Clean up duplicate includes in mm/
[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  * This file is released under the GPL.
18  */
19
20 /*
21  * This virtual memory filesystem is heavily based on the ramfs. It
22  * extends ramfs by the ability to use swap and honor resource limits
23  * which makes it a completely usable filesystem.
24  */
25
26 #include <linux/module.h>
27 #include <linux/init.h>
28 #include <linux/fs.h>
29 #include <linux/xattr.h>
30 #include <linux/exportfs.h>
31 #include <linux/generic_acl.h>
32 #include <linux/mm.h>
33 #include <linux/mman.h>
34 #include <linux/file.h>
35 #include <linux/swap.h>
36 #include <linux/pagemap.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/backing-dev.h>
40 #include <linux/shmem_fs.h>
41 #include <linux/mount.h>
42 #include <linux/writeback.h>
43 #include <linux/vfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/security.h>
46 #include <linux/swapops.h>
47 #include <linux/mempolicy.h>
48 #include <linux/namei.h>
49 #include <linux/ctype.h>
50 #include <linux/migrate.h>
51 #include <linux/highmem.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/div64.h>
55 #include <asm/pgtable.h>
56
57 /* This magic number is used in glibc for posix shared memory */
58 #define TMPFS_MAGIC     0x01021994
59
60 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
61 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
62 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
63
64 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
65 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
66
67 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
68
69 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
70 #define SHMEM_PAGEIN     VM_READ
71 #define SHMEM_TRUNCATE   VM_WRITE
72
73 /* Definition to limit shmem_truncate's steps between cond_rescheds */
74 #define LATENCY_LIMIT    64
75
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
78
79 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
80 enum sgp_type {
81         SGP_QUICK,      /* don't try more than file page cache lookup */
82         SGP_READ,       /* don't exceed i_size, don't allocate page */
83         SGP_CACHE,      /* don't exceed i_size, may allocate page */
84         SGP_WRITE,      /* may exceed i_size, may allocate page */
85         SGP_FAULT,      /* same as SGP_CACHE, return with page locked */
86 };
87
88 static int shmem_getpage(struct inode *inode, unsigned long idx,
89                          struct page **pagep, enum sgp_type sgp, int *type);
90
91 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
92 {
93         /*
94          * The above definition of ENTRIES_PER_PAGE, and the use of
95          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
96          * might be reconsidered if it ever diverges from PAGE_SIZE.
97          *
98          * __GFP_MOVABLE is masked out as swap vectors cannot move
99          */
100         return alloc_pages((gfp_mask & ~__GFP_MOVABLE) | __GFP_ZERO,
101                                 PAGE_CACHE_SHIFT-PAGE_SHIFT);
102 }
103
104 static inline void shmem_dir_free(struct page *page)
105 {
106         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
107 }
108
109 static struct page **shmem_dir_map(struct page *page)
110 {
111         return (struct page **)kmap_atomic(page, KM_USER0);
112 }
113
114 static inline void shmem_dir_unmap(struct page **dir)
115 {
116         kunmap_atomic(dir, KM_USER0);
117 }
118
119 static swp_entry_t *shmem_swp_map(struct page *page)
120 {
121         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
122 }
123
124 static inline void shmem_swp_balance_unmap(void)
125 {
126         /*
127          * When passing a pointer to an i_direct entry, to code which
128          * also handles indirect entries and so will shmem_swp_unmap,
129          * we must arrange for the preempt count to remain in balance.
130          * What kmap_atomic of a lowmem page does depends on config
131          * and architecture, so pretend to kmap_atomic some lowmem page.
132          */
133         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
134 }
135
136 static inline void shmem_swp_unmap(swp_entry_t *entry)
137 {
138         kunmap_atomic(entry, KM_USER1);
139 }
140
141 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
142 {
143         return sb->s_fs_info;
144 }
145
146 /*
147  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
148  * for shared memory and for shared anonymous (/dev/zero) mappings
149  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
150  * consistent with the pre-accounting of private mappings ...
151  */
152 static inline int shmem_acct_size(unsigned long flags, loff_t size)
153 {
154         return (flags & VM_ACCOUNT)?
155                 security_vm_enough_memory(VM_ACCT(size)): 0;
156 }
157
158 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
159 {
160         if (flags & VM_ACCOUNT)
161                 vm_unacct_memory(VM_ACCT(size));
162 }
163
164 /*
165  * ... whereas tmpfs objects are accounted incrementally as
166  * pages are allocated, in order to allow huge sparse files.
167  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
168  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169  */
170 static inline int shmem_acct_block(unsigned long flags)
171 {
172         return (flags & VM_ACCOUNT)?
173                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
174 }
175
176 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
177 {
178         if (!(flags & VM_ACCOUNT))
179                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
180 }
181
182 static const struct super_operations shmem_ops;
183 static const struct address_space_operations shmem_aops;
184 static const struct file_operations shmem_file_operations;
185 static const struct inode_operations shmem_inode_operations;
186 static const struct inode_operations shmem_dir_inode_operations;
187 static const struct inode_operations shmem_special_inode_operations;
188 static struct vm_operations_struct shmem_vm_ops;
189
190 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
191         .ra_pages       = 0,    /* No readahead */
192         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
193         .unplug_io_fn   = default_unplug_io_fn,
194 };
195
196 static LIST_HEAD(shmem_swaplist);
197 static DEFINE_SPINLOCK(shmem_swaplist_lock);
198
199 static void shmem_free_blocks(struct inode *inode, long pages)
200 {
201         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
202         if (sbinfo->max_blocks) {
203                 spin_lock(&sbinfo->stat_lock);
204                 sbinfo->free_blocks += pages;
205                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
206                 spin_unlock(&sbinfo->stat_lock);
207         }
208 }
209
210 /*
211  * shmem_recalc_inode - recalculate the size of an inode
212  *
213  * @inode: inode to recalc
214  *
215  * We have to calculate the free blocks since the mm can drop
216  * undirtied hole pages behind our back.
217  *
218  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
219  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
220  *
221  * It has to be called with the spinlock held.
222  */
223 static void shmem_recalc_inode(struct inode *inode)
224 {
225         struct shmem_inode_info *info = SHMEM_I(inode);
226         long freed;
227
228         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
229         if (freed > 0) {
230                 info->alloced -= freed;
231                 shmem_unacct_blocks(info->flags, freed);
232                 shmem_free_blocks(inode, freed);
233         }
234 }
235
236 /*
237  * shmem_swp_entry - find the swap vector position in the info structure
238  *
239  * @info:  info structure for the inode
240  * @index: index of the page to find
241  * @page:  optional page to add to the structure. Has to be preset to
242  *         all zeros
243  *
244  * If there is no space allocated yet it will return NULL when
245  * page is NULL, else it will use the page for the needed block,
246  * setting it to NULL on return to indicate that it has been used.
247  *
248  * The swap vector is organized the following way:
249  *
250  * There are SHMEM_NR_DIRECT entries directly stored in the
251  * shmem_inode_info structure. So small files do not need an addional
252  * allocation.
253  *
254  * For pages with index > SHMEM_NR_DIRECT there is the pointer
255  * i_indirect which points to a page which holds in the first half
256  * doubly indirect blocks, in the second half triple indirect blocks:
257  *
258  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
259  * following layout (for SHMEM_NR_DIRECT == 16):
260  *
261  * i_indirect -> dir --> 16-19
262  *            |      +-> 20-23
263  *            |
264  *            +-->dir2 --> 24-27
265  *            |        +-> 28-31
266  *            |        +-> 32-35
267  *            |        +-> 36-39
268  *            |
269  *            +-->dir3 --> 40-43
270  *                     +-> 44-47
271  *                     +-> 48-51
272  *                     +-> 52-55
273  */
274 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
275 {
276         unsigned long offset;
277         struct page **dir;
278         struct page *subdir;
279
280         if (index < SHMEM_NR_DIRECT) {
281                 shmem_swp_balance_unmap();
282                 return info->i_direct+index;
283         }
284         if (!info->i_indirect) {
285                 if (page) {
286                         info->i_indirect = *page;
287                         *page = NULL;
288                 }
289                 return NULL;                    /* need another page */
290         }
291
292         index -= SHMEM_NR_DIRECT;
293         offset = index % ENTRIES_PER_PAGE;
294         index /= ENTRIES_PER_PAGE;
295         dir = shmem_dir_map(info->i_indirect);
296
297         if (index >= ENTRIES_PER_PAGE/2) {
298                 index -= ENTRIES_PER_PAGE/2;
299                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
300                 index %= ENTRIES_PER_PAGE;
301                 subdir = *dir;
302                 if (!subdir) {
303                         if (page) {
304                                 *dir = *page;
305                                 *page = NULL;
306                         }
307                         shmem_dir_unmap(dir);
308                         return NULL;            /* need another page */
309                 }
310                 shmem_dir_unmap(dir);
311                 dir = shmem_dir_map(subdir);
312         }
313
314         dir += index;
315         subdir = *dir;
316         if (!subdir) {
317                 if (!page || !(subdir = *page)) {
318                         shmem_dir_unmap(dir);
319                         return NULL;            /* need a page */
320                 }
321                 *dir = subdir;
322                 *page = NULL;
323         }
324         shmem_dir_unmap(dir);
325         return shmem_swp_map(subdir) + offset;
326 }
327
328 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
329 {
330         long incdec = value? 1: -1;
331
332         entry->val = value;
333         info->swapped += incdec;
334         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
335                 struct page *page = kmap_atomic_to_page(entry);
336                 set_page_private(page, page_private(page) + incdec);
337         }
338 }
339
340 /*
341  * shmem_swp_alloc - get the position of the swap entry for the page.
342  *                   If it does not exist allocate the entry.
343  *
344  * @info:       info structure for the inode
345  * @index:      index of the page to find
346  * @sgp:        check and recheck i_size? skip allocation?
347  */
348 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
349 {
350         struct inode *inode = &info->vfs_inode;
351         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
352         struct page *page = NULL;
353         swp_entry_t *entry;
354
355         if (sgp != SGP_WRITE &&
356             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
357                 return ERR_PTR(-EINVAL);
358
359         while (!(entry = shmem_swp_entry(info, index, &page))) {
360                 if (sgp == SGP_READ)
361                         return shmem_swp_map(ZERO_PAGE(0));
362                 /*
363                  * Test free_blocks against 1 not 0, since we have 1 data
364                  * page (and perhaps indirect index pages) yet to allocate:
365                  * a waste to allocate index if we cannot allocate data.
