr_handler_get,
2682         .set = shmem_xattr_handler_set,
2683 };
2684 
2685 static const struct xattr_handler *shmem_xattr_handlers[] = {
2686 #ifdef CONFIG_TMPFS_POSIX_ACL
2687         &posix_acl_access_xattr_handler,
2688         &posix_acl_default_xattr_handler,
2689 #endif
2690         &shmem_security_xattr_handler,
2691         &shmem_trusted_xattr_handler,
2692         NULL
2693 };
2694 
2695 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2696 {
2697         struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
2698         return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
2699 }
2700 #endif /* CONFIG_TMPFS_XATTR */
2701 
2702 static const struct inode_operations shmem_short_symlink_operations = {
2703         .readlink       = generic_readlink,
2704         .get_link       = simple_get_link,
2705 #ifdef CONFIG_TMPFS_XATTR
2706         .setxattr       = generic_setxattr,
2707         .getxattr       = generic_getxattr,
2708         .listxattr      = shmem_listxattr,
2709         .removexattr    = generic_removexattr,
2710 #endif
2711 };
2712 
2713 static const struct inode_operations shmem_symlink_inode_operations = {
2714         .readlink       = generic_readlink,
2715         .get_link       = shmem_get_link,
2716 #ifdef CONFIG_TMPFS_XATTR
2717         .setxattr       = generic_setxattr,
2718         .getxattr       = generic_getxattr,
2719         .listxattr      = shmem_listxattr,
2720         .removexattr    = generic_removexattr,
2721 #endif
2722 };
2723 
2724 static struct dentry *shmem_get_parent(struct dentry *child)
2725 {
2726         return ERR_PTR(-ESTALE);
2727 }
2728 
2729 static int shmem_match(struct inode *ino, void *vfh)
2730 {
2731         __u32 *fh = vfh;
2732         __u64 inum = fh[2];
2733         inum = (inum << 32) | fh[1];
2734         return ino->i_ino == inum && fh[0] == ino->i_generation;
2735 }
2736 
2737 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2738                 struct fid *fid, int fh_len, int fh_type)
2739 {
2740         struct inode *inode;
2741         struct dentry *dentry = NULL;
2742         u64 inum;
2743 
2744         if (fh_len < 3)
2745                 return NULL;
2746 
2747         inum = fid->raw[2];
2748         inum = (inum << 32) | fid->raw[1];
2749 
2750         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2751                         shmem_match, fid->raw);
2752         if (inode) {
2753                 dentry = d_find_alias(inode);
2754                 iput(inode);
2755         }
2756 
2757         return dentry;
2758 }
2759 
2760 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2761                                 struct inode *parent)
2762 {
2763         if (*len < 3) {
2764                 *len = 3;
2765                 return FILEID_INVALID;
2766         }
2767 
2768         if (inode_unhashed(inode)) {
2769                 /* Unfortunately insert_inode_hash is not idempotent,
2770                  * so as we hash inodes here rather than at creation
2771                  * time, we need a lock to ensure we only try
2772                  * to do it once
2773                  */
2774                 static DEFINE_SPINLOCK(lock);
2775                 spin_lock(&lock);
2776                 if (inode_unhashed(inode))
2777                         __insert_inode_hash(inode,
2778                                             inode->i_ino + inode->i_generation);
2779                 spin_unlock(&lock);
2780         }
2781 
2782         fh[0] = inode->i_generation;
2783         fh[1] = inode->i_ino;
2784         fh[2] = ((__u64)inode->i_ino) >> 32;
2785 
2786         *len = 3;
2787         return 1;
2788 }
2789 
2790 static const struct export_operations shmem_export_ops = {
2791         .get_parent     = shmem_get_parent,
2792         .encode_fh      = shmem_encode_fh,
2793         .fh_to_dentry   = shmem_fh_to_dentry,
2794 };
2795 
2796 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2797                                bool remount)
2798 {
2799         char *this_char, *value, *rest;
2800         struct mempolicy *mpol = NULL;
2801         uid_t uid;
2802         gid_t gid;
2803 
2804         while (options != NULL) {
2805                 this_char = options;
2806                 for (;;) {
2807                         /*
2808                          * NUL-terminate this option: unfortunately,
2809                          * mount options form a comma-separated list,
2810                          * but mpol's nodelist may also contain commas.
