mm: memcg: only check for PageSwapCache when uncharging anon
[linux-3.10.git] / 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/pagemap.h>
29 #include <linux/file.h>
30 #include <linux/mm.h>
31 #include <linux/export.h>
32 #include <linux/swap.h>
33
34 static struct vfsmount *shm_mnt;
35
36 #ifdef CONFIG_SHMEM
37 /*
38  * This virtual memory filesystem is heavily based on the ramfs. It
39  * extends ramfs by the ability to use swap and honor resource limits
40  * which makes it a completely usable filesystem.
41  */
42
43 #include <linux/xattr.h>
44 #include <linux/exportfs.h>
45 #include <linux/posix_acl.h>
46 #include <linux/generic_acl.h>
47 #include <linux/mman.h>
48 #include <linux/string.h>
49 #include <linux/slab.h>
50 #include <linux/backing-dev.h>
51 #include <linux/shmem_fs.h>
52 #include <linux/writeback.h>
53 #include <linux/blkdev.h>
54 #include <linux/pagevec.h>
55 #include <linux/percpu_counter.h>
56 #include <linux/falloc.h>
57 #include <linux/splice.h>
58 #include <linux/security.h>
59 #include <linux/swapops.h>
60 #include <linux/mempolicy.h>
61 #include <linux/namei.h>
62 #include <linux/ctype.h>
63 #include <linux/migrate.h>
64 #include <linux/highmem.h>
65 #include <linux/seq_file.h>
66 #include <linux/magic.h>
67
68 #include <asm/uaccess.h>
69 #include <asm/pgtable.h>
70
71 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
72 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
73
74 /* Pretend that each entry is of this size in directory's i_size */
75 #define BOGO_DIRENT_SIZE 20
76
77 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
78 #define SHORT_SYMLINK_LEN 128
79
80 struct shmem_xattr {
81         struct list_head list;  /* anchored by shmem_inode_info->xattr_list */
82         char *name;             /* xattr name */
83         size_t size;
84         char value[0];
85 };
86
87 /*
88  * shmem_fallocate and shmem_writepage communicate via inode->i_private
89  * (with i_mutex making sure that it has only one user at a time):
90  * we would prefer not to enlarge the shmem inode just for that.
91  */
92 struct shmem_falloc {
93         pgoff_t start;          /* start of range currently being fallocated */
94         pgoff_t next;           /* the next page offset to be fallocated */
95         pgoff_t nr_falloced;    /* how many new pages have been fallocated */
96         pgoff_t nr_unswapped;   /* how often writepage refused to swap out */
97 };
98
99 /* Flag allocation requirements to shmem_getpage */
100 enum sgp_type {
101         SGP_READ,       /* don't exceed i_size, don't allocate page */
102         SGP_CACHE,      /* don't exceed i_size, may allocate page */
103         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
104         SGP_WRITE,      /* may exceed i_size, may allocate !Uptodate page */
105         SGP_FALLOC,     /* like SGP_WRITE, but make existing page Uptodate */
106 };
107
108 #ifdef CONFIG_TMPFS
109 static unsigned long shmem_default_max_blocks(void)
110 {
111         return totalram_pages / 2;
112 }
113
114 static unsigned long shmem_default_max_inodes(void)
115 {
116         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
117 }
118 #endif
119
120 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122                                 struct shmem_inode_info *info, pgoff_t index);
123 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
124         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
125
126 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
127         struct page **pagep, enum sgp_type sgp, int *fault_type)
128 {
129         return shmem_getpage_gfp(inode, index, pagep, sgp,
130                         mapping_gfp_mask(inode->i_mapping), fault_type);
131 }
132
133 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
134 {
135         return sb->s_fs_info;
136 }
137
138 /*
139  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
140  * for shared memory and for shared anonymous (/dev/zero) mappings
141  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
142  * consistent with the pre-accounting of private mappings ...
143  */
144 static inline int shmem_acct_size(unsigned long flags, loff_t size)
145 {
146         return (flags & VM_NORESERVE) ?
147                 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
148 }
149
150 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
151 {
152         if (!(flags & VM_NORESERVE))
153                 vm_unacct_memory(VM_ACCT(size));
154 }
155
156 /*
157  * ... whereas tmpfs objects are accounted incrementally as
158  * pages are allocated, in order to allow huge sparse files.
159  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
160  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
161  */
162 static inline int shmem_acct_block(unsigned long flags)
163 {
164         return (flags & VM_NORESERVE) ?
165                 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
166 }
167
168 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
169 {
170         if (flags & VM_NORESERVE)
171                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
172 }
173
174 static const struct super_operations shmem_ops;
175 static const struct address_space_operations shmem_aops;
176 static const struct file_operations shmem_file_operations;
177 static const struct inode_operations shmem_inode_operations;
178 static const struct inode_operations shmem_dir_inode_operations;
179 static const struct inode_operations shmem_special_inode_operations;
180 static const struct vm_operations_struct shmem_vm_ops;
181
182 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
183         .ra_pages       = 0,    /* No readahead */
184         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
185 };
186
187 static LIST_HEAD(shmem_swaplist);
188 static DEFINE_MUTEX(shmem_swaplist_mutex);
189
190 static int shmem_reserve_inode(struct super_block *sb)
191 {
192         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
193         if (sbinfo->max_inodes) {
194                 spin_lock(&sbinfo->stat_lock);
195                 if (!sbinfo->free_inodes) {
196                         spin_unlock(&sbinfo->stat_lock);
197                         return -ENOSPC;
198                 }
199                 sbinfo->free_inodes--;
200                 spin_unlock(&sbinfo->stat_lock);
201         }
202         return 0;
203 }
204
205 static void shmem_free_inode(struct super_block *sb)
206 {
207         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
208         if (sbinfo->max_inodes) {
209                 spin_lock(&sbinfo->stat_lock);
210                 sbinfo->free_inodes++;
211                 spin_unlock(&sbinfo->stat_lock);
212         }
213 }
214
215 /**
216  * shmem_recalc_inode - recalculate the block usage of an inode
217  * @inode: inode to recalc
218  *
219  * We have to calculate the free blocks since the mm can drop
220  * undirtied hole pages behind our back.
221  *
222  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
223  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
224  *
225  * It has to be called with the spinlock held.
226  */
227 static void shmem_recalc_inode(struct inode *inode)
228 {
229         struct shmem_inode_info *info = SHMEM_I(inode);
230         long freed;
231
232         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
233         if (freed > 0) {
234                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
235                 if (sbinfo->max_blocks)
236                         percpu_counter_add(&sbinfo->used_blocks, -freed);
237                 info->alloced -= freed;
238                 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
239                 shmem_unacct_blocks(info->flags, freed);
240         }
241 }
242
243 /*
244  * Replace item expected in radix tree by a new item, while holding tree lock.
245  */
246 static int shmem_radix_tree_replace(struct address_space *mapping,
247                         pgoff_t index, void *expected, void *replacement)
248 {
249         void **pslot;
250         void *item = NULL;
251
252         VM_BUG_ON(!expected);
253         pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
254         if (pslot)
255                 item = radix_tree_deref_slot_protected(pslot,
256                                                         &mapping->tree_lock);
257         if (item != expected)
258                 return -ENOENT;
259         if (replacement)
260                 radix_tree_replace_slot(pslot, replacement);
261         else
262                 radix_tree_delete(&mapping->page_tree, index);
263         return 0;
264 }
265
266 /*
267  * Sometimes, before we decide whether to proceed or to fail, we must check
268  * that an entry was not already brought back from swap by a racing thread.
269  *
270  * Checking page is not enough: by the time a SwapCache page is locked, it
271  * might be reused, and again be SwapCache, using the same swap as before.
272  */
273 static bool shmem_confirm_swap(struct address_space *mapping,
274                                pgoff_t index, swp_entry_t swap)
275 {
276         void *item;
277
278         rcu_read_lock();
279         item = radix_tree_lookup(&mapping->page_tree, index);
280         rcu_read_unlock();
281         return item == swp_to_radix_entry(swap);
282 }
283
284 /*
285  * Like add_to_page_cache_locked, but error if expected item has gone.
286  */
287 static int shmem_add_to_page_cache(struct page *page,
288                                    struct address_space *mapping,
289                                    pgoff_t index, gfp_t gfp, void *expected)
290 {
291         int error;
292
293         VM_BUG_ON(!PageLocked(page));
294         VM_BUG_ON(!PageSwapBacked(page));
295
296         page_cache_get(page);
297         page->mapping = mapping;
298         page->index = index;
299
300         spin_lock_irq(&mapping->tree_lock);
301         if (!expected)
302                 error = radix_tree_insert(&mapping->page_tree, index, page);
303         else
304                 error = shmem_radix_tree_replace(mapping, index, expected,
305                                                                  page);
306         if (!error) {
307                 mapping->nrpages++;
308                 __inc_zone_page_state(page, NR_FILE_PAGES);
309                 __inc_zone_page_state(page, NR_SHMEM);
310                 spin_unlock_irq(&mapping->tree_lock);
311         } else {
312                 page->mapping = NULL;
313                 spin_unlock_irq(&mapping->tree_lock);
314                 page_cache_release(page);
315         }
316         return error;
317 }
318
319 /*
320  * Like delete_from_page_cache, but substitutes swap for page.
321  */
322 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
323 {
324         struct address_space *mapping = page->mapping;
325         int error;
326
327         spin_lock_irq(&mapping->tree_lock);
328         error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
329         page->mapping = NULL;
330         mapping->nrpages--;
331         __dec_zone_page_state(page, NR_FILE_PAGES);
332         __dec_zone_page_state(page, NR_SHMEM);
333         spin_unlock_irq(&mapping->tree_lock);
334         page_cache_release(page);
335         BUG_ON(error);
336 }
337
338 /*
339  * Like find_get_pages, but collecting swap entries as well as pages.
340  */
341 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
342                                         pgoff_t start, unsigned int nr_pages,
343                                         struct page **pages, pgoff_t *indices)
344 {
345         unsigned int i;
346         unsigned int ret;
347         unsigned int nr_found;
348
349         rcu_read_lock();
350 restart:
351         nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
352                                 (void ***)pages, indices, start, nr_pages);
353         ret = 0;
354         for (i = 0; i < nr_found; i++) {
355                 struct page *page;
356 repeat:
357                 page = radix_tree_deref_slot((void **)pages[i]);
358                 if (unlikely(!page))
359                         continue;
360                 if (radix_tree_exception(page)) {
361                         if (radix_tree_deref_retry(page))
362                                 goto restart;
363                         /*
364                          * Otherwise, we must be storing a swap entry
365                          * here as an exceptional entry: so return it
366                          * without attempting to raise page count.
367                          */
368                         goto export;
369                 }
370                 if (!page_cache_get_speculative(page))
371                         goto repeat;
372
373                 /* Has the page moved? */
374                 if (unlikely(page != *((void **)pages[i]))) {
375                         page_cache_release(page);
376                         goto repeat;
377                 }
378 export:
379                 indices[ret] = indices[i];
380                 pages[ret] = page;
381                 ret++;
382         }
383         if (unlikely(!ret && nr_found))
384                 goto restart;
385         rcu_read_unlock();
386         return ret;
387 }
388
389 /*
390  * Remove swap entry from radix tree, free the swap and its page cache.
391  */
392 static int shmem_free_swap(struct address_space *mapping,
393                            pgoff_t index, void *radswap)
394 {
395         int error;
396
397         spin_lock_irq(&mapping->tree_lock);
398         error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
399         spin_unlock_irq(&mapping->tree_lock);
400         if (!error)
401                 free_swap_and_cache(radix_to_swp_entry(radswap));
402         return error;
403 }
404
405 /*
406  * Pagevec may contain swap entries, so shuffle up pages before releasing.
