thp: add tlb_remove_pmd_tlb_entry
[linux-2.6.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/splice.h>
57 #include <linux/security.h>
58 #include <linux/swapops.h>
59 #include <linux/mempolicy.h>
60 #include <linux/namei.h>
61 #include <linux/ctype.h>
62 #include <linux/migrate.h>
63 #include <linux/highmem.h>
64 #include <linux/seq_file.h>
65 #include <linux/magic.h>
66
67 #include <asm/uaccess.h>
68 #include <asm/pgtable.h>
69
70 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
71 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
72
73 /* Pretend that each entry is of this size in directory's i_size */
74 #define BOGO_DIRENT_SIZE 20
75
76 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
77 #define SHORT_SYMLINK_LEN 128
78
79 struct shmem_xattr {
80         struct list_head list;  /* anchored by shmem_inode_info->xattr_list */
81         char *name;             /* xattr name */
82         size_t size;
83         char value[0];
84 };
85
86 /* Flag allocation requirements to shmem_getpage */
87 enum sgp_type {
88         SGP_READ,       /* don't exceed i_size, don't allocate page */
89         SGP_CACHE,      /* don't exceed i_size, may allocate page */
90         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
91         SGP_WRITE,      /* may exceed i_size, may allocate page */
92 };
93
94 #ifdef CONFIG_TMPFS
95 static unsigned long shmem_default_max_blocks(void)
96 {
97         return totalram_pages / 2;
98 }
99
100 static unsigned long shmem_default_max_inodes(void)
101 {
102         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
103 }
104 #endif
105
106 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
107         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
108
109 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
110         struct page **pagep, enum sgp_type sgp, int *fault_type)
111 {
112         return shmem_getpage_gfp(inode, index, pagep, sgp,
113                         mapping_gfp_mask(inode->i_mapping), fault_type);
114 }
115
116 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
117 {
118         return sb->s_fs_info;
119 }
120
121 /*
122  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
123  * for shared memory and for shared anonymous (/dev/zero) mappings
124  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
125  * consistent with the pre-accounting of private mappings ...
126  */
127 static inline int shmem_acct_size(unsigned long flags, loff_t size)
128 {
129         return (flags & VM_NORESERVE) ?
130                 0 : security_vm_enough_memory_kern(VM_ACCT(size));
131 }
132
133 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
134 {
135         if (!(flags & VM_NORESERVE))
136                 vm_unacct_memory(VM_ACCT(size));
137 }
138
139 /*
140  * ... whereas tmpfs objects are accounted incrementally as
141  * pages are allocated, in order to allow huge sparse files.
142  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
143  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
144  */
145 static inline int shmem_acct_block(unsigned long flags)
146 {
147         return (flags & VM_NORESERVE) ?
148                 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
149 }
150
151 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
152 {
153         if (flags & VM_NORESERVE)
154                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
155 }
156
157 static const struct super_operations shmem_ops;
158 static const struct address_space_operations shmem_aops;
159 static const struct file_operations shmem_file_operations;
160 static const struct inode_operations shmem_inode_operations;
161 static const struct inode_operations shmem_dir_inode_operations;
162 static const struct inode_operations shmem_special_inode_operations;
163 static const struct vm_operations_struct shmem_vm_ops;
164
165 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
166         .ra_pages       = 0,    /* No readahead */
167         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
168 };
169
170 static LIST_HEAD(shmem_swaplist);
171 static DEFINE_MUTEX(shmem_swaplist_mutex);
172
173 static int shmem_reserve_inode(struct super_block *sb)
174 {
175         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
176         if (sbinfo->max_inodes) {
177                 spin_lock(&sbinfo->stat_lock);
178                 if (!sbinfo->free_inodes) {
179                         spin_unlock(&sbinfo->stat_lock);
180                         return -ENOSPC;
181                 }
182                 sbinfo->free_inodes--;
183                 spin_unlock(&sbinfo->stat_lock);
184         }
185         return 0;
186 }
187
188 static void shmem_free_inode(struct super_block *sb)
189 {
190         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
191         if (sbinfo->max_inodes) {
192                 spin_lock(&sbinfo->stat_lock);
193                 sbinfo->free_inodes++;
194                 spin_unlock(&sbinfo->stat_lock);
195         }
196 }
197
198 /**
199  * shmem_recalc_inode - recalculate the block usage of an inode
200  * @inode: inode to recalc
201  *
202  * We have to calculate the free blocks since the mm can drop
203  * undirtied hole pages behind our back.
204  *
205  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
206  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
207  *
208  * It has to be called with the spinlock held.
209  */
210 static void shmem_recalc_inode(struct inode *inode)
211 {
212         struct shmem_inode_info *info = SHMEM_I(inode);
213         long freed;
214
215         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
216         if (freed > 0) {
217                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
218                 if (sbinfo->max_blocks)
219                         percpu_counter_add(&sbinfo->used_blocks, -freed);
220                 info->alloced -= freed;
221                 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
222                 shmem_unacct_blocks(info->flags, freed);
223         }
224 }
225
226 /*
227  * Replace item expected in radix tree by a new item, while holding tree lock.
228  */
229 static int shmem_radix_tree_replace(struct address_space *mapping,
230                         pgoff_t index, void *expected, void *replacement)
231 {
232         void **pslot;
233         void *item = NULL;
234
235         VM_BUG_ON(!expected);
236         pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
237         if (pslot)
238                 item = radix_tree_deref_slot_protected(pslot,
239                                                         &mapping->tree_lock);
240         if (item != expected)
241                 return -ENOENT;
242         if (replacement)
243                 radix_tree_replace_slot(pslot, replacement);
244         else
245                 radix_tree_delete(&mapping->page_tree, index);
246         return 0;
247 }
248
249 /*
250  * Like add_to_page_cache_locked, but error if expected item has gone.
251  */
252 static int shmem_add_to_page_cache(struct page *page,
253                                    struct address_space *mapping,
254                                    pgoff_t index, gfp_t gfp, void *expected)
255 {
256         int error = 0;
257
258         VM_BUG_ON(!PageLocked(page));
259         VM_BUG_ON(!PageSwapBacked(page));
260
261         if (!expected)
262                 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
263         if (!error) {
264                 page_cache_get(page);
265                 page->mapping = mapping;
266                 page->index = index;
267
268                 spin_lock_irq(&mapping->tree_lock);
269                 if (!expected)
270                         error = radix_tree_insert(&mapping->page_tree,
271                                                         index, page);
272                 else
273                         error = shmem_radix_tree_replace(mapping, index,
274                                                         expected, page);
275                 if (!error) {
276                         mapping->nrpages++;
277                         __inc_zone_page_state(page, NR_FILE_PAGES);
278                         __inc_zone_page_state(page, NR_SHMEM);
279                         spin_unlock_irq(&mapping->tree_lock);
280                 } else {
281                         page->mapping = NULL;
282                         spin_unlock_irq(&mapping->tree_lock);
283                         page_cache_release(page);
284                 }
285                 if (!expected)
286                         radix_tree_preload_end();
287         }
288         if (error)
289                 mem_cgroup_uncharge_cache_page(page);
290         return error;
291 }
292
293 /*
294  * Like delete_from_page_cache, but substitutes swap for page.
295  */
296 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
297 {
298         struct address_space *mapping = page->mapping;
299         int error;
300
301         spin_lock_irq(&mapping->tree_lock);
302         error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
303         page->mapping = NULL;
304         mapping->nrpages--;
305         __dec_zone_page_state(page, NR_FILE_PAGES);
306         __dec_zone_page_state(page, NR_SHMEM);
307         spin_unlock_irq(&mapping->tree_lock);
308         page_cache_release(page);
309         BUG_ON(error);
310 }
311
312 /*
313  * Like find_get_pages, but collecting swap entries as well as pages.
314  */
315 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
316                                         pgoff_t start, unsigned int nr_pages,
317                                         struct page **pages, pgoff_t *indices)
318 {
319         unsigned int i;
320         unsigned int ret;
321         unsigned int nr_found;
322
323         rcu_read_lock();
324 restart:
325         nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
326                                 (void ***)pages, indices, start, nr_pages);
327         ret = 0;
328         for (i = 0; i < nr_found; i++) {
329                 struct page *page;
330 repeat:
331                 page = radix_tree_deref_slot((void **)pages[i]);
332                 if (unlikely(!page))
333                         continue;
334                 if (radix_tree_exception(page)) {
335                         if (radix_tree_deref_retry(page))
336                                 goto restart;
337                         /*
338                          * Otherwise, we must be storing a swap entry
339                          * here as an exceptional entry: so return it
340                          * without attempting to raise page count.
341                          */
342                         goto export;
343                 }
344                 if (!page_cache_get_speculative(page))
345                         goto repeat;
346
347                 /* Has the page moved? */
348                 if (unlikely(page != *((void **)pages[i]))) {
349                         page_cache_release(page);
350                         goto repeat;
351                 }
352 export:
353                 indices[ret] = indices[i];
354                 pages[ret] = page;
355                 ret++;
356         }
357         if (unlikely(!ret && nr_found))
358                 goto restart;
359         rcu_read_unlock();
360         return ret;
361 }
362
363 /*
364  * Remove swap entry from radix tree, free the swap and its page cache.
365  */
366 static int shmem_free_swap(struct address_space *mapping,
367                            pgoff_t index, void *radswap)
368 {
369         int error;
370
371         spin_lock_irq(&mapping->tree_lock);
372         error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
373         spin_unlock_irq(&mapping->tree_lock);
374         if (!error)
375                 free_swap_and_cache(radix_to_swp_entry(radswap));
376         return error;
377 }
378
379 /*
380  * Pagevec may contain swap entries, so shuffle up pages before releasing.
