check ATTR_SIZE contraints in inode_change_ok
[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                 /* XXX(truncate): truncate_setsize should be called last */
809                 truncate_setsize(inode, newsize);
810                 if (page)
811                         page_cache_release(page);
812                 shmem_truncate_range(inode, newsize, (loff_t)-1);
813         }
814
815         setattr_copy(inode, attr);
816 #ifdef CONFIG_TMPFS_POSIX_ACL
817         if (attr->ia_valid & ATTR_MODE)
818                 error = generic_acl_chmod(inode);
819 #endif
820         return error;
821 }
822
823 static void shmem_delete_inode(struct inode *inode)
824 {
825         struct shmem_inode_info *info = SHMEM_I(inode);
826
827         if (inode->i_mapping->a_ops == &shmem_aops) {
828                 truncate_inode_pages(inode->i_mapping, 0);
829                 shmem_unacct_size(info->flags, inode->i_size);
830                 inode->i_size = 0;
831                 shmem_truncate_range(inode, 0, (loff_t)-1);
832                 if (!list_empty(&info->swaplist)) {
833                         mutex_lock(&shmem_swaplist_mutex);
834                         list_del_init(&info->swaplist);
835                         mutex_unlock(&shmem_swaplist_mutex);
836                 }
837         }
838         BUG_ON(inode->i_blocks);
839         shmem_free_inode(inode->i_sb);
840         clear_inode(inode);
841 }
842
843 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
844 {
845         swp_entry_t *ptr;
846
847         for (ptr = dir; ptr < edir; ptr++) {
848                 if (ptr->val == entry.val)
849                         return ptr - dir;
850         }
851         return -1;
852 }
853
854 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
855 {
856         struct inode *inode;
857         unsigned long idx;
858         unsigned long size;
859         unsigned long limit;
860         unsigned long stage;
861         struct page **dir;
862         struct page *subdir;
863         swp_entry_t *ptr;
864         int offset;
865         int error;
866
867         idx = 0;
868         ptr = info->i_direct;
869         spin_lock(&info->lock);
870         if (!info->swapped) {
871                 list_del_init(&info->swaplist);
872                 goto lost2;
873         }
874         limit = info->next_index;
875         size = limit;
876         if (size > SHMEM_NR_DIRECT)
877                 size = SHMEM_NR_DIRECT;
878         offset = shmem_find_swp(entry, ptr, ptr+size);
879         if (offset >= 0)
880                 goto found;
881         if (!info->i_indirect)
882                 goto lost2;
883
884         dir = shmem_dir_map(info->i_indirect);
885         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
886
887         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
888                 if (unlikely(idx == stage)) {
889                         shmem_dir_unmap(dir-1);
890                         if (cond_resched_lock(&info->lock)) {
891                                 /* check it has not been truncated */
892                                 if (limit > info->next_index) {
893                                         limit = info->next_index;
894                                         if (idx >= limit)
895                                                 goto lost2;
896                                 }
897                         }
898                         dir = shmem_dir_map(info->i_indirect) +
899                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
900                         while (!*dir) {
901                                 dir++;
902                                 idx += ENTRIES_PER_PAGEPAGE;
903                                 if (idx >= limit)
904                                         goto lost1;
905                         }
906                         stage = idx + ENTRIES_PER_PAGEPAGE;
907                         subdir = *dir;
908                         shmem_dir_unmap(dir);
909                         dir = shmem_dir_map(subdir);
910                 }
911                 subdir = *dir;
912                 if (subdir && page_private(subdir)) {
913                         ptr = shmem_swp_map(subdir);
914                         size = limit - idx;
915                         if (size > ENTRIES_PER_PAGE)
916                                 size = ENTRIES_PER_PAGE;
917                         offset = shmem_find_swp(entry, ptr, ptr+size);
918                         shmem_swp_unmap(ptr);
919                         if (offset >= 0) {
920                                 shmem_dir_unmap(dir);
921                                 goto found;
922                         }
923                 }
924         }
925 lost1:
926         shmem_dir_unmap(dir-1);
927 lost2:
928         spin_unlock(&info->lock);
929         return 0;
930 found:
931         idx += offset;
932         inode = igrab(&info->vfs_inode);
933         spin_unlock(&info->lock);
934
935         /*
936          * Move _head_ to start search for next from here.
937          * But be careful: shmem_delete_inode checks list_empty without taking
938          * mutex, and there's an instant in list_move_tail when info->swaplist
939          * would appear empty, if it were the only one on shmem_swaplist.  We
940          * could avoid doing it if inode NULL; or use this minor optimization.
941          */
942         if (shmem_swaplist.next != &info->swaplist)
943                 list_move_tail(&shmem_swaplist, &info->swaplist);
944         mutex_unlock(&shmem_swaplist_mutex);
945
946         error = 1;
947         if (!inode)
948                 goto out;
949         /*
950          * Charge page using GFP_KERNEL while we can wait.
951          * Charged back to the user(not to caller) when swap account is used.
952          * add_to_page_cache() will be called with GFP_NOWAIT.
953          */
954         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
955         if (error)
956                 goto out;
957         error = radix_tree_preload(GFP_KERNEL);
958         if (error) {
959                 mem_cgroup_uncharge_cache_page(page);
960                 goto out;
961         }
962         error = 1;
963
964         spin_lock(&info->lock);
965         ptr = shmem_swp_entry(info, idx, NULL);
966         if (ptr && ptr->val == entry.val) {
967                 error = add_to_page_cache_locked(page, inode->i_mapping,
968                                                 idx, GFP_NOWAIT);
969                 /* does mem_cgroup_uncharge_cache_page on error */
970         } else  /* we must compensate for our precharge above */
971                 mem_cgroup_uncharge_cache_page(page);
972
973         if (error == -EEXIST) {
974                 struct page *filepage = find_get_page(inode->i_mapping, idx);
975                 error = 1;
976                 if (filepage) {
977                         /*
978                          * There might be a more uptodate page coming down
979                          * from a stacked writepage: forget our swappage if so.
980                          */
981                         if (PageUptodate(filepage))
982                                 error = 0;
983                         page_cache_release(filepage);
984                 }
985         }
986         if (!error) {
987                 delete_from_swap_cache(page);
988                 set_page_dirty(page);
989                 info->flags |= SHMEM_PAGEIN;
990                 shmem_swp_set(info, ptr, 0);
991                 swap_free(entry);
992                 error = 1;      /* not an error, but entry was found */
993         }
994         if (ptr)
995                 shmem_swp_unmap(ptr);
996         spin_unlock(&info->lock);
997         radix_tree_preload_end();
998 out:
999         unlock_page(page);
1000         page_cache_release(page);
1001         iput(inode);            /* allows for NULL */
1002         return error;
1003 }
1004
1005 /*
1006  * shmem_unuse() search for an eventually swapped out shmem page.
1007  */
1008 int shmem_unuse(swp_entry_t entry, struct page *page)
1009 {
1010         struct list_head *p, *next;
1011         struct shmem_inode_info *info;
1012         int found = 0;
1013
1014         mutex_lock(&shmem_swaplist_mutex);
1015         list_for_each_safe(p, next, &shmem_swaplist) {
1016                 info = list_entry(p, struct shmem_inode_info, swaplist);
1017                 found = shmem_unuse_inode(info, entry, page);
1018                 cond_resched();
1019                 if (found)
1020                         goto out;
1021         }
1022         mutex_unlock(&shmem_swaplist_mutex);
1023         /*
1024          * Can some race bring us here?  We've been holding page lock,
1025          * so I think not; but would rather try again later than BUG()
1026          */
1027         unlock_page(page);
1028         page_cache_release(page);
1029 out:
1030         return (found < 0) ? found : 0;
1031 }
1032
1033 /*
1034  * Move the page from the page cache to the swap cache.