407  */
408 static void shmem_deswap_pagevec(struct pagevec *pvec)
409 {
410         int i, j;
411
412         for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
413                 struct page *page = pvec->pages[i];
414                 if (!radix_tree_exceptional_entry(page))
415                         pvec->pages[j++] = page;
416         }
417         pvec->nr = j;
418 }
419
420 /*
421  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
422  */
423 void shmem_unlock_mapping(struct address_space *mapping)
424 {
425         struct pagevec pvec;
426         pgoff_t indices[PAGEVEC_SIZE];
427         pgoff_t index = 0;
428
429         pagevec_init(&pvec, 0);
430         /*
431          * Minor point, but we might as well stop if someone else SHM_LOCKs it.
432          */
433         while (!mapping_unevictable(mapping)) {
434                 /*
435                  * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
436                  * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
437                  */
438                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
439                                         PAGEVEC_SIZE, pvec.pages, indices);
440                 if (!pvec.nr)
441                         break;
442                 index = indices[pvec.nr - 1] + 1;
443                 shmem_deswap_pagevec(&pvec);
444                 check_move_unevictable_pages(pvec.pages, pvec.nr);
445                 pagevec_release(&pvec);
446                 cond_resched();
447         }
448 }
449
450 /*
451  * Remove range of pages and swap entries from radix tree, and free them.
452  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
453  */
454 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
455                                                                  bool unfalloc)
456 {
457         struct address_space *mapping = inode->i_mapping;
458         struct shmem_inode_info *info = SHMEM_I(inode);
459         pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
460         pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
461         unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
462         unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
463         struct pagevec pvec;
464         pgoff_t indices[PAGEVEC_SIZE];
465         long nr_swaps_freed = 0;
466         pgoff_t index;
467         int i;
468
469         if (lend == -1)
470                 end = -1;       /* unsigned, so actually very big */
471
472         pagevec_init(&pvec, 0);
473         index = start;
474         while (index < end) {
475                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
476                                 min(end - index, (pgoff_t)PAGEVEC_SIZE),
477                                                         pvec.pages, indices);
478                 if (!pvec.nr)
479                         break;
480                 mem_cgroup_uncharge_start();
481                 for (i = 0; i < pagevec_count(&pvec); i++) {
482                         struct page *page = pvec.pages[i];
483
484                         index = indices[i];
485                         if (index >= end)
486                                 break;
487
488                         if (radix_tree_exceptional_entry(page)) {
489                                 if (unfalloc)
490                                         continue;
491                                 nr_swaps_freed += !shmem_free_swap(mapping,
492                                                                 index, page);
493                                 continue;
494                         }
495
496                         if (!trylock_page(page))
497                                 continue;
498                         if (!unfalloc || !PageUptodate(page)) {
499                                 if (page->mapping == mapping) {
500                                         VM_BUG_ON(PageWriteback(page));
501                                         truncate_inode_page(mapping, page);
502                                 }
503                         }
504                         unlock_page(page);
505                 }
506                 shmem_deswap_pagevec(&pvec);
507                 pagevec_release(&pvec);
508                 mem_cgroup_uncharge_end();
509                 cond_resched();
510                 index++;
511         }
512
513         if (partial_start) {
514                 struct page *page = NULL;
515                 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
516                 if (page) {
517                         unsigned int top = PAGE_CACHE_SIZE;
518                         if (start > end) {
519                                 top = partial_end;
520                                 partial_end = 0;
521                         }
522                         zero_user_segment(page, partial_start, top);
523                         set_page_dirty(page);
524                         unlock_page(page);
525                         page_cache_release(page);
526                 }
527         }
528         if (partial_end) {
529                 struct page *page = NULL;
530                 shmem_getpage(inode, end, &page, SGP_READ, NULL);
531                 if (page) {
532                         zero_user_segment(page, 0, partial_end);
533                         set_page_dirty(page);
534                         unlock_page(page);
535                         page_cache_release(page);
536                 }
537         }
538         if (start >= end)
539                 return;
540
541         index = start;
542         for ( ; ; ) {
543                 cond_resched();
544                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
545                                 min(end - index, (pgoff_t)PAGEVEC_SIZE),
546                                                         pvec.pages, indices);
547                 if (!pvec.nr) {
548                         if (index == start || unfalloc)
549                                 break;
550                         index = start;
551                         continue;
552                 }
553                 if ((index == start || unfalloc) && indices[0] >= end) {
554                         shmem_deswap_pagevec(&pvec);
555                         pagevec_release(&pvec);
556                         break;
557                 }
558                 mem_cgroup_uncharge_start();
559                 for (i = 0; i < pagevec_count(&pvec); i++) {
560                         struct page *page = pvec.pages[i];
561
562                         index = indices[i];
563                         if (index >= end)
564                                 break;
565
566                         if (radix_tree_exceptional_entry(page)) {
567                                 if (unfalloc)
568                                         continue;
569                                 nr_swaps_freed += !shmem_free_swap(mapping,
570                                                                 index, page);
571                                 continue;
572                         }
573
574                         lock_page(page);
575                         if (!unfalloc || !PageUptodate(page)) {
576                                 if (page->mapping == mapping) {
577                                         VM_BUG_ON(PageWriteback(page));
578                                         truncate_inode_page(mapping, page);
579                                 }
580                         }
581                         unlock_page(page);
582                 }
583                 shmem_deswap_pagevec(&pvec);
584                 pagevec_release(&pvec);
585                 mem_cgroup_uncharge_end();
586                 index++;
587         }
588
589         spin_lock(&info->lock);
590         info->swapped -= nr_swaps_freed;
591         shmem_recalc_inode(inode);
592         spin_unlock(&info->lock);
593 }
594
595 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
596 {
597         shmem_undo_range(inode, lstart, lend, false);
598         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
599 }
600 EXPORT_SYMBOL_GPL(shmem_truncate_range);
601
602 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
603 {
604         struct inode *inode = dentry->d_inode;
605         int error;
606
607         error = inode_change_ok(inode, attr);
608         if (error)
609                 return error;
610
611         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
612                 loff_t oldsize = inode->i_size;
613                 loff_t newsize = attr->ia_size;
614
615                 if (newsize != oldsize) {
616                         i_size_write(inode, newsize);
617                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
618                 }
619                 if (newsize < oldsize) {
620                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
621                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
622                         shmem_truncate_range(inode, newsize, (loff_t)-1);
623                         /* unmap again to remove racily COWed private pages */
624                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
625                 }
626         }
627
628         setattr_copy(inode, attr);
629 #ifdef CONFIG_TMPFS_POSIX_ACL
630         if (attr->ia_valid & ATTR_MODE)
631                 error = generic_acl_chmod(inode);
632 #endif
633         return error;
634 }
635
636 static void shmem_evict_inode(struct inode *inode)
637 {
638         struct shmem_inode_info *info = SHMEM_I(inode);
639         struct shmem_xattr *xattr, *nxattr;
640
641         if (inode->i_mapping->a_ops == &shmem_aops) {
642                 shmem_unacct_size(info->flags, inode->i_size);
643                 inode->i_size = 0;
644                 shmem_truncate_range(inode, 0, (loff_t)-1);
645                 if (!list_empty(&info->swaplist)) {
646                         mutex_lock(&shmem_swaplist_mutex);
647                         list_del_init(&info->swaplist);
648                         mutex_unlock(&shmem_swaplist_mutex);
649                 }
650         } else
651                 kfree(info->symlink);
652
653         list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
654                 kfree(xattr->name);
655                 kfree(xattr);
656         }
657         BUG_ON(inode->i_blocks);
658         shmem_free_inode(inode->i_sb);
659         clear_inode(inode);
660 }
661
662 /*
663  * If swap found in inode, free it and move page from swapcache to filecache.
664  */
665 static int shmem_unuse_inode(struct shmem_inode_info *info,
666                              swp_entry_t swap, struct page **pagep)
667 {
668         struct address_space *mapping = info->vfs_inode.i_mapping;
669         void *radswap;
670         pgoff_t index;
671         gfp_t gfp;
672         int error = 0;
673
674         radswap = swp_to_radix_entry(swap);
675         index = radix_tree_locate_item(&mapping->page_tree, radswap);
676         if (index == -1)
677                 return 0;
678
679         /*
680          * Move _head_ to start search for next from here.
681          * But be careful: shmem_evict_inode checks list_empty without taking
682          * mutex, and there's an instant in list_move_tail when info->swaplist
683          * would appear empty, if it were the only one on shmem_swaplist.
684          */
685         if (shmem_swaplist.next != &info->swaplist)
686                 list_move_tail(&shmem_swaplist, &info->swaplist);
687
688         gfp = mapping_gfp_mask(mapping);
689         if (shmem_should_replace_page(*pagep, gfp)) {
690                 mutex_unlock(&shmem_swaplist_mutex);
691                 error = shmem_replace_page(pagep, gfp, info, index);
692                 mutex_lock(&shmem_swaplist_mutex);
693                 /*
694                  * We needed to drop mutex to make that restrictive page
695                  * allocation, but the inode might have been freed while we
696                  * dropped it: although a racing shmem_evict_inode() cannot
697                  * complete without emptying the radix_tree, our page lock
698                  * on this swapcache page is not enough to prevent that -
699                  * free_swap_and_cache() of our swap entry will only
700                  * trylock_page(), removing swap from radix_tree whatever.
701                  *
702                  * We must not proceed to shmem_add_to_page_cache() if the
703                  * inode has been freed, but of course we cannot rely on
704                  * inode or mapping or info to check that.  However, we can
705                  * safely check if our swap entry is still in use (and here
706                  * it can't have got reused for another page): if it's still
707                  * in use, then the inode cannot have been freed yet, and we
708                  * can safely proceed (if it's no longer in use, that tells
709                  * nothing about the inode, but we don't need to unuse swap).
710                  */
711                 if (!page_swapcount(*pagep))
712                         error = -ENOENT;
713         }
714
715         /*
716          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
717          * but also to hold up shmem_evict_inode(): so inode cannot be freed
718          * beneath us (pagelock doesn't help until the page is in pagecache).
719          */
720         if (!error)
721                 error = shmem_add_to_page_cache(*pagep, mapping, index,
722                                                 GFP_NOWAIT, radswap);
723         if (error != -ENOMEM) {
724                 /*
725                  * Truncation and eviction use free_swap_and_cache(), which
726                  * only does trylock page: if we raced, best clean up here.
727                  */
728                 delete_from_swap_cache(*pagep);
729                 set_page_dirty(*pagep);
730                 if (!error) {
731                         spin_lock(&info->lock);
732                         info->swapped--;
733                         spin_unlock(&info->lock);
734                         swap_free(swap);
735                 }
736                 error = 1;      /* not an error, but entry was found */
737         }
738         return error;
739 }
740
741 /*
742  * Search through swapped inodes to find and replace swap by page.
743  */
744 int shmem_unuse(swp_entry_t swap, struct page *page)
745 {
746         struct list_head *this, *next;
747         struct shmem_inode_info *info;
748         int found = 0;
749         int error = 0;
750
751         /*
752          * There's a faint possibility that swap page was replaced before
753          * caller locked it: caller will come back later with the right page.
754          */
755         if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
756                 goto out;
757
758         /*
759          * Charge page using GFP_KERNEL while we can wait, before taking
760          * the shmem_swaplist_mutex which might hold up shmem_writepage().
761          * Charged back to the user (not to caller) when swap account is used.
762          */
763         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
764         if (error)
765                 goto out;
766         /* No radix_tree_preload: swap entry keeps a place for page in tree */
767
768         mutex_lock(&shmem_swaplist_mutex);
769         list_for_each_safe(this, next, &shmem_swaplist) {
770                 info = list_entry(this, struct shmem_inode_info, swaplist);
771                 if (info->swapped)
772                         found = shmem_unuse_inode(info, swap, &page);
773                 else
774                         list_del_init(&info->swaplist);
775                 cond_resched();
776                 if (found)
777                         break;
778         }
779         mutex_unlock(&shmem_swaplist_mutex);
780
781         if (found < 0)
782                 error = found;
783 out:
784         unlock_page(page);
785         page_cache_release(page);
786         return error;
787 }
788
789 /*
790  * Move the page from the page cache to the swap cache.