381  */
382 static void shmem_pagevec_release(struct pagevec *pvec)
383 {
384         int i, j;
385
386         for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
387                 struct page *page = pvec->pages[i];
388                 if (!radix_tree_exceptional_entry(page))
389                         pvec->pages[j++] = page;
390         }
391         pvec->nr = j;
392         pagevec_release(pvec);
393 }
394
395 /*
396  * Remove range of pages and swap entries from radix tree, and free them.
397  */
398 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
399 {
400         struct address_space *mapping = inode->i_mapping;
401         struct shmem_inode_info *info = SHMEM_I(inode);
402         pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
403         unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
404         pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
405         struct pagevec pvec;
406         pgoff_t indices[PAGEVEC_SIZE];
407         long nr_swaps_freed = 0;
408         pgoff_t index;
409         int i;
410
411         BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
412
413         pagevec_init(&pvec, 0);
414         index = start;
415         while (index <= end) {
416                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
417                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
418                                                         pvec.pages, indices);
419                 if (!pvec.nr)
420                         break;
421                 mem_cgroup_uncharge_start();
422                 for (i = 0; i < pagevec_count(&pvec); i++) {
423                         struct page *page = pvec.pages[i];
424
425                         index = indices[i];
426                         if (index > end)
427                                 break;
428
429                         if (radix_tree_exceptional_entry(page)) {
430                                 nr_swaps_freed += !shmem_free_swap(mapping,
431                                                                 index, page);
432                                 continue;
433                         }
434
435                         if (!trylock_page(page))
436                                 continue;
437                         if (page->mapping == mapping) {
438                                 VM_BUG_ON(PageWriteback(page));
439                                 truncate_inode_page(mapping, page);
440                         }
441                         unlock_page(page);
442                 }
443                 shmem_pagevec_release(&pvec);
444                 mem_cgroup_uncharge_end();
445                 cond_resched();
446                 index++;
447         }
448
449         if (partial) {
450                 struct page *page = NULL;
451                 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
452                 if (page) {
453                         zero_user_segment(page, partial, PAGE_CACHE_SIZE);
454                         set_page_dirty(page);
455                         unlock_page(page);
456                         page_cache_release(page);
457                 }
458         }
459
460         index = start;
461         for ( ; ; ) {
462                 cond_resched();
463                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
464                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
465                                                         pvec.pages, indices);
466                 if (!pvec.nr) {
467                         if (index == start)
468                                 break;
469                         index = start;
470                         continue;
471                 }
472                 if (index == start && indices[0] > end) {
473                         shmem_pagevec_release(&pvec);
474                         break;
475                 }
476                 mem_cgroup_uncharge_start();
477                 for (i = 0; i < pagevec_count(&pvec); i++) {
478                         struct page *page = pvec.pages[i];
479
480                         index = indices[i];
481                         if (index > end)
482                                 break;
483
484                         if (radix_tree_exceptional_entry(page)) {
485                                 nr_swaps_freed += !shmem_free_swap(mapping,
486                                                                 index, page);
487                                 continue;
488                         }
489
490                         lock_page(page);
491                         if (page->mapping == mapping) {
492                                 VM_BUG_ON(PageWriteback(page));
493                                 truncate_inode_page(mapping, page);
494                         }
495                         unlock_page(page);
496                 }
497                 shmem_pagevec_release(&pvec);
498                 mem_cgroup_uncharge_end();
499                 index++;
500         }
501
502         spin_lock(&info->lock);
503         info->swapped -= nr_swaps_freed;
504         shmem_recalc_inode(inode);
505         spin_unlock(&info->lock);
506
507         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
508 }
509 EXPORT_SYMBOL_GPL(shmem_truncate_range);
510
511 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
512 {
513         struct inode *inode = dentry->d_inode;
514         int error;
515
516         error = inode_change_ok(inode, attr);
517         if (error)
518                 return error;
519
520         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
521                 loff_t oldsize = inode->i_size;
522                 loff_t newsize = attr->ia_size;
523
524                 if (newsize != oldsize) {
525                         i_size_write(inode, newsize);
526                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
527                 }
528                 if (newsize < oldsize) {
529                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
530                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
531                         shmem_truncate_range(inode, newsize, (loff_t)-1);
532                         /* unmap again to remove racily COWed private pages */
533                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
534                 }
535         }
536
537         setattr_copy(inode, attr);
538 #ifdef CONFIG_TMPFS_POSIX_ACL
539         if (attr->ia_valid & ATTR_MODE)
540                 error = generic_acl_chmod(inode);
541 #endif
542         return error;
543 }
544
545 static void shmem_evict_inode(struct inode *inode)
546 {
547         struct shmem_inode_info *info = SHMEM_I(inode);
548         struct shmem_xattr *xattr, *nxattr;
549
550         if (inode->i_mapping->a_ops == &shmem_aops) {
551                 shmem_unacct_size(info->flags, inode->i_size);
552                 inode->i_size = 0;
553                 shmem_truncate_range(inode, 0, (loff_t)-1);
554                 if (!list_empty(&info->swaplist)) {
555                         mutex_lock(&shmem_swaplist_mutex);
556                         list_del_init(&info->swaplist);
557                         mutex_unlock(&shmem_swaplist_mutex);
558                 }
559         } else
560                 kfree(info->symlink);
561
562         list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
563                 kfree(xattr->name);
564                 kfree(xattr);
565         }
566         BUG_ON(inode->i_blocks);
567         shmem_free_inode(inode->i_sb);
568         end_writeback(inode);
569 }
570
571 /*
572  * If swap found in inode, free it and move page from swapcache to filecache.
573  */
574 static int shmem_unuse_inode(struct shmem_inode_info *info,
575                              swp_entry_t swap, struct page *page)
576 {
577         struct address_space *mapping = info->vfs_inode.i_mapping;
578         void *radswap;
579         pgoff_t index;
580         int error;
581
582         radswap = swp_to_radix_entry(swap);
583         index = radix_tree_locate_item(&mapping->page_tree, radswap);
584         if (index == -1)
585                 return 0;
586
587         /*
588          * Move _head_ to start search for next from here.
589          * But be careful: shmem_evict_inode checks list_empty without taking
590          * mutex, and there's an instant in list_move_tail when info->swaplist
591          * would appear empty, if it were the only one on shmem_swaplist.
592          */
593         if (shmem_swaplist.next != &info->swaplist)
594                 list_move_tail(&shmem_swaplist, &info->swaplist);
595
596         /*
597          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
598          * but also to hold up shmem_evict_inode(): so inode cannot be freed
599          * beneath us (pagelock doesn't help until the page is in pagecache).
600          */
601         error = shmem_add_to_page_cache(page, mapping, index,
602                                                 GFP_NOWAIT, radswap);
603         /* which does mem_cgroup_uncharge_cache_page on error */
604
605         if (error != -ENOMEM) {
606                 /*
607                  * Truncation and eviction use free_swap_and_cache(), which
608                  * only does trylock page: if we raced, best clean up here.
609                  */
610                 delete_from_swap_cache(page);
611                 set_page_dirty(page);
612                 if (!error) {
613                         spin_lock(&info->lock);
614                         info->swapped--;
615                         spin_unlock(&info->lock);
616                         swap_free(swap);
617                 }
618                 error = 1;      /* not an error, but entry was found */
619         }
620         return error;
621 }
622
623 /*
624  * Search through swapped inodes to find and replace swap by page.
625  */
626 int shmem_unuse(swp_entry_t swap, struct page *page)
627 {
628         struct list_head *this, *next;
629         struct shmem_inode_info *info;
630         int found = 0;
631         int error;
632
633         /*
634          * Charge page using GFP_KERNEL while we can wait, before taking
635          * the shmem_swaplist_mutex which might hold up shmem_writepage().
636          * Charged back to the user (not to caller) when swap account is used.
637          */
638         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
639         if (error)
640                 goto out;
641         /* No radix_tree_preload: swap entry keeps a place for page in tree */
642
643         mutex_lock(&shmem_swaplist_mutex);
644         list_for_each_safe(this, next, &shmem_swaplist) {
645                 info = list_entry(this, struct shmem_inode_info, swaplist);
646                 if (info->swapped)
647                         found = shmem_unuse_inode(info, swap, page);
648                 else
649                         list_del_init(&info->swaplist);
650                 cond_resched();
651                 if (found)
652                         break;
653         }
654         mutex_unlock(&shmem_swaplist_mutex);
655
656         if (!found)
657                 mem_cgroup_uncharge_cache_page(page);
658         if (found < 0)
659                 error = found;
660 out:
661         unlock_page(page);
662         page_cache_release(page);
663         return error;
664 }
665
666 /*
667  * Move the page from the page cache to the swap cache.
668  */
669 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
670 {
671         struct shmem_inode_info *info;
672         struct address_space *mapping;
673         struct inode *inode;
674         swp_entry_t swap;
675         pgoff_t index;
676
677         BUG_ON(!PageLocked(page));
678         mapping = page->mapping;
679         index = page->index;
680         inode = mapping->host;
681         info = SHMEM_I(inode);
682         if (info->flags & VM_LOCKED)
683                 goto redirty;
684         if (!total_swap_pages)
685                 goto redirty;
686
687         /*
688          * shmem_backing_dev_info's capabilities prevent regular writeback or
689          * sync from ever calling shmem_writepage; but a stacking filesystem
690          * might use ->writepage of its underlying filesystem, in which case
691          * tmpfs should write out to swap only in response to memory pressure,
692          * and not for the writeback threads or sync.
693          */
694         if (!wbc->for_reclaim) {
695                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
696                 goto redirty;
697         }
698         swap = get_swap_page();
699         if (!swap.val)
700                 goto redirty;
701
702         /*
703          * Add inode to shmem_unuse()'s list of swapped-out inodes,
704          * if it's not already there.  Do it now before the page is
705          * moved to swap cache, when its pagelock no longer protects
706          * the inode from eviction.  But don't unlock the mutex until
707          * we've incremented swapped, because shmem_unuse_inode() will
708          * prune a !swapped inode from the swaplist under this mutex.
