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