366                  */
367                 if (sbinfo->max_blocks) {
368                         spin_lock(&sbinfo->stat_lock);
369                         if (sbinfo->free_blocks <= 1) {
370                                 spin_unlock(&sbinfo->stat_lock);
371                                 return ERR_PTR(-ENOSPC);
372                         }
373                         sbinfo->free_blocks--;
374                         inode->i_blocks += BLOCKS_PER_PAGE;
375                         spin_unlock(&sbinfo->stat_lock);
376                 }
377
378                 spin_unlock(&info->lock);
379                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
380                 if (page)
381                         set_page_private(page, 0);
382                 spin_lock(&info->lock);
383
384                 if (!page) {
385                         shmem_free_blocks(inode, 1);
386                         return ERR_PTR(-ENOMEM);
387                 }
388                 if (sgp != SGP_WRITE &&
389                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
390                         entry = ERR_PTR(-EINVAL);
391                         break;
392                 }
393                 if (info->next_index <= index)
394                         info->next_index = index + 1;
395         }
396         if (page) {
397                 /* another task gave its page, or truncated the file */
398                 shmem_free_blocks(inode, 1);
399                 shmem_dir_free(page);
400         }
401         if (info->next_index <= index && !IS_ERR(entry))
402                 info->next_index = index + 1;
403         return entry;
404 }
405
406 /*
407  * shmem_free_swp - free some swap entries in a directory
408  *
409  * @dir:        pointer to the directory
410  * @edir:       pointer after last entry of the directory
411  * @punch_lock: pointer to spinlock when needed for the holepunch case
412  */
413 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
414                                                 spinlock_t *punch_lock)
415 {
416         spinlock_t *punch_unlock = NULL;
417         swp_entry_t *ptr;
418         int freed = 0;
419
420         for (ptr = dir; ptr < edir; ptr++) {
421                 if (ptr->val) {
422                         if (unlikely(punch_lock)) {
423                                 punch_unlock = punch_lock;
424                                 punch_lock = NULL;
425                                 spin_lock(punch_unlock);
426                                 if (!ptr->val)
427                                         continue;
428                         }
429                         free_swap_and_cache(*ptr);
430                         *ptr = (swp_entry_t){0};
431                         freed++;
432                 }
433         }
434         if (punch_unlock)
435                 spin_unlock(punch_unlock);
436         return freed;
437 }
438
439 static int shmem_map_and_free_swp(struct page *subdir, int offset,
440                 int limit, struct page ***dir, spinlock_t *punch_lock)
441 {
442         swp_entry_t *ptr;
443         int freed = 0;
444
445         ptr = shmem_swp_map(subdir);
446         for (; offset < limit; offset += LATENCY_LIMIT) {
447                 int size = limit - offset;
448                 if (size > LATENCY_LIMIT)
449                         size = LATENCY_LIMIT;
450                 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
451                                                         punch_lock);
452                 if (need_resched()) {
453                         shmem_swp_unmap(ptr);
454                         if (*dir) {
455                                 shmem_dir_unmap(*dir);
456                                 *dir = NULL;
457                         }
458                         cond_resched();
459                         ptr = shmem_swp_map(subdir);
460                 }
461         }
462         shmem_swp_unmap(ptr);
463         return freed;
464 }
465
466 static void shmem_free_pages(struct list_head *next)
467 {
468         struct page *page;
469         int freed = 0;
470
471         do {
472                 page = container_of(next, struct page, lru);
473                 next = next->next;
474                 shmem_dir_free(page);
475                 freed++;
476                 if (freed >= LATENCY_LIMIT) {
477                         cond_resched();
478                         freed = 0;
479                 }
480         } while (next);
481 }
482
483 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
484 {
485         struct shmem_inode_info *info = SHMEM_I(inode);
486         unsigned long idx;
487         unsigned long size;
488         unsigned long limit;
489         unsigned long stage;
490         unsigned long diroff;
491         struct page **dir;
492         struct page *topdir;
493         struct page *middir;
494         struct page *subdir;
495         swp_entry_t *ptr;
496         LIST_HEAD(pages_to_free);
497         long nr_pages_to_free = 0;
498         long nr_swaps_freed = 0;
499         int offset;
500         int freed;
501         int punch_hole;
502         spinlock_t *needs_lock;
503         spinlock_t *punch_lock;
504         unsigned long upper_limit;
505
506         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
507         idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
508         if (idx >= info->next_index)
509                 return;
510
511         spin_lock(&info->lock);
512         info->flags |= SHMEM_TRUNCATE;
513         if (likely(end == (loff_t) -1)) {
514                 limit = info->next_index;
515                 upper_limit = SHMEM_MAX_INDEX;
516                 info->next_index = idx;
517                 needs_lock = NULL;
518                 punch_hole = 0;
519         } else {
520                 if (end + 1 >= inode->i_size) { /* we may free a little more */
521                         limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
522                                                         PAGE_CACHE_SHIFT;
523                         upper_limit = SHMEM_MAX_INDEX;
524                 } else {
525                         limit = (end + 1) >> PAGE_CACHE_SHIFT;
526                         upper_limit = limit;
527                 }
528                 needs_lock = &info->lock;
529                 punch_hole = 1;
530         }
531
532         topdir = info->i_indirect;
533         if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
534                 info->i_indirect = NULL;
535                 nr_pages_to_free++;
536                 list_add(&topdir->lru, &pages_to_free);
537         }
538         spin_unlock(&info->lock);
539
540         if (info->swapped && idx < SHMEM_NR_DIRECT) {
541                 ptr = info->i_direct;
542                 size = limit;
543                 if (size > SHMEM_NR_DIRECT)
544                         size = SHMEM_NR_DIRECT;
545                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
546         }
547
548         /*
549          * If there are no indirect blocks or we are punching a hole
550          * below indirect blocks, nothing to be done.
551          */
552         if (!topdir || limit <= SHMEM_NR_DIRECT)
553                 goto done2;
554
555         /*
556          * The truncation case has already dropped info->lock, and we're safe
557          * because i_size and next_index have already been lowered, preventing
558          * access beyond.  But in the punch_hole case, we still need to take
559          * the lock when updating the swap directory, because there might be
560          * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
561          * shmem_writepage.  However, whenever we find we can remove a whole
562          * directory page (not at the misaligned start or end of the range),
563          * we first NULLify its pointer in the level above, and then have no
564          * need to take the lock when updating its contents: needs_lock and
565          * punch_lock (either pointing to info->lock or NULL) manage this.
566          */
567
568         upper_limit -= SHMEM_NR_DIRECT;
569         limit -= SHMEM_NR_DIRECT;
570         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
571         offset = idx % ENTRIES_PER_PAGE;
572         idx -= offset;
573
574         dir = shmem_dir_map(topdir);
575         stage = ENTRIES_PER_PAGEPAGE/2;
576         if (idx < ENTRIES_PER_PAGEPAGE/2) {
577                 middir = topdir;
578                 diroff = idx/ENTRIES_PER_PAGE;
579         } else {
580                 dir += ENTRIES_PER_PAGE/2;
581                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
582                 while (stage <= idx)
583                         stage += ENTRIES_PER_PAGEPAGE;
584                 middir = *dir;
585                 if (*dir) {
586                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
587                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
588                         if (!diroff && !offset && upper_limit >= stage) {
589                                 if (needs_lock) {
590                                         spin_lock(needs_lock);
591                                         *dir = NULL;
592                                         spin_unlock(needs_lock);
593                                         needs_lock = NULL;
594                                 } else
595                                         *dir = NULL;
596                                 nr_pages_to_free++;
597                                 list_add(&middir->lru, &pages_to_free);
598                         }
599                         shmem_dir_unmap(dir);
600                         dir = shmem_dir_map(middir);
601                 } else {
602                         diroff = 0;
603                         offset = 0;
604                         idx = stage;
605                 }
606         }
607
608         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
609                 if (unlikely(idx == stage)) {
610                         shmem_dir_unmap(dir);
611                         dir = shmem_dir_map(topdir) +
612                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
613                         while (!*dir) {
614                                 dir++;
615                                 idx += ENTRIES_PER_PAGEPAGE;
616                                 if (idx >= limit)
617                                         goto done1;
618                         }
619                         stage = idx + ENTRIES_PER_PAGEPAGE;
620                         middir = *dir;
621                         if (punch_hole)
622                                 needs_lock = &info->lock;
623                         if (upper_limit >= stage) {
624                                 if (needs_lock) {
625                                         spin_lock(needs_lock);
626                                         *dir = NULL;
627                                         spin_unlock(needs_lock);
628                                         needs_lock = NULL;
629                                 } else
630                                         *dir = NULL;
631                                 nr_pages_to_free++;
632                                 list_add(&middir->lru, &pages_to_free);
633                         }
634                         shmem_dir_unmap(dir);
635                         cond_resched();
636                         dir = shmem_dir_map(middir);
637                         diroff = 0;
638                 }
639                 punch_lock = needs_lock;
640                 subdir = dir[diroff];
641                 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
642                         if (needs_lock) {
643                                 spin_lock(needs_lock);
644                                 dir[diroff] = NULL;
645                                 spin_unlock(needs_lock);
646                                 punch_lock = NULL;
647                         } else
648                                 dir[diroff] = NULL;
649                         nr_pages_to_free++;
650                         list_add(&subdir->lru, &pages_to_free);
651                 }
652                 if (subdir && page_private(subdir) /* has swap entries */) {
653                         size = limit - idx;
654                         if (size > ENTRIES_PER_PAGE)
655                                 size = ENTRIES_PER_PAGE;
656                         freed = shmem_map_and_free_swp(subdir,
657                                         offset, size, &dir, punch_lock);
658                         if (!dir)
659                                 dir = shmem_dir_map(middir);
660                         nr_swaps_freed += freed;
661                         if (offset || punch_lock) {
662                                 spin_lock(&info->lock);
663                                 set_page_private(subdir,
664                                         page_private(subdir) - freed);
665                                 spin_unlock(&info->lock);
666                         } else
667                                 BUG_ON(page_private(subdir) != freed);
668                 }
669                 offset = 0;
670         }
671 done1:
672         shmem_dir_unmap(dir);
673 done2:
674         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
675                 /*
676                  * Call truncate_inode_pages again: racing shmem_unuse_inode
677                  * may have swizzled a page in from swap since vmtruncate or
678                  * generic_delete_inode did it, before we lowered next_index.
