always call inode_change_ok early in ->setattr
[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) 2002-2005 Hugh Dickins.
10  * Copyright (C) 2002-2005 VERITAS Software Corporation.
11  * Copyright (C) 2004 Andi Kleen, SuSE Labs
12  *
13  * Extended attribute support for tmpfs:
14  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
16  *
17  * tiny-shmem:
18  * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
19  *
20  * This file is released under the GPL.
21  */
22
23 #include <linux/fs.h>
24 #include <linux/init.h>
25 #include <linux/vfs.h>
26 #include <linux/mount.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
29 #include <linux/mm.h>
30 #include <linux/module.h>
31 #include <linux/swap.h>
32
33 static struct vfsmount *shm_mnt;
34
35 #ifdef CONFIG_SHMEM
36 /*
37  * This virtual memory filesystem is heavily based on the ramfs. It
38  * extends ramfs by the ability to use swap and honor resource limits
39  * which makes it a completely usable filesystem.
40  */
41
42 #include <linux/xattr.h>
43 #include <linux/exportfs.h>
44 #include <linux/posix_acl.h>
45 #include <linux/generic_acl.h>
46 #include <linux/mman.h>
47 #include <linux/string.h>
48 #include <linux/slab.h>
49 #include <linux/backing-dev.h>
50 #include <linux/shmem_fs.h>
51 #include <linux/writeback.h>
52 #include <linux/blkdev.h>
53 #include <linux/security.h>
54 #include <linux/swapops.h>
55 #include <linux/mempolicy.h>
56 #include <linux/namei.h>
57 #include <linux/ctype.h>
58 #include <linux/migrate.h>
59 #include <linux/highmem.h>
60 #include <linux/seq_file.h>
61 #include <linux/magic.h>
62
63 #include <asm/uaccess.h>
64 #include <asm/div64.h>
65 #include <asm/pgtable.h>
66
67 /*
68  * The maximum size of a shmem/tmpfs file is limited by the maximum size of
69  * its triple-indirect swap vector - see illustration at shmem_swp_entry().
70  *
71  * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
72  * but one eighth of that on a 64-bit kernel.  With 8kB page size, maximum
73  * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
74  * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
75  *
76  * We use / and * instead of shifts in the definitions below, so that the swap
77  * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
78  */
79 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
80 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
81
82 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
83 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
84
85 #define SHMEM_MAX_BYTES  min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
86 #define SHMEM_MAX_INDEX  ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
87
88 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
89 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
90
91 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
92 #define SHMEM_PAGEIN     VM_READ
93 #define SHMEM_TRUNCATE   VM_WRITE
94
95 /* Definition to limit shmem_truncate's steps between cond_rescheds */
96 #define LATENCY_LIMIT    64
97
98 /* Pretend that each entry is of this size in directory's i_size */
99 #define BOGO_DIRENT_SIZE 20
100
101 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
102 enum sgp_type {
103         SGP_READ,       /* don't exceed i_size, don't allocate page */
104         SGP_CACHE,      /* don't exceed i_size, may allocate page */
105         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
106         SGP_WRITE,      /* may exceed i_size, may allocate page */
107 };
108
109 #ifdef CONFIG_TMPFS
110 static unsigned long shmem_default_max_blocks(void)
111 {
112         return totalram_pages / 2;
113 }
114
115 static unsigned long shmem_default_max_inodes(void)
116 {
117         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
118 }
119 #endif
120
121 static int shmem_getpage(struct inode *inode, unsigned long idx,
122                          struct page **pagep, enum sgp_type sgp, int *type);
123
124 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
125 {
126         /*
127          * The above definition of ENTRIES_PER_PAGE, and the use of
128          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
129          * might be reconsidered if it ever diverges from PAGE_SIZE.
130          *
131          * Mobility flags are masked out as swap vectors cannot move
132          */
133         return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
134                                 PAGE_CACHE_SHIFT-PAGE_SHIFT);
135 }
136
137 static inline void shmem_dir_free(struct page *page)
138 {
139         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
140 }
141
142 static struct page **shmem_dir_map(struct page *page)
143 {
144         return (struct page **)kmap_atomic(page, KM_USER0);
145 }
146
147 static inline void shmem_dir_unmap(struct page **dir)
148 {
149         kunmap_atomic(dir, KM_USER0);
150 }
151
152 static swp_entry_t *shmem_swp_map(struct page *page)
153 {
154         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
155 }
156
157 static inline void shmem_swp_balance_unmap(void)
158 {
159         /*
160          * When passing a pointer to an i_direct entry, to code which
161          * also handles indirect entries and so will shmem_swp_unmap,
162          * we must arrange for the preempt count to remain in balance.
163          * What kmap_atomic of a lowmem page does depends on config
164          * and architecture, so pretend to kmap_atomic some lowmem page.
165          */
166         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
167 }
168
169 static inline void shmem_swp_unmap(swp_entry_t *entry)
170 {
171         kunmap_atomic(entry, KM_USER1);
172 }
173
174 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
175 {
176         return sb->s_fs_info;
177 }
178
179 /*
180  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
181  * for shared memory and for shared anonymous (/dev/zero) mappings
182  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
183  * consistent with the pre-accounting of private mappings ...
184  */
185 static inline int shmem_acct_size(unsigned long flags, loff_t size)
186 {
187         return (flags & VM_NORESERVE) ?
188                 0 : security_vm_enough_memory_kern(VM_ACCT(size));
189 }
190
191 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
192 {
193         if (!(flags & VM_NORESERVE))
194                 vm_unacct_memory(VM_ACCT(size));
195 }
196
197 /*
198  * ... whereas tmpfs objects are accounted incrementally as
199  * pages are allocated, in order to allow huge sparse files.
200  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
201  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
202  */
203 static inline int shmem_acct_block(unsigned long flags)
204 {
205         return (flags & VM_NORESERVE) ?
206                 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
207 }
208
209 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
210 {
211         if (flags & VM_NORESERVE)
212                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
213 }
214
215 static const struct super_operations shmem_ops;
216 static const struct address_space_operations shmem_aops;
217 static const struct file_operations shmem_file_operations;
218 static const struct inode_operations shmem_inode_operations;
219 static const struct inode_operations shmem_dir_inode_operations;
220 static const struct inode_operations shmem_special_inode_operations;
221 static const struct vm_operations_struct shmem_vm_ops;
222
223 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
224         .ra_pages       = 0,    /* No readahead */
225         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
226         .unplug_io_fn   = default_unplug_io_fn,
227 };
228
229 static LIST_HEAD(shmem_swaplist);
230 static DEFINE_MUTEX(shmem_swaplist_mutex);
231
232 static void shmem_free_blocks(struct inode *inode, long pages)
233 {
234         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
235         if (sbinfo->max_blocks) {
236                 spin_lock(&sbinfo->stat_lock);
237                 sbinfo->free_blocks += pages;
238                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
239                 spin_unlock(&sbinfo->stat_lock);
240         }
241 }
242
243 static int shmem_reserve_inode(struct super_block *sb)
244 {
245         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
246         if (sbinfo->max_inodes) {
247                 spin_lock(&sbinfo->stat_lock);
248                 if (!sbinfo->free_inodes) {
249                         spin_unlock(&sbinfo->stat_lock);
250                         return -ENOSPC;
251                 }
252                 sbinfo->free_inodes--;
253                 spin_unlock(&sbinfo->stat_lock);
254         }
255         return 0;
256 }
257
258 static void shmem_free_inode(struct super_block *sb)
259 {
260         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
261         if (sbinfo->max_inodes) {
262                 spin_lock(&sbinfo->stat_lock);
263                 sbinfo->free_inodes++;
264                 spin_unlock(&sbinfo->stat_lock);
265         }
266 }
267
268 /**
269  * shmem_recalc_inode - recalculate the size of an inode
270  * @inode: inode to recalc
271  *
272  * We have to calculate the free blocks since the mm can drop
273  * undirtied hole pages behind our back.
274  *
275  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
276  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
277  *
278  * It has to be called with the spinlock held.
279  */
280 static void shmem_recalc_inode(struct inode *inode)
281 {
282         struct shmem_inode_info *info = SHMEM_I(inode);
283         long freed;
284
285         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
286         if (freed > 0) {
287                 info->alloced -= freed;
288                 shmem_unacct_blocks(info->flags, freed);
289                 shmem_free_blocks(inode, freed);
290         }
291 }
292
293 /**
294  * shmem_swp_entry - find the swap vector position in the info structure
295  * @info:  info structure for the inode
296  * @index: index of the page to find
297  * @page:  optional page to add to the structure. Has to be preset to
298  *         all zeros
299  *
300  * If there is no space allocated yet it will return NULL when
301  * page is NULL, else it will use the page for the needed block,
302  * setting it to NULL on return to indicate that it has been used.
303  *
304  * The swap vector is organized the following way:
305  *
306  * There are SHMEM_NR_DIRECT entries directly stored in the
307  * shmem_inode_info structure. So small files do not need an addional
308  * allocation.
309  *
310  * For pages with index > SHMEM_NR_DIRECT there is the pointer
311  * i_indirect which points to a page which holds in the first half
312  * doubly indirect blocks, in the second half triple indirect blocks:
313  *
314  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
315  * following layout (for SHMEM_NR_DIRECT == 16):
316  *
317  * i_indirect -> dir --> 16-19
318  *            |      +-> 20-23
319  *            |
320  *            +-->dir2 --> 24-27
321  *            |        +-> 28-31
322  *            |        +-> 32-35
323  *            |        +-> 36-39
324  *            |
325  *            +-->dir3 --> 40-43
326  *                     +-> 44-47
327  *                     +-> 48-51
328  *                     +-> 52-55
329  */
330 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
331 {
332         unsigned long offset;
333         struct page **dir;
334         struct page *subdir;
335
336         if (index < SHMEM_NR_DIRECT) {
337                 shmem_swp_balance_unmap();
338                 return info->i_direct+index;
339         }
340         if (!info->i_indirect) {
341                 if (page) {
342                         info->i_indirect = *page;
343                         *page = NULL;
344                 }
345                 return NULL;                    /* need another page */
346         }
347
348         index -= SHMEM_NR_DIRECT;
349         offset = index % ENTRIES_PER_PAGE;
350         index /= ENTRIES_PER_PAGE;
351         dir = shmem_dir_map(info->i_indirect);
352
353         if (index >= ENTRIES_PER_PAGE/2) {
354                 index -= ENTRIES_PER_PAGE/2;
355                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
356                 index %= ENTRIES_PER_PAGE;
357                 subdir = *dir;
358                 if (!subdir) {
359                         if (page) {
360                                 *dir = *page;
361                                 *page = NULL;
362                         }
363                         shmem_dir_unmap(dir);
364                         return NULL;            /* need another page */
365                 }
366                 shmem_dir_unmap(dir);
367                 dir = shmem_dir_map(subdir);
368         }
369
370         dir += index;
371         subdir = *dir;
372         if (!subdir) {
373                 if (!page || !(subdir = *page)) {
374                         shmem_dir_unmap(dir);
375                         return NULL;            /* need a page */
376                 }
377                 *dir = subdir;
378                 *page = NULL;
379         }
380         shmem_dir_unmap(dir);
381         return shmem_swp_map(subdir) + offset;
382 }
383
384 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
385 {
386         long incdec = value? 1: -1;
387
388         entry->val = value;
389         info->swapped += incdec;
390         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
391                 struct page *page = kmap_atomic_to_page(entry);
392                 set_page_private(page, page_private(page) + incdec);
393         }
394 }
395
396 /**
397  * shmem_swp_alloc - get the position of the swap entry for the page.
