tmpfs: convert to use the new truncate convention
[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         int error;
768
769         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
770                 loff_t newsize = attr->ia_size;
771                 struct page *page = NULL;
772
773                 if (newsize < inode->i_size) {
774                         /*
775                          * If truncating down to a partial page, then
776                          * if that page is already allocated, hold it
777                          * in memory until the truncation is over, so
778                          * truncate_partial_page cannnot miss it were
779                          * it assigned to swap.
780                          */
781                         if (newsize & (PAGE_CACHE_SIZE-1)) {
782                                 (void) shmem_getpage(inode,
783                                         newsize >> PAGE_CACHE_SHIFT,
784                                                 &page, SGP_READ, NULL);
785                                 if (page)
786                                         unlock_page(page);
787                         }
788                         /*
789                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
790                          * detect if any pages might have been added to cache
791                          * after truncate_inode_pages.  But we needn't bother
792                          * if it's being fully truncated to zero-length: the
793                          * nrpages check is efficient enough in that case.
794                          */
795                         if (newsize) {
796                                 struct shmem_inode_info *info = SHMEM_I(inode);
797                                 spin_lock(&info->lock);
798                                 info->flags &= ~SHMEM_PAGEIN;
799                                 spin_unlock(&info->lock);
800                         }
801                 }
802
803                 error = simple_setsize(inode, newsize);
804                 if (page)
805                         page_cache_release(page);
806                 if (error)
807                         return error;
808                 shmem_truncate_range(inode, newsize, (loff_t)-1);
809         }
810
811         error = inode_change_ok(inode, attr);
812         if (!error)
813                 generic_setattr(inode, attr);
814 #ifdef CONFIG_TMPFS_POSIX_ACL
815         if (!error && (attr->ia_valid & ATTR_MODE))
816                 error = generic_acl_chmod(inode);
817 #endif
818         return error;
819 }
820
821 static void shmem_delete_inode(struct inode *inode)
822 {
823         struct shmem_inode_info *info = SHMEM_I(inode);
824
825         if (inode->i_mapping->a_ops == &shmem_aops) {
826                 truncate_inode_pages(inode->i_mapping, 0);
827                 shmem_unacct_size(info->flags, inode->i_size);
828                 inode->i_size = 0;
829                 shmem_truncate_range(inode, 0, (loff_t)-1);
830                 if (!list_empty(&info->swaplist)) {
831                         mutex_lock(&shmem_swaplist_mutex);
832                         list_del_init(&info->swaplist);
833                         mutex_unlock(&shmem_swaplist_mutex);
834                 }
835         }
836         BUG_ON(inode->i_blocks);
837         shmem_free_inode(inode->i_sb);
838         clear_inode(inode);
839 }
840
841 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
842 {
843         swp_entry_t *ptr;
844
845         for (ptr = dir; ptr < edir; ptr++) {
846                 if (ptr->val == entry.val)
847                         return ptr - dir;
848         }
849         return -1;
850 }
851
852 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
853 {
854         struct inode *inode;
855         unsigned long idx;
856         unsigned long size;
857         unsigned long limit;
858         unsigned long stage;
859         struct page **dir;
860         struct page *subdir;
861         swp_entry_t *ptr;
862         int offset;
863         int error;
864
865         idx = 0;
866         ptr = info->i_direct;
867         spin_lock(&info->lock);
868         if (!info->swapped) {
869                 list_del_init(&info->swaplist);
870                 goto lost2;
871         }
872         limit = info->next_index;
873         size = limit;
874         if (size > SHMEM_NR_DIRECT)
875                 size = SHMEM_NR_DIRECT;
876         offset = shmem_find_swp(entry, ptr, ptr+size);
877         if (offset >= 0)
878                 goto found;
879         if (!info->i_indirect)
880                 goto lost2;
881
882         dir = shmem_dir_map(info->i_indirect);
883         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
884
885         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
886                 if (unlikely(idx == stage)) {
887                         shmem_dir_unmap(dir-1);
888                         if (cond_resched_lock(&info->lock)) {
889                                 /* check it has not been truncated */
890                                 if (limit > info->next_index) {
891                                         limit = info->next_index;
892                                         if (idx >= limit)
893                                                 goto lost2;
894                                 }
895                         }
896                         dir = shmem_dir_map(info->i_indirect) +
897                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
898                         while (!*dir) {
899                                 dir++;
900                                 idx += ENTRIES_PER_PAGEPAGE;
901                                 if (idx >= limit)
902                                         goto lost1;
903                         }
904                         stage = idx + ENTRIES_PER_PAGEPAGE;
905                         subdir = *dir;
906                         shmem_dir_unmap(dir);
907                         dir = shmem_dir_map(subdir);
908                 }
909                 subdir = *dir;
910                 if (subdir && page_private(subdir)) {
911                         ptr = shmem_swp_map(subdir);
912                         size = limit - idx;
913                         if (size > ENTRIES_PER_PAGE)
914                                 size = ENTRIES_PER_PAGE;
915                         offset = shmem_find_swp(entry, ptr, ptr+size);
916                         shmem_swp_unmap(ptr);
917                         if (offset >= 0) {
918                                 shmem_dir_unmap(dir);
919                                 goto found;
920                         }
921                 }
922         }
923 lost1:
924         shmem_dir_unmap(dir-1);
925 lost2:
926         spin_unlock(&info->lock);
927         return 0;
928 found:
929         idx += offset;
930         inode = igrab(&info->vfs_inode);
931         spin_unlock(&info->lock);
932
933         /*
934          * Move _head_ to start search for next from here.
935          * But be careful: shmem_delete_inode checks list_empty without taking
936          * mutex, and there's an instant in list_move_tail when info->swaplist
937          * would appear empty, if it were the only one on shmem_swaplist.  We
938          * could avoid doing it if inode NULL; or use this minor optimization.
939          */
940         if (shmem_swaplist.next != &info->swaplist)
941                 list_move_tail(&shmem_swaplist, &info->swaplist);
942         mutex_unlock(&shmem_swaplist_mutex);
943
944         error = 1;
945         if (!inode)
946                 goto out;
947         /*
948          * Charge page using GFP_KERNEL while we can wait.
949          * Charged back to the user(not to caller) when swap account is used.
950          * add_to_page_cache() will be called with GFP_NOWAIT.
951          */
952         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
953         if (error)
954                 goto out;
955         error = radix_tree_preload(GFP_KERNEL);
956         if (error) {
957                 mem_cgroup_uncharge_cache_page(page);
958                 goto out;
959         }
960         error = 1;
961
962         spin_lock(&info->lock);
963         ptr = shmem_swp_entry(info, idx, NULL);
964         if (ptr && ptr->val == entry.val) {
965                 error = add_to_page_cache_locked(page, inode->i_mapping,
966                                                 idx, GFP_NOWAIT);
967                 /* does mem_cgroup_uncharge_cache_page on error */
968         } else  /* we must compensate for our precharge above */
969                 mem_cgroup_uncharge_cache_page(page);
970
971         if (error == -EEXIST) {
972                 struct page *filepage = find_get_page(inode->i_mapping, idx);
973                 error = 1;
974                 if (filepage) {
975                         /*
976                          * There might be a more uptodate page coming down
977                          * from a stacked writepage: forget our swappage if so.
978                          */
979                         if (PageUptodate(filepage))
980                                 error = 0;
981                         page_cache_release(filepage);
982                 }
983         }
984         if (!error) {
985                 delete_from_swap_cache(page);
986                 set_page_dirty(page);
987                 info->flags |= SHMEM_PAGEIN;
988                 shmem_swp_set(info, ptr, 0);
989                 swap_free(entry);
990                 error = 1;      /* not an error, but entry was found */
991         }
992         if (ptr)
993                 shmem_swp_unmap(ptr);
994         spin_unlock(&info->lock);
995         radix_tree_preload_end();
996 out:
997         unlock_page(page);
998         page_cache_release(page);
999         iput(inode);            /* allows for NULL */
1000         return error;
1001 }
1002
1003 /*
1004  * shmem_unuse() search for an eventually swapped out shmem page.