791  */
792 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
793 {
794         struct shmem_inode_info *info;
795         struct address_space *mapping;
796         struct inode *inode;
797         swp_entry_t swap;
798         pgoff_t index;
799
800         BUG_ON(!PageLocked(page));
801         mapping = page->mapping;
802         index = page->index;
803         inode = mapping->host;
804         info = SHMEM_I(inode);
805         if (info->flags & VM_LOCKED)
806                 goto redirty;
807         if (!total_swap_pages)
808                 goto redirty;
809
810         /*
811          * shmem_backing_dev_info's capabilities prevent regular writeback or
812          * sync from ever calling shmem_writepage; but a stacking filesystem
813          * might use ->writepage of its underlying filesystem, in which case
814          * tmpfs should write out to swap only in response to memory pressure,
815          * and not for the writeback threads or sync.
816          */
817         if (!wbc->for_reclaim) {
818                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
819                 goto redirty;
820         }
821
822         /*
823          * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
824          * value into swapfile.c, the only way we can correctly account for a
825          * fallocated page arriving here is now to initialize it and write it.
826          *
827          * That's okay for a page already fallocated earlier, but if we have
828          * not yet completed the fallocation, then (a) we want to keep track
829          * of this page in case we have to undo it, and (b) it may not be a
830          * good idea to continue anyway, once we're pushing into swap.  So
831          * reactivate the page, and let shmem_fallocate() quit when too many.
832          */
833         if (!PageUptodate(page)) {
834                 if (inode->i_private) {
835                         struct shmem_falloc *shmem_falloc;
836                         spin_lock(&inode->i_lock);
837                         shmem_falloc = inode->i_private;
838                         if (shmem_falloc &&
839                             index >= shmem_falloc->start &&
840                             index < shmem_falloc->next)
841                                 shmem_falloc->nr_unswapped++;
842                         else
843                                 shmem_falloc = NULL;
844                         spin_unlock(&inode->i_lock);
845                         if (shmem_falloc)
846                                 goto redirty;
847                 }
848                 clear_highpage(page);
849                 flush_dcache_page(page);
850                 SetPageUptodate(page);
851         }
852
853         swap = get_swap_page();
854         if (!swap.val)
855                 goto redirty;
856
857         /*
858          * Add inode to shmem_unuse()'s list of swapped-out inodes,
859          * if it's not already there.  Do it now before the page is
860          * moved to swap cache, when its pagelock no longer protects
861          * the inode from eviction.  But don't unlock the mutex until
862          * we've incremented swapped, because shmem_unuse_inode() will
863          * prune a !swapped inode from the swaplist under this mutex.
864          */
865         mutex_lock(&shmem_swaplist_mutex);
866         if (list_empty(&info->swaplist))
867                 list_add_tail(&info->swaplist, &shmem_swaplist);
868
869         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
870                 swap_shmem_alloc(swap);
871                 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
872
873                 spin_lock(&info->lock);
874                 info->swapped++;
875                 shmem_recalc_inode(inode);
876                 spin_unlock(&info->lock);
877
878                 mutex_unlock(&shmem_swaplist_mutex);
879                 BUG_ON(page_mapped(page));
880                 swap_writepage(page, wbc);
881                 return 0;
882         }
883
884         mutex_unlock(&shmem_swaplist_mutex);
885         swapcache_free(swap, NULL);
886 redirty:
887         set_page_dirty(page);
888         if (wbc->for_reclaim)
889                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
890         unlock_page(page);
891         return 0;
892 }
893
894 #ifdef CONFIG_NUMA
895 #ifdef CONFIG_TMPFS
896 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
897 {
898         char buffer[64];
899
900         if (!mpol || mpol->mode == MPOL_DEFAULT)
901                 return;         /* show nothing */
902
903         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
904
905         seq_printf(seq, ",mpol=%s", buffer);
906 }
907
908 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
909 {
910         struct mempolicy *mpol = NULL;
911         if (sbinfo->mpol) {
912                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
913                 mpol = sbinfo->mpol;
914                 mpol_get(mpol);
915                 spin_unlock(&sbinfo->stat_lock);
916         }
917         return mpol;
918 }
919 #endif /* CONFIG_TMPFS */
920
921 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
922                         struct shmem_inode_info *info, pgoff_t index)
923 {
924         struct mempolicy mpol, *spol;
925         struct vm_area_struct pvma;
926
927         spol = mpol_cond_copy(&mpol,
928                         mpol_shared_policy_lookup(&info->policy, index));
929
930         /* Create a pseudo vma that just contains the policy */
931         pvma.vm_start = 0;
932         pvma.vm_pgoff = index;
933         pvma.vm_ops = NULL;
934         pvma.vm_policy = spol;
935         return swapin_readahead(swap, gfp, &pvma, 0);
936 }
937
938 static struct page *shmem_alloc_page(gfp_t gfp,
939                         struct shmem_inode_info *info, pgoff_t index)
940 {
941         struct vm_area_struct pvma;
942
943         /* Create a pseudo vma that just contains the policy */
944         pvma.vm_start = 0;
945         pvma.vm_pgoff = index;
946         pvma.vm_ops = NULL;
947         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
948
949         /*
950          * alloc_page_vma() will drop the shared policy reference
951          */
952         return alloc_page_vma(gfp, &pvma, 0);
953 }
954 #else /* !CONFIG_NUMA */
955 #ifdef CONFIG_TMPFS
956 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
957 {
958 }
959 #endif /* CONFIG_TMPFS */
960
961 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
962                         struct shmem_inode_info *info, pgoff_t index)
963 {
964         return swapin_readahead(swap, gfp, NULL, 0);
965 }
966
967 static inline struct page *shmem_alloc_page(gfp_t gfp,
968                         struct shmem_inode_info *info, pgoff_t index)
969 {
970         return alloc_page(gfp);
971 }
972 #endif /* CONFIG_NUMA */
973
974 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
975 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
976 {
977         return NULL;
978 }
979 #endif
980
981 /*
982  * When a page is moved from swapcache to shmem filecache (either by the
983  * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
984  * shmem_unuse_inode()), it may have been read in earlier from swap, in
985  * ignorance of the mapping it belongs to.  If that mapping has special
986  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
987  * we may need to copy to a suitable page before moving to filecache.
988  *
989  * In a future release, this may well be extended to respect cpuset and
990  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
991  * but for now it is a simple matter of zone.
992  */
993 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
994 {
995         return page_zonenum(page) > gfp_zone(gfp);
996 }
997
998 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
999                                 struct shmem_inode_info *info, pgoff_t index)
1000 {
1001         struct page *oldpage, *newpage;
1002         struct address_space *swap_mapping;
1003         pgoff_t swap_index;
1004         int error;
1005
1006         oldpage = *pagep;
1007         swap_index = page_private(oldpage);
1008         swap_mapping = page_mapping(oldpage);
1009
1010         /*
1011          * We have arrived here because our zones are constrained, so don't
1012          * limit chance of success by further cpuset and node constraints.
1013          */
1014         gfp &= ~GFP_CONSTRAINT_MASK;
1015         newpage = shmem_alloc_page(gfp, info, index);
1016         if (!newpage)
1017                 return -ENOMEM;
1018
1019         page_cache_get(newpage);
1020         copy_highpage(newpage, oldpage);
1021         flush_dcache_page(newpage);
1022
1023         __set_page_locked(newpage);
1024         SetPageUptodate(newpage);
1025         SetPageSwapBacked(newpage);
1026         set_page_private(newpage, swap_index);
1027         SetPageSwapCache(newpage);
1028
1029         /*
1030          * Our caller will very soon move newpage out of swapcache, but it's
1031          * a nice clean interface for us to replace oldpage by newpage there.
1032          */
1033         spin_lock_irq(&swap_mapping->tree_lock);
1034         error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1035                                                                    newpage);
1036         if (!error) {
1037                 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1038                 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1039         }
1040         spin_unlock_irq(&swap_mapping->tree_lock);
1041
1042         if (unlikely(error)) {
1043                 /*
1044                  * Is this possible?  I think not, now that our callers check
1045                  * both PageSwapCache and page_private after getting page lock;
1046                  * but be defensive.  Reverse old to newpage for clear and free.
1047                  */
1048                 oldpage = newpage;
1049         } else {
1050                 mem_cgroup_replace_page_cache(oldpage, newpage);
1051                 lru_cache_add_anon(newpage);
1052                 *pagep = newpage;
1053         }
1054
1055         ClearPageSwapCache(oldpage);
1056         set_page_private(oldpage, 0);
1057
1058         unlock_page(oldpage);
1059         page_cache_release(oldpage);
1060         page_cache_release(oldpage);
1061         return error;
1062 }
1063
1064 /*
1065  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1066  *
1067  * If we allocate a new one we do not mark it dirty. That's up to the
1068  * vm. If we swap it in we mark it dirty since we also free the swap
1069  * entry since a page cannot live in both the swap and page cache
1070  */
1071 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1072         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1073 {
1074         struct address_space *mapping = inode->i_mapping;
1075         struct shmem_inode_info *info;
1076         struct shmem_sb_info *sbinfo;
1077         struct page *page;
1078         swp_entry_t swap;
1079         int error;
1080         int once = 0;
1081         int alloced = 0;
1082
1083         if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1084                 return -EFBIG;
1085 repeat:
1086         swap.val = 0;
1087         page = find_lock_page(mapping, index);
1088         if (radix_tree_exceptional_entry(page)) {
1089                 swap = radix_to_swp_entry(page);
1090                 page = NULL;
1091         }
1092
1093         if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1094             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1095                 error = -EINVAL;
1096                 goto failed;
1097         }
1098
1099         /* fallocated page? */
1100         if (page && !PageUptodate(page)) {
1101                 if (sgp != SGP_READ)
1102                         goto clear;
1103                 unlock_page(page);
1104                 page_cache_release(page);
1105                 page = NULL;
1106         }
1107         if (page || (sgp == SGP_READ && !swap.val)) {
1108                 *pagep = page;
1109                 return 0;
1110         }
1111
1112         /*
1113          * Fast cache lookup did not find it:
1114          * bring it back from swap or allocate.