2811                          */
2812                         options = strchr(options, ',');
2813                         if (options == NULL)
2814                                 break;
2815                         options++;
2816                         if (!isdigit(*options)) {
2817                                 options[-1] = '\0';
2818                                 break;
2819                         }
2820                 }
2821                 if (!*this_char)
2822                         continue;
2823                 if ((value = strchr(this_char,'=')) != NULL) {
2824                         *value++ = 0;
2825                 } else {
2826                         printk(KERN_ERR
2827                             "tmpfs: No value for mount option '%s'\n",
2828                             this_char);
2829                         goto error;
2830                 }
2831 
2832                 if (!strcmp(this_char,"size")) {
2833                         unsigned long long size;
2834                         size = memparse(value,&rest);
2835                         if (*rest == '%') {
2836                                 size <<= PAGE_SHIFT;
2837                                 size *= totalram_pages;
2838                                 do_div(size, 100);
2839                                 rest++;
2840                         }
2841                         if (*rest)
2842                                 goto bad_val;
2843                         sbinfo->max_blocks =
2844                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2845                 } else if (!strcmp(this_char,"nr_blocks")) {
2846                         sbinfo->max_blocks = memparse(value, &rest);
2847                         if (*rest)
2848                                 goto bad_val;
2849                 } else if (!strcmp(this_char,"nr_inodes")) {
2850                         sbinfo->max_inodes = memparse(value, &rest);
2851                         if (*rest)
2852                                 goto bad_val;
2853                 } else if (!strcmp(this_char,"mode")) {
2854                         if (remount)
2855                                 continue;
2856                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2857                         if (*rest)
2858                                 goto bad_val;
2859                 } else if (!strcmp(this_char,"uid")) {
2860                         if (remount)
2861                                 continue;
2862                         uid = simple_strtoul(value, &rest, 0);
2863                         if (*rest)
2864                                 goto bad_val;
2865                         sbinfo->uid = make_kuid(current_user_ns(), uid);
2866                         if (!uid_valid(sbinfo->uid))
2867                                 goto bad_val;
2868                 } else if (!strcmp(this_char,"gid")) {
2869                         if (remount)
2870                                 continue;
2871                         gid = simple_strtoul(value, &rest, 0);
2872                         if (*rest)
2873                                 goto bad_val;
2874                         sbinfo->gid = make_kgid(current_user_ns(), gid);
2875                         if (!gid_valid(sbinfo->gid))
2876                                 goto bad_val;
2877                 } else if (!strcmp(this_char,"mpol")) {
2878                         mpol_put(mpol);
2879                         mpol = NULL;
2880                         if (mpol_parse_str(value, &mpol))
2881                                 goto bad_val;
2882                 } else {
2883                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2884                                this_char);
2885                         goto error;
2886                 }
2887         }
2888         sbinfo->mpol = mpol;
2889         return 0;
2890 
2891 bad_val:
2892         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2893                value, this_char);
2894 error:
2895         mpol_put(mpol);
2896         return 1;
2897 
2898 }
2899 
2900 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2901 {
2902         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2903         struct shmem_sb_info config = *sbinfo;
2904         unsigned long inodes;
2905         int error = -EINVAL;
2906 
2907         config.mpol = NULL;
2908         if (shmem_parse_options(data, &config, true))
2909                 return error;
2910 
2911         spin_lock(&sbinfo->stat_lock);
2912         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2913         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2914                 goto out;
2915         if (config.max_inodes < inodes)
2916                 goto out;
2917         /*
2918          * Those tests disallow limited->unlimited while any are in use;
2919          * but we must separately disallow unlimited->limited, because
2920          * in that case we have no record of how much is already in use.