1035  */
1036 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1037 {
1038         struct shmem_inode_info *info;
1039         swp_entry_t *entry, swap;
1040         struct address_space *mapping;
1041         unsigned long index;
1042         struct inode *inode;
1043
1044         BUG_ON(!PageLocked(page));
1045         mapping = page->mapping;
1046         index = page->index;
1047         inode = mapping->host;
1048         info = SHMEM_I(inode);
1049         if (info->flags & VM_LOCKED)
1050                 goto redirty;
1051         if (!total_swap_pages)
1052                 goto redirty;
1053
1054         /*
1055          * shmem_backing_dev_info's capabilities prevent regular writeback or
1056          * sync from ever calling shmem_writepage; but a stacking filesystem
1057          * may use the ->writepage of its underlying filesystem, in which case
1058          * tmpfs should write out to swap only in response to memory pressure,
1059          * and not for the writeback threads or sync.  However, in those cases,
1060          * we do still want to check if there's a redundant swappage to be
1061          * discarded.
1062          */
1063         if (wbc->for_reclaim)
1064                 swap = get_swap_page();
1065         else
1066                 swap.val = 0;
1067
1068         spin_lock(&info->lock);
1069         if (index >= info->next_index) {
1070                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1071                 goto unlock;
1072         }
1073         entry = shmem_swp_entry(info, index, NULL);
1074         if (entry->val) {
1075                 /*
1076                  * The more uptodate page coming down from a stacked
1077                  * writepage should replace our old swappage.
1078                  */
1079                 free_swap_and_cache(*entry);
1080                 shmem_swp_set(info, entry, 0);
1081         }
1082         shmem_recalc_inode(inode);
1083
1084         if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1085                 remove_from_page_cache(page);
1086                 shmem_swp_set(info, entry, swap.val);
1087                 shmem_swp_unmap(entry);
1088                 if (list_empty(&info->swaplist))
1089                         inode = igrab(inode);
1090                 else
1091                         inode = NULL;
1092                 spin_unlock(&info->lock);
1093                 swap_shmem_alloc(swap);
1094                 BUG_ON(page_mapped(page));
1095                 page_cache_release(page);       /* pagecache ref */
1096                 swap_writepage(page, wbc);
1097                 if (inode) {
1098                         mutex_lock(&shmem_swaplist_mutex);
1099                         /* move instead of add in case we're racing */
1100                         list_move_tail(&info->swaplist, &shmem_swaplist);
1101                         mutex_unlock(&shmem_swaplist_mutex);
1102                         iput(inode);
1103                 }
1104                 return 0;
1105         }
1106
1107         shmem_swp_unmap(entry);
1108 unlock:
1109         spin_unlock(&info->lock);
1110         /*
1111          * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1112          * clear SWAP_HAS_CACHE flag.
1113          */
1114         swapcache_free(swap, NULL);
1115 redirty:
1116         set_page_dirty(page);
1117         if (wbc->for_reclaim)
1118                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
1119         unlock_page(page);
1120         return 0;
1121 }
1122
1123 #ifdef CONFIG_NUMA
1124 #ifdef CONFIG_TMPFS
1125 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1126 {
1127         char buffer[64];
1128
1129         if (!mpol || mpol->mode == MPOL_DEFAULT)
1130                 return;         /* show nothing */
1131
1132         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1133
1134         seq_printf(seq, ",mpol=%s", buffer);
1135 }
1136
1137 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1138 {
1139         struct mempolicy *mpol = NULL;
1140         if (sbinfo->mpol) {
1141                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
1142                 mpol = sbinfo->mpol;
1143                 mpol_get(mpol);
1144                 spin_unlock(&sbinfo->stat_lock);
1145         }
1146         return mpol;
1147 }
1148 #endif /* CONFIG_TMPFS */
1149
1150 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1151                         struct shmem_inode_info *info, unsigned long idx)
1152 {
1153         struct mempolicy mpol, *spol;
1154         struct vm_area_struct pvma;
1155         struct page *page;
1156
1157         spol = mpol_cond_copy(&mpol,
1158                                 mpol_shared_policy_lookup(&info->policy, idx));
1159
1160         /* Create a pseudo vma that just contains the policy */
1161         pvma.vm_start = 0;
1162         pvma.vm_pgoff = idx;
1163         pvma.vm_ops = NULL;
1164         pvma.vm_policy = spol;
1165         page = swapin_readahead(entry, gfp, &pvma, 0);
1166         return page;
1167 }
1168
1169 static struct page *shmem_alloc_page(gfp_t gfp,
1170                         struct shmem_inode_info *info, unsigned long idx)
1171 {
1172         struct vm_area_struct pvma;
1173
1174         /* Create a pseudo vma that just contains the policy */
1175         pvma.vm_start = 0;
1176         pvma.vm_pgoff = idx;
1177         pvma.vm_ops = NULL;
1178         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1179
1180         /*
1181          * alloc_page_vma() will drop the shared policy reference
1182          */
1183         return alloc_page_vma(gfp, &pvma, 0);
1184 }
1185 #else /* !CONFIG_NUMA */
1186 #ifdef CONFIG_TMPFS
1187 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1188 {
1189 }
1190 #endif /* CONFIG_TMPFS */
1191
1192 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1193                         struct shmem_inode_info *info, unsigned long idx)
1194 {
1195         return swapin_readahead(entry, gfp, NULL, 0);
1196 }
1197
1198 static inline struct page *shmem_alloc_page(gfp_t gfp,
1199                         struct shmem_inode_info *info, unsigned long idx)
1200 {
1201         return alloc_page(gfp);
1202 }
1203 #endif /* CONFIG_NUMA */
1204
1205 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1206 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1207 {
1208         return NULL;
1209 }
1210 #endif
1211
1212 /*
1213  * shmem_getpage - either get the page from swap or allocate a new one
1214  *
1215  * If we allocate a new one we do not mark it dirty. That's up to the
1216  * vm. If we swap it in we mark it dirty since we also free the swap
1217  * entry since a page cannot live in both the swap and page cache
1218  */
1219 static int shmem_getpage(struct inode *inode, unsigned long idx,
1220                         struct page **pagep, enum sgp_type sgp, int *type)
1221 {
1222         struct address_space *mapping = inode->i_mapping;
1223         struct shmem_inode_info *info = SHMEM_I(inode);
1224         struct shmem_sb_info *sbinfo;
1225         struct page *filepage = *pagep;
1226         struct page *swappage;
1227         swp_entry_t *entry;
1228         swp_entry_t swap;
1229         gfp_t gfp;
1230         int error;
1231
1232         if (idx >= SHMEM_MAX_INDEX)
1233                 return -EFBIG;
1234
1235         if (type)
1236                 *type = 0;
1237
1238         /*
1239          * Normally, filepage is NULL on entry, and either found
1240          * uptodate immediately, or allocated and zeroed, or read
1241          * in under swappage, which is then assigned to filepage.
1242          * But shmem_readpage (required for splice) passes in a locked
1243          * filepage, which may be found not uptodate by other callers
1244          * too, and may need to be copied from the swappage read in.
1245          */
1246 repeat:
1247         if (!filepage)
1248                 filepage = find_lock_page(mapping, idx);
1249         if (filepage && PageUptodate(filepage))
1250                 goto done;
1251         error = 0;
1252         gfp = mapping_gfp_mask(mapping);
1253         if (!filepage) {
1254                 /*
1255                  * Try to preload while we can wait, to not make a habit of
1256                  * draining atomic reserves; but don't latch on to this cpu.