1115          */
1116         info = SHMEM_I(inode);
1117         sbinfo = SHMEM_SB(inode->i_sb);
1118
1119         if (swap.val) {
1120                 /* Look it up and read it in.. */
1121                 page = lookup_swap_cache(swap);
1122                 if (!page) {
1123                         /* here we actually do the io */
1124                         if (fault_type)
1125                                 *fault_type |= VM_FAULT_MAJOR;
1126                         page = shmem_swapin(swap, gfp, info, index);
1127                         if (!page) {
1128                                 error = -ENOMEM;
1129                                 goto failed;
1130                         }
1131                 }
1132
1133                 /* We have to do this with page locked to prevent races */
1134                 lock_page(page);
1135                 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1136                     !shmem_confirm_swap(mapping, index, swap)) {
1137                         error = -EEXIST;        /* try again */
1138                         goto unlock;
1139                 }
1140                 if (!PageUptodate(page)) {
1141                         error = -EIO;
1142                         goto failed;
1143                 }
1144                 wait_on_page_writeback(page);
1145
1146                 if (shmem_should_replace_page(page, gfp)) {
1147                         error = shmem_replace_page(&page, gfp, info, index);
1148                         if (error)
1149                                 goto failed;
1150                 }
1151
1152                 error = mem_cgroup_cache_charge(page, current->mm,
1153                                                 gfp & GFP_RECLAIM_MASK);
1154                 if (!error) {
1155                         error = shmem_add_to_page_cache(page, mapping, index,
1156                                                 gfp, swp_to_radix_entry(swap));
1157                         /* We already confirmed swap, and make no allocation */
1158                         VM_BUG_ON(error);
1159                 }
1160                 if (error)
1161                         goto failed;
1162
1163                 spin_lock(&info->lock);
1164                 info->swapped--;
1165                 shmem_recalc_inode(inode);
1166                 spin_unlock(&info->lock);
1167
1168                 delete_from_swap_cache(page);
1169                 set_page_dirty(page);
1170                 swap_free(swap);
1171
1172         } else {
1173                 if (shmem_acct_block(info->flags)) {
1174                         error = -ENOSPC;
1175                         goto failed;
1176                 }
1177                 if (sbinfo->max_blocks) {
1178                         if (percpu_counter_compare(&sbinfo->used_blocks,
1179                                                 sbinfo->max_blocks) >= 0) {
1180                                 error = -ENOSPC;
1181                                 goto unacct;
1182                         }
1183                         percpu_counter_inc(&sbinfo->used_blocks);
1184                 }
1185
1186                 page = shmem_alloc_page(gfp, info, index);
1187                 if (!page) {
1188                         error = -ENOMEM;
1189                         goto decused;
1190                 }
1191
1192                 SetPageSwapBacked(page);
1193                 __set_page_locked(page);
1194                 error = mem_cgroup_cache_charge(page, current->mm,
1195                                                 gfp & GFP_RECLAIM_MASK);
1196                 if (error)
1197                         goto decused;
1198                 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
1199                 if (!error) {
1200                         error = shmem_add_to_page_cache(page, mapping, index,
1201                                                         gfp, NULL);
1202                         radix_tree_preload_end();
1203                 }
1204                 if (error) {
1205                         mem_cgroup_uncharge_cache_page(page);
1206                         goto decused;
1207                 }
1208                 lru_cache_add_anon(page);
1209
1210                 spin_lock(&info->lock);
1211                 info->alloced++;
1212                 inode->i_blocks += BLOCKS_PER_PAGE;
1213                 shmem_recalc_inode(inode);
1214                 spin_unlock(&info->lock);
1215                 alloced = true;
1216
1217                 /*
1218                  * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1219                  */
1220                 if (sgp == SGP_FALLOC)
1221                         sgp = SGP_WRITE;
1222 clear:
1223                 /*
1224                  * Let SGP_WRITE caller clear ends if write does not fill page;
1225                  * but SGP_FALLOC on a page fallocated earlier must initialize
1226                  * it now, lest undo on failure cancel our earlier guarantee.
1227                  */
1228                 if (sgp != SGP_WRITE) {
1229                         clear_highpage(page);
1230                         flush_dcache_page(page);
1231                         SetPageUptodate(page);
1232                 }
1233                 if (sgp == SGP_DIRTY)
1234                         set_page_dirty(page);
1235         }
1236
1237         /* Perhaps the file has been truncated since we checked */
1238         if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1239             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1240                 error = -EINVAL;
1241                 if (alloced)
1242                         goto trunc;
1243                 else
1244                         goto failed;
1245         }
1246         *pagep = page;
1247         return 0;
1248
1249         /*
1250          * Error recovery.
1251          */
1252 trunc:
1253         info = SHMEM_I(inode);
1254         ClearPageDirty(page);
1255         delete_from_page_cache(page);
1256         spin_lock(&info->lock);
1257         info->alloced--;
1258         inode->i_blocks -= BLOCKS_PER_PAGE;
1259         spin_unlock(&info->lock);
1260 decused:
1261         sbinfo = SHMEM_SB(inode->i_sb);
1262         if (sbinfo->max_blocks)
1263                 percpu_counter_add(&sbinfo->used_blocks, -1);
1264 unacct:
1265         shmem_unacct_blocks(info->flags, 1);
1266 failed:
1267         if (swap.val && error != -EINVAL &&
1268             !shmem_confirm_swap(mapping, index, swap))
1269                 error = -EEXIST;
1270 unlock:
1271         if (page) {
1272                 unlock_page(page);
1273                 page_cache_release(page);
1274         }
1275         if (error == -ENOSPC && !once++) {
1276                 info = SHMEM_I(inode);
1277                 spin_lock(&info->lock);
1278                 shmem_recalc_inode(inode);
1279                 spin_unlock(&info->lock);
1280                 goto repeat;
1281         }
1282         if (error == -EEXIST)   /* from above or from radix_tree_insert */
1283                 goto repeat;
1284         return error;
1285 }
1286
1287 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1288 {
1289         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1290         int error;
1291         int ret = VM_FAULT_LOCKED;
1292
1293         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1294         if (error)
1295                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1296
1297         if (ret & VM_FAULT_MAJOR) {
1298                 count_vm_event(PGMAJFAULT);
1299                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1300         }
1301         return ret;
1302 }
1303
1304 #ifdef CONFIG_NUMA
1305 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1306 {
1307         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1308         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1309 }
1310
1311 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1312                                           unsigned long addr)
1313 {
1314         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1315         pgoff_t index;
1316
1317         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1318         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1319 }
1320 #endif
1321
1322 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1323 {
1324         struct inode *inode = file->f_path.dentry->d_inode;
1325         struct shmem_inode_info *info = SHMEM_I(inode);
1326         int retval = -ENOMEM;
1327
1328         spin_lock(&info->lock);
1329         if (lock && !(info->flags & VM_LOCKED)) {
1330                 if (!user_shm_lock(inode->i_size, user))
1331                         goto out_nomem;
1332                 info->flags |= VM_LOCKED;
1333                 mapping_set_unevictable(file->f_mapping);
1334         }
1335         if (!lock && (info->flags & VM_LOCKED) && user) {
1336                 user_shm_unlock(inode->i_size, user);
1337                 info->flags &= ~VM_LOCKED;
1338                 mapping_clear_unevictable(file->f_mapping);
1339         }
1340         retval = 0;
1341
1342 out_nomem:
1343         spin_unlock(&info->lock);
1344         return retval;
1345 }
1346
1347 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1348 {
1349         file_accessed(file);
1350         vma->vm_ops = &shmem_vm_ops;
1351         vma->vm_flags |= VM_CAN_NONLINEAR;
1352         return 0;
1353 }
1354
1355 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1356                                      umode_t mode, dev_t dev, unsigned long flags)
1357 {
1358         struct inode *inode;
1359         struct shmem_inode_info *info;
1360         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1361
1362         if (shmem_reserve_inode(sb))
1363                 return NULL;
1364
1365         inode = new_inode(sb);
1366         if (inode) {
1367                 inode->i_ino = get_next_ino();
1368                 inode_init_owner(inode, dir, mode);
1369                 inode->i_blocks = 0;
1370                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1371                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1372                 inode->i_generation = get_seconds();
1373                 info = SHMEM_I(inode);
1374                 memset(info, 0, (char *)inode - (char *)info);
1375                 spin_lock_init(&info->lock);
1376                 info->flags = flags & VM_NORESERVE;
1377                 INIT_LIST_HEAD(&info->swaplist);
1378                 INIT_LIST_HEAD(&info->xattr_list);
1379                 cache_no_acl(inode);
1380
1381                 switch (mode & S_IFMT) {
1382                 default:
1383                         inode->i_op = &shmem_special_inode_operations;
1384                         init_special_inode(inode, mode, dev);
1385                         break;
1386                 case S_IFREG:
1387                         inode->i_mapping->a_ops = &shmem_aops;
1388                         inode->i_op = &shmem_inode_operations;
1389                         inode->i_fop = &shmem_file_operations;
1390                         mpol_shared_policy_init(&info->policy,
1391                                                  shmem_get_sbmpol(sbinfo));
1392                         break;
1393                 case S_IFDIR:
1394                         inc_nlink(inode);
1395                         /* Some things misbehave if size == 0 on a directory */
1396                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1397                         inode->i_op = &shmem_dir_inode_operations;
1398                         inode->i_fop = &simple_dir_operations;
1399                         break;
1400                 case S_IFLNK:
1401                         /*
1402                          * Must not load anything in the rbtree,
1403                          * mpol_free_shared_policy will not be called.
1404                          */
1405                         mpol_shared_policy_init(&info->policy, NULL);
1406                         break;
1407                 }
1408         } else
1409                 shmem_free_inode(sb);
1410         return inode;
1411 }
1412
1413 #ifdef CONFIG_TMPFS
1414 static const struct inode_operations shmem_symlink_inode_operations;
1415 static const struct inode_operations shmem_short_symlink_operations;
1416
1417 #ifdef CONFIG_TMPFS_XATTR
1418 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1419 #else
1420 #define shmem_initxattrs NULL
1421 #endif
1422
1423 static int
1424 shmem_write_begin(struct file *file, struct address_space *mapping,
1425                         loff_t pos, unsigned len, unsigned flags,
1426                         struct page **pagep, void **fsdata)
1427 {
1428         struct inode *inode = mapping->host;
1429         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1430         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1431 }
1432
1433 static int
1434 shmem_write_end(struct file *file, struct address_space *mapping,
1435                         loff_t pos, unsigned len, unsigned copied,
1436                         struct page *page, void *fsdata)
1437 {
1438         struct inode *inode = mapping->host;
1439
1440         if (pos + copied > inode->i_size)
1441                 i_size_write(inode, pos + copied);
1442
1443         if (!PageUptodate(page)) {
1444                 if (copied < PAGE_CACHE_SIZE) {
1445                         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1446                         zero_user_segments(page, 0, from,
1447                                         from + copied, PAGE_CACHE_SIZE);
1448                 }
1449                 SetPageUptodate(page);
1450         }
1451         set_page_dirty(page);
1452         unlock_page(page);
1453         page_cache_release(page);
1454
1455         return copied;
1456 }
1457
1458 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1459 {
1460         struct inode *inode = filp->f_path.dentry->d_inode;
1461         struct address_space *mapping = inode->i_mapping;
1462         pgoff_t index;
1463         unsigned long offset;
1464         enum sgp_type sgp = SGP_READ;
1465
1466         /*
1467          * Might this read be for a stacking filesystem?  Then when reading
1468          * holes of a sparse file, we actually need to allocate those pages,
1469          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1470          */
1471         if (segment_eq(get_fs(), KERNEL_DS))
1472                 sgp = SGP_DIRTY;
1473
1474         index = *ppos >> PAGE_CACHE_SHIFT;
1475         offset = *ppos & ~PAGE_CACHE_MASK;
1476
1477         for (;;) {
1478                 struct page *page = NULL;
1479                 pgoff_t end_index;
1480                 unsigned long nr, ret;
1481                 loff_t i_size = i_size_read(inode);
1482
1483                 end_index = i_size >> PAGE_CACHE_SHIFT;
1484                 if (index > end_index)
1485                         break;
1486                 if (index == end_index) {
1487                         nr = i_size & ~PAGE_CACHE_MASK;
1488                         if (nr <= offset)
1489                                 break;
1490                 }
1491
1492                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1493                 if (desc->error) {
1494                         if (desc->error == -EINVAL)
1495                                 desc->error = 0;
1496                         break;
1497                 }
1498                 if (page)
1499                         unlock_page(page);
1500
1501                 /*
1502                  * We must evaluate after, since reads (unlike writes)
1503                  * are called without i_mutex protection against truncate
1504                  */
1505                 nr = PAGE_CACHE_SIZE;
1506                 i_size = i_size_read(inode);
1507                 end_index = i_size >> PAGE_CACHE_SHIFT;
1508                 if (index == end_index) {
1509                         nr = i_size & ~PAGE_CACHE_MASK;
1510                         if (nr <= offset) {
1511                                 if (page)
1512                                         page_cache_release(page);
1513                                 break;
1514                         }
1515                 }
1516                 nr -= offset;
1517
1518                 if (page) {
1519                         /*
1520                          * If users can be writing to this page using arbitrary
1521                          * virtual addresses, take care about potential aliasing
1522                          * before reading the page on the kernel side.