2921          */
2922         if (config.max_blocks && !sbinfo->max_blocks)
2923                 goto out;
2924         if (config.max_inodes && !sbinfo->max_inodes)
2925                 goto out;
2926 
2927         error = 0;
2928         sbinfo->max_blocks  = config.max_blocks;
2929         sbinfo->max_inodes  = config.max_inodes;
2930         sbinfo->free_inodes = config.max_inodes - inodes;
2931 
2932         /*
2933          * Preserve previous mempolicy unless mpol remount option was specified.
2934          */
2935         if (config.mpol) {
2936                 mpol_put(sbinfo->mpol);
2937                 sbinfo->mpol = config.mpol;     /* transfers initial ref */
2938         }
2939 out:
2940         spin_unlock(&sbinfo->stat_lock);
2941         return error;
2942 }
2943 
2944 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2945 {
2946         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2947 
2948         if (sbinfo->max_blocks != shmem_default_max_blocks())
2949                 seq_printf(seq, ",size=%luk",
2950                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2951         if (sbinfo->max_inodes != shmem_default_max_inodes())
2952                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2953         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2954                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2955         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2956                 seq_printf(seq, ",uid=%u",
2957                                 from_kuid_munged(&init_user_ns, sbinfo->uid));
2958         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2959                 seq_printf(seq, ",gid=%u",
2960                                 from_kgid_munged(&init_user_ns, sbinfo->gid));
2961         shmem_show_mpol(seq, sbinfo->mpol);
2962         return 0;
2963 }
2964 
2965 #define MFD_NAME_PREFIX "memfd:"
2966 #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
2967 #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
2968 
2969 #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
2970 
2971 SYSCALL_DEFINE2(memfd_create,
2972                 const char __user *, uname,
2973                 unsigned int, flags)
2974 {
2975         struct shmem_inode_info *info;
2976         struct file *file;
2977         int fd, error;
2978         char *name;
2979         long len;
2980 
2981         if (flags & ~(unsigned int)MFD_ALL_FLAGS)
2982                 return -EINVAL;
2983 
2984         /* length includes terminating zero */
2985         len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
2986         if (len <= 0)
2987                 return -EFAULT;
2988         if (len > MFD_NAME_MAX_LEN + 1)
2989                 return -EINVAL;
2990 
2991         name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
2992         if (!name)
2993                 return -ENOMEM;
2994 
2995         strcpy(name, MFD_NAME_PREFIX);
2996         if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
2997                 error = -EFAULT;
2998                 goto err_name;
2999         }
3000 
3001         /* terminating-zero may have changed after strnlen_user() returned */
3002         if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3003                 error = -EFAULT;
3004                 goto err_name;
3005         }
3006 
3007         fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3008         if (fd < 0) {
3009                 error = fd;
3010                 goto err_name;
3011         }
3012 
3013         file = shmem_file_setup(name, 0, VM_NORESERVE);
3014         if (IS_ERR(file)) {
3015                 error = PTR_ERR(file);
3016                 goto err_fd;
3017         }
3018         info = SHMEM_I(file_inode(file));
3019         file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3020         file->f_flags |= O_RDWR | O_LARGEFILE;
3021         if (flags & MFD_ALLOW_SEALING)
3022                 info->seals &= ~F_SEAL_SEAL;
3023 
3024         fd_install(fd, file);
3025         kfree(name);
3026         return fd;
3027 
3028 err_fd:
3029         put_unused_fd(fd);
3030 err_name:
3031         kfree(name);
3032         return error;
3033 }
3034 
3035 #endif /* CONFIG_TMPFS */
3036 
3037 static void shmem_put_super(struct super_block *sb)
3038 {
3039         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3040 
3041         percpu_counter_destroy(&sbinfo->used_blocks);
3042         mpol_put(sbinfo->mpol);
3043         kfree(sbinfo);
3044         sb->s_fs_info = NULL;
3045 }
3046 
3047 int shmem_fill_super(struct super_block *sb, void *data, int silent)
3048 {
3049         struct inode *inode;
3050         struct shmem_sb_info *sbinfo;
3051         int err = -ENOMEM;
3052 
3053         /* Round up to L1_CACHE_BYTES to resist false sharing */
3054         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3055                                 L1_CACHE_BYTES), GFP_KERNEL);
3056         if (!sbinfo)
3057                 return -ENOMEM;
3058 
3059         sbinfo->mode = S_IRWXUGO | S_ISVTX;
3060         sbinfo->uid = current_fsuid();
3061         sbinfo->gid = current_fsgid();
3062         sb->s_fs_info = sbinfo;
3063 
3064 #ifdef CONFIG_TMPFS
3065         /*
3066          * Per default we only allow half of the physical ram per
3067          * tmpfs instance, limiting inodes to one per page of lowmem;
3068          * but the internal instance is left unlimited.