1257                  */
1258                 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1259                 if (error)
1260                         goto failed;
1261                 radix_tree_preload_end();
1262         }
1263
1264         spin_lock(&info->lock);
1265         shmem_recalc_inode(inode);
1266         entry = shmem_swp_alloc(info, idx, sgp);
1267         if (IS_ERR(entry)) {
1268                 spin_unlock(&info->lock);
1269                 error = PTR_ERR(entry);
1270                 goto failed;
1271         }
1272         swap = *entry;
1273
1274         if (swap.val) {
1275                 /* Look it up and read it in.. */
1276                 swappage = lookup_swap_cache(swap);
1277                 if (!swappage) {
1278                         shmem_swp_unmap(entry);
1279                         /* here we actually do the io */
1280                         if (type && !(*type & VM_FAULT_MAJOR)) {
1281                                 __count_vm_event(PGMAJFAULT);
1282                                 *type |= VM_FAULT_MAJOR;
1283                         }
1284                         spin_unlock(&info->lock);
1285                         swappage = shmem_swapin(swap, gfp, info, idx);
1286                         if (!swappage) {
1287                                 spin_lock(&info->lock);
1288                                 entry = shmem_swp_alloc(info, idx, sgp);
1289                                 if (IS_ERR(entry))
1290                                         error = PTR_ERR(entry);
1291                                 else {
1292                                         if (entry->val == swap.val)
1293                                                 error = -ENOMEM;
1294                                         shmem_swp_unmap(entry);
1295                                 }
1296                                 spin_unlock(&info->lock);
1297                                 if (error)
1298                                         goto failed;
1299                                 goto repeat;
1300                         }
1301                         wait_on_page_locked(swappage);
1302                         page_cache_release(swappage);
1303                         goto repeat;
1304                 }
1305
1306                 /* We have to do this with page locked to prevent races */
1307                 if (!trylock_page(swappage)) {
1308                         shmem_swp_unmap(entry);
1309                         spin_unlock(&info->lock);
1310                         wait_on_page_locked(swappage);
1311                         page_cache_release(swappage);
1312                         goto repeat;
1313                 }
1314                 if (PageWriteback(swappage)) {
1315                         shmem_swp_unmap(entry);
1316                         spin_unlock(&info->lock);
1317                         wait_on_page_writeback(swappage);
1318                         unlock_page(swappage);
1319                         page_cache_release(swappage);
1320                         goto repeat;
1321                 }
1322                 if (!PageUptodate(swappage)) {
1323                         shmem_swp_unmap(entry);
1324                         spin_unlock(&info->lock);
1325                         unlock_page(swappage);
1326                         page_cache_release(swappage);
1327                         error = -EIO;
1328                         goto failed;
1329                 }
1330
1331                 if (filepage) {
1332                         shmem_swp_set(info, entry, 0);
1333                         shmem_swp_unmap(entry);
1334                         delete_from_swap_cache(swappage);
1335                         spin_unlock(&info->lock);
1336                         copy_highpage(filepage, swappage);
1337                         unlock_page(swappage);
1338                         page_cache_release(swappage);
1339                         flush_dcache_page(filepage);
1340                         SetPageUptodate(filepage);
1341                         set_page_dirty(filepage);
1342                         swap_free(swap);
1343                 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1344                                         idx, GFP_NOWAIT))) {
1345                         info->flags |= SHMEM_PAGEIN;
1346                         shmem_swp_set(info, entry, 0);
1347                         shmem_swp_unmap(entry);
1348                         delete_from_swap_cache(swappage);
1349                         spin_unlock(&info->lock);
1350                         filepage = swappage;
1351                         set_page_dirty(filepage);
1352                         swap_free(swap);
1353                 } else {
1354                         shmem_swp_unmap(entry);
1355                         spin_unlock(&info->lock);
1356                         if (error == -ENOMEM) {
1357                                 /*
1358                                  * reclaim from proper memory cgroup and
1359                                  * call memcg's OOM if needed.
1360                                  */
1361                                 error = mem_cgroup_shmem_charge_fallback(
1362                                                                 swappage,
1363                                                                 current->mm,
1364                                                                 gfp);
1365                                 if (error) {
1366                                         unlock_page(swappage);
1367                                         page_cache_release(swappage);
1368                                         goto failed;
1369                                 }
1370                         }
1371                         unlock_page(swappage);
1372                         page_cache_release(swappage);
1373                         goto repeat;
1374                 }
1375         } else if (sgp == SGP_READ && !filepage) {
1376                 shmem_swp_unmap(entry);
1377                 filepage = find_get_page(mapping, idx);
1378                 if (filepage &&
1379                     (!PageUptodate(filepage) || !trylock_page(filepage))) {
1380                         spin_unlock(&info->lock);
1381                         wait_on_page_locked(filepage);
1382                         page_cache_release(filepage);
1383                         filepage = NULL;
1384                         goto repeat;
1385                 }
1386                 spin_unlock(&info->lock);
1387         } else {
1388                 shmem_swp_unmap(entry);
1389                 sbinfo = SHMEM_SB(inode->i_sb);
1390                 if (sbinfo->max_blocks) {
1391                         spin_lock(&sbinfo->stat_lock);
1392                         if (sbinfo->free_blocks == 0 ||
1393                             shmem_acct_block(info->flags)) {
1394                                 spin_unlock(&sbinfo->stat_lock);
1395                                 spin_unlock(&info->lock);
1396                                 error = -ENOSPC;
1397                                 goto failed;
1398                         }
1399                         sbinfo->free_blocks--;
1400                         inode->i_blocks += BLOCKS_PER_PAGE;
1401                         spin_unlock(&sbinfo->stat_lock);
1402                 } else if (shmem_acct_block(info->flags)) {
1403                         spin_unlock(&info->lock);
1404                         error = -ENOSPC;
1405                         goto failed;
1406                 }
1407
1408                 if (!filepage) {
1409                         int ret;
1410
1411                         spin_unlock(&info->lock);
1412                         filepage = shmem_alloc_page(gfp, info, idx);
1413                         if (!filepage) {
1414                                 shmem_unacct_blocks(info->flags, 1);
1415                                 shmem_free_blocks(inode, 1);
1416                                 error = -ENOMEM;
1417                                 goto failed;
1418                         }
1419                         SetPageSwapBacked(filepage);
1420
1421                         /* Precharge page while we can wait, compensate after */
1422                         error = mem_cgroup_cache_charge(filepage, current->mm,
1423                                         GFP_KERNEL);
1424                         if (error) {
1425                                 page_cache_release(filepage);
1426                                 shmem_unacct_blocks(info->flags, 1);
1427                                 shmem_free_blocks(inode, 1);
1428                                 filepage = NULL;
1429                                 goto failed;
1430                         }
1431
1432                         spin_lock(&info->lock);
1433                         entry = shmem_swp_alloc(info, idx, sgp);
1434                         if (IS_ERR(entry))
1435                                 error = PTR_ERR(entry);
1436                         else {
1437                                 swap = *entry;
1438                                 shmem_swp_unmap(entry);
1439                         }
1440                         ret = error || swap.val;
1441                         if (ret)
1442                                 mem_cgroup_uncharge_cache_page(filepage);
1443                         else
1444                                 ret = add_to_page_cache_lru(filepage, mapping,
1445                                                 idx, GFP_NOWAIT);
1446                         /*
1447                          * At add_to_page_cache_lru() failure, uncharge will
1448                          * be done automatically.
