e64fa726a790e0c4162fec88db0e1032dd8a5be0
[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/config.h>
27 #include <linux/module.h>
28 #include <linux/init.h>
29 #include <linux/devfs_fs_kernel.h>
30 #include <linux/fs.h>
31 #include <linux/mm.h>
32 #include <linux/mman.h>
33 #include <linux/file.h>
34 #include <linux/swap.h>
35 #include <linux/pagemap.h>
36 #include <linux/string.h>
37 #include <linux/slab.h>
38 #include <linux/backing-dev.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/mount.h>
41 #include <linux/writeback.h>
42 #include <linux/vfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/security.h>
45 #include <linux/swapops.h>
46 #include <linux/mempolicy.h>
47 #include <linux/namei.h>
48 #include <linux/xattr.h>
49 #include <asm/uaccess.h>
50 #include <asm/div64.h>
51 #include <asm/pgtable.h>
52
53 /* This magic number is used in glibc for posix shared memory */
54 #define TMPFS_MAGIC     0x01021994
55
56 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
57 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
58 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
59
60 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
61 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
62
63 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
64
65 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
66 #define SHMEM_PAGEIN     VM_READ
67 #define SHMEM_TRUNCATE   VM_WRITE
68
69 /* Definition to limit shmem_truncate's steps between cond_rescheds */
70 #define LATENCY_LIMIT    64
71
72 /* Pretend that each entry is of this size in directory's i_size */
73 #define BOGO_DIRENT_SIZE 20
74
75 /* Keep swapped page count in private field of indirect struct page */
76 #define nr_swapped              private
77
78 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
79 enum sgp_type {
80         SGP_QUICK,      /* don't try more than file page cache lookup */
81         SGP_READ,       /* don't exceed i_size, don't allocate page */
82         SGP_CACHE,      /* don't exceed i_size, may allocate page */
83         SGP_WRITE,      /* may exceed i_size, may allocate page */
84 };
85
86 static int shmem_getpage(struct inode *inode, unsigned long idx,
87                          struct page **pagep, enum sgp_type sgp, int *type);
88
89 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
90 {
91         /*
92          * The above definition of ENTRIES_PER_PAGE, and the use of
93          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
94          * might be reconsidered if it ever diverges from PAGE_SIZE.
95          */
96         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
97 }
98
99 static inline void shmem_dir_free(struct page *page)
100 {
101         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
102 }
103
104 static struct page **shmem_dir_map(struct page *page)
105 {
106         return (struct page **)kmap_atomic(page, KM_USER0);
107 }
108
109 static inline void shmem_dir_unmap(struct page **dir)
110 {
111         kunmap_atomic(dir, KM_USER0);
112 }
113
114 static swp_entry_t *shmem_swp_map(struct page *page)
115 {
116         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
117 }
118
119 static inline void shmem_swp_balance_unmap(void)
120 {
121         /*
122          * When passing a pointer to an i_direct entry, to code which
123          * also handles indirect entries and so will shmem_swp_unmap,
124          * we must arrange for the preempt count to remain in balance.
125          * What kmap_atomic of a lowmem page does depends on config
126          * and architecture, so pretend to kmap_atomic some lowmem page.
127          */
128         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
129 }
130
131 static inline void shmem_swp_unmap(swp_entry_t *entry)
132 {
133         kunmap_atomic(entry, KM_USER1);
134 }
135
136 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
137 {
138         return sb->s_fs_info;
139 }
140
141 /*
142  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
143  * for shared memory and for shared anonymous (/dev/zero) mappings
144  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
145  * consistent with the pre-accounting of private mappings ...
146  */
147 static inline int shmem_acct_size(unsigned long flags, loff_t size)
148 {
149         return (flags & VM_ACCOUNT)?
150                 security_vm_enough_memory(VM_ACCT(size)): 0;
151 }
152
153 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
154 {
155         if (flags & VM_ACCOUNT)
156                 vm_unacct_memory(VM_ACCT(size));
157 }
158
159 /*
160  * ... whereas tmpfs objects are accounted incrementally as
161  * pages are allocated, in order to allow huge sparse files.
162  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
163  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
164  */
165 static inline int shmem_acct_block(unsigned long flags)
166 {
167         return (flags & VM_ACCOUNT)?
168                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
169 }
170
171 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
172 {
173         if (!(flags & VM_ACCOUNT))
174                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
175 }
176
177 static struct super_operations shmem_ops;
178 static struct address_space_operations shmem_aops;
179 static struct file_operations shmem_file_operations;
180 static struct inode_operations shmem_inode_operations;
181 static struct inode_operations shmem_dir_inode_operations;
182 static struct inode_operations shmem_special_inode_operations;
183 static struct vm_operations_struct shmem_vm_ops;
184
185 static struct backing_dev_info shmem_backing_dev_info = {
186         .ra_pages       = 0,    /* No readahead */
187         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
188         .unplug_io_fn   = default_unplug_io_fn,
189 };
190
191 static LIST_HEAD(shmem_swaplist);
192 static DEFINE_SPINLOCK(shmem_swaplist_lock);
193
194 static void shmem_free_blocks(struct inode *inode, long pages)
195 {
196         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
197         if (sbinfo->max_blocks) {
198                 spin_lock(&sbinfo->stat_lock);
199                 sbinfo->free_blocks += pages;
200                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
201                 spin_unlock(&sbinfo->stat_lock);
202         }
203 }
204
205 /*
206  * shmem_recalc_inode - recalculate the size of an inode
207  *
208  * @inode: inode to recalc
209  *
210  * We have to calculate the free blocks since the mm can drop
211  * undirtied hole pages behind our back.
212  *
213  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
214  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
215  *
216  * It has to be called with the spinlock held.
217  */
218 static void shmem_recalc_inode(struct inode *inode)
219 {
220         struct shmem_inode_info *info = SHMEM_I(inode);
221         long freed;
222
223         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
224         if (freed > 0) {
225                 info->alloced -= freed;
226                 shmem_unacct_blocks(info->flags, freed);
227                 shmem_free_blocks(inode, freed);
228         }
229 }
230
231 /*
232  * shmem_swp_entry - find the swap vector position in the info structure
233  *
234  * @info:  info structure for the inode
235  * @index: index of the page to find
236  * @page:  optional page to add to the structure. Has to be preset to
237  *         all zeros
238  *
239  * If there is no space allocated yet it will return NULL when
240  * page is NULL, else it will use the page for the needed block,
241  * setting it to NULL on return to indicate that it has been used.
242  *
243  * The swap vector is organized the following way:
244  *
245  * There are SHMEM_NR_DIRECT entries directly stored in the
246  * shmem_inode_info structure. So small files do not need an addional
247  * allocation.
248  *
249  * For pages with index > SHMEM_NR_DIRECT there is the pointer
250  * i_indirect which points to a page which holds in the first half
251  * doubly indirect blocks, in the second half triple indirect blocks:
252  *
253  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
254  * following layout (for SHMEM_NR_DIRECT == 16):
255  *
256  * i_indirect -> dir --> 16-19
257  *            |      +-> 20-23
258  *            |
259  *            +-->dir2 --> 24-27
260  *            |        +-> 28-31
261  *            |        +-> 32-35
262  *            |        +-> 36-39
263  *            |
264  *            +-->dir3 --> 40-43
265  *                     +-> 44-47
266  *                     +-> 48-51
267  *                     +-> 52-55
268  */
269 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
270 {
271         unsigned long offset;
272         struct page **dir;
273         struct page *subdir;
274
275         if (index < SHMEM_NR_DIRECT) {
276                 shmem_swp_balance_unmap();
277                 return info->i_direct+index;
278         }
279         if (!info->i_indirect) {
280                 if (page) {
281                         info->i_indirect = *page;
282                         *page = NULL;
283                 }
284                 return NULL;                    /* need another page */
285         }
286
287         index -= SHMEM_NR_DIRECT;
288         offset = index % ENTRIES_PER_PAGE;
289         index /= ENTRIES_PER_PAGE;
290         dir = shmem_dir_map(info->i_indirect);
291
292         if (index >= ENTRIES_PER_PAGE/2) {
293                 index -= ENTRIES_PER_PAGE/2;
294                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
295                 index %= ENTRIES_PER_PAGE;
296                 subdir = *dir;
297                 if (!subdir) {
298                         if (page) {
299                                 *dir = *page;
300                                 *page = NULL;
301                         }
302                         shmem_dir_unmap(dir);
303                         return NULL;            /* need another page */
304                 }
305                 shmem_dir_unmap(dir);
306                 dir = shmem_dir_map(subdir);
307         }
308
309         dir += index;
310         subdir = *dir;
311         if (!subdir) {
312                 if (!page || !(subdir = *page)) {
313                         shmem_dir_unmap(dir);
314                         return NULL;            /* need a page */
315                 }
316                 *dir = subdir;
317                 *page = NULL;
318         }
319         shmem_dir_unmap(dir);
320         return shmem_swp_map(subdir) + offset;
321 }
322
323 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
324 {
325         long incdec = value? 1: -1;
326
327         entry->val = value;
328         info->swapped += incdec;
329         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
330                 kmap_atomic_to_page(entry)->nr_swapped += incdec;
331 }
332
333 /*
334  * shmem_swp_alloc - get the position of the swap entry for the page.