709          */
710         mutex_lock(&shmem_swaplist_mutex);
711         if (list_empty(&info->swaplist))
712                 list_add_tail(&info->swaplist, &shmem_swaplist);
713
714         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
715                 swap_shmem_alloc(swap);
716                 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
717
718                 spin_lock(&info->lock);
719                 info->swapped++;
720                 shmem_recalc_inode(inode);
721                 spin_unlock(&info->lock);
722
723                 mutex_unlock(&shmem_swaplist_mutex);
724                 BUG_ON(page_mapped(page));
725                 swap_writepage(page, wbc);
726                 return 0;
727         }
728
729         mutex_unlock(&shmem_swaplist_mutex);
730         swapcache_free(swap, NULL);
731 redirty:
732         set_page_dirty(page);
733         if (wbc->for_reclaim)
734                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
735         unlock_page(page);
736         return 0;
737 }
738
739 #ifdef CONFIG_NUMA
740 #ifdef CONFIG_TMPFS
741 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
742 {
743         char buffer[64];
744
745         if (!mpol || mpol->mode == MPOL_DEFAULT)
746                 return;         /* show nothing */
747
748         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
749
750         seq_printf(seq, ",mpol=%s", buffer);
751 }
752
753 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
754 {
755         struct mempolicy *mpol = NULL;
756         if (sbinfo->mpol) {
757                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
758                 mpol = sbinfo->mpol;
759                 mpol_get(mpol);
760                 spin_unlock(&sbinfo->stat_lock);
761         }
762         return mpol;
763 }
764 #endif /* CONFIG_TMPFS */
765
766 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
767                         struct shmem_inode_info *info, pgoff_t index)
768 {
769         struct mempolicy mpol, *spol;
770         struct vm_area_struct pvma;
771
772         spol = mpol_cond_copy(&mpol,
773                         mpol_shared_policy_lookup(&info->policy, index));
774
775         /* Create a pseudo vma that just contains the policy */
776         pvma.vm_start = 0;
777         pvma.vm_pgoff = index;
778         pvma.vm_ops = NULL;
779         pvma.vm_policy = spol;
780         return swapin_readahead(swap, gfp, &pvma, 0);
781 }
782
783 static struct page *shmem_alloc_page(gfp_t gfp,
784                         struct shmem_inode_info *info, pgoff_t index)
785 {
786         struct vm_area_struct pvma;
787
788         /* Create a pseudo vma that just contains the policy */
789         pvma.vm_start = 0;
790         pvma.vm_pgoff = index;
791         pvma.vm_ops = NULL;
792         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
793
794         /*
795          * alloc_page_vma() will drop the shared policy reference
796          */
797         return alloc_page_vma(gfp, &pvma, 0);
798 }
799 #else /* !CONFIG_NUMA */
800 #ifdef CONFIG_TMPFS
801 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
802 {
803 }
804 #endif /* CONFIG_TMPFS */
805
806 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
807                         struct shmem_inode_info *info, pgoff_t index)
808 {
809         return swapin_readahead(swap, gfp, NULL, 0);
810 }
811
812 static inline struct page *shmem_alloc_page(gfp_t gfp,
813                         struct shmem_inode_info *info, pgoff_t index)
814 {
815         return alloc_page(gfp);
816 }
817 #endif /* CONFIG_NUMA */
818
819 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
820 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
821 {
822         return NULL;
823 }
824 #endif
825
826 /*
827  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
828  *
829  * If we allocate a new one we do not mark it dirty. That's up to the
830  * vm. If we swap it in we mark it dirty since we also free the swap
831  * entry since a page cannot live in both the swap and page cache
832  */
833 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
834         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
835 {
836         struct address_space *mapping = inode->i_mapping;
837         struct shmem_inode_info *info;
838         struct shmem_sb_info *sbinfo;
839         struct page *page;
840         swp_entry_t swap;
841         int error;
842         int once = 0;
843
844         if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
845                 return -EFBIG;
846 repeat:
847         swap.val = 0;
848         page = find_lock_page(mapping, index);
849         if (radix_tree_exceptional_entry(page)) {
850                 swap = radix_to_swp_entry(page);
851                 page = NULL;
852         }
853
854         if (sgp != SGP_WRITE &&
855             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
856                 error = -EINVAL;
857                 goto failed;
858         }
859
860         if (page || (sgp == SGP_READ && !swap.val)) {
861                 /*
862                  * Once we can get the page lock, it must be uptodate:
863                  * if there were an error in reading back from swap,
864                  * the page would not be inserted into the filecache.
865                  */
866                 BUG_ON(page && !PageUptodate(page));
867                 *pagep = page;
868                 return 0;
869         }
870
871         /*
872          * Fast cache lookup did not find it:
873          * bring it back from swap or allocate.
874          */
875         info = SHMEM_I(inode);
876         sbinfo = SHMEM_SB(inode->i_sb);
877
878         if (swap.val) {
879                 /* Look it up and read it in.. */
880                 page = lookup_swap_cache(swap);
881                 if (!page) {
882                         /* here we actually do the io */
883                         if (fault_type)
884                                 *fault_type |= VM_FAULT_MAJOR;
885                         page = shmem_swapin(swap, gfp, info, index);
886                         if (!page) {
887                                 error = -ENOMEM;
888                                 goto failed;
889                         }
890                 }
891
892                 /* We have to do this with page locked to prevent races */
893                 lock_page(page);
894                 if (!PageUptodate(page)) {
895                         error = -EIO;
896                         goto failed;
897                 }
898                 wait_on_page_writeback(page);
899
900                 /* Someone may have already done it for us */
901                 if (page->mapping) {
902                         if (page->mapping == mapping &&
903                             page->index == index)
904                                 goto done;
905                         error = -EEXIST;
906                         goto failed;
907                 }
908
909                 error = mem_cgroup_cache_charge(page, current->mm,
910                                                 gfp & GFP_RECLAIM_MASK);
911                 if (!error)
912                         error = shmem_add_to_page_cache(page, mapping, index,
913                                                 gfp, swp_to_radix_entry(swap));
914                 if (error)
915                         goto failed;
916
917                 spin_lock(&info->lock);
918                 info->swapped--;
919                 shmem_recalc_inode(inode);
920                 spin_unlock(&info->lock);
921
922                 delete_from_swap_cache(page);
923                 set_page_dirty(page);
924                 swap_free(swap);
925
926         } else {
927                 if (shmem_acct_block(info->flags)) {
928                         error = -ENOSPC;
929                         goto failed;
930                 }
931                 if (sbinfo->max_blocks) {
932                         if (percpu_counter_compare(&sbinfo->used_blocks,
933                                                 sbinfo->max_blocks) >= 0) {
934                                 error = -ENOSPC;
935                                 goto unacct;
936                         }
937                         percpu_counter_inc(&sbinfo->used_blocks);
938                 }
939
940                 page = shmem_alloc_page(gfp, info, index);
941                 if (!page) {
942                         error = -ENOMEM;
943                         goto decused;
944                 }
945
946                 SetPageSwapBacked(page);
947                 __set_page_locked(page);
948                 error = mem_cgroup_cache_charge(page, current->mm,
949                                                 gfp & GFP_RECLAIM_MASK);
950                 if (!error)
951                         error = shmem_add_to_page_cache(page, mapping, index,
952                                                 gfp, NULL);
953                 if (error)
954                         goto decused;
955                 lru_cache_add_anon(page);
956
957                 spin_lock(&info->lock);
958                 info->alloced++;
959                 inode->i_blocks += BLOCKS_PER_PAGE;
960                 shmem_recalc_inode(inode);
961                 spin_unlock(&info->lock);
962
963                 clear_highpage(page);
964                 flush_dcache_page(page);
965                 SetPageUptodate(page);
966                 if (sgp == SGP_DIRTY)
967                         set_page_dirty(page);
968         }
969 done:
970         /* Perhaps the file has been truncated since we checked */
971         if (sgp != SGP_WRITE &&
972             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
973                 error = -EINVAL;
974                 goto trunc;
975         }
976         *pagep = page;
977         return 0;
978
979         /*
980          * Error recovery.
981          */
982 trunc:
983         ClearPageDirty(page);
984         delete_from_page_cache(page);
985         spin_lock(&info->lock);
986         info->alloced--;
987         inode->i_blocks -= BLOCKS_PER_PAGE;
988         spin_unlock(&info->lock);
989 decused:
990         if (sbinfo->max_blocks)
991                 percpu_counter_add(&sbinfo->used_blocks, -1);
992 unacct:
993         shmem_unacct_blocks(info->flags, 1);
994 failed:
995         if (swap.val && error != -EINVAL) {
996                 struct page *test = find_get_page(mapping, index);
997                 if (test && !radix_tree_exceptional_entry(test))
998                         page_cache_release(test);
999                 /* Have another try if the entry has changed */
1000                 if (test != swp_to_radix_entry(swap))
1001                         error = -EEXIST;
1002         }
1003         if (page) {
1004                 unlock_page(page);
1005                 page_cache_release(page);
1006         }
1007         if (error == -ENOSPC && !once++) {
1008                 info = SHMEM_I(inode);
1009                 spin_lock(&info->lock);
1010                 shmem_recalc_inode(inode);
1011                 spin_unlock(&info->lock);
1012                 goto repeat;
1013         }
1014         if (error == -EEXIST)
1015                 goto repeat;
1016         return error;
1017 }
1018
1019 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1020 {
1021         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1022         int error;
1023         int ret = VM_FAULT_LOCKED;
1024
1025         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1026         if (error)
1027                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1028
1029         if (ret & VM_FAULT_MAJOR) {
1030                 count_vm_event(PGMAJFAULT);
1031                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1032         }
1033         return ret;
1034 }
1035
1036 #ifdef CONFIG_NUMA
1037 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1038 {
1039         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1040         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1041 }
1042
1043 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1044                                           unsigned long addr)
1045 {
1046         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1047         pgoff_t index;
1048
1049         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1050         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1051 }
1052 #endif
1053
1054 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1055 {
1056         struct inode *inode = file->f_path.dentry->d_inode;
1057         struct shmem_inode_info *info = SHMEM_I(inode);
1058         int retval = -ENOMEM;
1059
1060         spin_lock(&info->lock);
1061         if (lock && !(info->flags & VM_LOCKED)) {
1062                 if (!user_shm_lock(inode->i_size, user))
1063                         goto out_nomem;
1064                 info->flags |= VM_LOCKED;
1065                 mapping_set_unevictable(file->f_mapping);
1066         }
1067         if (!lock && (info->flags & VM_LOCKED) && user) {
1068                 user_shm_unlock(inode->i_size, user);
1069                 info->flags &= ~VM_LOCKED;
1070                 mapping_clear_unevictable(file->f_mapping);
1071                 /*
1072                  * Ensure that a racing putback_lru_page() can see
1073                  * the pages of this mapping are evictable when we
1074                  * skip them due to !PageLRU during the scan.