1449                          */
1450                         if (ret) {
1451                                 spin_unlock(&info->lock);
1452                                 page_cache_release(filepage);
1453                                 shmem_unacct_blocks(info->flags, 1);
1454                                 shmem_free_blocks(inode, 1);
1455                                 filepage = NULL;
1456                                 if (error)
1457                                         goto failed;
1458                                 goto repeat;
1459                         }
1460                         info->flags |= SHMEM_PAGEIN;
1461                 }
1462
1463                 info->alloced++;
1464                 spin_unlock(&info->lock);
1465                 clear_highpage(filepage);
1466                 flush_dcache_page(filepage);
1467                 SetPageUptodate(filepage);
1468                 if (sgp == SGP_DIRTY)
1469                         set_page_dirty(filepage);
1470         }
1471 done:
1472         *pagep = filepage;
1473         return 0;
1474
1475 failed:
1476         if (*pagep != filepage) {
1477                 unlock_page(filepage);
1478                 page_cache_release(filepage);
1479         }
1480         return error;
1481 }
1482
1483 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1484 {
1485         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1486         int error;
1487         int ret;
1488
1489         if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1490                 return VM_FAULT_SIGBUS;
1491
1492         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1493         if (error)
1494                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1495
1496         return ret | VM_FAULT_LOCKED;
1497 }
1498
1499 #ifdef CONFIG_NUMA
1500 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1501 {
1502         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1503         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1504 }
1505
1506 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1507                                           unsigned long addr)
1508 {
1509         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1510         unsigned long idx;
1511
1512         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1513         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1514 }
1515 #endif
1516
1517 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1518 {
1519         struct inode *inode = file->f_path.dentry->d_inode;
1520         struct shmem_inode_info *info = SHMEM_I(inode);
1521         int retval = -ENOMEM;
1522
1523         spin_lock(&info->lock);
1524         if (lock && !(info->flags & VM_LOCKED)) {
1525                 if (!user_shm_lock(inode->i_size, user))
1526                         goto out_nomem;
1527                 info->flags |= VM_LOCKED;
1528                 mapping_set_unevictable(file->f_mapping);
1529         }
1530         if (!lock && (info->flags & VM_LOCKED) && user) {
1531                 user_shm_unlock(inode->i_size, user);
1532                 info->flags &= ~VM_LOCKED;
1533                 mapping_clear_unevictable(file->f_mapping);
1534                 scan_mapping_unevictable_pages(file->f_mapping);
1535         }
1536         retval = 0;
1537
1538 out_nomem:
1539         spin_unlock(&info->lock);
1540         return retval;
1541 }
1542
1543 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1544 {
1545         file_accessed(file);
1546         vma->vm_ops = &shmem_vm_ops;
1547         vma->vm_flags |= VM_CAN_NONLINEAR;
1548         return 0;
1549 }
1550
1551 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1552                                      int mode, dev_t dev, unsigned long flags)
1553 {
1554         struct inode *inode;
1555         struct shmem_inode_info *info;
1556         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1557
1558         if (shmem_reserve_inode(sb))
1559                 return NULL;
1560
1561         inode = new_inode(sb);
1562         if (inode) {
1563                 inode_init_owner(inode, dir, mode);
1564                 inode->i_blocks = 0;
1565                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1566                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1567                 inode->i_generation = get_seconds();
1568                 info = SHMEM_I(inode);
1569                 memset(info, 0, (char *)inode - (char *)info);
1570                 spin_lock_init(&info->lock);
1571                 info->flags = flags & VM_NORESERVE;
1572                 INIT_LIST_HEAD(&info->swaplist);
1573                 cache_no_acl(inode);
1574
1575                 switch (mode & S_IFMT) {
1576                 default:
1577                         inode->i_op = &shmem_special_inode_operations;
1578                         init_special_inode(inode, mode, dev);
1579                         break;
1580                 case S_IFREG:
1581                         inode->i_mapping->a_ops = &shmem_aops;
1582                         inode->i_op = &shmem_inode_operations;
1583                         inode->i_fop = &shmem_file_operations;
1584                         mpol_shared_policy_init(&info->policy,
1585                                                  shmem_get_sbmpol(sbinfo));
1586                         break;
1587                 case S_IFDIR:
1588                         inc_nlink(inode);
1589                         /* Some things misbehave if size == 0 on a directory */
1590                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1591                         inode->i_op = &shmem_dir_inode_operations;
1592                         inode->i_fop = &simple_dir_operations;
1593                         break;
1594                 case S_IFLNK:
1595                         /*
1596                          * Must not load anything in the rbtree,
1597                          * mpol_free_shared_policy will not be called.
1598                          */
1599                         mpol_shared_policy_init(&info->policy, NULL);
1600                         break;
1601                 }
1602         } else
1603                 shmem_free_inode(sb);
1604         return inode;
1605 }
1606
1607 #ifdef CONFIG_TMPFS
1608 static const struct inode_operations shmem_symlink_inode_operations;
1609 static const struct inode_operations shmem_symlink_inline_operations;
1610
1611 /*
1612  * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1613  * but providing them allows a tmpfs file to be used for splice, sendfile, and
1614  * below the loop driver, in the generic fashion that many filesystems support.
1615  */
1616 static int shmem_readpage(struct file *file, struct page *page)
1617 {
1618         struct inode *inode = page->mapping->host;
1619         int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1620         unlock_page(page);
1621         return error;
1622 }
1623
1624 static int
1625 shmem_write_begin(struct file *file, struct address_space *mapping,
1626                         loff_t pos, unsigned len, unsigned flags,
1627                         struct page **pagep, void **fsdata)
1628 {
1629         struct inode *inode = mapping->host;
1630         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1631         *pagep = NULL;
1632         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1633 }
1634
1635 static int
1636 shmem_write_end(struct file *file, struct address_space *mapping,
1637                         loff_t pos, unsigned len, unsigned copied,
1638                         struct page *page, void *fsdata)
1639 {
1640         struct inode *inode = mapping->host;
1641
1642         if (pos + copied > inode->i_size)
1643                 i_size_write(inode, pos + copied);
1644
1645         set_page_dirty(page);
1646         unlock_page(page);
1647         page_cache_release(page);
1648
1649         return copied;
1650 }
1651
1652 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1653 {
1654         struct inode *inode = filp->f_path.dentry->d_inode;
1655         struct address_space *mapping = inode->i_mapping;
1656         unsigned long index, offset;
1657         enum sgp_type sgp = SGP_READ;
1658
1659         /*
1660          * Might this read be for a stacking filesystem?  Then when reading
1661          * holes of a sparse file, we actually need to allocate those pages,
1662          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1663          */
1664         if (segment_eq(get_fs(), KERNEL_DS))
1665                 sgp = SGP_DIRTY;
1666
1667         index = *ppos >> PAGE_CACHE_SHIFT;
1668         offset = *ppos & ~PAGE_CACHE_MASK;
1669
1670         for (;;) {
1671                 struct page *page = NULL;
1672                 unsigned long end_index, nr, ret;
1673                 loff_t i_size = i_size_read(inode);
1674
1675                 end_index = i_size >> PAGE_CACHE_SHIFT;
1676                 if (index > end_index)
1677                         break;
1678                 if (index == end_index) {
1679                         nr = i_size & ~PAGE_CACHE_MASK;
1680                         if (nr <= offset)
1681                                 break;
1682                 }
1683
1684                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1685                 if (desc->error) {
1686                         if (desc->error == -EINVAL)
1687                                 desc->error = 0;
1688                         break;
1689                 }
1690                 if (page)
1691                         unlock_page(page);
1692
1693                 /*
1694                  * We must evaluate after, since reads (unlike writes)
1695                  * are called without i_mutex protection against truncate
1696                  */
1697                 nr = PAGE_CACHE_SIZE;
1698                 i_size = i_size_read(inode);
1699                 end_index = i_size >> PAGE_CACHE_SHIFT;
1700                 if (index == end_index) {
1701                         nr = i_size & ~PAGE_CACHE_MASK;
1702                         if (nr <= offset) {
1703                                 if (page)
1704                                         page_cache_release(page);
1705                                 break;
1706                         }
1707                 }
1708                 nr -= offset;
1709
1710                 if (page) {
1711                         /*
1712                          * If users can be writing to this page using arbitrary
1713                          * virtual addresses, take care about potential aliasing
1714                          * before reading the page on the kernel side.
1715                          */
1716                         if (mapping_writably_mapped(mapping))
1717                                 flush_dcache_page(page);
1718                         /*
1719                          * Mark the page accessed if we read the beginning.
1720                          */
1721                         if (!offset)
1722                                 mark_page_accessed(page);
1723                 } else {
1724                         page = ZERO_PAGE(0);
1725                         page_cache_get(page);
1726                 }
1727
1728                 /*
1729                  * Ok, we have the page, and it's up-to-date, so
1730                  * now we can copy it to user space...
1731                  *
1732                  * The actor routine returns how many bytes were actually used..
1733                  * NOTE! This may not be the same as how much of a user buffer
1734                  * we filled up (we may be padding etc), so we can only update
1735                  * "pos" here (the actor routine has to update the user buffer
1736                  * pointers and the remaining count).