1005  */
1006 int shmem_unuse(swp_entry_t entry, struct page *page)
1007 {
1008         struct list_head *p, *next;
1009         struct shmem_inode_info *info;
1010         int found = 0;
1011
1012         mutex_lock(&shmem_swaplist_mutex);
1013         list_for_each_safe(p, next, &shmem_swaplist) {
1014                 info = list_entry(p, struct shmem_inode_info, swaplist);
1015                 found = shmem_unuse_inode(info, entry, page);
1016                 cond_resched();
1017                 if (found)
1018                         goto out;
1019         }
1020         mutex_unlock(&shmem_swaplist_mutex);
1021         /*
1022          * Can some race bring us here?  We've been holding page lock,
1023          * so I think not; but would rather try again later than BUG()
1024          */
1025         unlock_page(page);
1026         page_cache_release(page);
1027 out:
1028         return (found < 0) ? found : 0;
1029 }
1030
1031 /*
1032  * Move the page from the page cache to the swap cache.
1033  */
1034 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1035 {
1036         struct shmem_inode_info *info;
1037         swp_entry_t *entry, swap;
1038         struct address_space *mapping;
1039         unsigned long index;
1040         struct inode *inode;
1041
1042         BUG_ON(!PageLocked(page));
1043         mapping = page->mapping;
1044         index = page->index;
1045         inode = mapping->host;
1046         info = SHMEM_I(inode);
1047         if (info->flags & VM_LOCKED)
1048                 goto redirty;
1049         if (!total_swap_pages)
1050                 goto redirty;
1051
1052         /*
1053          * shmem_backing_dev_info's capabilities prevent regular writeback or
1054          * sync from ever calling shmem_writepage; but a stacking filesystem
1055          * may use the ->writepage of its underlying filesystem, in which case
1056          * tmpfs should write out to swap only in response to memory pressure,
1057          * and not for the writeback threads or sync.  However, in those cases,
1058          * we do still want to check if there's a redundant swappage to be
1059          * discarded.
1060          */
1061         if (wbc->for_reclaim)
1062                 swap = get_swap_page();
1063         else
1064                 swap.val = 0;
1065
1066         spin_lock(&info->lock);
1067         if (index >= info->next_index) {
1068                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1069                 goto unlock;
1070         }
1071         entry = shmem_swp_entry(info, index, NULL);
1072         if (entry->val) {
1073                 /*
1074                  * The more uptodate page coming down from a stacked
1075                  * writepage should replace our old swappage.
1076                  */
1077                 free_swap_and_cache(*entry);
1078                 shmem_swp_set(info, entry, 0);
1079         }
1080         shmem_recalc_inode(inode);
1081
1082         if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1083                 remove_from_page_cache(page);
1084                 shmem_swp_set(info, entry, swap.val);
1085                 shmem_swp_unmap(entry);
1086                 if (list_empty(&info->swaplist))
1087                         inode = igrab(inode);
1088                 else
1089                         inode = NULL;
1090                 spin_unlock(&info->lock);
1091                 swap_shmem_alloc(swap);
1092                 BUG_ON(page_mapped(page));
1093                 page_cache_release(page);       /* pagecache ref */
1094                 swap_writepage(page, wbc);
1095                 if (inode) {
1096                         mutex_lock(&shmem_swaplist_mutex);
1097                         /* move instead of add in case we're racing */
1098                         list_move_tail(&info->swaplist, &shmem_swaplist);
1099                         mutex_unlock(&shmem_swaplist_mutex);
1100                         iput(inode);
1101                 }
1102                 return 0;
1103         }
1104
1105         shmem_swp_unmap(entry);
1106 unlock:
1107         spin_unlock(&info->lock);
1108         /*
1109          * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1110          * clear SWAP_HAS_CACHE flag.
1111          */
1112         swapcache_free(swap, NULL);
1113 redirty:
1114         set_page_dirty(page);
1115         if (wbc->for_reclaim)
1116                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
1117         unlock_page(page);
1118         return 0;
1119 }
1120
1121 #ifdef CONFIG_NUMA
1122 #ifdef CONFIG_TMPFS
1123 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1124 {
1125         char buffer[64];
1126
1127         if (!mpol || mpol->mode == MPOL_DEFAULT)
1128                 return;         /* show nothing */
1129
1130         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1131
1132         seq_printf(seq, ",mpol=%s", buffer);
1133 }
1134
1135 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1136 {
1137         struct mempolicy *mpol = NULL;
1138         if (sbinfo->mpol) {
1139                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
1140                 mpol = sbinfo->mpol;
1141                 mpol_get(mpol);
1142                 spin_unlock(&sbinfo->stat_lock);
1143         }
1144         return mpol;
1145 }
1146 #endif /* CONFIG_TMPFS */
1147
1148 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1149                         struct shmem_inode_info *info, unsigned long idx)
1150 {
1151         struct mempolicy mpol, *spol;
1152         struct vm_area_struct pvma;
1153         struct page *page;
1154
1155         spol = mpol_cond_copy(&mpol,
1156                                 mpol_shared_policy_lookup(&info->policy, idx));
1157
1158         /* Create a pseudo vma that just contains the policy */
1159         pvma.vm_start = 0;
1160         pvma.vm_pgoff = idx;
1161         pvma.vm_ops = NULL;
1162         pvma.vm_policy = spol;
1163         page = swapin_readahead(entry, gfp, &pvma, 0);
1164         return page;
1165 }
1166
1167 static struct page *shmem_alloc_page(gfp_t gfp,
1168                         struct shmem_inode_info *info, unsigned long idx)
1169 {
1170         struct vm_area_struct pvma;
1171
1172         /* Create a pseudo vma that just contains the policy */
1173         pvma.vm_start = 0;
1174         pvma.vm_pgoff = idx;
1175         pvma.vm_ops = NULL;
1176         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1177
1178         /*
1179          * alloc_page_vma() will drop the shared policy reference
1180          */
1181         return alloc_page_vma(gfp, &pvma, 0);
1182 }
1183 #else /* !CONFIG_NUMA */
1184 #ifdef CONFIG_TMPFS
1185 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1186 {
1187 }
1188 #endif /* CONFIG_TMPFS */
1189
1190 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1191                         struct shmem_inode_info *info, unsigned long idx)
1192 {
1193         return swapin_readahead(entry, gfp, NULL, 0);
1194 }
1195
1196 static inline struct page *shmem_alloc_page(gfp_t gfp,
1197                         struct shmem_inode_info *info, unsigned long idx)
1198 {
1199         return alloc_page(gfp);
1200 }
1201 #endif /* CONFIG_NUMA */
1202
1203 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1204 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1205 {
1206         return NULL;
1207 }
1208 #endif
1209
1210 /*
1211  * shmem_getpage - either get the page from swap or allocate a new one
1212  *
1213  * If we allocate a new one we do not mark it dirty. That's up to the
1214  * vm. If we swap it in we mark it dirty since we also free the swap
1215  * entry since a page cannot live in both the swap and page cache
1216  */
1217 static int shmem_getpage(struct inode *inode, unsigned long idx,
1218                         struct page **pagep, enum sgp_type sgp, int *type)
1219 {
1220         struct address_space *mapping = inode->i_mapping;
1221         struct shmem_inode_info *info = SHMEM_I(inode);
1222         struct shmem_sb_info *sbinfo;
1223         struct page *filepage = *pagep;
1224         struct page *swappage;
1225         swp_entry_t *entry;
1226         swp_entry_t swap;
1227         gfp_t gfp;
1228         int error;
1229
1230         if (idx >= SHMEM_MAX_INDEX)
1231                 return -EFBIG;
1232
1233         if (type)
1234                 *type = 0;
1235
1236         /*
1237          * Normally, filepage is NULL on entry, and either found
1238          * uptodate immediately, or allocated and zeroed, or read
1239          * in under swappage, which is then assigned to filepage.
1240          * But shmem_readpage (required for splice) passes in a locked
1241          * filepage, which may be found not uptodate by other callers
1242          * too, and may need to be copied from the swappage read in.
1243          */
1244 repeat:
1245         if (!filepage)
1246                 filepage = find_lock_page(mapping, idx);
1247         if (filepage && PageUptodate(filepage))
1248                 goto done;
1249         error = 0;
1250         gfp = mapping_gfp_mask(mapping);
1251         if (!filepage) {
1252                 /*
1253                  * Try to preload while we can wait, to not make a habit of
1254                  * draining atomic reserves; but don't latch on to this cpu.