679                  * Also, though shmem_getpage checks i_size before adding to
680                  * cache, no recheck after: so fix the narrow window there too.
681                  *
682                  * Recalling truncate_inode_pages_range and unmap_mapping_range
683                  * every time for punch_hole (which never got a chance to clear
684                  * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
685                  * yet hardly ever necessary: try to optimize them out later.
686                  */
687                 truncate_inode_pages_range(inode->i_mapping, start, end);
688                 if (punch_hole)
689                         unmap_mapping_range(inode->i_mapping, start,
690                                                         end - start, 1);
691         }
692
693         spin_lock(&info->lock);
694         info->flags &= ~SHMEM_TRUNCATE;
695         info->swapped -= nr_swaps_freed;
696         if (nr_pages_to_free)
697                 shmem_free_blocks(inode, nr_pages_to_free);
698         shmem_recalc_inode(inode);
699         spin_unlock(&info->lock);
700
701         /*
702          * Empty swap vector directory pages to be freed?
703          */
704         if (!list_empty(&pages_to_free)) {
705                 pages_to_free.prev->next = NULL;
706                 shmem_free_pages(pages_to_free.next);
707         }
708 }
709
710 static void shmem_truncate(struct inode *inode)
711 {
712         shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
713 }
714
715 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
716 {
717         struct inode *inode = dentry->d_inode;
718         struct page *page = NULL;
719         int error;
720
721         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
722                 if (attr->ia_size < inode->i_size) {
723                         /*
724                          * If truncating down to a partial page, then
725                          * if that page is already allocated, hold it
726                          * in memory until the truncation is over, so
727                          * truncate_partial_page cannnot miss it were
728                          * it assigned to swap.
729                          */
730                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
731                                 (void) shmem_getpage(inode,
732                                         attr->ia_size>>PAGE_CACHE_SHIFT,
733                                                 &page, SGP_READ, NULL);
734                         }
735                         /*
736                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
737                          * detect if any pages might have been added to cache
738                          * after truncate_inode_pages.  But we needn't bother
739                          * if it's being fully truncated to zero-length: the
740                          * nrpages check is efficient enough in that case.
741                          */
742                         if (attr->ia_size) {
743                                 struct shmem_inode_info *info = SHMEM_I(inode);
744                                 spin_lock(&info->lock);
745                                 info->flags &= ~SHMEM_PAGEIN;
746                                 spin_unlock(&info->lock);
747                         }
748                 }
749         }
750
751         error = inode_change_ok(inode, attr);
752         if (!error)
753                 error = inode_setattr(inode, attr);
754 #ifdef CONFIG_TMPFS_POSIX_ACL
755         if (!error && (attr->ia_valid & ATTR_MODE))
756                 error = generic_acl_chmod(inode, &shmem_acl_ops);
757 #endif
758         if (page)
759                 page_cache_release(page);
760         return error;
761 }
762
763 static void shmem_delete_inode(struct inode *inode)
764 {
765         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
766         struct shmem_inode_info *info = SHMEM_I(inode);
767
768         if (inode->i_op->truncate == shmem_truncate) {
769                 truncate_inode_pages(inode->i_mapping, 0);
770                 shmem_unacct_size(info->flags, inode->i_size);
771                 inode->i_size = 0;
772                 shmem_truncate(inode);
773                 if (!list_empty(&info->swaplist)) {
774                         spin_lock(&shmem_swaplist_lock);
775                         list_del_init(&info->swaplist);
776                         spin_unlock(&shmem_swaplist_lock);
777                 }
778         }
779         BUG_ON(inode->i_blocks);
780         if (sbinfo->max_inodes) {
781                 spin_lock(&sbinfo->stat_lock);
782                 sbinfo->free_inodes++;
783                 spin_unlock(&sbinfo->stat_lock);
784         }
785         clear_inode(inode);
786 }
787
788 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
789 {
790         swp_entry_t *ptr;
791
792         for (ptr = dir; ptr < edir; ptr++) {
793                 if (ptr->val == entry.val)
794                         return ptr - dir;
795         }
796         return -1;
797 }
798
799 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
800 {
801         struct inode *inode;
802         unsigned long idx;
803         unsigned long size;
804         unsigned long limit;
805         unsigned long stage;
806         struct page **dir;
807         struct page *subdir;
808         swp_entry_t *ptr;
809         int offset;
810
811         idx = 0;
812         ptr = info->i_direct;
813         spin_lock(&info->lock);
814         limit = info->next_index;
815         size = limit;
816         if (size > SHMEM_NR_DIRECT)
817                 size = SHMEM_NR_DIRECT;
818         offset = shmem_find_swp(entry, ptr, ptr+size);
819         if (offset >= 0) {
820                 shmem_swp_balance_unmap();
821                 goto found;
822         }
823         if (!info->i_indirect)
824                 goto lost2;
825
826         dir = shmem_dir_map(info->i_indirect);
827         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
828
829         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
830                 if (unlikely(idx == stage)) {
831                         shmem_dir_unmap(dir-1);
832                         dir = shmem_dir_map(info->i_indirect) +
833                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
834                         while (!*dir) {
835                                 dir++;
836                                 idx += ENTRIES_PER_PAGEPAGE;
837                                 if (idx >= limit)
838                                         goto lost1;
839                         }
840                         stage = idx + ENTRIES_PER_PAGEPAGE;
841                         subdir = *dir;
842                         shmem_dir_unmap(dir);
843                         dir = shmem_dir_map(subdir);
844                 }
845                 subdir = *dir;
846                 if (subdir && page_private(subdir)) {
847                         ptr = shmem_swp_map(subdir);
848                         size = limit - idx;
849                         if (size > ENTRIES_PER_PAGE)
850                                 size = ENTRIES_PER_PAGE;
851                         offset = shmem_find_swp(entry, ptr, ptr+size);
852                         if (offset >= 0) {
853                                 shmem_dir_unmap(dir);
854                                 goto found;
855                         }
856                         shmem_swp_unmap(ptr);
857                 }
858         }
859 lost1:
860         shmem_dir_unmap(dir-1);
861 lost2:
862         spin_unlock(&info->lock);
863         return 0;
864 found:
865         idx += offset;
866         inode = &info->vfs_inode;
867         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
868                 info->flags |= SHMEM_PAGEIN;
869                 shmem_swp_set(info, ptr + offset, 0);
870         }
871         shmem_swp_unmap(ptr);
872         spin_unlock(&info->lock);
873         /*
874          * Decrement swap count even when the entry is left behind:
875          * try_to_unuse will skip over mms, then reincrement count.
876          */
877         swap_free(entry);
878         return 1;
879 }
880
881 /*
882  * shmem_unuse() search for an eventually swapped out shmem page.
883  */
884 int shmem_unuse(swp_entry_t entry, struct page *page)
885 {
886         struct list_head *p, *next;
887         struct shmem_inode_info *info;
888         int found = 0;
889
890         spin_lock(&shmem_swaplist_lock);
891         list_for_each_safe(p, next, &shmem_swaplist) {
892                 info = list_entry(p, struct shmem_inode_info, swaplist);
893                 if (!info->swapped)
894                         list_del_init(&info->swaplist);
895                 else if (shmem_unuse_inode(info, entry, page)) {
896                         /* move head to start search for next from here */
897                         list_move_tail(&shmem_swaplist, &info->swaplist);
898                         found = 1;
899                         break;
900                 }
901         }
902         spin_unlock(&shmem_swaplist_lock);
903         return found;
904 }
905
906 /*
907  * Move the page from the page cache to the swap cache.