398  * @info:       info structure for the inode
399  * @index:      index of the page to find
400  * @sgp:        check and recheck i_size? skip allocation?
401  *
402  * If the entry does not exist, allocate it.
403  */
404 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
405 {
406         struct inode *inode = &info->vfs_inode;
407         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
408         struct page *page = NULL;
409         swp_entry_t *entry;
410
411         if (sgp != SGP_WRITE &&
412             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
413                 return ERR_PTR(-EINVAL);
414
415         while (!(entry = shmem_swp_entry(info, index, &page))) {
416                 if (sgp == SGP_READ)
417                         return shmem_swp_map(ZERO_PAGE(0));
418                 /*
419                  * Test free_blocks against 1 not 0, since we have 1 data
420                  * page (and perhaps indirect index pages) yet to allocate:
421                  * a waste to allocate index if we cannot allocate data.
422                  */
423                 if (sbinfo->max_blocks) {
424                         spin_lock(&sbinfo->stat_lock);
425                         if (sbinfo->free_blocks <= 1) {
426                                 spin_unlock(&sbinfo->stat_lock);
427                                 return ERR_PTR(-ENOSPC);
428                         }
429                         sbinfo->free_blocks--;
430                         inode->i_blocks += BLOCKS_PER_PAGE;
431                         spin_unlock(&sbinfo->stat_lock);
432                 }
433
434                 spin_unlock(&info->lock);
435                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
436                 spin_lock(&info->lock);
437
438                 if (!page) {
439                         shmem_free_blocks(inode, 1);
440                         return ERR_PTR(-ENOMEM);
441                 }
442                 if (sgp != SGP_WRITE &&
443                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
444                         entry = ERR_PTR(-EINVAL);
445                         break;
446                 }
447                 if (info->next_index <= index)
448                         info->next_index = index + 1;
449         }
450         if (page) {
451                 /* another task gave its page, or truncated the file */
452                 shmem_free_blocks(inode, 1);
453                 shmem_dir_free(page);
454         }
455         if (info->next_index <= index && !IS_ERR(entry))
456                 info->next_index = index + 1;
457         return entry;
458 }
459
460 /**
461  * shmem_free_swp - free some swap entries in a directory
462  * @dir:        pointer to the directory
463  * @edir:       pointer after last entry of the directory
464  * @punch_lock: pointer to spinlock when needed for the holepunch case
465  */
466 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
467                                                 spinlock_t *punch_lock)
468 {
469         spinlock_t *punch_unlock = NULL;
470         swp_entry_t *ptr;
471         int freed = 0;
472
473         for (ptr = dir; ptr < edir; ptr++) {
474                 if (ptr->val) {
475                         if (unlikely(punch_lock)) {
476                                 punch_unlock = punch_lock;
477                                 punch_lock = NULL;
478                                 spin_lock(punch_unlock);
479                                 if (!ptr->val)
480                                         continue;
481                         }
482                         free_swap_and_cache(*ptr);
483                         *ptr = (swp_entry_t){0};
484                         freed++;
485                 }
486         }
487         if (punch_unlock)
488                 spin_unlock(punch_unlock);
489         return freed;
490 }
491
492 static int shmem_map_and_free_swp(struct page *subdir, int offset,
493                 int limit, struct page ***dir, spinlock_t *punch_lock)
494 {
495         swp_entry_t *ptr;
496         int freed = 0;
497
498         ptr = shmem_swp_map(subdir);
499         for (; offset < limit; offset += LATENCY_LIMIT) {
500                 int size = limit - offset;
501                 if (size > LATENCY_LIMIT)
502                         size = LATENCY_LIMIT;
503                 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
504                                                         punch_lock);
505                 if (need_resched()) {
506                         shmem_swp_unmap(ptr);
507                         if (*dir) {
508                                 shmem_dir_unmap(*dir);
509                                 *dir = NULL;
510                         }
511                         cond_resched();
512                         ptr = shmem_swp_map(subdir);
513                 }
514         }
515         shmem_swp_unmap(ptr);
516         return freed;
517 }
518
519 static void shmem_free_pages(struct list_head *next)
520 {
521         struct page *page;
522         int freed = 0;
523
524         do {
525                 page = container_of(next, struct page, lru);
526                 next = next->next;
527                 shmem_dir_free(page);
528                 freed++;
529                 if (freed >= LATENCY_LIMIT) {
530                         cond_resched();
531                         freed = 0;
532                 }
533         } while (next);
534 }
535
536 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
537 {
538         struct shmem_inode_info *info = SHMEM_I(inode);
539         unsigned long idx;
540         unsigned long size;
541         unsigned long limit;
542         unsigned long stage;
543         unsigned long diroff;
544         struct page **dir;
545         struct page *topdir;
546         struct page *middir;
547         struct page *subdir;
548         swp_entry_t *ptr;
549         LIST_HEAD(pages_to_free);
550         long nr_pages_to_free = 0;
551         long nr_swaps_freed = 0;
552         int offset;
553         int freed;
554         int punch_hole;
555         spinlock_t *needs_lock;
556         spinlock_t *punch_lock;
557         unsigned long upper_limit;
558
559         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
560         idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
561         if (idx >= info->next_index)
562                 return;
563
564         spin_lock(&info->lock);
565         info->flags |= SHMEM_TRUNCATE;
566         if (likely(end == (loff_t) -1)) {
567                 limit = info->next_index;
568                 upper_limit = SHMEM_MAX_INDEX;
569                 info->next_index = idx;
570                 needs_lock = NULL;
571                 punch_hole = 0;
572         } else {
573                 if (end + 1 >= inode->i_size) { /* we may free a little more */
574                         limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
575                                                         PAGE_CACHE_SHIFT;
576                         upper_limit = SHMEM_MAX_INDEX;
577                 } else {
578                         limit = (end + 1) >> PAGE_CACHE_SHIFT;
579                         upper_limit = limit;
580                 }
581                 needs_lock = &info->lock;
582                 punch_hole = 1;
583         }
584
585         topdir = info->i_indirect;
586         if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
587                 info->i_indirect = NULL;
588                 nr_pages_to_free++;
589                 list_add(&topdir->lru, &pages_to_free);
590         }
591         spin_unlock(&info->lock);
592
593         if (info->swapped && idx < SHMEM_NR_DIRECT) {
594                 ptr = info->i_direct;
595                 size = limit;
596                 if (size > SHMEM_NR_DIRECT)
597                         size = SHMEM_NR_DIRECT;
598                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
599         }
600
601         /*
602          * If there are no indirect blocks or we are punching a hole
603          * below indirect blocks, nothing to be done.
604          */
605         if (!topdir || limit <= SHMEM_NR_DIRECT)
606                 goto done2;
607
608         /*
609          * The truncation case has already dropped info->lock, and we're safe
610          * because i_size and next_index have already been lowered, preventing
611          * access beyond.  But in the punch_hole case, we still need to take
612          * the lock when updating the swap directory, because there might be
613          * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
614          * shmem_writepage.  However, whenever we find we can remove a whole
615          * directory page (not at the misaligned start or end of the range),
616          * we first NULLify its pointer in the level above, and then have no
617          * need to take the lock when updating its contents: needs_lock and
618          * punch_lock (either pointing to info->lock or NULL) manage this.
619          */
620
621         upper_limit -= SHMEM_NR_DIRECT;
622         limit -= SHMEM_NR_DIRECT;
623         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
624         offset = idx % ENTRIES_PER_PAGE;
625         idx -= offset;
626
627         dir = shmem_dir_map(topdir);
628         stage = ENTRIES_PER_PAGEPAGE/2;
629         if (idx < ENTRIES_PER_PAGEPAGE/2) {
630                 middir = topdir;
631                 diroff = idx/ENTRIES_PER_PAGE;
632         } else {
633                 dir += ENTRIES_PER_PAGE/2;
634                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
635                 while (stage <= idx)
636                         stage += ENTRIES_PER_PAGEPAGE;
637                 middir = *dir;
638                 if (*dir) {
639                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
640                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
641                         if (!diroff && !offset && upper_limit >= stage) {
642                                 if (needs_lock) {
643                                         spin_lock(needs_lock);
644                                         *dir = NULL;
645                                         spin_unlock(needs_lock);
646                                         needs_lock = NULL;
647                                 } else
648                                         *dir = NULL;
649                                 nr_pages_to_free++;
650                                 list_add(&middir->lru, &pages_to_free);
651                         }
652                         shmem_dir_unmap(dir);
653                         dir = shmem_dir_map(middir);
654                 } else {
655                         diroff = 0;
656                         offset = 0;
657                         idx = stage;
658                 }
659         }
660
661         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
662                 if (unlikely(idx == stage)) {
663                         shmem_dir_unmap(dir);
664                         dir = shmem_dir_map(topdir) +
665                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
666                         while (!*dir) {
667                                 dir++;
668                                 idx += ENTRIES_PER_PAGEPAGE;
669                                 if (idx >= limit)
670                                         goto done1;
671                         }
672                         stage = idx + ENTRIES_PER_PAGEPAGE;
673                         middir = *dir;
674                         if (punch_hole)
675                                 needs_lock = &info->lock;
676                         if (upper_limit >= stage) {
677                                 if (needs_lock) {
678                                         spin_lock(needs_lock);
679                                         *dir = NULL;
680                                         spin_unlock(needs_lock);
681                                         needs_lock = NULL;
682                                 } else
683                                         *dir = NULL;
684                                 nr_pages_to_free++;
685                                 list_add(&middir->lru, &pages_to_free);
686                         }
687                         shmem_dir_unmap(dir);
688                         cond_resched();
689                         dir = shmem_dir_map(middir);
690                         diroff = 0;
691                 }
692                 punch_lock = needs_lock;
693                 subdir = dir[diroff];
694                 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
695                         if (needs_lock) {
696                                 spin_lock(needs_lock);
697                                 dir[diroff] = NULL;
698                                 spin_unlock(needs_lock);
699                                 punch_lock = NULL;
700                         } else
701                                 dir[diroff] = NULL;
702                         nr_pages_to_free++;
703                         list_add(&subdir->lru, &pages_to_free);
704                 }
705                 if (subdir && page_private(subdir) /* has swap entries */) {
706                         size = limit - idx;
707                         if (size > ENTRIES_PER_PAGE)
708                                 size = ENTRIES_PER_PAGE;
709                         freed = shmem_map_and_free_swp(subdir,
710                                         offset, size, &dir, punch_lock);
711                         if (!dir)
712                                 dir = shmem_dir_map(middir);
713                         nr_swaps_freed += freed;
714                         if (offset || punch_lock) {
715                                 spin_lock(&info->lock);
716                                 set_page_private(subdir,
717                                         page_private(subdir) - freed);
718                                 spin_unlock(&info->lock);
719                         } else
720                                 BUG_ON(page_private(subdir) != freed);
721                 }
722                 offset = 0;
723         }
724 done1:
725         shmem_dir_unmap(dir);
726 done2:
727         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
728                 /*
729                  * Call truncate_inode_pages again: racing shmem_unuse_inode
730                  * may have swizzled a page in from swap since
731                  * truncate_pagecache or generic_delete_inode did it, before we
732                  * lowered next_index.  Also, though shmem_getpage checks
733                  * i_size before adding to cache, no recheck after: so fix the
734                  * narrow window there too.