1523                          */
1524                         if (mapping_writably_mapped(mapping))
1525                                 flush_dcache_page(page);
1526                         /*
1527                          * Mark the page accessed if we read the beginning.
1528                          */
1529                         if (!offset)
1530                                 mark_page_accessed(page);
1531                 } else {
1532                         page = ZERO_PAGE(0);
1533                         page_cache_get(page);
1534                 }
1535
1536                 /*
1537                  * Ok, we have the page, and it's up-to-date, so
1538                  * now we can copy it to user space...
1539                  *
1540                  * The actor routine returns how many bytes were actually used..
1541                  * NOTE! This may not be the same as how much of a user buffer
1542                  * we filled up (we may be padding etc), so we can only update
1543                  * "pos" here (the actor routine has to update the user buffer
1544                  * pointers and the remaining count).
1545                  */
1546                 ret = actor(desc, page, offset, nr);
1547                 offset += ret;
1548                 index += offset >> PAGE_CACHE_SHIFT;
1549                 offset &= ~PAGE_CACHE_MASK;
1550
1551                 page_cache_release(page);
1552                 if (ret != nr || !desc->count)
1553                         break;
1554
1555                 cond_resched();
1556         }
1557
1558         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1559         file_accessed(filp);
1560 }
1561
1562 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1563                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1564 {
1565         struct file *filp = iocb->ki_filp;
1566         ssize_t retval;
1567         unsigned long seg;
1568         size_t count;
1569         loff_t *ppos = &iocb->ki_pos;
1570
1571         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1572         if (retval)
1573                 return retval;
1574
1575         for (seg = 0; seg < nr_segs; seg++) {
1576                 read_descriptor_t desc;
1577
1578                 desc.written = 0;
1579                 desc.arg.buf = iov[seg].iov_base;
1580                 desc.count = iov[seg].iov_len;
1581                 if (desc.count == 0)
1582                         continue;
1583                 desc.error = 0;
1584                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1585                 retval += desc.written;
1586                 if (desc.error) {
1587                         retval = retval ?: desc.error;
1588                         break;
1589                 }
1590                 if (desc.count > 0)
1591                         break;
1592         }
1593         return retval;
1594 }
1595
1596 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1597                                 struct pipe_inode_info *pipe, size_t len,
1598                                 unsigned int flags)
1599 {
1600         struct address_space *mapping = in->f_mapping;
1601         struct inode *inode = mapping->host;
1602         unsigned int loff, nr_pages, req_pages;
1603         struct page *pages[PIPE_DEF_BUFFERS];
1604         struct partial_page partial[PIPE_DEF_BUFFERS];
1605         struct page *page;
1606         pgoff_t index, end_index;
1607         loff_t isize, left;
1608         int error, page_nr;
1609         struct splice_pipe_desc spd = {
1610                 .pages = pages,
1611                 .partial = partial,
1612                 .nr_pages_max = PIPE_DEF_BUFFERS,
1613                 .flags = flags,
1614                 .ops = &page_cache_pipe_buf_ops,
1615                 .spd_release = spd_release_page,
1616         };
1617
1618         isize = i_size_read(inode);
1619         if (unlikely(*ppos >= isize))
1620                 return 0;
1621
1622         left = isize - *ppos;
1623         if (unlikely(left < len))
1624                 len = left;
1625
1626         if (splice_grow_spd(pipe, &spd))
1627                 return -ENOMEM;
1628
1629         index = *ppos >> PAGE_CACHE_SHIFT;
1630         loff = *ppos & ~PAGE_CACHE_MASK;
1631         req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1632         nr_pages = min(req_pages, pipe->buffers);
1633
1634         spd.nr_pages = find_get_pages_contig(mapping, index,
1635                                                 nr_pages, spd.pages);
1636         index += spd.nr_pages;
1637         error = 0;
1638
1639         while (spd.nr_pages < nr_pages) {
1640                 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1641                 if (error)
1642                         break;
1643                 unlock_page(page);
1644                 spd.pages[spd.nr_pages++] = page;
1645                 index++;
1646         }
1647
1648         index = *ppos >> PAGE_CACHE_SHIFT;
1649         nr_pages = spd.nr_pages;
1650         spd.nr_pages = 0;
1651
1652         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1653                 unsigned int this_len;
1654
1655                 if (!len)
1656                         break;
1657
1658                 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1659                 page = spd.pages[page_nr];
1660
1661                 if (!PageUptodate(page) || page->mapping != mapping) {
1662                         error = shmem_getpage(inode, index, &page,
1663                                                         SGP_CACHE, NULL);
1664                         if (error)
1665                                 break;
1666                         unlock_page(page);
1667                         page_cache_release(spd.pages[page_nr]);
1668                         spd.pages[page_nr] = page;
1669                 }
1670
1671                 isize = i_size_read(inode);
1672                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1673                 if (unlikely(!isize || index > end_index))
1674                         break;
1675
1676                 if (end_index == index) {
1677                         unsigned int plen;
1678
1679                         plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1680                         if (plen <= loff)
1681                                 break;
1682
1683                         this_len = min(this_len, plen - loff);
1684                         len = this_len;
1685                 }
1686
1687                 spd.partial[page_nr].offset = loff;
1688                 spd.partial[page_nr].len = this_len;
1689                 len -= this_len;
1690                 loff = 0;
1691                 spd.nr_pages++;
1692                 index++;
1693         }
1694
1695         while (page_nr < nr_pages)
1696                 page_cache_release(spd.pages[page_nr++]);
1697
1698         if (spd.nr_pages)
1699                 error = splice_to_pipe(pipe, &spd);
1700
1701         splice_shrink_spd(&spd);
1702
1703         if (error > 0) {
1704                 *ppos += error;
1705                 file_accessed(in);
1706         }
1707         return error;
1708 }
1709
1710 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1711                                                          loff_t len)
1712 {
1713         struct inode *inode = file->f_path.dentry->d_inode;
1714         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1715         struct shmem_falloc shmem_falloc;
1716         pgoff_t start, index, end;
1717         int error;
1718
1719         mutex_lock(&inode->i_mutex);
1720
1721         if (mode & FALLOC_FL_PUNCH_HOLE) {
1722                 struct address_space *mapping = file->f_mapping;
1723                 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1724                 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1725
1726                 if ((u64)unmap_end > (u64)unmap_start)
1727                         unmap_mapping_range(mapping, unmap_start,
1728                                             1 + unmap_end - unmap_start, 0);
1729                 shmem_truncate_range(inode, offset, offset + len - 1);
1730                 /* No need to unmap again: hole-punching leaves COWed pages */
1731                 error = 0;
1732                 goto out;
1733         }
1734
1735         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1736         error = inode_newsize_ok(inode, offset + len);
1737         if (error)
1738                 goto out;
1739
1740         start = offset >> PAGE_CACHE_SHIFT;
1741         end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1742         /* Try to avoid a swapstorm if len is impossible to satisfy */
1743         if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1744                 error = -ENOSPC;
1745                 goto out;
1746         }
1747
1748         shmem_falloc.start = start;
1749         shmem_falloc.next  = start;
1750         shmem_falloc.nr_falloced = 0;
1751         shmem_falloc.nr_unswapped = 0;
1752         spin_lock(&inode->i_lock);
1753         inode->i_private = &shmem_falloc;
1754         spin_unlock(&inode->i_lock);
1755
1756         for (index = start; index < end; index++) {
1757                 struct page *page;
1758
1759                 /*
1760                  * Good, the fallocate(2) manpage permits EINTR: we may have
1761                  * been interrupted because we are using up too much memory.
1762                  */
1763                 if (signal_pending(current))
1764                         error = -EINTR;
1765                 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1766                         error = -ENOMEM;
1767                 else
1768                         error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1769                                                                         NULL);
1770                 if (error) {
1771                         /* Remove the !PageUptodate pages we added */
1772                         shmem_undo_range(inode,
1773                                 (loff_t)start << PAGE_CACHE_SHIFT,
1774                                 (loff_t)index << PAGE_CACHE_SHIFT, true);
1775                         goto undone;
1776                 }
1777
1778                 /*
1779                  * Inform shmem_writepage() how far we have reached.
1780                  * No need for lock or barrier: we have the page lock.
1781                  */
1782                 shmem_falloc.next++;
1783                 if (!PageUptodate(page))
1784                         shmem_falloc.nr_falloced++;
1785
1786                 /*
1787                  * If !PageUptodate, leave it that way so that freeable pages
1788                  * can be recognized if we need to rollback on error later.
1789                  * But set_page_dirty so that memory pressure will swap rather
1790                  * than free the pages we are allocating (and SGP_CACHE pages
1791                  * might still be clean: we now need to mark those dirty too).
1792                  */
1793                 set_page_dirty(page);
1794                 unlock_page(page);
1795                 page_cache_release(page);
1796                 cond_resched();
1797         }
1798
1799         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1800                 i_size_write(inode, offset + len);
1801         inode->i_ctime = CURRENT_TIME;
1802 undone:
1803         spin_lock(&inode->i_lock);
1804         inode->i_private = NULL;
1805         spin_unlock(&inode->i_lock);
1806 out:
1807         mutex_unlock(&inode->i_mutex);
1808         return error;
1809 }
1810
1811 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1812 {
1813         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1814
1815         buf->f_type = TMPFS_MAGIC;
1816         buf->f_bsize = PAGE_CACHE_SIZE;
1817         buf->f_namelen = NAME_MAX;
1818         if (sbinfo->max_blocks) {
1819                 buf->f_blocks = sbinfo->max_blocks;
1820                 buf->f_bavail =
1821                 buf->f_bfree  = sbinfo->max_blocks -
1822                                 percpu_counter_sum(&sbinfo->used_blocks);
1823         }
1824         if (sbinfo->max_inodes) {
1825                 buf->f_files = sbinfo->max_inodes;
1826                 buf->f_ffree = sbinfo->free_inodes;
1827         }
1828         /* else leave those fields 0 like simple_statfs */
1829         return 0;
1830 }
1831
1832 /*
1833  * File creation. Allocate an inode, and we're done..
1834  */
1835 static int
1836 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1837 {
1838         struct inode *inode;
1839         int error = -ENOSPC;
1840
1841         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1842         if (inode) {
1843                 error = security_inode_init_security(inode, dir,
1844                                                      &dentry->d_name,
1845                                                      shmem_initxattrs, NULL);
1846                 if (error) {
1847                         if (error != -EOPNOTSUPP) {
1848                                 iput(inode);
1849                                 return error;
1850                         }
1851                 }
1852 #ifdef CONFIG_TMPFS_POSIX_ACL
1853                 error = generic_acl_init(inode, dir);
1854                 if (error) {
1855                         iput(inode);
1856                         return error;
1857                 }
1858 #else
1859                 error = 0;
1860 #endif
1861                 dir->i_size += BOGO_DIRENT_SIZE;
1862                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1863                 d_instantiate(dentry, inode);
1864                 dget(dentry); /* Extra count - pin the dentry in core */
1865         }
1866         return error;
1867 }
1868
1869 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1870 {
1871         int error;
1872
1873         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1874                 return error;
1875         inc_nlink(dir);
1876         return 0;
1877 }
1878
1879 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1880                 bool excl)
1881 {
1882         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1883 }
1884
1885 /*
1886  * Link a file..
1887  */
1888 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1889 {
1890         struct inode *inode = old_dentry->d_inode;
1891         int ret;
1892
1893         /*
1894          * No ordinary (disk based) filesystem counts links as inodes;
1895          * but each new link needs a new dentry, pinning lowmem, and
1896          * tmpfs dentries cannot be pruned until they are unlinked.