1255                  */
1256                 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1257                 if (error)
1258                         goto failed;
1259                 radix_tree_preload_end();
1260         }
1261
1262         spin_lock(&info->lock);
1263         shmem_recalc_inode(inode);
1264         entry = shmem_swp_alloc(info, idx, sgp);
1265         if (IS_ERR(entry)) {
1266                 spin_unlock(&info->lock);
1267                 error = PTR_ERR(entry);
1268                 goto failed;
1269         }
1270         swap = *entry;
1271
1272         if (swap.val) {
1273                 /* Look it up and read it in.. */
1274                 swappage = lookup_swap_cache(swap);
1275                 if (!swappage) {
1276                         shmem_swp_unmap(entry);
1277                         /* here we actually do the io */
1278                         if (type && !(*type & VM_FAULT_MAJOR)) {
1279                                 __count_vm_event(PGMAJFAULT);
1280                                 *type |= VM_FAULT_MAJOR;
1281                         }
1282                         spin_unlock(&info->lock);
1283                         swappage = shmem_swapin(swap, gfp, info, idx);
1284                         if (!swappage) {
1285                                 spin_lock(&info->lock);
1286                                 entry = shmem_swp_alloc(info, idx, sgp);
1287                                 if (IS_ERR(entry))
1288                                         error = PTR_ERR(entry);
1289                                 else {
1290                                         if (entry->val == swap.val)
1291                                                 error = -ENOMEM;
1292                                         shmem_swp_unmap(entry);
1293                                 }
1294                                 spin_unlock(&info->lock);
1295                                 if (error)
1296                                         goto failed;
1297                                 goto repeat;
1298                         }
1299                         wait_on_page_locked(swappage);
1300                         page_cache_release(swappage);
1301                         goto repeat;
1302                 }
1303
1304                 /* We have to do this with page locked to prevent races */
1305                 if (!trylock_page(swappage)) {
1306                         shmem_swp_unmap(entry);
1307                         spin_unlock(&info->lock);
1308                         wait_on_page_locked(swappage);
1309                         page_cache_release(swappage);
1310                         goto repeat;
1311                 }
1312                 if (PageWriteback(swappage)) {
1313                         shmem_swp_unmap(entry);
1314                         spin_unlock(&info->lock);
1315                         wait_on_page_writeback(swappage);
1316                         unlock_page(swappage);
1317                         page_cache_release(swappage);
1318                         goto repeat;
1319                 }
1320                 if (!PageUptodate(swappage)) {
1321                         shmem_swp_unmap(entry);
1322                         spin_unlock(&info->lock);
1323                         unlock_page(swappage);
1324                         page_cache_release(swappage);
1325                         error = -EIO;
1326                         goto failed;
1327                 }
1328
1329                 if (filepage) {
1330                         shmem_swp_set(info, entry, 0);
1331                         shmem_swp_unmap(entry);
1332                         delete_from_swap_cache(swappage);
1333                         spin_unlock(&info->lock);
1334                         copy_highpage(filepage, swappage);
1335                         unlock_page(swappage);
1336                         page_cache_release(swappage);
1337                         flush_dcache_page(filepage);
1338                         SetPageUptodate(filepage);
1339                         set_page_dirty(filepage);
1340                         swap_free(swap);
1341                 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1342                                         idx, GFP_NOWAIT))) {
1343                         info->flags |= SHMEM_PAGEIN;
1344                         shmem_swp_set(info, entry, 0);
1345                         shmem_swp_unmap(entry);
1346                         delete_from_swap_cache(swappage);
1347                         spin_unlock(&info->lock);
1348                         filepage = swappage;
1349                         set_page_dirty(filepage);
1350                         swap_free(swap);
1351                 } else {
1352                         shmem_swp_unmap(entry);
1353                         spin_unlock(&info->lock);
1354                         if (error == -ENOMEM) {
1355                                 /*
1356                                  * reclaim from proper memory cgroup and
1357                                  * call memcg's OOM if needed.
1358                                  */
1359                                 error = mem_cgroup_shmem_charge_fallback(
1360                                                                 swappage,
1361                                                                 current->mm,
1362                                                                 gfp);
1363                                 if (error) {
1364                                         unlock_page(swappage);
1365                                         page_cache_release(swappage);
1366                                         goto failed;
1367                                 }
1368                         }
1369                         unlock_page(swappage);
1370                         page_cache_release(swappage);
1371                         goto repeat;
1372                 }
1373         } else if (sgp == SGP_READ && !filepage) {
1374                 shmem_swp_unmap(entry);
1375                 filepage = find_get_page(mapping, idx);
1376                 if (filepage &&
1377                     (!PageUptodate(filepage) || !trylock_page(filepage))) {
1378                         spin_unlock(&info->lock);
1379                         wait_on_page_locked(filepage);
1380                         page_cache_release(filepage);
1381                         filepage = NULL;
1382                         goto repeat;
1383                 }
1384                 spin_unlock(&info->lock);
1385         } else {
1386                 shmem_swp_unmap(entry);
1387                 sbinfo = SHMEM_SB(inode->i_sb);
1388                 if (sbinfo->max_blocks) {
1389                         spin_lock(&sbinfo->stat_lock);
1390                         if (sbinfo->free_blocks == 0 ||
1391                             shmem_acct_block(info->flags)) {
1392                                 spin_unlock(&sbinfo->stat_lock);
1393                                 spin_unlock(&info->lock);
1394                                 error = -ENOSPC;
1395                                 goto failed;
1396                         }
1397                         sbinfo->free_blocks--;
1398                         inode->i_blocks += BLOCKS_PER_PAGE;
1399                         spin_unlock(&sbinfo->stat_lock);
1400                 } else if (shmem_acct_block(info->flags)) {
1401                         spin_unlock(&info->lock);
1402                         error = -ENOSPC;
1403                         goto failed;
1404                 }
1405
1406                 if (!filepage) {
1407                         int ret;
1408
1409                         spin_unlock(&info->lock);
1410                         filepage = shmem_alloc_page(gfp, info, idx);
1411                         if (!filepage) {
1412                                 shmem_unacct_blocks(info->flags, 1);
1413                                 shmem_free_blocks(inode, 1);
1414                                 error = -ENOMEM;
1415                                 goto failed;
1416                         }
1417                         SetPageSwapBacked(filepage);
1418
1419                         /* Precharge page while we can wait, compensate after */
1420                         error = mem_cgroup_cache_charge(filepage, current->mm,
1421                                         GFP_KERNEL);
1422                         if (error) {
1423                                 page_cache_release(filepage);
1424                                 shmem_unacct_blocks(info->flags, 1);
1425                                 shmem_free_blocks(inode, 1);
1426                                 filepage = NULL;
1427                                 goto failed;
1428                         }
1429
1430                         spin_lock(&info->lock);
1431                         entry = shmem_swp_alloc(info, idx, sgp);
1432                         if (IS_ERR(entry))
1433                                 error = PTR_ERR(entry);
1434                         else {
1435                                 swap = *entry;
1436                                 shmem_swp_unmap(entry);
1437                         }
1438                         ret = error || swap.val;
1439                         if (ret)
1440                                 mem_cgroup_uncharge_cache_page(filepage);
1441                         else
1442                                 ret = add_to_page_cache_lru(filepage, mapping,
1443                                                 idx, GFP_NOWAIT);
1444                         /*
1445                          * At add_to_page_cache_lru() failure, uncharge will
1446                          * be done automatically.