335  *                   If it does not exist allocate the entry.
336  *
337  * @info:       info structure for the inode
338  * @index:      index of the page to find
339  * @sgp:        check and recheck i_size? skip allocation?
340  */
341 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
342 {
343         struct inode *inode = &info->vfs_inode;
344         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
345         struct page *page = NULL;
346         swp_entry_t *entry;
347
348         if (sgp != SGP_WRITE &&
349             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
350                 return ERR_PTR(-EINVAL);
351
352         while (!(entry = shmem_swp_entry(info, index, &page))) {
353                 if (sgp == SGP_READ)
354                         return shmem_swp_map(ZERO_PAGE(0));
355                 /*
356                  * Test free_blocks against 1 not 0, since we have 1 data
357                  * page (and perhaps indirect index pages) yet to allocate:
358                  * a waste to allocate index if we cannot allocate data.
359                  */
360                 if (sbinfo->max_blocks) {
361                         spin_lock(&sbinfo->stat_lock);
362                         if (sbinfo->free_blocks <= 1) {
363                                 spin_unlock(&sbinfo->stat_lock);
364                                 return ERR_PTR(-ENOSPC);
365                         }
366                         sbinfo->free_blocks--;
367                         inode->i_blocks += BLOCKS_PER_PAGE;
368                         spin_unlock(&sbinfo->stat_lock);
369                 }
370
371                 spin_unlock(&info->lock);
372                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
373                 if (page) {
374                         page->nr_swapped = 0;
375                 }
376                 spin_lock(&info->lock);
377
378                 if (!page) {
379                         shmem_free_blocks(inode, 1);
380                         return ERR_PTR(-ENOMEM);
381                 }
382                 if (sgp != SGP_WRITE &&
383                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
384                         entry = ERR_PTR(-EINVAL);
385                         break;
386                 }
387                 if (info->next_index <= index)
388                         info->next_index = index + 1;
389         }
390         if (page) {
391                 /* another task gave its page, or truncated the file */
392                 shmem_free_blocks(inode, 1);
393                 shmem_dir_free(page);
394         }
395         if (info->next_index <= index && !IS_ERR(entry))
396                 info->next_index = index + 1;
397         return entry;
398 }
399
400 /*
401  * shmem_free_swp - free some swap entries in a directory
402  *
403  * @dir:   pointer to the directory
404  * @edir:  pointer after last entry of the directory
405  */
406 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
407 {
408         swp_entry_t *ptr;
409         int freed = 0;
410
411         for (ptr = dir; ptr < edir; ptr++) {
412                 if (ptr->val) {
413                         free_swap_and_cache(*ptr);
414                         *ptr = (swp_entry_t){0};
415                         freed++;
416                 }
417         }
418         return freed;
419 }
420
421 static int shmem_map_and_free_swp(struct page *subdir,
422                 int offset, int limit, struct page ***dir)
423 {
424         swp_entry_t *ptr;
425         int freed = 0;
426
427         ptr = shmem_swp_map(subdir);
428         for (; offset < limit; offset += LATENCY_LIMIT) {
429                 int size = limit - offset;
430                 if (size > LATENCY_LIMIT)
431                         size = LATENCY_LIMIT;
432                 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
433                 if (need_resched()) {
434                         shmem_swp_unmap(ptr);
435                         if (*dir) {
436                                 shmem_dir_unmap(*dir);
437                                 *dir = NULL;
438                         }
439                         cond_resched();
440                         ptr = shmem_swp_map(subdir);
441                 }
442         }
443         shmem_swp_unmap(ptr);
444         return freed;
445 }
446
447 static void shmem_free_pages(struct list_head *next)
448 {
449         struct page *page;
450         int freed = 0;
451
452         do {
453                 page = container_of(next, struct page, lru);
454                 next = next->next;
455                 shmem_dir_free(page);
456                 freed++;
457                 if (freed >= LATENCY_LIMIT) {
458                         cond_resched();
459                         freed = 0;
460                 }
461         } while (next);
462 }
463
464 static void shmem_truncate(struct inode *inode)
465 {
466         struct shmem_inode_info *info = SHMEM_I(inode);
467         unsigned long idx;
468         unsigned long size;
469         unsigned long limit;
470         unsigned long stage;
471         unsigned long diroff;
472         struct page **dir;
473         struct page *topdir;
474         struct page *middir;
475         struct page *subdir;
476         swp_entry_t *ptr;
477         LIST_HEAD(pages_to_free);
478         long nr_pages_to_free = 0;
479         long nr_swaps_freed = 0;
480         int offset;
481         int freed;
482
483         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
484         idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
485         if (idx >= info->next_index)
486                 return;
487
488         spin_lock(&info->lock);
489         info->flags |= SHMEM_TRUNCATE;
490         limit = info->next_index;
491         info->next_index = idx;
492         topdir = info->i_indirect;
493         if (topdir && idx <= SHMEM_NR_DIRECT) {
494                 info->i_indirect = NULL;
495                 nr_pages_to_free++;
496                 list_add(&topdir->lru, &pages_to_free);
497         }
498         spin_unlock(&info->lock);
499
500         if (info->swapped && idx < SHMEM_NR_DIRECT) {
501                 ptr = info->i_direct;
502                 size = limit;
503                 if (size > SHMEM_NR_DIRECT)
504                         size = SHMEM_NR_DIRECT;
505                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
506         }
507         if (!topdir)
508                 goto done2;
509
510         BUG_ON(limit <= SHMEM_NR_DIRECT);
511         limit -= SHMEM_NR_DIRECT;
512         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
513         offset = idx % ENTRIES_PER_PAGE;
514         idx -= offset;
515
516         dir = shmem_dir_map(topdir);
517         stage = ENTRIES_PER_PAGEPAGE/2;
518         if (idx < ENTRIES_PER_PAGEPAGE/2) {
519                 middir = topdir;
520                 diroff = idx/ENTRIES_PER_PAGE;
521         } else {
522                 dir += ENTRIES_PER_PAGE/2;
523                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
524                 while (stage <= idx)
525                         stage += ENTRIES_PER_PAGEPAGE;
526                 middir = *dir;
527                 if (*dir) {
528                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
529                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
530                         if (!diroff && !offset) {
531                                 *dir = NULL;
532                                 nr_pages_to_free++;
533                                 list_add(&middir->lru, &pages_to_free);
534                         }
535                         shmem_dir_unmap(dir);
536                         dir = shmem_dir_map(middir);
537                 } else {
538                         diroff = 0;
539                         offset = 0;
540                         idx = stage;
541                 }
542         }
543
544         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
545                 if (unlikely(idx == stage)) {
546                         shmem_dir_unmap(dir);
547                         dir = shmem_dir_map(topdir) +
548                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
549                         while (!*dir) {
550                                 dir++;
551                                 idx += ENTRIES_PER_PAGEPAGE;
552                                 if (idx >= limit)
553                                         goto done1;
554                         }
555                         stage = idx + ENTRIES_PER_PAGEPAGE;
556                         middir = *dir;
557                         *dir = NULL;
558                         nr_pages_to_free++;
559                         list_add(&middir->lru, &pages_to_free);
560                         shmem_dir_unmap(dir);
561                         cond_resched();
562                         dir = shmem_dir_map(middir);
563                         diroff = 0;
564                 }
565                 subdir = dir[diroff];
566                 if (subdir && subdir->nr_swapped) {
567                         size = limit - idx;
568                         if (size > ENTRIES_PER_PAGE)
569                                 size = ENTRIES_PER_PAGE;
570                         freed = shmem_map_and_free_swp(subdir,
571                                                 offset, size, &dir);
572                         if (!dir)
573                                 dir = shmem_dir_map(middir);
574                         nr_swaps_freed += freed;
575                         if (offset)
576                                 spin_lock(&info->lock);
577                         subdir->nr_swapped -= freed;
578                         if (offset)
579                                 spin_unlock(&info->lock);
580                         BUG_ON(subdir->nr_swapped > offset);
581                 }
582                 if (offset)
583                         offset = 0;
584                 else if (subdir) {
585                         dir[diroff] = NULL;
586                         nr_pages_to_free++;
587                         list_add(&subdir->lru, &pages_to_free);
588                 }
589         }
590 done1:
591         shmem_dir_unmap(dir);
592 done2:
593         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
594                 /*
595                  * Call truncate_inode_pages again: racing shmem_unuse_inode
596                  * may have swizzled a page in from swap since vmtruncate or
597                  * generic_delete_inode did it, before we lowered next_index.
598                  * Also, though shmem_getpage checks i_size before adding to
599                  * cache, no recheck after: so fix the narrow window there too.
600                  */
601                 truncate_inode_pages(inode->i_mapping, inode->i_size);
602         }
603
604         spin_lock(&info->lock);
605         info->flags &= ~SHMEM_TRUNCATE;
606         info->swapped -= nr_swaps_freed;
607         if (nr_pages_to_free)
608                 shmem_free_blocks(inode, nr_pages_to_free);
609         shmem_recalc_inode(inode);
610         spin_unlock(&info->lock);
611
612         /*
613          * Empty swap vector directory pages to be freed?