1897          */
1898         ret = shmem_reserve_inode(inode->i_sb);
1899         if (ret)
1900                 goto out;
1901
1902         dir->i_size += BOGO_DIRENT_SIZE;
1903         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1904         inc_nlink(inode);
1905         ihold(inode);   /* New dentry reference */
1906         dget(dentry);           /* Extra pinning count for the created dentry */
1907         d_instantiate(dentry, inode);
1908 out:
1909         return ret;
1910 }
1911
1912 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1913 {
1914         struct inode *inode = dentry->d_inode;
1915
1916         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1917                 shmem_free_inode(inode->i_sb);
1918
1919         dir->i_size -= BOGO_DIRENT_SIZE;
1920         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1921         drop_nlink(inode);
1922         dput(dentry);   /* Undo the count from "create" - this does all the work */
1923         return 0;
1924 }
1925
1926 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1927 {
1928         if (!simple_empty(dentry))
1929                 return -ENOTEMPTY;
1930
1931         drop_nlink(dentry->d_inode);
1932         drop_nlink(dir);
1933         return shmem_unlink(dir, dentry);
1934 }
1935
1936 /*
1937  * The VFS layer already does all the dentry stuff for rename,
1938  * we just have to decrement the usage count for the target if
1939  * it exists so that the VFS layer correctly free's it when it
1940  * gets overwritten.
1941  */
1942 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1943 {
1944         struct inode *inode = old_dentry->d_inode;
1945         int they_are_dirs = S_ISDIR(inode->i_mode);
1946
1947         if (!simple_empty(new_dentry))
1948                 return -ENOTEMPTY;
1949
1950         if (new_dentry->d_inode) {
1951                 (void) shmem_unlink(new_dir, new_dentry);
1952                 if (they_are_dirs)
1953                         drop_nlink(old_dir);
1954         } else if (they_are_dirs) {
1955                 drop_nlink(old_dir);
1956                 inc_nlink(new_dir);
1957         }
1958
1959         old_dir->i_size -= BOGO_DIRENT_SIZE;
1960         new_dir->i_size += BOGO_DIRENT_SIZE;
1961         old_dir->i_ctime = old_dir->i_mtime =
1962         new_dir->i_ctime = new_dir->i_mtime =
1963         inode->i_ctime = CURRENT_TIME;
1964         return 0;
1965 }
1966
1967 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1968 {
1969         int error;
1970         int len;
1971         struct inode *inode;
1972         struct page *page;
1973         char *kaddr;
1974         struct shmem_inode_info *info;
1975
1976         len = strlen(symname) + 1;
1977         if (len > PAGE_CACHE_SIZE)
1978                 return -ENAMETOOLONG;
1979
1980         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1981         if (!inode)
1982                 return -ENOSPC;
1983
1984         error = security_inode_init_security(inode, dir, &dentry->d_name,
1985                                              shmem_initxattrs, NULL);
1986         if (error) {
1987                 if (error != -EOPNOTSUPP) {
1988                         iput(inode);
1989                         return error;
1990                 }
1991                 error = 0;
1992         }
1993
1994         info = SHMEM_I(inode);
1995         inode->i_size = len-1;
1996         if (len <= SHORT_SYMLINK_LEN) {
1997                 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1998                 if (!info->symlink) {
1999                         iput(inode);
2000                         return -ENOMEM;
2001                 }
2002                 inode->i_op = &shmem_short_symlink_operations;
2003         } else {
2004                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2005                 if (error) {
2006                         iput(inode);
2007                         return error;
2008                 }
2009                 inode->i_mapping->a_ops = &shmem_aops;
2010                 inode->i_op = &shmem_symlink_inode_operations;
2011                 kaddr = kmap_atomic(page);
2012                 memcpy(kaddr, symname, len);
2013                 kunmap_atomic(kaddr);
2014                 SetPageUptodate(page);
2015                 set_page_dirty(page);
2016                 unlock_page(page);
2017                 page_cache_release(page);
2018         }
2019         dir->i_size += BOGO_DIRENT_SIZE;
2020         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2021         d_instantiate(dentry, inode);
2022         dget(dentry);
2023         return 0;
2024 }
2025
2026 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2027 {
2028         nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2029         return NULL;
2030 }
2031
2032 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2033 {
2034         struct page *page = NULL;
2035         int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2036         nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2037         if (page)
2038                 unlock_page(page);
2039         return page;
2040 }
2041
2042 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2043 {
2044         if (!IS_ERR(nd_get_link(nd))) {
2045                 struct page *page = cookie;
2046                 kunmap(page);
2047                 mark_page_accessed(page);
2048                 page_cache_release(page);
2049         }
2050 }
2051
2052 #ifdef CONFIG_TMPFS_XATTR
2053 /*
2054  * Superblocks without xattr inode operations may get some security.* xattr
2055  * support from the LSM "for free". As soon as we have any other xattrs
2056  * like ACLs, we also need to implement the security.* handlers at
2057  * filesystem level, though.
2058  */
2059
2060 /*
2061  * Allocate new xattr and copy in the value; but leave the name to callers.
2062  */
2063 static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2064 {
2065         struct shmem_xattr *new_xattr;
2066         size_t len;
2067
2068         /* wrap around? */
2069         len = sizeof(*new_xattr) + size;
2070         if (len <= sizeof(*new_xattr))
2071                 return NULL;
2072
2073         new_xattr = kmalloc(len, GFP_KERNEL);
2074         if (!new_xattr)
2075                 return NULL;
2076
2077         new_xattr->size = size;
2078         memcpy(new_xattr->value, value, size);
2079         return new_xattr;
2080 }
2081
2082 /*
2083  * Callback for security_inode_init_security() for acquiring xattrs.
2084  */
2085 static int shmem_initxattrs(struct inode *inode,
2086                             const struct xattr *xattr_array,
2087                             void *fs_info)
2088 {
2089         struct shmem_inode_info *info = SHMEM_I(inode);
2090         const struct xattr *xattr;
2091         struct shmem_xattr *new_xattr;
2092         size_t len;
2093
2094         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2095                 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2096                 if (!new_xattr)
2097                         return -ENOMEM;
2098
2099                 len = strlen(xattr->name) + 1;
2100                 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2101                                           GFP_KERNEL);
2102                 if (!new_xattr->name) {
2103                         kfree(new_xattr);
2104                         return -ENOMEM;
2105                 }
2106
2107                 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2108                        XATTR_SECURITY_PREFIX_LEN);
2109                 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2110                        xattr->name, len);
2111
2112                 spin_lock(&info->lock);
2113                 list_add(&new_xattr->list, &info->xattr_list);
2114                 spin_unlock(&info->lock);
2115         }
2116
2117         return 0;
2118 }
2119
2120 static int shmem_xattr_get(struct dentry *dentry, const char *name,
2121                            void *buffer, size_t size)
2122 {
2123         struct shmem_inode_info *info;
2124         struct shmem_xattr *xattr;
2125         int ret = -ENODATA;
2126
2127         info = SHMEM_I(dentry->d_inode);
2128
2129         spin_lock(&info->lock);
2130         list_for_each_entry(xattr, &info->xattr_list, list) {
2131                 if (strcmp(name, xattr->name))
2132                         continue;
2133
2134                 ret = xattr->size;
2135                 if (buffer) {
2136                         if (size < xattr->size)
2137                                 ret = -ERANGE;
2138                         else
2139                                 memcpy(buffer, xattr->value, xattr->size);
2140                 }
2141                 break;
2142         }
2143         spin_unlock(&info->lock);
2144         return ret;
2145 }
2146
2147 static int shmem_xattr_set(struct inode *inode, const char *name,
2148                            const void *value, size_t size, int flags)
2149 {
2150         struct shmem_inode_info *info = SHMEM_I(inode);
2151         struct shmem_xattr *xattr;
2152         struct shmem_xattr *new_xattr = NULL;
2153         int err = 0;
2154
2155         /* value == NULL means remove */
2156         if (value) {
2157                 new_xattr = shmem_xattr_alloc(value, size);
2158                 if (!new_xattr)
2159                         return -ENOMEM;
2160
2161                 new_xattr->name = kstrdup(name, GFP_KERNEL);
2162                 if (!new_xattr->name) {
2163                         kfree(new_xattr);
2164                         return -ENOMEM;
2165                 }
2166         }
2167
2168         spin_lock(&info->lock);
2169         list_for_each_entry(xattr, &info->xattr_list, list) {
2170                 if (!strcmp(name, xattr->name)) {
2171                         if (flags & XATTR_CREATE) {
2172                                 xattr = new_xattr;
2173                                 err = -EEXIST;
2174                         } else if (new_xattr) {
2175                                 list_replace(&xattr->list, &new_xattr->list);
2176                         } else {
2177                                 list_del(&xattr->list);
2178                         }
2179                         goto out;
2180                 }
2181         }
2182         if (flags & XATTR_REPLACE) {
2183                 xattr = new_xattr;
2184                 err = -ENODATA;
2185         } else {
2186                 list_add(&new_xattr->list, &info->xattr_list);
2187                 xattr = NULL;
2188         }
2189 out:
2190         spin_unlock(&info->lock);
2191         if (xattr)
2192                 kfree(xattr->name);
2193         kfree(xattr);
2194         return err;
2195 }
2196
2197 static const struct xattr_handler *shmem_xattr_handlers[] = {
2198 #ifdef CONFIG_TMPFS_POSIX_ACL
2199         &generic_acl_access_handler,
2200         &generic_acl_default_handler,
2201 #endif
2202         NULL
2203 };
2204
2205 static int shmem_xattr_validate(const char *name)
2206 {
2207         struct { const char *prefix; size_t len; } arr[] = {
2208                 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2209                 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2210         };
2211         int i;
2212
2213         for (i = 0; i < ARRAY_SIZE(arr); i++) {
2214                 size_t preflen = arr[i].len;
2215                 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2216                         if (!name[preflen])
2217                                 return -EINVAL;
2218                         return 0;
2219                 }
2220         }
2221         return -EOPNOTSUPP;
2222 }
2223
2224 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2225                               void *buffer, size_t size)
2226 {
2227         int err;
2228
2229         /*
2230          * If this is a request for a synthetic attribute in the system.*
2231          * namespace use the generic infrastructure to resolve a handler
2232          * for it via sb->s_xattr.
2233          */
2234         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2235                 return generic_getxattr(dentry, name, buffer, size);
2236
2237         err = shmem_xattr_validate(name);
2238         if (err)
2239                 return err;
2240
2241         return shmem_xattr_get(dentry, name, buffer, size);
2242 }
2243
2244 static int shmem_setxattr(struct dentry *dentry, const char *name,
2245                           const void *value, size_t size, int flags)
2246 {
2247         int err;
2248
2249         /*
2250          * If this is a request for a synthetic attribute in the system.*
2251          * namespace use the generic infrastructure to resolve a handler
2252          * for it via sb->s_xattr.
2253          */
2254         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2255                 return generic_setxattr(dentry, name, value, size, flags);
2256
2257         err = shmem_xattr_validate(name);
2258         if (err)
2259                 return err;
2260
2261         if (size == 0)
2262                 value = "";  /* empty EA, do not remove */
2263
2264         return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
2265
2266 }
2267
2268 static int shmem_removexattr(struct dentry *dentry, const char *name)
2269 {
2270         int err;
2271
2272         /*
2273          * If this is a request for a synthetic attribute in the system.*
2274          * namespace use the generic infrastructure to resolve a handler
2275          * for it via sb->s_xattr.