908  */
909 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
910 {
911         struct shmem_inode_info *info;
912         swp_entry_t *entry, swap;
913         struct address_space *mapping;
914         unsigned long index;
915         struct inode *inode;
916
917         BUG_ON(!PageLocked(page));
918         BUG_ON(page_mapped(page));
919
920         mapping = page->mapping;
921         index = page->index;
922         inode = mapping->host;
923         info = SHMEM_I(inode);
924         if (info->flags & VM_LOCKED)
925                 goto redirty;
926         swap = get_swap_page();
927         if (!swap.val)
928                 goto redirty;
929
930         spin_lock(&info->lock);
931         shmem_recalc_inode(inode);
932         if (index >= info->next_index) {
933                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
934                 goto unlock;
935         }
936         entry = shmem_swp_entry(info, index, NULL);
937         BUG_ON(!entry);
938         BUG_ON(entry->val);
939
940         if (move_to_swap_cache(page, swap) == 0) {
941                 shmem_swp_set(info, entry, swap.val);
942                 shmem_swp_unmap(entry);
943                 spin_unlock(&info->lock);
944                 if (list_empty(&info->swaplist)) {
945                         spin_lock(&shmem_swaplist_lock);
946                         /* move instead of add in case we're racing */
947                         list_move_tail(&info->swaplist, &shmem_swaplist);
948                         spin_unlock(&shmem_swaplist_lock);
949                 }
950                 unlock_page(page);
951                 return 0;
952         }
953
954         shmem_swp_unmap(entry);
955 unlock:
956         spin_unlock(&info->lock);
957         swap_free(swap);
958 redirty:
959         set_page_dirty(page);
960         return AOP_WRITEPAGE_ACTIVATE;  /* Return with the page locked */
961 }
962
963 #ifdef CONFIG_NUMA
964 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
965 {
966         char *nodelist = strchr(value, ':');
967         int err = 1;
968
969         if (nodelist) {
970                 /* NUL-terminate policy string */
971                 *nodelist++ = '\0';
972                 if (nodelist_parse(nodelist, *policy_nodes))
973                         goto out;
974                 if (!nodes_subset(*policy_nodes, node_online_map))
975                         goto out;
976         }
977         if (!strcmp(value, "default")) {
978                 *policy = MPOL_DEFAULT;
979                 /* Don't allow a nodelist */
980                 if (!nodelist)
981                         err = 0;
982         } else if (!strcmp(value, "prefer")) {
983                 *policy = MPOL_PREFERRED;
984                 /* Insist on a nodelist of one node only */
985                 if (nodelist) {
986                         char *rest = nodelist;
987                         while (isdigit(*rest))
988                                 rest++;
989                         if (!*rest)
990                                 err = 0;
991                 }
992         } else if (!strcmp(value, "bind")) {
993                 *policy = MPOL_BIND;
994                 /* Insist on a nodelist */
995                 if (nodelist)
996                         err = 0;
997         } else if (!strcmp(value, "interleave")) {
998                 *policy = MPOL_INTERLEAVE;
999                 /* Default to nodes online if no nodelist */
1000                 if (!nodelist)
1001                         *policy_nodes = node_online_map;
1002                 err = 0;
1003         }
1004 out:
1005         /* Restore string for error message */
1006         if (nodelist)
1007                 *--nodelist = ':';
1008         return err;
1009 }
1010
1011 static struct page *shmem_swapin_async(struct shared_policy *p,
1012                                        swp_entry_t entry, unsigned long idx)
1013 {
1014         struct page *page;
1015         struct vm_area_struct pvma;
1016
1017         /* Create a pseudo vma that just contains the policy */
1018         memset(&pvma, 0, sizeof(struct vm_area_struct));
1019         pvma.vm_end = PAGE_SIZE;
1020         pvma.vm_pgoff = idx;
1021         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
1022         page = read_swap_cache_async(entry, &pvma, 0);
1023         mpol_free(pvma.vm_policy);
1024         return page;
1025 }
1026
1027 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
1028                           unsigned long idx)
1029 {
1030         struct shared_policy *p = &info->policy;
1031         int i, num;
1032         struct page *page;
1033         unsigned long offset;
1034
1035         num = valid_swaphandles(entry, &offset);
1036         for (i = 0; i < num; offset++, i++) {
1037                 page = shmem_swapin_async(p,
1038                                 swp_entry(swp_type(entry), offset), idx);
1039                 if (!page)
1040                         break;
1041                 page_cache_release(page);
1042         }
1043         lru_add_drain();        /* Push any new pages onto the LRU now */
1044         return shmem_swapin_async(p, entry, idx);
1045 }
1046
1047 static struct page *
1048 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
1049                  unsigned long idx)
1050 {
1051         struct vm_area_struct pvma;
1052         struct page *page;
1053
1054         memset(&pvma, 0, sizeof(struct vm_area_struct));
1055         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1056         pvma.vm_pgoff = idx;
1057         pvma.vm_end = PAGE_SIZE;
1058         page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
1059         mpol_free(pvma.vm_policy);
1060         return page;
1061 }
1062 #else
1063 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
1064 {
1065         return 1;
1066 }
1067
1068 static inline struct page *
1069 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
1070 {
1071         swapin_readahead(entry, 0, NULL);
1072         return read_swap_cache_async(entry, NULL, 0);
1073 }
1074
1075 static inline struct page *
1076 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
1077 {
1078         return alloc_page(gfp | __GFP_ZERO);
1079 }
1080 #endif
1081
1082 /*
1083  * shmem_getpage - either get the page from swap or allocate a new one
1084  *
1085  * If we allocate a new one we do not mark it dirty. That's up to the
1086  * vm. If we swap it in we mark it dirty since we also free the swap
1087  * entry since a page cannot live in both the swap and page cache
1088  */
1089 static int shmem_getpage(struct inode *inode, unsigned long idx,
1090                         struct page **pagep, enum sgp_type sgp, int *type)
1091 {
1092         struct address_space *mapping = inode->i_mapping;
1093         struct shmem_inode_info *info = SHMEM_I(inode);
1094         struct shmem_sb_info *sbinfo;
1095         struct page *filepage = *pagep;
1096         struct page *swappage;
1097         swp_entry_t *entry;
1098         swp_entry_t swap;
1099         int error;
1100
1101         if (idx >= SHMEM_MAX_INDEX)
1102                 return -EFBIG;
1103
1104         if (type)
1105                 *type = 0;
1106
1107         /*
1108          * Normally, filepage is NULL on entry, and either found
1109          * uptodate immediately, or allocated and zeroed, or read
1110          * in under swappage, which is then assigned to filepage.
1111          * But shmem_readpage and shmem_prepare_write pass in a locked
1112          * filepage, which may be found not uptodate by other callers
1113          * too, and may need to be copied from the swappage read in.
1114          */
1115 repeat:
1116         if (!filepage)
1117                 filepage = find_lock_page(mapping, idx);
1118         if (filepage && PageUptodate(filepage))
1119                 goto done;
1120         error = 0;
1121         if (sgp == SGP_QUICK)
1122                 goto failed;
1123
1124         spin_lock(&info->lock);
1125         shmem_recalc_inode(inode);
1126         entry = shmem_swp_alloc(info, idx, sgp);
1127         if (IS_ERR(entry)) {
1128                 spin_unlock(&info->lock);
1129                 error = PTR_ERR(entry);
1130                 goto failed;
1131         }
1132         swap = *entry;
1133
1134         if (swap.val) {
1135                 /* Look it up and read it in.. */
1136                 swappage = lookup_swap_cache(swap);
1137                 if (!swappage) {
1138                         shmem_swp_unmap(entry);
1139                         /* here we actually do the io */
1140                         if (type && !(*type & VM_FAULT_MAJOR)) {
1141                                 __count_vm_event(PGMAJFAULT);
1142                                 *type |= VM_FAULT_MAJOR;
1143                         }
1144                         spin_unlock(&info->lock);
1145                         swappage = shmem_swapin(info, swap, idx);
1146                         if (!swappage) {
1147                                 spin_lock(&info->lock);
1148                                 entry = shmem_swp_alloc(info, idx, sgp);
1149                                 if (IS_ERR(entry))
1150                                         error = PTR_ERR(entry);
1151                                 else {
1152                                         if (entry->val == swap.val)
1153                                                 error = -ENOMEM;
1154                                         shmem_swp_unmap(entry);
1155                                 }
1156                                 spin_unlock(&info->lock);
1157                                 if (error)
1158                                         goto failed;
1159                                 goto repeat;
1160                         }
1161                         wait_on_page_locked(swappage);
1162                         page_cache_release(swappage);
1163                         goto repeat;
1164                 }
1165
1166                 /* We have to do this with page locked to prevent races */
1167                 if (TestSetPageLocked(swappage)) {
1168                         shmem_swp_unmap(entry);
1169                         spin_unlock(&info->lock);
1170                         wait_on_page_locked(swappage);
1171                         page_cache_release(swappage);
1172                         goto repeat;
1173                 }
1174                 if (PageWriteback(swappage)) {
1175                         shmem_swp_unmap(entry);
1176                         spin_unlock(&info->lock);
1177                         wait_on_page_writeback(swappage);
1178                         unlock_page(swappage);
1179                         page_cache_release(swappage);
1180                         goto repeat;
1181                 }
1182                 if (!PageUptodate(swappage)) {
1183                         shmem_swp_unmap(entry);
1184                         spin_unlock(&info->lock);
1185                         unlock_page(swappage);
1186                         page_cache_release(swappage);
1187                         error = -EIO;
1188                         goto failed;
1189                 }
1190
1191                 if (filepage) {
1192                         shmem_swp_set(info, entry, 0);
1193                         shmem_swp_unmap(entry);
1194                         delete_from_swap_cache(swappage);
1195                         spin_unlock(&info->lock);
1196                         copy_highpage(filepage, swappage);
1197                         unlock_page(swappage);
1198                         page_cache_release(swappage);
1199                         flush_dcache_page(filepage);
1200                         SetPageUptodate(filepage);
1201                         set_page_dirty(filepage);
1202                         swap_free(swap);
1203                 } else if (!(error = move_from_swap_cache(
1204                                 swappage, idx, mapping))) {
1205                         info->flags |= SHMEM_PAGEIN;
1206                         shmem_swp_set(info, entry, 0);
1207                         shmem_swp_unmap(entry);
1208                         spin_unlock(&info->lock);
1209                         filepage = swappage;
1210                         swap_free(swap);
1211                 } else {
1212                         shmem_swp_unmap(entry);
1213                         spin_unlock(&info->lock);
1214                         unlock_page(swappage);
1215                         page_cache_release(swappage);
1216                         if (error == -ENOMEM) {
1217                                 /* let kswapd refresh zone for GFP_ATOMICs */
1218                                 congestion_wait(WRITE, HZ/50);
1219                         }
1220                         goto repeat;
1221                 }
1222         } else if (sgp == SGP_READ && !filepage) {
1223                 shmem_swp_unmap(entry);
1224                 filepage = find_get_page(mapping, idx);
1225                 if (filepage &&
1226                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1227                         spin_unlock(&info->lock);
1228                         wait_on_page_locked(filepage);
1229                         page_cache_release(filepage);
1230                         filepage = NULL;
1231                         goto repeat;
1232                 }
1233                 spin_unlock(&info->lock);
1234         } else {
1235                 shmem_swp_unmap(entry);
1236                 sbinfo = SHMEM_SB(inode->i_sb);
1237                 if (sbinfo->max_blocks) {
1238                         spin_lock(&sbinfo->stat_lock);
1239                         if (sbinfo->free_blocks == 0 ||
1240                             shmem_acct_block(info->flags)) {
1241                                 spin_unlock(&sbinfo->stat_lock);
1242                                 spin_unlock(&info->lock);
1243                                 error = -ENOSPC;
1244                                 goto failed;
1245                         }
1246                         sbinfo->free_blocks--;
1247                         inode->i_blocks += BLOCKS_PER_PAGE;
1248                         spin_unlock(&sbinfo->stat_lock);
1249                 } else if (shmem_acct_block(info->flags)) {
1250                         spin_unlock(&info->lock);
1251                         error = -ENOSPC;
1252                         goto failed;
1253                 }
1254
1255                 if (!filepage) {