614          */
615         if (!list_empty(&pages_to_free)) {
616                 pages_to_free.prev->next = NULL;
617                 shmem_free_pages(pages_to_free.next);
618         }
619 }
620
621 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
622 {
623         struct inode *inode = dentry->d_inode;
624         struct page *page = NULL;
625         int error;
626
627         if (attr->ia_valid & ATTR_SIZE) {
628                 if (attr->ia_size < inode->i_size) {
629                         /*
630                          * If truncating down to a partial page, then
631                          * if that page is already allocated, hold it
632                          * in memory until the truncation is over, so
633                          * truncate_partial_page cannnot miss it were
634                          * it assigned to swap.
635                          */
636                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
637                                 (void) shmem_getpage(inode,
638                                         attr->ia_size>>PAGE_CACHE_SHIFT,
639                                                 &page, SGP_READ, NULL);
640                         }
641                         /*
642                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
643                          * detect if any pages might have been added to cache
644                          * after truncate_inode_pages.  But we needn't bother
645                          * if it's being fully truncated to zero-length: the
646                          * nrpages check is efficient enough in that case.
647                          */
648                         if (attr->ia_size) {
649                                 struct shmem_inode_info *info = SHMEM_I(inode);
650                                 spin_lock(&info->lock);
651                                 info->flags &= ~SHMEM_PAGEIN;
652                                 spin_unlock(&info->lock);
653                         }
654                 }
655         }
656
657         error = inode_change_ok(inode, attr);
658         if (!error)
659                 error = inode_setattr(inode, attr);
660         if (page)
661                 page_cache_release(page);
662         return error;
663 }
664
665 static void shmem_delete_inode(struct inode *inode)
666 {
667         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
668         struct shmem_inode_info *info = SHMEM_I(inode);
669
670         if (inode->i_op->truncate == shmem_truncate) {
671                 shmem_unacct_size(info->flags, inode->i_size);
672                 inode->i_size = 0;
673                 shmem_truncate(inode);
674                 if (!list_empty(&info->swaplist)) {
675                         spin_lock(&shmem_swaplist_lock);
676                         list_del_init(&info->swaplist);
677                         spin_unlock(&shmem_swaplist_lock);
678                 }
679         }
680         BUG_ON(inode->i_blocks);
681         if (sbinfo->max_inodes) {
682                 spin_lock(&sbinfo->stat_lock);
683                 sbinfo->free_inodes++;
684                 spin_unlock(&sbinfo->stat_lock);
685         }
686         clear_inode(inode);
687 }
688
689 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
690 {
691         swp_entry_t *ptr;
692
693         for (ptr = dir; ptr < edir; ptr++) {
694                 if (ptr->val == entry.val)
695                         return ptr - dir;
696         }
697         return -1;
698 }
699
700 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
701 {
702         struct inode *inode;
703         unsigned long idx;
704         unsigned long size;
705         unsigned long limit;
706         unsigned long stage;
707         struct page **dir;
708         struct page *subdir;
709         swp_entry_t *ptr;
710         int offset;
711
712         idx = 0;
713         ptr = info->i_direct;
714         spin_lock(&info->lock);
715         limit = info->next_index;
716         size = limit;
717         if (size > SHMEM_NR_DIRECT)
718                 size = SHMEM_NR_DIRECT;
719         offset = shmem_find_swp(entry, ptr, ptr+size);
720         if (offset >= 0) {
721                 shmem_swp_balance_unmap();
722                 goto found;
723         }
724         if (!info->i_indirect)
725                 goto lost2;
726
727         dir = shmem_dir_map(info->i_indirect);
728         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
729
730         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
731                 if (unlikely(idx == stage)) {
732                         shmem_dir_unmap(dir-1);
733                         dir = shmem_dir_map(info->i_indirect) +
734                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
735                         while (!*dir) {
736                                 dir++;
737                                 idx += ENTRIES_PER_PAGEPAGE;
738                                 if (idx >= limit)
739                                         goto lost1;
740                         }
741                         stage = idx + ENTRIES_PER_PAGEPAGE;
742                         subdir = *dir;
743                         shmem_dir_unmap(dir);
744                         dir = shmem_dir_map(subdir);
745                 }
746                 subdir = *dir;
747                 if (subdir && subdir->nr_swapped) {
748                         ptr = shmem_swp_map(subdir);
749                         size = limit - idx;
750                         if (size > ENTRIES_PER_PAGE)
751                                 size = ENTRIES_PER_PAGE;
752                         offset = shmem_find_swp(entry, ptr, ptr+size);
753                         if (offset >= 0) {
754                                 shmem_dir_unmap(dir);
755                                 goto found;
756                         }
757                         shmem_swp_unmap(ptr);
758                 }
759         }
760 lost1:
761         shmem_dir_unmap(dir-1);
762 lost2:
763         spin_unlock(&info->lock);
764         return 0;
765 found:
766         idx += offset;
767         inode = &info->vfs_inode;
768         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
769                 info->flags |= SHMEM_PAGEIN;
770                 shmem_swp_set(info, ptr + offset, 0);
771         }
772         shmem_swp_unmap(ptr);
773         spin_unlock(&info->lock);
774         /*
775          * Decrement swap count even when the entry is left behind:
776          * try_to_unuse will skip over mms, then reincrement count.
777          */
778         swap_free(entry);
779         return 1;
780 }
781
782 /*
783  * shmem_unuse() search for an eventually swapped out shmem page.
784  */
785 int shmem_unuse(swp_entry_t entry, struct page *page)
786 {
787         struct list_head *p, *next;
788         struct shmem_inode_info *info;
789         int found = 0;
790
791         spin_lock(&shmem_swaplist_lock);
792         list_for_each_safe(p, next, &shmem_swaplist) {
793                 info = list_entry(p, struct shmem_inode_info, swaplist);
794                 if (!info->swapped)
795                         list_del_init(&info->swaplist);
796                 else if (shmem_unuse_inode(info, entry, page)) {
797                         /* move head to start search for next from here */
798                         list_move_tail(&shmem_swaplist, &info->swaplist);
799                         found = 1;
800                         break;
801                 }
802         }
803         spin_unlock(&shmem_swaplist_lock);
804         return found;
805 }
806
807 /*
808  * Move the page from the page cache to the swap cache.
809  */
810 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
811 {
812         struct shmem_inode_info *info;
813         swp_entry_t *entry, swap;
814         struct address_space *mapping;
815         unsigned long index;
816         struct inode *inode;
817
818         BUG_ON(!PageLocked(page));
819         BUG_ON(page_mapped(page));
820
821         mapping = page->mapping;
822         index = page->index;
823         inode = mapping->host;
824         info = SHMEM_I(inode);
825         if (info->flags & VM_LOCKED)
826                 goto redirty;
827         swap = get_swap_page();
828         if (!swap.val)
829                 goto redirty;
830
831         spin_lock(&info->lock);
832         shmem_recalc_inode(inode);
833         if (index >= info->next_index) {
834                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
835                 goto unlock;
836         }
837         entry = shmem_swp_entry(info, index, NULL);
838         BUG_ON(!entry);
839         BUG_ON(entry->val);
840
841         if (move_to_swap_cache(page, swap) == 0) {
842                 shmem_swp_set(info, entry, swap.val);
843                 shmem_swp_unmap(entry);
844                 spin_unlock(&info->lock);
845                 if (list_empty(&info->swaplist)) {
846                         spin_lock(&shmem_swaplist_lock);
847                         /* move instead of add in case we're racing */
848                         list_move_tail(&info->swaplist, &shmem_swaplist);
849                         spin_unlock(&shmem_swaplist_lock);
850                 }
851                 unlock_page(page);
852                 return 0;
853         }
854
855         shmem_swp_unmap(entry);
856 unlock:
857         spin_unlock(&info->lock);
858         swap_free(swap);
859 redirty:
860         set_page_dirty(page);
861         return WRITEPAGE_ACTIVATE;      /* Return with the page locked */
862 }
863
864 #ifdef CONFIG_NUMA
865 static struct page *shmem_swapin_async(struct shared_policy *p,
866                                        swp_entry_t entry, unsigned long idx)
867 {
868         struct page *page;
869         struct vm_area_struct pvma;
870
871         /* Create a pseudo vma that just contains the policy */
872         memset(&pvma, 0, sizeof(struct vm_area_struct));
873         pvma.vm_end = PAGE_SIZE;
874         pvma.vm_pgoff = idx;
875         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
876         page = read_swap_cache_async(entry, &pvma, 0);
877         mpol_free(pvma.vm_policy);
878         return page;
879 }
880
881 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
882                           unsigned long idx)
883 {
884         struct shared_policy *p = &info->policy;
885         int i, num;
886         struct page *page;
887         unsigned long offset;
888
889         num = valid_swaphandles(entry, &offset);
890         for (i = 0; i < num; offset++, i++) {
891                 page = shmem_swapin_async(p,
892                                 swp_entry(swp_type(entry), offset), idx);
893                 if (!page)
894                         break;
895                 page_cache_release(page);
896         }
897         lru_add_drain();        /* Push any new pages onto the LRU now */
898         return shmem_swapin_async(p, entry, idx);
899 }
900
901 static struct page *
902 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
903                  unsigned long idx)
904 {
905         struct vm_area_struct pvma;
906         struct page *page;
907
908         memset(&pvma, 0, sizeof(struct vm_area_struct));
909         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
910         pvma.vm_pgoff = idx;
911         pvma.vm_end = PAGE_SIZE;
912         page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
913         mpol_free(pvma.vm_policy);
914         return page;
915 }
916 #else
917 static inline struct page *
918 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
919 {
920         swapin_readahead(entry, 0, NULL);
921         return read_swap_cache_async(entry, NULL, 0);
922 }
923
924 static inline struct page *
925 shmem_alloc_page(unsigned int __nocast gfp,struct shmem_inode_info *info,
926                                  unsigned long idx)
927 {
928         return alloc_page(gfp | __GFP_ZERO);
929 }
930 #endif
931
932 /*
933  * shmem_getpage - either get the page from swap or allocate a new one
934  *
935  * If we allocate a new one we do not mark it dirty. That's up to the
936  * vm. If we swap it in we mark it dirty since we also free the swap
937  * entry since a page cannot live in both the swap and page cache
938  */
939 static int shmem_getpage(struct inode *inode, unsigned long idx,
940                         struct page **pagep, enum sgp_type sgp, int *type)
941 {
942         struct address_space *mapping = inode->i_mapping;
943         struct shmem_inode_info *info = SHMEM_I(inode);
944         struct shmem_sb_info *sbinfo;
945         struct page *filepage = *pagep;
946         struct page *swappage;
947         swp_entry_t *entry;
948         swp_entry_t swap;
949         int error;
950
951         if (idx >= SHMEM_MAX_INDEX)
952                 return -EFBIG;
953         /*
954          * Normally, filepage is NULL on entry, and either found
955          * uptodate immediately, or allocated and zeroed, or read
956          * in under swappage, which is then assigned to filepage.