1075                  */
1076                 smp_mb__after_clear_bit();
1077                 scan_mapping_unevictable_pages(file->f_mapping);
1078         }
1079         retval = 0;
1080
1081 out_nomem:
1082         spin_unlock(&info->lock);
1083         return retval;
1084 }
1085
1086 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1087 {
1088         file_accessed(file);
1089         vma->vm_ops = &shmem_vm_ops;
1090         vma->vm_flags |= VM_CAN_NONLINEAR;
1091         return 0;
1092 }
1093
1094 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1095                                      umode_t mode, dev_t dev, unsigned long flags)
1096 {
1097         struct inode *inode;
1098         struct shmem_inode_info *info;
1099         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1100
1101         if (shmem_reserve_inode(sb))
1102                 return NULL;
1103
1104         inode = new_inode(sb);
1105         if (inode) {
1106                 inode->i_ino = get_next_ino();
1107                 inode_init_owner(inode, dir, mode);
1108                 inode->i_blocks = 0;
1109                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1110                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1111                 inode->i_generation = get_seconds();
1112                 info = SHMEM_I(inode);
1113                 memset(info, 0, (char *)inode - (char *)info);
1114                 spin_lock_init(&info->lock);
1115                 info->flags = flags & VM_NORESERVE;
1116                 INIT_LIST_HEAD(&info->swaplist);
1117                 INIT_LIST_HEAD(&info->xattr_list);
1118                 cache_no_acl(inode);
1119
1120                 switch (mode & S_IFMT) {
1121                 default:
1122                         inode->i_op = &shmem_special_inode_operations;
1123                         init_special_inode(inode, mode, dev);
1124                         break;
1125                 case S_IFREG:
1126                         inode->i_mapping->a_ops = &shmem_aops;
1127                         inode->i_op = &shmem_inode_operations;
1128                         inode->i_fop = &shmem_file_operations;
1129                         mpol_shared_policy_init(&info->policy,
1130                                                  shmem_get_sbmpol(sbinfo));
1131                         break;
1132                 case S_IFDIR:
1133                         inc_nlink(inode);
1134                         /* Some things misbehave if size == 0 on a directory */
1135                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1136                         inode->i_op = &shmem_dir_inode_operations;
1137                         inode->i_fop = &simple_dir_operations;
1138                         break;
1139                 case S_IFLNK:
1140                         /*
1141                          * Must not load anything in the rbtree,
1142                          * mpol_free_shared_policy will not be called.
1143                          */
1144                         mpol_shared_policy_init(&info->policy, NULL);
1145                         break;
1146                 }
1147         } else
1148                 shmem_free_inode(sb);
1149         return inode;
1150 }
1151
1152 #ifdef CONFIG_TMPFS
1153 static const struct inode_operations shmem_symlink_inode_operations;
1154 static const struct inode_operations shmem_short_symlink_operations;
1155
1156 static int
1157 shmem_write_begin(struct file *file, struct address_space *mapping,
1158                         loff_t pos, unsigned len, unsigned flags,
1159                         struct page **pagep, void **fsdata)
1160 {
1161         struct inode *inode = mapping->host;
1162         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1163         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1164 }
1165
1166 static int
1167 shmem_write_end(struct file *file, struct address_space *mapping,
1168                         loff_t pos, unsigned len, unsigned copied,
1169                         struct page *page, void *fsdata)
1170 {
1171         struct inode *inode = mapping->host;
1172
1173         if (pos + copied > inode->i_size)
1174                 i_size_write(inode, pos + copied);
1175
1176         set_page_dirty(page);
1177         unlock_page(page);
1178         page_cache_release(page);
1179
1180         return copied;
1181 }
1182
1183 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1184 {
1185         struct inode *inode = filp->f_path.dentry->d_inode;
1186         struct address_space *mapping = inode->i_mapping;
1187         pgoff_t index;
1188         unsigned long offset;
1189         enum sgp_type sgp = SGP_READ;
1190
1191         /*
1192          * Might this read be for a stacking filesystem?  Then when reading
1193          * holes of a sparse file, we actually need to allocate those pages,
1194          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1195          */
1196         if (segment_eq(get_fs(), KERNEL_DS))
1197                 sgp = SGP_DIRTY;
1198
1199         index = *ppos >> PAGE_CACHE_SHIFT;
1200         offset = *ppos & ~PAGE_CACHE_MASK;
1201
1202         for (;;) {
1203                 struct page *page = NULL;
1204                 pgoff_t end_index;
1205                 unsigned long nr, ret;
1206                 loff_t i_size = i_size_read(inode);
1207
1208                 end_index = i_size >> PAGE_CACHE_SHIFT;
1209                 if (index > end_index)
1210                         break;
1211                 if (index == end_index) {
1212                         nr = i_size & ~PAGE_CACHE_MASK;
1213                         if (nr <= offset)
1214                                 break;
1215                 }
1216
1217                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1218                 if (desc->error) {
1219                         if (desc->error == -EINVAL)
1220                                 desc->error = 0;
1221                         break;
1222                 }
1223                 if (page)
1224                         unlock_page(page);
1225
1226                 /*
1227                  * We must evaluate after, since reads (unlike writes)
1228                  * are called without i_mutex protection against truncate
1229                  */
1230                 nr = PAGE_CACHE_SIZE;
1231                 i_size = i_size_read(inode);
1232                 end_index = i_size >> PAGE_CACHE_SHIFT;
1233                 if (index == end_index) {
1234                         nr = i_size & ~PAGE_CACHE_MASK;
1235                         if (nr <= offset) {
1236                                 if (page)
1237                                         page_cache_release(page);
1238                                 break;
1239                         }
1240                 }
1241                 nr -= offset;
1242
1243                 if (page) {
1244                         /*
1245                          * If users can be writing to this page using arbitrary
1246                          * virtual addresses, take care about potential aliasing
1247                          * before reading the page on the kernel side.
1248                          */
1249                         if (mapping_writably_mapped(mapping))
1250                                 flush_dcache_page(page);
1251                         /*
1252                          * Mark the page accessed if we read the beginning.
1253                          */
1254                         if (!offset)
1255                                 mark_page_accessed(page);
1256                 } else {
1257                         page = ZERO_PAGE(0);
1258                         page_cache_get(page);
1259                 }
1260
1261                 /*
1262                  * Ok, we have the page, and it's up-to-date, so
1263                  * now we can copy it to user space...
1264                  *
1265                  * The actor routine returns how many bytes were actually used..
1266                  * NOTE! This may not be the same as how much of a user buffer
1267                  * we filled up (we may be padding etc), so we can only update
1268                  * "pos" here (the actor routine has to update the user buffer
1269                  * pointers and the remaining count).