735                  *
736                  * Recalling truncate_inode_pages_range and unmap_mapping_range
737                  * every time for punch_hole (which never got a chance to clear
738                  * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
739                  * yet hardly ever necessary: try to optimize them out later.
740                  */
741                 truncate_inode_pages_range(inode->i_mapping, start, end);
742                 if (punch_hole)
743                         unmap_mapping_range(inode->i_mapping, start,
744                                                         end - start, 1);
745         }
746
747         spin_lock(&info->lock);
748         info->flags &= ~SHMEM_TRUNCATE;
749         info->swapped -= nr_swaps_freed;
750         if (nr_pages_to_free)
751                 shmem_free_blocks(inode, nr_pages_to_free);
752         shmem_recalc_inode(inode);
753         spin_unlock(&info->lock);
754
755         /*
756          * Empty swap vector directory pages to be freed?
757          */
758         if (!list_empty(&pages_to_free)) {
759                 pages_to_free.prev->next = NULL;
760                 shmem_free_pages(pages_to_free.next);
761         }
762 }
763
764 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
765 {
766         struct inode *inode = dentry->d_inode;
767         loff_t newsize = attr->ia_size;
768         int error;
769
770         error = inode_change_ok(inode, attr);
771         if (error)
772                 return error;
773
774         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)
775                                         && newsize != inode->i_size) {
776                 struct page *page = NULL;
777
778                 if (newsize < inode->i_size) {
779                         /*
780                          * If truncating down to a partial page, then
781                          * if that page is already allocated, hold it
782                          * in memory until the truncation is over, so
783                          * truncate_partial_page cannnot miss it were
784                          * it assigned to swap.
785                          */
786                         if (newsize & (PAGE_CACHE_SIZE-1)) {
787                                 (void) shmem_getpage(inode,
788                                         newsize >> PAGE_CACHE_SHIFT,
789                                                 &page, SGP_READ, NULL);
790                                 if (page)
791                                         unlock_page(page);
792                         }
793                         /*
794                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
795                          * detect if any pages might have been added to cache
796                          * after truncate_inode_pages.  But we needn't bother
797                          * if it's being fully truncated to zero-length: the
798                          * nrpages check is efficient enough in that case.
799                          */
800                         if (newsize) {
801                                 struct shmem_inode_info *info = SHMEM_I(inode);
802                                 spin_lock(&info->lock);
803                                 info->flags &= ~SHMEM_PAGEIN;
804                                 spin_unlock(&info->lock);
805                         }
806                 }
807
808                 error = simple_setsize(inode, newsize);
809                 if (page)
810                         page_cache_release(page);
811                 if (error)
812                         return error;
813                 shmem_truncate_range(inode, newsize, (loff_t)-1);
814         }
815
816         setattr_copy(inode, attr);
817 #ifdef CONFIG_TMPFS_POSIX_ACL
818         if (attr->ia_valid & ATTR_MODE)
819                 error = generic_acl_chmod(inode);
820 #endif
821         return error;
822 }
823
824 static void shmem_delete_inode(struct inode *inode)
825 {
826         struct shmem_inode_info *info = SHMEM_I(inode);
827
828         if (inode->i_mapping->a_ops == &shmem_aops) {
829                 truncate_inode_pages(inode->i_mapping, 0);
830                 shmem_unacct_size(info->flags, inode->i_size);
831                 inode->i_size = 0;
832                 shmem_truncate_range(inode, 0, (loff_t)-1);
833                 if (!list_empty(&info->swaplist)) {
834                         mutex_lock(&shmem_swaplist_mutex);
835                         list_del_init(&info->swaplist);
836                         mutex_unlock(&shmem_swaplist_mutex);
837                 }
838         }
839         BUG_ON(inode->i_blocks);
840         shmem_free_inode(inode->i_sb);
841         clear_inode(inode);
842 }
843
844 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
845 {
846         swp_entry_t *ptr;
847
848         for (ptr = dir; ptr < edir; ptr++) {
849                 if (ptr->val == entry.val)
850                         return ptr - dir;
851         }
852         return -1;
853 }
854
855 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
856 {
857         struct inode *inode;
858         unsigned long idx;
859         unsigned long size;
860         unsigned long limit;
861         unsigned long stage;
862         struct page **dir;
863         struct page *subdir;
864         swp_entry_t *ptr;
865         int offset;
866         int error;
867
868         idx = 0;
869         ptr = info->i_direct;
870         spin_lock(&info->lock);
871         if (!info->swapped) {
872                 list_del_init(&info->swaplist);
873                 goto lost2;
874         }
875         limit = info->next_index;
876         size = limit;
877         if (size > SHMEM_NR_DIRECT)
878                 size = SHMEM_NR_DIRECT;
879         offset = shmem_find_swp(entry, ptr, ptr+size);
880         if (offset >= 0)
881                 goto found;
882         if (!info->i_indirect)
883                 goto lost2;
884
885         dir = shmem_dir_map(info->i_indirect);
886         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
887
888         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
889                 if (unlikely(idx == stage)) {
890                         shmem_dir_unmap(dir-1);
891                         if (cond_resched_lock(&info->lock)) {
892                                 /* check it has not been truncated */
893                                 if (limit > info->next_index) {
894                                         limit = info->next_index;
895                                         if (idx >= limit)
896                                                 goto lost2;
897                                 }
898                         }
899                         dir = shmem_dir_map(info->i_indirect) +
900                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
901                         while (!*dir) {
902                                 dir++;
903                                 idx += ENTRIES_PER_PAGEPAGE;
904                                 if (idx >= limit)
905                                         goto lost1;
906                         }
907                         stage = idx + ENTRIES_PER_PAGEPAGE;
908                         subdir = *dir;
909                         shmem_dir_unmap(dir);
910                         dir = shmem_dir_map(subdir);
911                 }
912                 subdir = *dir;
913                 if (subdir && page_private(subdir)) {
914                         ptr = shmem_swp_map(subdir);
915                         size = limit - idx;
916                         if (size > ENTRIES_PER_PAGE)
917                                 size = ENTRIES_PER_PAGE;
918                         offset = shmem_find_swp(entry, ptr, ptr+size);
919                         shmem_swp_unmap(ptr);
920                         if (offset >= 0) {
921                                 shmem_dir_unmap(dir);
922                                 goto found;
923                         }
924                 }
925         }
926 lost1:
927         shmem_dir_unmap(dir-1);
928 lost2:
929         spin_unlock(&info->lock);
930         return 0;
931 found:
932         idx += offset;
933         inode = igrab(&info->vfs_inode);
934         spin_unlock(&info->lock);
935
936         /*
937          * Move _head_ to start search for next from here.
938          * But be careful: shmem_delete_inode checks list_empty without taking
939          * mutex, and there's an instant in list_move_tail when info->swaplist
940          * would appear empty, if it were the only one on shmem_swaplist.  We
941          * could avoid doing it if inode NULL; or use this minor optimization.
942          */
943         if (shmem_swaplist.next != &info->swaplist)
944                 list_move_tail(&shmem_swaplist, &info->swaplist);
945         mutex_unlock(&shmem_swaplist_mutex);
946
947         error = 1;
948         if (!inode)
949                 goto out;
950         /*
951          * Charge page using GFP_KERNEL while we can wait.
952          * Charged back to the user(not to caller) when swap account is used.
953          * add_to_page_cache() will be called with GFP_NOWAIT.
954          */
955         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
956         if (error)
957                 goto out;
958         error = radix_tree_preload(GFP_KERNEL);
959         if (error) {
960                 mem_cgroup_uncharge_cache_page(page);
961                 goto out;
962         }
963         error = 1;
964
965         spin_lock(&info->lock);
966         ptr = shmem_swp_entry(info, idx, NULL);
967         if (ptr && ptr->val == entry.val) {
968                 error = add_to_page_cache_locked(page, inode->i_mapping,
969                                                 idx, GFP_NOWAIT);
970                 /* does mem_cgroup_uncharge_cache_page on error */
971         } else  /* we must compensate for our precharge above */
972                 mem_cgroup_uncharge_cache_page(page);
973
974         if (error == -EEXIST) {
975                 struct page *filepage = find_get_page(inode->i_mapping, idx);
976                 error = 1;
977                 if (filepage) {
978                         /*
979                          * There might be a more uptodate page coming down
980                          * from a stacked writepage: forget our swappage if so.
981                          */
982                         if (PageUptodate(filepage))
983                                 error = 0;
984                         page_cache_release(filepage);
985                 }
986         }
987         if (!error) {
988                 delete_from_swap_cache(page);
989                 set_page_dirty(page);
990                 info->flags |= SHMEM_PAGEIN;
991                 shmem_swp_set(info, ptr, 0);
992                 swap_free(entry);
993                 error = 1;      /* not an error, but entry was found */
994         }
995         if (ptr)
996                 shmem_swp_unmap(ptr);
997         spin_unlock(&info->lock);
998         radix_tree_preload_end();
999 out:
1000         unlock_page(page);
1001         page_cache_release(page);
1002         iput(inode);            /* allows for NULL */
1003         return error;
1004 }
1005
1006 /*
1007  * shmem_unuse() search for an eventually swapped out shmem page.
1008  */
1009 int shmem_unuse(swp_entry_t entry, struct page *page)
1010 {
1011         struct list_head *p, *next;
1012         struct shmem_inode_info *info;
1013         int found = 0;
1014
1015         mutex_lock(&shmem_swaplist_mutex);
1016         list_for_each_safe(p, next, &shmem_swaplist) {
1017                 info = list_entry(p, struct shmem_inode_info, swaplist);
1018                 found = shmem_unuse_inode(info, entry, page);
1019                 cond_resched();
1020                 if (found)
1021                         goto out;
1022         }
1023         mutex_unlock(&shmem_swaplist_mutex);
1024         /*
1025          * Can some race bring us here?  We've been holding page lock,
1026          * so I think not; but would rather try again later than BUG()
1027          */
1028         unlock_page(page);
1029         page_cache_release(page);
1030 out:
1031         return (found < 0) ? found : 0;
1032 }
1033
1034 /*
1035  * Move the page from the page cache to the swap cache.