957          * But shmem_prepare_write passes in a locked filepage,
958          * which may be found not uptodate by other callers too,
959          * and may need to be copied from the swappage read in.
960          */
961 repeat:
962         if (!filepage)
963                 filepage = find_lock_page(mapping, idx);
964         if (filepage && PageUptodate(filepage))
965                 goto done;
966         error = 0;
967         if (sgp == SGP_QUICK)
968                 goto failed;
969
970         spin_lock(&info->lock);
971         shmem_recalc_inode(inode);
972         entry = shmem_swp_alloc(info, idx, sgp);
973         if (IS_ERR(entry)) {
974                 spin_unlock(&info->lock);
975                 error = PTR_ERR(entry);
976                 goto failed;
977         }
978         swap = *entry;
979
980         if (swap.val) {
981                 /* Look it up and read it in.. */
982                 swappage = lookup_swap_cache(swap);
983                 if (!swappage) {
984                         shmem_swp_unmap(entry);
985                         spin_unlock(&info->lock);
986                         /* here we actually do the io */
987                         if (type && *type == VM_FAULT_MINOR) {
988                                 inc_page_state(pgmajfault);
989                                 *type = VM_FAULT_MAJOR;
990                         }
991                         swappage = shmem_swapin(info, swap, idx);
992                         if (!swappage) {
993                                 spin_lock(&info->lock);
994                                 entry = shmem_swp_alloc(info, idx, sgp);
995                                 if (IS_ERR(entry))
996                                         error = PTR_ERR(entry);
997                                 else {
998                                         if (entry->val == swap.val)
999                                                 error = -ENOMEM;
1000                                         shmem_swp_unmap(entry);
1001                                 }
1002                                 spin_unlock(&info->lock);
1003                                 if (error)
1004                                         goto failed;
1005                                 goto repeat;
1006                         }
1007                         wait_on_page_locked(swappage);
1008                         page_cache_release(swappage);
1009                         goto repeat;
1010                 }
1011
1012                 /* We have to do this with page locked to prevent races */
1013                 if (TestSetPageLocked(swappage)) {
1014                         shmem_swp_unmap(entry);
1015                         spin_unlock(&info->lock);
1016                         wait_on_page_locked(swappage);
1017                         page_cache_release(swappage);
1018                         goto repeat;
1019                 }
1020                 if (PageWriteback(swappage)) {
1021                         shmem_swp_unmap(entry);
1022                         spin_unlock(&info->lock);
1023                         wait_on_page_writeback(swappage);
1024                         unlock_page(swappage);
1025                         page_cache_release(swappage);
1026                         goto repeat;
1027                 }
1028                 if (!PageUptodate(swappage)) {
1029                         shmem_swp_unmap(entry);
1030                         spin_unlock(&info->lock);
1031                         unlock_page(swappage);
1032                         page_cache_release(swappage);
1033                         error = -EIO;
1034                         goto failed;
1035                 }
1036
1037                 if (filepage) {
1038                         shmem_swp_set(info, entry, 0);
1039                         shmem_swp_unmap(entry);
1040                         delete_from_swap_cache(swappage);
1041                         spin_unlock(&info->lock);
1042                         copy_highpage(filepage, swappage);
1043                         unlock_page(swappage);
1044                         page_cache_release(swappage);
1045                         flush_dcache_page(filepage);
1046                         SetPageUptodate(filepage);
1047                         set_page_dirty(filepage);
1048                         swap_free(swap);
1049                 } else if (!(error = move_from_swap_cache(
1050                                 swappage, idx, mapping))) {
1051                         info->flags |= SHMEM_PAGEIN;
1052                         shmem_swp_set(info, entry, 0);
1053                         shmem_swp_unmap(entry);
1054                         spin_unlock(&info->lock);
1055                         filepage = swappage;
1056                         swap_free(swap);
1057                 } else {
1058                         shmem_swp_unmap(entry);
1059                         spin_unlock(&info->lock);
1060                         unlock_page(swappage);
1061                         page_cache_release(swappage);
1062                         if (error == -ENOMEM) {
1063                                 /* let kswapd refresh zone for GFP_ATOMICs */
1064                                 blk_congestion_wait(WRITE, HZ/50);
1065                         }
1066                         goto repeat;
1067                 }
1068         } else if (sgp == SGP_READ && !filepage) {
1069                 shmem_swp_unmap(entry);
1070                 filepage = find_get_page(mapping, idx);
1071                 if (filepage &&
1072                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1073                         spin_unlock(&info->lock);
1074                         wait_on_page_locked(filepage);
1075                         page_cache_release(filepage);
1076                         filepage = NULL;
1077                         goto repeat;
1078                 }
1079                 spin_unlock(&info->lock);
1080         } else {
1081                 shmem_swp_unmap(entry);
1082                 sbinfo = SHMEM_SB(inode->i_sb);
1083                 if (sbinfo->max_blocks) {
1084                         spin_lock(&sbinfo->stat_lock);
1085                         if (sbinfo->free_blocks == 0 ||
1086                             shmem_acct_block(info->flags)) {
1087                                 spin_unlock(&sbinfo->stat_lock);
1088                                 spin_unlock(&info->lock);
1089                                 error = -ENOSPC;
1090                                 goto failed;
1091                         }
1092                         sbinfo->free_blocks--;
1093                         inode->i_blocks += BLOCKS_PER_PAGE;
1094                         spin_unlock(&sbinfo->stat_lock);
1095                 } else if (shmem_acct_block(info->flags)) {
1096                         spin_unlock(&info->lock);
1097                         error = -ENOSPC;
1098                         goto failed;
1099                 }
1100
1101                 if (!filepage) {
1102                         spin_unlock(&info->lock);
1103                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1104                                                     info,
1105                                                     idx);
1106                         if (!filepage) {
1107                                 shmem_unacct_blocks(info->flags, 1);
1108                                 shmem_free_blocks(inode, 1);
1109                                 error = -ENOMEM;
1110                                 goto failed;
1111                         }
1112
1113                         spin_lock(&info->lock);
1114                         entry = shmem_swp_alloc(info, idx, sgp);
1115                         if (IS_ERR(entry))
1116                                 error = PTR_ERR(entry);
1117                         else {
1118                                 swap = *entry;
1119                                 shmem_swp_unmap(entry);
1120                         }
1121                         if (error || swap.val || 0 != add_to_page_cache_lru(
1122                                         filepage, mapping, idx, GFP_ATOMIC)) {
1123                                 spin_unlock(&info->lock);
1124                                 page_cache_release(filepage);
1125                                 shmem_unacct_blocks(info->flags, 1);
1126                                 shmem_free_blocks(inode, 1);
1127                                 filepage = NULL;
1128                                 if (error)
1129                                         goto failed;
1130                                 goto repeat;
1131                         }
1132                         info->flags |= SHMEM_PAGEIN;
1133                 }
1134
1135                 info->alloced++;
1136                 spin_unlock(&info->lock);
1137                 flush_dcache_page(filepage);
1138                 SetPageUptodate(filepage);
1139         }
1140 done:
1141         if (*pagep != filepage) {
1142                 unlock_page(filepage);
1143                 *pagep = filepage;
1144         }
1145         return 0;
1146
1147 failed:
1148         if (*pagep != filepage) {
1149                 unlock_page(filepage);
1150                 page_cache_release(filepage);
1151         }
1152         return error;
1153 }
1154
1155 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1156 {
1157         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1158         struct page *page = NULL;
1159         unsigned long idx;
1160         int error;
1161
1162         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1163         idx += vma->vm_pgoff;
1164         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1165         if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1166                 return NOPAGE_SIGBUS;
1167
1168         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1169         if (error)
1170                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1171
1172         mark_page_accessed(page);