1447                          */
1448                         if (ret) {
1449                                 spin_unlock(&info->lock);
1450                                 page_cache_release(filepage);
1451                                 shmem_unacct_blocks(info->flags, 1);
1452                                 shmem_free_blocks(inode, 1);
1453                                 filepage = NULL;
1454                                 if (error)
1455                                         goto failed;
1456                                 goto repeat;
1457                         }
1458                         info->flags |= SHMEM_PAGEIN;
1459                 }
1460
1461                 info->alloced++;
1462                 spin_unlock(&info->lock);
1463                 clear_highpage(filepage);
1464                 flush_dcache_page(filepage);
1465                 SetPageUptodate(filepage);
1466                 if (sgp == SGP_DIRTY)
1467                         set_page_dirty(filepage);
1468         }
1469 done:
1470         *pagep = filepage;
1471         return 0;
1472
1473 failed:
1474         if (*pagep != filepage) {
1475                 unlock_page(filepage);
1476                 page_cache_release(filepage);
1477         }
1478         return error;
1479 }
1480
1481 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1482 {
1483         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1484         int error;
1485         int ret;
1486
1487         if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1488                 return VM_FAULT_SIGBUS;
1489
1490         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1491         if (error)
1492                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1493
1494         return ret | VM_FAULT_LOCKED;
1495 }
1496
1497 #ifdef CONFIG_NUMA
1498 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1499 {
1500         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1501         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1502 }
1503
1504 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1505                                           unsigned long addr)
1506 {
1507         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1508         unsigned long idx;
1509
1510         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1511         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1512 }
1513 #endif
1514
1515 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1516 {
1517         struct inode *inode = file->f_path.dentry->d_inode;
1518         struct shmem_inode_info *info = SHMEM_I(inode);
1519         int retval = -ENOMEM;
1520
1521         spin_lock(&info->lock);
1522         if (lock && !(info->flags & VM_LOCKED)) {
1523                 if (!user_shm_lock(inode->i_size, user))
1524                         goto out_nomem;
1525                 info->flags |= VM_LOCKED;
1526                 mapping_set_unevictable(file->f_mapping);
1527         }
1528         if (!lock && (info->flags & VM_LOCKED) && user) {
1529                 user_shm_unlock(inode->i_size, user);
1530                 info->flags &= ~VM_LOCKED;
1531                 mapping_clear_unevictable(file->f_mapping);
1532                 scan_mapping_unevictable_pages(file->f_mapping);
1533         }
1534         retval = 0;
1535
1536 out_nomem:
1537         spin_unlock(&info->lock);
1538         return retval;
1539 }
1540
1541 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1542 {
1543         file_accessed(file);
1544         vma->vm_ops = &shmem_vm_ops;
1545         vma->vm_flags |= VM_CAN_NONLINEAR;
1546         return 0;
1547 }
1548
1549 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1550                                      int mode, dev_t dev, unsigned long flags)
1551 {
1552         struct inode *inode;
1553         struct shmem_inode_info *info;
1554         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1555
1556         if (shmem_reserve_inode(sb))
1557                 return NULL;
1558
1559         inode = new_inode(sb);
1560         if (inode) {
1561                 inode_init_owner(inode, dir, mode);
1562                 inode->i_blocks = 0;
1563                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1564                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1565                 inode->i_generation = get_seconds();
1566                 info = SHMEM_I(inode);
1567                 memset(info, 0, (char *)inode - (char *)info);
1568                 spin_lock_init(&info->lock);
1569                 info->flags = flags & VM_NORESERVE;
1570                 INIT_LIST_HEAD(&info->swaplist);
1571                 cache_no_acl(inode);
1572
1573                 switch (mode & S_IFMT) {
1574                 default:
1575                         inode->i_op = &shmem_special_inode_operations;
1576                         init_special_inode(inode, mode, dev);
1577                         break;
1578                 case S_IFREG:
1579                         inode->i_mapping->a_ops = &shmem_aops;
1580                         inode->i_op = &shmem_inode_operations;
1581                         inode->i_fop = &shmem_file_operations;
1582                         mpol_shared_policy_init(&info->policy,
1583                                                  shmem_get_sbmpol(sbinfo));
1584                         break;
1585                 case S_IFDIR:
1586                         inc_nlink(inode);
1587                         /* Some things misbehave if size == 0 on a directory */
1588                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1589                         inode->i_op = &shmem_dir_inode_operations;
1590                         inode->i_fop = &simple_dir_operations;
1591                         break;
1592                 case S_IFLNK:
1593                         /*
1594                          * Must not load anything in the rbtree,
1595                          * mpol_free_shared_policy will not be called.
1596                          */
1597                         mpol_shared_policy_init(&info->policy, NULL);
1598                         break;
1599                 }
1600         } else
1601                 shmem_free_inode(sb);
1602         return inode;
1603 }
1604
1605 #ifdef CONFIG_TMPFS
1606 static const struct inode_operations shmem_symlink_inode_operations;
1607 static const struct inode_operations shmem_symlink_inline_operations;
1608
1609 /*
1610  * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1611  * but providing them allows a tmpfs file to be used for splice, sendfile, and
1612  * below the loop driver, in the generic fashion that many filesystems support.
1613  */
1614 static int shmem_readpage(struct file *file, struct page *page)
1615 {
1616         struct inode *inode = page->mapping->host;
1617         int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1618         unlock_page(page);
1619         return error;
1620 }
1621
1622 static int
1623 shmem_write_begin(struct file *file, struct address_space *mapping,
1624                         loff_t pos, unsigned len, unsigned flags,
1625                         struct page **pagep, void **fsdata)
1626 {
1627         struct inode *inode = mapping->host;
1628         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1629         *pagep = NULL;
1630         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1631 }
1632
1633 static int
1634 shmem_write_end(struct file *file, struct address_space *mapping,
1635                         loff_t pos, unsigned len, unsigned copied,
1636                         struct page *page, void *fsdata)
1637 {
1638         struct inode *inode = mapping->host;
1639
1640         if (pos + copied > inode->i_size)
1641                 i_size_write(inode, pos + copied);
1642
1643         set_page_dirty(page);
1644         unlock_page(page);
1645         page_cache_release(page);
1646
1647         return copied;
1648 }
1649
1650 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1651 {
1652         struct inode *inode = filp->f_path.dentry->d_inode;
1653         struct address_space *mapping = inode->i_mapping;
1654         unsigned long index, offset;
1655         enum sgp_type sgp = SGP_READ;
1656
1657         /*
1658          * Might this read be for a stacking filesystem?  Then when reading
1659          * holes of a sparse file, we actually need to allocate those pages,
1660          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1661          */
1662         if (segment_eq(get_fs(), KERNEL_DS))
1663                 sgp = SGP_DIRTY;
1664
1665         index = *ppos >> PAGE_CACHE_SHIFT;
1666         offset = *ppos & ~PAGE_CACHE_MASK;
1667
1668         for (;;) {
1669                 struct page *page = NULL;
1670                 unsigned long end_index, nr, ret;
1671                 loff_t i_size = i_size_read(inode);
1672
1673                 end_index = i_size >> PAGE_CACHE_SHIFT;
1674                 if (index > end_index)
1675                         break;
1676                 if (index == end_index) {
1677                         nr = i_size & ~PAGE_CACHE_MASK;
1678                         if (nr <= offset)
1679                                 break;
1680                 }
1681
1682                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1683                 if (desc->error) {
1684                         if (desc->error == -EINVAL)
1685                                 desc->error = 0;
1686                         break;
1687                 }
1688                 if (page)
1689                         unlock_page(page);
1690
1691                 /*
1692                  * We must evaluate after, since reads (unlike writes)
1693                  * are called without i_mutex protection against truncate
1694                  */
1695                 nr = PAGE_CACHE_SIZE;
1696                 i_size = i_size_read(inode);
1697                 end_index = i_size >> PAGE_CACHE_SHIFT;
1698                 if (index == end_index) {
1699                         nr = i_size & ~PAGE_CACHE_MASK;
1700                         if (nr <= offset) {
1701                                 if (page)
1702                                         page_cache_release(page);
1703                                 break;
1704                         }
1705                 }
1706                 nr -= offset;
1707
1708                 if (page) {
1709                         /*
1710                          * If users can be writing to this page using arbitrary
1711                          * virtual addresses, take care about potential aliasing
1712                          * before reading the page on the kernel side.
1713                          */
1714                         if (mapping_writably_mapped(mapping))
1715                                 flush_dcache_page(page);
1716                         /*
1717                          * Mark the page accessed if we read the beginning.
1718                          */
1719                         if (!offset)
1720                                 mark_page_accessed(page);
1721                 } else {
1722                         page = ZERO_PAGE(0);
1723                         page_cache_get(page);
1724                 }
1725
1726                 /*
1727                  * Ok, we have the page, and it's up-to-date, so
1728                  * now we can copy it to user space...