1036  */
1037 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1038 {
1039         struct shmem_inode_info *info;
1040         swp_entry_t *entry, swap;
1041         struct address_space *mapping;
1042         unsigned long index;
1043         struct inode *inode;
1044
1045         BUG_ON(!PageLocked(page));
1046         mapping = page->mapping;
1047         index = page->index;
1048         inode = mapping->host;
1049         info = SHMEM_I(inode);
1050         if (info->flags & VM_LOCKED)
1051                 goto redirty;
1052         if (!total_swap_pages)
1053                 goto redirty;
1054
1055         /*
1056          * shmem_backing_dev_info's capabilities prevent regular writeback or
1057          * sync from ever calling shmem_writepage; but a stacking filesystem
1058          * may use the ->writepage of its underlying filesystem, in which case
1059          * tmpfs should write out to swap only in response to memory pressure,
1060          * and not for the writeback threads or sync.  However, in those cases,
1061          * we do still want to check if there's a redundant swappage to be
1062          * discarded.
1063          */
1064         if (wbc->for_reclaim)
1065                 swap = get_swap_page();
1066         else
1067                 swap.val = 0;
1068
1069         spin_lock(&info->lock);
1070         if (index >= info->next_index) {
1071                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1072                 goto unlock;
1073         }
1074         entry = shmem_swp_entry(info, index, NULL);
1075         if (entry->val) {
1076                 /*
1077                  * The more uptodate page coming down from a stacked
1078                  * writepage should replace our old swappage.
1079                  */
1080                 free_swap_and_cache(*entry);
1081                 shmem_swp_set(info, entry, 0);
1082         }
1083         shmem_recalc_inode(inode);
1084
1085         if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1086                 remove_from_page_cache(page);
1087                 shmem_swp_set(info, entry, swap.val);
1088                 shmem_swp_unmap(entry);
1089                 if (list_empty(&info->swaplist))
1090                         inode = igrab(inode);
1091                 else
1092                         inode = NULL;
1093                 spin_unlock(&info->lock);
1094                 swap_shmem_alloc(swap);
1095                 BUG_ON(page_mapped(page));
1096                 page_cache_release(page);       /* pagecache ref */
1097                 swap_writepage(page, wbc);
1098                 if (inode) {
1099                         mutex_lock(&shmem_swaplist_mutex);
1100                         /* move instead of add in case we're racing */
1101                         list_move_tail(&info->swaplist, &shmem_swaplist);
1102                         mutex_unlock(&shmem_swaplist_mutex);
1103                         iput(inode);
1104                 }
1105                 return 0;
1106         }
1107
1108         shmem_swp_unmap(entry);
1109 unlock:
1110         spin_unlock(&info->lock);
1111         /*
1112          * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1113          * clear SWAP_HAS_CACHE flag.
1114          */
1115         swapcache_free(swap, NULL);
1116 redirty:
1117         set_page_dirty(page);
1118         if (wbc->for_reclaim)
1119                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
1120         unlock_page(page);
1121         return 0;
1122 }
1123
1124 #ifdef CONFIG_NUMA
1125 #ifdef CONFIG_TMPFS
1126 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1127 {
1128         char buffer[64];
1129
1130         if (!mpol || mpol->mode == MPOL_DEFAULT)
1131                 return;         /* show nothing */
1132
1133         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1134
1135         seq_printf(seq, ",mpol=%s", buffer);
1136 }
1137
1138 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1139 {
1140         struct mempolicy *mpol = NULL;
1141         if (sbinfo->mpol) {
1142                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
1143                 mpol = sbinfo->mpol;
1144                 mpol_get(mpol);
1145                 spin_unlock(&sbinfo->stat_lock);
1146         }
1147         return mpol;
1148 }
1149 #endif /* CONFIG_TMPFS */
1150
1151 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1152                         struct shmem_inode_info *info, unsigned long idx)
1153 {
1154         struct mempolicy mpol, *spol;
1155         struct vm_area_struct pvma;
1156         struct page *page;
1157
1158         spol = mpol_cond_copy(&mpol,
1159                                 mpol_shared_policy_lookup(&info->policy, idx));
1160
1161         /* Create a pseudo vma that just contains the policy */
1162         pvma.vm_start = 0;
1163         pvma.vm_pgoff = idx;
1164         pvma.vm_ops = NULL;
1165         pvma.vm_policy = spol;
1166         page = swapin_readahead(entry, gfp, &pvma, 0);
1167         return page;
1168 }
1169
1170 static struct page *shmem_alloc_page(gfp_t gfp,
1171                         struct shmem_inode_info *info, unsigned long idx)
1172 {
1173         struct vm_area_struct pvma;
1174
1175         /* Create a pseudo vma that just contains the policy */
1176         pvma.vm_start = 0;
1177         pvma.vm_pgoff = idx;
1178         pvma.vm_ops = NULL;
1179         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1180
1181         /*
1182          * alloc_page_vma() will drop the shared policy reference
1183          */
1184         return alloc_page_vma(gfp, &pvma, 0);
1185 }
1186 #else /* !CONFIG_NUMA */
1187 #ifdef CONFIG_TMPFS
1188 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1189 {
1190 }
1191 #endif /* CONFIG_TMPFS */
1192
1193 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1194                         struct shmem_inode_info *info, unsigned long idx)
1195 {
1196         return swapin_readahead(entry, gfp, NULL, 0);
1197 }
1198
1199 static inline struct page *shmem_alloc_page(gfp_t gfp,
1200                         struct shmem_inode_info *info, unsigned long idx)
1201 {
1202         return alloc_page(gfp);
1203 }
1204 #endif /* CONFIG_NUMA */
1205
1206 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1207 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1208 {
1209         return NULL;
1210 }
1211 #endif
1212
1213 /*
1214  * shmem_getpage - either get the page from swap or allocate a new one
1215  *
1216  * If we allocate a new one we do not mark it dirty. That's up to the
1217  * vm. If we swap it in we mark it dirty since we also free the swap
1218  * entry since a page cannot live in both the swap and page cache
1219  */
1220 static int shmem_getpage(struct inode *inode, unsigned long idx,
1221                         struct page **pagep, enum sgp_type sgp, int *type)
1222 {
1223         struct address_space *mapping = inode->i_mapping;
1224         struct shmem_inode_info *info = SHMEM_I(inode);
1225         struct shmem_sb_info *sbinfo;
1226         struct page *filepage = *pagep;
1227         struct page *swappage;
1228         swp_entry_t *entry;
1229         swp_entry_t swap;
1230         gfp_t gfp;
1231         int error;
1232
1233         if (idx >= SHMEM_MAX_INDEX)
1234                 return -EFBIG;
1235
1236         if (type)
1237                 *type = 0;
1238
1239         /*
1240          * Normally, filepage is NULL on entry, and either found
1241          * uptodate immediately, or allocated and zeroed, or read
1242          * in under swappage, which is then assigned to filepage.
1243          * But shmem_readpage (required for splice) passes in a locked
1244          * filepage, which may be found not uptodate by other callers
1245          * too, and may need to be copied from the swappage read in.
1246          */
1247 repeat:
1248         if (!filepage)
1249                 filepage = find_lock_page(mapping, idx);
1250         if (filepage && PageUptodate(filepage))
1251                 goto done;
1252         error = 0;
1253         gfp = mapping_gfp_mask(mapping);
1254         if (!filepage) {
1255                 /*
1256                  * Try to preload while we can wait, to not make a habit of
1257                  * draining atomic reserves; but don't latch on to this cpu.
1258                  */
1259                 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1260                 if (error)
1261                         goto failed;
1262                 radix_tree_preload_end();
1263         }
1264
1265         spin_lock(&info->lock);
1266         shmem_recalc_inode(inode);
1267         entry = shmem_swp_alloc(info, idx, sgp);
1268         if (IS_ERR(entry)) {
1269                 spin_unlock(&info->lock);
1270                 error = PTR_ERR(entry);
1271                 goto failed;
1272         }
1273         swap = *entry;
1274
1275         if (swap.val) {
1276                 /* Look it up and read it in.. */
1277                 swappage = lookup_swap_cache(swap);
1278                 if (!swappage) {
1279                         shmem_swp_unmap(entry);
1280                         /* here we actually do the io */
1281                         if (type && !(*type & VM_FAULT_MAJOR)) {
1282                                 __count_vm_event(PGMAJFAULT);
1283                                 *type |= VM_FAULT_MAJOR;
1284                         }
1285                         spin_unlock(&info->lock);
1286                         swappage = shmem_swapin(swap, gfp, info, idx);
1287                         if (!swappage) {
1288                                 spin_lock(&info->lock);
1289                                 entry = shmem_swp_alloc(info, idx, sgp);
1290                                 if (IS_ERR(entry))
1291                                         error = PTR_ERR(entry);
1292                                 else {
1293                                         if (entry->val == swap.val)
1294                                                 error = -ENOMEM;
1295                                         shmem_swp_unmap(entry);
1296                                 }
1297                                 spin_unlock(&info->lock);
1298                                 if (error)
1299                                         goto failed;
1300                                 goto repeat;
1301                         }
1302                         wait_on_page_locked(swappage);
1303                         page_cache_release(swappage);
1304                         goto repeat;
1305                 }
1306
1307                 /* We have to do this with page locked to prevent races */
1308                 if (!trylock_page(swappage)) {
1309                         shmem_swp_unmap(entry);
1310                         spin_unlock(&info->lock);
1311                         wait_on_page_locked(swappage);
1312                         page_cache_release(swappage);
1313                         goto repeat;
1314                 }
1315                 if (PageWriteback(swappage)) {
1316                         shmem_swp_unmap(entry);
1317                         spin_unlock(&info->lock);
1318                         wait_on_page_writeback(swappage);
1319                         unlock_page(swappage);
1320                         page_cache_release(swappage);
1321                         goto repeat;
1322                 }
1323                 if (!PageUptodate(swappage)) {
1324                         shmem_swp_unmap(entry);
1325                         spin_unlock(&info->lock);
1326                         unlock_page(swappage);
1327                         page_cache_release(swappage);
1328                         error = -EIO;
1329                         goto failed;
1330                 }
1331
1332                 if (filepage) {
1333                         shmem_swp_set(info, entry, 0);
1334                         shmem_swp_unmap(entry);
1335                         delete_from_swap_cache(swappage);
1336                         spin_unlock(&info->lock);
1337                         copy_highpage(filepage, swappage);
1338                         unlock_page(swappage);
1339                         page_cache_release(swappage);
1340                         flush_dcache_page(filepage);
1341                         SetPageUptodate(filepage);
1342                         set_page_dirty(filepage);
1343                         swap_free(swap);
1344                 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1345                                         idx, GFP_NOWAIT))) {
1346                         info->flags |= SHMEM_PAGEIN;
1347                         shmem_swp_set(info, entry, 0);
1348                         shmem_swp_unmap(entry);
1349                         delete_from_swap_cache(swappage);
1350                         spin_unlock(&info->lock);
1351                         filepage = swappage;
1352                         set_page_dirty(filepage);
1353                         swap_free(swap);
1354                 } else {
1355                         shmem_swp_unmap(entry);
1356                         spin_unlock(&info->lock);
1357                         if (error == -ENOMEM) {
1358                                 /*
1359                                  * reclaim from proper memory cgroup and
1360                                  * call memcg's OOM if needed.