1737                  */
1738                 ret = actor(desc, page, offset, nr);
1739                 offset += ret;
1740                 index += offset >> PAGE_CACHE_SHIFT;
1741                 offset &= ~PAGE_CACHE_MASK;
1742
1743                 page_cache_release(page);
1744                 if (ret != nr || !desc->count)
1745                         break;
1746
1747                 cond_resched();
1748         }
1749
1750         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1751         file_accessed(filp);
1752 }
1753
1754 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1755                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1756 {
1757         struct file *filp = iocb->ki_filp;
1758         ssize_t retval;
1759         unsigned long seg;
1760         size_t count;
1761         loff_t *ppos = &iocb->ki_pos;
1762
1763         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1764         if (retval)
1765                 return retval;
1766
1767         for (seg = 0; seg < nr_segs; seg++) {
1768                 read_descriptor_t desc;
1769
1770                 desc.written = 0;
1771                 desc.arg.buf = iov[seg].iov_base;
1772                 desc.count = iov[seg].iov_len;
1773                 if (desc.count == 0)
1774                         continue;
1775                 desc.error = 0;
1776                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1777                 retval += desc.written;
1778                 if (desc.error) {
1779                         retval = retval ?: desc.error;
1780                         break;
1781                 }
1782                 if (desc.count > 0)
1783                         break;
1784         }
1785         return retval;
1786 }
1787
1788 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1789 {
1790         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1791
1792         buf->f_type = TMPFS_MAGIC;
1793         buf->f_bsize = PAGE_CACHE_SIZE;
1794         buf->f_namelen = NAME_MAX;
1795         spin_lock(&sbinfo->stat_lock);
1796         if (sbinfo->max_blocks) {
1797                 buf->f_blocks = sbinfo->max_blocks;
1798                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1799         }
1800         if (sbinfo->max_inodes) {
1801                 buf->f_files = sbinfo->max_inodes;
1802                 buf->f_ffree = sbinfo->free_inodes;
1803         }
1804         /* else leave those fields 0 like simple_statfs */
1805         spin_unlock(&sbinfo->stat_lock);
1806         return 0;
1807 }
1808
1809 /*
1810  * File creation. Allocate an inode, and we're done..
1811  */
1812 static int
1813 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1814 {
1815         struct inode *inode;
1816         int error = -ENOSPC;
1817
1818         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1819         if (inode) {
1820                 error = security_inode_init_security(inode, dir, NULL, NULL,
1821                                                      NULL);
1822                 if (error) {
1823                         if (error != -EOPNOTSUPP) {
1824                                 iput(inode);
1825                                 return error;
1826                         }
1827                 }
1828 #ifdef CONFIG_TMPFS_POSIX_ACL
1829                 error = generic_acl_init(inode, dir);
1830                 if (error) {
1831                         iput(inode);
1832                         return error;
1833                 }
1834 #else
1835                 error = 0;
1836 #endif
1837                 dir->i_size += BOGO_DIRENT_SIZE;
1838                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1839                 d_instantiate(dentry, inode);
1840                 dget(dentry); /* Extra count - pin the dentry in core */
1841         }
1842         return error;
1843 }
1844
1845 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1846 {
1847         int error;
1848
1849         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1850                 return error;
1851         inc_nlink(dir);
1852         return 0;
1853 }
1854
1855 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1856                 struct nameidata *nd)
1857 {
1858         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1859 }
1860
1861 /*
1862  * Link a file..
1863  */
1864 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1865 {
1866         struct inode *inode = old_dentry->d_inode;
1867         int ret;
1868
1869         /*
1870          * No ordinary (disk based) filesystem counts links as inodes;
1871          * but each new link needs a new dentry, pinning lowmem, and
1872          * tmpfs dentries cannot be pruned until they are unlinked.
1873          */
1874         ret = shmem_reserve_inode(inode->i_sb);
1875         if (ret)
1876                 goto out;
1877
1878         dir->i_size += BOGO_DIRENT_SIZE;
1879         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1880         inc_nlink(inode);
1881         atomic_inc(&inode->i_count);    /* New dentry reference */
1882         dget(dentry);           /* Extra pinning count for the created dentry */
1883         d_instantiate(dentry, inode);
1884 out:
1885         return ret;
1886 }
1887
1888 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1889 {
1890         struct inode *inode = dentry->d_inode;
1891
1892         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1893                 shmem_free_inode(inode->i_sb);
1894
1895         dir->i_size -= BOGO_DIRENT_SIZE;
1896         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1897         drop_nlink(inode);
1898         dput(dentry);   /* Undo the count from "create" - this does all the work */
1899         return 0;
1900 }
1901
1902 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1903 {
1904         if (!simple_empty(dentry))
1905                 return -ENOTEMPTY;
1906
1907         drop_nlink(dentry->d_inode);
1908         drop_nlink(dir);
1909         return shmem_unlink(dir, dentry);
1910 }
1911
1912 /*
1913  * The VFS layer already does all the dentry stuff for rename,
1914  * we just have to decrement the usage count for the target if
1915  * it exists so that the VFS layer correctly free's it when it
1916  * gets overwritten.
1917  */
1918 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1919 {
1920         struct inode *inode = old_dentry->d_inode;
1921         int they_are_dirs = S_ISDIR(inode->i_mode);
1922
1923         if (!simple_empty(new_dentry))
1924                 return -ENOTEMPTY;
1925
1926         if (new_dentry->d_inode) {
1927                 (void) shmem_unlink(new_dir, new_dentry);
1928                 if (they_are_dirs)
1929                         drop_nlink(old_dir);
1930         } else if (they_are_dirs) {
1931                 drop_nlink(old_dir);
1932                 inc_nlink(new_dir);
1933         }
1934
1935         old_dir->i_size -= BOGO_DIRENT_SIZE;
1936         new_dir->i_size += BOGO_DIRENT_SIZE;
1937         old_dir->i_ctime = old_dir->i_mtime =
1938         new_dir->i_ctime = new_dir->i_mtime =
1939         inode->i_ctime = CURRENT_TIME;
1940         return 0;
1941 }
1942
1943 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1944 {
1945         int error;
1946         int len;
1947         struct inode *inode;
1948         struct page *page = NULL;
1949         char *kaddr;
1950         struct shmem_inode_info *info;
1951
1952         len = strlen(symname) + 1;
1953         if (len > PAGE_CACHE_SIZE)
1954                 return -ENAMETOOLONG;
1955
1956         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1957         if (!inode)
1958                 return -ENOSPC;
1959
1960         error = security_inode_init_security(inode, dir, NULL, NULL,
1961                                              NULL);
1962         if (error) {
1963                 if (error != -EOPNOTSUPP) {
1964                         iput(inode);
1965                         return error;
1966                 }
1967                 error = 0;
1968         }
1969
1970         info = SHMEM_I(inode);
1971         inode->i_size = len-1;
1972         if (len <= (char *)inode - (char *)info) {
1973                 /* do it inline */
1974                 memcpy(info, symname, len);
1975                 inode->i_op = &shmem_symlink_inline_operations;
1976         } else {
1977                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1978                 if (error) {
1979                         iput(inode);
1980                         return error;
1981                 }
1982                 inode->i_mapping->a_ops = &shmem_aops;
1983                 inode->i_op = &shmem_symlink_inode_operations;
1984                 kaddr = kmap_atomic(page, KM_USER0);
1985                 memcpy(kaddr, symname, len);
1986                 kunmap_atomic(kaddr, KM_USER0);
1987                 set_page_dirty(page);
1988                 unlock_page(page);
1989                 page_cache_release(page);
1990         }
1991         dir->i_size += BOGO_DIRENT_SIZE;
1992         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1993         d_instantiate(dentry, inode);
1994         dget(dentry);
1995         return 0;
1996 }
1997
1998 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1999 {
2000         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
2001         return NULL;
2002 }
2003
2004 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2005 {
2006         struct page *page = NULL;
2007         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2008         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2009         if (page)
2010                 unlock_page(page);
2011         return page;
2012 }
2013
2014 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2015 {
2016         if (!IS_ERR(nd_get_link(nd))) {
2017                 struct page *page = cookie;
2018                 kunmap(page);
2019                 mark_page_accessed(page);
2020                 page_cache_release(page);
2021         }
2022 }
2023
2024 static const struct inode_operations shmem_symlink_inline_operations = {
2025         .readlink       = generic_readlink,
2026         .follow_link    = shmem_follow_link_inline,
2027 };
2028
2029 static const struct inode_operations shmem_symlink_inode_operations = {
2030         .readlink       = generic_readlink,
2031         .follow_link    = shmem_follow_link,
2032         .put_link       = shmem_put_link,
2033 };
2034
2035 #ifdef CONFIG_TMPFS_POSIX_ACL
2036 /*
2037  * Superblocks without xattr inode operations will get security.* xattr
2038  * support from the VFS "for free". As soon as we have any other xattrs
2039  * like ACLs, we also need to implement the security.* handlers at
2040  * filesystem level, though.