3069          */
3070         if (!(sb->s_flags & MS_KERNMOUNT)) {
3071                 sbinfo->max_blocks = shmem_default_max_blocks();
3072                 sbinfo->max_inodes = shmem_default_max_inodes();
3073                 if (shmem_parse_options(data, sbinfo, false)) {
3074                         err = -EINVAL;
3075                         goto failed;
3076                 }
3077         } else {
3078                 sb->s_flags |= MS_NOUSER;
3079         }
3080         sb->s_export_op = &shmem_export_ops;
3081         sb->s_flags |= MS_NOSEC;
3082 #else
3083         sb->s_flags |= MS_NOUSER;
3084 #endif
3085 
3086         spin_lock_init(&sbinfo->stat_lock);
3087         if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3088                 goto failed;
3089         sbinfo->free_inodes = sbinfo->max_inodes;
3090 
3091         sb->s_maxbytes = MAX_LFS_FILESIZE;
3092         sb->s_blocksize = PAGE_CACHE_SIZE;
3093         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
3094         sb->s_magic = TMPFS_MAGIC;
3095         sb->s_op = &shmem_ops;
3096         sb->s_time_gran = 1;
3097 #ifdef CONFIG_TMPFS_XATTR
3098         sb->s_xattr = shmem_xattr_handlers;
3099 #endif
3100 #ifdef CONFIG_TMPFS_POSIX_ACL
3101         sb->s_flags |= MS_POSIXACL;
3102 #endif
3103 
3104         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3105         if (!inode)
3106                 goto failed;
3107         inode->i_uid = sbinfo->uid;
3108         inode->i_gid = sbinfo->gid;
3109         sb->s_root = d_make_root(inode);
3110         if (!sb->s_root)
3111                 goto failed;
3112         return 0;
3113 
3114 failed:
3115         shmem_put_super(sb);
3116         return err;
3117 }
3118 
3119 static struct kmem_cache *shmem_inode_cachep;
3120 
3121 static struct inode *shmem_alloc_inode(struct super_block *sb)
3122 {
3123         struct shmem_inode_info *info;
3124         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3125         if (!info)
3126                 return NULL;
3127         return &info->vfs_inode;
3128 }
3129 
3130 static void shmem_destroy_callback(struct rcu_head *head)
3131 {
3132         struct inode *inode = container_of(head, struct inode, i_rcu);
3133         kfree(inode->i_link);
3134         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3135 }
3136 
3137 static void shmem_destroy_inode(struct inode *inode)
3138 {
3139         if (S_ISREG(inode->i_mode))
3140                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3141         call_rcu(&inode->i_rcu, shmem_destroy_callback);
3142 }
3143 
3144 static void shmem_init_inode(void *foo)
3145 {
3146         struct shmem_inode_info *info = foo;
3147         inode_init_once(&info->vfs_inode);
3148 }
3149 
3150 static int shmem_init_inodecache(void)
3151 {
3152         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3153                                 sizeof(struct shmem_inode_info),
3154                                 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3155         return 0;
3156 }
3157 
3158 static void shmem_destroy_inodecache(void)
3159 {
3160         kmem_cache_destroy(shmem_inode_cachep);
3161 }
3162 
3163 static const struct address_space_operations shmem_aops = {
3164         .writepage      = shmem_writepage,
3165         .set_page_dirty = __set_page_dirty_no_writeback,
3166 #ifdef CONFIG_TMPFS
3167         .write_begin    = shmem_write_begin,
3168         .write_end      = shmem_write_end,
3169 #endif
3170 #ifdef CONFIG_MIGRATION
3171         .migratepage    = migrate_page,
3172 #endif
3173         .error_remove_page = generic_error_remove_page,
3174 };
3175 
3176 static const struct file_operations shmem_file_operations = {
3177         .mmap           = shmem_mmap,
3178 #ifdef CONFIG_TMPFS
3179         .llseek         = shmem_file_llseek,
3180         .read_iter      = shmem_file_read_iter,
3181         .write_iter     = generic_file_write_iter,