1256                         spin_unlock(&info->lock);
1257                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1258                                                     info,
1259                                                     idx);
1260                         if (!filepage) {
1261                                 shmem_unacct_blocks(info->flags, 1);
1262                                 shmem_free_blocks(inode, 1);
1263                                 error = -ENOMEM;
1264                                 goto failed;
1265                         }
1266
1267                         spin_lock(&info->lock);
1268                         entry = shmem_swp_alloc(info, idx, sgp);
1269                         if (IS_ERR(entry))
1270                                 error = PTR_ERR(entry);
1271                         else {
1272                                 swap = *entry;
1273                                 shmem_swp_unmap(entry);
1274                         }
1275                         if (error || swap.val || 0 != add_to_page_cache_lru(
1276                                         filepage, mapping, idx, GFP_ATOMIC)) {
1277                                 spin_unlock(&info->lock);
1278                                 page_cache_release(filepage);
1279                                 shmem_unacct_blocks(info->flags, 1);
1280                                 shmem_free_blocks(inode, 1);
1281                                 filepage = NULL;
1282                                 if (error)
1283                                         goto failed;
1284                                 goto repeat;
1285                         }
1286                         info->flags |= SHMEM_PAGEIN;
1287                 }
1288
1289                 info->alloced++;
1290                 spin_unlock(&info->lock);
1291                 flush_dcache_page(filepage);
1292                 SetPageUptodate(filepage);
1293         }
1294 done:
1295         if (*pagep != filepage) {
1296                 *pagep = filepage;
1297                 if (sgp != SGP_FAULT)
1298                         unlock_page(filepage);
1299
1300         }
1301         return 0;
1302
1303 failed:
1304         if (*pagep != filepage) {
1305                 unlock_page(filepage);
1306                 page_cache_release(filepage);
1307         }
1308         return error;
1309 }
1310
1311 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1312 {
1313         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1314         int error;
1315         int ret;
1316
1317         if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1318                 return VM_FAULT_SIGBUS;
1319
1320         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_FAULT, &ret);
1321         if (error)
1322                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1323
1324         mark_page_accessed(vmf->page);
1325         return ret | VM_FAULT_LOCKED;
1326 }
1327
1328 #ifdef CONFIG_NUMA
1329 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1330 {
1331         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1332         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1333 }
1334
1335 struct mempolicy *
1336 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1337 {
1338         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1339         unsigned long idx;
1340
1341         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1342         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1343 }
1344 #endif
1345
1346 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1347 {
1348         struct inode *inode = file->f_path.dentry->d_inode;
1349         struct shmem_inode_info *info = SHMEM_I(inode);
1350         int retval = -ENOMEM;
1351
1352         spin_lock(&info->lock);
1353         if (lock && !(info->flags & VM_LOCKED)) {
1354                 if (!user_shm_lock(inode->i_size, user))
1355                         goto out_nomem;
1356                 info->flags |= VM_LOCKED;
1357         }
1358         if (!lock && (info->flags & VM_LOCKED) && user) {
1359                 user_shm_unlock(inode->i_size, user);
1360                 info->flags &= ~VM_LOCKED;
1361         }
1362         retval = 0;
1363 out_nomem:
1364         spin_unlock(&info->lock);
1365         return retval;
1366 }
1367
1368 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1369 {
1370         file_accessed(file);
1371         vma->vm_ops = &shmem_vm_ops;
1372         vma->vm_flags |= VM_CAN_NONLINEAR;
1373         return 0;
1374 }
1375
1376 static struct inode *
1377 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1378 {
1379         struct inode *inode;
1380         struct shmem_inode_info *info;
1381         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1382
1383         if (sbinfo->max_inodes) {
1384                 spin_lock(&sbinfo->stat_lock);
1385                 if (!sbinfo->free_inodes) {
1386                         spin_unlock(&sbinfo->stat_lock);
1387                         return NULL;
1388                 }
1389                 sbinfo->free_inodes--;
1390                 spin_unlock(&sbinfo->stat_lock);
1391         }
1392
1393         inode = new_inode(sb);
1394         if (inode) {
1395                 inode->i_mode = mode;
1396                 inode->i_uid = current->fsuid;
1397                 inode->i_gid = current->fsgid;
1398                 inode->i_blocks = 0;
1399                 inode->i_mapping->a_ops = &shmem_aops;
1400                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1401                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1402                 inode->i_generation = get_seconds();
1403                 info = SHMEM_I(inode);
1404                 memset(info, 0, (char *)inode - (char *)info);
1405                 spin_lock_init(&info->lock);
1406                 INIT_LIST_HEAD(&info->swaplist);
1407
1408                 switch (mode & S_IFMT) {
1409                 default:
1410                         inode->i_op = &shmem_special_inode_operations;
1411                         init_special_inode(inode, mode, dev);
1412                         break;
1413                 case S_IFREG:
1414                         inode->i_op = &shmem_inode_operations;
1415                         inode->i_fop = &shmem_file_operations;
1416                         mpol_shared_policy_init(&info->policy, sbinfo->policy,
1417                                                         &sbinfo->policy_nodes);
1418                         break;
1419                 case S_IFDIR:
1420                         inc_nlink(inode);
1421                         /* Some things misbehave if size == 0 on a directory */
1422                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1423                         inode->i_op = &shmem_dir_inode_operations;
1424                         inode->i_fop = &simple_dir_operations;
1425                         break;
1426                 case S_IFLNK:
1427                         /*
1428                          * Must not load anything in the rbtree,
1429                          * mpol_free_shared_policy will not be called.
1430                          */
1431                         mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1432                                                 NULL);
1433                         break;
1434                 }
1435         } else if (sbinfo->max_inodes) {
1436                 spin_lock(&sbinfo->stat_lock);
1437                 sbinfo->free_inodes++;
1438                 spin_unlock(&sbinfo->stat_lock);
1439         }
1440         return inode;
1441 }
1442
1443 #ifdef CONFIG_TMPFS
1444 static const struct inode_operations shmem_symlink_inode_operations;
1445 static const struct inode_operations shmem_symlink_inline_operations;
1446
1447 /*
1448  * Normally tmpfs avoids the use of shmem_readpage and shmem_prepare_write;
1449  * but providing them allows a tmpfs file to be used for splice, sendfile, and
1450  * below the loop driver, in the generic fashion that many filesystems support.
1451  */
1452 static int shmem_readpage(struct file *file, struct page *page)
1453 {
1454         struct inode *inode = page->mapping->host;
1455         int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1456         unlock_page(page);
1457         return error;
1458 }
1459
1460 static int
1461 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1462 {
1463         struct inode *inode = page->mapping->host;
1464         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1465 }
1466
1467 static ssize_t
1468 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1469 {
1470         struct inode    *inode = file->f_path.dentry->d_inode;
1471         loff_t          pos;
1472         unsigned long   written;
1473         ssize_t         err;
1474
1475         if ((ssize_t) count < 0)
1476                 return -EINVAL;
1477
1478         if (!access_ok(VERIFY_READ, buf, count))
1479                 return -EFAULT;
1480
1481         mutex_lock(&inode->i_mutex);
1482
1483         pos = *ppos;
1484         written = 0;
1485
1486         err = generic_write_checks(file, &pos, &count, 0);
1487         if (err || !count)
1488                 goto out;
1489
1490         err = remove_suid(file->f_path.dentry);
1491         if (err)
1492                 goto out;
1493
1494         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1495
1496         do {
1497                 struct page *page = NULL;
1498                 unsigned long bytes, index, offset;
1499                 char *kaddr;
1500                 int left;
1501
1502                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1503                 index = pos >> PAGE_CACHE_SHIFT;
1504                 bytes = PAGE_CACHE_SIZE - offset;
1505                 if (bytes > count)
1506                         bytes = count;
1507
1508                 /*
1509                  * We don't hold page lock across copy from user -
1510                  * what would it guard against? - so no deadlock here.
1511                  * But it still may be a good idea to prefault below.
1512                  */
1513
1514                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1515                 if (err)
1516                         break;
1517
1518                 left = bytes;
1519                 if (PageHighMem(page)) {
1520                         volatile unsigned char dummy;
1521                         __get_user(dummy, buf);
1522                         __get_user(dummy, buf + bytes - 1);
1523
1524                         kaddr = kmap_atomic(page, KM_USER0);
1525                         left = __copy_from_user_inatomic(kaddr + offset,
1526                                                         buf, bytes);
1527                         kunmap_atomic(kaddr, KM_USER0);
1528                 }
1529                 if (left) {
1530                         kaddr = kmap(page);
1531                         left = __copy_from_user(kaddr + offset, buf, bytes);
1532                         kunmap(page);
1533                 }
1534
1535                 written += bytes;
1536                 count -= bytes;
1537                 pos += bytes;
1538                 buf += bytes;
1539                 if (pos > inode->i_size)
1540                         i_size_write(inode, pos);
1541
1542                 flush_dcache_page(page);
1543                 set_page_dirty(page);
1544                 mark_page_accessed(page);
1545                 page_cache_release(page);
1546
1547                 if (left) {
1548                         pos -= left;
1549                         written -= left;
1550                         err = -EFAULT;
1551                         break;
1552                 }
1553
1554                 /*
1555                  * Our dirty pages are not counted in nr_dirty,
1556                  * and we do not attempt to balance dirty pages.
1557                  */
1558
1559                 cond_resched();
1560         } while (count);
1561
1562         *ppos = pos;
1563         if (written)
1564                 err = written;
1565 out:
1566         mutex_unlock(&inode->i_mutex);
1567         return err;
1568 }
1569
1570 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1571 {
1572         struct inode *inode = filp->f_path.dentry->d_inode;
1573         struct address_space *mapping = inode->i_mapping;
1574         unsigned long index, offset;
1575
1576         index = *ppos >> PAGE_CACHE_SHIFT;
1577         offset = *ppos & ~PAGE_CACHE_MASK;
1578
1579         for (;;) {
1580                 struct page *page = NULL;
1581                 unsigned long end_index, nr, ret;
1582                 loff_t i_size = i_size_read(inode);
1583
1584                 end_index = i_size >> PAGE_CACHE_SHIFT;
1585                 if (index > end_index)
1586                         break;
1587                 if (index == end_index) {
1588                         nr = i_size & ~PAGE_CACHE_MASK;
1589                         if (nr <= offset)
1590                                 break;
1591                 }
1592
1593                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1594                 if (desc->error) {
1595                         if (desc->error == -EINVAL)
1596                                 desc->error = 0;
1597                         break;
1598                 }
1599
1600                 /*
1601                  * We must evaluate after, since reads (unlike writes)
1602                  * are called without i_mutex protection against truncate
1603                  */
1604                 nr = PAGE_CACHE_SIZE;
1605                 i_size = i_size_read(inode);
1606                 end_index = i_size >> PAGE_CACHE_SHIFT;
1607                 if (index == end_index) {
1608                         nr = i_size & ~PAGE_CACHE_MASK;
1609                         if (nr <= offset) {
1610                                 if (page)
1611                                         page_cache_release(page);
1612                                 break;
1613                         }
1614                 }
1615                 nr -= offset;
1616
1617                 if (page) {
1618                         /*
1619                          * If users can be writing to this page using arbitrary
1620                          * virtual addresses, take care about potential aliasing
1621                          * before reading the page on the kernel side.