2041  */
2042
2043 static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
2044                                         size_t list_len, const char *name,
2045                                         size_t name_len, int handler_flags)
2046 {
2047         return security_inode_listsecurity(dentry->d_inode, list, list_len);
2048 }
2049
2050 static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2051                 void *buffer, size_t size, int handler_flags)
2052 {
2053         if (strcmp(name, "") == 0)
2054                 return -EINVAL;
2055         return xattr_getsecurity(dentry->d_inode, name, buffer, size);
2056 }
2057
2058 static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2059                 const void *value, size_t size, int flags, int handler_flags)
2060 {
2061         if (strcmp(name, "") == 0)
2062                 return -EINVAL;
2063         return security_inode_setsecurity(dentry->d_inode, name, value,
2064                                           size, flags);
2065 }
2066
2067 static const struct xattr_handler shmem_xattr_security_handler = {
2068         .prefix = XATTR_SECURITY_PREFIX,
2069         .list   = shmem_xattr_security_list,
2070         .get    = shmem_xattr_security_get,
2071         .set    = shmem_xattr_security_set,
2072 };
2073
2074 static const struct xattr_handler *shmem_xattr_handlers[] = {
2075         &generic_acl_access_handler,
2076         &generic_acl_default_handler,
2077         &shmem_xattr_security_handler,
2078         NULL
2079 };
2080 #endif
2081
2082 static struct dentry *shmem_get_parent(struct dentry *child)
2083 {
2084         return ERR_PTR(-ESTALE);
2085 }
2086
2087 static int shmem_match(struct inode *ino, void *vfh)
2088 {
2089         __u32 *fh = vfh;
2090         __u64 inum = fh[2];
2091         inum = (inum << 32) | fh[1];
2092         return ino->i_ino == inum && fh[0] == ino->i_generation;
2093 }
2094
2095 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2096                 struct fid *fid, int fh_len, int fh_type)
2097 {
2098         struct inode *inode;
2099         struct dentry *dentry = NULL;
2100         u64 inum = fid->raw[2];
2101         inum = (inum << 32) | fid->raw[1];
2102
2103         if (fh_len < 3)
2104                 return NULL;
2105
2106         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2107                         shmem_match, fid->raw);
2108         if (inode) {
2109                 dentry = d_find_alias(inode);
2110                 iput(inode);
2111         }
2112
2113         return dentry;
2114 }
2115
2116 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2117                                 int connectable)
2118 {
2119         struct inode *inode = dentry->d_inode;
2120
2121         if (*len < 3)
2122                 return 255;
2123
2124         if (hlist_unhashed(&inode->i_hash)) {
2125                 /* Unfortunately insert_inode_hash is not idempotent,
2126                  * so as we hash inodes here rather than at creation
2127                  * time, we need a lock to ensure we only try
2128                  * to do it once
2129                  */
2130                 static DEFINE_SPINLOCK(lock);
2131                 spin_lock(&lock);
2132                 if (hlist_unhashed(&inode->i_hash))
2133                         __insert_inode_hash(inode,
2134                                             inode->i_ino + inode->i_generation);
2135                 spin_unlock(&lock);
2136         }
2137
2138         fh[0] = inode->i_generation;
2139         fh[1] = inode->i_ino;
2140         fh[2] = ((__u64)inode->i_ino) >> 32;
2141
2142         *len = 3;
2143         return 1;
2144 }
2145
2146 static const struct export_operations shmem_export_ops = {
2147         .get_parent     = shmem_get_parent,
2148         .encode_fh      = shmem_encode_fh,
2149         .fh_to_dentry   = shmem_fh_to_dentry,
2150 };
2151
2152 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2153                                bool remount)
2154 {
2155         char *this_char, *value, *rest;
2156
2157         while (options != NULL) {
2158                 this_char = options;
2159                 for (;;) {
2160                         /*
2161                          * NUL-terminate this option: unfortunately,
2162                          * mount options form a comma-separated list,
2163                          * but mpol's nodelist may also contain commas.
2164                          */
2165                         options = strchr(options, ',');
2166                         if (options == NULL)
2167                                 break;
2168                         options++;
2169                         if (!isdigit(*options)) {
2170                                 options[-1] = '\0';
2171                                 break;
2172                         }
2173                 }
2174                 if (!*this_char)
2175                         continue;
2176                 if ((value = strchr(this_char,'=')) != NULL) {
2177                         *value++ = 0;
2178                 } else {
2179                         printk(KERN_ERR
2180                             "tmpfs: No value for mount option '%s'\n",
2181                             this_char);
2182                         return 1;
2183                 }
2184
2185                 if (!strcmp(this_char,"size")) {
2186                         unsigned long long size;
2187                         size = memparse(value,&rest);
2188                         if (*rest == '%') {
2189                                 size <<= PAGE_SHIFT;
2190                                 size *= totalram_pages;
2191                                 do_div(size, 100);
2192                                 rest++;
2193                         }
2194                         if (*rest)
2195                                 goto bad_val;
2196                         sbinfo->max_blocks =
2197                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2198                 } else if (!strcmp(this_char,"nr_blocks")) {
2199                         sbinfo->max_blocks = memparse(value, &rest);
2200                         if (*rest)
2201                                 goto bad_val;
2202                 } else if (!strcmp(this_char,"nr_inodes")) {
2203                         sbinfo->max_inodes = memparse(value, &rest);
2204                         if (*rest)
2205                                 goto bad_val;
2206                 } else if (!strcmp(this_char,"mode")) {
2207                         if (remount)
2208                                 continue;
2209                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2210                         if (*rest)
2211                                 goto bad_val;
2212                 } else if (!strcmp(this_char,"uid")) {
2213                         if (remount)
2214                                 continue;
2215                         sbinfo->uid = simple_strtoul(value, &rest, 0);
2216                         if (*rest)
2217                                 goto bad_val;
2218                 } else if (!strcmp(this_char,"gid")) {
2219                         if (remount)
2220                                 continue;
2221                         sbinfo->gid = simple_strtoul(value, &rest, 0);
2222                         if (*rest)
2223                                 goto bad_val;
2224                 } else if (!strcmp(this_char,"mpol")) {
2225                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2226                                 goto bad_val;
2227                 } else {
2228                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2229                                this_char);
2230                         return 1;
2231                 }
2232         }
2233         return 0;
2234
2235 bad_val:
2236         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2237                value, this_char);
2238         return 1;
2239
2240 }
2241
2242 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2243 {
2244         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2245         struct shmem_sb_info config = *sbinfo;
2246         unsigned long blocks;
2247         unsigned long inodes;
2248         int error = -EINVAL;
2249
2250         if (shmem_parse_options(data, &config, true))
2251                 return error;
2252
2253         spin_lock(&sbinfo->stat_lock);
2254         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2255         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2256         if (config.max_blocks < blocks)
2257                 goto out;
2258         if (config.max_inodes < inodes)
2259                 goto out;
2260         /*
2261          * Those tests also disallow limited->unlimited while any are in
2262          * use, so i_blocks will always be zero when max_blocks is zero;
2263          * but we must separately disallow unlimited->limited, because
2264          * in that case we have no record of how much is already in use.