2276          */
2277         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2278                 return generic_removexattr(dentry, name);
2279
2280         err = shmem_xattr_validate(name);
2281         if (err)
2282                 return err;
2283
2284         return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
2285 }
2286
2287 static bool xattr_is_trusted(const char *name)
2288 {
2289         return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2290 }
2291
2292 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2293 {
2294         bool trusted = capable(CAP_SYS_ADMIN);
2295         struct shmem_xattr *xattr;
2296         struct shmem_inode_info *info;
2297         size_t used = 0;
2298
2299         info = SHMEM_I(dentry->d_inode);
2300
2301         spin_lock(&info->lock);
2302         list_for_each_entry(xattr, &info->xattr_list, list) {
2303                 size_t len;
2304
2305                 /* skip "trusted." attributes for unprivileged callers */
2306                 if (!trusted && xattr_is_trusted(xattr->name))
2307                         continue;
2308
2309                 len = strlen(xattr->name) + 1;
2310                 used += len;
2311                 if (buffer) {
2312                         if (size < used) {
2313                                 used = -ERANGE;
2314                                 break;
2315                         }
2316                         memcpy(buffer, xattr->name, len);
2317                         buffer += len;
2318                 }
2319         }
2320         spin_unlock(&info->lock);
2321
2322         return used;
2323 }
2324 #endif /* CONFIG_TMPFS_XATTR */
2325
2326 static const struct inode_operations shmem_short_symlink_operations = {
2327         .readlink       = generic_readlink,
2328         .follow_link    = shmem_follow_short_symlink,
2329 #ifdef CONFIG_TMPFS_XATTR
2330         .setxattr       = shmem_setxattr,
2331         .getxattr       = shmem_getxattr,
2332         .listxattr      = shmem_listxattr,
2333         .removexattr    = shmem_removexattr,
2334 #endif
2335 };
2336
2337 static const struct inode_operations shmem_symlink_inode_operations = {
2338         .readlink       = generic_readlink,
2339         .follow_link    = shmem_follow_link,
2340         .put_link       = shmem_put_link,
2341 #ifdef CONFIG_TMPFS_XATTR
2342         .setxattr       = shmem_setxattr,
2343         .getxattr       = shmem_getxattr,
2344         .listxattr      = shmem_listxattr,
2345         .removexattr    = shmem_removexattr,
2346 #endif
2347 };
2348
2349 static struct dentry *shmem_get_parent(struct dentry *child)
2350 {
2351         return ERR_PTR(-ESTALE);
2352 }
2353
2354 static int shmem_match(struct inode *ino, void *vfh)
2355 {
2356         __u32 *fh = vfh;
2357         __u64 inum = fh[2];
2358         inum = (inum << 32) | fh[1];
2359         return ino->i_ino == inum && fh[0] == ino->i_generation;
2360 }
2361
2362 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2363                 struct fid *fid, int fh_len, int fh_type)
2364 {
2365         struct inode *inode;
2366         struct dentry *dentry = NULL;
2367         u64 inum = fid->raw[2];
2368         inum = (inum << 32) | fid->raw[1];
2369
2370         if (fh_len < 3)
2371                 return NULL;
2372
2373         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2374                         shmem_match, fid->raw);
2375         if (inode) {
2376                 dentry = d_find_alias(inode);
2377                 iput(inode);
2378         }
2379
2380         return dentry;
2381 }
2382
2383 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2384                                 struct inode *parent)
2385 {
2386         if (*len < 3) {
2387                 *len = 3;
2388                 return 255;
2389         }
2390
2391         if (inode_unhashed(inode)) {
2392                 /* Unfortunately insert_inode_hash is not idempotent,
2393                  * so as we hash inodes here rather than at creation
2394                  * time, we need a lock to ensure we only try
2395                  * to do it once
2396                  */
2397                 static DEFINE_SPINLOCK(lock);
2398                 spin_lock(&lock);
2399                 if (inode_unhashed(inode))
2400                         __insert_inode_hash(inode,
2401                                             inode->i_ino + inode->i_generation);
2402                 spin_unlock(&lock);
2403         }
2404
2405         fh[0] = inode->i_generation;
2406         fh[1] = inode->i_ino;
2407         fh[2] = ((__u64)inode->i_ino) >> 32;
2408
2409         *len = 3;
2410         return 1;
2411 }
2412
2413 static const struct export_operations shmem_export_ops = {
2414         .get_parent     = shmem_get_parent,
2415         .encode_fh      = shmem_encode_fh,
2416         .fh_to_dentry   = shmem_fh_to_dentry,
2417 };
2418
2419 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2420                                bool remount)
2421 {
2422         char *this_char, *value, *rest;
2423         uid_t uid;
2424         gid_t gid;
2425
2426         while (options != NULL) {
2427                 this_char = options;
2428                 for (;;) {
2429                         /*
2430                          * NUL-terminate this option: unfortunately,
2431                          * mount options form a comma-separated list,
2432                          * but mpol's nodelist may also contain commas.
2433                          */
2434                         options = strchr(options, ',');
2435                         if (options == NULL)
2436                                 break;
2437                         options++;
2438                         if (!isdigit(*options)) {
2439                                 options[-1] = '\0';
2440                                 break;
2441                         }
2442                 }
2443                 if (!*this_char)
2444                         continue;
2445                 if ((value = strchr(this_char,'=')) != NULL) {
2446                         *value++ = 0;
2447                 } else {
2448                         printk(KERN_ERR
2449                             "tmpfs: No value for mount option '%s'\n",
2450                             this_char);
2451                         return 1;
2452                 }
2453
2454                 if (!strcmp(this_char,"size")) {
2455                         unsigned long long size;
2456                         size = memparse(value,&rest);
2457                         if (*rest == '%') {
2458                                 size <<= PAGE_SHIFT;
2459                                 size *= totalram_pages;
2460                                 do_div(size, 100);
2461                                 rest++;
2462                         }
2463                         if (*rest)
2464                                 goto bad_val;
2465                         sbinfo->max_blocks =
2466                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2467                 } else if (!strcmp(this_char,"nr_blocks")) {
2468                         sbinfo->max_blocks = memparse(value, &rest);
2469                         if (*rest)
2470                                 goto bad_val;
2471                 } else if (!strcmp(this_char,"nr_inodes")) {
2472                         sbinfo->max_inodes = memparse(value, &rest);
2473                         if (*rest)
2474                                 goto bad_val;
2475                 } else if (!strcmp(this_char,"mode")) {
2476                         if (remount)
2477                                 continue;
2478                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2479                         if (*rest)
2480                                 goto bad_val;
2481                 } else if (!strcmp(this_char,"uid")) {
2482                         if (remount)
2483                                 continue;
2484                         uid = simple_strtoul(value, &rest, 0);
2485                         if (*rest)
2486                                 goto bad_val;
2487                         sbinfo->uid = make_kuid(current_user_ns(), uid);
2488                         if (!uid_valid(sbinfo->uid))
2489                                 goto bad_val;
2490                 } else if (!strcmp(this_char,"gid")) {
2491                         if (remount)
2492                                 continue;
2493                         gid = simple_strtoul(value, &rest, 0);
2494                         if (*rest)
2495                                 goto bad_val;
2496                         sbinfo->gid = make_kgid(current_user_ns(), gid);
2497                         if (!gid_valid(sbinfo->gid))
2498                                 goto bad_val;
2499                 } else if (!strcmp(this_char,"mpol")) {
2500                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2501                                 goto bad_val;
2502                 } else {
2503                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2504                                this_char);
2505                         return 1;
2506                 }
2507         }
2508         return 0;
2509
2510 bad_val:
2511         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2512                value, this_char);
2513         return 1;
2514
2515 }
2516
2517 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2518 {
2519         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2520         struct shmem_sb_info config = *sbinfo;
2521         unsigned long inodes;
2522         int error = -EINVAL;
2523
2524         if (shmem_parse_options(data, &config, true))
2525                 return error;
2526
2527         spin_lock(&sbinfo->stat_lock);
2528         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2529         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2530                 goto out;
2531         if (config.max_inodes < inodes)
2532                 goto out;
2533         /*
2534          * Those tests disallow limited->unlimited while any are in use;
2535          * but we must separately disallow unlimited->limited, because
2536          * in that case we have no record of how much is already in use.
2537          */
2538         if (config.max_blocks && !sbinfo->max_blocks)
2539                 goto out;
2540         if (config.max_inodes && !sbinfo->max_inodes)
2541                 goto out;
2542
2543         error = 0;
2544         sbinfo->max_blocks  = config.max_blocks;
2545         sbinfo->max_inodes  = config.max_inodes;
2546         sbinfo->free_inodes = config.max_inodes - inodes;
2547
2548         mpol_put(sbinfo->mpol);
2549         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2550 out:
2551         spin_unlock(&sbinfo->stat_lock);
2552         return error;
2553 }
2554
2555 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2556 {
2557         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2558
2559         if (sbinfo->max_blocks != shmem_default_max_blocks())
2560                 seq_printf(seq, ",size=%luk",
2561                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2562         if (sbinfo->max_inodes != shmem_default_max_inodes())
2563                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2564         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2565                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2566         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2567                 seq_printf(seq, ",uid=%u",
2568                                 from_kuid_munged(&init_user_ns, sbinfo->uid));
2569         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2570                 seq_printf(seq, ",gid=%u",
2571                                 from_kgid_munged(&init_user_ns, sbinfo->gid));
2572         shmem_show_mpol(seq, sbinfo->mpol);
2573         return 0;
2574 }
2575 #endif /* CONFIG_TMPFS */
2576
2577 static void shmem_put_super(struct super_block *sb)
2578 {
2579         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2580
2581         percpu_counter_destroy(&sbinfo->used_blocks);
2582         kfree(sbinfo);
2583         sb->s_fs_info = NULL;
2584 }
2585
2586 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2587 {
2588         struct inode *inode;
2589         struct shmem_sb_info *sbinfo;
2590         int err = -ENOMEM;
2591
2592         /* Round up to L1_CACHE_BYTES to resist false sharing */
2593         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2594                                 L1_CACHE_BYTES), GFP_KERNEL);
2595         if (!sbinfo)
2596                 return -ENOMEM;
2597
2598         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2599         sbinfo->uid = current_fsuid();
2600         sbinfo->gid = current_fsgid();
2601         sb->s_fs_info = sbinfo;
2602
2603 #ifdef CONFIG_TMPFS
2604         /*
2605          * Per default we only allow half of the physical ram per
2606          * tmpfs instance, limiting inodes to one per page of lowmem;
2607          * but the internal instance is left unlimited.