1173         return page;
1174 }
1175
1176 static int shmem_populate(struct vm_area_struct *vma,
1177         unsigned long addr, unsigned long len,
1178         pgprot_t prot, unsigned long pgoff, int nonblock)
1179 {
1180         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1181         struct mm_struct *mm = vma->vm_mm;
1182         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1183         unsigned long size;
1184
1185         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1186         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1187                 return -EINVAL;
1188
1189         while ((long) len > 0) {
1190                 struct page *page = NULL;
1191                 int err;
1192                 /*
1193                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1194                  */
1195                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1196                 if (err)
1197                         return err;
1198                 if (page) {
1199                         mark_page_accessed(page);
1200                         err = install_page(mm, vma, addr, page, prot);
1201                         if (err) {
1202                                 page_cache_release(page);
1203                                 return err;
1204                         }
1205                 } else if (nonblock) {
1206                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1207                         if (err)
1208                                 return err;
1209                 }
1210
1211                 len -= PAGE_SIZE;
1212                 addr += PAGE_SIZE;
1213                 pgoff++;
1214         }
1215         return 0;
1216 }
1217
1218 #ifdef CONFIG_NUMA
1219 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1220 {
1221         struct inode *i = vma->vm_file->f_dentry->d_inode;
1222         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1223 }
1224
1225 struct mempolicy *
1226 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1227 {
1228         struct inode *i = vma->vm_file->f_dentry->d_inode;
1229         unsigned long idx;
1230
1231         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1232         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1233 }
1234 #endif
1235
1236 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1237 {
1238         struct inode *inode = file->f_dentry->d_inode;
1239         struct shmem_inode_info *info = SHMEM_I(inode);
1240         int retval = -ENOMEM;
1241
1242         spin_lock(&info->lock);
1243         if (lock && !(info->flags & VM_LOCKED)) {
1244                 if (!user_shm_lock(inode->i_size, user))
1245                         goto out_nomem;
1246                 info->flags |= VM_LOCKED;
1247         }
1248         if (!lock && (info->flags & VM_LOCKED) && user) {
1249                 user_shm_unlock(inode->i_size, user);
1250                 info->flags &= ~VM_LOCKED;
1251         }
1252         retval = 0;
1253 out_nomem:
1254         spin_unlock(&info->lock);
1255         return retval;
1256 }
1257
1258 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1259 {
1260         file_accessed(file);
1261         vma->vm_ops = &shmem_vm_ops;
1262         return 0;
1263 }
1264
1265 static struct inode *
1266 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1267 {
1268         struct inode *inode;
1269         struct shmem_inode_info *info;
1270         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1271
1272         if (sbinfo->max_inodes) {
1273                 spin_lock(&sbinfo->stat_lock);
1274                 if (!sbinfo->free_inodes) {
1275                         spin_unlock(&sbinfo->stat_lock);
1276                         return NULL;
1277                 }
1278                 sbinfo->free_inodes--;
1279                 spin_unlock(&sbinfo->stat_lock);
1280         }
1281
1282         inode = new_inode(sb);
1283         if (inode) {
1284                 inode->i_mode = mode;
1285                 inode->i_uid = current->fsuid;
1286                 inode->i_gid = current->fsgid;
1287                 inode->i_blksize = PAGE_CACHE_SIZE;
1288                 inode->i_blocks = 0;
1289                 inode->i_mapping->a_ops = &shmem_aops;
1290                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1291                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1292                 info = SHMEM_I(inode);
1293                 memset(info, 0, (char *)inode - (char *)info);
1294                 spin_lock_init(&info->lock);
1295                 INIT_LIST_HEAD(&info->swaplist);
1296
1297                 switch (mode & S_IFMT) {
1298                 default:
1299                         inode->i_op = &shmem_special_inode_operations;
1300                         init_special_inode(inode, mode, dev);
1301                         break;
1302                 case S_IFREG:
1303                         inode->i_op = &shmem_inode_operations;
1304                         inode->i_fop = &shmem_file_operations;
1305                         mpol_shared_policy_init(&info->policy);
1306                         break;
1307                 case S_IFDIR:
1308                         inode->i_nlink++;
1309                         /* Some things misbehave if size == 0 on a directory */
1310                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1311                         inode->i_op = &shmem_dir_inode_operations;
1312                         inode->i_fop = &simple_dir_operations;
1313                         break;
1314                 case S_IFLNK:
1315                         /*
1316                          * Must not load anything in the rbtree,
1317                          * mpol_free_shared_policy will not be called.
1318                          */
1319                         mpol_shared_policy_init(&info->policy);
1320                         break;
1321                 }
1322         } else if (sbinfo->max_inodes) {
1323                 spin_lock(&sbinfo->stat_lock);
1324                 sbinfo->free_inodes++;
1325                 spin_unlock(&sbinfo->stat_lock);
1326         }
1327         return inode;
1328 }
1329
1330 #ifdef CONFIG_TMPFS
1331 static struct inode_operations shmem_symlink_inode_operations;
1332 static struct inode_operations shmem_symlink_inline_operations;
1333
1334 /*
1335  * Normally tmpfs makes no use of shmem_prepare_write, but it
1336  * lets a tmpfs file be used read-write below the loop driver.
1337  */
1338 static int
1339 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1340 {
1341         struct inode *inode = page->mapping->host;
1342         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1343 }
1344
1345 static ssize_t
1346 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1347 {
1348         struct inode    *inode = file->f_dentry->d_inode;
1349         loff_t          pos;
1350         unsigned long   written;
1351         ssize_t         err;
1352
1353         if ((ssize_t) count < 0)
1354                 return -EINVAL;
1355
1356         if (!access_ok(VERIFY_READ, buf, count))
1357                 return -EFAULT;
1358
1359         down(&inode->i_sem);
1360
1361         pos = *ppos;
1362         written = 0;
1363
1364         err = generic_write_checks(file, &pos, &count, 0);
1365         if (err || !count)
1366                 goto out;
1367
1368         err = remove_suid(file->f_dentry);
1369         if (err)
1370                 goto out;
1371
1372         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1373
1374         do {
1375                 struct page *page = NULL;
1376                 unsigned long bytes, index, offset;
1377                 char *kaddr;
1378                 int left;
1379
1380                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1381                 index = pos >> PAGE_CACHE_SHIFT;
1382                 bytes = PAGE_CACHE_SIZE - offset;
1383                 if (bytes > count)
1384                         bytes = count;
1385
1386                 /*
1387                  * We don't hold page lock across copy from user -
1388                  * what would it guard against? - so no deadlock here.
1389                  * But it still may be a good idea to prefault below.
1390                  */
1391
1392                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1393                 if (err)
1394                         break;
1395
1396                 left = bytes;
1397                 if (PageHighMem(page)) {
1398                         volatile unsigned char dummy;
1399                         __get_user(dummy, buf);
1400                         __get_user(dummy, buf + bytes - 1);
1401
1402                         kaddr = kmap_atomic(page, KM_USER0);
1403                         left = __copy_from_user_inatomic(kaddr + offset,
1404                                                         buf, bytes);
1405                         kunmap_atomic(kaddr, KM_USER0);
1406                 }
1407                 if (left) {
1408                         kaddr = kmap(page);
1409                         left = __copy_from_user(kaddr + offset, buf, bytes);
1410                         kunmap(page);
1411                 }
1412
1413                 written += bytes;
1414                 count -= bytes;
1415                 pos += bytes;
1416                 buf += bytes;
1417                 if (pos > inode->i_size)
1418                         i_size_write(inode, pos);
1419
1420                 flush_dcache_page(page);
1421                 set_page_dirty(page);
1422                 mark_page_accessed(page);
1423                 page_cache_release(page);
1424
1425                 if (left) {
1426                         pos -= left;
1427                         written -= left;
1428                         err = -EFAULT;
1429                         break;
1430                 }
1431
1432                 /*
1433                  * Our dirty pages are not counted in nr_dirty,
1434                  * and we do not attempt to balance dirty pages.