1361                                  */
1362                                 error = mem_cgroup_shmem_charge_fallback(
1363                                                                 swappage,
1364                                                                 current->mm,
1365                                                                 gfp);
1366                                 if (error) {
1367                                         unlock_page(swappage);
1368                                         page_cache_release(swappage);
1369                                         goto failed;
1370                                 }
1371                         }
1372                         unlock_page(swappage);
1373                         page_cache_release(swappage);
1374                         goto repeat;
1375                 }
1376         } else if (sgp == SGP_READ && !filepage) {
1377                 shmem_swp_unmap(entry);
1378                 filepage = find_get_page(mapping, idx);
1379                 if (filepage &&
1380                     (!PageUptodate(filepage) || !trylock_page(filepage))) {
1381                         spin_unlock(&info->lock);
1382                         wait_on_page_locked(filepage);
1383                         page_cache_release(filepage);
1384                         filepage = NULL;
1385                         goto repeat;
1386                 }
1387                 spin_unlock(&info->lock);
1388         } else {
1389                 shmem_swp_unmap(entry);
1390                 sbinfo = SHMEM_SB(inode->i_sb);
1391                 if (sbinfo->max_blocks) {
1392                         spin_lock(&sbinfo->stat_lock);
1393                         if (sbinfo->free_blocks == 0 ||
1394                             shmem_acct_block(info->flags)) {
1395                                 spin_unlock(&sbinfo->stat_lock);
1396                                 spin_unlock(&info->lock);
1397                                 error = -ENOSPC;
1398                                 goto failed;
1399                         }
1400                         sbinfo->free_blocks--;
1401                         inode->i_blocks += BLOCKS_PER_PAGE;
1402                         spin_unlock(&sbinfo->stat_lock);
1403                 } else if (shmem_acct_block(info->flags)) {
1404                         spin_unlock(&info->lock);
1405                         error = -ENOSPC;
1406                         goto failed;
1407                 }
1408
1409                 if (!filepage) {
1410                         int ret;
1411
1412                         spin_unlock(&info->lock);
1413                         filepage = shmem_alloc_page(gfp, info, idx);
1414                         if (!filepage) {
1415                                 shmem_unacct_blocks(info->flags, 1);
1416                                 shmem_free_blocks(inode, 1);
1417                                 error = -ENOMEM;
1418                                 goto failed;
1419                         }
1420                         SetPageSwapBacked(filepage);
1421
1422                         /* Precharge page while we can wait, compensate after */
1423                         error = mem_cgroup_cache_charge(filepage, current->mm,
1424                                         GFP_KERNEL);
1425                         if (error) {
1426                                 page_cache_release(filepage);
1427                                 shmem_unacct_blocks(info->flags, 1);
1428                                 shmem_free_blocks(inode, 1);
1429                                 filepage = NULL;
1430                                 goto failed;
1431                         }
1432
1433                         spin_lock(&info->lock);
1434                         entry = shmem_swp_alloc(info, idx, sgp);
1435                         if (IS_ERR(entry))
1436                                 error = PTR_ERR(entry);
1437                         else {
1438                                 swap = *entry;
1439                                 shmem_swp_unmap(entry);
1440                         }
1441                         ret = error || swap.val;
1442                         if (ret)
1443                                 mem_cgroup_uncharge_cache_page(filepage);
1444                         else
1445                                 ret = add_to_page_cache_lru(filepage, mapping,
1446                                                 idx, GFP_NOWAIT);
1447                         /*
1448                          * At add_to_page_cache_lru() failure, uncharge will
1449                          * be done automatically.
1450                          */
1451                         if (ret) {
1452                                 spin_unlock(&info->lock);
1453                                 page_cache_release(filepage);
1454                                 shmem_unacct_blocks(info->flags, 1);
1455                                 shmem_free_blocks(inode, 1);
1456                                 filepage = NULL;
1457                                 if (error)
1458                                         goto failed;
1459                                 goto repeat;
1460                         }
1461                         info->flags |= SHMEM_PAGEIN;
1462                 }
1463
1464                 info->alloced++;
1465                 spin_unlock(&info->lock);
1466                 clear_highpage(filepage);
1467                 flush_dcache_page(filepage);
1468                 SetPageUptodate(filepage);
1469                 if (sgp == SGP_DIRTY)
1470                         set_page_dirty(filepage);
1471         }
1472 done:
1473         *pagep = filepage;
1474         return 0;
1475
1476 failed:
1477         if (*pagep != filepage) {
1478                 unlock_page(filepage);
1479                 page_cache_release(filepage);
1480         }
1481         return error;
1482 }
1483
1484 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1485 {
1486         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1487         int error;
1488         int ret;
1489
1490         if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1491                 return VM_FAULT_SIGBUS;
1492
1493         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1494         if (error)
1495                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1496
1497         return ret | VM_FAULT_LOCKED;
1498 }
1499
1500 #ifdef CONFIG_NUMA
1501 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1502 {
1503         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1504         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1505 }
1506
1507 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1508                                           unsigned long addr)
1509 {
1510         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1511         unsigned long idx;
1512
1513         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1514         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1515 }
1516 #endif
1517
1518 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1519 {
1520         struct inode *inode = file->f_path.dentry->d_inode;
1521         struct shmem_inode_info *info = SHMEM_I(inode);
1522         int retval = -ENOMEM;
1523
1524         spin_lock(&info->lock);
1525         if (lock && !(info->flags & VM_LOCKED)) {
1526                 if (!user_shm_lock(inode->i_size, user))
1527                         goto out_nomem;
1528                 info->flags |= VM_LOCKED;
1529                 mapping_set_unevictable(file->f_mapping);
1530         }
1531         if (!lock && (info->flags & VM_LOCKED) && user) {
1532                 user_shm_unlock(inode->i_size, user);
1533                 info->flags &= ~VM_LOCKED;
1534                 mapping_clear_unevictable(file->f_mapping);
1535                 scan_mapping_unevictable_pages(file->f_mapping);
1536         }
1537         retval = 0;
1538
1539 out_nomem:
1540         spin_unlock(&info->lock);
1541         return retval;
1542 }
1543
1544 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1545 {
1546         file_accessed(file);
1547         vma->vm_ops = &shmem_vm_ops;
1548         vma->vm_flags |= VM_CAN_NONLINEAR;
1549         return 0;
1550 }
1551
1552 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1553                                      int mode, dev_t dev, unsigned long flags)
1554 {
1555         struct inode *inode;
1556         struct shmem_inode_info *info;
1557         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1558
1559         if (shmem_reserve_inode(sb))
1560                 return NULL;
1561
1562         inode = new_inode(sb);
1563         if (inode) {
1564                 inode_init_owner(inode, dir, mode);
1565                 inode->i_blocks = 0;
1566                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1567                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1568                 inode->i_generation = get_seconds();
1569                 info = SHMEM_I(inode);
1570                 memset(info, 0, (char *)inode - (char *)info);
1571                 spin_lock_init(&info->lock);
1572                 info->flags = flags & VM_NORESERVE;
1573                 INIT_LIST_HEAD(&info->swaplist);
1574                 cache_no_acl(inode);
1575
1576                 switch (mode & S_IFMT) {
1577                 default:
1578                         inode->i_op = &shmem_special_inode_operations;
1579                         init_special_inode(inode, mode, dev);
1580                         break;
1581                 case S_IFREG:
1582                         inode->i_mapping->a_ops = &shmem_aops;
1583                         inode->i_op = &shmem_inode_operations;
1584                         inode->i_fop = &shmem_file_operations;
1585                         mpol_shared_policy_init(&info->policy,
1586                                                  shmem_get_sbmpol(sbinfo));
1587                         break;
1588                 case S_IFDIR:
1589                         inc_nlink(inode);
1590                         /* Some things misbehave if size == 0 on a directory */
1591                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1592                         inode->i_op = &shmem_dir_inode_operations;
1593                         inode->i_fop = &simple_dir_operations;
1594                         break;
1595                 case S_IFLNK:
1596                         /*
1597                          * Must not load anything in the rbtree,
1598                          * mpol_free_shared_policy will not be called.
1599                          */
1600                         mpol_shared_policy_init(&info->policy, NULL);
1601                         break;
1602                 }
1603         } else
1604                 shmem_free_inode(sb);
1605         return inode;
1606 }
1607
1608 #ifdef CONFIG_TMPFS
1609 static const struct inode_operations shmem_symlink_inode_operations;
1610 static const struct inode_operations shmem_symlink_inline_operations;
1611
1612 /*
1613  * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1614  * but providing them allows a tmpfs file to be used for splice, sendfile, and
1615  * below the loop driver, in the generic fashion that many filesystems support.
1616  */
1617 static int shmem_readpage(struct file *file, struct page *page)
1618 {
1619         struct inode *inode = page->mapping->host;
1620         int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1621         unlock_page(page);
1622         return error;
1623 }
1624
1625 static int
1626 shmem_write_begin(struct file *file, struct address_space *mapping,
1627                         loff_t pos, unsigned len, unsigned flags,
1628                         struct page **pagep, void **fsdata)
1629 {
1630         struct inode *inode = mapping->host;
1631         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1632         *pagep = NULL;
1633         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1634 }
1635
1636 static int
1637 shmem_write_end(struct file *file, struct address_space *mapping,
1638                         loff_t pos, unsigned len, unsigned copied,
1639                         struct page *page, void *fsdata)
1640 {
1641         struct inode *inode = mapping->host;
1642
1643         if (pos + copied > inode->i_size)
1644                 i_size_write(inode, pos + copied);
1645
1646         set_page_dirty(page);
1647         unlock_page(page);
1648         page_cache_release(page);
1649
1650         return copied;
1651 }
1652
1653 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1654 {
1655         struct inode *inode = filp->f_path.dentry->d_inode;
1656         struct address_space *mapping = inode->i_mapping;
1657         unsigned long index, offset;
1658         enum sgp_type sgp = SGP_READ;
1659
1660         /*
1661          * Might this read be for a stacking filesystem?  Then when reading
1662          * holes of a sparse file, we actually need to allocate those pages,
1663          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1664          */
1665         if (segment_eq(get_fs(), KERNEL_DS))
1666                 sgp = SGP_DIRTY;
1667
1668         index = *ppos >> PAGE_CACHE_SHIFT;
1669         offset = *ppos & ~PAGE_CACHE_MASK;
1670
1671         for (;;) {
1672                 struct page *page = NULL;
1673                 unsigned long end_index, nr, ret;
1674                 loff_t i_size = i_size_read(inode);
1675
1676                 end_index = i_size >> PAGE_CACHE_SHIFT;
1677                 if (index > end_index)
1678                         break;
1679                 if (index == end_index) {
1680                         nr = i_size & ~PAGE_CACHE_MASK;
1681                         if (nr <= offset)
1682                                 break;
1683                 }
1684
1685                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1686                 if (desc->error) {
1687                         if (desc->error == -EINVAL)
1688                                 desc->error = 0;
1689                         break;
1690                 }
1691                 if (page)
1692                         unlock_page(page);
1693
1694                 /*
1695                  * We must evaluate after, since reads (unlike writes)
1696                  * are called without i_mutex protection against truncate
1697                  */
1698                 nr = PAGE_CACHE_SIZE;
1699                 i_size = i_size_read(inode);
1700                 end_index = i_size >> PAGE_CACHE_SHIFT;
1701                 if (index == end_index) {
1702                         nr = i_size & ~PAGE_CACHE_MASK;
1703                         if (nr <= offset) {
1704                                 if (page)
1705                                         page_cache_release(page);
1706                                 break;
1707                         }
1708                 }
1709                 nr -= offset;
1710
1711                 if (page) {
1712                         /*
1713                          * If users can be writing to this page using arbitrary
1714                          * virtual addresses, take care about potential aliasing
1715                          * before reading the page on the kernel side.