1270                  */
1271                 ret = actor(desc, page, offset, nr);
1272                 offset += ret;
1273                 index += offset >> PAGE_CACHE_SHIFT;
1274                 offset &= ~PAGE_CACHE_MASK;
1275
1276                 page_cache_release(page);
1277                 if (ret != nr || !desc->count)
1278                         break;
1279
1280                 cond_resched();
1281         }
1282
1283         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1284         file_accessed(filp);
1285 }
1286
1287 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1288                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1289 {
1290         struct file *filp = iocb->ki_filp;
1291         ssize_t retval;
1292         unsigned long seg;
1293         size_t count;
1294         loff_t *ppos = &iocb->ki_pos;
1295
1296         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1297         if (retval)
1298                 return retval;
1299
1300         for (seg = 0; seg < nr_segs; seg++) {
1301                 read_descriptor_t desc;
1302
1303                 desc.written = 0;
1304                 desc.arg.buf = iov[seg].iov_base;
1305                 desc.count = iov[seg].iov_len;
1306                 if (desc.count == 0)
1307                         continue;
1308                 desc.error = 0;
1309                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1310                 retval += desc.written;
1311                 if (desc.error) {
1312                         retval = retval ?: desc.error;
1313                         break;
1314                 }
1315                 if (desc.count > 0)
1316                         break;
1317         }
1318         return retval;
1319 }
1320
1321 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1322                                 struct pipe_inode_info *pipe, size_t len,
1323                                 unsigned int flags)
1324 {
1325         struct address_space *mapping = in->f_mapping;
1326         struct inode *inode = mapping->host;
1327         unsigned int loff, nr_pages, req_pages;
1328         struct page *pages[PIPE_DEF_BUFFERS];
1329         struct partial_page partial[PIPE_DEF_BUFFERS];
1330         struct page *page;
1331         pgoff_t index, end_index;
1332         loff_t isize, left;
1333         int error, page_nr;
1334         struct splice_pipe_desc spd = {
1335                 .pages = pages,
1336                 .partial = partial,
1337                 .flags = flags,
1338                 .ops = &page_cache_pipe_buf_ops,
1339                 .spd_release = spd_release_page,
1340         };
1341
1342         isize = i_size_read(inode);
1343         if (unlikely(*ppos >= isize))
1344                 return 0;
1345
1346         left = isize - *ppos;
1347         if (unlikely(left < len))
1348                 len = left;
1349
1350         if (splice_grow_spd(pipe, &spd))
1351                 return -ENOMEM;
1352
1353         index = *ppos >> PAGE_CACHE_SHIFT;
1354         loff = *ppos & ~PAGE_CACHE_MASK;
1355         req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1356         nr_pages = min(req_pages, pipe->buffers);
1357
1358         spd.nr_pages = find_get_pages_contig(mapping, index,
1359                                                 nr_pages, spd.pages);
1360         index += spd.nr_pages;
1361         error = 0;
1362
1363         while (spd.nr_pages < nr_pages) {
1364                 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1365                 if (error)
1366                         break;
1367                 unlock_page(page);
1368                 spd.pages[spd.nr_pages++] = page;
1369                 index++;
1370         }
1371
1372         index = *ppos >> PAGE_CACHE_SHIFT;
1373         nr_pages = spd.nr_pages;
1374         spd.nr_pages = 0;
1375
1376         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1377                 unsigned int this_len;
1378
1379                 if (!len)
1380                         break;
1381
1382                 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1383                 page = spd.pages[page_nr];
1384
1385                 if (!PageUptodate(page) || page->mapping != mapping) {
1386                         error = shmem_getpage(inode, index, &page,
1387                                                         SGP_CACHE, NULL);
1388                         if (error)
1389                                 break;
1390                         unlock_page(page);
1391                         page_cache_release(spd.pages[page_nr]);
1392                         spd.pages[page_nr] = page;
1393                 }
1394
1395                 isize = i_size_read(inode);
1396                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1397                 if (unlikely(!isize || index > end_index))
1398                         break;
1399
1400                 if (end_index == index) {
1401                         unsigned int plen;
1402
1403                         plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1404                         if (plen <= loff)
1405                                 break;
1406
1407                         this_len = min(this_len, plen - loff);
1408                         len = this_len;
1409                 }
1410
1411                 spd.partial[page_nr].offset = loff;
1412                 spd.partial[page_nr].len = this_len;
1413                 len -= this_len;
1414                 loff = 0;
1415                 spd.nr_pages++;
1416                 index++;
1417         }
1418
1419         while (page_nr < nr_pages)
1420                 page_cache_release(spd.pages[page_nr++]);
1421
1422         if (spd.nr_pages)
1423                 error = splice_to_pipe(pipe, &spd);
1424
1425         splice_shrink_spd(pipe, &spd);
1426
1427         if (error > 0) {
1428                 *ppos += error;
1429                 file_accessed(in);
1430         }
1431         return error;
1432 }
1433
1434 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1435 {
1436         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1437
1438         buf->f_type = TMPFS_MAGIC;
1439         buf->f_bsize = PAGE_CACHE_SIZE;
1440         buf->f_namelen = NAME_MAX;
1441         if (sbinfo->max_blocks) {
1442                 buf->f_blocks = sbinfo->max_blocks;
1443                 buf->f_bavail =
1444                 buf->f_bfree  = sbinfo->max_blocks -
1445                                 percpu_counter_sum(&sbinfo->used_blocks);
1446         }
1447         if (sbinfo->max_inodes) {
1448                 buf->f_files = sbinfo->max_inodes;
1449                 buf->f_ffree = sbinfo->free_inodes;
1450         }
1451         /* else leave those fields 0 like simple_statfs */
1452         return 0;
1453 }
1454
1455 /*
1456  * File creation. Allocate an inode, and we're done..
1457  */
1458 static int
1459 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1460 {
1461         struct inode *inode;
1462         int error = -ENOSPC;
1463
1464         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1465         if (inode) {
1466                 error = security_inode_init_security(inode, dir,
1467                                                      &dentry->d_name,
1468                                                      NULL, NULL);
1469                 if (error) {
1470                         if (error != -EOPNOTSUPP) {
1471                                 iput(inode);
1472                                 return error;
1473                         }
1474                 }
1475 #ifdef CONFIG_TMPFS_POSIX_ACL
1476                 error = generic_acl_init(inode, dir);
1477                 if (error) {
1478                         iput(inode);
1479                         return error;
1480                 }
1481 #else
1482                 error = 0;
1483 #endif
1484                 dir->i_size += BOGO_DIRENT_SIZE;
1485                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1486                 d_instantiate(dentry, inode);
1487                 dget(dentry); /* Extra count - pin the dentry in core */
1488         }
1489         return error;
1490 }
1491
1492 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1493 {
1494         int error;
1495
1496         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1497                 return error;
1498         inc_nlink(dir);
1499         return 0;
1500 }
1501
1502 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1503                 struct nameidata *nd)
1504 {
1505         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1506 }
1507
1508 /*
1509  * Link a file..
1510  */
1511 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1512 {
1513         struct inode *inode = old_dentry->d_inode;
1514         int ret;
1515
1516         /*
1517          * No ordinary (disk based) filesystem counts links as inodes;
1518          * but each new link needs a new dentry, pinning lowmem, and
1519          * tmpfs dentries cannot be pruned until they are unlinked.
1520          */
1521         ret = shmem_reserve_inode(inode->i_sb);
1522         if (ret)
1523                 goto out;
1524
1525         dir->i_size += BOGO_DIRENT_SIZE;
1526         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1527         inc_nlink(inode);
1528         ihold(inode);   /* New dentry reference */
1529         dget(dentry);           /* Extra pinning count for the created dentry */
1530         d_instantiate(dentry, inode);
1531 out:
1532         return ret;
1533 }
1534
1535 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1536 {
1537         struct inode *inode = dentry->d_inode;
1538
1539         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1540                 shmem_free_inode(inode->i_sb);
1541
1542         dir->i_size -= BOGO_DIRENT_SIZE;
1543         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1544         drop_nlink(inode);
1545         dput(dentry);   /* Undo the count from "create" - this does all the work */
1546         return 0;
1547 }
1548
1549 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1550 {
1551         if (!simple_empty(dentry))
1552                 return -ENOTEMPTY;
1553
1554         drop_nlink(dentry->d_inode);
1555         drop_nlink(dir);
1556         return shmem_unlink(dir, dentry);
1557 }
1558
1559 /*
1560  * The VFS layer already does all the dentry stuff for rename,
1561  * we just have to decrement the usage count for the target if
1562  * it exists so that the VFS layer correctly free's it when it
1563  * gets overwritten.
1564  */
1565 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1566 {
1567         struct inode *inode = old_dentry->d_inode;
1568         int they_are_dirs = S_ISDIR(inode->i_mode);
1569
1570         if (!simple_empty(new_dentry))
1571                 return -ENOTEMPTY;
1572
1573         if (new_dentry->d_inode) {
1574                 (void) shmem_unlink(new_dir, new_dentry);
1575                 if (they_are_dirs)
1576                         drop_nlink(old_dir);
1577         } else if (they_are_dirs) {
1578                 drop_nlink(old_dir);
1579                 inc_nlink(new_dir);
1580         }
1581
1582         old_dir->i_size -= BOGO_DIRENT_SIZE;
1583         new_dir->i_size += BOGO_DIRENT_SIZE;
1584         old_dir->i_ctime = old_dir->i_mtime =
1585         new_dir->i_ctime = new_dir->i_mtime =
1586         inode->i_ctime = CURRENT_TIME;
1587         return 0;
1588 }
1589
1590 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1591 {
1592         int error;
1593         int len;
1594         struct inode *inode;
1595         struct page *page;
1596         char *kaddr;
1597         struct shmem_inode_info *info;
1598
1599         len = strlen(symname) + 1;
1600         if (len > PAGE_CACHE_SIZE)
1601                 return -ENAMETOOLONG;
1602
1603         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1604         if (!inode)
1605                 return -ENOSPC;
1606
1607         error = security_inode_init_security(inode, dir, &dentry->d_name,
1608                                              NULL, NULL);
1609         if (error) {
1610                 if (error != -EOPNOTSUPP) {
1611                         iput(inode);
1612                         return error;
1613                 }
1614                 error = 0;
1615         }
1616
1617         info = SHMEM_I(inode);
1618         inode->i_size = len-1;
1619         if (len <= SHORT_SYMLINK_LEN) {
1620                 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1621                 if (!info->symlink) {
1622                         iput(inode);
1623                         return -ENOMEM;
1624                 }
1625                 inode->i_op = &shmem_short_symlink_operations;
1626         } else {
1627                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1628                 if (error) {
1629                         iput(inode);
1630                         return error;
1631                 }
1632                 inode->i_mapping->a_ops = &shmem_aops;
1633                 inode->i_op = &shmem_symlink_inode_operations;
1634                 kaddr = kmap_atomic(page, KM_USER0);
1635                 memcpy(kaddr, symname, len);
1636                 kunmap_atomic(kaddr, KM_USER0);
1637                 set_page_dirty(page);
1638                 unlock_page(page);
1639                 page_cache_release(page);
1640         }
1641         dir->i_size += BOGO_DIRENT_SIZE;
1642         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1643         d_instantiate(dentry, inode);
1644         dget(dentry);
1645         return 0;
1646 }
1647
1648 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
1649 {
1650         nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
1651         return NULL;
1652 }
1653
1654 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1655 {
1656         struct page *page = NULL;
1657         int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1658         nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
1659         if (page)
1660                 unlock_page(page);
1661         return page;
1662 }
1663
1664 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1665 {
1666         if (!IS_ERR(nd_get_link(nd))) {
1667                 struct page *page = cookie;
1668                 kunmap(page);
1669                 mark_page_accessed(page);
1670                 page_cache_release(page);
1671         }
1672 }
1673
1674 #ifdef CONFIG_TMPFS_XATTR
1675 /*
1676  * Superblocks without xattr inode operations may get some security.* xattr
1677  * support from the LSM "for free". As soon as we have any other xattrs
1678  * like ACLs, we also need to implement the security.* handlers at
1679  * filesystem level, though.