1622                          */
1623                         if (mapping_writably_mapped(mapping))
1624                                 flush_dcache_page(page);
1625                         /*
1626                          * Mark the page accessed if we read the beginning.
1627                          */
1628                         if (!offset)
1629                                 mark_page_accessed(page);
1630                 } else {
1631                         page = ZERO_PAGE(0);
1632                         page_cache_get(page);
1633                 }
1634
1635                 /*
1636                  * Ok, we have the page, and it's up-to-date, so
1637                  * now we can copy it to user space...
1638                  *
1639                  * The actor routine returns how many bytes were actually used..
1640                  * NOTE! This may not be the same as how much of a user buffer
1641                  * we filled up (we may be padding etc), so we can only update
1642                  * "pos" here (the actor routine has to update the user buffer
1643                  * pointers and the remaining count).
1644                  */
1645                 ret = actor(desc, page, offset, nr);
1646                 offset += ret;
1647                 index += offset >> PAGE_CACHE_SHIFT;
1648                 offset &= ~PAGE_CACHE_MASK;
1649
1650                 page_cache_release(page);
1651                 if (ret != nr || !desc->count)
1652                         break;
1653
1654                 cond_resched();
1655         }
1656
1657         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1658         file_accessed(filp);
1659 }
1660
1661 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1662 {
1663         read_descriptor_t desc;
1664
1665         if ((ssize_t) count < 0)
1666                 return -EINVAL;
1667         if (!access_ok(VERIFY_WRITE, buf, count))
1668                 return -EFAULT;
1669         if (!count)
1670                 return 0;
1671
1672         desc.written = 0;
1673         desc.count = count;
1674         desc.arg.buf = buf;
1675         desc.error = 0;
1676
1677         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1678         if (desc.written)
1679                 return desc.written;
1680         return desc.error;
1681 }
1682
1683 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1684 {
1685         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1686
1687         buf->f_type = TMPFS_MAGIC;
1688         buf->f_bsize = PAGE_CACHE_SIZE;
1689         buf->f_namelen = NAME_MAX;
1690         spin_lock(&sbinfo->stat_lock);
1691         if (sbinfo->max_blocks) {
1692                 buf->f_blocks = sbinfo->max_blocks;
1693                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1694         }
1695         if (sbinfo->max_inodes) {
1696                 buf->f_files = sbinfo->max_inodes;
1697                 buf->f_ffree = sbinfo->free_inodes;
1698         }
1699         /* else leave those fields 0 like simple_statfs */
1700         spin_unlock(&sbinfo->stat_lock);
1701         return 0;
1702 }
1703
1704 /*
1705  * File creation. Allocate an inode, and we're done..
1706  */
1707 static int
1708 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1709 {
1710         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1711         int error = -ENOSPC;
1712
1713         if (inode) {
1714                 error = security_inode_init_security(inode, dir, NULL, NULL,
1715                                                      NULL);
1716                 if (error) {
1717                         if (error != -EOPNOTSUPP) {
1718                                 iput(inode);
1719                                 return error;
1720                         }
1721                 }
1722                 error = shmem_acl_init(inode, dir);
1723                 if (error) {
1724                         iput(inode);
1725                         return error;
1726                 }
1727                 if (dir->i_mode & S_ISGID) {
1728                         inode->i_gid = dir->i_gid;
1729                         if (S_ISDIR(mode))
1730                                 inode->i_mode |= S_ISGID;
1731                 }
1732                 dir->i_size += BOGO_DIRENT_SIZE;
1733                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1734                 d_instantiate(dentry, inode);
1735                 dget(dentry); /* Extra count - pin the dentry in core */
1736         }
1737         return error;
1738 }
1739
1740 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1741 {
1742         int error;
1743
1744         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1745                 return error;
1746         inc_nlink(dir);
1747         return 0;
1748 }
1749
1750 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1751                 struct nameidata *nd)
1752 {
1753         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1754 }
1755
1756 /*
1757  * Link a file..
1758  */
1759 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1760 {
1761         struct inode *inode = old_dentry->d_inode;
1762         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1763
1764         /*
1765          * No ordinary (disk based) filesystem counts links as inodes;
1766          * but each new link needs a new dentry, pinning lowmem, and
1767          * tmpfs dentries cannot be pruned until they are unlinked.
1768          */
1769         if (sbinfo->max_inodes) {
1770                 spin_lock(&sbinfo->stat_lock);
1771                 if (!sbinfo->free_inodes) {
1772                         spin_unlock(&sbinfo->stat_lock);
1773                         return -ENOSPC;
1774                 }
1775                 sbinfo->free_inodes--;
1776                 spin_unlock(&sbinfo->stat_lock);
1777         }
1778
1779         dir->i_size += BOGO_DIRENT_SIZE;
1780         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1781         inc_nlink(inode);
1782         atomic_inc(&inode->i_count);    /* New dentry reference */
1783         dget(dentry);           /* Extra pinning count for the created dentry */
1784         d_instantiate(dentry, inode);
1785         return 0;
1786 }
1787
1788 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1789 {
1790         struct inode *inode = dentry->d_inode;
1791
1792         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1793                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1794                 if (sbinfo->max_inodes) {
1795                         spin_lock(&sbinfo->stat_lock);
1796                         sbinfo->free_inodes++;
1797                         spin_unlock(&sbinfo->stat_lock);
1798                 }
1799         }
1800
1801         dir->i_size -= BOGO_DIRENT_SIZE;
1802         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1803         drop_nlink(inode);
1804         dput(dentry);   /* Undo the count from "create" - this does all the work */
1805         return 0;
1806 }
1807
1808 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1809 {
1810         if (!simple_empty(dentry))
1811                 return -ENOTEMPTY;
1812
1813         drop_nlink(dentry->d_inode);
1814         drop_nlink(dir);
1815         return shmem_unlink(dir, dentry);
1816 }
1817
1818 /*
1819  * The VFS layer already does all the dentry stuff for rename,
1820  * we just have to decrement the usage count for the target if
1821  * it exists so that the VFS layer correctly free's it when it
1822  * gets overwritten.
1823  */
1824 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1825 {
1826         struct inode *inode = old_dentry->d_inode;
1827         int they_are_dirs = S_ISDIR(inode->i_mode);
1828
1829         if (!simple_empty(new_dentry))
1830                 return -ENOTEMPTY;
1831
1832         if (new_dentry->d_inode) {
1833                 (void) shmem_unlink(new_dir, new_dentry);
1834                 if (they_are_dirs)
1835                         drop_nlink(old_dir);
1836         } else if (they_are_dirs) {
1837                 drop_nlink(old_dir);
1838                 inc_nlink(new_dir);
1839         }
1840
1841         old_dir->i_size -= BOGO_DIRENT_SIZE;
1842         new_dir->i_size += BOGO_DIRENT_SIZE;
1843         old_dir->i_ctime = old_dir->i_mtime =
1844         new_dir->i_ctime = new_dir->i_mtime =
1845         inode->i_ctime = CURRENT_TIME;
1846         return 0;
1847 }
1848
1849 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1850 {
1851         int error;
1852         int len;
1853         struct inode *inode;
1854         struct page *page = NULL;
1855         char *kaddr;
1856         struct shmem_inode_info *info;
1857
1858         len = strlen(symname) + 1;
1859         if (len > PAGE_CACHE_SIZE)
1860                 return -ENAMETOOLONG;
1861
1862         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1863         if (!inode)
1864                 return -ENOSPC;
1865
1866         error = security_inode_init_security(inode, dir, NULL, NULL,
1867                                              NULL);
1868         if (error) {
1869                 if (error != -EOPNOTSUPP) {
1870                         iput(inode);
1871                         return error;
1872                 }
1873                 error = 0;
1874         }
1875
1876         info = SHMEM_I(inode);
1877         inode->i_size = len-1;
1878         if (len <= (char *)inode - (char *)info) {
1879                 /* do it inline */
1880                 memcpy(info, symname, len);
1881                 inode->i_op = &shmem_symlink_inline_operations;
1882         } else {
1883                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1884                 if (error) {
1885                         iput(inode);
1886                         return error;
1887                 }
1888                 inode->i_op = &shmem_symlink_inode_operations;
1889                 kaddr = kmap_atomic(page, KM_USER0);
1890                 memcpy(kaddr, symname, len);
1891                 kunmap_atomic(kaddr, KM_USER0);
1892                 set_page_dirty(page);
1893                 page_cache_release(page);
1894         }
1895         if (dir->i_mode & S_ISGID)
1896                 inode->i_gid = dir->i_gid;
1897         dir->i_size += BOGO_DIRENT_SIZE;
1898         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1899         d_instantiate(dentry, inode);
1900         dget(dentry);
1901         return 0;
1902 }
1903
1904 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1905 {
1906         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1907         return NULL;
1908 }
1909
1910 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1911 {
1912         struct page *page = NULL;
1913         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1914         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1915         return page;
1916 }
1917
1918 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1919 {
1920         if (!IS_ERR(nd_get_link(nd))) {
1921                 struct page *page = cookie;
1922                 kunmap(page);
1923                 mark_page_accessed(page);
1924                 page_cache_release(page);
1925         }
1926 }
1927
1928 static const struct inode_operations shmem_symlink_inline_operations = {
1929         .readlink       = generic_readlink,
1930         .follow_link    = shmem_follow_link_inline,
1931 };
1932
1933 static const struct inode_operations shmem_symlink_inode_operations = {
1934         .truncate       = shmem_truncate,
1935         .readlink       = generic_readlink,
1936         .follow_link    = shmem_follow_link,
1937         .put_link       = shmem_put_link,
1938 };
1939
1940 #ifdef CONFIG_TMPFS_POSIX_ACL
1941 /**
1942  * Superblocks without xattr inode operations will get security.* xattr
1943  * support from the VFS "for free". As soon as we have any other xattrs
1944  * like ACLs, we also need to implement the security.* handlers at
1945  * filesystem level, though.