3182         .fsync          = noop_fsync,
3183         .splice_read    = shmem_file_splice_read,
3184         .splice_write   = iter_file_splice_write,
3185         .fallocate      = shmem_fallocate,
3186 #endif
3187 };
3188 
3189 static const struct inode_operations shmem_inode_operations = {
3190         .getattr        = shmem_getattr,
3191         .setattr        = shmem_setattr,
3192 #ifdef CONFIG_TMPFS_XATTR
3193         .setxattr       = generic_setxattr,
3194         .getxattr       = generic_getxattr,
3195         .listxattr      = shmem_listxattr,
3196         .removexattr    = generic_removexattr,
3197         .set_acl        = simple_set_acl,
3198 #endif
3199 };
3200 
3201 static const struct inode_operations shmem_dir_inode_operations = {
3202 #ifdef CONFIG_TMPFS
3203         .create         = shmem_create,
3204         .lookup         = simple_lookup,
3205         .link           = shmem_link,
3206         .unlink         = shmem_unlink,
3207         .symlink        = shmem_symlink,
3208         .mkdir          = shmem_mkdir,
3209         .rmdir          = shmem_rmdir,
3210         .mknod          = shmem_mknod,
3211         .rename2        = shmem_rename2,
3212         .tmpfile        = shmem_tmpfile,
3213 #endif
3214 #ifdef CONFIG_TMPFS_XATTR
3215         .setxattr       = generic_setxattr,
3216         .getxattr       = generic_getxattr,
3217         .listxattr      = shmem_listxattr,
3218         .removexattr    = generic_removexattr,
3219 #endif
3220 #ifdef CONFIG_TMPFS_POSIX_ACL
3221         .setattr        = shmem_setattr,
3222         .set_acl        = simple_set_acl,
3223 #endif
3224 };
3225 
3226 static const struct inode_operations shmem_special_inode_operations = {
3227 #ifdef CONFIG_TMPFS_XATTR
3228         .setxattr       = generic_setxattr,
3229         .getxattr       = generic_getxattr,
3230         .listxattr      = shmem_listxattr,
3231         .removexattr    = generic_removexattr,
3232 #endif
3233 #ifdef CONFIG_TMPFS_POSIX_ACL
3234         .setattr        = shmem_setattr,
3235         .set_acl        = simple_set_acl,
3236 #endif
3237 };
3238 
3239 static const struct super_operations shmem_ops = {
3240         .alloc_inode    = shmem_alloc_inode,
3241         .destroy_inode  = shmem_destroy_inode,
3242 #ifdef CONFIG_TMPFS
3243         .statfs         = shmem_statfs,
3244         .remount_fs     = shmem_remount_fs,
3245         .show_options   = shmem_show_options,
3246 #endif
3247         .evict_inode    = shmem_evict_inode,
3248         .drop_inode     = generic_delete_inode,
3249         .put_super      = shmem_put_super,
3250 };
3251 
3252 static const struct vm_operations_struct shmem_vm_ops = {
3253         .fault          = shmem_fault,
3254         .map_pages      = filemap_map_pages,
3255 #ifdef CONFIG_NUMA
3256         .set_policy     = shmem_set_policy,
3257         .get_policy     = shmem_get_policy,
3258 #endif
3259 };
3260 
3261 static struct dentry *shmem_mount(struct file_system_type *fs_type,
3262         int flags, const char *dev_name, void *data)
3263 {
3264         return mount_nodev(fs_type, flags, data, shmem_fill_super);
3265 }
3266 
3267 static struct file_system_type shmem_fs_type = {
3268         .owner          = THIS_MODULE,
3269         .name           = "tmpfs",
3270         .mount          = shmem_mount,
3271         .kill_sb        = kill_litter_super,
3272         .fs_flags       = FS_USERNS_MOUNT,
3273 };
3274 
3275 int __init shmem_init(void)
3276 {
3277         int error;
3278 
3279         /* If rootfs called this, don't re-init */
3280         if (shmem_inode_cachep)
3281                 return 0;
3282 
3283         error = shmem_init_inodecache();
3284         if (error)
3285                 goto out3;
3286 
3287         error = register_filesystem(&shmem_fs_type);
3288         if (error) {
3289                 printk(KERN_ERR "Could not register tmpfs\n");
3290                 goto out2;
3291         }
3292 
3293         shm_mnt = kern_mount(&shmem_fs_type);