1729                  *
1730                  * The actor routine returns how many bytes were actually used..
1731                  * NOTE! This may not be the same as how much of a user buffer
1732                  * we filled up (we may be padding etc), so we can only update
1733                  * "pos" here (the actor routine has to update the user buffer
1734                  * pointers and the remaining count).
1735                  */
1736                 ret = actor(desc, page, offset, nr);
1737                 offset += ret;
1738                 index += offset >> PAGE_CACHE_SHIFT;
1739                 offset &= ~PAGE_CACHE_MASK;
1740
1741                 page_cache_release(page);
1742                 if (ret != nr || !desc->count)
1743                         break;
1744
1745                 cond_resched();
1746         }
1747
1748         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1749         file_accessed(filp);
1750 }
1751
1752 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1753                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1754 {
1755         struct file *filp = iocb->ki_filp;
1756         ssize_t retval;
1757         unsigned long seg;
1758         size_t count;
1759         loff_t *ppos = &iocb->ki_pos;
1760
1761         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1762         if (retval)
1763                 return retval;
1764
1765         for (seg = 0; seg < nr_segs; seg++) {
1766                 read_descriptor_t desc;
1767
1768                 desc.written = 0;
1769                 desc.arg.buf = iov[seg].iov_base;
1770                 desc.count = iov[seg].iov_len;
1771                 if (desc.count == 0)
1772                         continue;
1773                 desc.error = 0;
1774                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1775                 retval += desc.written;
1776                 if (desc.error) {
1777                         retval = retval ?: desc.error;
1778                         break;
1779                 }
1780                 if (desc.count > 0)
1781                         break;
1782         }
1783         return retval;
1784 }
1785
1786 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1787 {
1788         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1789
1790         buf->f_type = TMPFS_MAGIC;
1791         buf->f_bsize = PAGE_CACHE_SIZE;
1792         buf->f_namelen = NAME_MAX;
1793         spin_lock(&sbinfo->stat_lock);
1794         if (sbinfo->max_blocks) {
1795                 buf->f_blocks = sbinfo->max_blocks;
1796                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1797         }
1798         if (sbinfo->max_inodes) {
1799                 buf->f_files = sbinfo->max_inodes;
1800                 buf->f_ffree = sbinfo->free_inodes;
1801         }
1802         /* else leave those fields 0 like simple_statfs */
1803         spin_unlock(&sbinfo->stat_lock);
1804         return 0;
1805 }
1806
1807 /*
1808  * File creation. Allocate an inode, and we're done..
1809  */
1810 static int
1811 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1812 {
1813         struct inode *inode;
1814         int error = -ENOSPC;
1815
1816         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1817         if (inode) {
1818                 error = security_inode_init_security(inode, dir, NULL, NULL,
1819                                                      NULL);
1820                 if (error) {
1821                         if (error != -EOPNOTSUPP) {
1822                                 iput(inode);
1823                                 return error;
1824                         }
1825                 }
1826 #ifdef CONFIG_TMPFS_POSIX_ACL
1827                 error = generic_acl_init(inode, dir);
1828                 if (error) {
1829                         iput(inode);
1830                         return error;
1831                 }
1832 #else
1833                 error = 0;
1834 #endif
1835                 dir->i_size += BOGO_DIRENT_SIZE;
1836                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1837                 d_instantiate(dentry, inode);
1838                 dget(dentry); /* Extra count - pin the dentry in core */
1839         }
1840         return error;
1841 }
1842
1843 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1844 {
1845         int error;
1846
1847         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1848                 return error;
1849         inc_nlink(dir);
1850         return 0;
1851 }
1852
1853 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1854                 struct nameidata *nd)
1855 {
1856         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1857 }
1858
1859 /*
1860  * Link a file..
1861  */
1862 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1863 {
1864         struct inode *inode = old_dentry->d_inode;
1865         int ret;
1866
1867         /*
1868          * No ordinary (disk based) filesystem counts links as inodes;
1869          * but each new link needs a new dentry, pinning lowmem, and
1870          * tmpfs dentries cannot be pruned until they are unlinked.
1871          */
1872         ret = shmem_reserve_inode(inode->i_sb);
1873         if (ret)
1874                 goto out;
1875
1876         dir->i_size += BOGO_DIRENT_SIZE;
1877         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1878         inc_nlink(inode);
1879         atomic_inc(&inode->i_count);    /* New dentry reference */
1880         dget(dentry);           /* Extra pinning count for the created dentry */
1881         d_instantiate(dentry, inode);
1882 out:
1883         return ret;
1884 }
1885
1886 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1887 {
1888         struct inode *inode = dentry->d_inode;
1889
1890         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1891                 shmem_free_inode(inode->i_sb);
1892
1893         dir->i_size -= BOGO_DIRENT_SIZE;
1894         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1895         drop_nlink(inode);
1896         dput(dentry);   /* Undo the count from "create" - this does all the work */
1897         return 0;
1898 }
1899
1900 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1901 {
1902         if (!simple_empty(dentry))
1903                 return -ENOTEMPTY;
1904
1905         drop_nlink(dentry->d_inode);
1906         drop_nlink(dir);
1907         return shmem_unlink(dir, dentry);
1908 }
1909
1910 /*
1911  * The VFS layer already does all the dentry stuff for rename,
1912  * we just have to decrement the usage count for the target if
1913  * it exists so that the VFS layer correctly free's it when it
1914  * gets overwritten.
1915  */
1916 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1917 {
1918         struct inode *inode = old_dentry->d_inode;
1919         int they_are_dirs = S_ISDIR(inode->i_mode);
1920
1921         if (!simple_empty(new_dentry))
1922                 return -ENOTEMPTY;
1923
1924         if (new_dentry->d_inode) {
1925                 (void) shmem_unlink(new_dir, new_dentry);
1926                 if (they_are_dirs)
1927                         drop_nlink(old_dir);
1928         } else if (they_are_dirs) {
1929                 drop_nlink(old_dir);
1930                 inc_nlink(new_dir);
1931         }
1932
1933         old_dir->i_size -= BOGO_DIRENT_SIZE;
1934         new_dir->i_size += BOGO_DIRENT_SIZE;
1935         old_dir->i_ctime = old_dir->i_mtime =
1936         new_dir->i_ctime = new_dir->i_mtime =
1937         inode->i_ctime = CURRENT_TIME;
1938         return 0;
1939 }
1940
1941 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1942 {
1943         int error;
1944         int len;
1945         struct inode *inode;
1946         struct page *page = NULL;
1947         char *kaddr;
1948         struct shmem_inode_info *info;
1949
1950         len = strlen(symname) + 1;
1951         if (len > PAGE_CACHE_SIZE)
1952                 return -ENAMETOOLONG;
1953
1954         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1955         if (!inode)
1956                 return -ENOSPC;
1957
1958         error = security_inode_init_security(inode, dir, NULL, NULL,
1959                                              NULL);
1960         if (error) {
1961                 if (error != -EOPNOTSUPP) {
1962                         iput(inode);
1963                         return error;
1964                 }
1965                 error = 0;
1966         }
1967
1968         info = SHMEM_I(inode);
1969         inode->i_size = len-1;
1970         if (len <= (char *)inode - (char *)info) {
1971                 /* do it inline */
1972                 memcpy(info, symname, len);
1973                 inode->i_op = &shmem_symlink_inline_operations;
1974         } else {
1975                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1976                 if (error) {
1977                         iput(inode);
1978                         return error;
1979                 }
1980                 inode->i_mapping->a_ops = &shmem_aops;
1981                 inode->i_op = &shmem_symlink_inode_operations;
1982                 kaddr = kmap_atomic(page, KM_USER0);
1983                 memcpy(kaddr, symname, len);
1984                 kunmap_atomic(kaddr, KM_USER0);
1985                 set_page_dirty(page);
1986                 unlock_page(page);
1987                 page_cache_release(page);
1988         }
1989         dir->i_size += BOGO_DIRENT_SIZE;
1990         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1991         d_instantiate(dentry, inode);
1992         dget(dentry);
1993         return 0;
1994 }
1995
1996 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1997 {
1998         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1999         return NULL;
2000 }
2001
2002 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2003 {
2004         struct page *page = NULL;
2005         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2006         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2007         if (page)
2008                 unlock_page(page);
2009         return page;
2010 }
2011
2012 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2013 {
2014         if (!IS_ERR(nd_get_link(nd))) {
2015                 struct page *page = cookie;
2016                 kunmap(page);
2017                 mark_page_accessed(page);
2018                 page_cache_release(page);
2019         }
2020 }
2021
2022 static const struct inode_operations shmem_symlink_inline_operations = {
2023         .readlink       = generic_readlink,
2024         .follow_link    = shmem_follow_link_inline,
2025 };
2026
2027 static const struct inode_operations shmem_symlink_inode_operations = {
2028         .readlink       = generic_readlink,
2029         .follow_link    = shmem_follow_link,
2030         .put_link       = shmem_put_link,
2031 };
2032
2033 #ifdef CONFIG_TMPFS_POSIX_ACL
2034 /*
2035  * Superblocks without xattr inode operations will get security.* xattr
2036  * support from the VFS "for free". As soon as we have any other xattrs
2037  * like ACLs, we also need to implement the security.* handlers at
2038  * filesystem level, though.