1946  */
1947
1948 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1949                                         size_t list_len, const char *name,
1950                                         size_t name_len)
1951 {
1952         return security_inode_listsecurity(inode, list, list_len);
1953 }
1954
1955 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1956                                     void *buffer, size_t size)
1957 {
1958         if (strcmp(name, "") == 0)
1959                 return -EINVAL;
1960         return security_inode_getsecurity(inode, name, buffer, size,
1961                                           -EOPNOTSUPP);
1962 }
1963
1964 static int shmem_xattr_security_set(struct inode *inode, const char *name,
1965                                     const void *value, size_t size, int flags)
1966 {
1967         if (strcmp(name, "") == 0)
1968                 return -EINVAL;
1969         return security_inode_setsecurity(inode, name, value, size, flags);
1970 }
1971
1972 static struct xattr_handler shmem_xattr_security_handler = {
1973         .prefix = XATTR_SECURITY_PREFIX,
1974         .list   = shmem_xattr_security_list,
1975         .get    = shmem_xattr_security_get,
1976         .set    = shmem_xattr_security_set,
1977 };
1978
1979 static struct xattr_handler *shmem_xattr_handlers[] = {
1980         &shmem_xattr_acl_access_handler,
1981         &shmem_xattr_acl_default_handler,
1982         &shmem_xattr_security_handler,
1983         NULL
1984 };
1985 #endif
1986
1987 static struct dentry *shmem_get_parent(struct dentry *child)
1988 {
1989         return ERR_PTR(-ESTALE);
1990 }
1991
1992 static int shmem_match(struct inode *ino, void *vfh)
1993 {
1994         __u32 *fh = vfh;
1995         __u64 inum = fh[2];
1996         inum = (inum << 32) | fh[1];
1997         return ino->i_ino == inum && fh[0] == ino->i_generation;
1998 }
1999
2000 static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
2001 {
2002         struct dentry *de = NULL;
2003         struct inode *inode;
2004         __u32 *fh = vfh;
2005         __u64 inum = fh[2];
2006         inum = (inum << 32) | fh[1];
2007
2008         inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
2009         if (inode) {
2010                 de = d_find_alias(inode);
2011                 iput(inode);
2012         }
2013
2014         return de? de: ERR_PTR(-ESTALE);
2015 }
2016
2017 static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
2018                 int len, int type,
2019                 int (*acceptable)(void *context, struct dentry *de),
2020                 void *context)
2021 {
2022         if (len < 3)
2023                 return ERR_PTR(-ESTALE);
2024
2025         return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
2026                                                         context);
2027 }
2028
2029 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2030                                 int connectable)
2031 {
2032         struct inode *inode = dentry->d_inode;
2033
2034         if (*len < 3)
2035                 return 255;
2036
2037         if (hlist_unhashed(&inode->i_hash)) {
2038                 /* Unfortunately insert_inode_hash is not idempotent,
2039                  * so as we hash inodes here rather than at creation
2040                  * time, we need a lock to ensure we only try
2041                  * to do it once
2042                  */
2043                 static DEFINE_SPINLOCK(lock);
2044                 spin_lock(&lock);
2045                 if (hlist_unhashed(&inode->i_hash))
2046                         __insert_inode_hash(inode,
2047                                             inode->i_ino + inode->i_generation);
2048                 spin_unlock(&lock);
2049         }
2050
2051         fh[0] = inode->i_generation;
2052         fh[1] = inode->i_ino;
2053         fh[2] = ((__u64)inode->i_ino) >> 32;
2054
2055         *len = 3;
2056         return 1;
2057 }
2058
2059 static struct export_operations shmem_export_ops = {
2060         .get_parent     = shmem_get_parent,
2061         .get_dentry     = shmem_get_dentry,
2062         .encode_fh      = shmem_encode_fh,
2063         .decode_fh      = shmem_decode_fh,
2064 };
2065
2066 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2067         gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2068         int *policy, nodemask_t *policy_nodes)
2069 {
2070         char *this_char, *value, *rest;
2071
2072         while (options != NULL) {
2073                 this_char = options;
2074                 for (;;) {
2075                         /*
2076                          * NUL-terminate this option: unfortunately,
2077                          * mount options form a comma-separated list,
2078                          * but mpol's nodelist may also contain commas.
2079                          */
2080                         options = strchr(options, ',');
2081                         if (options == NULL)
2082                                 break;
2083                         options++;
2084                         if (!isdigit(*options)) {
2085                                 options[-1] = '\0';
2086                                 break;
2087                         }
2088                 }
2089                 if (!*this_char)
2090                         continue;
2091                 if ((value = strchr(this_char,'=')) != NULL) {
2092                         *value++ = 0;
2093                 } else {
2094                         printk(KERN_ERR
2095                             "tmpfs: No value for mount option '%s'\n",
2096                             this_char);
2097                         return 1;
2098                 }
2099
2100                 if (!strcmp(this_char,"size")) {
2101                         unsigned long long size;
2102                         size = memparse(value,&rest);
2103                         if (*rest == '%') {
2104                                 size <<= PAGE_SHIFT;
2105                                 size *= totalram_pages;
2106                                 do_div(size, 100);
2107                                 rest++;
2108                         }
2109                         if (*rest)
2110                                 goto bad_val;
2111                         *blocks = size >> PAGE_CACHE_SHIFT;
2112                 } else if (!strcmp(this_char,"nr_blocks")) {
2113                         *blocks = memparse(value,&rest);
2114                         if (*rest)
2115                                 goto bad_val;
2116                 } else if (!strcmp(this_char,"nr_inodes")) {
2117                         *inodes = memparse(value,&rest);
2118                         if (*rest)
2119                                 goto bad_val;
2120                 } else if (!strcmp(this_char,"mode")) {
2121                         if (!mode)
2122                                 continue;
2123                         *mode = simple_strtoul(value,&rest,8);
2124                         if (*rest)
2125                                 goto bad_val;
2126                 } else if (!strcmp(this_char,"uid")) {
2127                         if (!uid)
2128                                 continue;
2129                         *uid = simple_strtoul(value,&rest,0);
2130                         if (*rest)
2131                                 goto bad_val;
2132                 } else if (!strcmp(this_char,"gid")) {
2133                         if (!gid)
2134                                 continue;
2135                         *gid = simple_strtoul(value,&rest,0);
2136                         if (*rest)
2137                                 goto bad_val;
2138                 } else if (!strcmp(this_char,"mpol")) {
2139                         if (shmem_parse_mpol(value,policy,policy_nodes))
2140                                 goto bad_val;
2141                 } else {
2142                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2143                                this_char);
2144                         return 1;
2145                 }
2146         }
2147         return 0;
2148
2149 bad_val:
2150         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2151                value, this_char);
2152         return 1;
2153
2154 }
2155
2156 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2157 {
2158         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2159         unsigned long max_blocks = sbinfo->max_blocks;
2160         unsigned long max_inodes = sbinfo->max_inodes;
2161         int policy = sbinfo->policy;
2162         nodemask_t policy_nodes = sbinfo->policy_nodes;
2163         unsigned long blocks;
2164         unsigned long inodes;
2165         int error = -EINVAL;
2166
2167         if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2168                                 &max_inodes, &policy, &policy_nodes))
2169                 return error;
2170
2171         spin_lock(&sbinfo->stat_lock);
2172         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2173         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2174         if (max_blocks < blocks)
2175                 goto out;
2176         if (max_inodes < inodes)
2177                 goto out;
2178         /*
2179          * Those tests also disallow limited->unlimited while any are in
2180          * use, so i_blocks will always be zero when max_blocks is zero;
2181          * but we must separately disallow unlimited->limited, because
2182          * in that case we have no record of how much is already in use.
2183          */
2184         if (max_blocks && !sbinfo->max_blocks)
2185                 goto out;
2186         if (max_inodes && !sbinfo->max_inodes)
2187                 goto out;
2188
2189         error = 0;
2190         sbinfo->max_blocks  = max_blocks;
2191         sbinfo->free_blocks = max_blocks - blocks;
2192         sbinfo->max_inodes  = max_inodes;
2193         sbinfo->free_inodes = max_inodes - inodes;
2194         sbinfo->policy = policy;
2195         sbinfo->policy_nodes = policy_nodes;
2196 out:
2197         spin_unlock(&sbinfo->stat_lock);
2198         return error;
2199 }
2200 #endif
2201
2202 static void shmem_put_super(struct super_block *sb)
2203 {
2204         kfree(sb->s_fs_info);
2205         sb->s_fs_info = NULL;
2206 }
2207
2208 static int shmem_fill_super(struct super_block *sb,
2209                             void *data, int silent)
2210 {
2211         struct inode *inode;
2212         struct dentry *root;
2213         int mode   = S_IRWXUGO | S_ISVTX;
2214         uid_t uid = current->fsuid;
2215         gid_t gid = current->fsgid;
2216         int err = -ENOMEM;
2217         struct shmem_sb_info *sbinfo;
2218         unsigned long blocks = 0;
2219         unsigned long inodes = 0;
2220         int policy = MPOL_DEFAULT;
2221         nodemask_t policy_nodes = node_online_map;
2222
2223 #ifdef CONFIG_TMPFS
2224         /*
2225          * Per default we only allow half of the physical ram per
2226          * tmpfs instance, limiting inodes to one per page of lowmem;
2227          * but the internal instance is left unlimited.