1680  */
1681
1682 static int shmem_xattr_get(struct dentry *dentry, const char *name,
1683                            void *buffer, size_t size)
1684 {
1685         struct shmem_inode_info *info;
1686         struct shmem_xattr *xattr;
1687         int ret = -ENODATA;
1688
1689         info = SHMEM_I(dentry->d_inode);
1690
1691         spin_lock(&info->lock);
1692         list_for_each_entry(xattr, &info->xattr_list, list) {
1693                 if (strcmp(name, xattr->name))
1694                         continue;
1695
1696                 ret = xattr->size;
1697                 if (buffer) {
1698                         if (size < xattr->size)
1699                                 ret = -ERANGE;
1700                         else
1701                                 memcpy(buffer, xattr->value, xattr->size);
1702                 }
1703                 break;
1704         }
1705         spin_unlock(&info->lock);
1706         return ret;
1707 }
1708
1709 static int shmem_xattr_set(struct dentry *dentry, const char *name,
1710                            const void *value, size_t size, int flags)
1711 {
1712         struct inode *inode = dentry->d_inode;
1713         struct shmem_inode_info *info = SHMEM_I(inode);
1714         struct shmem_xattr *xattr;
1715         struct shmem_xattr *new_xattr = NULL;
1716         size_t len;
1717         int err = 0;
1718
1719         /* value == NULL means remove */
1720         if (value) {
1721                 /* wrap around? */
1722                 len = sizeof(*new_xattr) + size;
1723                 if (len <= sizeof(*new_xattr))
1724                         return -ENOMEM;
1725
1726                 new_xattr = kmalloc(len, GFP_KERNEL);
1727                 if (!new_xattr)
1728                         return -ENOMEM;
1729
1730                 new_xattr->name = kstrdup(name, GFP_KERNEL);
1731                 if (!new_xattr->name) {
1732                         kfree(new_xattr);
1733                         return -ENOMEM;
1734                 }
1735
1736                 new_xattr->size = size;
1737                 memcpy(new_xattr->value, value, size);
1738         }
1739
1740         spin_lock(&info->lock);
1741         list_for_each_entry(xattr, &info->xattr_list, list) {
1742                 if (!strcmp(name, xattr->name)) {
1743                         if (flags & XATTR_CREATE) {
1744                                 xattr = new_xattr;
1745                                 err = -EEXIST;
1746                         } else if (new_xattr) {
1747                                 list_replace(&xattr->list, &new_xattr->list);
1748                         } else {
1749                                 list_del(&xattr->list);
1750                         }
1751                         goto out;
1752                 }
1753         }
1754         if (flags & XATTR_REPLACE) {
1755                 xattr = new_xattr;
1756                 err = -ENODATA;
1757         } else {
1758                 list_add(&new_xattr->list, &info->xattr_list);
1759                 xattr = NULL;
1760         }
1761 out:
1762         spin_unlock(&info->lock);
1763         if (xattr)
1764                 kfree(xattr->name);
1765         kfree(xattr);
1766         return err;
1767 }
1768
1769 static const struct xattr_handler *shmem_xattr_handlers[] = {
1770 #ifdef CONFIG_TMPFS_POSIX_ACL
1771         &generic_acl_access_handler,
1772         &generic_acl_default_handler,
1773 #endif
1774         NULL
1775 };
1776
1777 static int shmem_xattr_validate(const char *name)
1778 {
1779         struct { const char *prefix; size_t len; } arr[] = {
1780                 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1781                 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1782         };
1783         int i;
1784
1785         for (i = 0; i < ARRAY_SIZE(arr); i++) {
1786                 size_t preflen = arr[i].len;
1787                 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1788                         if (!name[preflen])
1789                                 return -EINVAL;
1790                         return 0;
1791                 }
1792         }
1793         return -EOPNOTSUPP;
1794 }
1795
1796 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1797                               void *buffer, size_t size)
1798 {
1799         int err;
1800
1801         /*
1802          * If this is a request for a synthetic attribute in the system.*
1803          * namespace use the generic infrastructure to resolve a handler
1804          * for it via sb->s_xattr.
1805          */
1806         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1807                 return generic_getxattr(dentry, name, buffer, size);
1808
1809         err = shmem_xattr_validate(name);
1810         if (err)
1811                 return err;
1812
1813         return shmem_xattr_get(dentry, name, buffer, size);
1814 }
1815
1816 static int shmem_setxattr(struct dentry *dentry, const char *name,
1817                           const void *value, size_t size, int flags)
1818 {
1819         int err;
1820
1821         /*
1822          * If this is a request for a synthetic attribute in the system.*
1823          * namespace use the generic infrastructure to resolve a handler
1824          * for it via sb->s_xattr.
1825          */
1826         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1827                 return generic_setxattr(dentry, name, value, size, flags);
1828
1829         err = shmem_xattr_validate(name);
1830         if (err)
1831                 return err;
1832
1833         if (size == 0)
1834                 value = "";  /* empty EA, do not remove */
1835
1836         return shmem_xattr_set(dentry, name, value, size, flags);
1837
1838 }
1839
1840 static int shmem_removexattr(struct dentry *dentry, const char *name)
1841 {
1842         int err;
1843
1844         /*
1845          * If this is a request for a synthetic attribute in the system.*
1846          * namespace use the generic infrastructure to resolve a handler
1847          * for it via sb->s_xattr.
1848          */
1849         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1850                 return generic_removexattr(dentry, name);
1851
1852         err = shmem_xattr_validate(name);
1853         if (err)
1854                 return err;
1855
1856         return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
1857 }
1858
1859 static bool xattr_is_trusted(const char *name)
1860 {
1861         return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1862 }
1863
1864 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1865 {
1866         bool trusted = capable(CAP_SYS_ADMIN);
1867         struct shmem_xattr *xattr;
1868         struct shmem_inode_info *info;
1869         size_t used = 0;
1870
1871         info = SHMEM_I(dentry->d_inode);
1872
1873         spin_lock(&info->lock);
1874         list_for_each_entry(xattr, &info->xattr_list, list) {
1875                 size_t len;
1876
1877                 /* skip "trusted." attributes for unprivileged callers */
1878                 if (!trusted && xattr_is_trusted(xattr->name))
1879                         continue;
1880
1881                 len = strlen(xattr->name) + 1;
1882                 used += len;
1883                 if (buffer) {
1884                         if (size < used) {
1885                                 used = -ERANGE;
1886                                 break;
1887                         }
1888                         memcpy(buffer, xattr->name, len);
1889                         buffer += len;
1890                 }
1891         }
1892         spin_unlock(&info->lock);
1893
1894         return used;
1895 }
1896 #endif /* CONFIG_TMPFS_XATTR */
1897
1898 static const struct inode_operations shmem_short_symlink_operations = {
1899         .readlink       = generic_readlink,
1900         .follow_link    = shmem_follow_short_symlink,
1901 #ifdef CONFIG_TMPFS_XATTR
1902         .setxattr       = shmem_setxattr,
1903         .getxattr       = shmem_getxattr,
1904         .listxattr      = shmem_listxattr,
1905         .removexattr    = shmem_removexattr,
1906 #endif
1907 };
1908
1909 static const struct inode_operations shmem_symlink_inode_operations = {
1910         .readlink       = generic_readlink,
1911         .follow_link    = shmem_follow_link,
1912         .put_link       = shmem_put_link,
1913 #ifdef CONFIG_TMPFS_XATTR
1914         .setxattr       = shmem_setxattr,
1915         .getxattr       = shmem_getxattr,
1916         .listxattr      = shmem_listxattr,
1917         .removexattr    = shmem_removexattr,
1918 #endif
1919 };
1920
1921 static struct dentry *shmem_get_parent(struct dentry *child)
1922 {
1923         return ERR_PTR(-ESTALE);
1924 }
1925
1926 static int shmem_match(struct inode *ino, void *vfh)
1927 {
1928         __u32 *fh = vfh;
1929         __u64 inum = fh[2];
1930         inum = (inum << 32) | fh[1];
1931         return ino->i_ino == inum && fh[0] == ino->i_generation;
1932 }
1933
1934 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1935                 struct fid *fid, int fh_len, int fh_type)
1936 {
1937         struct inode *inode;
1938         struct dentry *dentry = NULL;
1939         u64 inum = fid->raw[2];
1940         inum = (inum << 32) | fid->raw[1];
1941
1942         if (fh_len < 3)
1943                 return NULL;
1944
1945         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1946                         shmem_match, fid->raw);
1947         if (inode) {
1948                 dentry = d_find_alias(inode);
1949                 iput(inode);
1950         }
1951
1952         return dentry;
1953 }
1954
1955 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1956                                 int connectable)
1957 {
1958         struct inode *inode = dentry->d_inode;
1959
1960         if (*len < 3) {
1961                 *len = 3;
1962                 return 255;
1963         }
1964
1965         if (inode_unhashed(inode)) {
1966                 /* Unfortunately insert_inode_hash is not idempotent,
1967                  * so as we hash inodes here rather than at creation
1968                  * time, we need a lock to ensure we only try
1969                  * to do it once
1970                  */
1971                 static DEFINE_SPINLOCK(lock);
1972                 spin_lock(&lock);
1973                 if (inode_unhashed(inode))
1974                         __insert_inode_hash(inode,
1975                                             inode->i_ino + inode->i_generation);
1976                 spin_unlock(&lock);
1977         }
1978
1979         fh[0] = inode->i_generation;
1980         fh[1] = inode->i_ino;
1981         fh[2] = ((__u64)inode->i_ino) >> 32;
1982
1983         *len = 3;
1984         return 1;
1985 }
1986
1987 static const struct export_operations shmem_export_ops = {
1988         .get_parent     = shmem_get_parent,
1989         .encode_fh      = shmem_encode_fh,
1990         .fh_to_dentry   = shmem_fh_to_dentry,
1991 };
1992
1993 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
1994                                bool remount)
1995 {
1996         char *this_char, *value, *rest;
1997
1998         while (options != NULL) {
1999                 this_char = options;
2000                 for (;;) {
2001                         /*
2002                          * NUL-terminate this option: unfortunately,
2003                          * mount options form a comma-separated list,
2004                          * but mpol's nodelist may also contain commas.