2039  */
2040
2041 static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
2042                                         size_t list_len, const char *name,
2043                                         size_t name_len, int handler_flags)
2044 {
2045         return security_inode_listsecurity(dentry->d_inode, list, list_len);
2046 }
2047
2048 static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2049                 void *buffer, size_t size, int handler_flags)
2050 {
2051         if (strcmp(name, "") == 0)
2052                 return -EINVAL;
2053         return xattr_getsecurity(dentry->d_inode, name, buffer, size);
2054 }
2055
2056 static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2057                 const void *value, size_t size, int flags, int handler_flags)
2058 {
2059         if (strcmp(name, "") == 0)
2060                 return -EINVAL;
2061         return security_inode_setsecurity(dentry->d_inode, name, value,
2062                                           size, flags);
2063 }
2064
2065 static const struct xattr_handler shmem_xattr_security_handler = {
2066         .prefix = XATTR_SECURITY_PREFIX,
2067         .list   = shmem_xattr_security_list,
2068         .get    = shmem_xattr_security_get,
2069         .set    = shmem_xattr_security_set,
2070 };
2071
2072 static const struct xattr_handler *shmem_xattr_handlers[] = {
2073         &generic_acl_access_handler,
2074         &generic_acl_default_handler,
2075         &shmem_xattr_security_handler,
2076         NULL
2077 };
2078 #endif
2079
2080 static struct dentry *shmem_get_parent(struct dentry *child)
2081 {
2082         return ERR_PTR(-ESTALE);
2083 }
2084
2085 static int shmem_match(struct inode *ino, void *vfh)
2086 {
2087         __u32 *fh = vfh;
2088         __u64 inum = fh[2];
2089         inum = (inum << 32) | fh[1];
2090         return ino->i_ino == inum && fh[0] == ino->i_generation;
2091 }
2092
2093 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2094                 struct fid *fid, int fh_len, int fh_type)
2095 {
2096         struct inode *inode;
2097         struct dentry *dentry = NULL;
2098         u64 inum = fid->raw[2];
2099         inum = (inum << 32) | fid->raw[1];
2100
2101         if (fh_len < 3)
2102                 return NULL;
2103
2104         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2105                         shmem_match, fid->raw);
2106         if (inode) {
2107                 dentry = d_find_alias(inode);
2108                 iput(inode);
2109         }
2110
2111         return dentry;
2112 }
2113
2114 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2115                                 int connectable)
2116 {
2117         struct inode *inode = dentry->d_inode;
2118
2119         if (*len < 3)
2120                 return 255;
2121
2122         if (hlist_unhashed(&inode->i_hash)) {
2123                 /* Unfortunately insert_inode_hash is not idempotent,
2124                  * so as we hash inodes here rather than at creation
2125                  * time, we need a lock to ensure we only try
2126                  * to do it once
2127                  */
2128                 static DEFINE_SPINLOCK(lock);
2129                 spin_lock(&lock);
2130                 if (hlist_unhashed(&inode->i_hash))
2131                         __insert_inode_hash(inode,
2132                                             inode->i_ino + inode->i_generation);
2133                 spin_unlock(&lock);
2134         }
2135
2136         fh[0] = inode->i_generation;
2137         fh[1] = inode->i_ino;
2138         fh[2] = ((__u64)inode->i_ino) >> 32;
2139
2140         *len = 3;
2141         return 1;
2142 }
2143
2144 static const struct export_operations shmem_export_ops = {
2145         .get_parent     = shmem_get_parent,
2146         .encode_fh      = shmem_encode_fh,
2147         .fh_to_dentry   = shmem_fh_to_dentry,
2148 };
2149
2150 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2151                                bool remount)
2152 {
2153         char *this_char, *value, *rest;
2154
2155         while (options != NULL) {
2156                 this_char = options;
2157                 for (;;) {
2158                         /*
2159                          * NUL-terminate this option: unfortunately,
2160                          * mount options form a comma-separated list,
2161                          * but mpol's nodelist may also contain commas.
2162                          */
2163                         options = strchr(options, ',');
2164                         if (options == NULL)
2165                                 break;
2166                         options++;
2167                         if (!isdigit(*options)) {
2168                                 options[-1] = '\0';
2169                                 break;
2170                         }
2171                 }
2172                 if (!*this_char)
2173                         continue;
2174                 if ((value = strchr(this_char,'=')) != NULL) {
2175                         *value++ = 0;
2176                 } else {
2177                         printk(KERN_ERR
2178                             "tmpfs: No value for mount option '%s'\n",
2179                             this_char);
2180                         return 1;
2181                 }
2182
2183                 if (!strcmp(this_char,"size")) {
2184                         unsigned long long size;
2185                         size = memparse(value,&rest);
2186                         if (*rest == '%') {
2187                                 size <<= PAGE_SHIFT;
2188                                 size *= totalram_pages;
2189                                 do_div(size, 100);
2190                                 rest++;
2191                         }
2192                         if (*rest)
2193                                 goto bad_val;
2194                         sbinfo->max_blocks =
2195                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2196                 } else if (!strcmp(this_char,"nr_blocks")) {
2197                         sbinfo->max_blocks = memparse(value, &rest);
2198                         if (*rest)
2199                                 goto bad_val;
2200                 } else if (!strcmp(this_char,"nr_inodes")) {
2201                         sbinfo->max_inodes = memparse(value, &rest);
2202                         if (*rest)
2203                                 goto bad_val;
2204                 } else if (!strcmp(this_char,"mode")) {
2205                         if (remount)
2206                                 continue;
2207                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2208                         if (*rest)
2209                                 goto bad_val;
2210                 } else if (!strcmp(this_char,"uid")) {
2211                         if (remount)
2212                                 continue;
2213                         sbinfo->uid = simple_strtoul(value, &rest, 0);
2214                         if (*rest)
2215                                 goto bad_val;
2216                 } else if (!strcmp(this_char,"gid")) {
2217                         if (remount)
2218                                 continue;
2219                         sbinfo->gid = simple_strtoul(value, &rest, 0);
2220                         if (*rest)
2221                                 goto bad_val;
2222                 } else if (!strcmp(this_char,"mpol")) {
2223                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2224                                 goto bad_val;
2225                 } else {
2226                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2227                                this_char);
2228                         return 1;
2229                 }
2230         }
2231         return 0;
2232
2233 bad_val:
2234         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2235                value, this_char);
2236         return 1;
2237
2238 }
2239
2240 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2241 {
2242         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2243         struct shmem_sb_info config = *sbinfo;
2244         unsigned long blocks;
2245         unsigned long inodes;
2246         int error = -EINVAL;
2247
2248         if (shmem_parse_options(data, &config, true))
2249                 return error;
2250
2251         spin_lock(&sbinfo->stat_lock);
2252         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2253         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2254         if (config.max_blocks < blocks)
2255                 goto out;
2256         if (config.max_inodes < inodes)
2257                 goto out;
2258         /*
2259          * Those tests also disallow limited->unlimited while any are in
2260          * use, so i_blocks will always be zero when max_blocks is zero;
2261          * but we must separately disallow unlimited->limited, because
2262          * in that case we have no record of how much is already in use.