3294         if (IS_ERR(shm_mnt)) {
3295                 error = PTR_ERR(shm_mnt);
3296                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
3297                 goto out1;
3298         }
3299         return 0;
3300 
3301 out1:
3302         unregister_filesystem(&shmem_fs_type);
3303 out2:
3304         shmem_destroy_inodecache();
3305 out3:
3306         shm_mnt = ERR_PTR(error);
3307         return error;
3308 }
3309 
3310 #else /* !CONFIG_SHMEM */
3311 
3312 /*
3313  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
3314  *
3315  * This is intended for small system where the benefits of the full
3316  * shmem code (swap-backed and resource-limited) are outweighed by
3317  * their complexity. On systems without swap this code should be
3318  * effectively equivalent, but much lighter weight.
3319  */
3320 
3321 static struct file_system_type shmem_fs_type = {
3322         .name           = "tmpfs",
3323         .mount          = ramfs_mount,
3324         .kill_sb        = kill_litter_super,
3325         .fs_flags       = FS_USERNS_MOUNT,
3326 };
3327 
3328 int __init shmem_init(void)
3329 {
3330         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
3331 
3332         shm_mnt = kern_mount(&shmem_fs_type);
3333         BUG_ON(IS_ERR(shm_mnt));
3334 
3335         return 0;
3336 }
3337 
3338 int shmem_unuse(swp_entry_t swap, struct page *page)
3339 {
3340         return 0;
3341 }
3342 
3343 int shmem_lock(struct file *file, int lock, struct user_struct *user)
3344 {
3345         return 0;
3346 }
3347 
3348 void shmem_unlock_mapping(struct address_space *mapping)
3349 {
3350 }
3351 
3352 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
3353 {
3354         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
3355 }
3356 EXPORT_SYMBOL_GPL(shmem_truncate_range);
3357 
3358 #define shmem_vm_ops                            generic_file_vm_ops
3359 #define shmem_file_operations                   ramfs_file_operations
3360 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
3361 #define shmem_acct_size(flags, size)            0
3362 #define shmem_unacct_size(flags, size)          do {} while (0)
3363 
3364 #endif /* CONFIG_SHMEM */
3365 
3366 /* common code */
3367 
3368 static struct dentry_operations anon_ops = {
3369         .d_dname = simple_dname
3370 };
3371 
3372 static struct file *__shmem_file_setup(const char *name, loff_t size,
3373                                        unsigned long flags, unsigned int i_flags)
3374 {
3375         struct file *res;
3376         struct inode *inode;
3377         struct path path;
3378         struct super_block *sb;
3379         struct qstr this;
3380 
3381         if (IS_ERR(shm_mnt))
3382                 return ERR_CAST(shm_mnt);
3383 
3384         if (size < 0 || size > MAX_LFS_FILESIZE)
3385                 return ERR_PTR(-EINVAL);
3386 
3387         if (shmem_acct_size(flags, size))
3388                 return ERR_PTR(-ENOMEM);
3389 
3390         res = ERR_PTR(-ENOMEM);
3391         this.name = name;
3392         this.len = strlen(name);
3393         this.hash = 0; /* will go */
3394         sb = shm_mnt->mnt_sb;
3395         path.mnt = mntget(shm_mnt);
3396         path.dentry = d_alloc_pseudo(sb, &this);
3397         if (!path.dentry)
3398                 goto put_memory;
3399         d_set_d_op(path.dentry, &anon_ops);
3400 
3401         res = ERR_PTR(-ENOSPC);
3402         inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
3403         if (!inode)
3404                 goto put_memory;
3405 
3406         inode->i_flags |= i_flags;
3407         d_instantiate(path.dentry, inode);
3408         inode->i_size = size;
3409         clear_nlink(inode);     /* It is unlinked */
3410         res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
3411         if (IS_ERR(res))
3412                 goto put_path;
3413 
3414         res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
3415                   &shmem_file_operations);
