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