1435                  */
1436
1437                 cond_resched();
1438         } while (count);
1439
1440         *ppos = pos;
1441         if (written)
1442                 err = written;
1443 out:
1444         up(&inode->i_sem);
1445         return err;
1446 }
1447
1448 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1449 {
1450         struct inode *inode = filp->f_dentry->d_inode;
1451         struct address_space *mapping = inode->i_mapping;
1452         unsigned long index, offset;
1453
1454         index = *ppos >> PAGE_CACHE_SHIFT;
1455         offset = *ppos & ~PAGE_CACHE_MASK;
1456
1457         for (;;) {
1458                 struct page *page = NULL;
1459                 unsigned long end_index, nr, ret;
1460                 loff_t i_size = i_size_read(inode);
1461
1462                 end_index = i_size >> PAGE_CACHE_SHIFT;
1463                 if (index > end_index)
1464                         break;
1465                 if (index == end_index) {
1466                         nr = i_size & ~PAGE_CACHE_MASK;
1467                         if (nr <= offset)
1468                                 break;
1469                 }
1470
1471                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1472                 if (desc->error) {
1473                         if (desc->error == -EINVAL)
1474                                 desc->error = 0;
1475                         break;
1476                 }
1477
1478                 /*
1479                  * We must evaluate after, since reads (unlike writes)
1480                  * are called without i_sem protection against truncate
1481                  */
1482                 nr = PAGE_CACHE_SIZE;
1483                 i_size = i_size_read(inode);
1484                 end_index = i_size >> PAGE_CACHE_SHIFT;
1485                 if (index == end_index) {
1486                         nr = i_size & ~PAGE_CACHE_MASK;
1487                         if (nr <= offset) {
1488                                 if (page)
1489                                         page_cache_release(page);
1490                                 break;
1491                         }
1492                 }
1493                 nr -= offset;
1494
1495                 if (page) {
1496                         /*
1497                          * If users can be writing to this page using arbitrary
1498                          * virtual addresses, take care about potential aliasing
1499                          * before reading the page on the kernel side.
1500                          */
1501                         if (mapping_writably_mapped(mapping))
1502                                 flush_dcache_page(page);
1503                         /*
1504                          * Mark the page accessed if we read the beginning.
1505                          */
1506                         if (!offset)
1507                                 mark_page_accessed(page);
1508                 } else
1509                         page = ZERO_PAGE(0);
1510
1511                 /*
1512                  * Ok, we have the page, and it's up-to-date, so
1513                  * now we can copy it to user space...
1514                  *
1515                  * The actor routine returns how many bytes were actually used..
1516                  * NOTE! This may not be the same as how much of a user buffer
1517                  * we filled up (we may be padding etc), so we can only update
1518                  * "pos" here (the actor routine has to update the user buffer
1519                  * pointers and the remaining count).
1520                  */
1521                 ret = actor(desc, page, offset, nr);
1522                 offset += ret;
1523                 index += offset >> PAGE_CACHE_SHIFT;
1524                 offset &= ~PAGE_CACHE_MASK;
1525
1526                 page_cache_release(page);
1527                 if (ret != nr || !desc->count)
1528                         break;
1529
1530                 cond_resched();
1531         }
1532
1533         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1534         file_accessed(filp);
1535 }
1536
1537 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1538 {
1539         read_descriptor_t desc;
1540
1541         if ((ssize_t) count < 0)
1542                 return -EINVAL;
1543         if (!access_ok(VERIFY_WRITE, buf, count))
1544                 return -EFAULT;
1545         if (!count)
1546                 return 0;
1547
1548         desc.written = 0;
1549         desc.count = count;
1550         desc.arg.buf = buf;
1551         desc.error = 0;
1552
1553         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1554         if (desc.written)
1555                 return desc.written;
1556         return desc.error;
1557 }
1558
1559 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1560                          size_t count, read_actor_t actor, void *target)
1561 {
1562         read_descriptor_t desc;
1563
1564         if (!count)
1565                 return 0;
1566
1567         desc.written = 0;
1568         desc.count = count;
1569         desc.arg.data = target;
1570         desc.error = 0;
1571
1572         do_shmem_file_read(in_file, ppos, &desc, actor);
1573         if (desc.written)
1574                 return desc.written;
1575         return desc.error;
1576 }
1577
1578 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1579 {
1580         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1581
1582         buf->f_type = TMPFS_MAGIC;
1583         buf->f_bsize = PAGE_CACHE_SIZE;
1584         buf->f_namelen = NAME_MAX;
1585         spin_lock(&sbinfo->stat_lock);
1586         if (sbinfo->max_blocks) {
1587                 buf->f_blocks = sbinfo->max_blocks;
1588                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1589         }
1590         if (sbinfo->max_inodes) {
1591                 buf->f_files = sbinfo->max_inodes;
1592                 buf->f_ffree = sbinfo->free_inodes;
1593         }
1594         /* else leave those fields 0 like simple_statfs */
1595         spin_unlock(&sbinfo->stat_lock);
1596         return 0;
1597 }
1598
1599 /*
1600  * File creation. Allocate an inode, and we're done..
1601  */
1602 static int
1603 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1604 {
1605         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1606         int error = -ENOSPC;
1607
1608         if (inode) {
1609                 if (dir->i_mode & S_ISGID) {
1610                         inode->i_gid = dir->i_gid;
1611                         if (S_ISDIR(mode))
1612                                 inode->i_mode |= S_ISGID;
1613                 }
1614                 dir->i_size += BOGO_DIRENT_SIZE;
1615                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1616                 d_instantiate(dentry, inode);
1617                 dget(dentry); /* Extra count - pin the dentry in core */
1618                 error = 0;
1619         }
1620         return error;
1621 }
1622
1623 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1624 {
1625         int error;
1626
1627         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1628                 return error;
1629         dir->i_nlink++;
1630         return 0;
1631 }
1632
1633 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1634                 struct nameidata *nd)
1635 {
1636         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1637 }
1638
1639 /*
1640  * Link a file..
1641  */
1642 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1643 {
1644         struct inode *inode = old_dentry->d_inode;
1645         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1646
1647         /*
1648          * No ordinary (disk based) filesystem counts links as inodes;
1649          * but each new link needs a new dentry, pinning lowmem, and
1650          * tmpfs dentries cannot be pruned until they are unlinked.
1651          */
1652         if (sbinfo->max_inodes) {
1653                 spin_lock(&sbinfo->stat_lock);
1654                 if (!sbinfo->free_inodes) {
1655                         spin_unlock(&sbinfo->stat_lock);
1656                         return -ENOSPC;
1657                 }
1658                 sbinfo->free_inodes--;
1659                 spin_unlock(&sbinfo->stat_lock);
1660         }
1661
1662         dir->i_size += BOGO_DIRENT_SIZE;
1663         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1664         inode->i_nlink++;
1665         atomic_inc(&inode->i_count);    /* New dentry reference */
1666         dget(dentry);           /* Extra pinning count for the created dentry */
1667         d_instantiate(dentry, inode);
1668         return 0;
1669 }
1670
1671 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1672 {
1673         struct inode *inode = dentry->d_inode;
1674
1675         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1676                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1677                 if (sbinfo->max_inodes) {
1678                         spin_lock(&sbinfo->stat_lock);
1679                         sbinfo->free_inodes++;
1680                         spin_unlock(&sbinfo->stat_lock);
1681                 }
1682         }
1683
1684         dir->i_size -= BOGO_DIRENT_SIZE;
1685         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1686         inode->i_nlink--;
1687         dput(dentry);   /* Undo the count from "create" - this does all the work */
1688         return 0;
1689 }
1690
1691 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1692 {
1693         if (!simple_empty(dentry))
1694                 return -ENOTEMPTY;
1695
1696         dir->i_nlink--;
1697         return shmem_unlink(dir, dentry);
1698 }
1699
1700 /*
1701  * The VFS layer already does all the dentry stuff for rename,
1702  * we just have to decrement the usage count for the target if
1703  * it exists so that the VFS layer correctly free's it when it
1704  * gets overwritten.