2005                          */
2006                         options = strchr(options, ',');
2007                         if (options == NULL)
2008                                 break;
2009                         options++;
2010                         if (!isdigit(*options)) {
2011                                 options[-1] = '\0';
2012                                 break;
2013                         }
2014                 }
2015                 if (!*this_char)
2016                         continue;
2017                 if ((value = strchr(this_char,'=')) != NULL) {
2018                         *value++ = 0;
2019                 } else {
2020                         printk(KERN_ERR
2021                             "tmpfs: No value for mount option '%s'\n",
2022                             this_char);
2023                         return 1;
2024                 }
2025
2026                 if (!strcmp(this_char,"size")) {
2027                         unsigned long long size;
2028                         size = memparse(value,&rest);
2029                         if (*rest == '%') {
2030                                 size <<= PAGE_SHIFT;
2031                                 size *= totalram_pages;
2032                                 do_div(size, 100);
2033                                 rest++;
2034                         }
2035                         if (*rest)
2036                                 goto bad_val;
2037                         sbinfo->max_blocks =
2038                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2039                 } else if (!strcmp(this_char,"nr_blocks")) {
2040                         sbinfo->max_blocks = memparse(value, &rest);
2041                         if (*rest)
2042                                 goto bad_val;
2043                 } else if (!strcmp(this_char,"nr_inodes")) {
2044                         sbinfo->max_inodes = memparse(value, &rest);
2045                         if (*rest)
2046                                 goto bad_val;
2047                 } else if (!strcmp(this_char,"mode")) {
2048                         if (remount)
2049                                 continue;
2050                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2051                         if (*rest)
2052                                 goto bad_val;
2053                 } else if (!strcmp(this_char,"uid")) {
2054                         if (remount)
2055                                 continue;
2056                         sbinfo->uid = simple_strtoul(value, &rest, 0);
2057                         if (*rest)
2058                                 goto bad_val;
2059                 } else if (!strcmp(this_char,"gid")) {
2060                         if (remount)
2061                                 continue;
2062                         sbinfo->gid = simple_strtoul(value, &rest, 0);
2063                         if (*rest)
2064                                 goto bad_val;
2065                 } else if (!strcmp(this_char,"mpol")) {
2066                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2067                                 goto bad_val;
2068                 } else {
2069                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2070                                this_char);
2071                         return 1;
2072                 }
2073         }
2074         return 0;
2075
2076 bad_val:
2077         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2078                value, this_char);
2079         return 1;
2080
2081 }
2082
2083 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2084 {
2085         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2086         struct shmem_sb_info config = *sbinfo;
2087         unsigned long inodes;
2088         int error = -EINVAL;
2089
2090         if (shmem_parse_options(data, &config, true))
2091                 return error;
2092
2093         spin_lock(&sbinfo->stat_lock);
2094         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2095         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2096                 goto out;
2097         if (config.max_inodes < inodes)
2098                 goto out;
2099         /*
2100          * Those tests disallow limited->unlimited while any are in use;
2101          * but we must separately disallow unlimited->limited, because
2102          * in that case we have no record of how much is already in use.
2103          */
2104         if (config.max_blocks && !sbinfo->max_blocks)
2105                 goto out;
2106         if (config.max_inodes && !sbinfo->max_inodes)
2107                 goto out;
2108
2109         error = 0;
2110         sbinfo->max_blocks  = config.max_blocks;
2111         sbinfo->max_inodes  = config.max_inodes;
2112         sbinfo->free_inodes = config.max_inodes - inodes;
2113
2114         mpol_put(sbinfo->mpol);
2115         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2116 out:
2117         spin_unlock(&sbinfo->stat_lock);
2118         return error;
2119 }
2120
2121 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2122 {
2123         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2124
2125         if (sbinfo->max_blocks != shmem_default_max_blocks())
2126                 seq_printf(seq, ",size=%luk",
2127                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2128         if (sbinfo->max_inodes != shmem_default_max_inodes())
2129                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2130         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2131                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2132         if (sbinfo->uid != 0)
2133                 seq_printf(seq, ",uid=%u", sbinfo->uid);
2134         if (sbinfo->gid != 0)
2135                 seq_printf(seq, ",gid=%u", sbinfo->gid);
2136         shmem_show_mpol(seq, sbinfo->mpol);
2137         return 0;
2138 }
2139 #endif /* CONFIG_TMPFS */
2140
2141 static void shmem_put_super(struct super_block *sb)
2142 {
2143         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2144
2145         percpu_counter_destroy(&sbinfo->used_blocks);
2146         kfree(sbinfo);
2147         sb->s_fs_info = NULL;
2148 }
2149
2150 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2151 {
2152         struct inode *inode;
2153         struct dentry *root;
2154         struct shmem_sb_info *sbinfo;
2155         int err = -ENOMEM;
2156
2157         /* Round up to L1_CACHE_BYTES to resist false sharing */
2158         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2159                                 L1_CACHE_BYTES), GFP_KERNEL);
2160         if (!sbinfo)
2161                 return -ENOMEM;
2162
2163         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2164         sbinfo->uid = current_fsuid();
2165         sbinfo->gid = current_fsgid();
2166         sb->s_fs_info = sbinfo;
2167
2168 #ifdef CONFIG_TMPFS
2169         /*
2170          * Per default we only allow half of the physical ram per
2171          * tmpfs instance, limiting inodes to one per page of lowmem;
2172          * but the internal instance is left unlimited.
2173          */
2174         if (!(sb->s_flags & MS_NOUSER)) {
2175                 sbinfo->max_blocks = shmem_default_max_blocks();
2176                 sbinfo->max_inodes = shmem_default_max_inodes();
2177                 if (shmem_parse_options(data, sbinfo, false)) {
2178                         err = -EINVAL;
2179                         goto failed;
2180                 }
2181         }
2182         sb->s_export_op = &shmem_export_ops;
2183 #else
2184         sb->s_flags |= MS_NOUSER;
2185 #endif
2186
2187         spin_lock_init(&sbinfo->stat_lock);
2188         if (percpu_counter_init(&sbinfo->used_blocks, 0))
2189                 goto failed;
2190         sbinfo->free_inodes = sbinfo->max_inodes;
2191
2192         sb->s_maxbytes = MAX_LFS_FILESIZE;
2193         sb->s_blocksize = PAGE_CACHE_SIZE;
2194         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2195         sb->s_magic = TMPFS_MAGIC;
2196         sb->s_op = &shmem_ops;
2197         sb->s_time_gran = 1;
2198 #ifdef CONFIG_TMPFS_XATTR
2199         sb->s_xattr = shmem_xattr_handlers;
2200 #endif
2201 #ifdef CONFIG_TMPFS_POSIX_ACL
2202         sb->s_flags |= MS_POSIXACL;
2203 #endif
2204
2205         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2206         if (!inode)
2207                 goto failed;
2208         inode->i_uid = sbinfo->uid;
2209         inode->i_gid = sbinfo->gid;
2210         root = d_alloc_root(inode);
2211         if (!root)
2212                 goto failed_iput;
2213         sb->s_root = root;
2214         return 0;
2215
2216 failed_iput:
2217         iput(inode);
2218 failed:
2219         shmem_put_super(sb);
2220         return err;
2221 }
2222
2223 static struct kmem_cache *shmem_inode_cachep;
2224
2225 static struct inode *shmem_alloc_inode(struct super_block *sb)
2226 {
2227         struct shmem_inode_info *info;
2228         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2229         if (!info)
2230                 return NULL;
2231         return &info->vfs_inode;
2232 }
2233
2234 static void shmem_destroy_callback(struct rcu_head *head)
2235 {
2236         struct inode *inode = container_of(head, struct inode, i_rcu);
2237         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2238 }
2239
2240 static void shmem_destroy_inode(struct inode *inode)
2241 {
2242         if (S_ISREG(inode->i_mode))
2243                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2244         call_rcu(&inode->i_rcu, shmem_destroy_callback);
2245 }
2246
2247 static void shmem_init_inode(void *foo)
2248 {
2249         struct shmem_inode_info *info = foo;
2250         inode_init_once(&info->vfs_inode);
2251 }
2252
2253 static int shmem_init_inodecache(void)
2254 {
2255         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2256                                 sizeof(struct shmem_inode_info),
2257                                 0, SLAB_PANIC, shmem_init_inode);
2258         return 0;
2259 }
2260
2261 static void shmem_destroy_inodecache(void)
2262 {
2263         kmem_cache_destroy(shmem_inode_cachep);
2264 }
2265
2266 static const struct address_space_operations shmem_aops = {
2267         .writepage      = shmem_writepage,
2268         .set_page_dirty = __set_page_dirty_no_writeback,
2269 #ifdef CONFIG_TMPFS
2270         .write_begin    = shmem_write_begin,
2271         .write_end      = shmem_write_end,
2272 #endif