2263          */
2264         if (config.max_blocks && !sbinfo->max_blocks)
2265                 goto out;
2266         if (config.max_inodes && !sbinfo->max_inodes)
2267                 goto out;
2268
2269         error = 0;
2270         sbinfo->max_blocks  = config.max_blocks;
2271         sbinfo->free_blocks = config.max_blocks - blocks;
2272         sbinfo->max_inodes  = config.max_inodes;
2273         sbinfo->free_inodes = config.max_inodes - inodes;
2274
2275         mpol_put(sbinfo->mpol);
2276         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2277 out:
2278         spin_unlock(&sbinfo->stat_lock);
2279         return error;
2280 }
2281
2282 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2283 {
2284         struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2285
2286         if (sbinfo->max_blocks != shmem_default_max_blocks())
2287                 seq_printf(seq, ",size=%luk",
2288                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2289         if (sbinfo->max_inodes != shmem_default_max_inodes())
2290                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2291         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2292                 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2293         if (sbinfo->uid != 0)
2294                 seq_printf(seq, ",uid=%u", sbinfo->uid);
2295         if (sbinfo->gid != 0)
2296                 seq_printf(seq, ",gid=%u", sbinfo->gid);
2297         shmem_show_mpol(seq, sbinfo->mpol);
2298         return 0;
2299 }
2300 #endif /* CONFIG_TMPFS */
2301
2302 static void shmem_put_super(struct super_block *sb)
2303 {
2304         kfree(sb->s_fs_info);
2305         sb->s_fs_info = NULL;
2306 }
2307
2308 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2309 {
2310         struct inode *inode;
2311         struct dentry *root;
2312         struct shmem_sb_info *sbinfo;
2313         int err = -ENOMEM;
2314
2315         /* Round up to L1_CACHE_BYTES to resist false sharing */
2316         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2317                                 L1_CACHE_BYTES), GFP_KERNEL);
2318         if (!sbinfo)
2319                 return -ENOMEM;
2320
2321         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2322         sbinfo->uid = current_fsuid();
2323         sbinfo->gid = current_fsgid();
2324         sb->s_fs_info = sbinfo;
2325
2326 #ifdef CONFIG_TMPFS
2327         /*
2328          * Per default we only allow half of the physical ram per
2329          * tmpfs instance, limiting inodes to one per page of lowmem;
2330          * but the internal instance is left unlimited.
2331          */
2332         if (!(sb->s_flags & MS_NOUSER)) {
2333                 sbinfo->max_blocks = shmem_default_max_blocks();
2334                 sbinfo->max_inodes = shmem_default_max_inodes();
2335                 if (shmem_parse_options(data, sbinfo, false)) {
2336                         err = -EINVAL;
2337                         goto failed;
2338                 }
2339         }
2340         sb->s_export_op = &shmem_export_ops;
2341 #else
2342         sb->s_flags |= MS_NOUSER;
2343 #endif
2344
2345         spin_lock_init(&sbinfo->stat_lock);
2346         sbinfo->free_blocks = sbinfo->max_blocks;
2347         sbinfo->free_inodes = sbinfo->max_inodes;
2348
2349         sb->s_maxbytes = SHMEM_MAX_BYTES;
2350         sb->s_blocksize = PAGE_CACHE_SIZE;
2351         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2352         sb->s_magic = TMPFS_MAGIC;
2353         sb->s_op = &shmem_ops;
2354         sb->s_time_gran = 1;
2355 #ifdef CONFIG_TMPFS_POSIX_ACL
2356         sb->s_xattr = shmem_xattr_handlers;
2357         sb->s_flags |= MS_POSIXACL;
2358 #endif
2359
2360         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2361         if (!inode)
2362                 goto failed;
2363         inode->i_uid = sbinfo->uid;
2364         inode->i_gid = sbinfo->gid;
2365         root = d_alloc_root(inode);
2366         if (!root)
2367                 goto failed_iput;
2368         sb->s_root = root;
2369         return 0;
2370
2371 failed_iput:
2372         iput(inode);
2373 failed:
2374         shmem_put_super(sb);
2375         return err;
2376 }
2377
2378 static struct kmem_cache *shmem_inode_cachep;
2379
2380 static struct inode *shmem_alloc_inode(struct super_block *sb)
2381 {
2382         struct shmem_inode_info *p;
2383         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2384         if (!p)
2385                 return NULL;
2386         return &p->vfs_inode;
2387 }
2388
2389 static void shmem_destroy_inode(struct inode *inode)
2390 {
2391         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2392                 /* only struct inode is valid if it's an inline symlink */
2393                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2394         }
2395         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2396 }
2397
2398 static void init_once(void *foo)
2399 {
2400         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2401
2402         inode_init_once(&p->vfs_inode);
2403 }
2404
2405 static int init_inodecache(void)
2406 {
2407         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2408                                 sizeof(struct shmem_inode_info),
2409                                 0, SLAB_PANIC, init_once);
2410         return 0;
2411 }
2412
2413 static void destroy_inodecache(void)
2414 {
2415         kmem_cache_destroy(shmem_inode_cachep);
2416 }
2417
2418 static const struct address_space_operations shmem_aops = {
2419         .writepage      = shmem_writepage,
2420         .set_page_dirty = __set_page_dirty_no_writeback,
2421 #ifdef CONFIG_TMPFS
2422         .readpage       = shmem_readpage,
2423         .write_begin    = shmem_write_begin,
2424         .write_end      = shmem_write_end,
2425 #endif
2426         .migratepage    = migrate_page,
2427         .error_remove_page = generic_error_remove_page,
2428 };
2429
2430 static const struct file_operations shmem_file_operations = {
2431         .mmap           = shmem_mmap,
2432 #ifdef CONFIG_TMPFS
2433         .llseek         = generic_file_llseek,
2434         .read           = do_sync_read,
2435         .write          = do_sync_write,
2436         .aio_read       = shmem_file_aio_read,
2437         .aio_write      = generic_file_aio_write,
2438         .fsync          = noop_fsync,
2439         .splice_read    = generic_file_splice_read,
2440         .splice_write   = generic_file_splice_write,
2441 #endif
2442 };
2443
2444 static const struct inode_operations shmem_inode_operations = {
2445         .setattr        = shmem_notify_change,
2446         .truncate_range = shmem_truncate_range,
2447 #ifdef CONFIG_TMPFS_POSIX_ACL
2448         .setxattr       = generic_setxattr,
2449         .getxattr       = generic_getxattr,
2450         .listxattr      = generic_listxattr,
2451         .removexattr    = generic_removexattr,
2452         .check_acl      = generic_check_acl,
2453 #endif
2454
2455 };
2456
2457 static const struct inode_operations shmem_dir_inode_operations = {
2458 #ifdef CONFIG_TMPFS
2459         .create         = shmem_create,
2460         .lookup         = simple_lookup,
2461         .link           = shmem_link,
2462         .unlink         = shmem_unlink,
2463         .symlink        = shmem_symlink,
2464         .mkdir          = shmem_mkdir,
2465         .rmdir          = shmem_rmdir,
2466         .mknod          = shmem_mknod,
2467         .rename         = shmem_rename,
2468 #endif
2469 #ifdef CONFIG_TMPFS_POSIX_ACL
2470         .setattr        = shmem_notify_change,
2471         .setxattr       = generic_setxattr,
2472         .getxattr       = generic_getxattr,
2473         .listxattr      = generic_listxattr,
2474         .removexattr    = generic_removexattr,
2475         .check_acl      = generic_check_acl,
2476 #endif
2477 };
2478
2479 static const struct inode_operations shmem_special_inode_operations = {
2480 #ifdef CONFIG_TMPFS_POSIX_ACL
2481         .setattr        = shmem_notify_change,
2482         .setxattr       = generic_setxattr,
2483         .getxattr       = generic_getxattr,
2484         .listxattr      = generic_listxattr,
2485         .removexattr    = generic_removexattr,
2486         .check_acl      = generic_check_acl,
2487 #endif
2488 };
2489
2490 static const struct super_operations shmem_ops = {
2491         .alloc_inode    = shmem_alloc_inode,
2492         .destroy_inode  = shmem_destroy_inode,
2493 #ifdef CONFIG_TMPFS
2494         .statfs         = shmem_statfs,
2495         .remount_fs     = shmem_remount_fs,