2228          */
2229         if (!(sb->s_flags & MS_NOUSER)) {
2230                 blocks = totalram_pages / 2;
2231                 inodes = totalram_pages - totalhigh_pages;
2232                 if (inodes > blocks)
2233                         inodes = blocks;
2234                 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2235                                         &inodes, &policy, &policy_nodes))
2236                         return -EINVAL;
2237         }
2238         sb->s_export_op = &shmem_export_ops;
2239 #else
2240         sb->s_flags |= MS_NOUSER;
2241 #endif
2242
2243         /* Round up to L1_CACHE_BYTES to resist false sharing */
2244         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2245                                 L1_CACHE_BYTES), GFP_KERNEL);
2246         if (!sbinfo)
2247                 return -ENOMEM;
2248
2249         spin_lock_init(&sbinfo->stat_lock);
2250         sbinfo->max_blocks = blocks;
2251         sbinfo->free_blocks = blocks;
2252         sbinfo->max_inodes = inodes;
2253         sbinfo->free_inodes = inodes;
2254         sbinfo->policy = policy;
2255         sbinfo->policy_nodes = policy_nodes;
2256
2257         sb->s_fs_info = sbinfo;
2258         sb->s_maxbytes = SHMEM_MAX_BYTES;
2259         sb->s_blocksize = PAGE_CACHE_SIZE;
2260         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2261         sb->s_magic = TMPFS_MAGIC;
2262         sb->s_op = &shmem_ops;
2263         sb->s_time_gran = 1;
2264 #ifdef CONFIG_TMPFS_POSIX_ACL
2265         sb->s_xattr = shmem_xattr_handlers;
2266         sb->s_flags |= MS_POSIXACL;
2267 #endif
2268
2269         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2270         if (!inode)
2271                 goto failed;
2272         inode->i_uid = uid;
2273         inode->i_gid = gid;
2274         root = d_alloc_root(inode);
2275         if (!root)
2276                 goto failed_iput;
2277         sb->s_root = root;
2278         return 0;
2279
2280 failed_iput:
2281         iput(inode);
2282 failed:
2283         shmem_put_super(sb);
2284         return err;
2285 }
2286
2287 static struct kmem_cache *shmem_inode_cachep;
2288
2289 static struct inode *shmem_alloc_inode(struct super_block *sb)
2290 {
2291         struct shmem_inode_info *p;
2292         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2293         if (!p)
2294                 return NULL;
2295         return &p->vfs_inode;
2296 }
2297
2298 static void shmem_destroy_inode(struct inode *inode)
2299 {
2300         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2301                 /* only struct inode is valid if it's an inline symlink */
2302                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2303         }
2304         shmem_acl_destroy_inode(inode);
2305         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2306 }
2307
2308 static void init_once(void *foo, struct kmem_cache *cachep,
2309                       unsigned long flags)
2310 {
2311         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2312
2313         inode_init_once(&p->vfs_inode);
2314 #ifdef CONFIG_TMPFS_POSIX_ACL
2315         p->i_acl = NULL;
2316         p->i_default_acl = NULL;
2317 #endif
2318 }
2319
2320 static int init_inodecache(void)
2321 {
2322         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2323                                 sizeof(struct shmem_inode_info),
2324                                 0, 0, init_once);
2325         if (shmem_inode_cachep == NULL)
2326                 return -ENOMEM;
2327         return 0;
2328 }
2329
2330 static void destroy_inodecache(void)
2331 {
2332         kmem_cache_destroy(shmem_inode_cachep);
2333 }
2334
2335 static const struct address_space_operations shmem_aops = {
2336         .writepage      = shmem_writepage,
2337         .set_page_dirty = __set_page_dirty_no_writeback,
2338 #ifdef CONFIG_TMPFS
2339         .readpage       = shmem_readpage,
2340         .prepare_write  = shmem_prepare_write,
2341         .commit_write   = simple_commit_write,
2342 #endif
2343         .migratepage    = migrate_page,
2344 };
2345
2346 static const struct file_operations shmem_file_operations = {
2347         .mmap           = shmem_mmap,
2348 #ifdef CONFIG_TMPFS
2349         .llseek         = generic_file_llseek,
2350         .read           = shmem_file_read,
2351         .write          = shmem_file_write,
2352         .fsync          = simple_sync_file,
2353         .splice_read    = generic_file_splice_read,
2354         .splice_write   = generic_file_splice_write,
2355 #endif
2356 };
2357
2358 static const struct inode_operations shmem_inode_operations = {
2359         .truncate       = shmem_truncate,
2360         .setattr        = shmem_notify_change,
2361         .truncate_range = shmem_truncate_range,
2362 #ifdef CONFIG_TMPFS_POSIX_ACL
2363         .setxattr       = generic_setxattr,
2364         .getxattr       = generic_getxattr,
2365         .listxattr      = generic_listxattr,
2366         .removexattr    = generic_removexattr,
2367         .permission     = shmem_permission,
2368 #endif
2369
2370 };
2371
2372 static const struct inode_operations shmem_dir_inode_operations = {
2373 #ifdef CONFIG_TMPFS
2374         .create         = shmem_create,
2375         .lookup         = simple_lookup,
2376         .link           = shmem_link,
2377         .unlink         = shmem_unlink,
2378         .symlink        = shmem_symlink,
2379         .mkdir          = shmem_mkdir,
2380         .rmdir          = shmem_rmdir,
2381         .mknod          = shmem_mknod,
2382         .rename         = shmem_rename,
2383 #endif
2384 #ifdef CONFIG_TMPFS_POSIX_ACL
2385         .setattr        = shmem_notify_change,
2386         .setxattr       = generic_setxattr,
2387         .getxattr       = generic_getxattr,
2388         .listxattr      = generic_listxattr,
2389         .removexattr    = generic_removexattr,
2390         .permission     = shmem_permission,
2391 #endif
2392 };
2393
2394 static const struct inode_operations shmem_special_inode_operations = {
2395 #ifdef CONFIG_TMPFS_POSIX_ACL
2396         .setattr        = shmem_notify_change,
2397         .setxattr       = generic_setxattr,
2398         .getxattr       = generic_getxattr,
2399         .listxattr      = generic_listxattr,
2400         .removexattr    = generic_removexattr,
2401         .permission     = shmem_permission,
2402 #endif
2403 };
2404
2405 static const struct super_operations shmem_ops = {
2406         .alloc_inode    = shmem_alloc_inode,
2407         .destroy_inode  = shmem_destroy_inode,
2408 #ifdef CONFIG_TMPFS
2409         .statfs         = shmem_statfs,
2410         .remount_fs     = shmem_remount_fs,
2411 #endif
2412         .delete_inode   = shmem_delete_inode,
2413         .drop_inode     = generic_delete_inode,
2414         .put_super      = shmem_put_super,
2415 };
2416
2417 static struct vm_operations_struct shmem_vm_ops = {
2418         .fault          = shmem_fault,
2419 #ifdef CONFIG_NUMA
2420         .set_policy     = shmem_set_policy,
2421         .get_policy     = shmem_get_policy,
2422 #endif
2423 };
2424
2425
2426 static int shmem_get_sb(struct file_system_type *fs_type,
2427         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2428 {
2429         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2430 }
2431
2432 static struct file_system_type tmpfs_fs_type = {
2433         .owner          = THIS_MODULE,
2434         .name           = "tmpfs",
2435         .get_sb         = shmem_get_sb,
2436         .kill_sb        = kill_litter_super,
2437 };
2438 static struct vfsmount *shm_mnt;
2439
2440 static int __init init_tmpfs(void)
2441 {
2442         int error;
2443
2444         error = init_inodecache();
2445         if (error)
2446                 goto out3;
2447
2448         error = register_filesystem(&tmpfs_fs_type);
2449         if (error) {
2450                 printk(KERN_ERR "Could not register tmpfs\n");
2451                 goto out2;
2452         }
2453
2454         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2455                                 tmpfs_fs_type.name, NULL);
2456         if (IS_ERR(shm_mnt)) {
2457                 error = PTR_ERR(shm_mnt);
2458                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2459                 goto out1;
2460         }
2461         return 0;
2462
2463 out1:
2464         unregister_filesystem(&tmpfs_fs_type);
2465 out2:
2466         destroy_inodecache();
2467 out3:
2468         shm_mnt = ERR_PTR(error);
2469         return error;
2470 }
2471 module_init(init_tmpfs)
2472
2473 /*
2474  * shmem_file_setup - get an unlinked file living in tmpfs
2475  *
2476  * @name: name for dentry (to be seen in /proc/<pid>/maps
2477  * @size: size to be set for the file
2478  *
2479  */
2480 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2481 {
2482         int error;
2483         struct file *file;
2484         struct inode *inode;
2485         struct dentry *dentry, *root;
2486         struct qstr this;
2487
2488         if (IS_ERR(shm_mnt))
2489                 return (void *)shm_mnt;
2490
2491         if (size < 0 || size > SHMEM_MAX_BYTES)
2492                 return ERR_PTR(-EINVAL);
2493
2494         if (shmem_acct_size(flags, size))
2495                 return ERR_PTR(-ENOMEM);
2496
2497         error = -ENOMEM;
2498         this.name = name;
2499         this.len = strlen(name);
2500         this.hash = 0; /* will go */
2501         root = shm_mnt->mnt_root;
2502         dentry = d_alloc(root, &this);
2503         if (!dentry)
2504                 goto put_memory;
2505
2506         error = -ENFILE;
2507         file = get_empty_filp();
2508         if (!file)
2509                 goto put_dentry;
2510
2511         error = -ENOSPC;
2512         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2513         if (!inode)
2514                 goto close_file;
2515
2516         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2517         d_instantiate(dentry, inode);
2518         inode->i_size = size;
2519         inode->i_nlink = 0;     /* It is unlinked */
2520         file->f_path.mnt = mntget(shm_mnt);
2521         file->f_path.dentry = dentry;
2522         file->f_mapping = inode->i_mapping;
2523         file->f_op = &shmem_file_operations;
2524         file->f_mode = FMODE_WRITE | FMODE_READ;
2525         return file;
2526
2527 close_file:
2528         put_filp(file);
2529 put_dentry:
2530         dput(dentry);
2531 put_memory:
2532         shmem_unacct_size(flags, size);
2533         return ERR_PTR(error);
2534 }
2535
2536 /*
2537  * shmem_zero_setup - setup a shared anonymous mapping
2538  *
2539  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2540  */
2541 int shmem_zero_setup(struct vm_area_struct *vma)
2542 {
2543         struct file *file;
2544         loff_t size = vma->vm_end - vma->vm_start;
2545
2546         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2547         if (IS_ERR(file))
2548                 return PTR_ERR(file);
2549
2550         if (vma->vm_file)
2551                 fput(vma->vm_file);
2552         vma->vm_file = file;
2553         vma->vm_ops = &shmem_vm_ops;
2554         return 0;
2555 }