2265          */
2266         if (config.max_blocks && !sbinfo->max_blocks)
2267                 goto out;
2268         if (config.max_inodes && !sbinfo->max_inodes)
2269                 goto out;
2270
2271         error = 0;
2272         sbinfo->max_blocks  = config.max_blocks;
2273         sbinfo->free_blocks = config.max_blocks - blocks;
2274         sbinfo->max_inodes  = config.max_inodes;
2275         sbinfo->free_inodes = config.max_inodes - inodes;
2276
2277         mpol_put(sbinfo->mpol);
2278         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2279 out:
2280         spin_unlock(&sbinfo->stat_lock);
2281         return error;
2282 }
2283
2284 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2285 {
2286         struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2287
2288         if (sbinfo->max_blocks != shmem_default_max_blocks())
2289                 seq_printf(seq, ",size=%luk",
2290                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2291         if (sbinfo->max_inodes != shmem_default_max_inodes())
2292                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2293         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2294                 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2295         if (sbinfo->uid != 0)
2296                 seq_printf(seq, ",uid=%u", sbinfo->uid);
2297         if (sbinfo->gid != 0)
2298                 seq_printf(seq, ",gid=%u", sbinfo->gid);
2299         shmem_show_mpol(seq, sbinfo->mpol);
2300         return 0;
2301 }
2302 #endif /* CONFIG_TMPFS */
2303
2304 static void shmem_put_super(struct super_block *sb)
2305 {
2306         kfree(sb->s_fs_info);
2307         sb->s_fs_info = NULL;
2308 }
2309
2310 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2311 {
2312         struct inode *inode;
2313         struct dentry *root;
2314         struct shmem_sb_info *sbinfo;
2315         int err = -ENOMEM;
2316
2317         /* Round up to L1_CACHE_BYTES to resist false sharing */
2318         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2319                                 L1_CACHE_BYTES), GFP_KERNEL);
2320         if (!sbinfo)
2321                 return -ENOMEM;
2322
2323         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2324         sbinfo->uid = current_fsuid();
2325         sbinfo->gid = current_fsgid();
2326         sb->s_fs_info = sbinfo;
2327
2328 #ifdef CONFIG_TMPFS
2329         /*
2330          * Per default we only allow half of the physical ram per
2331          * tmpfs instance, limiting inodes to one per page of lowmem;
2332          * but the internal instance is left unlimited.
2333          */
2334         if (!(sb->s_flags & MS_NOUSER)) {
2335                 sbinfo->max_blocks = shmem_default_max_blocks();
2336                 sbinfo->max_inodes = shmem_default_max_inodes();
2337                 if (shmem_parse_options(data, sbinfo, false)) {
2338                         err = -EINVAL;
2339                         goto failed;
2340                 }
2341         }
2342         sb->s_export_op = &shmem_export_ops;
2343 #else
2344         sb->s_flags |= MS_NOUSER;
2345 #endif
2346
2347         spin_lock_init(&sbinfo->stat_lock);
2348         sbinfo->free_blocks = sbinfo->max_blocks;
2349         sbinfo->free_inodes = sbinfo->max_inodes;
2350
2351         sb->s_maxbytes = SHMEM_MAX_BYTES;
2352         sb->s_blocksize = PAGE_CACHE_SIZE;
2353         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2354         sb->s_magic = TMPFS_MAGIC;
2355         sb->s_op = &shmem_ops;
2356         sb->s_time_gran = 1;
2357 #ifdef CONFIG_TMPFS_POSIX_ACL
2358         sb->s_xattr = shmem_xattr_handlers;
2359         sb->s_flags |= MS_POSIXACL;
2360 #endif
2361
2362         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2363         if (!inode)
2364                 goto failed;
2365         inode->i_uid = sbinfo->uid;
2366         inode->i_gid = sbinfo->gid;
2367         root = d_alloc_root(inode);
2368         if (!root)
2369                 goto failed_iput;
2370         sb->s_root = root;
2371         return 0;
2372
2373 failed_iput:
2374         iput(inode);
2375 failed:
2376         shmem_put_super(sb);
2377         return err;
2378 }
2379
2380 static struct kmem_cache *shmem_inode_cachep;
2381
2382 static struct inode *shmem_alloc_inode(struct super_block *sb)
2383 {
2384         struct shmem_inode_info *p;
2385         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2386         if (!p)
2387                 return NULL;
2388         return &p->vfs_inode;
2389 }
2390
2391 static void shmem_destroy_inode(struct inode *inode)
2392 {
2393         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2394                 /* only struct inode is valid if it's an inline symlink */
2395                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2396         }
2397         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2398 }
2399
2400 static void init_once(void *foo)
2401 {
2402         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2403
2404         inode_init_once(&p->vfs_inode);
2405 }
2406
2407 static int init_inodecache(void)
2408 {
2409         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2410                                 sizeof(struct shmem_inode_info),
2411                                 0, SLAB_PANIC, init_once);
2412         return 0;
2413 }
2414
2415 static void destroy_inodecache(void)
2416 {
2417         kmem_cache_destroy(shmem_inode_cachep);
2418 }
2419
2420 static const struct address_space_operations shmem_aops = {
2421         .writepage      = shmem_writepage,
2422         .set_page_dirty = __set_page_dirty_no_writeback,
2423 #ifdef CONFIG_TMPFS
2424         .readpage       = shmem_readpage,
2425         .write_begin    = shmem_write_begin,
2426         .write_end      = shmem_write_end,
2427 #endif
2428         .migratepage    = migrate_page,
2429         .error_remove_page = generic_error_remove_page,
2430 };
2431
2432 static const struct file_operations shmem_file_operations = {
2433         .mmap           = shmem_mmap,
2434 #ifdef CONFIG_TMPFS
2435         .llseek         = generic_file_llseek,
2436         .read           = do_sync_read,
2437         .write          = do_sync_write,
2438         .aio_read       = shmem_file_aio_read,
2439         .aio_write      = generic_file_aio_write,
2440         .fsync          = noop_fsync,
2441         .splice_read    = generic_file_splice_read,
2442         .splice_write   = generic_file_splice_write,
2443 #endif
2444 };
2445
2446 static const struct inode_operations shmem_inode_operations = {
2447         .setattr        = shmem_notify_change,
2448         .truncate_range = shmem_truncate_range,
2449 #ifdef CONFIG_TMPFS_POSIX_ACL
2450         .setxattr       = generic_setxattr,
2451         .getxattr       = generic_getxattr,
2452         .listxattr      = generic_listxattr,
2453         .removexattr    = generic_removexattr,
2454         .check_acl      = generic_check_acl,
2455 #endif
2456
2457 };
2458
2459 static const struct inode_operations shmem_dir_inode_operations = {
2460 #ifdef CONFIG_TMPFS
2461         .create         = shmem_create,
2462         .lookup         = simple_lookup,
2463         .link           = shmem_link,
2464         .unlink         = shmem_unlink,
2465         .symlink        = shmem_symlink,
2466         .mkdir          = shmem_mkdir,
2467         .rmdir          = shmem_rmdir,
2468         .mknod          = shmem_mknod,
2469         .rename         = shmem_rename,
2470 #endif
2471 #ifdef CONFIG_TMPFS_POSIX_ACL
2472         .setattr        = shmem_notify_change,
2473         .setxattr       = generic_setxattr,
2474         .getxattr       = generic_getxattr,
2475         .listxattr      = generic_listxattr,
2476         .removexattr    = generic_removexattr,
2477         .check_acl      = generic_check_acl,
2478 #endif
2479 };
2480
2481 static const struct inode_operations shmem_special_inode_operations = {
2482 #ifdef CONFIG_TMPFS_POSIX_ACL
2483         .setattr        = shmem_notify_change,
2484         .setxattr       = generic_setxattr,
2485         .getxattr       = generic_getxattr,
2486         .listxattr      = generic_listxattr,
2487         .removexattr    = generic_removexattr,
2488         .check_acl      = generic_check_acl,
2489 #endif
2490 };
2491
2492 static const struct super_operations shmem_ops = {
2493         .alloc_inode    = shmem_alloc_inode,
2494         .destroy_inode  = shmem_destroy_inode,
2495 #ifdef CONFIG_TMPFS
2496         .statfs         = shmem_statfs,
2497         .remount_fs     = shmem_remount_fs,
2498         .show_options   = shmem_show_options,
2499 #endif
2500         .delete_inode   = shmem_delete_inode,
2501         .drop_inode     = generic_delete_inode,
2502         .put_super      = shmem_put_super,
2503 };
2504
2505 static const struct vm_operations_struct shmem_vm_ops = {
2506         .fault          = shmem_fault,
2507 #ifdef CONFIG_NUMA
2508         .set_policy     = shmem_set_policy,
2509         .get_policy     = shmem_get_policy,
2510 #endif
2511 };
2512
2513
2514 static int shmem_get_sb(struct file_system_type *fs_type,
2515         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2516 {
2517         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2518 }
2519
2520 static struct file_system_type tmpfs_fs_type = {
2521         .owner          = THIS_MODULE,
2522         .name           = "tmpfs",
2523         .get_sb         = shmem_get_sb,
2524         .kill_sb        = kill_litter_super,
2525 };
2526
2527 int __init init_tmpfs(void)
2528 {
2529         int error;
2530
2531         error = bdi_init(&shmem_backing_dev_info);
2532         if (error)
2533                 goto out4;
2534
2535         error = init_inodecache();
2536         if (error)
2537                 goto out3;
2538
2539         error = register_filesystem(&tmpfs_fs_type);
2540         if (error) {
2541                 printk(KERN_ERR "Could not register tmpfs\n");
2542                 goto out2;
2543         }
2544
2545         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2546                                 tmpfs_fs_type.name, NULL);
2547         if (IS_ERR(shm_mnt)) {
2548                 error = PTR_ERR(shm_mnt);
2549                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2550                 goto out1;
2551         }
2552         return 0;
2553
2554 out1:
2555         unregister_filesystem(&tmpfs_fs_type);
2556 out2:
2557         destroy_inodecache();
2558 out3:
2559         bdi_destroy(&shmem_backing_dev_info);
2560 out4:
2561         shm_mnt = ERR_PTR(error);
2562         return error;
2563 }
2564
2565 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2566 /**
2567  * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2568  * @inode: the inode to be searched
2569  * @pgoff: the offset to be searched
2570  * @pagep: the pointer for the found page to be stored
2571  * @ent: the pointer for the found swap entry to be stored
2572  *
2573  * If a page is found, refcount of it is incremented. Callers should handle
2574  * these refcount.