3416         if (IS_ERR(res))
3417                 goto put_path;
3418 
3419         return res;
3420 
3421 put_memory:
3422         shmem_unacct_size(flags, size);
3423 put_path:
3424         path_put(&path);
3425         return res;
3426 }
3427 
3428 /**
3429  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
3430  *      kernel internal.  There will be NO LSM permission checks against the
3431  *      underlying inode.  So users of this interface must do LSM checks at a
3432  *      higher layer.  The users are the big_key and shm implementations.  LSM
3433  *      checks are provided at the key or shm level rather than the inode.
3434  * @name: name for dentry (to be seen in /proc/<pid>/maps
3435  * @size: size to be set for the file
3436  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3437  */
3438 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
3439 {
3440         return __shmem_file_setup(name, size, flags, S_PRIVATE);
3441 }
3442 
3443 /**
3444  * shmem_file_setup - get an unlinked file living in tmpfs
3445  * @name: name for dentry (to be seen in /proc/<pid>/maps
3446  * @size: size to be set for the file
3447  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
3448  */
3449 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
3450 {
3451         return __shmem_file_setup(name, size, flags, 0);
3452 }
3453 EXPORT_SYMBOL_GPL(shmem_file_setup);
3454 
3455 /**
3456  * shmem_zero_setup - setup a shared anonymous mapping
3457  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3458  */
3459 int shmem_zero_setup(struct vm_area_struct *vma)
3460 {
3461         struct file *file;
3462         loff_t size = vma->vm_end - vma->vm_start;
3463 
3464         /*
3465          * Cloning a new file under mmap_sem leads to a lock ordering conflict
3466          * between XFS directory reading and selinux: since this file is only
3467          * accessible to the user through its mapping, use S_PRIVATE flag to
3468          * bypass file security, in the same way as shmem_kernel_file_setup().
3469          */
3470         file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
3471         if (IS_ERR(file))
3472                 return PTR_ERR(file);
3473 
3474         if (vma->vm_file)
3475                 fput(vma->vm_file);
3476         vma->vm_file = file;
3477         vma->vm_ops = &shmem_vm_ops;
3478         return 0;
3479 }
3480 
3481 /**
3482  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3483  * @mapping:    the page's address_space
3484  * @index:      the page index
3485  * @gfp:        the page allocator flags to use if allocating
3486  *
3487  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3488  * with any new page allocations done using the specified allocation flags.
3489  * But read_cache_page_gfp() uses the ->readpage() method: which does not
3490  * suit tmpfs, since it may have pages in swapcache, and needs to find those
3491  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3492  *
3493  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3494  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3495  */
3496 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3497                                          pgoff_t index, gfp_t gfp)
3498 {
3499 #ifdef CONFIG_SHMEM
3500         struct inode *inode = mapping->host;
3501         struct page *page;
3502         int error;
3503 
3504         BUG_ON(mapping->a_ops != &shmem_aops);
3505         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3506         if (error)
3507                 page = ERR_PTR(error);
3508         else
3509                 unlock_page(page);
3510         return page;
3511 #else
3512         /*
3513          * The tiny !SHMEM case uses ramfs without swap
3514          */
3515         return read_cache_page_gfp(mapping, index, gfp);
3516 #endif
3517 }
3518 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
3519 

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