1716                          */
1717                         if (mapping_writably_mapped(mapping))
1718                                 flush_dcache_page(page);
1719                         /*
1720                          * Mark the page accessed if we read the beginning.
1721                          */
1722                         if (!offset)
1723                                 mark_page_accessed(page);
1724                 } else {
1725                         page = ZERO_PAGE(0);
1726                         page_cache_get(page);
1727                 }
1728
1729                 /*
1730                  * Ok, we have the page, and it's up-to-date, so
1731                  * now we can copy it to user space...
1732                  *
1733                  * The actor routine returns how many bytes were actually used..
1734                  * NOTE! This may not be the same as how much of a user buffer
1735                  * we filled up (we may be padding etc), so we can only update
1736                  * "pos" here (the actor routine has to update the user buffer
1737                  * pointers and the remaining count).
1738                  */
1739                 ret = actor(desc, page, offset, nr);
1740                 offset += ret;
1741                 index += offset >> PAGE_CACHE_SHIFT;
1742                 offset &= ~PAGE_CACHE_MASK;
1743
1744                 page_cache_release(page);
1745                 if (ret != nr || !desc->count)
1746                         break;
1747
1748                 cond_resched();
1749         }
1750
1751         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1752         file_accessed(filp);
1753 }
1754
1755 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1756                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1757 {
1758         struct file *filp = iocb->ki_filp;
1759         ssize_t retval;
1760         unsigned long seg;
1761         size_t count;
1762         loff_t *ppos = &iocb->ki_pos;
1763
1764         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1765         if (retval)
1766                 return retval;
1767
1768         for (seg = 0; seg < nr_segs; seg++) {
1769                 read_descriptor_t desc;
1770
1771                 desc.written = 0;
1772                 desc.arg.buf = iov[seg].iov_base;
1773                 desc.count = iov[seg].iov_len;
1774                 if (desc.count == 0)
1775                         continue;
1776                 desc.error = 0;
1777                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1778                 retval += desc.written;
1779                 if (desc.error) {
1780                         retval = retval ?: desc.error;
1781                         break;
1782                 }
1783                 if (desc.count > 0)
1784                         break;
1785         }
1786         return retval;
1787 }
1788
1789 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1790 {
1791         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1792
1793         buf->f_type = TMPFS_MAGIC;
1794         buf->f_bsize = PAGE_CACHE_SIZE;
1795         buf->f_namelen = NAME_MAX;
1796         spin_lock(&sbinfo->stat_lock);
1797         if (sbinfo->max_blocks) {
1798                 buf->f_blocks = sbinfo->max_blocks;
1799                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1800         }
1801         if (sbinfo->max_inodes) {
1802                 buf->f_files = sbinfo->max_inodes;
1803                 buf->f_ffree = sbinfo->free_inodes;
1804         }
1805         /* else leave those fields 0 like simple_statfs */
1806         spin_unlock(&sbinfo->stat_lock);
1807         return 0;
1808 }
1809
1810 /*
1811  * File creation. Allocate an inode, and we're done..
1812  */
1813 static int
1814 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1815 {
1816         struct inode *inode;
1817         int error = -ENOSPC;
1818
1819         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1820         if (inode) {
1821                 error = security_inode_init_security(inode, dir, NULL, NULL,
1822                                                      NULL);
1823                 if (error) {
1824                         if (error != -EOPNOTSUPP) {
1825                                 iput(inode);
1826                                 return error;
1827                         }
1828                 }
1829 #ifdef CONFIG_TMPFS_POSIX_ACL
1830                 error = generic_acl_init(inode, dir);
1831                 if (error) {
1832                         iput(inode);
1833                         return error;
1834                 }
1835 #else
1836                 error = 0;
1837 #endif
1838                 dir->i_size += BOGO_DIRENT_SIZE;
1839                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1840                 d_instantiate(dentry, inode);
1841                 dget(dentry); /* Extra count - pin the dentry in core */
1842         }
1843         return error;
1844 }
1845
1846 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1847 {
1848         int error;
1849
1850         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1851                 return error;
1852         inc_nlink(dir);
1853         return 0;
1854 }
1855
1856 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1857                 struct nameidata *nd)
1858 {
1859         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1860 }
1861
1862 /*
1863  * Link a file..
1864  */
1865 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1866 {
1867         struct inode *inode = old_dentry->d_inode;
1868         int ret;
1869
1870         /*
1871          * No ordinary (disk based) filesystem counts links as inodes;
1872          * but each new link needs a new dentry, pinning lowmem, and
1873          * tmpfs dentries cannot be pruned until they are unlinked.
1874          */
1875         ret = shmem_reserve_inode(inode->i_sb);
1876         if (ret)
1877                 goto out;
1878
1879         dir->i_size += BOGO_DIRENT_SIZE;
1880         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1881         inc_nlink(inode);
1882         atomic_inc(&inode->i_count);    /* New dentry reference */
1883         dget(dentry);           /* Extra pinning count for the created dentry */
1884         d_instantiate(dentry, inode);
1885 out:
1886         return ret;
1887 }
1888
1889 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1890 {
1891         struct inode *inode = dentry->d_inode;
1892
1893         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1894                 shmem_free_inode(inode->i_sb);
1895
1896         dir->i_size -= BOGO_DIRENT_SIZE;
1897         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1898         drop_nlink(inode);
1899         dput(dentry);   /* Undo the count from "create" - this does all the work */
1900         return 0;
1901 }
1902
1903 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1904 {
1905         if (!simple_empty(dentry))
1906                 return -ENOTEMPTY;
1907
1908         drop_nlink(dentry->d_inode);
1909         drop_nlink(dir);
1910         return shmem_unlink(dir, dentry);
1911 }
1912
1913 /*
1914  * The VFS layer already does all the dentry stuff for rename,
1915  * we just have to decrement the usage count for the target if
1916  * it exists so that the VFS layer correctly free's it when it
1917  * gets overwritten.
1918  */
1919 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1920 {
1921         struct inode *inode = old_dentry->d_inode;
1922         int they_are_dirs = S_ISDIR(inode->i_mode);
1923
1924         if (!simple_empty(new_dentry))
1925                 return -ENOTEMPTY;
1926
1927         if (new_dentry->d_inode) {
1928                 (void) shmem_unlink(new_dir, new_dentry);
1929                 if (they_are_dirs)
1930                         drop_nlink(old_dir);
1931         } else if (they_are_dirs) {
1932                 drop_nlink(old_dir);
1933                 inc_nlink(new_dir);
1934         }
1935
1936         old_dir->i_size -= BOGO_DIRENT_SIZE;
1937         new_dir->i_size += BOGO_DIRENT_SIZE;
1938         old_dir->i_ctime = old_dir->i_mtime =
1939         new_dir->i_ctime = new_dir->i_mtime =
1940         inode->i_ctime = CURRENT_TIME;
1941         return 0;
1942 }
1943
1944 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1945 {
1946         int error;
1947         int len;
1948         struct inode *inode;
1949         struct page *page = NULL;
1950         char *kaddr;
1951         struct shmem_inode_info *info;
1952
1953         len = strlen(symname) + 1;
1954         if (len > PAGE_CACHE_SIZE)
1955                 return -ENAMETOOLONG;
1956
1957         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1958         if (!inode)
1959                 return -ENOSPC;
1960
1961         error = security_inode_init_security(inode, dir, NULL, NULL,
1962                                              NULL);
1963         if (error) {
1964                 if (error != -EOPNOTSUPP) {
1965                         iput(inode);
1966                         return error;
1967                 }
1968                 error = 0;
1969         }
1970
1971         info = SHMEM_I(inode);
1972         inode->i_size = len-1;
1973         if (len <= (char *)inode - (char *)info) {
1974                 /* do it inline */
1975                 memcpy(info, symname, len);
1976                 inode->i_op = &shmem_symlink_inline_operations;
1977         } else {
1978                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1979                 if (error) {
1980                         iput(inode);
1981                         return error;
1982                 }
1983                 inode->i_mapping->a_ops = &shmem_aops;
1984                 inode->i_op = &shmem_symlink_inode_operations;
1985                 kaddr = kmap_atomic(page, KM_USER0);
1986                 memcpy(kaddr, symname, len);
1987                 kunmap_atomic(kaddr, KM_USER0);
1988                 set_page_dirty(page);
1989                 unlock_page(page);
1990                 page_cache_release(page);
1991         }
1992         dir->i_size += BOGO_DIRENT_SIZE;
1993         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1994         d_instantiate(dentry, inode);
1995         dget(dentry);
1996         return 0;
1997 }
1998
1999 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
2000 {
2001         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
2002         return NULL;
2003 }
2004
2005 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2006 {
2007         struct page *page = NULL;
2008         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2009         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2010         if (page)
2011                 unlock_page(page);
2012         return page;
2013 }
2014
2015 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2016 {
2017         if (!IS_ERR(nd_get_link(nd))) {
2018                 struct page *page = cookie;
2019                 kunmap(page);
2020                 mark_page_accessed(page);
2021                 page_cache_release(page);
2022         }
2023 }
2024
2025 static const struct inode_operations shmem_symlink_inline_operations = {
2026         .readlink       = generic_readlink,
2027         .follow_link    = shmem_follow_link_inline,
2028 };
2029
2030 static const struct inode_operations shmem_symlink_inode_operations = {
2031         .readlink       = generic_readlink,
2032         .follow_link    = shmem_follow_link,
2033         .put_link       = shmem_put_link,
2034 };
2035
2036 #ifdef CONFIG_TMPFS_POSIX_ACL
2037 /*
2038  * Superblocks without xattr inode operations will get security.* xattr
2039  * support from the VFS "for free". As soon as we have any other xattrs
2040  * like ACLs, we also need to implement the security.* handlers at
2041  * filesystem level, though.