2496         .show_options   = shmem_show_options,
2497 #endif
2498         .delete_inode   = shmem_delete_inode,
2499         .drop_inode     = generic_delete_inode,
2500         .put_super      = shmem_put_super,
2501 };
2502
2503 static const struct vm_operations_struct shmem_vm_ops = {
2504         .fault          = shmem_fault,
2505 #ifdef CONFIG_NUMA
2506         .set_policy     = shmem_set_policy,
2507         .get_policy     = shmem_get_policy,
2508 #endif
2509 };
2510
2511
2512 static int shmem_get_sb(struct file_system_type *fs_type,
2513         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2514 {
2515         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2516 }
2517
2518 static struct file_system_type tmpfs_fs_type = {
2519         .owner          = THIS_MODULE,
2520         .name           = "tmpfs",
2521         .get_sb         = shmem_get_sb,
2522         .kill_sb        = kill_litter_super,
2523 };
2524
2525 int __init init_tmpfs(void)
2526 {
2527         int error;
2528
2529         error = bdi_init(&shmem_backing_dev_info);
2530         if (error)
2531                 goto out4;
2532
2533         error = init_inodecache();
2534         if (error)
2535                 goto out3;
2536
2537         error = register_filesystem(&tmpfs_fs_type);
2538         if (error) {
2539                 printk(KERN_ERR "Could not register tmpfs\n");
2540                 goto out2;
2541         }
2542
2543         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2544                                 tmpfs_fs_type.name, NULL);
2545         if (IS_ERR(shm_mnt)) {
2546                 error = PTR_ERR(shm_mnt);
2547                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2548                 goto out1;
2549         }
2550         return 0;
2551
2552 out1:
2553         unregister_filesystem(&tmpfs_fs_type);
2554 out2:
2555         destroy_inodecache();
2556 out3:
2557         bdi_destroy(&shmem_backing_dev_info);
2558 out4:
2559         shm_mnt = ERR_PTR(error);
2560         return error;
2561 }
2562
2563 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2564 /**
2565  * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2566  * @inode: the inode to be searched
2567  * @pgoff: the offset to be searched
2568  * @pagep: the pointer for the found page to be stored
2569  * @ent: the pointer for the found swap entry to be stored
2570  *
2571  * If a page is found, refcount of it is incremented. Callers should handle
2572  * these refcount.
2573  */
2574 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2575                                         struct page **pagep, swp_entry_t *ent)
2576 {
2577         swp_entry_t entry = { .val = 0 }, *ptr;
2578         struct page *page = NULL;
2579         struct shmem_inode_info *info = SHMEM_I(inode);
2580
2581         if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2582                 goto out;
2583
2584         spin_lock(&info->lock);
2585         ptr = shmem_swp_entry(info, pgoff, NULL);
2586 #ifdef CONFIG_SWAP
2587         if (ptr && ptr->val) {
2588                 entry.val = ptr->val;
2589                 page = find_get_page(&swapper_space, entry.val);
2590         } else
2591 #endif
2592                 page = find_get_page(inode->i_mapping, pgoff);
2593         if (ptr)
2594                 shmem_swp_unmap(ptr);
2595         spin_unlock(&info->lock);
2596 out:
2597         *pagep = page;
2598         *ent = entry;
2599 }
2600 #endif
2601
2602 #else /* !CONFIG_SHMEM */
2603
2604 /*
2605  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2606  *
2607  * This is intended for small system where the benefits of the full
2608  * shmem code (swap-backed and resource-limited) are outweighed by
2609  * their complexity. On systems without swap this code should be
2610  * effectively equivalent, but much lighter weight.
2611  */
2612
2613 #include <linux/ramfs.h>
2614
2615 static struct file_system_type tmpfs_fs_type = {
2616         .name           = "tmpfs",
2617         .get_sb         = ramfs_get_sb,
2618         .kill_sb        = kill_litter_super,
2619 };
2620
2621 int __init init_tmpfs(void)
2622 {
2623         BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2624
2625         shm_mnt = kern_mount(&tmpfs_fs_type);
2626         BUG_ON(IS_ERR(shm_mnt));
2627
2628         return 0;
2629 }
2630
2631 int shmem_unuse(swp_entry_t entry, struct page *page)
2632 {
2633         return 0;
2634 }
2635
2636 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2637 {
2638         return 0;
2639 }
2640
2641 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2642 /**
2643  * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2644  * @inode: the inode to be searched
2645  * @pgoff: the offset to be searched
2646  * @pagep: the pointer for the found page to be stored
2647  * @ent: the pointer for the found swap entry to be stored
2648  *
2649  * If a page is found, refcount of it is incremented. Callers should handle
2650  * these refcount.
2651  */
2652 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2653                                         struct page **pagep, swp_entry_t *ent)
2654 {
2655         struct page *page = NULL;
2656
2657         if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2658                 goto out;
2659         page = find_get_page(inode->i_mapping, pgoff);
2660 out:
2661         *pagep = page;
2662         *ent = (swp_entry_t){ .val = 0 };
2663 }
2664 #endif
2665
2666 #define shmem_vm_ops                            generic_file_vm_ops
2667 #define shmem_file_operations                   ramfs_file_operations
2668 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2669 #define shmem_acct_size(flags, size)            0
2670 #define shmem_unacct_size(flags, size)          do {} while (0)
2671 #define SHMEM_MAX_BYTES                         MAX_LFS_FILESIZE
2672
2673 #endif /* CONFIG_SHMEM */
2674
2675 /* common code */
2676
2677 /**
2678  * shmem_file_setup - get an unlinked file living in tmpfs
2679  * @name: name for dentry (to be seen in /proc/<pid>/maps
2680  * @size: size to be set for the file
2681  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2682  */
2683 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2684 {
2685         int error;
2686         struct file *file;
2687         struct inode *inode;
2688         struct path path;
2689         struct dentry *root;
2690         struct qstr this;
2691
2692         if (IS_ERR(shm_mnt))
2693                 return (void *)shm_mnt;
2694
2695         if (size < 0 || size > SHMEM_MAX_BYTES)
2696                 return ERR_PTR(-EINVAL);
2697
2698         if (shmem_acct_size(flags, size))
2699                 return ERR_PTR(-ENOMEM);
2700
2701         error = -ENOMEM;
2702         this.name = name;
2703         this.len = strlen(name);
2704         this.hash = 0; /* will go */
2705         root = shm_mnt->mnt_root;
2706         path.dentry = d_alloc(root, &this);
2707         if (!path.dentry)
2708                 goto put_memory;
2709         path.mnt = mntget(shm_mnt);
2710
2711         error = -ENOSPC;
2712         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2713         if (!inode)
2714                 goto put_dentry;
2715
2716         d_instantiate(path.dentry, inode);
2717         inode->i_size = size;
2718         inode->i_nlink = 0;     /* It is unlinked */
2719 #ifndef CONFIG_MMU
2720         error = ramfs_nommu_expand_for_mapping(inode, size);
2721         if (error)
2722                 goto put_dentry;
2723 #endif
2724
2725         error = -ENFILE;
2726         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2727                   &shmem_file_operations);
2728         if (!file)
2729                 goto put_dentry;
2730
2731         return file;
2732
2733 put_dentry:
2734         path_put(&path);
2735 put_memory:
2736         shmem_unacct_size(flags, size);
2737         return ERR_PTR(error);
2738 }
2739 EXPORT_SYMBOL_GPL(shmem_file_setup);
2740
2741 /**
2742  * shmem_zero_setup - setup a shared anonymous mapping
2743  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2744  */
2745 int shmem_zero_setup(struct vm_area_struct *vma)
2746 {
2747         struct file *file;
2748         loff_t size = vma->vm_end - vma->vm_start;
2749
2750         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2751         if (IS_ERR(file))
2752                 return PTR_ERR(file);
2753
2754         if (vma->vm_file)
2755                 fput(vma->vm_file);
2756         vma->vm_file = file;
2757         vma->vm_ops = &shmem_vm_ops;
2758         return 0;
2759 }