2273         .migratepage    = migrate_page,
2274         .error_remove_page = generic_error_remove_page,
2275 };
2276
2277 static const struct file_operations shmem_file_operations = {
2278         .mmap           = shmem_mmap,
2279 #ifdef CONFIG_TMPFS
2280         .llseek         = generic_file_llseek,
2281         .read           = do_sync_read,
2282         .write          = do_sync_write,
2283         .aio_read       = shmem_file_aio_read,
2284         .aio_write      = generic_file_aio_write,
2285         .fsync          = noop_fsync,
2286         .splice_read    = shmem_file_splice_read,
2287         .splice_write   = generic_file_splice_write,
2288 #endif
2289 };
2290
2291 static const struct inode_operations shmem_inode_operations = {
2292         .setattr        = shmem_setattr,
2293         .truncate_range = shmem_truncate_range,
2294 #ifdef CONFIG_TMPFS_XATTR
2295         .setxattr       = shmem_setxattr,
2296         .getxattr       = shmem_getxattr,
2297         .listxattr      = shmem_listxattr,
2298         .removexattr    = shmem_removexattr,
2299 #endif
2300 };
2301
2302 static const struct inode_operations shmem_dir_inode_operations = {
2303 #ifdef CONFIG_TMPFS
2304         .create         = shmem_create,
2305         .lookup         = simple_lookup,
2306         .link           = shmem_link,
2307         .unlink         = shmem_unlink,
2308         .symlink        = shmem_symlink,
2309         .mkdir          = shmem_mkdir,
2310         .rmdir          = shmem_rmdir,
2311         .mknod          = shmem_mknod,
2312         .rename         = shmem_rename,
2313 #endif
2314 #ifdef CONFIG_TMPFS_XATTR
2315         .setxattr       = shmem_setxattr,
2316         .getxattr       = shmem_getxattr,
2317         .listxattr      = shmem_listxattr,
2318         .removexattr    = shmem_removexattr,
2319 #endif
2320 #ifdef CONFIG_TMPFS_POSIX_ACL
2321         .setattr        = shmem_setattr,
2322 #endif
2323 };
2324
2325 static const struct inode_operations shmem_special_inode_operations = {
2326 #ifdef CONFIG_TMPFS_XATTR
2327         .setxattr       = shmem_setxattr,
2328         .getxattr       = shmem_getxattr,
2329         .listxattr      = shmem_listxattr,
2330         .removexattr    = shmem_removexattr,
2331 #endif
2332 #ifdef CONFIG_TMPFS_POSIX_ACL
2333         .setattr        = shmem_setattr,
2334 #endif
2335 };
2336
2337 static const struct super_operations shmem_ops = {
2338         .alloc_inode    = shmem_alloc_inode,
2339         .destroy_inode  = shmem_destroy_inode,
2340 #ifdef CONFIG_TMPFS
2341         .statfs         = shmem_statfs,
2342         .remount_fs     = shmem_remount_fs,
2343         .show_options   = shmem_show_options,
2344 #endif
2345         .evict_inode    = shmem_evict_inode,
2346         .drop_inode     = generic_delete_inode,
2347         .put_super      = shmem_put_super,
2348 };
2349
2350 static const struct vm_operations_struct shmem_vm_ops = {
2351         .fault          = shmem_fault,
2352 #ifdef CONFIG_NUMA
2353         .set_policy     = shmem_set_policy,
2354         .get_policy     = shmem_get_policy,
2355 #endif
2356 };
2357
2358 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2359         int flags, const char *dev_name, void *data)
2360 {
2361         return mount_nodev(fs_type, flags, data, shmem_fill_super);
2362 }
2363
2364 static struct file_system_type shmem_fs_type = {
2365         .owner          = THIS_MODULE,
2366         .name           = "tmpfs",
2367         .mount          = shmem_mount,
2368         .kill_sb        = kill_litter_super,
2369 };
2370
2371 int __init shmem_init(void)
2372 {
2373         int error;
2374
2375         error = bdi_init(&shmem_backing_dev_info);
2376         if (error)
2377                 goto out4;
2378
2379         error = shmem_init_inodecache();
2380         if (error)
2381                 goto out3;
2382
2383         error = register_filesystem(&shmem_fs_type);
2384         if (error) {
2385                 printk(KERN_ERR "Could not register tmpfs\n");
2386                 goto out2;
2387         }
2388
2389         shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2390                                  shmem_fs_type.name, NULL);
2391         if (IS_ERR(shm_mnt)) {
2392                 error = PTR_ERR(shm_mnt);
2393                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2394                 goto out1;
2395         }
2396         return 0;
2397
2398 out1:
2399         unregister_filesystem(&shmem_fs_type);
2400 out2:
2401         shmem_destroy_inodecache();
2402 out3:
2403         bdi_destroy(&shmem_backing_dev_info);
2404 out4:
2405         shm_mnt = ERR_PTR(error);
2406         return error;
2407 }
2408
2409 #else /* !CONFIG_SHMEM */
2410
2411 /*
2412  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2413  *
2414  * This is intended for small system where the benefits of the full
2415  * shmem code (swap-backed and resource-limited) are outweighed by
2416  * their complexity. On systems without swap this code should be
2417  * effectively equivalent, but much lighter weight.
2418  */
2419
2420 #include <linux/ramfs.h>
2421
2422 static struct file_system_type shmem_fs_type = {
2423         .name           = "tmpfs",
2424         .mount          = ramfs_mount,
2425         .kill_sb        = kill_litter_super,
2426 };
2427
2428 int __init shmem_init(void)
2429 {
2430         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2431
2432         shm_mnt = kern_mount(&shmem_fs_type);
2433         BUG_ON(IS_ERR(shm_mnt));
2434
2435         return 0;
2436 }
2437
2438 int shmem_unuse(swp_entry_t swap, struct page *page)
2439 {
2440         return 0;
2441 }
2442
2443 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2444 {
2445         return 0;
2446 }
2447
2448 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2449 {
2450         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2451 }
2452 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2453
2454 #define shmem_vm_ops                            generic_file_vm_ops
2455 #define shmem_file_operations                   ramfs_file_operations
2456 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2457 #define shmem_acct_size(flags, size)            0
2458 #define shmem_unacct_size(flags, size)          do {} while (0)
2459
2460 #endif /* CONFIG_SHMEM */
2461
2462 /* common code */
2463
2464 /**
2465  * shmem_file_setup - get an unlinked file living in tmpfs
2466  * @name: name for dentry (to be seen in /proc/<pid>/maps
2467  * @size: size to be set for the file
2468  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2469  */
2470 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2471 {
2472         int error;
2473         struct file *file;
2474         struct inode *inode;
2475         struct path path;
2476         struct dentry *root;
2477         struct qstr this;
2478
2479         if (IS_ERR(shm_mnt))
2480                 return (void *)shm_mnt;
2481
2482         if (size < 0 || size > MAX_LFS_FILESIZE)
2483                 return ERR_PTR(-EINVAL);
2484
2485         if (shmem_acct_size(flags, size))
2486                 return ERR_PTR(-ENOMEM);
2487
2488         error = -ENOMEM;
2489         this.name = name;
2490         this.len = strlen(name);
2491         this.hash = 0; /* will go */
2492         root = shm_mnt->mnt_root;
2493         path.dentry = d_alloc(root, &this);
2494         if (!path.dentry)
2495                 goto put_memory;
2496         path.mnt = mntget(shm_mnt);
2497
2498         error = -ENOSPC;
2499         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2500         if (!inode)
2501                 goto put_dentry;
2502
2503         d_instantiate(path.dentry, inode);
2504         inode->i_size = size;
2505         clear_nlink(inode);     /* It is unlinked */
2506 #ifndef CONFIG_MMU
2507         error = ramfs_nommu_expand_for_mapping(inode, size);
2508         if (error)
2509                 goto put_dentry;
2510 #endif
2511
2512         error = -ENFILE;
2513         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2514                   &shmem_file_operations);
2515         if (!file)
2516                 goto put_dentry;
2517
2518         return file;
2519
2520 put_dentry:
2521         path_put(&path);
2522 put_memory:
2523         shmem_unacct_size(flags, size);
2524         return ERR_PTR(error);
2525 }
2526 EXPORT_SYMBOL_GPL(shmem_file_setup);
2527
2528 /**
2529  * shmem_zero_setup - setup a shared anonymous mapping
2530  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2531  */
2532 int shmem_zero_setup(struct vm_area_struct *vma)
2533 {
2534         struct file *file;
2535         loff_t size = vma->vm_end - vma->vm_start;
2536
2537         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2538         if (IS_ERR(file))
2539                 return PTR_ERR(file);
2540
2541         if (vma->vm_file)
2542                 fput(vma->vm_file);
2543         vma->vm_file = file;
2544         vma->vm_ops = &shmem_vm_ops;
2545         vma->vm_flags |= VM_CAN_NONLINEAR;
2546         return 0;
2547 }
2548
2549 /**
2550  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2551  * @mapping:    the page's address_space
2552  * @index:      the page index
2553  * @gfp:        the page allocator flags to use if allocating
2554  *
2555  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2556  * with any new page allocations done using the specified allocation flags.
2557  * But read_cache_page_gfp() uses the ->readpage() method: which does not
2558  * suit tmpfs, since it may have pages in swapcache, and needs to find those
2559  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2560  *
2561  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2562  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2563  */
2564 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2565                                          pgoff_t index, gfp_t gfp)
2566 {
2567 #ifdef CONFIG_SHMEM
2568         struct inode *inode = mapping->host;
2569         struct page *page;
2570         int error;
2571
2572         BUG_ON(mapping->a_ops != &shmem_aops);
2573         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2574         if (error)
2575                 page = ERR_PTR(error);
2576         else
2577                 unlock_page(page);
2578         return page;
2579 #else
2580         /*
2581          * The tiny !SHMEM case uses ramfs without swap
2582          */
2583         return read_cache_page_gfp(mapping, index, gfp);
2584 #endif
2585 }
2586 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);