2575  */
2576 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2577                                         struct page **pagep, swp_entry_t *ent)
2578 {
2579         swp_entry_t entry = { .val = 0 }, *ptr;
2580         struct page *page = NULL;
2581         struct shmem_inode_info *info = SHMEM_I(inode);
2582
2583         if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2584                 goto out;
2585
2586         spin_lock(&info->lock);
2587         ptr = shmem_swp_entry(info, pgoff, NULL);
2588 #ifdef CONFIG_SWAP
2589         if (ptr && ptr->val) {
2590                 entry.val = ptr->val;
2591                 page = find_get_page(&swapper_space, entry.val);
2592         } else
2593 #endif
2594                 page = find_get_page(inode->i_mapping, pgoff);
2595         if (ptr)
2596                 shmem_swp_unmap(ptr);
2597         spin_unlock(&info->lock);
2598 out:
2599         *pagep = page;
2600         *ent = entry;
2601 }
2602 #endif
2603
2604 #else /* !CONFIG_SHMEM */
2605
2606 /*
2607  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2608  *
2609  * This is intended for small system where the benefits of the full
2610  * shmem code (swap-backed and resource-limited) are outweighed by
2611  * their complexity. On systems without swap this code should be
2612  * effectively equivalent, but much lighter weight.
2613  */
2614
2615 #include <linux/ramfs.h>
2616
2617 static struct file_system_type tmpfs_fs_type = {
2618         .name           = "tmpfs",
2619         .get_sb         = ramfs_get_sb,
2620         .kill_sb        = kill_litter_super,
2621 };
2622
2623 int __init init_tmpfs(void)
2624 {
2625         BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2626
2627         shm_mnt = kern_mount(&tmpfs_fs_type);
2628         BUG_ON(IS_ERR(shm_mnt));
2629
2630         return 0;
2631 }
2632
2633 int shmem_unuse(swp_entry_t entry, struct page *page)
2634 {
2635         return 0;
2636 }
2637
2638 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2639 {
2640         return 0;
2641 }
2642
2643 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2644 /**
2645  * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2646  * @inode: the inode to be searched
2647  * @pgoff: the offset to be searched
2648  * @pagep: the pointer for the found page to be stored
2649  * @ent: the pointer for the found swap entry to be stored
2650  *
2651  * If a page is found, refcount of it is incremented. Callers should handle
2652  * these refcount.
2653  */
2654 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2655                                         struct page **pagep, swp_entry_t *ent)
2656 {
2657         struct page *page = NULL;
2658
2659         if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2660                 goto out;
2661         page = find_get_page(inode->i_mapping, pgoff);
2662 out:
2663         *pagep = page;
2664         *ent = (swp_entry_t){ .val = 0 };
2665 }
2666 #endif
2667
2668 #define shmem_vm_ops                            generic_file_vm_ops
2669 #define shmem_file_operations                   ramfs_file_operations
2670 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2671 #define shmem_acct_size(flags, size)            0
2672 #define shmem_unacct_size(flags, size)          do {} while (0)
2673 #define SHMEM_MAX_BYTES                         MAX_LFS_FILESIZE
2674
2675 #endif /* CONFIG_SHMEM */
2676
2677 /* common code */
2678
2679 /**
2680  * shmem_file_setup - get an unlinked file living in tmpfs
2681  * @name: name for dentry (to be seen in /proc/<pid>/maps
2682  * @size: size to be set for the file
2683  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2684  */
2685 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2686 {
2687         int error;
2688         struct file *file;
2689         struct inode *inode;
2690         struct path path;
2691         struct dentry *root;
2692         struct qstr this;
2693
2694         if (IS_ERR(shm_mnt))
2695                 return (void *)shm_mnt;
2696
2697         if (size < 0 || size > SHMEM_MAX_BYTES)
2698                 return ERR_PTR(-EINVAL);
2699
2700         if (shmem_acct_size(flags, size))
2701                 return ERR_PTR(-ENOMEM);
2702
2703         error = -ENOMEM;
2704         this.name = name;
2705         this.len = strlen(name);
2706         this.hash = 0; /* will go */
2707         root = shm_mnt->mnt_root;
2708         path.dentry = d_alloc(root, &this);
2709         if (!path.dentry)
2710                 goto put_memory;
2711         path.mnt = mntget(shm_mnt);
2712
2713         error = -ENOSPC;
2714         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2715         if (!inode)
2716                 goto put_dentry;
2717
2718         d_instantiate(path.dentry, inode);
2719         inode->i_size = size;
2720         inode->i_nlink = 0;     /* It is unlinked */
2721 #ifndef CONFIG_MMU
2722         error = ramfs_nommu_expand_for_mapping(inode, size);
2723         if (error)
2724                 goto put_dentry;
2725 #endif
2726
2727         error = -ENFILE;
2728         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2729                   &shmem_file_operations);
2730         if (!file)
2731                 goto put_dentry;
2732
2733         return file;
2734
2735 put_dentry:
2736         path_put(&path);
2737 put_memory:
2738         shmem_unacct_size(flags, size);
2739         return ERR_PTR(error);
2740 }
2741 EXPORT_SYMBOL_GPL(shmem_file_setup);
2742
2743 /**
2744  * shmem_zero_setup - setup a shared anonymous mapping
2745  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2746  */
2747 int shmem_zero_setup(struct vm_area_struct *vma)
2748 {
2749         struct file *file;
2750         loff_t size = vma->vm_end - vma->vm_start;
2751
2752         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2753         if (IS_ERR(file))
2754                 return PTR_ERR(file);
2755
2756         if (vma->vm_file)
2757                 fput(vma->vm_file);
2758         vma->vm_file = file;
2759         vma->vm_ops = &shmem_vm_ops;
2760         return 0;
2761 }