2042  */
2043
2044 static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
2045                                         size_t list_len, const char *name,
2046                                         size_t name_len, int handler_flags)
2047 {
2048         return security_inode_listsecurity(dentry->d_inode, list, list_len);
2049 }
2050
2051 static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2052                 void *buffer, size_t size, int handler_flags)
2053 {
2054         if (strcmp(name, "") == 0)
2055                 return -EINVAL;
2056         return xattr_getsecurity(dentry->d_inode, name, buffer, size);
2057 }
2058
2059 static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2060                 const void *value, size_t size, int flags, int handler_flags)
2061 {
2062         if (strcmp(name, "") == 0)
2063                 return -EINVAL;
2064         return security_inode_setsecurity(dentry->d_inode, name, value,
2065                                           size, flags);
2066 }
2067
2068 static const struct xattr_handler shmem_xattr_security_handler = {
2069         .prefix = XATTR_SECURITY_PREFIX,
2070         .list   = shmem_xattr_security_list,
2071         .get    = shmem_xattr_security_get,
2072         .set    = shmem_xattr_security_set,
2073 };
2074
2075 static const struct xattr_handler *shmem_xattr_handlers[] = {
2076         &generic_acl_access_handler,
2077         &generic_acl_default_handler,
2078         &shmem_xattr_security_handler,
2079         NULL
2080 };
2081 #endif
2082
2083 static struct dentry *shmem_get_parent(struct dentry *child)
2084 {
2085         return ERR_PTR(-ESTALE);
2086 }
2087
2088 static int shmem_match(struct inode *ino, void *vfh)
2089 {
2090         __u32 *fh = vfh;
2091         __u64 inum = fh[2];
2092         inum = (inum << 32) | fh[1];
2093         return ino->i_ino == inum && fh[0] == ino->i_generation;
2094 }
2095
2096 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2097                 struct fid *fid, int fh_len, int fh_type)
2098 {
2099         struct inode *inode;
2100         struct dentry *dentry = NULL;
2101         u64 inum = fid->raw[2];
2102         inum = (inum << 32) | fid->raw[1];
2103
2104         if (fh_len < 3)
2105                 return NULL;
2106
2107         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2108                         shmem_match, fid->raw);
2109         if (inode) {
2110                 dentry = d_find_alias(inode);
2111                 iput(inode);
2112         }
2113
2114         return dentry;
2115 }
2116
2117 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2118                                 int connectable)
2119 {
2120         struct inode *inode = dentry->d_inode;
2121
2122         if (*len < 3)
2123                 return 255;
2124
2125         if (hlist_unhashed(&inode->i_hash)) {
2126                 /* Unfortunately insert_inode_hash is not idempotent,
2127                  * so as we hash inodes here rather than at creation
2128                  * time, we need a lock to ensure we only try
2129                  * to do it once
2130                  */
2131                 static DEFINE_SPINLOCK(lock);
2132                 spin_lock(&lock);
2133                 if (hlist_unhashed(&inode->i_hash))
2134                         __insert_inode_hash(inode,
2135                                             inode->i_ino + inode->i_generation);
2136                 spin_unlock(&lock);
2137         }
2138
2139         fh[0] = inode->i_generation;
2140         fh[1] = inode->i_ino;
2141         fh[2] = ((__u64)inode->i_ino) >> 32;
2142
2143         *len = 3;
2144         return 1;
2145 }
2146
2147 static const struct export_operations shmem_export_ops = {
2148         .get_parent     = shmem_get_parent,
2149         .encode_fh      = shmem_encode_fh,
2150         .fh_to_dentry   = shmem_fh_to_dentry,
2151 };
2152
2153 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2154                                bool remount)
2155 {
2156         char *this_char, *value, *rest;
2157
2158         while (options != NULL) {
2159                 this_char = options;
2160                 for (;;) {
2161                         /*
2162                          * NUL-terminate this option: unfortunately,
2163                          * mount options form a comma-separated list,
2164                          * but mpol's nodelist may also contain commas.
2165                          */
2166                         options = strchr(options, ',');
2167                         if (options == NULL)
2168                                 break;
2169                         options++;
2170                         if (!isdigit(*options)) {
2171                                 options[-1] = '\0';
2172                                 break;
2173                         }
2174                 }
2175                 if (!*this_char)
2176                         continue;
2177                 if ((value = strchr(this_char,'=')) != NULL) {
2178                         *value++ = 0;
2179                 } else {
2180                         printk(KERN_ERR
2181                             "tmpfs: No value for mount option '%s'\n",
2182                             this_char);
2183                         return 1;
2184                 }
2185
2186                 if (!strcmp(this_char,"size")) {
2187                         unsigned long long size;
2188                         size = memparse(value,&rest);
2189                         if (*rest == '%') {
2190                                 size <<= PAGE_SHIFT;
2191                                 size *= totalram_pages;
2192                                 do_div(size, 100);
2193                                 rest++;
2194                         }
2195                         if (*rest)
2196                                 goto bad_val;
2197                         sbinfo->max_blocks =
2198                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2199                 } else if (!strcmp(this_char,"nr_blocks")) {
2200                         sbinfo->max_blocks = memparse(value, &rest);
2201                         if (*rest)
2202                                 goto bad_val;
2203                 } else if (!strcmp(this_char,"nr_inodes")) {
2204                         sbinfo->max_inodes = memparse(value, &rest);
2205                         if (*rest)
2206                                 goto bad_val;
2207                 } else if (!strcmp(this_char,"mode")) {
2208                         if (remount)
2209                                 continue;
2210                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2211                         if (*rest)
2212                                 goto bad_val;
2213                 } else if (!strcmp(this_char,"uid")) {
2214                         if (remount)
2215                                 continue;
2216                         sbinfo->uid = simple_strtoul(value, &rest, 0);
2217                         if (*rest)
2218                                 goto bad_val;
2219                 } else if (!strcmp(this_char,"gid")) {
2220                         if (remount)
2221                                 continue;
2222                         sbinfo->gid = simple_strtoul(value, &rest, 0);
2223                         if (*rest)
2224                                 goto bad_val;
2225                 } else if (!strcmp(this_char,"mpol")) {
2226                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2227                                 goto bad_val;
2228                 } else {
2229                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2230                                this_char);
2231                         return 1;
2232                 }
2233         }
2234         return 0;
2235
2236 bad_val:
2237         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2238                value, this_char);
2239         return 1;
2240
2241 }
2242
2243 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2244 {
2245         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2246         struct shmem_sb_info config = *sbinfo;
2247         unsigned long blocks;
2248         unsigned long inodes;
2249         int error = -EINVAL;
2250
2251         if (shmem_parse_options(data, &config, true))
2252                 return error;
2253
2254         spin_lock(&sbinfo->stat_lock);
2255         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2256         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2257         if (config.max_blocks < blocks)
2258                 goto out;
2259         if (config.max_inodes < inodes)
2260                 goto out;
2261         /*
2262          * Those tests also disallow limited->unlimited while any are in
2263          * use, so i_blocks will always be zero when max_blocks is zero;
2264          * but we must separately disallow unlimited->limited, because
2265          * in that case we have no record of how much is already in use.
2266          */
2267         if (config.max_blocks && !sbinfo->max_blocks)
2268                 goto out;
2269         if (config.max_inodes && !sbinfo->max_inodes)
2270                 goto out;
2271
2272         error = 0;
2273         sbinfo->max_blocks  = config.max_blocks;
2274         sbinfo->free_blocks = config.max_blocks - blocks;
2275         sbinfo->max_inodes  = config.max_inodes;
2276         sbinfo->free_inodes = config.max_inodes - inodes;
2277
2278         mpol_put(sbinfo->mpol);
2279         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2280 out:
2281         spin_unlock(&sbinfo->stat_lock);
2282         return error;
2283 }
2284
2285 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2286 {
2287         struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2288
2289         if (sbinfo->max_blocks != shmem_default_max_blocks())
2290                 seq_printf(seq, ",size=%luk",
2291                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2292         if (sbinfo->max_inodes != shmem_default_max_inodes())
2293                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2294         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2295                 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2296         if (sbinfo->uid != 0)
2297                 seq_printf(seq, ",uid=%u", sbinfo->uid);
2298         if (sbinfo->gid != 0)
2299                 seq_printf(seq, ",gid=%u", sbinfo->gid);
2300         shmem_show_mpol(seq, sbinfo->mpol);
2301         return 0;
2302 }
2303 #endif /* CONFIG_TMPFS */
2304
2305 static void shmem_put_super(struct super_block *sb)
2306 {
2307         kfree(sb->s_fs_info);
2308         sb->s_fs_info = NULL;
2309 }
2310
2311 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2312 {
2313         struct inode *inode;
2314         struct dentry *root;
2315         struct shmem_sb_info *sbinfo;
2316         int err = -ENOMEM;
2317
2318         /* Round up to L1_CACHE_BYTES to resist false sharing */
2319         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2320                                 L1_CACHE_BYTES), GFP_KERNEL);
2321         if (!sbinfo)
2322                 return -ENOMEM;
2323
2324         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2325         sbinfo->uid = current_fsuid();
2326         sbinfo->gid = current_fsgid();
2327         sb->s_fs_info = sbinfo;
2328
2329 #ifdef CONFIG_TMPFS
2330         /*
2331          * Per default we only allow half of the physical ram per
2332          * tmpfs instance, limiting inodes to one per page of lowmem;
2333          * but the internal instance is left unlimited.