1705  */
1706 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1707 {
1708         struct inode *inode = old_dentry->d_inode;
1709         int they_are_dirs = S_ISDIR(inode->i_mode);
1710
1711         if (!simple_empty(new_dentry))
1712                 return -ENOTEMPTY;
1713
1714         if (new_dentry->d_inode) {
1715                 (void) shmem_unlink(new_dir, new_dentry);
1716                 if (they_are_dirs)
1717                         old_dir->i_nlink--;
1718         } else if (they_are_dirs) {
1719                 old_dir->i_nlink--;
1720                 new_dir->i_nlink++;
1721         }
1722
1723         old_dir->i_size -= BOGO_DIRENT_SIZE;
1724         new_dir->i_size += BOGO_DIRENT_SIZE;
1725         old_dir->i_ctime = old_dir->i_mtime =
1726         new_dir->i_ctime = new_dir->i_mtime =
1727         inode->i_ctime = CURRENT_TIME;
1728         return 0;
1729 }
1730
1731 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1732 {
1733         int error;
1734         int len;
1735         struct inode *inode;
1736         struct page *page = NULL;
1737         char *kaddr;
1738         struct shmem_inode_info *info;
1739
1740         len = strlen(symname) + 1;
1741         if (len > PAGE_CACHE_SIZE)
1742                 return -ENAMETOOLONG;
1743
1744         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1745         if (!inode)
1746                 return -ENOSPC;
1747
1748         info = SHMEM_I(inode);
1749         inode->i_size = len-1;
1750         if (len <= (char *)inode - (char *)info) {
1751                 /* do it inline */
1752                 memcpy(info, symname, len);
1753                 inode->i_op = &shmem_symlink_inline_operations;
1754         } else {
1755                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1756                 if (error) {
1757                         iput(inode);
1758                         return error;
1759                 }
1760                 inode->i_op = &shmem_symlink_inode_operations;
1761                 kaddr = kmap_atomic(page, KM_USER0);
1762                 memcpy(kaddr, symname, len);
1763                 kunmap_atomic(kaddr, KM_USER0);
1764                 set_page_dirty(page);
1765                 page_cache_release(page);
1766         }
1767         if (dir->i_mode & S_ISGID)
1768                 inode->i_gid = dir->i_gid;
1769         dir->i_size += BOGO_DIRENT_SIZE;
1770         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1771         d_instantiate(dentry, inode);
1772         dget(dentry);
1773         return 0;
1774 }
1775
1776 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1777 {
1778         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1779         return 0;
1780 }
1781
1782 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1783 {
1784         struct page *page = NULL;
1785         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1786         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1787         return 0;
1788 }
1789
1790 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1791 {
1792         if (!IS_ERR(nd_get_link(nd))) {
1793                 struct page *page;
1794
1795                 page = find_get_page(dentry->d_inode->i_mapping, 0);
1796                 if (!page)
1797                         BUG();
1798                 kunmap(page);
1799                 mark_page_accessed(page);
1800                 page_cache_release(page);
1801                 page_cache_release(page);
1802         }
1803 }
1804
1805 static struct inode_operations shmem_symlink_inline_operations = {
1806         .readlink       = generic_readlink,
1807         .follow_link    = shmem_follow_link_inline,
1808 #ifdef CONFIG_TMPFS_XATTR
1809         .setxattr       = generic_setxattr,
1810         .getxattr       = generic_getxattr,
1811         .listxattr      = generic_listxattr,
1812         .removexattr    = generic_removexattr,
1813 #endif
1814 };
1815
1816 static struct inode_operations shmem_symlink_inode_operations = {
1817         .truncate       = shmem_truncate,
1818         .readlink       = generic_readlink,
1819         .follow_link    = shmem_follow_link,
1820         .put_link       = shmem_put_link,
1821 #ifdef CONFIG_TMPFS_XATTR
1822         .setxattr       = generic_setxattr,
1823         .getxattr       = generic_getxattr,
1824         .listxattr      = generic_listxattr,
1825         .removexattr    = generic_removexattr,
1826 #endif
1827 };
1828
1829 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1830 {
1831         char *this_char, *value, *rest;
1832
1833         while ((this_char = strsep(&options, ",")) != NULL) {
1834                 if (!*this_char)
1835                         continue;
1836                 if ((value = strchr(this_char,'=')) != NULL) {
1837                         *value++ = 0;
1838                 } else {
1839                         printk(KERN_ERR
1840                             "tmpfs: No value for mount option '%s'\n",
1841                             this_char);
1842                         return 1;
1843                 }
1844
1845                 if (!strcmp(this_char,"size")) {
1846                         unsigned long long size;
1847                         size = memparse(value,&rest);
1848                         if (*rest == '%') {
1849                                 size <<= PAGE_SHIFT;
1850                                 size *= totalram_pages;
1851                                 do_div(size, 100);
1852                                 rest++;
1853                         }
1854                         if (*rest)
1855                                 goto bad_val;
1856                         *blocks = size >> PAGE_CACHE_SHIFT;
1857                 } else if (!strcmp(this_char,"nr_blocks")) {
1858                         *blocks = memparse(value,&rest);
1859                         if (*rest)
1860                                 goto bad_val;
1861                 } else if (!strcmp(this_char,"nr_inodes")) {
1862                         *inodes = memparse(value,&rest);
1863                         if (*rest)
1864                                 goto bad_val;
1865                 } else if (!strcmp(this_char,"mode")) {
1866                         if (!mode)
1867                                 continue;
1868                         *mode = simple_strtoul(value,&rest,8);
1869                         if (*rest)
1870                                 goto bad_val;
1871                 } else if (!strcmp(this_char,"uid")) {
1872                         if (!uid)
1873                                 continue;
1874                         *uid = simple_strtoul(value,&rest,0);
1875                         if (*rest)
1876                                 goto bad_val;
1877                 } else if (!strcmp(this_char,"gid")) {
1878                         if (!gid)
1879                                 continue;
1880                         *gid = simple_strtoul(value,&rest,0);
1881                         if (*rest)
1882                                 goto bad_val;
1883                 } else {
1884                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1885                                this_char);
1886                         return 1;
1887                 }
1888         }
1889         return 0;
1890
1891 bad_val:
1892         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1893                value, this_char);
1894         return 1;
1895
1896 }
1897
1898 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1899 {
1900         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1901         unsigned long max_blocks = sbinfo->max_blocks;
1902         unsigned long max_inodes = sbinfo->max_inodes;
1903         unsigned long blocks;
1904         unsigned long inodes;
1905         int error = -EINVAL;
1906
1907         if (shmem_parse_options(data, NULL, NULL, NULL,
1908                                 &max_blocks, &max_inodes))
1909                 return error;
1910
1911         spin_lock(&sbinfo->stat_lock);
1912         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
1913         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1914         if (max_blocks < blocks)
1915                 goto out;
1916         if (max_inodes < inodes)
1917                 goto out;
1918         /*
1919          * Those tests also disallow limited->unlimited while any are in
1920          * use, so i_blocks will always be zero when max_blocks is zero;
1921          * but we must separately disallow unlimited->limited, because
1922          * in that case we have no record of how much is already in use.
1923          */
1924         if (max_blocks && !sbinfo->max_blocks)
1925                 goto out;
1926         if (max_inodes && !sbinfo->max_inodes)
1927                 goto out;
1928
1929         error = 0;
1930         sbinfo->max_blocks  = max_blocks;
1931         sbinfo->free_blocks = max_blocks - blocks;
1932         sbinfo->max_inodes  = max_inodes;
1933         sbinfo->free_inodes = max_inodes - inodes;
1934 out:
1935         spin_unlock(&sbinfo->stat_lock);
1936         return error;
1937 }
1938 #endif
1939
1940 static void shmem_put_super(struct super_block *sb)
1941 {
1942         kfree(sb->s_fs_info);
1943         sb->s_fs_info = NULL;
1944 }
1945
1946 #ifdef CONFIG_TMPFS_XATTR
1947 static struct xattr_handler *shmem_xattr_handlers[];
1948 #else
1949 #define shmem_xattr_handlers NULL
1950 #endif
1951
1952 static int shmem_fill_super(struct super_block *sb,
1953                             void *data, int silent)
1954 {
1955         struct inode *inode;
1956         struct dentry *root;
1957         int mode   = S_IRWXUGO | S_ISVTX;
1958         uid_t uid = current->fsuid;
1959         gid_t gid = current->fsgid;
1960         int err = -ENOMEM;
1961         struct shmem_sb_info *sbinfo;
1962         unsigned long blocks = 0;
1963         unsigned long inodes = 0;
1964
1965 #ifdef CONFIG_TMPFS
1966         /*
1967          * Per default we only allow half of the physical ram per
1968          * tmpfs instance, limiting inodes to one per page of lowmem;
1969          * but the internal instance is left unlimited.