2608          */
2609         if (!(sb->s_flags & MS_NOUSER)) {
2610                 sbinfo->max_blocks = shmem_default_max_blocks();
2611                 sbinfo->max_inodes = shmem_default_max_inodes();
2612                 if (shmem_parse_options(data, sbinfo, false)) {
2613                         err = -EINVAL;
2614                         goto failed;
2615                 }
2616         }
2617         sb->s_export_op = &shmem_export_ops;
2618         sb->s_flags |= MS_NOSEC;
2619 #else
2620         sb->s_flags |= MS_NOUSER;
2621 #endif
2622
2623         spin_lock_init(&sbinfo->stat_lock);
2624         if (percpu_counter_init(&sbinfo->used_blocks, 0))
2625                 goto failed;
2626         sbinfo->free_inodes = sbinfo->max_inodes;
2627
2628         sb->s_maxbytes = MAX_LFS_FILESIZE;
2629         sb->s_blocksize = PAGE_CACHE_SIZE;
2630         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2631         sb->s_magic = TMPFS_MAGIC;
2632         sb->s_op = &shmem_ops;
2633         sb->s_time_gran = 1;
2634 #ifdef CONFIG_TMPFS_XATTR
2635         sb->s_xattr = shmem_xattr_handlers;
2636 #endif
2637 #ifdef CONFIG_TMPFS_POSIX_ACL
2638         sb->s_flags |= MS_POSIXACL;
2639 #endif
2640
2641         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2642         if (!inode)
2643                 goto failed;
2644         inode->i_uid = sbinfo->uid;
2645         inode->i_gid = sbinfo->gid;
2646         sb->s_root = d_make_root(inode);
2647         if (!sb->s_root)
2648                 goto failed;
2649         return 0;
2650
2651 failed:
2652         shmem_put_super(sb);
2653         return err;
2654 }
2655
2656 static struct kmem_cache *shmem_inode_cachep;
2657
2658 static struct inode *shmem_alloc_inode(struct super_block *sb)
2659 {
2660         struct shmem_inode_info *info;
2661         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2662         if (!info)
2663                 return NULL;
2664         return &info->vfs_inode;
2665 }
2666
2667 static void shmem_destroy_callback(struct rcu_head *head)
2668 {
2669         struct inode *inode = container_of(head, struct inode, i_rcu);
2670         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2671 }
2672
2673 static void shmem_destroy_inode(struct inode *inode)
2674 {
2675         if (S_ISREG(inode->i_mode))
2676                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2677         call_rcu(&inode->i_rcu, shmem_destroy_callback);
2678 }
2679
2680 static void shmem_init_inode(void *foo)
2681 {
2682         struct shmem_inode_info *info = foo;
2683         inode_init_once(&info->vfs_inode);
2684 }
2685
2686 static int shmem_init_inodecache(void)
2687 {
2688         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2689                                 sizeof(struct shmem_inode_info),
2690                                 0, SLAB_PANIC, shmem_init_inode);
2691         return 0;
2692 }
2693
2694 static void shmem_destroy_inodecache(void)
2695 {
2696         kmem_cache_destroy(shmem_inode_cachep);
2697 }
2698
2699 static const struct address_space_operations shmem_aops = {
2700         .writepage      = shmem_writepage,
2701         .set_page_dirty = __set_page_dirty_no_writeback,
2702 #ifdef CONFIG_TMPFS
2703         .write_begin    = shmem_write_begin,
2704         .write_end      = shmem_write_end,
2705 #endif
2706         .migratepage    = migrate_page,
2707         .error_remove_page = generic_error_remove_page,
2708 };
2709
2710 static const struct file_operations shmem_file_operations = {
2711         .mmap           = shmem_mmap,
2712 #ifdef CONFIG_TMPFS
2713         .llseek         = generic_file_llseek,
2714         .read           = do_sync_read,
2715         .write          = do_sync_write,
2716         .aio_read       = shmem_file_aio_read,
2717         .aio_write      = generic_file_aio_write,
2718         .fsync          = noop_fsync,
2719         .splice_read    = shmem_file_splice_read,
2720         .splice_write   = generic_file_splice_write,
2721         .fallocate      = shmem_fallocate,
2722 #endif
2723 };
2724
2725 static const struct inode_operations shmem_inode_operations = {
2726         .setattr        = shmem_setattr,
2727 #ifdef CONFIG_TMPFS_XATTR
2728         .setxattr       = shmem_setxattr,
2729         .getxattr       = shmem_getxattr,
2730         .listxattr      = shmem_listxattr,
2731         .removexattr    = shmem_removexattr,
2732 #endif
2733 };
2734
2735 static const struct inode_operations shmem_dir_inode_operations = {
2736 #ifdef CONFIG_TMPFS
2737         .create         = shmem_create,
2738         .lookup         = simple_lookup,
2739         .link           = shmem_link,
2740         .unlink         = shmem_unlink,
2741         .symlink        = shmem_symlink,
2742         .mkdir          = shmem_mkdir,
2743         .rmdir          = shmem_rmdir,
2744         .mknod          = shmem_mknod,
2745         .rename         = shmem_rename,
2746 #endif
2747 #ifdef CONFIG_TMPFS_XATTR
2748         .setxattr       = shmem_setxattr,
2749         .getxattr       = shmem_getxattr,
2750         .listxattr      = shmem_listxattr,
2751         .removexattr    = shmem_removexattr,
2752 #endif
2753 #ifdef CONFIG_TMPFS_POSIX_ACL
2754         .setattr        = shmem_setattr,
2755 #endif
2756 };
2757
2758 static const struct inode_operations shmem_special_inode_operations = {
2759 #ifdef CONFIG_TMPFS_XATTR
2760         .setxattr       = shmem_setxattr,
2761         .getxattr       = shmem_getxattr,
2762         .listxattr      = shmem_listxattr,
2763         .removexattr    = shmem_removexattr,
2764 #endif
2765 #ifdef CONFIG_TMPFS_POSIX_ACL
2766         .setattr        = shmem_setattr,
2767 #endif
2768 };
2769
2770 static const struct super_operations shmem_ops = {
2771         .alloc_inode    = shmem_alloc_inode,
2772         .destroy_inode  = shmem_destroy_inode,
2773 #ifdef CONFIG_TMPFS
2774         .statfs         = shmem_statfs,
2775         .remount_fs     = shmem_remount_fs,
2776         .show_options   = shmem_show_options,
2777 #endif
2778         .evict_inode    = shmem_evict_inode,
2779         .drop_inode     = generic_delete_inode,
2780         .put_super      = shmem_put_super,
2781 };
2782
2783 static const struct vm_operations_struct shmem_vm_ops = {
2784         .fault          = shmem_fault,
2785 #ifdef CONFIG_NUMA
2786         .set_policy     = shmem_set_policy,
2787         .get_policy     = shmem_get_policy,
2788 #endif
2789 };
2790
2791 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2792         int flags, const char *dev_name, void *data)
2793 {
2794         return mount_nodev(fs_type, flags, data, shmem_fill_super);
2795 }
2796
2797 static struct file_system_type shmem_fs_type = {
2798         .owner          = THIS_MODULE,
2799         .name           = "tmpfs",
2800         .mount          = shmem_mount,
2801         .kill_sb        = kill_litter_super,
2802 };
2803
2804 int __init shmem_init(void)
2805 {
2806         int error;
2807
2808         error = bdi_init(&shmem_backing_dev_info);
2809         if (error)
2810                 goto out4;
2811
2812         error = shmem_init_inodecache();
2813         if (error)
2814                 goto out3;
2815
2816         error = register_filesystem(&shmem_fs_type);
2817         if (error) {
2818                 printk(KERN_ERR "Could not register tmpfs\n");
2819                 goto out2;
2820         }
2821
2822         shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2823                                  shmem_fs_type.name, NULL);
2824         if (IS_ERR(shm_mnt)) {
2825                 error = PTR_ERR(shm_mnt);
2826                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2827                 goto out1;
2828         }
2829         return 0;
2830
2831 out1:
2832         unregister_filesystem(&shmem_fs_type);
2833 out2:
2834         shmem_destroy_inodecache();
2835 out3:
2836         bdi_destroy(&shmem_backing_dev_info);
2837 out4:
2838         shm_mnt = ERR_PTR(error);
2839         return error;
2840 }
2841
2842 #else /* !CONFIG_SHMEM */
2843
2844 /*
2845  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2846  *
2847  * This is intended for small system where the benefits of the full
2848  * shmem code (swap-backed and resource-limited) are outweighed by
2849  * their complexity. On systems without swap this code should be
2850  * effectively equivalent, but much lighter weight.
2851  */
2852
2853 #include <linux/ramfs.h>
2854
2855 static struct file_system_type shmem_fs_type = {
2856         .name           = "tmpfs",
2857         .mount          = ramfs_mount,
2858         .kill_sb        = kill_litter_super,
2859 };
2860
2861 int __init shmem_init(void)
2862 {
2863         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2864
2865         shm_mnt = kern_mount(&shmem_fs_type);
2866         BUG_ON(IS_ERR(shm_mnt));
2867
2868         return 0;
2869 }
2870
2871 int shmem_unuse(swp_entry_t swap, struct page *page)
2872 {
2873         return 0;
2874 }
2875
2876 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2877 {
2878         return 0;
2879 }
2880
2881 void shmem_unlock_mapping(struct address_space *mapping)
2882 {
2883 }
2884
2885 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2886 {
2887         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2888 }
2889 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2890
2891 #define shmem_vm_ops                            generic_file_vm_ops
2892 #define shmem_file_operations                   ramfs_file_operations
2893 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2894 #define shmem_acct_size(flags, size)            0
2895 #define shmem_unacct_size(flags, size)          do {} while (0)
2896
2897 #endif /* CONFIG_SHMEM */
2898
2899 /* common code */
2900
2901 /**
2902  * shmem_file_setup - get an unlinked file living in tmpfs
2903  * @name: name for dentry (to be seen in /proc/<pid>/maps
2904  * @size: size to be set for the file
2905  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2906  */
2907 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2908 {
2909         int error;
2910         struct file *file;
2911         struct inode *inode;
2912         struct path path;
2913         struct dentry *root;
2914         struct qstr this;
2915
2916         if (IS_ERR(shm_mnt))
2917                 return (void *)shm_mnt;
2918
2919         if (size < 0 || size > MAX_LFS_FILESIZE)
2920                 return ERR_PTR(-EINVAL);
2921
2922         if (shmem_acct_size(flags, size))
2923                 return ERR_PTR(-ENOMEM);
2924
2925         error = -ENOMEM;
2926         this.name = name;
2927         this.len = strlen(name);
2928         this.hash = 0; /* will go */
2929         root = shm_mnt->mnt_root;
2930         path.dentry = d_alloc(root, &this);
2931         if (!path.dentry)
2932                 goto put_memory;
2933         path.mnt = mntget(shm_mnt);
2934
2935         error = -ENOSPC;
2936         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2937         if (!inode)
2938                 goto put_dentry;
2939
2940         d_instantiate(path.dentry, inode);
2941         inode->i_size = size;
2942         clear_nlink(inode);     /* It is unlinked */
2943 #ifndef CONFIG_MMU
2944         error = ramfs_nommu_expand_for_mapping(inode, size);
2945         if (error)
2946                 goto put_dentry;
2947 #endif
2948
2949         error = -ENFILE;
2950         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2951                   &shmem_file_operations);
2952         if (!file)
2953                 goto put_dentry;
2954
2955         return file;
2956
2957 put_dentry:
2958         path_put(&path);
2959 put_memory:
2960         shmem_unacct_size(flags, size);
2961         return ERR_PTR(error);
2962 }
2963 EXPORT_SYMBOL_GPL(shmem_file_setup);
2964
2965 /**
2966  * shmem_zero_setup - setup a shared anonymous mapping
2967  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2968  */
2969 int shmem_zero_setup(struct vm_area_struct *vma)
2970 {
2971         struct file *file;
2972         loff_t size = vma->vm_end - vma->vm_start;
2973
2974         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2975         if (IS_ERR(file))
2976                 return PTR_ERR(file);
2977
2978         if (vma->vm_file)
2979                 fput(vma->vm_file);
2980         vma->vm_file = file;
2981         vma->vm_ops = &shmem_vm_ops;
2982         vma->vm_flags |= VM_CAN_NONLINEAR;
2983         return 0;
2984 }
2985
2986 /**
2987  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2988  * @mapping:    the page's address_space
2989  * @index:      the page index
2990  * @gfp:        the page allocator flags to use if allocating
2991  *
2992  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2993  * with any new page allocations done using the specified allocation flags.
2994  * But read_cache_page_gfp() uses the ->readpage() method: which does not
2995  * suit tmpfs, since it may have pages in swapcache, and needs to find those
2996  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2997  *
2998  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2999  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3000  */
3001 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3002                                          pgoff_t index, gfp_t gfp)
3003 {
3004 #ifdef CONFIG_SHMEM
3005         struct inode *inode = mapping->host;
3006         struct page *page;
3007         int error;
3008
3009         BUG_ON(mapping->a_ops != &shmem_aops);
3010         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3011         if (error)
3012                 page = ERR_PTR(error);
3013         else
3014                 unlock_page(page);
3015         return page;
3016 #else
3017         /*
3018          * The tiny !SHMEM case uses ramfs without swap
3019          */
3020         return read_cache_page_gfp(mapping, index, gfp);
3021 #endif
3022 }
3023 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);