2334          */
2335         if (!(sb->s_flags & MS_NOUSER)) {
2336                 sbinfo->max_blocks = shmem_default_max_blocks();
2337                 sbinfo->max_inodes = shmem_default_max_inodes();
2338                 if (shmem_parse_options(data, sbinfo, false)) {
2339                         err = -EINVAL;
2340                         goto failed;
2341                 }
2342         }
2343         sb->s_export_op = &shmem_export_ops;
2344 #else
2345         sb->s_flags |= MS_NOUSER;
2346 #endif
2347
2348         spin_lock_init(&sbinfo->stat_lock);
2349         sbinfo->free_blocks = sbinfo->max_blocks;
2350         sbinfo->free_inodes = sbinfo->max_inodes;
2351
2352         sb->s_maxbytes = SHMEM_MAX_BYTES;
2353         sb->s_blocksize = PAGE_CACHE_SIZE;
2354         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2355         sb->s_magic = TMPFS_MAGIC;
2356         sb->s_op = &shmem_ops;
2357         sb->s_time_gran = 1;
2358 #ifdef CONFIG_TMPFS_POSIX_ACL
2359         sb->s_xattr = shmem_xattr_handlers;
2360         sb->s_flags |= MS_POSIXACL;
2361 #endif
2362
2363         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2364         if (!inode)
2365                 goto failed;
2366         inode->i_uid = sbinfo->uid;
2367         inode->i_gid = sbinfo->gid;
2368         root = d_alloc_root(inode);
2369         if (!root)
2370                 goto failed_iput;
2371         sb->s_root = root;
2372         return 0;
2373
2374 failed_iput:
2375         iput(inode);
2376 failed:
2377         shmem_put_super(sb);
2378         return err;
2379 }
2380
2381 static struct kmem_cache *shmem_inode_cachep;
2382
2383 static struct inode *shmem_alloc_inode(struct super_block *sb)
2384 {
2385         struct shmem_inode_info *p;
2386         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2387         if (!p)
2388                 return NULL;
2389         return &p->vfs_inode;
2390 }
2391
2392 static void shmem_destroy_inode(struct inode *inode)
2393 {
2394         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2395                 /* only struct inode is valid if it's an inline symlink */
2396                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2397         }
2398         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2399 }
2400
2401 static void init_once(void *foo)
2402 {
2403         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2404
2405         inode_init_once(&p->vfs_inode);
2406 }
2407
2408 static int init_inodecache(void)
2409 {
2410         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2411                                 sizeof(struct shmem_inode_info),
2412                                 0, SLAB_PANIC, init_once);
2413         return 0;
2414 }
2415
2416 static void destroy_inodecache(void)
2417 {
2418         kmem_cache_destroy(shmem_inode_cachep);
2419 }
2420
2421 static const struct address_space_operations shmem_aops = {
2422         .writepage      = shmem_writepage,
2423         .set_page_dirty = __set_page_dirty_no_writeback,
2424 #ifdef CONFIG_TMPFS
2425         .readpage       = shmem_readpage,
2426         .write_begin    = shmem_write_begin,
2427         .write_end      = shmem_write_end,
2428 #endif
2429         .migratepage    = migrate_page,
2430         .error_remove_page = generic_error_remove_page,
2431 };
2432
2433 static const struct file_operations shmem_file_operations = {
2434         .mmap           = shmem_mmap,
2435 #ifdef CONFIG_TMPFS
2436         .llseek         = generic_file_llseek,
2437         .read           = do_sync_read,
2438         .write          = do_sync_write,
2439         .aio_read       = shmem_file_aio_read,
2440         .aio_write      = generic_file_aio_write,
2441         .fsync          = noop_fsync,
2442         .splice_read    = generic_file_splice_read,
2443         .splice_write   = generic_file_splice_write,
2444 #endif
2445 };
2446
2447 static const struct inode_operations shmem_inode_operations = {
2448         .setattr        = shmem_notify_change,
2449         .truncate_range = shmem_truncate_range,
2450 #ifdef CONFIG_TMPFS_POSIX_ACL
2451         .setxattr       = generic_setxattr,
2452         .getxattr       = generic_getxattr,
2453         .listxattr      = generic_listxattr,
2454         .removexattr    = generic_removexattr,
2455         .check_acl      = generic_check_acl,
2456 #endif
2457
2458 };
2459
2460 static const struct inode_operations shmem_dir_inode_operations = {
2461 #ifdef CONFIG_TMPFS
2462         .create         = shmem_create,
2463         .lookup         = simple_lookup,
2464         .link           = shmem_link,
2465         .unlink         = shmem_unlink,
2466         .symlink        = shmem_symlink,
2467         .mkdir          = shmem_mkdir,
2468         .rmdir          = shmem_rmdir,
2469         .mknod          = shmem_mknod,
2470         .rename         = shmem_rename,
2471 #endif
2472 #ifdef CONFIG_TMPFS_POSIX_ACL
2473         .setattr        = shmem_notify_change,
2474         .setxattr       = generic_setxattr,
2475         .getxattr       = generic_getxattr,
2476         .listxattr      = generic_listxattr,
2477         .removexattr    = generic_removexattr,
2478         .check_acl      = generic_check_acl,
2479 #endif
2480 };
2481
2482 static const struct inode_operations shmem_special_inode_operations = {
2483 #ifdef CONFIG_TMPFS_POSIX_ACL
2484         .setattr        = shmem_notify_change,
2485         .setxattr       = generic_setxattr,
2486         .getxattr       = generic_getxattr,
2487         .listxattr      = generic_listxattr,
2488         .removexattr    = generic_removexattr,
2489         .check_acl      = generic_check_acl,
2490 #endif
2491 };
2492
2493 static const struct super_operations shmem_ops = {
2494         .alloc_inode    = shmem_alloc_inode,
2495         .destroy_inode  = shmem_destroy_inode,
2496 #ifdef CONFIG_TMPFS
2497         .statfs         = shmem_statfs,
2498         .remount_fs     = shmem_remount_fs,
2499         .show_options   = shmem_show_options,
2500 #endif
2501         .delete_inode   = shmem_delete_inode,
2502         .drop_inode     = generic_delete_inode,
2503         .put_super      = shmem_put_super,
2504 };
2505
2506 static const struct vm_operations_struct shmem_vm_ops = {
2507         .fault          = shmem_fault,
2508 #ifdef CONFIG_NUMA
2509         .set_policy     = shmem_set_policy,
2510         .get_policy     = shmem_get_policy,
2511 #endif
2512 };
2513
2514
2515 static int shmem_get_sb(struct file_system_type *fs_type,
2516         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2517 {
2518         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2519 }
2520
2521 static struct file_system_type tmpfs_fs_type = {
2522         .owner          = THIS_MODULE,
2523         .name           = "tmpfs",
2524         .get_sb         = shmem_get_sb,
2525         .kill_sb        = kill_litter_super,
2526 };
2527
2528 int __init init_tmpfs(void)
2529 {
2530         int error;
2531
2532         error = bdi_init(&shmem_backing_dev_info);
2533         if (error)
2534                 goto out4;
2535
2536         error = init_inodecache();
2537         if (error)
2538                 goto out3;
2539
2540         error = register_filesystem(&tmpfs_fs_type);
2541         if (error) {
2542                 printk(KERN_ERR "Could not register tmpfs\n");
2543                 goto out2;
2544         }
2545
2546         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2547                                 tmpfs_fs_type.name, NULL);
2548         if (IS_ERR(shm_mnt)) {
2549                 error = PTR_ERR(shm_mnt);
2550                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2551                 goto out1;
2552         }
2553         return 0;
2554
2555 out1:
2556         unregister_filesystem(&tmpfs_fs_type);
2557 out2:
2558         destroy_inodecache();
2559 out3:
2560         bdi_destroy(&shmem_backing_dev_info);
2561 out4:
2562         shm_mnt = ERR_PTR(error);
2563         return error;
2564 }
2565
2566 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2567 /**
2568  * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2569  * @inode: the inode to be searched
2570  * @pgoff: the offset to be searched
2571  * @pagep: the pointer for the found page to be stored
2572  * @ent: the pointer for the found swap entry to be stored
2573  *
2574  * If a page is found, refcount of it is incremented. Callers should handle
2575  * these refcount.
2576  */
2577 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2578                                         struct page **pagep, swp_entry_t *ent)
2579 {
2580         swp_entry_t entry = { .val = 0 }, *ptr;
2581         struct page *page = NULL;
2582         struct shmem_inode_info *info = SHMEM_I(inode);
2583
2584         if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2585                 goto out;
2586
2587         spin_lock(&info->lock);
2588         ptr = shmem_swp_entry(info, pgoff, NULL);
2589 #ifdef CONFIG_SWAP
2590         if (ptr && ptr->val) {
2591                 entry.val = ptr->val;
2592                 page = find_get_page(&swapper_space, entry.val);
2593         } else
2594 #endif
2595                 page = find_get_page(inode->i_mapping, pgoff);
2596         if (ptr)
2597                 shmem_swp_unmap(ptr);
2598         spin_unlock(&info->lock);
2599 out:
2600         *pagep = page;
2601         *ent = entry;
2602 }
2603 #endif
2604
2605 #else /* !CONFIG_SHMEM */
2606
2607 /*
2608  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2609  *
2610  * This is intended for small system where the benefits of the full
2611  * shmem code (swap-backed and resource-limited) are outweighed by
2612  * their complexity. On systems without swap this code should be
2613  * effectively equivalent, but much lighter weight.
2614  */
2615
2616 #include <linux/ramfs.h>
2617
2618 static struct file_system_type tmpfs_fs_type = {
2619         .name           = "tmpfs",
2620         .get_sb         = ramfs_get_sb,
2621         .kill_sb        = kill_litter_super,
2622 };
2623
2624 int __init init_tmpfs(void)
2625 {
2626         BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2627
2628         shm_mnt = kern_mount(&tmpfs_fs_type);
2629         BUG_ON(IS_ERR(shm_mnt));
2630
2631         return 0;
2632 }
2633
2634 int shmem_unuse(swp_entry_t entry, struct page *page)
2635 {
2636         return 0;
2637 }
2638
2639 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2640 {
2641         return 0;
2642 }
2643
2644 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2645 /**
2646  * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2647  * @inode: the inode to be searched
2648  * @pgoff: the offset to be searched
2649  * @pagep: the pointer for the found page to be stored
2650  * @ent: the pointer for the found swap entry to be stored
2651  *
2652  * If a page is found, refcount of it is incremented. Callers should handle
2653  * these refcount.
2654  */
2655 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2656                                         struct page **pagep, swp_entry_t *ent)
2657 {
2658         struct page *page = NULL;
2659
2660         if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2661                 goto out;
2662         page = find_get_page(inode->i_mapping, pgoff);
2663 out:
2664         *pagep = page;
2665         *ent = (swp_entry_t){ .val = 0 };
2666 }
2667 #endif
2668
2669 #define shmem_vm_ops                            generic_file_vm_ops
2670 #define shmem_file_operations                   ramfs_file_operations
2671 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2672 #define shmem_acct_size(flags, size)            0
2673 #define shmem_unacct_size(flags, size)          do {} while (0)
2674 #define SHMEM_MAX_BYTES                         MAX_LFS_FILESIZE
2675
2676 #endif /* CONFIG_SHMEM */
2677
2678 /* common code */
2679
2680 /**
2681  * shmem_file_setup - get an unlinked file living in tmpfs
2682  * @name: name for dentry (to be seen in /proc/<pid>/maps
2683  * @size: size to be set for the file
2684  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2685  */
2686 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2687 {
2688         int error;
2689         struct file *file;
2690         struct inode *inode;
2691         struct path path;
2692         struct dentry *root;
2693         struct qstr this;
2694
2695         if (IS_ERR(shm_mnt))
2696                 return (void *)shm_mnt;
2697
2698         if (size < 0 || size > SHMEM_MAX_BYTES)
2699                 return ERR_PTR(-EINVAL);
2700
2701         if (shmem_acct_size(flags, size))
2702                 return ERR_PTR(-ENOMEM);
2703
2704         error = -ENOMEM;
2705         this.name = name;
2706         this.len = strlen(name);
2707         this.hash = 0; /* will go */
2708         root = shm_mnt->mnt_root;
2709         path.dentry = d_alloc(root, &this);
2710         if (!path.dentry)
2711                 goto put_memory;
2712         path.mnt = mntget(shm_mnt);
2713
2714         error = -ENOSPC;
2715         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2716         if (!inode)
2717                 goto put_dentry;
2718
2719         d_instantiate(path.dentry, inode);
2720         inode->i_size = size;
2721         inode->i_nlink = 0;     /* It is unlinked */
2722 #ifndef CONFIG_MMU
2723         error = ramfs_nommu_expand_for_mapping(inode, size);
2724         if (error)
2725                 goto put_dentry;
2726 #endif
2727
2728         error = -ENFILE;
2729         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2730                   &shmem_file_operations);
2731         if (!file)
2732                 goto put_dentry;
2733
2734         return file;
2735
2736 put_dentry:
2737         path_put(&path);
2738 put_memory:
2739         shmem_unacct_size(flags, size);
2740         return ERR_PTR(error);
2741 }
2742 EXPORT_SYMBOL_GPL(shmem_file_setup);
2743
2744 /**
2745  * shmem_zero_setup - setup a shared anonymous mapping
2746  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2747  */
2748 int shmem_zero_setup(struct vm_area_struct *vma)
2749 {
2750         struct file *file;
2751         loff_t size = vma->vm_end - vma->vm_start;
2752
2753         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2754         if (IS_ERR(file))
2755                 return PTR_ERR(file);
2756
2757         if (vma->vm_file)
2758                 fput(vma->vm_file);
2759         vma->vm_file = file;
2760         vma->vm_ops = &shmem_vm_ops;
2761         return 0;
2762 }