1970          */
1971         if (!(sb->s_flags & MS_NOUSER)) {
1972                 blocks = totalram_pages / 2;
1973                 inodes = totalram_pages - totalhigh_pages;
1974                 if (inodes > blocks)
1975                         inodes = blocks;
1976                 if (shmem_parse_options(data, &mode, &uid, &gid,
1977                                         &blocks, &inodes))
1978                         return -EINVAL;
1979         }
1980 #else
1981         sb->s_flags |= MS_NOUSER;
1982 #endif
1983
1984         /* Round up to L1_CACHE_BYTES to resist false sharing */
1985         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
1986                                 L1_CACHE_BYTES), GFP_KERNEL);
1987         if (!sbinfo)
1988                 return -ENOMEM;
1989
1990         spin_lock_init(&sbinfo->stat_lock);
1991         sbinfo->max_blocks = blocks;
1992         sbinfo->free_blocks = blocks;
1993         sbinfo->max_inodes = inodes;
1994         sbinfo->free_inodes = inodes;
1995
1996         sb->s_fs_info = sbinfo;
1997         sb->s_maxbytes = SHMEM_MAX_BYTES;
1998         sb->s_blocksize = PAGE_CACHE_SIZE;
1999         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2000         sb->s_magic = TMPFS_MAGIC;
2001         sb->s_op = &shmem_ops;
2002         sb->s_xattr = shmem_xattr_handlers;
2003
2004         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2005         if (!inode)
2006                 goto failed;
2007         inode->i_uid = uid;
2008         inode->i_gid = gid;
2009         root = d_alloc_root(inode);
2010         if (!root)
2011                 goto failed_iput;
2012         sb->s_root = root;
2013         return 0;
2014
2015 failed_iput:
2016         iput(inode);
2017 failed:
2018         shmem_put_super(sb);
2019         return err;
2020 }
2021
2022 static kmem_cache_t *shmem_inode_cachep;
2023
2024 static struct inode *shmem_alloc_inode(struct super_block *sb)
2025 {
2026         struct shmem_inode_info *p;
2027         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2028         if (!p)
2029                 return NULL;
2030         return &p->vfs_inode;
2031 }
2032
2033 static void shmem_destroy_inode(struct inode *inode)
2034 {
2035         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2036                 /* only struct inode is valid if it's an inline symlink */
2037                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2038         }
2039         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2040 }
2041
2042 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
2043 {
2044         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2045
2046         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2047             SLAB_CTOR_CONSTRUCTOR) {
2048                 inode_init_once(&p->vfs_inode);
2049         }
2050 }
2051
2052 static int init_inodecache(void)
2053 {
2054         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2055                                 sizeof(struct shmem_inode_info),
2056                                 0, 0, init_once, NULL);
2057         if (shmem_inode_cachep == NULL)
2058                 return -ENOMEM;
2059         return 0;
2060 }
2061
2062 static void destroy_inodecache(void)
2063 {
2064         if (kmem_cache_destroy(shmem_inode_cachep))
2065                 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2066 }
2067
2068 static struct address_space_operations shmem_aops = {
2069         .writepage      = shmem_writepage,
2070         .set_page_dirty = __set_page_dirty_nobuffers,
2071 #ifdef CONFIG_TMPFS
2072         .prepare_write  = shmem_prepare_write,
2073         .commit_write   = simple_commit_write,
2074 #endif
2075 };
2076
2077 static struct file_operations shmem_file_operations = {
2078         .mmap           = shmem_mmap,
2079 #ifdef CONFIG_TMPFS
2080         .llseek         = generic_file_llseek,
2081         .read           = shmem_file_read,
2082         .write          = shmem_file_write,
2083         .fsync          = simple_sync_file,
2084         .sendfile       = shmem_file_sendfile,
2085 #endif
2086 };
2087
2088 static struct inode_operations shmem_inode_operations = {
2089         .truncate       = shmem_truncate,
2090         .setattr        = shmem_notify_change,
2091 #ifdef CONFIG_TMPFS_XATTR
2092         .setxattr       = generic_setxattr,
2093         .getxattr       = generic_getxattr,
2094         .listxattr      = generic_listxattr,
2095         .removexattr    = generic_removexattr,
2096 #endif
2097 };
2098
2099 static struct inode_operations shmem_dir_inode_operations = {
2100 #ifdef CONFIG_TMPFS
2101         .create         = shmem_create,
2102         .lookup         = simple_lookup,
2103         .link           = shmem_link,
2104         .unlink         = shmem_unlink,
2105         .symlink        = shmem_symlink,
2106         .mkdir          = shmem_mkdir,
2107         .rmdir          = shmem_rmdir,
2108         .mknod          = shmem_mknod,
2109         .rename         = shmem_rename,
2110 #ifdef CONFIG_TMPFS_XATTR
2111         .setxattr       = generic_setxattr,
2112         .getxattr       = generic_getxattr,
2113         .listxattr      = generic_listxattr,
2114         .removexattr    = generic_removexattr,
2115 #endif
2116 #endif
2117 };
2118
2119 static struct inode_operations shmem_special_inode_operations = {
2120 #ifdef CONFIG_TMPFS_XATTR
2121         .setxattr       = generic_setxattr,
2122         .getxattr       = generic_getxattr,
2123         .listxattr      = generic_listxattr,
2124         .removexattr    = generic_removexattr,
2125 #endif
2126 };
2127
2128 static struct super_operations shmem_ops = {
2129         .alloc_inode    = shmem_alloc_inode,
2130         .destroy_inode  = shmem_destroy_inode,
2131 #ifdef CONFIG_TMPFS
2132         .statfs         = shmem_statfs,
2133         .remount_fs     = shmem_remount_fs,
2134 #endif
2135         .delete_inode   = shmem_delete_inode,
2136         .drop_inode     = generic_delete_inode,
2137         .put_super      = shmem_put_super,
2138 };
2139
2140 static struct vm_operations_struct shmem_vm_ops = {
2141         .nopage         = shmem_nopage,
2142         .populate       = shmem_populate,
2143 #ifdef CONFIG_NUMA
2144         .set_policy     = shmem_set_policy,
2145         .get_policy     = shmem_get_policy,
2146 #endif
2147 };
2148
2149
2150 #ifdef CONFIG_TMPFS_SECURITY
2151
2152 static size_t shmem_xattr_security_list(struct inode *inode, char *list, size_t list_len,
2153                                         const char *name, size_t name_len)
2154 {
2155         return security_inode_listsecurity(inode, list, list_len);
2156 }
2157
2158 static int shmem_xattr_security_get(struct inode *inode, const char *name, void *buffer, size_t size)
2159 {
2160         if (strcmp(name, "") == 0)
2161                 return -EINVAL;
2162         return security_inode_getsecurity(inode, name, buffer, size);
2163 }
2164
2165 static int shmem_xattr_security_set(struct inode *inode, const char *name, const void *value, size_t size, int flags)
2166 {
2167         if (strcmp(name, "") == 0)
2168                 return -EINVAL;
2169         return security_inode_setsecurity(inode, name, value, size, flags);
2170 }
2171
2172 static struct xattr_handler shmem_xattr_security_handler = {
2173         .prefix = XATTR_SECURITY_PREFIX,
2174         .list   = shmem_xattr_security_list,
2175         .get    = shmem_xattr_security_get,
2176         .set    = shmem_xattr_security_set,
2177 };
2178
2179 #endif  /* CONFIG_TMPFS_SECURITY */
2180
2181 #ifdef CONFIG_TMPFS_XATTR
2182
2183 static struct xattr_handler *shmem_xattr_handlers[] = {
2184 #ifdef CONFIG_TMPFS_SECURITY
2185         &shmem_xattr_security_handler,
2186 #endif
2187         NULL
2188 };
2189
2190 #endif  /* CONFIG_TMPFS_XATTR */
2191
2192 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2193         int flags, const char *dev_name, void *data)
2194 {
2195         return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2196 }
2197
2198 static struct file_system_type tmpfs_fs_type = {
2199         .owner          = THIS_MODULE,
2200         .name           = "tmpfs",
2201         .get_sb         = shmem_get_sb,
2202         .kill_sb        = kill_litter_super,
2203 };
2204 static struct vfsmount *shm_mnt;
2205
2206 static int __init init_tmpfs(void)
2207 {
2208         int error;
2209
2210         error = init_inodecache();
2211         if (error)
2212                 goto out3;
2213
2214         error = register_filesystem(&tmpfs_fs_type);
2215         if (error) {
2216                 printk(KERN_ERR "Could not register tmpfs\n");
2217                 goto out2;
2218         }
2219 #ifdef CONFIG_TMPFS
2220         devfs_mk_dir("shm");
2221 #endif
2222         shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2223                                 tmpfs_fs_type.name, NULL);
2224         if (IS_ERR(shm_mnt)) {
2225                 error = PTR_ERR(shm_mnt);
2226                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2227                 goto out1;
2228         }
2229         return 0;
2230
2231 out1:
2232         unregister_filesystem(&tmpfs_fs_type);
2233 out2:
2234         destroy_inodecache();
2235 out3:
2236         shm_mnt = ERR_PTR(error);
2237         return error;
2238 }
2239 module_init(init_tmpfs)
2240
2241 /*
2242  * shmem_file_setup - get an unlinked file living in tmpfs
2243  *
2244  * @name: name for dentry (to be seen in /proc/<pid>/maps
2245  * @size: size to be set for the file
2246  *
2247  */
2248 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2249 {
2250         int error;
2251         struct file *file;
2252         struct inode *inode;
2253         struct dentry *dentry, *root;
2254         struct qstr this;
2255
2256         if (IS_ERR(shm_mnt))
2257                 return (void *)shm_mnt;
2258
2259         if (size < 0 || size > SHMEM_MAX_BYTES)
2260                 return ERR_PTR(-EINVAL);
2261
2262         if (shmem_acct_size(flags, size))
2263                 return ERR_PTR(-ENOMEM);
2264
2265         error = -ENOMEM;
2266         this.name = name;
2267         this.len = strlen(name);
2268         this.hash = 0; /* will go */
2269         root = shm_mnt->mnt_root;
2270         dentry = d_alloc(root, &this);
2271         if (!dentry)
2272                 goto put_memory;
2273
2274         error = -ENFILE;
2275         file = get_empty_filp();
2276         if (!file)
2277                 goto put_dentry;
2278
2279         error = -ENOSPC;
2280         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2281         if (!inode)
2282                 goto close_file;
2283
2284         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2285         d_instantiate(dentry, inode);
2286         inode->i_size = size;
2287         inode->i_nlink = 0;     /* It is unlinked */
2288         file->f_vfsmnt = mntget(shm_mnt);
2289         file->f_dentry = dentry;
2290         file->f_mapping = inode->i_mapping;
2291         file->f_op = &shmem_file_operations;
2292         file->f_mode = FMODE_WRITE | FMODE_READ;
2293         return file;
2294
2295 close_file:
2296         put_filp(file);
2297 put_dentry:
2298         dput(dentry);
2299 put_memory:
2300         shmem_unacct_size(flags, size);
2301         return ERR_PTR(error);
2302 }
2303
2304 /*
2305  * shmem_zero_setup - setup a shared anonymous mapping
2306  *
2307  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2308  */
2309 int shmem_zero_setup(struct vm_area_struct *vma)
2310 {
2311         struct file *file;
2312         loff_t size = vma->vm_end - vma->vm_start;
2313
2314         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2315         if (IS_ERR(file))
2316                 return PTR_ERR(file);
2317
2318         if (vma->vm_file)
2319                 fput(vma->vm_file);
2320         vma->vm_file = file;
2321         vma->vm_ops = &shmem_vm_ops;
2322         return 0;
2323 }